Traditional logic and the early history of sets, 1854–1908

Traditional Logic and the Early History of Sets, 1854-1908a The final publication is available at Springer via http://dx.doi.org/10.1007/BF00375789 JOSÉ FERREIRÓS a This article develops ideas taken from my dissertation, 'El nacimiento de la teoría de conjuntos, 1854-1908' (Universidad Autónoma de Madrid, 1991), written under the direction of Javier ORDÓÑEZ. The paper has profited from comments on earlier versions made by Ralf HAUBRICH, Gregory H. MOORE, Jeremy GRAY, and Volker PECKHAUS. Also useful were the comments after a presentation of the main ideas before the Logic Colloquium at Berkeley, particularly those of Charles CHIHARA. The English was revised and corrected by John W. ADDISON. I acknowledge gratefully the support of a grant from the Spanish Ministry of Education and Science. Contents I. INTRODUCTION II. TRADITIONAL LOGIC, CONCEPTS, AND SETS 2.1. The doctrine of concepts. 2.2. The diffusion of logic in 19th century Germany. 2.3. Logicism and other views. III. RIEMANN'S MANIFOLDS AS A FOUNDATION FOR MATHEMATICS 3.1. Origins of the notion of manifold. 3.2. RIEMANN's theory of manifolds. 3.2.1. Magnitudes and manifolds. 3.2.2. Arithmetic, topology, and geometry. 3.3. RIEMANN's influence on the history of sets. 1 3.3.1. Diffusion; CANTOR's manifolds. 3.3.2. Topology, continuity, and dimension. 3.3.3. On the way to abstraction. IV. DEDEKIND'S "LOGICAL THEORY OF SYSTEMS" 4.1. The algebraic origins of DEDEKIND's theory of sets. 4.1.1. Groups and fields, sets and maps. 4.1.2. Sets and mappings in a general setting. 4.2. The heuristic way towards ideals. 4.3. The logical theory of systems and mappings. 4.3.1. DEDEKIND's program for the foundations of mathematics. 4.3.2. Set and mapping as logical notions. 4.3.3. DEDEKIND's deductive method. V. INTERLUDE: THE QUESTION OF THE INFINITE 5.1. A reconstruction of RIEMANN's viewpoint. 5.2. DEDEKIND and the infinite. 5.2.1. The roots of DEDEKIND's infinitism. 5.2.2. The definition and the theorem of the infinite. VI. FROM MARRIAGE TO PARTNERSHIP: IMPACT OF THE ANTINOMIES 6.1. Pre-Russellian logicism. 6.2. Effect of the antinomies. 1 I. INTRODUCTION A long-standing tradition considers Georg CANTOR (1845-1918) as the 'creator' or founder of set theory.1 This view, correct when restricted to transfinite set theory, stands in need of revision within the context of a general account of the emergence of sets in mathematics. Such an account should face the broad questions: how did sets become important in the different branches of mathematics? how did they emerge as basic mathematical objects, as a fundamental language for mathematics?2 Simultaneously, one should try to establish how and why some mathematicians began to think of sets as the foundation for mathematics. In fact, within the context of the early history of sets during the second half of the nineteenth century, it is convenient to differentiate three aspects of our picture of set theory: the language of sets, the theory of sets, and the idea that sets form the foundation of mathematics. These aspects are so intertwined today that it may seem artificial to distinguish them, but in my opinion the distinction is useful, and even necessary, if we want to clarify the history of sets. During the second half of the nineteenth century, the interaction between the three aspects was quite complex. Along these lines, a discussion of the work of Bernhard RIEMANN (1826-66) and Richard DEDEKIND (1831-1916) forms the core of this paper. In my view, both mathematicians offer the most clear examples of the use of set language in mathematics, and the elaboration of foundational views based on the notion of set, prior to the emergence of CANTOR's work. This is why they should figure prominently in a general account of the emergence of sets. The triad RIEMANN/DEDEKIND/CANTOR constitutes a quite closed net offering basic clues for a satisfactory explanation of the ascent of sets. (In a detailed description, subsidiary authors, like GAUSS, DIRICHLET, WEIERSTRASS, and HANKEL, would also be of great importance.) RIEMANN, DEDEKIND, and CANTOR give the main examples of set-theoretical and set- 1 For early and authoritative instances, cf. the dedication of HAUSDORFF 1914, ZERMELO's words in the preface to CANTOR 1932, and several passages of HILBERT 1930. But see ZERMELO's crucial paper on the axiomatization of set theory, where he said that it had been "created by CANTOR and DEDEKIND" (ZERMELO 1908a, 200). 2 Around 1870, the language of sets was being applied in geometry, elementary arithmetic, algebra, and even complex function theory. During the last two decades, much has been done to consider these developments, at least for the cases of algebra and real analysis. For the emergence of point-sets within real analysis cf. HAWKINS 1975, DAUBEN 1979, COOKE 1992. DEDEKIND's structural view of algebra is studied in DUGAC 1976 and 1981, EDWARDS 1980, SCHARLAU 1981 and 1982, EDWARDS, NEUMANN & PURKERT 1982. A wealth of material on both topics can be found in MOORE 1982, and an attempt to synthesize the foregoing studies in FERREIRÓS 1993a. Viewed in the light of such previous work, the present paper offers additional material by showing that the notion of set had also geometrical and function-theoretical origins in the work of RIEMANN. 1 foundational conceptions within 19th century German mathematics. The mention of RIEMANN in this context is bound to cause some surprise, since it has escaped the attention of historians that he proposed a set-foundation for mathematics in his famous 1854 lecture on geometry. It will be my purpose to establish this point, but also to analyze the origins of this early proposal and its influence on other mathematicians. RIEMANN's influence on DEDEKIND is highly probable, since they were intimate friends from around 1855, i.e., from the time when RIEMANN had just developed his new conception. This influence was mainly on the level of general views─the nature of mathematical objects and questions of methodology─and it is related to a rather abstract conception of mathematics. DEDEKIND followed his own path through algebra and number theory on his way to set theory.3 As for traditional logic, it is constantly intertwined with our story. RIEMANN's general notion of a manifold was stated in implicit reference to common logical ideas of his time. Here lies the main difficulty for an interpretation of his definition of a manifold (as will be apparent later, when we consider how two important commentators have confronted RIEMANN's explanation). The difficulties encountered by commentators simply mirror the conceptual gap that separates traditional logic from contemporary mathematical logic. The dramatic changes which logic underwent in the period from 1850 to our time were mainly due to two processes: the emergence of first-order logic motivated by metalogical studies, and the set-theoretical antinomies. As we will see later on, it is the question of the antinomies which turns out to be crucial for our purposes. These changes in logical views are also the reason why DEDEKIND's logicism has never been taken seriously. This has not only affected the interpretation of the work of this important mathematician, but also our general understanding of the rise and fall of logicism. In my view, a correct interpretation of DEDEKIND's view will pave the way for a reassessment of the history of logicism as an influential nineteenth-century movement. The most important outcome, in this respect, is that the antinomies distorted the traditional view of logic in a way that, despite RUSSELL's efforts, made the logicist enterprise fail. The mention of logic and set theory in the nineteenth century brings to mind the British algebra of logic, since it was above all an algebra of classes or sets. This suggests the question whether the British mathematical logicians that worked during the 1840s-60s (particularly BOOLE) may have had an influence on RIEMANN or DEDEKIND. This question, however, will not be 3 The development of DEDEKIND's conception can be traced back to the late 1850s, as was shown by DUGAC in a pioneering and very useful book (DUGAC 1976, cf. also his 1981). But DEDEKIND's views still require further discussion, particularly as regards his systematic conception of the foundations of mathematics, and his logicism. Moreover, there are points of detail in which my account differs from DUGAC's, the most important being the evolution of DEDEKIND's notion of mapping. 1 discussed below, since I have found absolutely no evidence in favor of such a connection.4 Thus, it seems that the development of set theory was for the most part independent from that of logic until the 1890s. The similarities that can be found between them can be explained in terms of the same traditional conception of logic that we have to invoke in order to understand RIEMANN and DEDEKIND. This conception was widely held throughout the 19th century, and it should be considered a major source of the 'classical logic' of our century. To end this introduction, I would like to comment on questions which are mainly of a terminological character. The study of the origins of set theory in Germany is complicated by the fact that the German language did not suggest a privileged term for naming the notion of set. For example, both DEDEKIND and CANTOR accepted the French translation 'ensemble', while they used different German words: 'System' in the case of DEDEKIND, 'Mannigfaltigkeit' [manifold] or 'Menge' in that of CANTOR. Following CANTOR's usage in the 1890s, German has finally adopted the word 'Menge' to design sets; in the beginning, this word seems to have been considered somewhat unsatisfactory because it suggested a rather rough meaning, like 'mass'. It was also usual to speak about 'Inbegriffe', a compound from 'Begriff' [concept]; the reason for this will become clear in section II, but the fact is that apparently CANTOR and DEDEKIND were also dissatisfied with that word. Some other terms used during the 19th century are 'Klasse', 'Gesamtheit' [totality], 'Vielheit' [multiplicity], 'Gebiet' [domain], even 'Schaar' [host]. In my translations I will follow present-day usage, employing 'set' only as a translation of 'Menge', but the reader should keep in mind that other words, like 'system', 'class', 'collection', 'totality', or 'multitude', were no less appropriate for referring to the notion of set. Whether or not a particular word should be interpreted as denoting this notion is to be judged from the context. This terminological question becomes particularly critical when we come to the word 'Mannigfaltigkeit', on which the whole question of RIEMANN's place in the history of sets depends. In the context of his famous lecture on geometry, RIEMANN proposed his notion of a 'Mannigfaltigkeit' as the basis for topology and differential geometry; in this respect, the word has the same meaning as present-day 'manifold'. Surprisingly, we can also find the word 'Mannigfaltigkeit' in the work of CANTOR, where it undoubtedly means 'set'. I will defend the claim that there is a direct connection between both uses of the word: RIEMANN's 'manifolds' were nothing but sets, as is shown by his theory of 'discrete' and 'continuous' manifolds, and by his 4 Or, for that matter, in favor of connecting DEDEKIND or RIEMANN with the German and French authors who had contributed previously to the theory of classes (cf. KNEALE & KNEALE 1962, STYAZHKIN 1969). 1 definition of a manifold. Therefore, throughout this paper I will translate 'Mannigfaltigkeit' into 'manifold', but I will interpret this term as a synonym for 'set'. II. TRADITIONAL LOGIC, CONCEPTS, AND SETS The situation in logic around 1850 constitutes a crucial background for the purposes of this article, not only in connection with RIEMANN, but even more so for the case of DEDEKIND. Unfortunately, the attempt to identify the sources of their knowledge of logic confronts difficulties, since there is no direct evidence about it. Moreover, the German nineteenth century was surprisingly prolific in the production of logic treatises, and the views exposed in them were exceptionally varied (see UEBERWEG 1882, 47-79). It suffices to recall the influence exerted by HEGEL who, convinced that "everything rational is real, and everything real is rational", identified logic with metaphysics. In the case of RIEMANN, however, it is natural to assume that he followed the logical doctrines of the man whom he regarded as his master in philosophy: Johann Friedrich HERBART (1776-1841).5 Notably, HERBART was one of the most outstanding defenders of a strictly formal conception of logic against the abuses of the idealists.6 That is, HERBART championed the rejection of Hegelian logic and the return to Aristotelian syllogistics, understood as a purely formal doctrine. In fact, internal evidence from RIEMANN's and particularly DEDEKIND's writings suggests that they paid attention almost exclusively to logicians that followed the traditional Aristotelian doctrines. These logicians presented in their works a core of knowledge the roots of which lie in Antiquity, and which was also incorporated by all other non-idealist logicians (sometimes embedded into psychological, epistemological or metaphysical doctrines). The existence of such a core enables us to produce a schematic portrait of the basic logical doctrine of the time. This proves sufficient for our purposes and makes the question of exact sources irrelevant. 5 On RIEMANN and HERBART see the interesting paper SCHOLZ 1982a, although it does not consider HERBART's logical views, nor their connection with RIEMANN's manifolds, which concerns us here. 6 Such emphasis on the formal character of logic, and on its treatment as an autonomous discipline, comes from KANT (cf. UEBERWEG 1882, 47-51, especially on 48-49); nevertheless, scholars debate whether KANT really kept formal logic separate from the transcendental logic that he developed in KANT 1787. Be it as it may, HERBART rejected KANT's transcendental logic and went as far as possible in isolating the pure or formal content of logic (o.c., 51-53). In fact, UEBERWEG distinguishes a Kantian and a Herbartian school (o.c., v, 52). 1 However, it is worth mentioning that one of the most widely read logic treatises of the early nineteenth century was written by a Herbartian, the Leipzig professor of mathematics Moritz Wilhelm DROBISCH (1802-96). His work (DROBISCH 1836) saw four editions between 1836 and 1875, being "acknowledgedly" the best exposition of logic from the formal viewpoint (UEBERWEG 1882, 53). Therefore, it is perhaps the most natural candidate for having been read by RIEMANN and DEDEKIND. Interestingly, the first edition of DROBISCH's treatise included an appendix containing an attempt to establish a mathematical calculus of logic (see STYAZHKIN 1969). 2.1. The doctrine of concepts. In traditional logic, and even philosophy in general, concepts occupied a very central position, being considered as the nuclear constituents of knowledge. The origins of this conception are almost as old as philosophy: PLATO's dialogues used to discuss some series of related words in order to enable a contemplation of the general concept (idea) behind them. In the same spirit, ARISTOTLE considered concepts as the true object of knowledge, for "there is no science but of universals, as is clear from demonstrations and definitions" (Metaphysics, 1086b 32). Demonstrations and definitions were for ARISTOTLE the topic of the discipline for whose creation he is usually credited, namely logic. Below we will indeed see how the Aristotelian doctrine of syllogism shows the centrality of concepts to logic.7 It is important to realize that Occidental philosophy used to accept the primacy of concepts or ideas;8 HERBART and his disciple RIEMANN accepted it too. In 1837, HERBART wrote: "today [...] all sciences have long assumed the logical ordering, and so have recognized the dominion of general concepts in all matters of thought" (HERBART 1837, 68). Already in 1807, he had recommended that every mature science should be established around a central general concept (SCHOLZ 1982a, 424-426). Later on, RIEMANN would write: "natural science is the attempt to comprehend Nature through precise concepts" (RIEMANN 1892, 521), which implies that concepts are the essential building stones of human knowledge. 7 After having written this section, I noticed that the primacy of concepts in traditional logic was emphasized in CASSIRER 1910, 4. 8 Even the classical British empiricists considered ideas as the core elements of knowledge, although they emphasized their empirical origins. But we are considering the work of German mathematicians, and the German tradition was particularly close to the Greeks in this respect. Perhaps it is no coincidence that set theory was born in the land of rationalism, and precisely during the 19th century, when a close relation reigned between scientists and philosophers. 1 2.1.1. Syllogistic logic is based on that same centrality of concepts. As is well known, traditional logicians considered any reasoning to be reducible to a chain of syllogisms. The typical example of a syllogism is: 'Every A is B; and every B is C; therefore, every A is C'. Here we can observe how a syllogism consists of a structured connection of judgements, i.e., propositions affirmed or denied (like 'Every A is B'). Propositions, in their turn, are formed by means of the copula 'is', which connects a subject and a predicate. Finally, subject and predicate were considered to be nothing but concepts. Following this picture, the part devoted to pure logic in logic treatises of the 19th century was normally divided into three sections: 'On Concepts', 'On Judgements', and 'On Conclusions' (which include syllogisms). Reasoning was thus regarded as the result of different kinds of formal combinations of concepts, by means of particles like 'every', 'some', 'is', and 'not'. According to KANT and his followers, concepts contained the whole matter or content of an argument, everything else being purely formal or logical. It should be mentioned that the centrality of concepts was sometimes contested: some philosophers considered judgements as more fundamental pieces of knowledge, since they operate a combination or connection of concepts that they thought not reducible to the concepts themselves. For our purposes, however, this debate is of little import; what matters is that concepts, either alone or in combination, were the central pieces of reasoning. During the second half of the century, this became the basis for the view that the theory of classes is properly a logical theory. To understand this, however, we need to consider one further element belonging to the core of traditional logical doctrine: the distinction between the intension [Inhalt] and the extension [Umfang] of a concept. This was an inevitable component of the section 'On Concepts' in nineteenth-century logic treatises. 2.1.2. The origins of this distinction can be traced back to Antiquity, to the Isagogé of PORPHYRY, a third-century commentary on ARISTOTLE that was very influential during the Middle Ages.9 In modern times the customary reference was to the Port Royal logicians, ARNAULD and NICOLE, who wrote La Logique ou l'art de penser, one of the most widely read treatises of the seventeenth century (ARNAULD & NICOLE 1662). This is how they established the distinction between intension (or comprehension) and extension of concepts: In these universal ideas, it is very important to correctly distinguish the comprehension and the extension. By the comprehension of the idea we understand the attributes which it involves and which cannot be withdrawn without 9 Cf. FRISCH 1969, 108-114; WALTHER-KLAUS 1987, 9. 1 destroying the idea, as the comprehension of the idea of triangle involves extension, figure, three lines, three angles, equality of those angles summed up to two right angles, etc. By the extension of the idea we understand the subjects to which the idea applies, and which are also known as the inferiors of a general term which, in relation to them, is called superior; as the general idea of a triangle extends to all the different species of triangles. (ARNAULD & NICOLE 1662, 51) It turns out that the notion of extension made it possible to develop a formal analysis of the relations between concepts, i.e., of judgements and conclusions. In this way, ARNAULD and NICOLE were able to develop quasi-mathematically the theories of the conversion of judgements, and of the syllogistic figures of deduction─the core of Aristotelian syllogistics. For our purposes, it is important to remark that HERBART's expositions of logic followed exactly that early practice. A summary of HERBART's way of dealing with logic is given in HERBART 1808: this includes the distinction between intension and extension (p. 218) and the use of extensions for explaining the conversion of judgements (p. 220-221) and the syllogistic figures (p. 223-226). In 1837 he defined the intension of a concept as "the sum of its attributes"; the extension, as "the set [Menge] of the other concepts, in which the first appears as an attribute" (HERBART 1837, 71).10 In the same work, HERBART makes a nice application of these ideas to the notion of number: the natural numbers form the extension of the concept of number; moreover, nobody would know what number is without first knowing what one, two, three, four are. The intension of this concept [of number] consists therefore in its extension. [HERBART 1837, 73-74] 2.1.3. It may be doubted whether the idea of the extension of a concept, as handled by authors of the seventeenth to the early-nineteenth century, can be compared to the later notion of class employed by mathematical logicians, or to the mathematical notion of set. Among older logicians there was a tendency to concentrate upon the idea of hierarchies of concepts, analyzing the reciprocal relations genus-species (cf. even DROBISCH 1836). This turned intension and extension into reciprocal elements: at any point in the hierarchy of concepts, the intensional elements are found in that point and higher ones, while the extension is whatever lies behind. The advances made in the second half 10 This text shows a quite common trait of 19th-century German logic textbooks: there is no reference to classes of objects, since the elements that form the extension are also concepts. But this happened under the assumption that an object is a "bundle of properties" or concepts (HERBART 1964, vol.6, 199). On the variety of positions regarding this point, cf. PEIRCE 1931/60, vol.2, 242-246. 1 But second, the surface captured a good deal of information about its corresponding function, and RIEMANN was interested in ways of determining complex functions by means of a minimal set of properties. He was aware of the global dependence of analytic functions on their local behavior, and noticed how this meant that contemporary determinations of functions by means of global formulas involved redundant information (RIEMANN 1851, 38-39, 1857, 97). RIEMANN obtained minimal determinations by using a system of data which were partly analytic and partly geometrical: the geometrical information was contained in the Riemann surface, while the analytic part consisted in discrete facts, including the behavior of the function in singular points.18 In order to study the integrals of complex functions, he analyzed the Riemann surfaces from a topological viewpoint, developing new methods that enabled him to define the 'order of connectivity' of surfaces─essentially Euler's characteristic (RIEMANN 1851, 9-12, 1857, 91-96). And he showed how the topological invariants of surfaces are intimately connected to the properties of functions, an astounding example being the RIEMANN-ROCH theorem. In this way he opened a promising new area of mathematical research. Some documents which have been recently published by Erhard SCHOLZ enable us to trace the development of RIEMANN's ideas in the time between his doctoral thesis and his lecture of 1854 (SCHOLZ 1982). According to SCHOLZ's dating, in the period 1851-53 RIEMANN wrote down some manuscripts in which he dealt with the notion of a manifold and its relation to geometry, with n-dimensional continuous manifolds and n-dimensional topology. In my opinion, these manuscripts suggest that the notion of a manifold grew out of an attempt to find a satisfactory conceptualization of RIEMANN's new function-theoretical ideas. The main text that supports this conclusion is SCHOLZ's appendix 4 (SCHOLZ 1982, 228- 229); the purpose of this manuscript was to establish the relations between manifolds, geometry, and intuition. During this early period, RIEMANN explained manifolds in terms of what essentially 18 RIEMANN's tendency toward geometrical considerations in function theory is related to GAUSS more than any other mathematician. GAUSS's first (1799) and fourth (1849) proofs of the fundamental theorem of algebra were based on the geometrical representation of complex numbers (GAUSS 1863/1929, vol.3, 74 and 114). He regarded his argument as belonging essentially to "a higher domain of the abstract theory of magnitudes, independent from the spatial, whose object is the combinations of magnitudes linked by continuity", i.e., to topology (GAUSS 1863/1929, vol.3, 79). So he showed the interest and even necessity of a geometricotopological approach to complex functions. (From a modern viewpoint, the so-called fundamental theorem of algebra belongs to function theory.) Moreover, when he defended the geometrical representation of complex numbers, GAUSS spoke about these numbers as representing a "series of series" or "what comes to the same ... a manifold of two dimensions" (GAUSS 1863/1929, vol.2, 176; emphasis added). At the end of this paper, GAUSS mentioned manifolds of more than two dimensions, and in his lectures he came to speak about (n-k)-dimensional manifolds (SCHOLZ 1980, 16-17). All of this anticipates RIEMANN's reflections on manifolds, and thus it is only natural that RIEMANN referred to GAUSS as a source of his ideas (RIEMANN 1854, 255). 1 is the space of states of a physical substance or object: given a "variable object" [veränderlicher Gegenstand, Ding], the different "forms of determination" [Bestimmungsweisen] which the object can take form the points of the manifold (SCHOLZ 1982, 222). In the manuscript in question, RIEMANN illustrates this by taking the example of an experiment in which we measure both a temperature and a weight: the totality of possible results of the experiment forms a two-dimensional manifold. This idea involves no limitation of dimension, since we can also make experiments in which a higher number of variables has to be considered, and then we will find manifolds of more than three dimensions (SCHOLZ 1982, 228-229). Thus manifolds are abstract objects, and the notion of a manifold is independent of our geometrical intuitions. Furthermore, manifolds establish a frame within which it is possible to recover traditional geometry in an abstract way, without the least dependence on intuition (o.c., 229). Finally, this is employed to explain the use of manifolds in function theory: although in principle manifolds are abstract objects, in practice it is simpler and clearer to deal with them, not in an abstract way, but leaning on spatial intuition (ibid.). This is what RIEMANN had done with the surfaces, i.e., two-dimensional manifolds, that he considered in the context of function theory. If we follow this line of reasoning backwards, it offers a plausible reconstruction of the path followed by RIEMANN in his development of the notion of manifold. It would seem that he felt puzzled by his use of geometrical constructs (Riemann surfaces) in complex analysis. Riemann surfaces were not a part of traditional geometry, since they were not objects in three-dimensional space, and so their theoretical basis was unclear. Moreover, he might well have felt that the use of surfaces was not rigorous, since the contemporary tendency was to avoid resorting to geometry and intuition in the development of analysis.19 So he was led to the questions: What is the true foundation for higher-dimensional geometry and topology? Does that foundation explain the relations between geometry and functions encountered in complex analysis? The notion of manifold gave the answer to these questions, being an abstract concept on which both the theory of magnitudes and (abstract) geometry could be based.20 19 This was not only the influential approach of CAUCHY, but also the direction taken by GAUSS. (Both mathematicians had strongly influenced DIRICHLET, who was the main representative of rigorous analysis during this time; DIRICHLET was RIEMANN's professor in Berlin and advised him while working on his dissertation and Habilitation.) RIEMANN's approach to function theory seemed to contradict that tendency, and therefore posed a problem: Did analysis really depend on geometry? In the above mentioned appendix 4, RIEMANN was able to turn this question and show that, on the contrary, geometry depended on the abstract theory of magnitudes, which for him was the basis of analysis (it suffices to remember that he refers to function theory as "the theory of a variable, complex magnitude", cf. RIEMANN 1851). 20 In his manuscripts of the same period RIEMANN generalized his topological analysis of surfaces to the case of n-dimensional manifolds (cf. 'Fragment aus der Analysis Situs', in RIEMANN 1892, 479-482, and its dating in SCHOLZ 1982). This fact adds to the plausibility of the above reconstruction: if n-dimensional manifolds evolved as a theoretical basis for Riemann surfaces, it is only 1 Be that as it may, the fact is that RIEMANN was employing a notion of manifold by 1851- 53. His first definition of manifolds was essentially equivalent to the idea of the space of states of a physical substance, but in the 1854 lecture we find a different definition of manifolds, whose purport was to attain greater generality. Likewise, although in the lecture we find again a study of the interrelations between magnitudes, manifolds, and geometry, there is an important novelty: by 1854 RIEMANN had realized the richness of different metrics that can be given to one and the same topological substratum (manifold), and how this could be used to produce a clearer and richer picture of the fundamental notions of geometry. Here again, RIEMANN's geometrical ideas were rather abstract; it is well known that many of his contemporaries either found them difficult to accept or failed to see their implications. 3.2. RIEMANN's theory of manifolds. 3.2.1. Magnitudes and manifolds. A difficult point in RIEMANN's conception is the connection between manifolds and the theory of magnitudes. In searching for a foundation on which to base Riemann surfaces and their topological analysis, RIEMANN stuck to the traditional view of mathematics as the theory of magnitudes.21 Until the middle of the 19th century─and even today in some dictionaries─it was common to define mathematics as the science of magnitudes. This, of course, originated with the Greeks, who also distinguished two kinds of magnitudes: the discrete, and the continuous.22 This conception was devised so as to include arithmetic and geometry, the two historical roots of mathematics, under a common heading. RIEMANN preserved this general conception while mixing it with some new, abstract ideas, the main novelty being the notion of manifold, that he took as basic for the theory of magnitudes. The amount of this transformation, and the extent to which it involved a radical departure from the Greeks, is easily seen when we come to the idea of a topological analysis of manifolds. Meanwhile, RIEMANN's adherence to the traditional view also implies that his theory meant to embrace both geometry and arithmetic. As regards the relationship between magnitudes and manifolds, a careful reading of the natural that the topological methods employed for studying the latter were then applied to the former. 21 The early development of set-theoretical notions is intimately related with re-conceptualizations of the traditional theory of magnitudes. Examples of this are Hermann GRASSMANN (1809-77), RIEMANN, and WEIERSTRASS. The case of DEDEKIND, studied below, shows the motives that led mathematicians to divorce their new, abstract ideas from the old word 'magnitude', and to talk about sets properly. 22 ARISTOTLE included number among discrete magnitudes, and counted line, surface, body, time, and space as continuous magnitudes (Categories, 4b 20). 1 1854 lecture shows that RIEMANN was simply using both words as synonymous. For instance, the first section of his lecture, devoted to explaining the idea of an n-dimensional manifold, is entitled "Notion of a multiply expanded magnitude" (RIEMANN 1854, 255). This identification was possible because he─like GAUSS─conceived of magnitudes in a more abstract way than was customary: to him, quantitative relations were only one part of the general theory of magnitudes, the other being the topological study of magnitudes or manifolds (o.c., 256). Having said this, it is also important to emphasize the fact that RIEMANN conceived of his explanation of the notion of manifold as belonging to philosophy, rather than to mathematics: In trying to solve the first of these tasks, the development of the concept of a multiply expanded magnitude, I believe I can ask for an indulgent judgement all the more since I have little practice in such tasks of a philosophical nature, where the difficulties lie more in the concepts than in the constructions; and because, apart from very brief indications given by Gauss in [1831, 1849] and some philosophical investigations of Herbart, I have been unable to use any previous work. (RIEMANN 1854, 255) Given this context, it should certainly not be surprising that RIEMANN's definition of a manifold, given immediately below the preceding lines, was based on contemporary logical (i.e., philosophical) notions. The definition is contained in a short sentence: Notions of magnitude are only possible where a general concept is found which admits different forms of determination [Bestimmungsweisen]. According as a continuous transition does or not take place among those determinations [Bestimmungsweisen], from one to another, they form a continuous or discrete manifold [Mannigfaltigkeit]; in the first case the single determinations are called points, in the last one elements of that manifold. (RIEMANN 1854, 255) With section II in mind, it should be easy to interpret this explanation: we find a clear statement of the traditional concept/class relation. And in view of the evidence presented in that section, I regard it as proven that contemporary readers should have interpreted RIEMANN's manifolds as intending to be the classes of traditional logic.23 Nevertheless, the above definition has puzzled modern commentators, possibly because they were formed after the 'revolution of the antinomies', which above all affected logic, more than 23 Concerning the terminology, recall that German did not supply a privileged word for expressing the notion of set, as other languages did. Therefore, each author proposed a terminology of his own, making things complex. RIEMANN's terminology was accepted by CANTOR himself from 1878 to 1890. 1 mathematics or set theory itself (see below). BOURBAKI, for instance, chooses to simplify his reader's task, translating "element" where RIEMANN wrote "general concept" (BOURBAKI 1969, 176); in this way, BOURBAKI alters radically the original intention of the author. A greater misunderstanding can be found in SCHOLZ's interpretation that what RIEMANN called a manifold was the general concept itself (SCHOLZ 1980, 30). In order to further corroborate my own interpretation, I will mention that several other passages from RIEMANN's lecture can only be understood when we interpret his "manifolds" as being simply sets. This happens with an incidental comment, where he uses the term "manifold" to refer to what we would call the set of possible cases for the result of a given experiment.24 It is likely that the idea of connecting his manifolds with concepts, thus relying on logical conceptions, was suggested to RIEMANN by HERBART's work. In HERBART's writings it is easy to find comments on the relation between concepts and continua. An early example is his mention, in the course of a discussion of the right way to teach mathematics, of "the whole continuum that is contained under a general concept" (HERBART 1964, vol.1, 174; emphases in the original). In his theory of spatial concepts, HERBART used to give as examples the continua that fall under the concepts of tone and color (cf. SCHOLZ 1982a). In both cases HERBART's terminology and general conception are very close to RIEMANN's, although RIEMANN did not think that most ordinary concepts involve continuous, but discrete manifolds.25 3.2.2. Arithmetic, topology, and geometry. Compared to the approach of the manuscripts written in 1851-53, it seems clear that the 1854 definition was marked by a desire to generalize as much as possible, in order to include the whole of mathematics under the new conception. This is clearly shown by RIEMANN's comments on discrete manifolds, related of course to traditional views on magnitudes as the foundation of mathematics. Immediately after the definition of a manifold, RIEMANN mentions the fact that, for any given objects, cultivated languages always make it possible to find a concept under which they all fall. In this way, his definition gives us the highest degree of generalization, because any objects 24 Regarding the question to what extent are our hypotheses confirmed by experience, he says: "there is an essential difference between mere extensive [topological] relations and metric relations, in that among the former, where the possible cases form a discrete manifold, the dictates of experience, never completely sure, are not imprecise; while for the latter, where the possible cases form a continuous manifold, every determination from experience remains always imprecise──as high as may be its probability of being nearly exact" (RIEMANN 1854, 266). 25 Cf. SCHOLZ 1982a, 421-424, where HERBART's theory of the spatial is discussed, together with its similarities and important differences from RIEMANN's theory of manifolds. 1 whatsoever are intended to be mathematizable. The notion of number thus attains greater clarity, as RIEMANN suggests in a comment within brackets.26 And the sentence beginning RIEMANN's following paragraph confirms the relation between discrete manifolds and numbers: Determined parts of a manifold, distinguished by a characteristic or a boundary, are called quanta. Their comparison according to quantity occurs for discrete magnitudes by numbering, for continuous by measuring. (ibid.) The direct relation between discrete manifolds and numbers is clearly stated, and although there is a great distance between this and a set-theoretical foundation for the number system, it is nevertheless clear that the basic idea─sets as the basic objects of arithmetic─was suggested by RIEMANN. But, although he included considerations about discrete manifolds and their relation to numbers for the sake of completion, RIEMANN's main interest was in continuous manifolds, which formed the basis for his work on function theory, topology, and geometry.27 In a footnote at the end of the lecture RIEMANN indicated that the section we are discussing "constitutes also the preliminary work for contributions to analysis situs", i.e., topology (RIEMANN 1854, 268).28 According to RIEMANN, even if we cannot compare two continuous magnitudes by measuring, it is still possible to study them in an abstract, mathematical way: The investigations which can be carried out in this case constitute a general part of the theory of magnitudes, independent of metric determinations, where magnitudes are considered, not as existing independently of position nor as expressible by means of a unit, but as domains in a manifold. Such investigations have become a necessity for many branches of mathematics, particularly for the treatment of multi-valued analytic functions, and the dearth of them is one of the main reasons why the celebrated theorem of Abel and the achievements of Lagrange, Pfaff, and Jacobi in the general theory of differential equations have remained unfruitful for so long. (RIEMANN 1854, 256) 26 Since any given objects may constitute a single manifold, "in the theory of discrete magnitudes, mathematicians could therefore depart unhesitatingly from the postulate of considering given things as all of one kind" (RIEMANN 1854, 256). This is an obscure comment indeed, but in the light of the reference to discreteness and to several things being "all of one kind", I think we should read it as a discussion of the notion of unit. EUCLID had defined number as a 'collection of units', which was still a current definition around 1850. But mathematicians and philosophers of that time were puzzled by the fact that number-units should be simultaneously equal and unequal (cf. FREGE 1884). RIEMANN suggests that a formulation in terms of sets eliminates the problem: diverse objects can be regarded as equal insofar as they belong to one and the same manifold. 27 RIEMANN's assumption that continuous manifolds are always differentiable turned continuity into the only prerequisite of the differential geometry of manifolds. It is well known how WEIERSTRASS showed the extent to which this was erroneous. 28 On RIEMANN's topological ideas, see PONT 1974, SCHOLZ 1980. 1 Here we find the crucial point in which RIEMANN introduces the new program of topology within the apparently traditional framework of the theory of magnitudes.29 Furthermore, the relation between topology and function theory is clearly stated. The core of RIEMANN's lecture was the fact that a single topological basis (a continuous ndimensional manifold) may be given a great number of different metrics, and its consequences for the study of physical space. Therefore, the distinction between a topological and a metrical theory of manifolds was the basis of RIEMANN's geometrical conceptions. In fact, the origins of RIEMANN's ideas on differential geometry and physical space should be traced back to the moment when he realized the richness of different metrics which can apply to one and the same topological substratum (see SCHOLZ 1982). 3.3. RIEMANN's influence on the history of sets. RIEMANN's contribution gives us a partial answer to a question stated in the introduction: where does the approach to the foundations of mathematics in terms of sets come from? It seems that RIEMANN arrived at his new view by reflecting on the surfaces that he had introduced in function theory, which seemed to imply an intimate relation between complex analysis and geometry. His explanation of that phenomenon involved a revision of the classical notion of magnitude, and the idea that magnitudes constitute the foundation of mathematics and in particular of geometry. RIEMANN proposed a new view of these most basic mathematical objects: they were to be considered as manifolds, that is, classes associated to concepts. Both quantitative and topological relations emerged from the comparison of manifolds, with discrete manifolds giving rise to arithmetic, and continuous manifolds constituting the foundation of topology and geometry. As we see, RIEMANN understood the surfaces of his function theory as based on the notion of set, and the same is true for the n-dimensional manifolds that appear in his geometrical work. This amounts to the introduction of set language in both fields. But he also went one step further, since sets or manifolds turned out to be the basic objects of a new approach to magnitudes, and thus a foundation for mathematics. What we do not find in his work, however, is any development of an autonomous theory of sets, but the possibility remains open that his seminal ideas stimulated other authors to work on set theory. Since RIEMANN's conception was of a general and intuitive nature, we should expect it to have influenced further developments more on a general level than in the customary way of 29 His terminology is very similar to that of GAUSS, speaking of 'analysis situs' as an autonomous part of the general theory of magnitudes (see footnote 18). 1 particular mathematical results or techniques. The ideas that we have reviewed offered a new vision of mathematics and its foundations, but not particular methods or results that would be important later in the theory of sets or even of point-sets. Their effectiveness lay in their potential to suggest interesting ways of inquiry, or even a research program, thus leading to particular questions. Although a general influence of this kind is more difficult for the historian to assess, its importance should not be undervalued. As CANTOR wrote in 1868, "in mathematics the art of posing questions is more consequential than that of solving them" (CANTOR 1932, 31). 3.3.1. Diffusion; CANTOR's manifolds. RIEMANN's general vision of mathematics remained unknown outside Göttingen until 1868, when DEDEKIND published both of his Habilitation works in volume 13 of the Abhandlungen der Königlichen Gesellschaft der Wissenschaften zu Göttingen. During the 1850s and most of the 1860s, only close friends such as DEDEKIND could have known his speculations, discussing them with the author himself. The other paper that DEDEKIND edited in 1868 is also famous: it is RIEMANN's work on trigonometrical series, which included his definition of the integral. This and the paper on geometry appeared simultaneously, and they immediately caused a sensation in the German mathematical world. Notably, both played some part in the development of set theory. The impact of the paper on trigonometrical series can be judged from works by HANKEL, HEINE, CANTOR, and DU BOIS REYMOND that appeared two or three years later. RIEMANN's paper did not contain any direct contribution to the theory of point-sets, nor any reference to the notion of manifold. But his new definition of the integral opened the way for a systematic study of discontinuous functions, and therefore constituted a most important background for the study of point sets in connection with questions of real analysis, a study begun by the above mentioned authors (cf. HAWKINS 1975, DAUBEN 1979). As regards the lecture on geometry, its impact can be seen from the papers of HELMHOLTZ and BETTI, and from KLEIN's Erlanger Programm of 1872 (cf. SCHOLZ 1980). But it is not so easy to judge the impact of RIEMANN's reflections on the general notions of magnitude and manifold. HELMHOLTZ and KLEIN adopted the term 'manifold', but interpreted it in a rather restricted way, and they did not develop RIEMANN's considerations on the foundations of arithmetic and topology. Nevertheless, the possibility of an influence of this aspect of RIEMANN's lecture on authors related with the development of point-sets should not be overlooked. This is the more probable, since both papers were published together, and both were widely read. Among the names mentioned above, HANKEL and CANTOR are the authors for whom the 1 probability of such an influence seems higher.30 In the case of CANTOR, it is notable that from 1878 to 1890 he termed his field of study 'theory of manifolds' [Mannigfaltigkeitslehre]. And the first time it happened he established a direct relation to RIEMANN's 1854 lecture: it was in a paper of 1878 devoted to show that all continuous 'manifolds' have the same power or cardinal number. Here, CANTOR presented the matter in connection with the Riemannian notion of n-dimensional continuous manifold, and showed how RIEMANN's characterization of dimension was insufficient (see below). Afterwards, his decisive series of papers published from 1879 to 1884 was devoted to 'infinite, linear point-manifolds'; it gradually incorporated all the basic ideas of transfinite set theory. It has been suggested that CANTOR's use of the word 'manifold' could have come from WEIERSTRASS's lectures (Johnson 1979, 128-129), but an analysis of the use of this term in extant transcriptions of those lectures suggests that WEIERSTRASS never gave it the general meaning of set. WEIERSTRASS seems to have called certain subspaces of Rn 'manifolds', which is similar to GAUSS's use of the word, and also to what a superficial reading of RIEMANN's lecture might suggest.31 Such a usage would not allow for calling 'manifold' a set of points scattered in a line or space, which is the sense given to the word by CANTOR in his papers of the period 1879-84. On the other hand, RIEMANN's conception of manifolds as classes allows this kind of use.32 Moreover, it was RIEMANN, not GAUSS or WEIERSTRASS, who talked about manifolds in a systematic way, and in connection with arithmetic, analysis, and geometry. And it was to RIEMANN that CANTOR referred in the 1878 paper in which he first used the word 'manifold'. Notably, DEDEKIND himself understood CANTOR's terminology to be related to RIEMANN's work. In a letter of 1879, he proposed to replace the clumsy word 'Mannigfaltigkeit' by the shorter 'Gebiet' [domain], which, he said, is "also Riemannian" (CANTOR & DEDEKIND 1937, 47). Furthermore, in his 1888 book devoted to the set-theoretical foundations of natural numbers, DEDEKIND mentioned the word 'Mannigfaltigkeit' as a synonym for 'System', that is, set (DEDEKIND 1888, 344). So it seems clear that both DEDEKIND and CANTOR were aware of the 30 For Hermann HANKEL (1839-73), see his 1870. Apart from this early and interesting contribution, he left no other traces in the development of the theory of point-sets because of his early death. 31 Cf. WEIERSTRASS 1986 and 1988, also WEIERSTRASS 1894/1927, vol.7, 55-60, and PINCHERLE 1880, 234-237. 32 Although RIEMANN's association of manifolds with concepts might seem to set narrow limits on acceptable classes, it should be noted that "being a point at which a given function is discontinuous" would be considered as a concept by nineteenth-century logicians. It is in this sense that RIEMANN's definition is ampler than WEIERSTRASS's or GAUSS's use of the word. A difficulty still exists, though, when we consider that apparently no concept could be associated to an arbitrary set of points in the line. This might have been the reason why both DEDEKIND and CANTOR later tended to dilute or even abandon the concept/class connection (see below). 1 fact that RIEMANN's manifolds intended to be nothing but sets. A problem with ascribing a Riemannian ancestry to CANTOR's manifolds is that CANTOR failed to understand or to accept RIEMANN's vision of mathematics as based on manifolds. Until 1884, CANTOR seems not to have considered manifolds or sets as a foundation for mathematics; rather, he seems to be studying manifolds as the latest refinement of mathematics, making use of any previously developed tools (MEDVEDEV 1984). An example will clarify what I mean: if we are going to base analysis on the notion of set, we should develop the necessary parts of set theory without recourse to analysis itself; in this spirit, during the early 1880s STOLZ and HARNACK defined the (outer) content of point sets in such a way that the notion of content could be used to define the integral (see HAWKINS 1975). But when CANTOR presented his version of the same notion in 1884, he defined the outer content of a point-set essentially in terms of the integral of its closure (CANTOR 1932, 229-231), which of course precludes the use of his definition as a foundation for integration theory. This evidences the fact that he was not interested in or accustomed to consider sets as a foundation for mathematics.33 An interesting, if speculative, explanation for this deviation from RIEMANN's viewpoint can be obtained from the fact that CANTOR was trained under WEIERSTRASS. Like RIEMANN's theory of manifolds, WEIERSTRASS's rigorous theory of the number system was presented under the ambiguous language of magnitudes.34 This might have caused CANTOR to pass over RIEMANN's identification of magnitudes and manifolds, retaining a Weierstrassian view of magnitudes, together with a more Riemannian conception of manifolds. At least, that might be the implication of the following comment, written in 1882: Most of the difficulties of principle encountered in mathematics have their origin, it seems to me, in the ignorance of the possibility of a purely arithmetical theory of magnitudes and manifolds. (CANTOR 1882, 156) This text might imply that CANTOR considered magnitudes and manifolds as basically different kinds of objects, and furthermore as the two fundamental kinds of objects of (pure) mathematics. Such reading is coherent with his general attitude when studying set theory: he could rely on 33 Earlier examples of the use of analysis for the development of set theory are his first proof of the non-denumerability of the real numbers (1874), his proof of the equipollence of continua that differ in number of dimensions (1878), and some other theorems contained in the 1879-84 series of papers on point-manifolds (see CANTOR 1932, 153, 161-164). 34 In defining the real numbers, both WEIERSTRASS and CANTOR spoke about 'numerical magnitudes' [Zahlengrössen]; a rigorous theory of magnitudes was to be based upon the notion of natural number (see WEIERSTRASS 1986 and 1988, CANTOR 1882, 150). 1 that the number of substitutions is finite:50 therefore, their results will be equally valid for any domain of a finite number of elements, things, concepts θ, θ', θ''..., which from θ, θ' admit a composition θθ', defined arbitrarily but in such a way that θθ' is again a member of that domain, and that this kind of composition obeys the laws expressed in both fundamental theorems. In many parts of mathematics, but especially in number theory and algebra, we are continuously finding examples of this theory; the same methods of proof are valid here as there. (DEDEKIND 1981, 63; emphasis added) Some pages later, DEDEKIND applied this principle to the particular case given by the law of composition induced in the partition of a group by a normal subgroup (o.c., 68). In the passage just quoted, DEDEKIND used the word 'domain' to refer to sets, while in some other paragraphs of the lecture he used 'complex' with apparently the same meaning. Moreover, in working with composition laws on the classes which form the partition of a group, he was employing sets as concrete objects, submitting them to operations that are analogous to the traditional ones. The same can be found in other manuscripts of the time.51 DEDEKIND's abstract conception of groups is most notable and surprising when we take into account that the mathematical community began to adopt it around 1890. It should be noted that DEDEKIND's view of groups appeared after RIEMANN had formulated his abstract conception of manifolds: as we have seen, RIEMANN's definition ensured that manifolds could be formed by any elements whatsoever. Since DEDEKIND most probably knew about RIEMANN's ideas, the conclusion is that a real influence is acting here.52 We see how a methodological or even epistemological vision may explain the fact that a researcher anticipates, in some respects, by decades the general evolution of his subject. DEDEKIND's algebraic studies led him also to the general notion of mapping.53 This is obscured by the fact that he used a rather strange name for mappings, namely 'substitutions'. But the 50 This is correct, since the existence of neutral and inverse elements is a consequence of the two mentioned laws when we require the group to be finite. Modern axiomatization of groups began with an 1893 paper of H. WEBER, after LIE's work made new axioms for infinite groups necessary (WUSSING 1969, 223-251). 51 In 'Aus den Gruppen-Studien 1855-1858' (DEDEKIND 1930/32, vol.3, 439-445), and in a paper written in 1856, 'Abriss einer Theorie der höheren Kongruenzen in Bezug auf einen reellen Primzahl-Modulus' (DEDEKIND 1930/32, vol.1, 40-66). 52 This is further reinforced by the importance for DEDEKIND of RIEMANN's abstract methodological principles (see FERREIRÓS 1992). The relevant texts in this connection are a letter to LIPSCHITZ (in LIPSCHITZ 1986, 59-60), and DEDEKIND 1895, 54-55. 53 Explicit consideration of mappings was the most characteristic aspect of DEDEKIND's set theory as exposed in DEDEKIND 1888. CANTOR considered one-to-one correspondences from the beginnings of his set-theoretical work, and he dealt with orderisomorphisms since 1883, but he never considered mappings explicitly as theoretical objects, nor did he use a general notion of mapping, even after having read DEDEKIND's work. 1 definition presented at the very beginning of his 'Vorlesung über Algebra' makes it clear that 'substitutions' are not simple permutations, but more generally transformations of a 'complex' of elements into any other. And DEDEKIND makes the explicit restriction that, for the rest of the lecture, both complexes will be identical (DEDEKIND 1981, 60). If we disregard this restriction, it is clear that 'substitutions' are simply mappings. This fact is confirmed by a text written in 1879: It happens very frequently, in mathematics and other sciences, that when we find a system Ω of things or elements ω, each definite element ω is replaced by a definite element ω' which is made to correspond to it according to a certain law; we use to call such an act a substitution, and we say that by means of this substitution the element ω is transformed into the element ω', and also the system Ω is transformed into the system Ω' of the elements ω'. Terminology becomes somewhat more convenient if, as we shall do, one conceives of that substitution as a mapping [Abbildung] of the system Ω, and accordingly one calls ω' the image of ω, and also Ω' the image of Ω. (DEDEKIND 1879, 470)54 Internal reasons account for the fact that, in his algebraic researches, DEDEKIND only considered morphisms and not general mappings; the paragraph following our quotation introduces that restriction explicitly. More interesting, however, is how he came to consider mappings which were not bijective, and not even injective. This was a natural consequence of his studies of group homomorphisms, which appear in 'Aus den Gruppen-Studien 1855-1858'. Homomorphisms are not generally injective maps, and it is clear that DEDEKIND saw this point since he proved the theorem of isomorphism, considering the kernel of a group homomorphism.55 DEDEKIND's investigations of the late 1850s remained unpublished, and so his setconception of algebra remained unknown until the appearance of his ideal theory in 1871. But his algebraic work of the 1850s shows already a definite tendency towards employing sets as basic, concrete mathematical objects. RIEMANN's ideas may have suggested this, but there is no doubt that it was the real value of such language for particular algebraic problems, that convinced 54 A note to this text contains the first public announcement of DEDEKIND 1888. Concerning the notion of mapping [Abbildung], he says: "On this mental faculty of comparing a thing ω with a thing ω', relating ω with ω', or making ω' correspond to ω, without which it is not at all possible to think, rests also, as I shall try to prove in another place, the whole science of numbers." 'Abbildung' is even today the German word for mapping, but it would be tempting to translate it by 'representation', which suggests some connotations of the German word and makes it understandable that DEDEKIND could consider such an operation as a basic mental faculty (see below). 55 DEDEKIND 1930/32, vol.3, 440-442; here again, homomorphisms are called 'substitutions'. The relevance of general (noninjective) mappings was later reinforced by his discovery that the general theory of chains, including a generalized theorem of induction, does not require injective maps (DEDEKIND 1888). This can already be found in a manuscript written between 1872 and 1878 (see DUGAC 1976, 293-309). 1 DEDEKIND of its usefulness. Sets offered the possibility of dealing with mathematical questions in a way such that DEDEKIND's methodological preferences could be satisfied (on these methodological ideas, see FERREIRÓS 1992 and HAUBRICH 1992). DEDEKIND's use of set-language for dealing with groups was accompanied by an important novelty: in the context of Galois theory, he realized the usefulness of talking about fields, which during this period he called "rational domains" (cf. DEDEKIND 1981 and SCHARLAU's comments in his 1981).56 Again this amounted to the introduction of set-language, but it also involved an important new aspect: rational domains or fields, in DEDEKIND's sense, are always infinite sets. In this respect they differ from the finite 'complexes' or 'domains' that emerged in group theory (on the question of the infinite, see section V). DEDEKIND's familiarity with fields would later prove crucial in the context of algebraic number theory. It was in this context that the notion of field was first published, and just one single paragraph of DEDEKIND's 1871 exposition of ideal theory is enough to show the maturity he had attained in operating with set-structures through set-operations and mappings. Thus the following text, with which DEDEKIND's first exposition of ideal theory began, condensed all the achievements of the 1850s that we have reviewed: While we try to introduce the reader to these new ideas, we will place ourselves at a somewhat higher standpoint, and begin by introducing a notion which seems very appropriate to serve as a foundation for higher algebra and those parts of number theory connected with it. 1. We will understand by a field [Körper] every system of infinitely many real or complex numbers, which is so closed and complete in itself, that addition, substraction, multiplication, and division of any two of those numbers yields always a number of the same system. The simplest field is constituted by all rational numbers, the greatest field by all [complex] numbers. We call a field A divisor of field M, and this a multiple of that, if all the numbers contained in A are also found in M; it is easily seen that the field of rational numbers is a divisor of all other fields. The collection of all numbers simultaneously contained in two fields A, B constitutes again a field D, which may be called the greatest common divisor of both fields A, B, for it is evident that any divisor common to A and B is necessarily a divisor of D; similarly, there always exists a field M which may be called the least common multiple of A and B, for it is a divisor of all other common multiples of both fields. Moreover, if to any number a in the field A, there corresponds a number b=φ(a), in such a way that φ(a+a') = φ(a)+φ(a'), and φ(aa') = φ(a)φ(a'), the numbers b constitute also (if not all of them are zero) a field B=φ(A), which is 56 It is noteworthy that, many years later, DEDEKIND would recommend to CANTOR the word 'domain' as an "also Riemannian" substitute for 'manifold' (CANTOR & DEDEKIND 1937, 47). 1 conjugate to A and results from A through the substitution φ; inversely, in this case A=θ(B) is also a conjugate of B. Two fields conjugate to a third are also conjugates of each other, and every field is a conjugate of itself. (DEDEKIND 1871, 223- 224) This passage should be considered as decisive in the history of sets: it is extremely rich, incorporating as it does all crucial ideas related to the notions of set and mapping as used in algebra (cf. DUGAC 1976, 29). There is no doubt that it had to be difficult to understand for any reader of the time, since they were completely unaccustomed to any use of the notion of set in algebra. But for this very reason, it had to attract strongly the attention of readers to this new vision─whether to accept it or refuse it. DEDEKIND's terminology might seem strange, but from his definitions it is clear that 'divisor' and 'multiple' mean the two sides of an inclusion; 'greatest common divisor' designs the intersection; and 'least common multiple' refers to the union, in the sense of the smallest structure that contains both. The selection of these expressions is easy to understand if we consider that DEDEKIND was dealing with algebraic number theory. In the case of Z, the inclusion of principal ideals corresponds strictly to the divisibility of their generators; this led him to employ the analogy between inclusion and division throughout the supplement on ideal theory. Moreover, his intention was to establish terminology in such a way that theorems could be formulated with exactly the same wording as in elementary number theory.57 There exists evidence that DEDEKIND's terminology was understood by at least one outstanding mathematician as related to the general notion of set. CANTOR used DEDEKIND's terminology for inclusion, union and intersection of fields in his decisive series of papers 'Über unendliche, lineare Punktmannigfaltigkeiten', from 1880 to 1884.58 This is particularly striking because that terminology is rather inappropriate in a general set-theoretical setting such as the one to which CANTOR applied it─DEDEKIND himself replaced it in such a context (see below). In the text on fields, DEDEKIND's treatment of 'substitutions' or mappings, and particularly those which transfer structure,59 was equally a model. The fact that he is considering morphisms 57 For instance, the sense of 'divisor' and 'multiple', as defined for modules, are contrary to their counterparts for fields; the reason: to obtain the common wording for theorems. Regarding arbitrariness of terminology, see a letter to LIPSCHITZ from 1876 (LIPSCHITZ 1986, 79), another to KEFERSTEIN from 1890 (SINACEUR 1974, 274), and some passages of Was sind und was sollen die Zahlen? (DEDEKIND 1888, 360, 377-378, 348). 58 The terminology is introduced in the second part of the series (1880; CANTOR 1932, 145-146); the latest instances are found in the sixth part (1884; ibid. 214, 226, 228, etc.). 59 In later versions of ideal theory, DEDEKIND emphasized the need for restricting the general notion of mapping or 'substitution' in a way characteristic of algebra, by considering only mappings which preserve structure (see DEDEKIND 1879, 470-471; 1893, 456- 1 which may not be injective, is made clear by the comment on zero images: when he says that the numbers b constitute a field "if not all of them are zero", it means that his notion of mapping allows for the trivial case in which all elements of the original have 0 as their image. (We have seen that non-injective maps were already present in his 1856-58 'Gruppen-Studien'.) Moreover, the clarity with which the reflexive, symmetric and transitive properties of field 'conjugation' are set forth in the text deserves attention. The example we have considered here is only that of fields, discussed by DEDEKIND in the first paragraph of his 1871 ideal theory, but the whole exposition insisted on the importance of setstructures for algebraic number theory. The very viewpoint adopted involved a constant exercise in the translation of problems stated in terms of numbers to new and more abstract set-formulations: DEDEKIND replaced KUMMER's ideal numbers with the kind of sets of numbers that he called 'ideals' (cf. section 4.2). 4.1.2. Sets and maps in a general setting. The year after the publication of this first version of ideal theory, DEDEKIND began to write the draft for his later book Was sind und was sollen die Zahlen?. The 1872 draft begins with definitions of some set operations (DUGAC 1976, 293-294); he presented them in a general setting, abstracted from the structural restrictions that are necessary in an algebraic context. The word 'system' was defined in a way that undeniably pointed to the extensional notion of set, and the most characteristic part of DEDEKIND's algebraic terminology changed in accordance with the new abstract framework: the word 'divisor' was replaced by 'part' [Theil], meaning subset. Meanwhile, he maintained the expression 'least common multiple' for the union set, although he also wrote 'compound system', which was to become his final choice. Nevertheless, set operations were by no means his primary interest. They were completely clear in themselves, and DEDEKIND had long been accustomed to them. The whole draft was primarily devoted to a study of the notions of mapping and chain, the second being based on the first. The term 'mapping' [Abbildung] appears also at the very beginning of the draft (DUGAC 1976, 294), and is defined in its general sense. We will return to this essential part of Was sind und was sollen die Zahlen? later on. The translation of set-structural operations to a general set-theoretical framework, which DEDEKIND sketched in his 1872 draft, reveals clearly what I have already stated: that DEDEKIND's mature set theory has its roots in his algebraic ideas, that emerged during the late 1850s. 457). 1 It would be interesting to know about DEDEKIND's motivation for writing the 1872-78 draft. It probably was one or two of the following factors. In 1872, he published his theory of the real number system after learning for the first time about two other theories, those of WEIERSTRASS and CANTOR. The knowledge that some mathematicians had developed ideas similar to his own might have motivated him to publish his more general reflections on the foundations of the natural numbers, and arithmetic as a whole, since they established the general framework for his theory of the real numbers. CANTOR had also developed the notion of derived set (see CANTOR 1872), which perhaps was even closer to DEDEKIND's viewpoint, since it used the notion of set. On the other hand, it was during the summer of 1872 that DEDEKIND made CANTOR's acquaintance, although unfortunately there is no record of the content of their conversations. One is left to speculate about the possibility that they discussed such matters as the (set-)foundations of the number system, and the Riemannian theory of manifolds.60 4.1.3. Critique of magnitudes. A striking difference between DEDEKIND and RIEMANN is the complete absence of the notion of magnitude in the former's views. While RIEMANN's conception of mathematics and manifolds was still rooted in the traditional idea of magnitude, DEDEKIND's mathematics are based on numbers, sets, and mappings. DEDEKIND was perhaps the first mathematician who consciously avoided any kind of reliance on the traditional notion of magnitude. His reasons are clearly formulated in passages of his works on numbers: "I demand that arithmetic shall be developed out of itself" (DEDEKIND 1872, 321), that is, "without any admixture of foreign ideas (such as that of measurable magnitudes)" (DEDEKIND 1888, 338).61 DEDEKIND's thoughts on the matter are spelled out in his correspondence with LIPSCHITZ of 1876 (LIPSCHITZ 1986; partly in DEDEKIND 1930/32, vol.3, 469-479). Here it becomes clear that the problem with the theory of magnitudes was a lack of rigor. The notion of magnitude was ambiguous: there had been talk of 'named' and 'pure' magnitudes, depending on whether the magnitudes belonged to the real world of natural science, or were of a purely mathematical kind; it had even been customary to speak of 'variable magnitudes'. But magnitudes had never been rigorously defined, and so the theory was dependent on implicit, unwarranted assumptions. This affected above all the question of the continuity of the real number system. No author, from Euclid to the nineteenth century, had ever established a postulate that implied the 60 Concerning this and other episodes in the relations between CANTOR and DEDEKIND, see FERREIRÓS 1993. 61 More generally, cf. DEDEKIND 1872, 321-322; 1876/77, 269; 1888, 338-339. 1 continuity of the set of ratios between homogeneous magnitudes.62 Instead, they had relied on the assumed existence of incommensurable magnitudes (DEDEKIND 1930/32, vol.3, 476-477), making arithmetic depend on foreign matters. Previously it was thought that the continuity of the number system is imposed by the characteristics of physical or geometrical magnitudes. In DEDEKIND's eyes, this made arithmetic lose its autonomy, and even its a priori, logical character. (The changes in geometrical thought, and above all RIEMANN's contribution, supported this interpretation, since geometry was no longer an a priori theory.) Breaking with such tradition, DEDEKIND emphasized the fact that numerical continuity was abstractly definable, and that only such abstract definition allows us to postulate the continuity of a domain of magnitudes, or of geometrical space (cf. also CANTOR 1872).63 An autonomous development of the number system, "without any reference to the rather obscure and complex concept of magnitude" (DEDEKIND 1930/32, vol.3, 476), became the antidote to those diseases. Apparently the mathematical community agreed that the measure was appropriate: by the beginning of the twentieth century, despite the frequent talk of magnitudes in the writings of influential mathematicians such as WEIERSTRASS and his school, the word 'magnitude' had basically disappeared from mathematical language. 4.2. The heuristic way towards ideals. One further aspect of DEDEKIND's ideal theory is directly relevant to our subject. As we have seen in section II, during the 19th century sets were primarily conceived as logical classes correlated with concepts. It turns out that DEDEKIND's notion of ideal emerged from considerations in which the traditional concept/set relation was central. This is at least what the introduction to a paper of 1876/77, 'Sur la théorie des nombres entiers algébriques', suggests. In it DEDEKIND described the path which led him to formulate the whole theory in terms of ideals; it was meant to be a historical description of the process of his own reflections, in which he "wrote each word only after the most careful reflection" (LIPSCHITZ 1986, 59). We can be certain that any such account involves a measure of rationalization, but DEDEKIND's description is interesting as it shows his implicit assumptions, including underlying conceptions related to the notion of set, that he considered to have been central to the formulation of 62 Cf. DEDEKIND 1930/32, vol.3, 472-474, where DEDEKIND emphasizes this problem and outlines a rigorous development of the theory of magnitudes. 63 As we have seen, the idea that continuity was really a postulate, and not a fact, was central to RIEMANN's geometry (RIEMANN 1854). 1 the notion of ideal. Around 1860, DEDEKIND had tried to produce a completely general theory of the factorization of algebraic integers by generalizing KUMMER's approach in a rather direct way, through higher congruences. This approach turned out to be unsatisfactory, because there always appeared exceptions, numbers whose decompositions could not be obtained by means of higher congruences.64 DEDEKIND let the problem sleep for some years, until around 1869, when he began work on the second edition of Vorlesungen über Zahlentheorie. The result of his renewed efforts was a completely new viewpoint: the factorization problem was formulated in set-structural terms. DEDEKIND emphasized the importance of this change: I have not arrived at a general theory without exceptions [...] until having abandoned completely the old, more formal approach, and having replaced it by another which departs from the simplest basic conception, and fixes the eyes directly on the end. Within this approach, I do not have any more need of new creations, as that of Kummer's ideal number, and it is entirely sufficient to consider the system of really existing numbers which I call an ideal. The power of this notion resting on its extreme simplicity, and being my wish to inspire confidence in this concept, I shall try to develop the series of ideas which led me to this notion. (DEDEKIND 1876/77, 268) KUMMER had not defined ideal numbers in themselves; he just defined divisibility by ideal numbers. That is, in KUMMER's restricted theory, and in DEDEKIND's first approach, ideal numbers were introduced through one or more congruence relations, which determined when an algebraic integer was to be considered as divisible by the ideal number (cf. EDWARDS 1980). It is a small step to consider the congruences associated with each ideal number as its defining concept or property, which may or may not be satisfied by an integer; DEDEKIND took this step. DEDEKIND's aim was to produce a general definition of ideals, making a precise definition of their multiplication possible. To this end, it would be necessary and sufficient to determine what is common to all those properties A, B, C... which are used to define the different ideal numbers, and to establish how properties A, B associated with two ideal numbers determine the property C which defines their product (DEDEKIND 1876/77, 268-269). At this point, the relation between concept and class came into play, since the stated task is "essentially simplified" when we consider, not the property A itself, but the class, set, or system which it determines: As such a characteristic property A serves to define, not the ideal number in itself, but only the divisibility of numbers 64 See DEDEKIND 1930/32, 202-230, and HAUBRICH 1988 and 1992. 1 contained in [the ring of integers] o by an ideal number, one is naturally led to consider the system of all those numbers α in domain o which are divisible by a certain ideal number; from now on I shall call such a system a, for short, an ideal, and so to each particular ideal number corresponds a certain ideal a. But as, reciprocally, the property A [...] consists only in α's belonging to the corresponding ideal a, one may, instead of properties A, B, C... [...], consider the corresponding ideals a, b, c... in order to establish their common and exclusive character. (DEDEKIND 1876/77, 270) This step towards set language, which DEDEKIND regarded as 'natural', was difficult and strange for his contemporaries.65 What made it natural for DEDEKIND were, undoubtedly, two factors: his familiarity with the traditional logical conceptions, that established the concept/set relation; but above all his confidence in the set-theoretical approach to arithmetic and mathematics generally. Without the latter, the connection between those traditional conceptions and mathematics would not be clear at all; this, and the changes in method implied by DEDEKIND's step, explain why the innovation was difficult to accept. The next task was to establish a definition of ideals, i.e., to determine necessary and sufficient conditions for a set of integers to be an ideal. Through a consideration of principal ideals (sets of integers divisible by a given, 'existing' integer), DEDEKIND found two conditions apt for a general definition. An ideal was defined to be a system or set of algebraic integers in a field K, such that (DEDEKIND 1871, 251; 1876/77, 271): I. the sums and differences of any two numbers in the ideal are again numbers belonging to it, and II. any product of a number in the ideal by an integer of the field is again a number belonging to the ideal. Finally, DEDEKIND took pains to prove that any so-defined ideal was either the set of multiples of an integer, or of an ideal number in the old sense. This train of thought led him "naturally" to base the whole theory of algebraic integers on "this simple definition, entirely freed from any obscurity and the admission of ideal numbers" (DEDEKIND 1876/77, 271-272). DEDEKIND had a rather strict conception of arithmetic and how to treat it, including the requirement that definitions should be appropriate to base upon them a rigorously deductive theory. It is clear that the definition of ideals satisfied his preferences completely, and this seems to have been the reason why he adhered to it, despite its novelty and the 65 This is one of the reasons why ideal theory was not generally accepted until the 1890s (see HAUBRICH 1992). 1 difficulties it posed to the understanding of contemporary readers. 4.3. The logical theory of systems and mappings. The foregoing may suffice for the reader to obtain a grasp of the sense in which sets were at the basis of ideal theory. Ideal theory was genetical or "constructive" in DEDEKIND's sense: taking the set of algebraic numbers A as given, it employed set-constructions on it, i.e., subsets of A (the ideals), as its basic objects; and this made it possible to define rigorously the operations on ideals, especially multiplication. In this way, it was perfectly coherent with DEDEKIND's conception of the number system, to which we now have to turn. 4.3.1. DEDEKIND's program for the foundations of mathematics. In his works on the foundations of the number system, DEDEKIND expressed repeatedly the idea that this system was obtained from the natural numbers through step-by-step definitions or "constructions" (DEDEKIND 1854, 430-431; 1872, 317-318; 1888, 338). In manuscripts that are still preserved, he developed the well-known idea of defining the integers and their operations on the basis of equivalence classes of pairs of natural numbers; and similarly for the rationals on the basis of equivalence classes of pairs of integers. In 1872 he published the much more sophisticated idea of employing so-called 'DEDEKIND cuts' on the set of rational numbers for defining the real numbers and the operations on them. In all of these cases, the procedure is genetical in the above sense: taking a set of numbers as given, the next higher set is defined by means of set-constructions on the former, i.e., it is defined as (isomorphic to) the set of some specified subsets of the former. This, and the operations on the 'lower' number-set, suffice to define the operations on the new numbers. Since, according to DEDEKIND's reliable dating, his theory of real numbers was formulated in 1858, it is a mild assumption that the much simpler theories of the integers and rationals should be traced back, at least, to that same year.66 According to the foregoing, the arithmetical framework that embraces both the number system and ideal theory is that of a series of set-constructions which have the natural numbers as their basis. Natural numbers with their operations on the one hand, and sets and mappings on the other, are the basic elements of arithmetic. At some point in time, however, DEDEKIND was able to simplify this conception essentially; he discovered that the natural numbers and their basic operations could be defined by means of sets and maps alone. Thus sets and mappings alone are a sufficient foundation for arithmetic. This is the main thought that finds its detailed expression in the 66 Here I skip over the details of the evolution of DEDEKIND's views on the matter, which I plan to discuss in a new article. 1 opinion, this peculiarity is intimately related to DEDEKIND's logicist convictions, and to his conception of logic. The traditional conception of axioms regarded them as true propositions that do not admit of a proof. To the influential Kantian epistemology, axioms were "synthetic a priori principles, insofar as they are immediately true" (KANT 1787, 760). Since logic was a purely analytical science, in which no synthetic principle plays a role, it was radically foreign to the use of axioms. This Kantian conception can still be found in the work of the German mathematical logicians SCHRÖDER (cf. 1890/1905, vol.1, 441) and FREGE. It was for that reason that FREGE did not talk about arithmetical 'axioms', but about the 'fundamental laws' [Grundgesetze] of arithmetic (FREGE 1893/1903). Similarly, DEDEKIND seems to have thought that the logicist program demanded the development of arithmetic in a rigorously deductive way, without any recourse to axioms. The idea that a deductive theory can be based on definitions alone was not completely unknown in this period. In SCHRÖDER's first logical work, devoted to the Boolean calculus, we can read the following: All theorems of our discipline are intuitive; as soon as they are brought to conscience, they appear as immediately evident. Therefore, the statements that we introduce here as axioms could also justifiably be presented as consequences that are immediately given with the definitions. (SCHRÖDER 1877, 4) SCHRÖDER preferred to base the logical calculus on axioms, perhaps due to the influence of Hermann GRASSMANN, while DEDEKIND's was the opposite choice. DEDEKIND made a conscious effort to derive all immediate consequences of his definitions that would be needed later on. In a word, it seems that for DEDEKIND a theory can only be judged strictly logical when its propositions follow from basic logical notions, like those of set and map, without the use of any axioms. In this way, DEDEKIND's theory can be considered as strictly deductive but nonaxiomatic─definitional instead. From a modern viewpoint, this has to be considered a drawback, since the axiomatic treatment allows better control of the theory. V. INTERLUDE: THE QUESTION OF THE INFINITE It has frequently been pointed out that, until the time when CANTOR championed the actual infinite, it had been traditionally rejected by all mathematicians─a tradition that is traced back to ARISTOTLE, and which counts GAUSS and CAUCHY as 19th century examples. In this section, 1 however, we will see that this generalization does not hold water against the views held by RIEMANN and DEDEKIND from the 1850s. In fact, when we consider more closely the situation in 19th century Germany, there are good reasons to reject that generalization. At the beginning of the century, in the time of idealism, the potential infinite of mathematics was called the "bad infinite" by HEGEL and his followers (see BOLZANO 1851, 7). The implication was clear: there is a good infinite, that is actual in the highest sense: the Absolute. It is well-known that idealism affected an important sector of German scientists, and among them we find some mathematicians; the best example I know is that of the Berlin professor Georg STEINER (see his 1832). But idealism was not all in German philosophy, nor the only trend that affected scientists. An important characteristic of German philosophy and science during the 19th century was the influence of LEIBNIZ, and LEIBNIZ was an outstanding defender of the actual infinite.78 It is not by chance that both BOLZANO (1851 frontispiece) and CANTOR (1883, 179) quoted the following text by LEIBNIZ: I am so much for the actual infinite, that instead of admitting that Nature abhors it, as is vulgarly said, I defend that it affects her everywhere, in order to better mark the perfections of her Author. And so I believe there is no part of matter which is not, I do not say divisible, but actually divided; and consequently the least particle must be considered as a world full of an infinity of different creatures. Even HERBART, who in his later works argued against the actual infinite (cf. CANTOR 1932, 392-393), talked early in the century about the service that metaphysics had made to mathematics by eliminating the aversion of the notion of infinity.79 In the group of Leibnizians we can also count the Königsberg professor of mathematics Johann SCHULZ, who was a friend of KANT's and developed a mathematical theory of the infinitely great (see SCHUBRING 1982). And we should also consider here scientists (not mathematicians) that under LEIBNIZ's influence presented revised versions of the theory of monads, defending the idea that the number of monads was actually infinite. The most important examples are the scientist-philosophers Gustav Theodor FECHNER (see his 1864) and Hermann 78 At least in some of his writings, notably in the influential Monadologie (LEIBNIZ 1714), though not in some other works, such as the Nouveaux Essais. On this question, cf. KÖNIG 1990. 79 HERBART 1964, vol.1, 174. Such aversion had impulsed mathematicians to teach "in strange ways, without that fundamental concept, that which only through it was accessible to the discoverer himself" (probably meaning LEIBNIZ and the calculus). 1 LOTZE (cf. 1856/64). These examples taken from 19th century mathematicians and scientists suffice to show that the customary generalization on the rejection of the actual infinite is far from true. The existence in the 1870s of a group of theologians (GUTBERLET, FRANZELIN) that accepted the actual infinite, and that were important for CANTOR, points to the same (cf. MESCHKOWSKI 1967, PURKERT & ILGAUDS 1987). But for our purposes, the most important cases are those of RIEMANN and DEDEKIND. Since the acceptance of the actual infinite could be proposed as a criterion for a mature, mathematically interesting set theory, the question treated here might seem crucial for the purposes of this article. 5.1. A reconstruction of RIEMANN's viewpoint. Like LOTZE and FECHNER later, and like his philosophical master HERBART, RIEMANN accepted a version of LEIBNIZ's monadology: the ontology of the 'Realen' that HERBART had elaborated (see SCHOLZ 1982). This is strong evidence for a positive position towards the infinite on his side. On this point, however, RIEMANN's mathematical texts are not so clear as those of DEDEKIND that we will see below. In speaking of continuous manifolds as sets of points, RIEMANN was implicitly accepting completed infinities, but this does not go beyond what was traditionally common in geometry. More interesting information on his views is to be found in one of his philosophical texts, called 'Antinomien', and in a passage of his 1854 lecture. Beginning with the latter, in a surprising passage RIEMANN assumes the existence of infinite-dimensional spaces, giving two examples: the different possible functions on a given domain, and the possible configurations of a spatial figure. This might imply an acceptance of the actual infinite, although the evidence is not unequivocal. Incidentally, the text seems to show that RIEMANN did not identify the discrete and the finite: he distinguishes "an infinite series" from "a continuous manifold" (RIEMANN 1854, 258), probably because infinite series are discrete but infinite manifolds. RIEMANN's acceptance of the infinite is proven, I think, by his philosophical text 'Antinomien'. In it, he presents several pairs of contradictory ideas as thesis and antithesis, no doubt following the famous example of KANT's Kritik der reinen Vernunft. RIEMANN's theses are presented under the common heading "finite, representable", and the antitheses under "infinite, conceptual systems which lie on the borders of the representable" (RIEMANN 1892, 518-519). And a general comment on the relation between thesis and antithesis enables us to reconstruct his position. RIEMANN indicates that under the latter we find "concepts which are well determined by 1 means of negative predicates, but are not positively representable" (ibid.). Thus the notion of the infinite is well defined, and therefore seems to be completely acceptable.80 5.2. DEDEKIND and the infinite. DEDEKIND's position was much more explicit, since he formulated a definition of the infinite, and based on it his whole theory of sets and natural numbers. This definition was first published in 1888, after CANTOR's pathbreaking contributions, but it can be found at the very beginning of DEDEKIND's 1872-78 draft of his later book (see DUGAC 1976, 294). According to this, it has to be dated to 1872, and therefore it precedes CANTOR's papers on set theory, including his crucial non-denumerability result obtained late in 1873. Unfortunately, as far as I know there exist no other indications regarding the origins of DEDEKIND's definition.81 There is much independent evidence for DEDEKIND's acceptance of the actual infinite at this time: both his 1871 ideal theory, and his 1872 theory of the real numbers, explicitly employ infinite sets. But since the theory of real numbers was first formulated in 1858, it might seem that DEDEKIND was accepting the actual infinite already in the 1850s. This is corroborated by some others texts, and again it is coherent with the plausible influence of RIEMANN's views on him. 5.2.1. The roots of DEDEKIND's infinitism. The dependence of the theory of real numbers on infinite sets has, in fact, been proposed as the reason why DEDEKIND took the side of the actual infinite, and why he formulated his ideal theory in an infinitistic version.82 This, however, does not seem to be very plausible. First, DEDEKIND's acceptance of the infinite does not have the appearance of a more or less ad hoc position, adopted to safeguard his theory of the real numbers. It rather looks like a deep-rooted conviction: infinite sets are perfectly acceptable objects of thought, that involve no contradiction, and that play a crucial role in mathematics; this is why DEDEKIND took the side of the actual infinite. Second, DEDEKIND placed no special value on his theory of real numbers: he regarded it as a rather straightforward contribution that simply came to fill a gap in the elements of the number 80 It was precisely RIEMANN's friend, DEDEKIND, who showed how we can define directly the infinite, leaving the finite as the non-infinite. Judging from the sentence above, this should have greatly surprised RIEMANN. 81 Since in the 1872-78 draft it is formulated from the start in its final form, the possibility remains open that the definition might have been obtained some years earlier. 82 EDWARDS 1983, 16-18. It has to be said that EDWARDS is a convinced constructivist, and this viewpoint, which he has never hidden, undoubtedly influences his perception of the historical issues at stake. 1 system; other mathematicians would have formulated something similar had they devoted some effort to it (DEDEKIND 1930/32, vol.3, 470, 475). Therefore, it is highly implausible that he might have conditioned his cherished ideal theory on a trait of the much less valuable theory of real numbers. Third, it is even possible to find texts written by DEDEKIND prior to 1858, i.e., prior to the formulation of his theory of the real numbers, in which he employs infinite sets. This is what happens in a paper written in 1856 and published the following year, where DEDEKIND presented a theory of higher congruences.83 We saw above that around this time DEDEKIND was talking about finite sets in the context of group theory, and he also introduced (infinite) fields under the name of 'rational domains'. The paper on higher congruences gives independent corroboration of his use of sets, but above all it is interesting for it shows that he accepted infinite sets and equated them with the most concrete objects of traditional mathematics. The relevant text says: The preceding theorems correspond exactly to those of number divisibility, in the sense that the whole system of infinitely many functions of a variable, congruent with each other modulo p, behaves here as a single concrete number in number theory, for each function of that system substitutes completely any other in any respect; such a function is the representative of the whole class; each class possesses a definite degree, certain divisors, etc. and all those characteristics correspond in the same manner to each particular member of the class. The system of infinitely many incongruent classes─infinitely many, since the degree may grow indefinitely─corresponds to the series of whole numbers in number theory. To number congruence corresponds here the congruence of classes of functions with respect to a double modulo, [...] (DEDEKIND 1930/32, vol.1, 47) It should be noticed how DEDEKIND stresses the fact that he is considering infinitely many classes, each of which contains infinitely many elements. Despite this character of actual infinity, he compares those classes with the natural numbers, most concrete objects for a traditional mathematician. Fourth and finally, in the only text that I have been able to find where DEDEKIND tried to justify his use of infinite sets in ideal theory, he establishes no relation between this and the real numbers. Instead, he traces an interesting parallel between his work and GAUSS's theory of the composition of quadratic forms in his famous Disquisitiones Arithmeticae (1801). In a letter to LIPSCHITZ of June 1876, DEDEKIND wrote: 83 'Abriss einer Theorie der höheren Kongruenzen in Bezug auf einen reellen Primzahl-Modulus' (1857), in DEDEKIND 1930/32, vol.1, 40-66. 1 As we can conceive of a collection of infinitely many functions─which are still dependent on variables─as one whole, for instance when we collect in a form-class all equivalent forms, denote this by a single letter, and submit it to composition, I can with the same right conceive of a system A of infinitely many, completely determined numbers of O, which satisfies two extremely simple conditions I. and II., as one whole, and name it an ideal [...] (LIPSCHITZ 1986, 62) While speaking about classes of quadratic forms, GAUSS had been very careful to express himself in a way that did not imply the existence of actual infinities. But it is clear that his reader DEDEKIND was not worried by philosophical subtleties; from the 1850s on, he accepted actual infinities as natural mathematical objects, perfectly sound and consistent in themselves. 5.2.2. The definition and the theorem of the infinite. DEDEKIND's definition of the infinite was made possible by the consideration of mappings, and particularly of injective mappings or 'similar representations'. A system S is called infinite when it is similar [equipollent] to a proper part of itself (32); in the opposite case S is said to be a finite system. (DEDEKIND 1888, 356) In a footnote to this definition, DEDEKIND affirmed that it "forms the core of my whole investigation", and that he had communicated it to several mathematicians (SCHWARZ and H. WEBER in the 1870s, CANTOR in 1882). The definition should have been quite surprising for DEDEKIND's contemporaries: far from defining the infinite as that which cannot be counted, that is, through a negative property (as RIEMANN and everybody else considered necessary), DEDEKIND defines it positively; the finite, instead, emerges as that which is not infinite. The reason for this lay in direct connection with DEDEKIND's aim of establishing set theory as the basis for mathematics, and in particular as the basis for a definition of the natural numbers.84 From a modern viewpoint, to choose one definition of the infinite or another is only of technical interest, in relation to what axioms we need for developing the theory, or for connecting one definition with the other. But this should not lead us to underestimate the historical and conceptual difficulties implied in DEDEKIND's step. He took what GALILEO and others─CAUCHY among them─had considered a paradoxical property, and used it 84 Defining the infinite as that which is not countable implies that we take the natural numbers as given, and use them to establish a crucial set-theoretical notion. This was precisely CANTOR's position until he knew about DEDEKIND's ideas (see DEDEKIND's letter to Weber in DEDEKIND 1930/32, vol.3, 488). 1 for a simple, deductively useful definition. Still more paradoxical must have seemed the fact that his theory of the finite subsets of N was based on the theory of chains, which are infinite subsets. DEDEKIND's theory of natural numbers and finite sets was decidedly an infinitistic one. We thus see that the notion of infinite set was really crucial for DEDEKIND's project. Along this line, he came to notice that the theory would be incomplete unless he could establish a proposition securing the existence of infinite sets. This is how his famous theorem of the infinite came to light. The theorem can not be found in the 1872-78 draft, but only in a comment to the second draft written in 1887 (Nachlass DEDEKIND III, 1, II, p. 32). It is well known that, in the meantime, DEDEKIND read BOLZANO's Paradoxien des Unendlichen,85 where a similar proof is presented (BOLZANO 1851, §13). BOLZANO's proof seems to have motivated DEDEKIND to include this theorem, although it is almost certain that it was not the only reason behind it. During the 1880s, CANTOR's papers on set theory encountered criticism, especially from KRONECKER, who tried to eliminate the infinite altogether from mathematics (cf. EDWARDS 1989). KRONECKER considered CANTOR's ideas as nonsense and even perverted, proposed to find ways to do without the real numbers, and criticized DEDEKIND's ideal theory for being infinitistic (KRONECKER 1882, 1887). The polemical reception of infinite sets during the 1880s was certainly behind DEDEKIND's theorem of the infinite. BOLZANO had tried to show that "the set of propositions and truths in themselves" was infinite. His method was to select a particular truth─the proposition 'that there are truths'─and construct from it another proposition, and another one from this, and so on.86 Thus he obtained an unlimited series of propositions, producing a set similar to N, and therefore infinite. This proof was dissatisfactory to the extent that it lacked a formal definition of the infinite. Therefore DEDEKIND refined it through his definition, but he also reformulated it in accordance with his peculiar logicist epistemology. Unlike BOLZANO, DEDEKIND had a strong 'nominalist' bias, in the sense that he avoided any reference to seemingly self-existing abstract objects, such as BOLZANO's world of truths in themselves. For him, mathematical objects were mental creations of the human mind, not selfexisting objects. Therefore he tried to show that the "mental universe" S, or totality of all possible thoughts, is infinite (DEDEKIND 1888, 357). According to his definition of the infinite, what he needed was to establish an injective mapping of S such that the image set is a proper part of S. Given sεS, he selected as its image φ(s) 85 See DUGAC 1976, 81, 88, 256. 86 'It is true that there are truths', 'It is true that it is true that there are truths', etc. 1 the proposition 's can be an object of my thought', which is a new possible thought. All of the images thus obtained are propositions according to traditional logical theories, that is, they are composed from a subject and a predicate. Moreover, φ(a) and φ(b) are different, since they have different subjects, and so the mapping is an injective one. Now, he only needed to exhibit an element of S not belonging to φ(S), and he selected 'my own self', which undoubtedly─at least for Occidental people─is a possible object of thought. Since 'my own self' has no predicate, it is not a proposition and does not belong to φ(S). Therefore, S is infinite.87 Nobody today would accept this as a mathematical theorem. DUGAC has even written that it is "the only theorem of Dedekind whose «proof» is not coherent with Dedekind's mathematical thought" (DUGAC 1976, 88). This statement, however, seems to me quite misleading and anachronistic. A basic characteristic of DEDEKIND's mature mathematical thought was to consider mathematics as the science of number, and this as a part of logic. But the logicist project would have fallen down if he had simply postulated an axiom of infinity (see section 4.3.3 above). DEDEKIND tried to show that the existence of infinite sets was a logical truth. It is for this reason that I consider his theorem as completely coherent with his mathematical thought, that is, with his conception of mathematics. Moreover, the requirement of a proof for the existence of infinite sets was not only a question of elegance and theoretical completeness, but also a matter of consistency. In 1890, DEDEKIND wrote to KEFERSTEIN: does such a[n infinite] system exist at all in our mental universe? Without a logical proof of existence it would always remain doubtful whether the notion of such a system might not perhaps contain internal contradictions. Hence the need for such proofs [...] (VAN HEIJENOORT 1967, 101; original in SINACEUR 1974, 275) This is still an unsolved─and mathematically unsolvable─question, since no consistency proof for set theory is possible without relying on stronger theories. In spite of the shortcomings of DEDEKIND's proof, later undermined by the antinomies, he shares with BOLZANO the merit of having seen the necessity of a proposition securing the existence of infinite sets within the deductive structure of set theory. For this reason, ZERMELO used to call the axiom of infinity 'DEDEKIND's axiom' (ZERMELO 1908a, 204; 1909, 186). The attempts of BOLZANO and DEDEKIND must be counted among the rare serious ones aiming to show the acceptability and validity of the actual 87 See also 'Über den Begriff des Unendlichen' (published in SINACEUR 1974), where DEDEKIND stresses the relation between his proof and traditional logic. 1 infinite in mathematics and in human thought generally. VI. FROM MARRIAGE TO PARTNERSHIP: IMPACT OF THE ANTINOMIES 6.1. Pre-Russellian logicism. Logicism can be considered to have been in the air since modern mathematics in general, and set theory in particular began to develop. But it was especially during the 1880s, when FREGE and DEDEKIND published their independently developed views, that a growing logicist trend emerged within the mathematical community. In the 1890s and 1900s, followers of logicism appeared in all the main European countries: DEDEKIND, FREGE, and SCHRÖDER in Germany, JOURDAIN and RUSSELL in Great Britain, and COUTURAT in France are the most outstanding names. The early form of logicism, however, has been mostly forgotten since, from the 1900s on, RUSSELL's conceptions became the most influential. RUSSELL and the Principia Mathematica (1910/13) represent a second phase in the history of logicism, forced by the effect of the antinomies. Earlier than the works of FREGE and DEDEKIND, we also have RIEMANN's seminal contribution, that could easily be understood in a logicist sense. But as we have seen its reception is difficult to ascertain. The impact of RIEMANN's notion of manifold on DEDEKIND and CANTOR is almost certain, but the same notion seems to have been understood in a much narrower sense by most mathematicians (see above). It is important to consider, however, that FREGE began his career working on geometry, and so it might well be that RIEMANN's work influenced his evolution.88 Be that as it may, by the time when logicism began to have a considerable group of followers, set theory was mainly represented by CANTOR and DEDEKIND, and so we do not encounter the name of RIEMANN any more. DEDEKIND's pioneering contribution of 1888 is surprising when viewed in the context of contemporary logic. He was a 'pure' mathematician who had never published anything on logic, but in 1888 he dared to present a presumed logical theory as a foundation of arithmetic, and even the whole of mathematics. Judged from twentieth-century standards, it was indeed a strange 'logic': nothing but a general theory of sets and mappings, and no mention even of propositional logic! Of course, DEDEKIND's 'logic' is not so strange when compared with the then-reigning algebra of logic, but it is still interesting to examine the response to this proposal within the logical community. Among the diverse contemporary conceptions of logic, we are interested above all in 88 This is just a conjecture for which I have no proof, but it might be fruitful to review FREGE's early writings in order to corroborate or disprove it. FREGE's "royal road from geometry" is described in WILSON 1992. 1 mathematical logic, the principal names being FREGE, SCHRÖDER, and PEIRCE. And in fact, there is evidence that DEDEKIND's logic was taken as a part of logic by his contemporaries. Even if the logical ideas of FREGE, PEANO, PEIRCE, or SCHRÖDER were more advanced, it was not so for mathematical conceptions, and DEDEKIND's contribution seems to have played an important role in the diffusion of logicism. In section III we observed how the notion of set was traditionally ascribed to logic, and we have seen that DEDEKIND did not even feel the necessity to argue for the logical character of sets. But his 1888 preface showed a clear desire to convince the reader that the notion of mapping was a purely logical one. According to DEDEKIND, mappings express the mental faculty of relating things to things, making things correspond, or using one thing as a representation of the other; this faculty is an indispensable ingredient of thought, and mappings are also a necessary basis for arithmetic (DEDEKIND 1888, 336). The reaction to this proposal was surprisingly good. This was the time of the development of a general theory of relations in the hands of PEIRCE and SCHRÖDER, and therefore a particularly appropriate situation for the reception of DEDEKIND's notion of mapping. Although Charles S. PEIRCE (1839-1931) did not undertake a logicist position, in 1901, writing on 'Logic' for BALDWIN's Dictionary of Philosophy and Psychology, he confessed that DEDEKIND's work showed how the borderline between logic and pure mathematics is almost evanescent (PEIRCE 1931/60, vol.2, 124-125). In 1911 he would say that DEDEKIND's conception of mappings was "an early and significant acknowledgement that the so-called «logic of relatives» is an integral part of logic" (PEIRCE 1931/60, vol.3, 389). By and large, however, it was Ernst SCHRÖDER (1841-1902) who received DEDEKIND's foundational work more positively, becoming a partisan of it. His work on the algebra of relatives, and on the subsumption of DEDEKIND's theories under it, finally brought SCHRÖDER's conversion to logicism. Incidentally, it is significant that in SCHRÖDER's and PEIRCE's work, the logicist position is always associated with DEDEKIND, and not with FREGE. In 1898, SCHRÖDER expressed openly his logicist conviction, saying that "pure mathematics seems to me just a branch of general logic" (SCHRÖDER 1898, 149).89 But he had not always been so positive about it. In 1890, while emphasizing that arithmetical truths are purely analytical, he was careful to leave open the question whether arithmetic can be equated with logic; thus he distanced himself from "those who, with DEDEKIND, consider arithmetic as a branch of 89 This and related texts of the late 1890s have been analyzed in PECKHAUS 1991, especially 177 and 191-194. But PECKHAUS has not observed the Dedekindian origins of SCHRÖDER's logicism (cf. also his 1993). 1 for historical understanding of the emergence of the logicist movement. 1 REFERENCES Archival sources Nachlass DEDEKIND. DEDEKIND's scientific legacy, kept in the Niedersächsische Staats- und Universitätsbibliothek Göttingen (Handschriftenabteilung). Nachlass RIEMANN. RIEMANN's scientific legacy, kept in the Niedersächsische Staats- und Universitätsbibliothek Göttingen (Handschriftenabteilung). Secondary sources ARNAULD, A., NICOLE, P. 1662. La Logique, ou l'art de penser. Facsimile reprint: Stuttgart, F. Frommann, 1965. BENACERRAF, P. & H. PUTNAM (eds.) 1983. Philosophy of Mathematics: selected readings. Cambridge University Press. BOLZANO, B. 1851. Paradoxien des Unendlichen. Leipzig, Reclam. Reprints: Leipzig, 1920, and Hamburg, 1975. English translation: London, Routledge & Kegan Paul, 1950. BOOLE, G. 1847. The mathematical analysis of logic. Cambridge, Macmillan. Reprint: Oxford, Basil Blackwell, 1951. BORGA, M. FREGUGLIA, P. PALLADINO D. 1985. I contributi fondazionali della scuola di Peano. Milano, Angeli. BOTTAZZINI, U. 1986. The Higher Calculus: A History of Real and Complex Analysis from Euler to Weierstrass. New York, Springer. BOURBAKI, N. 1969. Elements d'histoire des mathématiques. Paris, Hermann. Reprint: Paris, Masson, 1984. CANTOR, G. 1872. Über die Ausdehnung eines Satzes aus der Theorie der trigonometrischen Reihen. In CANTOR 1932, 92-101. ── 1878. Ein Beitrag zur Mannigfaltigkeitslehre, in CANTOR 1932, 119-133. ── 1882. Über unendliche, lineare Punktmannichfaltigkeiten, part III. In CANTOR 1932, 149-157. ── 1883. Über unendliche, lineare Punktmannichfaltigkeiten, part V. In CANTOR 1932, 165-208. Also published separatly as Grundlagen einer allgemeinen Mannigfaltigkeitslehre (Leipzig, 1883). ── 1895/97 Beiträge zur Begründung der transfiniten Mengenlehre. In CANTOR 1932, 282-351. English translation as Contributions to the founding of the Theory of Transfinite Numbers. Chicago, 1 Open Court, 1915; reprint: New York, Dover, 1955. ── 1932. Gesammelte Abhandlungen mathematischen und philosophischen Inhalts. Berlin, Springer. References to the reprint: Hildesheim, Olms, 1966. CANTOR, G., DEDEKIND, R. 1937. Cantor-Dedekind Briefwechsel. Paris, Hermann. French translation, including the 1899 letters, in CAVAILLÈS 1962. CARNAP, R. 1931. The Logicist Foundations of Mathematics. In BENACERRAF & PUTNAM 1983. CASSIRER, E. 1910. Substance and Function. Published with Einstein's Theory of Relativity (1921), New York, Dover, 1953. CAVAILLES, J. 1962. Philosophie mathématique. Paris, Hermann. COOKE, R. 1993. Uniqueness of Trigonometric Series and Descriptive Set Theory, 1870-1985. Archive for History of Exact Sciences 45, 281-334. COUTURAT, L. 1896. De l'infinie mathématique. Paris, Alcan. ── 1905. Les principes des mathématiques. Paris, Alcan. Reprint: Paris, Blanchard, 1980. DAUBEN, J. 1979. Georg Cantor. His mathematics and philosophy of the infinite. Cambridge, Harvard University Press. DEAÑO, A. 1980. Las concepciones de la lógica. Madrid: Taurus. DEDEKIND, R. 1854. Über die Einführung neuer Funktionen in der Mathematik. In DEDEKIND 1930/32, vol.3, 428-438. ── 1871. Über die Komposition der binären quadratischen Formen. Supplement X to Dirichlet & Dedekind, Zahlentheorie, 2nd edition (Braunschweig, Vieweg, 1871). In DEDEKIND 1930/32, vol.3, 223-261. ── 1872. Stetigkeit und irrationale Zahlen. In DEDEKIND 1930/32, vol.3, 315-334. ── 1876/77. Sur la théorie des nombres entiers algébriques. Bulletin des Sciences mathématiques et astronomiques, ser.1, 11 (1876), ser.2, 1 (1877). Also published separatly: Paris, Gauthier-Villars, 1877. References to the partial reprint in DEDEKIND 1930/32, vol.3, 262-296. ── 1879. Über die Theorie der ganzen algebraischen Zahlen. Supplement XI to Dirichlet & Dedekind, Zahlentheorie, 3rd edition (Braunschweig, Vieweg, 1879). ── 1888. Was sind und was sollen die Zahlen?. Braunschweig, Vieweg. In DEDEKIND 1930/32, vol.3, 335-390. ── 1893. Über die Theorie der ganzen algebraischen Zahlen. Supplement XI to Dirichlet & Dedekind, Vorlesungen über Zahlentheorie, 4th edition (Braunschweig, Vieweg, 1893; reprint: New York, Chelsea, 1968), 434-657. ── 1895. Über die Begründung der Idealtheorie. In DEDEKIND 1930/32, vol.2, 50-58. ── 1930/32. Gesammelte mathematische Werke. 3 vols., Braunschweig, Vieweg. Reprint: New 1 York, Chelsea, 1969, 3 vols. in 2. ── 1981. Eine Vorlesung über Algebra. In SCHARLAU 1981, 59-100. DROBISCH, M.W. 1836. Neue Darstellung der Logik nach ihren einfachen Verhältnissen. Leipzig, Voss. DUGAC, P. 1976. Richard Dedekind et les fondements des mathématiques (avec de nombreux textes inédits). Paris, Vrin. ── 1981. Richard Dedekind et l'application comme fondement des mathématiques. In SCHARLAU 1981, 134-144. EDWARDS, H.M. 1980. The Genesis of Ideal Theory. Archive for History of Exact Sciences 23, 321-378. ── 1983. Dedekind's invention of ideals. Bulletin of the London Mathematical Society 15, 8-17. Reprinted in PHILLIPS 1987, 8-20. ── 1989. Kronecker's Views on the Foundations of Mathematics. In ROWE & MCCLEARY 1989, 67-77. EDWARDS, H.M., NEUMANN, O., PURKERT, W. 1982. Dedekind's «Bunte Bemerkungen» zu Kroneckers «Grundzüge». Archive for History of Exact Sciences 27, 49-85. FECHNER, G.T. 1823. Katechismus der Logik oder Denklehre. Leipzig, Baumgärtner. ── 1864. Über die physikalische und philosophische Atomenlehre. Leipzig, H. Mendelssohn. FERREIROS, J. 1992. Sobre los orígenes de la matemática abstracta: Richard Dedekind y Bernhard Riemann. Theoria 16-17-18, tomo A, 473-498. Shorter version in S. GARMA, D. FLAMENT, V. NAVARRO (eds.) Contra los titanes de la rutina/Contre les titanes de la routine (Madrid, CSIC/Comunidad de Madrid, 1994), p. 301-318. ── 1993. On the Relations between Cantor and Dedekind. Historia Mathematica 20, 343-363. ── 1993a. El nacimiento de la teoría de conjuntos, 1854-1908. Madrid, Publicaciones de la Universidad Autónoma. ── forthcoming. The Discovery of Transfinite Numbers, or the Fermentation of Cantorian Ideas. Historia Mathematica. FISCHER, F. 1838. Lehrbuch der Logik für academische Vorlesungen und Gymnasialvorträge. Stuttgart, J.B. Metzler. FORSGREN, K.A. 1985. Die deutsche Satzgliedlehre. Göteborg, Göteborger Germanistische Forschungen, no. 29. ── 1992 Satz, Satzarten, Satzglieder. Zur Gestaltung traditionellen Grammatik von K.F. Becker bis K. Duden 1830-1880. Münster, Nodus. FRANK, H.J. 1973. Geschichte des Deutschunterrichts. Von den Anfängen bis 1945. München, C. Hanser. 1 FREGE, G. 1879. Begriffschrift. Halle, Nebert. Reprint with related texts: Hildesheim, Olms, 1964. ── 1884. Die Grundlagen der Arithmetik. Breslau, Koebner. ── 1893. Grundgesetze der Arithmetik, vol.1. Jena, Pohle. Reprint with FREGE 1903: Hildesheim, G. Olms, 1966. ── 1895. Kritische Beleuchtung einiger Punkte in E. Schröders Vorlesungen über die Algebra der Logik. Archiv für systematische Philosophie 1, 433-456. English translation in G. FREGE, Collected papers on mathematics, logic and philosophy (Oxford, Basil Blackwell, 1984). ── 1903. Grundgesetze der Arithmetik, vol.2. Jena, Pohle. Reprint with FREGE 1893: Hildesheim, G. Olms, 1966. ── 1969. Nachgelassene Schriften. Hamburg, Meiner. FRIES, J.F. 1827. Grundriss der Logik. Heidelberg, C.F. Winter. ── 1837. System der Logik. Heidelberg, C.F. Winter. FRISCH, J.C. 1969. Extension and Comprehension in Logic. New York, Philosophical Library. GARCIADIEGO, A.R. 1992. Bertrand Russell and the Origins of the Set-theoretic Paradoxes. Basel, Birkhäuser. GAUSS, C.F. 1831. Selbstanzeige of Theoria residuorum biquadraticorum, commentatio secunda. In GAUSS 1863/1912, vol.2, 169-178. ── 1849. Beiträge zur Theorie der algebraischen Gleichungen. In GAUSS 1863/1912, vol.3, 71- 102. ── 1863/1929. Werke. 12 vols., Göttingen, Dieterich. Reprint: Hildesheim, G. Olms, 1973. GOLDFARB, W. 1979. Logic in the Twenties: The Nature of the Quantifier. Journal of Symbolic Logic 44, 351-368. ── 1988. Poincaré against the Logicists. In ASPRAY, W. & P. KITCHER (eds.), History and Philosophy of Modern Mathematics. Minneapolis, University Press. GRATTAN-GUINNESS, I. 1970. An unpublished paper by Georg Cantor: Principien einer Theorie der Ordnungstypen. Erste Mitteilung. Acta Mathematica 124, 65-107. ── (ed.) 1980. From the Calculus to Set Theory, 1630-1910. London, Duckworth. ── 1984. Notes on the Fate of Logicism from Principia Mathematica to Gödel's Incompletability Theorem. History and Philosophy of Logic 5, 67-78. GRAY, J. 1986. Linear ordinary differential equations and group theory from Riemann to Poincarè. Basel: Birkhäuser. ── 1992. The nineteenth-century revolution in mathematical ontology. In D. Gillies (ed.) Revolutions in Mathematics (Oxford: Clarendon Press, 1992), p. 226-248. HANKEL, H. 1870. Untersuchungen über die unendlich oft oszillierenden und unstetigen Funktionen. In Mathematische Annalen 20 (1882), 63-112. 1 HAUBRICH, R. 1988. Zur Entstehung der Idealtheorie Dedekinds. Diplomarbeit, Universität Göttingen. ── 1992. Zur Entstehung der algebraischen Zahlentheorie Richard Dedekinds. Dissertation, Universität Göttingen. HAUSDORFF, F. 1914. Grundzüge der Mengenlehre. Leipzig, Veit. Reprint: New York, Chelsea, 1949. HAWKINS, T.W. 1975. Lebesgue's theory of integration. Its origins and development. New York, Chelsea. HERBART, J.F. 1808. Hauptpunkte der Logik. In HERBART 1964, vol.2, 217-226. ── 1837. Lehrbuch zur Einleitung in die Philosophie. Zweyter Abschnitt. Die Logik. In HERBART 1964, vol.4, 67-104. ── 1964. Sämtliche Werke in chronologischer Reihenfolge. 19 vols., Langensalza, Beyer & Söhne, 1887-1912. Reprint: Aalen, Scientia, 1964. HILBERT, D. 1897. Die Theorie der algebraischen Zahlkörper. In Gesammelte mathematische Abhandlungen, vol.1 (Berlin, Springer, 1932), 63-363. ── 1904. Über die Grundlagen der Logik und Arithmetik. In HILBERT 1930, 247-261. English translation in VAN HEIJENOORT 1967, 129-138. ── 1930. Grundlagen der Geometrie. Leipzig/Berlin, Teubner. JOHNSON, D.M. 1979/81. The Problem of the Invariance of Dimension in the Growth of Modern Topology. Archive for History of Sciences 20, 97-188, 25, 85-266. JOURDAIN, P.E.B. 1913. The Nature of Mathematics. London, Jack. ── 1991. Selected Essays on the History of Set Theory and Logic (1906-1918). Bologna, CLUEB. KANT, I. 1787. Kritik der reinen Vernunft. Riga, Hartknoch. References correspond to the so-called edition B. ── 1885. Introduction to Logic. London, Longman's Green. KENNEDY, H.C. 1974. Peano's concept of number. Historia Mathematica 1, 387-408. ── 1980. Life and Works of Giuseppe Peano. Dordrecht, Reidel. KNEALE, W., KNEALE, M. 1962. The development of Logic. Oxford, Clarendon. KÖNIG, G. (ed.) 1990. Konzepte des mathematisch Unendlichen im 19. Jahrhundert. Göttingen, Vandenhoeck & Ruprecht. KRONECKER, L. 1882. Grundzüge einer arithmetischen Theorie der algebraischen Grössen. Journal für die reine und angewandte Mathematik 92, 1-122. Reprint in KRONECKER 1895/1930, 1 vol.2. ── 1887. Über den Zahlbegriff. In KRONECKER 1895/1930, vol.3/1, 249-274. ── 1895/1930. Werke. 5 vols., Leipzig, Teubner. Reprint: New York, Chelsea, 1968. LANDAU, E. 1917. Richard Dedekind ─ Gedächtnisrede. Nachrichten der Königliche Gesellschaft der Wissenschaften zu Göttingen. Geschäftliche Mitteilungen 1917, 50-70. LEIBNIZ, G.W. 1714. Monadologie. In Die philosophischen Schriften (ed. Gerhardt, 7 vols., Berlin, Weidmann, 1875/90; reprint: Hildesheim, Oms, 1960/61), vol. 6, 607-623. ── 1966. Logical Papers. Oxford, Clarendon Press. LIPSCHITZ, R. 1986. Briefwechsel mit Cantor, Dedekind, Helmholtz, Kronecker, Weierstrass und anderen. Braunschweig, Vieweg. LOTZE, H. 1856/64. Mikrokosmus. Ideen zur Naturgeschichte und Geschichte der Menschheit, Versuch einer Anthropologie. 3 vols., Leipzig, S. Hirzel. English translation in 2 vols., Edinburgh, T. & T. Clark, 1885. ── 1874. Drei Bücher der Logik. Leipzig, Hirzel. English translation: Oxford, Clarendon, 1884. MATTHIAS, A. 1907. Geschichte des Deutschen Unterrichts. München, Beck. MEDVEDEV, F.A. 1984. Über die Abstrakten Mengenlehren von Cantor und Dedekind. Berichte zur Wissenschaftsgeschichte 7, 195-200. MEHRTENS, H. 1979. Die Entstehung der Verbandstheorie. Hildesheim, Gerstenberg. ── 1979a. Das Skelett der modernen Algebra. Zur Bildung mathematischer Begriffe bei Richard Dedekind. In Disciplinae Novae (Göttingen, Vandenhoeck & Ruprecht), 25-43. MESCHKOWSKI, H. 1967. Probleme des Unendlichen. Werk und Leben Georg Cantors. Braunschweig, Vieweg. Reprinted as Georg Cantor. Leben, Werk und Wirkung, Mannheim, Bibliographisches Institut, 1983. MOORE, G.H. 1978. The Origins of Zermelo's Axiomatization of Set Theory. Journal of Philosophical Logic 7, 307-329. ── 1980. Beyond First-order Logic: The Historical Interplay between Mathematical Logic and Axiomatic Set Theory. History and Philosophy of Logic 1, 95-137. ── 1982. Zermelo's Axiom of Choice. Its Origins, Development and Influence New York/Berlin, Springer. ── 1987. A House Divided Against Itself: The Emergence of First-Order Logic as the Basis for Mathematics. In PHILLIPS 1987, 98-136. ── 1993. Introduction to B. Russell, Collected Papers, vol.3. London: Allen & Unwin. NAUMANN, B. 1986. Grammatik der deutschen Sprache zwischen 1781 und 1856. Berlin, E. Schmidt. 1 NEUMANN, O. 1980. Zur Genesis der algebraischen Zahlentheorie. NTM-Schriftenreihe 17, (1) 32-48, (2) 38-58. PAULSEN, F. 1897. Geschichte des gelehrten Unterrichts auf den deutschen Schulen und Universitäten ... mit besonderer Rücksicht auf den klassischen Unterricht. 2 vol., Leipzig, Veit. PEANO, G. 1889. Arithmetices principia, nova methodo exposita. Torino, Bocca. English translation in VAN HEIJENOORT 1967, 83-97. ── 1891. Sul concetto di numero. Rivista di matematica 1, 87-102, 256-267. ── 1957/59. Opere scelte. 3 vols., Roma, Cremonese. PEARS, D.F. (ed.) 1972. Bertrand Russell: A collection of critical essays. Garden City, N.Y.: Anchor Books. PECKHAUS, V. 1990. «Ich habe mich wohl gehütet, alle Patronen auf einmal zu verschiessen». Ernst Zermelo in Göttingen. History and Philosophy of Logic 11, 19-58. ── 1990a. Hilbertprogramm und kritische Philosophie. Göttingen: Vandenhoeck & Ruprecht. ── 1991. Ernst Schröder und die "pasigraphischen Systeme" von Peano und Peirce. Modern Logic 1, 174-203. ── 1991a. Fallstudien zur Begründung einer Sozialgeschichte der formalen Logik [7. Logikunterricht an Schulen und Hochschulen]. Unpublished report on a research project, Institut für Philosophie, Universität Erlangen-Nürnberg. ── 1993. Ernst Schröder und der Logizismus. In W. STELZNER (ed.) Philosophie und Logik (Berlin, W. de Gruyter, 1993), 108-119. PEIRCE, C.S. 1931/60. Collected Papers. 7 vols., Harvard University Press. PHILLIPS, E.R. 1987. Studies in the History of Mathematics. Mathematical Association of America. PINCHERLE, S. 1880. Saggio di una introduzione alla teoria delle funzioni analitiche secondo i principii del Prof. C. Weierstrass. Giornale di Matematica 18, 178-254, 317-357. PONT, J.C. 1974. La topologie algébrique des origines à Poincaré. Paris, P.U.F.. PURKERT, W. 1973. Zur Genesis des abstrakten Körperbegriffs. NTM-Schriftenreihe 10, (1) 23- 37, (2) 8-20. ── 1977. Ein Manuskript Dedekinds über Galois-Theorie. NTM-Schriftenreihe 13, (2) 1-16. PURKERT, W., ILGAUDS, H.J. 1987. Georg Cantor 1845-1918. Basel, Birkhäuser. RIEMANN, B. 1851. Grundlagen für eine allgemeine Theorie der Functionen einer veränderlichen complexen Grösse. In RIEMANN 1892, 3-45. 1 ── 1854. Über die Hypothesen, welche der Geometrie zu Grunde liegen. In RIEMANN 1892, 272- 287. ── 1854a. Über die Darstellbarkeit einer Function durch eine trigonometrische Reihe. In RIEMANN 1892, 227-265. ── 1857. Theorie der Abel'schen Functionen. In RIEMANN 1892, 88-142. ── 1892. Gesammelte mathematische Werke und wissenschaftlicher Nachlass. Leipzig, Teubner. Reprint: New York, Dover, 1953. RODRIGUEZ CONSUEGRA, F. 1987. Russell's Logicist Definitions of Numbers, 1898-1913: Chronology and Significance. History and Philosophy of Logic 8, 141-169. ── 1989. Russell's theory of types, 1901-1910: Its complex origins in the unpublished manuscripts. History and Philosophy of Logic 10, 131-164. ── 1991. The Mathematical Philosophy of Bertrand Russell: Origins and Development. Basel, Birkhäuser. RUSSELL, B. 1903. The principles of mathematics. Cambridge, University Press. Reprint: London, Allen & Unwin, 1948. ── 1908. Mathematical Logic as Based on the Theory of Types. American Journal of Mathematics 30, 222-262. Reprinted in VAN HEIJENOORT 1967, 150-182. RUSSELL, B. WHITEHEAD, A.N. 1910/13. Principia Mathematica. 3 vols., Cambridge, University Press. SAVAGE, C.W. ANDERSON, C.A. (eds.) 1989. Rereading Russell: Essays in Bertrand Russell's metaphysics and epistemology. University of Minnesota Press. SCHARLAU, W. (ed.) 1981. Richard Dedekind 1831/1981. Eine Würdigung zu seinem 150. Geburtstag. Braunschweig, Vieweg. ── 1982. Unveröffentliche algebraische Arbeiten Richard Dedekinds aus seiner Göttinger Zeit 1855-1858. Archive for History of Exact Sciences 27, 335-367. SCHOLZ, E. 1980. Geschichte des Mannigfaltigkeitsbegriff vom Riemann bis Poincaré. Basel, Birkhäuser. ── 1982. Riemanns frühe Notizen zum Mannigfaltigkeitsbegriff und zu den Grundlagen der Geometrie. Archive for History of Exact Sciences 27, 213-232. ── 1982a. Herbart's influence on Bernhard Riemann. Historia mathematica 9, 413-440. SCHRÖDER, E. 1877. Der Operationskreis des Logikkalküls. Leipzig, Teubner. Reprinted in 1966. ── 1890/1905. Vorlesungen über die Algebra der Logik. 3 vols., Leipzig, Teubner. Reprint: New York, Chelsea, 1966. ── 1898. Über Pasigraphie, ihren gegenwartigen Stand und die pasigraphische Bewegung in Italien. In F. RUDIO (ed), Verhandlungen des Ersten Internationalen Mathematiker-Kongresses (Leipzig, Teubner), 147-162. English version in The Monist 9 (1899), 44-62. 1 SCHUBRING, G. 1982. Die Theorie des Unendlichen bei Johann Schultz. Historia Mathematica 9, 441-484. SIGWART, H.C.W. 1835. Die Logik in ihrer Beziehungen zur allgemeinen Sprachlehre. Tübingen, C.F. Oslander. SINACEUR, M.A. 1971. Appartenance et inclusion: Un inedit de Richard Dedekind. Revue d'histoire des sciences et de leurs applications 24, 247-254. ── 1974. L'infini et les nombres. Commentaires de R. Dedekind a «Zahlen». La correspondance avec Keferstein. Revue d'histoire des sciences 27, 251-278. ── 1990. Dedekind et le programme de Riemann. Revue d'histoire des sciences 11, 221-296. STEINER, J. 1832. Systematische Entwickelung der Abhängigkeit geometrischer Gestalten von einander, mit Berücksichtigung ... Berlin, Fincke. STYAZHKIN, N.I. 1969. History of mathematical logic from Leibniz to Peano. Cambridge/London, MIT Press. UEBERWEG, F. 1882. System der Logik und Geschichte der logischen Lehren. 5th ed., edited by J. Bona Meyer. Bonn, A. Marcus. English translation from the 3rd ed., London, Longmans, Green and Co., 1871. VAN HEIJENOORT, J. 1967. From Frege to Gödel. A source book in mathematical logic, 1879- 1931. Cambridge/London, Harvard Univesity Press. VERLEY, J.L. 1978. Les fonctions analytiques. In J. DIEUDONNÉ (ed.), Abrégé d'histoire des mathématiques, 1700-1900 (2 vols., Paris, Hermann, 1978), chap.4. WALTHER-KLAUS, E. 1987. Inhalt und Umfang. Untersuchungen zur Geltung und zur Geschichte der Reziprozität von Extension und Intension. Hildesheim: Olms. WANG, H. 1974. The concept of set. In BENACERRAF & PUTNAM 1983. WEIERSTRASS, K. 1894/1927. Mathematische Werke. 7 vols., Berlin, Mayer & Müller. ── 1986. Einführung in die Theorie der analytischen Funktionen (Mitschrift Killing, 1868). Schriftenreihe des Math. Instituts der Univ. Münster, 2nd series, 38. ── 1988. Einleitung in die Theorie der analytischen Funktionen (Mitschrift Hurwitz, 1878). Braunschweig, Vieweg. WILSON, M. 1992. Frege: The Royal Road from Geometry. Noûs 26, 149-180. WINCHESTER, I. BLACKWELL, K. (eds.) 1989. Antinomies and paradoxes: Studies in Russell's early philosophy. McMaster University Library Press.