Sources and Studies in the History of Mathematics and Physical Sciences

Editorial Board

L. Berggren J.Z. Buchwald J. L¨utzen

Advisory Board

P.J. Davis T. Hawkins A.E. Shapiro D. Whiteside

Sources and Studies in the History of Mathematics and Physical Sciences K. Andersen Brook Taylor’s Work on Linear Perspective K. Andersen The Geometry of an Art: The History of the Mathematical Theory of Perspective from Alberti to Monge H.J.M. Bos Redefining Geometrical Exactness: Descartes’ Transformation of the Early Modern Concept of Construction J. Cannon/S. Dostrovsky The Evolution of Dynamics: Vibration Theory from 1687 to 1742 B. Chandler/W. Magnus The History of Combinatorial Group Theory A.I. Dale A History of Inverse Probability: From Thomas Bayes to Karl Pearson, Second Edition A.I. Dale Most Honourable Remembrance: The Life and Work of Thomas Bayes A.I. Dale Pierre-Simon de Laplace, Philosophical Essay on Probabilities, Translated from the fifth French edition of 1825, with Notes by the Translator P. Damerow/G. Freudenthal/P. McLaughlin/J. Renn Exploring the Limits of Preclassical Mechanics: A Study of Conceptual Development in Early Modern Science: Free Fall and Compounded Motion in the Work of Descartes. Galileo, and Beeckman, Second Edition P.J. Federico Descartes on Polyhedra: A Study of the De Solidorum Elementis G. Ferraro The Rise and Development of the Theory of Series up to the Early 1820s J. Friberg A Remarkable Collection of Babylonian Mathematical Texts B.R. Goldstein The Astronomy of Levi Ben Gerson (1288–1344) H.H. Goldstine A History of Numerical Analysis from the 16th Through the 19th Century H.H. Goldstine A History of the Calculus of Variations from the 17th Through the 19th Century G. Graßhoff The History of Ptolemy’s Star Catalogue Continued after Subject Index

Giovanni Ferraro

The Rise and Development of the Theory of Series up to the Early 1820s

Giovanni Ferraro Universit`a del Molise Dipartimento STAT c. da Fonte Lappone 86090 Pesche Isernia, Italy [emailprotected]

Sources and Series Editor: Jesper L¨utzen Institute for Mathematical Sciences University of Copenhagen DK-2100 Copenhagen Denmark

ISBN: 978-0-387-73467-5

e-ISBN: 978-0-387-73468-2

Library of Congress Control Number: 2007939827 Mathematics Subject Classification (2000): 01Axx 01A50 26-03 40xx c 2008 Springer Science+Business Media, LLC All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, 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 hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

A Pina, Maria Grazia, Serena, Giuseppe

Preface The theory of series in the 17th and 18th centuries poses several interesting problems to historians. Indeed, mathematicians of the time derived numerous results that range from the binomial theorem to the Taylor formula, from the power series expansions of elementary functions to trigonometric series, from Stirling’s series to series solution of diﬀerential equations, from the Euler–Maclaurin summation formula to the Lagrange inversion theorem, from Laplace’s theory of generating functions to the calculus of operations, etc. Most of these results were, however, derived using methods that would be found unacceptable today, thus, if we look back to the theory of series prior to Cauchy without reconstructing internal motivations and the conceptual background, it appears as a corpus of manipulative techniques lacking in rigor whose results seem to be the puzzling fruit of the mind of a magician or diviner rather than the penetrating and complex work of great mathematicians. For this reason, in this monograph, not only do I describe the entire complex of 17th- and 18th-century procedures and results concerning series, but also I reconstruct the implicit and explicit principles upon which they are based, draw attention to the underlying philosophy, highlight competing approaches, and investigate the mathematical context where the series theory originated. My aim is to improve the understanding of the framework of 17th- and 18th-century mathematics and avoid trivializing the complexity of historical development by bringing it into line with modern concepts and views and by tacitly assuming that certain results belong, in some unproblematic sense, to a uniﬁed theory that has come down to us today. The initial and ﬁnal points of my monograph require some clariﬁcation. The point of departure is the publication of a paper by Vi`ete, Variorum de rebus mathematicis responsorum. Liber VIII (1593), where geometrical series are discussed and π is expressed in the form of an inﬁnite product. Even though previous tracks of inﬁnite series can be found, Vi`ete’s paper, when considered in the context of the new rising symbolic algebra, appears to be a step forward in a path –very slow to begin with, but that developed much more rapidly after 1650– that has made series an essential instrument in mathematics. The point of arrival is the early 1820s when Cauchy ´ published Cours d’analyse and R´esum´e des le¸cons donn´ees a ` l’Ecole Royale Polytechnique sur le calcul inﬁnit´esimal, which can be considered to mark the deﬁnitive abandonment of the 18th-century formal approach to the series theory. My main arguments can be summarised as follows. The mathematicians who ﬁrst used series were interested in their capacity to represent geometrical quantities and had an intuitive idea of convergence. They thought that a series represented a quantity and had a quantitative meaning if, and only if, vii

viii

Preface

it was convergent to this quantity. However, a distinction between ﬁnite and inﬁnite sums was lacking, and this gave rise to formal procedures consisting of the inﬁnite extension of ﬁnite procedures. In the works of mathematicians such as Newton and Leibniz, the quantitative and the formal aspect coexisted and formal manipulations were a tool for deriving convergent series. As from the 1720s, several results began to upset the previously established balance between the quantitative and the formal. Mathematicians introduced recurrent series, which stressed the law of formation of coeﬃcients, independently of the convergence of series. The attempt to improve the acceleration of series subsequently led to the emergence of asymptotic series, which showed the possibility of using divergent series to obtain appropriate approximations. Furthermore, the investigation of continued fractions and inﬁnite products and certain applications of series (for instance, in numerical analysis and in number theory) increasingly stressed the formal aspects. In this context, Euler oﬀered a unitary interpretation of the complex of results concerning series, which even allowed the acceptance of those ﬁndings that did not form part of the early theory. A series was thought to be the result of a formal transformation of an analytical quantity expressed in a closed form. This transformation was considered suﬃcient to give a meaning to the series, even when the latter was not convergent. However, mathematicians were not free to invent transformations by a free creative act. They limited themselves to using the same transformations that were used in the original theory or at least were compatible with it. This seemed to guarantee that the new more formal conception was a generalization of earlier conception, which remained the essential basis from which all the parts of the series theory were subsequently generated. The more formal Eulerian approach was widely predominant during the second part of the 18th-century for two main reasons. First, mathematicians who were critical of it were not able to eliminate the formal aspects of the early concept and found a really new theory: They always used the formal methodology that had led to asymptotic series and to the combinatorial use of series. Second, the formal concept of series contributed to the growth of mathematics. It led to many new discoveries and even to a new branch of analysis: the calculus of operations. The formal approach became unsuited to most advanced mathematical research toward the end of the 18th century and the beginning of the 19th century. Applied mathematics encouraged investigations and introduction of new functions in analysis, but formal methodology was unable to treat quantities that were not elementary quantities and series that were not power series. The need to use trigonometric series to enable the analytical investigation of heat led Fourier to reject the formal concept of series and assert an entirely quantitative notion of series. Similarly, the need to introduce hypergeometric and gamma functions into analysis and to have an adequate analytical theory of them forced Gauss to highlight the quantitative meaning

Preface

ix

of the sum of series and to reject formal manipulations. The new approach based only upon convergence was the basis of Cauchy’s treatises. Given the purposes of this book, I cannot avoid dealing with some topics that are closely connected to series theory and are crucial to an understanding of its historical evolution: Not only do these include other inﬁnite processes (continued fractions and inﬁnite products) but also certain basic mathematical notions (quantity, numbers, functions) and the 18th-century concept of analysis. This book is divided into four parts. The ﬁrst part starts with a chapter devoted to the use of series prior to the rise of the calculus (Chapter 1), where I deal principally with Vi`ete, Gr´egoire de Saint-Vincent, Mengoli, Wallis, and Gregory. I then move on to investigate the conception of the founders of the calculus (Leibniz in Chapter 2; Newton in Chapter 4). On the basis of this examination, and after discussing the contributions of Johann and Jacob Bernoulli (Chapters 3 and 5) and the notion of a quantity and of a number (Chapter 7), I oﬀer an interpretative scheme of the early theory of series in Chapter 8. The ﬁrst part also includes the appearance of Taylor series in Newton and Taylor (Chapter 6) and the rise of the problem of the sum of a divergent series in one of Grandi’s writings and the ensuing debate in Leibniz, Varignon, Daniel Bernoulli, and Goldbach (Chapter 9). In the second part, I illustrate the development of series theory from the 1720s to the 1750s. De Moivre’s recurrent series and Bernoulli’s method for solving equations are the subject of Chapter 10. Chapter 11 deals with the attempt to improve the acceleration of series and Stirling’s series, the ﬁrst example of asymptotic series. Chapter 12 examines the geometric conception of Colin Maclaurin. Most of the second part is devoted to Euler, “the master of all us,” to use an expression that Libri [1846, 51] ascribes to Laplace. From 1730 to 1750, Euler obtained many important results, which I examine in Chapters 13 to 17. In particular, I shall concentrate on the problem of interpolation and some of Euler’s ﬁrst ﬁndings (Chapter 13), on Euler’s derivation of the Euler–Maclaurin summation formula (Chapter 14), on issues connected to the interpretation of asymptotic series (Chapter 15), on the theory of inﬁnite products and continued fractions (Chapter 16), and on the application of series to number theory (Chapter 17). Chapter 18 is a digression on some basic principles of analysis during the period from the 1740s to the 1810s, which is essential for understanding series theory in the second half of the 18th century. In particular, the relationship between analysis and geometry, the notion of a function, and the principle of generality of algebra are examined. In Chapter 19, I discuss some criticisms of certain procedures and how Euler rejected them by giving a merely formal interpretation of the notion of the sum. The third part is devoted to the period when formal conception held undisputed sway. I begin by illustrating some of the greatest successes of the formal approach during the second part of the 18th century: the La-

x

Preface

grange inversion theorem, which is discussed in Chapter 20, the calculus of operations, examined in Chapter 21, and Laplace’s theory of generating functions (the subject of Chapter 22). The problem of the representation of transcendental quantities and their analytical investigation is treated in Chapters 23, 24, and 25. Integration by series and series solutions to diﬀerential equations were already known by the beginning of the calculus, but they underwent a remarkable development after Euler: Some examples from Euler, Laplace and Legendre are given in Chapter 26. I then deal with trigonometric series for which mathematicians applied the same procedure as that used for power series. This prevented them from being fully understood (Chapter 27). The attempts to prove the binomial theorem, Lagrange’s view of the Taylor theorem, and other signiﬁcant developments that took place between the end of the 18th century and the beginning of 19th century are the subject matter of Chapter 28. Chapters 29 and 30 focus on the problematic attempt of Legendre to enlarge the realm of accepted functions and to the emergence of techniques of inequalities in d’Alembert’s and Lagrange’s work. The fourth and ﬁnal part is devoted to the crisis in formal methods. It deals with Fourier’s investigations of Fourier series (Chapter 31), Gauss’s work on hypergeometric and gamma functions (Chapter 32), and Cauchy’s contributions on series during the early 1820s (Chapter 33). The conceptions of these mathematicians diﬀer from all other mathematicians discussed in this book since they belong to a new historical phase. However, the discussion of their approach allows me to illustrate some hypotheses about the abandonment of 18th-century series theory. In order to write this monograph I have drawn on various papers of mine, in particular: Some parts of “True and ﬁctitious quantities in Leibniz’s theory of series”, published in Studia Leibnitiana, 32 (2000), pp. 43–67 (copyright Franz Steiner Verlag GmbH, Stuttgart) are reproduced in Chapters 2, 3, and 9. Some parts of “Functions, functional relations and the laws of continuity in Euler”, published in Historia Mathematica, 27 (2000), pp. 107– 132 (copyright Elsevier), and “Analytical symbols and geometrical ﬁgures. Eighteenth century analysis as nonﬁgural geometry”, published in Studies in History and Philosophy of Science Part A, 32 (2001), pp. 535–555 (copyright Elsevier), are reproduced in Chapter 18. Some parts of “Some aspects of Euler’s theory of series. Inexplicable functions and the Euler–Maclaurin summation formula”, published in Historia Mathematica, 25 (1998), pp. 290–317 (copyright Elsevier), are reproduced in Chapters 13, 14, and 24.

Preface

xi

Some parts of “The value of an inﬁnite sum. Some observations on the Eulerian theory of series”, published in Sciences et Techniques en Perspective, 4 (2000), pp. 73–113, are reproduced in Chapters 15 and 19. Some parts of “Convergence and formal manipulation in the theory of series from 1730 to 1815”, published in Historia Mathematica, 34 (2007) pp. 62–88 (copyright Elsevier), are reproduced in Chapters 26, 27 and 31. Some parts of “The foundational aspects of Gauss’s work on the hypergeometric, factorial and digamma functions”, published in Archive for History of Exact Sciences, 61 (2007), 457-518 (copyright SpringerVerlag) are reproduced in Chapters 29, 30, and 32. I would like to thank Studia Leibnitiana, Sciences et Techniques en Perspective, Studies in History and Philosophy of Science, Historia Mathematica, and Archive for History of Exact Sciences for their permission to include material from the above-mentioned articles. Finally, I would like to thank Craig Fraser and Jesper L¨ utzen for their suggestions that have been helpful in the preparation of this volume. Afragola, Italy May 2007

Contents I

From the beginnings of the 17th century to about 1720: Convergence and formal manipulation . . . . 1

1 Series before the rise of the calculus . . . . . . . . . . . . .

3

2 Geometrical quantities and series in Leibniz . . . . . . . . 25 2.1 The capacity of series to express quantities and their manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.2 Power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3 The Bernoulli series and Leibniz’s analogy . . . . . . . . .

45

4 Newton’s method of series . . . . . . . . . . . . . . . . . . . 53 4.1 The expansion of quantities into convergent series . . . . . . . 54 4.2 On Newton’s manipulations of power series . . . . . . . . . . 67 5 Jacob Bernoulli’s treatise on series . . . . . . . . . . . . . .

79

6 The Taylor series . . . . . . . . . . . . . . . . . . . . . . . . .

87

7 Quantities and their representations . . . . . . . . . . . . . 93 7.1 Quantity and abstract quantity . . . . . . . . . . . . . . . . . 93 7.2 Continuous quantities, numbers and ﬁctitious quantities . . . 100 8 The formal-quantitative theory of series . . . . . . . . . . .

115

9 The ﬁrst appearance of divergent series . . . . . . . . . . .

121

II

From the 1720s to the 1760s: The development of a more formal conception . . . . . . . . . . . . . . 131

10 De Moivre’s recurrent series and Bernoulli’s method . .

133

11 Acceleration of series and Stirling’s series . . . . . . . . . .

141

12 Maclaurin’s contribution . . . . . . . . . . . . . . . . . . . . .

147

13 The 13.1 13.2 13.3

young Euler between innovation and tradition The search for the general term . . . . . . . . . . . . Analytical and synthetical methods in series theory . The manipulation of the harmonic series and inﬁnite equations . . . . . . . . . . . . . . . . . . . . . . . . xiii

. . . . 155 . . . . . 155 . . . . . 160 . . . . . 165

xiv

Contents

14 Euler’s derivation of the Euler–Maclaurin summation formula . . . . . . . . . . . . . . . . . . . . . . . .

171

15 On the sum of an asymptotic series

. . . . . . . . . . . . .

181

16 Inﬁnite products and continued fractions . . . . . . . . . .

185

17 Series and number theory . . . . . . . . . . . . . . . . . . . .

193

18 Analysis after the 1740s . . . . . . . . . . . . . . . . . . . 18.1 Eighteenth-century analysis as nonﬁgural and symbolic investigation of the real . . . . . . . . . . . . . . . . . . 18.2 Functions, relations, and analytical expressions . . . . . 18.3 On the continuity of curves and functions . . . . . . . .

201

. . . 201 . . . 205 . . . 211

19 The 19.1 19.2 19.3

. . . .

III

formal concept of series . . . . . . . . . . . . Criticisms to the inﬁnite extension of ﬁnite rules The impossibility of the quantitative approach . Euler’s deﬁnition of the sum . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. .

. 215 . . 215 . . 219 . . 222

The theory of series after 1760: Successes and problems of the triumphant formalism . . . . . 231

20 Lagrange inversion theorem . . . . . . . . . . . . . . . . . . .

233

21 Toward the calculus of operations . . . . . . . . . . . . . . .

239

22 Laplace’s calculus of generating functions . . . . . . . . . .

245

23 The problem of analytical representation of nonelementary quantities . . . . . . . . . . . . . . . . . . .

251

24 Inexplicable functions . . . . . . . . . . . . . . . . . . . . . .

257

25 Integration and functions . . . . . . . . . . . . . . . . . . . .

263

26 Series and diﬀerential equations . . . . . . . . . . . . . . . .

267

27 Trigonometric series . . . . . . . . . . . . . . . . . . . . . . .

275

28 Further developments of the formal theory of series . . .

283

29 Attempts to introduce new transcendental functions . . .

297

30 D’Alembert and Lagrange and the inequality technique .

303

Contents

IV

xv

The decline of the formal theory of series . . . . . . 311

31 Fourier and Fourier series . . . . . . . . . . . . . . . . . . . .

315

32 Gauss and the hypergeometric series . . . . . . . . . . . . .

323

33 Cauchy’s rejection of the 18th-century theory of series .

347

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

363

Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

383

Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

387

1

Series before the rise of the calculus

Even though series were occasionally found earlier, it is only from the 17th century that they began to be a topic of importance in mathematics. Their use mainly arose in the context of the problem of quadratures and rectiﬁcations of curves. During the 17th century, mathematicians attempted to ﬁnd new methods for squaring curved lines, which avoided the diﬃculty of the so-called method of exhaustion.4 This method, which had been one of the greatest successes of Greek geometry, made it possible to determine the area A of a given ﬁgure by means of a complex procedure divisible in two phases. 1. One or two sequences of polygons were constructed so that the areas of these polygons approximated to the given ﬁgure and suggested that the sought area A was equal to a certain area P . 2. One proved A = P by means of a double reductio ad absurdum (namely, one showed that neither A > P nor A < P was true). A classic example is the quadrature of the parabolic segment obtained by Archimedes5 . As the ﬁrst step in the proof, one considers the triangle ABC with area F , which is greater than one-half of the parabolic segment ACB with area P (see Fig. 1). Then, one considers the diameters B1 V1 and B1 V2 such that AV1 = AV2 = AH 2 and constructs the triangles AB1 C and BB2 C. These triangles are greater than one-half of the corresponding parabolic segments AB1 C and BB2 C. Moreover, both the triangles AB1 C and BB2 C are equal to 18 F and, consequently, their sum is 14 F . The process can be continued so as to construct a sequence Hn such that • Hn is a polygon formed by the sum of the triangles, • at the nth step, the area of Hn is

1 F, 4n−1

• the polygons Hn exhaust (namely, ﬁll up entirely) the segment, • the sum Sn of the areas of all the triangles up to the nth step is given by the ﬁnite geometric progression Sn = F +

F F F F + + + . . . + n−1 . 4 16 64 4

(1)

After having shown that Sn +

1 F 4 = F, 3 4n−1 3

Archimedes proved that the area of the parabolic segment is reasoning as follows. 4 5

The name is due to Gr´egoire de Saint-Vincent [1647, 740]. See Archimedes [QA, 233–252].

3

(2) 4 3F

by

4

Convergence and Formal Manipulation

C B1

A

B2

V1

H

V2

B

Fig. 1

• If P > 34 F , then P − 34 F > 0 and one can continue the exhaustion process until one obtains a sum Sn such that P −Sn < P − 34 F . Hence, Sn > 43 F , which contradicts formula (2). • If P < 43 F , then 43 F − P > 0. Since the triangles constructed become F of the polygon increasingly smaller, at a certain step n, the area 4n−1 4 Hn becomes less than 3 F − P . From (2), one obtains 1 F F 4 4 F − Sn = < n−1 < F − P. 3 3 4n−1 4 3 Hence, Sn > P , which is impossible. During the 17th century, the method of exhaustion was always considered as a model of a rigorous mathematical reasoning, although it was thought to be too diﬃcult, especially because of the double reductio ad absurdum. It was also thought to be too particular, since it was connected to speciﬁc properties of certain geometrical ﬁgures and the reasoning used in a speciﬁc case could not be used in others. In eﬀect, the method of exhaustion was not a method of ﬁnding or discovery, but rather it was a method of justiﬁcation of known results. Consequently, mathematicians searched for new methods that were easier and had a more general application. This led in a very natural way to the consideration of series and even inﬁnite products and continued fractions. For instance, in the above-mentioned quadrature of

1

Series Before the Rise of the Calculus

5

the parabola, it is possible to avoid the double reductio ad absurdum by 1 using the series ∞ n=0 4n . It is no wonder that series are found in many 17th-century works concerning the quadratures of curves and almost all the precursors of the calculus run up against series. In particular, the attempt to merge Cavalieri’s geometrical method of indivisibles with the emerging use of algebra led to the investigation of several series. ∗ ∗

∗

Geometric series played a crucial role in earlier research on series. In the 1590s, geometric series6 were mentioned in a work by Francois Vi`ete, Variorum de rebus mathematicis responsorum, in which he tackled the problem of the quadrature of circle. In this paper Vi`ete determined the sum of a geometric series ∞ i=1 ai . His starting point was Proposition 12 in Book 5 of Euclid’s Elements: If any number of magnitudes are proportional, then one of the antecedents is to one of the consequents as the sum of the antecedents is to the sum of the consequents (see Euclid [E]). In modern symbols, if sn = ni=1 ai , then a1 : a2 = (sn − an ) : (sn − a1 ).

Hence, a1 − an a1 − a2 = . a1 sn − an By assuming that the terms of the geometric series were decreasing, Vi`ete obtained a1 a1 − a2 = , (3) a1 s where s = ∞ i=1 ai . He justiﬁed (3) by stating that the magnitudes an were changed into nothing (in nihil ) when the series was continued ad inﬁnitum.7 As an example, Vi`ete considered the series ∞ 1 4 = 4n 3

n=0

and explicitly noted that it ﬁtted the Archimedean quadrature of the parabola. 6 It worthwhile pointing out that geometric series had already appeared earlier. N. Oresme dealt with the nature and summation of geometric series in a manuscript, the Quaestiones super geometriam Euclidis, which was only published in 1961 (On Oresme’s treatment of series, see Mazet [2003]). Oresme’s results seem to have had little inﬂuence on the rise of series theory in 17th century. 7 See Vi`ete [1593, 397–398].

6

Convergence and Formal Manipulation

A few decades later, Gr´egoire de Saint-Vincent made geometric series a crucial instrument of his method of quadratures.8 He wrote a remarkable treatise, the Opus geometricum, devoted to the quadrature of conics, which was published in 1647 though its essential aspects dated back to before 1625. Gr´egoire observed that the classic problems inherited from the Ancients had not been solved after many centuries; he therefore thought that new techniques and new methods needed to be discovered (unde novas artes et methodo novas iudicam excogitandas) to ﬁll the lacunae of ancient geometry [1647, 51–52]. Such new methods were grounded precisely on inﬁnite geometric series, which he discussed at length in the second book of the Opus geometricum. Saint-Vincent deﬁned a geometric series to be “a ﬁnite quantity divided by an uninterrupted sequence according to a given ratio” and distinguished series from progressions [1647, 54]. He used the term “progression” to mean both a ﬁnite sequence of the terms of a geometric series (which he understood as inﬁnite) and the sum of this ﬁnite sequence. Saint-Vincent used the term “limit” to denote the sum of a geometric series and stated that the “limit” of a progression was the end of the series that the progression did not reach –even if it continued indeﬁnitely; however, the progression could approach this limit more than any given quantity [1647, 54]. Saint-Vincent, as well Vi`ete, had an intuitive but clear idea of what the sum of series was (whatever words they used to denote the sum). By using n opinion, a series ∞more recent terminology, we could state that, in their k=0 ak had a sum S if the sequence of nth sums Sn = k=0 ak was convergent to S; namely, if it approached S indeﬁnitely when n increased so that the diﬀerence between Sn and S (in absolute value) became less than any given quantity. As we shall see below, this idea of the sum lay at the heart of the series theory during both the 17th century and, in a more complicated form, the 18th century. Basing his argument on the concept of the sum, Saint-Vincent examined the famous paradox of Achilles and the turtle. He showed that Achilles gains on the turtle according to a decreasing geometric series, which has a ﬁnite sum. Therefore, Achilles does reach the turtle and one can also determine the point where the turtle is reached by summing the series [1647, 97–98]. Saint-Vincent obtained several results by applying geometric series.9 One of the most interesting is the following proposition concerning the quadrature of the hyperbola: Let AY and AX be the asymptote of the hyperbola HKM [see Fig. 2]. If the segment AX is divided into segments AB, AC, 8

On Gr´egrorie de Saint-Vincent, see Dhombres [1995]. ∞ kn n Saint-Vincent, in particular, determined the sums of ∞ , for an n=0 q and n=0 q integer k [1647, 115–149] and constructed two geometric series with diﬀerent n-th terms but with the same sum [1647, 97–98]. 9

1

7

Series Before the Rise of the Calculus CD, DE that are in continued proportion, then the areas BCKH, CDLK,DEM N are equal 10 (Saint Vincent [1647, 586]).

Y

H

K L M

A

B

C

D

E

X

Fig. 2

Saint-Vincent did not employ the term “logarithm”. This term had already been introduced at that time although it was used with a meaning that diﬀered considerably from the modern one. Indeed, the word “logarithms” denoted the terms of an arithmetical progression that were matched with the terms of a geometric progression in sequence11 . By using the term logarithm in this sense, Saint-Vincent’s theorem can be formulated by stating that the areas BCKH, CDLK, DEM N are the logarithms of the abscissas of the hyperbola HKM . This formulation was made explicit by de Sarasa in 1649.12 ∗ ∗

∗

Pietro Mengoli was another mathematician who made a remarkable contribution to the rising theory of series. He was taught mathematics by Cavalieri and was inﬂuenced by Saint-Vincent and Torricelli13 . In 1650, 10 In other words, if the abscissa are in a geometric progression, then the areas are in an arithmetic progression. 11 See Burn [2001, 4]. 12 As regards diﬀerent historical interpretations of the actual contributions of SaintVincent and de Serasa to the study of natural logarithms, see Burn [2001]. 13 I point out that Torricelli gave a geometric proof of the sum of a geometric series in his De dimensione Parabolae [1644]. For Torricelli’s proof, I refer to Panza [1992, 307–308]. A similar geometrical proof, given by Leibniz, is discussed in Chapter 2.

8

Convergence and Formal Manipulation

Mengoli published several results concerning series in Novae quadraturae arithmeticae, seu de additione fractionum, a treatise that stemmed from the examination of the Archimedean quadrature of parabola, as he stated in the introduction.14 Mengoli based his argument upon two axioms: 1. If inﬁnite magnitudes have an inﬁnite extension, then one can take a certain number of these magnitudes such that they exceed any ﬁnite extension (In modern terms, if the sum of a series is inﬁnite, then the partial sums become greater than any positive number) (Mengoli [1650, 18]). 2. If inﬁnite magnitudes have a ﬁnite extension and if they are thought of as being arranged and gathered together to form another extension, then these two extensions are equal (that is to say, if a series with positive terms15 converges to a ﬁnite number, then any rearrangement of the series converges to the same number) (Mengoli [1650, 19]). From these axioms Mengoli derived various properties of the series of magnitudes. In particular, a. if the sum of any number of a sequence of inﬁnite quantities is bounded, then the series has a ﬁnite extension (in modern words, if the partial sums of a series are bounded, the series is convergent) (Mengoli [1650, 18]); b. if a series has the ﬁnite extension S and A is a quantity less than S, then there is a ﬁnite number of the given magnitudes such that their sum exceeds A [namely, there exists a partial sum Sn of the series such that Sn < A (< Sn+1 )] (Mengoli [1650, 19]). Mengoli applied these axioms and properties to the determination of the sum of various numerical series by conceiving the numbers present in such series as speciﬁc values of geometric quantities. He represented the terms, partial sums and remainder of series by means of segments. In order to sum the series ∞ 1 , n(n + 1) n=1

Mengoli employed a relation, which he had proved in his Novae quadraturae arithmeticae [1650, 9] and which, using modern symbols, can be written as an − an−1 an − a1 a2 − a1 a3 − a2 a4 − a3 + + + ... + = , a1 a2 a2 a3 a3 a4 an−1 an a1 an 14

On Mengoli’s contribution to series theory, see Agostini [1941]. Since Mengoli referred to geometrical quantities, he tacitly assumed that the terms of series were positive. 15

1

Series Before the Rise of the Calculus

9

This makes it possible to establish that the partial sums of ∞ formula 1 n=1 n(n+1) are Sn =

1 1 1 n + + ... + = , 1·2 2·3 n(n + 1) n+1

(4)

n < 1, the series has a ﬁnite extension S. This extension is precisely Since n+1 equal to 1. Indeed, if S > 1, then there should exist a partial sum Sn such that Sn > 1, which is impossible. Now, let S < 1 be. Since the numbers n n+1 approach 1 indeﬁnitely when n increases, the partial sums

Sn =

n n+1

would become greater than S when n is large enough. This is also impossible. Consequently, S = 1. Similarly, Mengoli obtained the sum of many other series, such as ∞

n=1 ∞

n=1 ∞

n=1 ∞

n=1

3 1 = , n(n + 2) 4 1 11 = , n(n + 3) 18

1 1 = , n(n + 1)(n + 2) 4

1 1 = . (2n + 1)(2n + 3)(2n + 5) 12

Moreover, in the introduction to Novae quadraturae arithmeticae, Mengoli showed that the harmonic series did not converge.16 In modern terms, his proof can be formulated as follows. Since 1 1 3 1 + + > , n−1 n n+1 n

one has

1 1 1 1 1 1 1 1 1 + + + + + + + + + ... 2 3 4 5 6 7 8 9 10 1 1 1 1 1 1 1 1 1 + + + + + + + + + ... 1+ 2 3 4 5 6 7 8 9 10 3 3 3 3 1+ + + + + ... 3 6 9 12 1 1 1 1 + 1 + + + + ... 2 3 4 1 + S.

S = 1+ = > = = 16

This result was not new (see Oresme [A]).

10

Convergence and Formal Manipulation

Consequently, S cannot be a ﬁnite quantity. In the introduction to Novae quadraturae arithmeticae, Mengoli also took the series ∞ 1 n2 n=0

into consideration. He failed to calculate the sum of such a series, but this problem was subsequently tackled by Jakob Bernoulli, and became known as the Basel problem. It was considered a very interesting problem of pure mathematics and its solution was one of Euler’s most important successes. Mengoli also wrote Geometriae speciosae elementa 17 [1659] and Circolo [1672], where he rediscovered an inﬁnite product expansion for π/2, which had already been found by Wallis in a way that I shall now go on to examine. ∗ ∗

∗

The use of series was worked on extensively by John Wallis. In Arithmetica inﬁnitorum [1656] he tried to provide an arithmetical version of the method of indivisibles; this led him to deal with a large number of series by means of a peculiar methodology that had an enormous inﬂuence on later mathematicians.18 As Maier` u [1994, 118–119] noted, Wallis’s treatment of series developed in a number of particular cases and makes use of the speciﬁc geometric properties of particular ﬁgures. To illustrate Wallis’s method,19 consider the problem of ﬁnding the area under the curves y = xk (k = 1, 2, . . . ) and over the segment [0, a] (see Fig. 3, where the curve y = xk is represented by means of P SR, P Q = AB = a, and RQ = BC = ak ). Following Cavalieri, Wallis regarded the ﬁgure P QR as consisting of an inﬁnite number of parallel lines, every one of them having length equal to xk . Therefore, if one divides the segment P Q = AB = a into n pieces of length h = na , where n is inﬁnite, the sum of these inﬁnite lines is of the type 0k + hk + (2h)k + (3h)k + . . . + (nh)k , k = 1, 2, . . . .

(5)

Similarly, the area of the rectangle is ak + ak + ak + . . . + ak = (nh)k + (nh)k + (nh)k + . . . + (nh)k , k = 1, 2, . . . . The ratio between the parabola P QR and the rectangle ABCD is Area parabola P SR 0k + 1k + 2k + 3k + . . . + nk = k , k = 1, 2, . . . . (6) Area rectangle ABCD n + nk + nk + nk + . . . + nk This procedure led Wallis to consider the problem of determining the values 17

For this work, I refer to Massa [1997]. On Wallis’s method of quadrature, see Scott [1938], Panza [1995, 135–176], and Maier` u [1994], [1995], and [2000]. 19 See Wallis [1656, 1–52]. 18

1

11

Series Before the Rise of the Calculus

R

S P

Q

D

C

A

B Fig. 3

of 0k + 1k + 2k + 3k + . . . + nk , nk + nk + nk + nk + . . . + nk

(7)

for n = ∞ and k = 1, 2, 3, . . ..20 He stated The simplest method of investigation . . . is to consider a certain number of individual cases, and to observe the emergent ratios, and to compare these with one another, so that a universal proposition may be established by induction. (Wallis [1656, 1]) He ﬁrst considered the case k = 1 and observed that 0+1 1+1 0+1+2 2+2+2 0+1+2+3 3+3+3+3 0+1+2+3+4 4+4+4+4+4

20

= 12 = 12 = 12 = 12 .

In modern terms, he sought lim

n→∞

1k + 2k + 3k + . . . + nk . (n + 1)nk

k The divergent series appears in Wallis’s work as intermediate steps during the jj analytical manipulation of geometrical entities.

12

Convergence and Formal Manipulation

By induction, Wallis asserted that 1 0 + 1 + 2 + 3 + ... + n = . n + n + n + n + ... + n 2 Wallis then considered the case k = 2. Since 0+1 1+1 0+1+4 4+4+4 0+1+4+9 9+9+9+9 0 + 1 + 4 + 9+16 16+16+16+16+16

he stated

= = = =

1 2 5 12 14 36 30 80

= = = =

1 3 1 3 1 3 1 3

+ + + +

1 6 1 6·2 1 6·3 1 6·4 ,

1 1 02 + 12 + 22 + 32 + . . . + n2 = + . 2 2 2 2 2 n + n + n + n + ... + n 3 6n

The ratio approached

1 3

as the number of terms increased, and

1 02 + 12 + 22 + 32 + . . . + n2 = 2 2 2 2 2 n + n + n + n + ... + n 3 for n = ∞. In case k = 3, Wallis proceeded in a similar way and found 03 + 13 + 23 + 33 + . . . + n3 1 1 = + 3 3 3 3 3 n + n + n + n + ... + n 4 4n and

1 03 + 13 + 23 + 33 + . . . + n3 = n3 + n3 + n3 + n3 + . . . + n3 4

for n = ∞. By generalizing these results, Wallis asserted21 1 0k + 1k + 2k + 3k + . . . + nk = . k k k k k n + n + n + n + ... + n k+1

(8)

Wallis did not stop here. He continued to generalize in order to give a meaning to (8) even when k = 1, 2, 3, . . .. He ﬁrst stated that if the value 0 was assigned to k, then one obtained 00 + 10 + 20 + 30 + . . . + n0 1 = . n0 + n0 + n0 + n0 + . . . + n0 1 He then sought to justify the assignment of fractional values to k in the following way. If we denote22 the series 0k + 1k + 2k + 3k + . . . + nk by Ak , 21

Of course, from this formula and (6), one can deduce that the area under the parabola n k ak+1 = . j j=0 k+1

y = xk from 0 to a is 22

The symbolism is mine.

1

13

Series Before the Rise of the Calculus

and the reciprocal of their sums (the corresponding ratio, in Wallis’s terms) by bk (= k + 1), then formula (8) can be written in the form nk

+

nk

1 Ak = . k + ... + n bk

Wallis observed that √ √ √ √ 04 + 14 + 24 + 34 + . . . = 02 + 12 + 22 + 32 + . . . . √ The terms n2 = n4 of A2 are the square roots of the terms of A4 and, therefore, A2 can be viewed as the series “interpolating” A0 and A4 . The corresponding ratios of A0 , A2 , and A4 are the numbers b0 = 1, b2 = 3, and b4 = 5, which are in arithmetic progression (see table below). A0 = 00 + 10 + 20 + 30 + . . . + n0 –> b0 = 1 A2 = 02 + 12 + 22 + 32 + . . . + n2 –> b2 = 3 A4 = 04 + 14 + 24 + 34 + . . . + n4 –> b4 = 5 √ √ √ √ At this point Wallis considered A√ = 0 + 1 + 2 + 3 + . . . and stated that it was the series interpolating A0 and A1 since it behaved with respect to A0 and A1 as A2 behaved with respect to A0 and A4 . By analogy, the value of √ √ √ √ √ 0 + 1 + 2 + 3 + ... + n √ √ √ √ √ n + n + n + n + ... + n ought to be a number 1b such that b0 = 1, b and b1 = 2 (namely, the corresponding ratios of A0 = 1, A√ , and A1 = 2) were in arithmetic progression. Hence, b = 21 + 1. By observing that for k = 12 , formula (8) becomes 01/2 + 11/2 + 21/2 + 31/2 + . . . + n1/2 = n1/2 + n1/2 + n1/2 + n1/2 + . . . + n1/2

1 2

1 , +1

√ 1 Wallis concluded that n 2 = n. Similarly, Wallis observed that √ √ √ √ 3 3 3 3 A1 = 03 + 13 + 23 + 33 + . . . = 0 + 1 + 2 + 3 + . . . , √ 2 √ 2 √ 2 √ 2 3 3 3 3 03 + 13 + 23 + 33 + . . . = 02 + 12 + 22 + 32 + . . . , A2 =

√ 3 and that the terms n = n3 of A1 were the cube roots of the terms of A3 √ 2 3 and the terms n2 = n3 of A2 were the squares of cube roots. For this reason A1 and A2 could be viewed as the series interpolating A0 and A3 . The corresponding ratios of A0 , A1 , A2 , A3 (b0 = 1, b1 = 2, b2 = 3, b3 = 4) were in arithmetical progression. Wallis then considered √ √ √ √ √ 3 0 + 3 1 + 3 2 + 3 3 + ... + 3 n 1 √ √ √ √ √ (9) = 3 3 3 3 3 r n + n + n + n + ... n

14

Convergence and Formal Manipulation

and 2 √ 2 √ 2 √ 2 √ √ 2 3 0 + 3 1 + 3 2 + 3 3 + . . . + ( 3 n) 1 √ √ √ √ √ 2 2 2 2 2 = q 3 3 3 3 3 ( n) + ( n) + ( n) + ( n) + . . . ( n)

(10)

and assumed that √ 3

0+

√ 3

1+

√ 3

2+

√ 3

3 and

√ 2 √ 2 √ 2 √ 2 3 3 3 3 0 + 1 + 2 + 3

behaved with respect to A0

and

A1

in the same way as A1 and A2 behaved with respect to A0 and A3 . By analogy he concluded that b0 , r, q, b1 had to be in arithmetical progression as b0 , b1 , b2 , b3 . Therefore, r=

4 3

5 and q = . 3

This made it possible to write (9) and (10) as 1 01/3 + 11/3 + 21/3 + 31/3 + . . . + n1/3 = 1/3 1/3 1/3 1/3 1/3 n + n + n + n + ...n 1+

1 3

and 02/3 + 12/3 + 22/3 + 32/3 + . . . + n2/3 1 . = 2/3 2/3 2/3 2/3 2/3 n + n + n + n + ...n 1 + 32 √ 2 √ Consequently, 3 n was equal to n1/3 and ( 3 n) was equal to n2/3 . In this way Wallis was able to √ ﬁnd the meaning of the power xα , where α was a l k rational number (n k = nl ). He even considered the case in which α was an irrational and a negative number. Wallis’s analogical procedure (later known as Wallis’s interpolation) was of great importance in the 18th century. It can be considered as an answer to the following problem: Given a sequence Pk , deﬁned for integral values of k, ﬁnd the meaning of Pα where α is a nonintegral number.

1

Series Before the Rise of the Calculus

15

In the case Wallis considered, Pk were the sequences xk and 0k + 1k + 2k + 3k + . . . + nk , nk + nk + nk + nk + . . . + nk and the problem was reduced to the interpolation of the number sequence 1 , k+1

k = 0, 1, 2, . . . .

From a modern point of view, this problem is meaningless. A modern mathematician attributes meaning to new operations, formulas, or symbols using appropriate deﬁnitions. Operations, formulas, and symbols do not have a “natural” meaning. Thus, if xn is deﬁned only for an integer value of n, then any meaning can be assigned to a new symbol such as x1/2 . Wallis viewed the matter diﬀerently. New combinations of symbols, such as x1/2 and x0 , were not introduced arbitrarily. Mathematical objects were not given by deﬁnition, but they existed in nature (or were an idealization of natural objects). It seemed obvious to them that x1/2 and x0 had a “natural” meaning and that mathematicians had to discover it. When a new symbol or a new object had to be introduced, mathematicians asked “What is the value (or the meaning) of the symbol?” and not “How shall we deﬁne it?” For Wallis, interpolating xn required investigating the objects x, x2 , . . . and reconstructing the “nature” of these objects just as one reconstructed the nature of a physical phenomenon by interpolating physical data. When √ he met with the undeﬁned symbolic notation x1/2 , he did not take x1/2 = x by a useful but arbitrary deﬁnition; rather he “discovered” that the true √ meaning23 of x1/2 was x. In Arithmetica inﬁnitorum,24 Wallis reduced the problem of the quadrature of the circle to determining the corresponding ratio of the series whose

general term is ζ p = R2 − p2 a2 . To do this, he considered the series whose general terms are (R2 − p2 a2 )0 , (R2 − p2 a2 )1 , (R2 − p2 a2 )2 , (R2 − p2 a2 )3 , . . .

(11)

which have for their corresponding ratios 1,

2 8 48 , , , .... 3 15 105

If the series ζ p = R2 − p2 a2 is interpolated between the ﬁrst and second terms of (11), the corresponding ratio of

ζ p = R2 − p2 a2 23 24

See also Ferraro [1998, 291–293]. See Wallis [1656, 89–182].

16

Convergence and Formal Manipulation

8 48 is given by the interpolated value of 1, 32 , 15 , 105 , . . . between 1 and 23 . Wallis introduced the symbol to denote the sought-after number and constructed several numerical tables such as

1

1

1

1

1

2

3

4

1

3

6

10

1

4

10

20

where • the numbers in the ﬁrst row and column are 1, • those in the second row and column are the natural numbers {n}n=1,2,..., ∞ , (triangular • those in the third row and column are n(n+1) 1·2 n=1,2,...,∞

numbers),

• those in the fourth row and column are angular pyramidal number),

n(n+1)(n+2) 1·2·3

n=1,2,...,∞

(tri-

• ... After a long sequence of calculations, he succeeded in expressing as the inﬁnite product 3 · 3 · 5 · 5 · 7 · 7... 4 . (12) = = π 2 · 4 · 4 · 6 · 6 · 8... ∗ ∗

∗

Formula (12) was not the ﬁrst inﬁnite product to be found in the history of mathematics. In his Variorum de rebus mathematicis responsorum [1593], Vi`ete had already squared the circle by means of an inﬁnite product. He assumed the circle to be a polygon with inﬁnite sides and considered regular inscribed polygons of 4, 8, 16, . . . sides. By using geometric properties of these polygons he represented π in the form25

1 1 1 11 1 1 1 1 2 = + + + .... (13) π 2 2 2 2 2 2 2 2 2 25

See Vi`ete [1593, 400].

1

17

Series Before the Rise of the Calculus

In the Arithmetica inﬁnitorum,26 Wallis also published an expansion of π4 into continued fractions. He had submitted (12) to Lord Brouncker, who expressed π4 in the form 4 1 9 25 49 =1+ . π 2+ 2+ 2+ 2+

(14)

This formula was published by Wallis in the Arithmetica Inﬁnitorum, Proposition 191. On this occasion Wallis introduced the term “continued fraction”. However, he did not expound the procedure used by Lord Brouncker to derive (14). It should be emphasised that when Brouncker obtained (14), continued fractions were already known, at least since 1613 when Cataldi had shown √ how a root p could be expanded into a continued fraction. Earlier, in his Algebra [1572, 37–38], Bombelli had published a procedure for calculating the approximate value of a root which can be interpreted a posteriori as a procedure for developing numbers into continued fractions. √ 13, Bombelli ﬁrst observed that 3√is the greatTo compute the value of √ est integer less than 13. Then he considered the diﬀerence 13 − 3 = x (for the sake of simplicity, I use the letter x to denote this diﬀerence, though Bombelli did not use symbols of this kind). The ﬁrst approximation of x (say x1 ) is given by 23 because 13 − 32 = 4 and x1 =

4 2 4 = = . 2·3 6 3

To ﬁnd a second approximation x2 , he set x2 =

4 6+

x3 =

4 6+

2 3

3 = . 5

The third approximation is 3 5

=

20 . 33

Similarly, he found x4 =

66 , 109

x5 =

109 , 180

x6 =

720 . 1189

The approximation can be improved as desired. In modern terms, Bombelli’s procedure can be described as follows. If one sets √ p = n + x, 26

See Wallis [1665, 181–193].

18

Convergence and Formal Manipulation

where n is the maximum integer such that n2 ≤ p, then one has p − n2 = 2nx + x2 . By neglecting x2 , one obtains the ﬁrst approximation x1 =

r , 2n

where r = p − n2 . By approximating 2n + x to 2n + x1 , the equation r = (2n + x)x can be written as r = (2n + x1 )x2 . Hence, x2 =

r , 2n + x1

and so on. In this way, one derives a repeating continued fraction √

p=n+

r 2n +

r r 2n+ 2n+...

The absence of appropriate symbolism makes it hard to state whether Bombelli had grasped the idea of continued fractions. This idea is clear in Cataldi, who expounded the same procedure in his Trattato [1613]. He considered “the manner of ﬁnding roots, by adding step by step to the denominator of the fraction which is the last in the preceding root, a fraction equal the ﬁrst one” [1613, 70]. As an example, he expressed the square root of 18 as a continued fraction and used the following symbols: 4& and

2 8 & 28

& 82

2 2 2 4& & & 8 8 8

During the 17th century, other mathematicians used continued fractions. For instance, Christiaan Huygens used the convergents of continued fractions to approximate the correct design for the toothed wheels of a planetarium.27 Wallis himself returned to continued fractions later in his Algebra [1685], where he expounded some of the now-familiar properties of convergents. However, it was only during the 18th century that new important results concerning continued fractions were obtained. ∗ ∗ 27

See Huygens [P, 627–631].

∗

1

19

Series Before the Rise of the Calculus

In 1668, Nicholas Mercator published a book, entitled Logarithmotechnia, in which he illustrated a technique for facilitating the calculation of logarithms. This technique was based on the series expansion of the natural logarithm28 x2 x3 + − ... (15) log(1 + x) = x − 2 3 which he derived by using the results of Saint-Vincent and de Sarasa (namely, 1 the fact that the logarithm is the area under the hyperbola y = 1+x ) and by applying Wallis’s method of quadrature (see Mercator [1668, 28–33]). Mercator’s procedure can be summarized as follows.29 To compute the area 1 and above the segment [1, A], one divides this under the hyperbola 1+x segment into an equal number of parts (in aequales partes innumeras). Denoting each of the parts by h, the area of the hyperbola can be thought of as the sum of the ordinates 1 , 1+h

1 , 1 + 2h

1 , 1 + 3h

...,

1 , 1 + nh

where nh = A. By expanding these fractions by long division (I shall clarify the meaning of this term a few lines below), one obtains 1 1+h 1 1 + 2h 1 1 + 3h

= 1 − h + h2 − h3 + . . . , = 1 − 2h + 4h2 − 8h3 + . . . , = 1 − 3h + 9h2 − 27h3 + . . . , ....

By summing column by column, the sought-after area comes to be equal to (1 + 1 + 1 + . . .)

(16)

− (h + 2h + 3h + . . .)

+ (h2 + 4h2 + 9h2 + . . .)

− (h3 + 8h3 + 27h3 + . . .)

+ ....

The series n (na)k , k = 0, 1, 2, . . ., in formula (16) are the same series used by Wallis and can be thought of as representing the areas of the parabolas 28

The expansion of logarithm had already been found by Newton (see p. 56). However, Mercator was the ﬁrst to publish it in 1668. 29 Here I follow the simpliﬁed version of Mercator’s proof that was expounded by Wallis in the same year (1668) in a letter to Brouncker published in Philosophical Transactions (see Wallis [1668, 753–754]).

20

Convergence and Formal Manipulation

y = xk between 0 to A. Therefore, Ak+1 , (na)k = k+1 n and log(1 + A) = A −

k = 0, 1, 2, . . .

A2 A3 + − .... 2 3

Mercator’s result was improved by Gregory [1668, 9–13] and Halley [1695, 58–67], who provided the expansion log

1+x 2x3 2x5 = 2x + + + ..., 1−x 3 5

which is faster and more useful in calculations. In his Logarithmotechnia, Proposition 7, Mercator used the method of long division, today know as one of Mercator’s rules.30 This method was based on the idea that the usual algebraic operations could be used to generate series and that one could operate on a series in the same manner as one operated on a closed analytical expression. In the speciﬁc case of long division, the usual algorithm to ﬁnd the quotient between two polynomials was continued endlessly to obtain an inﬁnite series. For instance, consider the fraction 1 . 1+x By dividing 1 by 1 + x, one obtains the quotient 1 and the remainder −x. By dividing such a remainder by 1 + x, one obtains the quotient x and the remainder −x2 . By continuing in inﬁnitum, one obtains the series 1 − x + x2 − . . .. ∗ ∗

∗

The last mathematician I shall discuss in this chapter is James Gregory. He made a number of remarkable contributions to series theory and some of his results overlapped with the ﬁndings of Newton. In his Vera Circuli et Hyperbolae Quadratura [1667], while investigating the areas of conic sections, he introduced the expression “convergent series” [1667, 10]. However, his deﬁnition is rather diﬀerent from the modern one, even if it contains the basic idea of quantities approaching a limit. Gregory did not give the name of ’convergent series’ to a series or sequence, but to a pair of sequences an and bn so that 30 The other is the extraction of roots (see Chapter 4). On Leibniz’s treatment of long division, see Chapter 2.

1

21

Series Before the Rise of the Calculus

a. they were deﬁned by recurrence,31 namely an = f (an−1 , bn−1 ) and bn = g(an−1 , bn−1 ), n > 1 b. |an − bn | < |an−1 − bn−1 |. He also gave the name “convergent terms” to the pair of terms (an , bn ) of the convergent sequences. An example of convergent sequences in Gregory’s sense is the following. Given the positive number a, b, c, d, e, with b > a, c > d, c > e, consider the sequences an and bn thus deﬁned: a1 = a, an

b1 = b, d = an−1 + (bn−1 − an−1 ), c

e bn = bn−1 − (bn−1 − an−1 ). c

It is not diﬃcult to prove that an−1 < an < bn−1

and

an−1 < bn < bn−1 .

Hence, |bn − an | < |bn−1 − an−1 |. The two sequences therefore approach each other more and more. Since bn−1 − an−1 b1 − a1 = , bn − an b2 − a2 −an−1 is a constant value (< 1) and the diﬀerences |bn − an | the ratio bn−1 bn −an becomes less than each given quantity. Gregory remarked

Then, if we imagine that the series is continued in inﬁnitum, we can imagine the last convergent terms to be equal. We call the equal terms the termination of the series. (Gregory [1667, 18–19, my emphasis]) In Vera Circuli et Hyperbolae Quadratura, Gregory applied his notion of convergence to the investigation of the problem of the quadrature of the circle by means of appropriate sequences of regular n-polygons, inscribed and circumscribed to a sector of a conic section. 31 Condition a mainly depended on the special context of Gregory’s research (the quadrature of conic sections).

22

Convergence and Formal Manipulation

In a letter written on November 23, 1670 to Collins,32 Gregory provided the formula of interpolation, which today is named after him and Newton:33 f (x0 + ct) = f (x0 ) + tΔf (x0 ) + +

t(t − 1) 2 Δ f (x0 ) 2!

t(t − 1)(t − 2) 3 Δ f (x0 ) + . . . , 3!

(17)

This formula gives the interpolated value of a function f (x) at the point x0 + ct by using the forward diﬀerences34 Δh f (x0 ), h = 1, 2, 3, . . . at the point x0 . Gregory expressed (17) in the form αγ =

kh li ad bf + + + + etc., c c c c

where αγ is the interpolated value of the ordinate of a given curve ABH at the point ac , the letters d, f, k, . . . are the ﬁrst, second, third, etc. diﬀerences taken at the point x0 = 0, and cb , kc , . . . denote a(a − c) a(a − c)(a − 2c) , , ..., c · 2c c · 2c · 3c

respectively. He assumed that the ordinate of the curve at the point 0 is 0.

I H

B

A

Q Fig. 4

Gregory gave no proof of his formula and observed This method, as I apprehend, is both more easie and universal than either Briggs or Mercator’s,35 and also performed without table. (Turnbull [GT, 131]) 32

See Turnbull [GT, 131]. As regards the use of this formula before Gregory, see the ﬁrst chapter of Goldstine [1977]. 34 Forward diﬀerences are deﬁned as follows: 33

n

Δf (x0 ) = f (x0 + h) − f (x0 )

Δ f (x0 ) = Δ 35

nia.

n−1

f (x0 + h) − Δ

n−1

(h is a constant), f (x0 ), n = 2, 3, 4, . . . .

Gregory referred to Briggs’s Arithmetica logarithmica and Mercator’s Logarithmotech-

1

Series Before the Rise of the Calculus

23

Gregory applied the interpolation formula to logarithms. Indeed, he considered the problem: Given b, d, e = log b and e + c = log(b + d), ﬁnd the number whose logarithm is e + a and stated the the sought-after number is36 a(a − c) d2 a(a − c)(a − 2c) d3 a + + .... b+ d+ c c · 2c b c · 2c · 3c b2

(18)

To obtain (18), one can observe that a a e + a = log b + (e + c − e) = log b + (log(b + d) − log e) c c a c b = log b 1 + d a and then the sought-after number is b 1 + db c . If one considers the diﬀerences Δ(1 + p)z = p(1 + p)z , Δ2 (1 + p)z = p2 (1 + p)z , Δ3 (1 + p)z = p3 (1 + p)z , ... of the polynomial (1 + p)z , one obtains [Δn (1 + p)z ]z=0 = pn . By applying (17) to (1 + p)z for p = db and z = ac , one ﬁnds37 (18). In another letter to Collins written on February 15, 1671,38 Gregory referred to seven power series expansions for the quantities which, using modern symbols terminology, would be written as

36

arctan x, tan x, x π log tan + , 2 4

sec x, log sec x, √ 2ex , 2 arctan (tanh x) . sec−1

One could note that expansion (18) includes the binomial theorem. Newton became aware of this from the winter of 1664–65. On the priority of the discovery, see Pensivy [1987–88, 45–48]. 37 Cf. Pensivy [1987–88, 43] and Goldstine [1977, 76]. 38 See Turnbull [GT, 170–171].

24

Convergence and Formal Manipulation The ﬁrst three expansions were written by Gregory in the form t5 t7 t9 t3 + 4 − 6 + 9, 2 3r 5r 7r 9r a3 2a5 17a8 3233a9 + + , t = a+ 2 + 3r 15r4 315r6 181440r8 a2 5a4 61a6 277a8 s = r+ + + + , 3 5 2r 24r 720r 8064r7

a = t−

where r is the radius of a circle, a the arc of this circle cut oﬀ by the angle θ, rt the tangent of θ, and rs the secant of θ. In his James Gregory Tercentenary Memorial Volume [GT, 26], Turnbull hypothesized that Gregory might have used a procedure corresponding to the use of the Taylor formula to obtain some of his expansions. He based his argument upon an error of calculation of the series for arctangent (in 3968a9 3233a9 eﬀect the ﬁfth term is 181440r 8 and not 181440r 8 ). In his James Gregorie on Tangents and the “Taylor ” Rule of Series Expansions [1993], Malet states convincingly that Gregory calculated the coeﬃcient of any of the expansions sent to Collins by means of the following recursive procedure: Let

x3 x2 + A∗3 2 + etc. r r be any of Gregorie’s [expansions]. [. . .] The constant r and the term ri are introduced to preserve the homogeneity of variable y [. . .] the y always were the ordinates of some curves [. . .] y0 is the value taken of y at a given point, say x0 , and the coeﬃcients A∗i are the numerical values taken by the variables Ai at the point x0 . In order to ﬁnd the variable A1 , Gregory applied his new method of tangent [. . .] to the curve the ordinate of which are y, thereby determining its subtangent, say T [. . .] A1 was then determined as Ty r [. . .] Next Gregorie would assume A1 to represent the ordinates of a curve having the same axis as the ycurve and calculate its subtangent [. . .] and so on. (Malet [1993, 100–101]) y = y0 + A∗1 x + A∗2

2

Geometrical quantities and series in Leibniz

Leibniz began to study higher mathematics in 1672 following Huygens’ prompting. In a short time he became interested in inﬁnite series. It is well-known that the investigation of number sequences and their diﬀerence and sum sequences was of great importance in his discovery of the calculus.39 He himself asserted I arrived at the method of inassignables through the method of inﬁnite increments in the series of numbers, as the nature of the things requires. (Leibniz [GMS, 4:413]) Between 1675 and 1676, he wrote a treatise, entitled De quadratura arithmetica circuli ellipseos et hyperbolae cujus corollarium est trigonometria sine tabulis [KQA], which aimed to provide the quadrature of certain curves by means of series. Leibniz wrote at least six versions of this treatise, which nevertheless remained unpublished through his life. Only recently has Eberhard Knobloch published its last and most extensive version, which consists of 51 propositions and many scholia.40 In the years that followed, Leibniz published many of the results of the De quadratura arithmetica (but not the proofs and the solution methods) in De vera proportione circuli ad quadratum circumscriptum in numeris rationalibus expressa [1682] and in Quadratura arithmetica communis sectionum conicarum [1691]. Then Leibniz wrote other important papers concerning series, in particular Supplementum geometriae practicae sese ad problemata transcendentia extendens, ope novae methodi generalissimae per series inﬁnitas [1693] and Epistola ad V. Cl. Christianum Wolﬁum Grandi [1710]. This chapter is divided into two sections. The ﬁrst is devoted to the investigation of Leibniz’s notion of convergence and the way in which he manipulated series. In the second section, I shall examine how power series were employed in the geometrical context of the early calculus. Other aspects of Leibniz’s conception (Leibniz’s derivation of the Bernoulli series, Leibniz’s analogy, the rise of the question of divergent series) are discussed in Chapters 3 and 9.

2.1

The capacity of series to express quantities and their manipulation

In De quadratura arithmetica, Leibniz formulated the widelyknown theorem on the sum of a geometric series as follows: The greatest term of an inﬁnite geometric series is the mean proportional between the greatest sum and the greatest diﬀerence. (see Leibniz [KQA, 71]) 39 40

See Bos [1974, 13] and Guicciardini [1999, 137–138]. For an analysis of the manuscript, see Knobloch [1989] and [1991].

25

26

Convergence and Formal Manipulation

B1

B2 D1

B3 B4 D2

D3

B5 D5

D4

s A1

A2

A3

A4

A5

C

Fig. 5

Leibniz’s proof is geometric and runs as follows. Given a decreasing geometric series an and a straight line s, one draws a segment A1 B1 so that it equals the ﬁrst term a1 of the series and is perpendicular s (see Fig. 5). One takes A2 on the line s such that A1 A2 = A1 B1 and draws the segment A2 B2 so that it equals the second term a2 of the series and is perpendicular to s. Then one takes A3 on s such that A2 A3 = A2 B2 and so on. The point Bn fall on the line B1 B2 , which intersects s in C. Leibniz states that the segment A1 C is the sum of the series. Indeed the triangles Dn Bn Bn+1 are similar to An Bn C, and therefore the segments An C are proportional to Dn Bn+1 = An An+1 = An Bn . It is true that no An coincides with C; however, the segments An Bn become smaller than any quantity and the sequence An approaches C closer and closer, with an error less than any assignable quantity. Since A1 C : A1 B1 = A1 B1 : D1 B1 , one has S : a1 = a1 : (a1 − a2 ), where S is the sum of the series (see Leibniz [KQA, 71–73]). Leibniz, like the other mathematicians considered in chapter 1, thought that the sum of a series was a determinate quantity (the segment A1 C) to which the partial sums of the series approached increasingly. Leibniz’s

2

27

Geometrical Quantities and Series in Leibniz

notion of the sum resembles the modern one; however, it presents some aspect that makes these two notions diﬀerent.41 These aspects concern: 1. the relationship between a series and the quantity expressing this series, 2. the way Leibniz handled series. As regards the ﬁrst point, I would emphasise that, according to Leibniz, a series could express exact relationships and not merely approximations; however, a series expressed an exact relationship insofar as it is considered a whole and not a limit process. Indeed, in De vera proportione,42 Leibniz justiﬁed that π4 is equal to 1−

1 1 1 + − + ... 3 5 7

by observing that • if we take the ﬁrst term of this series, then error less than 13 ,

π 4

is approximated with an

• if we take the ﬁrst two terms of this series 1 − 31 , the error is less than 1 5, • if we take the ﬁrst three terms of this series 1 − 31 + 15 , the error is less than 71 , • etc. If the series is continued, the error becomes less than any given quantity. However, only the whole series (tota series), i.e., the actual inﬁnity of the terms of series, contains all (omnes) approximations and expresses the exact value (Quare tota series exactum exprimit valorem).43 Leibniz went on to state that although one could express the sum of the series by means of no (rational) number and this series was produced in inﬁnitum, one, however, conceived the whole series in his mind, since the law of progression was known (Leibniz [GMS, 5:120]). In other terms, Leibniz thought that the sum of a series was achieved when the aggregate of the series (namely, the inﬁnite terms of the series) was all together taken into account. He illustrated this concept of the sum by Fig.6. 41

Such aspects can also be noted in the works of previous mathematicians. (See, e.g., Wallis’s interpolation, p. 10, Mercator’s rule, p. 20, and Gregory’s reference to the last term, p. 21.) In Leibniz they appear in a clearer way. 42 See Leibniz [1682, 44]. 43 See Leibniz [GMS, 5:120].

28

Convergence and Formal Manipulation

Numero

Deus

1 1 1 1 1 − + − + 1 3 5 7 9 1 1 1 1 − + − + 11 13 15 17 1 etc. − 19

I

Impare

gaudet

Fig. 6 Leibniz explained that a circle is equal to an inﬁnite series in the same way as the segment of length 1 is equal to the sum of the segment 1 1 1 + + + ... 2 4 8 since (see Fig.7) the segment AB = 1 could be obtained by the juxtaposition of the segments 1 1 1 AC = , CD = , DE = , 2 4 8

...

(see Leibniz [GMS, 5:120–121]).

A

C

D

E

B

Fig. 7 While an inﬁnite series was equal to a quantity exactly only if it was conceived globally, the limit process of the partial sums provided approximations of the quantity represented by the series. Leibniz stressed the importance of series as the instrument that made it possible for numerical results to be computed with as small a margin of error as desired (Leibniz [KQA, 79]). According to Leibniz, if one moved from theoretical observations to practice, the results of De quadratura arithmetica circuli ellipseos et hyperbolae cujus corollarium est trigonometria sine tabulis (the title44 gives 44 It means: ”On the arithmetical quadrature of the circle, the ellipse, and the hyperbola. A corollary is a trigonometry without tables.”

2

29

Geometrical Quantities and Series in Leibniz

a signiﬁcant indication of the main scope of the treatise) enabled trigonometric operations to be performed without tables with as small a margin of error as desired; he argued that this was an incredible sign of the power of geometry45 (Leibniz [KQA, 79]). Although the partial sums were fundamental for practical applications, they did not deﬁne the sum in the same manner as it was thought that 1.4,

1.41,

1.414,

1.4142,

1.41421,

...

√ approximated 2 but did not deﬁne the square root of 2, which was the ratio of two incommensurable quantities.46 The partial sums could even be used to ﬁnd the sum of a series or to prove a result; however, in order to sum a series, one had to reach the ultimate value of series. It was not necessary that the ultimate value was conceived of as a real object, a really existing entity: It was suﬃcient that one conceived the possibility of it. According to Leibniz, the ultimate term was a ﬁctitious quantity. For instance, he stated: [Inﬁnite numbers, inﬁnitely small numbers, and ultimate terms of series] are nothing but ﬁctions. Every number is ﬁnite and assignable . . . and inﬁnite and inﬁnitely small [magnitudes] mean nothing but magnitudes that can be assumed as large or as small as desired so that it is demonstrated without doubt that error is less that any given number. (Leibniz [D, 1:107]) I shall discuss Leibniz’s notion of ﬁctitious quantities in the following section. ∗ ∗

∗

As concerns the second above-mentioned aspect –exactly how Leibniz handled series–, I ﬁrst of all observe that, in modern mathematics, given the series ∞ 1 , 2n + 1 n=1

45 In a letter to Gallois written in December 1678, Leibniz wrote “I left my manuscript on the arithmetical quadrature in Paris. The theorems which are contained in it are considerable in theory and very useful in practice. Since if one memorises just two very simple progressions which I give and which are almost impossible to forget, once one has learnt them, all problems of trigonometry could easily be solved without tables, without instruments, and without books, with the exactness one wishes [. . .] Having some tables is a convenience, but not being able to solve problems without tables is a defect of science, for which I claim to have found a remedy” (Leibniz [GMS 1:186]). 46 On the notion of number, see Section 7.2.

30

Convergence and Formal Manipulation

one can write 1 1 1 = −2 2n + 1 2n − 1 (2n)2 − 1 or ∞

n=1

∞

1 = 2n + 1

n=1

1 1 −2 2n − 1 (2n)2 − 1

.

The last equality merely means that the nthterm of the series ∞is a linear 1 1 combination of the nth terms of the series ∞ and n=1 2n−1 n=1 (2n)2 −1 . Instead Leibniz considered the relation between the terms of the series as a relation between the sums of series and assumed ∞

n=1

∞

∞

n=1

n=1

1 1 1 = −2 . 2n + 1 2n − 1 (2n)2 − 1

Indeed, in De quadratura arithmetica [KQA, 82], he set 1 1 1 1 + + + + . . . = A, 1 3 5 7 1 1 1 1 + + + + . . . = B, 3 15 35 63 2 2 2 2 + + + + . . . = 2B. 3 15 35 63 Subtracting term by term, he derived C=

1 1 1 1 + + + + . . . = A − 2B = A − 1. 3 5 7 9

At this juncture Leibniz did not merely consider A − 2B and A − 1 as two symbols to denote two ways for deriving the terms of series C by operating upon the terms of A and B. He handled A − 2B and A − 1 as if A and B were numbers and A − 2B = A − 1 an ordinary equation in the unknown B; thus, he derived B = 21 and ∞

n=1

1 1 = . 2 (2n) − 1 2

(19)

2

31

Geometrical Quantities and Series in Leibniz Leibniz often used this method.47 For instance, he set48 ∞ 1 =A n

n=1

and

∞

n=1

and derived

2 = 2B n(n + 1)

∞ 1 = A − B = A − 1. n

n=2

Hence, B = 1 and49 ∞

n=1

2 =2 n(n + 1)

(see Leibniz [KQA, 83]). Nowadays we realize that this method worked well because Leibniz employed series ∞ n=1 an such that limn→∞ an = 0 and transformed them into a series with the nth term of the type cn = an − an+s , where s ≥ 1 is a ﬁxed integer. Under this constraint, one has ∞

n=1

cn=

∞

n=1

(an − an+s ) =

s

n=1

an − lim

k→∞

s

ak+n =

n=1

s

an .

n=1

Leibniz was able to use this procedure correctly and, at least intuitively, understood that the numerical result depended on an = 0.50 This could give 47

In Rencensio libri, which was written at the height of the controversy with Leibniz, Newton [OO, 4:459] dealt with the presumed diﬃculty of Leibniz’s results rather ironically ∞ 2 and showed how to obtain n=1 n(n+1) = 2 by a procedure conceptually identical to Leibniz’s one. Indeed, he wrote 1=1−

48

1 1 1 1 1 1 1 1 + − + − + − + + ... 2 2 3 3 4 4 5 5 1 1 1 1 = + + + + ... 1×2 2×3 3×4 4×6

Leibniz knew that the harmonic series was divergent (see 1Leibniz [KQA, 103–104]). Note the diﬀerence with Mengoli’s derivation of ∞ n=1 n(n+1) = 1 (see p. 9). 50 This condition was explicitly formulated by Jacob Bernoulli (see Chapter 5) and by de Moivre [1730, 129]. 49

32

Convergence and Formal Manipulation

the idea that the use of divergent series was due to diﬃculties in symbolism or an inadequate formalization of the notion of convergence. In reality, Leibniz thought that series could be handled independently of any preliminary meaning that might be attributed to them. For instance, in De vera proportione [1682, 121], Leibniz disregarded the fact that ∞

1 1 + 4k

∞

1 3 + 4k

k=0

and

k=0

were divergent and stated 1 1 1 1 π = + + + + ... − 4 1 5 9 13

1 1 1 1 + + + + ... . 3 7 11 15

(20)

A proof of this relation is found in De quadratura arithmetica [KQA, 81] and merely consists in rearranging the series 1 − 31 + 15 − 71 + 91 − . . . in order to obtain (20). Equation (20) can be interpreted in the sense that the diﬀerence between Q1 (i) =

i k=0

1 1 + 4k

and

Q2 (i) =

i k=0

1 3 + 4k

approaches π/4 closer and closer and it is ultimately Q1 (i) − Q2 (i) =

π . 4

Today we should say that the sums Q1 (i) and Q2 (i) have the same asymptotic behavior (with respect to log i). This idea was widely used by Euler in the 1730s.51 Another interesting example is the theorem known today as Leibniz’s convergence criterion for alternating series: Given a decreasing sequence an > 0, if an goes to 0 when n → ∞, then the series ∞ (−1)n+1 an n=1

is convergent. 51

See Section 13.3.

2

Geometrical Quantities and Series in Leibniz

33

In a letter to Johann Bernoulli of October 25, 1713, Leibniz formulated it thus: If the terms of a series are continuously decreasing and, alternatively, positive and negative, then the series is “advergent” (namely, convergent) (see Leibniz [GMS, 3: 926])52 . To prove this criterion, Leibniz set S=

∞

(−1)n+1 an

n=1

and observed that s2n−1 = a1 − a2 + a3 − a4 + . . . + a2n−1

> (a1 − a2 + a3 − a4 + . . . + a2n−1 )

+(−a2n + a2n+1 ) + (−a2n+2 + a2n+3 ) + . . .

= S (sn is the nth sum) since an ≥ an+1 . Similarly s2n = a1 − a2 + a3 − a4 + . . . + a2n

< (a1 − a2 + a3 − a4 + . . . + a2n )

+(a2n−1 − a2n+2 ) + (a2n+3 − a2n+4 ) + . . .

= S. Since

s2n < S < s2n+1 , he derived that S was ﬁnite. Furthermore, since S − s2n < a2n+1

and s2n−1 − S < a2n ,

the diﬀerence between the ﬁnite series sn and the inﬁnite series S could be made smaller than an , namely it became as small as desired, provided one considered a suﬃcient number of terms. 52

The proof I give here was sent to Bernoulli on January 10, 1714. An initial version of Leibniz’s criterion is found in De quadratura arithmetica [KQA, 115], where it is formulated as follows (apart from the use of modern symbols): If the quantity S is equal to the series ∞ n+1 an , then n=1 (−1) m−1 |S − (−1)n+1 an | < am n=1

(see also Leibniz [GM 4:273]). The formulation of 1713–1714 shows traces of the problem of divergent series, which had surfaced at the beginnings of the 18th century (see Chapter 9). On Leibniz’s criterion, see Knobloch [1993].

34

Convergence and Formal Manipulation

From a modern point of view, the ﬁrst part of Leibniz’s demonstration consists of the statement: If the sum exists, it is ﬁnite; the second part is n+1 a , we mean precisely a vicious circle. Today, by setting S = ∞ n n=1 (−1) that the diﬀerence between sn and S is less than any quantity when n goes to inﬁnity. To understand this proof, we must admit that, according to Leibniz, ∞ (−1)n+1 an S= n=1

is not deﬁned by S = lim

m→∞

m

(−1)n+1 an ,

n=1

namely S does not denote the value to which the series a1 − a2 + a3 − a4 + . . . approaches closer and closer. The previous examples show that, according to Leibniz, the symbol a1 − a2 + a3 − a4 + . . . could be used independently of the possibility of associating it with a ﬁnite number or a ﬁnite quantity or before proving that it was possible to associate a number with certain series. He operated upon these series by assuming that if an algebraic operation (and also diﬀerentials or integrals, if one considers power series, as we shall see later) could be performed upon ﬁnite series then it could be performed upon inﬁnite series. As concerns the nature of the above series, Knobloch [by referring in particular to (19)] stated: [A] is a symbol, suﬃciently known by the known law of formation of the relative series. A does not depend on summation step by step: Leibniz takes the whole expression as a ﬁctive quantity like signs for imaginary numbers. (Knobloch [1991, 276]) I substantially agree. In the quotation on p. 29 we already saw that Leibniz justiﬁed inﬁnitesimals, inﬁnitely large numbers, and ultimate terms of series in this way. Leibniz’s use of ﬁctitious entities is largely known; for instance, Bos stated: “[Leibniz] could not invoke the existence of inﬁnitesimals in answer to objections to the validity of the calculus. Instead, he had to treat the inﬁnitesimals as ‘ﬁctions’ which need not correspond to actually existing quantities, but which nevertheless can be used in the analysis of problems” (Bos [1974, 54–55]).53 An observation about this point is appropriate. In a letter written on February 2, 1702, to Varignon, Leibniz refused to reduce the science of the 53 On Leibniz’s conception, see Horv` ath [1982 and 1986] and, above all, a very stimulating paper by Knobloch published in 1999 (Knobloch [1999]).

2

Geometrical Quantities and Series in Leibniz

35

inﬁnite to a “ﬁction”. In this letter, however, the word “ﬁction” is used to refer solely to invention, in other words, creation at our own discretion, such as in the rules of a game. According to Leibniz, imaginary numbers, inﬁnite numbers, inﬁnitesimals, the powers whose exponents were not “ordinary” numbers and other mathematical notions are not mere inventions; they are auxiliary and ideal quantities that have their foundation in nature and serve to shorten the path of thought: Without worry one can use inﬁnitely small and large lines as ideal concepts — even though they do not exist as real objects in the metaphysically rigorous sense — as a means to shorten calculation, just as √ the imaginary roots in ordinary analysis, such as for example −2. Regardless of whether one calls these “imaginary”, they are nonetheless useful and sometimes even indispensable, in order to express real magnitudes analytically; so, for example, it is impossible, without using them, to give an analytical expression for a line segment, which divides a given angle into three equal parts. Just so, one could not elaborate our calculus of transcendental curves, without talking about differences, which are in the act of vanishing, and introducing once and for all the concept of incomparably small magnitudes . . . It is always in the same way that one knows the dimensions above three, and the same is true for the powers whose exponents were not ordinary numbers; all this establishes ideas which are suitable for shortening reasoning and are based on reality. Nevertheless one must not imagine that the science of the inﬁnite is degraded for this reason and reduced to ﬁction; because it always remains a syncateromatic inﬁnite, to use the term used in the School, and it remains true, for example, that 2 is as much as 1 1 1 1 1 1 1 + 2 + 4 + 8 + 16 + 32 etc., which is an inﬁnite series, where all the fractions are included at the same time, though one employs nothing but ordinary numbers and though one considers no inﬁnitely small fraction, or fractions whose denominator is an inﬁnite number. Also the imaginary numbers have their foundation in reality (fundamentum in re). When I pointed out to

√ √ 2 2 1 + −3 + 1 − −3 is equal to the late Mr. Huygens that √ 2 6, he was so amazed and he answered that, for him, there is something incomprehensible in this. But just so, one can say, that the inﬁnite and inﬁnitely small have such a solid basis, that all results of geometry, and even the processes of Nature, behave as if both were complete realities . . . because everything obeys the Rule of Reason.54 (Leibniz [GMS 4:92–93]) 54 See also Leibniz’s Quid sit idea [GP, 7:263–264], where he stated that certain expressions have their foundation in nature, others in our discretion.

36

Convergence and Formal Manipulation

The use of ﬁctitious quantities could lead to the erroneous idea of objects whose existence is assured by their very deﬁnition and therefore to ascribing a modern conception to Leibniz. In reality, what ﬁnds its foundation in Nature cannot be created by the human mind by means of a deﬁnition55 . Modern mathematical theories do not make direct comparisons with nature but sound out our thought. They describe the logical consequences of certain axioms by means of clearly deﬁned rules of inference: A new object or symbol is always introduced by an (explicit or implicit) deﬁnition and its meaning derives entirely from that deﬁnition and the subject matter of the search is circumscribed a priori by the given axioms. Thus, one can manipulate an object O only after having deﬁned O and the operations that can be performed upon O. However, Leibniz believed that objects that have their foundation in nature have a meaning in their own right. This meaning could be not entirely clear so that these objects could √ be only partially known; however, one could manipulate the symbols −1 or 1 + 1/2 + 1/3 + . . . This was possible because Leibnizian ﬁctitious quantities were analogical extensions of ordinary quantities and were manipulated by analogical extensions of the rules of ordinary quantities. Thus, √ one could operate upon √ −1 in the same manner as one operated upon 2, and one could handle a1 − a2 + a3 − a4 + . . . in the same manner as one handled the ﬁnite sum a1 − a2 + a3 − a4 + . . . + an . I shall return to the notion of ﬁctitious quantities in Chapter 7, where I shall show that it can be used to improve our understanding of 18th-century mathematics.

2.2

Power series

Although Leibniz considered certain numerical series interesting in their own right, he concentrated above all on what is referred to today as power series. However, his investigations originated in the ﬁeld of geometry and this gave a peculiar aspect to series theory. Leibniz’s analysis was not based upon the notion of function (either in the modern or in the 18th-century sense) but upon curves. He did not consider a function y = f (x) and represented it diagrammatically; instead, he considered a certain ﬁgure F and adapted an appropriate system of coordinates (origins, unit, axes) to the ﬁgure F . The diﬀerence is very important. Today, given a function y = f (x), we know from the start the constants, the dependent variable and the independent variable. By contrast, given a ﬁgure F , Leibniz had some geometric quantities connected to the ﬁgure F and, according to the speciﬁc circumstances, chose what were to be considered as variables and what were to be considered as constants. To compute an area or length connected 55

“Deﬁnitions, according to Leibniz, are no mere conventions but have to show the possibility of the idea deﬁned” (Otte [1989, 15]). For instance, see Leibniz [1684, 537– 542].

2

Geometrical Quantities and Series in Leibniz

37

to the ﬁgure F , the series that expressed the area or length had to be convergent, but convergence could be obtained simply by changing constants into variables, modifying the value of the constant in an appropriate way, or choosing between diﬀerent expansions regarding the given quantities. For instance, let us consider the method of the long division.56 In De quadratura arithmetica, Leibniz expounded it as follows. He observed that a ac a = − 2 . b+c b b + bc If one replaces ac, b2 , bc by a1 , b1 , c1 , then

a1 b1 +c1

=

a1 b1

−

a1 c1 (b1 )2 +b1 c1

and

a ac a1 c1 a = − 2 + . b+c b b (b1 )2 + b1 c1 In a similar way, a ac ac2 a2 c2 a = − 2 + 3 − . b+c b b b (b2 )2 + b2 c2 The remainder of the series decreases as desired (quantumlibet decrescentes); if one continues the series to inﬁnity, then one ﬁnds the expansion a ac ac2 − 2 + 3 − ... b b b

(21)

a of b+c . In reality, this is true only if c < b; however, if one reverses b by c, a : then one obtains a diﬀerent expansion of b+c

a ab ab2 − 2 + 3 − ... c c c (see Leibniz [KQA, 76–77]).57 Therefore, if the ratio of certain geometrical quantities is a : (b + c), then it can always be expanded, except the isolated case b = c. There is no reason to refuse one of two diﬀerent expansions preliminarily. Only, a posteriori can one choose between two diﬀerent developments according to the geometric situation. This approach is entirely a diﬀerent from the modern expansion of the function f (c) = b+c , where the 56 57

See Chapter 1, p. 20. Of course, Leibniz assumed that the quantities a, b, and c are positive.

38

Convergence and Formal Manipulation

fact the c is a variable and a and b are constants is established a priori. In this case, it is not possible to change c into b without changing the function a f (c) (which can be expanded only if |c| < b) into the function g(b) = b+c (which can be expanded only if |b| < c).

To make this point clearer, now let us investigate how Leibniz squared the hyperbola in De quadratura arithmetica [KQA, 88–90]. He observed that, given the hyperbola GCH, the area of the square ABCD is equal to

G

D

C L

Q

H

A

B

M

E

Fig. 8 the rectangle AM LQ by the nature of the hyperbola (see Fig. 8). Therefore,

ML =

AB 2 AB 2 = . AM AB + BM

By expanding, he had

M L = AB − BM +

BM 2 BM 3 BM 4 BM 5 − + − + ... AB AB 2 AB 3 AB 4

under the condition BM < AB. Leibniz applied some theorems that he had proved earlier in De quadratura arithmetica and showed that the area of the ﬁgure BEHC (= to the sum all M L from BC to EH) was given from n the sum of the areas of the curves having ordinate equal to BM AB m (with M varying from B to E). Since AB is constant, the area AB is AB · AB; the 2 area of BM is a triangle of sides BE and BE, namely BE 2 , and so on. He,

2

39

Geometrical Quantities and Series in Leibniz

de facto, integrated term by term58 and obtained59 Area BEHC = AB·BE −

BE 2 BE 3 BE 4 BM 5 BE 6 + − + − +. . . . (22) 2 3AB 4AB 2 5AB 3 6AB 4

If one chooses BE appropriately so that BE < AB, then (22) makes it possible for any ﬁnite part of the given hyperbola to be squared: This means that, although the interval of convergence60 is ﬁnite, one can compute any ﬁnite part of the area by means of a change in the system of coordinates. The condition of convergence is not an a priori condition in the sense that it does not restrict the possibility of formal procedures. Indeed, even if Leibniz knew that (22) converged only if BE < AB, he posed AB = BE and obtained Area BEHC = AB 2 −

AB 2 AB 2 AB 2 AB 2 AB 2 + − + − + ... 2 3 4 5 6

and AB 2 2 Area BEHC = AB − 2 AB 2 AB 2 AB 2 AB 2 − − + + + ... 3 4 5 6 AB 2 AB 2 AB 2 + + + ... = 2 12 30 Hence, he found that the area of BEHC for BE =

1 2

is equal to

1 1 1 1 log 2 = + + + .... 4 8 48 120 58

In a modern form, B being the origin of the axes, if AB = b, BE = c, and BM = x, b2 , expanding it into Leibniz’s reasoning is equivalent to considering the hyperbola y = b+x the power series b − x + to obtain

x2 b

−

c

Area BEHC =

x3 b2

+

x4 b3

−

x5 b4

+ . . . and integrating it term by term in order

b2 c4 c5 c6 c2 c3 dx = bc − + 2 − 3 + 4 − 4 + ... b+x 2b 3b 4b 5b 6b

59

Leibniz derived several developments from expansion (22). For instance, for BE < AB and AB = 1, he reobtained Area BEHC =

BE 2 BE 3 BE 4 BM 5 BE 6 BE − + − + − + ... 1 2 3 4 5 6

(see p. 19). 60 Of course, Leibniz and the other mathematicians I discuss in this book did not possess the modern notion of interval of convergence, mainly because of the lacking of R (see Chapter 7). Nevertheless, the term “interval of convergence” can be used, conveniently i provided when referring to the interval over which the power series ∞ i=0 ai x converges, one does not refer to a subset of R, but simply that the series is convergent when the quantity x varies within appropriate limits.

40

Convergence and Formal Manipulation

The logical rule, according to which all the consequences of a theorem are subject to the same conditions of that theorem, does not seem to be applied. In my opinion, Leibniz did not break this law but merely thought that one could operate on a series independently of the conditions of convergence. What was important was that the results had a numerical or geometrical meaning, whereas the steps of the formal procedure that yielded a certain result might lack a numerical or geometrical meaning.61 The conditions of convergence were viewed as extrinsic to the question of ﬁnding the rule of the development of a quantity, whereas they actually concerned the application of a rule to the given problem. If one had to square a ﬁxed part of the hyperbola, then one ﬁtted the series to a speciﬁc situation: This was possible because the question of convergence could easily be circumvented by an appropriate choice of the coordinate system. It was precisely the geometrical context of De quadratura that made this approach possible. Not only was Leibniz’s main objective the computation of areas and other geometric quantities, but the theory of series also used ﬁgural representations in a crucial way. Here is another example. After having introduced logarithms,62 Leibniz showed that logc

AB + BX b+n AX = logc = logc AB AB b

logc

AZ AB − BZ b−m = logc = logc AB AB b

and

(where AB = b; BX = n; BZ = m) were proportional to the decreasing series n4 n5 n6 n3 n2 (23) + 2 − 3 + 4 − 5 + ... n− 2b 3b 4b 5b 6b and m+

m3 m4 m5 m6 m2 + 2 + 3 + 4 + 5 + ..., 2b 3b 4b 5b 6b

(24)

respectively (see Fig. 9). On p. 121 of De quadratura, Leibniz observed that if n > b, then the series (23) was not decreasing. However, in order to compute logc b+n b , namely to calculate the area BXLK, one could take the ﬁgure BKHZ such that the area BXLK was equal to BKHZ and calculate the area BKM Z (which is equal to logc b−m b ). Indeed, if n > b, then m < b, and (24) was decreasing. 61

Leibniz referred to this kind of derivation during the debate about the series 1 − 1 + 1 − 1 + . . . (see Chapter 9). 62 See Leibniz [KQA, 94–101].

2

Geometrical Quantities and Series in Leibniz

41

H

K L

A

Z

B

X

Fig. 9 In Quadratura arithmetica communis sectionum conicarum, a short paper written in 1691, Leibniz published his main results on the series expansions of geometrical quantities. In this article, apart from the expansion of logarithm (15), he provided the expansions of arctangent, sine, cosine, and exponential quantities:63 1 1 arctan x = x − x3 + x5 − . . . , 3 5

(25)

1 1 3 x + x5 − . . . , 3! 5! 1 1 1 − cos x = x2 − x4 + . . . , 2! 4! 1 1 ex − 1 = x + x2 + x3 + . . . . 2! 3! Later in Supplementum geometriae praticae [1693], Leibniz presented a method that made it possible to solve diﬀerential equations using series. This method, which is known today as the method of indeterminate coeﬃcients, is based upon the following principle: sin x = x −

63

Leibniz derived these expansions by methods similar to the ones we have just seen for logarithms (see [KQA]). For instance, the expansion 1 of arctan x can be brieﬂy summarized in modern terms as follows. Since arctan x = 1+x 2 dx, the term-by-term integration of 1 the expansion of 1+x provides (25). From this series one can obtain π/4. 2

42

Convergence and Formal Manipulation

(PIC) A series

∞

bk xαk is equal to 0 for every x on an interval

k=0

I if and only if all the coeﬃcients bk ( k = 0, 1, . . .) are separately equal to zero. Leibniz illustrated the method by means of several examples. I will give two of them here. The ﬁrst again concerns the expansion of logarithm. In Supplementum geometriae practicae [1693, 286], Leibniz stated that if y = a log a+x a , then adx dy = . a+x Hence, a

dy dy +x − a = 0. dx dx

(26)

He set y = Bx + Cx2 + Dx3 + Ex4 + . . . (he assumed the arbitrary constant to be equal 0). He diﬀerentiated term by term and obtained dy = B + 2Cx + 3Dx3 + 4Ex3 + . . . dx

(27)

By replacing (27) into (26), he derived (aB − a) + (2aC + B)x + (3aD + 2C)x2 + (4aE + 2D)x3 + . . . = 0. Equating the coeﬃcients to zero, he obtained aB − a = 0 2aC + B = 0 3aD + 2C = 0 4aE + 2D = 0 .... Hence, B = 1,

C=−

1 , 2a

D=

1 , 3a2

E=−

1 , 4a3

...,

and y =x−

x3 x4 x2 + 2 − 3 + ... . 2a 3a 4a

(28)

2

43

Geometrical Quantities and Series in Leibniz The second example concerns the equation a2 dx2 = a2 dy 2 + y 2 dx2 .

(29)

This equation has an easily understood geometric meaning. Indeed, given any circle, if the sine, the arc and the radius are denoted by y, x, and a, then (29) represents the sine in terms of the arc, and radius. In Supplementum geometriae practicae [1693, 287], Leibniz diﬀerentiated (29) and derived 0 = 2a2 dyd2 y + 2ydyd2 x (he supposed that dx was a constant, namely, in more modern terms, that y was the independent variable). Hence, a2

d2 y + y = 0. d2 x

(30)

Leibniz set y = Bx + Cx3 + Dx5 + Ex7 + . . . and obtained d2 y = 2 · 3Cx + 4 · 5Dx3 + 6 · 7Ex5 + . . . . d2 x

(31)

Substituting (31) into (30), he had (2 · 3a2 C + B)x + (4 · 5a2 D + C)x3 + (6 · 7a2 E + D)x5 + . . . = 0. By applying the principle of indeterminate coeﬃcients and assuming B = 1, he obtained B = 1,

C=−

and y =x−

1 , 2 · 3a2

D=−

C , 4 · 5a2

E=−

D , 6 · 7a2

...,

x3 x3 x7 + − + .... 2 · 3a2 2 · 3 · 4 · 5a2 2 · 3 · 4 · 5 · 6 · 7a2

(32)

In Supplementum geometriae praticae, Leibniz again stated that a series provided the exact solution to a diﬀerential equation only if it was considered as a whole; however, if one considered the partial sums of the series, then one obtained a solution that could be approximated as desired (Leibniz [1693, 286]). Leibniz made no further mention of convergence in this article, unlike in De quadratura and also Quadratura arithmetica communis sectionum conicarum [1691]. This stemmed from the fact that his aim in Supplementum geometriae praticae was to determine the form of the development, and questions concerning convergence therefore could be neglected.

44

Convergence and Formal Manipulation

It is worth noting that Leibniz regarded the variables x and y in Equation (26) as geometric quantities connected to the quadrature of hyperbola and the logarithmic curve. Similarly, the variables x and y in Equation (29) are geometric quantities connected to the circle. However, in Supplementum geometriae praticae, the immediate reference to geometrical ﬁgures is missing, and the reasoning regarded the way symbols were combined. Geometric applications were still the ﬁnal aim of his study of series, but these applications were not the heart of the Supplementum geometriae praticae: While the objective in De quadratura was to ﬁnd the area of the given ﬁgure, the aim in Supplementum geometriae praticae was restricted to ﬁnding the form of the development of the given quantity. However much Leibniz still intended to solve geometrical questions, the problem of the quadrature assumed a more analytical aspect; in other words, it became a question of manipulating symbolic expressions, which began to be isolated from the geometrical aspect. In this way, the search for the development of a quantity and the application of this development began to be viewed as two separate problems. Leibniz only dealt with the ﬁrst question in Supplementum geometriae praticae, whereas he examined both in De quadratura. Despite the diﬀerence in perspective, conditions of convergence were always considered as a posteriori conditions for the applicability of a development and not as a priori conditions for the development of quantities. Since convergence concerned the use of a speciﬁc series, conditions of convergence were not treated when the expansion of a quantity was sought, as is the case in Supplementum geometriae praticae. It can therefore be stated that the Leibnizian theory of series always contained a certain degree of formalism though it was hidden from the immediacy of geometrical reference. However, in some cases, the phase of application of the results is removed from the search for expansion and the formal character of series theory became more evident. In the years that followed, this trend toward a stronger type of formalism, which was still in an initial stage in Leibniz, was progressively emphasized, as we shall in Part II of this book.

3

The Bernoulli series and Leibniz’s analogy

In this chapter I follow the debate between Leibniz and Johann Bernoulli about the Bernoulli series and Leibniz’s analogy. This provides the possibility of highlighting the role of Johann Bernoulli, who, together with his brother Jacob, were the ﬁrst scholars who developed the work of Leibniz, and also of illustrating some other aspects of Leibniz’s work. In particular, I show the kind of results that were obtained when analysis did not concern concrete geometrical ﬁgures directly but concentrated on the manipulation of symbols. In 1694, following and developing Leibniz’s search of Supplementum geometriae praticae, Johann Bernoulli formulated the theorem:64

If n and z are two analytically expressed quantities,65 then

ndz = nz −

z 2 dn z 3 d2 n + + .... 2!dz 3!dz

(33)

This series, which was later known as the Bernoulli series, is equivalent to the Taylor theorem. According to Bernoulli, (33) was an improvement over the method of indeterminate coeﬃcients that Leibniz had used in Supplementum geometriae praticae [1693]. Bernoulli asserted that Leibniz’s method was “rather ingenious”; however for any application an individual calculation had to be undertaken by which only a particular series was produced. Instead, Bernoulli series was “a universal series which expressed generally all quadratures, rectiﬁcations, and integrals of other diﬀerentials” (Johann Bernoulli [1694, 437]). To derive (33) Bernoulli considered a quantity n and wrote z 2 dn z 2 dn ndz = ndz + (zdn − zdn) + − + 2!dz 2!dz

64

+

z 3 d2 n z 3 d2 n − 3!dz 3!dz

− .... (34)

On the Bernoulli series, see Feigenbaum [1985] and Panza [1992, 394–406]. Bernoulli stated that n is “a quantity formed in any way of indeterminates and constants”. This is a clear anticipation of the 18th-century notion of a function. It was precisely Johann Bernoulli who introduced this notion in his [1718], by using very similar words: “I call a function of a variable quantity, a quantity composed in whatever way of that variable quantity and constants” (Bernoulli [1718, 241]). It is appropriate to note that Bernoulli assumed dz to be a constant (z is an independent variable) and that he did not use the symbol of integral yet, but wrote integr. ndz. 65

45

46

Convergence and Formal Manipulation

By rearranging and integrating appropriately, he had

2 z dn z 3 d2 n z 2 dn zdn + + − ... + ndz = (ndz + zdn) − 2!dz 2!dz 3!dz z 2 dn z 3 d2 n = d(nz) − d + d + ... 2!dz 3!dz z 2 dn z 3 d2 n + + .... = nz − 2!dz 3!dz

The application of formula (33) allowed Bernoulli to obtain the expansions that had already been derived by Leibniz in Supplementum geometriae practicae [1693]. For example, in order to derive the expansion of the logarithm a+x , y = a log a Bernoulli observed that dy = and set n =

1 a+x

adx a+x

and dz = dx. He obtained dn dz d2 n dz 2

1 , (a + x)2 1 = , 2(a + x)3 .... = −

By applying (33), he had y = a

x3 x x2 dx + + .... = + 2 a+x a + x 2(a + x) 3(a + x)3

(35)

Bernoulli observed that this formula was diﬀerent from that of Leibniz [he referred to (28)], “nevertheless it has the same value”66 (Bernoulli [1694, 438]). Bernoulli also dealt with the diﬀerential equation for the sine. Indeed from (29) one had

a2 − y 2 dx dy = a and, by setting dz = dx and

a2 − y 2 , n= a 66

See Feigenbaum [1985, 83–84].

3

The Bernoulli Series and Leibniz’s Analogy

47

it is easy to obtain dn dz d2 n dz 2 d3 n dz 3

y dy y = − dx =− 2 2 2 a a a −y

a2 − y 2 1 dy = − 2 =− a dx a3 dy y y

dx = = 4 3 2 2 a a a −y ...

Hence, y=

a2 − y 2 dx = a

y 2 y 4 a2 − y 2 a2 − y 2 3 x + 2x − x − x + .... a 2a 3!a3 4!a3

Bernoulli divided this equation by y

Hence,

a2 − y 2

=

a2 − y 2 and obtained

x x3 yx2 yx4

− + .... + − a 2a2 a2 − y 2 3!a3 4!a3 a2 − y 2

x4 x x3 x2

+ . . . = − + ... 1− 2 + 2a 4!a3 a 3!a3 a2 − y 2 y

and

y

a2 − y 2

=

x3 x5 + 5!a 4 − ... 3!a2 . x2 x4 − 2a + 4!a3 + . . .

x− a

Bernoulli observed that the numerator of x5 x3 + 5!a 4 − ... 3!a2 x2 x4 − 2a + 4!a3 + . . .

x− a

is the sine series. This example shows that the Bernoulli series did not always succeed in isolating the sought-after quantity. Bernoulli himself noted this disadvantage of his method with respect to Leibniz’s: This is to remarked to the credit of the Incomparable Leibniz, that by his method what is sought results at once alone and not involved with other quantities. However, he will likewise not deny that mine is highly praiseworthy for its universality. (Johann Bernoulli [1694, 440–441])

48

Convergence and Formal Manipulation

Johann Bernoulli communicated this theorem to Leibniz on September 2, 1694 (see Leibniz [GMS, 3:150–152]). In response, in December 1694 (see Leibniz [GMS, 3:150–152]), Leibniz wrote that he had already found 3 d2 n 2 dn the expansion ndz = nz − z2!dz + z3!dz + . . . by means of techniques connected to his ﬁrst studies of Pascal’s triangle (see Leibniz [GMS 5, 396– 397]). Leibniz’s proof goes as follows. Given a decreasing sequence an , one can write ∞ Δak , a1 = a1 − a2 + a2 − a3 + a3 + . . . = k=1

where Δak = ak+1 − ak . Since Δak =

k−1

(−1)n

n=0

k−1 Δn+1 a1 , n

where Δn ak = Δn−1 ak+1 − Δn−1 ak , one has ∞ ∞ ∞ k k 2 a1 = Δ a1 + Δ3 a1 − . . . . Δa1 − 1 2 k=1

k=1

(36)

k=1

According to Leibniz, ∞

1 = x,

k=1 ∞

k=

k=1

∞ k(k + 1) k=1

2

x,

=

x,

...

Leibniz does not explain how these sums, which are rather similar to the sums (5) of Wallis and the sums (16) of Mercator, are derived. However, one might observe that, by considering x = 1 + 1 + 1 + . . . as an inﬁnite number, one has ⎞ ⎛ ⎞ ⎛ ∞ ∞ ∞ k ∞ ∞ ⎠ ⎝ ⎠ ⎝ x = x, etc. 1 = 1 = k= k=1

k=1

j=1

j=1

k=j

j=1

Leibniz sets y = a1 and dy = Δa1 , where dy is an inﬁnitesimal quantity, and assumes that dx is an inﬁnitely small quantity taken as a unity. He rewrote Equation (36) in the form x − .... y = xdy − d2 y x + d3 y

3

49

The Bernoulli Series and Leibniz’s Analogy

Since

x=

x2 2 ,

x=

x3 6 ,

. . . , Leibniz obtains

y = xdy −

x3 x2 2 d y + d3 y − . . . , 2 6

By taking dx = 1, one has y=x

x2 d2 y x3 d3 y dy − + − ... dx 2 dx2 6 dx3

(37)

and, “by moving ahead the y, dy, ddy, etc. into y, y, dy, etc., respectively”, x3 d2 y x2 dy + − .... (38) ydx = xy − 2 dx 6 dx2 Leibniz started from a decreasing sequence an , which was later identiﬁed with a continuous variable y. The proposition would have to be valid for y(x) decreasing. However, it is clear from the context that the conditions under which this series was derived were of little interest to Leibniz and Bernoulli. The result was acknowledged to be general: The condition regarding the decrease of the series did not limit the result. Bernoulli’s and Leibniz’s derivations of the Bernoulli series highlighted the fact that the formal aspect of the manipulation of series became much accentuated when the object of the study was not a speciﬁc series derived from speciﬁc geometrical quantities with reference to a concrete ﬁgure, but a generic series derived from a generic quantity. In this case, mathematicians could not refer to the properties of the speciﬁc geometrical quantities; rather they investigated what could be achieved through the mere combination of symbols that were not subjected to a priori conditions. This approach led Johann Bernoulli and Leibniz to obtain several interesting results. Indeed, in a letter written on February 28, 1695 (see Leibniz [GMS, 3:164–169]), Leibniz derived an expansion of xn dm y by reiterative applications of the formula xn dm y = xn dm−1 y − nxn−1 dxdm−1 y. In his proof there was an error of calculation; Bernoulli readily corrected it and gave the precise expansion: xn dm y = xn dm−1 y − nxn−1 dm−2 ydx + n(n − 1)xn−2 dm−3 ydx2 − . . .

(dx constant)

(see Leibniz [GMS, 3:169–174]). According to Leibniz, this formula implied diﬀerentials with negative exponents to be regarded as equivalent to integrations, namely d0 y = 0 y = y, d−1 y = 1 y, d−2 y = y = 2 y, . . .

50

Convergence and Formal Manipulation

(see Leibniz [GMS, 3:167]). This led Leibniz to formulate the famous analogy between the powers of a binomial and diﬀerentials of a product by observing that (x + y)1 = 1x + 1y = 1x1 y 0 + 1x0 y 1 , (x + y)2 = 1x2 + 2xy + 1y 2 , (x + y)2 = 1x3 + 3x2 y + 3xy 2 + 1y 3 , etc. and

d1 (xy) = 1ydx + 1xdy = 1d1 xd0 y + 1d0 xd1 y, d2 (xy) = 1d2 x + 2dxdy + 1d2 y, d3 (xy) = 1d3 x + 3d2 xdy + 3dxd2 y + 1d3 y, etc.

obey the same rule67 (see Leibniz [GMS, 3:174–177]). The combinatorial aspect, previously incorporated to diagrammatic representation, is now freely shown. In response, on June 1695, Bernoulli observed that d, d2 , d3 , . . . could be considered “as algebraic quantities and not just as characteristic letters” (see Leibniz [GMS, 3:180]). For instance, he considered the proportions d3 : d2 = d2 : d, d4 : d3 = d3 : d2 , . . . Later, in a letter on July 17, 1695 (see Leibniz [GMS, 3:197–205]), Bernoulli developed this idea. He stated that the third proportional between ndz = d−1 (ndz) and ndz = d0 (ndz) was d(ndz). Indeed, if one operated in the same way as in Leibniz’s analogy, one had d−1 (ndz) : d0 (ndz) = d0 (ndz) : d(ndz). By diﬀerentiating d(ndz), he obtained d(ndz) = d0 nd2 z + dndz and, therefore,

ndz = d−1 (ndz) =

d0 nd2 z d0 ndz (d0 (ndz))2 = 0 2 = 0 . d(ndz) d nd z + dndz d ndz + dn

67 In 1710 Leibniz published this analogy in a paper entitled Symbolismus memorabilis calculi algebraici et inﬁnitesimalis in comparatione potentiarum et diﬀerentiarum, et de lege homogeneorum transendentali [1710]. Here he used the symbols pe (x+y) and de (x+y), as in a letter to de l’Hˆ opital on September 30, 1695 [GMS, 1:301–302].

3

The Bernoulli Series and Leibniz’s Analogy

51

Since Mercator’s rule led one to ﬁnd two diﬀerent expansions of a quan 0 ndz a tity b+c , Bernoulli obtained two diﬀerent expansions of ndz = d0dndz+dn , namely d0 ndz = d0 nd0 z − d1 nd−1 z + d2 nd−2 z − d3 nd−3 z + . . . ndz = d0 ndz + dn = nz − dn z + d2 n 2 z − d3 n 3 z + . . . = nz −

z3 2 z2 z4 3 dn + d n− d n + . . . (dz constant) 2 2dz 6dz 24dz 3

and

d0 ndz = d1 zd−1 n − d2 zd−2 n + d3 zd−3 n − . . . dn + d0 ndz 1 2 2 3 = d z n−d z n + d z 3n − . . .

ndz =

=

n2 n3 2 n4 3 dz + d z− d z + . . . (dn constant) 2 2dn 6dn 24dn3

Thus, starting from Leibniz’s analogy, the Bernoulli series was once again obtained.68 Bernoulli stated: [W]hen I began to write these things, I did not indeed expect this result, thinking that I would arrive at a far diﬀerent series by this method. This elegant agreement wonderfully conﬁrms the probity of the methods, especially of this last, where so remarkably and contrary to all practice it is advanced with the letters d. (see Leibniz [GMS, 3:200], translation in Feigenbaum [1985, 89]) It is worthwhile noting that Leibniz also obtained a generalization of the Bernoulli series. In October 1695 he wrote to Johann Bernoulli that n (n + 1) 2 n n−1 n d y n+1 z + . . . (xy) = y z − ndy z+ 2! followed from dm (xy) = dm xd0 y + mdm−1 xd1 y +

m (m − 1) m−2 2 d xd y + . . . 2!

by letting m = −n and x = dz (see Leibniz [GMS, 3:221]). 68 Such an entirely formal procedure, which anticipated the typical aspects of the calculus of operations (see Chapter 21), was, however, a borderline case around 1700.

4

Newton’s method of series

In this chapter I investigate Newton’s contribution to the series theory. Newton obtained a large number of results concerning series, which he expounded upon in several papers but only partially published.69 Of these papers, the most interesting for my purpose are: • De analysi per æquationes numero terminorum inﬁnitas (see Newton [MP, 2:206–247]), which was probably written 1669, on the basis of previous results. It was published in 1711 in I. Newton, Analysis per quantitatum, series ac diﬀerentias cum enumeration linearum tertii ordinis, edited by W. Jones, London: ex oﬃcina Pærsoniana. • De Methodis serierum et ﬂuxionum (see Newton [MP, 3:32–329]). It was probably written in 1671 (based on a manuscript draft of 1666, see Newton [MP, 1:400-448]) but was published only nine years after his death by J. Colson in English translation with the title The Method of Fluxions and Inﬁnite Series; with Its Application to the Geometry [1736]. • Tractatus de quadratura curvarum [MP, 8:92–167]. It is a revision of De quadratura [MP, 7:48-129], which he wrote between December 1691 and the beginning of 1692. A wide-ranging abstract was also inserted by Wallis in his work Algebra (see Wallis [1693, 390–396]). It was published as one of the appendices of the Opticks: Or a Treatise of the Reﬂections, Refractions, Inﬂections & Colours of Light. Also Two Treatises of the Species and Magnitude of Curvilinear Figures, London: Smith and Walford (the other treatise being Enumeratio linearum tertii ordinis). • some pieces of his correspondence in Commercium epistolicum d. Johannis Collins et aliorium de analysi promota [CE]. This paper originated by the polemics with Leibniz, which Collins edited in 1712. In particular, it contains two very famous letters that are addressed to Oldenburg though they were actually written for Leibniz. They are usually known as epistola prior (the ﬁrst letter is dated June 13, 1676) and epistola posterior (the second is dated October 24, 1676). • some manuscripts of Newton published by Whiteside in [MP], especially the manuscripts that Whiteside entitled Annotations from Wallis [MP, 1:89–142]. Moreover, in discussing Newton’s concept of limit, I mainly refer to 69

For biographical information on Newton, see Westfall [1980] and Panza [2003].

53

54

Convergence and Formal Manipulation • Philosophia naturalis principia mathematica, whose ﬁrst edition dates back to 1687 and which was later republished in 1713 and 1726. It contains several mathematical propositions, in particular the muchdebated 11 mathematical lemmas of Book 1, Section 1, concerning the method of ﬁrst and last ratios.

Finally, in Chapter 6, while investigating the notion of quantity during the 17th and 18th centuries, I shall also deal with • Arithmetica universalis, sive de compositione et de resolutione arithmetica liber. It was based on lectures that Newton had left with the university library. It was published in 1707 by William Whiston, Newton’s successor as Lucasian Professor at Cambridge, without the name of the author. A second Latin edition was published in 1722 and an English version in 1720. My investigation of Newton’s method of series is subdivided into two sections. In Section 4.1, I ﬁrst deal with Newton’s procedures for expanding quantities —also expressed in the form of algebraic or diﬀerential equations— and stress the important role played by convergence, after which I examine the notion of limit and convergence. In Section 4.2, I discuss the manipulative aspects of the Newtonian theory of series and the relationship between manipulation and convergence. Newton’s contribution to the Taylor series is illustrated in Chapter 6.

4.1

The expansion of quantities into convergent series

The reading of Wallis’s Arithmetica inﬁnitorum 70 in 1664 had a great inﬂuence on Newton’s ﬁrst mathematical investigations. In the winter of 1664– 1665, he was able to derive his ﬁrst mathematical discovery, the binomial expansion: (p + q)r = pr +rpr−1 q +

r (r − 1) r−2 2 r (r − 1) (r − 2) r−3 3 p q + p q +. . . (39) 2! 3!

(r is any rational exponent) by applying Wallis’s method of interpolation71 . Newton’s procedure can be summarized as follows.72 At that time, mathematicians knew how to square the curves (1 − x2 ), (1 − x2 )2 , (1 − x2 )3 , . . . , but they did not possess a general method for squaring the irrational curves 1

(1 − x2 ) 2 , 70

3

(1 − x2 ) 2 ,

On Wallis’s inﬂuence on Newton, see Panza [1995]. See Chapter 1. 72 See Newton [CE, 67–86] and [MP, 1:104–111]. 71

5

(1 − x2 ) 2 ,

....

4

55

Newton’s Method of Series

1

Newton regarded the problem of the quadrature of the curves (1 − x2 ) 2 , 3 5 (1 − x2 ) 2 , (1 − x2 ) 2 , . . . as that of interpolating the following table: Curves 0 y = (1 − x2 ) 2 1 y = (1 − x2 ) 2 2 y = (1 − x2 ) 2 3 y = (1 − x2 ) 2 4 y = (1 − x2 ) 2 5 y = (1 − x2 ) 2 6 y = (1 − x2 ) 2

=1

Quadratures z=x

= (1 − x2 )

z = x − 13 x3

= (1 − x2 )2

z = x − 23 x3 + 51 x5

= (1 − x2 )3

z = x − 33 x3 + 53 x5 − 17 x7

In other terms, the problem was to ﬁnd an analytical expression Qm (x) such that, by giving the values 0, 1, 2, 3, . . . to m, the analytical expressions Q0 (x), Q1 (x), Q2 (x), Q3 (x), . . . provide the quadratures of 1, (1 − x2 ), (1 − x2 )2 , (1 − x2 )3 ,

....

Then the principle of Wallis’s interpolation made it possible to state that by giving the values 21 , 23 , 25 , . . . to m, the analytical expressions Q1/2 (x),

Q3/2 (x),

Q5/2 (x),

...

provide the quadratures of the curves 3

1

5

(1 − x2 ) 2 , (1 − x2 ) 2 , (1 − x2 ) 2 , . . . . Newton noted that the ﬁnite polynomials expressing the area under 1, (1 − x2 ), (1 − x2 )2 , (1 − x2 )3 . . ., were of the form Qm (x) = x −

m−2 m m−1 m 3 m m−1 2 2 3 x + x5 − x7 + . . . , 3 5 7

(40)

where m is an integer. He considered (40) valid even when m is not an integer; by setting m = 21 , he obtained the inﬁnite series Q1/2 (x) = x −

1 2

3

x3 −

1 8

5

x5 −

1 16

7

x7 + . . . ,

√ by which the semicircle 1 − x2 was squared. Newton went on to observe that y y y y

o

= (1 − x2 ) 2 2 = (1 − x2 ) 2 4 = (1 − x2 ) 2 6 = (1 − x2 ) 2

= 1, = 1 + x2 , = 1 − 2x2 + x4 , = 1 − 3x2 + 3x4 − x6 .

(41)

56

Convergence and Formal Manipulation

These expansions obeyed the rule (1 − x2 )m = 1 − mx2 + m

m−1m−3 6 m−1 4 x −m x + ... 2 2 3

(42)

By considering (42) valid for a noninteger m, he obtained the binomial expansion. In particular,73 by setting m = 12 , he had

1 1 1 1 − x2 = 1 − x2 − x4 − x6 + . . . 2 8 16

(43)

Newton also obtained the quadrature of the hyperbola (and therefore the expansion of the logarithm74 ) using the same method. Indeed, in his [MP, 1:112-115] he considered the following scheme Curve y = (1 + x)0 y = (1 + x)1 y = (1 + x)2 y = (1 + x)3

=1 =1+x = 1 + 2x + x2 = 1 + 3x + 3x2 + x3

Quadrature z=x z = x + 21 x2 z = x + x2 + 13 x3 z = x + 32 x2 + x3 + 41 x4

and noted that the general rule was Curve y = (1 + x)n = 1 + nx + Quadrature z = x + n2 x2 +

n(n−1) 3 1·2·3 x

n(n−1) 2 1·2 x

+

+

n(n−1)(n−2) 3 x 1·2·3

n(n−1)(n−2) 4 x 1·2·3·4

+ ...

+ ...

Then the quadrature of the hyperbola could be obtained by setting n = −1 Curve 1 = 1 − x + x2 − x3 + . . . y = 1+x

Quadrature z = x − 12 x2 + 31 x3 − 14 x4 + . . .

Three comments should be made at this point. First, making the step from an integer n to a fractional or negative n involved the step from a ﬁnite series to an inﬁnite series. Newton regarded this as unproblematic: He made no distinction between ﬁnite series and inﬁnite series. For instance, in De analysi he wrote: Whatever common analysis performs by means of an equation with a ﬁnite number of terms (whenever it may be possible) can be performed by means of inﬁnite equations. (Newton [1711, 1:280]) 73 Newton also understood the relationship between the coeﬃcients of the series (41) and (43) and so he poses the basis for the algorithm of the calculus (on this, see Panza [1989]). 74 It should be noted that while in his letter to Oldenburg on October 24, 1676, Newton explicitly identiﬁed the area of hyperbola with logarithms (cf. Newton [CE, 67–86]), in his young paper he made no mention of this fact (cf. Newton [MP, 1:104–111]).

4

57

Newton’s Method of Series

Second, even if Newton’s derivation of the binomial expansion (39) is formal, he payed attention to convergence. For instance, in writing to Leibniz, Newton considered several applications of the binomial expansion, which he wrote in the form m

(P + P Q) n = P

m n

+

m m m (m − n) m 2 P nQ+ P n Q + ... n 2n2

(44)

In the case of the function

5

he ﬁrst set P = c5

c5 + c4 x − x5 , and

Q=

c4 x − x5 c5

and obtained

c4 x − x5 2c8 x2 − 4c4 x6 + 2x10 5 − + ... c5 + c4 x − x5 = c + 5c4 25c9

He then set

P = −x5

and

Q=−

c4 x + c 5 x5

and obtained

c4 x + c5 2c8 x2 + 4c9 x + 2c10 5 c5 + c4 x − x5 = −x + + + .... 5x4 25x9

Finally, he observed that the ﬁrst procedure is preferable when x is very small, the second when it is very large (see Newton [OO, 1: 285-289]). Third, Newton used Wallis’s interpolation in an improved and simpliﬁed form, which can be schematized as follows: 1. Some speciﬁc cases P (1), P (2), . . . are given; 2. an appropriate formula P (k) is determined such that P (1), P (2), . . . are speciﬁc cases of P (n); 3. if the formula P (n) has an appropriate quantitative meaning for nonnatural numbers (usually a rational number, but in principle any number), then P (n) is considered valid for nonnatural numbers. The complicated argumentation used by Wallis, which was often based on geometrical reasoning and consisted of the investigation of speciﬁc tables of numerical values, had now disappeared. They are replaced by the search for a general formula. This interpretation of Wallis’s interpolation paved the way for the research of later mathematicians.

58

Convergence and Formal Manipulation

Newton invented other procedures for expanding analytically expressed quantities. Two of them — the long division and extraction of root— are known as Mercator’s rules, even if Mercator published only the long division in his Logarithmotechnia. I have already dealt with long division in Chapter 1, p. 20 and Section 2.2, p. 37. Concerning the extraction of roots, Newton noted that the common procedure of extraction of roots made it possible to calculate the terms of the development of an analytical

√ of the expression i of type f (x). For instance, to obtain the development ∞ a x a2 + x2 i=0 i (see Newton [MP, 3:45-50]), one can proceed as follows:75 i. The ﬁrst term a0 of the expansion was the square root of a2 , namely √ a0 = a2 = a; ii. one calculated the ﬁrst remainder R0 = (a2 + x2 ) − a2 = x2 ; iii. one looked for the second term a1 x such that (a0 + a1 x)2 = a2 +R0 + (a1 x)2 , and got x2 R0 = ; a1 x = 2a 2a 2 2 iv. one multiplied x2a by 2a + x2a and obtained Γ1 = x2 +

x4 ; 4a2

v. one calculated the second remainder R1 R1 = R0 − Γ1 = −

x4 ; 4a2

vi one repeated the step (iii ) on R1 in order to ﬁnd a2 x2 , that is, one 2 looked for a2 such that a0 + a2 x2 = a2 + R1 + x2 , and got a2 x2 =

x4 R1 = − 3; 2a 8a

vii. one considered x2 x4 x4 x2 x8 x4 − 3 =− 2 − 4 + ; Γ2 = − 3 2 a + 8a 2a 8a 4a 8a 16a6 viii. one calculated the third remainder R2 = R1 − Γ2 = − 75

See Ferraro and Panza [2003, 24–25].

x2 x8 + ; 4 8a 16a6

4

59

Newton’s Method of Series and so on. One thus obtained76

x4 x6 x2 − 3+ + .... a2 + x2 = a + 2a 8a 16a5

(45)

The following scheme clariﬁed the diﬀerent steps of the procedure: a2 + x2

a+

x2 x4 − 3 + ... 2a 8a

a2 + x2 ...

x4 4a2 x4 − 2 4a x2 x8 x4 − 2− 4+ 4a 8a 16a6 2 x8 x − 8a4 16a6 ...

x2 +

Another of Newton’s procedures is today known as Newton’s method of parallelogram. It makes it possible to solve a given algebraic equation ai,j xj y i = 0 P (x, y) = i,j

∞

k by means of a power means of a ∞series α k=0 bk x or, in certain cases, by 77 series of the kind k=0 bk x k , where αk is a rational number. Newton considers tables of the type

j x4 x3 x2 x 0

x4 y x3 y x2 y xy y

x4 y 2 x3 y 2 x2 y 2 xy 2 y2

x4 y 3 x3 y 3 x2 y 2 xy 3 y3

x4 y 4 x3 y 4 x2 y 4 xy 4 y4

i

and marks all the squares of coordinates (i, j) corresponding to the terms ai,j xj y i of the given equation i,j ai,j xj y i = 0. For instance, if the equation y 3 + a2 y + axy − 2a3 − x3 = 0

76 77

The same result could be derived by applying (44). See Newton [MP, 3:51–57] and [MP, 2:218–233].

60

Convergence and Formal Manipulation

is given, one marks the squares (3, 0), (1, 0), (1, 1), (0, 3), (0, 0) corresponding to the terms y 3 , y, xy, x3 , 0.

∗ ∗

∗ ∗

∗

One chooses the lowest marked square in the column farthest to the left, namely, one chooses the square (i1 , j1 ), where • i1 = min{i : (i, j) ∈ S}, • j1 = min{j : (i1 , j) ∈ S}, • S = {(i, j) such that (i, j) are the coordinates of marked squares}. One then draws a line from the square A ≡ (i1 , j1 ) to another marked square, say B ≡ (i2 , j2 ), such that all the rest of the marked squares either are in contact with the line AB or lie above it. Then one considers all the terms that are found to be in contact with the line AB and forms a “ﬁctitious” equation by making terms equal to zero. An appropriate root of this equation is the ﬁrst term, say y1 , of the sought-after expansion of P (x, y) = 0. In the example the line AB is the horizontal line joining (0, 0) and (3, 0) and the considered terms are −2a3 , a2 y, y 3 . He thus obtains the equation y 3 + a2 y − 2a3 = 0. One of the roots of this equation is y1 = a; then y=a is the ﬁrst term of the sought-after expansion

∞

bk xk .

k=0

By replacing y = y1 + p

into the given equation P (x, y) = 0, Newton obtains a new equation Q(x, p) = 0 and, by repeating the procedure, he ﬁnds the other terms of the expansion. In the example he replaces y = a + p in the given equation and obtains the equation p3 + 3ap2 + axp + 4a2 p + a2 x − x3 = 0. (46)

4

61

Newton’s Method of Series

The corresponding diagram is

∗ ∗

∗ ∗

∗

∗

By repeating the reasoning, it is possible to ﬁnd a new ﬁctitious equation a2 x + 4a2 p = 0. The root is p = − 41 x. This is the second term of the sought-after expansion. Then he sets 1 p=− x+q 4 and substitutes it into (46), etc. In this way Newton obtains y =a−

x2 131x3 x + + + .... 4 64a 512a2

The crucial idea of is the following. the By substituting this procedure k for y in P (x, y) = j y i , one should b x a x indeterminate series ∞ i,j k i,j k=0 obtain a new polynomial j+ki Q(x) = Br xr . Ai,j xj+ki = bk x = ai,j i,j

If the series

∞

k k=0 bk x

i,j,k

k

r

is the sought-after solution of P (x, y) = 0, then

Br xr = 0

r

and all the coeﬃcients Br of the powers xr have to be separately equal to zero, namely the terms xj+ki should cancel each other out. Therefore, Newton de facto uses the principle of indeterminate coeﬃcients.78 In order to justify how draws the lines, note that if xs is the Newton r term of lower degree in r Br x and if Bs is equal to 0, then the exponent s is obtained in at least two diﬀerent ways as a sum of the type j + ki, say j1 + ki1 and j2 + ki2 . Moreover, xs = xj1 +ki1 = xj2 +ki2

78

See Chapter 2.

62

Convergence and Formal Manipulation

is eﬀectively the term of lower degree only if there is no point (i, j) below the straight line joining ( i1 , j1 ) and (i2 , j2 ).79 Newton observed that if one did not consider small values of80 x, but values that diﬀered by a small amount from a number c, one could replace x by z = x − c into the equation. Further if x was large, one could replace x by z = x1 . In this way, he was sure to obtain a good approximation of y as a function of x since the series was convergent over at least an appropriate certain interval. In the 1692 version of De quadratura curvarum [MP, 7:93-96] Newton also illustrated a method to expand the solution of a diﬀerential equation into series that can be considered a generalization of the previous method. Indeed Proposition 12 of this treatise contains the following problem: Proposition 12. Given an equation involving two ﬂuent quantities (if any, together with their ﬂuxions), express the relation between these quantities by means of a convergent series. (Newton [MP, 7:93]) To solve this problem, Newton (see [MP, 7:93-100]) considered the differential equation F (z, y, y ′ , y ′′ , . . .) = 0, where F is an expression of the type pi z µi y αi (y ′ )β i (y ′′ )γ i . . . (y (k) )ωi . (47) F (z, y, y ′ , y ′′ , . . . , y (k) ) = i

az n

If one substitutes y = (a and n to be determined) into the equation F (z, y, y ′ , y ′′ , . . .) = 0, then one has F [z, az n , anz n−1 , an(n − 1)z n−2 , . . .] = 0. This equation can be written as F [z, az n , anz n−1 , an(n − 1)z n−2 , . . .] = Az α + Bz β + Cz γ + . . . + Sz σ = 0, for α < β < γ < . . . < σ. The coeﬃcients A, B, C, . . . , S depend on a, while the exponents α, β, γ, . . . , σ depend on n and have the form µi + (αi + β i + γ i + δ i + . . .)n − (β i + 2γ i + 3δ i + . . .). The number n is taken by considering the summands of F (x, y, y ′ , y ′′ , . . .) of the type Kz µi (without y, y ′ , y ′′ , . . .) and setting n = max i

λ − µi + β i + 2γ i + 3δ i + . . . , αi + β i + γ i + δ i + . . .

79 Newton gave examples where the points (i1 , j1 ) and (i2 , j2 ) are of the types (0, j1 ) and (i2 , 0). It can happen, however, that the distribution of the marked rectangles (i, j) gives rise to more complicated situations (see Chabert [1999, 195–196]). 80 The series k=0 bk xk is convergent if x is small.

4

63

Newton’s Method of Series

where λ is the minimum exponent µi . Thus, the lowest exponent α in Az α + Bz β + Cz γ + . . . + Sz σ is equal to λ. Since Az α + Bz β + Cz γ + . . . + Sz σ

(48)

= z α (A + Bz β−α + Cz γ−α + . . . + Sz σ−α ) = 0, one can ﬁnd a by solving the equation A = 0.81 As the next step in the procedure, one sets y = az n + p and replaces y, y ′ , . . . into F (x, y, y ′ , . . .): Thus one obtains a new function G(z, p, p′ , p′′ , . . .) and can repeat the same reasoning. In this way, step by step, one ﬁnds a series that is a solution of F (z, y, y ′ , y ′′ , . . .) = 0. Newton observed that this method worked well when the quantity x is small enough. This is equivalent to recognizing that the series is not necessarily always convergent. However, if one desires the development at a point x that is not small enough, it is suﬃcient to set x = w + v, where w is an appropriate constant, and to proceed as earlier with respect to the variable v. It is clear that in this case Newton found an expansion of the type ∞ Ai (x − w)i , i=0

convergent over a neighborhood of w.82 In a letter to Wallis (see Newton [OO, 1: 294]), the method was applied to y 2 − z 2 y ′ − d2 + dz = 0. In the ﬁrst step, the summand of the lowest degree (excepting the terms with y and y ′ ) is d2 , and therefore λ = 0. Since n = max i=1,2

λ − µi + β i = 0, αi + β i

81

Higher-order inﬁnitesimals are neglected near zero and so Equation (81) is reduced to Az α = 0, which is constantly equal to zero when A = 0. In this way an approximate solution is established. 82 Newton also considered the case when x is large (quantitas permagna). In this case, in order to determine the exponent n of y = az n one always considers the summands of F (x, y, y ′ , y ′′ , . . .) of the type Kz µi (without y, y ′ , y ′′ , . . .) but sets n = max i

ξ − µi + β i + 2γ i + 3δ i + . . . . , αi + β i + γ i + δ i + . . .

where ξ is the maximum exponent µi .

64

Convergence and Formal Manipulation

he had y = a and y ′ = 0. By substituting into the initial equation, he obtained a2 − d2 + dz = 0. The equation a2 − d2 = 0 allowed him to determine a = d. In the second step, he set y = d + p and obtained 2dp + p2 − z 2 p′ + dz = 0. It is easy to verify that λ = 1 and n = 1; hence, he set p = az and, substituting it into the last equation, found a = 12 . Consequently, the ﬁrst two terms of the expansion of y are d− z2 . In the third step, he set p = − z2 +q, 2 repeated the reasoning, and found q = − 3z 8d . Proceeding in this way, he obtained the expansion 3 z2 9 z3 1 + .... − y =d− z− 2 8 d 16 d2 At this juncture, I explicitly note that Newton’s procedures cannot be applied to a generic function, if one gives to this term the modern meaning (namely, if function is taken to mean a relation that associates an element y of a given set B to an element of another given set A). Newton’s procedures were applied to the analytical expressions of geometrical quantities; more precisely, they were applied to a very special kind of analytical expression: the ﬁnite analytical expressions obtained by the composition of elementary operations (addition, subtraction, multiplication, division, extraction of root, raising to power).83 Consequently, when Newton used the term “equation”, he was referring to the equality between the analytical expressions composed by means of elementary operations84 . For instance, the method of Proposition 12 is the general solution of the problem of ﬁnding the solution to a ﬂuxional equation since, in Newton’s opinion, any equation involving two ﬂuent quantities and their ﬂuxions had the form F (x, y, y ′ , y ′′ , . . .) = 0, where F is an analytical expression of type (47).85 I would also like to emphasize the fact that Newton thought that the power series that resulted from the expansion of any quantity was convergent 83 The expansion of quantities, such as trigonometric quantities, that lacked an analytical expression at the time, was derived by using their connection with certain analytical expressions. For instance, the expansion of the sine was derived by using the fact that the ﬂuxion of arcsine is an analytical expression (see Section 4.2). 84 One might distinguish between algebraic and ﬂuxional equations (but the distinction is not explicit in Newton). If the variables are ﬂuent and ﬂuxions, we can refer to them as ﬂuxional equations. 85 For this reason, if one might for the sake of discussion use the term “function” in reference to Newton (Newton, as is known, never used it), provided the word “function” is taken to mean an algebraic function. The 18th-century notion of a function is examined in Chapter 18.

4

Newton’s Method of Series

65

at least when the variable varied for an appropriate interval of values of x. This opinion, for instance, is the basis of the following proposition, which is found in the 1692 version of De quadratura (it is a corollary of Proposition 12): All curves can be squared by means of indeterminate convergent series. (Newton [MP, 7:96]) Newton explained that, given a curve F (x, y) = 0, it is suﬃcient to express the ordinate y = y(x) by a series using the above method and integrating it term by term. At this point, it is appropriate to try to clarify the meaning of the expression “convergent series” in Newton’s work. In order to do this, I investigate the closely connected concept of limit as formulated in Book 1, Section 1 of Principia mathematica, where Newton expounded the basic notions of the calculus in geometrical forms. In the ﬁnal scholium Newton stated: For 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 in inﬁnitum. (Newton [PN, 87])86 One might be tempted to consider this sentences as a deﬁnition of limit, and in this case the expression “ultimate ratio” would denote something very similar to the modern limit concept (the point of view, e.g., of Pourciau [1998] and [2001]). However, I believe that a diﬀerent interpretation is preferable. For instance, consider the ﬁrst of the 11 mathematical lemmas of Book 1 of Principia: Lemma 1. Quantities, and the ratios of quantities, which in any ﬁnite time converge continually to equality, and before the end of that time approach nearer the one to the other than by any given diﬀerence, become ultimately equal. (Newton [PN, 73]) This lemma contains the following explicit hypotheses: (H1) two quantities, say A(t) and B(t), approach closer and closer to one other, when t varies over a ﬁnite interval I, whose endpoints are a and c, and approach c. (H2) A(t) and B(t) approach so close to one other that their diﬀerence is less than any given quantity, namely it is |A(t) − B(t)| < ε, when t < c but near enough to c. 86

The translations from Principia are by Motte (see Newton [M]).

66

Convergence and Formal Manipulation

Hypothesis (H1) implies that A and B approach each other, but this does not necessarily mean that the distance between A and B becomes smaller than any quantity [for instance, A(t) = −t2 and B(t) = t2 + 1 (t → 0)] satisfy hypothesis (H1)). Hypothesis (H2) guarantees that the distance actually becomes smaller than any given quantity. The thesis is (T) A(t) and B(t) are ultimately equal. The thesis states that the two quantities eﬀectively reach each other when t = c. If we use the term “limit” in the same way as Newton does in the scholium, the thesis states that the quantities A(t) and B(t) have the same limit or that the limit of their diﬀerence is 0. In the proof, Newton87 assumes that D > 0 is the ultimate diﬀerence, namely, |A(c) − B(c)| = D; then |A(t) − B(t)| does not become less than D, contrary to hypothesis (H2). It is clear that if the proof is to be taken seriously, (H2) and (T) are not the same thing, and this implies that Newton did not think of (H2) as the deﬁnition of limit or ultimately equal.88 (H2) is an essential property of limit but not the deﬁnition. In eﬀect, Newton did not deﬁne the terms “limit” and “ultimate ratio”: These terms had a clear intuitive meaning to him.89 In the Principia he illustrated this intuitive meaning by referring to the “limit” as the last place or the last velocity of a motion.90 For Newton, the notion of limit or ultimate value was an idea borrowed from nature; it was not a mathematical notion determined by its deﬁnition. A translation into modern terminology would strain Newton’s concept.91 87 “If you deny it; suppose them to be ultimately unequal, and let D be their ultimate diﬀerence. Therefore they cannot approach nearer to equality than by that given diﬀerence D; which is against the supposition” (Newton [PN, 73]). 88 See also Panza [2003, 194–195]. 89 This behavior is similar to Leibniz’s, who did not deﬁne the “ultimate value” (see Chapter 2). 90 “Perhaps it may be objected, that there is no ultimate proportion of evanescent quantities; because the proportion, before the quantities have vanished, is not the ultimate, and when they are vanished, is none. But by the same argument it may be alleged, that a body arriving at a certain place, and there stopping, has no ultimate velocity; because the velocity, before the body comes to the place, is not its ultimate velocity; when it has arrived, is none. But the answer is easy; for by the ultimate velocity is meant that with which the body is moved, neither before it arrives at its last place and the motion ceases, nor after, but at the very instant it arrives; that is, that velocity with which the body arrives at its last place, and with which the motion ceases. And in like manner, by the ultimate ratio of evanescent quantities is to be understood the ratio of the quantities, not before they vanish, nor afterwards, but with which they vanish. In like manner the ﬁrst ratio of nascent quantities is that with which they begin to be. And the ﬁrst or last sum is that with which they begin and cease to be (or to be augmented or diminished). There is a limit which the velocity at the end of the motion may attain, but not exceed. This is the ultimate velocity. And there is the like limit in all quantities and proportions that begin and cease to be” (Newton [PN, 87]). 91 For a more general discussion of the diﬀerences between the 17th- and 18th-century

4

67

Newton’s Method of Series

Note that the absence of appropriate deﬁnitions was not restricted to this case. For instance, in Lemma 2 of Book 1 of Principia [PN, 73–74], Newton proved that the area under a curve was the limit of the circumscribed and inscribed rectangles. However, he did not deﬁne the area (nor the integral92 ). The area is a geometrical and physical entity that existed per se prior to any possible deﬁnition.93 Of course, this concept of limit has remarkable consequences for the notion of the sum. For Newton, summing a series ∞ i=0 ai meant gradually combining all its terms in order to obtain the last term a1 + a2 , (a1 + a2 ) + a3 , (a1 + a2 + a3 ) + a4 , ..., (a1 + a2 + . . . + an−1 ) + an , .... It made no diﬀerence whether the last term n was ﬁnite or inﬁnite. It was intuitively clear that if S was the sum of the series ∞ i=0 ai , the property (∗)

|S −

k i=0

ai | < ε,

for every ε and large k

held, and vice versa if property (∗) held, S was the sum of the series. However, Newton did not regard (*) as the deﬁnition of the sum. According to him, the ﬁniteness of human beings precluded them from naming (designating) and conceiving all the terms of a series and therefore determining the precise value of the quantity expressed by the series94 (Newton [1711, 280]). As we shall see below, this conception prevented the development of the notion of limit in a modern sense and favoured the growth of a more formal concept of the sum of a series.

4.2

On Newton’s manipulations of power series

Power series played a fundamental role in Newtonian calculus. For instance, in De analysi [MP, 2:206–207], Newton based his method of squaring a curve upon the following three rules: notion of limit and the modern one, see Chapter 7 92 For a diﬀerent opinion, see Pourciau [1998]. 93 On the role of deﬁnitions in the 17th- and 18th-century mathematics, see Chapter 18. 94 However, he stated that reasonings about series were certain and equations involving series were deﬁnitely correct (Newton [1711, 280]). In other words, a series could adequately represent a quantity in analytical calculations, but only the knowledge of the whole series provided the exact values of a quantity.

68

Convergence and Formal Manipulation 1. The area95 under the curve of equation y = axm/n is

m+n na ax n . m+n

2. If the equation y = y(x) of a curve is given by the sum of a ﬁnite or inﬁnite number of terms y1 + y2 + y3 + . . ., the area under the curve y is equal to the sum of the areas of all terms y1 , y2 , y3 , . . .. 3. If the curve has a more complicated form, then one αmust expand the equation of the curve into a series of the type ak x k , where αk is a rational number, and apply rules 1 and 2. In the third point, Newton assumed that any analytically expressed quantity could be expanded into a series of the type ak xαk , where αk is a rational number. This series can be squared by integrating term by term using rule 1. Similarly, in De Methodis [MP, 3:70–71], Newton reduced many geometrical problems (determination of tangents, maxima and minima, areas, surfaces, curvatures, arc lengths, centers of gravity, etc.) to two problems, which he formulated as follows: (1a) Given the length of the space continuously (that is, at every time), ﬁnd the velocity of motion at any time proposed. (2a) Given the speed of the motion continuously, ﬁnd the length of the space described at any time proposed. When Newton introduced the terms “ﬂuent” and “ﬂuxion”, these problems became (1b) Given the relation between the ﬂuents, ﬁnd the relation between the ﬂuxions. (2b) Given the relation between the ﬂuxions, ﬁnd the relation between the ﬂuents. In modern terms, problem (2b) consists of seeking the solution to a diﬀerential equation. Newton suggested two ways of solving it. 95

This rule was already known (see Chapter 1). On the rise of Newton’s calculus, see Panza [1995].

4

69

Newton’s Method of Series .

.

First, one tried to transform a ﬂuxional equation f (x, y, x, y) = 0 so that the problem could be reduced to seek the ﬂuent in a known catalogue of ﬂuents (in more modern terms, a table of integrals). Second, one sought a solution expressed using series. We have seen that in the 1692 version of De quadratura, Newton proposed a general method for solving equations by series. Instead, in de Methodis to solve an equation . . f (x, y, x, y) = 0, Newton transformed it in the form .

y . = g(x, y); x then he expanded g(x, y) into series and integrated term by term. It is worthwhile noting that, despite the huge importance of series in the calculus, solutions in closed form were preferred with respect to series solution. For instance, in Proposition 11 of De quadratura, Newton formulated the following proposition: Proposition 11. Given an equation involving the ﬂuxions of two ﬂuent quantities, ﬁnd the relation between these quantities. This problem is similar to the problem of the subsequent Proposition96 12, the diﬀerence is that, in Proposition 11, Newton sought a solution in ﬁnite form (see Newton [MP, 1:72–92]). Only when the equation could not be solved by means of a ﬁnite expression did Newton suggest solving this by means of convergent series, according to the method of Proposition 12, and then by trying to reduce the series to a ﬁnite equation. The series solution oﬀered a useful representation of a quantity, which could be handled and made it possible to compute the value of the quantity by approximation, but, where possible, one had to seek solutions in closed form. In the previous Section 4.1, I underlined the importance of convergence for Newton, although we also saw that Newton’s procedures were formal ones based upon the inﬁnite extension of ﬁnite procedures. I shall now move on to investigate the relationship between convergence and manipulation of series in more detail, and I shall show that the Newtonian conception was not substantially diﬀerent to Leibniz’s with regard to this topic (in particular, convergence was not a preliminary condition to manipulation of power series). For instance, to ﬁnd the area enclosed by the curve y=

1 1 + x2

and the axis of abscissa, Newton (see [OO, 1:264]) observed that 1 : (1 + x2 ) 96

See Proposition 12 on p. 62.

70

Convergence and Formal Manipulation

is equal to 1 − x2 + x4 + . . . by continued division. It is then easy to express the area by the series 1 1 y = x − x3 + x5 + . . . 3 5 by integrating term by term. According to Newton, however, one can also divide 1 : (x2 + 1). In this case the result is y = x−2 − x−4 + x−6 + . . . and the area is given from 1 1 y = −x−1 − x−3 + x−5 + . . . . 3 5 The former expansion, Newton said, was to be used when x was small (satis parva) enough whereas the latter was to be used when x was large enough (satis magna). The diﬀerence of the developments 1 : (1 + x2 ) and 1 : (x2 + 1) was not viewed as a defect but as a useful tool for facilitating the computation of areas. A preliminary analysis of convergence seemed not only superﬂuous but even counterproductive. It was preferable to expand a series formally: After determining the coeﬃcients and form of development, one veriﬁed if the interval of convergence was suitable for the speciﬁc problem and, if there were any, chose the series that ﬁtted it best.97 Convergence concerned the moment of the application, namely when one computed the values of the given series in order to ﬁnd the values of a certain geometrical (or physical) quantity. As we already saw in Leibniz,98 such an approach was encouraged by the facts that, for practical aims, the lack of convergence in a certain interval appeared fairly easy to avoid by employing appropriate strategies. For instance, consider the ﬂuxional equation .

2 y + 3 − x2 . . . . . = x x Newton [OO, 1:418] solved it by replacing x by z + 1 and expanding the right side into power series. He obtained .

2 y + 2 − 2z − z 2 = 4 − 4z + z 2 − 2z 3 + 2z 4 + . . . . = 1+z z and y = 4z − 2z 2 − 97 98

See also the example on p. 57. See Section 2.2.

z3 + .... 3

4

71

Newton’s Method of Series

Of course, this series converges only for |z| < 1 (0 < x < 2). However, this diﬃculty can be overcome by using a diﬀerent substitution b + z and computing y(x), for every |x − b| = |z| < b. This method of tackling convergence was based on the assumption, which I mentioned on p. 64, that power series were always convergent for appropriate values of variables. It went tacitly that a convergent series at a speciﬁc point x could be easily determined by means of appropriate substitutions, and this was suﬃcient to obtain numerical approximations.

I

E

D M

F

A

B

Fig. 10

This approach was closely connected with geometric aims of the early calculus. For instance, to calculate the area ABDE under the hyperbola

y=

a2 b+x

(see Fig. 10), Newton expanded this analytical expression so to obtain series a2 a2 a2 a2 = − 2 x + 3 x2 − . . . b+x b b b

(49)

72

Convergence and Formal Manipulation

By integrating this series, he obtained99 area (ABDE) =

a2 a2 a2 a2 x − 2 x2 + 3 x3 − 4 x4 + . . . . b 2b 3b 4b

Of course, this derivation is valid under the condition (0 1 + na. Bernoulli’s proof runs as follows. Since A : B = B : C, then A + C > 2B = A + F 111 and C > F . Moreover, A + D > B + C > B + F = A + G. Hence, D > G, so on. Then Bernoulli proved the following two statements. d. An increasing geometric progression112 a1 , a2 , . . . can be continued up to a term an greater than any given number Z. e. A decreasing geometric progression a1 , a2 , . . . can be continued up to a term an less than any given number Z. To prove theorem e, Bernoulli considered an increasing geometric progression b1 , b2 , . . . such that b1 = Z and the ratio b1 : b2 is the reciprocal of the ratio a1 : a2 . This progression can be continued up to a term bn > a1 (for proposition d). Now continue the progression a1 , a2 , . . . up to an . Since b1 an = a1 bn

and

a1 < bn ,

one obtains an < b1 = Z (see Bernoulli [P, 244]). Bernoulli gave two corollaries to the theorems d and e: f. The ultimate terms of an increasing geometric progression is ∞. g. The ultimate terms of a decreasing geometric progression is 0 (see Bernoulli [P, 244]).

110 This result had already been proved in a similar way by Barrow in his Lectiones geometricae [1670, 60]. 111 Bernoulli explicitly refers to Euclid’s Elements, Book 5, Proposition 25: If four magnitudes are proportional, then the sum of the greatest and the least is greater than the sum of the remaining two. 112 Bernoulli indicated the progression by A, B, C, D, E as in the previous theorem. I prefer to use a more modern symbolism.

5

81

Jacob Bernoulli’s Treatise on Series

Following Euclid, Bernoulli stated that if ai , i = 1, . . . , n, is a ﬁnite geometric progression then a1 : a2 =

n−1

ai :

n

ai .

i=2

i=1

Therefore, sn =

a1 (a1 − an ) + an , a1 − a2

where sn = ni=1 ai . Then, following Vi`ete(see p. 5), he showed that if the geometric progression is decreasing, then ∞ i=1 ai is equal to (a1 )2 , a1 − a2

since an becomes less than any given quantity and the last term of the sequence is 0 (for proposition g) (see Bernoulli [P, 244–245]). Bernoulli then proved that the ultimate term had to be zero in order for the sum to be ﬁnite (see Bernoulli [P, 249]). He also showed that the harmonic series n1 has an inﬁnite sum. To prove this theorem he observed that 1 1 1 1 + + + ... + 2 k k+1 k+2 k

> =

1 1 1 1 + + + ... + 2 k k2 k2 k 1 1 + (k 2 − k) 2 = 1. k k

Therefore, given any number N , it is possible to group the terms of the harmonic series as follows: 1 1 + ... + 2 a 1 1 + ... + a+1 (a + 1)2 1 1 + ... + (a + 1)2 + 1 ((a + 1)2 + 1)2 1+

> 1, > 1, > 1, ...

and so on until one obtains N groups. This means that the partial sums of series become greater than any number N 113 (see Bernoulli [P, 250–251]). While these theorems from the Positiones arithmeticae show that Bernoulli paid much attention to convergence, others show that he did not hesitate to 113 Bernoulli observed: “The sum of an inﬁnite series whose ﬁnal term vanishes perhaps may be both inﬁnite and ﬁnite” [P, 252].

82

Convergence and Formal Manipulation

2 use formal manipulation. For instance, to sum ∞ n=1 n(n+2) , Bernoulli [P, 252–253] used the same method used by Leibniz (see p. 30)114 He set115 ∞ 1 =A n

n=1

and

∞

n=1

1 =B n+2

and derived ∞

n=1

∞

∞

n=1

n=1

1 1 1 3 2 = − =A−B =1− = . n(n + 2) n n+2 2 2

Bernoulli was quite aware that the condition “an is an inﬁnitesimal” for n = ∞ was necessary to sum a series an . In 1689 he even showed what would happen if this was not satisﬁed. He set S=

2a 3a 4a 5a + + + + ... c 2c 3c 4c

and

3a 4a 5a + + + ... 2c 3c 4c and observed that the application of the above method led to T =

Q=S−T =

2a , c

whereas the sum is Q=S−T =

a a a a + + + ... = . 2c 6c 12c c

Bernoulli explained that the method failed because Q was equal to the ﬁrst term of S (namely 2a c ) minus the last term of T , which is not 0, but equals a . c Bernoulli limited himself to advising the use of such series with caution in these particular cases (an exhortation that is also found in the young Euler116 ). Bernoulli did not require a control of convergence of series as a 114 Bernoulli explicitly referred to Leibniz and stated that, in his [1682], Leibniz had not explained how he derived his results; he therefore thought that it was appropriate to illustrate how certain sums could be determined (see Bernoulli [P, 252]). 115 1 Equality ∞ n=1 n = A is found in the Positiones arithmeticae immediately after the proof that the sum of the harmonic series is inﬁnite. 116 See Chapter 15.

5

83

Jacob Bernoulli’s Treatise on Series

preliminary condition for the application of this procedure,117 but simply veriﬁed the soundness of the results that unconditioned manipulation led to and, if necessary, provided an explanation of failures, which were considered as anomalies or exceptional cases. This approach prevented him from rejecting the use of divergent series in a clear and1 deﬁnitive way, even when √ it resulted in a contradiction concerning ∞ n=1 n . Jacob Bernoulli set ∞ 1 X= nm

and Y =

∞

n=1

n=1

1 . (2n − 1)m

By subtracting term by term, he obtained X −Y =

∞

n=1

∞ 1 1 1 1 = = m X. m m m (2n) 2 n 2 n=1

Therefore, 2m X − X = 2m Y and X 1 1 X Y : (Y − X) = X − m : m = 1 − m : m = (2m − 1) : 1. 2 2 2 2 Hence, he stated that the ratio between two series can be known even without knowing their sums. Bernoulli applied the last proportion to the series ∞ 1 √ n n=1

(whose sum is inﬁnite, he argued, since118 its term √1n is larger than the 1 term of ∞ n=1 n ) and noted that the ratio between the series of odd terms Y =

∞

n=1

√

1 2n − 1

and the series of even terms X −Y =

∞

1 √ 2n n=1

117 It was only Gauss who dealt with telescopic series in a way that can be considered satisfying (see Gauss [1812b, 143]). 118 In his treatise Bernoulli applied what we term the “comparison test”. However, he did not use it to establish whether a series was convergent, ∞this as a preliminary considering condition for using a given series. Given two series ∞ n such that an > n=1 an and n=1 b bn , Bernoulli did not deduce the existence of ∞ fact that ∞ n=1 an from the n=1 bn , is ﬁnite. ∞ Rather he deduced that A > B where A = ∞ n=1 an and B = n=1 bn (independently of the fact whether A and B are ﬁnite or inﬁnite).

84 is, unexpectedly,

Convergence and Formal Manipulation

√ Y : (Y − X) = ( 2 − 1) : 1

and, therefore, ∞

√

∞

√

n=1

while it is clear that

n=1

∞

1 1 √ , < 2n − 1 n=1 2n ∞

1 1 √ > 2n − 1 n=1 2n

(see Bernoulli [P, 261–262]). While in the ﬁrst part of his treatise Bernoulli was mainly interested in summing numerical series, in the second he dealt with the expansion of analytical expressions (representing curves, such as hyperbola, etc.), into power series and used these expansions to square and rectify curves. He ﬁrst dealt with the method of long division (see p. 20). He then explained that there were three other “artiﬁces” (artiﬁcia) for expanding analytically expressed quantities: Wallis’s interpolation, binomial expansion, and the method of indeterminate coeﬃcients. Moreover, in order to square curves, Bernoulli applied term-by-term integration of power series in an unproblematic way. In eﬀect, Bernoulli’s conception is very similar to Newton’s and Leibniz’s. Therefore, I shall not dwell upon it and I restrict k . myself to some observations concerning the expansion of the quantity m∓n k Bernoulli stated that the method of long division applied to m∓n always yields a remainder: Only if m > n does the remainder “decrease and is ﬁnally less than any given quantity” in which case one has: kn2 k kn k = ± 2 + 3 ± .... m∓n m m m

(53)

2

kn k kn k −m Bernoulli also proved that the sum of m 2 + m3 − . . . = m+n varies k k between m and 2m for 0 < n < m. He observed that if m = n, the series gives rise to the “non inelegans” paradox

k k k k k = − + − + .... 2m m m m m

(54)

He explained that in this case the remainder does not decrease but is always k equal to ± 2m : The paradox actually originated in the fact that the result of division is not k k k k − + − + ..., m m m m

5

85

Jacob Bernoulli’s Treatise on Series

but

k k k k k − + − + . . . (−1)n , m m m m 2m where n is the number of terms we added. Bernoulli, therefore, interpreted (53) as a relation that is legitimate under the condition |m| < n.119 To end this chapter I would like to point out an interesting result obtained by Bernoulli which is found in a letter written to Leibniz in 1703. It concerns the series solution to the equation dy = y 2 dx + x2 dx.

(55)

This equation is of interest because it is the ﬁrst appearance of an equation of the Riccati type. Johann Bernoulli had already posed the problem of the solution to (55) in 1694. Jacob Bernoulli succeeded in solving it by transforming it into the simpler equation d2 z = −x2 dx2 z dz (see Leibniz [GMS, 3:74]). By using the method of by letting y = − zdx 2 indeterminate coeﬃcients, it is easy to solve the equation dzz = −x2 dx2 . Indeed one obtains

z =1−

x8 x12 x4 + − + ... 3 · 4 3 · 4 · 7 · 8 3 · 4 · 7 · 8 · 11 · 12

dz By replacing this series into y = − zdx , Bernoulli found the solution

y=

1−

x7 x11 x3 3 − 3·4·7 + 3·4·7·8·11 + . . . x4 x8 x12 3·4 + 3·4·7·8 − 3·4·7·8·11·12 +

...

to diﬀerential equation (55). This solution can be transformed into the series y=

x7 2x11 13x15 x3 + + + + .... 3 3 · 3 · 7 3 · 3 · 3 · 7 · 11 3 · 3 · 3 · 3 · 5 · 7 · 7 · 11

(see Leibniz [GMS, 3:75]).

119 Bernoulli’s treatise ends this poem, which illustrated the diﬃculty of the notion of inﬁnite series: Even as the ﬁnite encloses an inﬁnite series And in the unlimited limits appear, So the soul of immensity dwells in minutia And in the narrowest limits no limit in here. What joy to discern the minute in inﬁnity! The vast to perceive in the small, what divinity! (Bernoulli [P, 306], translation by Walker in [Smith, 1:271])

6

The Taylor series

The Taylor series df 1 d2 f (x − x0 )2 (x − x ) + 0 dx x=x0 2! dx2 x=x0 1 d3 f 1 d4 f 3 + (x − x0 ) + (x − x0 )4 + . . . (56) 3! dx3 x=x0 4! dx4 x=x0

f (x) = f (x0 ) +

is one of the most important series in mathematics. It bears the name of Brook Taylor, who ﬁrst published it in Methodus incrementorum [1715]. However, similar results were probably known to Gregory (see Chapter 1) and certainly to Newton, Johann Bernoulli (see Chapter 3), Leibniz (see Chapter 3), and de Moivre (see Wollenshl¨ ager [1933, 241–257]). Newton’s formulation of the Taylor series is found in the 1692 version of De quadratura curvarum (see Newton [MP, 7:96–98]). It is contained in the corollaries 3 and 4 to Proposition120 12. In modern terms, Corollary 3 stated that if the series that results from the application of the method contained in Proposition 12 is a power series f (z) =

∞

An z n ,

n=1

then

1 dn f An = . n! dxn z=0

In Corollary 4 Newton gave the more general series f (x + w) =

∞ 1 dn f n w . n! dxn

n=1

(In both cases, Newton assumed that the ﬁrst term of the series was equal to 0.) The brevity of Newton’s treatment made it unclear as to the importance that he gave to this formula, which he removed from the later version of De quadratura curvarum and never published. Taylor’s derivation of the Taylor theorem was based on what he called the “method of increments.”121 This was substantially a theory of ﬁnite diﬀerences. Taylor thought that the method of increment provided “a perfect knowledge of the method of ﬂuxions” (Taylor [1715d, 339]) and applied it to several mathematical and mechanical problems. Among the results he derived was the following: 120

See Section 4.1, p. 62. On Taylor’s contributions to the Taylor series, see Feigenbaum [1985] and Panza [1992, 364–375]. 121

87

88

Convergence and Formal Manipulation Proposition 7. Let z and x be two variable quantities, of which z increases uniformly with given increments z . Let nz = v, \

\

·

\\

·

v − z = v, v − z = v , and so on. Then I say that in the time · · that z increases to z + v, x will likewise increase to

\

\ \\

v vv vv v x+x +x + etc. +x .1·z .. 1 · 2 · z 2 ∵ 1 · 2 · 3z 3 .

.

.

(Taylor [1715, 21], translation in Feigenbaum [1985, 40])

By using a more modern symbolism and the notion of a function, the previous theorem can formulated by stating that if y(z) is a function of z, Δz and Δy are the increments of z and y, and vk = kΔz, then

vn vn−1 vn−2 vn vn−1 vn + Δ3 y + ... + Δ2 y 2 1 · Δz 1 · 2 · (Δz) 1 · 2 · 3(Δz)3 (57) To prove (57), Taylor observed that y(z + nΔz) = y(z) + Δy

y(z) = y(z), y(z + Δz) = y(z) + Δy, y(z + 2Δz) = y(z) + 2Δy + Δ2 y, y(z + 3Δz) = y(z) + 3Δy + 3Δ2 y + Δ3 y, y(z + 4Δz) = y(z) + 4Δy + 6Δ2 y + 4Δ3 y + Δ4 z, ...

(58)

He stated that the coeﬃcients of Δz in the table (58) “are formed in the same way as the coeﬃcients of corresponding terms in the binomial expansion” (Taylor [1715, 22]). This made it possible to apply Newton’s binomial expansion. In doing this Taylor passed from the ﬁnite series (58) to the inﬁnite

6

89

The Taylor Series

series122 n(n − 1) 2 n n(n − 1)(n − 2) 3 Δ y+ Δ y + . . . (60) y(z + nΔz) = y(z) + Δy + 1 1·2 1·2

vk , he obtained (57). Through the substitution k = ∆z In the ﬁrst corollary of this proposition, Taylor stated that, in the hypothesis of Proposition 7, if z decreases to z − v, x will decrease to \

\ \\

v vv vv v −x + &c. +x .1·z .. 1 · 2 · z 2 ∵ 1 · 2 · z3

x−x

.

.

.

In Corollary 2, Taylor obtained the Taylor series merely by assuming Δz = . z(t)o, where o is an evanescent increment. To use his own terms: Corollary 2. If we substitute for the evanescent increments \\ \

the ﬂuxions proportional to them, and if the v vv are now made equal, then in the time that z, ﬂowing uniformly becomes z + v, x will become .

x+x

v2 v3 v .. ... + x . +x .2 . 3 + &c. 1·z 1·2·z 1·2·z

(61)

and likewise, with z decreasing to z − v, x will decrease to .

x−x

v3 v v2 ... .. − x . +x .2 . 3 + &c. 1·z 1·2·z 1·2·z

(Taylor [1715, 23]) Taylor gave importance to (61) as a method for solving diﬀerential equations. According to him, one ﬁrst had to seek a solution in closed form. If this was impossible, then one looked for a series solution. In this case, (61) 122 Formula (57) is nothing but interpolation formula (17). A similar result was given by Newton in Principia. It can be formulated as follows. The nth degree interpolation polynomial y(x), passing through any given n + 1 points with abscissas a, a + h,..., a + nh and ordinates y(a), y(a + h), . . . , y(a + nh), is

y(x)

= +

x(x − h)(x − 2h) 3 x(x − h) 2 x Δ y(a) + Δ y(a) + . . . (59) Δy(a) + h 1 · 2 · h2 1 · 2 · 3 · h3 x(x − h)(x − 2h) . . . (x − (n − 1)h) n Δ y(a) 1 · 2 · 3 · . . . · n · hn

y(a) +

In his [1985, 42–43], Feigenbaum notes that Newton’s formula (59) explicitly provides a polynomial, whereas (57) was meant to be an inﬁnite series. However, the early series theory was based on the systematic transition from the ﬁnite to the inﬁnite, as Taylor’s proof of the Taylor theorem conﬁrms. For this reason it is diﬃcult to consider (59) and (57) as being really diﬀerent in the context of Taylor’s proof.

90

Convergence and Formal Manipulation

oﬀered a general method of solution. As an example Taylor [1715, 24–26] considered the equation that, in modern symbols, can be written as xy ′′ + nyy ′′ − y ′ − (y ′ )2 = 0.

(62)

It is easy to derive y ′′ =

y ′ + (y ′ )2 x + ny

from (62). By repeated diﬀerentiation Taylor found y ′ y ′′ , x + ny y ′ y (3) , = (3 − 2n) x + ny y ′ y (4) = (4 − 3n) , x + ny ...

y ′′′ = (2 − n) y (4) y (5)

If c = y(a) and c′ = y ′ (a) are the initial conditions, the solution is given by the series y(a + x) = y(a) + y ′ (a)x + y ′′ (a)

x3 x2 + y ′′′ (a) + . . . 2 3!

(63a)

He also showed that, for particular values of a, the series (63a) was ﬁnite and then one obtained the preferred solution in closed form. Among the other results contained in the Methodus, the following should be pointed out: . The ﬂuent of rs can be expressed by either the series \.

.

\\..

\\\...

rs = rs − rs + r s − r s + &c.

(63)

or .\

.

.. \\

...\\\

rs = rs − r s + r s + &c. (Taylor [1715, 38]) These formulas have often been considered as identical to the Bernoulli series.123 For instance, (63) can be written as 123

sdr = rs−

rdr·

ds + dr

dr

rdr·

d2 s − dr2

dr

Their proofs are also very similar, see [1715, 38–39].

dr

rdr·

d3 s +. . . . (65) dr3

6

91

The Taylor Series

Such a translation of (63) from Newtonian ﬂuxional language into the Leibnizian one was the basis for Bernoulli’s charge of plagiarism toward Taylor, which was ﬁrst raised by Burchard, one of Bernoulli’s students, in his Epistola ad Virum Clarissimum Brook Taylor [1721]. It is interesting noting that the charge of plagiarism only concerned series (63), not series (61). This is due to the fact that both Bernoulli and Taylor probably considered the Bernoulli and Taylor series as results diﬀering from each other.124 In her [1985, 117–125] Feigenbaum notes that the interpretation that Taylor gave to (63) makes it more general than (65), so that the charge of plagiarism appears only partially justiﬁed. Based upon Taylor’s explanation and the three examples he provided, Feigenbaum shows that (63) should be expressed in modern notation as follows: dr sdr = s dw dw ds d2 s = rs − dw rdw rdw + dw dw2 d3 s − 3 dw dw rdw + . . . , dw where s, r, and w are functions of the same variable z. The presence of the . additional quantity w made it possible to expand rs into series in various ways. Moreover, an appropriate choice of w could make calculations easier, wich was an advantage of (63) with respect to the Bernoulli series. In 1717, immediately after the publication of the Taylor’s Methodus, Stirling gave a diﬀerent proof of the Taylor series (56), or, to be precise, of the Maclaurin series, since he only considered the case x0 = 0 (see Stirling [1717, 76–77]). In modern symbolism, Stirling’s proof can be formulated as follows.125 If one sets ∞ Ai xi f (x) = i=0

and repeatedly diﬀerentiates this equality term by term, then one obtains ∞

df = Ai ixi−1 , dx i=0 ∞ d2 f Ai i (i − 1) xi−2 , = dx2 i=0 ∞ d3 f Ai i (i − 1) (i − 2) xi−3 , = dx3 i=0 ... 124 125

On this question, see Panza [1992, 406–421]. This might have been Newton’s proof. However, Newton was never explicit.

92

Convergence and Formal Manipulation

By setting x = a, one has df , dx x=0 1 d2 f A2 = , 2! dx2 x=0 3 1 d f , A3 = 3! dx3

A1 =

...

x=0

Stirling illustrated this result with several examples, among which is the expansion for the cosine. According to Feigenbaum: “This would appear to be the ﬁrst published instance of the explicit use of the Taylor theorem to generate a power series expansion of a well-known function” (see Feigenbaum [1985, 80]).

7

Quantities and their representations

An adequate understanding of 17th- and 18th-century series theory cannot avoid a discussion of the basic notion of quantity. In the ﬁrst part of this chapter, I shall describe the evolution of the concept of quantity until the end of 18th century and will then examine the diﬀerent ways - ﬁgural and symbolic- of representing quantities. In the second part, I shall illustrate the concepts of numbers and continuity and highlight the diﬃculties of translating the traditional notion of quantity into modern terms.

7.1

Quantity and abstract quantity

The 17th- and 18th-century notion of quantity had classic roots. In Metaphysics (V, 1020a7), Aristotle characterized quantity as “that which is divisible into two or more constituent parts”. In Physics (VIII, 7, 260a7f) he stated that a thing changes in quantity if it is increased or decreased. Greeks distinguished quantities as being continuous or discrete. Discrete quantity (or multitude) was made up of discontinuous parts, meaning there was no common boundary at which they joined. These parts formed a plurality, an aggregate of units. Number and speech were given as examples of discrete quantities (see Aristotle’s Categories, 4b20–22). Aristotle stated: In the case of the parts of a number, there is no common boundary at which they join. For example: two ﬁves make ten, but the two ﬁves have no common boundary, but are separate; the parts three and seven also do not join at any boundary. Nor, to generalize, would it ever be possible in the case of number that there should be a common boundary among the parts; they are always separate. Number, therefore, is a discrete quantity. (Categories, 4b20) A continuous quantity (or magnitude) consists of parts whose position is established in reference to each other, so that the limit of one is the limit of the next. Ancient Greeks considered several types of continuous quantities, such as time, movement, and various geometrical quantities. In his [1996, 363], Grattan-Guiness listed 10 diﬀerent types of geometrical quantities in Euclid’s Elements: straight lines, planar curved lines (arc of circle), planar rectilinear regions (rectangle), planar curvilinear regions (segment of circle), spatial rectilinear surfaces (pyramid), spatial curvilinear surfaces (sphere), rectilinear solids (cube), curvilinear solids (hemisphere), planar angles, solid planar angles. All these diﬀerences between geometrical quantities were of importance since Greek mathematicians did not study quantity as an abstract entity with the capacity of decreasing and increasing; rather, they considered speciﬁc and determinate quantities in speciﬁc and determinate contexts. For the 93

94

Convergence and Formal Manipulation

Greeks, quantity always referred to an object: it was “quantity of”. Quantity presupposed the material, or, at least, an idealization of the material (geometrical quantities, such as lines, were idealization of real objects) Similar to geometrical quantities, numbers were also thought of as speciﬁc and deﬁnite objects.126 According to Aristotle, “to be present in number” is to be some number of a given object (Physics, 4 221b14f). “[T]he assertion ‘three trees’ presupposes the assertion ‘three’, but what the assertion ‘three’ intends has no existence outside of the trees of which they are said to be three. For the number of the trees, i.e., ‘three’, has no proper, no independent ‘nature’. . . ” (Klein [1968, 101]). Number in Greek mathematics always meant “a deﬁnite number of deﬁnite objects”. These objects could be objects of sense or pure units or monads. It is worth noting that there was a diﬀerence between numbers, which were pluralities, and one, which was not a number but the unity by which the numbers were counted. According to Euclid’s Elements, “a number is a multitude composed of units” (Book 7, Deﬁnition 2), while “a unit is that by virtue of which each of the things that exist is called one” (Book 7, Deﬁnition 1). Following Aristotle, Euclid considered unity as “unity of measurement”.127 In the 17th and 18th centuries mathematicians continued to use quantity as the basic notion of mathematics —mathematics was regarded as the science of quantity— although the notion they used was markedly diﬀerent from the Greek one. While Greek mathematicians had only dealt with speciﬁc and determinate quantities,128 in the period around 1600, after a long evolution, mathematicians had ﬁnally become capable of investigating mathematical objects in a more general form. Quantities were now considered as the abstract objects of our thought. These abstract entities were reiﬁed in appropriate symbolic representations (algebraic and geometric), which allowed their mathematical treatment. The degree of abstraction and the forms of reiﬁcation were subjected to further gradual changes from the 1590s to the 1740s and afterwards. The ﬁnal result of this evolution was the mathematical treatment of quantity as such (general quantity). To make this point clearer, I shall now investigate the nature of algebraic and geometric representations and describe the development of the relationship between algebraic and geometric representations during the 17th century. ∗ ∗

∗

The evolution of the concept of quantity underwent a turning point with 126

In the quotation on p. 93, Aristotle speaks of “parts” of numbers that have no common boundary. This has a sense only if a number is deemed to refer to concrete or ideal objects. 127 See Klein [1968, 108–112]. 128 It is worth noting that while Greek mathematicians had always dealt with speciﬁc and determinate quantities, Greek philosophers discussed the concept of quantity in an abstract sense.

7

Quantities and Their Representations

95

the rise of algebraic or analytical symbolism. I specify that the decisive aspect of analytical or algebraic symbolism was not the use in itself of certain signs but the fact that those signs were the objects of manipulation in their own right. For instance, I can write a⊥b to indicate that the straight line a is perpendicular to b. However, if in the proof of a theorem of elementary geometry, for instance “Given a point A and a straight line a, there exists one and only one perpendicular b to a lead for A”, I write ⊥ in place of “perpendicular to”, I do not really manipulate the symbol ⊥ by itself, but work with the concept of “perpendicular to”. The sign ⊥ is employed as a mere shorthand symbol, unless one establishes a calculus upon and operates according to the rule of this calculus. In symbolic expressions, such as (a + b)2 = (a + b)(a + b) = a2 + 2ab + b2 , the letters a and b, are used as the concrete objects of a calculation (see Panza [1992, 68-69]), namely they are manipulated according to certain rules, but the fact that they represented numbers or lines or other objects is of no importance. According to Leibniz, a calculation is cogitatio caeca, blind reasoning. Operating blindly was conceived of as moving pebbles mechanically in an abacus: What is of importance is that the concrete objects of manipulation (pebbles or graphical signs) are handled according to certain rules (syntactically, in modern terms), not their meaning. Greeks did not manipulate algebraic symbols in their mathematical reasonings; rather, they reasoned upon ﬁgures. A ﬁgure is a symbolic representation as well; however, it has a diﬀerent nature with respect to algebraic symbolism. It is iconic and imititative and reproduces the features of various real bodies by analogy. Figures can also be used in modern mathematics. However, there is a huge diﬀerence between the modern and the ancient use of ﬁgures. In modern mathematics, ﬁgures are dispensable tools for facilitating the comprehension: Their role is essentially pedagogical or illustrative. A modern theory is a conceptual system, composed of explicit axioms and rules of inference, deﬁnitions and theorems derived by means of a merely linguistic deduction. For instance, consider the proposition Two equal circles of radius r intersect each other if the separation of their centers is less than 2r. In modern geometry one can state this proposition if an appropriate axiom (or a theorem based upon appropriate axioms) guarantees their intersection. Modern verbal formulation of geometry implies that terms such as circle, radius, and center, only have the properties that derive from their deﬁnitions and the axioms of the theory.

96

Convergence and Formal Manipulation

Instead, Greek geometry used ﬁgures as parts of reasoning (and not as a merely pedagogical or illustrative tool). Thus, in order to derive the existence of the intersection between two circles, say C and C ′ , Greek geometers could instead refer to the evidence of Fig. 12 and simply say: “Look!” This is precisely what Euclid did in the proof of his very ﬁrst proposition, where he constructed an equilateral triangle. There was no necessity to clarify

Fig. 12

precisely all the relationships between the objects of a theory, to make all axioms explicit and to deﬁne all terms. The mere inspection of ﬁgures provided information that we would now consider missing.129 Profound changes took place in the 17th and 18th centuries compared with the Greek world. The main change was that, whereas the diagram used by the Greeks represented a speciﬁc ﬁgure and only “this” ﬁgure could be the object of reasoning, 17th century ﬁgures were the concrete representations of abstract entities. A circle in a ﬁgure was the representation of the abstract concept of the circle and this abstract object that was the object of study.130 This led a kind of dematerialization of the ﬁgures, which were conceived as the reiﬁcation of the abstract notions under investigation (see Klein [1968, 206–207]).131 Nevertheless, a ﬁgure still consisted of being the icon of a given geometrical object: this “imitative” aspect of ﬁgures continued to be used by 17th- and 18th-century mathematicians when they dealt with quantity from a geometric perspective.132 129 In his The Shaping of Deduction in Greek Mathematics [1999a], Netz made a detailed analysis of the form of the proof in Greek geometry and concluded that ﬁgures did not just supplement the reasoning, but that the proof was reasoned using speciﬁc geometric quantities formalized by a speciﬁc ﬁgure. According to Netz [1999a, 240–270], Greek geometry was also based on repeatability, rather than generality. 130 See p. 98. 131 It even occurred that an area could be represented by means of a line. 132 On the role of diagrams in 18th-century mathematics, see also Friedman [1992].

7

Quantities and Their Representations ∗ ∗

97

∗

Vi`ete’s symbolism was a decisive step toward the mathematical treatment of abstract quantity. The subject of Vi`ete’s analysis was geometrical objects, but he conceived them in abstract entities that could be manipulated in a symbolic way. These abstract entities were not yet general quantities (which are characterized only due to the fact that they could be decreased and increased) but broad subclasses of general quantities. In his In artem analyticem isagoge, Vi`ete distinguished between a logistice numerosa, which operated upon numbers, and a logistice speciosa, which operated “with species or forms of things, as, for example, with the letters of the alphabet” (Vi`ete [1591, 328]). According to him, a species was a certain type of geometric quantity (magnitude) considered in the abstract and logistice speciosa was a general algebra that acted both on continuous “geometric” magnitudes and numbers divisible into “discrete units” (Klein [1968, 123]). This enabled Vi`ete to write analytical expressions such as (A plane) + (A in B) . B

(66)

The expression A plane denoted a two-dimensional unknown, and Z in B was the product of two one-dimensional quantities133 (see Vi`ete [1591, 338]). Consequently, (66) stood for a2 + ab . b Although Vi`ete’s work marked a fundamental step in the process that led to a conceptualization of mathematics diﬀering from the Greek one,134 it still contained aspects that restricted its aspiration toward universality. Indeed, Vi`ete attached dimension to the species in any given equation and thought that “only homogeneous magnitudes are to be compared with one another” (Vi`ete [1591, 324–325]). Homogeneity is a form of determination that prevents the reduction of all quantities to only one type of quantity. For instance, the square ABCD remained, for Vi`ete, substantially diﬀerent from the product of the measurements of the sides AB and CD and not be identiﬁed with the number AB 2 = AB × CD. Such a reduction was later used by Descartes by introducing a new deﬁnition of multiplication between segments. Taking an arbitrary line segment as the unit segment u, Descartes deﬁned the product of two quantities a and b as the quantity c satisfying the proportion u : a = b : c. 133

In Vi`ete’s symbolic system, the unknown quantity was designated by the vowels A, E, I, O, U , or Y and the given quantities by the letters B, G, and D or other consonants (Vi`ete [1591, 340]). 134 For a hypothesis about the causes of this change in relation to Greek mathematics, see Klein [1968, 120–121]. A partially diﬀerent viewpoint is oﬀered by Netz [1999b, 43–45].

98

Convergence and Formal Manipulation

This allowed the powers to be interpreted appropriately: For example, x2 was the quantity deﬁned by 1 : x = x : x2 . In this way dimensional homogeneity could be circumvented and any quantity could be reduced to a line. It is clear that when any quantity is reduced to a line, the line itself assumed a symbolic character: in other words, it is the symbolic representation of the abstract notion of quantity. Descartes attempted to combine these symbolic ﬁgures and algebraic symbols. He thought that the symbols of algebra helped the understanding of geometrical ﬁgures and, vice versa, that geometrical ﬁgures helped the understanding of the symbols of algebra.135 However, when a ﬁgure was used in reasoning, its iconic aspect continued to be employed.136 Newton and Leibniz continued to develop Descartes’ abstract and symbolic conception, both on the mathematical level in the strict sense as well as on the epistemological level. However, they did not break the link between geometrical and algebraic representations. For instance, consider the following proposition:137 Given a curve of equation y = axm/n , its area is

m+n na n . m+n ax

In his [1711, 281–283], Newton138 justiﬁed this proposition on the basis of various assumptions linked to the simple inspection of Fig. 13. Indeed, it is the ﬁgure that ensures the existence of the area ADB, the regular behavior of the curve (what is referred to today as the continuity of a curve), and the existence of the rectangle BbKH whose area is equal to the trapezoid BbdD. Certainly, Newton’s demonstration139 gives considerable attention 135

See Descartes [1637, 17–18 and 20]. On Descartes, see Bos [1993] and [2001], and Klein [1968, 197–211]. 137 See Chapter 1, p. 68. 138 Newton’s proof runs as follows. He posed AB = x, area ABD = z, DB = y, Ab = x + o, area Abd = z + ov. He set 2 z = x3/2 3 (even if he considered a particular example, his reasoning is valid in general). He replaced x by x + o and z by z + ov into z 2 = 49 x3 and derived 136

ov 2 + 2zv =

4 2 4 2 4 o + x + xo. 9 3 3

He supposed Bb = o to be inﬁnitely small and obtained 2zv = 43 x2 . Since z = 32 x3/2 and v = y, he had y = x1/2 (Newton [1711, 281–283]). 139 In this proof the role of the ﬁgure is, on the one hand, fundamental because it guarantees that the reasoning has a foundation but, on the other hand, only intervenes indirectly in the algorithmic game. The importance of the ﬁgure is clearly greater when dealing with the calculus with the synthetic method. For this method, my argument is valid a fortiori.

7

Quantities and Their Representations

99

Fig. 13

to the relational aspect and shows a clear algorithmic structure. However, the ﬁgure cannot be eliminated, although it is only used to represent several properties of an entire class of curves, while the other characteristics are now entrusted to algebraic symbolism. Newton’s demonstration displays a mixture of ﬁgures and analytical symbolism. This mixture is typical of the early years of the calculus: It is also found in Leibniz and his school. Nevertheless, Leibniz’s epistemological position140 is worthy of mention because it heralds some of the important developments that deﬁnitely inﬂuenced mathematicians in the mid-18th century. Leibniz placed great insistence on the fact that the calculus could be understood as an algorithm that enabled operations that did not require lines or inspection of ﬁgures.141 But it should be emphasised not only that Leibniz always embedded the algorithm into a geometrical interpretation142 but also that his aim was considerably diﬀerent from the 18th-century conception of a self-founding and self-meaning calculus. Leibniz aimed to exclude the inspection of the ﬁgure from certain algorithmic procedures concerning 140 If one restricts oneself to mathematical practice, Newton seems, in certain aspects (I refer principally to Newton’s De Methodis Serierum et Fluxionum, about which see Panza [1989, 168–170]), to anticipate 18th-century conceptions even more than Leibniz. If looked at from the epistemological point of view, the situation changes, since Newton gave priority to the synthetic method and this had an eﬀect on the development of his school of thought. 141 See, for example, Leibniz [GMS, 4:479]. 142 For instance, Guicciardini stated: “Leibniz always embedded the algorithm into a geometrical interpretation” (Guicciardini, [1999, 167–168]). See also Bos [1974, 8].

100

Convergence and Formal Manipulation

geometric objects, but he did not doubt that the algorithm had a meaning and could only be justiﬁed to the extent that it referred to such geometric objects. Analysis was an ars inveniendi, a method for treating entities that geometry oﬀered (curves and geometric quantities related to them): It was not a theory that had its own objects, in the way that Eulerian analysis had functions. The nature of Leibniz’s analysis clearly emerged when, for example, he dealt with the problem of the series 1 − 1 + 1 − . . .. In a letter to Wolﬀ on July 13, 1712 (Leibniz [GBLW, 147]), he stated, among other things, that the relation 1−1+1−. . . = 12 was well founded since a geometric demonstration (demonstratio linearis) of it existed.143 In conclusion, analysis, around 1700, developed as an algorithmic and symbolic method based on the notion of abstract quantity; nevertheless (and this is what distinguishes it from Eulerian and post-Eulerian analysis), it remained a method for studying geometry and did not constitute a selffounding theory: The idea that one could invent a mathematical theory whose aim was the study of quantity in an abstract sense, independent of any ﬁgural evidence, did not exist. The objects of Leibniz’s and Newton’s analysis remained the objects of geometry (analytically expressed), and the ﬁgure144 continued to play one of the fundamental functions of the ﬁgure in Greek geometry: A part of the reasoning was unloaded on to it.

7.2

Continuous quantities, numbers and ﬁctitious quantities

The preceding discussion of symbols and ﬁgures makes it possible to provide some clariﬁcations concerning the terms “geometrical quantity”, “analytical expression of geometrical quantity” (or “analytically expressed quantity”), and “general quantity”, which I have so far used in a rather intuitive sense. I refer to “geometrical quantities” or “ﬁgural quantities” when quantities were, or could be, referred to as a concrete and perceptible representation in a diagram so that it could be investigated, at least partially, by means of the diagram itself. Geometrical quantities were lines or other geometrical objects, e.g., the arclength, subtangent, and tangent of a curve, the area between given lines, and so on. I specify that 17th- and 18th-century mathematicians did not deal with geometrical quantities as speciﬁc lines in a given diagram, but as abstract entities of our thought, which were reiﬁed in the diagram. I use the terms “analytical expressions of a geometrical quantity” or “analytically expressed quantities” to refer to analytical symbols associated with geometrical quantities (which therefore had or could have an explicit diagrammatic representation). Analytical expressions of a geometrical quan143

See p. 124. This is true even if the ﬁgure sometimes had clearly a symbolic form, as in Leibniz’s characteristic triangle. 144

7

Quantities and Their Representations

101

tities were investigated symbolically or analytically (by manipulating symbols, according to given rules) and (if necessary) using ﬁgures. I use the term “general quantity” to refer to an abstract entity145 that can be increased and diminished (in a continuous way146 ). The investigation of general quantities consisted of the investigation of the relations between a certain quantity and other quantities (see Panza [1996, 241]). These relationships were only examined by symbolic notations (analytical expressions of general quantity or functions), without referring to ﬁgures (to emphasize the contrast with geometrical quantities, I also refer to them as “analytical quantities” or “nonﬁgural quantities”). The notion of general quantity lay at the basis of analysis after 1740.147 ∗ ∗

∗

Indeterminacy was a crucial characteristic underlying the 17th- and 18thcentury notion of quantity. A quantity (whether geometric or analytical) was an intrinsically indeterminate entity. Of course, it can be determined in inﬁnitely diﬀerent ways, but, as far as a quantity was considered in terms of its capacity to be increased and reduced, it was thought of as being indeterminate. For example, a segment is a quantity insofar as it has the capacity to become larger or smaller, whereas a ﬁxed segment is only a determination of the quantity. The indeterminacy of a quantity gave rise to the notion of a variable quantity (or, more simply, a variable). A variable was merely a (continuous) quantity represented by the symbols x, y, z, . . .. In the ﬁrst works on the calculus, a variable was a geometrical quantity.148 After the 1740s, a variable was a general quantity.149 Another terminological speciﬁcation is necessary at this point. Indeed, 17th- and 18th-century mathematicians used the term “quantity” not only to denote a variable entity capable of increasing or diminishing but also to indicate speciﬁc determinations of this variable entity (the values of the quantity). To avoid confusion, I shall hereafter use the term “quantity” (or, also, “indeterminate quantity” or “variable quantity”) to refer to an entity in the sense of its capability to increase or diminish while I use the term “quantum” or “determinate quantity” to denote a speciﬁc determination of quantity. 145 A general quantity was generated from particular quantities by means of a process of abstraction, which consisted of rendering as a general quantity what is common to all continuous quantities, just as the “greenness” consists of the speciﬁc shared attributes of all green individual objects, such as trees and grass. The shared attribute of all individual quantities was the capability of being increased or decreased. 146 The general quantity was thought of as an abstraction drawn from continuous quantities and was assumed that it varied in a continuous way. 147 See Chapter 18. 148 For instance, in de l’Hˆ opital’s Analyse des inﬁniment petits, pour l’intelligence des lignes courbes, variables were considered as lines denoted by the letters y, x, [1696, 1–2]. 149 See Chapter 18.

102

Convergence and Formal Manipulation

I stress the importance of the distinction between quantity and quantum in the calculus: The calculus referred to indeterminate quantities, subject to possible variations, whether increases or decreases, rather than to speciﬁc determinations of quantities or determinate quantities. A quantity could assume diﬀerent values or determinations, although a quantity was not reduced to the enumeration of these values (see Ferraro [2000a, 108]). The notion of quantity as an intrinsically indeterminate (or variable) entity did not prevent quantities from being divided into constants and variables.150 However, this distinction did not depend on the nature of quantities but on speciﬁc questions, quantities being variable in themselves.151 Abstract general quantities, as well as concrete geometrical quantities, were regarded as continuous quantities. Mathematicians did not discuss the properties of continuous quantity explicitly: The continuous was substantially a primitive, undeﬁned notion, based on physical reality (founded upon nature, as Newton says). It was tacitly assumed that continuous quantities behaved as a segment of a straight line or a piece of a curved line. In his Les continu chez Leibniz,152 H. Breger has given a remarkable description of Leibniz’s geometrical continuous quantity, which adequately clariﬁes 17thand 18th-century mathematicians concept of the geometrical continuum. The main features of this commonly held concept are as follows. First, a segment was divisible into parts, each of which was similar in kind to the original quantity, but it could not be reduced to an aggregate of points. Thus, the continuum was given as a whole and was not regarded as a set of points, even though it was possible to determine speciﬁc points in it. Second, for the precise reason that a segment was not considered as a set of points it was impossible to distinguish between an open and a closed 150 For instance, de l’Hˆ opital stated: “Variables quantities are called those which increase or decrease continually whereas constant quantities are those that remain the same while the others change” [1696, 1]. Similar deﬁnitions lasted during the whole 18th century, and Lacroix still wrote in 1797: “Quantities, considered as changing in value or capable of changing it, are called to be variables, and the name constants is given to those quantities that always maintain their value during the calculation” [1797–1800, 1:82]. 151 For instance, Euler explained: “[T]his calculus deals with variable quantities, even though every quantity, by its very nature, can be increased or diminished in inﬁnitum; however, as long as the calculus is addressed towards a certain goal, some quantities are designed to maintain the same magnitude constantly while others are truly changed for each amount of increase and decrease: the former quantities are usually termed constants, the latter variables so that this diﬀerence is not expressed so much in terms of the nature of the thing as in the character of the question to which the calculus refers” [1755, 3]. As an example, Euler observed that the trajectory of a bullet was determined by four quantities: the amount of gunpowder, the angle of ﬁre, the range, and the time. Each of them was a quantity in the sense that it could be increased or reduced. This property was never lost, though in certain calculations it was utilized and not in others: In this sense, a quantity could be imagined as a variable or constant according to the speciﬁc calculation. 152 See Breger [1992a, 76–84].

7

Quantities and Their Representations

103

segment. It is always thought of as including its endpoints: “One cannot, e.g., consider the interval from 0 to 1 without the point zero. Imagine a metre long thread without the left extremity of the thread. It is clearly an absurdity. Precisely, in the same way, the point zero is not a part of continuum but its extremity on the left: the point cannot be suppressed, not even in thought” (Breger [1992b, 77]). Third, a curve or a relation between quantities was not deﬁned pointwise. An equation, such as y = x2 , was viewed as a relation that assigns an interval on the y-axis to an interval on the x-axis in an appropriate way (Breger [1992a, 77]). Curves were not plotted; they were generated by motion153 (see also Mahoney’s description of the notion of curve in Fermat in Mahoney [1994, 82]). Unlike a geometrical quantity, a general quantity was not represented by a line in a diagram. It was, however, made up of what all the geometrical quantities have in common. This implies that the basic notions of continuous geometrical quantities were immediately transferred to general quantities and the analytical continuum. For instance, 1. the continuum was given as a whole and was not regarded as a set of elements (of any nature); however, it was possible to determine speciﬁc determinations of the continuum; 2. a variable quantity x always varied continuously and when it moved from a value x1 to a value x2 , it was impossible to think of this variation without the initial and ﬁnal values (or if one wishes, it was impossible to distinguish between an open and a closed interval); 3. a relation between general quantities was not deﬁned pointwise. Thus, throughout the 17th and 18th centuries the continuum used by mathematicians —even when they referred to general quantities— was substantially a geometrical continuum that reﬂected the properties of a segment. However, during this period, the substratum of the continuum changed. In other words, during the 17th and 18th centuries, the nature of the objects investigated by analysis underwent a signiﬁcant change (from geometrical 153 In Tractatus de quadratura curvarum, Newton expressed himself using the following terms: “I don’t here consider Mathematical Quantities as composed of Parts extremely small, but as generated by a continual motion. Lines are described, and by describing are generated, not by any apposition of Parts, but by a continual motion of Points. Surfaces are generated by the motion of Lines, Solids by the motion of Surfaces, Angles by the Rotation of their Legs, Time by a continual ﬂux, and so in the rest. These Geneses are founded upon Nature, and are every Day seen in the motion of Bodies” (see Newton [1704, 332] translation by Harris in Newton [QHW, 3]).

104

Convergence and Formal Manipulation

quantities into abstract quantities), whereas the concept of continuity remained unaltered.154 I will return to the concept of continuity in Chapter 18.3. ∗ ∗

∗

Seventeenth- and 18th-century mathematicians went beyond the Greek concept of number as a multiplicity of unities (number of unities). This evolution, which developed gradually during the Middle Ages and Renaissance, underwent its turning point in the works of Stevin and Vi`ete155 . After 1600, the notion of number had two main characteristics. 1. It was abstract. Numbers were abstracted from counted things and became autonomous entities. A number, such as 7, was no longer considered as the attribute of a group of material or ideal objects; rather it was an abstract entity that expressed what all the things that are seven times the unity had in common.156 2. It was symbolic. The abstract concept of number was reiﬁed into symbols upon which one could operate according to given rules (the rules of arithmetic). Thus, the number 7 was also a symbol that reiﬁed an ideal entity into ciphers upon which one manipulates directly. In this sense, a number had a full signiﬁcance only if it was considered an element of a set of signs governed by appropriate rules that allowed their manipulation (see Klein [1968, 193]). However, unlike modern mathematics, the rules governing the manipulation of this set of signs were not arbitrary (nor even potentially so).157 This abstract and symbolic concept of numbers made it possible to introduce new species of numbers in addition to natural numbers. Irrational numbers, negative numbers, and imaginary numbers entered the realm of numbers, although (and this is a very important point) they were epistemologically diﬀerent from natural numbers. To make this point clearer, I begin by observing that numbers were understood as the measurement of a given quantity with respect to another 154

See also Chapter 32. On Renaissance notions of number and magnitudo, see Malet [1996]. On the notion of number from Stevin to Wallis, I refer to Klein [1968, 186–224]. On Euler, see Ferraro [2004, 39–43]. 156 Using Menninger’s words, one can state that the relationship between numbers and counted objects is upset: It is not the operation of counting that originates numbers but numbers existed abstractly and the operation of counting assigned numbers to counted objects; “so long as there is no counting, it [the abstract number sequence] is merely there, detached from all concrete objects, unused but ready. But as soon as we count . . . the number words become assigned to the objects, and . . . the objects are placed in the empty boxes of the number sequences” (Menninger [1969, 7]). 157 See Section 18.1. 155

7

Quantities and Their Representations

105

taken with as a measure or unity. For instance, in Universal Arithmetick, Newton stated: By number we understand not so much a multitude of quantities, as the abstracted ratio of any quantity, to another quantity of the same kind, which we take for unity. (Newton [1720, 2])158 According to Newton: An integer is what is measured by unity, a fraction, that which a submultiple part of unity measures, and a surd, to which unity is incommensurable. (Newton [1707, 2]) Several decades later, Euler similarly stated that all the determinations or measures of any quantity are reduced to determining the relation that a given quantity has with a certain quantity of the same kind taken as a measure or unity: [This relation] is always indicated by numbers, so that a number is nothing but the relation of a quantity to another quantity, taken arbitrarily as a unity. (Euler [1770, 10]) Quantity was therefore considered as an entity that logically precedes number and number was viewed as a tool for treating quantity: According to Stevin, “Number is that by which the quantity of each thing is revealed” (Stevin [1585, 1]). The concept of number as a measurement made it possible to embody the discrete within the continuous. Thus, even though the traditional distinction between discrete and continuous quantity was maintained, it lost its former importance. In eﬀect, the discrete was thought of as originating from the continuous and was regarded as a particularization of the continuous. The sequence 1, 2, 3, . . . only identiﬁes discrete determinations of a continuous quantity, depending on the choice of the unity. A precise deﬁnition of measurement was not given. Mathematicians assumed implicitly that measuring meant repeating the operation of comparing with unity or one of its parts successively and ﬁnitely. This concept of measurement, which assumes that the sequence of natural numbers159 is intuitively known, made it possible to give an adequate treatment of fractions as numbers. For instance, one could have a clear idea of 7/3 by considering a segment 7 feet in length and by dividing it into 3 parts (see Euler [1770, 30]). 158

See also Stevin [1585, 1] and Wallis [1657, 183]. After Stevin, the idea that 1 was a number gained widespread acceptance. Klein views this development as one of the signs of the transition towards a more symbolic concept of numbers (see Klein [1968, 186–224]). 159

106

Convergence and Formal Manipulation

Diﬃculties arose with irrational numbers. During the 17th and 18th centuries, irrational numbers were accepted. Vi`ete referred to them as “numeri asymmetri” (asymmetrical numbers); Stevin [1585, 30] stated “Every root is a number”160 and, as we saw earlier, Newton considered numbers as having a threefold nature: integer, fractional, and surd. Nevertheless, irrational numbers were signiﬁcantly diﬀerent from natural and fractional numbers. Fractions had a meaning in terms of unity of measure and consequently were numbers in the strict sense of the term, or “true numbers”. In contrast, irrational numbers were not true numbers since they did not represent a process of measurement in a precise sense. For instance, in his Vollst¨ andige Anleitung zur Algebra, Euler observed that the root of 12 is not a fraction; nevertheless, it is a determinate quantity, which is greater than 3,

24 38 45 , , , ... 7 11 13

and smaller than 4,

7 52 , , ... : 2 15

√ 12 is a new species of number. He Therefore, √ then added that a correct √ idea of 12 can be gathered by observing that √ 12 is the number that, when multiplied by itself, makes 12 √ (this deﬁnes 12 operatively, as a symbolic entity) and that the value of 12 can be approximated as desired (Euler [1770, 50–51]). The domain of true numbers (natural and rational numbers) was not sufﬁcient to investigate quantity, nor was the addition of irrational numbers.161 Other numbers apart from rational and irrational ones were necessary: negative numbers, zero, and imaginary numbers.162 Negative numbers163 were usually introduced simply by stating that they were entities less than the nothing and that were represented by numbers with the sign − (in opposition, positive numbers were numbers greater than nothing and had the sign +).164 This deﬁnition were accompanied with rules that made it possible to operate upon them as symbolic entities: It was these rules that transformed the symbols −1, −2, . . . into numbers. As the other 160 Stevin even refused to call them absurd, surd, or irrational numbers, because incommensurability was not a cause of absurdity (see Stevin [1595]). 161 In 18th century, it is possible to grasp a distinction between irrational and transcendental numbers, where transcendental numbers are those numbers that derived from the application of transcendent operations (such as a logarithm) to a rational number. However, this distinction is not of importance for my purposes and I shall not take it into account. 162 In this book I do not consider inﬁnitesimal numbers. The following remarks also hold for them (see Ferraro [2004]). 163 Their use came up against resistance. Vi`ete did not yet acknowledge negative numbers. 164 See, e.g., Newton [1720, 3].

7

Quantities and Their Representations

107

above-mentioned species of numbers, they also had an intuitive meaning. For instance, Newton [1720, 3] gave the traditional interpretation of negatives as debits. He also stressed that a negative motion was a “regression”, i.e., proceeding backwards. They could also be represented by directed segments. However, they did not correspond to a notion of measurement of a quantity in the strict sense of the term: They were not true numbers. Zero was considered as the absence of quantity and was the name given to “nothing”, from which numbers were subtracted to obtain negative numbers. In his Arithm´etique [1585, 4] Stevin considered the zero to be the beginning of natural numbers (but not a number), in the same sense as the point is the beginning of the line (not an element of the line, see p. 102). Almost two centuries later, in Vollst¨ andige Anleitung zur Algebra Euler did not yet list zero as an integer (integers were the natural numbers +1, +2, +3, . . ., which are greater than nothing, and negative numbers were −1, −2, −3, . . ., which are less than nothing).165 Complex numbers had been introduced century and gradually √16th√ √ in the gained acceptance. Expressions such as −1, −2, −3, . . . were termed impossible or imaginary numbers: Nevertheless, mathematicians thought that they were useful for dealing with quantity and could have a role in our reasonings. Imaginary numbers were introduced as formal instruments for obtaining the root √ of certain numbers, in a similar way to irrational numbers. For instance, −4 merely meant a number that multiplied by itself equals166 −4. Imaginary numbers could not be reduced to the measurement of quantity, not even in an approximate sense. They diﬀered from other numbers and were generated by the symbolic mechanism of analysis; they had no intuitive meaning on their own but assumed a meaning within the overall context of analysis. At this point it is clear that even though all numbers were abstract and symbolic entities, only some adequately reﬂected the concept of number as the exact result of a process of measurement and were “true” numbers. Other types of numbers did not ﬁt the notion of a number (although for diﬀerent reasons). In the strict sense of the term they were not true numbers. I shall term them “ﬁctitious numbers” or “ﬁctions”. The idea of false or ﬁctitious numbers is an old one. For instance, many mathematicians referred to negative numbers as false numbers.167 In Sec165

See Euler [1770, 14]. See, e.g., Euler [1770, 56]. 167 Descartes wrote the following when commenting upon the roots of an equation: “But often it happens, that some of these roots are false, or less than anything, as if one supposes that x indicates also the defect of a quantity, which is 5, one has x + 5 = 0, which being multiplied by x3 − 9xx + 26x − 24 = 0 is x4 − 4x3 − 19xx + 106x − 120 = 0, for an equation in which there are four roots, namely three true ones which are 2, 3, 4, and a false one which is 5” (Descartes [1637, 56]). But, in 1545, Cardano had already stated: “If the square of a square is equal to a number and a square, there is always one true solution and another and ﬁctitious solution equal to it.” He gave the example x4 = 2x2 + 8, where 166

108

Convergence and Formal Manipulation

tion 2.1 we saw that Leibniz attempted to justify inﬁnitesimal and inﬁnite numbers as ﬁctions similar to other ﬁctions used in mathematics (imaginary numbers, the power whose exponents are not true numbers, etc.).168 In eﬀect, the idea of false numbers is at the basis of much of mathematical terminology regarding numbers, which we still partially retain today. Unlike Leibniz, most mathematicians did not employ the term “ﬁction” explicitly. Nevertheless, I use this expression because it expresses the nature of the 17th- and 18th-century approach, in particular because it implies an ontological diﬀerence between that which is ﬁctitious and that which is true. Seventeenth- and 18th-century mathematics eﬀectively presents an ontological diﬀerence between natural and rational numbers (true numbers) and the other species of numbers (which did not correspond to the idea of numbers and √ therefore were ﬁctitious numbers). To put it more clearly, nowadays −1 is an element of the set of complex numbers C and exists in the same 1 way as any other √ number in C, such as 1, 2, 2 , etc. During the 17th and 18th centuries, −1 was a useful symbol for studying certain aspects of quantity; it did not have an existence in the same sense as true numbers. Similarly, 0 was the symbol that represented the absence of quantity, the nothing, the non-existence; it was not a number, because it did not measure quantity and did not denote anything; however, 0 could be treated as a number. Mutatis mutandis, the same holds for irrational (unspeakable, inexpressible) numbers and negative numbers. I already pointed some aspects of ﬁctitious quantities with reference to Leibniz at the end of Section 2.1. Now, in a more systematic and general way, I observe that ﬁctions had the following ﬁve characteristics. (a) Fictions were a useful tool for shortening the path of thought and arriving at new results.169 It was of no importance whether ﬁctions appeared in nature or not, namely if they represented physical or geometrical objects. Irrational numbers appeared in nature (they represented the length of a segment); imaginary numbers did not appear. (b) Fictions, however, were always connected with reality, directly or indirectly. They were not arbitrary creations of the human mind but had to be well founded in reality and were needed for investigating reality (this is true even for imaginary numbers170 ). By the phrase “well founded in reality or in nature”,171 I intend to highlight the fact that certain mathematical objects did not originate from x “equals 2 or −2” (Cardano [1545, 11]). 168 For instance, see the quotation on p. 35. 169 See Leibniz’s quotation on p. 35. 170 See Leibniz [GMS, 4:92–93] and Euler [1770, 57]. 171 I derive this expression from Leibniz’s statement that imaginary quantities have their foundation in nature (see Section 2.1, p. 35).

7

Quantities and Their Representations

109

arbitrary deﬁnitions given in a theory based upon an arbitrary system of axioms; instead, they originated 1. from the need to express certain properties of quantities and 2. from the need to manipulate objects that directly expressed quantities or properties of quantities (an example would be the casus irreducibilus of the equation of third degree). In the ﬁrst case, a well-founded object had an intuitively obvious interpretation (e.g., irrational numbers). For this reason I would say that it was directly connected to reality.172 The second case was that of imaginary numbers, which did not have an intuitively obvious meaning173 . They were introduced in a merely formal way, but they made up for rational and irrational numbers when these did not suﬃce: They were always connected to reality, even though only indirectly. In any case, well-foundedness, used in this sense, excludes the possibility that mathematical objects could originate from a free act of will and required them to be rooted in reality, directly or indirectly, as elements of a theory that aimed to interpret the real. It should also be emphasized that ﬁctions were not of interest in themselves, but only insofar as they allowed one to solve problems concerning quantities. They were auxiliary instruments for dealing with quantities. (c) Fictions were manipulated as if they were true numbers. This means that a ﬁction was treated by analogical extensions of rules valid for true numbers or geometrical quantities.174 Therefore, a ﬁctitious number was more than a mere fa¸con de parler or a shorthand way of denoting a certain operation upon true numbers: It was a symbolic entity that formed part of the symbolic nature of true numbers and quantities. (d) An adequate theoretical construction for moving from ﬁctions as a sign for shortening the path of thought to the analogical use of ﬁctions as true numbers was completely lacking. Thus, well-foundedness in the above sense was the only justiﬁcation for ﬁctions. 172

However, this does not mean that there exists a geometrical or physical object corresponding to it. 173 See also footnote no. 263. 174 This statement is to be understood as follows. The principle was assumed that if an operation (not only an algebraic operation —sum, product, etc.— but also transcendental operations —logarithm, etc.) had true numbers or geometrical quantities as operands then it could have ﬁctitious numbers as operands. Of course, some adjustments might be necessary: They took the form of speciﬁc rules inherent to the peculiar nature of every distinct species of ﬁctitious numbers. For instance, the rule of signs was a speciﬁc rule for negative numbers. These speciﬁc rules were what distinguished a calculation involving a particular species of ﬁctitious numbers from a calculation involving true numbers or a diﬀerent species of ﬁctitious numbers.

110

Convergence and Formal Manipulation

(e) Even though quantity was an entity abstracted from geometrical quantity and had the same properties as lines, it could be determined by ﬁctitious values; in other words, one could assign ﬁctitious numbers to a variable x. Seventeenth- and 18th-century mathematicians used the term “irrational quantities” to refer to irrational numbers or irrational determinations of quantity. They referred similarly to negative quantities, imaginary quantities, etc. I maintain this terminology and, more generally, I use the term “ﬁctitious quantities” by referring to ﬁctitious determinations of quantity or ﬁctitious numbers. A general quantity has some determinations that can be represented by a nondirected segment, whereas others cannot. I use the term “real quantity” to denote a quantity that only assumes these determinations and that corresponds to the mental image of the geometrical or physical quantity. I do not therefore intend this term in opposition to ﬁctitious quantity, since a real quantity can have both true numbers and certain ﬁctitious numbers as its determinations.175 Finally, I wish to make some simple consequences of the above-described notions of quantity and numbers explicit. First, since a single number was a speciﬁc determination of quantity, a single number expressed a quantum rather than a quantity. Second, even though each speciﬁc determination of real quantity can be represented by means of numbers, the idea that quantity might be reduced to a set of numbers was not taken into account. Third, more generally, numbers were not conceived of as elements of a set, if by “set” one means an extensional entity, which is arbitrarily deﬁned, entirely characterized by the list of its elements and having a certain cardinality. Instead, 17th- and 18th-century mathematicians classiﬁed numbers into diﬀerent classes or species, where a “species” of numbers was intended as an intensional entity, which could not be reduced to an enumeration of objects: it was given by a nonarbitrary and nontrivial property and was not necessarily associated with cardinality. 175

Irrational numbers do not correspond to the idea of number and, therefore, are ﬁctions; however, they have a very special nature with respect to other ﬁctitious numbers since they can be represented by means of a nondirected segment and answer the question: What is the measure of a given (real) geometrical quantity? Rational numbers answer the same question (though in a more precise way), and thus these rational and irrational numbers might be grouped together to form the class of (positive) real numbers (and in eﬀect rational and irrational numbers were often taken together; for instance, as opposed to imaginary numbers, see (Euler [1748a, 18]). By so doing one obtains a diﬀerent classiﬁcation, which considers the capacity of numbers to express the determinations of geometrical quantities directly, but this is not relevant to my purposes in this book.

7

111

Quantities and Their Representations ∗ ∗

∗

The preceding sections should make it clear that the transformation of 17th- and 18th-century mathematics into modern terms may cause a certain degree of stretching in the meaning of certain results and procedures. I would like to emphasize this eﬀect of stretching by means of an example that highlights the diﬀerence between the 17th-century notion of limit and the modern one; it also aids in our understanding of Newton’s concept of limit, which has been discussed in Chapter 4. In Analyse des Inﬁniment petits, de l’Hˆ opital considered the following problem: Let x = AP and y = P M be the abscissa and the ordinate of a given curve AM D. Suppose that a = AB is a particular value of the abscissa and that the value of the ordinate y = P M is expressed by a fraction, whose denominator and numerator both become equal to 0 when the abscissa x becomes equal to a. It is required to ﬁnd the value of the ordinate y when x = a. (de l’Hˆ opital [1696, 145]) By using the notion of a function, the problem can be formulated as follows: Find the value of the function y = f (x) =

h(x) g(x)

for x = a, when h(a) = 0 and g(a) = 0. The problem is solved by means the rule today named after de l’Hˆ opital dh(x) f (a) = . dg(x) x=a For instance, to ﬁnd the value of y=

a2 − x2 √ a − ax

for x = a, de l’Hˆ opital set

a2 − ax √ a − ax

x=a

=

a

a 1 2

x

= 2a x=a

112

Convergence and Formal Manipulation

(see de l’Hˆopital [1696, 146]). One may be struck by the similarity between this procedure for “ﬁnding” the value of and the modern problem of extending the function a2 − ax √ F (x) = a − ax in a continuous way by setting a2 − ax √ . x→a a − ax

F (a) = lim

Indeed it is completely natural to translate as

a2 − ax √ a − ax

=

x=a

a

a 1 2

x

x=a

a2 − ax a √ = lim 1 a ; x→a a − ax x−a 2 x lim

however, such a translation into the language of limits may produce misina2 −ax √ terpretations. From a modern perspective, ﬁnding the value of a− when ax x = a (I assume a > 0) means that a. one considers the function (function in the modern sense of the term, not as an analytical expression) f (x) =

a2 − ax √ a − ax

deﬁned for x > 0 and x = a. b. the domain of f (x) has a point of accumulation at a so that we can attempt to calculate the limit as x → a; c. the application of l’Hˆ opital’s rule, under whose hypotheses our case falls, makes it possible to state that such a limit exists and is equal to 2a; d. ﬁnally, we deﬁne a new function f (x), which will be continuous at the point a, by setting f (x) =

a2 −ax √ a− ax

2a

for x ≥ 0 and x = a,

for x = a.

7

Quantities and Their Representations

113

In this procedure we use notions such as the limit, value, and extension of a function, whose meaning are opportunely and explicitly deﬁned. Indeed, λ = lim F (x), x→x0

where F (x) is a function with domain D in R, means Given any ε > 0 there exists a δ > 0 such that if x belongs to D and |x − x0 | < δ then |F (x) − λ| < ε. By λ = F (x0 ) we mean: λ is the number that the function F associates with the number x0 . If limx→x0 F (x) exists and is equal to λ, while the function F (x) is not deﬁned at the point x0 or F (x0 ) = λ, we can remove the discontinuity at x0 by deﬁning the new function f (x) x ∈ D − {x0 }, x→ λ x = x0 . These deﬁnitions presuppose knowledge of the notions of set, real numbers, function in the modern sense, continuity, etc. For this reason the above procedure is substantially meaningless for 17th and 18th century mathematicians. They did not consider a function as a pointwise correspondence between numerical sets but as a rule that linked two variables quantities and was embodied in one single analytical expression. They had no set of points or numbers, did not separate an interval of values (a segment) from its endpoints, etc., nor could they formulate the notion of extension of a 2 2 √ = 2a necessarily to be the value of function, but instead considered a −a 2 a2 −ax √ a− ax

a− a

when the variable x equals a.

8

The formal-quantitative theory of series

In this chapter, on the basis of the investigations carried out in the previous chapters, I would like to provide further clariﬁcation about the early theory of series. First of all, I emphasize that all the mathematicians examined hitherto considered it obvious that the equality Q=

qk

(67)

meant that q k and Q denoted the same quantity. It also seemed obvious that the series qk denoted the same quantity as Q if and only if the series qk was convergent to the quantity denoted by Q, the sum of theseries. For the sake of brevity, I shall refer to an equality of the kind Q = qk as a quantitative equality if the series qk converges to the quantity denoted by Q. (More generally, I shall refer to an equality of the kind Q = P as a quantitative equality if Q and P represent the same quantity.) The above investigations lead us to conclude that convergence was understood in the following sense: (C) A series ∞ k=0 ak converges to a quantity S, the sum of the series, if and only if the sequence of nnth sums Sn =

n

ak

k=0

approaches S indeﬁnitely when n increases so that Sn is ultimately equal to S, when n is inﬁnite. The following property was a trivial consequence of the fact that the sequence Sn approaches S indeﬁnitely: (AP) If the sequence of the nth sums Sn = nk=0 ak approaches S indeﬁnitely when n increases, then the diﬀerence between Sn and S (in absolute value) becomes less than any given quantity. I would like to point out that the sum of a series was the ultimate value of the series and that (AP) was only an obvious property of the notion of approaching a quantity. In Proposition (C) the symbols S and ak can denote both determinate quantities as well as variable quantities; in modern terms, nk=0 ak can be both numerical and function series. Referring explicitly to function series, (C) can be reformulated as follows: 115

116

Convergence and Formal Manipulation ∞ (CF S ) The series k=0 fk (x) is said to be convergent to the function f (x) on an interval176 Ix , over which x varies, if and only if the sequence of the nth sum nk=0 fk (α) approached f (α) indeﬁnitely when n increases, for any value α of x belonging to I, and it is ﬁnally equal to f (α), when n is a inﬁnite number.

Some words of caution are necessary regarding (CF S ). First, as should be clear from the previous function series chapters, k or, at most, with a x around 1700 were constituted by power series ∞ k k=0 series of the type ∞ ∞ ak xαk ak x−k or k=0

k=0

where a ﬁnite number of the exponents αk were rational numbers. The early theory of function series was substantially a theory of power series. Only in the second part of the 18th century did other types of function series appear (but power series were always largely dominant). However, the use of the f expression “function series” and the symbol ∞ k=0 k (x) in deﬁnition (CF S ) is useful in order to avoid repeating a similar deﬁnition of convergence with reference to other types of function series, when they come to be examined. Second, the word “function” is an anachronism before the 1720s (or, perhaps, even before the 1740s). However, the use of this term is appropriate and simpler than other expressions, provided it is borne in mind that, when referring to the period prior to 1720, the word “function” should be understood either as an analytical expression — composed of algebraic, exponential and logarithmic operations— or directly as geometric quantities — this is the case with trigonometric quantities, which were not considered as analytical expressions, but whose expansions into series were known. (The meaning of the term “function” during the 18th century is illustrated in Section 18.2. This meaning largely justiﬁes the anachronism.) We saw that convergence was one of the two cornerstones of the early theory of series, the other one being formal manipulation. To stress the coexistence of these two factors, I shall later refer to the early theory of series as the “formal-quantitative theory of series”. Some clariﬁcation is, however, appropriate with regard to the term “formal”. A ﬁrst good deﬁnition of “formal” is the following: A procedure or rule is formal if it is applied to (ﬁnite or inﬁnite) analytical expressions A(x, y, . . .), B(x, y, . . .), . . . regardless of the actual meaning of such expressions. In other words, when formal procedures or rules are employed, the quantitative meaning of certain expressions is neglected. 176

The interval was not usually speciﬁed.

8

The Formal-Quantitative Theory of Series

117

This meaning of the term “formal” can be made more precise when referred to 17th- and 18th-century mathematics. Indeed,

in the 17th and 18th centuries, procedures or rules were formal if they were based upon the principle of the inﬁnite extension from the ﬁnite to the inﬁnite or the principle of the generality of algebra.

I refer readers to Section 18.2 for an analysis of the generality of algebra. I shall now dwell upon the principle of the inﬁnite extension whose importance in the early series theory is immediately evident from the previous examination of the works of Leibniz, Newton, and other mathematicians. It consisted of the following assumption:

(IE) If a rule R was valid for ﬁnite expressions or if a procedure P depended on a ﬁnite number n of steps S1 , S2 , S3 , . . ., Sn , then it was legitimate to apply the rule R and the procedure P to inﬁnite expressions and in an unending number of steps S1 , S2 , S3 , . . .

Principle (IE) stemmed from the lack of any distinction between inﬁnite and ﬁnite series. For instance, since the following rules hold k

(an + bn ) = s1 + s2 ,

n=1 k

λan = λs1 , n=1 k k bn = s1 s2 , an n=1

k

n=2 k

n=1

n=1

an = s1 − a1 ,

an =

k

n=1

aj(n) = s1 ,

when s1 and s2 are the sum of the ﬁnite series kn=1 an and kn=1 an , λ is a number and aj(n) is any rearrangement of the ﬁnite sequence an , it seemed natural to extend these rules to inﬁnite series as well.

118

Convergence and Formal Manipulation Consequently, it was assumed that ∞

(an + bn ) = s1 + s2 ,

n=1 ∞

λan = λs1 , n=1 ∞ ∞ bn = s1 s2 , an n=1 ∞

n=2 ∞

n=1

(68)

n=1

an = s1 − a1 ,

an =

∞

n=1

aj(n) = s1 ,

∞ when s1 and s2 are the sum of the inﬁnite series ∞ n=1 an and n=1 an , λ is a number, and aj(n) is any rearrangement of the inﬁnite sequence an . It should be noted that even the symbolism masked the diﬀerence between ﬁnite and inﬁnite series. In the 17th and 18th centuries, a series was generally denoted by a written expression like a + b + c + etc. : this could mean both a + b + c + . . . ad inﬁnitum and a + b + c + . . . + p + q + r untill the term r. I would also like to emphasize that formal procedures were not freely invented or created: In series theory, mathematicians only considered rules deriving from the inﬁnite extension of the properties that were valid for ﬁnite expressions and a ﬁnite numbers of steps. The analysis set out in the previous chapters makes it possible to group the standard procedure for expanding a quantity into series into the following types:177 ( P1 ) The Mercator’s expansions of fractions and square roots of polynomials. ( P2 ) The binomial expansion for any (rational or irrational) exponent. ( P3 ) Any expansion following the method of indeterminate coeﬃcients. ( P4 ) Any expansion deriving from contemporary diﬀerentiation or integration of both the sides of a given equality f (x) =

∞

fi (x)

i=0

—the operations on the series being performed term by term.178 177

For more details on this classiﬁcation, see Ferraro and Panza [2003]. The procedures (P4 ) depend on an inﬁnite extension of the properties of linearity of diﬀerentiation and integration (today, we know that they do not follow from simple convergence). The Taylor series is an example. 178

8

The Formal-Quantitative Theory of Series

119

( P5 ) Any composition of the procedures ( P1 ), ( P2 ), ( P3 ), ( P4 ).179 The above investigations also make it possible to give an explicit explanation of how the formal and the quantitative were related to one other in the early theory of series. Indeed, the relationship between the formal and the quantitative was based upon the following three principles. First, it was assumed that the usual procedures transformed a function f (x) into a power series convergent to the function f (x) at least over an interval I of values of variable x. This was an unproved assumption: It was merely derived from noting that it occurred in all known cases. Second, the actual determination of the interval of convergence was an a posteriori question intervened only when one wished to apply which i of a function f (x), and, thus, one needed to a x the expansion ∞ i=0 i compute the value of f (x). In other terms, ﬁrst, results were formally derived and then were subjected to reinterpretations that adapted them to concrete circumstances and ﬁxed the bound of numerical vai lidity of the formal equality f (x) = ∞ i=0 ai x . Convergence was not a preliminary condition to the manipulation of function series. Third, even though the series

∞

ai xi was convergent to the function f (x)

i=0

over a certain interval I, mathematicians did not think that the manipulation of such an equality needed to be restricted to the values of x belonging to such an interval.180 In other words, the relation ∞ f (x) = i=0 ai xi was considered as valid in manipulations independently of the value of x. Of course, it did not have a quantitative meaning outside the interval of For instance, though it convergence. ∞ i xi converges only for |x| < 1, (−1) was well known that the series i=0 ∞ 1 i i = the relation 1+x i=0 (−1) x was freely used in manipulations, without being restricted to the condition |x| < 1. Finally, I would like to point out another important feature of the formalquantitative notion of series. Nowadays, the sum of a function series ∞ f (x) is a function f (x) that is deﬁned in an appropriate subset A k=0 k of ℜ by considering the sums of the convergent numerical series f (α) = ∞ f (α), where α belongs to the subset A. Function series are, in a k=0 k sense, subordinated and reduced to numerical series whereas 17th- and 18thcentury mathematicians expanded functions into series without referring to numerical series. For instance, Mercator’s rule enabled the development of 1 without being based upon numerical series. the function 1+x 179

Mathematicians were open to the possibility of ﬁnding other procedures and, in eﬀect, other, more particular procedures were applied in some particular cases. 180 This is also an example of the generality of algebra; see Section 18.3.

120

Convergence and Formal Manipulation

This fact makes it possible to understand why the problem of the development of a function into a power series was viewed as a direct problem within the framework of the formal-quantitative theory of series, whereas the problem of the sum of a power series was thought to be an inverse problem.181 In eﬀect, summing a power series meant returning from the given series to the function whose expansion generated the power series. For instance, the function 1 1+x was thought to be the sum of the series ∞

(−1)i xi

i=0

1 made it possible to just because the application of Mercator’s rule to 1+x ∞ 1 i i expand 1+x and obtain i=0 (−1) x . I would like to underline the diﬀerence between this way of conceiving the problem of the sum and the modern-day one. Nowadays, the problem of the sum is the direct problem, whereas the problem of development is the 1 i xi for is the sum of ∞ (−1) inverse problem. Indeed, we say that 1+x i=0 |x| < 1, because

lim

n→∞

∞ i=0

1 1 + (−1)n xn+1 = n→∞ 1+x 1+x

(−1)i xi = lim

i i for |x| < 1; instead, we say that ∞ development of the i=0 (−1) x is the 1 1 i i function 1+x for|x| < 1, just because 1+x is the sum of ∞ i=0 (−1) x . In Chapter 33 we shall see that only when the 17th- and 18th-century methodology went into a state of crisis did the problem of the sum of a given series stop being viewed as the problem of returning to the generating function and instead became the direct problem of series theory.

181

On this question I refer to Ferraro and Panza [2003].

9

The ﬁrst appearance of divergent series

We have already seen that the early approach to series admitted the possibility of using divergent series as a means of deriving information about certain quantities. However, the problem of summing a divergent series, namely the problem of eﬀectively associating a quantity with a divergent series, was not considered. It seemed obvious that a series such as 1 − 1 + 1 − 1 + . . . had no sum, as Jacob Bernoulli had explicitly noted in 1696.182 The situation changed when Guido Grandi proved a theorem that can be formulated as follows (see Fig. 14). Given a semicircle Γ with diameter IK, let us consider the tangent gI to circle at the point I, the intersection G of gI with the secant KZ. If GF0 is equal and parallel to IK and the points Fn and GF0 are such that IG2 Fn+1 G = , Fn G IK 2 then we have ∞ ∞ (F2k G − F2k+1 G) = KX = GD, F2k F2k+1 = k=0

(69)

k=0

where D is a point on the continuation of GF0 such that KX = GD (Grandi [1703, 27–28]).

If the point Z varies on the semicircle, the points Fn describe the parabolic curves IFn A of degree 2n and the point D describes a curve KDd (today it is named the witch of Agnesi). By using the principle of continuity,183 Grandi [1703, 29] stated that if GT coincides with the side gt of the square IKgt and IG becomes equal to IK, then the point D coincides with d and the segment GD is equal to gd = IK 2 . Therefore, the relation (69) gave rise to ∞ IK (70) (IK − IK) = 2 k=0

(see Grandi [1703, 29]). This series highlighted a contrast between geometric and algebraic principles. On the one side, the principle of continuity led to (70), which is the geometric version of 1 1 − 1 + 1 − 1 + ... = ; (71) 2 on the other hand, (70) could be read as the sum 0 + 0 + 0 + . . ., and it seemed trivial that 0 + 0 + 0 + . . . was equal to 0. 182

See Chapter 5. Leibniz formulated this principle as follows: “what is true up to the limit is true at the limit”. The connection with the description of the continuous given in Section 7.2 is clear, and, in particular, the lack of distinction between closed and open segments. 183

121

122

Convergence and Formal Manipulation A

F0

g

F1

F2 F3

d

G

t

D

T

Z Γ

I

X

K

Fig. 14 Grandi’s reasoning can be translated and summarized in more analytical terms (as 18th-century mathematicians did immediately) by stating that if the equation a3 y= 2 a + x2 of the witch of Agnesi (where a is the diameter of the circle generating the witch) is expanded into the series ∞

(−1)k

n=1

x2k , a2k−1

then one obtains 1 − 1 + 1 − . . . = 1/2 by setting a = x = 1. From an analytical perspective, the problem derived from the passage from being a power series to a numerical series and posed the crucial question of the relationship between the formal and quantitative aspect of the notion of the sum. The question of the series 1 − 1 + 1 − . . . gave rise to a certain debate, of which there are many traces in Leibniz’s correspondence. For example, on June 1, 1712, Hermann wrote to Leibniz, stating that Grandi’s idea that inﬁnite zeros could be gathered to form a ﬁnite quantity was clearly ridiculous. The mistake was due to treating ∞ k=0

(IK − IK)

as though it were a convergent series (see Leibniz [GMS, 4:369–370]). Since (71) involved a basic notion of mathematics, namely the notion of continuity, Leibniz immediately realized that it could not be rejected for

9

The First Appearance of Divergent Series

123

the reason that (71) was quantitatively false. Leibniz thought he had solved the question or had at least begun the solution in a paper184 published in 1713. In this article, he ﬁrst restated that (71) could be derived by using the same procedure commonly employed to derive valid results about convergent series.185 Leibniz then reproposed Grandi’s proof and regarded it as valid on the basis of the continuity principle [GMS, 5:385]. At the end of the article, he stated very clearly that if one takes G close to the segment IK, as desired, then the diﬀerence between gd and IK 2 can be made less than any given quantity and one therefore has gd = IK 2 by reasoning in the 186 Archimedean manner. Nevertheless, Leibniz did not solve the crucial matter of the failure of algebraic rules at inﬁnity. He limited himself to reaﬃrming that the rules of algebra are to be extended to inﬁnity and that one must merely explain this exception to the general rule. The new science, he said, is not discredited by this paradox (Leibniz [GMS, 5:387]). However, he also sought to justify (71) in a diﬀerent way so that an intuitively coherent meaning could be given from an algebraic viewpoint. The partial sums of the series 1−1+1−. . . are 0 or 1, attributing an even or odd number of terms to each sum, but if the series 1 − 1 + 1 − . . . is continued up to the inﬁnite, where the nature of the number vanishes, then the possibility of distinguishing between odd and even vanishes as well [GMS, 5:386]. According to Leibniz, the sum of the “whole” series cannot be either 0 or 1; since 0 and 1 have the same probability when the number of terms is ﬁnite, their arithmetic mean provides the sum of the series when the number of term is inﬁnite [GMS, 5:386]. Leibniz refused to justify (71) in an exclusively algebraic manner and rejected certain merely formal methods that could be used to sum other divergent series such as 1 − 2 + 4 − 8 + 16 − 32 + . . .. This is made clear by an exchange of letters with Wolﬀ.187 In a letter written on June 12, 1712, Wolﬀ noted that if one considered 1 = 1 − x + x2 − x3 + . . . 1+x 184

See Epistola ad V. Cl. Christianum Wolﬁum, Professorem matheseos Halensem, circa scientiam inﬁniti [1713]. In a letter to D. Bourget on May 3, 1715, by referring to (71), Leibniz stated: “it seems that there is a manifest absurdity. It is in the Transactions of Leipzig where I think I have given the solution to this enigma of the science of the inﬁnite” [D, 6.1:219]. 185 1 = 1 − x + x2 − . . . for |x| < 1, one He gave the following example. Since y = 1+x x 1 3 5 2 4 1 obtained 1+x2 = 1 − x + x − . . . and arctan x = 0 1+x2 = x − x3 + x5 − . . . . For x = 1, π 1 the ﬁrst series furnished 1−1+1−1+. . . = 2 , whereas the third gave 4 = 1− 31 + 15 − 71 +. . . (see [GMS, 5:382–383]). 186 “[B]y assuming G close to V as desired, we can also show that GD will also be close to 12 BV [ 12 gt, with reference to Fig. 14], so that the diﬀerence can be made less than any given quantity. Whence, by inferring in the Archimedean manner, it also follows that V S [= gd] is equal to 12 BV ” [GMS, 5:387]. 187 See Leibniz [GBLW, 143–149].

124

Convergence and Formal Manipulation

and set x = 2, then one obtained 1 = 1 − 2 + 4 − 8 + 16 − 32 + . . . , 3 namely the sum of the series 1 − 2 + 4 − 8 + 16 − 32 + . . ..was equal to 13 . Wolﬀ thought that it was possible to justify this as follows. The sum of the “positive terms” 1 + 4 + 16 + . . . + 22n is equal to

8 · 22n−1 − 1 , 3 the sum of the “negative terms” 2 + 8 + 32 + . . . + 22n−1 is equal to

4 · 22n−1 − 2 . 3

Therefore, s2n+1 = 1 − 2 + 4 − 8 + 16 − 32 + . . . − 22n−1 + 22n =

4 · 22n−1 + 1 . 3

In a similar way, s2n+2 = 1 − 2 + 4 − 8 + 16 − 32 + . . . + 22n − 22n+1 = 1 + 4 + 16 + . . . + 22n + 2 + 8 + 32 + . . . + 22n+1 =

−4 · 22n + 1 4 · 22n − 1 8 · 22n − 2 − = . 3 3 3

The arithmetic mean of the partial sums s2n+1 and s2n+2 is equal to 2 · 22n−1 − 2 · 22n + 1 . 3 At this juncture, Wolﬀ identiﬁed 22n−1 and 22n for n = ∞ and stated that the sum of the series was 31 . Leibniz’s answer188 on July 13, 1712, seems to be a summary of the principles that had inspired his work. First of all, he stated that the summation of inﬁnite series usually requires “decreasing terms” (see footnote no. 2). He then proposed three criteria to examine the correctness of a sum: 1. Is there a geometrical proof? 188

See Leibniz [GBLW, 147].

9

The First Appearance of Divergent Series

125

2. Does the summation of a ﬁnite number of the terms provide anything that concerns the problem and that agrees with the inﬁnite series or which at least approaches it continuously? 3. Is the demonstration suﬃciently rigorous?189 The ﬁrst criterion underlines the strongly geometrical character of Leib1 was viewed nizian analysis. Eﬀectively, an analytical expression such as 1+x as a geometrical quantity (represented symbolically) and, consequently, each analytical argumentation had to be reduced to a geometrical proof. The second criterion concerns the quantitative meaning of the sum: Partial sums had to give information about the sum of the series and ideally furnished the whole series at the inﬁnite or, at least, approached it. In this sense partial sums of 1 − 2 + 4 − 8 − 16 + . . . increase beyond any limit whereas the nth sums of 1 − 1 + 1 − . . . stop at around 21 . As far as the third criterion is concerned, Leibniz observed that one cannot assume A = 2A for A = ∞. The reference to a suﬃciently rigorous proof (satis accurate) is especially interesting. Leibniz believed that analysis did not have a self-suﬃcient demonstrative power and it was necessary to resort to a geometric proof. In the years that followed, the debate about this kind of series continued; however, it should be speciﬁed that the question of the sum of (71), though epistemologically interesting, was not of great importance for the growth of mathematics at the beginning of the 18th century, prior to the evolution that I will describe in the second part. Divergent series were explicitly rejected by Varignon. In his Pr´ecautions a prendre dans l’usage des suites ou series inﬁnies,190 Varignon indeed ` considered the equality ∞

ani a = (∓1)i i+1 m±n m

(72)

i=0

for m > 0 and n > 0 and stated that 1. if m > n, then equality (72) was true; 2. if m < n, then equality (72) was false; 3. if m = n and the denominator is the diﬀerence m − n, then equality (72) is simply ∞ = ∞: It is true but does not provide information; 4. if m = n and the denominator is the sum m + n, then equality (72) was false. 189 190

This is Knobloch’s formulation of Leibniz’s criteria (see Knobloch [1991, 290]). See Varignon [1715, 203–225].

126

Convergence and Formal Manipulation

Varignon’s proof consists in verifying that (72) is quantitatively valid in case 1 (and in case 3) and that the ﬁrst and the second sides of (72) are not equal in the other cases. In particular, in case 4, Varignon states that the sum of 1 − 1 + 1 − . . . can never be 1/2.191 . In the 1720s, Goldbach and D.Bernoulli discussed divergent series in a series of letters dating from July 23 to October 12, 1724.192 Goldbach opened the discussion by asking Bernoulli’s opinion about the series 1 − 2 + 4 − 8 + 16 − 32 + . . . . He recalled that Varignon disapproved of the use of such a series as it was derived from an invalid division, whereas he thought that it was to be tolerated (Fuss [1843, 2:210]). In response, Daniel Bernoulli stated that the sum of 1 + 2 + 4 + 8 + . . . was −1. He calculated the sum according to the following scheme: A 1 + 2 + 4 + 8 + . . . = s, a B

2 + 4 + 8 + . . . = s − 1, s−1 1 + 2 + 4 + ... = = s. 2

The equation

s−1 =s 2

provides the sum s = −1.193 Daniel Bernoulli used the same reasoning as Leibniz and Jacob Bernoulli did in summing194 ∞ 2 = 2. n(n + 1) n=1

However, he neglected the constraint a∞ = 0. It is very interesting to highlight similarities and diﬀerences between Daniel Bernoulli’s and Jacob Bernoulli’s behavior on this point. We saw 191

It is worth noting that Varignon’s view is the same as Jacob Bernoulli’s. As we saw in chapter 5, p. 84, in the third part of his Positiones arithmeticae de seriebus inﬁnitis (see [1689–1704, 749–752]), published in 1696 (before Grandi posed the question of the series 1 − 1 + 1 − . . ., Bernoulli had examined the non inelegans paradox (54) and had provided an explanation. 192 See Fuss [1843, 2:210–226]. Their main ﬁndings were later published (see Goldbach [1720], [1727], [1728] and Bernoulli [1728] and [1771]). 193 See Fuss [1843, 2:214–216] and Bernoulli [1771, 72]. Goldbach criticized this procedure, observing that the series A and B, though similar, did actually diﬀer. He considered two bodies that moved according to the laws expressed from series A and B (that is to say, in the ﬁrst hour A covered the space 1 and B covered no space, in the second hour A covered 2 and B covered 1, etc.) and noted that the distance between the two bodies became increasingly larger (see Fuss [1843, 2:220–221]). 194 See Chapters 2 and 5.

9

The First Appearance of Divergent Series

127

that mathematicians summed series by applying the rules (68) formally; only later did they verify whether the result was a convergent series. Thus Jacob Bernoulli did not consider a∞ = 0 as a preliminary condition; it rather explained why the procedure succeeded in summing a convergent series. In other words, Jacob Bernoulli applied the rules (68) independently of convergence; however, he was interested in convergent series, and this made the constraint a∞ = 0 crucial. The context of Daniel Bernoulli’s derivation is diﬀerent. He intended to give a meaning to the sum of a divergent series, a problem that derived from the observation that the relation a − a + a − a + . . . = a2 had a geometric meaning. In order to do this, the condition a∞ = 0 had to be neglected. Daniel Bernoulli showed that the mere transformation of a divergent series by the usual rules (68) allowed it to be summed and that the possibility of giving a sum to a divergent series was a natural consequence of the usual procedures in series theory. In a letter to Goldbach, Daniel Bernoulli also illustrated another method for summing divergent series (see Fuss [1843, 2:216]). He considered the partial sums x an D(x) = n=1

of a given series partial sums

∞

n=1 an ,

where x = 2m + 1 is an odd number, and the P (x) =

x

an ,

n=1

where x = 2m is an even number. He stated that the arithmetic mean of P (x) and D(x) is equal to the sum of series ad inﬁnitum. For instance, Bernoulli considered the series 1 − 2 + 4 − 8 + . . . and found D(x) =

x−1

(−2)n =

n=0

1 − (−2)x 1 − (−2)x = , 1 − (−2) 3

which is equal to

1 + 2x 3 since x is an odd number. Similarly, he derived P (x) =

x−1

(−2)n =

n=0

1 − 2x 1 − (−2)x = . 1 − (−2) 3

Consequently, the sum of the series is S=

D(x) + P (x) 1 = . 2 3

128

Convergence and Formal Manipulation In the same way, Bernoulli calculated the sum of the following series

a − a + a − a + ... 1 − 2 + 3 − 4 + 5 − ... 1 − 3 + 5 − 7 + 9 − ... 1 − 4 + 9 − 16 + 25 − . . .

D(x) = a D(x) = x+1 2 D(x) = x 2 D(x) = x 2+x

P (x) = 0 P (x) = − x2 P (x) = −x 2 P (x) = − x 2+x

S S S S

= = = =

D(x)+P (x) 2 D(x)+P (x) 2 D(x)+P (x) 2 D(x)+P (x) 2

= a2 , = 41 , = 0, = 0.

This procedure cannot be translated into modern terms by stating that Bernoulli, de facto, calculated sx + sx+1 , 2 where sx is the partial sum of the series ∞ n=1 an , because if x is odd and D(x) = sx , then P (x) is diﬀerent from sx+1 = P (x + 1).195 In a sense, Bernoulli took the mean of the analytical expressions P (x) and D(x) and the result represented the expectation that the sum assumed the form P (x) or the form D(x). Daniel Bernoulli’s procedure was essentially based upon the fact that the sum of the above series can be expressed as lim

x→∞

H + (−1)x f (x), for ﬁnite x, where H is a constant and f (x) is an appropriate function of the index x. He takes (−1)x f (x) equal to zero when x is inﬁnite. Daniel Bernoulli presented his ideas on the summation of series in a paper published only in 1771, where he observed that the sum of 1 − 1 + 1 − 1 + . . . had the form sx =

1 (−1)x − . 2 2

He stated that if one assumed that the sum of series to be 12 + k, then one could equally assume it to be 21 − k; therefore, one must assume k = 0 to avoid contradictions (D. Bernoulli [1771, 72–73]). 195

Thisprocedure is therefore diﬀerent from Hutton’s method, which deﬁnes the sum of a series ∞ n=1 an as follows. If we take (k−1)

Sn(k) =

(0)

with Sn =

n

i=1

(k−1)

Sn−1 + Sn 2

(n ≥ 0),

(k)

ai and S−1 = 0 (k ≥ 0), and if lim Sn(k)

n→∞

exists, then the series is said to be (Hu, k)-summable and its sum is S = lim Sn(k) . n→∞

On Hutton’s method, see Hardy [1949, 21–22].

9

The First Appearance of Divergent Series

129

In this paper, Daniel Bernoulli also observed that the partial sums of periodic series of the type an+k = an , with a0 + a1 + . . . + ak−1 = 0, assumes solely k values, S0 , S1 , . . . , Sk−1 . In this simple case he took ∞

n=1

an =

S0 + S1 + . . . + Sk−1 k

(see D. Bernoulli [1771, 77–80]). In his correspondence with Bernoulli, Goldbach, in his turn, suggested a method for summing divergent series which was based on the formal transformations of a divergent series using (68). He published his ﬁndings in De ∞transformatione serierum [1727]. In this paper he considered a series n=1 bn such that a. the “last” term bn is inﬁnitesimal, b. ∞ n=1 bn = 1. ∞ ∞ Given a series ∞ n=1 bn so to obtain n=1 an by n=1 an , he multiplied the series ∞ ∞ ∞ bn . cn = an n=1

n=1

n=1

∞ The sum of n=1 cn is precisely ∞ equal to the sum of n=1 an . Therefore, if one was able to calculate n=1 cn , then one also determined ∞

∞

n=1

In particular,

∞

n=1 bn

an =

∞

cn .

n=1

could be considered as being of the form

1 + d1 − d1 + d2 − d2 + d3 − d3 + d4 − d4 + . . . . n n Goldbach applied his method to the geometric series ∞ n=0 (−1) m , m > 1. 196 He chose m2 + m − 1 dn−1 d1 = , dn = , m+2 m+2 rearranged the product appropriately, and obtained n ∞ ∞ 1 n n (−1) m = . m+2 n=0

196

n=0

The method actually consists of adding ﬁctitious terms to a given series. This idea was probably the cue for Euler’s Delucidationes in capita postrema calculi mei diﬀerentialis de functionibus inexplicabilibus (see [EDel]).

130

Convergence and Formal Manipulation

Since

∞

n=0

Goldbach derived

∞

1 m+2

n

=

(−1)n mn =

n=0

1 , m+1

1 . m+1

It is worthwhile noting that Goldbach did not use ∞ ∞ ∞ ∞ an = an bn = cn n=1

n=1

n=1

n=1

∞ as a deﬁnition of the sum: He tackled the question of the sum of n=1 an ∞ as if the symbol n=1 an had a natural meaning that he merely had to discover. He never considered the idea that it was necessary to deﬁne the a before manipulating it. sum of ∞ n n=1

10

De Moivre’s recurrent series and Bernoulli’s method

In the ﬁrst edition of his The Doctrine of Chance [1718, 127–134] A. de Moivre dealt with a peculiar type of power series, which he later named as recurrent series in his De fractionibus algebricis [1722, 175–176]. According to de Moivre, the terms of a recurrent series are “so related to one another that each of them may have to the same number of preceding terms a certain given relation, always expressible by same index” [1718, 133].198 Even though this deﬁnition is very general, de Moivre in reality considered only power series and linear laws of relation; in other words, he considered a power series an xn as recurrent or recurring series if the coeﬃcient an of its general term an xn was a linear combination of a ﬁxed number of antecedent terms: an = b1 an−1 + b2 an−2 + . . . + bs an−s .

(73)

The constants (b1 , b2 ,

. . . , bs )

were called the scale of series, and the polynomial 1 − b1 x − b2 x2 − . . . − bs xs was termed the diﬀerential scale of series.199 For instance, 1 + 3x + 7x2 + 17x3 + 41x4 + 99x5 + . . . . is a recurrent series, its scale is (2, 1), and its diﬀerential scale is 1 − 2x − x2 . Recurrent series were a topic of remarkable interest during the 18th century. Even though they were power series, they were not of interest for their capability of representing analytical expressions in the quantitative sense. Mathematicians began to use them because they oﬀered the possibility of investigating the combinations of objects. In the theory of recurrent series, questions of convergence were non-existent. The letter x was treated in a merely combinatorial way, namely as a mere sign, a placeholder, and one operated upon series merely by combining and rearranging letters and numbers. Here are some examples. In his Miscellanea analytica, de Moivre formulated the following theorem: 198 In the second and third editions of The Doctrine of Chance, De Moivre gave the following deﬁnition: ”a recurring series ... is so constituted, that having taken at pleasure any number of its terms, each following term shall be related to the same number of preceding terms, according to a constant law of relation.” [1738, 183 and 1756, 220]. 199 Although 18-century mathematicians generalized this notion (see, for instance, Goldbach [1728, 164]), they maintained de Moivre’s use of the term “recurrent series” for series given by the law of recurrence (73).

133

134

The Development of a More Formal Conception If the series

∞

an xn

n=0

has the scale (b1 , −b2 ) and the equation x2 − b1 x + b2 = 0 has the roots ξ 1 = ξ 2 , then the given series can be divided into two geometric series with ratios ξ 1 x and ξ 2 x. (de Moivre [1730, 27–28]) The proof was based fact that the relation between the terms upon the n , whose ratio is µ, can be written not only C x of the geometric series ∞ n=0 n in the form Ci = µCi−1 but also in the form Ci = (µ + λ)Ci−1 − µλCi−2 , for any number λ. Therefore, if ξ 1 and ξ 2 are the roots of the equation x2 − b1 x + b2 = 0, the relation between the terms of each of the progressions Ai

∞

ξ ni xn

(i = 1, 2)

n=0

can be expressed by Ai ξ ni = (ξ 1 + ξ 2 )Ai (ξ i )n−1 − ξ 1 ξ 2 Ai (ξ i )n−2 , and the sum of these series A1

∞

ξ n1 xn + A2

n=0

∞

ξ n2 xn =

∞

(A1 ξ n1 + A2 ξ n2 )xn

n=0

n=0

has a scale (ξ 1 + ξ 2 , −ξ 1 ξ 2 ). Of course, b1 = ξ 1 + ξ 2 , b2 = −ξ 1 ξ 2 , and therefore the sum A1

∞

ξ n1 xn + A2

n=0

n=0

is exactly

∞

∞

n=0

an xn .

ξ n2 xn

10

De Moivre’s Recurrent Series and Bernoulli’s Method

135

De Moivre considered the analogous result for series whose scale has length 3 and 4, this allowed for the realization that a recurrence series, whose diﬀerential scale has no multiple roots, can generally be reduced to the sum of geometric series (see de Moivre [1730, 29–31]). De Moivre also showed that certain operations on series corresponded to operations upon diﬀerential scales. For instance, in his [1738, 193],200 he proved that if the series ∞

an xn

and

∞

bn xn

n=0

n=0

have the diﬀerential scales 1 − α1 x + α2 x2

and

1 − β 1 x + β 2 x2 ,

then the series ∞

n=0

cn xn =

∞

n=0

an xn +

∞

bn xn

n=0

has the diﬀerential scale (1 − α1 x + α2 x2 )(1 − β 1 x + β 2 x2 ). The heart of the theory of recurrent series is the observation that any rational function with a numerator whose degree is less than the denominator can be expanded into a recurrent series and, vice versa, that any recurrent series can be summed and that sum is a rational function with a numerator whose degree is less than the denominator (see de Moivre [1730, 27–35]). n In order to prove that the sum of a recurrent series z = ∞ n=0 an x , with scale (b1 , . . . , bs ), is a rational function de Moivre operated according to the following scheme:201 200

See also de Moivre [1756, 228]. The general scheme is a reconstruction. As often occurred, de Moivre considered some particular cases as arbitrary exampliﬁcations. An “arbitrary exempliﬁcation” is a demonstrative technique that consists of using an appropriate example rather than a general demonstration. It is based on the repeatability of the procedure employed, which is invariable with respect to the speciﬁc exempliﬁcation. 201

136

The Development of a More Formal Conception

z= = =

∞

n=0 s−1

n=0 s−1

an xn = an xn +

s−1

an xn +

∞

an xn

n=s

n=0 ∞

(b1 an−1 + b2 an−2 + . . . + bs an−s )xn

n=s ∞

an xn + b1

(an−1 xn−1 )x

n=s

n=0

+ b2

∞

(an−2 xn−2 )x2 + . . . + bs

n=s

=

s−1

∞

(an+s−1 xn+s−1 )x

n=0

+ b2

∞

(an+s−2 xn+s−2 )x2 + . . . + bs

n=s

=

s−1

n=0

(an−s xn−s )xs

n=s

an xn + b1

n=0

∞

an xn + b1 x(z −

∞

n=0 s−3

s−2

an xn ) + b2 x2 (z −

s−2

an xn ) + b2 x2 (z −

n=0

(an xn )xs an xn ) + . . . + bs xs z.

n=0

By solving the equation z=

s−1

n=0

an xn + b1 x(z −

n=0

s−3

an xn ) + . . . + bs xs z

n=0

with respect to the unknown z, we have that z is equal to fraction whose denominator is 1 − b1 x − b2 x2 − . . . − bs xs and the numerator has a degree less than s (see de Moivre [1730, 27–35]) The inverse statement (the expansion of any rational function is a recurring series) could be easily derived by applying the method of indeterminate coeﬃcients. For instance, in [1730, 35] de Moivre posed 1 = P + Qx + Rx2 + Sx3 + . . . . 1 − f x + gx2 − hx3 + kx4 Hence,

and

1 − f x + gx2 − hx3 + kx4 P + Qx + Rx2 + Sx3 + . . . = 1 Q = f P, R = f Q − gP, S = f R − gQ + hP, T = f S − gR + hQ − kP, ...

10

De Moivre’s Recurrent Series and Bernoulli’s Method

137

Consequently, P + Qx + Rx2 + Sx3 + . . . is a recurrence series with the scale of relation (f, −g, h, −k). These results pave the way for the identiﬁcation of any recurrent series with a function (later referred to as a generating function). In his [1748a, 1:175], Euler made this explicit and deﬁned the sum of a recurrent series to be the function that generated it. De Moivre and other 18th-century mathematicians attached importance to the determination of the general term of a recurring series. They observed that the general term an of a recurrent series could be written in the form an =

r

Pi (n)qin ,

i=1

where qi are the roots of equation 1 − b1 x − b2 x2 − . . . − bs xs = 0 and Pi (n) is a polynomial of degree equal to 1 less the multiplicity of qi (see de Moivre [1730], Bernoulli [1728], and Euler [1748a]). For instance, if we consider the Fibonacci sequence 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, . . . deﬁned by an = an−1 + an−2 , since the scale of recurrence is (1, 1) and the roots of equations x2 = x + 1 √ are 1±2 5 , the general term of the Fibonacci sequence is α

√ n √ n 1− 5 1+ 5 +β . 2 2

The numbers α and β are to be determined by taking into account a0 = 1 and a1 = 1: So one has 1 α=β= √ 5 (see Bernoulli [1728, 89–90]). In Observationes de seriebus, Daniel Bernoulli reversed the last result and, using several examples, showed how recurrent series could arise in numerical computations, involving only combinatorial properties of series. Indeed, if |r1 | > |r2 | > . . . > |rn |

138

The Development of a More Formal Conception

are the roots of the equation 1 = a1 x + a2 x + . . . + an xn , D. Bernoulli explained that in order to determine r1 , one had to write a recurrent sequence bn with the scale a1 , a2 , . . . , an and whose ﬁrst n terms are arbitrary. Then the ratio bn bn+1 “is nearly equal to a sought root”. For instance, consider 1 = −2x + 5x2 − 4x3 + x4 and form the sequence with the scale (−2, 5, −4, 1) and the ﬁrst four terms equal to 1, namely the series 1, 1, 1, 1, 0, 2, − 7, 25, − 93, 341, − 1254, . . . . Then one of the roots of the equation is approximately equal to −341 1254 (see D. Bernoulli [1728, 92]). This method diﬀered from other already known techniques for determining the roots (which consisted of the approximation of a numerical value by means of the nth partial sum of a convergent series). It oﬀered a sequence whose terms had a numerical value only indirectly connected to the result: These were the steps of a transformation that enabled one to arrive at the ﬁnal result. It was evident that the divergence of the sequence was of no importance for determining the numerical value of the root. Bernoulli’s method was later improved by Euler and Lagrange. In his [1748a, 1:339–361], Euler reformulated Bernoulli’s method as follows. Let us consider the equation F (x) = 1 − b1 x − b2 x2 − . . . − bs xs = 0, If we suppose that the roots qi are simple, then F (x) can be thought of as the denominator of a rational function s

G(x) Ai = , F (x) 1 − qi x i=1

where G(x) is an arbitrary polynomial with degree less than s. (This corresponds to the arbitrary choice of the ﬁrst n terms of the recurrent series in Bernoulli’s formulation.) Since each addend can be expanded into a power series ∞ Ai qih xh , h=0

10

De Moivre’s Recurrent Series and Bernoulli’s Method

139

by rearranging we have ∞ s ∞ G(x) Qh xh , Ai qih )xh = ( = F (x) h=0 i=1

h=0

where Qh =

s

Ai qih .

i=1

From the theory of recurrent series, we know that this is a recurrent series with the scale (b1 ,. . .,bs ). At this point Euler supposed that the root |q1 | was greater than |qi |, i = 2, . . . , s and stated that if h is an inﬁnite number, then we have Qh = A1 q1h ; “but if h is only a very large number, we have very nearly” Qh = A1 q1h . Similarly, Qh+1 approximates A1 q1h+1 for large h, and so Qh+1 Qh is an approximation of q1 . The polynomial G(x) has to be chosen so that the coeﬃcient A1 is different from 0 [then 1 − q1 x is not a factor of G(x)]. Once q1 has been determined, one can choose G(x) so that it contains the factor 1 − q1 x and in this way it becomes possible to determine the root |q2 | where |q2 | is the greatest in the set {|qi |, i = 3, . . . , s}. If there are two roots q1 and q2 such that q1 = −q2 and |q1 | = |q2 | > |qi |, Euler showed that

Qh+1 Qh

i = 3, . . . , s,

did not approximate q1 ; however, Qh+2 Qh

approximated |q1 |2 . He also dealt with the case where there are two complex conjugate roots and proposed a method for accelerating the convergence when there are multiple roots (the convergence might be very slow in this case). Euler even applied Bernoulli’s method to inﬁnite equations. Indeed, he sought the largest root of 1 1 1 5 1 7 = z − z3 + z − z + ... 2 6 120 5040

140

The Development of a More Formal Conception

(it is the equation

1 2

1 1 − 2z + 31 z 3 −

= sin z). He wrote

1 5 60 z

+ ...

= 1+

∞ h=1

h 1 1 2z − z 3 + z 5 − . . . 3 60

23 4 44 4 z + z 3 3 1681 5 2408 6 + z + z + ... 60 45

= 1 + 2z + 4z 2 +

Hence, he obtained z=

1681 2408 · ≈ 0.52356 60 45

(see Euler [1748a, 360]). At the end of the 18th century, Lagrange also made a contribution to Bernoulli’s method. In Trait´e de la resolution des `equations numeriques [1798], he assumed that the numerator G(x) of the fraction G(x) F (x) was chosen to be the derivative of the denominator F (x). In this way the denominator ′ (x) had no multiple roots. of the fraction FF (x)

11

Acceleration of series and Stirling’s series

In the ﬁrst part of this book, we saw that power series were used with a quantitative purpose, but mathematicians separated the investigation of convergence from the search of the law of formation of coeﬃcients. Once this law was found, it seemed easy to establish the interval of convergence (given the very simple nature of series employed), which consisted of ﬁtting analytical results to the speciﬁc problem by making an appropriate choice of the parameters. From this point of view, the crucial problem was not convergence per se but achieving fast convergence. Researchers indeed endeavored to speed up the convergence of series. However, this search was also undertaken through purely algebraic manipulations, and the usefulness of the results was only tested a posteriori. In this way, it led to results that could not be reduced to the formal-quantitative concept of series. These results are the subject matter of the present chapter. The acceleration of series underwent extensive investigation by James Stirling. In the preface to his Methodus diﬀerentialis [1730], Stirling explained that series often converged so slowly that they were de facto no more useful than if they had been divergent. For this reason, he provided various methods in the ﬁrst part of his treatise by means of which the sum of these series could be quickly obtained. I shall focus on three of Stirling’s theorems. The ﬁrst proposition concerns the sum of ﬁnite series. The second concerns the transformation of inﬁnite series into a more rapid convergent series. The third proposition is Stirling’s derivation of Stirling’s series,202 the ﬁrst example of an asymptotic series (as they were termed later203 ). To prove this last theorem, he used the same methodology as in the other two propositions. He obtained a series 202

Today, the expression

∞

n=1

1 1 1 1 1 B2n = + − ... (74) log 2π + z − − log z − z + 2n(2n − 1)z 2n−1 2 2 12z 360z 3 1260z 3

is called Stirling’s series. 203 An asymptotic series is a series expansion of a function in a variable z that may converge or diverge but whose partial sum can be made good approximation to an arbitrarily −i is the asymptotic expansion a given function for large enough z. For instance, ∞ i=0 ai x of f (x), if ∞ n −i ai x =0 lim x f (x) − x→∞

i=0

for ﬁxed n. The asymptotic series for the gamma function Γ(z) is given by √ 139 1 1 e−z z z−1/2 2π 1 + − + ... . + 12z 288z 2 51840z 3 Stirling’s series is the asymptotic expansion of log Γ(z).

141

142

The Development of a More Formal Conception

that was very eﬀective for approximate calculations, but, unlike the series in Proposition 2, was not convergent. This result was, in a sense, unexpected and diﬃcult to be set against in the formal-quantitative series theory. Proposition 1. The sum of the ﬁnite series

z

n=1 an ,

where

an = b 0 +b 1 n + b 2 n(n − 1 ) + b 3 n(n − 1 )(n − 2 ) + . . ., is z

ai

n=1

1 1 = b0 z + b1 (z + 1 )z + b2 (z + 1 )z (z − 1 ) 2 3 1 + b3 (z + 1 )z (z − 1 )(z − 2 ) + . . . 4

(Stirling [1730, 20]). z In the proof, Stirling assumes that the thesis is true and lets S(z) = n=1 ai and S(0) = a0 = 0. Then S(z − 1) =

z−1

n=1

1 1 ai = b0 (z − 1) − b1 z(z − 1) − b2 z(z − 1)(z − 2) 2 3

1 − b3 z(z − 1)(z − 2)(z − 3) + . . . 4 Since [(z + 1)z(z − 1) . . . (z − n + 2)] − [z(z − 1) . . . (z − n + 1)]

= nz(z − 1) . . . (z − n + 2), he has S(z) − S(z − 1) = az

1 1 = b0 z + b1 (z + 1)z + b2 (z + 1)z(z − 1) 2 3 1 + b3 (z + 1)z(z − 1)(z − 2) + . . . 4 1 1 −b0 (z − 1) − b1 z(z − 1) − b2 z(z − 1)(z − 2) 2 3 1 − b3 z(z − 1)(z − 2)(z − 3) − . . . 4 = b0 + b1 z + b2 z(z − 1) + b3 z(z − 1)(z − 2) + . . .

for all z > 0. Then he stated that since S(z) = S(z − 1) + az ,

11

143

Acceleration of Series and Stirling’s Series

if az is given, one inversely obtains the sum S(z).204 Using this theorem Stirling obtained 1 + 2 + ... + z =

z(z + 1) 2

and 1 + 8 + 27 + . . . + z 3 =

z2 (z + 1)2 4

[since z 3 = z + 3z(z − 1) + z(z − 1)(z − 2)]. Stirling employed a similar procedure to accelerate the convergence of an inﬁnite series. ∞ an , where Proposition 2. The sum S(z) of the inﬁnite series n=z

an =

b1 b2 b0 + + n(n + 1) n(n + 1)(n + 2) n(n + 1)(n + 2)(n + 3) b3 +..., + n(n + 1)(n + 2)(n + 3)(n + 4)

is

S (z ) =

b1 b2 b0 + + z 2z(z + 1) 3z(z + 1)(z + 2) b3 + ... + 4z(z + 1)(z + 2)(z + 3)

(75)

(Stirling [1730, 25]). If the thesis were true, we would have S(z + 1) =

b1 b2 b0 + + z + 1 2(z + 1)(z + 2) 3(z + 1)(z + 2)(z + 3) b3 + + .... 4(z + 1)(z + 2)(z + 3)(z + 4)

Since 1 1 k − = z(z + 1) . . . (z + k − 1) (z + 1) . . . (z + k) z(z + 1) . . . (z + k)

we obtain

b1 b0 + z(z + 1) z(z + 1)(z + 2) b3 b2 + + .... + z(z + 1)(z + 2)(z + 3) z(z + 1)(z + 2)(z + 3)(z + 4)

az = S(z) − S(z + 1) =

204 Stirling supposes that if the thesis were true for S(z − 1), it would be true for S(z) = S(z − 1) + az .

144

The Development of a More Formal Conception

Conversely if this is the general term, the sum is given by (75) (Stirling [1730, 25]).205 Stirling applied this theorem to determine the value of various sums. In particular, he found ∞ 1 ≈ 1.644934065. n2 i=1

The third proposition that I would like to consider is the following.

Proposition 3. If the numbers x + n, x + 3n, x + 5n, . . ., z − n are in an arithmetic progression, then lg(x + n) + lg(x + 3n) + lg(x + 5n) + . . . + lg(z − n) (76) 3 5 7 9 az an 7an 31an 127an 511an z lg z − − + − + − + ... = 3 5 7 2n 2n 12z 360z 1260z 1680z 1188z 9 an 7an3 31an5 127an7 511an9 x lg x ax − − + − + − + ... − 2n 2n 12x 360x3 1260x5 1680x7 1188x9 where a = log110 and lg X and log Y denote the logarithm to the base 10 and natural logarithm of the numbers X and Y , respectively (Stirling [1730, 135]). To prove this proposition, Stirling considered

A=

az an 7an3 31an5 z lg z − − + − + ... 2n 2n 12z 360z 3 1260z 5

and B =

(z − 2n) lg(z − 2n) a(z − 2n) − 2n 2n 7an3 31an5 an − + − + ... 12(z − 2n) 360(z − 2n)3 1260(z − 2n)5

The terms of the series B were reduced to the same form as the terms of the series A “by performing a suitable division”. This means that he considered 205

An alternative procedure can be found in Stirling [1719].

11

145

Acceleration of Series and Stirling’s Series

the expansion of

(z−2n) lg(z−2n) : 2n

(z − 2n) lg(z − 2n) 2n

z lg(z − 2n) − lg(z − 2n) 2n z 2an 4an2 8an3 = lg z − − − − . . . 2n z 2z 2 3z 3 2an 4an2 8an3 − − . . . − − lg z − z 2z 2 3z 3 an 4an2 z − ... lg z − a − − = 2n z 3z 2 2an 4an2 8an3 − lg z − − − − ... z 2z 2 3z 3 =

an , , − 12(z−2n) and the expansions of − a(z−2n) 2n

a(z − 2n) 2n an − 12(z − 2n) 7an3 360(z − 2n)3 31an5 − 1260(z − 2n)5 −

7an3 , 360(z−2n)3

5

31an − 1260(z−2n) 5 , . . .:

az + a, 2n an 2n 4n2 8n3 =− + 2 + 3 + ... , 1+ 12z z z z 2n 4n2 7an3 8n3 1 + 3 + 6 = + 10 + . . . , 360z 3 z z2 z3 31an5 2n 4n2 =− 1 + 5 + 20 + . . . . , 1260z 5 z z2 ... =−

Then he subtracted these expansions from A and obtained an an2 an3 an4 − 2 − 3 − 4 − ... D1 = A − B = lg z − z 2z 3z 4z n = lg z − lg(1 − ) = lg(n − z). z In a similar way, he found that log(z − 3n) is equal to D2 = B − C, where C =

an (z − 4n) lg(z − 4n) a(z − 4n) − − 2n 2n 12(z − 4n) 7an3 31an5 + − + .... 360(z − 4n)3 1260(z − 4n)5

So, generally, one can express any of the logarithms lg(z − 5n),

...,

lg(x + 3n),

by the diﬀerence of the type D1 , D2 , . . .. furnishes (76).

lg(x + n)

The sum of these diﬀerences

146

The Development of a More Formal Conception In a subsequent example, Stirling obtained the relation 1 1 1 1 lg x + −a x+ lg 2π + x + lg x! = 2 2 2 2 a 7a + − 3 − . . . 2 · 12 x + 21 8 · 360 x + 1

(77)

2

by setting x = 21 , n = 12 , and z = x + 21 . Series (77) is equivalent to series (74), and its ﬁrst few terms provide extremely accurate approximations to log x! for large x; however, it is divergent. I emphasize that Stirling’s proof employed the usual modality of manipulation of series:206 Asymptotic series emerged in a rather unconscious way.207 This means that, although series theory was designed to determine convergent series, the fact that the conditions of convergence were not determined a priori and manipulations were unrestricted made it possible to go from convergent to divergent series. This also means that, while today the notions of asymptotic series and convergent series are diﬀerent and grounded on diﬀerent deﬁnitions, in the 18th century they were not effectively distinct. Both notions were reducible to the idea that series were inﬁnite expressions derived from the ﬁnite analytical expressions of given quantities by formal manipulations; only under certain conditions could the values of the quantities be approximated by series that were generated by the quantities themselves. In the years that followed, Euler made this concept explicit.

206 “Stirling does not perceive any diﬃculty” (Guicciardini [1989, 35]). In eﬀect, none of this enables us to understand whether Stirling understood the real nature of what today is known as Stirling’s series. 207 The same thing occurred for de Moivre, who gave a similar result in his [1730, 127].

12

Maclaurin’s contribution

Nowadays, the name Maclaurin in series theory is linked to the Maclaurin series and the Euler–Maclaurin summation formula. The ﬁrst is merely a special case of the Taylor series, which in any case was explicitly dealt with by various mathematicians. The latter was a new result, which he derived independently and almost at the same time as Euler. Maclaurin’s proofs of these results are interesting because they were part of an extreme attempt at laying the foundations of calculus on a geometrical framework. In 1742, Maclaurin published the Treatise of Fluxions, a work partly written in response to an attack by Berkeley on the principles of the inﬁnitesimal calculus.208 Maclaurin’s aim was to obtain a secure proof of Newton’s results using the ancient method of exhaustion. The treatise is divided into two books. In the ﬁrst, he presented a geometry of ﬂuxions: He investigated ﬂuents and ﬂuxions in a purely geometric way, by referring to diagrammatic representations and without using any speciﬁc signs or characters. Maclaurin thought that this would provide a complete and clear description of the theorems and would avoid the suspicion that the symbols employed in the calculus served to cover defects in the principles and the demonstrations (see Maclaurin [1742, 575]). Maclaurin209 provided a kinematic model of Newtonian theory. He considered motion, space, and velocity as primitive notions and gave the following deﬁnition of ﬂuxions: “The velocity with which a quantity ﬂows, at any term of the time while it is supposed to be generated, is called its Fluxion, which is therefore measured by the increment or decrement that would be generated in a given time by the motion, if it was continued uniformly from that term without acceleration or retardation” [1742, 57]. Maclaurin represented the measures of the ﬂuxions geometrically, by means of segments or ﬁnite surfaces. For instance, he proved that (see Fig. 15), given the curve AKP , if KL is the tangent at the point K, then BC measures the ﬂuxion of AB, IL measures the ﬂuxion of the ordinate KB,210 and KL measures the ﬂuxion of the curve211 (see Maclaurin [1742, 181]). Moreover, the rectangles BCDE and KILM measure the ﬂuxions of the trapezoid ABEJ and the rectangle BCIK, respectively. This way of representing (the measures of) ﬂuxions made it possible to consider the theorems on ﬂuxion as geometrical theorems, which Maclaurin proved by reductio ad absurdum inspired by the method of exhaustion (to 208 In the Preface of the Treatise of Fluxions, Maclaurin stated: “A letter published in the year 1734, under the title of The Analyst, ﬁrst gave occasion to the ensuing treatise, and several reasons concurred to induce me to write on this subject at so great length.” 209 On Maclaurin’s Treatise of Fluxions, see Panza [1989, 213–240] and Guicciardini [1989, 47–51]. 210 In modern terms, IL represents the derivative of the function y(x) represented by the curve AKP , provided AD ﬂows uniformly. 211 The triangle KIL is Maclaurin’s version of Leibniz’s characteristic triangle.

147

148

The Development of a More Formal Conception

prove that A = P , where A and P are geometric quantities, he proved geometrically that neither A > P nor A < P ). In the second book, Maclaurin dealt with the calculus of ﬂuxions. Here, he investigated quantities abstractly using the analytical symbolism212 (obviously, he employed Newtonian symbolism, and speciﬁcally the symbolism used in De quadratura). His aim was to provide the analytical version of the geometric results given in the ﬁrst part. He attempted to give an analytical version of the double reductio ad absurdum (to prove that A = P , where A and P are analytical quantities, he proved analytically that neither A > P nor A < P ). In this context, he often made remarkable use of the technique of algebraic inequalities whose inﬂuence on later mathematics should not be underestimated. For example, in order. to demonstrate that if the ﬂuxion of . n A is A = a, then the ﬂuxion of An is A = naAn−1 , Maclaurin stated that if one supposed that the ﬂuxion of An to be greater than naAn−1 or less than naAn−1 , then one arrived at a contradiction. Indeed, he ﬁrst .showed n that, if A assumes the values213 A − u, A, A + u, then the ﬂuxion A of A satisﬁes the condition .n

An − (A − u)n < A < (A + u)n − An .

(78)

If the ﬂuxion of An is greater than naAn−1 , then it is equal to naAn−1 +σ, where σ > 0. If σ n−1 An−1 + − A, o= nA then one obtains naAn−1 + σ = na(A + o)n−1 . .n

.

Hence, the ratio between A and A is n(A + o)n−1 . Suppose now that u is any increment of A less than o; according to Maclaurin, since a is to u as n(A + o)a to n(A − o)u, it follows214 that if the ﬂuxion of A should be represented by u, the ﬂuxion of An would be represented by n(A + o)n−1 , which is greater than nu(A + u)n−1 . Since (A + u)n − An < n(A + u)n−1 , u 212

He explicitly referred to “quantities considered abstractly, or as represented by general characters in algebra” [1742, 575]. In the second book, ﬂuxions are deﬁned in the following way: “By the ﬂuxions of quantities we shall therefore now understand any measure of their respective rates of increase or decrease, while they vary (or ﬂow) together” [1742, 579]. 213 A and u are positive quantities. 214 In [1742, Section 706] Maclaurin stated: “As the ﬂuxion of quantities are any measures of the respective rates according to which they increase or decrease, [. . .] so it is of no importance how great or small soever those measures are [. . .] Therefore if the ﬂuxions of A and B may be supposed equal to a and b, respectively, they may be likewise supposed a and m b.” equal to 12 a and 21 b, or to m n n

149

12 Maclaurin’s Contribution .n

for A > 0 and u > 0; one easily obtains A > (A + u)n − An , contrary to (78). Similarly, he shows that the ﬂuxion of An is not less than naAn−1 (see Maclaurin [1742, Sections 704 and 713])

P N M

L

K

I

J

H G E

A

B

D

C

Fig. 15 I shall now move on to consider the Taylor series. In the ﬁrst part of the Treatise, Maclaurin gave the geometrical version of the Taylor theorem as a corollary of the following proposition. Proposition 20. The trapezoid ABEJ and the rectangle BCIK have the same area if, and only if, BE = IL; namely: the curve JE, whose ordinate BE is the ﬂuxion of the curve AK, whose ordinate is KB if, and only if, the area under the curve JE is equal to the ordinate KB. In modern terms, if BE = f (x) and KB = F (x), Proposition 20 can be formulated by stating that F ′ (x) = f (x) if, and only if, F (x) =

x

f (x)dx. 0

One of the corollaries of Proposition 20 asserts the following:

150

The Development of a More Formal Conception a. If the arc EH is a segment, then the increment IP of the ordinate of the curve AK can be divided into two parts, IL = BE and LP = 12 DG, where DG is (or, to use a more precise term, measures) the ﬂuxion of the ﬂuxion.

x That is to say, let ΔF be the increment of the ordinate F (x) = 0 f (x)dx of the curve; if one assumes that F (x) is of the type ax2 + bx + c, then ′′ ΔF = f ′ (x) + 21 f (x). The next corollary concerns the curves of the type F (x) = ax3 + bx2 + cx + d. b. If the arc EH is a parabola, then the increment IP of the ordinate of the curve AK can be divided into three parts, IL = BE, LN = 12 DG, N P = 13 GH, where DG is the ﬂuxion of BE (namely, DG is the second ﬂuxion of BK), and GH is 21 of the second ﬂuxion of BE (namely, GH is the third ﬂuxion of BK). In other terms, if F (x) = ax3 + bx2 + cx + d, then

1 1 1 ΔF = f ′ (x) + f ′′ (x) + · f ′′′ (x). 2 3 2

Maclaurin then generalized these corollaries. By assuming that the curve AKP could be expressed by a polynomial of degree n or by an inﬁnite polynomial, he stated: [T]he increment of a ﬂuent can be approximated continually by adding continually together the right line that measure the ﬁrst ﬂuxion [of BK,] . . . 12 of that which measures the second ﬂuxion of the ordinate, 16 of that which measures its third 1 ﬂuxion, 24 of that which measures its fourth ﬂuxion, and so on (Maclaurin. [1742, 223–224]) In the second volume of the Treatise, Maclaurin, following Newton, used series to solve diﬀerential equations or, to use his terminology, to ﬁnd the ﬂuent. He stated: When a ﬂuent cannot be represented exactly by in algebraic terms, it should then be expressed by a convergent series. (Maclaurin [1742, 604])

12

151

Maclaurin’s Contribution

In this context, he also provided an analytical proof of the Taylor theorem that reproduced Newton’s and Stirling’s proofs; it corresponds exactly to the above geometrical proof. Maclaurin assumed that a function could be expanded into a power series A + Bx + Cx2 + . . . and then determined the coeﬃcients A, B, C by means of step-by-step diﬀerentiation (see Maclaurin [1742, 610–612]). Maclaurin’s interest in the use of convergent series to express a ﬂuent led him to examine the problem of the velocity of convergence. Indeed, he noted that “when an area or a ﬂuent is reduced to a series . . . these series, in some cases, converge at so slow a rate as to be of little use for ﬁnding the area” [1742, 670]. As an example, Maclaurin considered the expansion 1 and stated: of y = 1+x [T]his series converges so that the sum of the ﬁrst 1000 terms of it was found deﬁcient from the true value of the area in the ﬁfth decimal; and other examples similar to this might be brought, wherein the area may be more easily computed by the inscribed polygons than from series. Some further artiﬁce is therefore necessary in order to compute the areas in such cases. [1742, 672] Thus, he determined a formula that might “be of use for this purpose”: the Euler–Maclaurin summation formula, which he formulated so: Suppose that the base AP = z [see Fig. 16], the ordinate P M = . y, and the base being supposed to ﬂow uniformly, let z = 1. Let the ﬁrst ordinate AF represented by a, AB = 1, and the area ABEF = A. Since A is the area generated by the ordinate y, so let B, C, D, E, etc. represent the areas upon the same base AB, generated by the respective ordinates .

..

...

....

y, y, y, y , etc. Then AF = a = A −

C E G B + − + − etc. 2 12 740 30240

(79)

(Maclaurin [1742, 672]). In order to obtain this formula, Maclaurin [1742, 672–673] applied the

152

The Development of a More Formal Conception f

L

E

K

F M v V

R A

P

r

B

C

D

G

Fig. 16

Maclaurin series repeatedly. In modern symbols:

A = B = C = D = E =

1

1 1 1 (4) 1 y (0) + . . . , y(z)dz = y(0) + y ′ (0) + y ′′ (0) + y ′′′ (0) + 2 6 24 120 0 1 1 1 1 y ′ (z)dz = y(1) − y(0)+y ′ (0) + y ′′ (0) + y ′′′ (0)+ y (4) (0)+. . . , 2 6 24 0 1 1 1 y ′′ (z)dz = y ′ (1) − y ′ (0) = y ′′ (0) + y ′′′ (0) + y (4) (0) + . . . , 2 6 0 1 1 y ′′′ (z)dz = y ′′ (1) − y ′′ (0) = y ′′′ (0) + y (4) (0) + . . . , 2 0 1 y (4) (z)dz = y ′′′ (1) − y ′′′ (0) = y (4) (0) + . . . ,

Maclaurin rearranged these equations, “exterminated” (to use his term) y ′ (0), y ′′ (0), y ′′′ (0), y (4) (0), . . . from the ﬁrst equation (namely, he made successive substitutions starting from the last written equation). In this way, he obtained 1

y(1) − y(0) y ′ (1) − y ′ (0) + 2 12 0 y ′′ (1) − y ′′ (0) y ′′′ (1) − y ′′′ (0) + − ... − 740 30240

y(1) = a =

y(z)dz −

12

153

Maclaurin’s Contribution

By repeating the same reasoning for y(2), y(3), . . . , he derived n−1 i=0

y(i) =

n

−

y(z)dz −

y(n) − y(0) y ′ (n) − y ′ (0) + 2 12

(80)

y ′′ (n) − y ′′ (0) y ′′′ (n) − y ′′′ (0) + − ... 740 30240

This is the Euler–Maclaurin summation formula in one of its asymmetrical forms.215 Series (80) is not necessarily convergent, but Maclaurin did not consider this fact. Like Stirling, Maclaurin came up against the problem of divergent series while he manipulated series to improve their convergence. Even though his aim was to estimate an area, the formal methodology he used led him unintentionally to stumble on divergent series.

215

Maclaurin also gave some variants of (80); see [1742, 292–293].

13

The young Euler between innovation and tradition

Series were one of Euler’s favourite topics. In his Opera omnia there are three volumes, the last of which was published in two parts, which are devoted to series (totaling more than 1800 pages). In reality, many other articles (which were not explicitly classiﬁed as belonging to series theory by the editors of Opera omnia) as well as his textbooks contain much material concerning series. The extent of his results is impressive and his inﬂuence on 18th-century mathematics was enormous. The treatment of Euler’s contributions is subdivided into various chapters in Part II and also Part III. In this chapter, I examine the main ﬁndings of Euler’s early work which appeared in issues 5 to 7 of the Transactions of the Saint-Petersburg Academy. (I explicitly note that these issues were dated 1730–1731, 1732–1733 and 1734–1735, respectively, but their actual publication occurred several years later, from 1738 to 1740.)

13.1

The search for the general term

Euler’s ﬁrst published article about series concerns the problem of ﬁnding the general term of a series. This problem had been tackled by various mathematicians in the 1720s. In Chapter 10 we saw that Daniel Bernoulli determined the general term of a recurrent series and applied this result to an approximate determination of the roots of equations.216 In Specimen methodi ad summas serierum [1720] and De terminis generalibus serierum [1728], Christian Goldbach expressed the general terms and the partial sums of certain series by means of ﬁnite diﬀerences. He also tried to generalize the results by examining the series derived by means of avariable “law of progression”.217 In [1728] Goldbach considered the series ∞ n=1 ai given by a recurrence relation and, using the formula (59), expressed its nth term as (n − 1)(n − 2) 2 n−1 ∆a1 + ∆ a1 1 1·2 (n − 1)(n − 2)(n − 3) 3 ∆ a1 + . . . , + 1·2·3

an = a1 +

(81)

where ∆an = an+1 − an and ∆n an = ∆n−1 an+1 − ∆n−1 an . His procedure was usual for this period:218 It consisted of giving non-integral values to n in (81) and so producing an infinite expression. Goldbach said that it 216

Cf. Bernoulli [1728]. “A law of progression is a formula by means of which, given one or more terms of the series we can ﬁnd another term antecedent or successive”; for instance, un+1 = kun and kun+1 +run are two constant laws of progression for k and r constants, whereas un+2 = un+1 +un un+1 = nun is variable (Goldbach [1728, 164]). 218 See, e.g., Chapters 4 and 8. 217

155

156

The Development of a More Formal Conception

approximated the general term an as desired and was highly suitable for 1 for n = 32 , he interpolation.219 For instance, in order to interpolate an = n! 1 expressed n! by means of (81) as 1 1 1 (n − 1)(n − 2) 1 (n − 1)(n − 2)(n − 3) = 1 − (n − 1) + + + ..., n! 2 6 2 24 6 where

For n =

1 1 1 a1 = 1, Δa1 = − , Δ2 a1 = , Δ3 a1 = , . . . 2 6 24 3 2

he had 1 3 2!

=1−

1 1 1 1 1 − · + · + ... 2 · 2 6 2 · 4 24 2 · 4 · 6

Goldbach also observed that if one interpolated the sequence n! for n = 32 , (81) provided the series 1+

1 3 11 265 − + − + ..., 2 8 6 128

which is divergent. He thought, however, that one could assume 3 != 2

1 3 2!

−1

=

1−

1 2·2

−

1 6

·

1 2·4

1 +

1 24

·

1 2·4·6

+ ...

.

By using (81), Goldbach simpliﬁed the procedure of interpolation that had enabled Wallis to obtain 4/π as inﬁnite products. This sort of problem attracted the attention of the young Euler, who at once obtained several important results. In a letter dated October 13, 1729 (see Fuss [1743, 1:3–7]), the ﬁrst in a long-lasting correspondence, Euler communicated to Goldbach the formula n! =

∞ k 1−n (k + 1)n , k+n

(82)

k=1

which he was about to publish in his De progressionibus transcendentibus [1730–31a] and asked him to give his opinion on it. As Euler made clear, however, he did not consider satisfactory the expression of the general term of a series by means of another series. He wrote to Goldbach (Fuss [1743, 1:4]) stating that, if one expresses the general term of a series by means of other series, then the intermediate terms, i.e., those 219

On the problem of interpolation, see Chapter 1, p. 14 and Chapter 4.1, p. 57.

13

The Young Euler between Innovation and Tradition

157

with a nonintegral index, are determined only approximately. For this reason, he stopped treating the matter using series and became interested in a method that would enable him to determine the real (and not approximate) intermediate terms. Furthermore, in De progressionibus transcendentibus [1730–31a, 3], he stated that he chose not to dwell upon (82) because he already had more suitable ways to express the nth term of the hypergeometric sequences n!. So doing, he posed the so-called problem of integration: to express the general term of a series by an integral, namely, by a formula of the kind220 b p(x, n)dx. 0

Euler explained that the function p depends on x and certain constants (of which one is n) and that the integration of p(x, n) from 0 to the real number b yields a “function of the index n and constant quantities,” namely, the general term (Fuss [1843, 1:12]). At this point, two observations are necessary. First, Euler regarded the notions of the general term and the function as formulas. In [1730–31a] he oﬀered this deﬁnition of the general term: A general term is a formula that consists of constant quantities or any other quantities like n, which gives the order of terms; thus, if one wishes the third term, 3 can be set in the place of n. (Euler [1730–31a, 4]) Second, Euler considered inﬁnite expressions as unsuitable for providing general terms. In De summatione and in his letter to Goldbach, Euler conceived (82) only as a tool for computing the terms of series approximately. Euler suggested that integration provided the “true” result, since integration was interpreted geometrically as a quadrature, i.e., the result was exact insofar as it was geometrically conceived. Euler’s method for solving the problem of integration involved considering a certain integral and ﬁnding the series whose general term corresponded to it. He later [1732–33] termed such a procedure a synthetic method, since a priori knowledge of the result was supposed and one simply veriﬁed that a certain integral (which was already known) expressed the general term of 220

It should be emphasised that when, in the Dedication of Arithmetica inﬁnitorum, Wallis tackled the problem of interpolating series of numbers, such as 1, 6, 30, 140, 630, . . . [namely, a0 = 1, an = an−1 (4 + n2 ), n = 1, 2, 3, . . .], the absence of integral formula at that time led him to use the complicated procedures I mentioned in Chapter 1. Following Euler, one may state that the sequence 1, 6, 30, 140, 630, . . . can be interpolated by 1 , n = 0, 1, 2, . . . However, I would underline that the means of the integral 1 x−x 2 )n dx 0( translation of Wallis’s procedure by means of integral formulas strains it.

158

The Development of a More Formal Conception

a series.221 For example, Euler [1730–31a, 7–12] considered

1

xq (1 − x)n dx.

By expanding (1 − x)n , he derived 1

q

n

x (1 − x) dx =

1 0

1 ∞ ∞ n n h (−x) dx = (−1)h xh+q dx x h h 0 q

h=0

h=0

and, by a term-by-term integration, he obtained

1 0

q

n

x (1 − x) dx =

∞ n h=0

h

1

h h+q

(−1) x 0

∞ (−1)h n . dx = h q+h+1

(83)

h=0

n! for Euler veriﬁed that the right-hand side of (83) equals (q+1)(q+2)...(q+n+1) 1 q n n = 0, 1, 2, 3 and concluded that 0 x (1 − x) dx is the nth term of the series n! an = . (q + 1)(q + 2) . . . (q + n + 1)

In order to determine an integral expression of n!, Euler multiplied

1 0

xq (1 − x)n dx =

n! (q + 1)(q + 2) . . . (q + n + 1)

by (q + n + 1) to get (q + n + 1)

1

xq (1 − x)n dx =

n! . (q + 1)(q + 2) . . . (q + n)

If q = f /g, then an =

f + (n + 1)g g n+1

1

xf /g (1 − x)n dx =

n! , (84) (f + g)(f + 2g) . . . (f + ng)

which becomes 1 0n+1 for f = 1 and g = 0. 221

See Section 13.2.

1

x1/0 (1 − x)n dx = n!,

(85)

13

The Young Euler between Innovation and Tradition

159

1 1/0 1 (1 − x)n dx.222 The At this point Euler sought the value of 0n+1 0 x problem is similar to problems such as “determining the value” of fractions f (x) g(x) , when they assume the form 0/0 for a certain x, which was examined at the end of Chapter 7, p. 111. Euler’s conception did not diﬀer from the one illustrated in that chapter. Indeed, he observed that the previous result also holds if we replace x by any function f (x) [provided f (0) = 0 and f (1) = 1] in the above integrals (and obviously with df in place of dx). g For f (x) = x f +g , the left-hand side of (84) becomes f + (n + 1)g g n+1 from which

1

1

g g n x f +g 1 − x f +g dx,

n 1 − x0 dx = n!, 0n

for f = 1, g = 0. Euler interpreted

1−x0 0

1 − xz z

as

z=0

and found that it is equal to − log x, by applying the so-called l’Hˆ opital rule. Consequently, we have the formula n! =

1

(− log x)n dx,

(86)

which enables us to attribute a value to sequence an = n! for n nonintegral. Historically, (86) was the ﬁrst integral expression of the function223 Π(x) = x! =

∞

tx e−t dt.

From the modern viewpoint, however, its derivation is problematic as is the formulation of the question (the interpolation) that leads to (86) or (82). Furthermore, the ﬁnal result (86) was not viewed as a function in its own right; it was seen merely as a tool for evaluating and representing n!. In De summatione innumerabilium progressionum, Euler also veriﬁed (86) in certain particular cases and noted that (86) did not allow for an easy calculation of the value of n!. It is possible, however, to reduce the (x) The problem is similar to problems such as ”determining the value” of fractions fg(x) , when they assume the form 0/0 for a certain x, which was examined at the end of chapter 7, p.7.2. Euler’s conception did not diﬀer from ∞ the one illustrated in that chapter. 223 The factorial function Π(x) = x! = 0 tx e−t dt is related to the gamma function ∞ Γ(x) = 0 tx−1 e−t dt by the equation xΓ(x) = Π(x). 222

160

The Development of a More Formal Conception

1 calculation of 0 (− log x)n dx to the quadrature of certain algebraic curves by formulas of the type 1 (f + g)(f + 2g) . . . (f + (n + 1)g) 1 f /g x (1 − x)n dx, (− log x)n dx = n+1 g 0 0 (87) i.e., in more modern notation, f (f + g)(f + 2g) . . . (f + (n + 1)g) B + 1, n + 1 , Γ(n + 1) = g n+1 g where B(x, y) is the beta function.224 The latter formulas enable us to compute some values of n! for noninteger n, if we assume the quadrature of the given algebraic curve to be known (Euler [1730–31a, 13–14]). In De progressionibus transcendentibus, Euler found other relations involving these integrals. For instance, he wrote (87) in the form 1 n g n+1 0 (−lgx)n dx (f + ig) = 1 f (f + (n + 1)g) 0 x g (1 − x)n dx i=1 and obtained 1 h n g n+1 (h + (n + 1)k) 0 x k (1 − x)n dx (f + ig) ni=1 = , 1 f i=1 (h + ik) k n+1 (f + (n + 1)g) 0 x g (1 − x)n dx

(88)

where n is a whole number (see Euler [1730–31a, 13–15]).

13.2

Analytical and synthetical methods in series theory

In his Methodus generalis summandi progressiones [1732–33] Euler made a remarkable observation about analytical and synthetic methods. He stated: Last year I proposed a method for summing innumerable progressions,225 which not only covers the series having an algebraic sum but also provides the sum of the series dependent on the quadrature of curves, which cannot be summed algebraically. I then used a synthetic method; indeed taking any general formula I asked myself what the series could be whose sums are expressed by that formula. In this way I obtained several series, whose sums I had been able to assign. In order to make it easier and clearer to ﬁnd the sum of any proposed series, provided this can be achieved, I communicate this analytical method, which allows the discovery of the summation term by the nature of the series. (Euler [1732–33, 42]) 224 225

1 It is deﬁned by B(m + 1, n + 1) = 0 xm (1 − x)n dx. He refers to De summatione innumerabilium progressionum [1730–31b].

13

The Young Euler between Innovation and Tradition

161

The synthetic method used in De summatione innumerabilium progressionum [1730–31b] started with a determinate formula [speciﬁcally, Euler b considers an integral formula of the kind c p(n, x)dx] and derived a (ﬁb nite or inﬁnite) series whose sum term S(n, x) was equal to c p(n, x)dx b [ c p(∞, x)dx = S(∞, x) was a special case]. Euler started from the result, supposed to be entirely determined, and arrived at the summation term. In the analytic method of Methodus generalis [1732–33], to which the previous quotation refers, Euler supposed that an unknown formula f (x) was the sum of the series and, operating on f (x), tried to derive ∞ f n=0 n (x) = f (x). In both [1730–31b] and [1732–33], Euler operated on the sum f (x), deriving certain results concerning the series ∞ n=0 fn (x), but in the synthetic method the function sum was already known, guessed at in some way, while for the analytic method the sum f (x) was unknown. In [1732–33] Euler adopted Pappus’s classical terminology226 to the analysis of the inﬁnite. The synthetic method of De summatione yields a synthetic solution (in Pappus’s sense) of the converse problem [i.e., given a function f (x) ﬁnd the series whose summation term is f (x)], which, read backwards, becomes the solution to the direct problem. With regard to the theory of series, the use of Pappus’s terminology implies the following elaborate scheme: To sum Analytical procedure Synthetic procedure

To develop

AS

AD S1 S

S2 S

S1 D

S2 D

In the above scheme: • AS is the analytical method of the sum. One operates upon the indeterminate object f (x),227 satisfying the condition of being the sum of the given series ∞ n=0 fn (x) (this is the method used in [1732–33]). • AD is the analytical method of development. Given f (x), one oper ates upon the indeterminate series228 ∞ f (x), which satisﬁes the n n=0 condition of being the development of f (x), and obtains a known series which is the development of f (x).

226

Such terminology already seems residual and relative to a phase in which Euler’s formalism was not yet entirely developed. It no longer appears in his later papers about series, where the term “analytical” has a diﬀerent meaning (see Chapter 18). 227 Usually, the indeterminate object on which one operates in any application of the analytical method is denoted by a symbol. For instance, we can denote the unknown sum ∞ f (x) simply by the symbol f (x). This of n n=0 ∞fact is irrelevant to my argument. Of course, we can directly operate by means of n=0 fn (x), which is always an unknown object as long as the sum of the series is unknown. 228 Even if the series is indeterminate, its form is assumed to be of a particular kind, usually a power series.

162

The Development of a More Formal Conception ∞ • S1 S is the synthetic method of the sum. The series n=0 fn (x) is given, ∞ the sum f (x) is guessed at in some way, and one shows f (x) = n=0 fn (x). As Euler put it in [1732–33, 42], to sum a series, one needs to compare it with known formulas and to investigate whether the series is derived from one of them (this is the method used in [1730–31b]). • S1 D is the synthetic method of development. One guesses at the expansion of the given function f (x) in some way and proceeds to derive f (x) = ∞ n=0 fn (x).

• S2 S is the synthetic method of the sum. The series is given, and one derives f (x). • S2 D is the synthetic method of development. The function f (x) is f (x). given and one derives ∞ n n=0

In De summatione innumerabilium progressionum,229 Euler applied the synthetic method of the sum as follows. Since

1 0

1 1 1 1 − xn dx = 1 + + + . . . + 1−x 2 3 n

for n an integer, this integral can be used to interpolate the series 1+

1 1 1 + + ... + 2 3 n

[since it makes up its nth term] and to sum a ﬁnite number of the terms of the harmonic series (since it expresses the summation term of n n1 ). More generally, Euler denoted a function of x by P (x) and considered

k

1 − P n (x) dx = k + 1 − P (x)

k

P (x)dx +

k

P 2 (x)dx + . . . +

to be the summation term of the series k P n (x)dx. an = 0

In particular, he studied P (x) = ∞

n=1

229

See Euler [1730–31b, 26–30].

xα aa

and reduced

b(n−1)i+1 c + (n − 1)a

k

P n−1 (x)dx

13

163

The Young Euler between Innovation and Tradition

to this case. If b = 1, the latter becomes ∞

1 , c + (n − 1)a

n=1

whose summation term is expressed by

1 0

1 − xna/c dx. (1 − xa/c )c

Euler then230 generalized this result, stating, for example, that the sum term of the series ∞ c (c + (n − 1)a) (c(α + 2) + (n − 1)a) n=1

is He noted that

1

α

x dx 0

1

1 − xna/c dx. (1 − xa/c )c

1 − xna/c 1 = 1 − xa/c 1 − xa/c

if n = ∞ and (0 0) has a ﬁnite sum, then h

an

n=i+1

is inﬁnitesimal for inﬁnitely large h and i; vice versa, if is a ﬁnite quantity for appropriate h and i, then ∞

h

n=i+1 an

an

n=1

has an inﬁnite sum. It is evident that Euler had a clear idea of convergence; followhowever, ∞ ing previous mathematicians, he thought the expression n=1 an could be manipulated even if the series was not convergent. Here are two remarkable examples. The ﬁrst is taken from the same paper, De progressionibus harmonicis, where Euler expounded the above principle. It shows how Euler obtained the approximate value of the constant named today after Euler and Mascheroni by using divergent series. In De progressionibus harmonicis observationes [1734–35b, 90], Euler had found the relation i 1 j=1

j

= C + log(1 + i), for i = ∞,

(90)

where C is a constant.233 Hence, he obtained ni

i

j=1

j=1

1 1 ni + 1 log n = log = log(ni + 1) − log(i + 1) = − , for i = ∞. i+1 j j By giving particular values to n and rearranging the harmonic series, he derived 1 1 1 1 1 1 1 + − + − + − + ..., 2 3 4 5 6 7 8 1 2 1 1 2 1 1 log 3 = 1 + − + + − + + − . . . , 2 3 4 5 6 7 8 1 1 3 1 1 1 3 log 4 = 1 + + − + + + − + . . . , 2 3 4 5 6 7 8 etc. log 2 = 1 −

233

On the procedures Euler used to determine Equation (90), see Chapter 14, p. 167.

13

167

The Young Euler between Innovation and Tradition

Moreover, by expanding log(1 + x), he obtained ∞ 1 (−1)h 1 = log 1 + + j j hj h h=2

Hence, i 1 j=1

j

=

i j=1

log

1+j j

+

∞ (−1)h h=2

hj h

.

Euler freely rearranged and obtained ⎡ ⎤ i ∞ h 1 (−1) ⎣ ⎦. = log (1 + i) + j h jh

i 1 j=1

He observed that the series summed, they generate 1+

j=1

h=2

i

1 j=1 j h

“are convergent and, if approximately

1 1 1 + + . . . + = log(1 + i) + 0.577218” 2 3 i

(see Euler [1734–35b, 94]). The second example concerns Euler’s solution of the Basel problem in De summis serierum reciprocarum [1734–35a]. In this paper, Euler succeeded 1 in summing the series ∞ n=1 n2k by dealing with inﬁnite algebraic equations as if they were ﬁnite. This was simply another application of the principle of inﬁnite extension, although it came in for some criticism, as we shall see in Section 19.1. Euler observed that the sine P M = y of the arcs AM = s (see the circle in Fig. 17, where AC = 1) could be expressed as y = sin s = s −

1 1 1 3 s + s5 − s7 + . . . . 3! 5! 7!

Hence, he obtained the equation 0=1− He factored 1−

1 3 1 5 1 7 s + s − s + s − .... y 3!y 5!y 7!y

1 5 1 7 1 3 s + s − s + s − ... y 3!y 5!y 7!y

as

s 1− A1

s 1− A2

s 1− A3

...,

(91)

168

The Development of a More Formal Conception

M C

P

A

Fig. 17 where A1 , A2 , A3 are the roots of Equation (92). At this point Euler considered some well-known properties of the roots of ﬁnite equations. Indeed, it is possible to prove that if one sets S−k =

n

α−k i ,

i=o

where αi are the roots of the equation a0 + a1 x1 + a2 x2 + . . . + an xn = 0, then a0 S−1 + a1 = 0,

a0 S−2 + a1 S−1 + 2a2 = 0, ...

(92)

a0 S−n + a1 S−n+1 + . . . + an−1 S−1 + na0 = 0, ....

1 4 1 2 s + 5!y s −. . . as an inﬁnite polynomial Euler considered the series 1− 3!y

a0 + a1 x1 + a2 x2 + . . . + an xn + . . . and applied formulas (92). Hence, a1 1 = , a0 y 1 a1 S−1 + 2a2 = 2, S−2 = a0 y ....

S−1 = −

13

169

The Young Euler between Innovation and Tradition

Then he set y = 1 so that Equation (91) became 0=1−

1 1 1 s + s3 − s5 + s7 − . . . = 1 − sin s, 1 3! 5! 7!

The roots An of the equation 1 = sin s are π , 2

π − , 2

3π , 2

−

3π , 2

5π , 2

−

5π , 2

7π , 2

−

7π , 2

...;

therefore, the sum S−1 = is equal to − aa01 =

1 y

∞ ∞ 1 2 (−1)n =2 An (2n + 1)π

n=1

n=0

= 1; namely, 2

∞

(−1)n

n=0

Hence,

∞

2 = 1. (2n + 1)π

(−1)n

n=0

Similarly,

π 1 = . 2n + 1 4

∞ ∞ 1 4 (−1)n = 2 = 1, A2n (2n + 1)2 π 2

n=1

namely,

n=0

∞

1 π2 . = 8 (2n + 1)2 n=0 Since

∞ ∞ ∞ 1 1 1 + , = 2 2 n (2n + 1) (2n)2

n=1

he had

n=1

n=0

∞ ∞ 1 π2 4 1 = = . 2 2 n 3 6 (2n + 1)

n=1

n=1

This procedure allowed Euler to ﬁnd the sum of ∞

n=0

(−1)n

1

, (2n + 1)2k+1

∞

n=0

1 (2n + 1)

, and 2k

∞ 1 , for k = 1, 2, 3, . . . n2k

n=1

14

Euler’s derivation of the Euler–Maclaurin summation formula

Euler ﬁrst mentioned the summation formula in Methodus generalis summandi progressiones [1732–33] and proved it, for the ﬁrst time, in Inventio summae cuiusque seriei ex dato termino generali [1736b]. It was again called analytical, but this term had a meaning diﬀerent from that given in Methodus generalis. In Inventio summae, Euler spoke of an analytical method in contrast to the geometric method used in Methodus universalis serierum convergentium summas quam proxime inveniendi [1736a]: When I gave more precise consideration to the mode of summing which I had dealt with by using by the geometrical method in the above dissertation234 and investigated it analytically, I discovered that what I had derived geometrically could be deduced from a peculiar method for summing that I mentioned three years before in a paper235 on the sum of series. (Euler [1736b, 108]) In Methodus universalis, Euler based the determination of an approximating evaluation of the sum of a series an on a geometric representation. He wrote the series as a+b+c+d+e+etc. and denoted the nth and (n+1)st terms by x and y, respectively. He considered the diagram shown in Fig. 18, where β

a

γ

b

δ c

ε

d

ξ

e

f

ρ q

p A B

C

D

E

F

P

Q

Fig. 18 aA = a1 , bB = a2 , cC = a3 , dD = a4 , . . . , pP = an , qQ = an+1 , and AB = BC = CD = DE = . . . = P Q = 1. 234

Methodus universalis serierum convergentium summas quam proxime inveniendi [1736a]. 235 Methodus generalis summandi progressiones [1732–33].

171

172

The Development of a More Formal Conception

Since “x is a quantity composed of n and constants,” pP = x provides, according to Euler, the equation between AQ and qQ that expresses the nature of the curved line ap, i.e., the equation of the curve is x = a(n) = an . Of course, n

ai >

i=1

n+1

a(n)dn 1

or, as Euler said, sn =

n

ai >

i=1

a(n + 1)dn,

with the condition that the value of a(t)dt is equal 0 at s0 .236 In order to improve this approximation, he observed that the curvilinear triangles abβ, bcγ,. . ., pqρ, which have been neglected, are greater than the rectilinear triangles abβ, bcγ,. . ., pqρ (the curved line aq is convex, at least for large enough n). Since the sum of the areas of the rectilinear triangles abβ, bcγ,. . ., pqρ is (Aa − Qq)AB : 2, we have237 n

ai >

i=1

a(t + 1)dt +

a1 an+1 − . 2 2

Finally, Euler considered the secant bc (Fig. 19) and approximated the arc ac by an appropriate arc of a parabola. Thus, the median bm is close to the tangent to the curve, and 1 1 Sab = T1 = T2 , 3 6 where the area between the curved line and the segment ab is denoted by Sab , and the areas of triangles abm and abn are denoted by T1 and T2 . Since 236

Euler called this inequality an upper limit of series. Similarly, he derived sn =

n

ai <

i=1

with the condition that the value of 237 Analogously, he obtained n i=1

ai <

n+1 1

a(t − 1)dn (=

a(t)dt,

a(t)dt is equal 0 at the origin), i.e., a lower limit.

n+1 1

a(t − 1)dt −

a1 an+1 + . 2 2

14

Euler’s Derivation of the Euler–Maclaurin Formula

173

na = Aa − 2Bb + Cc,238 we have T2 =

a1 − 2a2 + a3 Aa − 2Bb + Cc = 2 2

and

a1 − 2a2 + a3 . 12 Of course, the sum of all the areas Sab =

Sab + Sbc + Scd + . . . + Spq , which is neglected in the last inequality, is approximately equal to a1 − a2 an+1 − an+2 − . 12 12 Therefore, Euler determined the approximating formula: n a1 an+1 a1 − a2 an+1 − an+2 − + − . ai = a(t + 1)dt + 2 2 12 12

(93)

i=1

In proving (93), Euler implicitly assumed the convexity of the arc ab and hence of the curve. Furthermore, the errors are small when n is large and an is small. He actually applied (93) to sum the ﬁrst 1,000,000 terms of the a AB=1 m n b c

A

B

C

Fig. 19 harmonic series. For n =990,000, (93) yields 1 1 1 + + ... + 11 12 1,000,000 1,000,001 1 1 + + = log 11 22 132 1 1 1 1 − − + , − 144 2,000,002 12,000,012 12,000,024 238 Indeed, in the trapezoid ACcn, if a segment Bb is drawn parallel to the sides An and . Cc, and bisecting AC, then bB = An+Cc 2

174

The Development of a More Formal Conception

and therefore 1+

1 1 1 + + ... + = 14.392669. 2 3 1,000,000

When n = ∞, (93) becomes ∞ i=1

ai =

a(t + 1)dt +

a2 7a1 − , 12 12

and Euler illustrated this in the context of the example: ∞ 1 i2 i=1

=

10 ∞ 1 1 + 2 i i2 i=1

i=11

= 1.549768 +

7 1 1 + − = 1.644920. 11 12 · 121 12 · 144

The geometrical method of Methodus universalis hinged on using appropriate geometrical ﬁgures; some steps of the deduction were inferred by scrutinizing these ﬁgures. The analytical method of Inventio summae, i.e., the summation formula, dispensed with the geometrical representation.239 If, however, it is true that the analytical method actually dispensed with ﬁgures, this does not mean that it also lacked geometrical references. Indeed, if we look carefully at Methodus universalis and Inventio summae, we note that both papers are based upon the same principles and, in eﬀect, it would be easy to translate Methodus universalis into analytical symbols. Thus, in Methodus universalis, the sequence an was viewed as a curved line whose equation is y = a(n) and was assumed to be a continuous curve (both in the sense that it makes no jumps and in Euler’s sense240 ) possessing a tangent at each point. In Inventio summae, the sequence an was viewed as a continuous and inﬁnitely diﬀerentiable function a(n) (in modern sense). Both articles substantially turn the study of the series an into the study of the function a(n) and its integral an dn. As Euler put it, “I reduced summation to integration” [1736a, 101].241 With respect to Methodus universalis and Inventio summae, Euler stated that the analytical method was superior to the geometric one not only because it included the geometrically determined formula (93) but also because it led to an improvement of the sum so that the true result could be obtained through the addition of other terms (which were unlikely to be resolved by the geometric method). The power of the analytical method of Inventio summae derives from the concept of the summation term Sx = xn=1 an as a continuous and diﬀerentiable function of the index x. 239

See Chapter 7. See Chapter 18. 241 This integral expression of the summation diﬀers from those of the type sn = f (x, n)dx examined above (see Section 13.1). 240

14

Euler’s Derivation of the Euler–Maclaurin Formula

175

This concept appeared for the ﬁrst time in De summatione innumerabilium progressionum [1730–31b, 29–30]. Euler took the summation term of the series b(n−1)i+1 an = c + (n − 1)q

to be A and said that we can consider n and A as ﬂowing quantities (quantitates ﬂuentes) when n is almost inﬁnitely greater than 1, and therefore the ni+1 diﬀerential dn is to dA as 1 is to bc+nq . He thus obtained the diﬀerential equation bni+1 dn dA = c + nq whose integral gives A as a function of n. For i = 0, Euler actually solved this equation and found A=

b log(C(c + nq)), q

C being an indeterminate constant. This conception was later developed in De progressionibus harmonicis, where Euler considered the series c a + (i − 1)b

and stated that, when i increases by 1, the sum term s of the series c increases by a+ib and therefore di = ds

c a+ib

1

c . a+ib

This equation furnishes the summation term s=C+

c log(a + ib) b

(Euler [1734–35b, 90]).242 In Methodus universalis [1736a] Euler explicitly enunciated the general principle upon which the above results are based: dn : dsn = 1 : an , 242

(94)

For a = b = c = 1, we have i 1 = C + log(1 + i) = C + log i, j j=1

for i = ∞. In this way Euler proved that ij=1 1j and log(1 + i) diﬀered for a constant, whose values he determined as described on p. 167.

176

The Development of a More Formal Conception

i.e., dsn = andn (sn is the summation term of the series an ), and he derived sn = an dn from this. He assumed the validity of (94) under the condition that n is large and the increment of sn is very small.243 In Inventio summae [1736b], Euler proved the summation formula by using the Taylor series in a decisive way. He held that a function y(x) could be expanded in Taylor series “if y is given in whatever way by means of x and constants” [1736b, 109]. xThe general term X = Xi of a series and its summation term S(x) = i=1 Xi could also be expanded in Taylor series, because “both S and X, in the case that the series is determined, are composed of x and constants” [1736b, 112]. As a consequence, he wrote S(x − 1) =

x−1

Xi = S(x) −

i=1

d3 S d4 S d2 S dS − + − ... + 2 3 1!dx 2!dx 3!dx 4!dx4

and d2 S d3 S d4 S dS − + − − ..., 1!dx 2!dx2 3!dx3 4!dx4

S(x) − S(x − 1) = X =

(95)

where (95) expressed the general term as a function of the summation term. Now, Euler wanted to derive S as a function of X. Setting244 ∞

dn X dS = an n dx dx n=0

yielded S = a0

and

Xdx +

∞

an

n=1

dn−1 X dxn−1

∞

dn+h−1 X dh S an n+h−1 . = h dx dx n=0

Replacing the last formulas in (95), Euler inferred that X =

=

∞

(−1)h+1

h=1 ∞

n=0

243 244

∞ ∞ dh S (−1)h+1 dn+h−1 X an n+h−1 = h!dxh h! dx n=0

h=1

n+1 dn X

dxn

h=1

(−1)h+1 an+1−h h!

.

In later papers Euler used (94) independently of this condition. 0 Of course, ddxX 0 = X.

(96)

14

Euler’s Derivation of the Euler–Maclaurin Formula

177

By comparing the left-hand and the far right-hand sides of (96), he derived a0 = 1, an =

n (−1)i+1 an−i i=1

(i + 1)!

and the summation formula X d5 X d7 X dX d3 X S(x) = Xdx + + − + − 3 5 2! 3!2dx 5!6dx 7!6dx 9!10dx7 691d11 X 35d13 X 3617d15 X 5d9 X − + − + . . . . (97) + 9 11 13 11!6dx 13!210dx 15!2dx 17!30dx15 The indeﬁnite integral yields a constant that is determined by the condition S(0) = 0. Immediately following the completion of Inventio summae, Euler returned to (97) in Methodus universalis series summandi ulterius promota [1736c], where he modiﬁed the summation formula (now called the universal method) in order to make the calculation easier. He [1736c, 125] asserted that diﬃculties arose from the fact that the index of the general term increases by one unit at a time. He therefore considered a series Xa+ib i

and, setting S(x) =

x i

Xa+ib , obtained S(x − b) =

∞

(−1)h

h=0

bh dh S h!dxh

and S(x) =

bdX b3 d3 X b5 d 5 X 3b7 d7 X Xdx X + + − + − b 2! 3!2dx 5!6dx3 7!6dx5 9!10dx7 9 9 5b d X 691b11 d11 X 35b13 d13 X 3617b17 d15 X + − + − + ... 11!6dx9 13!210dx11 15!2dx13 17!30dx15

with S(a) = Xa . Euler also added another term to this formula, but it is incorrect. Euler’s research into the sum formula continued throughout his life, and he applied it to numerous series. In Institutiones calculi diﬀerentialis, he explicitly linked the coeﬃcients an with the Bernoulli numbers Br 245 (already studied in his De seriebus quibusdam considerationes [1740] and related to the sum of some remarkable series) and found ∞ 1 B2n d2n−1 z S(x) = z(x)dx + z + (−1)n−1 . (98) 2 (2n)! dx2n−1 n=1

245

Bernoulli numbers are deﬁned by the relation [x] denotes the integral part of x.

t et −1

= 1+

[r/2]+1 Br r t , r=1 (−1) r!

∞

where

178

The Development of a More Formal Conception

He speciﬁed that in z(x)dx “a constant must be added so that, if setting x = 0, we shall have S = 0”. The demonstration of the Institutiones calculi diﬀerentialis is remarkable because it indicates an evolution in Euler’s thought toward a more formal conception and procedures that seem to be a prelude to the calculus of operations. In [1755, 2: Sections 167–168], Euler denoted yx by Sy [with S(0) = 0] and interpreted S as a symbolic operation that enjoys certain formal properties: 1. ﬁnite and inﬁnite additivity, which he explicitly formulated in these terms: If yx = p x + q x + r x + . . . , then Sy = Sp + Sq + Sr + . . . , that is, x

yn =

x

pn +

qn +

n=1

n=1

n=1

x

x

rn + . . . ;

n=1

2. the commutativity of the operations S and

dn dxn ,

namely246

dn dn y (Sy) = S n . n dx dx Euler set v = y(x − 1) = yx−1 = vx (x = 2, 3, . . .) and v(1) = A to get Sv =

x

vn = A +

x

n=2

n=1

yn = A + Sy − y.

By applying the additivity of the operation S to vx = y(x − 1) =

∞

n=1

(−1)n

1 dn y n! dxn

and rearranging it, he derived S

∞

dn y dy =y−A+ (−1)n S dx n!dxn

and

n=2

246

From a modern viewpoint, this formula only has meaning if we interpret S as an dn y dn integral. In this case, it corresponds to dx dx. ydx) = n( n dx

14

Euler’s Derivation of the Euler–Maclaurin Formula

Sz =

zdx − A +

dy dx

∞

(−1)n+1 S

n=1

179

dn y (n + 1)!dxn

and the condition S(0) = 0 holds]. By diﬀerentiating and [where z = applying property (2), Euler found S

∞

dh z dh−1 z (−1)n dh+n−1 z S h+n−1 , h = 0, 1, 2, . . . = + dxh dxh−1 n! dx n=2

(here

d−1 z dx−1

=

zdx). He then expressed Sz as Sz =

zdx +

∞

n=1

an

dn−1 z dxn−1

and, proceeding as above, derived (98). Later, Euler provided other proofs of the sum formula. The basic principles did not change; however, there are some signiﬁcant diﬀerences.247

247

See, for instance, Euler [1761].

15

On the sum of an asymptotic series

We have previously seen the problematic nature of asymptotic series within the framework of 18th century mathematics. It is probable that, in deriving (77), neither Stirling nor de Moivre noted the divergence of series; however, the existence of series that had an anomalous behavior with respect to formal-quantitative notion of the sum was already clear to Euler in his Consideratio progressionis cuiusdam ad circuli quadraturam inveniendam idoneae [1739c], where he pointed out some diﬃculties of the theory t dt of series. In this paper, Euler considered arctan t = 0 1+t 2 as a sum of the dt type and, using relations concerning the sums of powers of integer 1+t2 numbers, derived π=

n

∞

k=1

h=0

1 4n 1 + (−1)h B4h+2 2h + . n n2 + k 2 2 (2h + 1)n4h+2

(99)

Euler understood the nature of asymptotic series. He observed that, although (99) seems to become increasingly convergent when n increases, it actually “converges” (i.e., its terms decrease in absolute value) only up to a certain term, after which it begins to diverge248 This statement is based upon the observation that (for h > 0) (h − 1)(2h − 3) B2h+2 > , B2h 2π 2 2

B

approaches asymptotiso that, if h = ∞, this ratio is equal to πh2 (i.e., h22h+2 B2h 1 cally π2 ). Euler suggested using (99) by calculating its ﬁrst terms (according to Euler’s terminology: the terms that “converged”) and neglecting them as soon as they began to “diverge”. More speciﬁcally, the terms of (99) had to πn be considered up to the index s ≈ √ (approximately s ≈ 2n) (Euler [1739c, 2 357]). Indeed, the ratio of the successive terms of the series ∞ h=0

(−1)h B4h+2

22h (2h

1 , + 1)n4h+2 4

for large enough h, is approximately equal to − n4h 4 π 4 (if we take bh as the bh+1 4 coeﬃcient of the series, − h4 bh is asymptotic to n4 π4 ). Euler also suggested approximating the terms an of (99) when the index n is greater than s by the geometric series r ∞ P π 4 n4 P 4s4 , ≈ = 4 4 P − 4 4 16s4 n π n π + 4s4 1 + 363n r=0 4 248 For this reason, these series were termed “semiconvergent series” by Legendre in his [1811–17].

181

182

The Development of a More Formal Conception

where P is the ﬁrst omitted term. By giving the values 1, 3, and 5 to n, Euler determined some approximations of π. He compared these approximations with a known approximation of π and observed that one moved away from the truth (a veritate). In his opinion, this fact was worthy of note because there was no error in the proof and formula (99) allowed π to be approximated easily. He thought that this departure from the truth was due to the divergence of series and advised using divergent series with caution [1739c, 359-360], as Jacob Bernoulli had also done in [1689–1704]. He even gave an example that showed ∞ a contradiction deriving from the use of divergent series. Given a series n=−∞ an , Euler assumed the validity of formula (81) for any number n and wrote a0 = a1 + (a1 − a2 ) + (a1 − 2a2 + a3 ) + . . . ,

a−1 = a1 + 2(a1 − a2 ) + 3(a1 − 2a2 + a3 ) + . . . ,

(100)

a−2 = a1 + 3(a1 − a2 ) + 6(a1 − 2a2 + a3 ) + . . . , ....

By summing column by column, and using the formal relations 1 1−1 1 (1 − 1)2 1 (1 − 1)3 which are the expansion of 0

= 1 + 1 + 1 + ..., = 1 + 2 + 3 + ..., = 1 + 3 + 6 + ..., ...,

1 1 1 1−x , (1−x)2 , (1−x)3 ,

1 1 1 (a1 − 2a2 + a3 ) + . . . a1 + 2 (a1 − a2 ) + 1−1 (1 − 1) (1 − 1)3 1 1 1 + + + ... = a1 1 − 1 (1 − 1)2 (1 − 1)3 2 3 1 + + + ...... −a2 (1 − 1)2 (1 − 1)3 (1 − 1)4 1 3 6 +a3 + + + . . . . . . + ... (1 − 1)3 (1 − 1)4 (1 − 1)5 1 1 1 + a3 + ... − a2 = a1 2 (1 − 1) − 1 ((1 − 1) − 1) ((1 − 1) − 1)3 = −a1 − a2 − a3 − . . .

an =

n=−∞

Hence,

. . ., for x = 1, Euler obtained

∞

n=−∞

an = 0.

15

183

On the Sum of an Asymptotic Series

Euler noted that this result sometimes diﬀered signiﬁcantly from the correct 1 one. For instance, when n is negative, if we consider the series ∞ n=−∞ n2 , ∞ 1 we obtain n=−∞ n2 = 0, whereas the sums of inﬁnite positive numbers would be diﬀerent from 0. Euler, however, stated that “this reasoning did not always fail”; indeed, ∞

n=−∞

hn =

∞

hn +

n=−1

n=−∞

hn =

h h + = 0. 1−h h−1

Euler ended the paper by observing that the exhibition of this example was no less useful than the exhibition of rigorously proved truths (see Euler [1739c, 360–363]). In the Consideratio, Euler sensed the necessity for a deeper analysis for the concept of the sum. In eﬀect, asymptotic series were puzzling to mathematicians’ eyes. Until then mathematicians had thought that series could be used to approximate a quantity only if convergence held and that the sum of a series was its unique, true, ultimate value. Instead asymptotic series showed that certain divergent series could be used to derive numerically acceptable results, even if the sum of an asymptotic series was not the ultimate value. For instance, in his letter “on a logarithmetical mistake of some eminent mathematicians”, Bayes explicitly observed that the series 1 1 1 log x! = log 2π + x + log x − x + 2 2 12x 1 1 1 1 + − + ... − 360x3 1260x5 1680x7 1188x9 “can never properly express any quantity at all” since “the whole series can have no ultimate value whatsoever” (Euler [1763, 279–271]). However, when Bayes wrote these words, Euler had overcome the initial scepticism, as one can note his Institutiones calculi diﬀerentialis, where asymptotic series were used in an unproblematic way. In this treatise he gave several examples of asymptotic series by deriving them from the Euler– Maclaurin summation formula [1755, 351–356]. In particular, he gave a slightly modiﬁed version of (99): π=

n

∞

k=1

h=0

1 4π 1 4n (−1)h B4h+2 2h + + − 2nπ (102) 2 2 4h+2 n n +k 2 (2h + 1)n e −1

4π , which was derived from the applicaThe addition of the constant − e2nπ −1 tion of the Euler–Maclaurin summation formula, does not suﬃce to justify the numerical discrepancy originating from (99) and does not essentially change the question that arose in the Consideratio. However, in the Institutiones, Euler no longer pointed out the strangeness of the behavior of such

184

The Development of a More Formal Conception

series: of importance to him now was the formal relation. Even if (102) failed as an exact numerical equality, it retained the formal validity. In the second part of the century amd into the new century, mathematicians, following Euler, accepted the use of asymptotic series (see, e.g., Laplace [1812], Legendre [1811–17]. However, this occurred by means of a substantial shift in the conception of series. According ∞ to the formalquantitative concept of series, the relations f (x) = i=0 fi (x) or S = ∞ a had a meaning that was independent of the procedure used for i=0 i ﬁnding f (x) and S: Indeed, it was the convergence of the series ∞ i=0

fi (x)

and

∞

ai

i=0

∞ that gave a meaning to the equalities f (x) = ∞ i=0 ai i=0 fi (x) and S = . In the case of asymptotic series, these equalities could not be regarded as convergence, since the ﬁnal value was not given. This poses the following question: How could mathematicians assert that the sum of the series was precisely S and not another number S ′ ?

∞

i=0 ai

The writings of Euler, Lagrange and Laplace show clearly thatmathematicians resorted to the formal derivation of the equalities S = ∞ i=0 ai ∞ ′ ) was the sum of f (x): They thought that S (and not S and f (x) = i i=0 ∞ ′ i=0 ai because S (and not S ) was the result of a formal procedure [in the 249 sense of proposition (IE)]. In this way, formal manipulation was no longer the tool for deriving convergent series, but the one and only justiﬁcation and guarantee of the exactness of certain relations.

249

See Chapter 8, p. 117.

16

Inﬁnite products and continued fractions

The investigations concerning the interpolation of sequences and the expression and approximation of integrals led Euler to deal with inﬁnite products. In De productis ex inﬁnitis factoribus ortis [1739a, 262–264], Euler sought to determine the value of the sequence a0 = 1, ai = (f + ig)ai−1 , i = 1, 2, . . . ,

(103)

for i = 1/2, 3/2, . . .. In order to do this, he deﬁned the sequence Ai by setting A2i = ai , A1 = z, A2i+1 = (f + 2i+1 2 g)Ai , where z is to be determined. He stated that, when n is inﬁnite, the sequence becomes indistinguishable from a geometric progression and, therefore, the formula √ αi = αi−1 αi+1 , (104) which is valid for a geometrical progression αi = hi , i = 0, 1, 2, . . ., can be applied to the sequence Ai . Thus, he obtained the following approximate expressions of z :

f + g, z = (f + g)2 (f + 2g) z = , 1(f + 32 g)2 (f + g)2 (f + 2g)2 (f + 3g) z = , 1(f + 32 g)2 (f + 25 g)2 ... and assumed that the value of z is z=

(f + g)(f + 2g)(f + 3g) . . . . (f + 32 g)(f + 52 g) . . .

Then Euler considered formula (87) and changed x into y g in this formula. Since √ 1 π 1/2 , (− log x) dx = 2 0 he obtained

(2f + g)(2f + 3g)(2f + 3g)(2f + 5g)(2f + 5g) . . . (2f + 2g)(2f + 2g)(2f + 4g)(2f + 4g)(2f + 6g)(2f + 6g).. 1 2

2f 2 (2f + g) f −1 g = 1 − y dy . y πg 0 185

186

The Development of a More Formal Conception

In the De productis, Euler that if a quantity X is transformed ∞ observed pi into an inﬁnite product i=1 qi , then it can be computed by logarithms pi namely, log X = ∞ and the more rapidly this series converges, log i=1 qi

the more the factors pqii go to 1. He also stated that inﬁnite products could be used to illustrate the nature of the quantity X [1739a, 261]. In eﬀect, Euler used inﬁnite (or ﬁnite) products as an intermediate step to derive various relations between integrals. For instance, he found 1 a+g−1 1 x xa−1 √ √ dx dx π = 2ag 1 − x2g 1 − x2g 0 0 and =

1 0 1 0

1 xa−1 xa+mb−1 dx dx b 1−n (1 − xb )1−m 0 (1 − x ) 1 xa−1 xa+nb−1 dx dx b 1−m (1 − xb )1−n 0 (1 − x )

(105)

(see Euler [1739a, 266 and 288]). These are de facto results concerning the beta function, which 1 was, however, not recognized as such, namely an integral of the type 0 xm (1 − x)n dx was not recognized as a new function. Moreover, in De productis, the above integrals are still viewed as the expression of geometrical quantities (areas). It is worth noting that Euler was often interested in determining convergent products so that they could be computed eﬀectively. However, even in this case, inﬁnite products were used without an a priori consideration of their convergence. Furthermore, he also used these products as formal instruments for deriving relations such as (105) and, in these cases, the convergence of products seems to be of no importance. ∗ ∗

∗

Euler systematically investigated continued fractions from the 1730s onwards. In De fractionibus continuis dissertatio, he observed that this species of inﬁnite expressions was not as usual as the other two types (series and inﬁnite products),250 although they were highly appropriate for expressing the values of certain quantities and for calculating them approximately [1737a, 188]. In De fractionibus continuis dissertatio [1737a, 189–196] and De fractionibus continuis observationes [1739b, 292–293], Euler showed that the continued fraction b1 b2 b3 ... C = a1 + a2 + a3 + a4 + 250 In practice, after the initial results —described in Chapter 1— mathematicians showed little interest in continued fractions.

16

187

Inﬁnite Products and Continued Fractions

can be transformed into C = C1 +

∞

ΔCk = a1 +

pk qk

k+1

(−1)

k=1

k=1

where Ck =

∞

k

i=1 bi , qk qk+1

and pk and qk are deﬁned for all k > 1 by the recursive rule pk = ak pk−1 + bk−1 pk−2 qk = ak qk−1 + bk−1 qk−2

with p1 = a1 , p0 = 1, q1 = 1, q0 = 0, and ΔCk = Ck+1 − Ck . Vice versa, given the series ∞

(−1)k+1

k=1

Euler obtained a1 +

k

i=1 bi , qk qk+1

b1 b2 b3 ... a2 + a3 + a4 +

by setting ai =

qi − ai−1 qi−2 . qi−1

In De fractionibus continuis dissertatio and De fractionibus continuis observationes, Euler paid attention to the convergence of the inﬁnite processes he used: Indeed, he mainly aimed to ﬁnd the value of certain integrals and convergence was essential. However, the link between series and continued fractions was viewed in a merely formal way, independently of the convergence of the series and fractions. A series and a continued fraction were considered to be equal simply if they were derived from each other by the formal transformation a1 +

k ∞ bi b1 b2 b3 (−1)k+1 i=1 . . . . = a1 + a2 + a3 + a4 + qk qk+1

(106)

k=1

For instance, in [1739b, 296], by using the binomial expansion and transformation (106), he derived

1 0

xn−1 (1 + xm )

h v

dx =

∞

1 h(h + v) . . . (h + (k − 1)v) + (−1)k n k!v k (km + n) k=1

hn2 v(h + v)(m + n)2 1 .... = n+ vm + (v − h)n+ (3v − h)m + (v − h)n

188

The Development of a More Formal Conception

Euler observed that if v = 1 and h is an integer, one had the fraction hn2 (h + 1)(m + n)2 1 ..., n+ m + (1 − h)n+ (3 − h)m + (1 − h)n whose series expression was not convergent. However, this was not considered to be a good reason to reject the general validity of equality (106). If one assumed (106) to be valid independently of the convergence of the series and of the continued fraction, then it was very natural to use (106) to give a value to divergent series. During the 1740s, neither Euler nor other mathematicians tackled this question. Subsequently, however, Euler drew the logical conclusions from this.251 In his [1784, 41], for instance, he regarded the transformation of divergent series into fractions as a secure way (and perhaps the only way) of ﬁnding the approximate value of a divergent series. Some examples are appropriate to clarify the use of continued fractions. In De fractionibus continuis observationes [1739b, 302–306], after haved interpolating the sequence (103), Euler argued that if the sequence Ak =

2k

ai

i=1

was given, then A1/2 was equal to a1 , A3/2 was equal to a1 a2 a3 , . . . . He applied this idea to the series ∞ k=1

Ak =

∞ k=1

p(p + 2r) . . . (p + 2(k − 1)r . (p + 2q)(p + 2q + 2r) . . . (p + 2q + 2((k − 1)r)

He took p , p + 2q p + 2r = , p + 2q + 2r ... p + 2(k − 1)r , = p + 2q + 2(k − 1)r ...

a1 a2 = a3 a4 a2k−1 a2k

and ak =

251

See Euler [1754–55] and [1784].

bk , p + 2q + 2(k − 1)r

(107)

16

189

Inﬁnite Products and Continued Fractions

where the numbers bk are to be determined appropriately, and found P p(p + 2q − r)(P + Q) (108) m+ 2rR+ (p + r)(p + 2q)P (P + 2Q) (p + 2r)(p + 2q + r)(P + Q)(P + 3Q) , 2r(R + S)+ 2r(R + 2S)+

b1 = m − r +

where m is an arbitrary quantity and P

= p2 + 2pq − pr − m2 + mr,

Q = 2r(p + q − m),

R = p2 + 2pq − mp − mq + qr, S = pr + qr − mr.

At this point Euler applied (88). Setting f + g = p, h + k = p + 2q, g = k = 2r, and substituting n with n − 1, he deduced 1 p+2q−2r x 2r (1 − x)n−1 dx p(p + 2r) . . . (p + 2(n − 1)r , = 0 1 p−2r (p + 2q)(p + 2q + 2r) . . . (p + 2q + 2((n − 1)r) 2r (1 − x)n−1 dx x 0

i.e., in more modern notation,

p(p + 2r) . . . (p + 2(n − 1)r = (p + 2q)(p + 2q + 2r) . . . (p + 2q + 2((n − 1)r)

p+2q 2r , n p B 2r , n

B

,

where B(x, y) is the beta function. It follows, through a simple substitution in the integral, that the general term of (107) is 1

y p+2q−1 (1 − y 2r )n−1 dy . 1 p−1 (1 − y 2r )n−1 dy 0 y

Hence,

1

y p+2q−1 (1 − y 2r )−1 dy . b1 = A 1 (p + 2q + r) = (p + 2q − r) 1

p−1 (1 − y 2r )−1 dy 2 0 y 0

(109)

Therefore, the continued fraction (108) is expressed by the ratio of two integrals, and the crucial step for arriving at (109) is the series (107).

190

The Development of a More Formal Conception More generally, Euler obtained252 2r2 + pr − q 2 pr − q 2 2(p + q + (k − 2)r)+ 2(p + q + (k − 2)r)+ 12r2 + pr − q 2 6r2 + pr − q 2 ... 2(p + q + (k − 2)r)+ 2(p + q + (k − 2)r)+

1 p+2q+(k−1)r−1 (1 − y 2r )−1 dy y = (p + 2q + (k − 2)r) 0 1 .

p+k(−1)r−1 (1 − y 2r )−1 dy 0 y

bk = p + q + (k − 2)r +

It should be noted that series were merely conceived as a mere instrument relating continued fractions and integrals. It was thus of no importance whether series on their own had a meaning as quantities (i.e., they were convergent). In [1739b, 324–325], Euler also considered the functions P and R satisfying the recursive relation (a + nα)

1

n

P (x)R (x)dx = (b + nβ)

1

P (x)Rn+1 (x)dx

(110)

+(c + nγ)

1

P (x)Rn+2 (x)dx;

for n = 0, 1, 2, . . ., he showed that 1 0

P (x)R(x)dx a (a + α)c (a + 2c)(c + γ) .... = 1 b+ b + β+ b + 2β+ 0 P (x)dx

(111)

The proof consisted in dividing the relation (110) by

1

P (x)Rn+1 (x)dx,

n = 0, 1, 2, . . .

to obtain the terms of the fraction step by step. Various interesting expansions follow from (111). For instance, 1

xex dx 1 2 3 1 = .... = 0 1 x 1 − e 1+ 2+ 3+ 0 e dx

(see Euler [1739b, 343–344]). Another example concerns the use of continued fractions to solve diﬀerential equations. In [1739a, 345–347], Euler considered the Riccati equation axm dx + bxm−1 ydx + cy 2 dx + dy = 0 252

See Euler [1739b, 319–322].

(112)

16

191

Inﬁnite Products and Continued Fractions

and sought a solution of the like kind y=

1 1 , − cx zx2

where z is to be determined. By replacing y = xm+3 = t, he obtained the equation

1 cx

− zx1 2 in (112) and setting

m+4 1 −c − m+3 b ac + b 2 t t m+3 zdt − z dt + dz = 0, dt − m+3 m+3 (m + 3)c

which is of the same type as the previous. The reasoning can be repeated. He set z = 2m+5

t m+3

1 −(m + 3)c − 2, (ac + b)t vt

= u

and replaced into the last equation and thus obtained a new equation of the same type. And so on. Euler stated the equation had the solution a1 x−1 +

1 1 1 .... −a2 x−1 + a3 x−1 + −a4 x−1 +

where a1 = c

−1

ac + b −1 , a2 = − , etc. (m + 3)c

He preferred to write the solution in the form cxy = 1 +

(ac + b)xm+2 (ac − (m + 2)b)xm+2 (ac + (m + 3)b)xm+2 .... −(m + 3)+ (2m + 5)+ −(3m + 7)+

He also noted that the solution satisﬁed the condition cxy = 1 for x = 0 if m + 2 > 0, and the condition cxy = 1 for x = ∞ if m + 2 < 0. The solution of a diﬀerential equation by continued fractions highlighted that continued fractions could be connected with totally divergent series (series with a radius of convergence equal to 0).253 For instance, in De seriebus divergentibus [1754–55b, 615], Euler observed that the solution of the diﬀerential equations xm dx = xq+1 dy + (p − m)xq ydx + ydx was given by y=

253

xm pxq qxq p(p + q)xq 2qxq p(p + 2q)xq 3qxq p(p + 3q)xq ... 1+ 1+ 1+ 1+ 1+ 1+ 1+ 1+

On series solutions to diﬀerential equations, see Chapter 26.

192

The Development of a More Formal Conception

under the condition y(0) = 0. However, the solution can also given by the integral x 1 dx − 1 q m−q qx e qxq p−q−1 e x x 0 and by the series

y = xm − pxm+q + p(m + q)xm+2q − p(p + q)(p + 2q)xm+3q + . . . . (113) which, in general, has the radius of convergence equal to zero. These three solutions were conceived of as equivalent and it seemed natural to assume the value of a continued fraction is the value of the series. Indeed, setting p = m = 1, q = 2, x = 1, Euler found 1−1+1·3−1·3·5+1·3·5·7−... =

1 1 2 3 4 5 . . . = 0.65568; 1+ 1+ 1+ 1+ 1+ 1+

hence, he assumed that the sum of 1 − 1 + 1 · 3 − 1 · 3 · 5 + 1 · 3 · 5 · 7 − . . . was 0.65568. I stress that this type of power series (totally divergent power series) was diﬀerent from the type of power series upon which the formal-quantitative use of series was based. As we saw in Chapter 8, the formal-quantitative concept of series meant that a power series always had a positive radius of convergence and had a meaning that was independent of the formal transformation by which the series was derived: They represented quantities independently of the modality of derivation (I shall refer to these series as ordinary power series). Instead, totally divergent series have no quantitative meaning by themselves: Their meaning derived only by a formal transformation that linked them to a diﬀerential equation and a continued fraction or another expression of quantity.

17

Series and number theory

In Variae observationes circa series inﬁnitas, Euler dealt with series that he argued were diﬀerent from other series. In his opinion, mathematicians had hitherto mainly considered either series whose general term was given (by an analytical expression) or series whose terms could be computed by a recurrence formula. Instead, in Variae observationes, he considered series where neither their general term (in Euler’s sense254 ) nor the rule of the continuation of terms was known, but whose nature was determined by means of other conditions. Moreover, their sum had to be computed by methods that diﬀered from the customary ones, which required knowledge of the general term or the rule of continuation [1737b, 217]. Euler refers to series such as 1 1 1 1 1 + + + + + ... 15 63 80 255 624 and

1 1 1 1 1 1 + + + + + + .... 2 3 5 7 11 13

The general term of the ﬁrst series is the fraction a1 “whose denominators are one less all perfect squares which simultaneously are other powers” (Euler [1737b, 226]). The general term of the second series is p1 , where p is a prime number. I note that the series are, in reality, given by the general term, although these general terms do not satisfy the Eulerian deﬁnition by which the general term an was given by one single formula composed of the commonly used analytical symbols. This prevented Euler from investigating such numerical series according to his usual procedures, which consisted of reducing the treatment of a numerical series to that of an appropriate function series, even transforming an into a function a(n) of a continuous variable. The ﬁrst series that Euler summed in Variae observationes was 1 1 1 1 1 + + + + ... = . n 3 7 8 15 m −1 m,n>1

He used a sophisticated application of Leibniz’s method described on p. 30. He set 1 1 1 1 (114) x = 1 + + + + + ... 2 3 4 5 and subtracted 1 1 1 1 + ... 1= + + + 2 4 8 16 254

See Section 13.1, p. 157.

193

194

The Development of a More Formal Conception

from (114). Hence, x−1=1+ Then he subtracted

1 1 1 1 1 + + + + + ... 3 5 6 7 9

(115)

1 1 1 1 = + 2 + 3 + ... 2 3 3 3

from (115) and obtained x−1−

1 1 1 1 1 =1+ + + + + .... 2 5 6 7 10

(116)

He repeated the reasoning. By subtracting 1 1 1 1 = + 2 + 3 + ... 4 5 5 5 from (116), he obtained x−1−

1 1 1 1 1 − =1+ + + + .... 2 4 6 7 10

Euler stated that, in a similar way, all the remaining terms could be elimi1 nated, leaving x − 1 − 12 − 14 − 15 − 61 − 19 − 10 − . . . = 1 or x−1=1+

1 1 1 1 1 1 + + + + + + .... 2 4 5 6 9 10

Finally, he subtracted this series from (114) and obtained 1 1 1 1 1 + + + + + . . . = 1. 3 7 8 15 24 The summation of 1 1 1 1 1 + + + + + ... 15 63 80 255 624 is similar but a little less hard to understand for the modern reader since ∞ 1 π2 1 the starting point is ∞ rather than = n=1 n = x. Euler then n=1 n2 6 considered the geometrical series n2

1 1 1 1 = 2 + 2 2 + 2 3 + ... −1 n (n ) (n )

and observed that 1 1 1 1 + + + + ... 4 9 16 25 1 1 1 1 1 1 1 1 + + + ... + + + + ... + + +. . . + . . . = 1+ 4 16 64 9 81 729 25 625 1 1 1 1 = 1+ + + + + .... 3 8 24 35 1+

17

195

Series and Number Theory

1 15

The series

1 63

+

1+

+

1 80

∞

n2

n=2

+

1 255

+

1 624

+ . . . is the diﬀerence between

1 1 1 1 1 1 =1+ + + + + + ... −1 3 8 15 24 35

and ∞ 1 1 1 1 1 1 1 1 =1+ + + + + ... = 1 + + + + .... 2 n 4 9 16 25 3 8 24

n=1

Since

∞

n=2

he obtained

1 4 = , n2 − 1 3

1 1 1 1 1 7 π2 + + + + + ... = − . 15 63 80 255 624 4 6 Then Euler derived various results concerning prime numbers. For example, 1+

2 · 3 · 5 · 7 · 11 · 13 · 17 · 19 · . . . 1 1 1 1 + + + + ... = , 2 3 4 5 1 · 2 · 4 · 6 · 10 · 12 · 16 · 18 · . . .

(117)

where the numerator is the product of all the primes and the denominator is the product of all numbers one less than the primes [1737b, 228], and 3 · 5 · 7 · 11 · 13 · 17 · 19 · . . . π = , 4 4 · 4 · 8 · 12 · 12 · 16 · 20 · . . .

where the denominator is the product of the numbers one greater or less than the corresponding numerator [1737b, 233]. In Variae observationes [1737b, 242–244], Euler also proved that the 1 1 + 13 + . . . is inﬁnite. sum of the reciprocal of all the primes 21 + 13 + 51 + 17 + 11 He set 1 1 1 1 1 + + + + + ..., 2 3 5 7 11 1 1 1 1 1 + + + + + ..., B = 22 32 52 72 112 1 1 1 1 1 C = + + + + + ..., 23 33 53 73 113 .... A =

(118)

Euler summed (118) column by column after multiplying the ﬁrst row by 1, the second by 12 , the third by 31 , the fourth by 14 , etc. Since log(1 − x)−1 = x +

x2 x3 x4 + + + ..., 2 3 4

196

The Development of a More Formal Conception

he obtained 1 1 2 3 5 7 1 A + B + C + D + . . . = log + log + log + log + . . . . 2 3 4 1 2 4 6 Hence, 1

1

1

eA+ 2 B+ 3 C+ 4 D+... = Euler applied (117) and wrote 1

1

2 · 3 · 5 · 7 · 11 · 13 · 17 · 19 · . . . . 1 · 2 · 4 · 6 · 10 · 12 · 16 · 18 · . . .

1

eA+ 2 B+ 3 C+ 4 D+... = 1 +

1 1 1 1 + + + + .... 2 3 4 5

(119)

Since 21 B + 13 C + 41 D + . . . has a ﬁnite value255 and 1

1

1

eA+ 2 B+ 3 C+ 4 D+... = ∞, it follows that A=

1 1 1 1 1 1 + + + + + + . . . = ∞. 2 3 5 7 11 13

Moreover, 12 B + 31 C + 41 D + . . . can be neglected with respect to A = ∞; therefore, (119) can be written as eA = 1 +

1 1 1 1 + + + + .... 2 3 4 5

By passing to logarithms, 1 1 1 1 1 1 + + + + + + ... 2 3 5 7 11 13 1 1 1 1 = log 1 + + + + + . . . = log log ∞. 2 3 4 5

A =

Euler concluded that 1 1 1 1 1 1 + + + + + + ... 2 3 5 7 11 13 is inﬁnite but is inﬁnitely less than 1+

1 1 1 1 + + + + .... 2 3 4 5

255 ∞ ∞ ∞ 1 1 1 1 n+1 1 log B + C + D + ... < − = log(n + 1) − = C + 1, 2 3 4 n k k n=1 k=2

where C is the Euler–Mascheroni constant.

k=2

17

197

Series and Number Theory

Some years later Euler used series in tackling the problem of the partitions of numbers; this consists of determining the number of ways of writing the integer n as a sum of a ﬁxed number ν of addends or any number of addends (the addends, belonging to a ﬁxed set of positive integers, can all diﬀer or not, as the case may be, but their order is not considered significant). The technique used in this case diﬀered considerably from that of Variae observationes circa series inﬁnitas. He now sought the power series whose coeﬃcients provided the numbers of partitions (in modern terms, he was trying to ﬁnd the generating function, though in this case the term “function” is not quite appropriate for Euler). Here, Euler made a purely combinatorial use of series; they were needed for counting objects and he could disregard convergence completely. In his Introductio in analysin inﬁnitorum 256 [1748a, 313–337], given a sequence nj of integers, Euler sets 1+

∞

Pi z i =

∞

(1 + xnj z).

(120)

j=1

i=1

By the method of indeterminate coeﬃcients, he obtains xnj , P1 = ∞ j=1 ∞ ∞ P2 = j=1 k>j xnj +nk , ∞ ∞ nj +n +nr k P3 = ∞ , j=1 k>j r>k x .... It is clear that Pi is the sum of the inﬁnite numbers xα , where α is equal to the sum of i diﬀerent terms of the sequence ns . Since some sums nj + nk + nr + . . . might be equal, Euler sets Pv =

∞

Nv,mj xmj ,

j=1

where Nv,mj gives the number of ways of writing the integer mj as the sum of v diﬀerent addends chosen between the terms of the sequence nj . If one now denotes the set of the integers mj (which are equal to the sum of i diﬀerent terms of sequences ns ) by M , one can write Pv =

∞

Nv,m xm .

m∈M

Then the number of ways of writing the integer m as the sum of v diﬀerent addends chosen between the terms of the sequence nj can be calculated by 256

See also Euler [1750–51d].

198

The Development of a More Formal Conception

determining the coeﬃcients of xm z n in the series ∞ ∞ ∞ m zn = 1+ (1 + xnj z). Nν,m x n=1

(121)

j=1

m∈M

As an example, Euler considered the sequence of natural numbers 1, 2, 3, 4, . . .. In this case, formula (121) can be written as (1 + xz)(1 + x2 z)(1 + x3 z)(1 + x4 z)(1 + x5 z) . . . . (122) = 1 + z x + x2 + x3 + x4 + x5 + x6 + x7 + x8 + . . . + z 2 x3 + x4 + 2x5 + 2x6 + 3x7 + 3x8 + 4x9 + 4x10 + . . . + z 3 x6 + x7 + 2x8 + 3x9 + 4x10 + 5x11 + 7x12 + 8x13 + . . . + z 4 x10 + x11 + 2x12 + 3x13 + 5x14 + 6x15 + 9x16 + 11x17 + . . . + z 5 x15 + x16 + 2x17 + 3x18 + 5x19 + 7x20 + 10x21 + 13x22 + . . . + z 6 x21 + x22 + 2x23 + 3x24 + 5x25 + 7x26 + 11x27 + 14x28 + . . . + z 7 x28 + x29 + 2x30 + 3x31 + 5x32 + 7x33 + 11x34 + 15x35 + . . . ....

He thus found that the number of ways of writing the number 35 as the sum of 7 addends taken from the sequence of the natural number 1, 2, 3, 4, . . . is 15, since 15 is the coeﬃcient of z 7 x35 . If we set z = 1 and rearrange the series (121), we have 1+

∞

m

Rm x

=

∞

(1 + xnj ),

j=1

m∈M

where Rm = N1,m + N2,m + . . . . The coeﬃcient Rm gives the numbers of ways of writing the integer m as the sum of any number of diﬀerent addends belonging to the sequence nj . For example, for z = 1, we have (1 + x)(1 + x2 )(1 + x3 )(1 + x4 )(1 + x5 ) . . . . = 1 + x + x2 + 2x3 + 2x4 + 3x5 + 4x6 + 5x7 + 6x8 + . . . . We thus found that the number of ways of writing the number 8 as the sum of any number of diﬀerent addends taken from the sequence 1, 2, 3, 4, 5, . . . is 6. Euler continued with the case whereby the addends can equal each other. Given a sequence nj of integers, he considered 1 (1 + xnj z) j=1

∞

17

199

Series and Number Theory

and expanded this product into a power series 1+

∞ ∞

Pn,m xm z n ,

n=1 m∈M

where M is the set of the integers that are equal to the sum of i terms of sequences nj . Reasoning in the way described above, Euler showed that the number of ways of writing an appropriate integer mj as the sum of v addends (not necessarily diﬀerent), chosen from the terms of the sequence calculated by determining the coeﬃcients of xm z n of the series nj , could ∞ ∞ be 1 + n=1 m∈M Pn,m xm z n . In particular, when nj = j, j = 1, 2, 3, . . . he obtained 1 (1 + xz)(1 + x2 z)(1 + x3 z)(1 + x4 z)(1 + x5 z) . . . . = 1 + z x + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + . . . +z 2 x2 + x3 + 2x4 + 2x5 + 3x6 + 3x7 + 4x8 + . . . +z 3 x3 + x4 + 2x5 + 3x6 + 4x7 + 5x8 + 7x9 + . . . . . . +z 4 x4 + x5 + 2x6 + 3x7 + 5x8 + 6x9 + 9x10 + . . . +z 5 x5 + x6 + 2x7 + 3x8 + 5x9 + 7x10 + 10x11 + . . . ...

For example, he found that the number of ways of writing the number 11 as the sum of 5 integer positive numbers is 10. By setting z = 1, Euler obtained 1 (1 + x)(1 + x2 )(1 + x3 )(1 + x4 )(1 + x5 ) . . . = 1 + x + 2x2 + 3x3 + 5x4 + 7x5 + 11x6 + 15x7 + 22x8 + . . . The coeﬃcients of the latter series (which today is considered as deﬁning the partition function) give the number of ways of writing an integer n as the sum of any number of (both diﬀerent or equal) positive integers. For instance, 6 can be written in 11 ways as the sum of positive integers.

18

Analysis after the 1740s

In the previous chapters of this second part, I described the growth of the theory of series from the 1720s to the 1750s. However, this evolution was part of a more general change in analysis which, during the 18th century, became an autonomous discipline, independent of geometry and arithmetic.257 This change matured in the 1730s and 1740s and was made manifest by the publication of Euler’s Introductio in analysin inﬁnitorum in 1748. In this chapter, I shall discuss the basic principles of the 18th-century concept of analysis, which lasted through to the ﬁrst decades of the 19th century. The success and decline of the formal theory of series would be unintelligible if it was not considered within this context.

18.1

Eighteenth-century analysis as nonﬁgural and symbolic investigation of the real

In the preface to the Institutiones calculi diﬀerentialis, Euler made two remarkable observations about the nature of the diﬀerential calculus. First, he explicitly rejected geometrical conﬁrmation as a means of testing the validity of the calculus, namely, he refused to accept proofs of the calculus’ correctness based solely on the fact that the calculus reached the same conclusions as elementary geometry: The calculus cannot have its own foundation in a geometrical reference [1755, 6]. He then observed: I mention nothing of the use of this calculus in the geometry of curved lines: that will be least felt, since this part has been investigated so comprehensively that even the ﬁrst principles of the diﬀerential calculus are, so to speak, derived from geometry and, as soon as they had been suﬃciently developed, were applied with extreme care to this science. Here, instead, everything is contained within the limits of pure analysis so that no ﬁgure is necessary to explain the rules of this calculus. (Euler [1755, 9; my emphasis]) Similar statements can be found in Lagrange’s writings. Indeed, in 1773, he wrote: I hope that the solutions I shall give will interest geometers both in terms of the methods and the results. These solutions are purely analytical and can be understood without ﬁgures. (Lagrange [1773, 661]) And, in his Trait´e de m´ecanique analytique, he stated: 257

On 18th-century analysis, see Fraser [1989] and [1997].

201

202

The Development of a More Formal Conception One will ﬁnd no ﬁgures in this work. The methods that I present require neither constructions nor geometrical or mechanical reasonings, but only algebraic operations, subject to a regular and uniform course. Those who admire analysis will with pleasure see mechanics become a new branch of it and will be grateful to me for having extended its domain (Lagrange [1788, 2]).

The insistence on ﬁgures can be easily understood if one thinks of the role that ﬁgures played in geometry (I refer to Chapter 7).258 In eﬀect, when Euler and Lagrange claimed that ﬁgures were absent from their treatises, they were claiming the absence of inference derived by the mere inspection of a ﬁgure and therefore the independence of analysis from geometry, understood as a ﬁgural study of curves. This gives rise to a crucial question: What basic principles and instruments were used by 18th-century analysts to make analysis truly independent of geometry? To answer this question, I shall begin by observing that d’Alembert considered the principles of analysis to be “based upon merely intellectual notions, upon ideas that we ourselves shaped by abstraction, by simplifying and generalising the ‘ﬁrst’ ideas”.259 In other terms, analysis was considered as a system of merely intellectual notions, where the term “intellectual” referred to a form of knowledge that was not based on material awareness but was conceptual and mediated; it functioned in a discursive way along abstract notions. Whereas geometry was entrusted, to a certain extent, to the intuitive immediacy of an inspection of the ﬁgure and the perception of the relationships shown in the diagram, analysis was understood as a conceptual system where deduction was merely linguistic and mediated, or to put it another way, proceeded from one proposition to another discursively. Eighteenth-century analysis was not simply the linear continuation of Leibniz’s or Newton’s analysis but was based on a new way of doing mathematics. This new concept of analysis is undoubtedly closer to modern analysis than the previous one, even though it presents some aspects that signiﬁcantly distinguish it from the modern concept. One of these aspects was the very notion of mathematical theory, which I shall examine in the remainder of this section. ∗ ∗

∗

In Chapter 7, we already saw that the decisive aspect of analytical symbolism was the fact that signs were the concrete objects of a calculation, 258

In Chapter 14, we also saw that this conception of the function of ﬁgures was still true for Euler. 259 See d’Alembert [1773, 5:154]. By contrast, geometry and mechanics were “material and sensible” science; in particular, geometry was “the science of the properties of extension as it is considered as merely extended and ﬁgured” (d’Alembert [1773, 5:158]).

18

Analysis after the 1740s

203

namely of a manipulation performed according to certain rules, independently of the meaning of the symbols (syntactically, in modern terms). Of course, analysis cannot be reduced to the mechanical or blind manipulation of letters. It is not only a matter of the inventiveness necessary to derive formulas that are not reduced to a simple exercise; rather, the point is that doing mathematics does not merely consist of deriving formulas but of deriving formulas that have an interest or a sense in a certain context. This is also true for modern formal theories. A theorem T of a formal theory is the last proposition of a sequence of propositions Pi , i = 1, . . . , n, where Pn = T and Pi , i = 1, . . . , n − 1, is an axiom or is deduced by a rule of inference from the preceding propositions. While all derivable propositions in the given theory are theorems in this sense, in mathematical praxis, only some propositions (signiﬁcant for whatever reason) are theorems. The decision that Pn is a theorem, while Pn−1 is not, is not part of the formal structure of theory. However, the goal of a formal theory is to yield theorems in this more restricted sense.260 I would argue that the nature of analytical or algebraic derivations is necessarily syntactical and, as such, one handles signs associated with certain rules regardless of the meaning of the objects of calculation; however, the syntactical rules that govern analytical signs must make sense for the mathematician and must yield results that make sense or have some interest. Eighteenth-century analysis was symbolical in the sense that it dealt with quantities that were reiﬁed into concrete signs and were manipulated according to certain ﬁxed transformations. However, the way in which the syntactical structure was constructed diﬀered profoundly from the way it is conceived today. Today the rules261 used in a theory are explicit axioms, which in principle are freely chosen, or, to use a widely employed term, arbitrary.262 Within the limits of the given system of axioms, mathematical objects can freely be created by arbitrary deﬁnitions. In this way, the development of a theory is entirely syntactical and it is possible to make a distinction between syntactical correctness and semantic truth. This is not the case for 18th-century mathematicians. The idea of the free creation of mathematical objects was lacking. Mathematical objects did not exist in virtue of implicit or explicit deﬁnitions. They were always connected 260

On this, see Panza [1997, 366–367]. It is clear that by “rules of manipulation”, I do not intend rules of inference, but rules of the type ab = ba, which in modern formal theory are axioms (or theorems derived from axioms). 262 Here freedom and arbitrariness do not mean that one chooses the system of axioms and gives deﬁnitions without reason; rather, it means that axioms and deﬁnitions are ﬁxed by an act of will determined by the targets that one wants to achieve, without other restraint to the achievement of such targets. Axioms and deﬁnitions have no intrinsic necessity, neither do they consist of a description of physical or geometrical reality; however, they must have the capability of representing certain concepts adequately. 261

204

The Development of a More Formal Conception

with reality, directly or indirectly.263 The rules of manipulation were not arbitrary: They were derived from the notion of quantity and expressed properties of quantities (or of numbers). For instance, a + b = b + a was not an arbitrary axiom associated with the operation + (which we may or may not choose, according to the objectives of our theory); it was a mere consequence of the concept of joining two quantities. A system of explicit axioms in the modern sense and an accurate construction of certain mathematical objects (e.g., the construction of the different species of numbers) were lacking. In their place, 18th-century mathematics admitted the reference to the intuitive knowledge of the mathematical notions drawn from pre-mathematical experiences. This depended on the 18th-century concept of mathematics as a “science of nature”.264 Analysis was considered as a mirror of reality; its objects were idealizations derived from the physical world and had an intrinsic existence before and independently of their deﬁnition. Mathematical propositions were not merely hypothetical but concerned reality, and were true or false accordingly to whether or not they corresponded to the facts. For instance, d’Alembert stated: “The physicist ignorant of mathematics considers the truths of geometry as if they were grounded upon arbitrary hypotheses and as mere whims (jeux d’esprit) that entirely lack any applications.”265 This led to a lack of distinction between syntax and semantics and to the impossibility of distinguishing a syntactically correct theory from semantically true theory. Today, stating that a proposition “p” of the mathematical theory T is syntactically correct is not the same as saying that it is semantically true. The truth can be predicated of “p” if and only if we specify what universes of objects constitute the models of the theory T . In this case, we say that “p” is true if the event p occurs in the model M where T has been interpreted. Given the theory L1 containing the statement p and the theory L2 containing the statement non-p, if one asks: “May L1 and L2 be correct simultaneously?”, we today answer that L1 and L1 can be syntactically correct at the same time and, even, both true provided they are interpreted by two diﬀerent models. In the 18th century, mathematicians thought that a theory was acceptable only if it conformed to the reality. Since reality is unique, two alternative theories based on alternative deﬁnitions of certain notions (e.g., the sum of a series and limit of a sequence) or diﬀerent axioms could not be correct simultaneously.

263 If no intuitive interpretation of them was known —e.g., imaginary numbers—, they were viewed as tools for improving the analytical theory of quantity, in the same manner as the sign 0 improves the notation of natural numbers that counts objects even though it denotes no object. 264 See, for example, the preliminary discourse to the Encyclop´edie of d’Alembert. 265 See d’Alembert [1773, 5:121].

18

Analysis after the 1740s

18.2

205

Functions, relations, and analytical expressions

The transformation of analysis into a system based on linguistic deduction was made possible by the notion of function.266 By “function” Leibniz initially denoted a line that performs a special duty in a given ﬁgure (Youschkevitch [1976, 56]). Later, Leibniz used this term to denote a part of a straight line that is cut oﬀ by straight lines drawn solely by means of a ﬁxed point and points of a given curve.267 Therefore, functions were merely geometric variables. The calculus, however, expressed geometric quantities analytically (by “indeterminates and constants”) and, already during the ﬁrst decades of the calculus, mathematicians felt the need to give a name to such analytical expressions of geometric quantities. Thus, while investigating the isoperimetric problem that consists of minimizing the area enclosed by a curve, Johann Bernoulli termed them “functions”, with Leibniz’s agreement (see Leibniz [GMS, 3:506–507 and 526]).268 However, it was only as a result of Euler’s work that the notion of a function assumed a crucial role in mathematics. According to Euler, “A function of a variable quantity is an analytical expression composed in any way of that variable and numbers or constant quantities” [1748a, 1:18].269 At ﬁrst glance, this deﬁnition seems to reduce a function to an analytical expression. In reality, the problem is considerably more complex. In order to make this point clear, let us examine Euler’s deﬁnition for functions of more than one variable: 77. Even though we have so far examined more than one variable quantity, they were connected so that each of them was the function of only one variable and once the value of one variable was determined, the others would be simultaneously determined at the same time. We shall now consider certain variable quantities that do not depend on one another; if a determined value is given to one of these variables, the others remain indeterminate and variable. It would be convenient to denote such variables with x, y, z, because they comprise all determined values; if they are compared with each other, they will completely unconnected, since it is legitimate to replace any value of one of them such as z, and the others, x and y, remain entirely free as before. This is the diﬀerence between dependent variable quantities and independent variable quantities. In the ﬁrst case, if we determine one, all the others are determined. In the second case, the de266

On the 18th-century concept of a function, see Fraser [1989] and Panza [1996]. For instance, see Leibniz [GMS, 5:268 and 316]. 268 In his [1718, 241], Bernoulli gave the following deﬁnition: “I call a function of a variable quantity, a quantity composed in whatever way of that variable quantity and constants.” 269 See the deﬁnition of the general term (Section 13.1, p. 157). 267

206

The Development of a More Formal Conception termination of a variable in no way restricts the meanings of the others. 78. Therefore a function of two or more variable quantities x, y, z is an expression composed of these quantities in whatever manner. (Euler [1748a, 1: 91])

Euler ﬁrst, in Section 77, spoke of “dependence” among variables; he later, in Section 78, deﬁned a function of more than one variable as an analytical expression. This seems to be a contradiction. This apparent contradiction can often be found in 18th-century texts (see Panza [1992, 695–696]). Thus, in his Th´eorie des fonctions analytiques, Lagrange ﬁrst stated: “The term function of one or more quantities shall be given to every expression of calculus to which these quantities belong, with or without other quantities which are considered as given and invariable, so that the quantities of the function can have all possible values” (Lagrange [1797, 15]). However, he was later to assert: “In general, by the characteristic f or F placed before a variable, we shall denote any function of this variable, that is to say, any quantity dependent on this variable and that vary according to it following a given law” (Lagrange [1797, 21]). In my opinion, the 18th-century concept of function eﬀectively contained both the idea of dependence or relation among variables and the idea of analytical expression. A function was intended as the analytical expression of a relation between general quantities: It was a pair consisting of a relation between quantities and of the formula that analytically expressed this relation. Not only were the notions of analytical expressions and relations between quantities not contrasted with each others, but they were closely intertwined. An analytical expression was a function since it reiﬁes a relation between quantities; conversely, a relation between quantities could be the object of study in analysis only insofar as it was expressed by an analytical expression or formula. I also specify that 18th-century mathematicians often referred to a function as a quantity. For instance, Euler stated: “A function itself of a variable quantity is a variable” [1748a, 1:18]. Here the word “quantity” denoted a quantity depending on other quantities (and therefore the word “quantity” denoted what I have termed as a “relation between quantities”). By using this terminology, one can state that a function was a pair consisting of a quantity —depending on other quantities— and the analytical expression of this quantity.270 A crucial aspect of the 18th-century concept of function was that only the relationships that were analytically expressed by means of certain determined formulas were actually accepted as functions. The following excerpt 270 Afterwards I shall often conform to this use and speak of a “quantity” in place of a “relation between quantities.”

18

Analysis after the 1740s

207

from Lagrange’s Le¸cons sur le calcul des fonctions is useful to make this clear: The functions that we have considered in the last three lessons [they are: xm , ax , loga x, sin x, cos x, arcsin x, arccos x] are as the elements of which all functions, which can be formed by algebraic operations, are composed. For this reason I have thought to start by trying the derived functions of these simple functions; now I go on to illustrate how one can ﬁnd the derived functions of the functions that are composed by means of the simple functions in any way (Lagrange [1806, 48]).271 Lagrange identiﬁed a set of functions that acted as the basic building blocks. For the sake of simplicity, I shall term them “basic functions”. In his opinion, all other functions were constructed by means of basic functions using only the operations of addition, subtraction, multiplication, division and composition of functions. This concept was widely shared in the 18th century. Following Fraser,272 one can therefore state that (F) a function was given by one analytical expression constructed from variables in a ﬁnite number of steps using basic functions, algebraic operations and composition of functions. In the remainder of this book I shall refer to the functions included in this notion as “elementary functions”. I specify that functions could also be given in an implicit form. For example, in the ﬁrst chapter of [1748a], Euler wrote: [A]lgebraic functions can not often be exhibited explicitly, a function of z of this type is Z if it is deﬁned by an equation such as Z 5 = azzZ 3 − bz 4 Z 2 + cz 3 Z − 1. Indeed, although this equation cannot be solved, it is however known that Z is equal to any expression composed of the variable z and constants and, therefore, Z is a certain function of z (Euler [1748a, 19–20]). If F (x, y) was an elementary function of x and y, then the equation F (x, y) = 0 expressed a function y of x, even when the explicit form of the function y = y(x) was not known. Similarly, if y = f (x) was a function of x, it was assumed that x was a function of y, even though the analytical expression x = g(y) was unknown. In both cases, it was considered suﬃcient to have an analytical expression, F (x, y) or f (x), upon which one could operate. 271 272

The same concept is expressed in Lagrange [1813, 25–26]. See Fraser [1989, 325].

208

The Development of a More Formal Conception

There were historical reasons for the choice of the above-mentioned set of basic functions. At the end of the 17th century, only algebraic relations between quantities were expressed using formulas. Transcendental relations “were expressed by means of certain circumlocutions in prose” (see Bos [1974, 5]). Subsequently, logarithmic and exponential relations were expressed analytically by formulas involving numbers, letters, and abbreviations; trigonometric functions followed around 1740.273 In the 1740s, the set of basic functions was created and was explicitly described in Euler [1748a]. In the second half of the century, it remained substantially unchanged. The speciﬁc set of basic functions was also connected with the widespread notion of analysis as a unitary theory based upon the step-by-step extensions of arithmetical rules.274 For instance, Lagrange stated that analysis considered the functions that resulted from the generalization and symbolic representation of arithmetical operations (Lagrange [1806, 10]).275 This was due to the fact that functions were not really deﬁned if the term “deﬁnition” is taken to mean a free act of will by which we create the deﬁniendum. Thus, in the case of the exponential function ax , mathematicians did not deﬁne it but assumed the existence of a quantity with the following properties: (1) It interpolated an ; (2) it possessed the same properties as the arithmetical operation of raising to a power; (3) it could be represented by a “nice” curve. The function ax was merely the expression of this quantity in an abstract form using symbolic notation. It should be noted that basic functions were thought to satisfy the following conditions, which made them diﬀerent from other relations between quantities: (C1) A special calculus concerning these functions existed (i.e., a group of algorithmic rules related to the analytical expression, such as the rule of the calculus of trigonometric functions).276 273 They were introduced later, when their link with the exponential function had been established and it had been highlighted that they occurred as solutions to certain diﬀerential equations (see Katz [1987]). 274 See Panza [1992, 701–702] and Jahnke [1993, 281]. 275 This idea even made the introduction of trigonometric functions problematic (see Panza [1992, 701]). 276 In [1754–55a], Euler wrote: “The diﬀerent kinds of quantities, which Analysis deals with, generate diﬀerent types of calculus, where rules had to be adapted to any kind of quantities. Thus one teaches the special algorithm of both fractions and irrational quantities in elementary Analysis. The same use occurs in higher Analysis. There, since logarithmic and exponential quantities, which formed a new kind of transcendental quantities,

18

Analysis after the 1740s

209

(C2) The values of basic functions were considered as given since they could be calculated by performing algebraic operations and using tables of values. Conditions (C1) and (C2) were considered to be preserved by algebraic operations and compositions of functions and to be shared by all elementary functions. Eighteenth-century mathematicians regarded elementary functions as satisfying these conditions. It should be noted that (C1) and (C2) were precisely those conditions that allowed the object “function” to be accepted as the solution to a problem. In general, it is necessary to exhibit a known object in order to solve a problem. During the 18th century, only elementary functions (in the above sense) were thought to be known objects277 to the point that they could be accepted as the ﬁnal solution to a problem. Conditions (C1) and (C2) were vague and did not specify when a relation was to be considered as known; they did not imply that the set of accepted functions was deﬁnitively ﬁxed. In eﬀect, 18th-century mathematicians were prepared to extend it by introducing new functions, once certain (relations between) quantities were considered as known, but the introduction of new functions only took place very slowly and with great uncertainty. Only around 1800 did many mathematicians, such as Legendre,278 accept new functions, but Lagrange did not consider them in any edition of his treatises published from 1797 to 1813. I will return on this question in Chapter 29. Moreover, 18th-century functions were characterized in an essential way by the use of a formal methodology that it made it possible to operate upon analytical expressions, independently of their meaning. In Chapter 7, we saw that formal methodology was based upon two closely connected principles and examined one of these principles, the extension of rules and procedures from the ﬁnite to the inﬁnite. I shall now illustrate the other principle, the generality of algebra. It consisted of the following assumption: (GA) If an analytical formula was derived by using the rules of algebra,279 enter in computations, one usually teaches a special type of algorithm concerning both symbols and rules. It was termed exponential calculus by the inventor Joh. Bernoulli and also treats the theory of logarithm and their diﬀerentiation and integration. In addition to the logarithmic and exponential quantities there occurs in analysis a very important type of transcendental quantity, namely the sine, cosine and tangent of angles, whose use is certainly the most frequent. Therefore this type rightly merits, or rather demands, that a special calculus be given, whose invention in so far as the special signs and rules are comprised, the celebrated author of this dissertation [Euler] is able rightly to claim all for himself, and of which he gave examples in his Introduction to Analysis and Institutions of Diﬀerential Calculus” [1754–55a, 542–543]. 277 In analysis, an object was considered as known if it had an analytical expression on which one could operate and if one could at least partially calculate its values. 278 See Chapter 29. 279 Here the expression “the rules of algebra” is meant in a general sense. It includes not only the rules of the algebra of ﬁnites but also the rules of analysis of inﬁnite quantities.

210

The Development of a More Formal Conception then it was thought to be valid in general.

In his Calculus as Algebraic Analysis, Fraser expressed this principle by stating “The existence of an equation among variables implies the global validity of the relation in question” (Fraser [1989, 329]). The generality of algebra made it possible to view a function as a whole. Its behavior became a global matter that could not be reduced to the sum of the behavior of the points of its domain: It could not have a property P here, and a diﬀerent property there. This does not mean that 18th-century mathematicians merely considered functions that had the property P in every point: Rather they assumed rules that were valid over an interval Ix (or, more precisely, for certain values that this variable x assumed moving with continuity) as globally valid. For this reason, if one proved that a function f (x) had the property P in the interval Ix , then one could extend this property beyond the interval Ix . For instance, the rules concerning the function log x were derived for positive values of quantity x; however, it was assumed that the properties of the expression “ log x” lasted beyond the original interval of deﬁnition, even when x is negative or even imaginary.280 Of course, if what was valid in an interval was generally valid, not only did a function possess the same properties everywhere but also it maintained the same form everywhere since the form embodied all its properties.281 Therefore, one function necessarily consisted of one single formula282 and a relation such as 2 x for x ≥ 0 f (x) = x3 for x < 0 was never considered a function.283 In this context, it is necessary to emphasise the fact that the equality ∞ i f (x) = i=0 ai x was not restricted to the values of x where the series was The generality of algebra implied that the relation f (x) = ∞ convergent. i could not be regarded as a local relation, only valid for an interval. a x i i=0 Vice versa, a statement of the type 280 Euler stated: “For, as this calculus concerns variable quantities, that is quantities considered in general, if it were not generally true that d(log x) = dx/x, whatever value we give to x, either positive, negative or even imaginary, we would never able to make use of this rule, the truth of the diﬀerential calculus being founded on the generality of the rules it contains” [1749b, 143–144]. For two examples of the generality of algebra from Legendre, see Chapter 29. 281 One may ask to what extent this conception also belonged to pre-Eulerian analysis. The generality of algebra was also part of pre-Eulerian analysis insofar as it used analytical expressions, but its impact proved to be somewhat restricted by the vicinity of geometric reference, the disappearance of which led to the explosion of formalism. 282 See Fraser [1989, 325]. 283 I note that an analytical expression of the type y 2 = f (x) was considered as a manyvalued implicit function (see p. 207).

18

211

Analysis after the 1740s

f (x) =

∞

i=0

ai xi for x ∈ (a, b) and f (x) =

∞

i=0

ai xi for x ∈ / (a, b)

involved a rejection of the generality of algebra. A heavy reliance on formal methods prevented 18th-century mathematicians from appreciating the diﬀerence between complex and real variables and, therefore, between complex and real analysis. The transition from real to imaginary values usually occurred only by applying the generality of algebra, without proving speciﬁc rules for imaginary values of variables. Complex functions were not an autonomous object of study but, instead, were useful tools for the theory of real functions, and their use was restricted to exceptional circumstances. Finally, it is also worthwhile noting that the generality of algebra was restricted to analysis, where functions were studied without a priori restrictions concerning variables. In arithmetic, geometry, and mechanics, functions and variables have a natural range; therefore, mathematicians were obliged to take into consideration the restrictions that the nature of the speciﬁc problem under examination imposed. When the results derived from the use of the generality of algebra were applied to other sciences, they had to be subjected to appropriate reinterpretations that adapted them to concrete circumstances. This approach is an aspect of the mathematical method for studying natural science in the 18th century, which Dhombres [1988] referred to as the “functional method.” By solving a problem mathematically, appropriate symbols replaced concrete quantities and their relations come to be conceived as formulas and equations. The solutions to these equations were to be interpreted in relation to the speciﬁc problem and by eliminating anything that was meaningless for this particular problem. The whole theory of series is an example of this conception, since the convergence was studied a posteriori as a condition for the applicability of the formally derived results.

18.3

On the continuity of curves and functions

When referring to 18th-century functions or curves, the term “continuity” can be understood in two diﬀerent (though connected) ways. First, continuity can be understood as the absence of jumps or the assumption of any intermediate state between two given states or gradual change. I shall refer to this sense of continuity as local284 continuity, or L-continuity for short. In the 18th century, functions were thought to be intrinsically continuous in this sense. This concept depended on the fact that a function y = f (x) was a relation between the general quantities y and x, or in other words, the general quantity y was considered as depending on the 284 Of course, local continuity does not mean pointwise continuity. According to the description on p. 102, the continuous was not reducible to points and a function —in the sense of a relation between general quantities— was not deﬁned pointwise.

212

The Development of a More Formal Conception

quantity x. Since a general quantity was continuous, any function y = f (x) was considered as such. In analytical terms, this implied that the following property held for any function:285 (LC) Δf = f (x + ω) − f (x) is inﬁnitesimal if ω is inﬁnitesimal.286 I explicitly emphasize that this property was not the deﬁnition of the continuity of a function but merely a trivial consequence of the notion of continuous quantity.287 Second, continuity can be thought of as coinciding with uniqueness. The basic idea behind this concept is that an object is continuous if it is an unbroken object, i.e., if it is not broken in two objects and is therefore one object.288 Even though 18th-century mathematicians always considered functions and curves as locally continuous, the usually accepted deﬁnition of the continuity of a function or curve was based on the property of uniqueness. I shall refer to this way of understanding the continuous as global continuity, or G-continuity or Euler’s continuity for short. I shall investigate global continuity in this section, as concerns continuous curves, and in Chapter 23, as concerns G-continuous functions. First of all, I observe that if continuity is the same thing as uniqueness, one curved line was G-continuous merely because it was one. Therefore, the notion of a continuous curve may appear superﬂuous and useless. However, in analytical geometry, a curve is represented by an analytical expression and one analytical expression does not necessarily correspond to an unbroken curve. For instance, the function y = xk is G-continuous since it is one, but its geometrical counterpart, the hyperbola of the equation y = xk , is broken into two pieces: It is then very natural to ask whether the hyperbola is continuous, i.e., whether its two pieces form an unique curve. Put in more general terms, how does one recognize that an object is one? The most obvious answer is that an object is one if it retains its properties. Now, if we study a curve analytically, its properties are included within its analytic expression. If we accepted this view, then it is entirely natural that the criterion of uniqueness must be applied to the analytical expression, as Euler did in classifying curves.289 Indeed, he stated that although some curves could be described mechanically, he aimed to study curves insofar as they were originated by functions because this method was the most general and best suited to calculation. According to Euler, from such an idea about curved lines, it immediately follows that they should be divided 285

See, for instance, Euler [1755, 82] and Lagrange [1797, 28]. The property (LC) held at least for over one interval. However, the generality of algebra made it possible to consider L-continuity as a property of the whole function. 287 See Chapter 7. 288 This is precisely the Aristotelian concept of continuity. On the notion of continuity in Aristotle, see Panza [1989, 39–65]. 289 See Euler [1748a, 2: Section 8]. 286

18

Analysis after the 1740s

213

into continuous and discontinuous or mixed. A curve was continuous if its nature was determined by only one function, and discontinuous or mixed if it was described piecewise by more than one function and, consequently, was not formed according to an unique law. Uniqueness did not apply to the course of a curve, which was seen as an outward manifestation, but to the function itself as a primary object. The number of the branches of a curve was therefore of no importance. Euler also subdivided curves into complex and not complex ones using a similar criterion. He noted that the equation of certain algebraic curves could be broken down into rational factors: Such equations include not one but many continuous curves, each of which can be expressed by a particular equation. They are connected with each other only because their equations are multiplied mutually. Since their link depends upon our discretion, such curved lines cannot be classiﬁed as constituting a single continuous line. Such equations (referred to above as complex) do not give rise to continuous curves, although they are composed of continuous lines. For this reason, we shall call these curves complex. (Euler [1748a, 2: Section 61]) The complex curves (like mixed ones) were discontinuous because their equation was characterized by arbitrariness; in other words, they are not determined by exactly one analytical law. Their diﬀerence is that the complex curves were composed of more than one whole curve, whereas mixed curves were composed of pieces of more than one curve.290 In his Introductio, Euler only considered G-discontinuous curves rather than G-discontinuous functions. This is entirely coherent with the 18thcentury principle being developed by Euler. Indeed, the application of global continuity to functions required that an essential property of functions did not hold anywhere, whereas the principle of the generality of algebra made it possible to consider a function as a unitary object: The properties valid on an interval were considered valid anywhere. In this context it seemed impossible to attribute a meaning to the term “discontinuous” when referring to functions. Nevertheless, when the controversy about the vibrating string arose, the existence of noncontinuous functions was admitted and the relationship between quantities and analytical expressions appeared to be √ On the basis of these subdivisions, the curve of equation y = x2 is not continuous. Although it appears to be a G-continuous curve since it derives from one two-valued function, it is, in reality, the complex curve corresponding to the (implicit function) equation y 2 − x2 = 0. According to Euler, uniqueness did not refer to the “apparent”, complex form, but to the essential, irreducible form. In the light of this observation, Cauchy’s objection to Euler’s classiﬁcation in [1844] should also be considered. 290

214

The Development of a More Formal Conception

problematic. The way this occurred is dealt with subsequently when I illustrate the emergence of certain diﬃculties in the structure of 18th-century analysis.

19

The formal concept of series

The evolution of series theory after 1720 revealed tension between the formal and quantitative aspects. This led to the formulation of a diﬀerent concept of series, where the previously existing balance was upset. I term this new concept formal. In the present chapter, I illustrate how Euler made the formal concept of series explicit.

19.1

Criticisms to the inﬁnite extension of ﬁnite rules

1 The sum of ∞ n=1 n2k in De summis serierum reciprocarum was one of Euler’s most important successes,291 though the method he had used also became the object of severe criticism. Some mathematicians292 noted that such a method could be applied to other quantities and not only to sin s: In this case the series ∞ 1 n2 n=1

would have sums that were diﬀerent from those derived by Euler. Indeed, given an ellipse of axes m, l (m > l), let s be the length of the arc AP , where A has coordinates m , 0 and P is an arbitrary point (Fig. 20). If 2 one considered the ordinate of P as a function y(s) of s and applied Euler’s method to the equation 1 1 13 y(s) = s − 2 s3 + − s5 + 6c 10rc3 120c4 71 493 1 + − s7 + . . . = 0, + − 14r2 c4 420rc5 5040c6 where r =

m 2

and c =

l2 2m ,

one obtained

∞ 1 S 2 m2 1 S2 = = 2· , 2 n 6c 4 6l4

n=1

1 where S is the circumference of the ellipse, in contradiction with ∞ n=1 n2 = 2 π 6 . In De summis serierum reciprocarum dissertatio altera, Euler293 replied to this criticism by stating that the equation y(s) = 0 had imaginary roots, unlike sin s = 0, and it was therefore not possible to deduce the sum of series 291

See Section 13.3. See Fuss [1843, 2:477 and 683]. 293 It is worth observing that Euler also responded to criticisms by seeking new proofs of the results of [1734–35a] (see, e.g., D´emonstration de la somme de cette suite 1 + 41 + 19 + 1 1 1 + 25 + 36 + . . . [1743b]). 16 292

215

216

The Development of a More Formal Conception

P(x(s),y(s)) s A(m/2,0)

Fig. 20

from this equation (Euler [1743a, 139–140]). In a letter to Euler, however, Nikolaus Bernoulli II observed that, even if sin s = 0 did not have imaginary roots, his reasoning was illegitimate unless “it had been demonstrated that 1 5 s −etc. was convergent and gave the sinus of the the series s − 61 s3 + 120 arc s accurately, whatever value was assigned to s” (Fuss [1843, 2:683]). In Bernoulli’s opinion, if one was not sure that series were convergent, one might be wrong, and this fact occurred when one reasoned on y(s) instead of sin s. In eﬀect, the validity of the factorization s s s 1 3 1 5 s 1− + 1− 1− ... s − s + ... = 1 − y 3!y 5!y A1 A2 A3 depended on the convergence of s − 61 s3 +

1 5 120 s

− . . .:

and for this reason the objection concerning the series of the s3 is solved. This series is sinus of the elliptic arc s − 6c 2 +etc. divergent for increasing s; from which it cannot be deduced, as in the circle where the series is not divergent, that the coeﬃcient, taken as a negative, of the second term in the inﬁnite equation s3 1 1 0 = s − 6c 2 + etc, i.d.. 6c2 expresses the sum of all ss , namely it 1 1 1 is = ππ + 4ππ + 9ππ +etc. (Fuss [1843, 2: 691]) On April 6, 1743, Bernoulli explicitly wrote to Euler: I cannot persuade myself that you [Euler] think that a divergent series . . . provides the exact value of a quantity which is 1 is not = 1 + x + xx + expanded into the series. Indeed, e.g., 1−x x3 + . . . + x∞ , but = 1 + x + xx + x3 + . . . + x∞ + [1843, 2:701–702])

x∞+1 1−x .

(Fuss

19

217

The Formal Concept of Series

Later, Bernoulli clariﬁed: “I think the idea of the sum or aggregation of more terms cannot be associated with the idea of terms advancing endlessly, and I regard these two ideas as contradictory” and argued that the properties applied to a ﬁnite equation could not be applicable to an inﬁnite equation (Fuss [1843, 2:708–709]). Although N. Bernoulli’s concept might seem close to modern ideas, in reality he reproposed the traditional formal-quantitative concept of series. 1 1 1 1 Indeed, in his Inquisitio in summa seriei ∞ 1 + 4 + 9 + 16 + 25 + etc. [1738], in order to sum the recurrent series n=1 an , with an+1 =

An(n − 1) + Bn + C an E(n + 1)n + F (n + 1) + G

(A, B, C, E, F , and G are constants), N. Bernoulli considered y(x) =

∞

an xn

n=1

and derived the expression of dy and d2 y by diﬀerentiating term by term. By using principle294 (IE), he obtained Gydx2 + F xdydx + Ex2 d2 y ∞ ∞ ∞ n n 2 n 2 En(n − 1)an x dx2 F nan x dx + Gan x dx + = =

Ga1 x +

+

∞

∞

n=2

n+1

Can x

n=1

n=2

n=1

n=1

2

dx +

An(n − 1)an xn+1

F a1 x +

∞

n+1

Bnan x

n=1

dx2

dx2

= Ga1 xdx2 + F a1 xdx2 + Cxydx2 + Bx2 dydx + Ax3 d2 y. The diﬀerential equation Gydx2 + F xdydx + Ex2 d2 y = Ga1 xdx2 + F a1 xdx2 + Cxydx2 + Bx2 dydx + Ax3 d2 y provided the sum y(x). Surely N. Bernoulli intended to apply this procedure to a convergent series, but any a priori discussion of the convergence and of the legitimacy of the steps of the proof is lacking. He diﬀerentiated term by term and freely rearranged the series. However, by diﬀerentiating and integrating term by term and freely rearranging, a convergent series can be turned into a divergent one, and vice versa. This had already been noted by 294

See p. 117.

218

The Development of a More Formal Conception

Leibniz in his Epistola ad V. Cl. Christianum Wolﬁum 295 [1713] and was to be one of the reasons for Daniel Bernoulli in his De summationibus serierum [1771, 76] to use of the sum of divergent series. D. Bernoulli observed that, by integrating ∞ 1 (−1)n (n + 1)xn = , (1 + x)2 n=0

one had

∞

(−1)n xn+1 =

n=0

x . 1+x

If one divided this equality by x and integrated again, one obtained ∞

(−1)n

n=0

xn+1 = log(1 + x). n+1

Setting x = 1, the three series yielded ∞

(−1)n (n + 1) =

n=0 ∞

(−1)n =

n=0 ∞

n=0

(−1)n

1 n+1

1 , 4 1 , 2

= log 2,

respectively. Daniel Bernoulli thought that the validity of the last relation, in a sense, guaranteed the validity of the two others. Of course, Nikolaus Bernoulli could have replied that in the ﬁrst two cases the series were not convergent; however, it had to be observed that he never analyzed the condition of convergence a priori and, above all, did not provide the conditions of applicability of certain procedure. In his Inquisitio he limited himself to observing a posteriori that a formal theorem was applied to a particular, convergent series. The formal study of a theorem and consideration of the conditions of its applicability in concrete cases a posteriori is just one of the characteristics of a concept that contains formal components. For this reason, his criticisms of the inﬁnite extension of ﬁnite laws were weak: He used the same principle (IE) that Euler had, even though Euler used it in a stronger form. N. Bernoulli’s ideas could be developed in a more modern sense, but this did not occur during the 18th century. People who criticised Euler were unable to avoid formal methods entirely. When considered seriously, N. Bernoulli’s arguments involved a rethinking of the whole analysis, 295

See footnote no. 185.

19

The Formal Concept of Series

219

which went beyond the intention of the mathematicians of the time and the state of the art. There was a second reason why N. Bernoulli’s standpoint was weak. It did not produce results of wide interest; by contrast, divergent series gave rise to a number of signiﬁcant ﬁndings and were to prove fertile ground for further investigations in later decades. I would like to emphasize this point: The formal point of view contributed to the development of mathematical knowledge, whereas the approach of their opponents was substantially sterile during the middle of 18th century. I think that this was the heart of question. The formal approach was triumphant because it was capable of producing new mathematics.296

19.2

The impossibility of the quantitative approach

In the 1750s, the criticisms of his opponents led Euler to make the formal concept of series explicit and to go beyond the formal-quantitative approach. He sought to give a deﬁnition of the sum that generalized the old notion and provided a basis for new ﬁndings. Before examining the formal concept, however, it is necessary to make some preliminary remarks. In writing to Euler, N. Bernoulli asserted that the “idea of the sum” was to be clariﬁed in order to avoid the situation whereby “the dispute becomes a logomachy” (Fuss [1843, 2:708]). Euler observed that the problem lay in the word “sum” and thought it was appropriate to provide an adequate deﬁnition of this term [1754–55b, 589]. These statements should not give the idea that the disputants gave a “deﬁnition” in the modern sense of the term, i.e., in the sense of the numerous deﬁnitions of the sum of a divergent series given after 1890. As we saw, in the 18th century, a mathematical theory was not the set of the logical deduction of axioms and deﬁnitions, subject only to the principle of contradiction; rather, it was an idealization derived from physical reality. In this context, deﬁnitions were clariﬁcations of the unique and necessary “truth” that already existed in nature. For this reason, Euler believed that 1 − 1 + 1 − 1 + . . . was equal to 1/2 not because of an arbitrary deﬁnition, but because this equality was the unique possible equality that could be derived from the laws of analysis. In a similar way, Euler thought that 1 + 1 + 1 + ... = ∞ and could not assert 1 + 1 + 1 + . . . = −1/2, unlike Ramanujan’s modern deﬁnition of the sum.297 In Euler’s opinion, ∞ setting n=0 an = C without a justiﬁcation but only by deﬁnition was 296 297

See Part III. On this deﬁnition of the sum of a divergent series sum, see Hardy [1949, 327 and 333].

220

The Development of a More Formal Conception

unacceptable. By deﬁning the sum of a series, Euler wanted to indicate there was a “really existent” link between that series and a certain quantity; he did not want to create an arbitrary notion whose logical value would be independent from any veriﬁcation of utility or correspondence to external truth. Since Euler was seeking the “true” notion of the sum, in De seriebus divergentibus [1754–55b] and Institutiones calculi diﬀerentialis [1755], he provided a long analysis that allowed him to exclude the possibility of a quantitative interpretation of the sum of divergent series. In De seriebus divergentibus, he subdivided divergent series into four classes according to the sign and behavior of the nth term: ∞

an and an < M,

(123)

n=0 ∞

(−1)n an and an < M,

n=0 ∞

n=0 ∞

n=0

an and a∞ = ∞,

(124)

(−1)n an and a∞ = ∞,

where M denotes a positive constant and an > 0 is a nondecreasing sequence.298 Divergent series with positive terms gave rise to the most serious problems: Unlike alternating divergent series, it would be natural to assign the value ∞ as their sum. For Euler, only the series of type (123) actually had an inﬁnite sum.299 The series of the type (124) could have a ﬁnite sum; however, the real meaning of equalities such as 1 + 2 + 4 + . . . = −1 and

1 2 was problematic. These series had negative sums but, at the same time, had to have sums greater than 1 + 3 + 9 + ... = −

1 + 1 + 1 + . . . = ∞; 298

According to the usual terminology of 18th-century, Euler termed a series ∞ n=0 an convergent if an > 0 is a decreasing sequence and a∞ = 0. Euler did not discuss series of this kind, whose sum can be inﬁnite. He probably thought that “convergent” series (in this sense) had always a quantitative meaning even when their sum was inﬁnite. 299 He stated: “There is no doubt that the sum of these series can be shown by means of expressions of the type a0 ”[1754–55b, 589]

19

221

The Formal Concept of Series

hence, negative numbers should be greater than inﬁnity. Euler believed that the concept according to which negative numbers are both greater than inﬁnite and less than zero is not far from the truth; using a modern image, we could say that the set of real numbers is closed by the inﬁnite. In Institutiones calculi diﬀerentialis, Euler justiﬁed this idea by examining several sequences of the kind 1 1 1 1 1 1 1 1 ...,− , − ,− ,− , + , + ,+ ,+ ,..., 4 3 2 1 0 1 2 3 . . . , −4, − 3, − 2, − 1, + 0, + 1, + 2, + 3, . . . . He highlighted analogies between the zero and the inﬁnite, showing that the “transition” from the negative to the positive occurred via both the zero and the inﬁnite, and appealed to the law of continuity and geometry [1755, 2: Sections 98–101]. This idea was compatible with a quantitative interpretation of the sum; however, it did not prevent diﬃculties. Further, in his Institutiones calculi diﬀerentialis, Euler observed that 1 = 1 + 2x + 3x2 + 4x3 + . . . (1 − x)2 yielded ∞= and 1=

1 = 1 + 2 + 3 + 4 + . . . for x = 1 (1 − 1)2

1 = 1 + 4 + 12 + 32 + . . . (1 − 2)2

for x = 2;

therefore, even 1 should be greater than ∞ [1755, 2: Section 104]. He was thus obliged to reject the quantitative concept of the sum of divergent series, as it was unsuitable for any relation of order between numbers and the inﬁnite. With respect to alternating divergent series, the main objection to their use (a valid objection for positive term series as well) concerned the possibility of neglecting the remainder. Euler criticized mathematicians such as n+1 Nikolaus Bernoulli II, who thought that the remainder a1+a of an+1 1 = 1 − a + a2 − a3 + . . . ± an ∓ 1+a 1+a had not to be neglected for n = ∞, except for (0 1 (see [1755, 2: Chapter 1]). If there is no numberk such that Δn a1 = 0 for n > k, (125) is unsuitn able for approximating ∞ n=1 an x . Indeed, if x < 1 (“in which case only summation in the proper sense of the word can take place” [1755, 221]), then x > x, 1−x n and (125) does not improve the convergence of ∞ n=1 an x . Furthermore, for x = 1, (125) does not yield a numerical value. However, by changing x into −x in (125), Euler derived ∞

∞

(−1)n−1 an xn =

n=1

n=1

Since

x 1+x

(−1)n Δn−1 a1

< 1 and

x 1+x

x 1+x

n

.

< x for 0 < x < 1, ∞

n

(−1) Δ

n−1

n=1

a1

x 1+x

n

sped up ∞

(−1)n−1 an xn

n=1

and provided an appropriate “approximation of the value”. Setting x = 1, he obtained ∞ ∞ (−1)n−1 2−n △n−1 a1 . (126) (−1)n−1 an = n=1

n=1

Euler stated that, in certain cases, (126) could provide the sum of divergent series. Of course, it did not provide the sum in the quantitative sense but in the formal sense or, to use his words, the value of the ﬁnite expression whose expansion generates the given series (Euler [1755, 222–223]). This could occur either because (126) changed the divergent series into a ﬁnite one or because it transformed a divergent series into a convergent one.310 310

By applying (126) repeatedly, Euler found that 1! − 2! + 3! − 4! + 5! − . . . was approximately equal to 0, 40082038: He said that it was a value not much diﬀerent from 0, 4036524077, the more precise value he determined elsewhere [1755, 222–223]. Euler probably referred to the last of the values of 1 − (1! − 2! + 3! − 4! + 5! − . . .), which he had determined in his [1754–55b].

19

225

The Formal Concept of Series

For instance, 1 − 2 + 4 − 8 + ... =

1 1 1 1 1 − + − = . 2 4 8 16 3

Euler also used (126) to transform slowly convergent series into rapidly convergent series, such as 1−

1 1 1 1 1 1 1 1 + − + − ... = + + + + ..., 2 3 4 5 2 2 · 4 3 · 8 5 · 32

without appreciating the crucial diﬀerence (from a modern viewpoint) between the application of (126) to a divergent or convergent series. Euler thought that the essence of the method was to derive a series ∞

bn

n=1

from a given series ∞

(−1)n−1 an

n=1

such that |bn | < |an | independently of the fact that we speed up the convergence or go from the divergence to the convergence or decelerate the divergence. The similarity to the modern summation method (E, 1) is therefore only apparent.311 Nowadays, if ∞ (−1)n−1 an n=1

is divergent, (126) conventionally deﬁnes its sum, which exists by means of this deﬁnition; on the other hand, if the series is convergent, it allows us to calculate its ordinary sum more easily, which exists by virtue of another deﬁnition. By contrast, for Euler, it externalized the true sum but did not “create” it. ∗ ∗

∗

In Institutiones calculi diﬀerentialis, Euler argued that divergent series never led to an error being made. This was one of the reasons in favo of divergent series, because if their sums were false, they would not always lead us 311

On this method of summation, see Hardy [1949].

226

The Development of a More Formal Conception

to the truth.312 Such a justiﬁcation must have already been communicated to N. Bernoulli, who, in response, pointed out some sums of divergent series that led to an error being made (see Fuss [1843, 2:709]). On the other hand, in Consideratio progressionis cuiusdam ad circuli quadraturam inveniendam idoneae [1739a], Euler himself had pointed out that divergent series could cause discrepancies. As we saw in Chapter 15, he had noted that ∞ (a) the theorem “given any series ∞ n=1 an , then n=−∞ an = 0, where a0 , a−1, . . . are deﬁned by (100)” was not always true; (b) Equation (99) provided a good approximation of π but was not a quantitative equality in the sense of convergent series. In De seriebus divergentibus and Institutiones calculi diﬀerentialis, Euler did not mention the above diﬃculties313 and limited himself to inviting mathematicians to correct any accidental mistake as the absence of systematic errors was certain [1755, 81].314 Euler therefore interpreted the results of the Consideratio diﬀerently. 1 I think that Euler merely considered the counterexample ∞ n=−∞ n2 = 0 ∞ to the theorem n=−∞ an = 0 as an exception. This approach was already implicit in the Consideratio when he stated that ∞ n=−∞ an = 0 did not fail in any case but led to the truth in numerous cases [1739a, 362]. From a modern viewpoint, only one counterexample is suﬃcient to reject a proposition. According to Euler, theexistence of a counterexample on its own did not mean that the proof of ∞ n=−∞ an = 0 is incorrect. Euler conceived counterexamples (if they were sporadic) as exceptions to the rule, which had to be examined and explained, but did not invalidate a formal result. This approach was widespread in 18th-century analysis. Mathematicians thought that a theorem concerning a generic object X could admit exceptional cases in which it fails. In other words, the theorem the generic object X has the property P did not imply that all the propositions the object X1 has the property P , the object X2 has the property P , the object X3 has the property P , ... 312

See Euler [1755, 2: Section109]. It is appropriate to note that the results of Consideratio were based upon the same principles Euler used in De seriebus divergentibus and Institutiones calculi diﬀerentialis: the interpretation of an as a function of n (in Euler’s sense), the extrapolation by means 1 1 1 , (1−1) of (100); the replacement of 1−1 2 , (1−1)3 with 1 + 1 + 1 + . . ., 1 + 2 + 3 + . . ., 1 + 3 + 6 + . . ., . . ., the possibility of rearranging the terms of a series. 314 See Chapter 15, p. 183. 313

19

227

The Formal Concept of Series

where X1 , X2 , X3 , . . ., are speciﬁc cases of X, was true. In the above example, the theorem concerns the behavior of a generic ∞ 1 series ∞ n=−∞ n2 . n=1 an and the exceptional case was the speciﬁc series More frequently, the theorem had the form a function f (x) has the property P . In this case, the theorem was thought to be valid and rigorous as long as the variable x remained indeterminate; but this was no longer the case if one gave a determinate value to x. As Fraser315 puts it, “isolated exceptional values at which the relation fails are not signiﬁcant”. For instance, in his Le¸con sur le calcul des functions, Lagrange proved that, given a generic function f (x), the development f (x) + ai + bi2 + ci3 + . . . + qin + . . . of f (x + i) included no fractional power of i.316 When referring to this theorem, Lagrange asserted: This demonstration is general and rigorous as long as x and i remain indeterminate; but this is no longer the case if one gives a determinate value to x. (Lagrange [1797, 23]) In his [1771], Daniel Bernoulli explained such a concept by stating that the formal sum may exclude certain points: whose existence and location cannot be indicated by abstract analysis. Thus the tangent method cannot indicate cuspidal points if they are in the given curve. As a consequence, however, neither can the tangent method be disproved nor can one be convinced of its falsity. (D. Bernoulli [1771, 84]) I explicitly emphasize that, in this context, no speciﬁc counterexample could make the derivation of a theorem (and in particular a theorem on series) invalid. ∗ ∗

∗

In the early theory of series, we saw that the investigation of the eﬀective convergence of a numerical series or the determination of the interval of convergence of a power series was an a posteriori question; it concerned the moment of the application of the results found using analytical techniques.317 315

Cf. Fraser [1989, 331]. See Chapter 28. 317 See Chapter 8, p. 119. 316

228

The Development of a More Formal Conception

This is also true for the formal concept, but with diﬀerences regarding the mathematical context in which series theory was included. Previously, the context had been geometric: Series served to investigate geometric quantities and the moments of determination of a certain expansion and that of its application were two moments of the same ﬁeld of mathematics. The context was now analytical and series served to investigate general quantities. Geometric and numerical applications of series were not part of analysis; they belonged to diﬀerent ﬁelds of mathematics and this increasingly separated the two moments and made the formal more evident. Analysis on its own did not deal with convergence explicitly,318 but when series were applied to geometry, mechanics, and to what today is termed numerical analysis, they were treated quantitatively, i.e., one observed whether, and within what limits, a formal, general equality became a quantitative, special equality. In these applications,319 convergence was important. This concept explains the observations of historians such as Golland and Golland, according to whom Euler was more concerned with convergence issues than he has traditionally been credited.320 The Gollands’ analysis is based on the papers concerning the application of trigonometric series to mechanics: This is one of the cases in which 18th-century mathematicians paid attention to questions of convergence, approximations, and the evaluations of error. I recall the words of Lagrange written in a letter regarding astronomy: It is not diﬃcult to reduce the problem to an equation but since this equation was diﬀerential, it required some integrations. Such integrations can only be obtained through series. The entire question came down to whether they were convergent. (Lagrange [Oeuvres, 14: 280]) For instance, in Institutiones calculi diﬀerentialis [1755], Euler ﬁrst derived the Taylor theorem formally, without regard to the interval of convergence, and only subsequently took care to consider convergence when he applied the Taylor series to the calculation of the sth roots of a number c. He321 considered the Taylor expansion of y = xn : n (n − 1) n−2 2 x ω 2! n (n − 1) (n − 2) n−3 3 + x ω + ... 3!

(x + ω)n = xn + nxn−1 ω +

(127)

318 It should be recalled that the quantitative was masked within the accepted procedures and the notion of variable quantity (see Chapter 8). 319 Applications (especially numerical applications) were sometimes contained in the treatises of the analysis of the inﬁnite, but they were conceptually separate (the clearest example is Lagrange [1797]). 320 See Golland and Golland [1993, 55]. 321 See Euler [1755, 277–279].

19

The Formal Concept of Series

Setting ω = u, n = r/s, x = as in (127), he derived r (r − s) u2 ru r/s s r = a 1+ s + (a + u) sa s · 2s · a2s r (r − s) (r − 2s)u3 + + ... . s · 2s · 3s · a3s

229

(128)

This formula could be applied to the calculation of c1/s by an appropriate decomposition of c of the type as + u. However, Euler was not satisﬁed su and improved the convergence by changing ω into − aas +u and n into − rs in formula (127). He obtained ru r (r + s) u2 r/s s r = a 1+ (129) + (a + u) s(as + u) s · 2s · (as + u)2 r (r + s) (r + 2s)u3 + ... . + s · 2s · 3s · (as + u)3 Euler observed:

the latter series [i.e., (129)] converges more than the former [i.e., (128)], since its terms still decrease if u > as , in which case the latter instead series diverges. (Euler [1755, 277]) Then he showed that one could further improve the convergence of (129) by means of an appropriate decomposition of c in as + u or by calculating the roots hc1/s = (hs c)1/s in place of c1/s and decomposing hs c in bs + u, with b > a. In the context of the actual use of series for numerical calculations, Euler also provided two criteria that series had to satisfy in order to be suitable for numerical calculations. First, the terms of series should not be complex. Second, “the series must be vehemently convergent322 or of such a kind that any term is much less than the preceding one.” In this way not many terms provided a close enough approximation to the sought sum (see Euler [1737c, 247]). Finally, I note that when analysis (abstract analysis, to use Bernoulli’s words [1771, 71 and 84]) was applied to geometry, mechanics, and numerical analysis, convergence criteria were even used: They served to avoid useless calculations when one had to compute the value of a series by approximation.323 322 As usual in the 18th century, he was interested in rapid convergence rather than the convergence on its own: A slowly convergent series is of little use for calculating its values in almost the same way as a divergent series. 323 For instance, see Euler [1794a].

20

Lagrange inversion theorem

During the second part of the 18th century, the bulk of research on series placed increasing emphasis on the formal aspect. One of the most remarkable results was the “Lagrange series”, which I examine in this chapter.324 In Nouvelle m´ethode pour r´esoudre le ´equations litt´erales par le moyen des s´eries [1768, 14–25] Lagrange proved the following theorem: Given the equation α − x + ϕ(x) = 0, where ϕ(x) is any function, if p is a root of this equation and f (p) is a function of p, then 1 d2 ′ 1 d ′ f (x)ϕ3 (x) + . . . f (x)ϕ2 (x) + 2! dx 3! dx2 (130) by changing x into α after diﬀerentiation. In order to prove this theorem, Lagrange considered the equation a1 − a2 x+a3 x2 −. . .+(−1)n an+1 xn = 0, where a1 , a2 , a3 , . . . , an+1 are unspeciﬁed coeﬃcients, and set f (p) = f (x) + ϕ(x)f ′ (x) +

x x x 1− ... 1 − . a1 −a2 x+a3 x2 −. . .+(−1)n an+1 xn = a1 1 − p1 p2 pn He divided by a2 x and obtained a1 a3 x − . . . + (−1)n an+1 xn−1 − a2 x a2 x x x a1 1− ... 1 − 1− = − a2 x p1 p2 pn a1 p1 x x = 1− ... 1 − . 1− a2 p1 x p2 pn 1−

Then he set ξ=

a3 x − . . . + (−1)n an+1 xn−1 a2

and wrote p1 a1 a1 + log 1 − −ξ = log log 1 − a2 x a2 p1 x x x + . . . + log 1 − , + log 1 − p2 pn

(131)

324 In the reconstruction of the topics described in Chapter 20, 21, and 22, I follow Panza [1992].

233

234

Successes and Problems of the Triumphant Formalism

where log N denotes the natural logarithm of the numbers N . Lagrange expanded the right-hand side of (131): x p1 a1 x + log 1 − + log 1 − log + log 1 − a2 p1 x p2 p3 x + . . . + log 1 − pn ∞ ∞ ∞ 1 p1 r 1 x r 1 x r a1 − − − = log a2 p1 r x r p2 r p3 r=1 r=1 r=1 ∞ 1 x r −... − . r pn r=1

By rearranging, he had

x x a1 p1 + log 1 − + . . . + log 1 − log + log 1 − a2 p1 x p2 pn ∞ n ∞ r 1 p1 r 1 1 a1 xr . − (132) − = log a2 p1 r x r ps r=1

r=1

s=2

Then Lagrange wrote the left-hand side of (131) in the form ξ a1 a1 + log 1 − −ξ = log 1 − log 1 − a2 x a2 x 1 − aa21x r ∞ ∞ 1 a1 r 1 ξ − . (133) = − r a2 x r 1 − aa21x r=1 r=1 i He set ξ r = ∞ i=0 Ar,i x (r > 1). Consequently, r ξ 1 r a1 −r 1 ξ 1 − = r 1 − aa21x r a2 x ⎤ ⎡ ∞ ∞ j 1 a1 r+j−1 ⎦ = Ar,i xi ⎣ r a2 x j j=0 i=0 ⎡ ⎤ j ∞ ∞ a1 r+j−1 1⎣ Ar,i xi−j ⎦ (134) = j r a2 i=0 j=0

and

∞ 1 a1 r a1 −ξ = − (135) log 1 − a2 x r a2 x r=1 ⎡ ⎤ j ∞ ∞ ∞ r+j−1 a1 1⎣ Ar,i − x⎦ , j r a2 r=1

i=0 j=0

20

235

Lagrange Inversion Theorem

By comparing (135) and (132), he obtained ⎡ ⎤ j r ∞ ∞ ∞ ∞ a1 r+j−1 1⎣ 1 a1 An,i + xi−j ⎦ j r a2 x n a2 n=1 r=1 i=0 j=0 n ∞ ∞ a2 p1 1 p1 r 1 1 r xr . = log + + a1 r x r ps r=1

r=1

s=2

Lagrange assumed that the coeﬃcients of xr had to be equal. In particular, he considered the coeﬃcient of x0 , x−1 , x−2 , . . . and obtained ∞ i ∞ a1 1 n+i−1 a2 p1 An,i , = log i n a2 a1 n=1 i=0 ∞ i+1 ∞ a1 a1 1 n + i An,i + = p1 , i+1 a2 n a2 n=1 i=0 ∞ i+2 2 ∞ a1 1 a1 1 n+i+1 (p1 )2 , = An,i + i+2 a2 n a2 2 n=1 i=0 ... In this way, Lagrange obtained formulas for log p1 = log aa2 p1 1 + log aa12 and every power of p1 . Finally, Lagrange wrote these formulas by using diﬀerentials. He set p = p1 , x = aa12 and had ∞ ∞ 1 r+i−1 i Ar,i x log p = log x + i r r=1 i=0 ∞ ∞ 1 A0,i xi + = log x + (i + 1)A1,i xi 2 i=0 i=0 ∞ (i + 2)(i + 1) 1 i A2,i x + . . . + 3 2! i=0

= log x + ξ + Similarly, pk = xk + k

$

Since

xk ξ +

1 d 2 1 d2 2 3 xξ + x ξ + .... 2! dx 3! dx2

% 1 d k+1 2 1 d2 k+2 3 ξ x + . . . . x ξ + 2! dx 3! dx2

(136)

a1 − a2 x + a3 x2 − . . . + (−1)n an+1 xn = a1 − a2 x + ξa2 x, by setting α =

a1 a2

and ϕ(x) =

ξ(x) x ,

one had

α − x + ϕ(x) = 0.

(137)

236

Successes and Problems of the Triumphant Formalism

Lagrange [1768, 24] considered this as a generic equation, where ϕ(x) is any function, and stated that if p is a root of Equation (137), then pk is given by (136) by setting x = α after the diﬀerentiation. He also stated that it is easy to see that any function f (p) of the root p can be expressed in the form (130).325 An important aspect of Lagrange’s proof is that he used the properties dk i j of dx k (x y ) to represent the terms of certain series in a compact form. i j dk Lagrange was not interested in the meaning of dx but in how x and y k x y k

d 326 combined after the application of dx k , which is understood as an operator. In this sense, it resembles Euler’s derivation of the Euler–Maclaurin formula in Institutiones calculi diﬀerentialis [1755]. Lagrange devoted a section of the Nouvelle m´ethode to convergence. In this section he clearly shows his adherence to the view commonly held by 18th-century mathematicians. Convergence was not considered in the proof. However, the actual application of the series to the solution of the equation required series to be convergent; only in this case it represented actually the values of (ﬁnite) quantity [1768, 60–61]. For this reason, Lagrange made some a posteriori considerations on convergence and proved that 1 dn ′ n =0 n! dxn f (x)ϕ (x) x=α

for n = ∞ (under certain conditions concerning the function ϕ). In Sur le probl`eme de Kepler [1769], Lagrange used his inversion theorem to provide an approximate solution to Kepler’s equation M = E − e sin E.

(138)

Kepler’s equation arose from the description of the motion of celestial bodies. It gives the relation between the polar coordinates of a planet and the time elapsed from a given initial point. In Equation (138), M denotes the mean anomaly (a parameterization of time) and E the eccentric anomaly of a body orbiting on an ellipse with eccentricity e. Lagrange showed that the solution E (E is a function of M ) is given by E = M − e sin M +

1 d 2 2 1 d2 3 3 e sin M + . . . e sin M − 2! dM 3! dM 2

In order to compute E, Lagrange wrote the power sinn M as functions of cos kM and sin kM and obtained E = M − e sin M + 325

' e2 e3 & [2 sin 2M ] − 2 3 sin M − 32 sin 3M + . . . 2 · 2! 2 3!

Lagrange explicitly proved this only in his [1798, 252–253]. The proof depended on the assumption that any function could be expanded into a power series. 326 See Panza [1992, 556–557].

20

Lagrange Inversion Theorem

237

Lagrange did not discuss the interval of convergence. The radius of convergence was determined by Laplace in his [1799–1825, 5:479]). According to Lacroix, the Lagrange inversion theorem “was an epochmaking discovery in the history of analysis with regard to the expansion of functions into series” (Lacroix [1810–19, 1:286]). By 1770 it had become the subject matter of several studies. In 1772, Lambert had already published a new proof of the theorem by deriving it from an application of Taylor series (see Lambert [1770]). Laplace generalized it to several variables in M´emoire sur l’usage du calcul aux diﬀ´erences partielles dans la th´eorie des suites [1777] and Th´eorie du mouvement et de la ﬁgure elliptique des plan`etes [1784]. In Trait´e de m`ecanique c´eleste [1799–1825], he applied the theorem to the calculation of the orbits of planets. In his Trait`e de la r´esolution des ´equations num´eriques de tous les degr´es [1798], Lagrange returned to the inversion theorem and gave a new proof of it. It was also subject of investigation by the Combinatorial School327 (see p. 286) and many other studies followed in the next century.

327 In his [1793], Rothe dealt with the problem of reversion of series, which I discussed ∞ ∞ β+nk b z = a xα+nh (a1 , α > 0), Rothe in Chapter 4. Given the equality n n n=1 n=1 λ [1793] represented an arbitrary power x as a power series in y by using the polynomial formula (see Chapter 28) to calculate the coeﬃcients of this power series. In 1795, Pfaﬀ and Rothe showed that Rothe’s results and the Lagrange inversion theorem are equivalent if the Taylor series is presupposed (see Jahnke [1993, 273]).

21

Toward the calculus of operations

Leibniz’s analogy,328 which Lagrange rediscovered in 1754,329 was the starting point of numerous signiﬁcant results. In his Sur une nouvelle esp`ece de calcul,330 Lagrange stated that du

Δu = e dx

ξ+ du ϕ+ du ζ+... dy dz

− 1,

(139)

where u(x, y, z, . . .) is a function of the variables x, y, z, . . ., ξ, ϕ, ζ, . . . are increments of these variables, and Δu = u(x + ξ) − u(x). Lagrange’s reasoning can be so summarized [For the sake of simplicity, I consider the case of a function u(x) of the sole variable x]. By changing x into du dx ξ in the 2 3 expansion ex − 1 = x1 + x2! + x3! + . . ., one can write du 1 ξ+ dx 2! If one now identiﬁes

du dx

2

du k du

ξ2 +

with

1 3!

dk. u du. ,

du dx

3

du

ξ 3 + . . . = e dx ξ − 1.

then one obtains

du 1 d2 2 1 d3 u 3 ξ+ ξ + ξ + . . . = u(x + ξ) − u(x) = Δu. dx 2! dx2 3! dx3 Hence, du

Δu = e dx ξ − 1.

(140)

λ du du du ξ+ ϕ+ +... −1 Δλ u = e dx dy dz

(141)

From (139) Lagrange derived

if λ is a positive integer, and Σλ u =

1 e

du ξ+ du ϕ+ du +... dx dy dz

−1

−λ −λ = Δ u

(142)

if λ is a negative integer. Lagrange stated that the operation by which one goes from Δu to Σλ u and Δλ u was not based on clear and rigorous principles; nevertheless, it is exact, as can be ensured a posteriori. However, it would have probably been extremely diﬃcult to provide a direct and analytical demonstration of it (see Lagrange [1772, 451]). 328

See Chapter 3, p. 50. See Lettera a Giulio Carlo di Fagnano (Lagrange [1754]). 330 See Lagrange [1772, 448–452]. 329

239

240

Successes and Problems of the Triumphant Formalism

In his paper, Lagrange showed some applications of these formulas. He considered the expansion (ew − 1)λ = wλ (1 + Aw + Bw2 + Cw3 + . . .),

(143)

where A, B, C, . . . are to be determined, and calculated the logarithmic derivative of both sides. He obtained w 1 Aw + 2Bw2 + 3Cw3 + . . . e − . = λ w e −1 w 1 + Aw + Bw2 + Cw3 + . . . Since 1 ew = = w e −1 1 − e−w

∞ i=1

i+1 w

(−1)

i

i!

−1

,

Lagrange substituted this expression into the last equation and compared the coeﬃcients. He thus found λ , 2! λ λ2 1 (λ + 1)A λ B = − + , = 2 2! 3! 24 8 1 (λ + 2)B (λ + 1)A λ2 λ3 λ C = = − + + , 3 2! 3! 4! 48 48 .... λ du By a comparison between Δλ u = e dx ξ − 1 and (143), he obtained A =

dλ u λ λ dλ+1 u λ+1 λ2 dλ+2 u λ+2 λ Δ u = ξ + ξ + + ξ dxλ 2 dxλ+1 24 8 dxλ+2 2 λ3 dλ+3 u λ+3 λ + ξ + ... + 48 48 dxλ+3 λ

(144)

Finally, Lagrange replaced λ by −λ and obtained the analogous expression for Σλ u. For λ = −1, (144) provided the Euler–Maclaurin summation formula. Lagrange did not go into detail and referred to Euler and Maclaurin. He probably considered the coincidence of the results as a proof of the correctness of the method (see Lagrange [1772, 456-457]). Lagrange went on to obtain new formulas. In [1772, 457-458], he derived du Δ2 u Δ3 u Δ4 u ξ = log(1 + Δu) = Δu − + − + ... dx 2 3 4

(145)

21

241

Toward the Calculus of Operations

from (140). This equation is to be interpreted as a series in powers of Δu where the symbols (Δu)λ are replaced by Δλ u. Then Lagrange assumed dλ u λ ξ = logλ (1 + Δu) dxλ

(146)

and obtained dλ u λ ξ = Δλ u + M Δλ+1 u + N Δλ+2 u + P Δλ+3 u + . . . , dxλ where M N P

λ = − , 2! λ 1 (λ + 1)M − = − , 2 2 2·3 (λ + 1)M λ 1 (λ + 2)N − + , = − 3 2 2·3 3·4 ....

In a similar way, he derived λ λ−1 λ−2 udx λ = u+µ u+ν u + ..., λ ξ where

λ , 2! λ 1 (λ − 1)µ , − ν = 2 2 2·3 ....

µ =

For λ = 1, he obtained Gregory’s formula: udx = u + µu + νΔu + . . . . ξ

In [1792, 663–684], Lagrange also applied similar procedures to obtain some results concerning interpolation. He considered the diﬀerences m m Tm−r , m > 1, and Δ0 = T0 (−1)r Δm = r r=0

of a given sequence Tr . He noted that T1 = Δ0 + Δ1 and Tn = (T1 )n = (Δ0 + Δ1 )n = Δ0 + nΔ1 + +

n(n − 1)(n − 2) 3 Δ + ..., 1·2·3

n(n − 1) 2 Δ 1·2

242

Successes and Problems of the Triumphant Formalism

which is Newton’s interpolation formula. Then he stated that Δ0 Δ1 = Δ0+1 = Δ1 1

and T1 = Δ0 + Δ1 = Δ0 (1 + Δ1 ). Since 1 + Δ1 = elog(1+∆ ) , he derived 1

Tn = (T1 )n = Δ0 en log(1+∆ ) . Since the law of exponents implied that Δ0 [log(1 + Δ1 )]m = Δ0 [log(1 + Δ1 )]m 1 = [Δ0 [Δ1 − (Δ1 )2 + . . .]]m 2 1 1 2 1 = [Δ − (Δ ) + . . .]m = [log(1 + Δ1 )]m , 2 Lagrange expanded en log(1+∆ Tn = Δ0 en log(1+∆

1

1)

and obtained:

)

n3 0 n2 0 Δ [log(1 + Δ1 )]2 + Δ [log(1 + Δ1 )]3 + . . . 2 6 n2 n3 = Δ0 + n log(1 + Δ1 ) + [log(1 + Δ1 )]2 + [log(1 + Δ1 )]3 + . . . . 2 6

= Δ0 + nΔ0 log(1 + Δ1 ) +

Setting P0 = Δ0 = T0

and

he derived Tn = P0 + nP1 + and

Pr = [log(1 + Δ1 )]r ,

n3 n2 (P2 )2 + (P3 )3 + . . . . 2 6

1 1 1 + Δ1 = eP1 = 1 + P1 + P2 + P3 + . . . 2 6 In this way, Lagrange had various relationships between Tn , Pn , and Δn . Lagrange’s results were the starting point for some of Laplace’s research. In his M´emoire sur l’inclinaison moyenne des orbites des com`etes, sur la ﬁgure de la Terre et sur les fonctions, Laplace recalled Lagrange’s words on the diﬃculty of justifying the analogy between positive powers and diﬀerences, and between negative powers and sums; however, he thought he had found a method that was as direct and simple as possible and that showed the reason of the analogy a priori (see Laplace [1773, 314]). For this purpose, Laplace wrote u(x + ξ) = u(x) + α(x, ξ) (147)

21

243

Toward the Calculus of Operations

[with α(x, 0) = 0, for every x]. By diﬀerentiating with respect to variable ξ and applying the theorem of mixed partial diﬀerentials, he obtained ∂u(x + ξ) ∂α(x, ξ) = . ∂x ∂ξ Hence, α(x, ξ) =

ξ 0

∂u(x + ξ) dξ. ∂x

By replacing in (147), he had u(x + ξ) = u(x) +

ξ 0

∂u(x + ξ) dξ. ∂x

(148)

Hence, du(x) ∂u(x + ξ) = + ∂x dx

ξ

∂ 2 u(x + ξ) dξ. ∂x2

(149)

He substituted ∂u(x + ξ) dξ ∂x by du(x) + dx

ξ

ξ

∂ 2 u(x + ξ) dξ ∂x2

into (148), resulting in ξ 2 du(x) ∂ u(x + ξ) + dξ dξ dx ∂x2 0 0 ξ ξ 2 ∂ u(x + ξ) du(x) ξ+ dξ. dξ = u(x) + dx ∂x2 0 0

u(x + ξ) = u(x) +

(150)

Then Laplace diﬀerentiated (149) with respect to x and obtained an expres2 sion of ∂ u(x+ξ) , which he replaced into (150). By repeating the procedure, ∂x2 he derived Δu(x) =

d2 u(x) 2 d3 u(x) 3 du(x) ξ + A1 ξ + A ξ + .... 2 dx dx2 dx3

(151)

The coeﬃcients Ai depended neither on x, u, nor ξ. Laplace did not determine the coeﬃcients (since it did not serve his purpose); however, he de facto provided a proof of the Taylor series by repeated integration. In

244

Successes and Problems of the Triumphant Formalism

the following steps, he even provided a generalization of (151). Indeed, he substituted x by x + ξ into (151). Since d(x + ξ) = dx, he wrote ∂u(x + ξ) du(x) 2 Δ u(x) = ξ (152) − ∂x dx 2 ∂ u(x + ξ) d2 u(x) 2 − ξ +A1 ∂x2 dx2 3 ∂ u(x + ξ) d3 u(x) 3 ξ + ... +A2 − ∂x3 dx3 From

dh u(x) ∂ h u(x + ξ) = + ∂xh dxh

it follows that Δ2 u(x) =

ξ 0

∂ h+1 u(x + ξ) dξ, ∂xh+1

ξ 3 ∂ u(x + ξ) ∂ 2 u(x + ξ) dξ ξ + A dξ ξ2 1 2 3 ∂x ∂x 0 0 ξ 3 ∂ u(x + ξ) +A2 dξ ξ 3 + . . . ∂x3 0

ξ

After various calculations, he obtained Δ2 u(x) =

d3 u(x) 3 d4 u(x) 4 d2 u(x) 2 ξ + B ξ + A ξ + ..., 1 2 dx2 dx3 dx4

where the numerical coeﬃcients depend neither on x, u, nor ξ. Laplace [1773, 317] stated that, in general, the following formula held: Δk u(x) =

dk+1 u(x) k+1 dk+2 u(x) k+2 dk u(x) k ξ + K ξ + K ξ + . . . , (153) 1 2 dxk dxk+1 dxk+2

where the numerical coeﬃcients Ki depended neither on x, u, nor ξ. Laplace observed that, if one took a particular function, such as u = ex , one could determine the coeﬃcients Ki and thus reobtain Lagrange’s results (see Laplace [1773, 318–321]). In this way, Laplace thought he had provided an explanation of Leibniz’s analogy. In eﬀect, he was still very far oﬀ (see Panza [1992, 595–604]). In the years that followed, however, the subject underwent a remarkable development that gave rise to the calculus of operations and to a real explanation of the analogy.

22

Laplace’s calculus of generating functions

The research into Leibniz’s analogy and the calculus of ﬁnite diﬀerences led Laplace to the calculus of generating functions.331 Laplace formulated the calculus of generating functions in his M´emoire sur les suites [1779], and later he returned to it in on several occasions, in particular in M´emoire sur diverses points d’analyse [1809] and Th´eorie analytique des probabilit´es [1812], without modifying the basic principles expounded in the ﬁrst paper. In M´emoire sur les suites, Laplace considered a function yx of t and formed the inﬁnite series y 0 + y 1 t + y2 t 2 + . . . + yx t x + . . . + y∞ t ∞

(154)

He denoted the sum of the series by u and termed the function u as the generating function of the sequence yx . Laplace speciﬁed that he was using the term “sum” to mean the function whose expansion generates the series (see Laplace [1779, 5]).332 He then stated: A generating function of any variable yx is thus generally a function of t, which, expanded according to the powers of t, has this variable yx for the coeﬃcient of tx ; and, reciprocally, the corresponding variable of a generating function is the coeﬃcient of tx in the expansion of this function according to the powers of t. (Laplace [1779, 5–6]) In modern terms, Laplace identiﬁed an operation O(yx ) = u that associates the function u to the sequence yx such that the expansion of u is yx . It follows from the deﬁnition that • if u = u(t) is the generating function of yx , then z = tr u(t) is the generating function of yx−r , where r is any positive or negative integer [namely, if O(yx ) = u(t) then O(yx−r ) = tr u(t)]. Laplace justiﬁed this theorem by observing that the coeﬃcient of tx in the expansion of z = tr u(t) is equal to the coeﬃcient of tx−r in the expansion of u(t) and, therefore, it is equal to yx−r (see Laplace [1779, 6]). He then observed that the general term of (t−1 − 1)u(t) is the diﬀerence of the general term of t−1 u(t) —which is equal to yx+1 for the previous theorem— and u(t). Therefore, 331

For a slightly diﬀerent reconstruction, see Panza [1992, 550–584]. The explicit reference to the Eulerian deﬁnition is lacking in Th´ eorie analytique des probabilit´ es, where Laplace stated that (154) “can always be conceived as a function of t, which is developed according to the power of t” (Laplace [1812, 7]). However, there are no further substantial changes, and the basic concepts on which Laplace’s calculus is grounded remain the same. 332

245

246

Successes and Problems of the Triumphant Formalism • the generating function of △yx = yx+1 − yx is (t−1 − 1)u(t). More generally, Laplace stated that • if u = u(t) is the generating function of yx , then (t−1 − 1)n u(t) is the generating function of △n yx , where △n yx = △(△n−1 yx ) (see Laplace [1779, 6]).

At this point, Laplace set ∇yx = a0 yx + a1 yx+1 + a2 yx+2 + . . . + an yx+n , where a0 , a1 , a2 , . . ., an are constants, and ∇k yx = ∇(∇k−1 yx ) and showed that • the generating function of ∇yx is u(t)(a0 + a1 t + a2 t−2 + . . . . + an t−n ), • the generating function of ∇k yx = ∇(∇k−1 yx ) is u(t)(a0 +a1 t+a2 t−2 + . . . . + an t−n )k , • the generating function of Δs ∇k yx+r is u(t)(a0 + a1 t + a2 t−2 + . . . . + an t−n )k (t−1 − 1)s tr (see Lapace [1779, 6–7]).333 Finally, Laplace denoted the inverse operation of Δ by Σ. He proved that if u = u(t) is the generating function of yx , then u(t)tn + ni=1 Ai tn−i u(t) + ni=1 Ai t−i = , (t−1 − 1)n (1 − t)n where Ai are arbitrary constants, is the generating function of Σn yx (Laplace [1779, 8]). Laplace stated that by setting aside the arbitrary constants, one could obtain the generating function of Σn yx by changing n into −n in the generating function of Δn yx (Laplace [1779, 8]). In M´emoire sur les suites and Th´eorie analytique des probabilit´es, Laplace used the above theorems in a extremely powerful way: He showed how to derive a vast number of formal identities and provided several applications. In M´emoire sur les suites,334 he observed that u(t) ti

333

i 1 = u(t) 1 + ( − 1) t i(i − 1) −1 = u 1 + i(t−1 − 1) + (t − 1)2 1·2 i(i − 1)(i − 2) −1 (t − 1)3 + . . . . + 1·2·3

∞ Laplace also generalized these results for case where ∇yx = i=0 ai yx+i and the exponents are rational or irrational numbers (see Laplace [1779, 7] and [1812, 9]). 334 See Laplace [1779, 9].

22

247

Laplace’s Calculus of Generating Functions

By passing from the generating functions to the corresponding variables, he obtained335 the well-known formula336 yx+i = yx + iΔyx +

i(i − 1)(i − 2) 3 i(i − 1) 2 Δ yx + Δ yx + . . . 1·2 1·2·3

According to Laplace, the various diﬀerent ways of expanding the power give a corresponding number of methods for interpolating the series. For example, in M´emoire sur les suites, he set t−1 = 1 + αt−r and expanded t−i as a function of α. He obtained: 1 ti

u ti

i(i + 2r − 1) 2 i(i + 3r − 1)(i − 3r − 2) 3 α + α 2! 3! i(i + 4r − 1)(i − 4r − 2)(i − 4r − 3) 4 α + ... + 4!

= u + iα +

Hence, by applying the rules of calculus, he derived the interpolation formula i(i + 2r − 1) 2 Δ yx−2r 2! i(i + 3r − 1)(i − 3r − 2) 3 Δ yx−3r + 3! i(i + 4r − 1)(i − 4r − 2)(i − 4r − 3) 4 Δ yx−4r + . . . + 4!

yx+i = yx + iΔyx−r +

(see Laplace [1779, 9–10]). Another interesting instance of Laplace’s methodology is the following. Laplace337 set z = t(t−1 − 1)2 and sought the expansion of t−i as a power series in z. He observed that t−i is equal to the coeﬃcient of θi in the expansion of 1 . 1 − θt Since

1 t

+ t = 2 + z, one had

1 1− 335

θ t

=

∞

z i θi 1 − θt 1 − θt = . = (1 − θ)2 − zθ (1 − θ)2i+2 1 − θ( 1t + t) + θ2 i=0

Laplace brieﬂy mentioned convergence: “This equation, having place whatever be i, will serve to interpolate the series of which the diﬀerences of the terms go by decreasing” [1772, 10]. 336 See p. 22 and p. 88. 337 See Laplace [1779, 11–12].

248

Successes and Problems of the Triumphant Formalism

He named Z the coeﬃcient of θi in the expansion ∞ i=0

z i θi (1 − θ)2i+2

1 1−θt of (1−θ) 2 −zθ . By expanding the fractions (1−θ)2i+2 and rearranging, Lagrange obtained ∞ (i + k + 1)2k+1 k z , Z= (2k + 1)! k=0

where (α)n = α(α − 1) . . . (α − n + 1) or Z =1+i+

∞

(1 + i)

k=1

k

j=1 [(i

+ 1)2 − j 2 ]

(2k + 1)!

zk .

(155)

After that Laplace named Z ′ the coeﬃcient of θi in the expansion of (1−θ)θ2 −zθ . He remarked that it can be obtained by Z by changing i into i − 1 in (155) 1−θt ′ and, thus, the sought coeﬃcient of θi in the expansion of (1−θ) 2 −zθ is Z −tZ . Hence, u(t)t−i = u(t)(Z − tZ ′ ). By applying the principles of his calculus, Laplace obtained the formula today named after Everett: yx+i = (1 + i)yx + −iyx−1 +

∞

(1 + i)

k

k=1 ∞ k 2 j=1 [i

i

k=1

j=1 [(i

(2k + 1)!

− j2]

(2k + 1)!

+ 1)2 − j 2 ]

Δ2k yx−k

(156)

Δ2k yx−k−1 .

Laplace subsequently transformed it and obtained the so-called Newton– Stirling formula (see Laplace [1779, 13]). In a similar way Laplace derived many interpolation formulas, such as the Laplace summation formula:

n

yx dx =

1 1 1 y0 + y1 + y2 + . . . + yn−1 + yn − [Δyn−1 − Δy0 ] 2 2 2 19 3 1 2 2 [Δ yn−3 − Δ3 y0 ] + . . . − [Δ yn−2 − Δ y0 ] − 24 720

(see [1799–1825, 4:205–208]). To end this chapter, I mention Laplace’s treatment of diﬀerence equations by means of generating functions. In [1812, 80], Laplace showed that if

22

249

Laplace’s Calculus of Generating Functions

the sequence yx is given by the ﬁnite diﬀerence equation then its generating function has the form m−1

m

n=0 Kn yx+n

= 0,

Rn tn

n=0 m

,

Kn tm−n

n=0

where Rn are arbitrary constants. He then338 stated that if the generating functions of diﬀerence equations are known, one can obtain the solutions to these equations in terms of deﬁnite integrals. Indeed, he set u(t, z) =

m

yx (z)tx ,

n=0

t = eiw , U (w, z) = u(eiw , z), π where i is the imaginary unity, and integrated −π U (w, z)e−iwx dw. Hence, π 1 yx = U (w, z)(cos xw − i sin xw)dw. 2π −π This formula has the disadvantage of introducing complex numbers. To eliminate them, Laplace supposed that b yx = t−x−1 T (t)dt, a

where the function T (t) is to be determined, as are the limits of integration. He assumed that yx was given by the equation m

an yx+n +

b a

bn yx+n = 0.

(157)

n=0

n=0

On substituting

m

t−x−1 T (t)dt for yx in (157), he obtained

b ) ( m m m b d −n −n −x −n −x−1 T (t) bn t dt = 0. bn t + an t + t t T (t) −T (t)t dt a n=0

n=0

n=0

a

Lagrange set 338

b

t a

−x−1

T (t)

See Laplace [1812, 83].

m

n=0

an t

−n

d +t dt

(

T (t)

m

n=0

bn t

−n

)

dt = 0

250

Successes and Problems of the Triumphant Formalism

and considered the diﬀerential equation ) ( m m d −n −n T (t) bn t = 0, T (t) an t + t dt n=0

n=0

which determines T (t). In order to obtain the limits of the integral, Lagrange set b m −n −x bn t = 0. −T (t)t n=0

a

Therefore, the limits = ∞ and a = tα , where tα is one of the solutions are b−n to the equation m b t = 0 (see Laplace [1812, 83–85]). n=0 n

23

The problem of analytical representation of nonelementary quantities

In the previous chapters, I illustrated some of the most remarkable results obtained by formal methodology. In this chapter and the following three chapters, I shall examine a diﬀerent group of results derived during the 18th century. They were also based on a formal methodology, although they contain within them the germ of future diﬃculties. Such diﬃculties, which emerged for the most part at the end of the historical period considered here, mainly stemmed from the fact that series theory (in the Eulerian and Lagrangian sense) was not suitable for representing and dealing with certain quantities, especially quantities linked to the investigation of certain physical phenomena. The complications involved the notion of a function and other basic concepts related to it. Therefore, I shall ﬁrst return to the concept of a function. We saw that analysis investigated quantities insofar as they were expressed by functions and that the set of functions was restricted. During the second part of the 18th century, mathematicians felt the need to investigate certain (relations between) quantities that could not be expressed using elementary functions (later I will refer to them as nonelementary transcendental functions) and sometimes —though not always339 — termed them “functions”.340 In this way the term “function” underwent various terminological shifts. This shift did not aﬀect the concept of analysis. The real subject of analysis remained elementary functions, which were to be treated by formal methodology: nonelementary transcendental functions were not considered as functions in the strict sense of term and had a status diﬀerent from the former. To justify this statement, I note that 18th-century evidence shows that the term “function” was used in three diﬀerent senses (apart from the sense of elementary functions). First, the term “function”, in certain cases, denoted certain quantities that did not have an analytical expression or that were analytically expressed by means of more than one analytical expression. 339

The expression “(transcendental) quantity” or the words “formula” and “expression” (with reference to integral functions) were often used. 340 To distinguish between diﬀerent senses of the term, the adjective “analytical” was at times added to the noun “function” in order to denote the functions that had an appropriate analytical expression and were the true object of analysis. According to Youschkevitch [1976, 75–76], Condorcet was the ﬁrst to use it in his unpublished Trait´ e du calcul. On the notion of a function and of an analytical function in Condorcet, see Gilain [1988], especially p. 103, where Gilain stated that Condorcet’s analytical functions correspond to explicit elementary functions. For Lagrange, analytical functions were all the functions studied in analysis, namely elementary functions; indeed, there are only elementary functions in his two foundational treatises, which are entitled Th´ eorie des fonctions analytiques and Le¸con sur le calcul des fonctions.

251

252

Successes and Problems of the Triumphant Formalism

Second, the term “function” was associated with certain quantities derived by Wallis’s interpolation (inexplicable functions). Third, the term “function” was associated with certain quantities that were analytically expressed by integrals or diﬀerential equations. I refer to Chapters 24 and 25 for a discussion of the second and third meaning. I shall now dwell upon the ﬁrst meaning. The word “function” was employed for the ﬁrst time to denote a relation without an analytical expression in the controversy regarding the vibrating string. In his Recherches sur la courbe que forme une corde tendu¨e mise en vibration [1747], d’Alembert described the motion of a stretched elastic string by an equation equivalent to a partial diﬀerential equation 2 ∂2y 2∂ y = a . ∂x2 ∂t2

He solved this equation and found y = f (t + x) + F (t − x), for a = 1, f and F being two arbitrary functions. D’Alembert thought that the solution to the problem had to be interpreted only by means of relations that have one only analytical expression, because the rules of the calculus had been formulated with reference to such functions. In contrast, Euler tried to eliminate this restriction in the geometric or mechanical applications but without prejudicing the nature of the calculus. In the summary of De usu functionum discontinarum in Analysi, Euler explained: The solutions that Geometers gave to the problem of the vibrating motion of strings include nothing but the assumption that the ﬁgure, which is given to the string to the beginning of the motion, is regular and can be represented by a certain equation. Instead they denied that the other case (if this ﬁgure is discontinuous or irregular) was of relevance for analysis or that the motion that originated from this conﬁguration might be reasonably deﬁned. (Euler [1765, 7]) He thought that similar problems involved the use of discontinuous functions necessarily but merely added the new G-discontinuous functions to old continuous functions, without changing the concept of the latter. Euler obtained this result by a change in terminology and a peculiar interpretation of the constants resulting from the integration of partial diﬀerential equations. In Introductio in analysin inﬁnitorum [1748a], the term “function” always denoted an analytical written expression (embodying a relation between quantities), and the word “curve” had an obvious geometrical meaning; any function could be represented geometrically by a curve; the converse was not true, since some curves were not analytically expressible. For this reason a function had to be continuous and a curve could be discontinuous. In De usu functionum discontinarum, every curve was instead viewed as

23

Analytical Representation of Nonelementary Quantities

253

analytically expressible by a function341 and Euler denoted the analytical expression by the term “equation”, while he indicated the relation between quantities by the terms “curve” and “function” (the one was often used in place of the other in the paper). In this way, Euler could introduce the notion of discontinuous functions: Curves or functions were said to be discontinuous if they were unions of more than one equation (see Euler [1765, 4–5]). After having deﬁned discontinuous functions, Euler had to explain how these new functions entered in the calculus (he indeed agreed that the calculus concerned single analytical expressions, i.e., continuous functions). He resorted to a special interpretation of the constants produced by integration. He observed that these new functions, absolutely indeﬁnite and dependent upon our discretion, originate from the integration of a function of two variables, a new and little-developed ﬁeld of the integral calculus, which “diﬀers very much from the common integral calculus, where functions of a variable only occur. It demands entirely special rules, even if it also used the devices of the ﬁrst part [of the calculus]” (Euler [1765, 20]). In his Institutiones calculi integralis,342 Euler explained that if one integrates a function X(x) of one variable x, one obtains X(x) = F (x) + C, where F (x) is a function such that dF (x) = X(x), dx and the constant C is determined by the nature of problem whose integration gives the solution. In the same way, if one integrates a function Z(x, y) of the variables x and y with respect to x, one obtains Z(x, y)dx = F (x, y) + f (y), where F (x, y) was a function such that dF (x, y) = Z(x, y)dx dx and f (y) as an arbitrary quantity dependent on y.343 The character of the quantity f (y) is determined by the nature of the problem and could even be 341

See Euler [1765, 3]. See Euler [1768–70, 3:35–37]. 343 On this question, see also Euler [1765, 20]. 342

254

Successes and Problems of the Triumphant Formalism

a quantity that is not expressible by a formula but can be thought of as the ordinate of a curve whose abscissa is y. ∂2z 2 ∂2z Thus, if one considers the wave equation ∂y 2 = a ∂x2 , by a change of variable t = x + ay, u = x − ay, one obtains ∂2z = 0. ∂t∂u By integrating with respect t, one has a function ∂z = f (u); ∂u hence, z=

h(u)du + f (t) = F (u) + f (t)

and z = f (x + ay) + F (x − ay).

(158)

According to Euler, the functions f and F could be discontinuous. Since integration naturally contains an element of arbitrariness, Euler believed that the integral calculus of functions of more than one variable could directly provide a relation, without the intermediate step of the formula. Of course, in order to give a sense to this interpretation of integration, it was necessary to explain what the diﬀerential ratio of a G-discontinuous function is. Euler merely used the geometric meaning of a function and stated that if f (x) represented a curve, then f ′ (x) was the slope of the tangent, whereas if f (x) was interpreted as an area, then f ′ (x) was a curve (see Euler [1768–70, 3:69]). In his [1768–70, 3:192–193], Euler was, however, obliged to admit that the use of an immediately geometrical notion in an analytical context gave rise to a remarkable lack. He indeed observed that if one applied (158) to the equation ∂2z ∂2z + a2 2 = 0, 2 ∂y ∂x then one obtained the complex solution √ √ z = f (x + ay −1) + F (x − ay −1). √ Euler passes to an equation having a complex coeﬃcient ±a −1 without any special hypothesis. As was customary in the 18th century, he did not appreciate the diﬀerence between complex and real analysis. An interpretation of this solution, which was obviously inﬂuenced by a weak knowledge

23

Analytical Representation of Nonelementary Quantities

255

of the conditions of diﬀerentiability of a function of a complex variable, is beyond the scope of this book. I limit √ myself to illustrate √ how Euler derived “real solutions” from z = f (x + ay −1) + F (x − ay −1) provided f and F were continuous. He indeed observed that √ if f and F are continuous, then they can be reduced to the form P ±Q −1. Hence, it is easy, he said, to obtain solutions in the real form344 √ √ √ √ 1 1 z = [f (x+ay −1)+f (x−ay −1)]+ √ [F (x−ay −1)−F (x−ay −1)]. 2 2 −1 (159) √ ∂2z ∂2z 2 He probably realized that if P ± Q −1 satisﬁes ∂y2 + a ∂x2 = 0, then √ √ P + Q = Re[f (x + ay −1) + F (x − ay −1)] √ √ + Im[f (x + ay −1) + F (x − ay −1)] + 1* = f (ω) + f (ω) + F (ω) + F (ω) 2 + 1 * + √ f (ω) − f (ω) + F (ω) − F (ω) 2 −1

√ also does (here, I take ω = x + ay −1 and denote the conjugate, real part and imaginary part of the complex number ω by ω, Re(ω), Im(ω), respectively). Euler assumed that, for every continuous function, h(ω) = h(ω); therefore, P +Q =

1 [f (ω) + f (ω) + F (ω) + F (ω)] 2 1 [f (ω) − f (ω) + F (ω) − F (ω)] . + √ 2 −1

Since f and F are two generic continuous functions, the latter expression furnished (159). Euler justiﬁed the equality h(ω) = h(ω) as follows. Setting x = s cos φ and ay = s sin φ, one has √ √ (x ± ay −1)n = sn (cos nφ ± −1 sin nφ), and since h is a continuous functions, namely it is composed of analytical (algebraic or elementary transcendental) operations, its values can be exhibited by means of the sine and cosine (Euler assumed that every continuous function can be expanded in power series with real coeﬃcients). If f and F are discontinuous, then they cannot be reduced to a real form: 344

See Euler [1768-70, 3:192].

256

Successes and Problems of the Triumphant Formalism In any curve traced by a free stroke of the hand, what meaning will one give the ordinates corresponding to the abscissas √ √ x + ay −1 and x − ay −1 according to the nature of imaginaries and their real sums [the real part of their sums] or the diﬀerence which will also be real √ if it is divided by −1? Therefore we note this not slight lack of calculus, for which one can make up in no way yet. (Euler [1768–1770, 3:193])

Despite this fact, Euler’s solution to the problem of the vibrating string was substantially accepted in the 18th century. G-discontinuous functions were considered as tools that made up for a local insuﬃciency of analysis, just as imaginary quantities made up for local insuﬃciencies of real quantities. With hindsight, the controversy of the vibrating string posed the question of the lack of analytical tools for describing certain more complicated phenomena: It actually showed the restricted nature of 18th-century analysis and its overall inadequacy for more sophisticated investigations. Mathematicians had to resort to a direct geometrical interpretation in order to give meaning to discontinuous functions, so one of the presuppositions of Euler’s and Lagrange’s analysis (analysis as an autonomous theory) failed. To avoid a “return to geometry”345 and to make G-discontinuous functions true analytical objects, it was necessary to restructure analysis, but this did not happen. The result was that the discontinuous functions were on the margins of analysis, which remained grounded upon single analytical expressions.

345

See Grattan-Guiness [1970, 11].

24

Inexplicable functions

The second type of quantities to which the name “function” was given was inexplicable functions. Euler termed inexplicable those functions having neither a determinate expression nor an expression by means of an equation (namely, an implicit algebraic x xexpression). He actually considered the sums n=1 an and studied the functions obtained by n=1 an and the products giving nonintegral values to x. This was eﬀectively a new form of the old problem of Wallis’s interpolation. In Institutiones calculi diﬀerentialis, Euler attempted to establish a theory of inexplicable functions and mainly aimed to determine the diﬀerentials of these functions. He ﬁrst operated in general — in other words, he de346 termined the form x x of diﬀerentials of the general function of the kinds n=1 an ; he then applied the obtained results to particular n=1 an and cases. The two most important particular cases were 1+

1 1 1 1 + + + ... + 2 3 4 x

and 1 · 2 · 3 · . . . · x, and, while investigating these two cases, Euler derived some interesting series expansions of digamma347 and gamma functions [such as (161), (162) and (168)]. Euler348 set S(x) = xn=1 ax and T (x) = S(x + ω) and considered the sequence S(x), S(x + 1), S(x + 2), S(x + 3), . . . 346

See formulas (160), (164), and (165). The inexpicable function H(x) = 1 + 12 + 31 + 14 + . . . + function for a constant. The digamma function is deﬁned as 347

Ψ(z) =

1 x

diﬀers from the digamma

d log Γ(z), dz

where Γ(z) is the gamma function, or as F (z) =

d log z!, dz

where z! is the factorial function. It is well known that F (z) = Ψ(z + 1) = H(ω) = −γ −

∞

n=1

z , n(n + z)

where γ is the Euler constant. A comparison with (161) shows that Ψ(n) = −γ − H(n − 1). 348

See Euler [1755, 2: Sections 369-374].

257

258

Successes and Problems of the Triumphant Formalism

Since S(x + n) − S(x + n − 1) = ax+n , if one assumes that the numbers ax+n converge to a number L, one has S(∞ + 1) − S(∞) = a∞+1 = L,

S(∞ + 2) − S(∞ + 1) = a∞+2 = L,

...,

i.e., S(∞), S(∞ + 1), S(∞ + 2), S(∞ + 3), . . . are an arithmetic progression. Consequently, S(∞ + ω) − S(∞) = L · ω = a∞+1 · ω. Since T (∞) = S(∞ + ω) = S(∞) + ωa∞+1 , one can write T (∞) = S(x) +

∞

n=1

As T (∞) = T (x) +

∞

n=1 ax+n+ω ,

ax+n + ωa∞+1 .

one has

S(x + ω) = S(x) + ωa∞+1 +

∞

n=1

ax+n −

∞

ax+n+ω .

(160)

n=1

If ω is inﬁnitesimal, S(x + ω) − S(x) is the expression of the diﬀerential dS of the function. This aspect of (160) attracted Euler’s attention, not the possibility that one could obtain a representation of the function S(x) by means of an inﬁnite series from (160). The inﬁnite expression of S(x) was only viewed as a tool for deriving the diﬀerential of S(x) or for applying the function to the problem of interpolation. In no case did Euler conceive of the possibility of deﬁning S(x) by means of (160) or (165). For instance, in Section 371, he considered H(x) = 1 + 12 + 13 + 41 + . . . + x1 and derived H(x + ω) = H(x) + = H(x) + = H(x) + = H(x) +

∞

∞

1 1 − x+n x+n+ω

n=1 ∞

ω ω h−1 (−1)h−1 x+n (x + n)h

n=1 ∞

n=1 ∞

n=1

ω (x + n)(x + n + ω) ∞

h=1

(−1)h−1 ω h

∞

n=1

h=1

1 . (x + n)h+1

For ω = dx, he had dH =

∞ h=1

(−1)h−1 dxh

∞

n=1

1 . (x + n)h+1

24

259

Inexplicable Functions

Only subsequently, in Section 372, did he derive the inﬁnite expression of H(x) in order to interpolate 1 + 21 + 13 + 41 + . . . + x1 . Since H(0) = 0, he obtained the expansions H(ω) =

∞

n=1

and H(ω) =

∞

ω n(n + ω)

h−1 h

(−1)

ω

∞

n=1

h=1

(161)

1 . nh+1

(162)

These provide the sum H, even if ω is not an integer, but they do not deﬁne H(x). [In Euler’s opinion, the deﬁnition of H(x) was: H(x) is the quantity interpolating 1 + 12 + 13 + 41 + . . . + x1 .] For a general inexplicable function S(x), Euler considered the Taylor series of an+x+ω = a(n + x + ω) =

∞ ω h dh a(x + n) h=0

h!dxh

,

for n = 0, 1, 2, 3, . . .

(he applied the methodology already used to seek the summation formula349 ) and obtained S(x + ω) = S(x) + ωa∞+1 + = S(x) + ωa∞+1 − = S(x) + ωa∞+1 − As a∞+1 = a(1) +

∞

h=1 [a(h

∞

a(x + n) − a(x + n + ω)

n=1 ∞ ∞

ω h dh a(x + n) h!dxh

h=1

n=1

n=1 h=1 ∞ ∞ h

ω h!

dh a(x + n) . dxh

+ 1) − a(h)], he obtained

∞ ∞ ω h dh a(x + n) S(x + ω) = S(x) + ωa(1) + ω [a(h + 1) − a(h)] − , h! dxh n=1 h=1 h=1 (163) which provides the “complete diﬀerential”

dS = a(1)dx + dx

∞

∞ h=1

349

See Chapter 14.

[a(h + 1) − a(h)] −

∞ ∞ dh a(x + n) h=1 n=1

h!

(164)

260

Successes and Problems of the Triumphant Formalism

Euler later derived another expression of dS using the power series of S(x). Setting x = 0 and ∞ 1 dh a(x + n) Gh = , h! dxh n=1

we have

S(ω) = ωa(1) + ω

∞ h=1

∞

[a(h + 1) − a(h)] −

Gh ω h ,

h=1

and, changing ω into x, we obtain S(x) = xa(1) +

∞ h=1

∞

[a(h + 1) − a(h)] x −

Gh xh .

(165)

h=1

Euler generalized these achievements (derived subject to the condition that, for n = ∞, the sequence an becomes a constant L), considering the case in which the second or third diﬀerences of S(∞), S(∞+1), S(∞+2), . . . equal 0. He then reduced the products S(x) = xn=1 ax to sums by means of logarithms. Under the conditions log a∞ = 0 and log a∞+1 − log a∞ = 0, he derived S(x + ω) = S(x)

∞ an+x an+x+ω

and S(x + ω) = S(x)aωx+1

∞ aωx+n+1 a1−ω x+n , an+x+ω

n=1

n=1

respectively. For x = 0, one has S(0) = 1; changing ω into x, one obtains S(ω) =

∞ an an+x

and

S(ω) = ax1

n=1

∞ axn+1 a1−x n an+x

n=1

(see Euler [1755, 2: Sections 381–382]). The latter may be applied to G(x) = 1 · 2 · 3 · . . . · x (in this case, the ﬁrst diﬀerence of the logarithms of the terms whose index is inﬁnite is 0) to obtain (82) (see Euler [1755, 2: Section 402]). For log S(x) = log

∞

n=1

an =

∞

log an

n=1

and log a∞+1 − log a∞ = 0, (163) is transformed into log S(x + ω) = log S(x) + ω log a(1) + ω

∞ h=1

−

∞ ∞ ω h dh log a(x + n) h=1

h!

n=1

dxh

log

a(x + h + 1) (166) a(x + h)

24

261

Inexplicable Functions

and then ∞

∞

∞

a(x + h + 1) dh log a(x + n) dS = dx log a(x + 1) + dx − log S a(x + h) h! h=1 n=1

h=1

(see Euler [1755, 2: Section 385]). Euler applied this formula to G(x) = 1 · 2 · 3 · . . . · x to obtain ∞ ∞ ∞ 1 x+h+1 dxh dG log (−1)h = dx log(x + 1) + dx + . G x+h h (x + n)h n=1 h=1 h=1 (167) He did not write the inﬁnite expression of log G(x) by deriving it from (166); he merely sought the diﬀerential of log G(x) and did not consider the expression of log G(x) of importance. Finally, from (165), Euler derived

log S(x) = x log a(1) + x

∞ h=1

where Gh =

1 h!

∞

n=1

dh log a(x+n) dxh

log [a(h + 1) − a(h)] −

∞

Gh xh ,

h=1

and log a∞+1 − log a∞ = 0, and then

∞

∞

h=1

h=1

dS hGh xh−1 dx. log [a(h + 1) − a(h)] − = dx log a(1) + dx S If G(x) = 1 · 2 · 3 · . . . · x, he obtained log G(x) = ∞

∞

h=1

(−1)h h h Ch x

dG (−1)h Ch xh−1 , = G

and (168)

h=1

1 where C1 is Euler’s constant and Ch = ∞ n=1 nh , for h = 2, 3, 4, . . .. However, even in this case, the power series of log G(x) is only an intermediate step for arriving at (168) and in particular for calculating C1 .350 350

By applying the sum formula (98) to the inexplicable functions S(x) = Euler [1755, 2: Sections 386–388] inferred dS = a(x)dx +

x

n=1

an ,

∞ B2n d2n−1 a(x) 1 da(x) + da(x). (−1)n−1 2 (2n)! dx2n−1 n=1

As a particular case, he considered G = 1 · 2 · 3 · . . . · x and, using the logarithms, derived ∞

B2n dx dx dG (−1)n−1 = log xdx + − G 2x n=1 2n x2n The latter is more suitable than (167) or (168) for calculating dS/S when x is very large.

262

Successes and Problems of the Triumphant Formalism

Euler was not entirely satisﬁed by this exposition of the theory of inexplicable functions and later tried to clarify it in his Delucidationes in capita postrema calculi mei diﬀerentialis de functionibus inexplicabilibus (see Euler [1787]). Despite his eﬀorts, during the second half the 18th century, inexplicable functions did not actually enter into analysis in the same ways as the other known transcendental (logarithmic and trigonometric) functions.351 The reasons for this are closely connected with the limits of the formal methodology, which I will discuss in Chapter 29.

351

See Chapter 25.

25

Integration and functions

I shall now go on to examine the case in which the term “function” was associated with certain quantities that were analytically expressed in a closed form by integrals. The stimulus for the investigation of such quantities stemmed mainly from geometry and applied mathematics and from the attempt to mathematicize nature. For instance, probabilists ran up against gamma and beta functions, elliptic integrals arose from questions regarding physics and geometry, while astronomy in general required new instruments other than analytical expressions and their power series, etc. In the last 30 years of the century, nonelementary transcendental functions became an important ﬁeld of research in analysis. Many results were derived; although nonelementary transcendental functions were not considered well known enough to be accepted as true functions. For example, in his Institutiones calculi integralis, Euler argued that the solution to an integral f (x)dx could be a transcendental function, although only logarithmic and trigonometric functions — among all transcendental functions — were to be placed on the same plane as algebraic functions (see Euler [1768-70, 1:14]). He investigated various nonelementary transcendental functions in this treatise; the ﬁrst was z dz . log z 0 z dz According to him, if the integral 0 log z could be assigned, it should have been of a very wide use in analysis (Euler [1768-70, 1:122]. It merited a more careful investigation, however, for the time, “the nature of this function is not known enough” (Euler [1768-70, 1:128]). Similarly, in his Sur l’int´egration de quelques ´equations diﬀ´erentielles, Lagrange suggested that investigation of integrability of such expressions as √

a + bx +

cx2

1 + cx3 + dx4 + . . .

opened a vast ﬁeld to research (see Lagrange [1766, 33]). However, as Fraser noted,352 Lagrange limited himself to stating the existence of this possibility, which was not developed in his work and, in eﬀect, such expressions did not enter into the list of his analytical functions. This approach can be better understood if we brieﬂy examine the 18thcentury concept of integration. Leibniz regarded integration as summation of inﬁnitesimals. In the geometrical context of the ﬁrst years of the calculus, integration was an operation that, given the analytical expression of a geometrical quantity, made it possible to determine the analytical expression of another quantity, the existence of which was geometrically self-evident (e.g., 352

See Fraser [1987, 40].

263

264

Successes and Problems of the Triumphant Formalism

given the ordinate of a curve, integration made it possible to calculate the area or the length of the curve — the area and the length existed on the basis of geometrical evidence). In the second part of 18th century, integration was usually deﬁned as the inverse operation of diﬀerentiation or derivatio,353 namely that the integral f (x)dx of the function f (x) was a function F (x) such that dF = f (x)dx. For instance, Lagrange gave the name “direct analysis of functions” to that part of analysis that investigated the rules by which it was possible to move from a function to its derivatives of diﬀerent orders; instead, the part of the calculus that scrutinized the return from derivatives to their primitive function was termed the “inverse analysis of functions” (see Lagrange [1797, 140–141]). Euler regarded diﬀerentiation as the direct operation while he considered integration to be the inverse operation by which one returned from the diﬀerential to the function generating the diﬀerential (see Institutiones calculi integralis, [1768–70, 1:5]). Similarly, Lacroix stated that the integral calculus was the inverse of the diﬀerential calculus and consisted of returning from diﬀerential coeﬃcients to functions from which they were derived (see Lacroix [1797–1800, 2:1]). Mathematicians were aware that many simple functions could not be integrated by means of elementary functions and that this concept of integration posed the problem of nonelementarily integrable functions. Euler and Lagrange brieﬂy mentioned the question in their treatises. They compared integration with inverse arithmetical operations. According to Lagrange, there were certain cases in which the operations of division and extraction of a root could not be performed in an exact manner but only by approximation (Lagrange [1797, 141]). Similarly, inverse analysis did not always succeed in ﬁnding primitives, although it was still possible to ﬁnd an approximate solution to diﬀerential equations (by series) when this occurred. According to Euler, integration led to new transcendental “quantities” in the same manner as the operations of subtraction, division, and extracting a root led to negative, rational, and irrational numbers (Euler [1768-1770, 1:13]). This analogy was signiﬁcant for the status of transcendental quantities. Indeed, we saw that only integers and fractions were numbers in the strict sense of the term.354 The extension of the term “number” to incommensurable ratios was considered incorrect because “number” presupposed an exact denotation. Nevertheless, incommensurable ratios were similar to numbers and could therefore be viewed as numbers; in fact, √ (a) an irrational number could be handled as a symbol; e.g., 2 was a √ 2 symbol such that 2 = 2, 353

The notion of integration as the operational inverse of diﬀerentiation and the signiﬁcance of this conception within 18th-century analysis are discussed in Fraser [2003]. 354 See Section 7.2, p. 106.

25

Integration and Functions

265

(b) it could be approached by numbers as desired, (c) even though it could not be represented rigorously by means of arithmetic, it at least had a geometrical meaning.355 In the same way as irrational numbers were not true numbers, transcendental quantities were not functions in the strict sense of the term and diﬀered from elementary functions, which were the only genuine object of analysis. Nevertheless, transcendental quantities had properties that allowed one to view them as entities similar to functions; in fact, (α) transcendental quantities f (x)dx arose from the integration of an elementary function f (x) and could be manipulated formally under the condition d[ f (x)dx] = f (x)dx so as to transform f (x)dx into other integrals or into inﬁnite expressions, (β) even when the values of these transcendental quantities were not already given and tabulated, they could be computed in an approximate way by resorting to series or other techniques, (γ) integrals could be thought of as geometrical (or physical) quantities:356 so their existence at least had a geometrical visualization or physical basis.

355 356

See d’Alembert [1773, 5:217–218]. See Euler [1768–1770, 2:220].

26

Series and diﬀerential equations

In this chapter, I wish to investigate some important results concerning the series solution to diﬀerential equations which were derived in the second part of the 18th century. These results were of great importance for the subsequent development of mathematics since, while seeking the series solution to diﬀerential equations, mathematicians came up against many series that were not the expansions of elementary functions. Looking back with the advantage of hindsight, we might state that they gave a de facto treatment of many new transcendental functions. However, this is only an a posteriori interpretation. We saw that mathematicians were aware that certain expansions did not derive from elementary functions; however, they did not modify their approach and, in particular, did not abandon the use of the formal methodology, which had its foundations in elementary functions. They continued to avoid considering series as autonomous objects (as Lagrange still stated in 1797, “an expression in series can always be regarded as the development of a ﬁnite expression” (Lagrange [1797, 93])). Series solutions were only viewed as an instrument for dealing with certain quantities expressed in the form of diﬀerential equations. They had an inferior status to solutions in a closed formed (which were elementary functions or nonelementary functions given in integral form). Thus, while it is sometimes possible to ﬁnd the term “function” associated with quantities given by integral formulas, it is never associated with quantities given by series. Mathematicians resorted to series solutions when they were unable to integrate a function f (x) or a diﬀerential equation F (x, y, dy/dx, . . .) = 0 in other ways. In this case, they applied the usual procedures that enabled them to expand elementary functions into integrals and diﬀerential equations.357 Operating formally, one could determine the series solution an xn that was a useful instrument in the investigation of the quantities expressed by integrals f (x)dx or diﬀerential equations F (x, y, dy/dx, . . .) = 0. I emphasize that this did not mean that the solution to a given diﬀerential ∞ xn equation was deﬁned by the series.358 In the same manner as n=0 n! x represented the quantity allowed one to investigate it but did not e and deﬁne it, a power series an xn represented a quantity of the types f (x)dx or F (x, y, dy/dx, . . .) = 0 and allowed one to investigate it but did not deﬁne it. When they were not reducible to closed expressions, series solutions to diﬀerential equations played two roles. 357

Usual procedures could be applied since integrals and diﬀerential equations were expressions of the type f (x)dx or F (x, y, dy/dx, . . .) = 0, where f and F were elementary functions. 358 The solution is deﬁned by the diﬀerential equation, but this did not mean that it was a function in the proper sense of the term (see Chapter 23, p. 251).

267

268

Successes and Problems of the Triumphant Formalism

First, series were instruments that provided the approximate values of a quantity expressed by a diﬀerential equation. It was commonplace that a solution by series was not an exact solution. According to Lagrange [1776, 301], the method of series was a method “for integrating by approximation the diﬀerential equations whose ﬁnite integral was impossible or, at the very least, extremely diﬃcult.” And, in his [1780, 522–523], Euler regarded the representation of a quantity by a series as an approximate representation. Series did not provide the exact solution and did not express a quantity exactly. Second, series could be instruments for expressing a link between diﬀerent analytical expressions (in this role, convergence was not of importance and even totally divergent series359 could be used). An example of the ﬁrst role is the Eulerian treatment of the diﬀerential equation x2 (a + bxn )

dy d2 y + x(c + exn ) + (f + gxn )y = 0. dx2 dx

(169)

In Institutiones calculi integralis [1768–1770, 2:177–185], he tried a solution in the form ∞ Aj xλ+jn , (170) y= j=0

where A0 = 0. By replacing (170) into (169), he found β 0 A0 +

∞

(αj−1 Aj−1 + β j Aj )xλ+jn = 0,

j=1

where β 0 = λ(λ − 1)a + λc + f ,

α0 = λ(λ − 1)b + λe + g, αj

βj

= jn(jn + 2λ − 1)b + jne + α0 (for j > 0), = jn(jn + 2λ − 1)a + jnc (for j > 0).

Hence, λ(λ − 1)a + λc + f = 0 and β j Aj = αj−1 Aj−1 (for j > 0). If one chooses A0 arbitrarily, the previous relations allow one to determine λ by solving the equation λ(λ − 1)a + λc + f = 0 and Aj (for j > 0) by recurrence. 359

See p. 191.

26

269

Series and Diﬀerential Equations

Therefore, if a = 0, we can determine two values λ1 and λ2 of λ, and so, we have two series of the form (169), which furnish complete solutions to the diﬀerential equation. However, if λ(λ − 1)a + λc + f = 0 has only one root or when the diﬀerence between the two values of λ is divisible by n,360 the general integral cannot be expressed as the sum of two series of the form (169). In this case Euler found that the general integral had the form log x

∞

λ1 +jn

Cj x

+

∞

λ2 +jn

Bj x

Dj xλ1 +jn ,

j=0

j=0

j=0

+

∞

where λ2 ≤ λ1 and the coeﬃcients Aj , Bj , Cj depend on two arbitrary constants. An example of the second role of series is found in Specimen transformationis singularis serierum [1794a], where Euler showed that the solution of d2 y dy x(1 − x) 2 + [γ − (α + β + 1)x − αβy = 0 (171) dx dx was the hypergeometric series α(α + 1)β(β + 1) 2 α(α + 1)(α + 2)β(β + 1)(β + 2) 3 αβ x+ x + x +.... 1·γ 1 · 2 · γ(γ + 1) 1 · 2 · 3 · γ(γ + 1)(γ + 2)

1+

By using this fact and the relation between the hypergeometric series and certain appropriate expansions of and

π

π 0

(1 + α2 − 2α cos φ)n cos pφdφ

(1 + α2 − 2α cos φ)−n−1 cos pφdφ,

Euler proved the following equality between integrals π n+p (1 + α2 − 2α cos φ)n cos pφdφ (172) (1 − α2 )−n p 0 π p−n−1 (1 − α2 )n+1 = (1 + α2 − 2α cos φ)−n−1 cos pφdφ. p 0 ∗ ∗ 360

∗

In this case the coeﬃcients β j become equal to inﬁnity for one of the two roots.

270

Successes and Problems of the Triumphant Formalism

I would now like to examine some of the more interesting examples of series connected in some way with integral or diﬀerential equations. In his Theoremata de oscillationibus corporum ﬁlo ﬂexili connexorum et catenae verticaliter suspensae, Daniel Bernoulli enunciated some theorems on the oscillations of heavy chains. In one of these361 he considered a uniform heavy ﬂexible chain AC of length l, ﬁxed at the upper end A and free at the lower end C. He supposed that when the chain was slightly disturbed from its position of equilibrium in a vertical plane, it underwent small oscillations. Denoted the new position of the chain by AM F and F M by x, Bernoulli took the maximum distance of the endpoint F from the vertical line to be equal to 1. He asserted that, in this case, the oscillation y of the point M from the vertical line is given by the series 1−

x3 x4 x5 x2 x + − + ..., + 2− f 4f 4 · 9f 3 4 · 9 · 16f 4 4 · 9 · 16 · 25f 5

(173)

where f is the solution of the equation362 1−

l2 l3 l4 l5 l + 2− + − + ... = 0 f 4f 4 · 9f 3 4 · 9 · 16f 4 4 · 9 · 16 · 25f 5

(174)

Bernoulli also stated that f is approximately equal to 0.691l and that f can assume inﬁnitely other values [namely, Equation (174) has an inﬁnite number of zeros]. He gave a proof of this theorem later in [1734-35]. In De oscillationibus minimis funis libere suspensi [1781a], Euler dealt with the same problem. He considered the forces acting on an element of the chain of length dx and obtained the partial diﬀerential equation, which, using modern symbols, is written as ∂ ∂y ∂2y =g x , (175) ∂t2 ∂x ∂x where y is the horizontal displacement of a point P of chain at the time t, x is the height of P above the bottom of the chain in its undisturbed position, and g is the gravitational constant. He assumed , the oscillation y was essentially sinusoidal with angular frequency ω = fg and wrote

g y = Av sin ξ + t , (176) f where A and ξ are constants and v = Φ fx is an appropriate function of the only variable x. Substituting (176) into (175), he found that Φ fx is a 361 362

See D. Bernoulli [1732–33, 116]. f is the length of the simple equivalent pendulum.

26

271

Series and Diﬀerential Equations

solution of the ordinary diﬀerential equation d dv v x + = 0. dx dx f

(177)

In this way, Euler again found Bernoulli’s solution (173), which he wrote in the form v =1−u+

u3 u4 u5 u2 − + − + ..., 4 4 · 9 4 · 9 · 16 4 · 9 · 16 · 25

where u = fx . Later Euler dealt with the series solution of Equation (177) in [1768-80, 2: Section 977] and in [1781b].

Today we (173) as a Bessel function J0 (2 nx ) of order zero and

recognize argument 2 nx ; however, I emphasize that for Bernoulli and Euler (173) was merely a tool for obtaining an approximate solution to a problem of physics. The determination of f gives the solution to Equation (174) in terms of l . In [1781a], Euler found that the three smallest roots are fl = 1.445795, f l l l f = 7.6658, f = 18.63 (the values of f are suﬃciently accurate, especially n un the ﬁrst). In order to ﬁnd the roots of ∞ n=0 (−1) (n!)2 = 0, Euler used a technique already employed in [1734–35a]. He assumed that ∞ ∞ n u n u 1− . = (−1) an (n!)2 n=1 n=0 Hence, ∞ n u d log (−1)n du (n!)2 n=0 ∞ d n un (−1) 2 n=0 du (n!) ∞ = n un (−1) n=0 (n!)2 ∞ u 1− = log an n=1

∞ ∞ um 1 =− a −u (an )m+1 n=1 m=0 n=1 n ∞ ∞ ∞ 1 m u = − σ m+1 um , = − (an )m+1

= −

∞

n=1

m=0

where σ m+1 = d du

∞

∞

1 n=1 (an )m+1 .

un (−1) (n!)2 n=0 n

m=0

Therefore,

∞

un =− (−1) (n!)2 n=0 n

∞

m=0

σ m+1 um .

272

Successes and Problems of the Triumphant Formalism

By applying the method of indeterminate coeﬃcients, he found 1 1 11 19 473 σ 1 = 1, σ 2 = , σ 3 = , σ 4 = , σ 5 = , σ6 = ,... 2 3 48 120 4320 If a1 is the smallest of the root and 0 < a1 < a2 < a3 < . . ., then σm 1 . < a1 < 1/m σ (σ m ) m+1 m At this point, Euler computed the ﬁrst values of (σ 1)1/m and σσm+1 and m derived a1 = 1, 445795. Using similar reasoning, he found a2 and a3 (see Euler [1781a, 317–323]). Another problem, which was signiﬁcant to our purpose, was that of the vibrations of a stretched membrane. Euler investigated it in De motu vibratorio tympanorum [1764] and derived an equation equivalent to

1 ∂2z ∂ 2 z 1 ∂z 1 ∂2z + 2 = 2+ , 2 2 c ∂t ∂r r ∂r r ∂ϕ2

(178)

where z is the transverse displacement at time t at the point whose polar coordinates are (r, ϕ) and c is an appropriate constant. He assumed that the solutions had the form z = u(r) sin(ωt + A) sin(κϕ + B), where ω, A, κ, B are constants. By replacing u(r) sin(ωt + A) sin(βϕ + B) in (178), he derived the equation d2 u 1 du κ2 2 + ) = 0, + (α − dr2 r dr r2

(179)

2

where α = ωc2 . Euler assumed the existence of a power series solution of this equation and obtained u(r) =

∞

n=0

(−1)n αr κ+2n n!(κ + 1)n 2

(180)

(see Euler [1764, 344–359]).363 Series were crucial for the development of planetary mechanics, a very important subject in 18th-century mathematics.364 For instance, in his Recherches sur l’attraction des sph´ero¨ıdes homog`enes [1785], Legendre investigated the attraction of ellipsoids and, in particular, showed that if the 363 Today Equation (179) is called Bessel’s equation, and the solution u(r) is the Bessel function Jβ (αr). I again emphasize that Euler considered (180) as a tool for obtaining an approximate solution to a problem of physics. 364 On this topic, see Todhunter [1873] and Wilson [1980] and [1985].

26

273

Series and Diﬀerential Equations

attraction of a solid of revolution S is known for every external point that is on the prolongation of the axis, then it is known for every external point P . In the proof, Legendre employed power series as an intermediate step. Using more modern symbols and the potential function V , his proof can be summarized as follows. Let (r, θ, ϕ) and (ρ, Θ, Ψ) be the spherical polar coordinates of the attracted particle365 and of the element of the attracting body S, respectively. The potential function is V =

ρ2 r

−1/2 ρ2 ρ 1 − 2 cos γ + 2 sin ΘdΘdωdρ, r r

where cos γ = cos θ cos Θ+sin θ sin Θ sin ω and ω is the diﬀerence of longitude of the attracted point and the attracting element. If one sets 1 (1 − 2xα −

α2 )1/2

=

∞

Pn (x)αn ,

n=0

−1/2 namely Pn (x)denotes the coeﬃcient of αn in the expansion of 1− 2xα−α2 in ascending powers of α, namely one can write 2 ρ ρ2 ρ3 ρ 1 + P1 (cos γ) + P2 (cos θ) 2 + P3 (cos θ) 3 + . . . V = r r r r · sin ΘdΘdωdρ. The coeﬃcients Pn (x) are today named Legendre’s polynomials.366 If one now supposes that the body is symmetrical with respect its equator and integrates with respect to ρ between the limits −s and s, where s(Θ) is the radius vector of the solid corresponding to a colatitude Θ, then one obtains 2 s2 s4 1 s + P2 (cos θ) 2 + P4 (cos θ) 4 + . . . sin ΘdΘdω. V =2 r 3 r r Since cos γ = cos θ cos Θ + sin θ sin Θ sin ω, Pn is a function of θ,Θ, and ω. Legendre shows that if one integrates with respect to ω from 0 to 2π, Pn is a function of θ and Θ. It is of the form 2πfn (cos θ)fn (cos Θ) for an appropriate function f . By using such properties, one obtains a series whose ﬁrst term depends only on r and where the nth term is 4π f2n (cos θ) (2n + 3)r2n+1 365 366

π/2

f2n (cos Θ)s2n+3 sin ΘdΘ,

One can assume ϕ = 0 for the particular symmetry of the problems. On Legendre’s functions, see Todhunter [1873].

274

Successes and Problems of the Triumphant Formalism

where f2n (cos θ) is a known function that is independent of the form of the body and is diﬀerent from zero for θ = 0. If the attraction is known at all points that are all on the prolongation of the axis, V is known for all such points. Therefore, the integrals π/2 I2n = f2n (cos Θ)s2n+3 sin ΘdΘ 0

are known. Consequently, the attraction is also known for every external point P . In the same period, Laplace was working on similar questions. In particular, in Th´eorie des attractions des sph´ero¨ıdes et de la ﬁgure des plan`etes [1785], Laplace considered what today is known as Laplace’s equation: ∂2V ∂2V ∂2V + + = 0, ∂x2 ∂y 2 ∂z 2 which he expressed in the form $ % 1 ∂2V ∂ 2 (rV ) ∂ 2 ∂V (1 − µ ) + + r = 0, ∂µ ∂µ 1 − µ2 ∂ω 2 ∂r2

(181)

where V is the potential function, r, ω, θ are spherical polar coordinates, and µ = cos θ [1785, 362]. Laplace sought a series expansion for V . If the point is external, he assumed that the expansion is of the type V (r, ω, θ) =

∞ Ui . ri+1

(182)

i=0

Laplace substituted this expression for V into in (181) and obtained $ % 1 ∂ 2 Ui ∂ 2 ∂Ui (1 − µ ) + + i(i + 1)Ui = 0. (183) ∂µ ∂µ 1 − µ2 ∂ω 2 He succeeded in giving an integral form to Ui by using Legendre’s polynomial [1785, 362–369]. To calculate the potential of spheroids that diﬀer only slightly from spheres, Laplace [1785, 371–373] considered the equation r = a(1 + αy), where α is a small constant coeﬃcient and y is a function of ω and µ (a is the radius of a sphere). He showed that y(ω, µ) can be developed into a series ∞ Y0 , (184) y= i=0

* + ∂ i (1 − µ2 ) ∂Y where the functions Yi satisﬁed the equation ∂µ ∂µ + i(i + 1)Yi = 0, and that 2i + 1 Ui . Yi = 4παai+3

1 ∂ 2 Yi 1−µ2 ∂ω 2

+

27

Trigonometric series

The formal methodology, with regard to series, had been conceived with reference to power series. Power series are a particularly simple type of series, especially as regards the relationship between formal manipulations and convergence. In particular, an interval of convergence Ix0 can be associated with a power series an (x − x0 )n .

The whole theory of ordinary power series was grounded upon this fact, which excluded the appearance of pathological phenomena, and on the assumptionthat, if an (x−x0 )n converged to f (x) near x0 , it was possible to identify an (x − x0 )n with f (x) in analytical calculations, without further regard for the interval of convergence. The situation was certainly more complicated for other types of series and, in particular, for trigonometric series. Trigonometric series began to appear in some mathematical and physical investigation from the 1730s onwards. Astronomy was the ﬁeld where the need for such series was most strongly felt; indeed, they seemed well suited to describing periodic phenomena, clearly relevant to the subject matter of astronomy. In his paper of 1749 on irregularities of the orbits of Saturn and Jupiter [1748b], Euler investigated the following diﬀerential equations: m2 dζ + 2mndx + m2 nrdζ − n =

d2 r dζ

n(λ − cos ω) 1+ν 2nr ω dζ 3 dζ − 3 dζ + 2r cos λ dζ + h(1 − g cos ω)3/2 λ λ

and 2mdrdζ + d2 x = −n sin

sin ω ω dζ + dζ. λ h(1 − g cos ω)3/2

Euler observed that the integration of these two equations depended on the integral of (1 − gcosω)−3/2 , but it was not possible to give a closed form for it. Thus, he dealt with the question of the integration by series of (1 − gcosω)−µ . According to Euler, the greatest diﬃculty in ﬁnding the solutions to these equations was the determination of a suﬃciently fast expansion of (1 − gcosω)−µ . Indeed, he observed that the expansion of this formula, “following ordinary rules”, is (1 − gcosω)−µ = 1 +

µ(µ + 1) 2 µ g cos ω + g cos2 ω + . . . , 1 1·2

275

276

Successes and Problems of the Triumphant Formalism

“but this series is not suitable for my purpose, in as much as it is not suﬃciently convergent, since it contains powers of cos ω. As for the last disadvantage, one can remedy it by reducing the powers of the cosine of the angle ω, to the cosines of multiples of the angle.”367 Since 2 cos2 ω = cos 2ω + 1, 4 cos3 ω = cos 3ω + 3 cos ω, 8 cos4 ω = cos 4ω + 4 cos 2ω + 3, ... Euler stated that (1 − gcosω)−µ must have an expansion of the type ∞

ai cos iω.

i=0

Then Euler found a recursive formula for ai (i > 1) and calculated a0 and a1 approximately, a process that involved a long sequence of calculations (see Golland and Golland [1993, 58–64]). In De serierum determinatione seu nova methodus inveniendi terminos generales serierum [1750–51b], Euler returned to his search for the expression of the general term of a series Σan deﬁned by recurrence.368 He studied several cases. The ﬁrst was the series 1 + 1 + 1 + . . .. Euler reduced the problem to the study of the functional equation y(x + 1) = y(x), where y(x) is a function — obviously in the 18th-century sense of the term. It was solved by expanding y(x + 1) in Taylor series y(x + 1) =

∞ 1 dn y . n! dxn

n=0

Since y(x+1) = y(x), he obtained the diﬀerential equation of inﬁnite degree: ∞ 1 dn y = 0. n! dxn

(185)

n=1

This enabled Euler to apply the technique for solving diﬀerential equations that he had developed in Methodus equationes diﬀerentiales altiorum graduum integrandi ulterius promota [1750–51a]. Indeed, by using the sub 1 n z stitution y = ezx , he derived the auxiliary equation ∞ n=1 n! z = e − 1 = 0, 367 368

See Euler [1748b, 61], translation in Golland and Golland [1993, 58]. See Chapter 13.1.

27

277

Trigonometric Series

which has the inﬁnitely many roots z = 2kπi, k ∈ Z. Since y(0) = 1, he found that the solution to 185 was y =1+

∞

αi sin 2iπx +

∞ i=1

i=1

β i (cos 2iπx − 1)

(the coeﬃcients αi and β i are to be determined subject to the condition an = 1, for every integer n).369 It should be noted that Euler, however, did not conceive of this trigonometric series as the ﬁnal result. Euler, in principle, imagined that such trigonometric series could be expressed as ﬁnite functions of sin πn and cos πn. In the same way, he showed how the general term of the series a + (a + g) + (a + 2g) + . . . , for a and g constants, could be determined by solving the functional equation y(x + 1) = y(x) + g and found an = A + gn +

∞

αi sin2iπn +

∞

β i cos2iπn

i=1

i=1

[A, αi , and β i are determined subject to the condition an = a + (n − 1)g]. He then solved eight further problems, which he reduced to the following equations: y(x + 1) = ay(x); y(x) = ay(x − 1) + b;

y(x) = ay(x − 1) + by(x − 2);

y(x) = a1 y(x − 1) + a2 y(x − 2) + . . . + a2 y(x − n);

y(x) = c + a1 y(x − 1) + a2 y(x − 2) + . . . + an y(x − n);

y(x) = my(x) + a + bx;

y(x) = y(x − 1) + F (x);

y(x + 1) = xy(x);

where a, b, c, g, m, a1 , a2 , . . ., an are constants and F (x) is an elementary function. In particular, he found that the solution of the equation y(x) = y(x − 1) + F (x) was x x ∞ cos2πnx F (x)dx + 2 F (x)cos2πnxdx y(x) = 0

+2

∞

sin2πnx

n=1 369

See Euler [1750–51b, 470–480].

n=1 x

F (x)sin2πnxdx.

278

Successes and Problems of the Triumphant Formalism

Trigonometric series were also used in the problem of the vibrating string.370 This controversy is relevant to our purpose for two reasons. The ﬁrst reason reason is that it showed the need to introduce quantities that were diﬀerent from elementary functions in order to mathematize the study of natural phenomena; at the same time, it revealed the diﬃculties that 18thcentury analysis (and in particular of the theory of series) had in treating nonelementary functions. The treatment of these new quantities required a change in the concept of analysis, although this was not forthcoming. The result was that analysis could not fully develop its potential. The second, and more speciﬁc, reason for which the controversy is of importance to our purpose is that, in his R´eﬂexions et ´eclaircissemens sur les nouvelles vibrations des cordes [1753], Daniel Bernoulli stated that all initial positions could be represented in the form y=

∞

an sin

n=1

nπx . l

(186)

In his opinion, the trigonometric solution was general. He did not base this opinion on mathematical arguments, nor did he tackle the problem of calculating the coeﬃcients. He only assumed that all sonorous bodies contained potentially an inﬁnity of sounds and an inﬁnity of corresponding ways of making their regular vibrations.371 Instead, in Remarques sur les m´emoires pr´ec´edens de M. Bernoulli [1753], Euler, who had already discussed the possibility of a solution of the type (186) in his De vibratione chordarum exercitatio; 372 rejected Bernoulli’s opinion by noting that “all the curves contained in that equation, although one increases the number of the terms to inﬁnity, have certain characteristics which distinguish them from all other curves” (see Euler [1753, 236–237]). Indeed, Euler thought that a function could be represented by a trigonometric series only if it was periodic (and that it could be represented by a sine series if it was also odd). As Grattan-Guiness and Ravetz noted,373 periodicity was an insuperable diﬃculty (186). Euler’s belief derived from basic concepts of 18th-century analysis, particularly the generality of algebra374 and the concept of function. For example, consider the equality 1 1 1 x = sin x − sin 2x + sin x + . . . . 2 2 3

(187)

Today we think of it as an equality that is valid on a certain interval since we think that the sum of series is diﬀerent from 12 x when x varies on an 370

See See 372 See 373 See 374 See 371

Chapter 23. Bernoulli [1753, 147–172]. Euler [1749a, 50–62]. Grattan-Guiness and Ravetz [1972, 245]. Section 18.2, p. 209.

27

279

Trigonometric Series

interval diﬀerent from (−π, π) and that the series gives rise to a function that is deﬁned piecewise. However, equality (187) was not considered to be valid on a certain interval, but according to the principles of 18th-century analysis375 to be valid for every value of x. As a consequence, this relation could not be thought of as a quantitative relation. A trigonometric series was understood as a formally derived relation: it might have a quantitative meaning, though this was not necessarily the case. At any event, it was not deﬁned quantitatively, namely as the limit of the partial sums. In his [1773, 169], Euler stated this view in a resolute way. He set √ √ cos x − −1 sin x = q. cos x + −1 sin x = p and Hence,

pn + q n 2

and

sin nx =

p(1 − pn ) 1−p

and

q + q2 + q3 + . . . + qn =

cos nx = He observed that p + p2 + p3 + . . . + pn =

pn − q n √ 2 −1

He then summed these ﬁnite geometric progressions, replaced and obtained

q(1 − q n ) . 1−q

pn +q n 2

by cos x,

1 cos nx − cos(n + 1)x . (188) cos x + cos 2x + cos 3x + .. + cos nx = − + 2 2(1 − cos x) Similarly, he derived sin x + sin 2x + . . . + sin nx =

sin x + sin nx − sin(n + 1)x . 2(1 − cos x)

(189)

Euler stated that the values of sin nx and cos nx varied between −1 and 1 when n was ﬁnite and that the last term of the sequences sin nx and cos nx when n = ∞ was equal to 0. Therefore, 1 cos x + cos 2x + cos 3x + . . . = − , 2 sin x + sin 2x + . . . + sin nx =

sin x . 2(1 − cos x)

(190) (191)

He took care to show that these results corresponded to his deﬁnition of the sum. For instance, by considering cos x − 1 1 − = 2 2(1 − cos x) 375

See Chapters 18 and 19.

280

Successes and Problems of the Triumphant Formalism

and expanding the last fraction, Euler obtained 1 cos x + cos 2x + cos 3x + . . . = − . 2 Equation (190) was also derived by Lagrange in his [1759, 111–112] using similar procedures.376 The result had been criticized by d’Alembert, who denied that cos x+cos 2x+cos 3x+. . . = − 12 . In response, Lagrange recalled the principles of power series on which trigonometric series were rooted and stated: I would pose the question whether every time one encounters an inﬁnite geometric series in an algebraic formula, for example 1 + x + x2 + x3 + . . . , one can substitute

1 , 1−x

though this quantity is really equal to the sum of proposed series only when one supposes the last term x∞ to be zero (Lagrange [1760–61, 323]). This approach to trigonometric series prevented 18th-century mathematicians from releasing the potential of trigonometric series even though they derived many results that, in a sense, seem to anticipate Fourier’s series. For instance, in his [1759], Lagrange derived the functional solution ∂2y 1 ∂2y to equation ∂x 2 = c2 ∂t2 by a diﬀerent approach. He considered n equal bodies spaced regularly along a weightless string between the ﬁxed point and supposed that the masses satisﬁed the equations d 2 yk = Δ2 yk−1 , dt2

1 ≤ k ≤ n,

with y0 = yn+1 = 0. He solved these equations and obtained y(x, t) =

∞

∞

2 rπx rπct rπx sin Yk sin dx cos l l l l r=1

k=1

∞

∞

k=1

r=1

rπx rπct rπx 2 dx sin dx cos , sin Vk + πc l l l 376

He justiﬁed the step from (188) to (190) by stating that, in the case where n is an inﬁnite number, the number 1 vanishes relative to n, and hence the term cos(n + 1)x becomes equal to cos nx (Lagrange [1759, 111]).

27

281

Trigonometric Series

where Yk and Vk are the initial positions and velocities of the kth mass. At this point he moved from the discrete to the continuous: He replaced Yk and Vk by Y (x) and V (x), and the sum ∞

Y (xk )

∞

sin

∞

sin

r=1

k=1

rπx dx l

and

rπx dx l

and

∞

V (xk )

k=1

∞

sin

r=1

rπx dx l

by the integrals

l

Y (x)

r=1

l

V (x) 0

∞ r=1

sin

rπx dx. l

Thus, he [1759, 100–101] obtained ∞ rπx rπct rπx 2 l dx sin cos sin Y (x) y(x, t) = l 0 l l l r=1 ∞ l rπx rπct rπx 2 dx sin cos . (192) sin V (x) + πc 0 l l l r=1

Here Lagrange seems to be very close to Fourier series. By interchanging and in ∞ 2 l rπx rπx dx sin , sin Y (x) l 0 l l

r=1

one obtained the Fourier sine series. In reality, as Grattan-Guiness [1972, 16] noted, Lagrange was persuaded a priori of the impossibility of representing any function through trigonometric series (in his paper he also rejected Daniel Bernoulli’s solution). Equation (192) “was for him only a step on the road to the Eulerian functional solution” (Grattan-Guiness and Ravetz [1972, 248]). Lagrange used trigonometric series in a formal way according to typical 18th-century procedures and never considered them as autonomous objects, capable of deﬁning a quantity by themselves. Another particularly interesting result was obtained by Clairaut in his Sur l’orbite apparente du Soleil autour de la terre [1754, 544–564]. He returned to the problem that Euler had examined in [1749a] and sought the expansion of the function f (x) in the form a0 + 2

∞

ak cos kx.

k=1

In so doing he produced expressions for what later would be called the Fourier coeﬃcients of the series. Indeed, he wanted to interpolate the given

282

Successes and Problems of the Triumphant Formalism

function f (x) for x = 2π/n and obtained the interpolation formula n−1

1 f n

a0 =

h=0

and

n−1

ak =

1 f n h=0

2hπ n

cos

2hπ n

2hkπ (k > 0). n

For n = ∞, Clairaut obtained ak =

1 2π

2π

f (x) cos nxdx. 0

In 1777, in dealing with an astronomical problem, Euler showed that the coeﬃcients of the trigonometric series f (x) = a0 + 2

∞

ak cos kx

k=1

could be obtained in a very quick way. He multiplied both sides of the last equality by cos mx and integrated the series term by term by observing that 2π cos mx cos kxdx = 0 0

for m = k (see Euler [1793]). These results did not change the common approach to trigonometric series. The expansion of a function into a trigonometric series was always recognized as being the result of applying a formal procedure. However, while the capacity to expand a function into a power series, which was convergent on an interval, was considered to be guaranteed a priori, the capacity to expand a function into a trigonometric series was not to be guaranteed a priori by usual procedures and had to be justiﬁed even by referring to the physical meaning of the trigonometric series.377

377

Kline observed: “[Euler] did not accept the general fact that quite arbitrary functions could be so represented [by using trigonometric series]; the existence of such a representation, where he used it, was assured by other means” [1972, 517].

28

Further developments of the formal theory of series

In this chapter, I shall illustrate some further developments of the formal theory of series. I begin by discussing the role of the binomial theorem in 18th-century analysis. I then go on to illustrate Lagrange’s demonstrations of the Taylor theorem. Finally, I concentrate upon the research that led from the investigation of Leibniz’s analogy to the eﬀective rise of the calculus of operations. A rigorous proof of the binomial theorem was the object of many attempts in the second part of the 18th century. In Chapters 4 and 8 we saw that Newton had derived the binomial theorem by interpolation and that it was accepted on the basis of the principle of inﬁnite extension. However, when analysis developed as a deductive and self-founding system, mathematicians felt the need to prove it in a rigorous way instead of accepting it by induction or analogy. Of course, the proof had to be rigorous in the sense of the principles of 18th-century analysis; in other words, it could be grounded neither on geometric arguments nor on an arithmetical basis, but on the mere manipulation of general quantities. Moreover, a satisfying proof of the binomial theorem could not use diﬀerential methods (and, in particular, the consideration of the binomial theorem as a particular case of the Taylor theorem). Indeed, the binomial theorem was considered to belong to that part of analysis deﬁned by Euler as the introduction to the analysis of inﬁnities and which later became known as algebraic analysis. The introduction to the analysis of inﬁnities occupied an intermediate position between the analysis of ﬁnites and diﬀerential and integral calculi. It investigated functions and their expansion into power series without using diﬀerential calculus; rather mathematicians employed the ﬁndings of the introduction to the analysis of inﬁnities in the construction of the calculus. In particular, the binomial theorem was essential to the calculation of the ﬂuxions or differentials of certain quantities.378 Therefore the use of diﬀerential methods for proving the binomial theorem involved a petitio principii. 378 For example, in his De quadratura, to prove that the ﬂuxion of xn is nxn−1 , Newton considers the increment (x + o)n of xn and “by the Method of Inﬁnite Series” obtains

xn + noxn−1 +

n(n − 1) 2 n−2 + ... o x 2

n−1 The ratio between the increments + n(n−1) o2 xn−2 + ... of xn is equal to 2 o of x and nox 1 : nxn−1 + n(n−1) oxn−2 + ... . When the increments vanish their last ratio is 1 : nxn−1 . 2

Therefore, the ﬂuxion of the quantity x is to the ﬂuxion of the quantity xn as 1 to nxn−1 . [1704, 336–337]. Similarly, in the Eulerian construction of the calculus, the notion of diﬀerential ratios was introduced by assuming that a function could be expanded into series and therefore by assuming the validity of the binomial theorem.

283

284

Successes and Problems of the Triumphant Formalism

Nevertheless, during the 18th century, there were a number of proofs that used diﬀerential methods (even Euler did this). Almost all the scholars who provided such proofs admitted the vicious circle.379 Most of these proofs (especially in the second part of the century) were mere exercises in didactic works. In some cases, they were the result of the attempts to improve some speciﬁc point of the building of analysis (see Pensivy [1987–88, 99]). Finally, in other cases, such as Euler [1755], one can hypothesize that they served to verify a posteriori the ﬁndings of the calculus. Among the most interesting nondiﬀerential proofs,380 I would like to examine an algebraic proof by Landen. In his Residual Analysis [1758], Landen set m (1 + x) n = 1 + Ax + Bx2 + Cx3 + . . . and m

(1 + y) n = 1 + Ay + By 2 + Cy 3 + . . . and derived m

m

x−y x2 − y 2 x3 − y 3 (1 + x) n − (1 + y) n =A +B +C + .... x−y x−y x−y x−y Landen set u = 1 + x and v = 1 + y and, using algebraic transformations, obtained u

m −1 n

m−1 1 + uv + ( uv )2 + . . . + uv m (n−1) m m n 1 + uv n + ( uv )2 n + . . . + uv

= A + B(x + y) + C(x2 + xy + y 2 ) + . . . .

Since this equation holds for any value of x and y, he posed that x = y and had m m −1 = A + 2B + 3Cx2 + . . . un n By multiplying this equation by (1 + x), he obtained m m (1 + x) n = (1 + x)(A + 2B + 3Cx2 + . . . .). n

Hence, m (1 + Ax + Bx2 + Cx3 + . . .) = (1 + x)(A + 2B + 3Cx2 + . . . .) n

379 380

On Maclaurin’s proof, see Pensivy [1987–88, 102–105]. I refer to Pensivy [1987–88] for a detailed examination.

28

285

Further Developments of the Formal Theory

Equating the coeﬃcients, he found A=

m , n

2B + A =

m A, n

...

At this point Landen derived the coeﬃcients of the sought-after expansion recursively. A diﬀerent type of proof of the binomial theorem made use of functional equations. For example, in his Demonstratio theorematis newtoniani de binomio [1757], Aepinus posed that (x + 1)m = Axm + Bxm−1 + Cxm−2 + . . . and considered the second coeﬃcient B as a function B(m) of the exponent m. Then he showed that the following relation between the second coeﬃcients B(m + n), B(m), B(n) of the expansions of (x + 1)m+n , (x + 1)m , (x + 1)n held: B(m + n) = B(m) + B(n).

(193)

Aepinus solved Equation (193) under the condition B(1) = 1 and found B(m) = m (in the solution he assumed that if s is an inﬁnitesimal number, any real number r is the type r = ns, where n is an inﬁnite number). Hence, it is easy to derive the binomial expansion. In his Demonstratio theorematis newtoniani [1774–75], Euler attempted to prove the binomial expansion using another type of functional equation. He considered the series 1 + nx +

n (n − 1) 2 n (n − 1) (n − 2) 3 x + x + ... 2! 3!

2 n(n−1)(n−2) n3 +. . . and denoted it by [n] (in modern terms, 1+nx+ n(n−1) 2! x + 3! is conceived as a function of the exponent n). He did not take convergence into consideration but showed that the series [n] satisﬁed the functional relation

[m + n] = [m][n]. The solution to this equation allowed him to obtain the binomial expansion. This method was later used by Cauchy in an entirely diﬀerent context.381 As early as the turn of the 17th century, some mathematicians had already tried to generalize the binomial theorem and seek the expansion into power series of the mth power of a polynomial (and even an inﬁnite polynomial), the so-called polynomial theorem. In 1697, de Moivre published 381

See Chapter 33.

286

Successes and Problems of the Triumphant Formalism

382 He showed a paper an inﬁnite on raising m multinomial to a given power. 2 3 that az + bz + cz + . . . is equal to a power series

Az m + Bz m+1 + Cz m+2 + . . . ,

where the coeﬃcient of the term xm+r is equal to the sum of all products of the form ap bq cs . . . (p + q + r + . . . . = m and p + 2q + 3r + . . . = m + r); each n! (n is the number of factors in product ap bq cs . . . being multiplied by p!q!s!... p q s the product a b c . . .). The question was also studied by other mathematicians: Johann Bernoulli, Leibniz, Jacob Bernoulli, Colson, and K¨ astner. De Moivre and Leibniz also suggested the possibility of generalizing to polynomial formula in the case m is a fraction. However, it was only Carl Friedrich Hindenburg who succeeded in giving a formulation of the polynomial theorem, making it possible to apply it to the case of fractional exponents. Hinderburg’s proof can be summarized as follows. Given 1 + a1 x + a2 x2 + . . ., one sets m 1 + a1 x + a2 x2 + . . . = (1 + z)m and expands (1 + z)m = 1 + mz + m(m−1) z 2 + . . . . Then one calculates the 2 powers h z h = a1 x + a2 x2 + . . . , where h is a natural number. By rearranging, one obtains m 1 + a1 x + a2 x2 + . . . = 1 + A1 x + A2 x2 + . . . .

The coeﬃcients are given by the formula r m Qr,n . Ar = h

(194)

h=1

Qr,n denotes the sum of all products of the type a1 n1 a2 n2 . . . ak nk , where the factors ai belong to the set {a1, . . . , ar }; each product a1 n1 a2 n2 . . . ak nk is multiplied by a coeﬃcient n1 !n2n!!...nk ! . Since the exponent m is not found in Qr,n but only in the binomial coeﬃcients m h , formula (194) can be applied to fractional and negative exponents.383 382

∗ ∗

∗

See de Moivre [1697]. It is worthwhile noting that this proof of the polynomial theorem convinced Hinderburg that it was possible to systematize the analysis of inﬁnities by grounding it on combinatorics. He was a clever organizer and managed to enlist the assistance of several mathematicians who agreed with the idea of the project. They founded the Combinatorial School, which developed from 1780 to 1810. The school did not include any great mathematicians (the most important were H. B¨ urmann, C.H. Eschenbach, C.S. Kl¨ ugel, M. von Prasse, H.A. Roth, I.K. Tetens, H.A. T¨ opfer, C. Kramp, and Gauss’s teacher 383

28

287

Further Developments of the Formal Theory

In the 1770s, Lagrange and Laplace dealt with the Taylor series. In their investigations, the Taylor series appears as an intermediate step for obtaining new results concerning ﬁnite diﬀerences and the emergent calculus of operations. We saw Laplace’s proof in Chapter 21. I now examine Lagrange’s demonstration of 1772. In the Sur une nouvelle esp`ece de calcul,384 Lagrange considered a function f (x) and took for granted that f (x + i) = f (x) + p1 i + p2 i2 + . . . ,

(195)

where pi (x) are functions of x. From f ((x + o) + i) = f (x + (o + i)), he derived f (x + o) +

∞

pk (x + o)ik = f (x) +

∞

pk (x)(i + o)k .

k=1

k=1

Then, he set f (x + o) = f (x) +

∞

pk (x)ok

k=1

and pk (x + o) = pk (x) +

∞

pk,n (x)on ,

n=1

and obtained k ∞ ∞ ∞ k n k−n k n i o pk (x) + pk (x) pk,n (x)o i = f (x)+ . f (x+o)+ n k=1

n=1

k=1

n=0

Hence, 1 p1,1 (x), 2 1 p3 (x) = p2,1 (x), 3 1 p3,1 (x), p4 (x) = 4 ....

p2 (x) =

and friend J.F. Pfaﬀ), nor did its members obtain particularly important results (compared to those of other mathematicians of the period). However, their approach “was inﬂuential in Germany at the turn of the 19th century and became the basis for the mathematical syllabus of the Prussian gymnasium in the Humboldt educational reform” (see Jahnke [1993, 265]). I refer to Jahnke [1993] and Panza [1992, 651–659] for both a general background of the school and a detailed mathematical investigation of the methods employed. 384 See Lagrange [1772, 443–445].

288

Successes and Problems of the Triumphant Formalism

According to Lagrange, this showed that all the coeﬃcients pi could be determined, starting from the preceding coeﬃcient pi−1 , by using the same operation that made it possible to determine p1 starting from f (x). He denoted this operation by means of the symbol ′ and the result of the operation performed on f (x) by means of f ′ (x). Consequently, the result of the operation performed on f ′ (x) was denoted by f ′′ (x), and so on. Hence, pk (x) =

1 [k] f (x) k!

and f (x + i) = f (x) +

∞ 1 [k] f (x)ik . k!

(196)

k=1

Here Lagrange “explicitly considered a mathematical entity that would later be referred to as an operator” (Panza [1992, 577]). Indeed, he was particularly interested in the operation O, which made it possible to calculate the coeﬃcient p1 . He even argued that the diﬀerential calculus could be based on this operation [1772, 443]. Subsequently, Lagrange however concluded his reasoning using inﬁnitesimals and proving that the series (196) was the same as the Taylor series. He took i = dx as an inﬁnitely small quantity and obtained df = f (x + i) − f (x) = f (x + dx) − f (x). =

∞ 1 [k] f (x)dxk . k! k=1

Using the principle of cancellation, he derived du = u′ (x)dx. Hence, du , u (x) = dx ′

d u (x) = (u ) = dx ′′

′ ′

du dx

=

d2 u ,. . . dx2

k

By replacing ddxuk in (196), Lagrange obtained the Taylor theorem u(x + i) = 1 dk u k u(x) + ∞ k=1 k! dxk (x)i (see [1772, 446–447]). Lagrange developped his idea on the possibility of basing the calculus on the operation O in his Th´eorie des fonctions analytiques [1797].385 In the ﬁrst theorem of this treatise, he tried to prove equality (195): Given a function f (x), the series f (x + i) = f (x) +

∞

pk (x)iαk

(197)

k=1

385 For a detailed examination of Lagrange’s theory of analytical functions, see Ferraro and Panza [A].

28

289

Further Developments of the Formal Theory contains no fractional or negative power of i, except for particular isolated values of x, namely f (x + i) = f (x) +

∞

pk (x)ik .

(198)

k=1

The notion that any function could be expanded into a power series had always been considered obvious by 18th-century mathematicians; Lagrange now assumed a weaker hypothesis [any function had an expansion of type (197)] and attempted to prove (198) in a general, algebraic way. In the ﬁrst part of the demonstration, Lagrange tried to show that there were no fractional powers in the √ expansion (197). He stated that, if the series (197) has a term of the type n im , the only possibility is that it derives from radicals that are found in the function f (x). In fact, √ √ a. The operation n im does not assign only one value to i, namely n im is an n-valued function of i. b. If the expansion of f (x + i) contains an irrational term, then the expansion is many-valued; therefore, f (x + i) is also many-valued and has the same number of values as the expansion. c. The substitution of x + i in place of x changes neither the number nor the nature of radicals,√as long as x and i are indeterminate

[namely, the n-valued radical n xm becomes the n-valued radical n (x + i)m ]. Consequently, if f (x+i) assumes n values, then f (x) assumes the same αk p number of values as f (x+i) and as the expansion f (x)+ ∞ k=1 k (x)i .

d. In√conclusion, if the expansion of f (x + i) contains an irrational term n m i , then f (x +√ i) and f (x) have n values. However, the combination of each value of n im with each value of f (x) gives rise to a number of values of f (x+i) greater than the values of f (x). This is contradictory. To prove that there are no negative powers, Lagrange noted that if the expansion of f (x+i) has a term i−k with a negative exponent, then i−k = ∞, for i = 0. Therefore, f (x + i) is also equal to inﬁnity for i = 0, namely f (x + 0) = f (x) = ∞ when x is indeterminate. This is absurd since f (x) can be equal to inﬁnity only for particular values of x. Lagrange knew that certain determinate values of x exist where (198) fails, but, according to the 18th-century analysis, these cases were considered exceptions that did not invalidate the rigor of the proof concerning general quantity386 (Langrange [1797, 23]). A simple example is the expansion of 386

On the treatment of exceptional values, cf. Chapter 19.3, p. 227).

290

Successes and Problems of the Triumphant Formalism

the function f (x) = expansion is √

x+i=

√

x for x = 0. As Lagrange showed in [1797, 26], its

√

1 1 1 √ i3 − . . . , x + √ i − √ i2 + 2 x 8x x 16x2 x

(199)

and the theorem does not hold for x = 0. In his opinion, it was the existence of such values that made it necessary to give an a priori demonstration of (198), even though it was veriﬁed by all known functions (Lagrange [1797, 22]). Starting from this theorem, Lagrange [1797] constructed the theory of analytical functions with the aim of basing the calculus only on formal considerations. He ﬁrst gave a reasoning similar to the one used in 1772 to show that the operation that enabled the calculation of the ﬁrst coeﬃcient p1 starting from f (x) also made it possible to calculate the other coeﬃcients pi starting from pi−1 . Thus, he could deﬁne the derived (the function n of f (x). p i derivative) of f (x) as the coeﬃcient p1 of the expansion ∞ n=0 n He denoted p1 by f ′ (x). He then termed the derived function of f ′ (x) as the second derived function of f (x) and denoted it by f ′′ (x) and, in general, termed the derived function of f (k−1) (x) as the kth derived function of f (x) and denoted it byf (k) (x). This allowed him to write (198) in the form 1 (k) f (x + i) = f (x) + ∞ (x)ik . k=1 k! f After that Lagrange calculated the derivatives of elementary functions387 (the only functions whose existence he recognized) and introduced the notion of integration as antidiﬀerentiation.388 Finally, Lagrange examined some applications of the calculus to geometry and mechanics. These applications required the consideration of the quantitative, a problem that Lagrange tackled in an explicit and innovative form, as we shall see in Chapter 30. ∗ ∗

∗

Lagrange’s approach inﬂuenced several mathematicians at the turn of the 18th century. Some of them sought to provide a more convincing proof of the Taylor theorem (see Poisson389 [1805] and Amp`ere [1806]). Others were stimulated to carry out research that led to an understanding of Leibniz’s analogy and the rise of the calculus of operations. In his Du calcul des d´erivations [1800], Arbogast attempted to provide a new and more general calculus that included the diﬀerential calculus as a special case. Arbogast posed the question of ﬁnding the coeﬃcients of the expansion of f (a + bx + cx2 + . . .) for any function f . 387

He expanded elementary functions into series using algebraic methods (see Ferraro and Panza [A]) and deﬁned the derivatives of this functions to be the ﬁrst coeﬃcient of these expansions. 388 See Chapter 25. 389 For Poisson, I refer to Grattan-Guiness [1990, 201–202].

28

Further Developments of the Formal Theory

291

He ﬁrst considered a function + x) and sought the coeﬃcients An of ∞ Afn(a xn its power series expansion n=0 n! . He denoted the operation that had to be made on f (a) to derive A1 with the symbol D. By using this symbol, he wrote A1 = Df (a), A2 = DA1 = D2 f (a), A3 = DA2 = D3 f (a), ... By replacing in the expansion

∞

n=0

f (a + x) =

An x n n!

∞ xn

n=0

n!

of f (a + x), he had Dn f (a).

Arbogast then considered increments of the form bx instead of x and obtained ∞ xn Dcn. f (a), f (a + bx) = n=0

where Dcn. f (a) =

1 n D. f (a), n!

D.n f (a) = D. (D.n−1 f (a)), and D. f (a) is the operation that had to be made on f (a) to derive the ﬁrst coeﬃcient of the expansion of f (a + bx). Finally, Arbogast considered functions of polynomials and functions of ∞ n functions and showed that the coeﬃcients A of the expansion A x n n=0 n ∞ s of f ( s=0 as x ) were given by An =

Dcn. f (a0 )

=

n h=1

ah1 ) . Dch. f (a0 ) Dcn−h .

The introduction of the operation D allowed Arbogast to take a step forward in the calculus of operations. He explicitly spoke of the method of separating symbols of operation from quantities, which consisted of the fact that symbols of operations could be handled separately from the subjects on which they operate and treated as if they were symbols of quantities. This method, which had already been used de facto by Lagrange in his [1792],390 390

See Chapter 21.

292

Successes and Problems of the Triumphant Formalism

allowed him to simplify and clarify the content of Lagrange’s theorem (140). du ξ dx − 1 in the form Arbogast set δu = du dx and expressed the equation Δu = e 1 + Δ = eδξ

(200)

The diﬀerence between (200) and the Lagrangian du

Δu = e dx ξ − 1 is that the purely operational components are stated separately in 1 + Δ = eδξ . Arbogast applied the calculus of operations to diﬀerential equations. Many mathematicians did the same thing, including Brisson, who, in his [1808], tackled the problem of the solution to linear partial diﬀerential equations with constant coeﬃcients of any order and in any number of variables. He considered the equation Az + B

∂z ∂2z ∂2z ∂2z ∂z +C + ... + G 2 + H + I 2 + . . . . = 0, ∂x ∂y ∂x ∂x∂y ∂y

(201)

where z is a function of the variables x, y, . . ., and A, B, C, . . . , G, H, I, . . . are constants. Brisson wrote it in the form ∇z = 0

(202)

and used ∇ as a linear operator on the possible solution functions. He observed that the rule m ∂ z ∂ n B ∂y m ∂ n+m z = AB A ∂xn ∂xn ∂y m held for repeated partial diﬀerentiations (if A and B are constant) and that the same result could be obtained by considering ∂, x, y as quantities raised to the powers n, m, n + m. For this reason he replaced (202) with the equation ∇′ z = 0, which denoted the algebraic equation that resulted from (201) after substituting repeated diﬀerentiation m ∂ z ∂ n ∂y m ∂xn

by multiplication:

∂nz ∂mz · . ∂xn ∂y m

28

Further Developments of the Formal Theory

293

He factored ∇′ z = 0 and obtained ∇′ z = δ ′0 δ ′1 . . . δ ′n−1 z. At this point he wrote ∇z = δ 0 δ 1 . . . δ n−1 z = 0, where the δ i are obtained ′ from the δ i by inverting the previous substitution, namely replacing multiplication by iteration. Brisson stated that the solution of the given equation can be found by solving δ i z = 0 and showed how one could solve δ i z = 0 when δ i is linear. A further contribution to the rise of the calculus of operations was made by the mathematician Jacques-Fr´ederic Fran¸cais. In his M´emoire rendant a d´emontrer la l´egitimit´e de la s`eparation des ´echelles, Fran¸cais applied the ` method of the separation of symbols to solve problems concerning diﬀerential equations and ﬁnite diﬀerence. He denoted dϕ dx by δϕ and ϕ(x + 1) by Eϕ and wrote the Taylor series for ϕ in the form 1 Eϕ(x) = ϕ(x + 1) = ϕx + δϕ(x) + δ 2 ϕ(x) + . . . = eδϕ(x) 2 By separating symbols, he obtained E = eδ (see Fran¸cais [1812–13, 249–250]). A simple example of how the method of the separations of symbol works is the following. Fran¸cais considered the diﬀerential equation dϕ = aϕ, dx which he wrote as (δ − a) ϕ = 0.

By separating symbols, he had δ − a = 0 and eδ = ea . Since E = ea , he obtained 1 = eak E −k . Hence, ϕ(x) = eak E −k ϕ(x) = eak ϕ(x − k) For x = k, he had ϕ(k) = eak ϕ(0) = Ceak . Finally, by changing k into x in the last equation, he obtained the solution to the given diﬀerential equation ϕ(x) = Ceax

294

Successes and Problems of the Triumphant Formalism

(see Fran¸cais [1812–13, 244–276]). Several mathematicians attempted to discover the mechanism that ruled these procedures. As early as 1787, Lorgna had observed that if y was a function and y ′ was the diﬀerence or diﬀerential of y, then the iteration of the operation of diﬀerence or diﬀerential, which he denoted by the symbol ′ y n , obeyed the same law of combination as the symbol xn that represented raising to the power of the quantity x (see Lorgna [1787, 413]). In M´emoire rendant ` a d´emontrer la l´egitimit´e de la s`eparation des ´echelles, Fran¸cais compared equations of the type aF (x, y) + bF (x, y) + cF (x, y) + . . . = 0,

(203)

where a, b, c, . . . are constants, with equations of the types ∂ n F (x, y) + a∂ n−1 F (x, y) + b∂ n−2 F (x, y) + . . . = 0

(204)

∂ n F (x, y) + aΔ∂ n−1 F (x, y) + bΔ2 ∂ n−2 F (x, y) + . . . . = 0

(205)

and

He stated that the symbols of operations behaved in the same manner as the constants a, b, c, . . . . Consequently, since one could derive (a + b + c + . . .)F (x, y) = 0 from (203), one could, in the same way, derive (∂ n + a∂ n−1 + b∂ n−2 + . . .)F (x, y) = 0 and

∂ n + aΔ∂ n−1 + bΔ2 ∂ n−2 + . . . F (x, y) = 0

from (204) and (205) (see Fran¸cais [1812–13, 245–246]). These observations paved the way for Servois, who gave the ﬁrst satisfactory explanation of the Leibniz’s analogy in two papers, entitled Essai sur un nouveau mode d’exposition des principes du calcul diﬀ´erentiel [1814–15a] and R´eﬂexion sur les divers syst`emes d’exposition des principes du calcul diﬀ´erentiel [1814–15b], in which, following Lagrange, he attempted391 to give an algebraic basis to the calculus and reject the use of inﬁnitesimals.392 391 Lagrange’s concept had been attacked by Wronski. The Polish philosopher and mathematician asserted that all mathematics was rooted in the “absolute law”, according to which F (x) = ∞ n=0 An Ωn (x), where Ωn (x) were arbitrary generating function. Servois’s paper is also a response to Wronski, who never actually attempted to prove his claims. 392 In [1814–15b, 148] Servois stated: “[The inﬁnitesimals] neither have nor can have theory; in practice it is a dangerous instrument in the hands of beginners anticipating, for my part, the judgement of posterity, I would predict that this method will be accused one day, and rightly, of having retarded the progress of the mathematical sciences” (translation in Grattan-Guiness [1990, 137]).

28

Further Developments of the Formal Theory

295

In [1814–15a, 98], Servois introduced the notion of distributivity and commutativity. He deﬁned a function f as distributive 393 if f (x + y + z + . . .) = f (x) + f (y) + . . . , and the functions h and g as commutative 394 if f (g(x)) = g(f (x)). Servois proved several results on distributive and commutative functions (even though he used the term “function” in all cases, he made a clear distinction between functions of functions and usual functions). In particular, he stated that if F (z) = f (z) + g(z) + . . . , where f and g are distributive and pairwise commutative, then F n is distributive for any integer n and one can ﬁnd F n applying the ordinary laws of algebra to (f (z) + g(z) + . . . .)n (see Servois [1814–15a]). Servois went on to examine the operator Δf (x) = f (x + Δx) − f (x) and the diﬀerential operator, which he deﬁned as 1 1 dz = Δz − Δ2 z + Δ3 z − . . . . 2 3 According to Servois, the reason why the symbols of operations Δ and d could be handled as if they were the symbols of quantities was that they were distributive and commutative with each other and with constant factors. In his opinion, this provided a simple explanation for the analogy between the iteration and powers. Servois’s work was the basis of later developments in the calculus of operations, which underwent remarkable development during the 19th century, mainly in Great Britain.395 It is worth emphasizing that while the algebraic aspects were clear, Servois (and all those who followed him) considered the inﬁnite series of operators and ignored problems of convergence. They considered the distributivity to be valid for inﬁnite series and assumed that a rule that was valid for power series also remained valid for inﬁnite series of operations. For this reason the calculus of operations can be considered a legacy (and perhaps an extreme result) of the 18th-century theory of series. 393 “The functions which . . . are such that the function of the sum of any number of quantities is equal to the sum of the corresponding functions of each of these quantities, will be called distributive” (Servois [1814a, 98]). 394 “The functions which . . . are such that they give identical results, whatever be the order in which one applies them to the subject, will be called commutative between each other” (Servois [1814a, 98]). 395 On this topic and its link to the rise of abstract algebra, I refer to Goldstine [1977] and Koppelman [1971].

29

Attempts to introduce new transcendental functions

In the previous chapters, I dwelled on the aspects of analysis that remained unaltered throughout the 18th century and that, according to Fraser’s expression, “constitute evidence of a shared conception signiﬁcantly diﬀerent from the modern one” (see Fraser [1989, 318]). However, there was a remarkable growth in mathematical knowledge during the century that gave rise to an accumulation of results, problems, and techniques. In this chapter and the one that follows, I shall deal with two aspects of this evolutionary process: the introduction of some new basic functions around the year 1800 and the use of the inequality technique in order to determine error estimates. I shall argue that such developments remained within the overall structure of 18th-century analysis: New ﬁndings accumulated and were added to older ones without challenging or changing key concepts. I have already noted that the 18th-century notion of a function did not exclude the possibility of introducing new transcendental functions that had the same status as elementary functions, provided that they were considered as known objects.396 Euler mentioned the question in the De plurimis quantitatibus transcendentibus quas nullo modo per formulas integrales exprimere licet [1780], a short note published in 1784. In this paper, Euler suggested the consideration of elliptic integrals as new basic functions. Indeed, he stated that logarithm and arctangent arose from the formulas 1 dx x and

1 dx; 1 + x2

however, they were considered similar to algebraic quantities since they could be treated as easily as algebraic quantities. In the same way, quantities concerning the rectiﬁcation of conics (which were included in certain integral formulas of the type f + gx2 dx b + kx2 had been analyzed to the extent that, if a problem was reduced to these quantities, it could then be regarded as completely solved (Euler [1780, 522]). Euler went further and stated that, according to a widely held opinion, all transcendental quantities could be represented (geometrically) by the 396

See Chapter 18.

297

298

Successes and Problems of the Triumphant Formalism

quadrature of certain curves or could be reduced (analytically) to integral formulas f (x)dx. He thought that this opinion was to be revised. Indeed, he pointed out the existence of transcendental quantities that could not be expressed by integral formulas although their values could be determined by approximation, an example being the “curve” expressed by the “equation” y=

∞

x

n(n+1) 2

n=0

for |x| < 1 (see Euler [1780, 523–525]). De plurimis quantitatibus transcendentibus contains some elements that enriched 18th-century analysis and shows that mathematicians were open to add on new elements to the structure of analysis, provided — and this is of fundamental importance — the foundations of that structure were not undermined. In eﬀect, the paper retained the main points of the 18thcentury approach: (a) At any moment, the number of basic functions that were eﬀectively accepted was ﬁxed; (b) it was problematic to increase this number because every new function had to be an entirely known entity; (c) integral formulas on their own did not generate new functions; (d) there is a diﬀerence in status between the expressions of quantities using elementary functions and using integral functions; (e) the representation of a quantity by a series was regarded as approximate and as having a status inferior even to that of integral representation, though Euler’s paper suggested that it was the only possible analytical representation in certain cases. De plurimis quantitatibus transcendentibus had no practical consequences on Euler’s work. However, from the last decades of the 18th century to the beginning of the new century, various mathematicians thought that the acquired knowledge made it possible to consider certain quantities as new functions. For instance, Lacroix stated that all expressions involving integrals of the types dx x2 dx dx √ √ √ , , (x2 + a) a + bx2 + cx4 a + bx2 + cx4 a + bx2 + cx4 were to be regarded as integrated in the same way as an integration that gave rise to expressions composed of logarithms and inverse trigonomet dx , ric functions was regarded as complete. The integrals (x2 +a)√a+bx 2 +cx4

29

Attempts to Introduce New Transcendental Functions

299

x2 dx and √a+bx were therefore new transcendental functions 2 +cx4 to be introduced in the calculus (see Lacroix [1797–1800, 2:59]). Legendre was the scholar who introduced a more accurate analysis of elliptic integrals. In his Exercises de calcul int´egral ( the three volumes of the treatise were published in 1811, 1816, and 1817,397 at roughly the same time as Gauss’s paper on the hypergeometric function,398 which I will deal with below), he explained that the scope of his research was to compare the integrals of the type PRdx (where P is a rational function and R = √ a + bx + cx2 + dx3 + ex4 ) with one another, to classify them into diﬀerent species, to reduce each species to the simplest form, to evaluate them by fastest and easiest approximations, and to develop an algorithm concerning these integrals in order to contribute to the extension of the domain of analysis (Legendre [1811, 1:3–4]). In other words, he aimed to reduce all the integrals to certain basic integrals that satisﬁed the conditions399 (C1) and (C2). Legendre also investigated the gamma function and other closely connected functions. In the ﬁrst and second volumes of his Exercises de calcul int´egral, many pages were devoted to them although his approach remained traditional. For instance, the gamma function was deﬁned as 1 1 a−1 log Γ(a) = x 0

√

dx , a+bx2 +cx4

with 0 < a < ∞ (Legendre [1811–17, 2: 4]). He easily derived Γ(a + 1) = aΓ(a) and other relations concerning Γ. Shortly afterwards, he stated that, when a was positive, the function Γ(a) could be viewed as the area under the a−1 between x = 0 and x = 1 [1811–17, 2:60]. However, this curve y = log x1 representation was geometrical, in a sense (en quelque sorte g´eom´etrique), and gave no idea of the function when a was negative. In such cases, the meaning of the function was derived by considering the formula Γ(a + 1) = aΓ(a) (which contained a general property of the function) to be valid beyond its original interval 0 < a < ∞. In [1811–17, 1:295], Legendre considered the quantity400 M interpolating the discrete sequence 1+

1 1 1 1 + + + ... + 2 3 4 x

397 He had already published an essay M´emoire sur les transcendantes elliptiques (see Legendre [1794]) on this topic in 1793 or 1794 (the booklet is dated “the second year of the republic”). 398 See Gauss [1812b]. 399 See p. 208. 400 It is the function Ψ(x) + γ, where γ is Euler’s constant.

300

Successes and Problems of the Triumphant Formalism

and regarded M as a “continuous function”401 of x since it could be expressed in the form M = γ + log x +

∞

1 B2n + (−1)n . 2x 2nx2n

(206)

n=1

In his opinion, (206) furnished the approximate values of M for x > 1.402 In order to ﬁnd a relation between M and Γ(x), Legendre used the formula ∞ 1 B2n 1 log x − x + log 2π + (−1)n+1 . log Γ(x) = x − 2 2 2n(2n − 1)x2n−1 n=1 (207) He said that (207) could be employed to calculate the value of the function gamma for x > 1 (Legendre [1811–17, 1:294]). By diﬀerentiating (207), he obtained ∞ B2n 1 d (log Γ(x)) (−1)n . (208) = log x − + dx 2x 2nx2n n=1

According to Legendre, equations (208) and (206) provided the value of M and d(logdxΓ(x)) for x > 1. However, by comparing them he obtained 1 d (log Γ(x)) = − + M − γ; dx x “an equation that must take place whatever the value of x, since M can be regarded as a continuous function of x” (Legendre [1811–17, 1:297]). It is clear that Legendre was referring to the generality of algebra,403 and thus, the equation 1 d (log Γ(x)) =− +M −γ dx x was conceived to be valid in general, although numerically it only held for certain values of x (and in an approximate way). The deﬁnition lacked Γ(a) for negative numbers a, but the generality of algebra also provided the values in this case. Furthermore, Legendre gave no theoretical importance to the diﬀerence between divergent and convergent series. The only condition he imposed 401

See Section 18.3. Legendre knew the divergent and asymptotic nature of the series (206), (207), and (208). He termed them semiconvergent because they ﬁrst decrease (converge, in the language of his time) and then increase (diverge) [1811-17, 1: 267]. He also knew that they could be used in computing the values of gamma and digamma functions. He did not clarify why he assumed 1 as the bound of the interval of approximation. 403 See Section 18.2, p. 209. 402

29

Attempts to Introduce New Transcendental Functions

301

was that series provided approximate values of the expressed quantities for certain values of variables, independently of their divergence or convergence. In his [1984, 103] and [1986, 16], Gray observed that it was not true that Legendre never considered the inverse functions of elliptic integrals404 but that he did not think of them as functions of complex variables and, for this reason, did not realize the importance of such an inversion. This is true: In eﬀect, Legendre followed the principle of the generality of algebra and did not consider functions of complex variables as entities enjoying properties which diﬀered from those of real variables (e.g., double periodicity, which does not hold for functions of real variables, was beyond the boundaries of Legendre’s mathematical world). Legendre’s treatment of new functions was an integral part of many studies about transcendental quantities. These developed, enriched, and attempted to consolidate the 18th-century theory, and Legendre’s work probably represented the utmost advance of this theory (this is even true of his Trait´e des fonctions elliptiques, which was published in 1825 and 1826). Those mathematicians who followed the traditional line of investigation were not aware of the need to rethink 18th-century methods. They were convinced that additions had to be made to the traditional theory of functions and that the set of basic functions had to be enlarged; nevertheless, they thought that the core of analysis (the formal methodology connected to elementary functions) could and needed to remain unaltered. In eﬀect, they investigated new transcendental quantities f (x)dx formally by manipulating and recombining the basic components of the analytical expressions f (x) with the help of the theory of integration. It should also be noted that mathematicians often resorted to extraanalytical arguments and, in particular, geometrical interpretations when dealing with transcendental quantities (see the notion of discontinuous functions and the integral in Chapters 23 and 25, Legendre’s deﬁnition of the gamma function in this chapter), but this contradicted the declared independence of analysis from geometry, one of the cornerstones of 18th-century mathematics. Around 1800, a satisfactory analytical theory, which organized the acquired knowledge unitarily, was lacking.

404

In the 18th century, the consideration of the inverse of a given function y = f (x) was not a problem, even though the analytical expression x = g(y) could not be made explicit. In this case mathematicians operated on the analytical expression f (x) and, of course, if a table of values of x and y was given, it was understood to express both the relation y = f (x) and the relation x = g(y).

30

D’Alembert and Lagrange and the inequality technique

I conclude the third part of the present work by discussing another interesting feature of the evolutionary process of 18th-century analysis, namely the use of the inequality technique to estimate errors in approximation. Prior to the late 1760s, many methods of approximations were known. They mainly consisted in deriving inﬁnite analytical expressions or in giving a recursive process for estimating a certain quantities. According to Grabiner [1981, 56]: “Most mathematicians preferred to write down inﬁnite expressions, which, since they appeared in equations, seemed to give the solution exactly, rather than to write inequalities.” In general, little attention was paid to computing general bounds on the errors made in approximations (see Grabiner [1981, 57]). In 1768, d’Alembert published an innovative paper, R´eﬂexions sur les suites et sur les racines imaginaires; its novel feature was the fact that the problem of approximation was associated with the determination of an explicit error estimate. The paper was composed of two parts. The ﬁrst part dealt with the approximate calculation of the values of the function (1 + x)m by means of the series ∞ m n x . n

n=0

The second is devoted to questions concerning imaginary numbers (I shall not treat it). m n D’Alembert started by observing that the series ∞ n=0 n x had to be decreasing405 in order to compute the values of (1 + x)m . Consequently, the ratio between the nth and (n + 1)th terms of the series had to be less than 1 (in absolute value) and then |x| <

n if n > m + 1. n−m−1

Setting n = ∞ , he had that the last terms of series, at least, are decreasing (convergent) if |x| < 1. On the contrary, the last terms are increasing (divergent) when |x| > 1 (d’Alembert [1768, 173]). 405

D’Alembert used the term “convergent.”

303

304

Successes and Problems of the Triumphant Formalism Then d’Alembert determined the bounds of errors. He observed that if |x| < 1,

ν > 1 + m, ∞ m n S = x , n n=ν−1 m ν−1 x , A = ν−1

one had

S<

∞

A|x|i =

i=0

and S>

∞ i=0

A

v−1−m ν

i

A 1 − |x|

|x|i =

(209)

A . 1 − |x| v−1−m ν

He stated that the sum of the series from A on lay between the bounds A 1 − |x| and

nA , ν − (ν − 1 − m)|x|

which gave “a practicable enough way” of summing the series by approximation, and that the error was less than A(m + 1)|x| , (1 − |x|)(ν − (ν − 1)|x| + m|x|) if one assumed S to be equal to a value between these bounds (d’Alembert [1768, 177–178]). D’Alembert then discussed the improvement of convergence of the series and, ﬁnally, criticized the use of divergent series in certain demonstrations (see d’Alembert [1768, 181–183]). Some remarks are appropriate. First, d’Alembert did not depart from the basic tenets of the 18th-century conception: a series was not an autonomous object but the result of a transformation of a given closed analytical expression. Indeed, d’Alembert did not determine the sum of ∞ m n x : n

n=0

30

D’Alembert and Lagrange and the Inequality Technique

305

∞ m n He assumed the development of the function (1 + x)m is n=0 n x (it is to be imagined that, according to d’Alembert, this relation was derived by usual formal methods). Second, it is true that d’Alembert used the technique of inequalities, but this technique is a tool for numerical evaluation of a function. In no case did he use this technique to prove the existence of a limit.406 D’Alembert’s did not know the ratio test, if by this term we intend a convergence criterion by which we establish if the series has a ﬁnite sum. For him the condition an+1 0) increases, then |V (x, i)| also increases, and this occurs at least up to a certain value i′ of i, after which |V (x, i)| might begin to decrease.412 Moreover, if the increase of i is small, the increase of |V (x, i)| is small as well.413 Therefore, if D > 0 is a given quantity, there exists an i′ such that |V (x, i)| < D,

i < i′ .

Consequently, if i < i′ , then i[f ′ (x) − D] < f (x + i) − f (x) < i[f ′ (x) + D].

(210)

Lagrange replaces the indeterminate x by x + i, x + 2i, . . ., x + (n − 1)i, where i is taken to be suﬃciently small. He thus has i[f ′ (x + ki) − D] < f (x + (k + 1)i) − f (x + ki) < i[f ′ (x + ki) + D] (211) for k = 0, 1, 2, . . . , n − 1. Of course, this can be done if (210) holds for i < i′ and i′ is independent of x. In this case, one can ﬁx i such that ni < i′ (for a given n) and for this ﬁxed i the inequalities (211) hold, for every k. Here Lagrange implicitly assumes a condition of uniformity in the behavior of functions, which resembles uniform continuity.414 411

On the proof contained in the Th´ eorie, see Ferraro and Panza [A]. Lagrange imagines that, given a function h(x), if h(x) = ∞ when i varies between certain values T1 and T2 , then there exists a partition of [T1 , T2 ] such that h(x) is monotone over any interval [ti , ti+1 ] of the partition. 413 Here Lagrange uses the principle of continuity (LC). 414 Lagrange’s assumption of uniformity cannot simply be confused with modern uniform continuity, which presupposes a diﬀerent concept of analysis. Instead, it seems to depend on the fact that x is an indeterminate quantity. I also note the similarity between Lagrange’s condition of uniformity and Cauchy’s assumption of uniform behavior of continuous functions (see p. 350). 412

308

Successes and Problems of the Triumphant Formalism By summing (211) term by term, Lagrange derives i

n−1 k=0

f ′ (x + ki) − niD < f (x + ni) − f (x) < i

n−1

f ′ (x + ki) + niD

k=0

Let D= then one has

n−1 ′ f (x + ki) k=0

n

0 < f (x + ni) − f (x) < 2i or 2i

n−1 k=0

n−1

;

f ′ (x + ki)

k=0

f ′ (x + ki) < f (x + ni) − f (x) < 0.

If we denote the maximum of |f ′ (x + ki)|, for k = 0, 1, 2, . . . , n − 1, by P we have 0 < f (x + ni) − f (x) < 2inP or − 2inP < f (x + ni) − f (x) < 0. Since i can be taken as small as desired, and n as large, Lagrange assumed that in is equal to any quantity z and f (x + ni) − f (x) = f (x + z) − f (x). The quantity f (x + z) − f (x) can represent any function of z that vanishes for z = 0 and f ′ (x + in) = f ′ (x + z) is the derivative of f (x + z), with respect both to x and z. Consequently,415 0 < f (x + z) − f (x) < 2zP or − 2zP < f (x + z) − f (x) < 0, and, for x = 0, 0 < f (z) < 2zP or − 2zP < f (z) < 0, provided f ′ (z) is ﬁnite over a certain interval. From this lemma, Lagrange deduced the remainder theorem for which, in the Le¸cons, he gave the following formulation: 415 It is clear that P must now be interpreted as the supremum of f ′ (x + z) over an appropriate interval.

30

D’Alembert and Lagrange and the Inequality Technique

309

Given a function f (x), if f (n) (q) and f (n) (q) are the minimum and the maximum of the derivative f (n) (x + i), for i = 0 and i < i′ , the following inequalities holds: in i2 ′′ f (x) + . . . + f (n) (p) (212) 2 n! i2 in < f (x + i) < f (x) + if ′ (x) + f ′′ (x) + . . . + f (n) (q), 2 n!

f (x) + if ′ (x) +

provided f (n) (q) and f (n) (q) are not inﬁnite. The proof runs as follows. Let p and q be the values of x + i such that f ′ (p) and f ′ (q) are the least and greatest value of f ′ (x + i), taking x as given, and letting i vary from 0 to a given value i′ . By integrating f ′ (x + i) − f ′ (p) > 0 and f ′ (q) − f ′ (x + i) > 0 with respect to the variable i (between 0 and i), one obtains f (x + i) − f (x) − if ′ (p) and if ′ (q) − f ′ (x + i) + f (x). If f ′ (x + i) never becomes inﬁnite for i < i′ [which surely occurs when f ′ (p) and f ′ (q) are not inﬁnite], an application of the above lemma, with positive i, made it possible to obtain f (x + i) > f (x) + if ′ (p) and f (x + i) < f (x) + if ′ (q). We now repeat the reasoning by assuming that p and q are the values where the second derivative assumes the maximum and minimum. We have f ′′ (x + i) − f ′′ (p) > 0 and f ′′ (q) − f ′′ (x + i) < 0, provided that f ′′ (x + i) is not inﬁnite, which certainly occurs when f ′′ (p) and f ′′ (q) are not inﬁnite. By applying the lemma twice, we have f (x) + if ′ (x) +

i2 ′′ i2 f (p) < f (x + i) < f (x) + if ′ (x) + f ′′ (q). 2 2

The reasoning can be repeated again, and it is possible to derive (212).416

416

In the Th´ eorie, Lagrange supposed that any function takes all the intermediate values between two of its values. By applying this form of the principle of continuity, Lagrange obtained in+1 Rn (x, i) = f (n+1) (x + λi), (n + 1)! where 0 < λ < 1.

31

Fourier and Fourier series

Fourier began his mathematical work on the theory of heat in the early 1800s. In December 1807, he submitted a memoir, entitled Sur la propagation de la chaleur dans le solides [1807], to the Institut de France. Lagrange, Laplace, Monge, and Lacroix were chosen as referees. Fourier’s approach was too diﬀerent from Lagrange’s and Laplace’s, especially with regard to trigonometric series, and, thus, although Fourier made some attempts to clarify his views, the memoir remained a manuscript (it was published by Grattan-Guiness and Ravetz in Joseph Fourier 1768–1830422 ). In the following years, the Institut announced a competition on the propagation of heat in solid bodies. Fourier submitted a new version of his memoir containing additional work. He won the prize, but the commission (Lagrange, Laplace, Malus, Ha¨ uy, and Legendre) stated that the work “leaves something to be desired on the score of generality and even rigor.” (see Fuorier [Œuvres, 1: viii]). Thus, this version of the memoir was not published on this occasion either. Fourier wrote another version that was published by the Acad´emie des Sciences in 1822 with the title Th`eorie analytique de la chaleur.423 Fourier’s treatise is of great interest in the history of mathematics and physics and the relationships between these sciences. Given the scope of the present book, I concentrated on Fourier’s use of trigonometric series in his mathematical investigation of heat diﬀusion. In Sur la propagation de la chaleur, Fourier considered the steady-state diﬀusion in a lamina and obtained the diﬀusion equation ∂2z ∂2z + 2 =0 2 ∂x ∂y

(x ≥ 0,

− 1 ≤ y ≤ 1).

He solved this equation by separating variables and superposing simple states. He obtained ∞ ak e−nk x cos nk y z= k=1

By considering the boundary condition z = 0 (when y = ±1) and z = 1 (for x = 0) and replacing y by 2u π , he derived the trigonometric series 1 = a1 cos u + a2 cos 3u + a3 cos 5u + . . . , where − 12 π < u < 21 π (see Fourier [1807, 134–144]). 422

See Grattan-Guiness and Ravetz [1972]. For more details about the story of the publication of Fourier’s treatise, see GrattanGuiness and Ravetz [1972]. 423

315

316

The Decline of the Formal Theory of Series Fourier found the constants ak by diﬀerentiating 1=

∞ k=1

ak cos(2k − 1)u

term by term inﬁnitely many times: 1 = a1 cos u + a2 cos 3u + a3 cos 5u + . . . , 0 = a1 sin u + 3a2 sin 3u + 5a3 sin 5u + . . . , 0 = a1 cos u + 32 a2 cos 3u + 52 a3 cos 5u + . . . , 0 = a1 sin u + 33 a2 sin 3u + 53 a3 sin 5u + . . . , 0 = a1 cos u + 34 a2 cos 3u + 54 a3 cos 5u + . . . , .... He set u = 0 in all derived equations and obtained an inﬁnite number of equations in the unknowns ak : 1 = a1 + a2 + a3 + a4 + a5 + . . . , 0 = a1 + 32 a2 + 52 a3 + 72 a4 + 92 a5 + . . . , 0 = a1 + 34 a2 + 54 a3 + 74 a4 + 94 a5 + . . . , 0 = a1 + 36 a2 + 56 a3 + 76 a4 + 96 a5 + . . . , 0 = a1 + 38 a2 + 58 a3 + 78 a4 + 98 a5 + . . . , 0 = a1 + 310 a2 + 510 a3 + 710 a4 + 910 a5 + . . . , 0 = a1 + 312 a2 + 512 a3 + 712 a4 + 912 a5 + . . . , .... To solve this system, Fourier considered the ﬁrst seven equations in the ﬁrst seven unknowns and found that a1 =

3 · 3 · 5 · 5 · 7 · 7 · 9 · 9 · 11 · 11 · 13 · 13 . 2 · 4 · 4 · 6 · 6 · 8 · 8 · 10 · 10 · 12 · 12 · 14

At this point Fourier stated that if one considered more equations, one would have found an expression of a1 similar to the previous one. In the case of eight equations, he found a1 =

3 · 3 · 5 · 5 · 7 · 7 · 9 · 9 · 11 · 11 · 13 · 13 · 15 · 15 . 2 · 4 · 4 · 6 · 6 · 8 · 8 · 10 · 10 · 12 · 12 · 14 · 14 · 16

According to Fourier, if one considered all the inﬁnite equations, then one had 4 3 · 3 · 5 · 5 · 7 · 7... = . a1 = 2 · 4 · 4 · 6 · 6 · 8... π

31

317

Fourier and Fourier series

In the same way, he found the other coeﬃcients: 14 1 · 1 · 5 · 5 · 7 · 7 · ... =− , 2 · 4 · 2 · 8 · 4 · 10 · . . . 3π 1 · 1 · 3 · 3 · 7 · 7 · ... 14 = = , 4 · 6 · 2 · 8 · 2 · 12 · . . . 5π 14 1 · 1 · 5 · 5 · 9 · 9 · ... =− , = − 6 · 8 · 4 · 10 · 2 · 12 · . . . 7π 1 · 1 · 5 · 5 · 7 · 7 · 11 · 11 · . . . 14 = = , 8 · 10 · 6 · 12 · 4 · 14 · 2 · 16 . . . 9π ...

a2 = − a3 a4 a5

(see Fourier [1807, 147–156]). This procedure is rather close to typical 18th-century procedures. However, Fourier changed the interpretation of the relation ∞

1 (an cos nx + bn sin nx), f (x) = a0 + 2

(213)

n=1

and this is the crucial point. He viewed this relation as a purely quantitative relation. Thus, he [1807, 158] made clear that the equality ∞ (−1)k k=0

π 1 cos(2k − 1)u = 2k − 1 4

did not hold when the variable u does not lie between − π2 and the function ∞ 1 (−1)k cos(2k + 1)u 2k + 1

π 2.

Indeed,

k=0

π 4,

(− π2 , + π2 );

over it is 0 for u = ± π2 : it is equal to − π4 , over is equal to π 3π ( 2 , 2 ). To make this clear, he also provided a geometrical interpretation of the equation ∞ 1 y= cos(2k − 1)u, (−1)k+1 2k − 1 k=0

the curve having this equation being viewed as the limiting curve of the curves424 n (−1)k+1 y= k=0

1 cos(2k − 1)u, n = 1, 2, 3, . . . . 2k − 1

424 See Grattan-Guiness and Ravetz [1972, 169–171] and Grattan-Guiness [1990, 594– 601].

318

The Decline of the Formal Theory of Series

k+1 1 cos(2k − 1)u directly by Then he found the sum of ∞ k=0 (−1) 2k−1 showing that the nth sum (n even) of the series is 1 u sin 2nx dx 2 0 cos x and that it tended to π4 as n went to inﬁnity. Fourier applied a similar procedure to ∞

1 1 (−1)k+1 sin kx x= 2 k k=1

and

∞

1 1 (−1)k+1 cos kx log(2 cos x) = 2 k k=1 ∞ 1 For example, as concerns 2 x = k=0 (−1)k k1 sin kx, he considered the nth sum m 1 (−1)k+1 sin kx y= k k=1

and diﬀerentiated it in order to obtain m

dy (−1)k+1 cos kx. = dx k=1

By multiplying the last equation by 2 sin x and by performing the appropriate manipulations, he obtained 1 cos(m + 21 )x dy = − dx 2 2 cos 12 x By integrating by parts, 1 y = x− 2 and

cos(m + 21 )x 1 1 1 =C + x+ 2 2m+ 2 cos 12 x 1 y = x + C, 2

for

1 2

sin(m + 12 )x + ... cos 21 x

m = ∞.

Since y(0) = 0, he obtained y = 21 x. k+1 1 sin kx At this point425 Fourier stated that the equation y = ∞ k=1 (−1) k could be represented as a sequence of vertical and oblique straight lines426 and 425 426

See Grattan-Guiness and Ravetz [1972, 165]. Fourier’s diagram includes vertical lines ba′ , cb′ , dc′ , ed′ , f e, . . . (see Fig. 21).

31

319

Fourier and Fourier series

a'

a

b

b'

c'

c

d

d'

e'

e

f

f'

Fig. 21

made it clear that trigonometric series oﬀered an analytical representation of Eulerian discontinuous functions. Then Fourier investigated the behavior of the integral 1 2

u 0

sin 2nx dx cos x

to clarify that the sum ∞ (−1)k k=0

1 cos(2k + 1)u 2k + 1

π 4

only for certain values of the variable (see Fourier [1807, 159–173]). Fourier regarded the convergence of series as lying at the heart of the question. He thought it was easy to derive trigonometric series by diﬀerent procedures: is

but the essential point is to distinguish the limits within which the value of the variable is to be taken. For instance the equation 12 x = sin x − 21 sin 2x − 13 sin 3x + . . . , given by Euler, holds as long as the values of x lie between 0 and π or between 0 and −π. For all other values of x the right-hand side has a determined value very diﬀerent from 21 x . . .. It is by means of these observations that the contradictory consequences found by combining diﬀerent series of sine and cosine are explained. (Fourier [1807, 169]) On this question, Fourier’s view diﬀered enormously from Lagrange’s. Fourier also tried to explain his approach in a short note submitted to the Institut de France, which is still unpublished.427 The discussion must have 427

For a description, see Grattan-Guiness and Ravetz [1972, 169–173].

320

The Decline of the Formal Theory of Series

been continued privately; today Fourier’s response to an objection of Lagrange is kept. Lagrange had observed that if one diﬀerentiated428 1 1 π−x = sin x + sin 2x + sin 3x + . . . , 2 2 3

(214)

then one obtained 1 − = cos x + cos 2x + cos 3x + .. 2 Hence, −

x 1 1 + C = sin x + sin 2x + sin 3x + . . . 2 2 3

By setting x = 0, one had C = 0 and −

1 1 x = sin x + sin 2x + sin 3x + . . . , 2 2 3

which contradicted formula (214). It is probable (we can only reconstruct Lagrange’s observations through Fourier’s words) that Lagrange wanted to show a contradiction in Fourier’s approach by proving that if we assume that trigonometric series represent functions, then one series may represent two diﬀerent functions. According to Fourier, Equation (214) did not hold for x = 0, and the function π−x 2 is represented by the series sin x +

1 1 sin 2x + sin 3x + . . . 2 3

only for a certain interval of values of x. In the following pages of his treatise, Fourier generalized the above results and showed that an odd arbitrary function f (x) can be expanded into a sine series ∞ ak sin kx. f (x) = k=1

He assumed that f (x) could be expanded into the series f (x) =

∞ f (2k+1) (0) k=0

(2k + 1)!

428 This equation follows from the convergent series by replacing x by π − x.

x2k+1

1 x 2

= sin x −

1 2

sin 2x −

1 3

sin 3x + . . .

31

321

Fourier and Fourier series

and wrote f ′ (0) = a1 + 2a2 + 3a3 + 4a4 + 5a5 + . . . , f (3) (0) = a1 + 23 a2 + 33 a3 + 43 a4 + 53 a5 + . . . , f (5) (0) = a1 + 25 a2 + 35 a3 + 45 a4 + 55 a5 + . . . , f (7) (0) = a1 + 27 a2 + 37 a3 + 47 a4 + 57 a5 + . . . , .... After a long sequence of formal and rather complex manipulations, Fourier obtained 1 1 1 πf (x) = f (π) − 2 f (2) (π) − 4 f (4) (π) − . . . sin x 2 1 1 1 1 − f (π) − 2 f (2) (π) − 4 f (4) (π) − . . . sin 2x 2 2 .... Then he set 1 1 sn (x) = f (x) − 2 f (2) (x) − 4 f (4) (x) − . . . n n n and showed that sn satisﬁed the diﬀerential equation sn (x) +

1 d2 sn (x) = f (x). n2 dx2

Hence, sn (π) = (−1)n+1 n and

∞

1 πf (x) = 2 k=1

π

f (x) sin kudu

π

f (x) sin kudu sin kx.

Fourier felt it necessary to explain the meaning of the integral π f (x) sin kudu. 0

Indeed, the function f (x) can be a discontinuous function and the notion of integration as antidiﬀerentiation cannot applied to these functions. Fourier π stated the integral 0 f (x) sin kudu was the area under the function f (x) sin ku

322

The Decline of the Formal Theory of Series

and over the segment [0, π]. He did not provide an analytical notion of integration that also held for discontinuous function and seemed to be satisﬁed with geometrical interpretation, an approach that cannot have been shared by those who regarded analysis as separate from geometry. Only at this point did Fourier oﬀer an eﬀective method for ﬁnding the coeﬃcients of the sine series, namely the now-standard method based upon the orthogonality of sine terms of the series (see Fourier [1807, 216–217]).429 By this method he also obtained the general cosine series (see Fourier [1807, 223–224]) and the coeﬃcients of the full series ∞

1 f (x) = a0 + (an cos nx + bn sin nx) 2 n=1

for an arbitrary function (see Fourier [1807, 260–262]). It is interesting to note that while studying the annulus, he applied the method, in particular, to the function 1, 0≤x≤π f (x) = 0, π ≤ x ≤ 2π

Fourier’s treatise contained many other interesting results, some of which concerned those that were later referred to as Bessel’s functions. I do not illustrate these results, since the above description is suﬃcient to clarify the novelty of his approach. Although Fourier employed formal manipulations and geometrical arguments to support his theses, he used series based on the idea that the relation between series and its sum was only a quantitative relation. In so doing he succeeded in giving an analytical form to certain discontinuous functions and in enlarging the bounds of analysis. Fourier’s treatise opened up a series of new problems that his followers were to pursue vigorously.

429 This method was already known to Euler and had been published posthumously in 1798 (see p. 282).

32

Gauss and the hypergeometric series

The concept of analysis as a theory of functions based upon a formal methodology was predominant in Germany during the second half of the 18th century and the ﬁrst decades of the 19th century.430 Gauss was brought up within this cultural context. He also had a ﬁrsthand knowledge of the works of Euler and Lagrange: Many abstracts of their writings, which dated back to the years in which Gauss studied at Brunswick Collegium Carolinum (1792– 1795), were found in his Nachlass (see Gauss [WW, 64]). One insight into views during the years of his cultural development is provided by an early work, the De Integratione formulae diﬀerentialis (1 + n cos ψ)ν dψ (Gauss [WO]), whose ﬁnal draft dates back to the years at G¨ ottingen University (see Gauss [WW, 8:64]). In this paper, Gauss stated that integration of a function f (x) was often obtained by expanding f (x) into ﬁnite or inﬁnite series, but while ﬁnite series provided a genuine expression of the integral, inﬁnite series furnished a less perfect relation. It is evident that Gauss still thought that inﬁnite series had a status diﬀerent from that of closed expressions. Nonetheless, in his opinion, inﬁnite series provided approximations of the sought-after relations; they were therefore suﬃcient for common uses (ad vulgares usus) and, in a certain sense, were equivalent to equations (see Gauss [WO, 35]). In De Integratione, Gauss reformulated Chapter VI, vol. I of Euler’s Institutionum calculi integralis (Euler [1768–70, 1:159–182]). He sought to express the coeﬃcients of the expansions of (1 + n cos ψ)ν

and

(1 + n cos ψ)ν dψ

in a more concise and elegant form as well as provide a rigorous proof (rigida demonstratione) of them. According to Gauss, this rigorous proof was required by the risk of making mistakes, due to the complexity of these coeﬃcients, and by the dignity of science (see Gauss [WO, 35–36]). In this article, Gauss followed the style of the Combinatorial School,431 and the rigor that he claimed was entirely based upon the use of formal methods. A diﬀerent approach lies at the basis of Disquisitiones generales circa α(α+1)(α+2)β(β+1)(β+2) 3 αβ x + etc. x + α(α+1)β(β+1) seriem inﬁnitam 1 + 1·γ 1·2·γ(γ+1) xx+ 1·2·3·γ(γ+1)(γ+2) [1812b] and Determinatio seriei nostrae per aequationem diﬀerentialem secondi ordinis [WA]. In the ﬁrst paper, Gauss listed some frequently used functions that could be obtained from the hypergeometric series 1+

430 431

α(α + 1)β(β + 1) α(α + 1)(α + 2)β(β + 1)(β + 2) 3 αβ x+ xx + x +... 1·γ 1 · 2 · γ(γ + 1) 1 · 2 · 3 · γ(γ + 1)(γ + 2) (215) On this topic, see Jahnke [1993]. See Chapter 28.

323

324

The Decline of the Formal Theory of Series

by giving particular values to the parameters α, β, γ 432 and then explained the goal of his research in the following way: the previous functions are algebraic, logarithmic and circular transcendent. In no case, however, do we undertake our general inquiry because of these, but rather to promote the theory of higher transcendental functions, our series containing a very large number of them. (Gauss [1812b, 128], italics in original) Similar expressions are found in his Anzeige der Disquisitiones generales circa seriem inﬁnitam [1812a, 197], where Gauss announced his [1812b]. To achieve his aim, Gauss began by trying to deﬁne hypergeometric functions appropriately. In eﬀect, he gave two diﬀerent deﬁnitions (the ﬁrst is found in [1812b], the second in [WA]); both are of great interest since they per se implied the rejection of the principle of the generality of algebra433 and a radical change of the notions of functions, series, integrals, and complex numbers. In Disquisitiones generales, Gauss denoted series (215) by F (α, β, γ, x) and sought the conditions under which (215) could actually be considered as a function of x. He ﬁrst observed that γ cannot be either 0 or a negative integer (to avoid inﬁnitely large terms) and that, when α and β are either negative integers or zero, the series exhibits a rational function. Then Gauss [1812b, 126] determined the convergence of the series by the ratio test. Indeed, the ratio of the coeﬃcients of xm and xm+1 is α+β γ+1 γ αβ (m + 1)(γ + m) (216) = 1+ + 2 : 1+ + 2 , (α + m)(β + m) m m m m and (216) becomes ever closer to unity as m increases. Therefore, if x is a real number such that |x| < 1, the series is decreasing to zero,434 at least 432 Gauss provided 23 examples of elementary functions that are particular cases of (215). Among them:

(t + u)n = tn F (−n, γ, γ, −u/t), log(1 + t) = tF (1, 1, 2, −t),

et = F (1, k, 1, t/k),

3 −t2 ), sin t = tF (k, k ′ , , 2 4kk′ 2 1 −t cos t = F (k, k ′ , , ), 2 4kk′ where γ is any number, k and k′ denote inﬁnitely large numbers (he intended k → ∞ and k′ → ∞), and the variable t assumes appropriate values (Gauss [1812b, 127]). 433 See Section 18.2, p. 209. 434 Here Gauss used the expression “convergent”. In [WC, 400], Gauss deﬁned a series an to be convergent if the sequence an approaches zero. As stated in footnote no. 2, I use the term “convergent” in Cauchy’s sense.

32

Gauss and the Hypergeometric Series

325

after √ a certain term, and has a ﬁnite sum. The same occurs if x has the form a + b −1, provided |x|2 = a2 + b2 < 1. Instead, if x is a real or complex number such that |x| > 1, then the series is increasing (if not initially, at least after a certain term); consequently, one cannot refer to the sum of the series. After postponing the more complicated case x = 1 to Section 3 of [1812b],435 Gauss stated: It is clear that since our function is deﬁned to be the sum of the series, the inquiry as to its nature is restricted to the case in which the series actually converges. Therefore it is inappropriate to ask about the value of the series for x greater than unity. (Gauss [1812b, 126]) Even though Gauss only brieﬂy mentioned that the function F (α, β, γ, x) was deﬁned as the sum of the series (215) and that he restricted himself to considering the values of x for which the series (215) was convergent,436 this statement implied a radical shift in the 18th-century concept of function and series. Gauss rejected the restricted conditions a relation had to satisfy in order to be considered a function; rather than regarding the hypergeometric series as the expansion of a generating function, he viewed it as a function in its own right. Thus, the Eulerian notion of the sum was discarded and was replaced by the idea that the sum of a series was the limit of partial sums.437 The role of convergence also changed: from being an a posteriori condition for the application of formally derived results, it became the preliminary condition for using a series (for this reason Gauss used convergence criteria in a modern sense). Moreover, unlike the few 18th-century mathematicians who criticized the use of divergent series, Gauss actually developed an adequate methodology for making this rejection concrete. The second deﬁnition of hypergeometric functions also introduced new features into 18th-century analysis. In Determinatio seriei nostrae, Gauss observed that αβ dF (α, β, γ, x) = F (α + 1, β + 1, γ + 1, x) dx γ and αβ(α + 1)(β + 1) d2 F (α, β, γ, x) = F (α + 2, β + 2, γ + 2, x). dx2 γ(γ + 1) 435

See p.333. Gauss underlined this point in a letter to Bessel on November 21, 1811, where he considered the deﬁnition of a function by means of a divergent series to be utterly despicable (see Gauss [WW, 10:362–363]). 437 See also Gauss [1812b, 143]. 436

326

The Decline of the Formal Theory of Series

Since γ(γ + 1)F (α, β, γ, x) − (γ + 1)(γ − (α + β + 1)x)F (α + 1, β + 1, γ + 1, x) − (α + 1)(β + 1)x(1 − x)F (α + 2, β + 2, γ + 2, x) = 0,

he set P = F (α, β, γ, x) and derived αβP − (γ − (α + β + 1)x)

d2 P dP − x(1 − x) 2 = 0. dx dx

(217)

Gauss stated that this diﬀerential equation could be considered as another deﬁnition of (215). However, P = F (α, β, γ, x) was not the general integral but a particular solution; therefore, it was appropriate to add the conditions αβ(α+1)(β+1) αβ d2 P to this deﬁnition (Gauss [WA, P (0) = 1, dP dx = γ , and dx2 = γ(γ+1) 207]). According to Gauss, the solution P to (217) was a more general function than F (α, β, γ, x), which was deﬁned for |x| < 1. Indeed, P is a manyvalued function deﬁned in all of the plane except for 0 and 1.438 Gauss described this by stating that if one moved from 0 to x in a continuous way but did not arrive at 1 (where the coeﬃcient x − x2 was null), then P was a perfectly determinate quantity. However, one could achieve the real values of x greater than 1 only if one followed a path that went around the point 1 by passing through imaginary numbers. Since this could be performed in a continuous way according to diﬀerent paths, then P assumed many diﬀerent values. Euler already knew that the hypergeometric series was the solution to diﬀerential equation (217). In [1794a], Euler considered the hypergeometric series without taking convergence into account and showed that it satisﬁed (217). He then used the hypergeometric diﬀerential equation to prove439 (172). Euler did not use (217) to introduce a new type of functions (as we saw, a relation had to satisfy certain conditions in order to be considered a function). Instead, Gauss thought that a relation by itself was a function: Since (217) provided a relation between x and P, he could consider it as a deﬁnition of hypergeometric functions. It is clear that, since Gauss deﬁned hypergeometric functions by (217), his notion of integration diﬀered from those of Euler and Lagrange.440 A brief reference to his interpretation of the integral can be found in his letter to Bessel dated December 18, 1811. Even though Gauss was discussing 438

Gauss did not consider the case x = ∞. See p. 269. 440 The notion of integration as antidiﬀerentiation presupposed that the function (which was the solution to the diﬀerential equation) was known before it was recognized as the solution to the equation. Apart from the considerations in Chapter 25, this notion did not allow an integral or an diﬀerential equation per se to be used as the deﬁnition of a new function. Gauss’s concept of integral removed this diﬃculty. 439

32

Gauss and the Hypergeometric Series

327

complex integration and the letter was mainly devoted to the problem of many-valued functions, the idea of the integral that emerges is a general one. Gauss wrote: What should one think of for ϕxdx = a + ib? Evidently, if one wants to proceed by clear concepts, one must assume that x passes by inﬁnitely small increments (each of the form α + βi) from the value for which the integral is equal 0 to a + bi; and then that all the products ϕxdx are summed. In this way, the previous written expression is meaningful . . . . The transition from one point to another occurs without touching a point where ϕx = ∞. I therefore require thatone avoids the points where the original concept of the integral ϕxdx (the basis of the whole) manifestly loses its own clarity and easily arouses contradictions. Apart from this, it is evident that if a function is introduced by means of ϕxdx, the value corresponding to the same x can be more than one. (Gauss [WW, 10:366]) Gauss expressed the ancient Leibnizian deﬁnition of the integral in an abstract form441 and assumed that the relation x x→ ϕ(x)dx a

led to a new function provided ϕ(x) = ∞ along the path of integration. Similarly, Equation (217) provided a relationship between certain quantities x and P except for the points 0 and 1, where the coeﬃcients were inﬁnite and therefore led to a new function. Both deﬁnitions of hypergeometric functions highlight other novelties in Gauss’s conception concerning the principle of the generality of algebra. For instance, in Disquisitiones generales, Gauss explicitly veriﬁed that hypergeometric functions were also deﬁned for complex values of variables. In this way, Gauss revealed his awareness of the diﬀerence between complex and real variables and rejected the automatic extension of formulas valid for real quantities to imaginary quantities.442 This was possible because his approach to complex numbers diﬀered from that of 18th-century mathematicians. To use Gauss’s words, hitherto imaginary numbers were conceived of 441 In a manuscript entitled Bestimmung der convergenz der reihen, in welche die periodischen functionen einer anderlichen gr¨ osse entwickelt werden [WC, 407–419], Gauss ver¨ stated that the integral f (t)dt could be deﬁned in two diﬀerent senses. In a ﬁrst sense, the integral could be considered to be a function of t such that f (t)dt resulted from its diﬀerentiation. In the second sense, a deﬁnite integral from t0 to T = tn was deﬁned to be the limit of the sums n i=1 (ti − ti−1 )f (ti−1 ) as (ti − ti−1 ) → 0. This paper was written after 1831 and might have been inﬂuenced by Cauchy. In 1811, Gauss seems to be close to the idea of integral as the sum of inﬁnitely many inﬁnitesimals. 442 See Legendre’s deﬁnition of gamma function, Chapter 29.

328

The Decline of the Formal Theory of Series

¨ as “excrescences” (Uberbein) of real quantities (see Gauss [WW, 10:366]). In [1831], he stated: The imaginary quantities — which are contrasted with the real ones, and which were formerly, and are still occasionally (although improperly) called impossible — are still tolerated rather than fully accepted, and therefore appear more like a game with symbols, in itself empty of content, to which one unconditionally denies a thinkable substrate — without, however, wishing to scorn the rewards which this game with symbols achieved for our understanding of the relationships of the real quantities. (Gauss [1831, 175]) Gauss thought that complex numbers were to be regarded as just as possible as real quantities and that they had the same legitimacy as real numbers.443 He justiﬁed his view by stating: One does not discuss practical aims, since, in my opinion, analysis is an autonomous science; by leaving out imaginary magnitudes, it would lose its beauty and its truths, which are valid in general, and would be strongly bounded. (Gauss [WW, 10:366]) According to Gauss, complex numbers have no practical scope. The fact that he used them in scrutinizing hypergeometric functions may appear to contradict the above observation that remarkable stimuli for the introduction of new functions derived from mathematical applications. However, while practical incentives for pure mathematics are one matter, pure mathematics is another question. Gauss required analysis to be general and beautiful, where generality and beauty seem to indicate the capability of developing a deductive theory that derived from a few clear principles, had an internal coherence and harmony,444 and was linked to the real by means of the intuitiveness of its principles. Complex numbers allowed a higher degree of generality and beauty; for this reason, they were to be retained in analysis — independently of the fact that they were used in applications or not — provided they could be established on a suﬃciently solid basis (see p. 32). ∗ ∗

∗

The introduction of the factorial and digamma functions in Section 3 of Disquisitiones generales and Gauss’s criticisms of their previous “deﬁnitions” 443

See Gauss [1799, 6], [1831, 171], and [WW, 10:366]. This idea of analysis was widely shared at the turn of the 18th century, in particular in the cultural environment of Germany (see Jahnke [1993]). 444

32

329

Gauss and the Hypergeometric Series

oﬀers further remarkable points of interest for understanding his approach. In [1812b, 144–145], Gauss set Π(k, z) =

1 · 2 · 3 · ... · k kz , (z + 1)(z + 2)(z + 3) . . . (z + k)

(218)

where k is a positive integer, and proved that the limit of (218) for k → ∞ exists. Indeed, given a number h > |z|, he obtained log Π(h + 1, z) − log Π(h, z) =

n ∞ z(1 + (−1)j+1 z j ) j=1

j+1

i=1

1 (h + 1)j+1

Since this series is ﬁnite as n goes ad inﬁnitum, limk→∞ Π(k, z) is ﬁnite445 and exhibits a determinate function of z, provided z diﬀers from a negative integer number [1812b, 145]. At this point Gauss deﬁned the factorial function446 by setting Π(z) = lim Π(k, z) k→∞

or Π(z) =

∞ (k + 1)z+1 . k z (z + k)

(219)

k=1

Gauss then noted that, since

Π(k, z + 1) = Π(k, z)

1+z , 1 + 1+z k

one had Π(z + 1) = Π(z + 1)Π(z). Hence, Π(z + n) = (z + 1)(z + 2)(z + n)Π(z) and Π(z) = 1 · 2 · . . . · z (see Gauss [1812b, 145–146]). He stated: [T]his property of our function has badly been passed oﬀ as its deﬁnition, since it is limited to integer values by its own nature and, beside our function, is common to an inﬁnity of others (e.g., Π(z) cos 2πz, (z)cos2n πz,. . . ).447 (Gauss [1812b, 146]) 445

Gauss used the symbol Π(∞, z). See footnote no. 223. 447 1 existed and deﬁned the Similarly, Gauss proved that the limit of log k − kj=1 z+j k 1 digamma function as Ψ(z) = limk→∞ log k − j=1 z+j (see [1812b, 153–154]). In this chapter, by “digamma function”, I mean the logarithmic derivative of the factorial function (see footnote no. 347). Gauss denoted this function by the sign Ψ(z) but gave no name to it. 446

330

The Decline of the Formal Theory of Series

Gauss’s criticism was directed toward the deﬁnition of the factorial function as a quantity that derived from the process of interpolation of the discrete sequence448 n!. Gauss explicitly referred to Eulerian inexplicable functions in [1812a, 200–201] and in his correspondence.449 He dealt diplomatically with Euler’s work. However, he was more explicit in his criticisms of Kramp, who followed Euler’s approach in his Analyse des r´efractions astronomiques et terrestres [1799]. For instance, in [WW, 10:362], Gauss stated that if one did not wish to expose oneself to endless paralogism, paradoxes, and contradictions, as Kramp had done, then 1 · 2 · . . . · z could not be used as the deﬁnition of Π(z). It is worthwhile noting that, in the second part of the 18th century, the objection that the function interpolating n! was not unique was already known. Indeed, in his De serierum determinatione seu nova methodus inveniendi terminos generales serierum, Euler already observed that if a sequence an were interpolated by a function f (n), it may also be interpolated by y = f (n) + g(n) sin πn (Euler [1750–51b, 465–467]), where g(n) is any function. In De curva hypergeometrica hac aequatione expressa y = 1 · 2 · 3 · . . . · x, Euler explicitly stated that numberless curves interpolated the points (n, n!), for n = 0, 1, 2, . . ., but the nature of the equation (ratio equationis) y(n) = n! required y(x + 1) = (x + 1)y(x) or y(x − 1) =

y(x) , x

and therefore the curve had to satisfy this condition. In eﬀect, Euler rethought the problem of Wallis’s interpolation in a form that can be expressed as follows:450 (INT) If a sequence an is given in a recurring form an = F (an−1 , an−2 , . . . , an−r ), for n > r, ﬁnd the function y(x) such that y(x) = F (y(x − 1), y(x − 2), . . . , y(x − r)). 448

See Chapters 1, 4, and 13. See the letters to Olbers on October 17, 1811, and to Bessel on November 11, 1811, in Gauss [WW, 10: 361–363]. 450 See Euler [1750–51b]. 449

32

Gauss and the Hypergeometric Series

331

It might seem that Gauss’s criticism cannot be applied to this way of approaching the question. However, there is another sense in which this criticism can be interpreted. According to Euler, the interpolating function derived from “the nature” of the sequence n!. He made this assertion because a sequence an was not viewed during the 18th century as an arbitrary relation between the set of natural numbers and the set of real numbers but as the result of some ﬁxed analytical operations upon natural numbers;451 in other words, the numbers an were regarded as being generated by an analytical formula f (n) = an . Consequently, “ﬁnding the interpolating function f (n) of an ” meant “ﬁnding the function f (n) that generates an .” For instance, in his De termino generali serierum hypergeometricarum, Euler derived Δ(n) =

1 · 2 · ... · i (i + α)n , (n + 1)(n + 2) . . . (n + i)

(220)

where i is an inﬁnite number and α is an arbitrary number (Euler [1789, 141–146]). Formula (220) is de facto the relation Gauss used to deﬁne the factorial function; however, for Euler, it was not a deﬁnition of the quantity interpolating 1 · 2 · . . . · n, but only a property of this quantity. To give a ﬁrmer foundation to this procedure, in his [1768], [1789], and [1819], Euler illustrated the problem of interpolation by referring to curved lines. In De termino generali [1789], for instance, he stated that the points (n, n!) could be considered as points belonging to a curved line: Clearly, there is no doubt that determinate ordinates correspond to abscissas expressed by means of fractions or surd [irrational] numbers as well. (Euler [1789, 140]) However, he noted that the equation y = 1 · 2 · . . . · n could not be used to determine these ordinates but rather that it was necessary to have an appropriate expression of them. Apart from any considerations about the possibility of a geometrical construction of the curve, it should be observed that an explicitly geometrical foundation was not satisfactory for mathematicians such as Euler who wanted to separate analysis from geometry. Gauss viewed the matter diﬀerently. Functions were our own creations and their deﬁnition was a question of suitability. No function f (x) was the extension of a given sequence an = f (n) as a result of its own nature (see Gauss [WW, 10:363]). Thus, the deﬁnition of the factorial function implied a choice between the numberless functions satisfying the condition f (n) = 1 · 2 · 3 · . . . · n. This choice originated from our requirement that the factorial function satisﬁed certain constraints such as f (n+1) = (n+1)f (n), 451

See Section 13.1.

332

The Decline of the Formal Theory of Series

although this did not mean that the function derived from the “nature” of the given sequence. Gauss did not consider a sequence to be generated by an analytical expression. He explicitly discussed the notion of a sequence in Grundbegriﬀe der Lehre von den Reihen, where he observed that, in a very general sense, the totality of any number of arbitrary quantities might be referred to as a sequence (Gauss [WC, 390]). According to Gauss, this deﬁnition was of little use and he therefore gave a less general deﬁnition: (D1) A sequence is the totality of quantities which are determined by a law that established the ﬁrst, the second, . . . (Gauss [WC, 390]). Gauss compared (D1) with the common deﬁnition: (D2) A sequence is the totality of the values of a function of a variable, when one makes this variable equal to 1, 2, 3, . . . (Gauss [WC, 390]). In this case Gauss used the term “function” to refer to 18th-century functions. He argued that (D1) was more general than (D2), unless the term “function” was understood in a broader sense than previously was the case. Gauss noted that, according to the common deﬁnition, the sequence of prime numbers 1, 2, 3, 5, 7, 11, . . . and each sequence deriving from it 1 1 1 , 49 , 121 , . . .) were not to be considered as sequences since (such as 1, 41 , 91 , 25 they were not obtained from analytical expressions. He thought that this was not acceptable and therefore that deﬁnition (D1) was a better one. To diﬀerentiate the traditional and more restricted sense of the term from his wider meaning, Gauss proposed the name “analytical sequences” to those sequences whose general term was an analytical function452 of the index (see Gauss [WC, 390–391]). ∗ ∗

∗

In the 18th century, inﬁnitely large and inﬁnitesimal quantities were thought to be entities that could be manipulated in the same manner as numbers except for the principle of cancellation. Instead, according to Gauss, a lot of circumspection was required in treating inﬁnite quantities. In our opinion, these are to be allowed in analytical reasoning in so far as they can be reduced to the theory of limits. (Gauss [1812b, 159])453 452

On the meaning of term “analytical function”, see footnote no. 340. More explicitly, in a letter to Schumacher, Gauss stated: “I protest in the ﬁrst place against the use of inﬁnite magnitude as something completed, which is never allowed in mathematics. The inﬁnite is but a fa¸con de parler in that one actually speaks of limits to which certain relations come as close as one desires, while others are allowed to increase without bound” (Gauss [WW, 8:216] translated in Ewald [1996, 303]). 453

32

333

Gauss and the Hypergeometric Series

In Disquisitiones generales, these words annotated a paradoxical situa 1 1−x λ tion concerning the integral 0 1−x dx. By the substitution x = y k , Gauss obtained 1 1 k−1 1 − xλ ky − ky λk+k−1 dx = dy. 1 − yλ 0 1−x 0 Hence, by changing x into y, he had

1

1 − xλ dx = 1−x

1

kxk−1 − kxλk+k−1 dx. 1 − xλ

(221)

1 λ Gauss observed that equality (221) is legitimate only if 0 1−x 1−x dx is a ﬁnite quantity (namely, if λ + 1 > 0). Indeed, the diﬀerence between the two integrals is inﬁnitely large for λ + 1 ≤ 0 and, therefore, Equation (221) is paradoxical if one considers it without constraints. Gauss’s notion of limit is similar to the modern notion in many aspects but also presents crucial diﬀerences. To make this clear, let us consider the following proposition. Proposition (A). Given a sequence Mk , k = 0, 1, 2, . . . such that M are of the type the ratios Mk+1 k Mk+1 Pλ (m + k) = , Mk pλ (m + k)

k = 0, 1, 2, ..,

where m is an positive integer, Pλ (t) = tλ + A1 tλ−1 + A2 tλ−2 + . . . + Aλ , pλ (t) = tλ + a1 tλ−1 + a2 tλ−2 + . . . + aλ , then the series Mk converges if A1 − a1 < −1 (see Gauss [1812b, 139–143]). According to Gauss, this proposition enables one to investigate convergence of the hypergeometric series at x = 1 “with all rigor, for the beneﬁt of those who are inclined to rigorous methods of the ancient geometers” [1812b, 139]. In proving Proposition (A), Gauss [1812b, 139–143] employed inequalities in a clear and precise way. In the ﬁrst step of the proof, Gauss demonstrated that one has |Mk+1 | > |Mk | or |Mk+1 | < |Mk | for large enough k. He used the following unproved assumption, which contains an implicit deﬁnition of limt→∞ f (t) = ∞. Lemma 1. Given a number a, if F (t) → ∞ as t → ∞, it is possible to determine a number L such that F (t) > a for t > L.

334

The Decline of the Formal Theory of Series

Indeed, Gauss observed that if Pλ (t) and pλ (t) are not identical, then one of the diﬀerences Ak − ak , k = 1, 2, 3, . . ., is not equal to 0. If one supposes Ak − ak = 0 for k = 1, 2, 3, . . . , r − 1 and Ar − ar > 0, then one can ﬁnd a bound (limes) L such that Pλ (t) > pλ (t) > 0 for t > L. It is suﬃcient to consider L equal to the greatest real root of the equation pλ (t) (Pλ (t) − pλ (t)) = 0.

If this root does not exist, then pλ (t) (Pλ (t) − pλ (t)) and pλ (t) are positive for every value of t. Hence, Mk+1 > 1, Mk Mk always assumes the same sign, and |Mk+1 | > |Mk |. Similarly, if Ak −ak = 0 for k = 1, 2, 3, . . . , r − 1 and Ar − ar < 0, then |Mk+1 | < |Mk |. In the second step of the proof, Gauss showed that |Mk | → ∞

for A1 − a1 > 0

|Mk | → 0

for A1 − a1 < 0

and (see [1812b, 140]). In particular, he used the implicit lemma: Lemma 2. Given the sequences αk → ∞ and β k > 0 and the number Θ > 0, if β k > Θ, for large enough k, then αk β k → ∞; if 0 < αk β k < Θ, for large enough k, then β k → 0. In third step, Gauss demonstrated that if A1 − a1 = 0, then |Mk | goes to a nonzero finite limit, and he implicitly used the following unproved assumption: Lemma 3. Given an increasing sequence λk , if this sequence has an upper bound, then there exists a real number Λ that is the limit of λk for k → ∞. He indeed supposed, for instance, that the first difference Ai −ai different from zero is A2 − a2 and that A2 − a2 is greater than 0. He considered a positive integer h such that h + a2 − A2 > 0 and posed h m+k Mk = Nk , k = 0, 1, 2, . . . . m+k−1 Q = (t2 − 1)h Pλ (t) = tλ+2h + A1 tλ+2h−1 + (A2 − h)tλ+2h−2 + . . . q = t2h pλ (t) = tλ+2h + a1 tλ+2h−1 + a2 tλ+2h−2 + . . .

32

Gauss and the Hypergeometric Series

335

By applying the ﬁrst step (it is A2 − a2 − h < 0), he had that the series |Nk | becomes decreasing. Since |Mk | increases (A2 − a2 > 0) and |Mk | < |Nk |, he concluded that |Mk | is convergent to a ﬁnite limit. In the successive steps, Gauss completed the proof by using the comparison test. This proof is exemplary of Gauss’s conception for two reasons. First, he rejected inﬁnite quantities as something complete and replaced them with the inequality technique. As we saw previously, the inequality technique was not new, but d’Alembert and Lagrange considered it only in relation to approximations. Instead, Gauss used them to prove the existence of a limit, and this represented a crucial breakthrough. This does not mean that the importance of 18th-century studies on inequalities should be underestimated. They were part of Gauss’s background, and a certain knowledge of them might have helped Gauss in developing his approach by providing an important technical tool. Nevertheless, Gauss’s approach diﬀered conceptually from d’Alembert and Lagrange’s. It is impossible to reduce the former to the latter. Second, Gauss’s idea of the continuum diﬀered from the modern notion. Consider Lemma 3. Gauss did not provide the proof of the existence of the number Λ (the supremum of the series). A theorem concerning the existence of the supremum is found in his Grundbegriﬀe der lehre von den Reihen, but it is based upon reasoning that does not provide a solution to the question. Indeed, Gauss considered a series an such that an < λ, for a given number λ. He deﬁned this number to be the upper bound of a sequence (in Gauss’s terms: limes supra seriem, or obere Grenze or une limite en plus). He stated that if λ is an upper bound, then any number greater than λ is also an upper bound. Moreover, given a number κ less than a ﬁxed upper bound λ, if κ was not an upper bound and was allowed to assume the value λ by moving through all the intermediate quantities in a continuous way, it necessarily reached the least upper bound L (the supremum of the series). Gauss noted that no term of the series could be greater than L, but, when n was suﬃciently large, there are terms that exceed any number ξ < L (Gauss [WC, 391]). Gauss characterized the supremum adeptly; nevertheless, an adequate construction of real numbers was lacking. (Note that the requirement of a proof of Lemma 3 is precisely the starting point of Dedekind’s theory of real numbers in his Stetigkeit und irrationale Zahlen [1872].) Gauss did not feel the need to construct the set of real numbers, as his concept of the continuum was still linked to a revised notion of continuous quantity, which was a legacy of the 18th century. ∗ ∗ ∗ In the previous chapters, it was observed that from about the 1740s, mathematicians based analysis upon the notion of a general (continuous)

336

The Decline of the Formal Theory of Series

quantity, which was merely the result of a process of abstraction practized upon geometrical quantities (which, in turn, were abstracted from physical quantities). There was a close link between the abstract notion of general quantity and concrete and speciﬁc geometrical quantities. Thus, even if analysis was conceived to be independent of geometry, continuous quantities were substantially intuited as segments of a straight line or a piece of a curved line (in Kantian terms, they derived from empirical intuition). Instead, Gauss’s analysis no longer dealt with a notion of quantity drawn from geometrical sources but was based on the intuitively given numerical continuum:454 continuous quantity was reduced to the mere intuition of a sequence of numbers, which ﬂows from one value to another assuming all intermediate values, without jumps. In Gauss’s opinion, the numerical continuum was the basic and intuitive object of analysis and was connected with any sort of a priori intuition (namely, with a form of knowledge that does not appeal to any particular experience). His conception was based upon modiﬁed Kantian notions.455 Kant believed that space and time were a priori intuitions from which the certainty of geometry and arithmetic derived. It is well known that Gauss did not accept the whole of Kant’s theory of mathematical knowledge and, in particular, the idea that space is an a priori intuition (he wrote to Bessel on January 27, 1829: “We cannot establish geometry entirely a priori ” [WW, 8:200]). However, he agreed with some crucial points of Kantian philosophy: the importance of intuition, the existence of an a priori intuition that guarantees the certainty of mathematical knowledge, the fact that mathematical intuition concerns abstract objects (universal objects) but can only be disclosed by empirical objects (universal in concreto). Thus, in his review of J.C. Schwab’s Commentatio in primum elementorum Euclidis librum, Gauss stressed the importance of the intuition: A great part of the text [of Schwab] turns on the contention against Kant that the certainty of geometry is not based on intuition but on deﬁnitions and on the principium identitatis and the principium contradictionis. Kant certainly did not wish to deny that use is constantly made in geometry of these logical aids to the presentation and linking of truths: but anybody who is acquainted with the essence of geometry knows that they are able to accomplish nothing by themselves, and they put forth only sterile blossoms unless the fertilizing living intuition of the 454 Here, I use the term “intuition” in the sense that Gauss gave to it in Zur Metaphysik der Mathematik : “We can have an idea of quantity in two ways, either by immediate intuition (an immediate idea), or by comparison with other quantities given by immediate intuition (mediate idea)” (Gauss [WD, 57] translated in Ewald [1996, 294]). 455 Gauss wrote little about this question, although he displayed knowledge of Kant’s mathematical concept, which he criticized only partially.

32

Gauss and the Hypergeometric Series

337

object itself prevails everywhere. (Gauss [1816, 172] translated in Ewald [1996, 299–300]) Intuition as an immediate representation of an object seems to be what actually allows one to produce and understand mathematics. According to Gauss, this intuition is, at least partially, a priori. On this point he viewed the matter diﬀerently from Kant. On September 4, 1830, he wrote: It is my deepest conviction that the theory of space has a completely diﬀerent position in our a priori knowledge than does the pure theory of quantity. Our knowledge of the former utterly lacks the complete conviction of necessity (and also of absolute truth) that belongs to the latter, for number is merely the product of our mind. (Gauss [WW, 8:201] translated in Ewald [1996, 299–302]) It is probable that, following Kant,456 Gauss thought that the numerical continuum originated from a more basic intuition: the a priori intuition of time. In Zur Metaphysik der Mathematik, Gauss stated that quantities are time, number,457 and geometrical quantities (Gauss [WD, 57]). He thought that a quantity could be represented by numbers by measuring it with regard to a ﬁxed quantity, which was considered to be the unit. Gauss did not clarify what he meant by the measure of an incommensurable quantity and irrational numbers (his Zur Metaphysik der Mathematik stops after a few pages). However, he seems to have considered irrational numbers as determinations of a continuum that can be approximated by rationals and individuated by means of an endless number of steps. Thus, time is crucial for avoiding geometrical quantity and for giving the impression of analysis as a mere creation of our mind. Nevertheless, in mathematical practice, time can be easily replaced by its numerical representation, and Gauss directly operated upon a continuum viewed as an unbroken ﬂow of numbers. By describing the continuum as an unbroken ﬂow of numbers, I do not oﬀer a deﬁnition but merely a circumlocution to suggest Gauss’s idea of the continuum: Gauss’s continuum cannot be really deﬁned but only intuited. It is worthwhile noting that this notion deﬁnitively enables a continuous quantity to be separated from a geometrical substratum; continuity, which earlier was referred to as an empirical intuition (by means of geometry), is now reduced to the realm of numbers, even though by means of the a priori intuition of time. This does not mean that the numerical continuum is reduced to the discrete. There is no construction of the continuum: It is immediately given 456

Kant had provided a theoretical basis for the numerical continuum by means of the notion of time as a pure intuition. 457 Here number means whole number or, at most, rational number.

338

The Decline of the Formal Theory of Series

(with its ordinate structure and the property of completeness). Numbers do not generate the continuum; rather, the numerical continuum is given and one can cut it and determine any single, speciﬁc number (see Bussotti [2000]). Apart from Kant’s epistemological inﬂuence, Gauss’s concept of the continuum is similar to the notion of the continuum that was widely held in the 17th and 18th centuries.458 For this reason, the main consequences of the 17th and 18th centuries concept of the continuum also held for Gauss’s numerical continuum, which, in principle, embodies the ﬂow of time. For instance, an interval was always closed, pointwise deﬁnitions of function were impossible, and functions were considered on closed intervals. At this juncture, a remark on Gauss’s concept of analysis is appropriate. Gauss thought that analysis regarded arithmetic relations between quantities. He intended arithmetic relations in a broad sense, in contrast to geometrical relations. The latter concerned geometrical quantities and considered them with respect to location; the former considered quantities “only insofar as they are quantities”. In Gauss’s opinion, analysis was grounded upon some quantities of which we have an immediate —intuitive— idea. Natural numbers 1, 2, 3, . . . and continuous quantity (the continuum) seem to be the only immediate quantities. However, Gauss employed other simple intuitive principles and notions (to be part of, greater than, relationship, to be right or left, . . .). Basic principles and notions of analysis were not provided by axioms in the modern sense (arbitrary propositions) but were provided by intuition.459 More precisely, with regard to analysis, they were provided by a priori intuition. This is a sophisticated, Kantian way of saying that mathematics expressed reality, where reality is taken as the mental interpretation of sensorial perception.460 According to Gauss, other quantities must be reduced to immediate quantities: The duty of the mathematician is accordingly either actually to represent the sought-for quantity (geometric representation or construction), or to indicate the way and manner in which, from the idea of an immediately given quantity, one can achieve the idea of the sought quantity (arithmetical representation). (Gauss [WD, 57–58]) Thus, one can introduce new quantities, but these are not arbitrary creations. They are connected with original intuition and must have a model (a sensible representation that allows one to understand them intuitively). An example is the introduction of complex numbers. 458

See Section 7.2, p. 102. Analysis has a diﬀerent status from geometry (see [WW, 8:201]). 460 See also Gray [1992, 229]. 459

32

Gauss and the Hypergeometric Series

339

In his [1831], Gauss assumed that certain objects A, B, C, D, . . ., were ordered into a sequence and that this was unbounded on both sides. He stated that if the relation (or the transition) of A to B counted as +1, then the relation of B to A had to be represented by −1. So every real integer represented the relation of an arbitrary member chosen as the origin and a determinate member of the sequence. Then he supposed the objects were of such a sort that they could not be ordered in a sequence but only in sequences of sequences. In this case, for the measurement of the transition from one term of the system to another one needed besides the previous units +1 and −1 two others, also inverse to each other, +i and −i (Gauss [1831, 176]). Complex numbers were viewed as abstract relations among abstract objects. However, these relationships could “be brought to intuition only by a representation in space” [1831, 176–177] and, for this reason, Gauss constructed the model of complex number in the plane today associated with this name: √ Here, therefore, an intuitive meaning of −1 is completely established, and one needs no other to admit these quantities in the domain of arithmetic (Gauss [1831, 177] translated in Ewald [1996, 313]). In such a way new quantities were not arbitrary creations but were connected with original intuition and had a model (a concrete representation). In his [1799, 6], Gauss distinguished between imaginary quantities, which were to be accepted, and impossible quantities, which were to be rejected. He compared these impossible quantities with a right-angle equilateral triangle461 and deﬁned them as those quantities that satisﬁed conditions that could not be satisﬁed even by imaginary quantities. He realized that his argument about impossible quantities could be employed for imaginary quantities and proposed to vindicate them elsewhere. This vindication was published in [1831] and [1832] and consisted of providing the concrete model mentioned above. The existence of a model seemed to guarantee that, unlike the right-angle equilateral triangle, certain objects were possible. ∗ ∗

∗

Gauss considered functions as deﬁned over a segment of the real line or over a region of the whole complex plane (and, if necessary, over the whole real line or the whole complex plane) except for a certain (ﬁnite or inﬁnite) number of isolated singularities. However, as the continuum was not reduced to a set of numbers, Gauss’s functions were not conceived of as relations between elements of two sets, namely, they were not pointwise-deﬁned 461 It is clear that Gauss did not consider the possibility of a geometry where such triangles exist.

340

The Decline of the Formal Theory of Series

relations. In Gauss’s writings, a function462 y = f (z) was a relation between the variable quantity z and the variable quantity y; both variables were intended as continuous ﬂows; of course, if a value of the variable quantity z was determinate, then there was (at least) a value of the corresponding quantity y. This concept is particularly evident in reference to the continuity of functions. In Disquisitiones generales, Gauss stated: Continuity of the function Πz breaks, every time that the value is inﬁnitely large, i.e. for negative integer value of z. (Gauss [1812b, 147]) The ﬂow that the variable y describes can be broken for certain (exceptional) values of z; thus, y = (z) is discontinuous when z = −1, −2, −3, . . .. A point can break continuity though continuity is not a property of a point: Any reasoning implying continuity did not concern a single point, but a part of the continuum. What is important for Gauss is that the independent variable z could move from a point A of the domain of deﬁnition to another B (including the endpoints) in a continuous way and that the corresponding variation of the dependent variable y is continuous as well.463 One could argue that this idea is unclear, in particular the relationship between the single number and the ﬂow of numbers. This is true, and I think that this is probably linked to some problematic derivations of Disquisitiones generales. For instance, Gauss established that F (α, β, γ, 1) =

(γ − α)k (γ − β)k F (α, β, γ + k, 1), (γ)k (γ − α − β)k

(222)

where (x)n = x(x + 1)(x + 2)(x + n − 1). Applying (222) repeatedly, he had F (α, β, γ, 1) =

Π(k, γ − 1) · Π(k, γ − α − β − 1) F (α, β, γ + k, 1). Π(k, γ − α − 1) · Π(k, γ − β − 1)

As F (α, β, γ + k, 1) went to 1 as k tended to inﬁnity, he derived F (α, β, γ, 1) =

Π(γ − 1) · Π(γ − α − β − 1) . Π(γ − α − 1) · Π(γ − β − 1)

462 In his arithmetical works, Gauss introduced some arithmetical functions, although he avoided using the term “function” with reference to them. For instance, in [1801, 30–31], he designated the number of positive numbers which are relatively prime to the given number A and smaller than it by the “preﬁx” or the “symbol” φ, but did not give the name function to φA (see also [1801, 43]). In [1801] he spoke of functions if the variables included in certain expressions could be thought of as continuous. In the present paper, I always refer to functions of a continuous variable. 463 See also Ferraro [2007b, Section 14].

32

Gauss and the Hypergeometric Series

341

Gauss stated that it was evident that F (α, β, γ + k, 1) went to 1 as k → ∞ [1812b, 147]. Since F (α, β, γ + k, 1) =

∞ (α)k (β)k , n!(γ + k)k i=1

he interchanged the operation of limit; however, he did not justify this interchange, as if it was immediate and obvious. Similarly, on p. 151 of his Disquisitiones generales, Gauss considered the equality ν y Π(ν, λ) y λ−1 (1 − )ν dx = . ν λ 0 For ν → ∞ he obtained ∞ 0

y λ−1 (1 − e−y )dx = Π(λ − 1)

(see Gauss [1812b, 151]). Gauss performed some operations that are legitimate in these speciﬁc cases, but their legitimacy was not discussed. He behaved as if certain theorems on the interchange of limits, integrals, etc. were evident. Of course, in contrast to 18th-century mathematicians, Gauss rejected the inﬁnite extension of ﬁnite rules.464 It is likely that Gauss viewed the matter in a manner that resembled Cauchy’s approach;465 however, we lack the evidence for a more precise hypothesis.466 In any case, Gauss’s concept of the continuum made it impossible to distinguish among continuity and uniform continuity, convergence and uniform convergence, which are necessary for a better formulation of the theorems to which I referred above. Although Gauss had a fairly general concept of a function, he was interested in “special” functions that had “particular” characteristics and that played a special role in mathematics, such as Π(z). This special role is due to the utility and simplicity of certain functions. In [1812b, 146–147], Gauss noted that there were other functions similar to Π(z), such as the function of the variables a, b, c that Kramp denoted by ab|c and that was equal to cb Π( ac + b − 1) ; Π( ac − 1) 464

See p. 117. See Chapter 33. 466 Gauss possessed sophisticated knowledge of complex analysis —see, e.g., [WW, 10:368]— though he did not expound it in a systematic way. It is therefore diﬃcult to state which theorems he was aware of. 465

342

The Decline of the Formal Theory of Series

however, Π(z) was preferable since a function of one variable was easier than a function of three variables. In the letter to Bessel on December 18, 1811, Gauss asserted that the function x x e −1 dx x 0 was simpler and could give more incisive results than x 1 dx 0 log x because the former was a single-valued function (see Gauss [WW, 10:368]). Gauss seems to think that a useful function is linked to any form of analytical representation: in his work, a function is given by a series, by the limit of a sequence [for example, Π(z) and Ψ(z)], by an integral,467 by a diﬀerential equation or by the inversion of another function.468 However, Gauss is aware that a function can have diﬀerent representations that are valid in diﬀerent domains. Thus, he notes that the digamma function Ψ(λ) = k 1 limk→∞ log k − j=1 λ+j has the integral representation Ψ(λ) =

1

−

1 zλ − log z 1 − z

dz

(223)

and that (223) is valid if λ + 1 > 0, whereas the function Ψ(λ) is deﬁned for every λ but −1, −2, −3, . . .. In Determinatio seriei nostrae, Gauss explicitly discussed a seeming paradox connected with diﬀerent representations of F (α, β, γ, x).469 Setting γ =α+β+

1 2

and x = 4y − 4y 2 ,

Gauss transformed Equation (217) into 1 d2 P dP 4αβP − α + β + − (2α + 2β + 1) − (y − y 2 ) 2 = 0. 2 dy dy This equation has the particular solution F (2α, 2β, α + β + 21 , y). Hence, 1 1 F (α, β, α + β + , 4y − 4y 2 ) = F (2α, 2β, α + β + , y). 2 2 +∞ x He stated the best deﬁnition of the factorial function is Π(m) = −∞ e(m+1)x ee dx (see Gauss [WW, 3:230]). 468 It is the case of the sinus lemniscaticus, which was deﬁned as the inverse function of x lemniscatic integral 0 √ 1 4 dt. 467

469

1−t

On this paradox, see Gray [1986, 13–14].

32

Gauss and the Hypergeometric Series

343

By changing y by 1 − y, he had 1 1 F (2α, 2β, α + β + , y) = F (2α, 2β, α + β + , 1 − y), 2 2 which is certainly false470 (Gauss [WA, 225–226]). Gauss explained that one had to distinguish between two meanings of the symbol F (α, β, γ, x), namely between F (α, β, γ, x) as the sum of inﬁnite series (215) and F (α, β, γ, x) as the solution to diﬀerential equation (217). If one considers F (α, β, γ, x) as the sum of a series, it has a meaning only in the circle of convergence with radius 1 and center (0, 0) and F (α, β, α + β + 12 , 4y − 4y 2 ) only in the circle of convergence with radius

1 1 2 2 and center

1 1 − 2 2

1 , 0 2

(Gauss [WA, 226–227]). ∗ ∗

∗

In Disquisitiones generales, Gauss tried to oﬀer a reformulation of 18thcentury results (and, especially, Euler’s results) about gamma and digamma functions, accordingly his new methodology. I refer to Ferraro [2007b, Section 13] for an account. Here I limit myself to pointing out that Gauss did not always meet with success in rederiving 18th-century results. Indeed, in [1812b, 152 and 154], he discussed the Euler–Maclaurin expansions of log Π(z) and Ψ(z): 1 1 log z − z + log 2π log Π(z) = z+ 2 2 ∞ B2n (−1)n+1 + 2n(2n − 1)z 2n−1

(224)

n=1

Ψ(z) = log z +

∞

B2n 1 + (−1)n , 2z 2nz 2n−1

(225)

n=1

where Bn are the Bernoulli numbers. He stated that the ﬁrst terms of a series of this sort decrease in absolute value promptly enough for large values of z: 470

See also the paradox connected to formula (221) on p. (333).

344

The Decline of the Formal Theory of Series

This allowed one to compute the approximate values of the functions with relative ease and with a suﬃcient degree of precision. Nevertheless, for any value z, such series were certainly increasing, if they continued suﬃciently. “But one cannot deny that the theory of these divergent series has been hidden up to now by certain diﬃculties, of which I, perhaps, shall discuss more fully in other circumstances” (Gauss [1812b, 152]). Gauss seems to have planned a theory of series that also included divergent series but did not give any further consideration to this question.471 In his Disquisitiones generales, Gauss made another reference to divergent series. Indeed, he had treated the expansion of hypergeometric series into continued fractions and had proved that F (α, 1, γ, x) =

1 a1 x a2 x a3 x a4 x ..., 1− 1− 1− 1− 1−

(226)

where a2n+1 =

(α + n)(γ + n − 1) n(γ + n − 1 − α) , a2n = , n = 0, 1, 2, . . . (γ + 2n − 1)(γ + 2n) (γ + 2n − 2)(γ + 2n − 1)

(see Gauss [1812b, 134–138]). By applying (226), for inﬁnitely large γ, he derived F(

m , 1, γ, −γnt) = 1 − mx + m(m + n)x2 − m(m + n)(m + 2n)x3 + . . . n 1 nx nx (m + n)x 2nx (m + 2n)x 3nx ... = 1+ 1+ 1+ 1+ 1+ 1+ 1+

The series 1 − mx + m(m + n)x2 − m(m + n)(m + 2n)x3 + . . .

(227)

did not satisfy the condition of convergence |x| = | − γnt| < 1; however, Gauss mentioned it without annotations (see [1812b, 138]). In Part III, I noted that both the series of type (224) and those of type (227) had been investigated in the 18th century because of their usefulness and that such a use of divergent series had represented a remarkable success of formal methodology, which helped to allay d’Alembert’s criticisms. In Gauss’s eyes, these successes had become insuﬃcient (when he wrote Disquisitiones generales [1812b], they were already dated; rather, formal methods were an obstacle to the introduction of new functions in analysis 471

Some attempts at summing divergent series can be found in Gauss’s manuscripts; however, they are merely youthful attempts following the 18th-century style (see, for instance, Neue methode die summe der divergierenden reihe 1−1+2−6+24−etc.[=]0, 5963 zu ﬁnden, in [WC, 382-385]). This material is only an example of the approach Gauss used in his youth.

32

Gauss and the Hypergeometric Series

345

and to the development of mathematics). Even if Gauss was aware of the historical importance of formal methodology and of the applications of divergent series in certain ﬁelds of the exact sciences, he marginalized those ﬁndings of the old theory that were diﬃcult to adapt to new concepts in Disquisitiones generales.

33

Cauchy’s rejection of the 18th-century theory of series

This ﬁnal chapter is an analysis of how Cauchy introduced innovations to the 18th-century theory of series; it is therefore not a comprehensive study of all diﬀerent aspects of his series theory. I limit myself to examining Cauchy’s ´ works in the early 1820s and, in particular, to the Cours d’analyse a l’Ecole ´ Royal Polytechnique and R´esum´e des le¸cons a l’Ecole Royal Polytechnique sur le calcul inﬁnit´esimale, published in 1821 and 1823, respectively. These treatises represented the deﬁnitive abandonment of the 18th-century concept and provided a systematic exposition of the new quantitative approach, which we have already seen taking shape in the work of Fourier and Gauss. Unlike Gauss, who almost hid the radical novelty of his approach in his Disquisitiones generales [1812b], Cauchy openly declared it in the introduction to the Cours d’analyse: As for methods, I have sought to give them all the rigor that one requires in geometry, so as never to have recourse to the reasons drawn from the generality of algebra. Reasons of this kind, although commonly admitted, particularly in the passage from convergent series to divergent series, and from real quantities to imaginary expressions, can, it seems to me, only sometimes be considered as inductions suitable for presenting the truth, but which are little suited to the exactitude so vaunted in the mathematical sciences. We must at the same time observe that they tend to attribute an indeﬁnite extension to algebraic formulas, whereas in reality the larger part of these formulas exist only under certain conditions and for certain values of the quantities that they contain. Determining these conditions and these values, and ﬁxing in a precise way the sense of the notations I use, I make any uncertainty vanish; and then the diﬀerent formulas involve nothing more than relations among real quantities, relations which are always easy to verify on substituting numbers for the quantities themselves. In order to remain faithful to these principles, I admit that I was forced to accept several propositions which seem slightly hard at ﬁrst sight. For example . . . a divergent series has no sum. (Cauchy [1821, ii–iii]) This excerpt clearly shows that Cauchy regarded the rejection of divergent series as a consequence of the rejection of the formal methodology and of the acceptance of an entirely quantitative conception. As I have explained at some length, the use of the divergent was unavoidable within the context of 18th-century methodology. 347

348

The Decline of the Formal Theory of Series

Though Cauchy’s inﬁnitesimal analysis was profoundly innovative, it was still ﬁrmly based on the notion of variable quantity. Variable quantities were the primary objects of analysis and functions were deﬁned as relations between variable quantities.472 According to Cauchy, a variable quantity is a quantity that can be considered capable of receiving several diﬀerent values successively. (Cauchy [1821, 19]) Cauchy’s use of this notion shows that variable quantities were regarded as entities capable of varying with continuity between ﬁxed limits. In Cauchy’s view, variable quantities could be represented as sequences of numbers that varied without jumps, although a variable quantity did not vary upon sets of numbers. The legacy of 18th-century analysis is particularly strong in this matter. Cauchy continued to consider numbers as the ratios of the variable quantity to a ﬁxed unity. The continuum was a primitive notion and was not constituted by numbers; on the contrary, numbers expressed speciﬁc determinations of quantity. Variable quantities varied on this primitive continuum and could potentially receive all the speciﬁc determinations one wished.473 Laugwitz has expressed this concept as follows: “Cauchy’s intervals are not sets but loci where a variable can move freely” (Laugwitz [1987, 273]).474 If Cauchy’s notion of a variable quantity is similar to the 18th-century one, a crucial diﬀerence emerges as concerns the way quantities were dealt with. When 18th-century mathematicians examined a relation between quantities, they made the values of these quantities abstract and only considered the way that they combined with each other (see, e.g., Lagrange [1806, 1]). In other words, they operated upon analytical expressions. Instead, “Cauchy’s proof and other concepts . . . did not rely on the concept of an analytic expression” (L¨ uzten [2003a, 157]). According to Cauchy, an equality A(x) = B(x) was not a formal relation that was valid when the variable quantity x was indeterminate; it was a quantitative relation that was valid only for the speciﬁc, determinate values of the variable that satisﬁed it. This did not mean that analytical expressions played no role in Cauchy’s analysis. As L¨ utzen observes, Cauchy always conceived of functions expressed by analytical expressions in his mind (see L¨ utzen [2003a, 157]). However, unlike 18th-century mathematicians, Cauchy thought that the 472 “If variable quantities are so joined between themselves that, the values of one of these being given, one can derive the values of all the others, one ordinarily conceives these diverse quantities expressed by means of the one among them, which then takes the name of independent variable; and the other quantities expressed by means of the independent variable are those which one calls functions of this variable” (Cauchy [1821, 31], translation in R¨ uthing [1984, 74]). 473 This concept is clearly similar to Gauss’s, which is illustrated in the previous chapter. 474 See also Breger [1992a, 251].

33

Cauchy’s Rejection of the 18th-Century Theory of Series

349

analytical expressions that were suitable for representing a function were constituted not only by the composition of elementary functions but also by integrals475 and series (see below). The above concept of the continuum is reﬂected in the notion of continuous functions. Laugwitz [1987, 273] stated: “[C]ontinuity (at least piecewise) of any function appearing in the calculus is a deep-rooted conviction with Cauchy” [1987, 273]. Indeed, Cauchy’s functions, like Euler’s and Lagrange’s functions, are intrinsically continuous and, at most, have isolated points of discontinuity. While Euler and Lagrange viewed these points as exceptional points that were ignored in the formulation of the theorems, Cauchy speciﬁed the limits of continuity and investigated functions only on the interval where they were continuous. He felt the need for an explicit deﬁnition of continuous functions: Let f (x) be a function of the variable x and suppose that for each value of x between two given limits this functions always takes a unique and ﬁnite values. If, having a value of x between these limits, one can attribute to the variable x an inﬁnitely small increase α, the function itself increases by the diﬀerence f (x + α) − f (x), which depends simultaneously on the new variable α and the value of x. This done, the function f (x) will be, between the two limits assigned to the variable x, a continuous function of this variable if, for each value of x intermediate between these limits the numerical value of diﬀerence f (x + α) − f (x) decreases indeﬁnitely with α. In other words, the function f(x) will remain continuous with respect to x between the given limits if, between these limits an inﬁnitely small increase in the variable always produces an inﬁnitely small increase in the function itself. (Cauchy [1821, 43], Cauchy’s emphasis, translation in Fauvel– Gray [1987, 566–567]) This deﬁnition merely served to give an explicit description of the property476 (LC) that 18th-century mathematicians assumed held for any function f (x) and that was used without referring to the speciﬁc intervals where the property actually held. Cauchy also deﬁned continuity in the neighborhood of a particular value of x: 475

For this reason, Cauchy considered π2 Euler’s notions of continuity (see p. 351). 476 See Chapter 18.3, p. 212.

∞ 0

x2 dt t2 +x2

as an adequate counterexample to

350

The Decline of the Formal Theory of Series One says furthermore that the function f (x) is, in the neighborhood of a particular value attributed to x, a continuous function of this variable, whenever it is continuous between two limits of x, however close, which contain that value of x. (Cauchy [1821, 43])

Two remarks should be made here. First, providing a deﬁnition of continuity in the neighborhood is diﬀerent from deﬁning continuity at a point. There is no room for pointwise continuity in Cauchy’s concept, which still belonged to the time when the continuous was merely the unbroken. Continuity was a property connected to variables (and intervals), rather than to single points. It implied a variation, without jumps, of the variable x which corresponded to a variation, without jumps, of the variable y = f (x). Second, Cauchy’s deﬁnition of continuity in the neighborhood of a point —or, to use Cauchy’s words, a particular value— x0 , means that there exists an interval (a, b) containing x0 , such that if x and x + α vary over (a, b) and α is inﬁnitesimal, then f (x + α) − f (x) is inﬁnitesimal. Setting z = x + α, Cauchy’s deﬁnition can be formulated: (Ca) f (x) is continuous in the neighborhood of a particular value x0 if an interval (a, b) containing x0 exists such that when the variables x and z vary over (a, b) and their diﬀerence x − z is inﬁnitesimal, then f (z) − f (x) is inﬁnitesimal. Cauchy assumed that a continuous function displayed uniform behavior, with respect to the variables x and z, in a whole interval containing x0 . There is a very strong temptation to interpret (Ca) as (UC) f (x) is continuous in the neighborhood of a point x0 if an interval (a, b) containing x0 exists such that if x ∈ (a, b), z = x + α ∈ (a, b), and x − z is inﬁnitesimal, then f (z) − f (x) is inﬁnitesimal. This is the uniform continuity in the interval (a, b). But this interpretation presupposes the notion of belonging, the concept of the interval (a, b) as (a set of) made of points, pointwise-deﬁned functions, etc. Even though Cauchy’s deﬁnition resembles uniform continuity, it cannot be interpreted as such unless notions are ascribed to Cauchy that he did not possess.477 Cauchy’s assumption of uniform behavior recalls the similar assumption to be found in Lagrange’s proofs of the remainder theorem.478 It seems to be more a legacy of the past (closely connected with the notion of variable quantity) than an anticipation of the future. Of course, when I underline the similarity between Cauchy’s notion of continuity and the 18th-century one, I do not refer to Euler’s deﬁnition of a 477 478

On this question, see the interesting analysis of Laugwitz in his [1987, 273]. See p. 307 and, for more details, Ferraro and Panza [A].

33

351

Cauchy’s Rejection of the 18th-Century Theory of Series

continuous function479 but to property (LC). Euler’s deﬁnition, which was closely connected with the generality of algebra,480 was explicitly rejected by Cauchy.481 In his [1844, 145], Cauchy based his rejection on the observation that a simple change of notation is often enough to transform a continuous function (in Euler’s sense) into a discontinuous function (in Euler’s sense). For instance, the function482 x for x ≥ 0 f (x) = −x for x ≤ 0 can be represented both as 2 π and as

∞

√

x2 dt t2 + x2 x2 .

∗ ∗

∗

Cauchy’s treatment of series was one of crucial novelties of his Cours d’analyse and R´ esum´e des le¸cons. He gave the following deﬁnition of the sum of a series ∞ n=0 un : Let

sn = u0 + u1 + u2 + . . . + un−1

(228)

the sum of the ﬁrst n terms, where n is an arbitrary integer. If the sums sn approach a certain limit s indeﬁnitely for increasing values of n, the series is said to be convergent, and the limit in question is called the sum of the series. (Cauchy [1821, 114], translation in L¨ uzten [2003a, 159]) n As an example, he considered the geometric series ∞ n=0 x and showed that it converges or diverges according to whether the absolute value of x is less than or greater than 1. Immediately afterwards, Cauchy provided 479

See Section 18.3. See Section 18.2, p. 209. 481 I would like to point out that, at the end of the 18th century, Arbogast had also discussed the meaning of the term “continuity”. His point of view diﬀered slightly from that of Euler. In particular, he had stated that the continuity of a curve can be broken in two ways: 1. If the analytical expression of the curve changes; 2. if the curve is composed of diﬀerent parts and these parts are not joined to each other (see Arbogast [1791, 9–11]). 482 See footnote no. 290. 480

352

The Decline of the Formal Theory of Series

certain necessary and suﬃcient conditions for convergence.483 The ﬁrst is trivial: 1. The series is convergent if and only if the sequence of the partial sums is convergent toward a ﬁxed limit s; Cauchy thought that this condition could be reformulated as 2a. the series is convergent toward s if and only if sn+k − s ≈ 0, for inﬁnitely large n and every k, or 2b. the series is convergent if and only if sn+k − sn+h ≈ 0, for inﬁnitely large n and every k and h. Moreover from 2a Cauchy deduced that 3. the series is convergent if and only if, for every ε, one has |sn+k − sn | = rn+k = |un + un+1 + un+2 + . . . + un+k−1 | < ε, for enough large n and every k.484 483

“[I]n order for series u0 , u1 , u2 , . . . , un , un+1 , . . .

to be convergent, it is necessary and it suﬃces that increasing values of n make the sum sn = u0 + u1 + u2 + . . . + un−1 converge indeﬁnitely toward a ﬁxed limit s; in other words, it is necessary and it suﬃces that, for inﬁnitely large values of the number n, the sums sn , sn+1 , sn+2 , . . . diﬀer from the limit s, and in consequence among each other, by inﬁnitely small quantities” [1821, 115]. 484 In his own words: “[T]he successive diﬀerences between the ﬁrst sum sn and the following ones are determined by the equations sn+1 − sn = un ,

sn+2 − sn = un + un+1 ,

sn+3 − sn = un + un+1 + un+2 , ....

Therefore, in order for the series [228] to be convergent, it is ﬁrstly necessary that the general term un decreases indeﬁnitely, when n increases; but this condition is not suﬃcient,

33

Cauchy’s Rejection of the 18th-Century Theory of Series

353

Cauchy considered 1, 2a, 2b, and 3 as trivial and gave no proof of them. To modern eyes, 1 and 2a are really trivial, whereas the suﬃciency of 2b and 3 is not. The diﬀerence between 2b and 3, on one side, and 2a, on the other, is that in the formulation 2a the existence of the number s satisfying the condition sn+k − s ≈ 0, for large n and every k, is assumed by the hypothesis. If such a number exists, it is trivial to state that it is the sum. Instead, in the formulations 2b and 3 the existence of this number is not assumed by the hypothesis. The condition |sn+k − sn | = rn+k = |un + un+1 + un+2 + . . . + un+k−1 | < ε, for every ε > 0, for every k, and for large n

and the equivalent one in terms of inﬁnite and inﬁnitesimal numbers sn+k − sn ≈ 0, for every k and h, and inﬁnitely large n do not imply that the sum s exists, but only that the series is bounded.485 Instead, according to Cauchy, 2a and 2b are trivially equivalent. The only possible explanation is that he presumed that the condition sn+k − sn ≈ 0 implies the existence of a value s such that sn+k − s ≈ 0. This implicit assumption is grounded on Cauchy’s concept of continuum and, more precisely, on the principle that a variable quantity, by passing from a value A to a value B, must receive all intermediate values between A and B. For instance, if sn > 0 and sn+k − sn ≈ 0, for every k and inﬁnitely large n, then the sequence sn is bounded and a variable quantity by passing from the value A = s1 to a value B > sn , for all n, must become equal to the value s that separates the values greater than sn from the other values.486 An emblematic theorem of Cauchy’s theory of series is the following one: Theorem 1. When the diﬀerent terms of the series [228] are functions of the same variable x, and continuous with respect to and it must also be true for increasing values of n the diﬀerent sums un + un+1 , un + un+1 + un+2 , ..., that is, the sums of the quantities un , un+1 , un+2 , . . . taken from the ﬁrst, in whatever number we wish, will always end up having numerical values [absolute values] that are constantly smaller than any assignable limit. Conversely, when these various condition are satisﬁed, the convergence of series is assured.” [1821, 115–116] 485 The only trivial deduction that follows from sn+k − sn ≈ 0 is that the series is not inﬁnite. 486 The situation is similar to the existence of maximum in Gauss’s proof of the Gauss’s criterion; see p. 335.

354

The Decline of the Formal Theory of Series this variable in the neighborhood of a particular value for which the series is convergent, them the sum s is also a continuous function in the neighborhood of this particular value. (Cauchy [1821, 120], translation in L¨ uzten [2003a, 169])

A theorem of such a kind was not conceivable in the 18th century for at least two reasons. First, in the 18th century, series theory concerned the expansion of functions into series by formal methods or, if a series was given, the problem was to ﬁnd the generating function of that series.487 Theorem 1 presupposes that (a) a series is an autonomous object, (b) it is suitable for providing a function, and (c) one is interested in knowing whether the function is continuous. Second, in the 18th century, functions were thought to be always continuous at least over an interval, and the generality of algebra made it possible to consider the whole function as if it were continuous. A theorem that provided conditions for continuity over an interval was useless. Cauchy’s proof runs as follows. Let s be the sum of the series un and let sn be the nth sum of that series. The sum s can be written in the form s = sn + rn . Consider the equality s(x + α) − s(x) = sn (x + α) − sn (x) + rn (x + α) − rn (x). If α is an inﬁnitesimal,then sn (x + α) − sn (x) is inﬁnitesimal, for all n. Moreover, if the series un is convergent in the neighborhood of x, then rn (x + α) and rn (x) are inﬁnitesimal488 for large n. This implies that s(x) is continuous. Cauchy’s theorem and his proof are ambiguous. Indeed, one can note that a. Cauchy uses inﬁnitesimal neighborhoods of x in a decisive way.489 Inﬁnitesimals are not thought as a mere fa¸con de parler, but they are conceived as numbers, though a theory of inﬁnitesimal numbers is lacking.490 b. Cauchy does not give separate deﬁnitions of the sum of function series and of the sum of numerical series. In the above-mentioned deﬁnition, 487

See Chapter 8, p. 120. As Laugwitz [1987, 264–265] notes, Cauchy assumes that if the series un is convergent in the neighborhood of x, then it is convergent for x + α, where α is an inﬁnitesimal number. 489 On this question, see Laugwitz [1987]. 490 Cauchy tried to provide a classiﬁcation of inﬁnitesimal quantities in [1823, 250–256]. It remains rather vague and any attempt to transform it into the modern theory of inﬁnitesimals implies the rejection of Cauchy’s notion of quantity. Even Laugwitz, who had made some attempts of such kind, was forced to admit that such attempts “should be taken with reluctance and reserve” [1987, 273]. 488

33

Cauchy’s Rejection of the 18th-Century Theory of Series

355

the terms un can denote both numbers and functions. Cauchy considered function series as a series of variable quantities that vary over an interval. In his view, convergence is not pointwise,491 but a matter that concerns a variable over a whole interval.492 The combination of a and b leads one to think that the pointwise convergence of functions (for every real number in an interval [a, b]) is not the same as the convergence of functions (between the limits a and b) in Cauchy’s sense, which involved convergence in inﬁnitesimal neighborhood of a value of x. Unfortunately, Cauchy did not clarify this point: It is probable that his notion of variable quantity and number prevented him from doing so.493 ∗ ∗

∗

Having rejected formal methodology, Cauchy could not accept the principle of inﬁnite extension494 and so, in his [1821], he, for the ﬁrst time in the history of series theory, gave an appropriate deﬁnition of operations between series and also proved many results that had previously been considered obvious on the basis of the analogy between ﬁnite and inﬁnite sums. For instance, in the 18th century, it seemed obvious that (a + b + c + d + etc.) · (A + B + C + D + etc.) was equal to aA+ (aB + Ab) + (aC + bB + cA) + (aD + bC + cB + dA) + etc. independently of the meaning of “etc.” in such expressions: The rule of the ordinary multiplication between two polynomials was extended to inﬁnite series without the diﬀerence between ﬁnite and inﬁnite series having been pointed out. Instead, Cauchy gave explicit theorems for the sum and multiplication of series: 491 Of course, this also follows from the fact that Cauchy’s functions are not deﬁned pointwise. 492 I emphasize that the condition 2a, when applied to function series, does not imply the consideration of thespeciﬁc value of the independent variable. It can be interpreted as follows: The series un is convergent toward s if and only if sn+k − s ≈ 0, for inﬁnitely large n and every k, independently of x. 493 On Abel’s interpretations of Cauchy’s theorem, see Pensivy [1987–88, 20–28]. 494 See p. 117.

356

The Decline of the Formal Theory of Series

(S) If the series

∞

n=0 un

and

∞

∞

n=0 vn

un = s

are convergent and ∞

and

vn = S,

n=0

n=0

then the series

∞

(un + vn )

n=0

is convergent and its sum is s + S (Cauchy [1821, 127 and 132]). ∞ ∞ (P) Given the series ∞ n=0 un and n=0 vn , let n=0 an , deﬁne as follows: a0 = u0 v0 , a1 = u0 v1 + u1 v0 , a2 = u0 v2 + u1 v1 + u2 v0 , ..., an = u0 vn + u1 vn−1 + . . . + un−1 vn + un v0 , ... ∞ If n=0 |un | and ∞ n=0 |vn | are convergent and ∞

un = s and

∞

vn = S,

n=0

n=0

then the series

∞

an

n=0

is convergent and its sum is sS (Cauchy [1821, 127–128 and 133]). Cauchy also provided a counterexample that shows how the theorem on the series was not valid if one only assumed convergence of series ∞product of ∞ u and n n=0 n=0 vn . He showed that ∞

1 (−1)n+1 √ n n=1

is convergent, whereas the product ∞ ∞ n+1 1 n+1 1 √ √ (−1) (−1) n n n=1 n=1 is divergent (Cauchy [1821, 134–135]). Cauchy’s new approach (as well as Gauss’s) placed great importance on convergence tests. According to Cauchy,

33

Cauchy’s Rejection of the 18th-Century Theory of Series

357

before summing any series, I had to examine when the series can be summed, or, in other terms, what are the conditions for their convergence. (Cauchy [1821a, v]) Some convergence tests were already used in the 18th century, but their use occasionally appears in few speciﬁc proofs to ascertain the conditions that enabled the use a series to approximate a quantity. They were not employed to guarantee the a priori existence of a sum (in eﬀect, this had no sense in the formal conception, according to which summing a series meant exhibiting the generating functions). Instead, in Cauchy’s view, convergence tests served to guarantee the existence of the sum in the quantitative sense.495 In [1821], Cauchy considered several other tests (apart from the Cauchy criterion discussed above). Indeed, he proved

n |un | < 1 • the root test (the series ∞ n=0 un is convergent if lim sup n and divergent if lim sup |un | > 1) (Cauchy [1821, 121–123 and 129]),

• the ratio test, already used by Gauss (Cauchy [1821, 123 and 129]), ∞ n • the condensation test (if un+1 > un > 0, then ∞ n=0 2 u2n n=0 un and are convergent or divergent together) (Cauchy [1821, 123–124]), un • the logarithmic test (if un > 0 and lim log < 1, then the series is log 1 n

un > 1) (Cauchy [1821, 125–127]), convergent, it is divergent if lim log log 1 n

• the alternating test (Cauchy [1821, 130]). Later, in his [1827], he also published the integral test.496 ∗ ∗

∗

The expansion of elementary functions into power series was the main problem in 18th-century series theory. This problem continued to be important for Cauchy, although it changed. In the 18th century, the problem of the expansion was a direct problem that was solved by applying the procedures P1–P4, while the problem of the sum was a converse problem that consisted of seeking the generating function.497 The problem of the sum now became the direct problem, which consisted of seeking the limit of the partial sums, while the problem of expansion was the inverse problem, which consisted of seeking a series whose sum is the given function. 495

See also p. 325. ∞ Given u(x), ∞if un = u(n) > 0, for n = 1, 2, . . ., then n=1 un is convergent [divergent] if and only if 1 u(x)dx is convergent [divergent]. This criterion was already known as a sum estimator. 497 See Chapter 8. 496

358

The Decline of the Formal Theory of Series

Cauchy’s treatment of the expansions of elementary functions is based on various theorems concerning power series. In particular, he proved that 1 n a power series ∞ n=0 an x is convergent if the values of x are between − A 1 and A (where

A = lim sup n |an |

is the radius of convergence) and that it diverges if x < − A1 and x > A1 [1821, 136]. Cauchy also showed that the expansion into power series is unique [1821, 144–145]. His proof is similar to the proof provided by Euler in [1740, 417–418] and [1748a, 1:230–231]. As an example of how Cauchy dealt with the expansion of elementary functions, I shall brieﬂy illustrate his proof of the binomial theorem (Cauchy [1821, 146–147]). Given the series ∞ µ n x , n

n=0

Cauchy n proved that it is convergent for |x| < 1. Then he considered ∞ µﬁrst a series of the variable µ for a ﬁxed x (|x| < 1). The sum n=0 n xµ as n of ∞ n=0 n x is a continuous function for any value of µ. Cauchy denoted this function by φ(µ) and observed that φ(µ + µ′ ) = φ(µ)φ(µ′ ).

(229)

He had already proved in Chapter 5 of his Cours d’analyse [1821, 100–103] that the solution of the functional equation (229) is φ(µ) = [φ(1)]µ . Since φ(1) = (1 + x), he obtained φ(µ) = (1 + x)µ ; therefore, (1 + x)µ =

∞ µ

n=0

n

xn

for |x| < 1. Cauchy’s rejection of the generality of algebra also implies his refusal to extending a formula from the real values to the complex values of variables. In contrast to 18th-century mathematicians, Cauchy considered functions of complex variables as a subject in its own right and oﬀered appropriate

33

Cauchy’s Rejection of the 18th-Century Theory of Series

359

demonstrations for theorems concerning functions of complex terms. For example, he provided a proof of the binomial theorem (1 + x)µ =

∞ µ

n

n=0

xn

in the case that x is a complex variable (|x| < 1) and µ is a real number (Cauchy [1821, 243–247]). To prove this theorem, he considered the functional equation ω(x + y) = ω(x)ω(y), √ where ω(x) = φ(x) + −1ζ(x) is continuous complex-valued function. In [1821, 222–226] Cauchy proved that the solution of this equation is ω(x) = Ax (cos bx +

√

−1 sin bx),

where A and b are real numbers and A > 0. He applied this result to the series ∞ √ µ n z (cos nθ + −1 sin nθ), (230) n n=0

and showed that the sum ω(µ) of (230) is ω(µ) = rµ (cos µθ +

√

−1 sin µθ)

(for − 1 < z < 1),

where r and t depend on z and θ. He set r cos t = 1 + z cos θ and r sin t = z sin θ and proved that rµ (cos µθ + Of consequence,

√

√ & 'µ −1 sin µθ) = 1 + z(cos θ + −1 sin θ) .

∞ √ √ & 'µ µ n z (cos nθ + −1 sin nθ) = 1 + z(cos θ + −1 sin θ) , n

n=0

for −1 < z < 1. While Cours d’analyse is devoted to algebraic analysis, the subject matter of R´esum´e des le¸cons is the diﬀerential and integral calculus. In this book, Cauchy dealt with the Taylor series. He had already investigated this series in his Sur le d´eveloppement des fonctions en s´eries et sur l’int´egration des ´equations diﬀ´erentielles ou aux diﬀ´erences partielles, where he had given a counterexample to the uniqueness of the expansion in the Taylor series, a cornerstone of Lagrange’s theory of analytical functions (at least, he thought so). Indeed, Cauchy considered a function h(x) such that its derivatives were

360

The Decline of the Formal Theory of Series 1

1

equal to 0 for x = 0 (for instance, e− x , e− x2 , . . .). The expansion of h(x + i) for x = 0 is 1 h(i) = h(0) + h′ (0)i + h′′ (0)i2 + . . . = 0. (231) 2 Cauchy observed that the function f (x) + h(x) has the same expansion as f (x) so that a single Taylor series can represent more than one function. For this reason, “one cannot substitute series for functions indistinctly” (Cauchy [1822, 278]). It is worth noting that Cauchy’s criticism assumed that (198) is a quantitative relation, valid for any value of x (whereas Lagrange thought that it held when x is indeterminate). If we assume that (198) is a formal relation, derived by means of one of the 18th-century methods of expansion, this criticism is diﬃcult to accept, not just for the reason that Lagrangian formulas could fail at some exceptional points, but also because Lagrange’s aim was to show that it was possible to obtain the function f ′ (x) from the function f (x) by means of formal procedures. Cauchy did not prove that this was impossible.498 For instance, as Panza noted in [1992, 724], if one considers 2 the function e−1/x , one can easily obtain e−(x+i)

−2

= e−[x

−2 −2x−3 i+3x−4 i2 +...]

−x−2

= e

−x−2

+ 2e

−2

= (e−x )(e+2x −x−2

x−3 i + e

−3 i

)(e−3x

−4 i2

)...

[2x−6 − 3x−4 ]i2 + . . . 2

This expansion provides the derivative of the function e−1/x when x is indeterminate, but it fails to provide a quantitative representation of the 2 function e−1/x in a neighborhood of x = 0. In his R´esum´e des le¸cons, Cauchy gave his interpretation of the Taylor theorem. While this theorem had previously only concerned the derivation n of the coeﬃcients of the power series ∞ a n=0 n (x − x0 ) , namely in proving that f (n) (x0 ) , an = n! Cauchy mainly focused his attention on proving that the series ∞ f (n) (x0 )

n=0

n!

(x − x0 )n

converged to f (x0 + x) for appropriate values of x. He [1823, 214–217] showed499 that h n−1 f (k) (x) (h − z)n−1 (n) hk + f (x + z)dz. f (x + h) = (232) k! (n − 1)! 0 k=0

498 Consequently, Cauchy’s counterexample is not actually a counterexample if it is referred to as Lagrange’s theory. 499 He used repeated integrations according to a method that dated back to Johann Bernoulli (see Chapter 3).

33

Cauchy’s Rejection of the 18th-Century Theory of Series

Since

h 0

361

hn (n) (h − z)n−1 (n) f (x + z)dz = f (x + θh), (n − 1)! n!

with 0 < θ < 1, he rewrote (232) in the form f (x + h) =

n−1 k=0

f (k) (x) k hn (n) h + f (x + θh). k! n!

Afterwards, Cauchy [1823, 220–221] observed that the Taylor series ∞ f (k) (x)

k!

k=0

hk

was convergent to f (x + h) if h (h − z)n−1 (n) hn (n) lim f (x + z)dz = lim f (x + θh) = 0. n→∞ 0 n→∞ n! (n − 1)!

(233)

Only when this condition is satisﬁed could one write f (x + h) =

∞ f (k) (x) k=0

k!

hk .

For instance, the functions sin x and cos x satisfy (233) and, therefore, they admit the expansions sin x =

∞ (−1)k 2k+1 x (2k + 1)! k=0

and cos x =

∞ (−1)k k=0

(2k)!

x2k .

In conclusion, I shall discuss how, in his R´esum´e, Cauchy dealt with the problem of integration of series. This method had been fundamental in the rise and development of the calculus but had been used de facto as a trivial consequence of the principle of inﬁnite extension. Cauchy attempted to provide a proof of it, and this, in itself, is an innovative step. He presented the following theorem: If x0 and X are ﬁnite numbers, un (x) are continuous functions over [x0 , X] and the series s(x) = ∞ n=0 un (x) is convergent for all values of x over [x0 , X], then X ∞ X s(x)dx = un (x)dx (234) x0

n=0 x0

362

The Decline of the Formal Theory of Series In the proof, Cauchy set rn (x) = s(x) −

∞

un (x).

n=0

By integrating, he obtained

X

s(x)dx = x0

n i=0

X

x0

ui (x)dx +

X

rn (x)dx.

x0

Cauchy stated that since

X

x0

rn (x)dx = rn (ξ)(X − x0 ),

where x0 < ξ < X, and the series ∞ n=0 un (x) was convergent, the sequence rn (ξ) became zero for n = ∞. Hence, he obtained (234) (see Cauchy [1823, 237–238]). This theorem is subject to the same diﬃculties as in other of Cauchy’s theorems. Here, Cauchy assumes that the way rn (x) tends toward 0, for n → 0, does not depend on the choice of a particular value of the variable x, but only on the variable itself. Once again he presupposes a uniformity in the behavior of functions and function sequences. A few years later, mathematicians considered this theorem, as well as Theorem 1 and other of Cauchy’s theorems, as mistaken and began the process that was to lead toward the concept of uniform convergence and pointwise convergence. However, the investigation of the development of this notion is beyond the scope of the present book.

References Aepinus, F.U.T. [1757] Demonstratio theorematis newtoniani de binomio ad potentiam indeﬁnitam elevando, Novi Commentarii Academiae Scientiarum Petropolitanae, 8 (1757), pp. 169–180. Agostini, A. [1941] Le serie sommate da Pietro Mengoli, Bollettino Unione Matematica Italiana, (2) 3 (1941), pp. 231–251. d’Alembert, J.B. [1747] Recherches sur la courbe que forme une corde tendu¨e mise en vibration, M´emoires des l’Acad´emie Royale de Berlin, 3 (1747), pp. 214–219. — [1768] R´eﬂexions sur les suites et sur les racines imaginaires in Opuscles math´ematiques, Paris: David Briasson, Jombert, t.V, 1768, pp. 171–215. — [1773] M´elange de litt´erature, d’histoire, et de philosophie, 2nd edition, Amsterdam: Z. Chatelain et ﬁls, 5 vols. Amp` ere, A.M. [1806] Recherches sur quelques point de la th´eorie des fonctions d´eriv´es qui conduisent a` une nouvelle d´emonstration de la s´erie de Taylor et `a l’expression ﬁnie des termes qu’on n´eglige lorsqu’on arrˆete ´ cette s´erie `a travers un terme quelconque, Journal de l’Ecole Polytechnique, 6 (1806), pp. 148–181. Arbogast, L.-F.-A. [1791] M´emoire sur la nature des fonctions arbitraires qui entrent dans les int´egrales des ´equations aux d´eriv´ees partielles, St. Petersburg: de l’imprimerie a l’Academie Imperiale des Sciences, 1791. — [1800] Du calcul des d´erivations, Strasbourg: Levrault, 1800. Archimedes [QA] Quadrature of parabola, in The Works of Archimedes, translated by T. L. Heath, Cambridge: University Press, 1897, pp. 233–252. Aristotle, Aristotle’s Categories and De Interpretatione, translated by L. Ackrill, Oxford: Clarendon Press, 1963. — Metaphysics, translated by W. D. Ross, in The Basic Works of Aristotle, Richard McKeon (ed.), New York: Random House, 1941. — Aristotle’s Physics, translated by Joe Sachs. Rutgers, NY: University Press, 1994. Barrow, I. [1670] Lectiones geometricae: in quibus (praesertim) generalia curvarum linearum symptomata declarantur, Londini: Typis Gulielmi Godbid, & prostant venales apud Johannem Dunmore, 1670. Bayes, T. [1763] A letter from the late Reverend Mr. Thomas Bayes, F.R.S. to John Canton, M.A. and F.R.S, Philosophical Transactions, 53 (1763), pp. 269–271. Bernoulli, Daniel [1728] Observationes de seriebus quae formantur ex additione vel subtractione quacunque terminorum se mutuo consequentium, 363

364

References

Commentarii Academiae Scientiarum Imperialis Petropolitanae, 3 (1728), published in 1732, pp. 85–100. — [1732–33] Theoremata de oscillationibus corporum ﬁlo ﬂexili connexorum et catenae verticaliter suspensae, Commentarii Academiae Scientiarum Petropolitanae, 6 (1732–1733), published in 1738, pp. 108–122. — [1734–35] Demonstrationes Theorematum suorum de Oscillationibus corporum ﬁlo ﬂexili connexorum et catenae verticaliter suspensae, Commentarii Academiae Scientiarum Petropolitanae, 7 (1734–1735), published in 1740, pp. 162–173. — [1753] R´eﬂexions et ´eclaircissemens sur les nouvelles vibrations des cordes expos´ees dans les m´emoires de l’Acad´emie de 1747 et 1748, M´emoires des l’Acad´emie Royale de Berlin, 9 (1753), pp. 147–172. — [1771] De summationibus serierum quarunduam incongrue veris earumque interpretatione atque usu, Novi Commentarii Academiae Scientiarum Imperialis Petropolitanae, 16 (1771), published in 1772, pp. 71–90 (Summarium 12–14). Bernoulli, Jacob [1689–1704] Positiones arithmeticae de seriebus inﬁnitis, earumque summa ﬁnita, J. Conradi a Mechel, Basileæ, 1689–1704. In Jacobi Bernoulli basileensis Opera, Genevae: Cramer et fratum Philibert, 1744, 2 vols., vol. 1: pp. 375–402 and 517–542, vol. 2: pp. 745–767, 849–867 and 955–975. — [P] Ars conjectandi. Opus posthuma. Accedit Tractatus de seriebus inﬁnitis et Epistola Gallice scripta de ludo pilae reticularis, Basilae: Thurnisiorum Fratrum, 1713. Bernoulli, Johann [1694] Additamentum eﬀectionis omnium Quadraturarum et Rectiﬁcationum Curvarum per seriem quandam generalissimam, Acta Eruditorum, 1694, pp. 437–441. — [1718] Remarques sur ce qu’on a donn´e jusqu’ici de solutions de problˆemes sur les isoperim´etres, in Opera omnia, Lausannae et Genevae: Marci-Michaelis Bousquet et Sociorum, 1742, vol. 2, pp. 235–269. 1 + Bernoulli, Nikolaus II [1738] Inquisitio in summa seriei 1+ 14 + 19 + 16 1 25 +etc., Commentarii Academiae Scientiarum Imperialis Petropolitanae, 10 (1738), published in 1747, pp. 19–21. Bolzano, B [1816] Der binomische Lehrsatz, und als Folgerung aus ihm der polynomische, und die Reihen, die zur Berechnung der Logarithmen und Exponentialgr¨ oßen dienen, genauer als bisher erwiesen, Prague: Enders 1816. Bombelli, R. [1572] L’Algebra parte maggiore dell’aritmetica divisa in tre libri. Nuovamente posta in luce. Bologna: Giovanni Rossi 1572. Reprinted in L’Algebra, Opera di Rafael Bombelli da Bologna, Prima edizione integrale, Milano: Feltrinelli 1966. Bos, H. [1974]. Diﬀerentials, higher order diﬀerentials and the derivative in the Leibnizian Calculus, Archive for History of Exact Sciences, 14, pp. 1–90.

References

365

— [1993] Lectures in the history of mathematics, Providence, RI: American Mathematical Society, 1993. — [2001] Redeﬁning Geometrical Exactness: Descartes’ Transformation of the Early Modern Concept of Construction, New York: Springer, 2001. Bossut, C and Lagrange, J.L. [1801–02] Rapport sur un m´emoire pr´esent´e `a la classe par le citoyen Callet, M´emoires de l’Institut national des sciences et arts. Sciences Math´ematiques et Physiques, 3 (An IX (1801–02)): pp. 1–11. Breger, H. [1992a] A restoration that failed: Paul Finsler’s theory of sets. In Gillies [1992, 249–264]. — [1992b] Les Continu chez Leibniz. In Salanskis–Sinaceur [1992, 75– 84]. Brezinski, C. [1980] History of Continued Fractions and Pad´e Approximants, New York: Springer–Verlag, 1980. Brisson, B. [1808] Sur l’int´egration des ´equations diﬀ´erentielles par´ tielles, JEP [Journal de l’Ecole Polytechnique, Paris.] 7, 191–261. Burchard, J. [1721] Epistola ad Virum Clarissimum Brook Taylor, Acta Eruditorum, (1721), pp. 195–228. Burn, R.P. [2001] Alphonse Antonio de Sarasa and Logarithms, Historia Mathematica, 28 (2001), pp. 1–17. Bussotti, P. [2000] Aritmetica e aritmetizzazione. La via indicata da Gauss e Kronecker, Epistemologia, 23 (2000), pp. 23–50. Cardano, G. [1545] Artis Magnae, sive de regulis algebraicis liber unus, Norimbergae: apud Iohannem Petreium, 1545. English translation by T. Richard Witmer, The great art or the rules of algebra, Cambridge: Cambridge University Press, 1968. Cataldi, Pietro Antonio [1613] Trattato del modo brevissimo di trovare la radice quadra delli numeri, et regole da approssimarsi di coninuo al vero nelle radici de’ numeri non quadrati, con le cause, & inuentioni loro, et anco il modo di pigliarne la radice cuba, Bologna: Bartolomeo Cochi, 1613 Cauchy, A.–L. [1821] Cours d’analyse de l’´ecole royale polytechnique. 1re partie. Analyse alg´ebrique, Paris: Debure fr`eres, 1821. In [ŒCa, (2) 3]. — [1822] Sur le d´eveloppement des fonctions en s´eries et sur l’int´egration des ´equations diﬀ´erentielles ou aux diﬀ´erences partielles, Bulletin de la Soci´et´e Philomatique, 1822, pp. 49–54. In [ŒCa, (2) 2: 276–282]. ´ royale polytechnique sur — [1823] R´esum´e des le¸cons donn´ees a ` l’Ecole le calcul inﬁnit´esimal. Tome Premier, Paris: Debure fr`eres, 1823. In [ŒCa, (2) 4: 9–261]. — [1827] Sur la convergence des s´eries. In [ŒCa, (2) 7: 267–279]. — [1844] M´emoire sur les fonctions continues, Comptes rendus de l’Acad´emie des Science, 18 (1844), pp. 116–131. In [ŒCa, (1), 8:145–160]. — [ŒCa] Œuvres compl`etes de Augustin Cauchy, ed. by Acad`emie des Sciences, Paris: Gauthiers–Villars, 1882–1974, 27 vols.

366

References

Chabert, J.–L. (Ed.) [1999], A History of Algorithms. From the Pebble to the Microchip, Berlin: Springer–Verlag, 1999. Clairaut, A.C. [1754] Sur l’orbite apparente du Soleil autour de la terre, en ayant ´egard aux perturbations produites par les actions de la lune et des plan`etes principales, Histoire de l’Academie des Science de Paris, 1754, pp. 521–564. Dedekind, R. [1872] Stetigkeit und irrationale Zahlen, Braunschweig: Vieweg, 1872. Descartes, R. [1637] Discours de la m´ethode, Leyden: J. Maire, 1637. Dhombres, J. [1988] Un texte d’Euler sur les fonctions continues et les fonction discontinues, v´eritable programme d’organisation de l’analyse au XVIII`eme si`ecle, Cahiers du S´eminaire d’Historie des Math´ematiques, 9 (1988), pp. 23–97. — [1995] L’innovation comme produit captif de la tradition. Entre Apollonius et Descartes, une th´eorie des courbers chez Gr´egoire de Saint– Vincent, in Panza-Roero [1995, 13–83]. Dutka, J. [1982] Wallis’s product, Brouncker’s continued fraction, and Leibniz’s series, Archive for History of Exact Sciences, 26 (1982), pp. 115– 126. — [1984–85] The early history of the hypergeometric function, Archive for History of Exact Sciences, 31 (1984–85), pp. 15–34. — [1995] On the early history of Bessel functions, Archive for History of Exact Sciences, 49 (1995), pp. 105–134. Engelsman, S.B. [1984] Families of Curves and Origins of Partial Diﬀerentiation, Amsterdam: North–Holland, 1984. Euclid [E], The Thirteen Books of Euclid’s Elements, Translated by Thomas L. Heath, 2nd ed., Cambridge: Cambridge University Press, 1926. Euler, L. [1730–31a] De progressionibus transcendentibus seu quarum termini generales algebraice dari nequeunt, Commentarii Academiae Scientiarum Imperialis Petropolitanae, 5 (1730–31), published in 1738, pp. 36–57. In [EOO, (1) 14:1–24]. — [1730–31b] De summatione innumerabilium progressionum, Commentarii Academiae Scientiarum Imperialis Petropolitanae, 5 (1730–1731), published in 1738, pp. 91–105. In [EOO, (1) 14:25–41]. — [1732–33] Methodus generalis summandi progressiones, Commentarii Academiae Scientiarum Petropolitanae, 6 (1732–1733), published in 1738, pp. 68–97. In [EOO, (1) 14:42–72]. — [1734–35a] De summis serierum reciprocarum, Commentarii Academiae Scientiarum Petropolitanae, 7 (1734–1735), published in 1740, pp. 123–134. In [EOO, (1) 14:73–86]. — [1734–1735b] De progressionibus harmonicis observationes, Commentarii Academiae Scientiarum Petropolitanae, 7 (1734–1735), published in 1740, pp. 150–156. In [EOO, (1) 14:87–100].

References

367

— [1736a] Methodus universalis serierum convergentium summas quam proxime inveniendi, Commentarii Academiae Scientiarum Petropolitanae, 8 (1736), published in 1741, pp. 3–9. In [EOO, (1) 14:101–107]. — [1736b] Inventio summae cuiusque seriei ex dato termino generali, Commentarii Academiae Scientiarum Petropolitanae, 8 (1736), published in 1741, pp. 9–22. In [EOO, (1) 14:108–123]. — [1736c] Methodus universalis series summandi ulterius promota, Commentarii Academiae Scientiarum Petropolitanae, 8 (1736), published in 1741, pp. 147–158. In [EOO, (1) 14:124–137]. — [1737a] De fractionibus continuis dissertatio, Commentarii Academiae Scientiarum Petropolitanae, 9 (1737), published in 1744, pp. 98–137. In [EOO, (1), 14:187–215]. — [1737b] Variae observationes circa series inﬁnitas, Commentarii Academiae Scientiarum Petropolitanae, 9 (1737), published in 1744, pp. 160–188. In [EOO, (1) 14:216–244]. — [1737c] De variis modis circuli quadraturam numeris proxime exprimendi, Commentarii Academiae Scientiarum Petropolitanae, 9 (1737), published in 1744, pp. 222–236. In [EOO, (1) 14:245–259]. — [1739a] De productis ex inﬁnitis factoribus ortis, Commentarii Academiae Scientiarum Petropolitanae, 11 (1739), published in 1750, pp. 3–31. In [EOO, (1) 14:260 – 290]. — [1739b] De fractionibus continuis observationes, Commentarii Academiae Scientiarum Petropolitanae, 11 (1739), published in 1750, pp. 32–81. In [EOO, (1) 14:291–349]. — [1739c] Consideratio progressionis cuiusdam ad circuli quadraturam inveniendam idoneae, Commentarii Academiae Scientiarum Petropolitanae, 11 (1739), published in 1750, pp. 116–127. In [EOO, (1) 14:350–365]. — [1740] De seriebus quibusdam considerationes, Commentarii Academiae Scientiarum Imperialis Petropolitanae, 12 (1740), published in 1750, pp. 53–96. In [EOO, (1) 14:407–462]. — [1743a] De summis serierum reciprocarum ex potestatibus numerorum naturalium ortarum dissertatio altera in qua eaedem summationes ex fonte maxime diverso derivantur, Miscellanea Berolinensia, 7 (1743), published in 1743, pp. 172–192. In [EOO, (1) 14:138–159]. 1 + — [1743b] D´emonstration de la somme de cette suite 1 + 41 + 19 + 16 1 1 + + . . ., Journ. lit. d’Allemange, de Suisse et du Nord, 2:1 (1743), 25 36 published in 1743, pp. 145–127. In [EOO, (1) 14:176–186]. — [1748a] Introductio in analysin inﬁnitorum, Lausannae: M. M. Bousquet et Soc., 1748. In [EOO, (1), vols. 8–9]. — [1748b] Recherches sur la question des in´egalit´es du mouvement de Saturne de Jupiter, Piece qui a remporte le prix de l’academie royale des sciences (1748), published in 1751, pp. 1–123. In [EOO, (2), 25: 45–157]. — [1749a] De vibratione chordarum exercitatio, Nova Acta Eruditorum (1749), published in 1749, pp. 512–527. In Euler [EOO, (2), 10:50–62].

368

References

— [1749b] De la controverse entre Mrs. Leibnitz et Bernoulli sur les logarithmes des nombres negatifs et imaginaires, M´emoires des l’Acad´emie des Sciences de Berlin, 5 (1749), published in 1749, pp. 139–179. In [EOO, (1), 17:195–232]. — [1750–51a] Methodus aequationes diﬀerentiales altiorum graduum integrandi ulterius promota, Novi Commentarii Academiae Scientiarum Imperialis Petropolitanae, 3 (1750–51), published in 1753, pp. 3–35. In [EOO, (1) 22:181–213]. — [1750–51b] De serierum determinatione seu nova methodus inveniendi terminos generales serierum, Novi Commentarii Academiae Scientiarum Imperialis Petropolitanae, 3 (1750–51), published in 1753, pp. 8–10 and 36–85. In [EOO, (1), 14:463–515]. — [1750–51c] Consideratio quarundam serierum quae singularibus proprietatibus sunt praeditae, Novi Commentarii Academiae Scientiarum Imperialis Petropolitanae, 3 (1750–1751), published in 1753, pp. 10–12 and 86–108. In [EOO, (1) 14:516–541]. — [1750–51d] De partitione numerorum, Novi Commentarii Academiae Scientiarum Petropolitanae, 3 (1750–1751), published in 1753, pp. 125–169. In [EOO, (1), 2:254–294]. — [1753] Remarques sur les m´emoires pr´ec´edens de M. Bernoulli, M´emoires des l’Acad´emie Royale des Sciences de Berlin, 9 (1753), pp. 196–222. In [EOO, (2), 10:233–255]. — [1754–55a] Subsidium calculi sinuum, Novi Commentarii Academiae Scientiarum Petropolitanae, 5 (1760), published in 1760, pp. 164–204. In [EOO, (1) 14:542–582]. — [1754–55b] De seriebus divergentibus, Novi Commentarii Academiae Scientiarum Imperialis Petropolitanae 5 (1754–1755), published in 1760, pp. 19–23 and 205–237. In [EOO, (1) 14: 583–617]. — [1755] Institutiones calculi diﬀerentialis cum eius usu in analysi ﬁnitorum ac doctrina serierum, Berolini: 1755. In [EOO, (1), 10]. — [1761] Remarques sur un beau rapport entre les s´eries des puissances tant directes que r´eciproques, M´emoires de l’Acad´emie des Sciences de Berlin, 17 (1761), published in 1766, pp. 83–106. In Euler [EOO, (1) 14:83–106] — [1764] De motu vibratorio tympanorum, Novi Commentarii Academiae Scientiarum Petropolitanae, 10 (1764), published in 1766, pp. 243– 260. In [EOO, (2) 10:344–359]. — [1765] De usu functionum discontinarum in Analysi, Novi Commentarii Academiae Scientiarum Petropolitanae, 11 (1765), published in 1767, pp. 3–27. In [EOO, (1), 23:74–91]. — [1768] De curva hypergeometrica hac aequatione expressa y = 1 · 2 · 3 · . . . · x, Novi Commentarii Academiae Scientiarum Petropolitanae 13 (1768), published in 1769, pp. 3–66. In [EOO, (1) 28:41–98].

References

369

— [1768–70] Institutionum calculi integralis, Petropoli: Impensis Academiae Imperialis Scientiarum, 1768–1770. In [EOO, (1), vols. 11–13]. — [1770] Vollst¨ andige Anleitung zur Algebra. In [EOO, (1), vol. 1]. — [1773] Summatio progressionum sin ϕλ + sin 2ϕλ + sin 3ϕλ + . . . + sin nϕλ , cos ϕλ + cos 2ϕλ + cos 3ϕλ + . . . + cos nϕλ , Novi Commentarii Academiae Scientiarum Petropolitanae, 18 (1773), published in 1774, pp. 24–36. In [EOO, (1), 15:168–190]. — [1774–75] Demonstratio theorematis newtoniani de evolutione potestatum binomii pro casibus quibus exponentes non sunt numeri integri, Novi Commentarii Academiae Scientiarum Petropolitanae, 19 (1774–75), published in 1775, pp. 103–111. In [EOO, (1), 15:207–216]. — [1780] De plurimis quantitatibus transcendentibus quas nullo modo per formulas integrales exprimere licet, Acta Academiae Scientiarum Petropolitanae, 4 pt. 2 (1780), published in 1780, pp. 31–37. In [EOO, (1) 15: 522–527]. — [1781a] De oscillationibus minimis funis libere suspensi, Acta Academiae Scientiarum Petropolitanae (1781), pt. 1, pp. 157–177. In [EOO, (2) 11:307–323]. — [1781b] De perturbatione motus chordarum ab earum pondere oriunda, Acta Academiae Scientiarum Petropolitanae (1781), pt. 1, pp. 178– 190. In [EOO, (2) 11:324–334]. — [1784] De transformatione seriei divergentis 1 − mx + m(m + n)x2 − m(m + n)(m + 2n)x3 + m(m + n)(m + 2n)(m + 3n)x4 −etc. in fractionem continuam Nova Acta Academiae Scientiarum Petropolitanae, 2 (1784), pp. 36–45. In Euler [EOO, (1) 16:34–46). — [1787] Dilucidationes in capita postrema calculi mei diﬀerentalis de functionibus inexplicabilibus, in L. Euler, Institutiones calculi diﬀerentialis cum eius usu in analysi ﬁnitorum ac doctrina serierum, 2nd edition, Pavia: Galeati, 1787, pp. 797–819. In [EOO, (1) 16 pt. 1:1–33]. — [1789] De termino generali serierum hypergeometricarum, Nova Acta Academiae Scientiarum Pretopolitanae, 7 (1789), pp. 42–63. In [EOO, (1) 16 pt. 1:139–162]. — [1793] Disquisitio ulterior super seriebus secundum multipla cuiusdem anguli progendientibus, Nova Acta Academiae Scientiarum Petropolitanae, 11 (1793), published in 1798, pp. 114–132. In [EOO, (1) 16 pt. 1:333–355]. — [1794a] Specimen transformationis singularis serierum Nova Acta Academiae Scientiarum Pretopolitanae, 12 (1794), pp.58–70. In [EOO, (1), 16 pt. 2:41–55]. — [1794b] De resolutione formulae integralis xm−1 dx(Δ + xn )λ in seriem semper convergentem, ubi simul plura insignia artiﬁcia circa serierum explicantur, Nova Acta Academiae Scientarum Imperialis Petropolitanae 12 (1801), published in 1801, pp. 58-70. In [EOO, (1), 19:110–143].

370

References

— [1812] Methodus succincta summas serierum inﬁnitarum per formulas diﬀerentiales investigandi, M´emoires de l’acad´emie des Sciences de St.P´etersboug, 5 (1812), pp. 45–56. In [EOO, (1) 16 part 2:200–221]. — [1819] De unciis potestatum binomii earumque interpolatione; M´emoires de l’acad´emie des Sciences de St.P´etersboug, 9 (1819–20), pp. 57–76. In [EOO, 162:241–260]. — [EOO] Leonhardi Euleri Opera omnia. Soc. Sci. Nat. Helveticæ, Leipzig, Berlin, Basel, 1911–. . . Ewald, W.B. (ed.) [1996] From Kant to Hilbert. A Source Book in the Foundation of Mathematics, Oxford Science Publications, Oxford: Claredon Press, 1996. Feigenbaum, L. [1985] Brook Taylor and the method of increments, Archive for History of Exact Sciences, 34 (1985), 1–140. Ferraro, G. [1998] Some aspects of Euler’s series theory. Inexplicable functions and the Euler–Maclaurin summation formula, Historia Mathematica, 25 (1998), pp. 290–317. — [1999] Rigore e dimostrazione in Matematica alla met`a del Settecento, Physis, (2) 36 (1999), pp. 137–163. — [2000a] Functions, functional relations and the laws of continuity in Euler, Historia Mathematica, 27 (2000), pp. 107–132. — [2000b] The value of an inﬁnite sum. Some observations on the Eulerian theory of series, Sciences et techniques en perspective, 4 (2000), pp. 73–113. — [2000c] True and ﬁctitious quantities in Leibniz’s theory of series, Studia Leibnitiana, 32 (2000), pp. 43–67. — [2002] Convergence and formal manipulation of series from the origins of calculus to about 1730, Annals of Science, 59 (2002), pp. 179–199. — [2007a] Convergence and formal manipulation in the theory of series from 1730 to 1815, Historia Mathematica, 34 (2007), pp. 62–88. — [2007b] The foundational aspects of Gauss’s work on the hypergeometric, factorial and digamma functions, Archive for History of Exact Sciences, 61 (2007), pp. 457-518. — [2007c] L’evoluzione della matematica. Alcuni momenti critici. Napoli. Ernesto Ummarino Editore, 2007. — [2007d] La teoria formale delle serie in Euler: gli anni dal 1730 al 1755 in Ferraro [2007c, 183-208]. Ferraro, G. and Panza, M. [2003] Developing into series and returning from series. A note on the foundation 18th-century analysis, Historia Mathematica, 30 (2003), pp. 17–46. — [A] Lagrange’s theory of analytical functions and his ideal of purity of method (1797–1813), unpublished manuscript. Fraser, C. [1985] J.L. Lagrange’s changing approach to the foundations of the calculus of variations, Archive for History of Exact Sciences, 32 (1985), pp. 151–191.

References

371

— [1987] Joseph Louis Lagrange’s algebraic vision of the calculus, Historia Mathematica, 14 (1987), pp. 38–53. — [1989] The calculus as algebraic analysis: Some observations on mathematical analysis in the 18th century, Archive for History of Exact Sciences, 39 (1989), pp. 317–335. — [1997] The background to and early emergence of Euler’s analysis, in Otte and Panza [1997, 47–78]. — [2003] “Mathematics” in The Cambridge History of Science Volume 4 The Eighteenth Century; Cambridge: Cambridge University Press (2003). Friedman, M.C [1992], Kant and the Exact Science, Cambridge (Mass.): Harvard University Press, 1992. Fourier, J. [1807] Sur la propagation de la chaleur dans le solides, manuscript 1807, published by Grattan-Guiness and Ravetz, in GrattanGuiness and Ravetz [1972]. — [1822] Th`eorie Analytique de la Chaleur, Paris: F. Didot, p`ere et ﬁls, 1822. — [Œuvres] Œuvres de Fourier, edited by G. Darboux, Paris: GauthierVillars et ﬁls, 1888–1890. Fran¸ cais, J.–F. [1812–13] M´emoire rendant a` d´emontrer la l´egitimit´e de la s`eparation des ´echelles de diﬀ´erentiation et d’integration des fonctions qu’elles aﬀectent, Annales de Math´ematiques Pures et Appliqu´ees, 3 (1812– 13), pp. 244–273. Fuss, P.H. [1843] Correspondance Math´ematique et Physique de Quelque C´el`ebres G´eom`etres du XVIII`eme Si`ecle, St. P´etersbourg: Acad´emie imp´eriale des sciences, 1843. Gauss, C.F. [1799] Demonstratio nova theorematis omnem functionem algebraicam rationalem integram unius variabilis in factores reales primi vel secundi gradus resolvi posse, in [WW, 3:1–39]. — [1801] Disquisitiones Arithmeticae, Lipsiae: Gerh. Fleischer, 1801. In [WW, 1:1–475]. — [1812a] Anzeige der Disquisitiones generales circa seriem inﬁnitam G¨ ottingische gelehrte Anzeigen, 10 February 1812. In [WW, 3:197–202]. αβ x+ — [1812b] Disquisitiones generales circa seriem inﬁnitam 1 + 1·γ α(α+1)β(β+1) α(α+1)(α+2)β(β+1)(β+2) 3 x +etc. 1·2·γ(γ+1) xx+ 1·2·3·γ(γ+1)(γ+2)

Pars Prior, Commentationes Societatis Regiae Scientiarum Gottingensis recentiores, 2 (1813). In [WW, 3:125–162]. — [1816] Review of J.C. Schwab, Commentatio in primum elementorum Euclidis librum, qua veritatem geometriae principiis ontologicis niti evincitur, omnesque propositiones, axiomatum geometricorum loco habitae, demonstrantur (Stuttgart, 1814) and of Matthias Metternich, Vollst¨ andige Theorie der Parallel-Linien (Mainz, 1815). G¨ ottingische gelehrte Anzeigen, 20 April 1816. In [WW, 8:170–174]. — [1831] Anzeige der Theoria residuorum biquadraticorum. Commen-

372

References

tatio secunda, G¨ ottingische gelehrte Anzeigen, 23 April 1831. In [WW, 2:169–178]. — [1832] Theoria residuorum biquadraticorum. Commentatio secunda, Commentationes Societatis Regiae Scientiarum Gottingensis recentiores, 7 (1832). In [WW, 2:93–148]. — [WO] De integratione formulae diﬀerentialis (1 + n cos ψ)ν dψ. In [WW, 8:35–64]. — [WA] Determinatio seriei nostrae per aequationem diﬀerentialem secondi ordinis. In [WW, 3: 207–229]. — [WB] Zur Theorie der unendlichen Reihe F (α, β, γ, x). In [WW, 10 pt 1: 326–359]. — [WC] Zur Lehre von den Reihen. In [WW, 10 pt 1: 382–428]. — [WD] Zur Metaphysik der Mathematik. In [WW, 12:57–61]. — [WW] Werke, G¨ ottingen, Leipzig, and Berlin: K¨ onigliche Gesellschaft der Wissenchaften, 1863–1929, 12 vols. Giusti, E. [1991] Le prime ricerche di Pietro Mengoli: la somma delle serie, Proceedings of the international meeting “Geometry and Complex Variables”, New York: Dekker, 1991, pp. 195–213. Gilain, Ch. [1998] Condorcet et le calcul integral, in Sciences a ´ l’Epoque de la R´evolution Fran¸caise. Recherches Historiques, R. Rashed (ed.), Paris: Blanchard, 1988. Gillies, D. (eds.) [1992] Revolutions in Mathematics, Oxford: Claredon Press, 1992. Goldbach, C. [1720] Specimen methodi ad summas serierum, Acta eruditorum (1720), pp. 27–31. — [1727] De transformatione serierum, Commentarii Academiae Scientiarum Imperialis Petropolitanae, 2 (1727), published in 1729, pp. 30–34. — [1728] De terminis generalibus serierum, Commentarii Academiae Scientiarum Imperialis Petropolitanae, 3 (1728), published in 1732, pp. 164– 173. Goldstine, H. [1977] A History of Numerical Analysis from the 16th Century Through the 19th Century, New York: Springer-Verlag, 1977. Golland, L.A. and Golland, R.W. [1993] Euler’s troublesome series: An early example of the use of trigonometric series, Historia Mathematica, 20 (1993), pp. 54–67. Grabiner, J.V. [1981] The Origins of Cauchy Rigorous Calculus, Cambridge: MIT Press, 1981. Grandi, G. [1703] Quadratura circuli et hyperbolae, Pisis: ex typographia F. Bindi, 1703. Grattan–Guinness, I. [1990] Convolutions in French Mathematics, 1800–1840, Basel: Birkh¯ auser Verlag,1990. — [1996] Numbers, magnitudes, ratios and proportions in Euclid’s Elements: How did he handle them?, Historia Mathematica, 23 (1996), pp. 355–375.

References

373

Grattan–Guinness, I.and Ravetz, J.R. [1972] Joseph Fourier 1768– 1830, Cambridge: MIT Press, 1972. Gray, J. [1984] A commentary on Gauss’s mathematical diary, 1796– 1814, with an English translation, Expositiones Mathematicae, 2 (1984), pp. 97–130. — [1986] Linear Diﬀerential Equations and Group Theory from Reimann to Poincar´e, Boston: Birkh¨ auser, 1986. (2nd edition Boston: Birkh¨ auser, 2000). — [1992] The nineteenth-century revolution in mathematical ontology. In Gillies [1992, 226–248]. Gr´ egoire de Saint-Vincent [1647] Opus qeometricum quadraturae circuli et sectionum coni decem libris comprehensum, Antwerp: J. and J. Meursios, 1647. Gregory, J. [1667] Vera circuli et hyperbola quadratura, Padova: Cadorini, 1667. — [1668] Exercitationes Geometricae, London: Godbid and Pitt, 1668. Guicciardini, N. [1989] The Development of Newtonian Calculus in Britain 1700–1800, Cambridge: Cambridge University Press, 1989. — [1999] Reading the Principia. The Debate on Newton’s Mathematical Methods for Natural Philosophy from 1687 to 1736, Cambridge: Cambridge University Press, 1999. Halley, E. [1695] A most compendius and facile method for constructing the logarithms, exempliﬁed and demonstrated from the nature of numbers, without any regard to the hyperbola, with a speedy method for ﬁnding the number from the logarithm given, Philosophical Transactions, 19 (1695– 97), pp. 58–67. Hardy, G.H. [1949] Divergent Series, Oxford: Clarendon Press, 1949. Hayashi, T. [2002] Leibniz’s construction of Mathesis Universalis: A consideration of the relationship between the plan and his mathematical contributions, Historia Scientiarum, 12 (2002), pp. 121–141. Horv` ath, M. [1982] The problem of inﬁnitesimal small quantities in the Leibnizian mathematics, Studia Leibnitiana, Supplementa 22 (1982), pp. 150–157. — [1986] On the attempts made by Leibniz to justify his calculus, Studia Leibnitiana, 18 (1986), pp. 60–71. Huygens, C. [P] Descriptio automati planetarii in Oevres Compl`etes de Christiaan Huygens, La Haye: M. Nijhoﬀ, 1888–1950, vol.21, pp. 579– 647. Jahnke, H.N. [1987] Motive und Probleme der Arithmetisierung der Mathematik in der ersten H¨ alfte des 19. Jahrhunderts – Cauchys Analysis in der Sicht des Mathematikers Martin Ohm, Archive for History of Exact Sciences, 37 (1987), pp. 101–182. — [1993] Algebraic analysis in Germany, 1780–1840: Some mathematical and philosophical issues, Historia Mathematica, 20 (1993), pp. 265–284.

374

References

Jahnke, H.N. (ed.) [2003] A History of Analysis. Providence: American Mathematical Society, 2003. Katz, V.J. [1987] The Calculus of the Trigonometric Functions, Historia Mathematica 14 (1987), pp. 311–324. Klein, J. [1968] Greek Mathematical Thought and the Origin of Algebra, translation by E. Brann, Cambridge: MIT Press, 1968. Kline, M. [1972] Mathematical Thought from Ancient to Modern Times, New York: Oxford University Press, 1972. Knobloch, E. [1991] Leibniz and Euler: Problems and solutions concerning inﬁnitesimal geometry and calculus, in: Giornate di Storia della Matematica, M.Galuzzi (ed.), Commenda di Rende: Editel, 1991, pp. 271– 293. — [1989] Leibniz et son manuscrit in´edit´e sur la quadrature des sections coniques, in The Leibniz Renaissance, M. Mugnai (ed.), Firenze: 1989, pp. 127–151. — [1999] Galileo and Leibniz: Diﬀerent approaches to inﬁnity, Archive for History of Exact Sciences, 54 (1999), pp. 87–99. Koppelman, E. [1971] The calculus of operations and the rise of abstract algebra, Archive for History of Exact Sciences, 8 (1971), pp. 155–241. Kramp, C. [1799] Analyse des r´efractions astronomiques et terrestres. Strasbourg: Philippe Jacques Dannbach, 1799. Lacroix, S.F. [1797–1800] Trait´e du calcul diﬀ´erentiel et du calcul int´egral, Paris: Duprat, 1797–1800 (3 vols). — [1810–19] Trait´e du calcul diﬀ´erentiel et du calcul int´egral, Second edition, revue et augmentee. Paris: Courcier, 1810–1819, 3 vols. Lagrange, J.–L. [1754] Lettera a Giulio Carlo di Fagnano. Torino: Stamperia Reale, 1754. — [1759] Recherches sur la nature, et la propagation du son, Miscellanea Taurinensia, 1, classe math´ematique, pp. i–x and 1–112. In [Œuvr, 1, 39–148]. — [1760–61] Addition aux premi`eres recherches sur la nature, et la propagation du son, Miscellanea Taurinensia, 2 (1760–61), classe math´ematique, pp. i–x and 1–112. In [Œuvr, 1:319–333]. — [1766] Sur l’int´egration de quelques ´equations diﬀ´erentielles dont les ind´etermin´ees sont s´epar´ees, mais dont chaque membre en particulier n’est point int´egrable, Miscellanea Taurinensia, 4 (1766). In [Œuvr 9, 5–33]. — [1768] Nouvelle m´ethode pour r´esoudre le ´equations litt´erales par le moyen des s´eries, Histoire de l’Acad´emie Royale des Sciences et des BellesLettres de Berlin (1768), pp. 251–326. In [Œuvr, 3: 5–76]. — [1769] Sur le probl`eme de Kepler, Histoire de l’Acad´emie Royale des Sciences et des Belles-Lettres de Berlin (1769), pp. 204–233. In [Œuvr, 3:113–140]. — [1772] Sur une nouvelle esp`ece de calcul relatif `a la diﬀ´erentiation et `a l’int´egration des quantit`es variable, Nouveaux M´emoires de l’Acad´emie

References

375

Royale des Sciences et de Belles-Lettres de Berlin (1772), pp. 185–221. In [Œuvr, 3: 441–476]. — [1773] Solutions analytiques de quelques probl`emes sur les pyramides triangulaires, Nouveaux M´emoires de l’Acad´emie Royale des Sciences et de Belles–Lettres de Berlin (1773), pp. 149–176. In [Œuvr, 3: 661–692]. — [1776] Sur l’usage des fractions continues dans le calcul integrals, Nouveaux M´emoires de l’Acad´emie Royale des Sciences et de Belles–Lettres de Berlin, 1776. In [Œuvr, 4:301–332]. — [1788] Trait´e de m´ecanique analytique, Paris: Desaint 1788. In [Œuvr, 12]. — [1792] M´emoire sur le m´ethode d’interpolation, Nouveaux M´emoires de l’Acad´emie Royale des Sciences et de Belles–Lettres de Berlin, 40 (1792– 1793), pp. 267–288. In [Œuvr, 5:663–684]. — [1797] Th´eorie des fonctions analytiques, Paris: Impr. De la R´epublique, 2nd. ed. Paris: Courcier 1813. In [Œuvr, 9]. — [1798] Trait`e de la r´esolution des ´equations num´eriques de tous les degr´es, Paris: Duprat 1798. 2nd ed., Paris: Courcier 1806. In [Œuvr, 8]. — [1806] Le¸con sur le calcul des fonctions, 2nd ed., Paris: Courcier, 1806. In [Œuvres, 10]. — [Œuvr] Œuvres de Lagrange, M.J.-A. Serret [et G. Darboux] (ed.), Paris: Gauthier–Villars, 1867–1892, 14 vols. Lambert, J.H. [1770] Observations analytiques, Nouveaux M´emoires de l’Acad´emie Royale des Sciences et de Belles-Lettres de Berlin (1770), pp. 225–244. Landen, J. [1758] A Discourse Concerning the Residual Analysis: A New Branch of Algebraic Art, of Very Extensive Use Both in Pure Mathematicks and Natural Philosophy, London: 1758. Laplace, P.S. [1773] M´emoire sur l’inclinaison moyenne des orbites des com`etes, sur la ﬁgure de la Terre et sur les fonctions, M´emoires de l’Acad´emie royale des sciences de Paris (Savants ´etrangers), 7 (1773), pp. 503–540. In [ŒC, 8:279–321]. — [1777] M´emoire sur l’usage du calcul aux diﬀ´erences partielles dans la th´eorie des suites, M´emoires de l’Acad´emie royale des sciences de Paris (1777). In [ŒC, 9:313–335]. — [1779] M´emoire sur les suites, M´emoires de l’Acad´emie royale des sciences de Paris. M´emoires math´ematiques et physique ( 1779–1782), pp. 207–309. In [ŒC, 10:1–89]. — [1784] Th´eorie du mouvement et de la ﬁgure elliptique des plan`etes, Paris: Ph.–D. Pierres, 1784. — [1785] Th´eorie des attractions des sph´ero¨ıdes et de la ﬁgure des plan`etes, M´emoires de l’Academie Royal des Sciences de Paris. M´emoires math´ematiques et physique ( 1782–1785), pp. 207–309. In [ŒC, 10:339–419]. — [1799–1825] Trait´e de m`ecanique c´eleste, Paris: Courcier 1799–1825. In [ŒC, vols. 1–5]

376

References

´ — [1809] M´emoire sur diverses points d’analyse, Journal de l’Ecole Polytechnique, XV Cahier, Tome VIII; 1809. In [ŒC, 14:178–214]. — [1812] Th´eorie analytique des probabilit´es, Paris: Courcier, 1812. — [ŒC] Œuvres Compl`ete de Laplace, Paris: Gauthier–Villars, 1867– 1892, 14 vols. Laugwitz, D. [1987] Inﬁnitely small quantities in Cauchy’s textbooks, Historia Mathematica, 14 (1987), pp. 258–274. Legendre, A.M. [1785] Recherches sur l’attraction des sph´ero¨ıdes homog`enes, M´emoires de math´ematique et de physique, pr´esent´es a ` l’Acad´emie Royale des Sciences par divers savans et lus dans ses assembl´ees, Acad´emie des Sciences (Paris), 10 (1785), pp. 411–434. — [1794] M´emoire sur les transcendantes elliptiques, Paris: DuPont and Didot. The second year of the Republic. — [1811–17] Exercices de Calcul Int´egral, sur divers ordres de Transcendantes et sur les quadratures, Paris: Courcier, 1811–1817. Leibniz, G. W. [1682] De vera proportione circulis ad quadratum circumscriptum in numeris rationalibus expressa, Acta Eruditorum (1682), pp. 41–46. In Leibniz [GMS, 5:118–122]. — [1684] Meditationes de cognitione, veritate et ideis, Acta Eruditorum (1684) pp. 537–542. In Leibniz [GP, 4:422–426]. — [1691] Quadratura arithmetica communis sectionum conicarum, quae centrum habent, indeque ducta trigonometria canonica ad quantamcumque in numeris exactitudinem a tabularum necessitate liberata, cum usu speciali ad lineam rhomborum nauticam, aptatumque illi planisphaerium, Acta Eruditorum (1691), pp. 178–182. In Leibniz [GMS, 5:128–132]. — [1693] Supplementum geometriae practicae sese ad problemata transcendentia extendens, ope novae methodi generalissimae per series inﬁnitas, Acta Eruditorum (1693) pp. 178–180. In Leibniz [GMS, 5:285–288]. — [1710] Symbolismus memorabilis calculi algebraici et inﬁnitesimalis in comparatione potentiarum et diﬀerentiarum, et de lege homogeneorum transcendentali, Miscellanea Berolinensia (1710), pp. 160–165. In Leibniz [GMS, 5:377–382]. — [1713] Epistola ad V. Cl. Christianum Wolﬁum, Professorem matheseos Halensem, circa scientiam inﬁniti, Acta Eruditorum Supplem., 5 (1713), pp. 264–270. In Leibniz [GMS, 5:382–487]. — [D] Gothofridi Guillemi Leibnitii Opera Omnia, nunc primum collecta, in classes distribuita, praefationibus et indicibus exhornata, studio Ludovici Dutens, Genevae: Fratres de Tournes, 1768 (6 vols.). — [GP] Die philosophischen Schriften von G.W. Leibniz, C.I. Gerhardt (ed.), Berlin: Weidmann, 1875–1890. — [GMS] Leibnizes mathematische Schriften, C.I. Gerhardt (ed.), Berlin: Asher & Comp. 1849–1850: vols. I–II; Halle: H. W. Schmidt, 1855–1863: vols. III–VII.

References

377

— [GBLW] Briefwechsel zwischen Leibniz und Christian Wolﬀ, C.I. Gerhardt (ed.), Halle: 1860, reprinted Hildesheim: 1963. — [KQA] De quadratura arithmetica circuli ellipseos et hyperbolae cujus corollarium est trigonometria sine tabulis, E. Knobloch (ed.), G¨ ottingen: Vandenhoeck & Ruprecht, 1993. Leibniz, G.W.–Bernoulli, Johann [LB] Virorum celeberr. Got. Gul. Leibnitii et Johan. Bernoulli Commercium philosophicarum et mathematicarum, Lousannae et Genevae: M.M.Bousquet, 1745. Libri, G [1846] Review of P.H. Fuss, Correspondance math´ematique et physique de quelque c´el`ebres g´eom`etres du XVIII`eme si`ecle, Journal des Savants, Janvier 1846, pp. 50–62. Lorgna, A–M, [1787] Th´eorie d’une novelle esp`ece de calcul ﬁni et inﬁnit´esimal, M´emoire de l’academie Royal de Turin, 8 (1786–1787), pp. 411–430. L¨ utzen, J. [1993] Euler vision of a general partial–diﬀerential calculus for a generalized kind of functions, Mathematics Magazine, 56 (1983), pp. 299–306. — [2003a] The foundation of analysis in the 19th Century, in Jahnke [2003, 155–195]. — [2003b] Between rigor and applications. Developments in the concept of function in mathematical analysis, The Cambridge History of Science, Marry Jo Nye (ed.), vol. 5, pp. 468–487, Cambridge: Cambridge University Press, 2003. de l’Hˆ opital, G.F.A. [1696] Analyse des Inﬁniment petits, pour l’intelligence des lignes courbes, 2nd ed., Paris: F. Montalant, 2 ed., 1716. Maclaurin, C. [1742] A Treatise of Fluxions in Two Books, Edinburgh: Ruddimans 1742, 2 vols. Mahoney, M.S. [1994] The Mathematical Career of Pierre de Fermat 1601–1665, 2nd ed., Princeton: Princeton University Press, 1994. Maier` u, L. [1994] Fra Descartes e Newton. Isaac Barrow e John Wallis, Cosenza: Rubbettino, 1994. — [1995] Metodi degli “Antichi ” e dei “Moderni ” nell’algebra di John Wallis. Considerazioni circa i nessi tra loro esistenti, in Panza–Roero [1995, 85–130]. — [2000] Il “continuo” nella Arithmetica Inﬁnitorum di John Wallis, Atomismo e Continuo nel XVII secolo in Atti del Convegno Internazionale “Atomisme et Continuum au XVIII Si`ecle”, Napoli, 28–29–30 aprile 1997, E. Festa and R. Gatto (eds.), Napoli: Vivarium, Napoli, 2000, pp. 151–181. Malet, A. [1991] Mathematics and Mathematization in the Seventeenth Century, Studies in History and Philosophy of Science, 22 (1991) pp. 673–678. — [1993] James Gregorie on Tangents and the “Taylor” Rule of Series Expansions, Archive for History of Exact Sciences, 46 (1993), pp. 97–137.

378

References

— [2006] Renaissance notions of number and magnitude, Historia Mathematica, 33 (2006), pp. 63–81. Massa, M.R. [1997], Mengoli on “quasi proportions”, Historia Mathematica, 24 (1997), pp. 257–280. Mazet, E. [2003] La th´eorie des s´eries de Nicole Oresme dans sa perspective aristot´elicienne. “Questions 1 et 2 sur la G´eom´etrie d’Euclide”, Revue d’histoire des math´ematiques 9 (2003), pp. 33–80. Mengoli, P. [1650] Novae quadraturae arithmeticae, seu de additione fractionum. Bologna: Iacobi Montii, 1650. — [1659] Geometriae speciosae elementa. Bologna: 1659. — [1672] Circolo. Bologna: 1672. Menninger, K. [1969] Number Words and Number Symbols: A Cultural History of Numbers, revised German edition translated by Paul Broneer, Cambridge: MIT Press, 1969. Mercator, N. [1668] Logarithmotechnia; sive Methodus construendi logarithmos nova, accurata et facilis, London: G. Godbid, 1668. de Moivre, A. [1697] A method of raising an inﬁnite multinomial to any given power, or extracting any given roots of the same, Philosophical Transaction, 9 (no. 230, 1697), pp. 619–625. — [1698] A method of extracting the root of an inﬁnite equation, Philosophical Transaction, 20 (no. 240, 1698), pp. 190–193. — [1718] The Doctrine of Chances or a Method of Calculating the Probabilities of Events in Play, London 1718. — [1722] De fractionibus algebraicis radicalitate immunibus ad fractionibus simpliciores reducendis, de que summandis terminis quarundum serierum aequali intervallo a se distantibus, Philosophical Transaction, 32 (1722), pp. 167–172. — [1730] Miscellanea analytica de seriebus et quadraturis, London: J. Tomson and J. Watts, 1730. — [1738] The Doctrine of Chances or a Method of Calculating the Probabilities of Events in Play, 2nd ed., London, 1738. — [1756] The Doctrine of Chances or a Method of Calculating the Probabilities of Events in Play, 3rd ed., London, 1756. Netz, R. [1999a] The Shaping of Deduction in Greek Mathematics, Cambridge: Cambridge University Press, 1999. — [1999b] Archimedes transformed: The case of a result stating a maximum for a cubic equation, Archive for History of Exact Sciences, 54 (1999), pp. 1–47. Newton, I. [1704] Tractatus de quadratura curvarum, in [OO, 1:332– 386]. — [1707] Arithmetica universalis, sive de compositione et de resolutione arithmetica liber, in [OO, 1:1:234]. — [1711] De analysi per æquationes numero terminorum inﬁnitas, in [OO, 1:257–283].

References

379

— [CE] Commercium epistolicum d. Johannis Collins et aliorium de analysi promota, London: Iussu Soc. Regoae, typis Paeersoniani, 1712. — [1720] Universal Arithmetick, or, a treatise of arithmetical composition and resolution, translated by J. Raphson. London: Senex, Taylor, Warner and Osborn, 1720. — [1736] The Method of Fluxions and inﬁnite series; with its Application to the geometry, J. Colson (ed.), London: H.Woodfall and J.Nourse, 1736. — [OO] Opera quae exstant omnia commentariis illustrabat Samuel Horsley, Londini: J. Nichols, 1779–1885, 5 vols. — [PN] Philosophia naturalis principia mathematica. The third edition (1726) with variant readings assembled and edited by Alexandre Koyr´e and I. Bernard Cohen, with the assistence of Anne Whitman, Cambridge: Cambridge University Press, 1972. — [M] The Mathematical Principles of Natural Philosophy, book two, translated by Andrew Motte, London: Benjamin Motte, 1729. — [C] The Correspondance of Isaac Newton, H.W. Turnbull, J.W. Scott and A.R. Hall (eds.), Cambridge: Cambridge University Press, 1959–1977, 7 vols. — [MP] The Mathematical Papers of Isaac Newton, D.T. Whiteside (ed.), Cambridge: Cambridge University Press, 1967–1981, 8 vols. — [QHW] Introduction to the quadrature of the curves, translated by John Harris, D.R. Wilkins (ed.), http://www.maths.tcd.ie/pub/HistMath/ People /Newton/ Quadratura/ HarrisIQ.pdf. Oresme, N. [A] Quaestiones super geometriam Euclidis, H.L.L. Busard (ed.), Leiden: E.J. Brill, 1961. Otte, M. [1989] The ideas of Hermann Grassmann in the context of the mathematical and philosophical tradition since Leibniz, Historia Mathematica 16 (1989), pp. 1–35. Otte, M. and Panza. M. [1997] (eds.) Analysis and synthesis in Mathematics. History and Philosophy, Dordrecht: Kluwer A.P., 1997. Panza, M. [1992] La forma della quantit` a. Analisi algebrica e analisi superiore: il problema dell’unit` a della matematica nel secolo dell’illuminismo, vols. 38–39 of the Cahiers d’Historie et de Philosophie des Sciences, 1992. — [1995] Da Wallis a Newton: una via verso il calcolo. Qudrature, serie e rappresentazioni inﬁnite delle quantit` a e delle forme transcendenti, in Panza–Roero [1995, 131–219]. — [1996] Concept of function, between quantity and form, in the eighteenth century, in H.N. Jahnke, N. Knoche and M. Otte (eds.), History of Mathematics and Education: Ideas and Experiences, G¨ ottingen: Vandenhoeck & Ruprecht, 1996, pp. 241–274. — [1997a] Classical sources for the concepts of analysis and synthesis, in Otte and Panza [1997, 365–414].

380

References

— [1997b] Quelques distinctions a` l’usage de l’historiographie des math´ematiques, in J.–M. Salanskis, F. Rastier, R. Scheps (eds.), Herm´eneutique: texts, sciences, Paris: PUF, 1997, pp. 357–382. — [2003] Newton, Paris: Les belles lettres, 2003. Panza, M. and Roero, S. (eds.) [1995] Geometria, ﬂussioni e differenziali. Tradizione e innovazione nella matematica. Napoli: La citt` a del sole, 1995. Pensivy, M. [1987–88] Jalons historique pour une ´epist´emologie de la s´erie du binˆ ome, Sciences et Techniques en Perspective, 14 (1987–1988). Poisson, S.D. [1805] D´emonstration du th´eor`eme de Taylor, Correspondance sur l’Ecole Polytechnique, 1 (1804–1808), pp. 52–55. — [1811] Note sur le d´eveloppement des puissances des sinus et des cosinus en s´eries es sinus ou des cosinus d’arcs multiples, Correspondance sur l’Ecole Polytechnique, 2 (1809–1813), pp. 212–217. Pourciau, B. [1998] The preliminary mathematical lemmas of Newton’s Principia, Archive for History of Exact Sciences, 52 (1998), pp. 279–295. — [2001] Newton and the notion of limit, Historia Mathematica 28 (2001), pp. 18–30. Reiﬀ, R. [1889] Geschichte der unendlichen Reihen, T¨ ubingen: H. Laupp, 1889. Rothe, H.A. [1793] Formulae de serierum reversione demostratio universalis, signis localibus, combinatorio-analyticorum vicariis exhibita, Leipzig: Sommer 1793. R¨ uthing, D. [1984] Some deﬁnitions of the concept of function from Joh. Bernoulli to N. Bourbaki, Mathematical Intellingencer, 6(4) (1984), pp. 72–77. Salanskis, J.-M. and Sinaceur, H. (Eds.) [1992] Le Labyrinthe du Continu, Paris: Springer–Verlag France, 1992. Scott, J.F. [1938] The Mathematical Works of John Wallis, 2nd ed., New York: Chelsea Publishing Company, 1981. Servois, F.J. [1814–15a] Essai sur un nouveau mode d’exposition des principes du calcul diﬀ´erentiel, Annales des Math´ematiques Pures et Appliqu´ees, 5 (1814), pp. 93–140. — [1814–15b] R´eﬂexion sur les divers syst`emes d’exposition des principes du calcul diﬀ´erentiel, Annales des math´ematiques pures et appliqu´ees, 5 (1814), pp. 141–170. Smith, D. [1959] A Source Book in Mathematics, New York: Dover, 1959. Stevin, S. [1585] L’Arithm´etique de Simon Stevin de Bruges, Leyden: 1585. Stirling, J. [1717] Lineae Tertii Ordinis Neutonianae, Oxford: e Thetro Sheldoniano, 1717. — [1719] Methodus diﬀerentialis Newtoniana illustrata, Philosophical Transactions of the Royal Society, 30 (1719), pp. 1050–1070.

References

381

— [1730] Methodus diﬀerentialis sive Tractatus de Summatione et Interpolatione Serierum Inﬁnitarum, London: G. Bowyer and Straham, 1730. Taylor, B. [1715] Methodus incrementorium directa et inversa, London: G. Innys, 1715. Todhunter, I. [1873] A History of the Mathematical Theories of Attraction and the Figure of the Earth from Newton to Laplace, London : Macmillan, 1873. Torricelli, E. [1644] Opera geometrica, Firenze: Masse et de Laudis, 1644. Turnbull, H.W. [GT] James Gregory Tercentenary Memorial Volume, London, 1939. Tweddle, I. [2003] James Stirling’s Methodus Diﬀerentialis. An Annotated Translation of Stirling’s Text. London: Springer, 2003. Varignon, P. [1715] Pr´ecautions `a prendre dans l’usage des suites ou series inﬁnies r´esultantes, tant da la division inﬁnie des fractions, que du d´eveloppement `a l’inﬁni des puissance d’exposants n´egatifs entiers, M´emoires de l’Academie Royal des Sciences. M´emoires Math´ematiques et Physique (1715), pp. 203–225. Vi` ete, F. [1591] In artem analyticem isagoge, Turonis: J. Mettayers, 1691. English translation in Klein [1968, 315–353]. — [1593] Variorum de rebus mathematicis responsorum. Liber VIII, Tours 1593, in Opera mathematica: in unum volumen congesta ac recognita opera atque studio Francisci a Schooten, Lyede: B. et A. Elzeviriorum, 1646. Wallis, J. [1656] Arithmetica Inﬁnitorum sive Mova Methodus Inquirendi in Curvilineorum Quadraturam, aliaque diﬃciliora Matheseos Problemata, Oxon: Lichﬁeld, 1656. — [1657]. Mathesis universalis, sive Arithmeticum Opus Integrum, in Wallis [1693–1699, vol. 1]. — [1668] Logarithmotechnia Nicolae Mercatoris. Concerning which we shall here deliver the account of the Judicious Dr. I. Wallis given in a Letter to the Viscount Brouncker, Philosophical Transactions, 1668 (3), pp. 753– 768. — [1693] De Algebra Tractatus historicus et practicus, in Wallis [1693– 1699], vol. 2. — [1693–1699] Opera Mathematica, Oxford: Lichﬁeld, 1693–1699. Westfall, R. [1980] Never at Rest. A Biography of Isaac Newton, Cambridge: Cambridge University Press, 1980. Wilson, C. [1980] Perturbations and solar tables from Lacaille to Delambre: The rapprochement of observation and theory, Part II, Archive for History of Exact Sciences, 22 (1980), pp. 53–188 and 189–304. — [1985] The great inequality of Jupiter and Saturn: from Kepler to Laplace, Archive for the History of Exact Sciences, 33 (1985), pp. 215–290.

382

References

Wollenshl¨ ager, K. [1933] Der Mathematische Briefweschel zwischen Johann I Bernoulli und Abraham de Moivre, Verhandlungen der Naturforschenden Gesellschaft in Basel, 43 (1933), pp. 151–317. Youschkevitch, A.P. [1976] The concept of function up to the middle of the 19th century, Archive for History of Exact Sciences, 16 (1976), pp. 37–84.

Author Index Eschenbach, C.H., 286 Euclid, 5, 81, 94, 96 Euler, L., 32, 102, 104–108, 110, 131–132, 137–140, 146, 147, 151, 153, 155–169, 171–179, 181–199, 201–202, 205–210, 212–213, 215–216, 218–222, 224–226, 228–229, 231, 236, 240, 252–253, 254–262, 263–265, 269–272, 275–282, 283–285, 297–298, 312, 319, 322, 323, 326, 330–331, 343, 351, 358

Aepinus, F.U.T., 285 Agostini, A., 8 d’Alembert, J.B., 202, 204, 231, 252, 280, 303–309, 335, 344 Amp`ere, A.M., 290 Arbogast, L.-F.-A., 290–292, 351 Archimedes, 3 Aristotle, 93–94 Barrow, I., 80 Bayes, T., 183 Bernoulli, D., 90–91, 126, 127, 128, 129, 137, 138, 155, 182, 218, 227, 270, 278 Bernoulli, Jacob, 10, 31, 79–85, 121, 127, 286 Bernoulli, Johann, 2, 33, 45–51, 205, 209, 286, 360 Bernoulli, N., 216, 217, 218, 219, 221, 226 Bessel, F.W., 271, 272, 326, 336, 342 Bolzano, B., 311 Bombelli, R., 17, 18 Bos, H., 34, 208 Bossut, C., 312 Breger, H., 102, 103 Briggs, H., 22 Brisson, B., 292, 293 Brouncker, W., 17 Burchard, J., 91 B¨ urmann, H., 286 Burn, R.P., 7 Bussotti, P., 338

Feigenbaum, L., 45, 46, 51, 87, 88, 89, 91, 92 Ferraro, G., 15, 58, 102, 104, 106, 118, 120, 288, 290, 306, 307, 340, 343, 350 Fourier, J., 281, 311, 315–322, 347 Fran¸cais, J.-F., 293–294 Fraser, C., 207, 210, 222, 227, 263, 264, 297 Friedman, M.C., 96 Fuss, P.H., 126, 127, 156, 157, 216, 217, 219, 222, 226 Gauss, C.F., 311, 323–345, 347, 357 Gilain, C., 251 Goldbach, C., 126–127, 129–130, 155–157, 222 Goldstine, H., 22, 23, 295 Golland, L.A., 228, 276 Golland, R.W., 228, 276 Grabiner, J.V., 303, 305 Grandi, G., 25, 121, 126 Grattan-Guiness, I., 93, 278, 281, 315 Gray, J., 301, 338, 342, 349 Gr´egoire de Saint-Vincent, 2, 3, 6 Gregory [or Gregorie], J., 2, 20–24, 87 Guicciardini, N., 25, 99, 146, 147

Callet, J.F., 312 Cardano, G., 107, 108 Cataldi, P. A., 17, 18 Cauchy, A.L., 1, 165, 285, 311, 327, 347–362 Cavalieri, B., 7, 10 Clairaut, A.C., 281, 282 Colson, J., 53, 286 Condorcet, M. J.-A. N., 251

Halley, E., 20 Hardy, G.H., 128, 219, 225 Hindenburg, C. F., 286 Horsley, S., 76 Horv` ath, M., 34 Huygens, C., 18, 25, 35

Dedekind, R., 335 Descartes, R., 97, 98, 107 Dhombres, J., 6, 211

383

384 von Humboldt, W., 287 Jahnke, H.N., 208, 237, 287, 312, 313, 323, 328

Author Index Menninger, K., 104 Mercator, N., 2, 19, 20, 48, 58 de Moivre, A., 87, 131, 133–137, 146, 285, 286

Kant, I., 336–337 K¨ astner, A. G., 286 Katz, V.J., 208 Klein, J., 94, 96, 97, 98, 104–105 Kline, M., 282 Kl¨ ugel, C. S., 286 Knobloch, E., 25, 33–34, 125 Koppelman, E., 295 Kramp, C., 286, 330, 341

Netz, R., 96, 97 Newton, I., 1, 2, 19, 20, 22, 23, 31, 53, 54–76, 79, 87, 89, 91, 98, 99, 102, 103, 105, 106, 117, 150, 248, 283

Lacroix, S.F., 102, 237, 264, 298, 299, 315 Lagrange, J.-L., 138, 140, 184, 201–202, 206–209, 212, 227–228, 231, 233–237, 239–242, 244, 249, 250, 251, 256, 263, 264, 267, 268, 280, 281, 283, 287–291, 294, 303, 305–309, 312, 315, 319, 320, 323, 326, 335, 348–350, 359, 360 Lambert, J.H., 237 Landen, J., 284–285 Laplace, P.S., 184, 231, 237, 242–250, 274, 287, 315 Laugwitz, D., 348, 349, 350, 354 Legendre, A.M., 184, 209, 210, 231, 272–274, 299–301, 315, 327 Leibniz, G. W., 7, 1 2, 20, 25–53, 57, 66, 69, 70, 77, 79, 82, 84, 85, 87, 91, 95, 98–100, 102, 108, 117, 121–126, 193, 202, 205, 218, 239, 244, 245, 263, 283, 286, 290, 294, 327 Libri, G, 31, 270 Lorgna, A-M., 294 L¨ utzen, J., 348 de l’Hˆ opital, G.F.A., 50, 101, 111–112, 159

Panza, M., 7, 10, 45, 53, 54, 56, 58, 66, 68, 76, 87, 91, 95, 99, 101, 118, 120, 147, 203, 205, 206, 208, 212, 233, 236, 244, 245, 287, 288, 290, 306, 307, 350, 360 Pfaﬀ, J. F., 237, 287 Pensivy, M., 23, 284, 355 Poisson, S. D., 290, 312, 313 Pourciau, B., 65, 67 von Prasse, M., 286

Maclaurin, C., 91, 131, 147–153, 171–179, 183, 236, 240, 343 Mahoney, M.S., 103 Maier` u, L., 10 Malet, A., 24, 104 Massa, M.R., 10 Mazet, E., 5 Mengoli, P., 2, 7, 8–10

Olbers, H. W., 330 Oresme, N., 5, 9 Otte, M., 36

Ramanujan, S. A., 219 Ravetz, J. R, 278, 281, 315, 317, 318, 319 Rothe, H. A., 45, 237 R¨ uthing, D., 348 Schumacher, H. C., 332 Schwab, J. C., 336 Scott, J. F., 10 Servois, F. J., 294–295 Stevin, S., 104–107 Stirling, J., 91–92, 131, 141–144, 146, 151, 153, 181, 248 de Serasa, A. A., 7 Taylor, B., 2, 24, 45, 54, 87–92, 147, 149, 151, 176, 228, 231, 237, 243, 259, 276, 283, 287, 288, 290, 293, 359–361 Tetens, I. K., 286 Todhunter, I., 272, 273 T¨opfer, H. A., 286 Torricelli, E., 7, 22, 23, 24 Turnbull, H. W., 22, 23, 24 Varignon, P., 34, 125, 126

385

Author Index Vi`ete, F. 2, 5, 6, 16, 81, 97, 104, 106 Wallis, J., 2, 10–15, 17–19, 48, 53, 57, 63, 104, 105, 156, 157 Westfall, R., 53

Wilson, C., 272 Wollenshl¨ ager, K., 87 Wolﬀ C., 100, 123, 124 Wronski, J., 294 Youschkevitch, A. P., 205, 251

Subject Index Algebraic symbolism, 95, 99 Analytical expressions, 55, 64, 84, 97, 100, 101, 116, 128, 133, 146, 205–211, 213, 253, 256, 263, 268, 301, 303, 332, 348, 349 Axiom, 8, 36, 79, 95–96, 109, 203–204, 219

Euler-Maclaurin summation formula, 147, 151, 153, 171–179, 183, 240 Euler-Mascheroni constant, 196 Everett’s formula, 248

Bernoulli’s method, 133–140 Bernoulli numbers, 177, 343

Figure, 3–4, 10, 36, 37, 38, 40, 44, 45, 49, 53, 72, 95, 96, 98–101, 174, 201–202, 205, 237, 242, 252, 274 Finite diﬀerences, 87, 155, 223, 245, 287 Formal manipulation, 1, 2, 82, 116, 131, 146, 184, 275, 306, 311, 322 Formal procedure, 39, 40, 51, 116, 118, 184, 282, 360 Forward diﬀerences, 22 Function analytical, 251, 263, 288, 290, 305, 306, 332, 359 basic, 207–209, 297, 298, 301 Bessel, 271, 272 beta, 160, 186, 189, 263 of complex variable, 358, 359 digamma, 257, 300, 328, 329, 342, 343 discontinuous, 93, 213, 252–256, 301, 319, 321–322, 340, 351 elementary, 207, 209, 251, 264, 265, 267, 277–278, 290, 297, 298, 311, 324, 349, 357–358 gamma, 141, 159, 257, 299–301, 311, 327–329, 342, 343 generating, 120, 122, 137, 197, 222, 231, 245–249, 264, 325, 354, 357 inexplicable, 252, 257, 259, 261, 262, 330 potential, 273, 274, 278, 280, 311, 348 transcendental, 35, 106, 109, 208, 209, 231, 251, 255, 262, 263–265, 267, 297–301, 324 trigonometric, 208, 262, 263

Calculus of operations, 51, 178, 231, 239, 241, 243, 244, 283, 287, 290–293, 295 Cauchy criterion, 357 Combinatorial School, 237, 286, 323 Continued fraction, 2, 4, 17–18, 131, 132, 185–192, 344 Continuity, 93, 98, 104, 113, 121–123, 163, 210–213, 221, 307, 309, 337, 340–341, 348–351, 354 Continuum, 102–103, 335–341, 348, 349, 353 Convergence, 1–2, 21, 25, 32, 37, 39–40, 43–44, 54, 57, 69–72, 75–76, 81–82, 115, 116, 118, 127, 131, 133, 139, 141, 143, 146, 151, 153, 156, 165, 166, 183, 184, 186–188, 191, 192, 197, 211, 216–218, 224, 225, 227–229, 236–237, 247, 268, 275, 285, 295, 301, 304, 305, 319, 324–326, 333, 341, 344, 351–358, 362 Convergence criterion (or test), 32, 305, 356, 357 Deﬁnition (notion of), 25, 36, 66–67, 95–96, 108–109, 203–204, 219– 220, 225, 328–332 Diagram, 36, 50, 61, 96, 100, 103, 147, 171, 202, 318 Diﬀerential equation, 292–293, 321, 326, 342, 343 Divergence, 138, 165, 181, 182, 225, 301 Elliptic integral, 263, 297, 299, 301

387

388 Generality of algebra, 117, 119, 209–213, 300–301, 324, 327, 347, 351, 354, 358 General term, 15, 132, 133, 137, 144, 155–157, 176–177, 189, 193, 205, 212, 245, 276, 277, 290, 332, 352 Hutton’s method, 128 Inequality technique, 297, 303, 305, 307, 309, 335 Inﬁnite extension, 8, 69, 79, 117, 118, 167, 215, 218, 223, 283, 341, 355, 361 Inﬁnite product, 2, 4, 10, 16, 131, 132, 156, 185–187, 189, 191 Integration, 41, 49, 84, 118, 157, 158, 174, 209, 228, 243, 249, 252–254, 263–265, 275, 290, 298, 301, 321–323, 326, 327, 360, 361 Interpolation, 14, 15, 22, 23, 27, 54, 55, 57, 84, 89, 156, 159, 185, 241, 242, 247, 248, 252, 257, 258, 282, 283, 330, 331 Interval of convergence, 39, 70, 72, 76, 119, 131, 141, 227, 228, 237, 275 Laplace’s equation, 274 Legendre’s polynomials, 273 Leibniz’s criterion, 33 Limit (notion of), 54, 66–67, 111–113, 333–335 Long division, 19, 20, 37, 58, 84 Mercator’s rules, 20, 58 Method of exhaustion, 3, 4, 147

Newton–Stirling formula, 248 Number imaginary, 34, 35, 104, 106–110, 204, 303, 326, 327 irrational, 104, 106–110, 246, 264–265, 331, 337 natural, 16, 57, 104, 105, 107, 198, 204, 286, 331, 338 negative, 14, 104, 106–109, 221, 300

Subject Index Quantitative equality, 115, 226 Quantity abstract, 93, 97, 100 analytical or nonﬁgural, 101 continuous, 93, 101, 102, 105, 212, 335–338 discrete, 93 ﬁctitious, 29, 110 general, 94, 100, 101, 103, 110, 211, 289, 336 geometrical or ﬁgural, 78, 100, 101, 103, 110, 125, 263, 337

Remainder, 8, 20, 37, 58, 84, 202, 207, 221, 306, 308, 350

Series asymptotic, 132, 141, 146, 181, 183–184 Bernoulli, 25, 45, 47, 49, 51, 90, 91 convergent, 1, 20, 54, 62, 65, 69, 71, 122, 123, 127, 131, 138, 141, 146, 150, 151, 163, 181, 184, 217, 218, 220, 223, 225, 226, 229, 300, 305, 320, 347 divergent, 2, 11, 25, 32, 33, 83, 121, 123, 125–127, 129, 131, 146, 153, 166, 182, 183, 188, 191, 192, 216, 218–229, 268, 304, 325, 344–345, 347 Fourier, 281, 315, 317, 319, 321 Grandi’s, 2, 122, 123 harmonic, 9, 31, 81, 82, 162, 165, 166, 173 hypergeometric, 157, 269, 299, 323–331, 333–345 Maclaurin, 91, 147, 152 power, 23, 25, 34, 36, 39, 59, 64, 67, 69, 70, 71, 74, 78, 84, 87, 92, 116, 119, 120, 122, 133, 138, 141, 151, 161, 165, 192, 197, 199, 222, 223, 227, 236, 237, 247, 255, 260, 261, 263, 267, 272, 273, 275, 280, 282, 283, 285, 289, 291, 295, 311, 357, 358, 360 recurrent, 131–135, 137–139, 155, 217, 222

389

Subject Index Stirling’s, 141–146 Taylor, 54, 87–92, 118, 147, 149, 176, 228, 237, 243, 259, 276, 287, 288, 293, 359, 360, 361 trigonometric, 228, 231, 275–282, 311, 315, 319 Summation method (E, 1), 225 Theory of series formal, 132,

201,

283–295,

311–313, 316–322, 324–345, 348–362 formal-quantitative, 115–120 Ultimate value, 29, 66, 115, 183 Upper bound, 334–335 Wallis’s interpolation, 14, 27, 55, 57, 84, 252, 257, 330 Wave equation, 254

Sources and Studies in the History of Mathematics and Physical Sciences Continued from page ii A.W. Grootendorst Jan de Witt’s Elementa Curvarum Linearum, Liber Primus A. Hald A History of Parametric Statistical Inference from Bernoulli to Fischer 1713–1935 T. Hawkins Emergence of the Theory of Lie Groups: An Essay in the History of Mathematics 1869–1926 A. Hermann/K. von Meyenn/V.F. Wcisskopf (Eds.) Wolfgang Pauli: Scientific Correspondence I: 1919–1929 C.C. Heyde/E. Seneta I.J. Bienaym´e: Statistical Theory Anticipated J.P. Hogendijk Ibn Al-Haytham’s Completion of the Conics J. Høyrup Length, Widths, Surfaces: A Portrait of Old Babylonian Algebra and Its Kin B. Hughes Fibonacci’s De Practica Geometrie A. Jones (Ed.) Pappus of Alexandria, Book 7 of the Collection E. Kheirandish The Arabic Version of Euclid’s Optics, Volumes I and II J. Liitzen Joseph Liouville 1809–1882: Master of Pure and Applied Mathematics J. Liitzen The Prehistory of the Theory of Distributions G.H. Moore Zermelo’s Axiom of Choice O. Neugebauer A History of Ancient Mathematical Astronomy O. Neugebauer Astronomical Cuneiform Texts F.J. Ragep Nasır al-Dın al-Tusı’s Memoir on Astronomy ˙ al-hay’a) ˙ (al-Tadhkira f ı cilm B.A. Rosenfeld A History of Non-Euclidean Geometry G. Schubring Conflicts Between Generalization, Rigor and Intuition: Number Concepts Underlying the Development of Analysis in 17th-19th Century France and Germany

Sources and Studies in the History of Mathematics and Physical Sciences Continued from the previous page J. Sesiano Books IV to VII of Diophantus’ Arithmetica: In the Arabic Translation Attributed to Qusta ibn ˙ Luqa L.E. Sigler Fibonacci’s Liber Abaci: A Translation into Modern English of Leonardo Pisano’s Book of Calculation J.A. Stedall The Arithmetic of Infinitesimals: John Wallis 1656 B. Stephenson Kepler’s Physical Astronomy N.M. Swerdlow/O. Neugebauer Mathematical Astronomy in Copernicus’s De Revolutionibus G.J. Toomer (Ed.) Appolonius Conics Books V to VII: The Arabic Translation of the Lost Greek Original in the Version of the Banu Musa, Edited, with English Translation and Commentary by G.J. Toomer G.J. Toomer (Ed.) Diocles on Burning Mirrors: The Arabic Translation of the Lost Greek Original, Edited, with English Translation and Commentary by G.J. Toomer C. Truesdell The Tragicomical History of Thermodynamics, 1822–1854 K. von Meyenn/A. Hermann/V.F. Weisskopf (Eds.) Wolfgang Pauli: Scientific Correspondence II: 1930–1939 K. von Meyenn (Ed.) Wolfgang Pauli: Scientific Correspondence 111: 1940–1949 K. von Meyenn (Ed.) Wolfgang Pauli: Scientific Correspondence IV, Part I: 1950–1952 K. von Meyenn (Ed.) Wolfgang Pauli: Scientific Correspondence IV, Part II: 1953–1954