sign in sign up
<<
Home , Mathematical analysis

Is Mathematical History Written by the Victors?

Jacques Bair, Piotr Błaszczyk, Robert Ely, Valérie Henry, Vladimir Kanovei, Karin U. Katz, Mikhail G. Katz, Semen S. Kutateladze, Thomas McGaffey, David M. Schaps, David Sherry, and Steven Shnider

The ABCs of the History of Infinitesimal Klein’s sentiment was echoed by the philosopher G. Granger, in the context of a discussion of Leibniz, The ABCs of the history of infinitesimal mathemat- in the following terms: ics are in need of clarification. To what extent does Aux yeux des détracteurs de la nouvelle the famous dictum “history is always written by Analyse, l’insurmontable difficulté vient de the victors” apply to the as ce que de telles pratiques font violence well? A convenient starting point is a remark made aux règles ordinaires de l’Algèbre, tout en by Felix Klein in his book Elementary Mathematics conduisant à des résultats, exprimables en from an Advanced Standpoint (Klein [72, p. 214]). termes finis, dont on ne saurait contester Klein wrote that there are not one but two separate l’exactitude. Nous savons aujourd’hui que tracks for the development of : deux voies devaient s’offrir pour la solution du problème: (A) the Weierstrassian approach (in the context [A] Ou bien l’on élimine du langage of an Archimedean ) and mathématique le terme d’infiniment petit, (B) the approach with indivisibles and/or in- et l’on établit, en termes finis, le sens à finitesimals (in the context of what we will donner à la notion intuitive de ‘valeur limite’. refer to as a Bernoullian continuum).1 … [B] Ou bien l’on accepte de maintenir, Jacques Bair is professor of mathematics at the University tout au long du Calcul, la présence d’objets of Liege, Belgium. His email address is [email protected]. portant ouvertement la marque de l’infini, Piotr Błaszczyk is professor of mathematics at the Peda- mais en leur conférant un statut propre qui gogical University of Cracow, Poland. His email address is [email protected]. Mikhail G. Katz is professor of mathematics at Bar Ilan Uni- Robert Ely is professor of mathematics at the University of versity, Israel. His email address is [email protected]. Idaho. His email address is [email protected]. il. Valérie Henry is professor of mathematics, University of Semen S. Kutateladze is professor of mathematics at the Namur and of Liege, Belgium. Her email address is valerie. Sobolev Institute of Mathematics, Novosibirsk State Univer- [email protected]. sity, Russia. His email address is [email protected]. Vladimir Kanovei is professor of mathematics at IPPI, Thomas McGaffey teaches mathematics at Rice University. Moscow, and MIIT, Moscow. His email address is kanovei@ His email address is thomasmcgaffey@sbcglobal.. rambler.ru. David M. Schaps is professor of classical studies at Bar Karin U. Katz teaches mathematics at Bar Ilan University, Ilan University, Israel. His email address is dschaps@mail. Israel. Her email address is karin.usadi.katz@gmail. biu.ac.il. 1Systems of quantities encompassing infinitesimal ones were David Sherry is professor of philosophy at Northern Arizona used by Leibniz, Bernoulli, Euler, and others. Our choice of University. His email address is [email protected]. the term is explained in the subsection “Bernoulli, Johann”. Steven Shnider is professor of mathematics at Bar Ilan Uni- It encompasses modern non-Archimedean systems. versity, Israel. His email address is [email protected]. DOI: http://dx.doi.org/10.1090/noti1026 il.

886 Notices of the AMS 60, Number 7 les insère dans un système dont font aussi from to (see Herzberg 2013 partie les grandeurs finies.… [110]). One example is the visionary work of En- C’est dans cette seconde voie que les riques exploiting infinitesimals, recently analyzed vues philosophiques de Leibniz l’ont orienté. in an article by David Mumford, who wrote: 2 (Granger 1981 [43, pp. 27–28]) In my own education, I had assumed that Thus we have two parallel tracks for con- Enriques [and the Italians] were irrevocably ceptualizing infinitesimal , as shown in stuck.…As I see it now, Enriques must be Figure 1. credited with a nearly complete geometric proof using, as did Grothendieck, higher B-continuum order infinitesimal deformations.…Let’s be careful: he certainly had the correct ideas about infinitesimal , though he had no idea at all how to make precise A-continuum definitions. (Mumford 2011 [89]) Figure 1. Parallel tracks: a thick continuum and a Another example is important work by Cauchy thin continuum. (see the subsection “Cauchy, Augustin-Louis” be- low) on singular and Fourier using At variance with Granger’s appraisal, some of infinitesimals and infinitesimally defined “Dirac” the literature on the history of mathematics tends delta functions (these precede Dirac by a century), to assume that the A-approach is the ineluctably which was forgotten for a number of decades be- “true” one, while the infinitesimal B-approach was, cause of shifting foundational biases. The presence at best, a kind of evolutionary dead end or, at worst, of Dirac delta functions in Cauchy’s oeuvre was altogether inconsistent. To say that infinitesimals noted in (Freudenthal 1971 [40]) and analyzed by provoked passions would be an understatement. Laugwitz (1989 [74]), (1992a [75]); see also (Katz Parkhurst and Kingsland, writing in The Monist, and Tall 2012 [69]) and (Tall and Katz 2013 [107]). proposed applying a saline solution (if we may be Recent papers on Leibniz (Katz and Sherry [67], allowed a pun) to the problem of the infinitesimal: [68]; Sherry and Katz [100]) argue that, contrary [S]ince these two words [infinity and infini- to widespread perceptions, Leibniz’s system for tesimal] have sown nearly as much faulty infinitesimal calculus was not inconsistent (see the logic in the fields of mathematics and meta- subsection “Mathematical Rigor” for a discussion physics as all other fields put together, of the term). The significance and coherence of they should be rooted out of both the fields Berkeley’s critique of infinitesimal calculus have which they have contaminated. And not only been routinely exaggerated. Berkeley’s sarcastic should they be rooted out, lest more errors tirades against infinitesimals fit well with the be propagated by them: a due amount of salt ontological limitations imposed by the A-approach should be ploughed under the infected ter- favored by many historians, even though Berkeley’s ritory, that the damage be mitigated as well opposition, on empiricist grounds, to an infinitely as arrested. (Parkhurst and Kingsland 1925 divisible continuum is profoundly at odds with the [91, pp. 633–634]) [emphasis added—the A-approach. authors] A recent study of Fermat (Katz, Schaps, and Writes P. Vickers: Shnider 2013 [66]) shows how the nature of his So entrenched is the understanding that contribution to the calculus was distorted in the early calculus was inconsistent that recent Fermat scholarship, similarly due to an many authors don’t provide a reference to “evolutionary dead-end” bias (see the subsection the claim, and don’t present the “Fermat, Pierre de”). of inconsistent propositions they have The Marburg school of Hermann Cohen, Cassirer, in mind. (Vickers 2013 [108, section 6.1, p. Natorp, and others explored the philosophical foun- 146]) dations of the infinitesimal method underpinning Such an assumption of inconsistency can influ- the mathematized natural . Their versatile, ence one’s appreciation of historical mathematics, and insufficiently known, contribution is analyzed make a scholar myopic to certain significant devel- in (Mormann and Katz 2013 [88]). opments due to their automatic placement in an A number of recent articles have pioneered “evolutionary dead-end” track, and inhibit potential a reevaluation of the history and philosophy fruitful applications in numerous fields ranging of mathematics, analyzing the shortcomings of received views, and shedding new light on the 2Similar views were expressed by M. Parmentier in (Leibniz deleterious effect of the latter on the philosophy, 1989 [79, p. 36, note 92]). the practice, and the applications of mathematics.

August 2013 Notices of the AMS 887 Some of the conclusions of such a reevaluation are Next, it appears in the papers of as the presented below. following lemma (see Archimedes [2, I, Lamb. 5]): Of unequal lines, unequal surfaces, and un- to Chimeras equal solids [ a, b, c ], the greater exceeds the Some topics from the history of infinitesimals illus- lesser [ a < b ] by such a magnitude [b − a] trating our approach appear below in alphabetical as, when added to itself [ n(b − a) ], can order. be made to exceed any assigned magnitude [c] among those which are comparable with Adequality one another. (Heath 1897 [47, p. 4]) Adequality is a technique used by Fermat to solve This can be formalized as follows: problems of tangents, problems of , and other variational problems. The term (2) (∀a, b, c)(∃n ∈ N) [a < b → n(b − a) > c]. adequality derives from the παρισ oτης´ of Dio- Note that Euclid’s definition V.4 and the lemma of phantus (see the subsection “Diophantus”). The Archimedes are not logically equivalent (see the technique involves an of subsection “Euclid’s Definition V.4”, footnote 11). and “smallness”, represented by a small varia- The Archimedean axiom plays no role in the + − tion E, as in the familiar difference f (A E) f (A). plane geometry as developed in Books I–IV of The El- Fermat used adequality in particular to find the ements.4 Interpreting geometry in ordered fields, or tangents of transcendental curves such as the in geometry over fields in short, one knows that F2 cycloid that were considered to be “mechanical” is a model of Euclid’s plane, where (F, +, ·, 0, 1, <) is curves off-limits to geometry by Descartes. Fermat a Euclidean field, i.e., an ordered field closed under also used it to solve the variational problem of the square root operation. Consequently, R∗ × R∗ the refraction of light so as to obtain Snell’s law. (where R∗ is a hyperreal field) is a model of Adequality incorporated a procedure of discarding Euclid’s plane as well (see the subsection below higher-order terms in E (without setting them on modern implementations). Euclid’s definition equal to zero). Such a heuristic procedure was V.4 is discussed in more detail in the subsection ultimately formalized mathematically in terms of “Euclid’s Definition V.4”. the standard part principle (see the subsection “Standard Part Principle”) in Robinson’s theory of rediscovered the Archimedean axiom infinitesimals starting with (Robinson 1961 [94]). for mathematicians, making it one of his axioms Fermat’s adequality is comparable to Leibniz’s tran- for magnitudes and giving it the following form: if a > b, then there is a multiple of b such scendental law of homogeneity (see the subsection 5 “Lex homogeneorum transcendentalis”). that nb > a (Stolz 1885 [106, p. 69]). At the same time, in his development of the , Archimedean Axiom Stolz implicitly used the Archimedean axiom. Stolz’s visionary realization of the importance of What is known today as the Archimedean axiom the Archimedean axiom and his work on non- first appears in Euclid’s Elements, Book V, as Archimedean systems stand in sharp contrast Definition 4 (Euclid [34, Definition V.4]). It is to Cantor’s remarks on infinitesimals (see the exploited in (Euclid [34, Proposition V.8]). We include bracketed symbolic notation so as to subsection below on mathematical rigor). clarify the definition: In modern mathematics, the theory of or- dered fields employs the following form of the Magnitudes [ a, b ] are said to have a ratio Archimedean axiom (see, e.g., Hilbert 1899 [51, with respect to one another which, being p. 27]): multiplied [ na ] are capable of exceeding one another [ na > b ]. (∀x > 0)(∀ > 0)(∃n ∈ N) [n > x] 3 It can be formalized as follows: or equivalently (1) (∀a, b)(∃n ∈ N) [na > b], where na=a + · · · + a . (3) (∀ > 0)(∃n ∈ N) [n > 1]. | {z } n−times A number system satisfying (3) will be referred to as an Archimedean continuum. In the contrary case, 3See, e.g., the version of the Archimedean axiom in (Hilbert there is an element  > 0 called an infinitesimal 1899 [51, p. 19]). Note that we have avoided using “0” in formula (1), as in “∀a > 0”, since 0 was not part of the con- 4 ceptual framework of the Greeks. The term “multiplied” in With the exception of Proposition III.16, where so-called the English translation of Euclid’s definition V.4 corresponds horn angles appear that could be considered as non- to the Greek term πoλλαπλασ ιαζoµενα´ . A common Archimedean magnitudes relative to rectilinear angles. formalization of the noun “multiple”, πoλλαπλασ´ ιoν, 5See (Ehrlich [32]) for additional historical details concern- is na = a + · · · + a. ing Stolz’s account of the Archimedean axiom.

888 Notices of the AMS Volume 60, Number 7 such that no finite sum  +  + · · · +  will ever Berkeley’s Logical Criticism reach 1; in other words, Berkeley’s logical criticism of the calculus amounts h i (4) (∃ > 0)(∀n ∈ N)  ≤ 1 . to the contention that the evanescent increment is n first assumed to be nonzero to set up an algebraic A number system satisfying (4) is referred to as expression and then is treated as zero in discarding a Bernoullian continuum (i.e., a non-Archimedean the terms that contained that increment when the continuum); see the subsection “Bernoulli, Johann”. increment is said to vanish. In modern terms, Berkeley was claiming that the calculus was based Berkeley, George on an inconsistency of type George Berkeley (1685–1753) was a cleric whose empiricist (i.e., based on sensations, or sensa- (dx 6= 0) ∧ (dx = 0). tionalist) metaphysics tolerated no conceptual innovations, such as infinitesimals, without an The criticism, however, involves a misunderstand- empirical counterpart or referent. Berkeley was ing of Leibniz’s method. The rebuttal of Berkeley’s similarly opposed, on metaphysical grounds, to criticism is that the evanescent increment need not infinite divisibility of the continuum (which he be “treated as zero” but, rather, is merely discarded referred to as extension), an idea widely taken through an application of the transcendental law for granted today. In addition to his outdated of homogeneity by Leibniz, as illustrated in the sub- metaphysical criticism of the infinitesimal calculus section “Product Rule” in the case of the product of Newton and Leibniz, Berkeley also formulated a rule. logical criticism.6 Berkeley claimed to have detected While consistent (in the sense of the subsection a logical fallacy at the basis of the method. In terms “Mathematical Rigor”, level (2)), Leibniz’s system of Fermat’s E occurring in his adequality (see the unquestionably relied on heuristic principles, such subsection “Adequality”), Berkeley’s objection can as the laws of continuity and homogeneity, and be formulated as follows: thus fell short of a standard of rigor if measured by The increment E is assumed to be nonzero today’s criteria (see the subsection “Mathematical at the beginning of the calculation, but zero Rigor”). On the other hand, the consistency and at its conclusion, an apparent logical fallacy. resilience of Leibniz’s system is confirmed through the development of modern implementations of However, E is not assumed to be zero at the end of the calculation, but rather is discarded at the Leibniz’s heuristic principles (see the subsection end of the calculation (see the subsection “Berke- “Modern Implementations”). ley’s Logical Criticism” for more details). Such a technique was the content of Fermat’s adequality Bernoulli, Johann (see the subsection “Adequality”) and Leibniz’s Johann Bernoulli (1667–1748) was a disciple of transcendental law of homogeneity (see the sub- Leibniz’s who, having learned an infinitesimal section “Lex homogeneorum transcendentalis”), methodology for the calculus from the master, where the relation of equality has to be suitably never wavered from it. This is in contrast to Leibniz interpreted (see the subsection “Relation [\ ”). The himself, who throughout his career used both technique is equivalent to taking the (of a typ- (A) an Archimedean methodology (proof by ical expression such as f (A+E)−f (A) for example) in E exhaustion) and Weierstrass’s approach and to taking the standard (B) an infinitesimal methodology part (see the subsection “Standard Part Principle”) in Robinson’s approach. in a symbiotic fashion. Thus Leibniz relied on Meanwhile, Berkeley’s own attempt to explain the A-methodology to underwrite and justify the calculation of the of x2 in The the B-methodology, and he exploited the B- Analyst contains a logical circularity: namely, methodology to shorten the to discovery Berkeley’s argument relies on the determination of (Ars Inveniendi). Historians often name Bernoulli the tangents of a parabola by Apollonius (which is as the first mathematician to have adhered sys- equivalent to the calculation of the derivative). This tematically to the infinitesimal approach as the circularity in Berkeley’s argument was analyzed in basis for the calculus. We refer to an infinitesimal- (Andersen 2011 [1]). enriched number system as a B-continuum, as opposed to an Archimedean A-continuum, i.e., a 6Berkeley’s criticism was dissected into its logical and continuum satisfying the Archimedean axiom (see metaphysical components in (Sherry 1987 [98]). the subsection “Archimedean Axiom”).

August 2013 Notices of the AMS 889 Bishop, Errett Cantor, Georg Errett Bishop (1928–1983) was a mathematical (1845–1918) is familiar to the modern constructivist who, unlike his fellow intuition- reader as the underappreciated creator of the ist7 Arend Heyting (see the subsection “Heyting, “Cantorian paradise”, out of which Arend”), held a dim view of classical mathematics in would not be expelled, as well as the tragic hero, general and Robinson’s infinitesimals in particular. allegedly persecuted by Kronecker, who ended Discouraged by the apparent nonconstructivity of his days in a lunatic asylum. Cantor historian J. Dauben notes, however, an underappreciated his early work in , Bishop believed aspect of Cantor’s scientific activity, namely, his he had found the culprit in the law of the excluded principled persecution of infinitesimalists: middle (LEM), the key logical ingredient in every Cantor devoted some of his most vituper- proof by contradiction. He spent the remaining ative correspondence, as well as a portion eighteen years of his life in an effort to expunge the of the Beiträge, to attacking what he de- reliance on LEM (which he dubbed “the principle scribed at one point as the ‘infinitesimal of omniscience” in [11]) from analysis and sought Cholera bacillus of mathematics’, which had to define meaning itself in mathematics in terms spread from Germany through the work of of such LEM-extirpation. Thomae, du Bois-Reymond, and Stolz, to in- Accordingly, he described classical mathematics fect Italian mathematics.…Any acceptance as both a debasement of meaning (Bishop 1973 of infinitesimals necessarily meant that [13, p. 1]) and sawdust (Bishop 1973 [13, p. 14]), his own theory of number was incomplete. and he did not hesitate to speak of both crisis Thus to accept the work of Thomae, du (Bishop 1975 [11]) and schizophrenia (Bishop 1973 Bois-Reymond, Stolz, and Veronese was to [13]) in contemporary mathematics, predicting an deny the perfection of Cantor’s own cre- imminent demise of classical mathematics in the ation. Understandably, Cantor launched a following terms: thorough campaign to discredit Veronese’s work in every way possible. (Dauben 1980 Very possibly classical mathematics will [27, pp. 216–217]) cease to exist as an independent discipline. A discussion of Cantor’s flawed investigation (Bishop 1968 [10, p. 54]) of the Archimedean axiom (see the subsection His attack in (Bishop 1977 [12]) on calculus “Archimedean Axiom”) may be found in the subsection “Mathematical Rigor”.8 pedagogy based on Robinson’s infinitesimals was a natural outgrowth of his general opposition to the Cauchy, Augustin-Louis logical underpinnings of classical mathematics, as analyzed in (Katz and Katz 2011 [63]). Robinson Augustin-Louis Cauchy (1789–1857) is often viewed in the history of mathematics literature as a pre- formulated a brief but penetrating appraisal of cursor of Weierstrass. Note, however, that contrary Bishop’s ventures into the history and philosophy to a common misconception, Cauchy never gave an of mathematics, when he noted that , δ definition of either limit or continuity (see the the sections of [Bishop’s] book that attempt subsection “ Quantity” for Cauchy’s defini- to describe the philosophical and historical tion of limit). Rather, his approach to continuity background of [the] remarkable endeavor was via what is known today as microcontinuity [of Intuitionism] are more vigorous than (see the subsection “Continuity”). Several recent articles, (Błaszczyk et al. [14]; Borovik and Katz accurate and tend to belittle or ignore the [16]; Bråting [20]; Katz and Katz [62], [64]; Katz and efforts of others who have worked in the Tall [69]), have argued that a proto-Weierstrassian same general direction. (Robinson 1968 [95, view of Cauchy is one-sided and obscures Cauchy’s p. 921]) important contributions, including not only his See the subsection “Chimeras” for a related infinitesimal definition of continuity but also such criticism by Alain Connes. innovations as his infinitesimally defined (“Dirac”) delta , with applications in Fourier anal- ysis and evaluation of singular integrals, and his 7Bishop was not an intuitionist in the narrow sense of the study of orders of growth of infinitesimals that term, in that he never worked with Brouwer’s continuum or anticipated the work of Paul du Bois-Reymond, “free choice ”. We are using the term “intuitionism” in a broader sense (i.e., mathematics based on intuitionistic 8Cantor’s dubious claim that the infinitesimal leads to con- logic) that incorporates constructivism, as used for example tradictions was endorsed by no less an authority than by in the comment quoted at the end of B. Russell; see footnote 15 in the subsection “Mathematical this subsection. Rigor”.

890 Notices of the AMS Volume 60, Number 7 R Borel, Hardy, and ultimately (Skolem [102], [103], trace − (featured on the front cover of Connes’s [104]) Robinson. magnum opus) and the Hahn–Banach , in To elaborate on Cauchy’s “Dirac” delta function, Connes’s own framework; see also [65]. note the following formula from (Cauchy 1827 [23, See the subsection “Bishop, Errett” for a related p. 188]) in terms of an infinitesimal α: criticism by Errett Bishop. 1 Z a+ α dµ π (5) F(µ) = F(a). 2 2 Continuity to Indivisibles 2 a− α + (µ − a) 2 α Continuity Replacing Cauchy’s expression α2+(µ−a)2 by δa(µ), one obtains Dirac’s formula up to trivial modifica- Of the two main definitions of continuity of a tions (see Dirac [30, p. 59]): function, Definition A (see below) is operative in Z ∞ either a B-continuum or an A-continuum (satisfy- f (x)δ(x) = f (0). ing the Archimedean axiom; see the subsection −∞ “Archimedean Axiom”), while Definition B works Cauchy’s 1853 paper on a notion closely related only in a B-continuum (i.e., an infinitesimal-enriched to was recently examined or Bernoullian continuum; see the subsection in (Katz and Katz 2011 [62]) and (Błaszczyk et “Bernoulli, Johann”). al. 2012 [14]). Cauchy handles the said notion • Definition A (, δ approach): A real func- using infinitesimals, including one generated by tion f is continuous at a real point x if and the null ( 1 ). n only if

Chimeras (∀ > 0)(∃δ > 0)(∀x0) Alain Connes (1947–) formulated criticisms of  |x − x0| < δ → |f (x) − f (x0)| <  . Robinson’s infinitesimals between the years 1995 • Definition B (microcontinuity): A real func- and 2007 on at least seven separate occasions tion f is continuous at a real point x if and (see Kanovei et al. 2012 [57, Section 3.1, Table 1]). only if These range from pejorative epithets such as 0  0 0  “inadequate,” “disappointing,” “chimera,” and “irre- (6) (∀x ) x [\ x → f (x ) [\ f (x) . mediable defect,” to “the end of the rope for being In formula (6) the natural extension of f is still ‘explicit’.” denoted f , and the symbol “ [\ ” stands for the Connes sought to exploit the Solovay model S 0 relation of being infinitely close; thus, x [\ x if and (Solovay 1970 [105]) as ammunition against only if x0 − x is infinitesimal (see the subsection , but the model tends to “Relation ”). boomerang, undercutting Connes’s own earlier [\ work in analysis. Connes described Diophantus the hyperreals as both a “virtual theory” and a “chimera”, yet acknowledged that his argument Diophantus of Alexandria (who lived about 1,800 relies on the transfer principle (see the subsection years ago) contributed indirectly to the develop- “Modern Implementations”). In S, all definable sets ment of infinitesimal calculus through the tech- of reals are Lebesgue measurable, suggesting that nique called παρισ oτης´ , developed in his work Connes views a theory as being “virtual” if it is not Arithmetica, Book Five, problems 12, 14, and 17. definable in a suitable model of ZFC. If so, Connes’s The term παρισ oτης´ can be literally translated as claim that a theory of the hyperreals is “virtual” is “approximate equality”. This was rendered as adae- refuted by the existence of a definable model of qualitas in Bachet’s Latin translation [4] and adé- the hyperreal field (Kanovei and Shelah [59]). Free galité in French (see the subsection “Adequality”). ultrafilters aren’t definable, yet Connes exploited The term was used by Fermat to describe the such ultrafilters both in his own earlier work on comparison of values of an algebraic expression, the classification of factors in the 1970s and 80s or what would today be called a function f , at and in his magnum opus Noncommutative Geom- nearby points A and A + E and to seek extrema etry (Connes 1994 [26, Ch. V, Sect. 6.δ, Def. 11]), by a technique closely related to the vanishing of f (A+E)−f (A) raising the question whether the latter may not E after discarding the remaining E-terms; be vulnerable to Connes’s criticism of virtuality. see (Katz, Schaps, and Shnider 2013 [66]). The article [57] analyzed the philosophical under- pinnings of Connes’s argument based on Gödel’s Euclid’s Definition V.4 incompleteness theorem and detected an apparent Euclid’s Definition V.4 has already been discussed circularity in Connes’s logic. The article [57] also in the subsection “Archimedean Axiom”. In addition documented the reliance on nonconstructive foun- to Book V, it appears in Books X and XII and is dational material, and specifically on the Dixmier used in the (see Euclid [34,

August 2013 Notices of the AMS 891 Propositions X.1, XII.2]). The method of exhaustion Euler, Leonhard was exploited intensively by both Archimedes Euler’s Introductio in Analysin Infinitorum (1748 and Leibniz (see the subsection “Leibniz’s De [35]) contains remarkable calculations carried out Quadratura” on Leibniz’s work De Quadratura). It in an extended number system in which the basic was revived in the nineteenth century in the theory algebraic operations are applied to infinitely small of the Riemann . and infinitely large quantities. Thus, in Chapter 7, Euclid’s Book V sets the basis for the theory “Exponentials and Expressed through of similar figures developed in Book VI. Great Series”, we find a derivation of the mathematicians of the seventeenth century, such for az starting from the formula aω = 1 + kω, as Descartes, Leibniz, and Newton, exploited for ω infinitely small, and then raising the Euclid’s theory of similar figures of Book VI while to the infinitely great power12 j = z for a finite paying no attention to its axiomatic background.9 ω (appreciable) z to give Over time Euclid’s Book V became a subject of interest for historians and editors alone. az = ajω = (1 + kω)j To formalize Definition V.4, one needed a and finally expanding the right-hand side as a formula for Euclid’s notion of “multiple” and an power series by of the binomial formula. idea of . Some progress in this direction 10 In the chapters following, Euler finds infinite was made by Robert Simson in 1762. In 1876 product expansions factoring the power series provided a modern reconstruction expansion for transcendental functions (see the of Book V. Combining his own historical studies subsection “Euler’s Infinite Product Formula for with an idea of order compatible with addition Sine”). By Chapter 10 he has the tools to sum the developed by Hermann Grassmann (1861 [44]), he series for ζ(2) where ζ(s) = P n−s . He explicitly gave a formula that to this day is accepted as a n calculates ζ(2k) for k = 1,..., 13 as well as many formalization of Euclid’s definition of proportion other related infinite series. in V.5 (Hankel 1876 [46, pp. 389–398]). Euclid’s In Chapter 3 of his Institutiones Calculi Differen- proportion is a relation among four “magnitudes”, tialis (1755 [37]), Euler deals with the methodology such as of the calculus, such as the nature of infinitesimal A : B :: C : D. and infinitely large quantities. We will cite the It was interpreted by Hankel as the relation English translation [38] of the Latin original [37].  (∀m, n) (nA >1 mB → nC >2 mD) Here Euler writes that ∧ (nA = mB → nC = mD) even if someone denies that infinite num- ∧ (nA < mB → nC < mD), bers really exist in this world, still in 1 2 mathematical speculations there arise ques- where n, m are natural numbers. The indices tions to which answers cannot be given on the inequalities emphasize the fact that the unless we admit an infinite number. (ibid., “magnitudes” A, B have to be of “the same kind”, §82) [emphasis added—the authors] e.g., line segments, whereas C,D could be of Euler’s approach, countenancing the possibility another kind, e.g., triangles. of denying that “infinite numbers really exist,” is In 1880 J. L. Heiberg, in his edition of Archimedes’ consonant with a Leibnizian view of infinitesimal Opera omnia in a comment on a lemma of and infinite quantities as “useful fictions” (see Katz Archimedes, cites Euclid’s Definition V.4, not- and Sherry [67]; Sherry and Katz [100]). Euler then ing that these two are the same axioms (Heiberg notes that “an infinitely small quantity is nothing 1880 [49, p. 11]).11 This is the reason why Eu- but a vanishing quantity, and so it is really equal clid’s Definition V.4 is commonly known as the Archimedean axiom. Today we formalize Euclid’s to 0.” (ibid., §83) Definition V.4 as in (1), while the Archimedean Similarly, Leibniz combined a view of infinitesi- lemma is rendered by formula (2). mals as “useful fictions” and inassignable quantities, with a generalized notion of “equality” that was 9Leibniz and Newton apparently applied Euclid’s conclu- an equality up to an incomparably negligible term. sions in a context where the said conclusions did not, Leibniz sought to codify this idea in terms of his technically speaking, apply: namely, to infinitesimal fig- transcendental law of homogeneity (TLH); see the ures such as the characteristic triangle, i.e., triangle with subsection “Lex homogeneorum transcendentalis”. sides dx, dy, and ds. Thus, Euler’s formulas such as 10See Simson’s axioms that supplement the definitions of Book V as elaborated in (Simson 1762 [101, p. 114–115]). (7) a + dx = a 11In point of fact, Euclid’s axiom V.4 and Archimedes’ lemma are not equivalent from the logical viewpoint. Thus, 12Euler used the symbol i for the infinite power. Blanton re- the additive semigroup of positive appreciable limited placed this by j in the English translation so as√ to avoid a hyperreals satisfies V.4 but not Archimedes’ lemma. notational clash with the standard symbol for −1.

892 Notices of the AMS Volume 60, Number 7 (where a “is any finite quantity” ibid., §§86, 87) Euler proceeds to present the usual rules of are consonant with a Leibnizian tradition (cf. for- infinitesimal calculus, which go back to Leibniz, mula (17) in the subsection “Lex homogeneorum L’Hôpital, and the Bernoullis, such as transcendentalis”). To explain formulas like (7), (9) a dxm + b dxn = a dxm Euler elaborated two distinct ways ( and geometric) of comparing quantities in the following provided m < n “since dxn vanishes compared terms: with dxm” (ibid., §89), relying on his “geometric” equality. Euler introduces a distinction between Since we are going to show that an infinitely infinitesimals of different order and directly small quantity is really zero, we must 13 meet the objection of why we do not computes a ratio of the form always use the same symbol 0 for infinitely dx ± dx2 = 1 ± dx = 1 small quantities, rather than some special dx ones…[S]ince we have two ways to compare of two particular infinitesimals, assigning the them, either arithmetic or geometric, let us value 1 to it (ibid., §88). Euler concludes: look at the quotients of quantities to be Although all of them [infinitely small quan- compared in order to see the difference. tities] are equal to 0, still they must be If we accept the notation used in the carefully distinguished one from the other analysis of the infinite, then dx indicates if we are to pay attention to their mutual a quantity that is infinitely small, so that relationships, which has been explained both dx = 0 and a dx = 0, where a is any through a geometric ratio. (ibid., §89). finite quantity. Despite this, the geomet- The Eulerian hierarchy of orders of infinitesimals ric ratio a dx : dx is finite, namely a : 1. harks back to Leibniz’s work (see the subsection For this reason, these two infinitely small “Nieuwentijt, Bernard” for a historical dissenting quantities, dx and a dx, both being equal view). to 0, cannot be confused when we consider their ratio. In a similar way, we will deal Euler’s Infinite Product Formula for Sine with infinitely small quantities dx and dy. (ibid., §86, pp. 51–52) [emphasis added—the The fruitfulness of Euler’s infinitesimal approach authors] can be illustrated by some of the remarkable applications he obtained. Thus, Euler derived an Euler proceeds to clarify the difference between the infinite product decomposition for the sine and arithmetic and geometric comparisons as follows: sinh functions of the following form: Let a be a finite quantity and let dx be ! ! x2 x2 infinitely small. The arithmetic ratio of (10) sinh x = x 1 + 1 + equals is clear: Since ndx = 0, we have π 2 4π 2 ! ! ± − = x2 x2 a ndx a 0. · 1 + 1 + ,... 9π 2 16π 2 On the other hand, the geometric ratio is ! ! clearly of equals, since x2 x2 (11) sin x = x 1 − 1 − 2 2 a ± ndx π 4π (8) = 1. ! ! a x2 x2 · 1 − 1 − ... From this we obtain the well-known rule that 9π 2 16π 2 the infinitely small vanishes in comparison Decomposition (11) generalizes an infinite product with the finite and hence can be neglected. π formula for 2 due to Wallis [109]. Euler also (Euler 1755 [38, §87]) [emphasis in the 1 1 summed the inverse square series: 1 + 4 + 9 + original—the authors] 2 1 + · · · = π (see [86]) and obtained additional Like Leibniz, Euler considers more than one way of 16 6 comparing quantities. Euler’s formula (8) indicates identities. A common feature of these formulas that his geometric comparison is identical with is that Euler’s computations involve not only the Leibnizian TLH; namely, Euler’s geometric infinitesimals but also infinitely large natural comparison of a pair of quantities amounts to their numbers, which Euler sometimes treats as if they ratio being infinitely close to 1. The same is true 13 for TLH. Thus one has a + dx = a in this sense Note that Euler does not “prove that the expression is equal to 1”; such indirect proofs are a trademark of the , δ for an appreciable a 6= 0, but not dx = 0 (which is approach. Rather, Euler directly computes (what would to- true only arithmetically in Euler’s sense). Euler’s day be formalized as the standard part of) the expression, “geometric” comparison was dubbed “the principle illustrating one of the advantages of the B-methodology over of cancellation” in (Ferraro 2004 [39, p. 47]). the A-methodology.

August 2013 Notices of the AMS 893 were ordinary natural numbers.14 Similarly, Euler Euler’s Sine Factorization Formalized treats infinite series as of a specific Euler’s argument in favor of (10) and (11) was infinite degree. formalized in terms of a “nonstandard” proof in The derivation of (10) and (11) in (Euler 1748 (Luxemburg 1973 [82]). However, the formalization [35, §156]) can be broken up into seven steps as in [82] deviates from Euler’s argument, beginning follows. with Steps 3 and 4, and thus circumvents the more Step 1. Euler observes that problematic Steps 5 and 6. !j !j A proof in the framework of modern non- x x (12) 2 sinh x = ex − e−x = 1 + − 1 − , standard analysis formalizing Euler’s argument j j step-by-step throughout appeared in (Kanovei 1988 where j (or “i” in Euler [35]) is an infinitely large [56]); see also (McKinzie and Tuckey 1997 [86]) . To motivate the next step, note and (Kanovei and Reeken 2004 [58, §2.4a]). This that the expression xj − 1 = (x − 1)(1 + x + x2 + formalization interprets problematic details of j−1 Qj−1 · · · + x ) can be factored further as k=0(x − Euler’s argument on the basis of general principles ζk), where ζ = e2πi/j ; conjugate factors can then of modern nonstandard analysis, as well as general be combined to yield a decomposition into real analytic facts that were known in Euler’s time. Such quadratic terms. principles and facts behind some early proofs in Step 2. Euler uses the fact that aj −bj is the product infinitesimal calculus are sometimes referred to of the factors as “hidden lemmas” in this context; see (Laugwitz [73], [74]), (McKinzie and Tuckey 1997 [86]). For 2 2 2kπ (13) a + b − 2ab cos , where k ≥ 1 , instance, the “hidden lemma” behind Step 4 above j is the fact that, for a fixed x, the terms of the together with the factor a − b and, if j is an even Maclaurin expansion of cos x tend to 0 faster than number, the factor a + b as well. a convergent geometric series, allowing one to x x Step 3. Setting a = 1 + j and b = 1 − j in (12), infer that the effect of the transformation of Step Euler transforms expression (13) into the form 4 on the product of the factors (14) is infinitesimal. x2  x2  2kπ Some “hidden lemmas” of a different kind, related (14) 2 + 2 − 2 1 − cos . to basic principles of nonstandard analysis, are j2 j2 j discussed in [86, pp. 43ff.]. Step 4. Euler then replaces (14) by the expression What clearly stands out from Euler’s argument 4k2π 2  x2 x2  is his explicit use of infinitesimal expressions (15) 1 + − , j2 k2π 2 j2 such as (14) and (15), as well as the approximate justifying this step by means of the formula formula (16), which holds “up to” an infinitesimal of higher order. Thus, Euler used infinitesimals 2kπ 2k2π 2 (16) cos = 1 − . par excellence, rather than merely ratios thereof, 2 j j in a routine fashion in some of his best work. Step 5. Next, Euler argues that the difference ex − Euler’s use of infinite integers and their associ- − e x is divisible by the expression ated infinite products (such as the decomposition x2 x2 of the sine function) were interpreted in Robin- 1 + − k2π 2 j2 son’s framework in terms of hyperfinite sets. Thus 2 Euler’s product of j-infinitely many factors in (11) from (15), where “we omit the term x since even j2 is interpreted as a hyperfinite product in [58, when multiplied by j, it remains infinitely small.” formula (9), p. 74]. A hyperfinite formalization of (English translation from [36].) Euler’s argument involving infinite integers and Step 6. As there is still a factor of a − b = 2x/j, their associated products illustrates the successful Euler obtains the final equality (10), arguing that remodeling of the arguments (and not merely the then “the resulting first term will be x” (in order to results) of classical infinitesimal mathematics, as conform to the Maclaurin series for sinh x). discussed in the subsection “Mathematical Rigor”. Step 7. Finally, formula (11) is obtained from (10) Fermat, Pierre de by means of the substitution x , ix.  (1601–1665) developed a pio- We will discuss modern formalizations of Euler’s neering technique known as adequality (see the argument in the next subsection. subsection “Adequality”) for finding tangents to curves and for solving problems of maxima and 14Euler’s procedure is therefore consonant with the Leib- nizian law of continuity (see the subsection “Lex continui- minima. Katz, Schaps, and Shnider (2013 [66]) tatis”), though apparently Euler does not refer explicitly to analyze some of the main approaches in the lit- the latter. erature to the method of adequality, as well as

894 Notices of the AMS Volume 60, Number 7 its source in the παρισ oτης´ of Diophantus (see [Y]ou connected this extremely abstract part the subsection “Diophantus”). At least some of of with a theory apparently the manifestations of adequality, such as Fermat’s so far apart as the elementary calculus. In treatment of transcendental curves and Snell’s law, doing so you threw new light on the history amount to variational techniques exploiting a small of the calculus by giving a clear sense to (alternatively, infinitesimal) variation E. Fermat’s Leibniz’s notion of infinitesimals. (ibid) treatment of geometric and physical applications Intuitionist Heyting’s admiration for the appli- suggests that an aspect of approximation is inher- cation of Robinson’s infinitesimals to calculus ent in adequality, as well as an aspect of smallness pedagogy is in stark contrast with the views of his on the part of E. fellow constructivist E. Bishop (subsection “Bishop, Fermat’s use of the term adequality relied on Errett”). Bachet’s rendering of Diophantus. Diophantus coined the term parisotes for mathematical pur- Indivisibles versus Infinitesimals poses. Bachet performed a semantic calque in Commentators use the term infinitesimal to refer passing from par-iso¯o to ad-aequo. A historically to a variety of conceptions of the infinitely small, significant parallel is found in the similar role but the variety is not always acknowledged. It is of, respectively, adequality and the transcenden- important to distinguish the infinitesimal methods tal law of homogeneity (see the subsection “Lex of Archimedes and Cavalieri from those employed homogeneorum transcendentalis”) in the work of, by Leibniz and his successors. To emphasize this respectively, Fermat and Leibniz on the problems distinction, we will say that tradition prior to of maxima and minima. Leibniz employed indivisibles. For example, in his Breger (1994 [21]) denies that the idea of heuristic proof that the of a parabolic segment “smallness” was relied upon by Fermat. However, a is 4/3 the area of the inscribed triangle with the detailed analysis (see [66]) of Fermat’s treatment of same base and vertex, Archimedes imagines both the cycloid reveals that Fermat did rely on issues figures to consist of perpendiculars of various of “smallness” in his treatment of the cycloid heights erected on the base. The perpendiculars and reveals that Breger’s interpretation thereof are indivisibles in the sense that they are limits of contains both mathematical errors and errors of division and so one dimension less than the area. textual analysis. Similarly, Fermat’s proof of Snell’s In the same sense, the indivisibles of which a line law, a variational principle, unmistakably relies on consists are points, and the indivisibles of which a ideas of “smallness”. solid consists are planes. Cifoletti (1990 [25]) finds similarities between Leibniz’s infinitesimals are not indivisibles, for Fermat’s adequality and some procedures used they have the same dimension as the figures that in smooth infinitesimal analysis of Lawvere and compose them. Thus, he treats curves as comprising others. Meanwhile, J. Bell (2009 [9]) seeks the his- infinitesimal line intervals rather than indivisible torical sources of Lawvere’s infinitesimals mainly points. The strategy of treating infinitesimals as in Nieuwentijt (see the subsection “Nieuwentijt, dimensionally homogeneous with the objects they Bernard”). compose seems to have originated with Roberval Heyting, Arend or Torricelli, Cavalieri’s student, and to have been explicitly arithmetized in (Wallis 1656 [109]). Arend Heyting (1898–1980) was a mathematical Zeno’s paradox of extension admits resolution intuitionist whose lasting contribution was the in the framework of Leibnizian infinitesimals formalization of the intuitionistic logic underpin- (see the subsection “Zeno’s Paradox of Extension”). ning the Intuitionism of his teacher Brouwer. While Furthermore, only with the dimensionality retained Heyting never worked on any theory of infinites- is it possible to make sense of the fundamental imals, he had several opportunities to present theorem of calculus, where one must think about an expert opinion on Robinson’s theory. Thus, the rate of change of the area under a curve, another in 1961, Robinson made public his new idea of reason why indivisibles had to be abandoned nonstandard models for analysis and “communi- in favor of infinitesimals so as to enable the cated this almost immediately to…Heyting” (see development of the calculus (see Ely 2012 [33]). Dauben [28, p. 259]). Robinson’s first paper on the subject was subsequently published in Proceedings Leibniz to Nieuwentijt of the Netherlands Royal Academy of Sciences [94]. Heyting praised nonstandard analysis as “a stan- Leibniz, Gottfried dard model of important mathematical research” Gottfried Wilhelm Leibniz (1646–1716), the co- (Heyting 1973 [50, p. 136]). Addressing Robinson, inventor of infinitesimal calculus, is a key player in he declared: the parallel infinitesimal track referred to by Felix

August 2013 Notices of the AMS 895 Klein [72, p. 214] (see the section “The ABCs of the place, Leibniz wrote the treatise at a time when History of Infinitesimal Mathematics”). infinitesimals were despised by the French Acad- Leibniz’s law of continuity (see the subsection emy, a society whose approval and acceptance “Lex continuitatis”) and his transcendental law of he eagerly sought. More importantly, as (Jesseph homogeneity (which he had already discussed in 2013 [55]) has shown, De Quadratura depends on his response to Nieuwentijt in 1695, as noted by infinitesimal resources in order to construct an M. Parmentier [79, p. 38], and later in greater detail approximation to a given curvilinear area less than in a 1710 article [78] cited in the seminal study any previously specified error. This problem is of Leibnizian methodology by H. Bos [17]) form a reminiscent of the difficulty that led to infinitesimal basis for implementing the calculus in the context methods in the first place. Archimedes’ method of of a B-continuum. exhaustion required one to determine a value for Many historians of the calculus deny significant the in advance of showing, by reductio continuity between infinitesimal calculus of the argument, that any departure from that value seventeenth century and twentieth-century devel- entails a contradiction. Archimedes possessed opments such as Robinson’s theory (see further a heuristic, indivisible method for finding such discussion in Katz and Sherry [67]). Robinson’s hy- values, and the results were justified by exhaustion, perreals require the resources of modern logic; thus but only after the fact. By the same token, the many commentators are comfortable denying a his- use of infinitesimals is “just” a shortcut only if it torical continuity. A notable exception is Robinson is entirely eliminable from quadratures, tangent himself, whose identification with the Leibnizian constructions, etc. Jesseph’s insight is that this is tradition inspired Lakatos, Laugwitz, and others not the case. to consider the history of the infinitesimal in a Finally, the syncategorematic interpretation mis- more favorable light. Many historians have over- represents a crucial aspect of Leibniz’s mathemat- estimated the force of Berkeley’s criticisms (see the ical philosophy. His conception of mathematical subsection “Berkeley, George”) by underestimating fiction includes imaginary numbers, and he often the mathematical and philosophical resources sought approbation for his infinitesimals by com- available to Leibniz. paring them to imaginaries, which were largely Leibniz’s infinitesimals are fictions—not logical uncontroversial. There is no suggestion by Leibniz fictions, as (Ishiguro 1990 [54]) proposed, but that imaginaries are eliminable by long-winded rather pure fictions, like imaginaries, which are not paraphrase. Rather, he praises imaginaries for eliminable by some syncategorematic paraphrase; their capacity to achieve universal harmony by the see (Sherry and Katz 2012 [100]) and the subsection greatest possible systematization, and this char- “Leibniz’s De Quadratura” below. acteristic is more central to Leibniz’s conception In fact, Leibniz’s defense of infinitesimals is more of infinitesimals than the idea that they are mere firmly grounded than Berkeley’s criticism thereof. shorthand. Just as imaginary roots both unified and Moreover, Leibniz’s system for differential calculus extended the method for solving cubics, likewise was free of logical fallacies (see the subsection infinitesimals unified and extended the method “Berkeley’s Logical Criticism”). This strengthens the for quadrature so that, e.g., quadratures of general conception of modern infinitesimals as a formaliza- parabolas and hyperbolas could be found by the tion of Leibniz’s strategy of relating inassignable same method used for quadratures of less difficult to assignable quantities by means of his transcen- curves. dental law of homogeneity (see the subsection “Lex homogeneorum transcendentalis”). Lex continuitatis Leibniz’s De Quadratura A heuristic principle called The law of continuity In 1675 Leibniz wrote a treatise on his infinitesi- (LC) was formulated by Leibniz and is a key mal methods, On the Arithmetical Quadrature of to appreciating Leibniz’s vision of infinitesimal the Circle, the Ellipse, and the Hyperbola, or De calculus. The LC asserts that whatever succeeds in Quadratura, as it is widely known. However, the the finite succeeds also in the infinite. This form treatise appeared in print only in 1993 in a text of the principle appeared in a letter to Varignon edited by Knobloch (Leibniz [80]). (Leibniz 1702 [77]). A more detailed form of LC in De Quadratura was interpreted by R. Arthur terms of the concept of terminus appeared in his [3] and others as supporting the thesis that Leib- text Cum Prodiisset: niz’s infinitesimals are mere fictions, eliminable In any supposed continuous transition, by long-winded paraphrase. This so-called syn- ending in any terminus, it is permissible categorematic interpretation of Leibniz’s calculus to institute a general reasoning, in which has gained a number of adherents. We believe the final terminus may also be included. this interpretation to be a mistake. In the first (Leibniz 1701 [76, p. 40])

896 Notices of the AMS Volume 60, Number 7 ( ) ( ) ( ) assignable LC inassignable TLH assignable The TLH associates to an inassignable quantity Ž Ž quantities quantities quantities (such as a + dx) an assignable one (such as a); see Figure 2 for a relation between LC and TLH. Figure 2. Leibniz’s law of continuity (LC) takes one from assignable to inassignable quantities, Mathematical Rigor while his transcendental law of homogeneity There is a certain lack of clarity in the historical (TLH; the subsection “Lex homogeneorum literature with regard to issues of fruitfulness, transcendentalis”) returns one to assignable consistency, and rigorousness of mathematical quantities. writing. As a rough guide and to be able to formulate useful distinctions when it comes to evaluating mathematical writing from centuries To elaborate, the LC postulates that whatever past, we would like to consider three levels of properties are satisfied by ordinary or assignable judging mathematical writing: quantities should also be satisfied by inassignable (1) potentially fruitful but (logically) inconsis- quantities (see the subsection “Variable Quantity”) tent, such as infinitesimals (see Figure 2). Thus the (2) (potentially) consistent but informal, trigonometric formula sin2 x + cos2 x = 1 should (3) formally consistent and fully rigorous ac- cording to currently prevailing standards. be satisfied for an inassignable (e.g., infinitesi- mal) input x as well. In the twentieth century As an example of level (1) we would cite the work this heuristic principle was formalized as the of Nieuwentijt (see the subsection “Nieuwentijt, Bernard” for a discussion of the inconsistency). transfer principle (see the subsection “Modern Our prime example of level (2) is provided by the Implementations”) of Ło´s–Robinson. Leibnizian laws of continuity and homogeneity (see The significance of LC can be illustrated by the subsections “Lex continuitatis” and “Lex homoge- fact that a failure to take note of the law of conti- neorum transcendentalis”), which found rigorous nuity often led scholars astray. Thus, Nieuwentijt implementation at level (3) only centuries later (see the subsection “Nieuwentijt, Bernard”) was led (see the subsection “Modern Implementations”). into something of a dead end with his nilpotent A foundational rock of the received history of 1 infinitesimals (ruled out by LC) of the form ∞ . mathematical analysis is the belief that mathemat- J. Bell’s view of Nieuwentijt’s approach as a pre- ical rigor emerged starting in the 1870s through cursor of nilsquare infinitesimals of Lawvere (see the efforts of Cantor, Dedekind, Weierstrass, and Bell 2009 [9]) is plausible, though it could be noted others, thereby replacing formerly unrigorous that Lawvere’s nilsquare infinitesimals cannot be work of infinitesimalists from Leibniz onward. The 1 of the form ∞ . philosophical underpinnings of such a belief were analyzed in (Katz and Katz 2012a [64]), where it Lex homogeneorum transcendentalis was pointed out that in mathematics, as in other Leibniz’s transcendental law of homogeneity, or sciences, former errors are eliminated through a lex homogeneorum transcendentalis in the origi- process of improved conceptual understanding, evolving over time, of the key issues involved in nal Latin (Leibniz 1710 [78]), governs that . involving differentials. Leibniz historian H. Bos Thus no scientific development can be claimed interprets it as follows: to have attained perfect clarity or rigor merely on A quantity which is infinitely small with the grounds of having eliminated earlier errors. respect to another quantity can be neglected Moreover, no such claim for a single scientific if compared with that quantity. Thus all development is made either by the practitioners terms in an equation except those of the or by the historians of the natural sciences. It was highest order of infinity, or the lowest order further pointed out in [64] that the term mathemat- of infinite smallness, can be discarded. For ical rigor itself is ambiguous, its meaning varying instance, according to context. Four possible meanings for the term were proposed in [64]: + = (17) a dx a (1) it is a shibboleth that identifies the speaker dx + ddy = dx as belonging to a clan of professional mathematicians; etc. The resulting equations satisfy this…re- (2) it represents the idea that, as a scien- quirement of homogeneity. (Bos 1974 [17, tific field develops, its practitioners attain p. 33]) greater and more conceptual understand- For an interpretation of the equality in the ing of key issues and are less prone to formulas above, see the subsection “Relation [\ ”. error;

August 2013 Notices of the AMS 897 (3) it represents the idea that a search for by remodeling both its propositions and its pro- greater correctness in analysis inevitably cedures and reasoning. Using Weierstrassian , δ led Weierstrass specifically to epsilontics techniques, one can recover only the propositions (i.e., the A-approach) in the 1870s; but not the proof procedures. Thus, Euler’s result (4) it refers to the establishment of what are giving an infinite product formula for sine (see the perceived to be the ultimate foundations subsection “Euler’s Infinite Product Formula for for mathematics by Cantor, Sine”) admits numerous proofs in a Weierstrassian explicitly expressed in axiomatic form by context, but Robinson’s framework provides a Zermelo and Fraenkel. suitable context in which Euler’s proof, relying on Item (1) may be pursued by a fashionable infinite integers, can also be recovered. This is the academic in the social sciences but does not get to crux of the historical debate concerning , δ versus infinitesimals. In short, Robinson did for Leibniz the bottom of the issue. Meanwhile, item (2) would what Hilbert did for Euclid. Meanwhile, epsilontists be agreed upon by historians of the other sciences. failed to do for Leibniz what Robinson did for In this context it is interesting to compare Leibniz, namely, formalizing the procedures and the investigation of the as reasoning of the historical infinitesimal calculus. performed by the would-be rigorist Cantor, on the This theme is pursued further in terms of the one hand, and the infinitesimalist Stolz on the internal/external distinction in the subsection other. Cantor sought to derive the Archimedean “Variable Quantity”. property as a consequence of those of a linear continuum. Cantor’s work in this area was not only Modern Implementations unrigorous but actually erroneous, whereas Stolz’s work was fully rigorous and even visionary. Namely, In the 1940s Hewitt [48] developed a modern imple- Cantor’s arguments “proving” the inconsistency of mentation of an infinitesimal-enriched continuum infinitesimals were based on an implicit assumption extending R by means of a technique referred to today as the ultrapower construction. We will of what is known today as the Kerry-Cantor axiom denote such an infinitesimal-enriched continuum (see Proietti 2008 [92]). Meanwhile, Stolz was by the new symbol IR (“thick-R”).16 In the next the first modern mathematician to realize the decade, Ło´s(Ło´s1955 [81]) proved his celebrated importance of the Archimedean axiom (see the theorem on ultraproducts, implying in particular subsection “Archimedean Axiom”) as a separate that elementary (more generally, first-order) state- axiom in its own right (see Ehrlich 2006 [32]) ments over R are true if and only if they are true and, moreover, to develop some non-Archimedean over IR, yielding a modern implementation of the systems (Stolz 1885 [106]). Leibnizian law of continuity (see the subsection In his Grundlagen der Geometrie (Hilbert 1899 “Lex continuitatis”). Such a result is equivalent to [51]), Hilbert did not develop a new geometry, what is known in the literature as the transfer but rather remodeled Euclid’s geometry. More principle; see Keisler [70]. Every finite element of IR specifically, Hilbert brought rigor into Euclid’s is infinitely close to a unique ; see the geometry in the sense of formalizing both Euclid’s subsection “Standard Part Principle”. Such a princi- propositions and Euclid’s style of procedures and ple is a mathematical implementation of Fermat’s style of reasoning. adequality (see the subsection “Adequality”), of Note that Hilbert’s system works for Leibniz’s transcendental law of homogeneity (see built over a non-Archimedean field, as Hilbert the subsection “Lex homogeneorum transcenden- was fully aware. Hilbert (1900 [52, p. 207]) cites talis”), and of Euler’s principle of cancellation (see Dehn’s counterexamples to Legendre’s theorem the discussion between formulas (7) and (9) in the in the absence of the Archimedean axiom. Dehn subsection “Euler, Leonhard”). planes built over a non-Archimedean field were used to prove certain cases of the independence Nieuwentijt, Bernard of Hilbert’s axioms (see Cerroni 2007 [24]).15 In Nieuwentijt’s Analysis Infinitorum (1695), the Robinson’s theory similarly formalized 17 Dutch philosopher (1654–1718) proposed a sys- seventeenth- and eighteenth-century analysis tem containing an infinite number, as well as

15 infinitesimal quantities formed by dividing finite It is a melancholy comment to note that fully three years numbers by this infinite one. Nieuwentijt postu- later the philosopher-mathematician Bertrand Russell was lated that the product of two infinitesimals should still claiming, on Cantor’s authority, that the infinitesimal “leads to contradictions” (Russell 2003 [97, p. 345]). This be exactly equal to zero. In particular, an infini- set the stage for several decades of anti-infinitesimal vitriol, tesimal quantity is nilpotent. In an exchange of including the saline solution of Parkhurst and Kingsland 16 ∗ ∗ (see the section “The ABCs of the History of Infinitesimal A more traditional symbol is R or R . Mathematics”). 17Alternative spellings are Nieuwentijdt or Nieuwentyt.

898 Notices of the AMS Volume 60, Number 7 publications with Nieuwentijt on infinitesimals (see B-continuum Mancosu 1996 [84, p. 161]), Leibniz and Hermann st claimed that this system is consistent only if all infinitesimals are equal, rendering differential  A-continuum calculus useless. Leibniz instead advocated a sys- Figure 3. Thick-to-thin: applying the law of tem in which the product of two infinitesimals homogeneity or taking standard part (the is incomparably smaller than either infinitesimal. thickness of the top line is merely conventional). Nieuwentijt’s objections compelled Leibniz in 1696 to elaborate on the hierarchy of infinite and infini- tesimal numbers entailed in a robust infinitesimal of the product rule. The issue is the justification system. of the last step in the following calculation: Nieuwentijt’s nilpotent infinitesimals of the 1 d(uv) = (u + du)(v + dv) − uv form ∞ are ruled out by Leibniz’s law of continuity (see the subsection “Lex continuitatis”). J. Bell’s (18) = udv + vdu + du dv view of Nieuwentijt’s approach as a precursor of = udv + vdu. nilsquare infinitesimals of Lawvere (see Bell 2009 The last step in the calculation (18), namely, [9]) is plausible, though it could be noted that Lawvere’s nilsquare infinitesimals cannot be of the udv + vdu + du dv = udv + vdu, 1 19 form ∞ . is an application of the TLH. In his 1701 text Cum Prodiisset [76, pp. 46–47], Product Rule to Zeno Leibniz presents an alternative justification of the Product Rule product rule (see Bos [17, p. 58]). Here he divides by dx and argues with differential quotients rather In the area of Leibniz scholarship, the received than differentials. Adjusting Leibniz’s notation to view is that Leibniz’s infinitesimal system was logi- fit with (18), we obtain an equivalent calculation:20 cally faulty and contained internal contradictions d(uv) (u + du)(v + dv) − uv allegedly exposed by the cleric George Berkeley = (see the subsection “Berkeley, George”). Such a view dx dx udv + vdu + du dv is fully compatible with the A-track-dominated = outlook, bestowing supremacy upon the recon- dx udv + vdu du dv struction of analysis accomplished through the = + efforts of Cantor, Dedekind, Weierstrass, and their dx dx udv + vdu rigorous followers (see the subsection “Mathemati- = . cal Rigor”). Does such a view represent an accurate dx   appraisal of Leibniz’s system? du dv Under suitable conditions the term dx is The articles (Katz and Sherry 2012 [67], [68]; infinitesimal, and therefore the last step Sherry and Katz [100]), building on the earlier work udv + vdu du dv dv du (Sherry 1987 [98]), argued that Leibniz’s system (19) + = u + v dx dx dx dx was in fact consistent (in the sense of level (2) of the is legitimized as a special case of the TLH. The TLH subsection “Mathematical Rigor”)18 and featured interprets the equality sign in (19) and (17) as the resilient heuristic principles such as the law of relation of being infinitely close, i.e., an equality continuity (see the subsection “Lex continuitatis”) up to infinitesimal error. and the transcendental law of homogeneity (TLH) (see the subsection “Lex homogeneorum transcen- Relation [\ dentalis”), which were implemented in the fullness Leibniz did not use our equality symbol but rather of time as precise mathematical principles guiding the symbol “ ” (see McClenon 1923 [85, p. 371]). the behavior of modern infinitesimals. [\ Using such a symbol to denote the relation of being How did Leibniz exploit the TLH in developing infinitely close, one could write the calculation the calculus? We will now illustrate an application of the TLH in the particular example of the derivation 19Leibniz had two laws of homogeneity: one for dimen- sion and the other for the order of infinitesimalness. Bos 18Concerning the status of Leibniz’s system for differential notes that they “disappeared from later developments” [17, calculus, it may be more accurate to assert that it was not p. 35], referring to Euler and Lagrange. Note, however, the inconsistent, in the sense that the contradictions alleged similarity to Euler’s principle of cancellation (see Bair et by Berkeley and others turn out not to have been there in al. [5]). the first place once one takes into account Leibniz’s gen- 20The special case treated by Leibniz is u(x) = x. This eralized notion of equality and his transcendental law of limitation does not affect the conceptual structure of the homogeneity. argument.

August 2013 Notices of the AMS 899 enlargement of the number system, as the standard part of the value uH of the natural extension of the sequence at an infinite hypernatural index n = H. Thus,

(20) lim un = st(uH ). n→∞ Here the standard part function “st” associates to Figure 4. Zooming in on infinitesimal ε (here each finite hyperreal the unique finite real infinitely st(±ε) = 0). The standard part function close to it (i.e., the difference between them is associates to every finite hyperreal the unique infinitesimal). This formalizes the natural intuition real number infinitely close to it. The bottom that for “very large” values of the index, the terms line represents the “thin” real continuum. The in the sequence are “very close” to the limit value line at top represents the “thick” hyperreal of the sequence. Conversely, the standard part of a continuum. The “infinitesimal microscope” is hyperreal u = [un] represented in the ultrapower used to view an infinitesimal neighborhood of 0. construction by a (un) is simply The derivative f 0(x) of f (x) is then defined by the limit of that sequence: the relation f 0(x) f (x+ε)−f (x) . [\ ε (21) st(u) = lim un. n→∞ Formulas (20) and (21) express limit and standard of the derivative of y = f (x) where f (x) = x2 as part in terms of each other. In this sense, the follows: procedures of taking the limit and taking the dy standard part are logically equivalent. f 0(x) [\ dx (x + dx)2 − x2 Variable Quantity = dx The mathematical term µεγεϑoς´ in ancient Greek (x + dx + x)(x + dx − x) has been translated into Latin as quantitas. In = dx modern languages it has two competing counter- = 2x + dx parts: in English, quantity, magnitude;22 in French, quantité, grandeur; in German, Quantität, Grösse. [\ 2x. The term grandeur with the meaning real number Such a relation is formalized by the standard is still in use in (Bourbaki 1947 [19]). Variable part function; see the subsection “Standard Part quantity was a primitive notion in analysis as Principle” and Figure 3. presented by Leibniz, l’Hôpital, and later Carnot and Cauchy. Other key notions of analysis were Standard Part Principle defined in terms of variable quantities. Thus, in In any totally ordered field extension E of R, every Cauchy’s terminology a variable quantity becomes finite element x ∈ E is infinitely close to a suitable an infinitesimal if it eventually drops below any unique element x0 ∈ R. Indeed, via the total order, given (i.e., constant) quantity (see Borovik and Katz the element x defines a on R, and the [16] for a fuller discussion). Cauchy notes that the cut specifies a real number x0 ∈ R ⊂ E. The number limit of such a quantity is zero. The notion of limit x0 is infinitely close to x ∈ E. The subring Ef ⊂ E itself is defined as follows: consisting of the finite elements of E therefore Lorsque les valeurs successivement at- admits a tribuées à une même variable s’approchent st : Ef → R, x , x0, indéfiniment d’une valeur fixe, de manière called the standard part function. à finir par en différer aussi peu que l’on The standard part function is illustrated in voudra, cette dernière est appelée la lim- Figure 3. A more detailed graphic representation ite de toutes les autres. (Cauchy, Cours may be found in Figure 4.21 d’Analyse [22]) The key remark, due to Robinson, is that the limit Thus, Cauchy defined both infinitesimals and in the A-approach and the standard part function limits in terms of the primitive notion of a in the B-approach are essentially equivalent tools. variable quantity. In Cauchy, any variable quantity q More specifically, the (un) that does not tend to infinity is expected to can be expressed, in the context of a hyperreal 22The term “magnitude” is etymologically related to 21For a recent study of optical diagrams in nonstandard µεγεϑoς´ . Thus, µεγεϑoς´ in Greek and magnitudo in Latin analysis, see (Dossena and Magnani [31], [83]) and (Bair and both “bigness”, “big” being mega (µεγα´ ) in Greek Henry [8]). and magnum in Latin.

900 Notices of the AMS Volume 60, Number 7 decompose as the sum of a given quantity c and magnitude as though it consists of infinitely many an infinitesimal α: indivisibles; see (Sherry 1988 [99]), (Kirk et al. 1983 [71]). If the indivisibles have no magnitude, then (22) q = c + α. an extension (such as or time) composed In his 1821 text [22], Cauchy worked with a of them has no magnitude; but if the indivisibles hierarchy of infinitesimals defined by polynomials have some (finite) magnitude, then an extension in a base infinitesimal α. Each such infinitesimal composed of them will be infinite. There is a further decomposes as puzzle: If a magnitude is composed of indivisibles, (23) αn(c + ε) then we ought to be able to add or concatenate them in order to produce or increase a magnitude. for a suitable n and infinitesimal ε. Cauchy’s But indivisibles are not next to one another: as expression (23) can be viewed as a generalization limits or boundaries, any pair of indivisibles is of (22). separated by what they limit. Thus, the concept of In Leibniz’s terminology, c is an assigna- addition or concatenation seems not to apply to ble quantity, while α and ε are inassignable. indivisibles. Leibniz’s transcendental law of homogeneity (see The paradox need not apply to infinitesimals the subsection “Lex homogeneorum transcen- in Leibniz’s sense however (see the subsection dentalis”) authorized the replacement of the “Indivisibles versus Infinitesimals”) for, having = + inassignable q c α by the assignable c, since α neither zero nor finite magnitude, infinitely many is negligible compared to c: of them may be just what is needed to produce (24) q [\ c a finite magnitude. And in any case, the addition or concatenation of infinitesimals (of the same (see the subsection “Relation ”). Leibniz empha- [\ dimension) is no more difficult to conceive of than sized that he worked with a generalized notion of adding or concatenating finite magnitudes. This equality where expressions were declared “equal” is especially important, because it allows one to if they differed by a negligible term. Leibniz’s pro- apply arithmetic operations to infinitesimals (see cedure was formalized in Robinson’s B-approach the subsection “Lex continuitatis” on the law of by the standard part function (see the subsection continuity). See also (Reeder 2012 [93]). “Standard Part Principle”), which assigns to each finite the unique real number to which it is infinitely close. As such, the standard Acknowledgments part allows one to work “internally” (not in the The work of Vladimir Kanovei was partially sup- technical NSA sense but) in the sense of exploiting ported by RFBR grant 13-01-00006. M. Katz was concepts already available in the toolkit of the partially funded by the Israel Science Foundation, historical infinitesimal calculus, such as Fermat’s grant No. 1517/12. We are grateful to Reuben adequality (see the subsection “Adequality”), Leib- Hersh and Martin Davis for helpful discussions niz’s transcendental law of homogeneity (see the and to the anonymous referees for a number subsection “Lex homogeneorum transcendentalis”), of helpful suggestions. The influence of Hilton and Euler’s principle of cancellation (see Bair et Kramer (1928–2012) is obvious. al. [5]). Meanwhile, in the A-approach as formalized by Weierstrass, one is forced to work with “exter- References nal” concepts such as the multiple-quantifier , δ [1] K. Andersen, One of Berkeley’s arguments on compensating errors in the calculus, Historia definitions (see the subsection “Continuity”) which Mathematica 38 (2011), no. 2, 219–231. have no counterpart in the historical infinitesimal [2] Archimedes, De Sphaera et Cylindro, in Archimedis calculus of Leibniz and Cauchy. Opera Omnia cum Commentariis Eutocii, vol. I, ed. Thus, the notions of standard part and epsilontic J. L. Heiberg, B. G. Teubner, Leipzig, 1880. limit, while logically equivalent (see the subsec- [3] R. Arthur, Leibniz’s Syncategorematic Infinitesi- tion “Standard Part Principle”), have the following mals, Smooth Infinitesimal Analysis, and Newton’s difference between them: the standard part princi- Proposition 6 (2007). See http://www.humanities. mcmaster.ca/∼rarthur/papers/LsiSiaNp6.rev. ple corresponds to an “internal” development of pdf the historical infinitesimal calculus, whereas the [4] Bachet, Diophanti Alexandrini, Arithmeticorum epsilontic limit is “external” to it. Liber V (Bachet’s Latin translation). [5] J. Bair, P. Błaszczyk, R. Ely, V. Henry, V. Kanovei, Zeno’s Paradox of Extension K. Katz, M. Katz, S. Kutateladze, T. McGaffey, D. Schaps, D. Sherry, and S. Shnider, Interpreting (who lived about 2,500 years ago) Euler’s infinitesimal mathematics (in preparation). raised a puzzle (the paradox of extension, which [6] J. Bair and V. Henry, From Newton’s fluxions is distinct from his better-known paradoxes of to virtual microscopes, Teaching Mathematics and motion) in connection with treating any continuous V (2007), 377–384.

August 2013 Notices of the AMS 901 [7] , From mixed angles to infinitesimals, The [24] C. Cerroni, The contributions of Hilbert and Dehn to College Mathematics Journal 39 (2008), no. 3, non-Archimedean geometries and their impact on the 230–233. Italian school, Revue d’Histoire des Mathématiques [8] , Analyse infinitésimale - Le calculus redécou- 13 (2007), no. 2, 259–299. vert, Editions Academia Bruylant Louvain-la-Neuve [25] G. Cifoletti, La méthode de Fermat: son statut (Belgium) 2008, 189 pages. D/2008/4910/33. et sa diffusion, Algèbre et comparaison de figures [9] J. L. Bell, Continuity and infinitesimals, Stanford dans l’histoire de la méthode de Fermat, Cahiers Encyclopedia of Philosophy, revised 20 July 2009. d’Histoire et de Philosophie des Sciences, Nouvelle [10] E. Bishop, Mathematics as a Numerical Language, Série 33, Société Française d’Histoire des Sciences et 1970 Intuitionism and Proof Theory (Proc. Conf., des Techniques, Paris, 1990. Buffalo, NY, 1968), pp. 53–71. North-Holland, [26] A. Connes, Noncommutative Geometry, Academic Amsterdam. Press, Inc., San Diego, CA, 1994. [11] , The crisis in contemporary mathematics. Pro- [27] J. Dauben, The development of the Cantorian set ceedings of the American Academy Workshop on theory. In (Bos et al. 1980 [18]), pp. 181–219. the Evolution of Modern Mathematics (Boston, Mass., [28] , Abraham Robinson. 1918–1974, Biographical 1974), Historia Mathematica 2 (1975), no. 4, 507–517. Memoirs of the National Academy of Sciences [12] E. Bishop, Review: H. Jerome Keisler, Elementary 82 (2003), 243-284. Available at the addresses Calculus, Bull. Amer. Math. Soc. 83 (1977), 205–208. http://www.nap.edu/html/biomems/arobinson. [13] , Schizophrenia in contemporary mathematics. pdf and http://www.nap.edu/catalog/10683. In Errett Bishop: Reflections on Him and His Research html (San Diego, Calif., 1983), 1–32, Contemp. Math., vol. [29] A. De Morgan, On the early history of 39, Amer. Math. Soc., Providence, RI, 1985 [originally infinitesimals in England, Philosophical Mag- distributed in 1973]. azine, Ser. 4 4 (1852), no. 26, 321–330. [14] P. Błaszczyk, M. Katz, and D. Sherry, See http://www.tandfonline.com/doi/abs/ Ten misconceptions from the history of 10.1080/14786445208647134 analysis and their debunking, Foundations [30] P. Dirac, The Principles of , 4th of Science 18 (2013), no. 1, 43–74. See edition, Oxford, The Clarendon Press, 1958. http://dx.doi.org/10.1007/s10699-012-9285 [31] R. Dossena and L. Magnani, Mathematics through -8 and http://arxiv.org/abs/1202.4153 diagrams: Microscopes in non-standard and smooth [15] P. Błaszczyk and K. Mrówka, Euklides, Elementy, analysis, Studies in Computational Intelligence (SCI) Ksiegi V–VI, Tłumaczenie i komentarz [Euclid, Ele- 64 (2007), 193–213. ments, Books V–VI, Translation and commentary], [32] P. Ehrlich, The rise of non-Archimedean mathe- Copernicus Center Press, Kraków, 2013. matics and the roots of a misconception. I. The [16] A. Borovik and M. Katz, Who gave you emergence of non-Archimedean systems of magni- the Cauchy–Weierstrass tale? The dual his- tudes, Archive for History of Exact Sciences 60 (2006), tory of rigorous calculus, Foundations of no. 1, 1–121. Science 17 (2012), no. 3, 245–276. See [33] R. Ely, Loss of dimension in the history of cal- http://dx.doi.org/10.1007/s10699-011-9235 culus and in student reasoning, The Mathematics -x and http://arxiv.org/abs/1108.2885 Enthusiast 9 (2012), no. 3, 303–326. [17] H. J. M. Bos, Differentials, higher-order differentials [34] Euclid, Euclid’s Elements of Geometry, edited and the derivative in the Leibnizian calculus, Archive and provided with a modern English transla- for History of Exact Sciences 14 (1974), 1–90. tion by Richard Fitzpatrick, 2007. See http:// [18] H. Bos, R. Bunn, J. Dauben, I. Grattan-Guinness, farside.ph.utexas.edu/euclid.html T. Hawkins, and K. Pedersen, From the Calculus [35] L. Euler, Introductio in Analysin Infinitorum, Tomus to , 1630–1910. An Introductory History, primus, SPb and Lausana, 1748. edited by I. Grattan-Guinness, Gerald Duckworth and [36] , Introduction to Analysis of the Infinite. Book I, Co. Ltd., London, 1980. translated from the Latin and with an introduction [19] N. Bourbaki, Théorie de la mesure et de l’intégration, by John D. Blanton, Springer-Verlag, New York, 1988 Introduction, Université Henri Poincaré, Nancy, 1947. [translation of (Euler 1748 [35])]. [20] K. Bråting, A new look at E. G. Björling and the [37] , Institutiones Calculi Differentialis, SPb, 1755. Cauchy sum theorem. Arch. Hist. Exact Sci. 61 (2007), [38] , Foundations of Differential Calculus, English no. 5, 519–535. translation of Chapters 1–9 of (Euler 1755 [37]) by [21] H. Breger, The mysteries of adaequare: a vindication D. Blanton, Springer, New York, 2000. of Fermat, Archive for History of Exact Sciences 46 [39] G. Ferraro, Differentials and differential coefficients (1994), no. 3, 193–219. in the Eulerian foundations of the calculus, Historia [22] A. L. Cauchy, Cours d’Analyse de L’École Roy- Mathematica 31 (2004), no. 1, 34–61. ale Polytechnique. Première Partie. Analyse [40] H. Freudenthal, Cauchy, Augustin-Louis, in Dictio- algébrique. Paris: Imprimérie Royale, 1821. On- nary of Scientific Biography, ed. by C. C. Gillispie, line at http://books.google.com/books?id= vol. 3, Charles Scribner’s Sons, New York, 1971, pp. _mYVAAAAQAAJ&dq=cauchy&lr=&source=gbs_ 131–148. navlinks_s [41] C. I. Gerhardt (ed.), Historia et Origo calculi dif- [23] , Théorie de la propagation des ondes à la sur- ferentialis a G. G. Leibnitio conscripta, Hannover, face d’un fluide pesant d’une profondeur indéfinie, 1846. (published 1827, with additional Notes), Oeuvres, [42] (ed.), Leibnizens mathematische Schriften, Series 1, Vol. 1, 1815, pp. 4–318. Berlin and Halle: Eidmann, 1850–1863.

902 Notices of the AMS Volume 60, Number 7 [43] G. Granger, Philosophie et mathématique leibnizi- [62] , Cauchy’s continuum. Perspectives ennes, Revue de Métaphysique et de Morale 86e on Science 19 (2011), no. 4, 426–452. See Année, No. 1 (Janvier-Mars 1981), 1–37. http://www.mitpressjournals.org/toc/posc/ [44] Hermann Grassmann, Lehrbuch der Arithmetik, 19/4 and http://arxiv.org/abs/1108.4201 Enslin, Berlin, 1861. [63] , Meaning in classical mathematics: Is it at [45] Rupert Hall and Marie Boas Hall, eds., Unpublished odds with intuitionism? Intellectica 56 (2011), no. 2, Scientific Papers of , Cambridge Uni- 223-302. See http://arxiv.org/abs/1110.5456 versity Press, 1962, pp. 15–19, 31–37; quoted in I. [64] , A Burgessian critique of nominalistic Bernard Cohen, Richard S. Westfall, Newton, Norton tendencies in contemporary mathematics and Critical Edition, 1995, pp. 377–386. its historiography, Foundations of Science 17 [46] Hermann Hankel, Zur Geschichte der Mathematik (2012), no. 1, 51–89. See http://dx.doi.org/ in Alterthum und Mittelalter, Teubner, Leipzig, 1876. 10.1007/s10699-011-9223-1 and http://arxiv. [47] T. Heath (ed.), The Works of Archimedes, Cambridge org/abs/1104.0375 University Press, Cambridge, 1897. [65] M. Katz and E. Leichtnam, Commuting and [48] E. Hewitt, Rings of real-valued continuous functions. noncommuting infinitesimals, American Mathemat- I, Trans. Amer. Math. Soc. 64 (1948), 45–99. ical Monthly 120 (2013), no. 7, 631–641. See [49] J. Heiberg (ed.), Archimedis Opera Omnia cum http://arxiv.org/abs/1304.0583 Commentariis Eutocii, Vol. I. Teubner, Leipzig, 1880. [66] M. Katz, D. Schaps, and S. Shnider, Almost equal: [50] A. Heijting, Address to Professor A. Robin- The method of adequality from Diophantus to Fermat son, at the occasion of the Brouwer memorial and beyond, Perspectives on Science 21 (2013), no. 3, lecture given by Professor A. Robinson on 283–324. See http://arxiv.org/abs/1210.7750 the 26th April 1973, Nieuw Arch. Wisk. (3) [67] M. Katz and D. Sherry, Leibniz’s infinitesimals: 21 (1973), 134–137. MathSciNet Review at Their fictionality, their modern implementations, http://www.ams.org/mathscinet-getitem?mr and their foes from Berkeley to Russell and beyond, =434756 Erkenntnis 78 (2013), no. 3, 571–625; see http:// [51] D. Hilbert, Grundlagen der Geometrie, Festschrift dx.doi.org/10.1007/s10670-012-9370-y and zur Enthüllung des Gauss-Weber Denkmals, Göttin- http://arxiv.org/abs/1205.0174 gen, Leipzig, 1899. [68] , Leibniz’s laws of continuity and homogene- [52] , Les principes fondamentaux de la géométrie, ity, Notices of the American Mathematical Society Annales scientifiques de l’E.N.S. 3e série 17 (1900), 59 (2012), no. 11, 1550–1558. See http://www. 103–209. ams.org/notices/201211/ and http://arxiv. [53] O. Hölder, Die Axiome der Quantität und die org/abs/1211.7188 Lehre vom Mass, Berichte über die Verhandlungen [69] M. Katz and D. Tall, A Cauchy-Dirac delta function. der Königlich Sächsischen Gesellschaft der Wis- Foundations of Science 18 (2013), no. 1, 107–123. senschaften zu Leipzig, Mathematisch-Physische See http://dx.doi.org/10.1007/s10699-012- Classe, 53, Leipzig, 1901, pp. 1–63. 9289-4 and http://arxiv.org/abs/1206.0119 [54] H. Ishiguro, Leibniz’s Philosophy of Logic and Lan- [70] H. J. Keisler, The ultraproduct construction. Ultrafil- guage, second edition, Cambridge University Press, ters across mathematics, Contemp. Math., vol. 530, Cambridge, 1990. pp. 163–179, Amer. Math. Soc., Providence, RI, 2010. [55] D. Jesseph, Leibniz on the elimination of infinitesi- [71] G. S. Kirk, J. E. Raven, and M. Schofield, Presocratic mals: Strategies for finding truth in fiction, 27 pages. Philosophers, Cambridge Univ. Press, 1983, §316. In Leibniz on the Interrelations between Mathematics [72] F. Klein, Elementary Mathematics from an Advanced and Philosophy, edited by Norma B. Goethe, Philip Bee- Standpoint. Vol. I, Arithmetic, , Analysis. ley and David Rabouin, Archimedes Series, Springer Translation by E. R. Hedrick and C. A. Noble [Macmil- Verlag, 2013. lan, New York, 1932] from the third German edition [56] V. Kanovei, The correctness of Euler’s method for [Springer, Berlin, 1924]. Originally published as El- the factorization of the sine function into an infinite ementarmathematik vom höheren Standpunkte aus product, Russian Mathematical Surveys 43 (1988), (Leipzig, 1908). 65–94. [73] D. Laugwitz, Hidden lemmas in the early history of [57] V. Kanovei, M. Katz, and T. Mormann, Tools, ob- infinite series, Aequationes Mathematicae 34 (1987), jects, and chimeras: Connes on the role of hyperreals 264–276. in mathematics, Foundations of Science (online first). [74] , Definite values of infinite sums: Aspects of See http://dx.doi.org/10.1007/s10699-012- the foundations of infinitesimal analysis around 9316-5 and http://arxiv.org/abs/1211.0244 1820, Archive for History of Exact Sciences 39 (1989), [58] V. Kanovei and M. Reeken, Nonstandard Analysis, no. 3, 195–245. Axiomatically, Springer Monographs in Mathematics, [75] , Early delta functions and the use of infinites- Springer, Berlin, 2004. imals in research, Revue d’histoire des sciences 45 [59] V. Kanovei and S. Shelah, A definable nonstandard (1992), no. 1, 115–128. model of the reals, Journal of Symbolic Logic 69 [76] G. Leibniz Cum Prodiisset…mss “Cum prodiisset (2004), no. 1, 159–164. atque increbuisset Analysis mea infinitesimalis... ” [60] K. Katz and M. Katz, Zooming in on infinitesi- in Gerhardt [41, pp. 39–50]. mal 1 − .9.. in a post-triumvirate era, Educational [77] , Letter to Varignon, 2 Feb 1702, in Gerhardt Studies in Mathematics 74 (2010), no. 3, 259–273. [42, vol. IV, pp. 91–95]. See http://rxiv.org/abs/arXiv:1003.1501 [78] , Symbolismus memorabilis calculi algebraici [61] , When is .999 ... less than 1? The Montana et infinitesimalis in comparatione potentiarum Mathematics Enthusiast 7 (2010), no. 1, 3–30.

August 2013 Notices of the AMS 903 et differentiarum, et de lege homogeneorum [96] , Selected Papers of Abraham Robinson, Vol. II. transcendentali, in Gerhardt [42, vol. V, pp. 377-382]. Nonstandard Analysis and Philosophy, edited and [79] , La naissance du calcul différentiel, 26 articles with introductions by W. A. J. Luxemburg and S. des Acta Eruditorum. Translated from the Latin and Körner, Yale University Press, New Haven, CT, 1979. with an introduction and notes by Marc Parmentier. [97] B. Russell, The Principles of Mathematics, Vol. I, With a preface by Michel Serres. Mathesis, Librairie Cambridge Univ. Press, Cambridge, 1903. Philosophique J. Vrin, Paris, 1989. [98] D. Sherry, The wake of Berkeley’s analyst: Rigor [80] , De quadratura arithmetica circuli ellipseos mathematicae? Stud. Hist. Philos. Sci. 18 (1987), no. 4, et hyperbolae cujus corollarium est trigonome- 455–480. tria sine tabulis. Edited, annotated, and with a [99] , Zeno’s metrical paradox revisited, Philosophy foreword in German by Eberhard Knobloch, Ab- of Science 55 (1988), no. 1, 58–73. handlungen der Akademie der Wissenschaften in [100] D. Sherry and M. Katz, Infinitesimals, imaginaries, Göttingen, Mathematisch-Physikalische Klasse, Folge ideals, and fictions, Studia Leibnitiana, to appear. See 3 [Papers of the Academy of Sciences in Göt- http://arxiv.org/abs/1304.2137 tingen, Mathematical-Physical Class. Series 3], 43, [101] R. Simson, The Elements of Euclid: Viz. the First Six Vandenhoeck and Ruprecht, Göttingen, 1993. Books, together with the Eleventh and Twelfth. The [81] J. Ło´s, Quelques remarques, théorèmes et problèmes Errors by which Theon and others have long ago sur les classes définissables d’algèbres, in Mathemat- vitiated these Books are corrected and some of Eu- ical Interpretation of Formal Systems, pp. 98–113, clid’s Demonstrations are restored. Robert & Andrew North-Holland Publishing Co., Amsterdam, 1955. Foulis, Glasgow, 1762. [82] W. Luxemburg, What is nonstandard analysis? Pa- [102] T. Skolem, Über die Unmöglichkeit einer vollständi- pers in the foundations of mathematics, American gen Charakterisierung der Zahlenreihe mittels eines Mathematical Monthly 80 (1973), no. 6, part II, 38–67. endlichen Axiomensystems, Norsk Mat. Forenings [83] L. Magnani and R. Dossena, Perceiving the infinite Skr., II. Ser. No. 1/12 (1933), 73–82. and the infinitesimal world: Unveiling and optical [103] , Über die Nicht-charakterisierbarkeit der diagrams in mathematics, Foundations of Science 10 Zahlenreihe mittels endlich oder abzählbar un- (2005), no. 1, 7–23. endlich vieler Aussagen mit ausschliesslich Zahlen- [84] P. Mancosu, Philosophy of Mathematics and Math- variablen, Fundamenta Mathematicae 23 (1934), ematical Practice in the Seventeenth Century, The 150–161. Clarendon Press, Oxford University Press, New York, [104] , Peano’s axioms and models of arithmetic, 1996. in Mathematical Interpretation of Formal Systems, [85] R. McClenon, A contribution of Leibniz to the his- pp. 1–14, North-Holland Publishing Co., Amsterdam, tory of complex numbers, American Mathematical 1955. Monthly 30 (1923), no. 7, 369–374. [105] R. Solovay, A model of set-theory in which ev- [86] M. McKinzie and C. Tuckey, Hidden lemmas in Eu- ery set of reals is Lebesgue measurable, Annals of ler’s summation of the reciprocals of the squares, Mathematics (2) 92 (1970), 1–56. Archive for History of Exact Sciences 51 (1997), 29–57. [106] O. Stolz, Vorlesungen über Allgemeine Arithmetik, [87] H. Meschkowski, Aus den Briefbuchern Georg Can- Teubner, Leipzig, 1885. tors, Archive for History of Exact Sciences 2 (1965), [107] D. Tall and M. Katz, A cognitive analysis of Cauchy’s 503–519. conceptions of function, continuity, limit, and infini- [88] T. Mormann and M. Katz, Infinitesimals as an issue tesimal, with implications for teaching the calculus, of neo-Kantian philosophy of science, HOPOS: The Educational Studies in Mathematics, to appear. Journal of the International Society for the History of [108] P. Vickers, Understanding Inconsistent Science, Philosophy of Science, to appear. Oxford University Press, Oxford, 2013. [89] D. Mumford, Intuition and rigor and Enriques’s [109] J. Wallis, The Arithmetic of Infinitesimals (1656). quest, Notices Amer. Math. Soc. 58 (2011), no. 2, Translated from the Latin and with an introduc- 250–260. tion by Jacqueline A. Stedall, Sources and Studies [90] I. Newton, Methodus Fluxionum (1671); English ver- in the History of Mathematics and Physical Sciences, sion, The Method of Fluxions and Infinite Series Springer-Verlag, New York, 2004. (1736). [110] Frederik S. Herzberg, with in- [91] W. Parkhurst and W. Kingsland, Infinity and the finitesimals, Lecture Notes in Mathematics, 2067, infinitesimal, The Monist 35 (1925), 633–666. Springer, Heidelberg, 2013. [92] C. Proietti, Natural numbers and infinitesimals: a discussion between Benno Kerry and Georg Cantor, History and Philosophy of Logic 29 (2008), no. 4, 343– 359. [93] P. Reeder, Infinitesimals for Metaphysics: Conse- quences for the Ontologies of Space and Time, Ph.D. thesis, Ohio State University, 2012. [94] A. Robinson, Non-standard analysis, Nederl. Akad. Wetensch. Proc. Ser. A 64 = Indag. Math. 23 (1961), 432–440 [reprinted in Selected Works; see (Robinson 1979 [96, pp. 3–11]). [95] , Reviews: Foundations of Constructive Analy- sis, American Mathematical Monthly 75 (1968), no. 8, 920–921.

904 Notices of the AMS Volume 60, Number 7