Arxiv:1908.01571V5 [Math.HO] 3 May 2020 Contents

Alexander Aycock

May 5, 2020 arXiv:1908.01571v5 [math.HO] 3 May 2020 Contents

1 Euler and the Γ-Function 1 1.1 Introduction ...... 1 1.1.1 Motivation ...... 1 1.1.2 Euler’s Idea concerning the Γ-function ...... 2 1.1.3 Organisation of the Paper ...... 3 1.1.3.1 General Overview ...... 3 1.1.3.2 Notation ...... 3 1.2 Historical Overview of the Γ-function ...... 5 1.2.1 Brief historical Overview ...... 5 1.2.1.1 1. Phase: Discovery by Euler ...... 5 1.2.1.2 2. Phase: Gauß’s Investigations ...... 6 1.2.1.3 3. Phase: Development of complex Analysis ...... 7 1.2.1.4 4. Phase: Formulas following from complex Analysis ...... 7 1.2.1.5 5. Phase: Classification as transcendentally transcendental Function 8 1.2.1.6 6. Phase: Towards the axiomatic Introduction ...... 9 1.2.1.7 7. Phase: Transcendence Questions ...... 10 1.2.2 More detailed Discussion on the Discovery of the Γ-function ...... 11 1.2.3 Overview on Euler’s Contributions ...... 14 1.2.3.1 Euler’s Papers on the Γ-function ...... 14 1.2.3.2 Euler’s Results on the Γ-function ...... 17 1.3 Short modern Introduction to the Γ-Function ...... 20 1.3.1 Definition and simple Properties ...... 20 1.3.1.1 Definition ...... 20 1.3.1.2 Simple Properties ...... 21 1.3.2 Classification Theorems ...... 22 1.3.2.1 Wielandt’s Theorem ...... 22 1.3.2.2 Bohr-Mollerup Theorem ...... 23 1.4 Solution of the Difference Equation F(x + 1) = xF(x) ...... 26 1.4.1 Weierstraß’s Definition of the Γ-function ...... 26 1.4.2 A Remark concerning the Solution of the Difference Equation ...... 26 1.4.3 General Solution of the Equation F(x + 1) = xF(x) ...... 26 1.4.3.1 Introduction of an Auxiliary Function ...... 26 1.4.3.2 Product Representation of the Function F ...... 27 1.4.3.3 Finding the Constant K ...... 28 1.4.4 Application to Γ(x)- The Weierstraß Product Representation ...... 29 1.4.5 Euler on Weierstraß’s Condition ...... 30 1.5 Euler’s direct Solution of the Equation Γ(x + 1) = xΓ(x) - The Moment Ansatz . . . 32

ii 1.5.1 Origin of the Idea ...... 32 1.5.2 Euler’s Idea ...... 32 1.5.2.1 Application to the Γ-function - Finding the Integral Representation 34 1.5.2.2 Some Remarks on the Ansatz ...... 37 1.5.3 Examples of other Equations which can be solved by this Method ...... 37 1.5.3.1 1. Example: Legendre Polynomials ...... 37 1.5.3.2 Historical Remark on Legendre Polynomials ...... 39 1.5.3.3 2. Example: Hermite Polynomials ...... 40 1.5.3.4 3. Example: Beta Function ...... 41 1.5.3.5 Some Remarks on the Relation of Γ and B...... 43 1.5.3.6 4. Example: Hypergeometric Series ...... 44 1.5.3.7 Historical Note on the Integral Representation ...... 45 1.5.4 Euler and the Mellin-Transform ...... 45 1.5.4.1 Definition of the Mellin-Transform and inverse Mellin-Transform . 45 1.5.4.2 1. Example: Γ-function ...... 49 1.5.4.3 Remark ...... 49 1.5.4.4 2. Example: Generalized Γ-function ...... 50 1.5.4.5 3. Example: Hermite Polynomials ...... 50 1.5.4.6 4. Example: Legendre Polynomials ...... 51 1.5.4.7 Final Results ...... 52 1.5.5 Modern Idea - Intersection Theory ...... 52 1.5.5.1 Main Ingredient of Euler’s Approach ...... 53 1.5.5.2 Overview ...... 53 1.5.5.3 Necessity for the Concept of a Twist ...... 54 1.5.5.4 Towards Stokes’ Theorem for Integrals of multi-valued Functions . 54 1.5.5.5 Towards the twisted Homology Group ...... 55 1.5.5.6 Locally finite Twisted Holonomy Group ...... 58 1.5.5.7 de Rham Theory ...... 59 1.5.5.8 Twisted de Rham Theory ...... 60 1.5.5.9 Construction of twisted Cycles ...... 61 1.5.5.10 Intersection Number for Twisted Cycles ...... 64 1.5.5.11 Intersection Numbers for Twisted Cocycles ...... 66 1.5.5.12 Calculating Examples - Using a different Notation ...... 67 1.5.5.13 Derivation of the Decomposition Formula ...... 67 1.5.5.14 The case of 1-forms ...... 69 1.5.5.15 1. Example: B-function ...... 69 1.5.5.16 2. Example: Hypergeometric Function ...... 72 1.5.5.17 3. Example: Γ-function ...... 74 1.5.5.18 Remarks ...... 75 1.6 Solution of the Equation log Γ(x + 1) − log Γ(x) = log x by Conversion into a Differential Equation of infinite Order ...... 76 1.6.1 Overview ...... 76 1.6.2 Euler’s Idea ...... 76 1.6.2.1 Presentation of his Idea ...... 76 1.6.2.2 Actual Solution ...... 77 1.6.2.3 Mistake in Euler’s Approach ...... 77 1.6.3 Correction of Euler’s Approach ...... 78 1.6.3.1 Preparations - Lemmata and Definitions ...... 78 1.6.3.2 Solution Algorithm for the general difference equation - Finding a particular solution ...... 80 1.6.3.3 Case of multiple Roots ...... 81 1.6.3.4 Application to the difference equation f (x + 1) − f (x) = g(x).... 82 1.6.3.5 Explicit Solution of the general Difference Equation via Fourier- Transform ...... 82 1.6.4 A Solution Euler could have given ...... 84 1.6.4.1 An Application - Derivation of the Stirling Formula for the Factorial 85 1.6.5 Generalized Factorials ...... 90 1.6.5.1 Euler’s Results on generalized factorials ...... 90 1.6.5.2 Evaluation of the Constant A ...... 90 1.6.5.3 Finding the other Constants ...... 92 1.6.5.4 Euler’s Relations among the Constants ...... 93 1.7 Solution of log Γ(x + 1) − log Γ(x) = log x via the Euler-Maclaurin Formula . . . . . 95 1.7.1 Overview on the Section ...... 95 1.7.2 Euler’s Derivation of the Formula ...... 95 1.7.2.1 Several Remarks ...... 97 1.7.2.2 Purely formal Derivation ...... 97 1.7.3 Historical Overview ...... 98 1.7.3.1 Formal Approaches - From Euler to Bourlet ...... 98 1.7.3.2 Rigorous Results - From Carmichel to today ...... 99 1.7.4 Modern Derivation ...... 100 1.7.4.1 Euler Maclaurin Formula for C1−functions ...... 100 1.7.4.2 General Euler-Maclaurin Summation Formula ...... 101 1.7.4.3 Comparison to Euler’s Idea ...... 101 1.8 Interpolation Theory and Difference Calculus ...... 102 1.8.1 Overview ...... 102 1.8.2 Euler’s Idea ...... 102 1.8.3 Examples ...... 103 1.8.3.1 Harmonic Series ...... 103 1.8.3.2 Γ(x)-function ...... 103 1.8.4 Difference Calculus according to Euler ...... 104 1.8.5 Modern Idea - Weierstraß Product ...... 107 1.8.5.1 Introduction to the Problem ...... 107 1.8.5.2 Comparison to Euler’s Idea ...... 109 1.9 Relation among Γ and B ...... 110 1.9.1 From B to Γ - Euler’s first Way to the Integral Representation ...... 110 1.9.1.1 Euler’s Thought Process ...... 110 1.9.1.2 Euler’s Mathematical Argument ...... 110 1.9.1.3 Using the B-function ...... 112 1.9.1.4 Gauß’s Idea ...... 112 1.9.2 Expressing the B-function via Γ-functions - Fundamental Relation ...... 113 1.9.2.1 Euler’s Proof ...... 113 1.9.2.2 Jacobi’s Proof ...... 114 1.9.2.3 Dirichlet’s Proof ...... 114 1.9.2.4 Discussion ...... 115 1.9.3 Relation to the Beta Function - Expressing Γ via B ...... 116 1.9.4 Reflection Formula ...... 117 1.9.4.1 Euler’s Proof of the Sine Product Formula ...... 117 1.9.4.2 Euler’s Proof of the Reflection Formula ...... 120 1.9.4.3 Proof Euler could have given ...... 121 1.9.5 A slight Detour: Gamma Class Conjectures ...... 122 1.9.5.1 Gamma Class ...... 122 1.9.5.2 Gamma Conjectures ...... 122 1.9.6 Multiplication Formula ...... 124 p 1.9.6.1 The Function q ...... 124 1.9.6.2 Applying the Reflection Formula ...... 125 1.9.6.3 Euler’s Version of the Multiplication Formula ...... 126 1.9.6.4 Euler’s Proof of the Multiplication Formula ...... 127 1.9.6.5 Discussion of Euler’s Result ...... 128 1.10 Summary ...... 130 1.10.1 Overview of Euler’s Results on the Γ-Function ...... 130 1.10.1.1 General Overview ...... 130 1.10.1.2 Results not generally associated with Euler ...... 133 1.10.1.3 Results erroneously ascribed to Euler ...... 133 1.11 Appendix ...... 135 1.11.1 A Solution of the Basel Problem that Euler missed (but could have found) . 135 1.11.2 Euler and the Partial Fraction Decomposition of Transcendental Functions . 136 π 1.11.2.1 An Example: sin πx ...... 136 1 1.11.2.2 An Example in which Euler’s Method fails - ez−1 ...... 137 1.11.3 Euler on the Euler-Mascheroni Constant ...... 138 1.11.3.1 Euler’s Discovery of γ ...... 138 1.11.3.2 Euler’s Formulas for γ ...... 140

Appendices 155

I On the Translations of Euler’s Papers 155 I.1 Possible Questions about the Translations ...... 155 I.1.1 Where can I ﬁnd the original Papers? ...... 155 I.1.2 Why these three Papers? ...... 155 I.1.3 What was changed compared to Euler’s Originals? ...... 155 I.1.4 Why were certain Things not changed? ...... 156 I.1.5 What about the Footnotes? ...... 156

II Translation of E19 - On transcendental Progressions or those whose general terms cannot be given algebraically 157

III Translation of E368 - On the hypergeometric curve expressed by the equation y = 1 · 2 · 3 ··· x 175

R f −1 m IV Translation of E421 - Expansion of the integral x dx(log x) n having extended the integration from the value x = 0 to x = 1 233 List of Figures

1.1 Letter from Bernoulli to Goldbach ...... 12 1.2 Letter from Euler to Goldbach - first Part ...... 12 1.3 Letter from Euler to Goldbach - second Part ...... 13 1.4 Moment Method ...... 33 1.5 Euler finds the Integral Representation of Γ(x) via the Moment Method ...... 35 1.6 Euler and the Legendre Polynomials ...... 40 1.7 Euler and the hypergeometric Series ...... 46 1.8 Euler and the hypergeometric Differential Equation ...... 46 1.9 Integral Representation of the hypergeometric Series ...... 47 1.10 2-simplex with Orientation in the two-dimensional Case ...... 56 1.11 Construction of twisted Cycles in the one-dimensional Case ...... 62 1.12 Triangulation of the Circle ...... 62 1.13 Intersections of ∆(ω) and ∆ ...... 65 1.14 Euler and Stirling’s Constant ...... 89 1.15 Euler’s generalized Factorials ...... 91 1.16 Finding the Euler-Maclaurin Summation Formula ...... 96 1.17 The Euler-Maclaurin Sum Formula ...... 97 1.18 Euler’s Key Formula in the Calculus of Differences ...... 104 1.19 Solution of the Basel Problem ...... 118 1.20 Sine Product Formula ...... 119 1.21 Reflection Formula for the Γ-function ...... 120 1.22 Euler’s Version of the Multiplication Formula for the Γ-function ...... 124 1.23 Formula for the Fourier Coefficients ...... 131 1.24 Euler’s Formulas for Γ(x) - Part 1 ...... 131 1.25 Euler’s Formulas for Γ(x) - Part 2 ...... 132 1.26 Discovery of the Euler-Mascheroni Constant ...... 139 1.27 Formulas for the Euler-Mascheroni Constant - Part 1 ...... 141 1.28 Formulas for the Euler-Mascheroni Constant - Part 2 ...... 142

vi Chapter 1

Euler and the Γ-Function

This is the main body of the work. We present and discuss Euler’s results on the Γ-function and will explain how Euler obtained them and how Euler’s ideas anticipate more modern approaches and theories.

1.1 Introduction

1.1.1 Motivation

According to Leibniz, there are two arts in mathematics: The "Ars Inveniendi" (Art of Finding) and the "Ars Demonstrandi" (Art of Proving). Nowadays, the latter is dominating the mathematical education, whereas the first is often neglected. Leonhard Euler’s mathematical works are special not only for their quality and quantity, but also for his pedagogical style rendering them easily understandable. This is summarized by two famous quotes, the first due to Laplace: "Read Euler, read Euler, he is the Master of us all"1, the other due to Gauß: "The study of Euler’s works cannot be replaced by anything else." Moreover, Euler does not only provide the proofs of certain theorems, but also tells us how he found the theorem in the first place. In other words, Euler’s work gives us both, the Ars Demonstrandi2 and the Ars Inveniendi3. In this article we want to review and discuss some of the properties of the Γ-function, defined as4

Z∞ Γ(x) := tx−1e−tdt for Re(x) > 0, (1.1) 0

1W. Dunham even wrote a book with a title alluding to this quote [Du99]. 2Being written in the 18th century, Euler’s works do not meet the modern standards of mathematical rigor, but in most cases it is not a lot of work to formulate Euler’s proofs rigorously. 3His famous book on the calculus of variations [E65], the ﬁrst ever written on the subject, bearing the title "Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes, sive solutio problematis isoperimetrici lattissimo sensu accepti", even contains Methodus inveniendi in the title. 4Most books devoted to the Γ-function start from the integral representation. We mention [Ar15], [Fr06] as an example. [Ni05] is an exception. It starts the investigation from the difference equation satisﬁed by Γ(x), i.e. Γ(x + 1) = xΓ(x).

1 2 Main Thesis that were already discovered by Euler himself. More precisely, we will explain how Euler, the discoverer of (1.1) [E19]5, derived several different expressions for the Γ-function which are usually attributed to others.

1.1.2 Euler’s Idea concerning the Γ-function Euler found his expressions for the Γ-function basically from two different sources. 1. Interpolation theory 2. The functional equation Γ(x + 1) = x · Γ(x). But eventually they all boil down to the solution of the functional equation by different methods6. The first approach, outlined in [E212] (Chapter 16 and 17 of the second part) and [E613], is based 1 on difference calculus and led him to the Weierstaß product expansion of Γ(x) and also to the Taylor series expansion of log Γ(1 + x). Indeed, Euler implicitly uses the functional equation even in this approach. But we want to separate it from the other approaches, in which he tried to solve the functional equation explicitly and said so. Concerning the direct solution of the functional equation, first, Euler solved the functional equation satisfied by Γ(x),

Γ(x + 1) = xΓ(x), or, equivalently, the difference equation

log Γ(x + 1) − log Γ(x) = log x by applying the Euler-Maclaurin formula, which he had discovered in [E25] and proved in [E47]. We will discuss this below in section in section 1.7. Later, he tried a solution7 by conversion of the difference equation into a differential equation of inﬁnite order with constant coefﬁcients, see [E189], which he then wanted to solve by the methods he had developed in [E62], [E188]. Both ideas led him to the Stirling-formula for the factorial, i.e.,

xx √ x! = Γ(x + 1) = 2πx for x → ∞. ex The sign = is to understood as follows here: We have

Γ(x + 1) lim √ = 1. x→∞ xx ex 2πx In section 1.6 we will also explain that this formula follows from an asymptotic series. In [E123], he explained a method how to solve the functional equation by an educated guess. Euler applied it to the factorial in §13 of [E594], basically adding some more examples to those of [E123]; we will see in section 1.5 that his ideas lead to the integral representation (1.1).

1 x−1 5 R 1 To be completely precise, in [E19] Euler introduced the integral log t dt which transforms into the above 0 representation by the substitution t = e−u. 6This is the more natural approach, since the functional equation is one of the characteristic properties of the factorial. 7We say "tried", since the solution Euler found in this way is incorrect. He even realized this. But he tried to argue it away by another incorrect argument. This will be discussed in more detail in the corresponding section, i.e. in section 1.6.4.1. Euler and the Γ-Function 3

Finally, in [E652], he uses the functional equation to derive a product representation he ﬁrst stated without proof in [E19].

1.1.3 Organisation of the Paper

1.1.3.1 General Overview This paper is organised as follows: It can be subdivided into three parts, referred to as part I., part II. and part III., respectively. Part I. will be a brief historical overview on the Gamma function, explicitly stressing Euler’s contributions in section 1.2. Part II. contains the modern introduction of the Γ-function and the classification theorems in section 1.3. Furthermore, we dedicate a whole section to the rigorous solution of the difference equation F(x + 1) = xF(x) in section 1.4. Finally, part III. is devoted to Euler’s several approaches he used or could have used to arrive at the Γ-function (most of the time, he intended something entirely different and the Γ-function just was a special case). We will discuss Euler’s idea to solve general homogeneous difference equations with linear coefficients by an idea we will refer to as "moment ansatz" in section 1.5, and we will discuss how he solved the general difference equation by converting it into a differential equation of infinite order in section 1.6. Another idea of Euler was to derive solutions of the general difference equation by difference calculus, which we will also discuss in detail in section 1.8. We also devoted a complete section to the relation among the Γ- and B-function and how Euler found those connections in section 1.9. After this we will conclude and try to summarize Euler’s vast output in section 1.10. Given the time in which Euler wrote his papers, some of his arguments are not completely rigorous in the modern sense and some are even incorrect. Therefore, when necessary or appropriate, we will show, how his ideas can be formulated in a modern setting and at some points also give the rigorous proof. Sometimes, we will not give all the details, since this would simply take too long and would carry us too far away from our actual intention. We will always try to put Euler’s results and ideas in contrast to these modern ideas and it will turn out that Euler actually anticipated a lot that came after him (if understood in the modern context). Thus, we also added some sections just containing some historical notes. Furthermore, since this is mainly a paper on Euler’s works, we also included some quotes from his papers, translated from his original Latin into English. They will help to understand better Euler’s way of thinking and his persona.

1.1.3.2 Notation

Euler invented many of the modern notation, e.g., ∑ for a sum, f : x for a function of x etc. Nevertheless, most of the times he did not use the compact notation, but wrote things out x 1 1 1 explicitly, e.g., for ∑k=1 k he wrote 1 + 2 + ··· + x . When referring to Euler’s papers, we will retain his notation as closely as possible and only resort to a modern compact notation, if things become clearer that way. Euler also never used the symbol Γ(x) - This notation was introduced by Legendre in [Le26a] p. 365 - to denote the factorial nor did he write x!, his notation varies from paper to paper. We will always use the modern notation. Furthermore, the notion of limits as understood today did not exist at that time. Euler often speaks of inﬁnitely small or inﬁnitely large numbers. In this case, we will use the modern symbol 4 Main Thesis

limn→0 and limn→∞, respectively. At some occasions we also included scans of Euler’s writings (all available at the Euler Archive) such that the reader can compare Euler’s notation to the modern one. Finally, a reader going through the whole text will inevitably encounter some repetitions, which could not be avoided in the attempt to present each part independently from the previous ones. Finally, we do not always give a complete rigorous proof, but just explain the general idea how to prove a certain theorem. For, otherwise this would carry us too far away from our actual intention, presenting Euler’s contributions to the theory of the Γ-function. Euler and the Γ-Function 5

1.2 Historical Overview of the Γ-function

It will be convenient to give a short overview of the history of the Γ-function from its ﬁrst appearance in Euler’s 1738 paper [E19] to the axiomatic introduction in Bourbaki more than 200 years later [Bo51]. While doing this, we will mainly follow the history as described by Davis in his 1959 article on the Γ-function [Da59] and Sandifer’s articles in his book [Sa07]. But [Du91] and [Gr03] are also recommended. The ﬁrst goes into more detail on the investigations on the factorials before the Γ-function was actually introduced. The second discusses the correspondence between Euler, Goldbach and Bernoulli on the matter. We will talk about this after the general overview in section 1.2.2. We will especially stress Euler’s contributions and add some information, if necessary.

1.2.1 Brief historical Overview In his 1959 article [Da59], Davis described the history of the Γ-function from its discovery by Euler in 1729 [Eu29] in his correspondence with Goldbach8 to the axiomatic introduction in Bourbaki in 1951 [Bo51]. His article can be divided into the following phases: 1. Discovery by Euler in 1729 (Davis seems to be unaware of Bernoulli’s priority on the first definition. We will discuss this in more detail in the next section in section 1.2.2.) 2. Gauß’s investigations in 1812 on the hypergeometric series and the factorials in 1812 [Ga28]. 3. Development of the notion of an analytic function by Riemann and Weierstraß around 1850. 4. Application of analytic function theory to the Γ-function. 5. Classification as a transcendentally transcendent function by Hölder in 1887. 6. Uniqueness theorems, especially the Bohr-Mollerup theorem. We will also prove his theorem later in section 1.3.2.2. 7. Transcendence questions.

1.2.1.11 . Phase: Discovery by Euler As already mentioned, the Γ-function appeared in print the ﬁrst time in 1737 in [E19]. In that paper, Euler gave two expressions for the Γ-function. On the one hand, he found

2n 1 3n 2 4n 3 · · ··· = n!. 1 n + 1 2 n + 2 3 n + 3 This is equation (1) in [Da59]. In [E19], Euler does not give a proof for the formula, but just shows its correctness for n = 1, 2, 3. A proof had to wait until his 1755 book [E212], a completely elaborated version then also appeared in [E613]. This proof will be discussed in section 1.8.3.2. Therefore, we will not discuss this any further here. But we want to mention that it is more illustrative to write the product as follows:

8The first ever definition is due to Daniel Bernoulli, who in a letter written to Goldbach on the 6th of October 1729 [Be29] defines the factorial as:

x x−1 2 3 4 A n! = A + · · ··· 2 1 + x 2 + x 3 + x A − 1 + x for an infinitely large number A. Euler’s definition, also stated in [E19], appeared also in a letter to Goldbach [Eu29], written one week later than Bernoulli’s. However, Bernoulli’s definition did not catch on. 6 Main Thesis

11−x2x 21−x3x 31−x4x ∞ k1−x(k + 1)x x! = Γ(x + 1) = 11−x · · · · etc. = . + + + ∏ + x 1 x 2 x 3 k=1 x k This is also how Euler gave it already in [E19] and in his letter to Goldbach [Eu29], for reasons of convergence, which he then elaborated on in [E613]. In [E19], Euler also found the famous integral representation:

Z1 n! = (− log x)ndx for Re(x) > 0. 0 He arrived at it starting from the integral:

1 Z 1 · 2 ··· n xe(1 − x)ndx = (e + 1)(e + 2) ··· (e + n + 1) 0 by some clever substitutions and application of L’Hospital’s rule. We will discuss his ideas in more detail below in section 1.9. Davis [Da59] and Sandifer [Sa07] also give Euler’s original proof, i.e. also sticking to his notation. We will present it, when discussing the B-function in section 1.9.1.3.

1.2.1.22 . Phase: Gauß’s Investigations

In 1812 Gauß wrote the inﬂuential paper [Ga28], mainly on the hypergeometric series, but the last few paragraphs contain a discussion of the factorial. Gauß, in contrast to Euler, did not set the task for himself to ﬁnd a formula for the factorial, but he simply started his investigations from the product:

m!(m + 1)n n! = lim , m→∞ (n + 1)(n + 2) ··· (n + m) written in Davis’s notation; it is equation (3) in [Da59]. Although immediate from the above formula, Euler did not represent the Γ-function in this form in [E19], but arrived at this formula by a different reasoning in [E652]. We will see this below in section 1.9.1.3, when discussing the connection of the Γ-function to the B-function, i.e. the integral

Z1 B(x, y) = tx−1(1 − t)y−1dt for Re(x), Re(y) > 0. 0 Anyway, Gauß realized that the above product formula is easier to work with. First, since it is deﬁned for all non-negative integer numbers, whereas the integral is valid only for Re(y) > 0. Second, since it allows to prove some special identities of the Γ-function more easily than by using the integral representation. We mention the reﬂection formula:

π Γ(x)Γ(1 − x) = sin(πx) and the multiplication formula Euler and the Γ-Function 7

s x x + 1 x + 2 x + n − 1 (2π)n−1 Γ Γ Γ ··· Γ = n1−xΓ(x) , n n n n n as it is presented in modern form. Gauß proved this formula in [Ga28] using the product formula representation, which hence also often has its name attached to it. But in his 1772 paper [E421], Euler already heuristically arrived at a formula that can be shown to be equivalent to the Gauß multiplication formula. This will also be discussed below in section 1.9.6.

1.2.1.33 . Phase: Development of complex Analysis We do not want to go into much detail here. We just mention that in this phase functions of a complex variable became the focus of the studies and replaced the older notion of an analytical expression in the sense Euler introduced in his 1748 book [E101]. One of the most important results in complex analysis, due to the works of Riemann [Ri51] and Weierstaß [We78], is the concept of analytic continuation, i.e. the extension of a domain of the analytic function. Below we will see that the function defined via the integral of the Γ-function is an analytic function, if negative integer numbers are excluded, as is the function defined from the product representation. But on first sight, those expressions do not seem to have to do anything with one another at all. But using the functional equation of the Γ-function, the integral can be extended to the same domain as the product formula and thus the product formula can be seen to be just an analytic continuation of the integral representation. One might argue, that this is what Euler did in [E652] to obtain the product formula. This will also be discussed in more detail below in section 1.4.5. Furthermore, general results from complex analysis allow simple proofs of otherwise very difficult to prove identities.

1.2.1.44 . Phase: Formulas following from complex Analysis As already indicated, by extending the domain of a function from the real to complex numbers, many interesting identities emerge and moreover they are proved more easily. As an example, let us again mention the reﬂection formula π Γ(z)Γ(1 − z) = . sin(πz) Euler gave a proof in his 1772 paper [E421]. We will present his proof in section 1.9.4.2 and another one, he could have given, below in section 1.9.4.3. Euler’s proof is based on the product formula and the product for the sine, which he discovered in [E41]9 and proved in [E61] for the ﬁrst time. Using complex analysis, this identity is easily proved applying Liouville’s theorem. Another, more prominent example, is the functional equation of the Riemann ζ-function:

1 ζ(z) = ζ(1 − z)Γ(1 − z)2zπz−1 sin πz 2 with

9 + 1 + 1 + ··· = π2 This is the paper, in which he solved the Basel problem, i.e. summed the series 1 12 32 6 . 8 Main Thesis

1 1 ζ(z) = 1 + + + etc. 2z 3z This formula is attributed to Riemann who proved it in his famous 1859 paper [Ri59] using complex analysis. But Euler, using his deﬁnition of the sum of a divergent series which he gave in [E212], arrived at an equivalent identity in his 1768 paper [E352]. An equivalent formula also appears in the 1773 paper [E432]. Euler’s arguments are heuristic, as he also admitted. The story of Euler and the ζ-function is an interesting subject for itself, but we will not discuss it here. The reader is referred to [Ha48], who devoted a section to this discussion, and Ayoub’s 1974 article [Ay74], also reprinted in [Du07]. Let us mention that the functional equation for the ζ-function can easily be deduced from the results of Malmsten’s 1846 paper [Ma46], confer, e.g., [Ay13]. Returning to formulas for the Γ-function, let us mention the following formula found by Newman in 1848.

1 γz −z h z − z i = ze (1 + z) e 1 + e 2 ··· , where γ = 0.5772156649 ··· Γ(z) 2

γ is the Euler-Mascheroni constant. Confer the appendix, section 1.11.3, for more information on this constant. This formula is often referred to as Weierstraß product representation of the Γ-function, since it is a special case Weierstraß’s factorisation theorem proved in 1878 in [We78]. But this formula was found even earlier by Schlömilch in 1844 [Sc44]. We will review Weierstraß’s ideas in section 1.8.5 and show that it was in some sense anticipated by Euler in [E613] in section 1.8.5.2. Thus, it comes as no surprise that Newman’s or Schlömilch’s formula follows already from Euler’s. Finally, in this context we mention the recent paper [Pe20], in which another definition for the Γ-function is derived using complex analysis. This definition originates by generalising Laplace’s definition of the Γ-function given in [La82], i.e.

∞ 1 1 Z ex+iy = for Re(s) > 0, x > 0 Γ(s) 2π (x + iy)s −∞ to Hankel’s contour integral deﬁnition [Ha64], i.e.

1 1 Z = z−sezdz. Γ(s) 2πi η

Here, η is a curve from −∞ to +∞ surrounding the negative real axis R− =] − ∞, 0] and where for −s −s log z z ∈ C \ R− in z = e the principal branch of the logarithm is taken. This integral converges for all s ∈ C and thus deﬁnes an entire functions. Both formulas, Laplace’s and Hankel’s, also appear in [Ni05], but there Hankel’s formula is attributed to Weierstraß, who actually discovered it later than Hankel.

1.2.1.55 . Phase: Classification as transcendentally transcendental Function As the title of [E19] already suggests, the Γ-function is not an algebraic function. An algebraic function, already at the time of [E19], is a function that can be obtained by finitely many algebraic operations. Algebraic operations, according to Euler, were addition, subtraction, multiplication, division, raising to a natural power and taking root of integer powers. The simplest, or rather most familiar, non-algebraic or transcendental functions are exponentials, Euler and the Γ-Function 9 logarithms, sines and cosines. They cannot be constructed from finitely many algebraic operations but have to defined by a limit procedure somehow. For example, z n ez := lim 1 + . n→∞ n But all of the above functions share the property that they satisfy at least an algebraic differential equation. In 1887 [Ho87], Hölder showed that the Γ-function does not even satisfy an algebraic differential equation. Thus, the Γ-function is not only not an algebraic function, but has an even higher degree of transcendence than, e.g., sin(x), log x etc. The Γ-function is a so-called transcendentally transcendent function.

1.2.1.66 . Phase: Towards the axiomatic Introduction Up to this point we already mentioned several different expressions for the Γ-function, and initially there is no reason for them to be identical for all values. Euler claimed that all the different expressions he found while solving the interpolation problem of the factorial are identical. At least, it seems that he never addressed the issue, confer, e.g., [E368]. On the other hand, he was well aware that, given points of a function only at the natural numbers, this does not determine the function. On the contrary, there are still inﬁnitely many solutions. He stated this explicitly in his 1753 paper [E189]. We will come back to some of the results of this paper below, when we will discuss the solution of difference equations by converting them into solvable differential equations of inﬁnite order. The non-uniqueness of the interpolation theorem for the Γ-function raises the question, what makes it unique among all other solutions for the same problem. In other words: Which properties determine the Γ-function uniquely? The way to answer this question is an example for the Methodus inveniendi, whence we want to describe a possible line of thought to get to the characteristics of the function. As mentioned above, the sole condition

f (n) = (n − 1)! ∀ n ∈ N is not enough to conclude f (x) = Γ(x) for all x ∈ C. For,

f (x) = Γ(x)p(x), with p(x) being a periodic function with period 1 and p(1) = 1, is also a solution to the same question. Moreover, Hadamard in 1894 [Ha94] found the following nice solution:

" − # 1 d Γ 1 x f (x) = log 2 . ( − ) x Γ 1 x dx Γ 1 − 2 One can check that

1 f (x + 1) = x f (x) + and f (1) = 1. Γ(1 − x) This also f (n) = (n − 1)! ∀ n ∈ N. Those two examples show that we need additional conditions. First, it seems necessary to demand 10 Main Thesis f (1) = 1 and f (x + 1) = x f (x) for all x ∈ C. This excludes Hadamard’s solution, as it does not satisfy the recurrence relation. Moreover, one has to exclude solutions just being a product of the Γ-function by a function with period 1. One possible way to do this, was found by Weierstraß, who in his 1856 paper [We56] introduced the Γ−function as the unique function satisfying: f (1) = 1 and f (x + 1) = x f (x) and

f (x + n) lim = 1 ∀x ∈ C. n→∞ (n − 1)!nx In his paper, he proves that the function satisfying those conditions is unique and mentioned that he was using Gauß’s 1812 paper [Ga28] and the product representation for the Γ-function as motivation for this deﬁnition. Although this formulation solves the uniqueness question, as we will also show below in section 1.4, mathematicians kept looking for other, more "natural" properties, as Davis calls it. In other words, we want to replace the third of Weierstraß’s conditions by another one, which is somehow more intuitive. Trying so, one immediately notices the Γ-function to be convex for x > 0. But this is still not enough, as the following example, equation (35) in [Da59], shows. Deﬁne    1   for 0 < x ≤ 1,  x      1 for 1 ≤ x ≤ 2,     f (x) = (x − 1) for 2 ≤ x ≤ 3, .       (x − 1)(x − 2) for 3 ≤ x ≤ 4,       etc. 

This function is convex and has the two other two properties. The correct formulation had to wait until 1922 [Mo22], when the Bohr-Mollerup theorem was formulated. We will meet it again below several times, but let us already formulate it here: The Γ-function is the only function deﬁned for x > 0 which is positive, satisﬁes Γ(1) = 1 and the functional equation xΓ(x) = Γ(x + 1) and is logarithmically convex. In [Bo51], the Γ-function is introduced as the function satisfying all the above properties. The uniqueness is then proved afterwards.

1.2.1.77 . Phase: Transcendence Questions Having established the higher transcendence of the Γ-function, it is only natural to ask, whether the values of the Γ−function at rational values are always transcendental and if so, if they can be expressed in terms of already known constants. We will meet the Euler-Mascheroni constant γ, defined as " # n 1 γ := lim − log(n) n→∞ ∑ k=1 k again in section 1.4.3.3, and discuss Euler’s contribution to the constant in the appendix in section 1.11.3. It was discovered by Euler in his 1740 paper [E43] and appeared in Newman’s infinite Euler and the Γ-Function 11 product formula for Γ(x). To this day, it is not known whether γ is rational or not. Concerning values of the Γ-function itself, we know that Γ(n + 1) = n! and thus a natural number. From the reflection formula, we conclude

1 √ Γ = π 2 and thus transcendental. Euler already found this value in [E19] from his product formula which reduces to the Wallis product formula for π. And this encouraged him to look for an integral representation for Γ(x). More on this below in section 1.9.1.1. Using results on periods as deﬁned in [Ko01] and the relation of the Γ−function to the B-function, which√ we will also study below and mention Euler’s contributions to this, one can conclude that 1 4 1 Γ 4 π is a transcendental number, whereas it is not known whether only Γ 5 is rational or not. 1 1 1 3 5 But in 1975 Chudnovsky [Ch84] showed that Γ 6 , Γ 4 , Γ 3 , Γ 4 , Γ 6 are transcendental and algebraically independent from π.

1.2.2 More detailed Discussion on the Discovery of the Γ-function First, we want to stress here again that the ﬁrst representation for the factorial was given by Daniel Bernoulli, not by Euler. Confer [Dut91], [Gr03]. In [Be29] Bernoulli gave the formula:

x x−1 2 3 4 A x! = A + · · ··· 2 1 + x 2 + x 3 + x A − 1 + x for an infinitely large number A. Euler’s wrote the letter containing his definition to Goldbach one week later. Euler’s definition is just the first that appeared in print. Euler and Daniel Bernoulli were friends since their childhood and communicated about math via letters and even both had positions in the Imperial Russian Academy of Sciences in Saint Petersburg at 1729 and met daily, probably discussing various things, including the problem of interpolating the factorial. Thus, it is natural to ask, whether and, if so, how they influenced each other concerning the interpolation problem of the factorial. This is rather difficult to answer, since they are not explicit about this in their respective letter to Goldbach. But, for the sake of completeness, we mention, that in [Ju65], a book containing the letters exchanged between Euler and Goldbach from 1729-1764, in the footnote on page 21 it is stated that Euler knew the contents of the letters from Bernoulli to Goldbach and thus also Bernoulli’s formula. Bernoulli just stated the formula at the end of [Be29] without any explanation.

Euler opened his letter with the formula

1 2x 2 3x 3 4x 4 5x x! = · · · ··· 1 + x 1 2 + x 2 3 + x 3 4 + x 4 but dis also not explain how he arrived at this formula. In fact, a proof had to wait until his 1755 book [E212]. 12 Main Thesis

Figure 1.1: Letter from Bernoulli to Goldbach Part of Bernoulli’s letter [Be29] written 6th October 1729 to Goldbach. It contains the ﬁrst ever deﬁnition of the Γ-function.

Figure 1.2: Letter from Euler to Goldbach - ﬁrst Part First page of Euler’s letter [Eu29] written to Goldbach 13th October 1729 - one week letter than Bernoulli’s. The ﬁrst page already contains the product representation. Euler and the Γ-Function 13

Figure 1.3: Letter from Euler to Goldbach - second Part Part of the second page of Euler’s letter [Eu29] written to Goldbach 13th October 1729. It already contains what would become known as the Gauß’s product representation of the factorial.

In the same letter, he also said that:

1 · 2 · 3 ··· n m! = (n + 1)m (1 + m)(2 + m)(3 + m) ··· (n + m) for inﬁnite m.

He proved this later in his 1769 paper [E368] and again in his 1793 paper [E652]. Nevertheless, as mentioned, this expression is attributed to Gauß, who used it as a definition for the factorial in his influential 1812 paper [Ga28]. In his letter, concerning his and Bernoulli’s expressions, Euler wrote, translated from the Latin original: I communicated these discoveries to Bernoulli, who in a peculiar way found almost the same last term10, different from mine in that regard that he uses another power instead of (n + 1)m, determining which he maybe took into account the neglected factors. This sentence reveals two things: First, Euler knew about Bernoulli’s formula, most probably from private communication. For, the subject was never mentioned in letters they exchanged. Second, he did not know how Bernoulli arrived at his result. It is speculation, whether Euler found his result influenced by Bernoulli’s formula and, not knowing how the latter arrived at his formula, found his expression or whether he had found it independently. The letter seems to indicate the first option. Dutka in his overview article [Dut91] argues in favour of this possibility, stating the opinion that Bernoulli obtained his formula first and talked to Euler about it and then Euler, after this, found his solution. This is also endorsed by [Gr03]. But one might argue that Euler’s second formula is more natural, since it can be easily

10 x x−1 m Euler means the factor A + 2 in Bernoulli’s formula, in contrast to the factor (n + 1) in his formula. 14 Main Thesis found from the recursive property of the factorial and taking the limit. This is also how Euler then proved it in [E652], as we will see below in section 1.4.5. Unfortunately, the question how Bernoulli actually found his formula, remains unsettled. But, having already stated the Bohr-Mollerup theorem above, it is easily checked that also Bernoulli’s expression is another expression for the Γ-function; Γ(x + 1), to be precise. Applying the Bohr-Mollerup theorem, we also see that we get another expression for the Γ-function:

2 3 4 n Γ(x + 1) = lim A(x, n)x−1 · · ··· n→∞ 1 + x 2 + x 3 + x n − 1 + x if just A(x, n) is logarithmically convex for x > 1 and for n ∈ N and

lim A(n, 0) · n = 1. n→∞ x And Bernoulli’s choice A(n, x) = n + 2 obviously satisﬁes those requirements. But, e.g., also x−1 xk n + 2 for a natural number k works as well. Bernoulli probably preferred his choice, since it converges rapidly to its limit. Thus, the particular choice for A(n, x) is probably a result of extensive calculations trying out several similar formulas.

1.2.3 Overview on Euler’s Contributions Having now described the history of the Γ-function in general, we want to give an overview more focused on Euler’s work. To this end, let us ﬁrst mention that most historical overviews, e.g., [Da59] mostly consider only Euler’s ﬁrst paper on the Γ−function [E19] and discuss mainly the two presentations of the Γ-function Euler gave in that same paper, i.e. his product representation and the integral representation. But the Γ−function appears in Euler’s work at several places throughout his career and he devoted several papers to it. Additionally, as we will show below in section 1.5, some of his results he discovered while investigating other subjects can also be applied to the Γ-function, but Euler did not always do so. To give an example, in his 1750 paper [E123], he considered difference equations of the form:

(αx + a) f (x + 1) = (βx + b) f (x) + (γx + c) f (x − 1) but not does not apply it to the functional equation of the Γ−function, being a special case of the above equation. He did this only in the 1785 paper [E594] on the same subject in section 1.5.2.1. Both papers are actually devoted to continued fractions and the results on the Γ-function are just a byproduct. Thus, it will be convenient to give a comprehensive list of Euler’s papers on the Γ-function and group them into categories, whether they are directly devoted to the Γ-function or the results are just byproducts of other investigations. Furthermore, in the second part, we want to distinguish, whether Euler realized the consequences from his investigations or not, since on some occasions he did, on others he did not.

1.2.3.1 Euler’s Papers on the Γ-function In total, Euler wrote 13 papers either directly devoted to the Γ-function or containing insights on the Γ−function as byproducts. The papers devoted to the Γ-functions are: [E19], [E652], [E661], Euler and the Γ-Function 15

[E421], [E662], [E816], [E368]. The papers containing results on the Γ-function as a byproduct are: [E122], [E123], [E189], [E613], [E594], chapter 4, 16 and 17 of the second part of the book [E212]. For the sake of convenience, we want to give a brief summary of each paper and its title already here, although we will mention them in more detail in the corresponding sections. On this occasion, it will be convenient to mention Eneström’s 1910 book [En10], listing all of Euler’s writings in chronological order with a short description of the content. Eneström attributed a number from 1 to 866 to each of Euler’s papers or books, referred to a the Eneström number. The smaller the Eneström number is, the earlier the paper was published. We always cite Euler’s papers by their Eneström number.

1.[E19]: De progressionibus transcendentibus seu quarum termini generales algebraice dari nequeunt, published in 1738. This is Euler’s ﬁrst paper on the Γ-function. As mentioned above in section 1.2.1, Euler found the integral representation and the product representation.

2.[E122]: De productis ex inﬁnitis factoribus ortis, published in 1750. Euler studied B-function integrals and derives several relations among them.

3.[E123]: De fractionibus continuis observationes, published in 1750. This is Euler’s longest paper on continued fractions he ever wrote. It contains many topics, including the transformation of series into continued fractions and the solution of the Riccati differential equation via continued fractions. Additionally, Euler described a method to solve homogeneous difference equations with linear coefﬁcients. See section 1.5.

4.[E189]: De serierum determinatione seu nova methodus inveniendi terminos generales serierum, published in 1753. Euler solved several difference equations by converting them into a differential equation of inﬁnite order and solving those differential equations then. The factorial is one example he considered. We devote a whole section to the explanation of his ideas. See section 1.6.

5.[E212]: Institutiones calculi differentialis cum eius usu in analysi ﬁnitorum ac doctrina serierum, published 1755: This is Euler’s book on differential calculus. It contains many topics. We mention the introduction of differential quotients, the Euler-Maclaurin summation formula, and many transformation formulas for series as examples. We will mainly focus on his interpolation theory based on difference calculus. See section 1.8.

6.[E368]: De curva hypergeometrica hac aequatione expressa y = 1 · 2 · 3 ··· x, published 1769. This is an overview article on the Γ-function. Euler stated all formulas he ever found for x! - most of them without out proof -, evaluated the factorial and its derivative for special values.

R − m 7.[E421]: Evolutio formulae integralis x f 1dx (log(x)) n integratione a valore x = 0 ad x = 1 extensa, published 1772. This is an overview paper on the integral representation of the Γ-function. It contains many results 16 Main Thesis

- with proofs - e.g., the relation among the Γ- and B-function, the derivation of the sine product formula, even formulas attributed to Gauß. All this will be discussed in 1.9.

8.[E594]: Methodus inveniendi formulas integrales, quae certis casibus datam inter se teneant rationem, ubi sumul methodus traditur fractiones continuas summandi, published 1785. This paper can be considered as an addendum to [E123]. It contains several examples not discussed in the latter, including the Γ-function. Thus, [E594] contains a derivation of the integral representation of the Γ-function from its functional equation.

9.[E613]: Dilucidationes in capita postrema calculi mei differentalis de functionibus inexplicabilibus, published in 1787. This is an addendum to chapter 16 and 17 in [E212]. It is a more detailed discussion of his solution of the interpolation problem of a sum using difference calculus. He applied his general formulas to the factorial this time.

10.[E652]: De termino generali serierum hypergeometricarum, published in 1793. This paper contains the proof of the Gaußian representation of the factorial. The proof is based on the functional equation. We discuss this in 1.4.5.

11.[E661]: Variae considerationes circa series hypergeometricas, published in 1794. Euler uses the Euler-Maclaurin summation formula to study the factorial.

R 1 n 12.[E662]: De vero valore formulae integralis ∂x log x a termino x = 0 usque ad terminum x = 1 extensae, published in 1894, but presented already in 1776. In this paper, Euler tried to evaluate the integral for several rational numbers. Due to the transcendence of the Γ-function, his efforts are eventually in vain. Nevertheless, he was able to express products of several values in terms of familiar quantities, like π and sin πx.

13.[E816]: Considerations sur quelques formules integrales dont les valeurs peuvent etre exprimees, en certains cas, par la quadrature du cercle, published in 1862. This paper was written in the same spirit as [E662].

Having described all papers briefly, we want to note that the papers [E19], [E368] and [E421] together contain all formulas and insights on the Γ-function that Euler ever made and had. Thus, we provided a translation of the Latin originals into English of all three papers11. They are found in the appendix. As mentioned, the 1738 paper [E19] was the first appearance of the Γ-function in print ever. The 1772 paper [E421] and the 1769 paper [E368] are rather overview papers. The first is devoted to the integral representation, whereas the second focuses more on the product representations. Thus, they are elaborations on both the representations that appeared already in [E19]. Unfortunately, the proofs of the formulas are not always found in those same papers, but in others. In in some cases, Euler even only gave a heuristic proof.

11Meanwhile, all 12 (of 13) papers, which were written in Latin, have been translated into English by the author of this thesis. The translations are available at online the Euler-Kreis Mainz. Euler and the Γ-Function 17

1.2.3.2 Euler’s Results on the Γ-function Here, we want to give a list of Euler’s results on the Γ-function together with the paper, in which it can be found. How he obtained his result, will be explained below in the corresponding sections in more detail.

1. Integral Representation: Euler found the function deﬁned as:

1 Z 1 n log dx x 0 as a solution to the problem of interpolating of the factorial. In [E19] he obtained it rather indirectly from a B-function integral. We will see this below in section 1.9.1.3. But confer also [Da59], [Sa07]. The paper [E123], actually devoted to the continued fractions, contains the tools to obtain the integral formula directly. He did this in the paper [E594], also a paper on continued fractions.

2. Euler Product Representation: By this we mean the formula

1 2x 2 3x 3 4x 4 5x x! = · · · ··· 1 + x 1 2 + x 2 3 + x 3 4 + x 4 which he ﬁrst stated without proof in [E19]. It is also stated in the ﬁrst paragraph of [E368]. But neither of those papers contains a proof. A proof had to wait until 1755 in chapter 16 and 17 of the second part of his book from 1755. There, he solved interpolation problems via difference calculus. The idea is presented in more detail in [E613]. The above formula is just a corollary of more general formulas. Below in section 1.8.4 we will discuss Euler’s ideas from [E613] and show, how they anticipated the fundamental ideas of Weierstraß-products in section 1.8.5.

3. Gaußian product representation, i.e.

1 2 n x! = lim · ··· (n + 1)x. n→∞ 1 + x 2 + x n + x This formula is often attributed to Gauß, since it was the starting point of his investigations on the factorial in the influential paper [Ga28]. But Euler, arguing via infinitely large numbers, proved this formula in [E652]. We will discuss later in section 1.4.5 how he found this product. Additionally, Euler gave this definition in his 1729 letter to Goldbach [Eu29], as we have already seen in section 1.2.2.

4. Stirling formula for Γ(x + 1): The Stirling formula says

√ xx x! = Γ(x + 1) = 2πx for x → ∞. ex This formula is attributed to Stirling for his 1730 investigations [St30]. But we will have to say several things concerning priorities below in section 1.6.4.1. Anyhow, Euler proved this formula 18 Main Thesis on two occasions. First, in his 1755 book [E212]. There, the formula is derived as a consequence from his more general results on the Euler-Maclaurin summation formula. We will represent his proof, when discussing the Stirling formula in section 1.6.4.1.[E661] is then solely devoted to investigations of slight generalisations of the factorial functions via the Euler-Maclaurin formula. He also stated the Stirling formula in [E368]. He also tried to give another proof in [E189]. In this paper, he solved difference equations including f (x + 1) − f (x) = log(x), which is solved by log Γ(x) of course, by converting them into differential equations of inﬁnite order ﬁrst. But his general investigations contain a conceptual error and thus his proof is incorrect. We will elaborate on this below in section 1.6.2.3, discuss his error and give a method how to correct his arguments.

5. Series expansions for log Γ(x + 1): Euler found several series expansions for the logarithm of the Γ−function, he gave a list in [E368]. Let us mention two striking examples:

1 1 1 log Γ(x + 1) = −γx + x + x + x + x + etc. 2 3 4 x x x x − log 1 + − log 1 + − log 1 + − log 1 + + etc. 1 2 3 4

Taking the exponentials and applying the functional equation Γ(x + 1) = xΓ(x), we arrive at Newman’s formula which we saw above already

1 γz −z h z − z i = ze (1 + z) e 1 + e 2 ··· . Γ(z) 2 As a second example, we mention the Taylor series expansion of log Γ(x + 1):

ζ(2) ζ(3) ζ(4) log Γ(x + 1) = −γx + x2 − x3 + x4 − etc. 2 3 4 where

∞ 1 ( ) = ζ z ∑ z . k=1 k A proof by Euler can be found in chapter 16 and 17 of the second part of [E212].

6. Reﬂection formula π = Γ(x)Γ(1 − x). sin(πx) He used it in his investigations on the ζ-function in [E352], but provided a proof based on the product formula of the Γ−function and the sine product formula, discovered in [E41] but proved later in [E61], just in [E421]. Below, in section 1.9.4.2, we will present this proof and also discuss Euler’s proof of the sine product formula in section 1.9.4.1.

7. Relation among the Γ− and B−function: Euler and the Γ-Function 19

Γ(x)Γ(y) B(x, y) = . Γ(x + y) Euler stated this formula on several occasions, e.g. [E421], [E816], [E662]. But he never gave a satisfactory proof.

8. Gaußian multiplication formula:

n−1 x x + 1 x + n − 1 (2π) 2 Γ Γ ··· Γ = · Γ(x) x− 1 n n n n 2 Euler did not state it in the above form, but, in [E421], he gave the formula: s h m i m 1 2 3 n − 1 = n nn−m · 1 · 2 · 3 ··· (m − 1) ··· . n n m m m m Here is the explanation of Euler’s notation from [E421]:

Z1 p 1 p −1 q −1 1 p q [x] = x! = Γ(x + 1) and = dxx n (1 − x) n = B , . q n n n n 0 In section 1.9.6, we will show that this formula is equivalent to the Gaußian formula. Euler did not prove this formula and only obtains it heuristically by pattern recognition.

p 9. Expression of Γ q in terms of integrals of algebraic functions: In [E19] and [E122], Euler gave the formula

1 s Z p 2p 3p 4p qp (− log x) q dx = q 1 · 2 · 3 ··· p + 1 + 1 + 1 ··· + 1 q q q q 0 v u 1 1 1 1 1 uZ p Z p Z p Z p Z p uq q 2 3 q 3 4 q 4 5 q q−1 q q ×t dx(x − xx) · dx(x − x ) · dx(x − x ) · dx(x − x ) ··· dx(x − x ) . 0 0 0 0 0 20 Main Thesis

1.3 Short modern Introduction to the Γ-Function

We brieﬂy mention the modern deﬁnition of Γ(x) following [Fr06] (pp. 194-197). We start from the integral representation and derive the characteristic properties from it. Furthermore, we will obtain two equivalent characterisations of the Γ-function, based on Wielandt’s theorem in section 1.3.2.1 and the Bohr-Mollerup theorem in section 1.3.2.2.

1.3.1 Definition and simple Properties 1.3.1.1 Definition Definition 1.3.1 (Γ-integral). We define the Γ-function as a function in the complex plane as the following integral

Z∞ Γ(z) := tz−1e−tdt. 0 Here tz−1 := e(z−1) log(t), log t ∈ R, Re(z) > 0. We have the following simple theorem: Theorem 1.3.1. The Γ-integral

Z∞ Γ(z) := tz−1e−tdt 0 converges absolutely for Re(z) > 0 and represents an analytic function on the domain. The derivatives are given (for k ∈ N) by

Z∞ Γ(k)(z) = tz−1(log t)ke−tdt. 0 Proof. We split the Γ integral into the two integrals

Z1 Z∞ Γ(z) = tz−1e−tdt + tz−1e−tdt 0 1 and use the relation

z−1 −t x−1 −t t e = t e where we wrote x for Re(z). Let us consider both integrals separately. In general, for each x0 > 0 there is a number C > 0 with

x−1 t t ≤ Ce 2 ∀ x with 0 < x ≤ x0 and t ≥ 1. Thus, the integral

Z∞ tz−1e−tdt 1 Euler and the Γ-Function 21 converges absolutely for all z ∈ C. For the other integral, we use the estimate

z−1 −t x−1 t e < t for t > 0 and the existence of the integral

1 Z 1 dt for s < 1. ts 0 From these estimates it follows that the sequence of functions

n Z z−1 −t fn(z) := t e dt

1 n converges uniformly to Γ for n → ∞. Therefore, Γ is an analytic function. The formula for the k−th derivative follows from the application of the Leibniz rule (for differentiation) and then taking the limit n → ∞.

1.3.1.2 Simple Properties We have Theorem 1.3.2 (Elementary properties of the Γ-integral). The Γ-function can be analytically continued to the whole complex plane except at the points

z ∈ S := {0, −1, −2, −3, ···} and at C \ S satisﬁes the functional equation:

Γ(z + 1) = zΓ(z). All singularities are poles of ﬁrst order with the residues:

(−1)n Res(Γ, −n) = n! Proof. We show the functional equation ﬁrst. Obviously, we have

Z∞ ( ) = −t = − −t∞ = Γ 1 e dt e 0 1. 0 By integration by parts one arrives at the functional equation

Γ(z + 1) = zΓ(z) for Re(z) > 0. Using the functional equation iteratively, we ﬁnd

Γ(z + n + 1) Γ(z) = . z · (z + 1) ··· (z + n) The right-hand side of the equation has a large domain where it can be deﬁned, i.e. 22 Main Thesis

Re(z) > −(n + 1) and z 6= 0, −1, −2, −3, ··· , −n. Therefore, the above equation is an analytic continuation of Γ into a larger domain. Finally, let us consider the residues. Using the functional equation, we have

Γ(1) (−1)n Res(Γ; −n) = lim (z + n)Γ(z) = = . z→−n (−n)(−n + 1) ··· (−1) n!

1.3.2 Classiﬁcation Theorems The Γ-function was invented by Euler to interpolate the factorial in 1738 [E19]. The integral representation obviously fulﬁlls this task, since Γ(n + 1) = n!. The factorial has these two properties 0! = 1 and n! = n(n − 1)!. Therefore, this automatically raises the question, whether the Γ-function is the only holomorphic function with Γ(z + 1) = zΓ(z) and Γ(1) = 1. As already mentioned in section 1.2.1.6, the answer to this question is no, since, e.g.,

f (z) := (1 + sin(2πz))Γ(z) also has these two properties. Below we will encounter several other expressions also satisfying the functional equation and f (1) = 1 and above we already did in section 1.2.1.6. Therefore, it will be useful to have theorems that tell us immediately that the new expression is indeed the Γ-function without showing the equality to the integral representation directly. This is provided by classification theorems. They state that the Γ-function can be uniquely defined by the two obvious properties Γ(1) = 1 and Γ(z + 1) = zΓ(z) and an additional third one. We will present two theorems, Wielandt’s theorem and the Bohr-Mollerup theorem. In Bourbaki [Bo51], the Bohr-Mollerup theorem is the starting point for theory of the Γ-function. There, one does not start from a specific representation.

1.3.2.1 Wielandt’s Theorem Wielandt’s Theorem is one possible characterisation of the Γ-function. Wielandt’s original proof can be found in his collected papers [Wi96]. Other proofs can be found, e.g., in the books [Kn41] (pp. 47-49) and [Fr06] (pp. 198-199) which we will present here, and in the paper [Re96]. We have: Theorem 1.3.3 (Wielandt’s Theorem). Let D ⊆ C be a domain containing the vertical strip

1 ≤ x < 2. Let f : D → C be a function with the following properties: 1) f is bounded in the vertical strip 2) We have

f (z + 1) = z f (z) for z, z + 1 ∈ D Then we have:

f (z) = f (1)Γ(z) for z ∈ D. Euler and the Γ-Function 23

Proof. Applying the functional equation, it is easily seen that the function f can be analytically continued to the whole complex plane except at the points:

z ∈ S = {0, −1, −2, −3, ···} and satisﬁes

f (z + 1) = z f (z). All z ∈ S are either poles of ﬁrst order or removable singularities, and we have:

(−1)n Res( f ; −n) = f (1). n! Therefore, the function h(z) := f (z) − f (1)Γ(z) is an entire function. Furthermore, it is bounded in the vertical strip 0 ≤ x ≤ 1, which follows immediately from the boundedness in the strip 1 ≤ x < 2 and the functional equation for | Im(z)| ≥ 1. The domain | Im(z)| ≤ 1, 0 ≤ Re(z) ≤ 1 is compact. We want to use Liouville’s theorem and observe that from the functional equation for h, i.e. h(z)z = h(z + 1), if we deﬁne

H(z) := h(z)h(1 − z), we ﬁnd H(z + 1) = −H(z). But the strip 0 ≤ x ≤ 1 is not changed under the transformation z → 1 − z. Thus, H is bounded on this strip and, because of the periodicity, it is bounded on C. Therefore, Liouville’s theorem implies that H is constant. But h(1) = 0, so H = 0 and hence also h = 0 for all z ∈ C.

The Γ-integral obviously satisﬁes all three properties. We will see this below, when we ﬁnd the integral representation from the moment ansatz in section 1.5.

1.3.2.2 Bohr-Mollerup Theorem The Bohr-Mollerup Theorem, first proved in the 1922 book [Mo22], also states that the Gamma function can be uniquely classified by three properties. In other words, aside from the two obvious ones Γ(x + 1) = xΓ(x), Γ(1) = 1, we, as in the case of Wielandt’s theorem, need one additional property. This is the so-called logarithmic convexity. For the sake of completeness, let us define convexity first and show that Γ(x) has the property of logarithmic convexity, before we get to the theorem. Definition 1.3.2 (Logarithmic Convexity). Let X, Y be open subsets of the real numbers R. Further, let f : X → Y be a function. Then, f is called convex, if the following inequality holds:

f (tx + (1 − t)y) ≤ t f (x) + (1 − t) f (y) ∀ x, y ∈ X, ∀t ∈ [0, 1] Furthermore, f is called logarithmically convex , if log f (x) is convex. Let us state a theorem which can be used if f additionally is twice continuously differentiable. Theorem 1.3.4. If the second derivative of a twice continuously differentiable function is always ≥ 0 in the interval (a, b), then the function f is convex in this interval. The converse of this theorem is also true. The proof can be found in every book on analysis of one variable, one can also ﬁnd a proof in [Ar15] (pp. 6-7). We will need the following corollary. 24 Main Thesis

Corollary 1.3.4.1. If f : R → R is twice continuously differentiable and the following inequalities are satisﬁed for all x ∈ (a, b)

f (x) > 0, f (x) f 00(x) − ( f 0(x))2 ≥ 0, then log f is convex, i.e. f is logarithmically convex, in this interval.

For a proof one just has to apply the previous theorem to log f . Further, we have

Corollary 1.3.4.2. The sum of two logarithmically convex function is also logarithmically convex.

We will not prove this statement here. For a proof the reader is referred to [Ar15]. Instead, we want to go over to the logarithmic convexity of Γ. For this, consider f (t, x), continuous in both variables x and t. Let a ≤ t ≤ b and x live in another interval. If f (t, x) now is logarithmically convex for all t and twice continuously differentiable with respect to x, deﬁne:

b − a F (x) = h f (a, x) + f (a + h, x) + f (a + 2h, x) + ··· + f (a + (n − 1)h, x), h = n n

Then, Fn(x) is also logarithmically convex for all n ∈ N. Therefore, also

b Z lim Fn(x) = f (t, x)dt n→∞ a is logarithmically convex. This also holds for improper integrals, if the integral exists. Therefore, we have:

Theorem 1.3.5. The Γ- function, given as

Z∞ tx−1e−tdt, 0 is logarithmically convex for x > 0.

Now, having mentioned all this in advance, we can ﬁnally state the Bohr-Mollerup theorem.

Theorem 1.3.6 (Bohr-Mollerup Theorem). If a function f : R+ → R satisfies the three properties 1) f (x + 1) = x f (x) 2) f is logarithmically convex on the whole domain where it is defined 3) f (1) = 1, it is identical to the Γ-function in the region where it is defined.

Proof. We have shown that Γ satisﬁes all conditions. Therefore, let f be another function with the above properties. From the functional equation we ﬁnd

f (x + n) = (x + n − 1)(x + n − 2) ··· (x + 1)x f (x). Since f (1) = 1 we have f (n) = Γ(n) ∀n ∈ N. We only need to show f = Γ for the interval 0 < x ≤ 1, because of the functional equation. Thus, let x be a number in that interval and n a natural number ≥ 2. Then, we have the following inequality Euler and the Γ-Function 25

log( f (−1 + n)) − log( f (n)) log( f (x + n)) − log( f (n)) log( f (1 + n)) − log( f (n)) ≤ ≤ (−1 + n) − n (x + n) − n (1 + n) − n which follows from the logarithmic convexity. We can simplify the last equation:

log( f (x + n)) − log( f (n)) log(n − 1) ≤ ≤ log n (x + n) − n or

log((n − 1)x(n − 1)!) ≤ f (x + n) ≤ log(nx(n − 1)!). Using the above equation for f (x + n):

(n − 1)x(n − 1)! nx(n − 1)! nxn! x + n ≤ f (x) ≤ = · . x(x + 1) ··· (x + n − 1) x(x + 1) ··· (x + n − 1) x(x + 1) ··· (x + n) n

Since we assumed n ≥ 2, we can replace n by n + 1 and ﬁnd:

nxn! nxn! x + n ≤ f (x) ≤ · . x(x + 1) ··· (x + n) x(x + 1) ··· (x + n) n Therefore,

n nxn! f (x) ≤ ≤ f (x). n + x x(x + 1) ··· (x + n) Taking the limit n → ∞:

nxn! f (x) = lim . n→∞ x(x + 1) ··· (x + n) Since the function f only had to satisfy the three conditions in the theorem and was arbitrary otherwise, we conclude f (x) = Γ(x).

We have the following corollary:

Corollary 1.3.6.1. nxn! Γ(x) = lim . n→∞ x(x + 1) ··· (x + n) We will ﬁnd other ways to get to this product representation below. It is interesting that it follows directly from the proof. Additionally, we already pointed out in section 1.2.1.2 that Gauß in 1812 [Ga28] used the last corollary as a deﬁnition for the Γ-function. However, we already saw in section 1.2.2 that this product formula had already been discovered by Euler in [Eu29]. 26 Main Thesis

1.4 Solution of the Difference Equation F(x + 1) = xF(x)

In this section we will solve the functional equation in general and from there descend to the Γ-function. This will lead us to the Weierstraß product expansion of the Γ-function. Our exposition follows [Ni05].

1.4.1 Weierstraß’s Deﬁnition of the Γ-function It was Weierstraß’s [We56] idea to deﬁne the Γ-function as solution of the difference equation

F(x + 1) = xF(x) with the additional condition12

F(x + n) lim = 1. n→∞ (n − 1)!nx The above condition, as we will see soon, excludes solutions of the form Γ(x)p(x) with p a periodic function with period 1. Anyhow, in this section we want to solve the difference equation in general and want to show that it indeed deﬁnes the Γ-function as claimed.

1.4.2 A Remark concerning the Solution of the Difference Equation Let us begin with the following remark:

Remark 1. In order to solve the difference equation F(x + 1) = xF(x), we essentially only need one particular solution.

Proof. For, let F1(x) and F2(x) be two particular solutions of the difference equation. Then, one has

F (x + 1) xF (x) F (x) 2 = 2 = 2 , F1(x + 1) xF1(x) F1(x) i.e. the quotient of the two solutions is a periodic function with period +1. In other words, if F1(x) is a solution of the difference equation, then every other solution F2(x) is connected to it by

F2(x) = ω(x)F1(x) with ω(x + 1) = ω(x).

1.4.3 General Solution of the Equation F(x + 1) = xF(x)

1.4.3.1 Introduction of an Auxiliary Function Keeping the remark of the previous section in mind, we can now proceed to ﬁnd a solution of the difference equation and show that the Γ-function is actually the only one. For this aim, let us introduce the following function:

12Weierstraß added the condition F(1) = 1 which, however, is not necessary to deﬁne the Γ-function uniquely. Euler and the Γ-Function 27

Deﬁnition 1.4.1. We deﬁne a function13 Σ : C \{0, −1, −2, −3, · · · } → C by the sum

∞ 1 1 Σ(x) := ∑ − . s=0 s + 1 x + s This series is easily seen to converge uniformly. Thus, we are allowed to integrate it term by term with respect to x, provided the path of integration is of ﬁnite length and does not pass through any of the poles. Furthermore, we have

Theorem 1.4.1. Σ satisﬁes the functional equation

1 Σ(x + 1) = Σ(x) + . x Proof. Consider the difference Σ(x + 1) − Σ(x); it reads

1 1 1 1 ∑∞ − − ∑∞ − s=0 s + 1 x + 1 + s s=0 s + 1 x + s 1 1 1 1 1 = ∑∞ − − (1 − ) − ∑∞ − s=0 s + 1 x + 1 + s x s=0 s + 2 x + 1 + s 1 1 = 1 − 1 + = , x x since the sums involving x cancel and the sums involving only s are telescoping sums.

1.4.3.2 Product Representation of the Function F Theorem 1.4.2. Every meromorphic function F satisfying the equation F(x + 1) = xF(x) has a product expansion of the form

Kx ∞ x e e s F(x + 1) = ω(x) · · , ∏ + x x s=1 1 s where K is a constant and ω : C → C is a integrable and periodic function with period +1.

Proof. Let ω(x) be as above, and let it be integrable on (0, x) with x > 0. Deﬁne

x Z ω1(x) := ω(x)dx. 0

Then ω1(x) satisﬁes the functional equation:

ω1(x + 1) = ω1(x) + K, K being a constant. Now, recalling the deﬁnition of the function Σ from section 1.4.3.1, by integration we ﬁnd

13We will meet this function again in section 1.8.3.1, when we talk about Euler’s ideas on interpolation of so-called inexplicable functions, a term he coined in 1755 in chapter 16 of [E212]. 28 Main Thesis

x Z log F(x + 1) = Σ(x + 1)dx + K · (x + 1) 0 or, equivalently substituting the series for Σ and integrating it term by term, we have

∞ x x log F(x + 1) = ∑ − log 1 + + K · (x + 1). s=0 s + 1 s + 1 Thus, taking the exponentials, we ﬁnd

Kx ∞ x e e s F(x) = ω(x) · · . ∏ + x x s=1 1 s

1 Thus, to summarize the proof: We basically solved the simpler equation f (x + 1) − f (x) = x ﬁrst. A particular solution is given by our function Σ. Thus, by integrating, we can then deduce the solution of g(x + 1) − g(x) = log(x) and by taking the exponentials we arrive at the functional equation for F(x). It is helpful to keep this in mind, since this is also basically what Euler did in 1780 in [E613] and in 1755 in [E212] to ﬁnd the product representation of the Γ-function. In other words, this proof can easily be constructed from Euler’s ideas in that paper.

1.4.3.3 Finding the Constant K Finally, we need to ﬁnd the constant K, which was introduced by an integration in the previous section. Hence let us introduce the sequence of functions Gn : C \{0, −1, −2, · · · − (n − 1)} → C deﬁned by

Kx n−1 x e e s G (x) := . n ∏ + x x s=1 1 s From this

Kx K n−1 x 1 e · e e s · e s s G (x + 1) = · · n + ∏ + x + x 1 s=1 1 s+1 s 1 We want to rewrite this as

Kx K n−1 x e · e e s 1 −log(1+ 1 ) G (x + 1) = · · e s s . n + ∏ + x x 1 s=1 1 s+1 Now using the well-known result that

n−1 1 1 lim − log 1 + = γ n→∞ ∑ s=1 s s where γ is the Euler-Mascheroni constant, we know that the limit exists14. Deﬁne

1 1 1 γ = + + ··· + − log n n 1 2 n 14This constant is discussed in the appendix in section 1.11.3. There it is also shown that the limit exists. Euler and the Γ-Function 29

such that also γ = limn→∞ γn. Then, we have

−γ x+ x n−1 x e n n e s Gn(x) = ∏ x x 1 + x s=1 1+ s and the functional equation n G (x + 1) = x · G (x) · . n n n + x Therefore, we have:

Theorem 1.4.3. Let γ be the Euler-Mascheroni constant. Further, let ω : C → C be an arbitrary function satisfying ω(x + 1) = ω(x); then, the most general solution of the difference equation F(x + 1) = xF(x) is given by

−γx ∞ x e e s F(x) = ω(x) · · . ∏ + x x s=1 1 s

1.4.4 Application to Γ(x)- The Weierstraß Product Representation Now that we found the most general solution of the difference equation F(x + 1) = xF(x), we want to descend to Γ(x) from this. This is, e.g., possible by an application of the Bohr-Mollerup theorem. Doing so, we will arrive at the following theorem

Theorem 1.4.4 (Weierstraß Product Expansion of Γ(x)). The Γ-function has the following product expansion:

−γx ∞ x e e s Γ(x) = · . ∏ + x x s=1 1 s Proof. We need to check whether the three conditions in the Bohr-Mollerup theorem are fulﬁlled. Therefore, let us check Γ(1) = 1 ﬁrst. We have

∞ 1 e s ( ) = −γ · Γ 1 e ∏ 1 s=1 1 + s or, by taking logarithms,

∞ 1 1 log Γ(1) = −γ + ∑ − log 1 + s+1 s s But, as we have seen above, the sum evaluates to γ, whence log Γ(1) = 1 or Γ(1) = 1. Hence the first condition is satisfied. The second condition of the Bohr-Mollerup Theorem, i.e. Γ(x + 1) = xΓ(x) is satisfied, since it solves the general difference equation F(x + 1) = xF(x). Finally, let us check logarithmic convexity. Obviously, log Γ(x) is twice continuously differentiable, since the resulting sum converges uniformly. We find 30 Main Thesis

d 1 ∞ 1 1 1 log Γ(x) = −γ − + − · ∑ + x dx x s=1 s 1 s s and

d2 1 ∞ 1 1 log Γ(x) = + · , 2 2 ∑ 2 ( + x )2 dx x s=1 s 1 s d2 Therefore, obviously log Γ(x) > 0 ∀x > 0. dx2 Hence the Bohr-Mollerup theorem applies and the function deﬁned by the inﬁnite product is indeed the familiar Γ-function.

1.4.5 Euler on Weierstraß’s Condition We want to show how Euler already arrived at the condition for the Γ-function that Weierstraß used to introduce it. Note that since we obtained that, if in the above theorem, we chose the periodic function ω(x) to be = 1, F(x) = Γ(x), we proved that: Theorem 1.4.5. The Γ-function can also be deﬁned by Weierstraß’s conditions, i.e. the Γ function is the unique meromorphic function satisfying

Γ(x + 1) = xΓ(x) and

Γ(x + n) lim = 1. ∀ x ∈ C \ {0, −1, −2, −3, ··· } n→∞ (n − 1)!nx We will present how Euler obtained this condition in [E652]. Weierstraß in 1856 in [We56] attributed it to Gauß who introduced this condition in [Ga28]. We will show it for Γ(x + 1), since in 1793 paper [E652]15 Euler also did it for x!. Euler’s idea was to consider n as a very large natural number and x as a ﬁxed ﬁnite natural number with x n and evaluate Γ(x + n + 1) in two ways. Using the functional equation x times, we have

Γ(x + 1 + n) = (x + n)(x + n − 1) ··· (n + 1)Γ(n + 1). But, since x n, the ﬁnite parts added to n in each factor can be ignored such that

Γ(x + 1 + n) ≈ nxΓ(n + 1). On the other hand, we can use the functional equation n times, to ﬁnd:

Γ(x + n + 1) = (n + x)(n + x − 1) ··· (x + 1)Γ(x + 1). Therefore, dividing both expressions expressions for Γ(x + n + 1), using Γ(n + 1) = n! and solving for Γ(x + 1), we arrive at the formula:

nxn! Γ(x + 1) ≈ for x n. (x + 1) ··· (n + x − 1)(n + x)

15It was published after Euler’s death in 1783. Euler and the Γ-Function 31

In modern notation, this is the condition in the theorem. One just has to use the functional equation on Γ(x + n) n times to arrive at Euler’s and Gauß’s formula. Note that although the proof required x and n to be natural numbers, the right-hand side does not require x to be a natural number. Therefore, it can also be used to interpolate x! = Γ(x + 1). Additionally, we point out again that Gauß [Ga28] used this formula to introduce the Γ-function, but he did not motivate it at all. Euler’s reasoning, using inﬁnitely large numbers, is obviously not rigorous enough for modern times. But, in possession of the Bohr-Mollerup theorem, one could start from this expression and check, whether all conditions are satisﬁed or not. But since we already did it for the Weierstraß product and know this expression to be equivalent to it, we do not want to repeat this here. 32 Main Thesis

1.5 Euler’s direct Solution of the Equation Γ(x + 1) = xΓ(x) - The Moment Ansatz

We now go over to Euler’s different approaches leading him to an explicit formula of the Γ- function. We will start with the "moment ansatz", a name that will become clear later. First, we want to explain brieﬂy, where the method actually originated.

1.5.1 Origin of the Idea Euler uses a technique, which we will refer here to as moment ansatz, to solve difference equations of the kind:

(a + αx) f (x) = (b + βx) f (x + 1) + (c + γx) f (x + 2), where α, β, γ ∈ R \{0} and a, b, c ∈ R16, in his papers [E123] and [E594]. His actual intention was to derive continued fractions from this. For, dividing the above equation by (a + αx) and f (x + 1), one will ﬁnd

f (x) b + βx c + γx f (x + 2) = + f (x + 1) a + αx a + αx f (x + 1) or

f (x) b + βx c + γx 1 = + . f (x + 1) a + αx a + αx f (x+1) f (x+2) + f (x+1) Replacing x by x 1 one will get a similar equation for the quotient f (x+2) which can be inserted in the above equation. Repeating this procedure inﬁnitely often, one will get a continued fraction f (x) for f (x+1) . Euler was interested in the continued fraction arising from this and he tried to solve the difference equation. In the following, we will explain how he did this.

A(x) f (x) = B(x) f (x + 1) + C(x) f (x + 2) with more general functions A(x), B(x), C(x) etc. also leads to continued fractions. But Euler only considered the case in which those functions are linear functions in his papers [E123] and [E594]. Indeed, his investigations do not go beyond the case of linear functions in any of his papers.

1.5.2 Euler’s Idea Let us discuss his idea on the concrete example of the above difference equation. The generalisation to the general difference equation with linear coefﬁcients is immediate. Euler assumed that the solution is given as an integral of the form

b Z tx−1P(t)dt, a

16Euler, of course did not state the condition on α, β, γ explicitly. But the condition can be inferred from the following calculations in [E123]. Euler and the Γ-Function 33

Figure 1.4: Moment Method One page of Euler’s paper [E123]. Euler explains his idea how to solve the difference equation (a + αx) f (x) = (b + βx) f (x + 1) + (c + γx) f (x + 2) by assuming it to be a certain integral (§ 49). In the following paragraphs he applies it to continued fractions, which was his actual intention in that paper. 34 Main Thesis whence we have to determine the limits of the integration and the function P(t). Euler, in modern formulation, assumed the solution of the difference equation to be the x-th moment of the function P. This is why we gave the method the name moment ansatz. In order to do so, Euler considered the auxiliary equation

t t t Z Z Z (a + αx) tx−1P(t)dt = (b + βx) txP(t)dt + (c + γx) tx+1P(t)dt + txQ(t);

t here, R is supposed to denote the indeﬁnite integral over t and Q(t) is another function we have to determine; the use of this function will become clear in a moment. Euler then differentiated the auxiliary equation with respect to t:

(a + αx)tx−1P(t) = (b + βx)txP(t) + (c + γx)tx+1P(t) + xtx−1Q(t) + txQ0(t). Now divide by tx−1:

(a + αx)P(t) = (b + βx)tP(t) + (c + γx)t2P(t) + xQ(t) + tQ0(t). Comparing the coefﬁcients of the powers of x, we will get the following systems of coupled equations:

1. aP(t) = btP(t) + ct2P(t) + tQ0(t) 2. αP(t) = βtP(t) + γt2P(t) + Q(t)

Solving both equations for P, we ﬁnd

tQ0(t) 1. P(t) = a − bt − ct2 Q(t) 2. P(t) = α − βt − γt2 Therefore, we obtain the following equation for Q(t)

tQ0(t) a − bt − ct2 = Q(t) α − βt − γt2 Although this equation can be solved in general, we will not do this here, because it will be more illustrative to consider examples. Anyhow, having found Q(t), we can also ﬁnd P(t) substituting the value of Q(t) in one of the above equations. Finally, we need the term txQ(t) to vanish in the auxiliary equation. Hence the limits of integration are found from the solutions of the equation txQ(t) = 0.

1.5.2.1 Application to the Γ-function - Finding the Integral Representation As mentioned, everything becomes a lot clearer in certain examples. Therefore, let us consider the Γ-function, i.e. the functional equation f (x + 1) = x f (x). Euler considered the factorial explicitly in §13 of [E594], a paper published in 1785. We make the ansatz Euler and the Γ-Function 35

Figure 1.5: Euler ﬁnds the Integral Representation of Γ(x) via the Moment Method This is one of the examples that Euler considered. It led him to the familiar integral representation of the the Γ-function. Euler called the sequence of the factorial "hypergeometric series". The scan is taken from [E594].

b Z f (x) = tx−1P(t)dt. a Hence we need to determine P(t) and the limits of integration a and b. Let us introduce the auxiliary equation:

t t Z Z txP(t)dt = x tx−1P(t)dt + txQ(t)

Differentiating with respect to t gives

txP(t) = xtx−1P(t) + xtx−1Q(t) + txQ0(t). Division by tx−1 gives

tP(t) = xP(t) + xQ(t) + tQ0(t). Therefore, comparing the coefﬁcients of x:

1. tP(t) = tQ0(t) 2. 0 = P(t) + Q(t).

Solving both equations for P(t): 36 Main Thesis

1. P(t) = Q0(t) 2. P(t) = −Q(t). Hence we obtain the following differential equation for Q(t):

Q0(t) = −1. Q(t) This equation is easily integrated and gives

log(Q(t)) = C − t or Q(t) = Ce−t, where C 6= 0 is an arbitrary constant of integration. From this P is found to be

P(t) = −e−t. Finally, we need to find the limits of integration. For this, we consider the equation txQ(t) = Ctxe−t = 0. For x > 017 we find the two solution t = 0 and t = ∞. Therefore, the term txQ(t) in the auxiliary equation vanishes in these cases and we find:

Z∞ Z∞ C txe−tdt = xC tx−1e−tdt. 0 0 In other words, the equation f (x + 1) = x f (x) is satisﬁed by:

Z∞ f (x) = C tx−1e−tdt. 0 This is, of course, almost the famous integral representation of the Γ-function. (There the constant C is one.)

Finding the Integral Representation of Γ(x) We can force the function f to be the Γ by demanding it to satisfy all conditions of Wielandt’s theorem. The condition Γ(1) = 1 forces C = 1. More precisely, we have the theorem: Theorem 1.5.1 (Integral Representation of Γ(x)). The Γ-function is given by the following integral:

Z ∞ Γ(x) = tx−1e−tdt for Re x > 0. 0 Proof. We have to check all conditions of Wielandt’s theorem18. First, ﬁnd Γ(1).

Z ∞ ( ) = −t = − t∞ = − (− ) = Γ 1 e dt e 0 0 1 1. 0 Secondly, the functional equation is satisﬁed, as demonstrated in the last section. Finally, we have to check holomorphy (which is obvious) and that Γ(x) is bounded in the strip S := {x|1 ≤ x < 2}. Hence consider

17Note that this is precisely the condition on x we need for the integral to converge! 18Therefore, at this point, we basically prove that the integral representation is indeed a correct deﬁnition for Γ(x). Euler and the Γ-Function 37

Z ∞ Z ∞ − − − − |Γ(x)| = tx 1e tdt ≤ tx 1 e tdt 0 0 In other words, we have

|Γ(x)| ≤ Re(Γ(x)) for Re x > 0. For checking Wielandt’s theorem we have to consider

Z ∞ tx−1e−tdt for 1 ≤ x < 2. 0 But these integrals are obviously bounded, whence Wieldlandt’s theorem applies.

1.5.2.2 Some Remarks on the Ansatz

We assume the solution to have the form R tx−1P(t)dt, which explains the name moment ansatz19. But one can, of course, make other choices for the integrand. For the sake of an example, one can set R (R(t))x−1. By the same procedure, one would then arrive at the equation

1 − Z 1x 1 Γ(x) = log dt. t 0 This was Euler’s preferred integral representation and actually the ﬁrst he found in 1738 in [E19]. It follows from the representation just found by setting e−t = u. Furthermore, one can even generalize the ansatz to Z f (x) = R(t)x−1P(t)dt, where R(t) is another function to be determined20. Carrying out the procedure as above, one would arrive at certain conditions on the function R(t) which are trivially satisﬁed by R(t) = t. Indeed, Euler tried this most general ansatz in his 1750 paper [E123], but realizing that R(t) = t meets all requirements, he quickly focused on that special case.

1.5.3 Examples of other Equations which can be solved by this Method Having found the integral representation of the Γ-function from its difference equation, let us apply Euler’s method to more complicated but still familiar difference equations in order to ﬁnd some interesting integral representations.

1.5.3.11 . Example: Legendre Polynomials The Legendre polynomials satisfy the following difference equation

(n + 1)Pn+1(x) = (2n + 1)xPn(x) − nPn−1(x).

b 19 R n A moment is deﬁned as Mn := t dµ, µ being some integration measure. a 20It is indeed convenient to use this ansatz in the case of hypergeometric series, for example. 38 Main Thesis

Together with the conditions P0(x) = 1 and P1(x) = x, this difference equation determines them completely. The moment ansatz can be used to ﬁnd an explicit formula for Pn(x). More precisely, we have the theorem:

Theorem 1.5.2 (Integral representation for the n-th Legendre Polynomial). We have √ x+ x2−1 1 Z tn Pn(x) = √ dt. log(−1) √ 1 − 2xt + t2 x− x2−1 where the principal branch of log(−1) is to be taken.

Proof. This expression can be found by the moment ansatz. Since this is our ﬁrst concrete example of a second order difference equation and one has to be more careful than in the case of the Γ-function, let us present the calculation in detail. We start with the auxiliary equation again which reads

t t t Z Z Z (n + 1) tnR(x, t)dt = (2n + 1)x tn−1R(x, t)dt − n tn−2R(x, t)dt + Q(x, t)tn.

We wish to ﬁnd R(x, t)21 and Q(x, t) and the limits of integration. Let us differentiate that equation with respect to t, we ﬁnd:

(n + 1)tnR(x, t) = (2n + 1)xtn−1R(x, t) − ntn−2R(x, t) + Q0(x, t)tn + ntn−1Q(x, t).

Dividing by tn−2 and comparing the coefﬁcients of the powers of n, we obtain the following system of equations

1. t2R(x, t) = 2tR(x, t)x − R(x, t) + tQ(x, t) 2. t2R(x, t) = tR(x, t)x + Q0(x, t)t2 Solving both for R(x, t)

tQ(x, t) 1. R(x, t)= t2 − 2tx + 1 t2Q0(x, t) 2. R(x, t)= t2 − xt Therefore,

tQ(t) t2Q0(x, t) = t2 − 2xt + 1 t2 − xt whence we ﬁnd p Q(t) = C(x) t2 − 2xt + 1. C(x) being an arbitrary function of x that entered via integration with respect to t. Therefore,

21Although we wrote R(x, t) = R(t) instead of R(t), this does not alter the procedure at all. It will just turn out that R depends also on x which is to be considered as a parameter in the difference equation. The same goes for Q. Euler and the Γ-Function 39

t R(x, t) = C(x)√ . 1 − 2xt + t2 Integrating the differentiated auxiliary equation again from a to b, we would have

b Z tn Pn(x) = C(x) √ dt 1 − 2xt + t2 a if we determine a and b in such a way that Q(x, t)tn vanishes for a and b for all n. Since t0 = 1 has no zeros, we have to put Q(x, t) = 0. This gives

p p a = x − x2 − 1 and b = x + x2 − 1.

Therefore, it remains to ﬁnd C(x). For this we use the special case P0(x) = 1. We calculate √ x+ x2−1 √ Z t0 h p ix+ x2−1 C(x) √ dt = C(x) log 1 − 2xt + t2 + t − x √ 2 √ 1 − 2xt + t2 x− x −1 x− x2−1 Therefore,

1 C(x) = , log(−1) where the principal branch of the logarithm is to be taken, of course. The explicit formula for the n-th Legendre polynomial hence reads

√ x+ x2−1 1 Z tn Pn(x) = √ dt. log(−1) √ 1 − 2xt + t2 x− x2−1

It is easily checked that the explicit formula also gives P1(x) = x. Therefore, both initial conditions and the functional equations are satisﬁed and hence the above formula gives the n−th Legendre polynomial. The formula for the n-th Legendre polynomial seems to be ambiguous, but one always arrives at the same value for Pn(x) as long as the same branch of the logarithm is taken.

1.5.3.2 Historical Remark on Legendre Polynomials

The Legendre polynomials were named after Legendre because of his 1785 paper [Le85]; he discovered them in his investigations on the gravitational potential. Nowadays, they are important in electrodynamics, more precisely, the multipole expansion. But they were in fact already discovered by Euler in his 1783 paper [E551] in a completely different context. In his 1782 paper [E606], Euler even gave the explicit formula for the n−th Legendre polynomial we derived above. But since in that work he was mainly interested in continued fractions for the quotients of two ( ) = 1 consecutive integrals, he did not ﬁnd the constant C x log(−1) . 40 Main Thesis

Figure 1.6: Euler and the Legendre Polynomials Taken from [E606]. Euler solved the difference equation in terms of an integral. He was interested in the continued fraction of two consecutive Legendre polynomials. Thus, he did not ﬁnd any of their more interesting properties, e.g., their orthogonality relation. This had to wait until Legendre’s paper [Le85].

Further, it seems that he did not notice the connection between the findings of [E551] published in 1783 and [E606] published in 1782. In other words, he was not aware that he already obtained an explicit formula for Pn(x). This is even more interesting, because in [E551] he said that it is not possible for him to find such an explicit formula, although he provided all necessary tools in his earlier papers in 1750 in [E123] and in 1785 in [E594]22. In summary, it seems that Euler was not aware that in those papers he basically discovered a general method to find a particular solution of the general homogeneous difference equation with linear coefficients.

1.5.3.32 . Example: Hermite Polynomials

The Hermite polynomials satisfy the recurrence relation

Hn+1(x) = 2xHn(x) − 2nHn−1(x) with the additional conditions H0(x) = 1, H1(x) = 2x. We then have the formula

Theorem 1.5.3 (Explicit Formula for the n-th Hermite Polynomial). The following formula holds:

2 ∞ n x Z 1 t2 i e n 2 − 2 −2xit Hn(x) = √ t e dt 2 π −∞

Proof. We use the moment ansatz. We will not carry out the calculation since it is similar to the case of the Legendre polynomials. We will only state the intermediate results. Of course, we start from the auxiliary equation:

t t t Z Z Z tn+1P(t)dt = 2x tnP(t)dt − 2n tn−1P(t)dt + tnQ(t).

From this we derive the following equations for P(t) and Q(t):

22Although [E594] was published later than [E606] and [E551], they were chronologically written according to their Eneström numbers, i.e. [E551] was written ﬁrst, [E606] last. Euler and the Γ-Function 41

Q0(t)t 1. P(t) = t2 − 2xt Q(t) 2. P(t) = , 2 whence

2 2 1 t −2xt 1 1 t −2xt Q(t) = C(x)e 2 2 and P(t) = C(x)e 2 2 2 2 1 t −2xt In order to ﬁnd the limits of integration, we need to solve tne 2 2 = 0, which leads to t = ±i∞, if we want n to be an arbitrary integer number. Therefore, up to this point we have:

i∞ 2 C(x) Z 1 t −2xt H (x) = tne 2 2 dt. n 2 −i∞ It is convenient to get rid of the imaginary limits by the substitution t = iy. This gives

∞ 2 C(x) Z 1 − y −2xiy H (x) = in+1 yne 2 2 dy. n 2 −∞

From the initial condition H0(x) = 1 we ﬁnd

2 C(x) ex = √ . 2 2i π Therefore, we arrive at

2 ∞ x n Z 1 t2 e i n 2 − 2 −2itx Hn(x) = √ t e dt. 2 π −∞

The condition H1(x) = 2x is easily checked to be satisfied by the explicit formula. It is obvious that one can find similar explicit formulas for other orthogonal polynomials defined by second order homogeneous difference equations with linear coefficients, like, e.g., the Laguerre and Chebyshev polynomials.

1.5.3.43 . Example: Beta Function Let us consider the B-function, which is defined as Definition 1.5.1 (B-function). The B-function, also referred to as Eulerian integral of the first kind, is defined as:

Z1 B(x, y) := tx−1(1 − t)y−1dt for Re x, Re y > 0 0 From its deﬁnition it is immediate that B satisﬁes the functional equation x B(x + 1, y) = B(x, y)(∗). x + y 42 Main Thesis

And one could start from this functional equation to obtain the integral representation via the moment ansatz. Euler did this in 1785 in § 17 of [E594]23. Having already given several examples of this method, we do not want to do this here. Here we want to use the results obtained up to this point to show:

Theorem 1.5.4. We have

Γ(x)Γ(y) B(x, y) = . Γ(x + y)

Proof. We start from the functional equation (∗), of course. Further, we assume that B can be written as product of two functions B1, B2. We demand those to satisfy the equations:

B1(x + 1, y) = xB1(x, y) and (x + y)B2(x + 1, y) = B2(x, y). For the sake of brevity, we will drop the y in the argument in the following; we consider the functional equation only in x, and y can be seen as a parameter. It is easily seen that B1(x) · B2(x) satisﬁes the functional equation for B(x, y). Therefore, we need to solve the equations for B1 and B2 to ﬁnd an expression for B(x, y).

Let us consider B1 ﬁrst. It satisﬁes the functional equation of the Γ-function in x. Therefore, we immediately have B1(x) = C1(y)Γ(x)ω1(x),

C1(y) being an arbitrary function of y, ω1(x) being a periodic function with period +1. 1 To solve the functional equation for B2(x), let us introduce D(x) = . Then, D(x) satisﬁes the B2(x) functional equation:

D(x + 1) = (x + y)D(x). This equation is easily seen to be solved by

D(x) = C2(y)Γ(x + y) · ω2(x).

C2(y) is an arbitrary function of y, ω2(x) being a periodic function of period 1. And hence

1 B2(x) = . C2(y)Γ(x + y) Combining the results, we have found

Γ(x) B(x, y) = ω (x)ω (x)C(y) , 1 2 Γ(x + y) ( ) where C(y) = C1 y . C2(y) We can omit the periodic factor in this solution, since otherwise we would have:

1 Z Γ(x)Γ(y) B(x, y) = dttx−1(1 − t)y−1 = Ω (x) , 1 Γ(x + y) 0

23He even considered a slightly more general example. Euler and the Γ-Function 43

where Ω1(x) is a function with period 1. But Ω1(x) can be found from the special case B(x, 1). For, in this case we have on the one hand

1 Z 1 B(x, 1) = dttx−1 = . x 0 But on the other hand

Γ(x) Ω (x) B(x, 1) = Ω (x) = 1 . Γ(x + 1) 1 x Here we used the functional equation of the Γ-function and Γ(1) = 1. This already implies Ω1(x) = 1 for all x. Hence the periodic function is simply = 1. It remains to determine the function C(y). From the deﬁnition of B(x, y) we ﬁnd:

1 Z 1 B(1, y) = (1 − t)y−1dt = . y 0 First, from our solution we ﬁnd

Γ(1) 1 B(1, y) = C(y) = C(y) , Γ(y + 1) y · Γ(y) where we used Γ(1) = 1 and Γ(y + 1) = yΓ(y). Therefore, C(y) must satisfy:

1 1 1 C(y) = or C(y) = . yΓ(y) y Γ(y) Therefore, we ﬁnally arrived at the formula:

Γ(x)Γ(y) B(x, y) = . Γ(x + y)

1.5.3.5 Some Remarks on the Relation of Γ and B. The relation among the B- and Γ-function was already discovered by Euler essentially in 1738 in [E19]24. He states it explicitly, e.g., in 1772 in [E421]. But the argument he gave there is not a rigorous proof, as we mentioned above in section 1.2.3, he only proved the formula for integers x and y and then, without any further explanation, replaced the factorials by the integral representation of Γ(x). Rigorous proofs were ﬁrst given by Jacobi in 1834 in [Ja34] and Dirichlet in 1839 in [Di39], but they both use the theory of double integrals, which Euler did not know. We will discuss Euler’s argument, Dirichlet’s and Jacobi’s proof below in section 1.9.2.1, in section 1.9.2.3 and in section 1.9.2.2, respectively. Our reasoning to obtain this fundamental relation does not require double integrals and our proof certainly was within Euler’s grasp. In [E594], he even considered similar questions, but never made the connection to the B- and Γ-functions.

24We say "essentially" here, because Euler did not state it explicitly in that paper. 44 Main Thesis

1.5.3.64 . Example: Hypergeometric Series Finally, let us mention the hypergeometric series. It was ﬁrst deﬁned by Euler in [E710], a paper published only in 1801,

Deﬁnition 1.5.2 (Hypergeometric Series). For a, b, c ∈ C \{0, −1, −2, −3, ···} the hypergeometric series is deﬁned by

ab z a(a + 1)b(b + 1) z2 F (a, b, c; z) = 1 + + + ··· for |z| < 1. 2 1 c 1! c(c + 1) 2! We will drop the subscripts 2 and 1, and write simply F, if there is no chance for confusion. The first systematic study was done by Gauß in 1812 in [Ga28], whence the above series is often referred to as Gaussian hypergeometric series. Many people contributed to the nowadays highly developed theory of this function. We mention Kummer [Ku36] and Riemann [Ri57] as some of the contributors. The Gaußian hypergeometric series has been generalized in several ways. The number of parameters in the coefficients has been increased, leading to the Tomae functions pFq, see, e.g., [Sl09]. The number of variables has been increased, e.g., leading to Lauricella functions, which are also discussed in [Sl09]. Furthermore, we want to mention the GKZ systems and the modern text [Yo97]. Here we are mainly interested in hypergeometric integrals. A modern treatise on the subject is [Ao11]. What is of interest for us is that one can derive the integral representation of the hypergeometric series which is usually attributed to Euler25 from a certain difference equation which is satisfied by the the hypergeometric series. Gauß in his paper called them contiguity relations and gave a complete list of 15 of such equations. We will need one equation that follows from those he gave in the mentioned paper.

Theorem 1.5.5. We have the following equation:

b B(b + 2, c − b)F(a, b + 2, c + 2; x) = B(b, c − b)F(a, b, c; x) x(a − c − 1) (b − a + 1)x + c + B(b + 1, c − b)F(a, b + 1, c + 1; x) x(c − a + 1)

The proof is simply done by expanding each hypergeometric function into a power series and comparing coefficients. We will not do this here26. We will consider the above equation as an equation in b. Note that by dividing both sides by B(b, c − b) and applying the relation to the Γ-function and its functional equation, the coefficients become linear functions in b. Then, it is a homogeneous difference equation with linear coefficients: a, c, x are considered as parameters. Thus, we can solve this equation by the moment ansatz. Indeed, proceeding as in the previous cases (with the condition F(a, 0, c; x) = 1), after a long and tedious calculation we arrive at:

1 Γ(c) Z F(a, b, c; x) = tb−1(1 − t)c−b−1(1 − xt)−adt. Γ(a)Γ(c − b) 0 This is the Eulerian integral representation of the hypergeometric series.

25We have to say some things about that later. 26Below, in section 1.5.5, we will arrive precisely at this relation starting from the integral representation. Euler and the Γ-Function 45

Finally, let us mention one drawback of the moment ansatz. In Gauß’s paper one also ﬁnds contiguous relations, relating F(a, b, c; x), F(a + 1, b, c; x) and F(a − 1, b, c; x) (see, e.g. equation [1] in §7 of [Ga28]). The coefﬁcients are also linear in a. But in this case the moment ansatz does not produce a nice solution, if one just uses the ansatz

Z ta−1P(t)dt, since P(t), as we have seen, also depends on a.

1.5.3.7 Historical Note on the Integral Representation

We wish to make some remarks on the origin of the the integral representation of the hypergeometric function. It is often ascribed to Euler that he found this representation, see, e.g., [An10], but it is actually not that simple. What can be said for certain is that we do not find the above equality explicitly in any of Euler’s publications. Therefore, let us briefly discuss, what Euler actually did. First, as we already mentioned in the previous section, Euler studied the hypergeometric series in its most general form and defined it as the power series above in [E710], a paper written in 1778, but just published in 1801. He proceeded to find the differential equation satisfied by it and found a transformation, now bearing his name, of the hypergeometric series in that paper. But he did NOT state the above integral representation anywhere. Nevertheless, on several instances, he did more general investigations, from which the formula would easily follow. Those investigations were mainly concerned with differential equations. We mention his 1763 paper [E274]27 and especially chapter 12 of his second book on integral calculus [E366] published in 1769. In both works, he derived differential equations for parameter integrals, depending on one ore more variables, by differentiating them under the integral sign. His paper on the hypergeometric series, [E710], was written later, but nevertheless he did not make the connection to his earlier investigations. In conclusion, Euler could have written down the above equation, but he did not do so in any of his works. Therefore, let us turn to the people, who actually stated the integral representation. The first to write down the integral representation appears to be Legendre in 1817 in [Le17], although Abel in 1827 in [Ab27] is often credited for it, confer, e.g., [Ni05]28. Kummer also found it in 1837 in [Ku37a] and [Ku37b]. In those papers, he gave a general method to convert certain integrals into series and vice versa. Having mentioned all this, it seems to be up to personal preference to call the integral formula the Eulerian integral representation or not.

1.5.4 Euler and the Mellin-Transform

1.5.4.1 Deﬁnition of the Mellin-Transform and inverse Mellin-Transform

Euler’s ansatz

27Unfortunately, the second half of his paper is lost. But reading the ﬁrst few paragraphs, it is clear that Euler’s investigations would have led him to a formula containing the integral representation of the hypergeometric series as a special case 28In some sense this is true, since Abel was the ﬁrst to prove that uniformly convergent series can be integrated term-by term. 46 Main Thesis

Figure 1.7: Euler and the hypergeometric Series Deﬁnition of the hypergeometric series by Euler in [E710].

Figure 1.8: Euler and the hypergeometric Differential Equation Euler ﬁnds the differential equation for the hypergeometric series in [E710]. Euler and the Γ-Function 47

Figure 1.9: Integral Representation of the hypergeometric Series Euler got close to the integral representation of the hypergeometric series, taken from chapter 10 of [E366]. For a suitable choice of variables and after some simple substitutions, one arrives at the desired formula. Euler never related this result to the findings in [E710], the paper containing the first ever definition of the hypergeometric series as a power series. 48 Main Thesis

b Z F(x) = tx−1P(t)dt a for the solution of a homogeneous difference equation with linear coefﬁcients bears quite a resemblance to the Mellin-Transform of a function f (x). The Mellin-Transform M( f )(s) is deﬁned as

Z∞ M( f )(s) := ts−1 f (t)dt, 0 provided the integral exists29. Although in Euler’s approach we had to calculate the limits of integration, we will argue that Euler’s method can be considered as a precursor of the Mellin- Transform, a concept introduced by Mellin in his 1895 paper [Me95]. To this end, let us introduce the Mellin-Transform properly.

Deﬁnition 1.5.3 (Mellin-Transform). Let ts−1 f (t) ∈ L1(0, ∞), then the Mellin-Transform of the function f is deﬁned as:

Z∞ F(s) := ts−1 f (t)dt. 0 The inverse Mellin-Transform allows to recover the function f from its Mellin-Transform. We have the ∞ Theorem 1.5.6 (Inverse Mellin-Transform). If F(s) = R ts−1 f (t)dt exist, then we have 0 c+∞ 1 Z f (t) = F(s)(t)−sds 2πi c−i∞ with c > 0, provided the integral exists.

We omit the proof here and refer the interested reader, e.g., to [Ti48]. We will not use this formula, but just stated it for the sake of completeness. Obviously, one can apply the inverse Mellin-Transform to find the function P(t) in his ansatz. Thus, Euler’s method to find the function P(t) is also a method to find the inverse Mellin-Transform. In this section, we want to show that one can also start from the differential equation satisfied by P(t) and using Euler’s ansatz one can recover the difference equation. We want to do this in Eulerian fashion and consider several examples of increasing complexity. Before doing so, let us state the following theorem in advance:

Theorem 1.5.7. Let P(t)ts−1−n ∈ L1(0, ∞),P ∈ Cn(R) and n ∈ N a ﬁxed natural number. If

Z∞ F(s) = ts−1P(t)dt, 0

29We will give the precise deﬁnition in a moment. Euler and the Γ-Function 49 then Z∞ ts−1P(n)(t)dt = (−1)(s − 1)(s − 2) ··· (s − n)F(s − n). 0

The proof is by induction and integration by parts.

1.5.4.21 . Example: Γ-function

Let us start with the most familiar example of a Mellin-Transfrom. We want to solve the equation

Γ(x + 1) = xΓ(x) with Γ(1) = 1 for x > 0 again. To this end, we start from the differential equation

P(t) = −P0(t) at which we arrived using Euler’s method in section 1.5.2.1. Thus, multiplying this equation by tx and integrating from 0 to inﬁnity, formally

Z∞ Z∞ txP0(t)dt = − txP(t)dt, 0 0 whence by theorem 2.5.7

Z∞ Z∞ −x tx−1P(t)dt = − txP(t)dt. 0 0

∞ Therefore, calling Γ(x) = R tx−1P(t)dt, we recover the equation 0 Γ(x + 1) = xΓ(x).

Since the differential equation for P is solved by P(t) = Ce−t with C 6= 0, as we also saw above, using the initial condition Γ(1) = 1, we arrive at the integral representation of the Γ-function again.

1.5.4.3 Remark

Therefore, we used the differential equation, we found for P(t) via the moment ansatz to recon- struct the difference equation we wanted to solve. Generally speaking, we used that the Mellin-Transform converts a differential equation into a difference equation. Thus, vice versa the inverse Mellin-Transform converts a difference equation into a differential equation, which turned out to be the differential equation equation that we found using Euler’s method. The latter even works for general limits of integration as we will see in the following examples, whereas the Mellin-Transform is actually just deﬁned for the limits 0 and ∞. 50 Main Thesis

1.5.4.42 . Example: Generalized Γ-function We now want to solve the difference equation

f (x + 1) = (αx + β) f (x) with, f (1) = 1, α > 0, β > 0 for x > 0.

One can use Euler’s method again to solve this difference equation or simply guess the solution. Either way, the solution is given by

Z∞ x−1+ β − t f (x) = C t α e α dt. 0 β − t Thus, one ﬁnds that P(t) = Ct α e α in the moment ansatz. The constant C is not important at the moment. We will determine it later in section 1.6.5. P satisﬁes the following the differential equation30:

P0(t) · tα = βP(t) − tP(t). Now assume that we did not know about the origin of that equation and we wanted to derive the corresponding difference equation. To this end, we multiply both sides by tx−1 and integrate from 0 to ∞:

Z∞ Z∞ Z∞ α txP0(t)dt = β tx−1P(t)dt − txP(t)dt. 0 0 0 By theorem 2.5.7, this simpliﬁes to

Z∞ Z∞ Z∞ −xα tx−1P(t)dt = β tx−1P(t)dt − txP(t)dt. 0 0 0 ∞ Introducing f (x) = R tx−1P(t)dt, the equation reads. 0 −xα f (x) = β f (x) − α f (x + 1) or, equivalently

f (x + 1) = (αx + β) f (x), which is the initial difference equation again. In the next example we want to see how one can derive the limits of integration, if they are not known in advance.

1.5.4.53 . Example: Hermite Polynomials We start from the differential equation

P(t)(t − 2x) = 2P0(t), multiply it by tn, and integrate over t from a and b:

30We do not mention the initial condition, since we do not care about the constant C here. Euler and the Γ-Function 51

b b b Z Z Z P(t)tn+1dt − 2x P(t)tndt = 2 tnP0(t)dt. a a a Since theorem 2.5.7 only holds for a = 0 and b = ∞, we have to keep the absolute part while integrating the right-hand side by parts:

b b b Z Z Z n+1 n n b n−1 P(t)t dt − 2x P(t)t dt = 2t P(t)|a − 2n t P(t)dt. a a a b R n+1 Introducing the notation Hn(x) = P(t)t dt, we have: a n b Hn+1(x) − 2xHn(x) = 2t P(t)|a − 2nHn−1(x). Therefore, to recover the difference equation for the Hermite polynomials, see section 1.5.3.3, it has to be:

n b n n 0 = 2t P(t)|a or 0 = b P(b) − a P(a) n ∈ N. Furthermore, we require a 6= b, since this solution is trivial. As we also saw in 1.5.3.3, P(t) is given by

2 t −tx P(t) = C(x)e 4 . C(x) is just the integration constant. Therefore, a solution of the above equation is given by

(a, b) = (−i∞, +i∞), whence, proceeding with this result, we recover the solution found in 1.5.3.3. Therefore, we actually do not need Euler’s auxiliary function Q(t) to ﬁnd the limits of integration, but can also use the function P(t) to do so.

1.5.4.64 . Example: Legendre Polynomials Finally, let us reconsider the Legendre polynomials already discussed in 1.5.3.1. We start from the differential equation31

R0(t)(t − 2xt2 + t3) = (1 − tx)R(t), multiply it by tn−2, and then integrate it from a to b, whence:

b Z Z R0(t)(tn−1 − 2xtn + tn+1)dt = (tn−2 − tn−1x)R(t)dt. a Integration by parts on the left-hand side yields:

b b b Z Z R(t)tn−1(t2 − 2xt + 1) − R(t)((n − 1)tn−2 − 2xntn−1 + (n + 1)tn)dt = R(t)(tn−2 − tn−1x)dt. a a a

31This equation can be derived from the explicit formula for the function R in 1.5.3.1. 52 Main Thesis

We want the absolute terms to vanish; using the explicit formula for R(t) from 1.5.3.1, i.e. √ t , 1−2xt+t2 this implies:

p p 0 = bn 1 − 2xb + b2 − an 1 − 2xa + a2 ∀n ∈ N.

A solution, aside from a = b, is given by

p p (a, b) = x − x2 − 1, x + x2 + 1 , which is found, if each term in the above condition is set = 0. Finally, let us introduce the notation b R n−1 Pn(x) = t P(t)dt with the solutions we just found for a and b. Hence we arrive at a

−(n − 1)Pn−1(x) + 2xnPn(x) − (n + 1)P(n + 1)(x) = Pn−1(x) − xPn(x), which, after some rearrangement, reads as:

(n + 1)Pn+1(x) = (2n + 1)xPn(x) − nPn−1(x) which is the difference equation from which we started in 1.5.3.1.

1.5.4.7 Final Results

The examples we reconsidered illustrate lucidly that Euler, in some sense, anticipated the Mellin- Transform and found a way to calculate the inverse Mellin-Transform without using contour integration. As we already indicated, Euler’s approach is more general then Mellin’s, since Euler’s method works for general limits of integration. Moreover, it is now evident that Euler’s results obtained via the moment ansatz can be justiﬁed rigorously using the theory of the Mellin-Transform and therefore, Euler’s results are indeed correct. And Euler’s momentum method is not just a precursor of the Mellin-Transform, but it is actually the same idea. Euler just discovered it in a completely different context, i.e. the solution of difference equations, see section 1.5.2, whereas Mellin introduced it in his studies of the hypergeometric differential equation. see, e.g., Mellin’s paper [Me95].

1.5.5 Modern Idea - Intersection Theory

Euler’s approach, referred to as moment method in section 1.5.2, started from the propounded difference equation and reduced it to a differential equation, whereas the Mellin-Transform approach outlined in section 1.5.4 started from a differential equation and derived the initially propounded difference equation. In this section, we want to introduce another method; it starts from a hypergeometric integral of the Eulerian type, i.e.

Z I = tn−1(1 − t)α(1 − xt)βdt, n, α, β ∈ C \ Z, C and derives a difference equation for it. We, following the treatise [Ao11], will brieﬂy mention the fundamental theorems and ideas of the theory and give some examples, partially also found in [Fr19]. Before doing so, let us summarize the main ingredient of Euler’s approach. Euler and the Γ-Function 53

1.5.5.1 Main Ingredient of Euler’s Approach As mentioned in section 1.5.2, Euler explained his idea to solve homogeneous difference equations with linear coefﬁcients in his 1750 paper [E123] §§49 − 53, mainly focusing on continued fractions. But his main idea, aside from the ansatz

b Z tx−1P(t)dt, a is the assumption of an auxiliary equation, i.e. adding the extra term txQ(t). This was necessary to ﬁnd the limits of integration32. Then, we differentiated this auxiliary equation just to integrate it again later. Therefore, what we have essentially done, is to express the integral

b Z tx+1P(t)dt a as a linear combination of one or two others of the same kind. More precisely, we found an equation

b b b Z Z Z x+1 x x−1 t P(t)dt = C1(x) t P(t)dt + C2(x) t P(t)dt. a a a

C1(x) and C2(x) are functions of x, linear in x in our case. In physics, see, e.g., [Fr19], such equations are referred to as integration-by-parts-identities, or IBPs for short, and are frequently employed in the calculation of Feynman integrals of Feynman diagrams. The name IBP stems from the fact that the exponent of t is lowered by 1 by integrating by parts once. Therefore, the extra term txQ(t) can be understood as a term that has to be added to ensure that we arrive at the IBP given by the propounded difference equation. Intersection theory now turns this on its head and starts from an integral like the one above. It tells how many integrals of the same type one needs to express the given integral (e.g., it would immediately tell us that b one requires two integrals to express R tx+1P(t)dt) and furthermore it provides us with a tool to a calculate the coefﬁcients (C1(x) and C2(x) in our example). But we will use it here to solve difference equations again, which will once more lead us to differential equations for the function P(t). Thus, let us explain the origin of intersection theory ﬁrst.

1.5.5.2 Overview In the above integral, products of powers appear. Since we assume the powers to be non-integer numbers in general, we will modify de Rham’s theory presented in section 1.5.5.7, formalizing the ordinary theory of integrals of single-valued functions, to the multi-valued case. This modiﬁed de Rham theory, presented in section 1.5.5.8, is referred to as twisted de Rham theory. Since the key to ordinary de Rham theory is Stokes’ theorem, will we explain the analogue of this theorem for integrals of multi-valued functions ﬁrst. We will do this in section 1.5.5.4.

32It will turn out that it is very difﬁcult to ﬁnd the limits using intersection theory, since that theory actually starts from the integrals, i.e. the integration domain is given from the start. 54 Main Thesis

1.5.5.3 Necessity for the Concept of a Twist In this section, we want to motivate the concept of a twist. To this end, let M be a n-dimensional n afﬁne variety obtained as the complement, in C , of zeros Dj := {Pj(u) = 0} of a ﬁnite number of polynomials Pj(u) = Pj(u1, ··· , un), 1 ≤ j ≤ m; hence

m n [ M := C \ D, D := Dj, j=1 and consider the multi-valued function on M:

m αj U(u) = ∏ Pj(u) , αj ∈ C \ Z, 1 ≤ j ≤ m. j=1 Treating this U(u), one way to get rid of its multi-valuedness such that the concepts of ordinary analysis can be applied to it, is to lift this U to a covering manifold M˜ of M and consider it on this M˜ . But, in general the relation among M˜ and M is so complicated that this approach is not preferable for practical calculations as we intend to perform them here. Thus, instead of lifting U to M˜ , we introduce some quantities introducing a twist arising from the multi-valuedness, and then analyze U. In order to explain the key concepts as simply as possible, let us take a smooth triangulation K on M. As mentioned earlier, we want to extend Stokes’ theorem to multi-valued differential forms Uϕ with ϕ ∈ A•(M) and ﬁnd the twisted version of de Rham’s theory afterwards.

1.5.5.4 Towards Stokes’ Theorem for Integrals of multi-valued Functions We will argue rather intuitively, since the explanation of every mathematical detail is beyond the scope of this article. Let ∆ be one of the p-simplexes of the smooth triangulation K of M and let ϕ be a smooth p-form on M. To determine the integral of a multi-valued p-form we have to ﬁx the branch of U on ∆. Then, we get to the following deﬁnition:

Definition 1.5.4. Let the symbol ∆ ⊗ U∆ denote the pair of ∆ and one of the branches U∆ of U. Then, for ϕ ∈ Ap(M), the integral R U · ϕ is defined as: ∆⊗U∆ Z Z U · ϕ := [the fixed branch U∆ of U on ∆] · ϕ.

∆⊗U∆ ∆

Since U∆ can be continued analytically on a sufﬁciently small neighborhood of ∆, on this neighborhood, for a single-valued p-form U∆ · ϕ and a p-simplex ∆, the ordinary Stokes theorem holds and reads:

Theorem 1.5.8 (Stokes’s Theorem on ∆). For ϕ ∈ Ap−1(M) one has Z Z d (U∆ · ϕ) = U∆ · ϕ. ∆ ∂∆ On the other hand, since dU∆ d (U∆ · ϕ) = dU∆ ∧ ϕ + U∆dϕ = U∆ dϕ + ∧ ϕ U∆ Euler and the Γ-Function 55

dU on ∆ and ω = U is a single-value holomorphic 1-form, setting

∇ω ϕ := dϕ + ω ∧ ϕ, we have:

d (U∆ · ϕ) = U∆∇ω ϕ.

∇ω ϕ can be regarded as the defining equation of the covariant differential operator ∇ω associated to the connection form ω. It satisfies ∇ω · ∇ω = 0 and in such a case ∇ω is said to be integrable or to define a Gauß-Manin connection of rank 1 according to [De70]. Thus, we can rewrite Stokes’ theorem, using the symbol ∆ ⊗ U∆, as:

Theorem 1.5.9 (Stokes’ Theorem rewritten). For ϕ ∈ Ap−1(M) we have: Z Z d (U∆ · ϕ) = U · ∇ω ϕ,

∆ ∆⊗U∆ where

dU ∇ ϕ := dϕ + ω ∧ ϕ and ω = . ω U Next, we want to rewrite the right-hand side of the ordinary Stokes’ theorem (2.5.8). We will illustrate the idea by considering examples in low dimension.

1.5.5.5 Towards the twisted Homology Group We start with the one-dimensional case. For an oriented 1-simplex, with starting point p and ending point q, and a smooth function (since 0-forms are identiﬁed with functions) ϕ on M, Stokes’ theorem (2.5.8.) becomes: Z d (U∆ · ϕ) = U∆(q)ϕ(q) − U∆(p)ϕ(p). ∆

Here we have ∂∆ = hqi − hpi and U∆(q) is the value of the germ of the function Uq, determined by the branch U∆ at the boundary q of ∆, at the point q. Therefore, the symbol hqi ⊗ Uq is determined and we have:

U∆(q)ϕ(q) = the integral of U · ϕ on hqi ⊗ Uq, whence we conclude the formula: Z Z U · ∇ω ϕ = U · ϕ.

∆⊗U∆ hqi⊗Uq−hpi⊗Up Since, in the context of the rewritten Stokes theorem, the right-hand side is supposed to be the boundary of ∆ ⊗ U∆, we can deﬁne a boundary operator by:

Deﬁnition 1.5.5 (Boundary Operator for one dimensional 1-simplex). The boundary operator for this case can be deﬁned as: ∂ω(∆ ⊗ U∆) := hqi ⊗ Uq − hpi ⊗ Up. 56 Main Thesis

Figure 1.10: 2-simplex with Orientation in the two-dimensional Case Auxiliary Figure for the determination of the boundary operator in the two-dimensional case. This is Figure 2.2 from [Ao11].

Stokes’ theorem then reads: Z Z U · ∇ω ϕ = U · ϕ.

∆⊗U∆ ∂ω (∆⊗U∆) Let us also consider the two-dimensional case and let ∆ be an oriented 2-simplex with vertices h1i, h2i, h3i.

For a smooth 1-form ϕ on M, the right-hand side of the ordinary Stokes theorem reads: Z Z Z Z U∆ · ϕ = U∆ · ϕ + U∆ · ϕ + U∆ · ϕ. ∂∆ h12i h23i h31i

Let Uh12i be the branch on the boundary h12i of ∆ determined by U∆. Using the symbol h12i ⊗ Uh12i, the above formula can be rewritten as follows: Z Z Z Z U∆ · ϕ = U · ϕ + U · ϕ + U · ϕ. ∂∆ h12i⊗Uh12i h23i⊗Uh23i h31i⊗Uh31i This leads to the following deﬁnition of the boundary operator in this case:

Deﬁnition 1.5.6 (Boundary Operator for 2-simplex in two-dimensional case). The boundary operator in this case reads:

∂ω(∆ ⊗ U∆) := h12i ⊗ Uh12i + h23i ⊗ Uh23i + h31i ⊗ Uh31i. Hence Stokes’ theorem becomes: Z Z U · ∇ω ϕ = Uϕ.

∆⊗U∆ ∂(∆⊗U∆) Euler and the Γ-Function 57

After these introductory examples, we can generalize them to the higher-dimensional case. Note that the boundary operators defined above are precisely those appearing in algebraic topology when one defines homology with coefficients in a local system. We will explain this below. For now, consider the following differential equation on M

m dPj ∇ωh = dh + ∑ αj h = 0 j=1 Pj which has a general solution, formally expressible as:

m − αj C h = c ∏ Pj , c ∈ . j=1 The space generated by the local solutions of the differential equation has dimension 1. Cover the manifold M by a sufficiently fine locally finite open cover M = ∪Uν and fix a single-valued non-zero solution hν on each Uν. Since, provided µ 6= ν, hµ and hν are solutions of the same differential function on the non-empty set Uµ ∩ Uν, setting

hµ(u) = gµν(u)hν, u ∈ Uµ ∩ Uν the transition function gµν is a constant on Uµ ∩ Uν. Since a solution h(u) on Uµ ∩ Uν is expressed −1 in two ways, i.e. as h = ξµhµ = h = ξνhν with ξµ, ξν ∈ C, we find ξµ = gµν ξν. Therefore, the set of all local solutions of the given differential equation defines a flat line bundle n −1o Lω obtained by gluing the fibers C by the transition functions gµν . Let us denote the flat line ∨ bundle obtained from the transition functions gµν by Lω and call it the dual line bundle of −1 the line bundle Lω. By the above equation relating hν and hµ, hµ becomes a local section of ∨ ∨ Lω. Moreover, from the initial differential equation, Lω can be considered as the flat line bundle generated by the set of all local solutions of

m dPj ∇−ωh = dh − ∑ αj h = 0, j=1 Pj

αj and U(u) = ∏ Pj is its local section. Often, e.g., in [Ao11], the term "flat line bundle" is used in the same sense as "system of rank 1". ∨ Therefore, we call Lω the dual local system of Lω. Hence we see that the boundary operators ∨ ∨ defined in the first two examples are those of the chain groups C1(K, Lω) and C2(K, Lω) with ∨ coefficients in Lω. Naturally, if using M instead of the simplicial complex K, we denote the chain ∨ group by Cq(M, Lω) for the q-dimensional case. More precisely, we were lead to the definition:

Deﬁnition 1.5.7 (p-dimensional twisted Chain Group). We call

 for a p-simplex ∆ of M,  ∨   Cp(M, Lω) = the complex vector space   with basis ∆ ⊗ U∆ the p-dimensional twisted chain group.

Moreover, as a generalization of the above examples, we have: 58 Main Thesis

Deﬁnition 1.5.8 (Boundary Operator of the p-simplex). We deﬁne the boundary operator

∨ ∨ ∂ω : Cp(M, Lω) → Cp−1(M, Lω) for the p-simplex ∆ = h012 ··· pi as

p D E ( ⊗ U ) = (− )j ··· j ··· p ⊗ U ∂ω ∆ ∆ : ∑ 1 01 b h01···bj···pi. j=0 where bj indicates that this term is omitted. Similarly to the ordinary homology theory, one has:

Theorem 1.5.10. The boundary operator ∂ω satisﬁes:

∂ω ◦ ∂ω = 0. Having given this theorem, we can deﬁne the twisted homology group: Deﬁnition 1.5.9 (Twisted Homology Group). The qoutient vector space

∨ ∨ ∨ ∨ Hp(M, Lω) := Ker ∂ω : Cp(M, Lω) → Cp−1(M, Lω) /∂ωCp+1(M, Lω) is called the p-dimensional twisted homology group, and an element of Ker ∂ω is called a twisted cycle.

1.5.5.6 Locally finite Twisted Holonomy Group Having arrived at the definition of the twisted homology group, we can now, in analogy to the ordinary homology group, introduce some related concepts. We start with the locally finite twisted homology group. Since in general M is not compact, it would require infinitely many simplices for its triangulation, hence it is natural to consider an infinite chain group which is locally finite. More precisely, we set: Definition 1.5.10 (Locally finite p-dimensional twisted chain group). The group

l f ∨ Cp (M, Lω) := {Σc∆∆ ⊗ U∆|the ∆’s are locally ﬁnite} is called the locally ﬁnite Twisted chain Group.

l f ∨ Since ∂ω ◦ ∂ω = 0, (C• (M, Lω)) forms a complex, which in turn leads to the next Definition 1.5.11 (Locally finite Twisted Holonomy Group). The homology group obtained from the l f ∨ l f ∨ complex (C• (M, Lω)) is denoted by (H• (M, Lω)) and called the locally finite Twisted Holonomy Group.

Let us now consider Stokes’ theorem. Thus, let ∆ be a p-simplex of M, U∆ a branch of U on ∆, ϕ ∈ Ap−1(M), then the right-hand side of Stokes’ theorem becomes:

R p i R U∆ · ϕ = ∑i=0(−1) U∆ · ϕ ∂∆ h0···bi···pi p i R = ∑i=0(−1) U · ϕ ···i···p ⊗U h0 b i h0···bi···pi = R U · ϕ. ∂(∆⊗U∆) Euler and the Γ-Function 59

For a general twisted chain, one can just extend the deﬁnition C-linearly and arrives at the following form of Stokes’ theorem:

∨ l f ∨ Theorem 1.5.11 (Stokes’ Theorem for Cp(M, Lω) and Cp (M, Lω)).

∨ p−1 1. For σ ∈ Cp(M, Lω) and ϕ ∈ A (M) we have:

Z Z U · ∇ω ϕ = U · ϕ.

σ ∂ω σ

l f ∨ p−1 2. For τ ∈ Cp (M, Lω) and ϕ ∈ Ac (M) we have:

Z Z U · ∇ω ϕ = U · ϕ.

τ ∂ω τ

Next, since we already see the similarity to the de Rham theory here, we want to construct the twisted version of the theory. To this end, using the last theorem, we have to establish relations among the cohomology of

• • (A (M), ∇ω) , Ap(M), ∇ω and the twisted holonomy groups

∨ l f ∨ H• M, Lω , H• M, Lω .

1.5.5.7 de Rham Theory As mentioned in the preceding section, we want to establish that the cohomology of the complex • ∨ (A (M), ∇ω) is dual to the twisted homology H• (M, Lω). Since the proof of this fact is too long to present it here, we will simply state the theorem. The reader interested in the proof is, e.g., referred to [Ao11]. Furthermore, let us point out that ordinary de Rham theory corresponds to the case ω = 0 and thus twisted de Rham theory will be its natural generalization. De Rham theory is summarized by de Rham’s theorem and duality theorems.

Theorem 1.5.12 (de Rham’s Theorem). The following isomorphisms exist:

 p •  H (A (M), d) p ∼ p • ∼ l f (1) H (M, C) ' ↑ o −→ H (K (M), d) ←− H2n−p(M, C) Hp(Ω•(M), d)

p p • ∼ p • ∼ (2) Hc (M, C) ' H (Ac (M), d) −→ H (Kc (M), d) ←− H2n−p(M, C).

p Here, Kc is the sheaf of germs of currents of degree p with compact support on M. A current is an element of the space containing both C∞ differential forms and smooth singular chains. Furthermore, we have the following 60 Main Thesis

Theorem 1.5.13 (Duality Theorems). The following bilinear forms are non-degenerate

p • (1) Hp(M, C) × H (A (M), d) → C

([σ] , [ϕ]) 7−→ R ϕ, σ

2n−p • p • (2) H (Ac M, C) × H (A (M), d) → C

([α] , [β]) 7−→ R α ∧ β. M

These are the fundamental theorems in de Rham theory. These results can be extended to twisted de Rham theory.

1.5.5.8 Twisted de Rham Theory

We state the most important theorems necessary to develop twisted de Rham theory

Theorem 1.5.14. There is an natural isomorphism

p ∨ H (M, Lω) ' HomC(Hp(M, Lω), C)

p Via this isomorphism, the cohomology group H (M, Lω) can be considered as the dual ∨ space of the homology group Hp(M, Lω). Therefore, the value h[σ] , [ϕ]i of a cohomology class p [ϕ] ∈ H (M, Lω) at a homology class [σ] ∈ Hp(M, Lω) is well-deﬁned and we have the following

Theorem 1.5.15. The following bilinear form is non-degenerate

∨ p Hp(M, Lω) × H (M, Lω) → C ([σ] , [ϕ]) → h[σ] , [ϕ]i .

In analogy to de Rham’s Theorem, we have the

Theorem 1.5.16. The following isomorphisms exist

 p •  H (A (M), ∇ω) p ∼ p • ∼ l f (1) H (M, Lω) ' ↑ o −→ H (K (M), ∇ω) ←− H2n−p(M, Lω) p • H (Ω (M), ∇ω)

p p • ∼ p • ∼ (2) Hc (M, Lω) ' H (Ac (M), ∇ω) −→ H (Kc (M), ∇ω) ←− H2n−p(M, Lω).

Finally, we can give the most important theorem for our investigations.

Theorem 1.5.17. The following bilinear forms are non-degenerate Euler and the Γ-Function 61

∨ p • (1) Hp(M, Lω) × H (A (M), ∇ω) → C

([σ] , [ϕ]) 7→ R U · ϕ. σ

p • l f ∨ (2) H (Ac (M), ∇ω) × Hp (M), Lω) → C

([ψ] , [τ]) 7→ R U · ψ. τ

2n−p • p • (3) H (Ac (M), ∇−ω) × H (A (M), ∇ω) → C

([α] , [β]) 7→ R α ∧ β. M ∨ l f (4) Hp(M, Lω) × H2n−p(M, Lω) → C

([σ] , [τ]) 7→ intersection number [σ] · [τ].

In the next sections we will explain the meaning of equation (4), i.e. the intersection number which is crucial for investigations of hypergeometric integrals.

1.5.5.9 Construction of twisted Cycles In this section, we want to construct some twisted cycles explicitly. Starting from the one- dimensional case and the twisted cycle around one point, we will generalize this and will arrive ∨ at an explicit formula for H1(M, Lω). We will explicitly construct m − 1 independent twisted cycles associated to the multi-valued function

m αj U(u) = ∏(u − xj) , αj ∈/ Z j=1 which is defined M = C \{x1, ··· , xm}. Since later we will also be mainly interested in this case, we assume all the points xj, 1 ≤ j ≤ m to lie on the real axis such that x1 < ··· < xm. We set dU ∨ ω = U such that this U, as we explained in section 1.5.5.5, defines the rank 1 local system Lω. We choose the following branch of U(u): we take the one that is single-valued on the lower half plane and that satisfies

0 (1 ≤ j ≤ p) arg(u − x ) = j −π (p + 1 ≤ j ≤ m) on each interval ∆p := (xp, xp+1), 1 ≤ p ≤ m − 1. From basic algebraic topology we know that ∨ (1) H•(M, Lω) is homotopy invariant (2) The Mayer-Vietoris sequence holds ∨ Thus, using (1) it sufﬁces to calculate H•(K, Lω) for the one-dimensional simplicial complex K ∨ seen in the ﬁgure above. Note that, for the sake of simplicity, we also write Lω for the restriction ∨ of Lω on M to K. To apply the Mayer-Vietoris sequence (2), let us write K = K1 ∪ K2, i.e. as the union of two subcomplexes 62 Main Thesis

Figure 1.11: Construction of twisted Cycles in the one-dimensional Case Auxiliary ﬁgure for the construction of a twisted cycles for the function m αj U(u) = ∏j=1(u − xj) , αj ∈/ Z. This is ﬁgure 2.5 from [Ao11].

Figure 1.12: Triangulation of the Circle 1 ∨ Auxiliary ﬁgure for the calculation of Hq(Sε (xj), Lω). This is ﬁgure 2.6 from [Ao11].

m m−1 [ 1 [ K1 = Sε (xj), K2 = [xj + ε, xj+1 − ε]. j=1 j=1

1 In the ﬁrst step, we calculate the twisted homology for each circle Sε (xj), i.e. we prove:

Theorem 1.5.18 (Twisted homology for a circle). For αj ∈/ Z, 1 ≤ j ≤ m, we have

1 ∨ Hq(Sε (xj), Lω) = 0, q = 0, 1, ···

Proof. We triangulate the circle as seen in the ﬁgure above

1 ∨ Thus, the chain group Cq(Sε (xj), Lω) and the boundary operator ∂ω are: Euler and the Γ-Function 63

1 ∨ C0(Sε (xj), Lω) = Ch0i + Ch1i + Ch2i,

1 ∨ C1(Sε (xj), Lω) = Ch01i + Ch12i + Ch20i,

∂ωh01i = h1i − h0i, ∂ωh12i = h2i − h0i, ∂ωh20i = cjh0i − h2i, with cj = exp(2πiαj). Writing these relations as a vectorial equation, we have

h01i −1 1 0  h0i ∂ω h12i =  0 −1 1  h1i h20i cj 0 −1 h2i and det(∂ω) = cj − 1. This is not zero, since αj ∈/ Z. Therefore, the map

1 ∨ 1 ∨ ∂ωC1(Sε (xj), Lω) → C0(Sε (xj), Lω) is an isomorphism and hence the theorem is proven.

Next, we have the

Theorem 1.5.19. Provided αj ∈/ Z, 1 ≤ j ≤ m, we have

∨ H0(M, Lω) = 0,

∨ H1(M, Lω) ' C · ∆j(ω) with the twisted cycle ∆j(ω)

1 1 1 1 ∆j(ω) := Sε (xj) + [xj + ε, xj+1 − ε] − Sε (xj+1). cj − 1 cj+1 − 1

We sketch the idea of the proof. One applies the Mayer-Vietoris sequence to the union K = K1 ∪ K2 to ﬁnd the exact sequence

∨ ∨ ∨ δ 0 → H1(K1, Lω) ⊕ H1(K2, Lω) → H1(K, Lω) −→ ∨ ∨ ∨ ∨ → H0(K1 ∩ K2, Lω) → H0(K1, Lω) ⊕ H0(K2, Lω) → H0(K, Lω) → 0. ∨ Therefore, the task is to ﬁnd δ such that one can determine H1(K, Lω). To this end, note that K2 is homotopically equivalent to m − 1 points and K1 ∩ K2 is homotopically equivalent to 2m − 2 points. Furthermore, using the last theorem, many terms in the sequence are seen to vanish, and we are left with the exact sequence:

∨ δ ∨ i∗ ∨ 0 → H1(K, Lω) −→ H0(K1 ∩ K2, Lω) −→ H0(K2, Lω) → ||o ||o C2m−2 Cm−1 ∨ → H0(K,Lω) → 0

The map i∗ is induced from he inclusion i : K1 ∩ K2 → K2 and we ﬁnd 64 Main Thesis

m−1 ! m−1 i∗ ∑ ajhxj + εi + bjhxj+1 − εi = ∑ (aj + bj)hxj + εi, aj, bj ∈ C. j=1 j=1

Thus, i∗ is surjective and we ﬁnd:

∨ H0(K, Lω) = 0

Lm−1 Im δ = Ker i∗ = j=1 C(hxj − εi − hxj+1 − εi). ∨ One finds δ, which is necessary to find H1(K, Lω), in the same way; we state the result. Defining

1 1 1 1 ∆j(ω) := Sε (xj) + [xj + ε, xj+1 − ε] − Sε (xj+1), cj − 1 cj+1 − 1 one has

δ(∆j(ω)) = hxj + εi − hxj+1 − εi Therefore,

m−1 ∨ M H1(K, Kω) = C · ∆j(ω). j=1

1.5.5.10 Intersection Number for Twisted Cycles We will explain the intersection number by means of a simple example which can then be generalized to the general case. First, we mention the m Theorem 1.5.20. Under the assumption ∑j=1 αj ∈/ Z, twisted cycles ∆ν ⊗ U∆ν deﬁned by bounded m−1 l f ∨ chambers ∆ν, 1 ≤ ν ≤ ( n ), form a basis of Hn (M, Lω):

m−1 ( n ) l f ∨ M Hn (M, Lω) ' C [∆ν ⊗ U∆ν ] . ν=1 Concerning bounded champers: The arrangement of m real hyperplanes in Rn decomposes that space into several chambers, which are referred to as bounded chambers. Having stated this theorem in advance, let us explain what the intersection number between ∨ l f ∨ H1(M, Lω) and H1 (M, Lω) means. We will consider the case m = 2, the general case can then be understood similarly. Let M = C \{0, 1} and choose a branch of the multi-valued function U(u) = uα(u − 1)β, α, β, α + ∨ β ∈/ Z. H1(M, Lω) is one dimensional and is basis is given by

1 1 1 1 ∆(ω) = S (0) ⊗ U 1 + [ε, 1 − ε] ⊗ U − S (1) ⊗ U 1 , ε Sε (0) [ε,1−ε] ε Sε (1) c1 − 1 c2 − 1 √ c1 = exp(2πi α), c2 = exp(2πiβ)

1 Sε (0) is a circle around 0 with radius p with starting point ε, which turns in anticlockwise direction, stopping at an inﬁnitely near point before ε, and U 1 is the branch obtained by analytic Sε (0) Euler and the Γ-Function 65

Figure 1.13: Intersections of ∆(ω) and ∆ 1 Figure showing the three intersection of ∆(ω) and ∆ at the three points p, 2 , q. This is Figure 2.7 from [Ao11].

1 continuation along Sε (0), and similarly for the rest. l f Now, using the above theorem, H1 (M, Lω) is the one dimensional vector space with basis ( ) ⊗ −1 ( ) 0, 1 U(0,1), where U(0,1) is the restriction of the branch determined above to 0, 1 . Deﬁne −1 ∆ := γ ⊗ Uγ , obtained by deforming the interval (0, 1) as in the ﬁgure above, this is homologous ( ) ⊗ −1 l f ( L ) to 0, 1 U(0,1) in H1 M, ω .

1 The geometric intersections of ∆(ω) and ∆ are the three points p, 2 , q and the signatures are −, −, +. Since the difference of the branches is cancelled by U and U−1, we ﬁnally arrive at:

1 1 ∆(ω) · ∆ = × (−1) + (−1) + − × (+1) c1 − 1 c2 − 1 c c − 1 = − 1 2 (c1 − 1)(c2 − 1) for the intersection number.

As already indicated, one can compute the intersection matrix for a general m in a similar way. −1 l f The result will look as follows: We set ∆j := (xj, xj+1) ⊗ U ∈ H (M, Lω), 1 ≤ j ≤ m − 1, (xj,xj+1) 1 cj = exp(2πiαj), dj := cj − 1, djk = cjck − 1, then we have:   cj  , j = k + 1,  d  j   dj,j+1 − , j = k, djdj+1 ∆j(ω) · ∆k =  1 − + =  , j 1 k  dk   0, otherwise

Therefore, the intersection matrix, consisting of each intersection number, reads: 66 Main Thesis

  ∆1(ω) · ∆1 ··· ∆1(ω) · ∆m−1  . .   . .  ∆m−1(ω) · ∆1 ··· ∆m−1(ω) · ∆m−1  d −1  12 0 ··· 0 0  d1d2 d2     −c2 d23   0 0   d2 d2d3   .  = −  .   .   d −1   m−2,m−1   0 0   dm−2dm−1 dm−1   −cm− dm− m  0 0 ··· 0 1 1, dm−1 dm−1dm

1.5.5.11 Intersection Numbers for Twisted Cocycles Although the intersection number of twisted cycles is of great importance in the general theory of hypergeometric functions, for us, more interested in difference equations, we need the intersection number of twisted cocycles. Above, in the main theorem of section 1.5.5.8, equation (3),

n n ∨ H (M, Lω) × Hc (M, Lω) → C can be considered to define an intersection number of twisted cocycles. Indeed, the calculation works similarly to the case of twisted cycles and was carried out explicitly in [Ch95] such that it will enough to mention the necessary definitions briefly and consider examples after this. 1 Here, we set M = P \{x0, ··· , xm}, where the case xi = ∞ is not excluded. We define

m du ω := α , ∇ := d + ω∧, ∇∨ := d − ω∧ ∑ j − ω ω j=0 u xj

1 and let D be the divisor of P determined by the m + 1 points x0, ··· , xm, and denote the sheaf of p logarithmic p-forms that may have logarithmic poles only along D by ΩP1 (log D) and the sub- p p sheaf of ΩP1 (log D) of p-forms which have poles along D of at least order 1 by ΩP1 (log D)(−D), then we have the

Theorem 1.5.21. The following bilinear form is non-degenerate

1 1 1 1 I : Γ P , ΩP1 (log D) × H (P , OP1 (−D)) → C.

This deﬁnes the intersection number

Deﬁnition 1.5.12 (Intersection Number for twisted Cocycles). We deﬁne the intersection number of ϕ+, ψ− by

+ − −1 − Ic(ϕ , ψ ) := I(ϕ, i (ψ )), where

1 1 + − ϕ, ψ ∈ Γ P , ΩP1 (log D) and ϕ = ϕ mod C · ω, ψ = ψ mod C · (−ω) Euler and the Γ-Function 67 and i is the inclusion map

• ∨ i • ∨ Ω (log D)(−D), ∇ω −→ Ω (log D), ∇ω

1.5.5.12 Calculating Examples - Using a different Notation Having given all necessary concepts to understand the notion of a twist, we finally want to calculate some examples. To do this as conveniently as possible, we want to introduce a more compact notation, i.e. the one used in the application-oriented paper [Fr19]. There, the concept of twisted cohomology etc. is not explicitly introduced. But having introduced all the necessary concepts and theorems, the reader can now always add the corresponding mathematical concepts to the more physicist-oriented paper. Let us mention in advance that in the following calculations the manifold of interest is always M = P1 \P, where P is the set of poles of the 1-form ω. Now, let us go through the most important definitions of [Fr19]. We will use the notation and symbols of that paper such that it is easier for the reader to compare. First they pair hϕ|, which is a cocylce, and |C], which is a domain in P, to obtain an integral of the above form. More precisely, they define Z hϕ| C] := uϕ. C This defines a bilinear form in the spirit of the main theorem of section 1.5.5.8 and hence can be used to find linear relations among hypergeometric functions. To see this, let ν be the number of linearly independent cocycles, and denote an arbitrary basis of differential forms by

he1| , he2| , ··· , heν| . Note that we know from our considerations that the dimension ν of the space of cocylcles is ﬁnite. Then, as known from linear algebra, a decomposition is achieved by expressing the arbitrary cocycle hϕ| as a linear combination of the base elements. This can be done as follows. We need a dual space of twisted cocycles, whose basis we want to denote by |hii with i ∈ {1, 2, ··· ν}. This will lead us to the deﬁnition of the intersection number from [Fr19]33:

Definition 1.5.13 (Intersection Number in this new Notation). Let hei| with i ∈ {1, 2, ··· , ν} be a base element of the space of cocycles and |hii with i ∈ {1, 2, ··· , ν} a basis element of a second space dual to the first of the space of cocycles34, then we define the intersection numbers

Cij := ei|hj .

These are the entries of a ν × ν matrix.

1.5.5.13 Derivation of the Decomposition Formula Having defined the intersection numbers, our next task is to find the decomposition formula, i.e. how to find the coefficients in the linear combination of hϕ| in terms of the basis elements. This is the most important formula in [Fr19]. It is used in all examples there. To derive the decomposition formula, let us define a (ν + 1) × (ν + 1) matrix M as follows:

33This is seen to agree with the deﬁnition of the intersection number of twisted cocylces we have given in 1.5.5.11. 34From twisted de Rham Theory we know this dual space to exist. 68 Main Thesis

But by deﬁnition the matrix C cannot be zero, whence the other factor must be zero. From this we conclude

hϕ|ψi = A|C−1B or writing out the products

ν −1 hϕ|ψi = ϕ|hj C hei|ψi ∑ ji i,j=1 Therefore, since |ψi is arbitrary, we arrive at the the decomposition formula:

Theorem 1.5.22 (Decomposition Formula). Let hei| with i ∈ {1, 2, ··· ν} be a basis element of the space of cocycles and hei| with i ∈ {1, 2, ··· ν} a basis element of the dual space of the space of cocycles, then we have the decomposition formula ν −1 hϕ| = ϕ|hj C hei| ∑ ji i,j=1

The decomposition formula provides us with a projection of hϕ| onto the basis elements hei|. Contracting both sides with the twisted cocycle |C] (this means multiplying by u and integrating over C), we ﬁnd the formula

Corollary 1.5.22.1. Z ν Z −1 uϕ = ϕ|hj C uei. ∑ ji C i,j=1 C

Therefore, we ﬁnd the coefﬁcients of the decomposition

Corollary 1.5.22.2. Given ν I = K hϕ|C] = ∑ ci Ji, i=1 where

Ji ≡ KEi with Ei ≡ hei|C] , Euler and the Γ-Function 69 we have

ν −1 ci = ϕ|hj C . ∑ ji j=1

1.5.5.14 The case of 1-forms

In this section, we descend to 1-forms and show how to calculate intersection numbers and to apply the decomposition formula. First, we need to ﬁnd the dimension ν of the vector space under consideration. This is addressed by the following theorem.

Theorem 1.5.23. Let ω be a differential 1-form deﬁned on X = P1 \P, where P is the set of poles of ω. Then, the dimension ν of the vector space of cocycles is

ν = {number of solutions of ω = 0}.

Following [Ch95] and [Ma98], we deﬁne the intersection number hϕL|ϕRiω as follows:

Deﬁnition 1.5.14 (Intersection Number hϕL|ϕRiω). Let P be the set of poles of ω, a meromphic 1-form on P, i.e

P = {z|z is a pole of ω}

Then, we have:

hϕL|ϕRiω = ∑ Resz=p ψp ϕR , p∈P where ψp is a function and solves the differential equation ∇ωψ = ψL around p, i.e.

∇ωp ψp = ϕL,p.

In general, fp denotes the Laurent expansion of the function f around p.

1.5.5.151 . Example: B-function

We will start with the simplest hypergeometric integral - the B-function, deﬁned as the integral

Z1 B(x, y) = dzzx−1(1 − z)y−1. 0 Moreover, in section 1.5.3.4 we found the relation:

Γ(x)Γ(y) B(x, y) = . Γ(x + y) We will discuss the B-function in more detail to see how the method works in contrast to other more familiar methods discussed in this chapter. 70 Main Thesis

1. Approach: Direct Integration Let us put Z n γ γ In = uz dz, u = z (1 − z) , C = [0, 1]. C

We can use the relation among Γ and B to express In directly:

Γ(1 + γ)Γ(1 + γ + n) I = . n Γ(2 + 2γ + n) Therefore,

Γ(1 + γ + n)Γ(2 + 2γ) I = I . n Γ(1 + γ)Γ(2 + 2γ + n) 0 For n = 1, this reduces to

1 I = I . 1 2 0

2. Approach: Integration-by-Parts

Let us see, whether we can ﬁnd the last relation among I1 and I0 by integration by parts. For this aim, note that Z d (z(1 − z))γ+1zn−1 = 0. C Expanding the integrand gives:

(γ + n)In−1 − (1 + 2γ + n)In = 0. Hence

(γ + n) I = I − , n (1 + 2γ + n) n 1 which for n = 1 gives

1 I = I . 1 2 0

3. Approach: Intersection Numbers We want to ﬁnd the relation among the B-integrals again. We deﬁne Z n In = uφn+1 ≡ω hφn+1|C] , φn+1 ≡ z dz.

Additionally,

1 1 u = zγ(1 − z)γ, ω = d log u = γ + dz. z z − 1 Euler and the Γ-Function 71

The equation ω = 0 obviously has only 1 solution. Hence the dimension of the space under consideration is ν = 1. Further, we need to ﬁnd the poles of ω. The poles z = 0 and z = 1 are = = 1 = − 1 immediate. To consider z ∞, perform the substitution t z , whence dt z2 . Therefore, t dt ω = −γ t + . 1 − t t2 Since this tends to ∞ for t → 0, the set of poles is:

P = {0, 1, ∞}. we want to express I1 in terms of some integrals of the same class. Since ν = 1, we only need one integral and choose it to be I0. Therefore, we know

I1 = c1 I0 ⇔ ω hφ2|C] = c1 · ω hφ1|C] and we need to ﬁnd c1. Since ν = 1, the matrix Cij is simply a number, i.e. hφ1|φ1i. Hence, in total we need to evaluate the numbers hφ1|φ1i, hφ2|φ1i. For this we need the residues. For each pole p ∈ P we need φi,p - the series expansion of φi about z = p - and ψi,p - the series expansion of ψi around z = p, which is found from the differential equation

∇ωp ψi,p = φi,p. Note that the following Laurent expansions are known

= (k) k = k k φi,p ∑ φi,p z and ωp ∑ ωpz k=min −1 k=−1 and we need to ﬁnd

max k ψp = ∑ αkz . k=min

In other words, the coefﬁcients αk are to be found. max(φi) = ordp(φi) + 1 and min(φi) = − ordp(φi) − 1. We introduced min and max, since those determine the range of coefﬁcients necessary to determine the residue we need. We do not need the rest of the Laurent series. Writing out the above differential equation for differential forms, we arrive at the following simpler differential equation for functions:

d ψ + ω ψ = φ . dz p p p i,p

Now we can begin with the calculation of the intersection numbers. We start with hφ1|φ1i. For p = 0 and p = 1 we ﬁnd that min = 1 > max = −1. Therefore, the resulting residues are zero. For p = ∞ on the other hand, we ﬁnd min = −1 and min = 1. Therefore, our series for ψ∞ has the following three terms:

1 ψ = α + α + α z1. ∞ 1 z 0 1 Substituting this series in the above differential equations together with the series expansion of φ1,∞ and ω∞ and comparing coefﬁcients, we ﬁnd 72 Main Thesis

1 1 γ α− = , α = − , α = . 1 2γ + 1 0 2(2γ + 1) 1 2(2γ − 1)(2γ + 1) Therefore, the intersection number is

γ hφ |φ i = Res = (ψ φ ) = . 1 1 z ∞ ∞ 1 2(2γ − 1)(2γ + 1)

Let us go over to hφ2|φ1i. As in the ﬁrst case, the differential equation has no solution for p = 0 and p = 1. For p = ∞ we ﬁnd that the series must look as follows:

1 1 ψ = α− + α− + α + α z. ∞ 2 z2 1 z 0 1 The coefﬁcients are found by the same procedure as above and are calculated to be

1 γ 1 γ α− = , α− = , α = , α = − . 2 2(γ + 1) 1 2(γ + 1)(γ + 2) 0 4(2γ + 1) 1 4(2γ − 1)(2γ + 1)

Thus,

γ hφ |φ i = Res = (ψ φ ) = . 2 1 z ∞ ∞ 1 4(2γ − 1)(2γ + 1) Now we can apply the decomposition formula and ﬁnd

1 c = hφ |φ i (hφ |φ i)−1 = , 1 2 1 1 1 2 i.e.

1 I = I 1 2 0 in agreement with the approaches above.

1.5.5.162 . Example: Hypergeometric Function

Above in section 1.5.3.6 we saw that we can express the Gaussian hypergeometric series as

Z1 b−1 c−b−1 −a B(b, c − b)2F1(a, b, c; x) = z (1 − z) (1 − xz) dz. 0 Hence the integration contour is C = [0, 1]. B(a, b) is the B-function. We want to use intersection theory to ﬁnd a difference equation for hypergeometric series. We write:

Z B(b, c − b)2F1(a, b, c; x) = uϕ = ω hϕ|C] , C with the letters u, ω and ϕ being Euler and the Γ-Function 73

u = zb−1(1 − xz)−a(1 − z)−b+c+1

xz2(c − a − 2) + z(ax − c + x + 2) − bxz + b − 1 ω = d log u = dz (z − 1)z(xz − 1)

ϕ = dz.

Therefore, we have

1 ν = 2 and P = 0, 1, , ∞ . x This indicates that we can express one hypergeometric integrals by two others. This is not surprising, since we know Gauß’s contiguous relations. But let us ﬁnd a speciﬁc one by using intersection theory. i We chose the basis {hφ1| , hφ2|}, where, as above φi+1 = z dz. In this example, the matrix C is a 2 × 2 matrix and looks as follows:

− c + 1)(a − c + 2)(a − c + 3) 3 2 hφ1|φ2i = x (−(a − b + 1)(a − b + 2)(b − c + 1)) − ax (−b + c + 1)(2a − 3b + c + 2) + ax(a + 2c − 5)(−b + c + 1) − a(c − 3)(c − 2) / x3(a − c + 1) (a − c + 2)(a − c + 3)(a − c + 4) , 3 2 hφ2|φ1i = x (−(a − b))(a − b + 1)(b − c + 1) − ax (−b + c − 1)(2a − 3b + c) + ax(a + 2c − 3)(−b + c + 1) − a(c − 2)(c − 1) / x3(a − c)(a − c + 1) (a − c + 2)(a − c + 3) , 2 2 2 2 2 2 2 hφ2|φ2i = − ax (a b − a c + a − 3ab + 7abc − 8ab − 4ac + 9ac − 5a − 3b c

+ 6b2 + 4bc2 − 10bc + 6b − c3 + 2c2 − c) + x4(−(a3 − 3a2b + 3a2 + 3ab2 − 6ab + 2a − b3 + 3b2 − 2b))(b − c + 1) + 2ax3(a − b + 1)(ab − ac + a − 2b2 + 3bc − 2b − c2 + c) + 2a(c − 2)x(a + c − 2)(b − c + 1) + a(c3 − 6c2 + 11c − 6) / x4(a − c)(a − c + 1)(a − c + 2)(a − c + 3)(a − c + 4) . 74 Main Thesis

Now we found everything necessary for the application of the decomposition formula, which in this case reads

2 −1 hφn| = φn|φj C hφi| . ∑ ji i,j=1

As an example, let us take hφ3|C] = B(b + 2, c − b)2F1(a, b + 2, c + 2; x) in terns of B(b, c − b)2F1(a, b, c; x) and B(b + 1, c − b)2F1(a, b + 1, c + 1; x). Then, we ﬁnd

b B(b + 2, c − b) F (a, b + 2, c + 2; x) = B(b, c − b) F (a, b, c; x) 2 1 x(a − c − 1) 2 1 (b − a + 1)x + c + B(b + 1, c − b) F (a, b + 1, c + 1; x) x(c − a + 1) 2 1 which relation can also be derived from the contiguous relations for hypergeometric functions.

1.5.5.173 . Example: Γ-function Having explained the idea, let us apply the idea to the Γ-function and recover its integral representation from its functional equation one more time. We still assume that it is possible to write Γ(x) as an integral of the form

Z∞ Γ(x) = tx−1P(t)dt 0 and we have to determine P(t). We want to solve Γ(x + 1) = xΓ(x). In the language of intersection theory, we have

x P0(t) u = txP(t) and hence ω = d log u = + dt. t P(t) Since we seek to express Γ(x + 1) by one other integral, this forces us to arrange that

x P0(t) + = 0 t P(t) has precisely one solution for t. This in turn implies

P0(t) = C or P(t) = BeCt. P(t) for some constants B and C 6= 0 we have to determine from the remaining conditions. We need to ﬁnd the poles of ω, i.e. the 1-form deﬁned above35. Inserting the result we found for P(t), we have x ω = + C dt. t 1 One pole is obviously given by t = 0. But we also have to consider t = ∞. Therefore, put u = t . Hence locally around t = ∞

35Of course, at this point we could use Wielandt’s theorem directly to force the integral to become Γ(x) - it would imply B = 1 and C = −1. But we will use intersection theory to the end and see how we have to ﬁnd B and C his way. Euler and the Γ-Function 75

du ω = − (xu + C) u2 which is inﬁnite for u = 0 and hence indicates a pole at t = ∞. Therefore,

P = {0, ∞}.

As in the first example, we need the intersection numbers hφ1|φ1i and hφ2|φ1i. Since we forced ν = 1, the intersection matrix is just C11 = hφ1|φ1i, i.e a 1 × 1-matrix. The two necessary intersection numbers are calculated as in the preceding examples. As in the first example only the point p = ∞ actually contributes and we find:

x hφ |φ i = 1 1 C2 x2 hφ |φ i = − 2 1 C3 Therefore, the decomposition formula yields

− x c = hφ |φ i hφ |φ i 1 = − . 1 2 1 1 1 C Recall that we want to solve the equation

Γ(x + 1) = xΓ(x). Therefore, this gives the condition x − = x ⇒ C = −1. C Hence, we ﬁnally arrive at the solution:

Z∞ Γ(x) = B tx−1e−tdt. 0 B is a constant different from 0. Using the initial condition Γ(1) = 1, one ﬁnds the integral representation of Γ(x) again.

1.5.5.18 Remarks This last example shows that intersection theory can also be used backwards to solve certain difference equations, although it is quite a lot of work compared to Euler’s method. But note the difference, how we found the function P(t) here in contrast to the other methods. We recovered the differential equation for P by requiring an equation to have only one solution in t. This is quite different from Euler’s method or the inverse Mellin-Transform. Furthermore, one has to insert the integration limits from the beginning in contrast to Euler’s method, where you calculate them explicitly. But since we showed above in section 1.5.4 how to circumvent this problem, this is not to be considered an defect of intersection theory in this regard. Therefore, we have now presented three methods to solve homogeneous difference equations with linear difference equations, where intersection theory is clearly the most general. 76 Main Thesis

1.6 Solution of the Equation log Γ(x + 1) − log Γ(x) = log x by Conversion into a Differential Equation of inﬁnite Order

1.6.1 Overview We solve the general difference equation

f (x + 1) − f (x) = g(x) by converting it into a differential equation of inﬁnite order via Taylor’s theorem. This was done by Euler in 1753 in [E189]36. Unfortunately, there is an error in Euler’s approach that we will explain in section 1.6.2.3 and correct in section 1.6.3, before we solve the equation by the more modern approach of Fourier analysis in section 1.6.3.2. But we will also present a solution that Euler could have given in section 1.6.4, and present Euler’s, corrected, derivation of the Stirling-Formula given in [E189] in section 1.6.4.1.

1.6.2 Euler’s Idea Euler’s reasoning involves some purely formal operations. Therefore, we will not try to give the conditions under which the operations are valid here. Nevertheless, it is a beautiful example of the "Ars Inveniendi" (Art of Finding).

1.6.2.1 Presentation of his Idea As mentioned, the main source for this section is [E189]. Here, Euler actually intended to solve various interpolation problems. One of them is to interpolate the function f deﬁned for positive integers:

n−1 f (n) := ∑ g(k), k=0 g being an arbitrary function. Obviously, f satisﬁes the functional equation:

f (n + 1) − f (n) = g(n) ∀n ∈ N. One possible way to interpolate f is to solve the above difference equation for general n ∈ C which was Euler’s intention [E189]37. For this he used Taylor’s theorem to write, now for x ∈ C:

∞ f (n)(x) f (x + 1) = ∑ . n=0 n! Therefore, the above difference equation becomes:

∞ f (n)(x) ∑ = g(x). n=1 n! This can be seen as an inhomogeneous ordinary differential equation of inﬁnite order with constant coefﬁcients.

36He considered other examples of differential equations of inﬁnite order in [E62] and his second book on integral calculus [E366]. 37Euler actually assumed n to be real, but this restriction is not necessary at all. Euler and the Γ-Function 77

In [E62] and [E188], Euler explained a method to solve such differential equations, if the order is ﬁnite38. That method is still the same we use today and can be found in any modern textbook on differential equations.

1.6.2.2 Actual Solution Let us present Euler’s solution. This ﬁrst step to solve an equation of such a kind (at least in the case of ﬁnite order) is to consider the characteristic polynomial, i.e., the polynomial resulting by dn n substituting dxn for z in the differential operator acting on f . This "polynomial" in our case reads:

P(z) = ez − 1. Next, Euler wants to ﬁnd the zeros of P. Using the theory of complex logarithms, which he had developed in [E168] and [E807], he found the zeros to be

zk = 2kπi, k ∈ Z. All zeros are easily seen to be simple. From this Euler (assuming that the case of inﬁnite order can be treated as the case of ﬁnite order) concluded that each zero will lead to a term

x Z ezk x e−zktg(t)dt. x R means that we have to put x = t after the integration. Therefore, Euler claimed that the solution of the general difference equation reads

x ∞ Z f (x) = ∑ e2kπix e−2kπitg(t)dt. k=−∞

Note that each integral leads to an integration constant ck, whence this solution is not a particular but the complete solution.

1.6.2.3 Mistake in Euler’s Approach Unfortunately, Euler’s solution is incorrect. It gives the correct solution only in the homogeneous case. One ﬁnds

∞ 2kπix f (x) = ∑ cke k=−∞ as solution of f (x + 1) = f (x). ck are arbitrary constants of integration. Interestingly, Euler found that each periodic function has Fourier series. But Euler did not realize what a broad ﬁeld of mathematics he had entered and did not pursue this any further. But his solution formula already fails to give the correct result in the case g(t) = 1. The correct formula, as we will prove in the following, reads:

∞ 1 2kπix f (x) = − g(x) + ∑ cke , 2 k=−∞

38[E62] considers the homogeneous case, whereas [E188] treats the inhomogeneous case 78 Main Thesis or, presented in the more convenient form,

x x Z 1 Z f (x) = g(t)dt − g(x) + ∑ e2kπix e−2kπitg(t)dt. 2 k∈Z\{0} Furthermore, Euler’s mistake is not a computational but a conceptual one. His approach to construct the solution from the zeros of the characteristic "polynomial" (in analogy to the ﬁnite case where this is possible) simply does not work in the case of inﬁnite order. Instead of the zeros one has to use the partial fraction decomposition39:

1 1 1 1 = − + + . ez − z ∑ z − k i 1 2 k∈Z\{0} 2 π

x 1 2kπix R −2kπit Comparing this to the solution it is easily seen that each term z−2kπi leads to a term e e g(t)dt 1 1 in the solution, whereas − 2 explains the term − 2 g(x). We will prove this in the following section, but need to give some deﬁnitions and state some auxiliary theorems in advance.

1.6.3 Correction of Euler’s Approach

1.6.3.1 Preparations - Lemmata and Deﬁnitions

n Theorem 1.6.1 (Fundamental Theorem of Algebra). Let P(z) = a0 + a1z + ··· + anz be a non- constant polynomial of degree n with complex coefﬁcients, then P(z) has exactly n zeros in C, where the zeros have to be counted with multiplicity.

This theorem is usually proved by applying Liouville’s theorem. We refer the reader to any modern book on complex analysis for a proof and mention [Fr06] as an example. As a historical note, we add that Euler also tried to prove the fundamental theorem of Algebra in [E170]. But his proof is incomplete, as pointed out by Gauß in his 1799 dissertation [Ga99]. We refer to [Du91] for a review of Euler’s paper [E170] from the modern perspective.

But let us go over to the correction of Euler’s approach concerning the solution of the difference equation. We need to introduce some deﬁnitions.

Deﬁnition 1.6.1 (Schwartz Space). We denote by S the set of all functions f ∈ C∞(Rn) such that

sup max |xαDβ f (x)| < ∞ x∈Rn |α|,|β|

α β fj → 0 :⇔ sup max |x D fj(x)| → 0 x∈Rn |α|,|β|

Next, we introduce the Fourier transform and state the Fourier inversion formula.

39The following partial fraction decomposition is proved in the appendix in section 1.11.2 Euler and the Γ-Function 79

Deﬁnition 1.6.2 (Fourier Transform and Fourier Inversion Formula). We deﬁne the Fourier transfor- n mation of a function f ∈ L1(R ) as: Z fb(p) := e−ix·p f (x)dx, Rn n where x · p = ∑i=1 xk pk. If also fb is integrable, the following formula, the Fourier inversion formula holds: 1 Z f (x) = eix·p fb(p)dp. (2π)n Rn Next, we state a theorem, often very useful in calculations involving Fourier transforms. Theorem 1.6.2 (Convolution Theorem). Let F be the operator of the Fourier transform, that F{ f } = fb and F {g} = gb are the Fourier transforms of the functions f and g ∈ S, then we have

n n F { f ∗ g} = (2π) 2 F { f } ·F {g} and (2π) 2 F { f · g} = F { f } ∗ F {g} , where ( f ∗ g)(x) means the convolution product deﬁned via

Z ∞ Z ∞ ( f ∗ g)(x) := f (τ)g(x − τ)dτ = f (x − τ)g(τ)dτ −∞ −∞ The theorem is straight-forward by a simple application of Fubini’s theorem, so that we omit it here. One ﬁnds a proof in most textbook that covers the Fourier transform. We refer to [Ti48]. Theorem 1.6.3 (Convolution Theorem for the inverse Fourier Transform). Let F −1 be the operator of the inverse Fourier transform, so that F −1 { f } and F −1 {g} the inverse Fourier transforms f and g ∈ S, then we have

n −1 n −1 f ∗ g = (2π) 2 F {F { f } ·F {g}} and (2π) 2 f · g = F {F { f } ∗ F {g}} The proof is just by applying the inverse Fourier transform to the equations in the convolution theorem.

Before we present the solution algorithm for the general difference equation, we need two more properties of the Fourier Transform. First that the Fourier transform changes differential operators D into −ip, i.e. an algebraic object. More precisely, we have: Theorem 1.6.4. Let f ∈ S(Rn), then

F(Dα f )(x) = (ip)α fb(p). α denotes a multi-index again. The proof is just by integration by parts. Furthermore, we have Theorem 1.6.5. If f ∈ S(Rn), then fb ∈ S(Rn). This is an easy application of the fact that the Fourier transform interchanges differentiation and multiplication. Finally, we have the following property of the inverse Fourier transform: Theorem 1.6.6. The Fourier transform is a bijective mapping on the Schwartz space. The proof of this (and the two previous theorems) can be found, e.g., in [St02]. 80 Main Thesis

1.6.3.2 Solution Algorithm for the general difference equation - Finding a particular solution Having stated all the definitions and theorems, we can now finally give the solution for the general difference equation with constant coefficients.

Theorem 1.6.7. Let the following equation be propounded:

N ∑ ak f (x + k) = g(x), k=0 with f and g ∈ S(R) and let denote zk the zeros, which we, for the sake of brevity, assume to be simple, of the polynomial 40

N k P(z) = ∑ akz . k=0

And let pl be a solutions of the equation

ipl e = zk. Further, ﬁnd the partial fraction decomposition:

N h i−1 b a e+ip·k = C + l . ∑ k ∑ ( − ) k=0 l∈Z i p pl Then, a particular solution of the difference equation is given by

x Z −ipl (x−τ) f (x) = Cg(x) + ∑ bl e g(τ)dτ, l∈Z

x where R f (τ)dτ means that we integrate with respect to τ and then write x for τ after the integration and the constant of integration is to be kept. This solution is unique up to the addition of a function, satisfying the equation:

N ∑ ak f (x + k) = 0. k=0 Proof. To give a proof, it is easier to start from the solution and derive the difference equation. We do not present the calculation in detail. All steps are justiﬁed, by using the theorems in the preparations. With pk deﬁned as above, consider

x Z ipl x −ipl τ f (x) := Cg(x) + ∑ ble e g(τ)dτ. l∈Z Taking the Fourier transform of this expression, simplifying it by using the convolution theorem and the inverse convolution theorem41, we arrive at the following expression:

40The fundamental theorem of algebra guarantees that we always have exactly N solutions. 41In section 1.6.3.5 we will also present a solution of the simple difference equation f (x + 1) − f (x) = g(x) without resorting to the convolution theorem. Euler and the Γ-Function 81

! 1 fb(p) = g(p) C + b . b ∑ l ( − ) l∈Z i p pl The expression in brackets is just the partial fraction decomposition of the polynomial in eipk, hence we have

−1 " N # ipk fb(p) = gb(p) ∑ ake . k=0 Solving for gb(p) gives

" N # −ipk gb(p) = fb(p) ∑ ake . k=0 Finally, taking the inverse Fourier transform, we get

N g(x) = ∑ ak f (x + k), k=0 which is the difference equation propounded and completes our proof. That we can add a function satisfying

N ∑ ak f (x + k) = 0, k=0 is obvious.

1.6.3.3 Case of multiple Roots Although we only proved the theorem for simple zeros of the polynomial, we can directly generalize it to multiple zeros. Suppose, that zk is a multiple zero of order m of the polynomial P(z) deﬁned above. Then in the formula

N h i−1 b a eip·k = C + l ∑ k ∑ ( − ) k=0 l∈Z i p pl we just have to replace

m b bj l by . − ∑ j p pl j=1 (p − pl) In general, we have that

j! ( ) j+1 gdp (ip − ipl) leads to the term

x Z (τ − x)jeipl (τ−x)g(τ)dτ in the ﬁnal solution. 82 Main Thesis

1.6.3.4 Application to the difference equation f (x + 1) − f (x) = g(x). The difference equation we are interested in is a special case of the theorem we just proved. Here,

P(z) = z − 1. Considering the solutions of

ipl e = zk = 1, pl is found to satisfy:

ipl = 2kπi k ∈ Z. Therefore, we need to ﬁnd the partial fraction decomposition of

1 eip − 1 which (comparing to the result we mentioned above in 1.6.2.3) is found to be

1 ∞ 1 − + ∑ − 2 l=∞ ip 2lπi or in terms of pl, see also section 1.11.2:

1 1 ∞ 1 = − + . ip ∑ ( − ) e − 1 2 l=∞ i p pl Therefore, the solution of the difference equation is concluded to be

x 1 ∞ Z f (x) = − + ∑ e2lπix e−2πiltg(t)dt. 2 l=∞ Precisely, as we stated the solution above in section 1.6.2.3.

1.6.3.5 Explicit Solution of the general Difference Equation via Fourier-Transform Above in section 1.6.3.2 we just explained how to proceed in general. Therefore, we want to add the explicit calculation in the example of the simple difference equation, i.e.

f (x + 1) − f (x) = g(x). Assuming the function f , g have all the necessary properties, i.e. are ∈ S(R), we now present the solution of the above equation. First, we rewrite this equation as

∞ f (n)(x) ∑ = g(x) n=1 n! using Taylor’s theorem and apply the Fourier-Transform; then, our equation becomes

ip e − 1 fb(p) = gb(p). Thus, Euler and the Γ-Function 83

1 fb(p) = g(p). eip − 1 b 1 Using the partial fraction decomposition of ez−1 given in 1.6.2.3, we ﬁnd ! 1 1 1 fb(p) = − + g(p). ip ∑ ip − k i b 2 k∈Z\{0} 2 π Next, we want to take the inverse Fourier transform of the last equation. The inverse Fourier − 1 ( ) − 1 ( ) transform of 2 gb p is simply 2 g x . Therefore, let us consider ∞ 1 1 Z 1 F −1 g(p) = g(p)eixpdp ip b 2π ip b −∞ Now note that one can write x x Z eiyp eixp eiypdy = = . ip i∞ ip i∞ Inserting this into the above formula, we have

 x  Z∞ Z − 1 1 F 1 g(p) =  eiypdy g(p)dp. ip b 2π b −∞ i∞ Applying Fubini’s theorem, this becomes

x  ∞  1 Z Z =  g(p)eiypdp dy. 2π b i∞ −∞ 1 ( ) The inner integral times 2π is just the inverse Fourier transform of gb p . Therefore, in total x 1 Z F −1 g(p) = g(y)dy. ip b i∞ Let us now consider the more general case

∞ 1 1 Z g(p) F −1 g(p) = b eixpdp. ip − 2kπi b 2π i(p − 2kπ) −∞ Substituting p − 2kπ = q, this becomes

∞ 1 Z g(q + 2kπ) = b eix(q+2kπ)dq. 2π iq −∞

eixq Rewriting iq as above and applying Fubini’s theorem, we ﬁnd

x   Z Z∞ − 1 1 F 1 g(p) = e2πix  eiyqg(q + 2kπ)dq dy. ip b 2π b i∞ −∞ 84 Main Thesis

1 ( + ) The inner integral times 2π is just the inverse Fourier transform of gb q 2kπ . To ﬁnd this integral, one just has to set p = q + 2kπ again. More precisely,

∞ ∞ 1 R iyq 1 R iy(p−2kπ) e gb(q + 2kπ) = e gb(p)dp 2π −∞ 2π −∞ ∞ 1 −2kπiy R iyp = = e e gb(p)dp 2π −∞ 1 = e−2kπiyg(y), 2π 1 ( ) since the last integral times 2π is the inverse Fourier transform of gb p . Therefore, in total, we found

x g(p) Z F −1 b = e2kπix e−2kπyg(y). ip − 2kπi i∞ Finally, inserting everything in the complete formula

x x Z 1 Z f (x) = g(y)dy − g(x) + ∑ e+2kπix e−2kπiyg(y)dy. 2 ∈Z∈{ } i∞ k 0 i∞ So, there are no integration constants in this formula and the solution we found is a particular solution, as we claimed. But since the full solution is obtained by adding an arbitrary function of period 1, we hence derive the general solution

x x Z 1 Z f (x) = g(y)dy − g(x) + ∑ e2kπix e−2kπiyg(y)dy, 2 k∈Z\{0} as we stated it above in 1.6.2.3.

1.6.4 A Solution Euler could have given Fourier analysis came after Euler42, i.e. he had no access to it. But here we argue that he could have derived the correct solution from his results, if only we overlook some details of mathematical rigor. d The ﬁrst idea is to consider dx and the higher derivatives as operators acting on the function f (x). Then, solving a differential equation is equivalent to ﬁnding the inverse operator to the d operator acting on f and applying it to both sides of the equation. Let us again write dx = z. The fundamental theorem of calculus implies

1 Z f = f dx. z And the difference equation we want to solve then becomes

1 f (x) = X. ez − 1 42Fourier published his book, in which he introduced Fourier series, in 1822 [Fo22]. Euler and the Γ-Function 85

We can now simplify this equation inserting the partial fraction decomposition of (ez − 1)−1. We obtain ! 1 1 1 f (x) = − + X. z ∑ z − k i 2 k∈Z\{0} 2 π 1 We only need to ﬁnd out what z−α X is. For this, let us write

1 1 X = X − α z α z 1 − z 1 αn = ∑∞ X, z n=0 zn 1 + where we used the geometric series in the second step. zn+1 X is the n 1 times iterated integral of X. Euler considered integrals of this kind in [E679], a paper written in 1778. His formulas yield

n x Z Z (x − t)n−1 Xdx = X(t)dt. (n − 1)!

n x Here, R denotes the n times iterated integral, R indicates that one has to put x = t after the integration. Substituting this formula and using the Taylor series for ex, we ﬁnd

x x 1 ∞ αn Z Z X = ∑ (x − t)nX(t)dt = eα(x−t)X(t)dt. z − α n=0 n! Therefore, we can write our solution as

x x R 1 R ( − ) f (x) = Xdx − X(x) + ∑ e2kπi x t X(t)dt 2 k∈Z\{0} x x R 1 ∞ R = Xdx − X(x) + 2 ∑ cos (2kπ(x − t)) X(t)dt, 2 k=1

eix+e−ix where we used the formula cos x = 2 in the second line. 1 Note that this is almost the formula Euler gave, in Euler’s formula only the term − 2 X is missing, making his result incorrect. Although we have operated completely non-rigorously, the formula, as we found it, is correct, see [We14]. This formal calculus is a beautiful example of the "Ars inveniendi" and is validated by the techniques from Fourier analysis.

1.6.4.1 An Application - Derivation of the Stirling Formula for the Factorial In the last paragraphs of [E189], Euler tried to derive the Stirling formula for Γ(x) from his general 1 solution of the general difference equation. But since he missed the term − 2 g(x) in the solution 86 Main Thesis of the difference equation, his formula is incorrect43. He gave a correct derivation of the formula in [E212]. We will present his idea from [E189] and derive the Stirling formula from the solution of the difference equation:

log Γ(x + 1) − log Γ(x) = log(x). More precisely, we prove the formula Theorem 1.6.8 (Stirling Formula for Γ(x + 1)). √ xx Γ(x + 1) ∼ 2πx for x → ∞. ex Before we start with the proof, we want to mention that we will, presenting Euler’s argument, argue formally. For, we will write =, although we will encounter a series which actually is an asymptotic series and thus = is not the correct sign to use here. For deﬁnition of asymptotic series, the reader is, e.g., referred to [Ha48], [Ba09] and the chapter on divergent series in [Va06]. Having said this in advance, let us go over to Euler’s proof.

Proof. His idea was to simplify the solution of the difference equation we derived; applied to log Γ(x) this solution reads

x x 1 Z Z log Γ(x) = − log x + log(t)dt + ∑ e2kπix e−2πit log(t)dt, 2 k∈Z\{0} We simplify this general solution following Euler in [E189]. First, we have

x Z log tdt = x log x − x + C,

C being the constant of integration. Secondly, by iterated partial integration, we obtain

x Z log xe−2kπix ∞ (−1)n (n − 1)! log te−2kπitdt = C − + e−2kπix, k ∑ ( )n+1 n 2kπi n=1 2kπi x

Ck being the constant of integration. Therefore, " # 1 log x ∞ (−1)n (n − 1)! log Γ(x) = x log x − x + Π(x) − log x + − + , ∑ k i ∑ ( k i)n+1 xn 2 k∈Z\{0} 2 π n=1 2 π where Π(x) is an arbitrary periodic function with period 1, i.e., a solution of the homogeneous equation y(x + 1) − y(x) = 0. Now note that for a natural number m

1 1 ∞ 1 = 0 and = 2 . ∑ k2m−1 ∑ k2m ∑ k2m k∈Z\{0} k∈Z\{0} k=1 Therefore,

43He was even aware of this but tried to argue it away by another incorrect argument. More precisely, he arrived at the 1 asymptotic formula x! = x log x − x + P, i.e. the term − 2 log x is missing. But he argued that his formula is nevertheless 1 correct, since P has to be a periodic function - which statement is incorrect - and he just found P = 2 log(2π) instead 1 of the correct value P = 2 log(2πx). Euler and the Γ-Function 87

1 ∞ ∞ (−1)2m−1 2(2m − 2)! log Γ(x) = x log x − x + Π(x) − log x + . ∑ ∑ 2m−1 ( )2m 2 k=1 m=1 x 2kπi Now recall Euler’s famous formula for the even ζ-values44

∞ 1 (−1)m−1B (2π)2m = 2m , ∑ 2m ( ) k=1 k 2 2m ! where Bn are the Bernoulli numbers. Hence, having substituted those explicit values for the sums, we ﬁnally arrive at

1 ∞ B log Γ(x) = x log x − x − log x + Π(x) + 2m . ∑ ( − ) 2m−1 2 k=1 2m 2m 1 x 1 This is where Euler stopped. But in his case, the term − 2 log x was missing, as we mentioned in section 1.6.2.3. But having simplified the solution this far, let us now check the conditions of the Bohr-Mollerup theorem. Demanding them to be satisfied by the solution, we can prove that indeed Π(x) = 1 2 log(2π). We start with logarithmic convexity. One way of arguing that Π(x) = A, i.e., a constant, is as follows. Recall the following condition for convexity. Definition 1.6.3 (Convexity for a differentiable function). A differentiable function y(x) is convex on the interval I if and only if

y(a) − y(b) ≥ y0(b) ∀ a, b ∈ I. a − b Let us denote the convex part of the right-hand side in the ﬁnal equation, i.e., the sum of every function except the periodic function Π(x) since all of them are easily seen to be convex on the positive real axis, by C(x). Then, since y(x) = Π(x) + C(x) and y(x) is convex by assumption, we have

Π(a) − Π(b) + C(a) − C(b) Π(a) − Π(b) ≥ + C0(b) a − b a − b and

Π(a) − Π(b) + C(a) − C(b) ≥ C0(b) + Π0(b) a − b Therefore, we must have either

Π(a) − Π(b) + C0(b) ≥ C0(b) + Π0(b), a − b in which case Π(x) would already be convex and everything works out nicely, since the only differentiable periodic function that is convex on the whole positive real axis is the constant function, or we have

44He never wrote it down like this explicitly but gave a list of the ﬁrst 13 values, e.g., in [E130] and [E212]. He certainly was aware of this general formula. 88 Main Thesis

Π(a) − Π(b) C0(b) + Π0(b) ≥ C0(b) + , a − b or equivalently

Π(a) − Π(b) Π0(b) ≥ . a − b Since this inequality must hold for all positive a and b, let us put b = a + 1. Then, since Π(x) and hence also Π0(x) is periodic,

Π0(a) ≥ 0. This condition must hold for all positive a. Hence Π(x) is a monotonically increasing function on the positive real axis. But the only differentiable periodic function that is monotonically increasing on the whole positive real axis is, again, the constant function. This completes the proof that Π(x) is indeed a constant. In the next step, we have to determine the constant explicitly. This is done using the condition y(1) = 0. Therefore,

∞ B 0 = y(1) = −1 + A + 2m . ∑ ( − ) k=1 2m 2m 1 This equation, at least in principle, allows to find the constant A. We say "in principle", since the sum diverges because of the rapid growth of the Bernoulli numbers. Nevertheless, applying techniques to sum divergent√ series, see, e.g., [Ha48] written in 1948, the sum can be evaluated and we find A = log 2π. The evaluation of this constant was Stirling’s great contribution to the Stirling-formula - he gave a more general formula in 1730 in [St30] and essentially used the Wallis product formula for π to find the constant A. But the formula we want to prove and was named after him, was actually discovered by de Moivre in 1718 in [dM18], who could at that time only evaluate the constant numerically. See, e.g, [Du91], [Le86], [Pe24] for a discussion on this matter. In order to avoid the use of divergent series here, let us present Euler’s strategy to find A, taken from [E212] §158 chapter 6 of the second part. Euler argues that, since A is constant, we can determine it from any case. First, let us assume that x is an integer and x 1 so that the term ∞ B2m ∑k=1 2m(2m−1)x2m−1 is negligible. Then, we have the summation x 1 ∑ log k = A + x + log x − x k=1 2 and hence

2x 1 ∑ log k = A + 2x + log(2x) − 2x. k=1 2 Moreover,

x x x 1 ∑ log(2k) = ∑ log k + ∑ log 2 = x log 2 + A + x + log x − x. k=1 k=1 k=1 2 Euler and the Γ-Function 89

Figure 1.14: Euler and Stirling’s Constant Taken from [E212]. Euler√ uses the Wallis product formula for π to ﬁnd the Stirling constant, log 2π, in the Stirling formula. Euler wrote l for log.

Therefore, ﬁnally

x 2x x 1 ∑ log(2k − 1) = ∑ log k − ∑ log(2k) = x log x + x + log 2 − x. k=1 k=1 k=1 2 Now, Euler’s idea also was to use the Wallis product formula for π, i.e.

π 2 2 · 4 4 · 6 6 · 8 8 · 10 = · · · · · etc. 2 1 3 · 3 5 · 5 7 · 7 9 · 9 Therefore, ! π x x log = lim 2 log(2k) − log(2x) − 2 log(2k − 1) . x→∞ ∑ ∑ 2 k=1 k=1 Thus, combining the corresponding equations and taking the limit, we ﬁnd π log = 2A − 2 log 2 2 and hence √ A = log 2π. Finally, we can write down our expression for y(x) = log Γ(x) in its ﬁnal form:

1 √ ∞ B log Γ(x) = x log x − x − log x + log 2π + 2m . ∑ ( − ) 2m−1 2 k=1 2m 2m 1 x Finally, taking exponentials we arrive at the limit formula. Concerning this formula, we stress here again that this is an asymptotic expansion and thus one rather should write: 90 Main Thesis

1 √ ∞ B log Γ(x) ∼ x log x − x − log x + log 2π + 2m . ∑ ( − ) 2m−1 2 k=1 2m 2m 1 x

1.6.5 Generalized Factorials

1.6.5.1 Euler’s Results on generalized factorials

In [E661], as the title Variae considerationes circa series hypergeometricas45 suggests, Euler studied more general factorials. The generalisation is that the difference equation f (n + 1) = n f (n) is now replaced by the slightly more general one f (n + 1) = (a + bn) f (n) with positive real numbers a, b. Having discussed the moment ansatz, we can easily solve such equations, but Euler had other ideas. He wanted to ﬁnd Stirling-like formulas and used the Euler-Maclaurin summation formula46. Otherwise, the ideas in this paper are not new. Regardless, we want to state his results. He introduced the following functions:

Γ(i) = a(a + b)(a + 2b)(a + 3b) ··· (a + (i − 1)b) ∆(i) = a(a + 2b)(a + 4b) ··· (a + (2i − 2)b) θ(i) = (a + b)(a + 3b) ··· (a + (2i − 1)b).

He established several relations among them and found the Stirling-like formulas, i.e. asymptotic expansions,

−i a +i− 1 Γ(i) ∼ Ae (a − b + bi) b 2 −i a +i− 1 ∆(i) ∼ Be (a − 2b + 2bi) 2b 2 −i a +i θ(i) ∼ Ce (a − b + 2bi) 2b .

All equations hold only for i → ∞. In these formulas, Γ(i) is not to be confused with our Γ-function. We simply adopted Euler’s notation from [E661]. Euler could not determine the constants A, B, C as in the case of Stirling’s formula, since he did not have access to a Wallis-like product and could not argue analogously as we showed in section 1.6.4.1. But we, in possession of all necessary tools, will evaluate these constants in the next sections.

1.6.5.2 Evaluation of the Constant A

We will present the evaluation of the constant A in the ﬁrst of the three asymptotic expansions. We claim that the following calculations could all have been done by Euler himself, since all steps involved concepts or techniques that he used elsewhere in a similar way. We have:

45The term "hypergeometric series" that Euler used can be translated as "factorial series" in this context. 46The Euler-Maclaurin summation formula will be discussed in the following section 1.7, but it will turn out that the summation formula is just a special case of the formula we found for the solution of the difference equation in this section. Euler and the Γ-Function 91

Figure 1.15: Euler’s generalized Factorials Euler’s deﬁnition of the generalized factorials he considered in [E661]. The scan is taken from the same paper.

Theorem 1.6.9. Let F be a meromorphic function that is logarithmically convex47 on the positive real axis. Furthermore, let F satisfy:

F(x + 1) = (a + bx)F(x) ∀ a, b, x > 0 and F(1) = 1. Then F has the following asymptotic expansion:

a − 1 +x − F(x) ∼ A (a − b + bx) b 2 e x for x → ∞ with √ − a − 1 1− a b b 2 2πe b A = a . Γ 1 + b Proof. That the general asymptotic formula is correct follows from an application of the Euler- Maclaurin summation formula, which will be discussed in section 1.7, and was already shown by Euler in [E661], as we mentioned above. Thus, we will only determine the constant A. To this end, recall the solution for the propounded difference equation we found in section 1.5.4.4, i.e.

Z∞ x−1+ a − t a F(x) = C t b e b dt for x − 1 + > 0. b 0 C is a constant that we did not need in the previous investigations. But we will determine it here. t Before doing so, let us express F in terms of the familiar Γ-function. Substituting b = y in the above integral, we ﬁnd:

Z∞ x+ a x−1+ a −y x+ a a F(x) = C b b y b e dy = Cb b Γ x + , b 0 where we used the integral representation of the Γ-function in the last step. Therefore, from F(1) = 1:

47Of course, Euler did not know the deﬁnition of a logarithmically convex function. But we state the result as general as possible. 92 Main Thesis

− a −1 b b C = a . Γ 1 + b Finally, we have to use the Stirling-formula from section 1.6.4.1 yielding the asymptotic expansion a of Γ(x + 1) for large x; we just have to replace x by x + b in that formula and ﬁnd the asymptotic expansion:

− a −1 + a + 1 b b a √ a x b 2 a x+1+ b −x− b F(x + 1) ∼ a b 2π x + e for x → ∞ Γ 1 + b b Therefore,

− a −1 + a − 1 b b a √ a x b 2 a x+ b −x− b +1 F(x) ∼ a b 2π x − 1 + e x → ∞. Γ 1 + b b Massaging this into a form resembling Euler’s:

− a − 1 √ b b 2 x+ a − 1 a b 2 −x 1− b F(x) ∼ a 2π (bx − b + a) e · e x → ∞. Γ 1 + b Therefore, comparing this to Euler’s result we ﬁnd: √ 1− a 2π · e b a 1 − b − 2 A = a · b . Γ 1 + b

1.6.5.3 Finding the other Constants Having found A in this way, one can proceed analogously to ﬁnd the others. Concerning the function ∆(x), satisfying

∆(x + 1) = (a + 2bx)∆(x), a, b > 0 with ∆(1) = 1, we can derive the asymptotic expansion just replacing b by 2b in the expansion found for F. Hence we ﬁnd:

− a − 1 √ (2b) 2b 2 x+ a − 1 a 2b 2 −x 1− 2b ∆(x) ∼ a 2π (2bx − 2b + a) e · e x → ∞. Γ 1 + 2b Thus, the constant B in Euler’s formulas from section 1.6.5.1 is:

− a − 1 (2b) 2b 2 √ a 1− 2b B = a 2πe . Γ 1 + 2b Finally, Euler introduced the function θ(x) satisfying

θ(x + 1) = (a + (2b + 1)x)θ(x), a, b > 0 with θ(0) = 1. Additionally, it is logarithmically convex, as it follows from Euler’s deﬁnition given in 1.6.5.1. Here, one is inclined to derive the asymptotic expansion from the one of F again. But this does not lead to a form resembling Euler’s. To get to such a form, note that Euler and the Γ-Function 93

1 ∆ x + = kθ(x) 2 for a constant k, since both sides satisfy the same functional equation and are logarithmically 1 convex on the positive real axis. From the special case x = 0 the constant k is found to be = 1 . ∆( 2 ) Let us evaluate k. First, we have:

− a −1 (2b) 2b a a x+ 2b ∆(x) = a · (2b) · Γ x + . Γ 1 + 2b 2b Hence

1 a 1 1 Γ + − 2 2 2b ∆ = (2b) · a . 2 Γ 1 + 2b Therefore, we can now use the asymptotic expansion of ∆ to ﬁnd the one for θ :

1 x+ a −x− 1 ( ) ∼ · ( − + ) 2b 2 θ x 1 B 2bx b a e . ∆ 2 1 Thus, the constant C in Euler’s formulas from 1.6.5.1 can be expressed in terms of k = 1 and ∆( 2 ) B, which we already found:

− 1 C = k · Be 2

1.6.5.4 Euler’s Relations among the Constants As mentioned in section 1.6.5.1, in §17 of [E661], Euler also proved some relations among the constants A, B and C in the asymptotic expansions. More precisely, he proved:

s 1 1 1 B = ∆ e 2 C and B = A∆ e. 2 2 The relation among B follows directly from our calculations so that we focus on the derivation of the other one, among A and B. To prove it from our explicit formulas for A and B, we need the following

Theorem 1.6.10 (Legendre’s duplication formula for Γ). The following equation holds for all z ∈ C:

1 √ Γ(z)Γ z + = 21−2z πΓ(2z). 2 This formula is a special case of the Gaußian multiplication formula for the Γ-function, which will be discussed in quite some detail below in section 1.9.6. 1 1 But, for now, let us consider A · ∆ 2 · e; inserting the values for A and ∆ 2 : √ − a − 1 1− a 1 a 1 b b 2 2πe b 1 Γ + − 2 2 2b A∆ · e = a · (2b) a · e. 2 Γ 1 + b Γ 1 + 2b 1 a Applying Legendre’s duplication formula for z = 2 + 2b , this equation becomes 94 Main Thesis

√ √ − a − 1 2− a − 1 − a 1 b b 2 2π · e b (2b) 2 · π · 2 b 1 a A∆ · e = Γ + . 2 1 a a 2 2 2b Γ 2 + 2b Γ 1 + 2b This expression can be simpliﬁed to

− a −1 2− a 1 (2b) b · 2π · e b A∆ · e = 2 a 2 Γ 1 + 2b The right-hand side is just B2; thus, the relation is proven. Finally, we want to point out that Euler in [E661] found the relations among A, B, C from the relations among the functions ∆, F and θ. Therefore, Legendre’s duplication formula for the Γ-function follows from the relations among A, B and C. Euler did not notice this, of course. But below in 1.9.6.3 we will show that later in his career Euler found a formula equivalent to the multiplication formula for the Γ-function containing Legendre’s formula as a special case. Euler and the Γ-Function 95

1.7 Solution of log Γ(x + 1) − log Γ(x) = log x via the Euler-Maclaurin Formula

1.7.1 Overview on the Section

We will present Euler’s derivation in section 1.7.2 and the modern derivation of the Euler- Maclaurin summation formula in section 1.7.4. Euler also saw the formula as the solution of the difference equation, i.e., as a tool to calculate ﬁnite sums.

1.7.2 Euler’s Derivation of the Formula

Euler derived the Euler-Maclaurin summation formula on various occasions. The ﬁrst occasion was in 1738 in [E25], but he also gave derivations in 1741 in [E47] and in his book [E212] published 1755. He saw it as a tool to calculate ﬁnite sums and this helps to understand his derivation. For, we already mentioned that

n f (n) := ∑ g(k) k=1 satisﬁes the functional equation

f (n) − f (n − 1) = g(n) ∀n ∈ N.

And the idea to ﬁnd the function f is still the same as in section 1.6.2 (or in [E189]), i.e. to solve the above difference equation. Using Taylor’s theorem, he wrote

∞ (−1)k f (k)(n) f (n − 1) = ∑ . k=0 k! This, as above, led him to a differential equation of inﬁnite order, i.e.

∞ (−1)k f (k)(n) ∑ = g(n). k=1 k! Now, it is important to note that at the time of his ﬁrst proof in 1738 in [E25] and later in 1741 in [E47], Euler had not developed the theory how to solve differential equations with constant coefﬁcients yet. But Euler had another idea: He made an educated guess. More precisely, he assumed the solution to be of the following form:

n Z dg(n) d2g(n) d3g(n) d4g(n) f (n) = α g(k)dk + β + γ + δ + ε + etc. dn dn2 dn3 dn4

Inserting this ansatz into the differential equation of infinite order gives recursive relations to define the coefficients α, β, γ etc. 96 Main Thesis

Figure 1.16: Finding the Euler-Maclaurin Summation Formula Scan taken from [E47]. Euler makes the ansatz with undetermined coefficients α, β, γ etc. to solve the differential equation of infinite order (derived from the difference equation S(x + 1) − S(x) = X) for S. In the following paragraphs, he found the recursive relation for the coefficients. This led him to the Euler-Maclaurin summation formula.

In his 1755 book [E212], Euler then also proved that the coefficients are generated by: z e−z − 1 if it is expanded into a Taylor series around the origin48. This is almost the modern definition of the Bernoulli numbers49. Definition 1.7.1 (Bernoulli Numbers). The Bernoulli numbers are defined via a generating function. More precisely, we define the n-th Bernoulli number Bn via

z ∞ B = n n z ∑ z . e − 1 n=0 n! Euler calculated many of the Bernoulli numbers, which were deﬁned by Jacob Bernoulli in the context of combinatorics in his 1713 book [Be13], and proved some elementary properties about 1 them, e.g., that all odd Bernoulli numbers except B1 = − 2 vanish. He was mainly interested in ( ) = 1 + 1 + 1 + ··· them, since they appear in his formula for ζ 2n 12n 22n 32n , a connection he realized for the ﬁrst time in 1750 in [E130]. See also Sandifer’s article in [Br07] (pp. 279 - 303). Using the Bernoulli numbers, one can write the Euler-Maclaurin formula as

n n ∞ k Z B + d g(n) f (n + 1) = g(k) = g(k)dk + k 1 . ∑ ∑ ( + ) k k=0 k=0 k 1 ! dn The constant of integration must be determined in such a way that the condition f (1) = g(0) is satisﬁed. This is at least, how Euler used the formula. The modern formula avoids this problem, essentially by subtracting two sums from each other. Furthermore, we stress that the inﬁnite sum does in general not converge for any value of n and is to be understood as an asymptotic series. We addressed this issue already, when we discussed the Stirling formula 1.6.4.1.

48In his earlier papers on the summation formula, he did not realize this. 49Indeed, Euler introduced that name for those numbers in 1755 in [E212]. Furthermore, in [E746], a paper just z published in 1815, he arrived at the generating function ez−1 , which is used nowadays to introduce the Bernoulli numbers. Euler and the Γ-Function 97

Figure 1.17: The Euler-Maclaurin Sum Formula Taken from [E47]. Euler states the Euler-Maclaurin formula explicitly for the ﬁrst time. α, β, γ etc. are deﬁned recursively in the same paper. They are the Bernoulli numbers, a fact Euler did not realize at that time.

1.7.2.1 Several Remarks

The derivation of the above series is purely formal and the convergence of the sum is not guaranteed. Indeed, since the Bernoulli numbers increase rapidly (roughly as (2n)!, which follows from Euler’s formula for ζ(2n).), the series actually converges very rarely. Euler was aware of this and only used it for numerical calculations truncating the sum after a certain number of terms. The Euler-Maclaurin summation formula leads to the notion of a semi-convergent series, a term coined by Gauß in 1812 in [Ga28]. Bn is small for the first few n, whence the series seems to converge taking only a few terms50, although the sum if continued to infinity must ultimately diverge, if the derivatives of g do not vanish. An explicit formula for the remainder term, if the sum is truncated at some point, was given by Jacobi in 1834 in [Ja34a] and Poisson. The issue of semi-convergence troubled Euler and many others, including Gauß in [Ga28]. Nowadays, the right-hand side of the above series is understood as an asymptotic expansion of the sum on the right. We mention [Ha48], [Ba09] and [Va06] again as references discussing the notion of an asymptotic expansion. Leaving the issues of convergence aside, let us discuss the nature of the formula. Recalling its origin, the solution of a differential equation of infinite order, it has to be a particular solution of that equation. In other words, it has to be a special case of the solution given above in section 1.6. Staying purely formal, it is easier to understand this connection, what we will do in the following section.

1.7.2.2 Purely formal Derivation

d 1 51 We will use the idea that we can replace dx by z and an integral by z and vice versa . Above we saw that a solution of

f (x + 1) − f (x) = g(x) is given by

1 f (x) = g(x). ez − 1

50Indeed, the value found that way is often very accurate. 51Note that using the language of Fourier analysis this can be made completely rigorous, as we have seen above. 98 Main Thesis

1 And next, we expanded ez−1 into partial fractions and derived the complete solution of the simple 1 difference equation. But we can also expand ez−1 differently, i.e. into a Laurent series around z = 0. Using the deﬁnition of the Bernoulli numbers above, we ﬁnd

1 ∞ B B ∞ B = n n−1 = 0 + n+1 n z ∑ z ∑ z . e − 1 n=0 n! z n=0 (n + 1)! 1 n dn Therefore, replacing ez−1 by the right-hand side of this equation and then replacing z by dxn and x 1 R z by in the above equation, we have x Z ∞ B dng(x) ( ) = ( ) + n+1 f x B0 g t dt ∑ n . n=0 (n + 1)! dx It is easy to see that this is the Euler-Maclaurin summation formula again. Therefore, the naive and formal derivation easily reproduces both, the general solution and the particular solution provided by the Euler-Maclaurin series, from the general difference equation.

1.7.3 Historical Overview For the sake of completeness, it will be convenient to give a least a short overview about the results in the theory of differential equations of inﬁnite order. The overview can be subdivided in formal results and rigorously proven results. Naturally, the formal results extend much further, but will we see that the general formulas, derived in purely formal manner, do not hold in every case. And the study of the ﬁrst rigorous results will reveal, why the the formal approaches are not correct in the most general case.

1.7.3.1 Formal Approaches - From Euler to Bourlet Although many people contributed to the formal theory of differential equations of inﬁnite order, including Lagrange [La72], Laplace [La20] and many others, the most general result was derived by Bourlet [Bo97] and [Bo99], who basically treated the problem of solving a differential equation as a problem to ﬁnd the left-inverse operator of the corresponding differential operator as we did d in the case of the general difference equation. He wrote z = dx , and considered the equation ∞ dn ( ) ( ) = ( ) ( ) = ( ) F x, z f x ∑ an x n f x g x , n=0 dx where Bourlet, like Euler, did not specify the functions an(x), f (x), g(x). Now Bourlet’s simple idea was that, since F(x, z) is an operator, it has (in modern language), a left inverse X(x, z). And treating z as a variable quantity, he derived a partial differential equation that determines X(x, z). It reads as follows:

∞ 1 ∂nX ∂nF · = ∑ n n 1. n=0 n! ∂z ∂x As appealing as this formula might look, it is not true in general. To see this, this consider the equation:

∞ (−x)n dn ( ) = ( ) ∑ n f x g x . n=0 n! dx Euler and the Γ-Function 99

On the one hand we see, that the left hand side is by Taylor’s theorem equal to:

f (0). Thus, g(x) cannot be chosen arbitrarily. On the other hand, Bourlet’s formula would lead to an inverse function, because the corresponding differential equation can be solved. So it gives a solution to an ill-deﬁned question.

1.7.3.2 Rigorous Results - From Carmichel to today Therefore, formally the theory is established by Bourlet’s formula, a beautiful account of this is given in [Da36], where the main focus is put on solving the equations by mostly formal means. Because of the problems with the formal procedure, mathematicians considered more special problems and considered more special classes of functions, for which rigorous results can be proven. One of the ﬁrst overview papers on the rigorous results was [Ca36]. Most of the results described in that paper only consider the existence of a solution, but do not give explicit solution formulas. One exception is the theorem we want to quote here.

Theorem 1.7.1. In the linear differential equation of inﬁnite order

0 a0y + a1y + ··· = φ(x) let the constants aν be such constants that the function

2 F(z) = a0 + a1z + a2z + ··· is analytic in the region |z| ≤ q, i.e. inside a disc52, where q is a given positive constant or zero, and let φ(x) be a function of exponential type not exceeding q. If F(z) vanishes at least once in the region |z| ≤ q, let n be the number of its zeros in this region (each counted according to its multiplicity) and let P(z) be F(z) ( ) the polynomial of degree n with leading coefﬁcient unity, that P(z) does not vanish in the region. If F z does not vanish in the region, let P(z) be identically equal to 1. When P(z) ≡ 1, let Pn−1(z) be identically equal to zero; otherwise, let it be an arbitrary polynomial of degree n − 1 (including the case of an arbitrary constant when n = 1). Then the general solution y(x), subject to the condition that it shall be a function of exponential type not exceeding q, may be written in the form

1 Z Ψ(s) 1 Z P − (s) y(x) = ds + n 1 ds 2πi Cρ F(s) 2πi Cρ Pn(s) where

∞ φ(ν)(0) ( ) = Ψ s ∑ ν+1 , ν=0 s and where Cρ is a circle of radius ρ about 0 as center, ρ being greater than q and such that F(z) is analytic in the region q > |z| ≤ ρ and does not vanish there. If φ(x) is exactly of exponential type q, then the named solution y(x) is also of exponential type q.

52To be precise, we should rather write |z| < q, because for differentiation we need open domains, but Carmichel’s paper states the theorem with ≤, which we simply adopted here. 100 Main Thesis

Having stated this theorem, we see, how carefully it is formulated and that the class of functions is quite restricted. But we have to keep in mind that in the time of this paper, i.e. 1936, the notion and theory of distributions did not exist, making the theorems quite difﬁcult to state. The more recent treatments, e.g., [Du10] start with constructing the appropriate function spaces, before solving any differential equations.

1.7.4 Modern Derivation

It will be illustrative to compare a modern proof of the Euler-Maclaurin summation formula to Euler’s idea. Most modern proofs in modern introductory textbooks are similar. We will present it as it is found in [Koe00] (pp. 223-226). The proof in [Va06] is the same. The proof of the general formula proceeds in several steps, slowly ascending from special cases to the general formula.

1.7.4.1 Euler Maclaurin Formula for C1−functions

We ﬁrst have to deﬁne an auxiliary function

Deﬁnition 1.7.2. We deﬁne a function H : R → R as follows

x − [x] − 1 for x ∈ R \ Z H(x) := 2 0 for x ∈ Z

[x] is the Gauß bracket of x and expresses the integer part of x.

This function obviously has period 1. Having introduced H in this way, we can state the elementary form of the summation formula

Theorem 1.7.2 (Euler summation formula (simple version)). Let f : [1, n] → C, n ∈ N be a once continuously differentiable function, then

n n n Z 1 Z ∑ f (k) = f (x)dx + ( f (1) + f (n)) + H(x) f 0(x)dx. = 2 k 1 1 1

Proof. By integration by parts over the interval [k, k + 1], one has

k+1 + k+1 Z 1 k 1 Z 1 1 · f (x)dx = x − k − f (x) − x − k − f 0(x)dx. 2 k 2 k k

1 0 0 53 Since the function (x − k − 2 ) f is identical to H f in the interval [k; k + 1] , their integrals over the same intervals are identical, thus,

k+1 k+1 Z 1 Z 1 · f (x)dx = ( f (k + 1) − f (k)) − H(x) f 0(x)dx. 2 k k

1 Summation over k from 1 to n − 1 and addition of 2 ( f (1) + f (n)) then gives the formula. 53Maybe not at the end points of the intervals, but this does not matter in the following. Euler and the Γ-Function 101

1.7.4.2 General Euler-Maclaurin Summation Formula As in the simple case, we need to introduce some auxiliary functions

Deﬁnition 1.7.3. We deﬁne functions Hk : R → R recursively as follows: 1)Hk is a primitive of Hk−1, k ≥ 2 ∈ N and H1 := H R1 2) Hk(x)dx = 0. 0 Now we can state the general Euler-Maclaurin summation formula

Theorem 1.7.3 (Euler-Maclaurin Summation Formula). Let f : [1, n] → C be a C2k+1-function and k ≥ 1. Then we have

n n n Z " k # 1 (2κ−1) ∑ = f (x) + ( f (1) + f (n)) + ∑ Hκ(0) f + R( f ); ν=1 2 κ=1 1 1 with

Z n (2k+1) R( f ) = H2k+1 f dx. 1

Proof. One just has to note that Hk is a periodic function for all k, which is seen as follows by induction. We know that H1 = H is periodic. Thus, consider

x+ Z 1 Z1 Hk+1(x + 1) − Hk+1(x) = Hk(t)dt = Hk(t)dt = 0, x 0 where we used the deﬁning properties of Hk and the induction assumption that Hk is periodic. Having noticed this, one proceeds just as in the case of the simple Euler-Maclaurin formula, but integrates by parts 2k + 1-times to arrive at the formula.

1.7.4.3 Comparison to Euler’s Idea First, we want to mention that this idea of the modern proof is basically Jacobi’s, see [Ja34a]. Furthermore, we want to point out that we have in general

1 H (0) = B . k k! k Thus, Euler’s result and the modern result agree. Anyhow, the modern proof does not obtain the formula from the solution of a difference equation but rather starts from the results and proves it to be correct. In Euler’s case, the formula resulted as an answer to a more general question. Finally, we used the procedure of iterated integration by parts above in our derivation of the Stirling formula for n! in section 1.6.4.1 and simply ignored R( f ) in the calculation. As already mentioned in the same section, this is not justiﬁed. A rigorous reasoning requires the theory of asymptotic expansions which we did not want to discuss at that point and hence operated on a formal level. Nevertheless, if understood as an asymptotic expansion, dropping R( f ) in the Stirling-formula is justiﬁed. 102 Main Thesis

1.8 Interpolation Theory and Difference Calculus

This section mainly discusses chapter 16 and 17 of [E212] and the paper [E613], which Euler stated to be an elaboration of the mentioned chapters.

1.8.1 Overview Euler again tried to solve the difference equation

f (x + 1) − f (x) = g(x) in order to ﬁnd an explicit formula for

x ∑ g(k − 1). k=1 The sum, only deﬁned for integer x, is interpolated by the solution of the difference equation. But this time, Euler used the rules of difference calculus which he had developed in [E212] to solve the difference equation. Interestingly, addressing issues of convergence, this led him to the concept of Weierstraß products in section 1.8.5.

1.8.2 Euler’s Idea Euler’s idea, outlined in [E613], is best explained by an example. Euler tried to ﬁnd the sum

x ∑ g(k). k=1 This time, he simply added 0 in a clever way. For, we formally have

x ∑k=1 g(k) = g(1) + g(2) + g(3) + g(4) + etc. − g(x + 1) − g(x + 2) − g(x + 3) − g(x + 4) − etc. Or, in short notation

x ∞ ∑ g(k) = ∑ (g(k) − g(x + k)) . k=1 k=1 As Euler observed, this can only work, if g(k) − g(x + k) converges to zero for k → ∞. If the series does not converge, one can again add 0 in a clever way. If the resulting series does still not converge, one can do it again. Indeed, one can repeat the process arbitrarily often to increase the convergence as much as one wants. In [E613] Euler divided sums into classes according to the behaviour of their infinitesimal terms. More precisely, the first class contains those series, in which we have limk→∞ g(k) = 0. The second class contains those series whose differences of infinitesimal terms vanish, i.e. limk→∞ g(x + k + 1) − g(x + k) = 0. The third class contains the series, whose second differences vanish etc. From this definition it immediately follows that the series of the i + 1-th class are a subset of the series of the i-th class. Euler then gave formulas for the first, second and third class and explains how the general formula for the i−th class can be constructed. We will only need the first and second class for our Euler and the Γ-Function 103 discussion. The formula for the first class was stated above. Therefore, we will give the formula for the second class. It reads

x ∑k=1 g(k) = (1 − x)g(1) + (1 − x)g(2) + (1 − x)g(3) + etc. +xg(1) + xg(2) + xg(3) + xg(4) + etc. − g(x + 1) − g(x + 2) − g(x + 3) − etc. Or in short notation:

x ∞ ∑ g(k) = xg(1) + ∑ ((1 − x)g(k) + xg(k + 2) − g(x + k)) . k=1 k=1 Whereas the left-hand side only makes sense for integer x, this restriction is not necessary on the right-hand side. Therefore, we can interpolate the sum in this way. But let us consider some examples ﬁrst.

1.8.3 Examples 1.8.3.1 Harmonic Series Euler chose the harmonic series as his ﬁrst example, i.e.

x 1 1 1 1 1 f (x) = ∑ = 1 + + + + ··· + k=1 k 2 3 4 x This series belongs to the ﬁrst class and hence we can write

1 1 1 1 f (x) = + + + + etc. 1 2 3 4 1 1 1 1 − − − − − etc. x + 1 x + 2 x + 3 x + 4 And adding each two terms written above each other, we ﬁnd:

∞ x x x x x f (x) = = + + + + etc., ∑ ( + ) ( + ) ( + ) ( + ) ( + ) k=1 k x k 1 x 1 2 x 2 3 x 3 4 x 4 Trying to ﬁnd the derivative of the function, in his book [E212], Euler arrived at the Taylor series x expansion of this series. For this, he just expanded each term k(x+k) into a power series via the geometric series and summed the resulting series columnwise.

1.8.3.2 Γ(x)-function But we are mainly interested in the expression for the Γ-function arising from this idea. Since we have Γ(x + 1) = xΓ(x), we also have log Γ(x + 1) − log Γ(x) = log x. Therefore, we can, as we did above, consider log Γ(x) as the solution of the simple difference equation, and for integer x we have

x log Γ(x + 1) = ∑ log(k). k=1 104 Main Thesis

Figure 1.18: Euler’s Key Formula in the Calculus of Differences Taken from [E613]. Euler’s main formula from the paper. He used the word "summatory" to describe the sum of a series up to n terms and writes (x) to denote the sum from 1 to x. ∆k denotes the k-th difference. Furthermore, Euler did not use a separate letter to denote a function. This is suppressed in his notation, e.g., (x + 3) is to be understood a ( f (x + 3)) for a general function f .

1 Since log(k + 1) − log(k) = log 1 + k tends to zero for inﬁnite k, the series belongs to the second class. Applying the corresponding formula, we obtain

x ∑k=1 log(k) = (1 − x) log(1) + (1 − x) log(2) + (1 − x) log(3) + etc. +x log(1) + x log(2) + x log(3) + x log(4) + etc. − log(x + 1) − log(x + 2) − log(x + 3) − etc. Collecting the columns, we arrive at:

11−x2x 21−x3x 31−x4x log(Γ(x + 1)) = log(11−x) + log + log + log + etc. x + 1 x + 2 x + 3

Taking exponentials, we arrive at Euler’s ﬁrst formula for x! in [E19]54.

11−x2x 21−x3x 31−x4x ∞ k1−x(k + 1)x x! = Γ(x + 1) = 11−x · · · · etc. = + + + ∏ + x 1 x 2 x 3 k=1 x k

1.8.4 Difference Calculus according to Euler Here, we will explain in a bit more detail how Euler arrived at his formulas. To this end, we need to explain his results on difference calculus. He outlined his ideas in his book [E212] and also in [E613]. We want to state and prove his main formula from [E613] for the ﬁnite sum of x terms.

For this, we need to introduce some notation.

54There, he did not prove the formula, but just gives an heuristic argument why it is true. Euler and the Γ-Function 105

Deﬁnition 1.8.1 (n−th Difference). We denote the n-th difference by ∆ng(k) and deﬁne it recursively by

∆ng(k) := ∆n−1g(k + 1) − ∆n−1g(k) k being an arbitrary number55 and

∆0g(k) := g(k). Additionally, we sometimes write simply ∆ for ∆1. It is easily seen that we can express each term g(k) using only the differences of g(1). More precisely, we have: Theorem 1.8.1. We have for a natural number k:

k − 1 k − 1 k − 2 k − 1 k − 2 k − 3 g(k) = g(1) + ∆g(1) + · ∆2g(1) + · · ∆3g(1) + etc. 1 1 2 1 2 3 or in compact notation

∞ k − 1 g(k) = ∑ ∆ng(1). n=0 n n where (k) is the binomial coefficient. Proof. Since the binomial coefficients vanish if n > k − 1, the sum is finite and hence converges. The proof by induction. The formula is obviously true for k = 1. So let us assume that it is true for a fixed k ∈ N. Consider

∞ k − 1 g(k + 1) = g(k) + ∆g(k) = g(k) + ∑ ∆ng(k), n=0 n where we used the deﬁnition of ∆ and the induction assumption. Using the formula 56

∞ k − 1 ∆g(k) = ∆ng(1), ∑ − n=1 n 1 we arrive at ! ∞ k − 1 ∞ k − 1 g(k + 1) = ∆ng(1) + ∆ng(1) , ∑ ∑ − n=0 n n=1 n 1 or

k−1 k − 1 k k − 1 g(k + 1) = ∆ng(1) + ∆ng(1). ∑ ∑ − n=0 n n=1 n 1 Let us extract the ﬁrst term of the ﬁrst sum and the last term of the second sum such that

k−1 k − 1 k−1 k − 1 g(k + 1) = g(1) + ∆ng(1) + ∆ng(1) + ∆kg(k). ∑ ∑ − n=1 n n=1 n 1

55In most cases k will be a natural number. 56The formula follows by taking ∆ on both sides of the equation for g(k). 106 Main Thesis

k−1 k−1 k Contracting the sums and using the well-known identity ( n ) + (n−1) = (n), we have

k−1 k g(k + 1) = g(1) + ∑ ∆ng(1) + ∆kg(k). n=1 n Finally, absorbing the two isolated terms into the sum and using the vanishing of the binomial coefﬁcients for n > k we arrive at

∞ k g(k + 1) = ∑ ∆ng(1). n=0 n

x Next, we want to determine the sum ∑k=1 g(k). This is done in the following theorem.

Theorem 1.8.2. We have

x ∞ x ∑ g(k) = ∑ ∆kg(k). k=1 k=1 k

The proof is by induction again along the same lines as the last. Therefore, we omit it here. we want to make some remarks instead. Although the proof explicitly assumes x to be a natural number, the right-hand side of the above equation does not require x to be an integer57 and hence interpolates the sum on the left-hand side. In other words, the right-hand side solves the functional equation satified by the finite sum, x x−1 1 i.e. ∑k=1 g(k) − ∑k=1 g(k) = g(x) with the initial condition ∑k=1 g(k) = g(1). And Euler precisely did this replacement, even addressing the issue of convergence of the then infinite series. But let us now go over to Euler’s idea of adding zero in a clever way to enforce convergence. First note that from the above theorem we also have:

∞ x g(x + 1) = ∑ ∆kg(1). k=0 k But we can ﬁnd similar expressions for g(x + 2) using g(2). We have:

∞ x g(x + 2) = ∑ ∆kg(2). k=0 k And in general for n:

∞ x g(x + n) = ∑ ∆kg(n). k=0 k Therefore, using the sum representation we found above, we arrive at the following equation:

57The binomial coefﬁcients can be deﬁned for non-integer numbers replacing the factorials by Γ-functions. Indeed, this is a consequence of Newton’s generalized binomial theorem proved by Euler in [E465]. Euler and the Γ-Function 107

Theorem 1.8.3 (Equation for sum in terms of differences).

x ∑k=1 g(k) =

x x 1 x 2 (1)g(1) + (2)∆ g(1) + (3)∆ g(1) + ···

x 1 x 2 x 3 +g(1) + (1)∆ g(1) + (2)∆ g(1) + (3)∆ g(1) + · · · − g(x + 1)

x 1 x 2 x 3 +g(2) + (1)∆ g(2) + (2)∆ g(2) + (3)∆ g(2) + · · · − g(x + 2)

x 1 x 2 x 3 +g(3) + (1)∆ g(3) + (2)∆ g(3) + (3)∆ g(3) + · · · − g(x + 3)

......

x 1 x 2 x 3 +g(n) + (1)∆ g(n) + (2)∆ g(n) + (3)∆ g(n) + · · · − g(x + n)

Proof. The proof is immediate from the preceding. The ﬁrst row is just the alternate representation of the sum. Each following row is simply = 0 by the results we stated.

And this is Euler’s fundamental formula. For, he proceeded to sum the series column by column from n = 1 to n = ∞. And hence it is easily seen that the definition of the classes we mentioned above makes sense. For, if any iterated difference vanishes, one only has a finite number of columns to sum and the series converges58. It is easily seen, how the examples we gave (from the first and second class) result from this formula. In the next section, we will explain how this idea actually anticipates the idea of Weierstraß factors.

1.8.5 Modern Idea - Weierstraß Product We introduce, but not prove, Weierstraß’s product theorem and show the connection to Euler’s ideas outlined in the last sections. The exposition of Weierstraß’s theory follows [Fr06] (pp. 213 -217).

1.8.5.1 Introduction to the Problem Weierstraß considered the following problem in his 1878 paper [We78]: Given a domain D ⊂ C and a discrete subset S in D, can one construct an analytic function f : D → C with zeros of a given order ms precisely in S?. The answer to this question is yes and the task can be solved by Weierstraß products. Let us see how we arrive at the concept. For the sake of simplicity, let us take D = C. First, we note that closed disks are compact sets and hence there are only ﬁnitely many s ∈ S with |s| ≤ N ∈ N. Thus, S is a countable set and the elements can be ordered with respect to their magnitude

S = {s1, s2, · · · } |s1| ≤ |s2| ≤ |s3| ≤ · · · . If S is a ﬁnite set, we know how to solve the problem, the solution is given by the polynomial

58Under some mild additional assumptions. 108 Main Thesis

∏(z − s)ms . s∈S For inﬁnite sets on the other hand, the product obtained in this way cannot converge in general. But we can assume S to not contain zero, since we can multiply by zm0 at the end. We want to do this, since we can focus on products of the form

∞ z mn − = ∏ 1 , mn : msn . n=1 sn Indeed, Euler already had this idea and it led him to the discovery of the sine product in [E41] and its proof in [E61]59. 2 This product still does not always converge (it converges, e.g., for sn = n , mn = 1 but diverges for sn, mn = 1). Weierstraß had the idea to multiply it by factors not changing the zeros but forcing the product to converge. He made the ansatz:

∞ z mn f (z) := ∏ 1 − · ePn(z). n=1 sn

Pn(z) is a polynomial still to be determined. We have to ensure that

z mn lim 1 − ePn(z) = 1 ∀z ∈ C. n→∞ sn

This is possible, since it is easily seen that we can ﬁnd an analytic function An(z)

mn z An(z) 1 − e = 1 ∀z ∈ U|sn|(0) sn with An(0) = 0. The power series An converges uniformly in each compact subset of the disk

U|sn|(0). Therefore, truncating this power series for An, we can easily ﬁnd a polynomial Pn(z) with

mn z ( ) 1 1 − − Pn z ≤ | | ≤ | | 1 1 e 2 for all z with z sn . sn n 2

1 1 Since the series 1 + 4 + 9 + ··· converges, we arrive at the theorem:

Theorem 1.8.4. The series

∞ mn z ( ) 1 1 − − Pn z ≤ | | ≤ | | ∑ 1 1 e 2 for all z with z sn n=1 sn n 2 converges normally.

Concerning the problem propounded initially, we can now formulate Weierstraß’s factorisation theorem.

59The discussion of his proof can be found in section 1.9.4.1 Euler and the Γ-Function 109

Theorem 1.8.5 (Weierstraß’s factorisation theorem). Let S ∈ C be a discrete subset. Further, let the following map be given

m : S → N, s 7→ ms. Then, there exists an analytic function

f : C → C with the properties: 1)S := {z ∈ C| f (z) = 0} 2) ms = ord( f ; s). (Order of the zero.)

1.8.5.2 Comparison to Euler’s Idea Let us compare both, Weierstraß’s idea from 1878 in [We78] and Euler’s idea from [E613], a paper written in 1780 and published in 1813. In his paper, Euler added zero and expressed the series under consideration in an alternative way to obtain a more convergent series, whereas Weierstraß told us to multiply by additional exponential of polynomials to ensure the convergence. But it is easily seen that both ideas are actually equivalent. The following quote from the Introduction written by G. Faber in Volume 16,2 of the ﬁrst series of Euler’s Opera Omnia conﬁrms this. See p. XLIII.

Tatsächlich hat Euler nicht nur die Produktdarstellung (12) [this means the product expansion of the Γ-function], sondern sogar den Gedanken der Konvergenz erzeugenden Faktoren von Weierstraß vor- weggenommen. Denn es bedeuted keinen Unterschied, ob man den Gliedern des divergenten Produktes x ∞ x − ν ∏ν=1 1 + ν die Konvergenz erzeugenden Fakoren e oder den Gliedern der divergenten unendlichen ∞ x x 1 Reihe ∑ν=1 log 1 + ν die Konvergenz erzeugenden Summanden − ν oder auch −x log 1 + ν beifügt. Das tat aber Euler mit voller Absicht in der Abhandlung 613.

For, Euler’s idea translates into the one of Weierstraß by considering sums of logarithms. Euler even told us how to find those factors you need to enforce convergence60. Therefore, Euler actually anticipated the idea of Weierstraß factors without actually intending it. His intention, as we saw above, was the interpolation of a sum, which is defined for positive integer numbers, to all numbers. Nevertheless, Weierstraß’s name is attached to the idea, since he constructed a rigorous theory of infinite products and he, by solving the problem propounded above 1.8.5, also provided the mathematical community with a large class of analytic functions. Euler’s contribution was maybe overlooked, since he was interested in something completely different and did not point out the generality of his method clearly enough.

60Indeed, there is even a prescription how to ﬁnd those factors for the Weierstraß product. Confer, e.g. [Fr06]. 110 Main Thesis

1.9 Relation among Γ and B

This section is entirely devoted to the connection between the Γ- and B-function.

1.9.1 From B to Γ - Euler’s ﬁrst Way to the Integral Representation 1.9.1.1 Euler’s Thought Process The 1738 paper [E19] is interesting, since Euler described his thought process how he got the idea that the Γ-function can be expressed as an integral. This provides us with a beautiful example of the Ars inveniendi (Art of Finding). He explains his thoughts in §§3 − 7. He wrote:

I had believed before that the general term of the series 1, 2, 6, 24 etc., if not algebraic, is nevertheless given as an exponential. But after I had understood that certain terms depend on the quadrature of the circle, I realized that neither algebraic nor exponential quantities sufﬁce to express it. [...]. But after I had considered that among differential quantities there are formulas of such a kind, which admit an integration in certain cases and then yield algebraic quantities, but in others do not admit an integration and then exhibited quantities depending on quadratures, it came to mind that maybe formulas of this kind are apt to express the general terms of the mentioned and other progressions.[...]. But the differential formula must contain a certain variable quantity.[...] For the sake of clarity, I say that R pdx is the general term of the progression to be found as follows from it; but let p denote a function of x and constants, amongst which here n must be contained61. Imagine pdx to be integrated and such a constant to be added that for x = 0 the whole integral vanishes; then set x equal to a certain known quantity. Having done this, if in the found integral only quantities extending to the progression remain, it will express the term, whose index is n. In other words, the integral determined like this will be the general term.[...] Therefore, I considered many differential formulas only admitting an integration, if one takes n to be a positive integer number, so that the principal terms become algebraic, and hence formed progressions. To summarize Euler’s idea in modern formulation: He propounded that there is a function p(x, n) such that

b Z n! = p(x, n)dx for n ∈ N. a In the following paragraphs, he really tried out several different functions and integrals (which essentially all boil down to the B-function) and eventually arrived at the integral representation of the Γ-function. We will discuss his proof in the following sections, in section 1.9.1.2 and in section 1.9.1.3. But here we want to stress that the theory of parameter integrals did not really exist at the time. Euler was the first to develop this theory. Furthermore, this is one, if not the first paper, in which parameter integrals or functions defined through integrals have been discussed.

1.9.1.2 Euler’s Mathematical Argument It is interesting, how Euler arrived at the integral representation of Γ(x) for the ﬁrst time in his 1738 paper [E19]. He repeated his argument in 1772 in [E421]. For a review of the following

61Euler wanted the parameter integral to express the n-th term of the progression. Thus, the parameter integral must contain n, which is not to be integrated over. Euler and the Γ-Function 111 argument, using Euler’s notation, confer, e.g., the article on the Γ-function in [Sa15], [Va06] or [Du99]. First, he showed that

1 1 · 2 · 3 ··· n f + (n + 1)g Z f = x g dx(1 − x)n. ( f + g)( f + 2g) ··· ( f + ng) gn+1 0 This is easily proved by induction. Now it is easy to see that one arrives at an expression for 1 · 2 ··· n = n!, if one sets g = 0, i.e.

1 f n! Z x g (1 − x)n = f lim dx. f n g→0 gn+1 0 g To get rid of the g in the denominator, Euler set x = y f +g (and uses then y = x again) and arrived at:

g n 1 f +g n! Z g 1 − x = f lim dx. f n g→0 f + g gn+1 0 We are mainly interested in the case f = 1, i.e.

g n 1 1+g Z g 1 − x n! = lim dx. g→0 1 + g gn+1 0 But this is the same as:

g n Z1 1 − x 1+g n! = lim dx. g→0 gn 0 We can pull the limit into the integral, i.e.

1 Z (1 − xg)n n! = lim dx. g→0 gn 0 Note that we already took the limit62 in the denominator of the power of x. Let us rewrite this as

1 Z (1 − xg) n n! = lim dx. g→0 g 0 This is allowed, since xn is a continuous function. But for natural n this limit can be found by L’Hospital’s rule. And one ﬁnds:

1 Z 1 n n! = log dx. x 0 This is the integral representation of Γ(n + 1). Hence it is easily understood why Euler preferred to work with this representation. He was led naturally to it.

62This is allowed since all functions are continuous. 112 Main Thesis

1.9.1.3 Using the B-function Euler did not consider the B-function as an independent function at the time he wrote [E19] (in 1738) and [E421] is devoted to the integral representation of the Γ-function. Therefore, let us see how Euler’s argument can be formulated using the known properties of B. We also start from

1 1 Z x g (1 − x)n n! = lim dx. g→0 gn+1 0 Let us rewrite the integral as a B-function:

1 B g + 1, n + 1 n! = lim dx. g→0 gn+1 x Using the functional equation B(x + 1, y) = x+y B(x, y), we have 1 B 1 n + g g , 1 n! = lim . g→0 1 gn+1 g + n + 1

1 It is more convenient to put g = h and consider the following limit h n! = lim hn+1B (h, n + 1) . h→∞ n + h + 1 This is the same limit as

n! = lim hn+1B (h, n + 1) . h→∞ ( ) = Γ(x)Γ(y) Let us use the relation B x, y Γ(x+y) : Γ(h)Γ(n + 1) n! = lim hn+1 . h→∞ Γ(n + h + 1) Using the functional equation of the Γ-function h times in the denominator, we ﬁnd

Γ(h)Γ(n + 1) n! = lim hn+1 . h→∞ Γ(n + 1)(n + h)(n + h − 1) ··· (n + 1) Or equivalently

hn+1Γ(h) n! = lim . h→∞ (n + h)(n + h − 1) ··· (n + 1) Interestingly, we arrived at the condition Weierstraß used to deﬁne Γ.

1.9.1.4 Gauß’s Idea Now that we have seen that it is possible to get to the Γ-function from the B-function, for the sake of completeness, let us also brieﬂy mention Gauß’s idea, which he presented in his 1828 paper Euler and the Γ-Function 113

[Ga28]. His idea is basically the same as Euler’s. We already mentioned that Gauß deﬁned the Γ-function as Weierstraß did later63 as the above limit, i.e. he deﬁned

nxn! Γ(x) = lim . n→∞ (x + 1) · (x + 2) ··· (x + n) And Gauß observed, essentially as Euler did in 1738 in [E19], that

n − nxn! Z t n 1 = tx−1 1 − dt (x + 1) · (x + 2) ··· (x + n) n 0 and hence

n − nxn! Z t n 1 Γ(x) = lim = tx−1 1 − dt for Re(x) > 0. n→∞ (x + 1) · (x + 2) ··· (x + n) n 0

Hence he concluded:

Z∞ Γ(x) = e−ttx−1dt for Re(x) > 0. 0 It is interesting that, although starting from the same idea, the same identity even, Gauß and Euler got to the integral representation in such different ways. Euler’s proof can even be considered rigorous by today’s standards64, whereas Gauß’s approach is harder to make it rigorous. A rigorous proof was given by Schlömilch [Sc79] and is also presented in Nielsen’s book [Ni05].

1.9.2 Expressing the B-function via Γ-functions - Fundamental Relation We want to start with the formula, already proved above, relating the Γ- and B-function, i.e. the formula

Γ(x)Γ(y) B(x, y) = . Γ(x + y)

1.9.2.1 Euler’s Proof As already indicated above in section 1.5.5.18, Euler’s proof is not rigorous by modern standards, and based on the extension of the validity of the formula for natural numbers x and y to all numbers. Nevertheless, we present Euler’s arguments here, since it is interesting to see how he discovered the result. In [E421] (§26), he stated the following equation

1 1 R 1 n−1 R 1 m−1 dx log · dx log 1 x x Z 0 0 = k xmk−1dx(1 − xk)n−1, 1 m+n−1 R 1 0 dx log x 0

63Weierstaß followed Gauß’s example in [Ga28] and explicitly said so in [We78]. 64Euler did not know how to reason rigorously that the limit can be pulled inside the integral, he simply did it without any reasoning. But it can be justiﬁed by means of the Lebesgue integral. 114 Main Thesis which reduces to the desired relation for m = 1. But in his paper, he only established this equation for natural numbers n and m and not in general. In the following paragraphs, he then established the formula for some more fractional numbers, but not for the general case. In the next two sections, we will consider Dirichlet’s proof from [Di39] and Jacobi’s proof from [Ja34] and see why it was difﬁcult for Euler to prove the identity in general.

1.9.2.2 Jacobi’s Proof We present Jacobi’s proof of the fundamental relation from this 1834 paper [Ja34], the proof is also given in [Ni05].

Proof. Let x, y > 0 and consider:

Z∞ Z∞ Z∞ Z∞ Γ(x)Γ(y) = tx−1e−tdt · uy−1e−udu = e−(t+u)tx−1uy−1dtdu. 0 0 0 0 Make the substitution:

t + u = α(u, t), u = α(u, t)β(u, t). This gives:

t = α(1 − β), u(1 − β) = tβ, dt = (1 − β)dα, (1 − β)du = αdβ, which immediately leads to the formula:

Z∞ Z1 Γ(x)Γ(y) = e−ααx+y−1dα · βx−1(1 − β)y−1dβ. 0 0 This implies the desired formula.

Obviously, the hard part is to ﬁnd the substitution and actually it is only possible, if you know in advance how the ﬁnal result looks like.

1.9.2.3 Dirichlet’s Proof Dirichlet’s proof [Di39] is a bit more straight-forward.

Proof. We start from the following expression of B

∞ Z tx−1dt B(x, y) = . (1 + t)x+y 0 1 R x−1 y−1 1 It is obtained from B(x, y) = dtt (1 − t) by setting t = z and then z = u + 1. Furthermore, 0 setting t = ky, provided k > 0, in the integral representation of the Γ-function, we have

∞ Z Γ(x) e−tktx−1dt = . kx 0 Euler and the Γ-Function 115

Therefore, from the above representation of the B-function

∞ ∞ 1 Z Z B(x, y) = · tx−1dt e−(1+t)u · ux+y−1du. Γ(x + y) 0 0 We can exchange the order of integration here by Fubini’s theorem such that

∞ ∞ 1 Z Z B(x, y) = · e−uux−y−1du · e−tutx−1dt. Γ(x + y) 0 0 Performing the integral over t:

∞ 1 Z Γ(x) B(x, y) = e−uux+y−1 · . Γ(x + y) ux 0 Finally, we arrive at:

Γ(x)Γ(y) B(x, y) = . Γ(x + y)

1.9.2.4 Discussion Considering these proofs, Jacobi’s proof was out of Euler’s reach. For, Euler did not know how to perform a substitution in the case of several variables; in other words, he did not know the concept of the Jacobi determinant. This was only explained in 1841 by Jacobi in [Ja41], as we already mentioned in section 1.5.3.5. Additionally, the anticommutativity of the wedge product puzzled him in his only paper devoted explicitly to double integrals [E391]65. See also Katz’s article in [Du07] on the subject. Dirichlet’s proof on the other hand was deﬁnitely within Euler’s reach. Especially, since Euler also knew the formula

∞ Γ(x) Z = e−kttx−1dt kx 0 and derived many extraordinary integrals from it in [E675], including the Fresnel integrals. Although Fubini’s theorem was proven only a lot later, this would not have troubled Euler, since he basically considered integrals as ordinary sums, just over inﬁnitesimally small numbers. In general, the exchange of limits did not bother him at all. The necessity to prove the validity of such a procedure only came after Abel’s proof of what we now call Abel’s limit theorem for series, i.e. more than 30 years after Euler’s death. Interestingly, although Euler obviously knew the formula

Z1 B(x, y) = tx−1(1 − t)y−1dt 0

65The concept of the wedge product was introduced much later by Cartan. Indeed, Euler, could not explain why his results seemed to indicate that dxdy = −dydx and tried to argue it away. In hindsight, we might say that he encountered the wedge product for the ﬁrst time. 116 Main Thesis and devoted some of his papers to examine it, e.g., [E321], [E640] and even a chapter in his second book on integral calculus [E366], he never explicitly stated the formula:

1 ∞ Z Z tx−1 B(x, y) = tx−1(1 − t)y−1dt = , (1 + t)x+y 0 0 which would have simplified his investigations and results on definite integrals a lot. Indeed, almost all his papers on definite integrals (contained in the Opera Omnia, Series 1, Volumes 17-19) can be understood very easily using the B- and Γ-functions and their derivatives and the fundamental relation connecting them that we considered in this section. But it seems that Euler did not realize this, at least he does not mention it in any of his papers or books.

1.9.3 Relation to the Beta Function - Expressing Γ via B p In [E19] and [E122], Euler stated, but not proved, a formula which expresses Γ( q ) in terms of a product of several B-functions66. This is interesting, since the Γ-function involves a transcendental integrand, whereas B is an integral over an algebraic function (if the variables are rational numbers) and hence is a period in the sense of Zagier and Kontsevich [Ko01]. Euler stated the formula as follows in [E19]

1 s Z p 2p 3p 4p qp (− log x) q dx = q 1 · 2 · 3 ··· p + 1 + 1 + 1 ··· + 1 q q q q 0 v u 1 1 1 1 1 uZ p Z p Z p Z p Z p uq q 2 3 q 3 4 q 4 5 q q−1 q q ×t dx(x − xx) · dx(x − x ) · dx(x − x ) · dx(x − x ) ··· dx(x − x ) . 0 0 0 0 0

1 p R q p Replacing the integral (− log x) dx by Γ q + 1 and 1 · 2 · 3 ··· p by Γ(p + 1), we can formulate 0 the theorem as follows: Theorem 1.9.1. Let p > 0 and q ∈ N. Then, the following formula holds: v u q q−1 1 p u kp Z p uq i−1 i q Γ + 1 = tΓ(p + 1) × ∏ + 1 × ∏ dx(x − x ) . q = q = k 2 i 1 0 Proof. The idea is to express the integrals as B-functions, and then rewrite them in terms of Γ-functions. Let us consider the q-th power of the right-hand side and call it G(p, q). Then

q q 1 Z p p kp (i−1) G(p, q) = Γ(p + 1) × ∏ + 1 × ∏ x q dx(1 − x) q . = q = k 2 i 2 0 The integrals can be rewritten in terms of B:

q kp q (i − 1)p p G(p, q) = Γ(p + 1) × ∏ + 1 × ∏ B + 1, + 1 . k=2 q i=2 q q

66Euler wanted p and q to be natural numbers, but we will see that this restriction is not necessary. Euler and the Γ-Function 117

Next, express the B-functions via Γ-functions:

( − ) q q i 1 p + p + kp Γ q 1 Γ q 1 G(p, q) = Γ(p + 1) × + 1 × . ∏ q ∏ ip k=2 i=2 Γ q + 2

p Γ q does not depend on the multiplication index. Therefore, we can pull it out of the product. Additionally, let us use the functional equation of the Γ-function to rewrite the denominator. Then, we arrive at:

(i−1)p q−1 q q + p kp Γ q 1 G(p, q) = Γ(p + 1) · Γ + 1 × + 1 × . q ∏ q ∏ ip ip k=2 i=2 q + 1 Γ q + 1 Finally, writing everything as one product

(ip (i−1)p q−1 q + · + p q 1 Γ q 1 G(p, q) = Γ(p + 1) · Γ + 1 × . q ∏ ip ip i=2 q + 1 · Γ q + 1 Almost every factor in the product cancels. Indeed, it just remains

p p Γ q + 1 Γ q + 1 = . p pq Γ(p + 1) q + 1 · Γ q + 1 Hence we arrive at p q G(p, q) = Γ + 1 . q Therefore, we arrived at the desired result.

1.9.4 Reflection Formula Theorem 1.9.2 (Reflection formula for the Γ-function). We have: π Γ(x)Γ(1 − x) = . sin(πx) We want to discuss the proof of the theorem that Euler gave in [E421] and one he could have given. For Euler’s proof, we need to prove the product formula for the sine first.

1.9.4.1 Euler’s Proof of the Sine Product Formula The product formula for the sine, i.e.

∞ x2 ( ) = − sin πx xπ ∏ 1 2 , k=1 k is one of Euler’s most beautiful discoveries. He discovered it in 1740 in [E41] and used it in the solution of the Basel problem, i.e. the summation over the reciprocals of the squares in the same paper. 118 Main Thesis

Figure 1.19: Solution of the Basel Problem + 1 + 1 + ··· = π2 Taken from [E41] §11. Euler arrives at the famous formula 1 22 32 6 for the ﬁrst time. He writes simply p for π.

Due to criticism from Bernoulli and Cramer, he gave a proof in 1741 in [E61], which he repeated in [E101], a book published in 1748.

Concerning the criticism, Euler in [E61] §6 wrote:

This almost completely oppressed worry67 has recently been renewed by letters from Daniel Bernoulli, in which he shared the same worries concerning my method and also mentioned that Cramer has the same doubts whether my method is right.

The main point of criticism was that in [E41] Euler did not prove that the sine only has the integer numbers times π as roots. In other words, Euler could not rule out complex roots at that time. But in [E61], one year after [E41], he then offered a proof for the sine product formula. For this, he considered the expression

an − bn, n being a natural number, and its factorization into real factors. The complex factors are easily found by de Moivre’s theorem. They appear in complex conjugate pairs and hence, combing each into a real factor, each real factor reads

2k a2 − 2ab cos π + b2, n k being a natural number with 2k < n. If n is odd, one has the additional factor a − b. In the next step, Euler used the famous identity named after him to write

67The worry concerns the correctness of his product formula for the sine and the solution of the Basel problem derived from it. Euler and the Γ-Function 119

Figure 1.20: Sine Product Formula Taken from the book [E101]. Euler arrives at the sine product formula. His proof given there is rigorous even by modern standards.

eis − e−is sin s = 2i and the deﬁnition of es s n es := lim 1 + . n→∞ n Therefore, Euler considered the expression

is n is n 1 + n − 1 − n 2i and noted that for inﬁnite n this expression goes over into sin s. This expression has the form of the general expression above and hence each factor is given by is si the form we ascribed to it above. Here, a = 1 + n , b = 1 − n and hence each factor has the form

2s2 s2 2kπ 2 − − 2 1 + cos . n2 n2 n Therefore, all the factors of the sine are obtained, if all natural numbers are substituted for k. But since now n is inﬁnite we have

2kπ 2k2π2 cos = 1 − , n n2 at least approximately, whence our factor becomes

4s2 4k2π2 − + , n2 n2 or, if we want each factor to have the form 1 − ak, the general factor will be 120 Main Thesis

Figure 1.21: Reﬂection Formula for the Γ-function Taken from E421. Euler proves the reﬂection formula. He uses the symbol [λ] to denote λ!. Thus, in modern notation [λ] = Γ(λ + 1).

s2 1 − . k2π2 Therefore, we already have

∞ s2 = − sin s As ∏ 1 2 2 . k=1 k π sin x The constant A is easily seen to be = 1 from the well-known limit limx→0 x = 1. This completes Euler’s proof of the sine product formula. Although some arguments are not completely rigorous, the general idea is correct and one can work out a proof meeting today’s standard of rigor. Confer, e.g., [Va06] for a modern proof based on Euler’s ideas. Let us mention that in [E664] Euler started from the right-hand side of the last equation68 and proved that it is equal to sin x.

1.9.4.2 Euler’s Proof of the Reﬂection Formula The proof, we are about to give, is the one from [E421].

Proof. Above we had the formula

11−x2x 21−x3x 31−x4x ∞ k1−x(k + 1)x Γ(x + 1) = 1x · · · · etc. = . + + + ∏ + x 1 x 2 x 3 k=1 x k

68More precisely, he used the corresponding expression of cos x, since the proof is easier working with the product of cos x. Euler and the Γ-Function 121

Therefore, replacing x by −x:

11+x2−x 21+x3−x 31+x4−x ∞ k1+x(k + 1)−x Γ(1 − x) = 11+x · · · · etc. = . − − − ∏ − 1 x 2 x 3 x k=1 k x Therefore,

12 22 32 Γ(1 + x)Γ(1 − x) = · · ··· x2 − 12 x2 − 22 32 − x2 or in compact notation:

∞ k2 ∞ 1 Γ(1 + x)Γ(1 − x) = = ∏ k2 − x2 ∏ − x2 k=1 k=1 1 k2 The product is the well-known product formula for the sine such that

xπ Γ(x + 1)Γ(1 − x) = . sin(πx) Using the functional equation of the Γ-function, one arrives at the desired formula.

1.9.4.3 Proof Euler could have given

In [E59], Euler stated and, in [E60], [E61] and [E462], then proved the following identity:

1 Z tx−1 − t−x π dt = 0 < x < 1. 1 + t sin πx 0 He did this by considering integrals over rational fractions and solving them by partial fraction π decomposition. But using the partial fraction decomposition of sin πx , one will easily see the identity to be true by expanding the denominator into a geometric series and integrating term by term. Therefore, we will not give the proof here. Instead, we want to prove the reﬂection formula in the most simple way using only formulas at Euler’s disposal. The theorem to be proved is still the same, i.e that

π Γ(x)Γ(1 − x) = . sin(πx)

Proof. We have

∞ Z tx−1 Γ(x)Γ(1 − x) = B(x, 1 − x) = dt 1 + t 0 by using the relation among the Γ- and B-function discussed in section 1.9.2 and using the integral 1 ∞ representation with limits 0 and ∞ for B. Next, we split the integral into two integrals R + R and 0 1 1 put t = u in the second. Then, calling u t again, we arrive at: 122 Main Thesis

1 Z tx−1 − t−x π Γ(x)Γ(1 − x) = dt = . 1 + t sin πx 0

1.9.5 A slight Detour: Gamma Class Conjectures At this point, it seems appropriate to mention a rather recent application of the Γ-function, more precisely the reﬂection formula, which was the subject of the last section. We will follow [Ga18]. We want to mention brieﬂy the so-called Gamma Conjecture and how the Γ-function is connected to it. This will also show that there is a recent interest in the Γ-function.

1.9.5.1 Gamma Class Let us describe brieﬂy what a Gamma class is. The description follows [Ga16]. The Gamma class of a complex manifold X is the cohomology class

n • ΓbX = ∏ Γ(1 + δi) ∈ H (X, R) i=1 where δ1, δ2, ··· , δn are the Chern roots of the tangent bundle TX and Γ(x) is our Γ-function. Above we have seen the Taylor series for log(Γ(1 + x)), i.e.

∞ log(Γ(1 + x)) = −γx + ∑ (−1)kζ(k)xk. k=2 γ is the Euler-Mascheroni constant, ζ(s) denotes the Riemann zeta function. This Taylor series, which we gave in section 1.2.3.2, implies the following formula for the Gamma Class:

∞ ! k ΓbX = exp −γc1(X) + ∑ (−1) (k − 1)!ζ(k) chk(TX) k=2 chk is the k-th Chern character.

1.9.5.2 Gamma Conjectures The Gamma Conjecture, stated, e.g., in [Ga16] and [Ga18], is a conjecture relating quantum cohomology of a Fano manifold with its topology. The small quantum cohomology of F deﬁnes a ﬂat connection, often referred to as a quantum connection, over C× and its solution is a cohomology- valued function JF(t) called J-function. Under certain conditions, not to be discussed here, the limit of the J-function

JF(t) • AF := lim ∈ H (F) t→∞ h[pt] , JF(t)i exists and deﬁnes the principal asymptotic class AF of F. Gamma Conjecture I states that AF equals the Gamma class ΓbF = Γb(TF) of the tangent bundle of F, i.e. that

AF = ΓbF. Euler and the Γ-Function 123

Now suppose that the quantum cohomology of F is semisimple. Then, one can deﬁne higher • asymptotic classes AF,i with 1 ≤ i ≤ N = dim H (F) from exponential asymptotics of ﬂat sections of the quantum connection.

··· b ( ) Gamma Conjecture II states that there exists a full exceptional collection E1, E2, , En of Dcoh F such that

AF,i = ΓbF · Ch(Ei) i = 1, ··· , N. Here,

dim F deg p Ch(E) := (2πi) 2 ch(E) = ∑ (2πi) chp(E) p=0 is the Chern character. The principal asymptotic class AF corresponds to the exceptional object E = OF. Now let us see how the Γ-function enters into this: One can say that the Gamma conjecture is something like the square root of the index theorem. To explain this, ﬁrst recall the Hirzebuch- Riemann-Roch formula: Z ∨ χ(E1, E2) = ch E1 · ch(E2) · tdF F dim F i i for vector bundles E1, E2 of F. χ(E1, E2) = ∑i=0 (−1) dim Ext (E1, E2) is the Euler-Pairing and tdF = td(TF) is the Todd class of F. We have seen in 1.9.4 that Euler proved πx Γ(1 − x)Γ(1 + x) = . sin(πx) x Rewriting the sines in terms of exponential functions and then writing 2πi instead of x, one arrives at

x x x x = e 2 Γ 1 − Γ 1 + . 1 − e−x 2πi 2πi Recall that we have seen the left-hand side of this equation while discussing the Bernoulli numbers in section 1.7.2.2. Anyhow, the factorisation of the Γ-functions gives the factorisation of the Todd class:

deg πic1(F) ∗ (2πi) 2 tdF = e · ΓbF · ΓbF. This in turn, factorises the Hirzebuch-Riemann-Roch formula: h χ(E1, E2) = Ch(E1) · ΓbF, Ch(E2) · ΓbF .

∗ deg dim F p dim F In these formulas: ΓbF = (−1) 2 ΓbF = ∑p=0 (−1) γp, if one writes ΓbF = ∑p=0 γp with γp ∈ H2p(F), and [·, ·) is a non-symmetric pairing of H•(F) given by

1 Z [α, β) = eπic1(X)eπiµα ∪ β (2π)dim F F dim F with µ ∈ End H•(F) deﬁned by µ(φ) = p − φ for φ ∈ H2p(F). 2 124 Main Thesis

Figure 1.22: Euler’s Version of the Multiplication Formula for the Γ-function This formula, found in §53 of [E421], is equivalent to the Gaußian multiplication formula above. The restriction to natural numbers m is not necessary. Although Euler states it as a theorem, his proof is not rigorous.

1.9.6 Multiplication Formula One of the fundamental properties of the Γ-function is the so-called multiplication formula that reads in modern notation n−1 x x + 1 x + n − 1 (2π) 2 Γ Γ ··· Γ = · Γ(x). x− 1 n n n n 2 For n = 2 one obtains the duplication formula that is usually ascribed to Legendre, who proved it in [Le26]; we met this formula already in 1.6.5.3. The multiplication formula was ﬁrst proven by Gauß in his inﬂuential 1828 paper [Ga28] on the hypergeometric series, in which he also gave a complete account of the factorial function Π(x) := Γ(x + 1) = x!. Gauß cited Euler’s results very often, but apparently he was not aware of the lesser-known paper [E421] of Euler. At least, he did not cite it.

In that paper Euler presented a formula that is equivalent to the above equation (see ﬁgure), as we will explain now. We need to introduce some notation.

p 1.9.6.1 The Function q In §3 of [E321] and §44 of [E421], Euler studied properties of the function

1 p Z xp−1dx := n−q . q (1 − xn) n 0 In his notation, the variable n is left implicit, and Euler showed the nice symmetry property p q = . q p By the substitution xn = y it is clear that this function is just the Beta function in disguise:

Z1 p 1 p −1 q −1 1 p q = y n dy(1 − y) n = B , , q n n n n 0 Euler and the Γ-Function 125 where the Beta function still is deﬁned as

Z1 B(x, y) = tx−1dt(1 − t)y−1 for Re(x), Re(y) > 0. 0 Euler implicitly assumed p and q to be natural numbers, but this restriction is not necessary. As mentioned several times, see, e.g. section 1.9.2, Euler already knew the relation among Beta integral and the Γ-function: Γ(x) · Γ(y) B(x, y) = . Γ(x + y)

1.9.6.2 Applying the Reflection Formula Euler’s version of the reflection formula for the Γ-function, π = Γ(x)Γ(1 − x), sin πx can be found in §43 of [E421] and reads πλ [λ] · [−λ] = , sin πλ where, in Euler’s notation, [λ] stands for λ!, that is Γ(1 + λ). i If one applies the reflection formula for x = n , i = 1, 2, ··· , n − 1, one obtains 1 n − 1 π Γ Γ = π , n n sin n 2 n − 2 π Γ Γ = , n n 2π sin n 3 n − 3 π Γ Γ = , n n 3π sin n ... = ... n − 1 1 π Γ Γ = . n n (n−1)π sin n Multiplying these equations gives our first auxiliary formula

n−1 i 2 πn−1 Γ = . ∏ n n−1 iπ i=1 ∏i=1 sin n Our second auxiliary formula is n−1 iπ n = ∏ sin n−1 , i=1 n 2 which was certainly known to Euler. For example, in §7 of [E562] and in §240 of [E101], he stated the more general formula π π sin nϕ = 2n−1 sin ϕ sin − ϕ sin + ϕ n n 2π 2π sin − ϕ sin + ϕ · etc. n n 126 Main Thesis

n−1 π(n−i) iπ The product has n factors in total. If we divide by 2 sin ϕ, use sin n = sin n and take the limit ϕ → 0, we obtain the second auxiliary formula. The ﬁrst and the second auxiliary formula were also given by Gauß in [Ga28] and are used in his proof of the multiplication formula69. Combining them and taking the square root, we obtain the beautiful formula r 1 2 n − 1 (2π)n−1 Γ Γ ··· Γ = . n n n n This formula was also found by Euler in §46 of [E816], where he stated it in the form

1 1 1 2 1 n−1 s Z 1 n Z 1 n Z 1 n 1 · 2 · 3 ··· (n − 1) 2n−1πn−1 dx log dx log ··· dx log = . x x x nn−1 n 0 0 0

1.9.6.3 Euler’s Version of the Multiplication Formula In §53 of [E421], Euler gave the formula s h m i m 1 2 3 n − 1 = n nn−m · 1 · 2 · 3 ··· (m − 1) ··· . n n m m m m m m As before, [λ] is Euler’s notation for the factorial of λ such that n = Γ n + 1 . Euler assumed m and n to be natural numbers, but it is easily seen that we can interpolate 1 · 2 · 3 ··· (m − 1) by Γ(m). Therefore, if we assume x to be real and positive and write x instead of m in the above formula and express it in terms of the Beta function, Euler’s formula becomes s x 1 1 x 2 x n − 1 x Γ = n nn−xΓ(x) B , B , ··· B , . n nn−1 n n n n n n Expressing the B-function in terms of the Γ-function, then after some rearrangement under the √ n -sign we obtain

s − x Γ 1 Γ x Γ 2 Γ x Γ n 1 Γ x Γ = n n1−xΓ(x) n n · n n ··· n n . n x+1 x+2 x+n−1 Γ n Γ n Γ n Bringing all Γ-functions of fractional argument to the left-hand side, the expression simpliﬁes to

x x + 1 x + 2 x + n − 1 1 n − 1 Γ Γ Γ ··· Γ = n1−xΓ(x)Γ ··· Γ . n n n n n n

1 n−1 The product on the right-hand side, Γ n ··· Γ n , was evaluated above and thus we obtain s x x + 1 x + 2 x + n − 1 (2π)n−1 Γ Γ Γ ··· Γ = n1−xΓ(x) . n n n n n Thus, we arrived at the multiplication formula.

69Gauß’s proof follows the same lines as ours concerning the use of the auxiliary formulas. Euler and the Γ-Function 127

1.9.6.4 Euler’s Proof of the Multiplication Formula As mentioned, Euler never had a complete proof for the multiplication formula. Basically, this was because he did not prove the fundamental relation to the B-function for all real numbers. Nevertheless, for the sake of completeness, we present Euler’s proof which he offered in the ﬁrst supplement to [E421]. His proof is based on the formula that he wrote as m λ λm λ n n = . λ+m λ + m m n This is just the fundamental relation among the B- and Γ-function. In modern notation, the formula reads: Γ m + 1 Γ λ + 1 λm 1 λ m n n = · B , . λ+m λ + m n n n Γ n + 1 His idea was to substitute all natural numbers from 1 to n for λ and multiply the resulting expressions. Then, the multiplication formula will follow from simple manipulation of the product. On the right-hand side he ﬁnds, in modern notation:

m 2m nm 1 1 m 1 m n m · ··· · B , B , ··· B , 1 + m 2 + m n + m nn n n n n n n which simpliﬁes to

mn m! 1 m 1 m n m · B , B , ··· B , nn (n + 1)(n + 2) ··· (n + m) n n n n n n On the left-hand side on the other hand:

Γ m + 1 Γ 1 + 1 Γ m + 1 Γ 2 + 1 Γ m + 1 Γ m + 1 n n · n n ··· n n 1+m 2+m n+m + Γ n + 1 Γ n + 1 Γ n 1 m n Γ 1 + 1 Γ 2 + 1 ··· Γ n + 1 = Γ + 1 · n n n . n 1+m 2+m m+n Γ n + 1 · Γ n + 1 ··· Γ n + 1 This simpliﬁes to m n Γ 1 + 1 Γ 2 + 1 ··· Γ m + 1 Γ + 1 · n n n n 1+n 2+n m+n Γ n + 1 · Γ n + 1 ··· Γ n + 1 Finally, using the recursive relation in the denominator and cancelling all the Γ-functions:

m n nm = Γ + 1 · . n (n + 1)(n + 2) ··· (n + m) Therefore, equating the ﬁnal result of the left-hand and right-hand side, we arrive at the equation:

mn m! 1 m 1 m n m · B , B , ··· B , nn (n + 1)(n + 2) ··· (n + m) n n n n n n m n nm = Γ + 1 · . n (n + 1)(n + 2) ··· (n + m) Therefore, replacing also m! by Γ(m + 1), we ﬁnd 128 Main Thesis

m n 1 m n m Γ + 1 = mnΓ(m + 1)n−n−mB , ··· B , . n n n n n This is just Euler’s version of the multiplication formula in modern notation and thus ﬁnishes his proof.

1.9.6.5 Discussion of Euler’s Result From the above sketch it is apparent that in [E421] Euler had a result that is equivalent to the p multiplication formula for the Γ-function. He expressed it in terms of the symbol q , which in modern notation is the Beta function. One may wonder why Euler did not express his result in terms of the Γ-function itself. Reading his paper, it becomes clear that his main motivation was to express the factorial of rational numbers in terms of integrals of algebraic functions, and the formula given by Euler fulﬁlls this purpose. For the same reason, he probably did not replace 1 · 2 · 3 ··· (m − 1) by Γ(m). p Euler also expressed Γ( q ), as we saw in 1.9.3 in terms of integrals of algebraic algebraic functions in §23 [E19] and §5 of [E122]. That formula reads

1 s Z p 2p 3p 4p qp (− log x) q dx = q 1 · 2 · 3 ··· p + 1 + 1 + 1 ··· + 1 q q q q 0 v u 1 1 1 1 1 uZ p Z p Z p Z p Z p uq q 2 3 q 3 4 q 4 5 q q−1 q q ×t dx(x − xx) · dx(x − x ) · dx(x − x ) · dx(x − x ) ··· dx(x − x ) . 0 0 0 0 0 Despite the similarity to the ﬁrst formula expressing Γ via B, that formula is not as general as the multiplication formula70. It appears that Euler was aware that the proofs he indicated in [E421] were not completely convincing. He expressed that with characteristic honesty in a concluding SCHOLIUM:

Hence inﬁnitely many relations among the integral formulas of the form

Z xp−1dx p n−q = (1 − xn) n q follow, which are even more remarkable, because we were led to them by a completely singular method. And if anyone does not believe them to be true, he or she should consult my observations on these integral formulas71 and will then hence easily be convinced of their truth for any case. But even if this consideration provides some conﬁrmation, the relations found here are nevertheless of even greater importance, because a certain structure is noticed in them and they are easily generalized to all classes, whatever number was assumed for the exponent n, whereas in the ﬁrst treatment the calculation for the higher classes becomes

70We want to mention here that in the foreword of the Opera Omnia, series 1, volume 19, p. LXI A. Krazer and G. Faber claim that these two formulas are equivalent and both are a special case of the multiplication formula. This is incorrect, as it was shown in the preceding sections. The formula given in section 1.9.3 does not lead to the p multiplication formula, it only interpolates Γ q in terms of algebraic integrals. 71Here Euler refers to his paper [E321]. Euler and the Γ-Function 129 continuously more cumbersome and intricate. 130 Main Thesis

1.10 Summary

1.10.1 Overview of Euler’s Results on the Γ-Function

1.10.1.1 General Overview As we have seen, Euler basically already found all common representations of the Γ-function ranging from the integral representation to the product representation, although he never got credit for the latter. The common denominator of all his ways to these representations was an attempt to solve the functional equation Γ(x + 1) = xΓ(x). He basically had four different approaches: The moment method (section 1.5), which gave the integral represantation, solution by conversion into a differential equation (section 1.6), which led to the Euler-Maclaurin formula (section 1.7) and from it to the Stirling formula (section 1.6.4.1), difference calculus (section 1.8), which led him to Weierstraß product formula (section 1.8.4), and direct iterative application of the functional equation, which led him to the Gauß product formula (section 1.1.2). Furthermore, he discovered several special properties like the reﬂection formula (section 1.9.4), the multiplication formula 1.9.6, and the relation to the B-function (section 1.9.2). Thus, it is safe to say that throughout his career he discovered all basic properties of the Γ-function.

He could have found some more results that were then later discovered by others. As the foremost example, we want to mention the Fourier series expansion of log(Γ(x)). This is usually attributed to Kummer [Ku47], but an equivalent result was obtained one year earlier by Malmsten [Ma46] (p. 25)72. Fourier series were introduced later in Fourier’s monumental treatise on heat [Fo22] in 1822. But certain examples appear at seemingly random places in Euler’s work and in [E189] he actually showed that every periodic function has a Fourier expansion. But at that time, he did not pay any further attention to it. He gave the formula for the Fourier coefficients then later in [E704] but did not make the connection to his earlier findings and did not realize the importance of those findings as Fourier later did.

But there are also results that were out of his reach. Those concern all results which require the theory of complex functions, like the classification theorems discussed in section 1.3.2. Those are necessary in order to see, whether two different expressions for Γ are actually identical. Euler never felt the need to do so and never even addressed that issue. Indeed, it is quite difficult to show directly that all the different expressions are identical. Another property that Euler rarely talked about is that the Γ-function is a transcendental function 1 and even such values as Γ 2 are transcendental. Although Euler had a notion of what a transcendental number is, he never actually gave a definition.

Unfortunately, Euler never pointed out the connections between all his ﬁndings concerning the Γ-function and never organized all his results with the exception of [E421]. But that paper mainly focused on the integral representation and the other representations are not discussed. Probably, the closest to an overview article might be the paper [E368], in which he listed almost all of the formulas73, which we discussed throughout the text.

72Malmsten’s paper was published later than Kummer’s, but written earlier in 1846. Both authors clearly obtained their results independently, since the techniques applied are very different. 73Compare also Figure 2.24 and 2.25 Euler and the Γ-Function 131

Figure 1.23: Formula for the Fourier Coefﬁcients Taken from [E704]. Euler states the formula for the Fourier series of a symmetric function. He proves it formally in the following paragraphs of this paper.

Figure 1.24: Euler’s Formulas for Γ(x) - Part 1 First two of Euler’s formulas for the factorial from his paper [E368].

But it seems that in this paper he was more interested in the evaluation of other values of the 1 Γ-function than Γ 2 which can be expressed in terms of known constants. His attemps were fruitless at the end due to the higher transcendality of the Γ-function pointed out in 1.2.1.7. It would have been really intriguing to see, what Euler had to say about all the connections we pointed out. But despite all this, it was interesting to see what Euler actually already knew about the Γ-function and how many ideas he actually anticipated and that this is not generally known. Thus, we have added some interesting details to the history of the Γ-function, wonderfully told in [Da59]. Therefore, I hope, our article on Euler and the Γ-function provides a motivation to go through other historical texts and see what treasures they might contain. 132 Main Thesis

Figure 1.25: Euler’s Formulas for Γ(x) - Part 2 Remainder of Euler’s formulas for the factorial from his paper [E368]. It contains all the formulas he ever found for the factorial. He writes simply l for log, ∆ is the Euler-Mascheroni constant γ, A, B, C denote the rational part of ζ(2n), i.e. ζ(2) = Aπ2, ζ(4) = Bπ4 etc. Euler and the Γ-Function 133

1.10.1.2 Results not generally associated with Euler In this section, we wish to give an overview of the results that Euler already obtained, but that are not usually attributed to him. Or in other words, some of his lesser-known results:

1. Moment Method: In [E123] and [E594], two papers actually devoted to continued fractions, Euler finds a method which can be used to solve homogeneous difference equations with linear coefficients in terms of definite integrals (section 1.5). In this context, we mention that in [E551] he already found the recursive definition of the Legendre polynomials (see section 1.5.3.2). Using the moment ansatz, one can find an explicit formula for the n-th Legendre polynomial in terms of definite integrals. This formula was found (up to a multiplicative constant) by Euler in [E606]. But it seems that Euler was not aware of the connections between [E551] and [E606]. Moreover, we explained that the moment method can be considered as the first appearance of the Mellin-Transform 1.5.4.

2. Anticipation of convergence inducing factors: In [E613], a paper actually devoted to interpolation problems, Euler anticipated Weierstraß’s idea of introducing factors forcing a product to converge (see section 1.8.4).

3. Gauß’s deﬁnition of the factorial and multiplication formula Gauß’s deﬁnition of the factorial [Ga28] now bearing his name had already been found by Euler in [Eu29] (see also in section 1.2.2). He proved it in [E652] (see section 1.4.5). Additionally, he also found a formula equivalent to the Gaußian multiplication formula (see section 1.9.6).

4. Solution of differential equations of infinite order: In [E189], Euler considered and solved differential equations of infinite order with constant coefficients (see section 1.6). Unfortunately, his formulas are incorrect in general due to a conceptual error (see section 1.6.2.3).

5. Cauchy’s definition for convergence of infinite series: In [E43], Euler gave the necessary and sufficient criterion for the convergence of an infinite series usually attributed to Cauchy (see section 1.11.3).

6. Partial Fraction Decomposition of transcendental functions: In [E592], Euler explained his method to ﬁnd the partial fraction decomposition of simple 1 ( ) transcendental functions as, e.g., sin(x) , tan x (see section in section 1.11.2). Anyhow, although all examples given in that paper are correct, his method does not work in general (see section 1.11.2.2).

1.10.1.3 Results erroneously ascribed to Euler There are also some results that are wrongly ascribed to Euler that we learnt about. We also list them here. 134 Main Thesis

1. Discovery of the Γ-function: Euler was not the first to discover an expression for the Γ-function. The first was Daniel Bernoulli [Be29]. But Euler was indeed the first to discover the integral representation one year later. Any- how, his paper [E19] is the first paper published in a scientific journal containing an expression for Γ-function. See section 1.2.2 for a more detailed discussion on the subject.

2. Eulerian Integral Representation for the hypergeometric series: It seems that Euler never wrote down the integral representation for the hypergeometric series in his papers, although it can easily be derived from his ideas and results in [E254] and [E366]. Legendre seems to be the ﬁrst to write it down explicitly in his 1812 book [Le17]. See section 1.5.3.7. Euler and the Γ-Function 135

1.11 Appendix

1.11.1 A Solution of the Basel Problem that Euler missed (but could have found) The solution of the Basel problem, i.e. the summation of the series

∞ 1 1 1 = + + + ··· ∑ 2 1 k=1 k 4 9 was Euler’s first great claim to fame. He gave the correct result the first time in [E41]. But his proof is based on the sine-product that he could not prove at that time. Therefore, the first complete solution of the problem he gave was in [E61] one year later. In his career, he gave at least 5 different proofs of this formula. Here, we want to show, how Euler could have given another proof, i.e. we give a proof using only Euler’s results. We showed that Euler had an explicit solution of the difference equation

f (x + 1) − f (x) = X(x). But the same equation, for integer x, is also solved by

x f (x) = ∑ X(n − 1). n=1 To solve the Basel problem, let us consider the special case X = x2, i.e. the equation

f (x + 1) − f (x) = x2. For integer x the solution is

x x−1 f (x) = ∑ (n − 1)2 = ∑ n2. n=1 n=1 But applying the Euler-Maclaurin summation formula this sum is easily found to be

1 x3 x2 x f (x) = x(x − 1)(2x − 1) = − + . 6 3 2 6 But this solution is seen to satisfy the propounded difference equation for all x ∈ C! (The general solution is obtained by adding an arbitrary periodic function to that solution as we discussed in section 1.6.3.4.). But now let us also use the explicit solution of the general difference equation that we found in 1.6.3.4. We ﬁnd

x 2 x R x 0 R − f (x) = t2dt − + ∑ e2πix e 2πitt2dt 2 k∈Z x3 x2 x 1 = + C − + ∑∞ + ∑ C e2πix 3 2 π2 k=1 k2 k∈Z\{0} k x3 x2 x 1 = − + ∑∞ + A(x). 3 2 π2 k=1 k2 136 Main Thesis

0 The indicates that the term for k = 0 is left out in the summation. The Ck are integration constants. A is just an arbitrary periodic function with period one, i.e., a solution of the homogeneous difference equation f (x + 1) − f (x) = 0. Since the solution of a difference equation must be unique (if the solution of the homogeneous equation is subtracted), comparing the coefﬁcients of x and the last equations we ﬁnd

1 ∞ 1 1 = 2 ∑ 2 . π k=1 k 6 And this is the Basel sum again. Although the way in which we obtained this result was rather non-straightforward, it nevertheless only used results that Euler knew; hence we claim that he also could have given that solution. Anyhow, this solution seems to be new, since it does not appear in [Ch03] which is a list of several different solutions for the Basel problem.

1.11.2 Euler and the Partial Fraction Decomposition of Transcendental Functions Euler devoted the whole paper [E592] to the expansion of transcendental functions into partial fractions. Although his method is not correct in general, all formulas he stated in his paper are correct. The ﬂaw in his method can easily be ﬁxed. We will explain his method in the example of 1 sin x , where his method works, but then will also give an example in which his method fails and will explain the reason.

π 1.11.2.1 An Example: sin πx In [E592], Euler wrote down the formula

π 1 1 1 1 1 − = − + − + etc. valid for λ ∈ C \ Z. 2λ sin λπ 2λ2 1 − λ2 4 − λ2 9 − λ2 16 − λ2

1 To find the partial fraction decomposition of sin ϕ and other similar transcendental functions, Euler argued as if they were rational functions. For those he outlined the procedure in [E101], [E162], [E163] and gave an improved version in the last chapter of [E212]. Sandifer wrote a nice paper on this method in [Sa07]. The first step is, as usual, to find the zeros of the denominator, i.e., sin ϕ. They are, as it was demonstrated in section 1.9.4.1, ϕi = iπ with i ∈ Z. Furthermore, all of those zeros are simple. Hence Euler made the ansatz

1 A = i + R (ϕ), sin ϕ ϕ − iπ i

Ai being a constant to be determined, Ri(ϕ) being a function that does not contain the factor ϕ − iπ or any positive or negative powers of it. To determine the constant Ai, he wrote

ϕ − iπ A = − R (ϕ)(ϕ − iπ). i sin ϕ i

Since Ri(ϕ) does not contain ϕ − iπ or any power of it, we have Euler and the Γ-Function 137

ϕ − iπ 1 i Ai = lim = lim = (−1) , ϕ→iπ sin ϕ ϕ→iπ cos ϕ where L’Hospital’s rule was used in the second step. Since this method works for all zeros Euler then concluded

1 1 1 1 1 1 1 1 = + − − + + − − + ··· sin ϕ ϕ ϕ − π ϕ + π ϕ − 2π ϕ + 2π ϕ − 3π ϕ + 3π This follows from the integral representation we used above to prove the sine product formula in section 1.9.4.3 and can vice versa be used to show that the integral formula is correct. Although this formula turns out to be right, Euler’s reasoning is not quite correct. We will elaborate on this in the next section. For now, let us simplify the result. Adding each two terms we ﬁnd

1 1 2ϕ 2ϕ 2ϕ 2ϕ − = − + − + ··· sin ϕ ϕ π2 − ϕ2 4π2 − ϕ2 9π2 − ϕ2 16π2 − ϕ2 Diving by 2ϕ and then setting ϕ = λπ we arrive at

π 1 1 1 1 1 − = − + − + ··· λ sin λπ 2λ2 1 − λ2 4 − λ2 9 − λ2 16 − λ2 This is the claimed formula and hence completes the proof.

1 1.11.2.2 An Example in which Euler’s Method fails - ez−1 In the preceding section, we already mentioned that Euler’s method to find the partial fraction decomposition of a transcendental function does not work in general. We want to illustrate this with an example that we actually already needed above in section 1.6.3. Let us try to find the partial fraction decomposition of (ez − 1)−1. The first step is again to find all the zeros of the denominator. They are zk = 2kπi with an integer number k. Furthermore, they are all simple. Using Euler’s ansatz let, us set

1 A = k + R (z), ez − 1 z − 2kπi k where Ak is a constant we want to determined and Rk is a function not containing z − 2kπi or any powers of it. Proceeding as above we ﬁnd Ak = 1. Therefore, we can write

1 1 = + R(z), z − ∑ − e 1 k∈Z z 2kπi and we have to determine R(z). Euler simply would have assumed R(z) to be zero according to the procedure outlined in [E592]; this will turn out to be not true. To see this, consider

1 1 R(z) = − . z − ∑ − e 1 k∈Z z 2kπi 138 Main Thesis

From this we ﬁnd that R(z) is a bounded holomorphic function and hence constant by Liouville’s theorem. To ﬁnd the value of R consider the Laurent series of (ez − 1)−1 around z = 0. This series turns out to be

1 1 ∞ B = n n z ∑ z , e − 1 z n=0 n! where Bn are the Bernoulli numbers. As we mentioned in section 1.7.2, Euler essentially found this generating function for the Bernoulli number already in [E25], but especially pointed it out in his studies concerning the Euler-Maclaurin summation formula, see, e.g., [E47], [E55] and in his book [E212]. Using this series we have

B 1 ∞ B 1 0 1 R = 0 + B + n zn − − . z 1 z ∑ n z ∑ z − k i n=2 ! k∈Z\{0} 2 π

1 But from the deﬁnition of the Bernoulli numbers one easily ﬁnds B0 = 1 and B1 = − 2 . Hence putting z = 0 in the equation for R we obtain

1 R(z) = R(0) = R = B = − . 1 2 Therefore, we obtain the partial fraction decomposition

1 1 1 = − + . z − ∑ − e 1 2 k∈Z z 2kπi 1 As we saw above in section 1.6.3, the extra term of − 2 turns out to be of major importance in the derivation of the solution of the simple difference equation.

1.11.3 Euler on the Euler-Mascheroni Constant We needed the Euler-Mascheroni74 constant γ in the derivation of the Weierstraß product formula for Γ(x) in section 1.4.3.3; hence we want to take the opportunity to consider Euler’s contribution to the constant. The γ constant appears frequently at various places in mathematics. Therefore, it is natural that it also appears at various places in Euler’s works. But we want to focus mainly on three papers [E43], the ﬁrst occurrence of γ,[E583], a paper devoted to γ, and [E629] considering one singular expression for γ in more detail75.

1.11.3.1 Euler’s Discovery of γ Euler discovered the constant, nowadays called γ, in 1740 in [E43]. That paper is also interesting for a another reason. Euler stated the modern convergence criterion for the convergence of an inﬁnite series in § 2; he wrote:

74Mascheroni’s name was added to the constant, since in his elaborations to Euler’s book on integral calculus he found a new formula for γ and used it to calculate several digits of it. Additionally, he found several other functions, in whose Taylor series expansion it appears. Mascheroni’s ideas are reprinted in the Opera Omnia version of Euler’s book [E366]. 75For an interesting historical account, confer also Sandifer’s article in [Sa15]. Euler and the Γ-Function 139

Figure 1.26: Discovery of the Euler-Mascheroni Constant First appearance of the Euler-Mascheroni constant in print. In this paper ([E43]) Euler denotes it by C. He uses i to denote an inﬁnitely large number.

A series, which continued to infinity has a finite sum, even though it is continued twice as far, will never gain any increment, but that what is added after infinity, will actually be infinitely small. For, if it would not be like this, the sum, even though it is continued to infinity, would not be defined and hence not finite. Hence it follows, if that, what results beyond the infinitesimal term, is of finite magnitude, that the sum of the series is necessarily finite.

This is essentially Cauchy’s deﬁnition for the convergence of a series. But let us discuss how Euler discovered γ. He encounters it the ﬁrst time in §6, where he noted, if

i c s := ∑ + ( − ) n=1 a n 1 b and i is a very large number,

that

ds c c = and hence s = C + log(a + ib). di a + bi b The constant C will turn out to be γ in the case a = b = c = 1. By this reasoning, Euler found that ! n 1 lim − log(n + 1) n→∞ ∑ k=1 k is a finite number, although both the harmonic series and the logarithm become infinite for n → ∞. Euler was certainly intrigued by this and tried to find γ. For this, he noted that:

2 3 n + 1 log + log + ··· + log = log(n + 1). 1 2 n Next, he noted the series expansion: 140 Main Thesis

n + 1 1 1 1 1 log = log 1 + = − + − etc. n n n 2n2 3n3

1 which is is just the Taylor series expansion of log(1 + x) around x = 0 applied for x = n . Since n ∈ N, the convergence condition |x| ≤ 1 is met. Therefore, we have

1 1 1 1 = log 2 + − + − etc. 2 3 4 1 3 1 1 1 = log + − + − etc. 2 2 2 · 22 3 · 23 4 · 24 1 4 1 1 1 = log + − + − etc. 3 3 2 · 32 3 · 33 4 · 34 1 5 1 1 1 = log + − + − etc. 4 4 2 · 42 3 · 43 4 · 44 . .

1 n + 1 1 1 1 = log + − + − etc. n n 2 · n2 3 · n3 4 · n4 Therefore, adding the columns, we arrive at:

1 1 1 1 1 1 1 1 + + + ··· + = log(n + 1) + 1 + + + + etc. 2 3 n 2 22 32 42 1 1 1 1 − 1 + + + + etc. 3 23 33 43 1 1 1 1 − 1 + + + + etc. 4 24 34 44

etc.

∞ 1 Therefore, if we call ζ(m) = ∑n=1 nm , we conclude:

∞ (−1)m γ = ∑ ζ(m). m=2 m By the Leibniz criterion, this series is seen to converge and hence Euler proved that γ indeed exists and is ﬁnite.

1.11.3.2 Euler’s Formulas for γ

As mentioned above [E583] is solely devoted to the determination of alternate expressions for γ. Euler and the Γ-Function 141

Figure 1.27: Formulas for the Euler-Mascheroni Constant - Part 1 First part of Euler’s formulas for the Euler-Mascheroni constant from [E583]. He wrote α, β, γ etc. for ζ(2), ζ(3), ζ(4) etc. and C for the Euler-Mascheroni constant γ.

We do not want to prove them here, but want to mention some of the results and ideas. Probably the most interesting formula is (formula I. in the ﬁgure):

∞ 1 1 − γ = ∑ (ζ(m) − 1) . m=2 m Euler listed all other formulas, involving series, at the end of this paper.

He also gave an integral formula76, namely

1 Z 1 1 γ = + dz. 1 − z log z 0 He found it in true Eulerian fashion. First, he notes that

1 − xn lim log = n x→1 1 − x and

Z xn−1dx log (1 − xn) = −n ; 1 − xn in particular, for n = 1

Z dx log (1 − x) = − . 1 − x Moreover,

76[E629] is completely devoted to this one formula. 142 Main Thesis

Figure 1.28: Formulas for the Euler-Mascheroni Constant - Part 2 Second part of Euler’s formulas for the Euler-Mascheroni constant from [E583]. He writes α, β, γ etc. for ζ(2), ζ(3), ζ(4) etc. and C for the Euler-Mascheroni constant γ.

1 Z (1 − xn) 1 1 dx = 1 + + ··· + 1 − x 2 n 0 which follows integrating the geometric sum,

1 − xn = 1 + x + x2 + ··· + xn−1, 1 − x term by term. Therefore, we have:

 1 1 1  Z (1 − xn)dx Z xn−1dx Z dx γ = lim − + n −  , n→∞ 1 − x 1 − xn 1 − x 0 0 0 or

 1 1  Z xndx Z xn−1dx γ = lim − + n  . n→∞ 1 − x 1 − xn 0 0 Next, Euler set xn = z to arrive at

 1 1 1  1 Z z n dz Z dz γ = lim − + . →  1  n ∞ n 1 − z n 1 − z 0 0 Note the known limit

1 log(z) = lim n z n − 1 . n→∞

1 Pull the limit inside the integral and use the above limit and z n → 0 for n → ∞ and we ﬁnally arrive at Euler and the Γ-Function 143

1 Z 1 1 γ = + dz. 1 − z log z 0 Therefore, we arrived at the formula, which is subject of [E629]. But the formulas Euler presented in that paper are not as useful as those in [E583]. So we will not discuss this here. Finally, we want to add Euler’s conjecture that γ is a logarithm of an important number (confer §2 of [E583]). Unfortunately, Euler did not further explain what "important" means in this context. In general, not much is known about the Euler-Mascheroni constant. Not even whether it is irrational or not. See, e.g., [Ha17] for an entertaining account on the history and current state of the art concerning Euler’s constant γ. Bibliography

[Ab27] N. Abel Sur quelques integrales deﬁnies, Journal für Mathmatik Bd. 2, p. 22-30; reprint: Werke, Bd. 1, p. 251-262

[An10] G. Andrews, R. Askey, R. Roy, Special Functions, Cambridge University Press; Edition: New Ed (8th September 2010)

[Ao11] K. Aomoto and M. Kita, Theory of Hypergeometric Functions. Springer Monographs in Mathe- matics, Springer Japan, 2011

[Ao77] K. Aomoto On the structure of integrals of power product of linear functions, Sci. Papers College Gen. Ed. Univ. Tokyo 27 (1977), no. 2 49–61

[Ar15] E. Artin The Gamma Function, Dover Publications; Reprint edition (January 28, 2015)

[Ay13] A. Aycock Note on Malmstèn’s paper De Integralibus quibusdam deﬁnitis seriebusque inﬁnitis, arXiv:1306.4225

[Ay74] R. Ayoub Euler and the Zeta Function, American Mathematical Monthly, Vol. 81, December, 1974, pp. 1067-1086.

[Ba09] W. Balser From Divergent Power Series to Analytic Functions: Theory and Application of Multisummable Power Series, Springer; reprint from 2009 of the ﬁrst edition from 1994

[Be29] D. Bernoulli, Letter on the factorial written to Goldbach 6th October 1729, reprinted in: Correspondance mathématique et physique de quelques célèbres géomètres du XVIIIème siècle. (Band 2), St.-Pétersbourg 1843, p. 324–325 edited by Paul Fuss.

[Bo51] Éléments de mathématique IV. Fonctions d’une variable réelle Hermann, Paris 1951

[Bo97] C. Bourlèt Sur les operationes en general et les equations differentielles lineaires d’ordre inﬁni, Annales de l’Ecole Normale Superieure, 3rd series, vol. 14 (1897), pp. 133-190

[Be13] J.Bernoulli, Ars Conjectandi, published as a book in 1713 in Basel

[Bo99] C. Bourlèt Sur certain equations analogues aux differentielles, Annales de l’Ecole Normale Superieure, 3rd series, vol. 16 (1899), pp. 333-375

[Br07] R. Bradley, E. Sandifer (editors), Leonhard Euler: Life, Work and Legacy, Elsevier Science; 1 edition (January 30, 2007)

[Ca36] R.D. Carmichael Linear Differential Equations of inﬁnite Order, Bulletin American Math. Soc., vol. 34 (1936), pp. 193-218

144 Euler and the Γ-Function 145

[Ch03] R. Chapman Evaluating ζ(2), available online http://empslocal.ex.ac.uk/people/staff/rjchapma/etc/zeta2.pdf

[Ch84] G. Chudnovsky Contributions to the theory of transcendental numbers, American Mathematical Society, 1984

[Ch95] K. Cho and K. Matsumoto Intersection theory for twisted cohomologies and twisted Riemann’s period relations I, Nagoya Math. J. 139 (1995) 67–86

[Da36] H.T. Davis The Theory of linear operators, The Principia Press, Bloomington, Indiana 1936

[Da59] P. Davis Leonhard Euler’s Integral: A Historical Proﬁle of the Gamma Function, American Mathematical Monthly, Vol. 66, December, 1959, pp. 849-869, reprint in the book: The Genius of Euler: Reﬂections on his Life and Work edited by W. Dunham, The Mathematical Association of America; 1st edition (January 1, 2007)

[De70] P. Deligne Equations differentielles a points singuliers reguliers, Lecture Notes in Math 163, Springer-Verlag, 1970

[Di39] P. Dirichlet, Über eine neue Methode zur Bestimmung vielfacher Integrale, Berichter der Berliner Akademie 1389, p. 18-25, Abhandlungen der Berliner Akad. 1839, p. 61-79. See also Dirichlet’s Werke Band U, pp. 381-390, 391-410

[dM18] A. de Moivre The Doctrine of Chances, ﬁrst edition published in 1718 by W. Pearson, third edition (1756) has been reprinted by Chelsea, New York (1967)

[Du75] Y.A. Dubinskii Sobolev spaces of inﬁnite order and the bahavior of solutions of some boundary value problems with unbounded increase of the order of the equation, Mat. Sb. 98(140):2(1975); English translation in Math. USSR Sb. 27(1975), 143-162

[Du10] J. Duistermaat, J. Kolk Distributions Theory and Applications, Birkhäuser; Auﬂage: 2010 (18. August 2010)

[Du91] W. Dunham Euler and the Fundamental Theorem of Algebra, The College Mathematics Journal, Vol. 22, No. 4,(1991), pp. 282-293

[Du99] W. Dunham Euler: The Master of Us All, Dolciani Mathematical Expositions (Book 22), American Mathematical Society (December 31, 1999)

[Du07] W. Dunham (ed.) The Genius of Euler: Reﬂections on his Life and Work, The Genius of Euler: Reﬂections on his Life and Work (2007)

[Dut91] J. Dutka, The early history of the factorial function, Archive for History of Exact Sciences, September 1991, Volume 43, Issue 3, pp 225–249

[E19] L. Euler De progressionibus transcendentibus seu quarum termini generales algebraice dari nequeunt, Commentarii academiae scientiarum Petropolitanae 5, 1738, pp. 36-57, reprint: Opera Omnia: Series 1, Volume 14, pp. 1 - 24

[E25] L. Euler Methodus generalis summandi progressiones, Commentarii academiae scientiarum Petropolitanae 6, 1738, pp. 68-97, reprint: Opera Omnia: Series 1, Volume 14, pp. 42 - 72

[E41] L. Euler De summis serierum reciprocarum, Commentarii academiae scientiarum Petropoli- tanae 7, 1740, pp. 123-134, reprint: Opera Omnia Series 1, Volume 14, pp. 73 - 86 146 Main Thesis

[E43] L. Euler De progressionibus harmonicis observationes, Commentarii academiae scientiarum Petropolitanae 7, 1740, pp. 150-161, reprint: Opera Omnia: Series 1, Volume 14, pp. 87 - 100

[E47] L. Euler Inventio summae cuiusque seriei ex dato termino generali, Commentarii academiae scientiarum Petropolitanae 8, 1741, pp. 9-22, reprint: Opera Omnia: Series 1, Volume 14, pp. 108 - 123

[E55] L. Euler Methodus universalis series summandi ulterius promota, Commentarii academiae scientiarum Petropolitanae 8, 1741, pp. 147-158, reprint: Opera Omnia: Series 1, Volume 14, pp. 124 - 137

[E59] L. Euler Theoremata circa reductionem formularum integralium ad quadraturam circuli, Miscel- lanea Berolinensia 7, 1743, pp. 91-129, reprint: Opera Omnia: Series 1, Volume 17, pp. 1 - 34

[E60] L. Euler De inventione integralium, si post integrationem variabili quantitati determinatus valor tribuatur, Miscellanea Berolinensia 7, 1743, pp. 129-171, reprint: Series 1, Volume 17, pp. 35 - 69

[E61] L. Euler De summis serierum reciprocarum ex potestatibus numerorum naturalium ortarum dissertatio altera, in qua eaedem summationes ex fonte maxime diverso derivantur, Miscellanea Berolinensia 7, 1743, pp. 172-192, reprint: Opera Omnia Series 1, Volume 14, pp. 138-155

[E62] L. Euler De integratione aequationum differentialium altiorum graduum, Miscellanea Berolinensia 7, 1743, pp. 193-242, reprint: Opera Omnia: Series 1, Volume 22, pp. 108 - 149

1 1 1 [E63] L. Euler Demonstration de la somme de cette suite 1 + 4 + 9 + 16 + etc., Journ. lit. d’Allemange, de Suisse et du Nord, 2:1, 1743, p. 115-127, reprint: Opera Omnia Series 1, Volume 14, pp. 177 - 186

[E65] L. Euler Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes, sive solutio problematis isoperimetrici lattissimo sensu accepti, Originally published as a book in 1744, reprint: Opera Omnia: Series 1, Volume 24

[E72] L. Euler Variae observationes circa series inﬁnitas, Commentarii academiae scientiarum Petropolitanae 9, 1744, pp. 160-188, reprint: Opera Omnia: Series 1, Volume 14, pp. 217 - 244

[E101] L. Euler Introductio in analysin inﬁnitorum, Tomus primus, Originally published as a book in 1748, printed by M. M. Bousquet in Lausanne, reprint: Opera Omnia, Series 1, Volume 8

[E122] L. Euler De productis ex inﬁnitis factoribus ortis, Commentarii academiae scientiarum Petropolitanae 11, 1750, pp. 3-31, reprint: Opera Omnia: Series 1, Volume 14, pp. 260 - 290

[E123] L. Euler De fractionibus continuis observationes, Commentarii academiae scientiarum Petropolitanae 11, 1750, pp. 32-81, reprint: Opera Omnia: Series 1, Volume 14, pp. 291 - 349

[E130] L. Euler, De seriebus quibusdam considerationes, Commentarii academiae scientiarum Petropolitanae 12, 1750, pp. 53-96, reprint: Series 1, Volume 14, pp. 407 - 462

[E162] L. Euler Methodus integrandi formulas differentiales rationales unicam variabilem involventes, Commentarii academiae scientiarum Petropolitanae 14, 1751, pp. 3-91, reprint: Opera Omnia: Series 1, Volume 17, pp. 70 - 148 Euler and the Γ-Function 147

[E163] L. Euler Methodus facilior atque expeditior integrandi formulas differentiales rationales, Memoires de l’academie des sciences de Berlin 5, 1751, pp. 222-288, reprint: Opera Omnia: Series 1, Volume 17, pp. 149 - 194

[E168] L. Euler De la controverse entre Mrs. Leibniz et Bernoulli sur les logarithmes des nombres negatifs et imaginaires, Memoires de l’academie des sciences de Berlin 5, 1751, pp. 139-179, reprint: Opera Omnia: Series 1, Volume 17, pp. 195 - 232

[E170] L. Euler Recherches sur les racines imaginaires des equations, Commentarii academiae scientiarum Petropolitanae 14, 1751, pp. 99-150, reprint: Opera Omnia: Series 1, Volume 6, pp. 78 - 150

[E188] L. Euler Methodus aequationes differentiales altiorum graduum integrandi ulterius promota, Novi Commentarii academiae scientiarum Petropolitanae 3, 1753, pp. 3-35, reprint: Opera Omnia: Series 1, Volume 22, pp. 181 - 213

[E189] L. Euler De serierum determinatione seu nova methodus inveniendi terminos generales serierum, Novi Commentarii academiae scientiarum Petropolitanae 3, 1753, pp. 36-85, reprint: Opera Omnia Series 1, Volume 14, pp. 463 - 515

[E212] L. Euler Institutiones calculi differentialis cum eius usu in analysi ﬁnitorum ac doctrina serierum, Originally published as a book in 1755, reprint: Opera Omnia Series 1, Volume 10

[E254] L. Euler De expressione integralium per factores, Novi Commentarii academiae scientiarum Petropolitanae 6, 1761, pp. 115-154, reprint: Opera Omnia: Series 1, Volume 17, pp. 233 - 267

[E274] L. Euler Constructio aequationis differentio-differentialis Aydu2 + (B + Cu)dudy + (D + Eu + Fuu)ddy = 0, sumto elemento du constante, Novi Commentarii academiae scientiarum Petropoli- tanae 8, 1763, pp. 150-156, reprint: Opera Omnia: Series 1, Volume 22, pp. 395 - 402

R p−1 n q −1 [E321] L. Euler Observationes circa integralia formularum x dx(1 − x ) n posito post integrationem x = 1, Melanges de philosophie et de la mathematique de la societe royale de Turin 3, 1766, pp. 156-177, reprint: Opera Omnia: Series 1, Volume 17, pp. 269 - 288

[E352] L. Euler Remarques sur un beau rapport entre les series des puissances tant directes que reciproques, Memoires de l’academie des sciences de Berlin 17, 1768, pp. 83-106, reprint: Opera Omnia: Series 1, Volume 15, pp. 70 - 90

[E366] L. Euler Institutionum calculi integralis volumen secundum, Originally published as a book in 1769, reprint: Opera Omnia: Series 1, Volume 12

[E368] L. Euler De curva hypergeometrica hac aequatione expressa y = 1 · 2 · 3 ··· x, Novi Commentarii academiae scientiarum Petropolitanae 13, 1769, pp. 3-66, reprint: Opera Omnia: Series 1, Volume 28, pp. 41 - 98

[E391] L. Euler De formulis integralibus duplicatis, Novi Commentarii academiae scientiarum Petropolitanae 14, 1770, pp. 72-103, reprint: Opera Omnia: Series 1, Volume 17, pp. 289 - 315

[E393] L. Euler De summis serierum numeros Bernoullianos involventium, Novi Commentarii academiae scientiarum Petropolitanae 14, 1770, pp. 129-167, reprint: Opera Omnia: Se- ries 1, Volume 15, pp. 91 - 130 148 Main Thesis

R − m [E421] L. Euler Evolutio formulae integralis x f 1dx (log(x)) n integratione a valore x = 0 ad x = 1 extensa , Novi Commentarii academiae scientiarum Petropolitanae 16, 1772, pp. 91-139 , reprint: Opera Omnia: Series 1, Volume 17, pp. 316 - 357

[E432] L. Euler Exercitationes analyticae, Novi Commentarii academiae scientiarum Petropolitanae 17, 1773, pp. 173-204, reprint: Opera Omnia: Series 1, Volume 15, pp. 131 - 167

R zm−1±zm−n−1 [E462] L. Euler De valore formulae integralis 1±zn dz casu quo post integrationem ponitur z = 1 , Novi Commentarii academiae scientiarum Petropolitanae 19, 1775, pp. 3-29, reprint: Opera Omnia: Series 1, Volume 17, pp. 358 - 383

[E465] L. Euler Demonstratio theorematis Neutoniani de evolutione potestatum binomii pro casibus, quibus exponentes non sunt numeri integri, Novi Commentarii academiae scientiarum Petropolitanae 19, 1775, pp. 103-111, reprint: Opera Omnia: Series 1, Volume 15, pp. 207 - 216

[E551] L. Euler Varia artiﬁcia in serierum indolem inquirendi, Opuscula Analytica 1, 1783, pp. 48-63, reprint: Opera Omnia: Series 1, Volume 15, pp. 383 - 399

[E562] L. Euler, Quomodo sinus et cosinus angulorum multiplorum per producta exprimi queant, Opus- cula Analytica 1, 1783, pp. 353-363, reprint: Opera Omnia: Series 1, Volume 15, p. 509-521.

[E583] L. Euler De numero memorabili in summatione progressionis harmonicae naturalis occurrente, Acta Academiae Scientarum Imperialis Petropolitinae 5, 1785, pp. 45-75, reprint: Series 1, Volume 15, pp. 569 - 603

[E592] L. Euler De resolutione fractionum transcendentium in inﬁnitas fractiones simplices, Opuscula Analytica 2, 1785, pp. 102-137, reprint: Opera Omnia: Series 1, Volume 16, pp. 112 - 121

[E594] L. Euler Methodus inveniendi formulas integrales, quae certis casibus datam inter se teneant rationem, ubi sumul methodus traditur fractiones continuas summandi, Opuscula Analytica 2, 1785, pp. 178-216, reprint: Opera Omnia: Series 1, Volume 18, pp. 209 - 243

n [E606] L. Euler Speculationes super formula integrali R √ x dx , ubi simul egregiae observationes aa−2bx+cxx circa fractiones continuas occurrunt, Acta Academiae Scientarum Imperialis Petropolitinae 1782, 1786, pp. 62-84, reprint: Opera Omnia: Series 1, Volume 18, pp. 244 - 264

[E613] L. Euler Dilucidationes in capita postrema calculi mei differentalis de functionibus inexplicabilibus, Memoires de l’academie des sciences de St.-Petersbourg 4, 1813, pp. 88-119, reprint: Opera Omnia: Series 1, Volume 16,1, pp. 1 - 33

R 1 1 [E629] L. Euler Evolutio formulae integralis dx 1−x + log x a termino x = 0 ad x = 1 extensae

[E637] L. Euler Nova demonstratio, quod evolutio potestatum binomii Newtoniana etiam pro exponentibus fractis valeat, Nova Acta Academiae Scientarum Imperialis Petropolitinae 5, 1789, pp. 52-58, reprint: Opera Omnia: Series 1, Volume 15, pp. 621 - 660

p−1 [E640] L. Euler Comparatio valorum formulae integralis R √ x dx a termino x = 0 usque ad x = 1 n (1−xn)n−q extensae, Nova Acta Academiae Scientarum Imperialis Petropolitinae 5, 1789, pp. 86-117, reprint: Opera Omnia: Series 1, Volume 18, pp. 392 - 423 Euler and the Γ-Function 149

[E652] L. Euler De termino generali serierum hypergeometricarum, Nova Acta Academiae Scientarum Imperialis Petropolitinae 7, 1793, pp. 42-82, reprint: Opera Omnia: Series 1, Volume 16, pp. 139 - 162

[E661] L. Euler Variae considerationes circa series hypergeometricas, Nova Acta Academiae Scientarum Imperialis Petropolitinae 8, 1794, pp. 3-14, reprint: Opera Omnia: Series 1, Volume 16, pp. 178 - 192

R 1 n [E662] L. Euler De vero valore formulae integralis ∂x log x a termino x = 0 usque ad terminum x = 1 extensae, Nova Acta Academiae Scientarum Imperialis Petropolitinae 8, 1894, pp. 15-31, reprint: Opera Omnia: Series 1, Volume 19, pp. 63 - 83

[E664] L. Euler Exercitatio analytica, Nova Acta Academiae Scientarum Imperialis Petropolitinae 8, 1794, pp. 69-72, reprint: Opera Omnia: Series 1, Volume 16.1, pp. 235 - 240

[E675] L. Euler De valoribus integralium a termino variabilis x = 0 usque ad x = ∞ extensorum, Institutiones calculi integralis 4, 1794, pp. 337-345, reprint: Opera Omnia: Series 1, Volume 19, pp. 217 - 227

[E679] L. Euler De formulis integralibus implicatis earumque evolutione et transformatione, Institutiones calculi integralis 4, 1794, pp. 544-563, reprint: Opera Omnia: Series 1, Volume 23, pp. 250 - 267

[E704] L. Euler Disquisitio ulterior super seriebus secundum multipla cuiusdam anguli progredientibus, Nova Acta Academiae Scientarum Imperialis Petropolitinae 11, 1798, pp. 114-132, reprint: Opera Omnia: Series 1, Volume 16,2, pp. 333 - 353

[E710] L. Euler Specimen transformationis singularis serierum, Nova Acta Academiae Scientarum Imperialis Petropolitinae 12, 1801, pp. 58-70, reprint: Opera Omnia: Series 1, Volume 16, pp. 41 - 55

[E746] L. Euler Methodus succincta summas serierum inﬁnitarum per formulas differentiales investigandi, Memoires de l’academie des sciences de St.-Petersbourg 5, 1815, pp. 45-56, reprint: Opera Omnia: Series 1, Volume 16, pp. 200 - 213

[E807] L. Euler Sur les logarithmes des nombres negativs et imaginaires, Opera Postuma 1, 1862, pp. 269-281, reprint: Opera Omnia: Series 1, Volume 19, pp. 417 - 438

[E816] L. Euler Considerations sur quelques formules integrales dont les valeurs peuvent etre exprimees, en certains cas, par la quadrature du cercle, Opera Postuma 1, 1862, pp. 408-438, reprint: Opera Omnia: Series 1, Volume 19, pp. 439 - 490

[En10] G. Eneström, Verzeichnis der Schriften Leonhard Eulers, Teubner-Verlag 1910

[Eu29] L. Euler, letter to Goldbach on the factorial on 13th October 1729, reprinted in: Correspon- dance mathématique et physique de quelques célèbres géomètres du XVIIIème siècle. (Band 1), St.-Pétersbourg 1843, p. 3–7 edited by Paul Fuss

[Fo22] J. Fourier Théorie analytique de la chaleur, Chez Firmin Didot, père et ﬁls 1822

[Fr19] H. Frellesvig et al. Decomposition of Feynman Integrals on the Maximal Cut by Intersection Numbers, arXiv:1901.11510 [hep-ph] 150 Main Thesis

[Fr06] E. Freitag, R. Busam Funktionentheorie 1 4., korrigierte und erweiterte Ausgabe, Springer, Heidelberg (2006)

[Ga16] S. Galkin, V. Golyshev, H. Iritani Gamma classes and quantum cohomology of Fano manifolds: Gamma conjectures, Duke Math. J. 165 (2016), no. 11, 2005–2077, arXiv:1404.6407

[Ga18] S. Galkin, H. Iritani, Gamma Conjecture via Mirror Symmetry, arXiv:1508.00719

[Ga99] C. Gauss Demonstratio nova theorematis omnem functionem algebraicam rationalem integram unius variabilis in factores reales primi vel secundi gradus resolvi posse, Dissertation written in 1799, reprint: Carl Friedrich Gauss Werke, Band III, herausgegeben von der Königlichen Gesellschaft der Wissenschaften zu Göttingen (1866)

+ a·b + a(a+1)·b(b+1) 2 + ··· [Ga28] C. Gauss Disquisitiones generales circa seriem inﬁnitam 1 1·c x 1·2·c(c+1) x , Com- mentationes recentiores Bd. II. , Göttingen 1813

[Gr03] D. Gronau Why is the gamma function so as it is?, Teaching Mathematics and Computer Science, 1/1 (2003), p. 43-53

[Ha94] J. Hadamard, Sur L’Expression Du Produit 1 · 2 · 3 ··· (n − 1) Par Une Fonction Entière, reprinted in: Œuvres de Jacques Hadamard, Centre National de la Recherche Scientiﬁques, Paris, 1968

[Ha64] H. Hankel, Die Euler’schen Integrale bei unbeschränkter Variabilit¨at des Argumentes, Zeitschr. Math. Phys., 9, p.1-21, 1864

[Ha48] G. Hardy, Divergent Series, AMS Chelsea Publishing (1948)

[Ha17] J. Havil Gamma: Exploring Euler’s Constant, Princeton University Press; New edition (October 31, 2017)

[Ho87] O. Hölder Ueber die Eigenschaft der Gammafunction keiner algebraischen Differentialgleichung zu genügen, written 26th of June 1886, published in Mathematische Annalen 28, 1887, S. 1–13

[Ja34a] C. Jacobi De Usu legitimo Formulae summatoriae Maclaurianae, ﬁrst published in Crelle Journal für die reine und angewandte Mathematik, Band 12, pp. 263-272, 1834; reprinted in C.G.J. Jacobi’s Gesammelte Werke, Volume 6, pp. 64-75

1 [Ja34] C. Jacobi, Demonstratio formulae R ωa−1(1 − ω)b−1dω = Γ(a) · Γ(b) : Γ(a + b), Journal für 0 Mathematik, Bd. 11, p.307; 1834

[Ja41] C. Jacobi, De Determinantibus Functionalibus, Crelle Journal für die reine und angewandte Mathematik, Bd. 22. p. 319-352

[Ju65] A.P. Juskevic, E. Winter, Leonhard Euler und Christian Goldbach Briefwechsel 1729-1764, Akademie-Verlag Berlin 1965

[Kn41] Funktionentheorie II. (5. Auﬂage), de Gruyter, Berlin 1941, S. 47–49.

[Koe00] Analysis I (5. Auﬂage), Springer-Verlag Berlin Heidelberg New York

[Ko01] M. Kontsevich, D. Zagier Periods printed in Engquist, Björn; Schmid, Wilfried (eds.), Mathematics unlimited—2001 and beyond, Berlin, New York: Springer-Verlag, pp. 771–808 Euler and the Γ-Function 151

[Ku36] E. Kummer Über die hypergeometrische Reihe, Journal für Mathematik Bd. 15, p. 39 -83; 1836

[Ku37a] E. Kummer De integralibus deﬁnitis et seriebus inﬁnitis Journal für Mathematik Bd. 17, p. 210-227; 1837

[Ku37b] E. Kummer De integralibus quibusdam inﬁnitis et seriebus inﬁnitis Journal für Mathematik, Bd. 17, p. 228-242; 1837

[Ku47] E. Kummer Beitrag zur Theorie der Funktion Γ(x) Journal für Mathematik Bd. 35, pp. 1-4; 1847

[La72] J.L. Lagrange Sur une nouvelle espece du calcul relatif a la differentiation et l’integration des quantites , Berlin nouv. mem., 3, 1772(74), p.185 also in Oeuvres, vol. 3, pp. 441-476

[La20] P. S. Laplace Theorie analytique des probalities,(3rd edition), Paris(1820)

[La82] P. S. Laplace Mémoire sur les approximations des formules qui sont des fonctions de tr‘es grands nombres, Oeuvres, Vol. 10, p.209-291. Mémoires de l’Académie des Sciences de Paris, 1782

[Le86] L. Le Cam, The central limit theorem around 1935, Statistical Science, 1 (1): 78–96 (1986)

[Le26] A. Legendre traité des fonctions elliptiques et des intégrales Eulériennes (vol 2), Huzard-Courcier, Paris 1826

[Le17] A. Legendre, Exercises de Calcul integral, Paris 1811-1819 (three volumes)

[Le85] A. Legendre, Recherches sur l’attraction des sphéroïdes homogènes, Mémoires de Mathématiques et de Physique, présentés à l’Académie Royale des Sciences, par divers savans, et lus dans ses Assemblées, X. Paris. pp. 411–435 (1785)

[Le09] A. Legendre Recherches sur diverses sortes d’intégrales déﬁnies, Mémoires de la classe des sciences mathématiques et physiques de l’Institut de France 10, 1809

[Le26a] A. Legendre Traité des fonctions elliptiques et des intégrales Eulériennes (Volume 2), Huzard- Courcier, Paris 1826

[Ma46] C. Malmsten De integralibus quibusdam deﬁnitis, seriebusque inﬁnitis, Journal für die reine und angewandte Mathematik 38, 1849

[Ma98] K. Matsumoto, Intersection numbers for logarithmic k-forms, Osaka J. Math. 35 (1998), no. 4 873–893

[Me95] H. Mellin, Über die fundamentale Wichtigkeit des Satzes von Cauchy für die Theorien der Gamma- und hypergeometrischen Functionen, Acta Soc. Scient. Fennicæ, Band (1895)

[Mo22] J. Mollerup, H. Bohr Lærebog i Kompleks Analyse vol. III, Analysis Textbook printed in Copenhagen (1922)

[Ni05] N. Nielsen, Die Gammafunction, AMS Chelsea Publishing, American Mathematical Society (June 26, 2005)

[Pe24] K. Pearson Historical note on the origin of the normal curve of errors, Biometrika, 16 (3/4): 402–404 (1924)

[Pe20] R. Perez-Marco On the deﬁnition of Euler Gamma function, arXiv:2001.04445 152 Main Thesis

[Re96] R. Remmert Wielandt’s theorem about the Γ-function, The American Mathematical Monthly 103, 1996, S. 214–220

[Ri57] B. Riemann Beiträge zur Theorie der durch die Gauss’sche Reihe F(α, β, γ, x) darstellbaren Functionen. Abhandlungen der Mathematischen Classe der Königlichen Gesellschaft der Wis- senschaften zu Göttingen (in German). Göttingen: Verlag der Dieterichschen Buchhandlung. 7: 3–22 (1857)

[Ri51] B. Riemann "Grundlagen für eine allgemeine Theorie der Funktionen einer veränderlichen kom- plexen Grösse, Disseratation written in 1851, reprinted in: Riemann’s gesammelte math. Werke, Dover (published 1953), pp. 3–48

[Ri59] B. Riemann Ueber die Anzahl der Primzahlen unter einer gegebenen Grösse, Monatsberichte der Königlichen Preußischen Akademie der Wissenschaften zu Berlin. Berlin 1859, S. 671–680

[Sa07] E. Sandifer How Euler did it, Spectrum, The Mathematical Association of America (July 3, 2007)

[Sa072] E. Sandifer The Early Mathematics of Leonhard Euler, Mathematical Association of America, Washington, DC, 2007

[Sa15] E. Sandifer How Euler did even more, Spectrum, The Mathematical Association of America (2015)

[Sc44] O. Schlömilch Einiges über die Eulerischen Integrale der zweiten Art, Archiv der Mathematik und Physik 4, 1844

[Sc79] O. Schlömilch Kompendium der höheren Analysis Bd. II, Braunschweig, Viewg & Sohn (1879)

[Sl09] L. Slater Generalized Hypergeometric Functions, Cambridge University Press; reprint (March 2009)

[St02] E. Stein, R. Shakarchi, Fourier Analysis, Fourier Analyis - An Introduction, Princeton University Press (2002)

[St30] J. Stirling Methodus differentialis sive tractatus de summatione et interpolatione serierum inﬁni- tarum, London 1730

[Ti48] E. Titchmarsh Introduction to Theory of Fourier Integrals, Oxford University Press; 2nd Revised edition (December 1948)

[Va06] S. Varadarajan Euler Through Time: A New Look at Old Themes, American Mathematical Society; 1st edition (April 18, 2006)

[We14] J. Wedderburn Note on the simple Difference Equation, Annals of Mathematics, vol. 16 (2nd ser.) (1914-1915), pp. 82-85

[We56] K. Weierstraß Über die Theorie der analytischen Fakultäten, Journal für Mathematik Bd. 51, p. 1-60 (1856)

[We78] K. Weierstraß, Einleitung in die Theorie der analytischen Funktionen. Vorlesung, Berlin 1878. Vieweg, Braunschweig Wiesbaden 1988

[Wi96] H. Wielandt Mathematische Werke Vol. 2, de Gruyter, Berlin New York 1996 Euler and the Γ-Function 153

[Yo97] M. Yoshida Hypergeometric Functions, My Love: Modular Interpretations of Conﬁguration Spaces, Vieweg, edition from 1997 Appendices

154 Appendix I

On the Translations of Euler’s Papers

I.1 Possible Questions about the Translations

We answer some possible question the reader might ask about the translations provided in the following.

I.1.1 Where can I ﬁnd the original Papers? Scanned Versions of the original papers can be found at the Euler Archive online. Moreover, they are all reprinted in the Opera Omnia, Euler’s collected works published by Birkhäuser123.

I.1.2 Why these three Papers? As mentioned in the main text, the three papers E19, E368, E421 contain all formulas that Euler contributed to the Γ-Function. Moreover, E19 was the ﬁrst published paper containing the deﬁnition of the Γ-function. In the papers E368 and E421 Euler summarised everything he had discovered on the theory of the Γ-function throughout the years.

I.1.3 What was changed compared to Euler’s Originals? In the translations we stayed as close as possible to the original, concerning both the translation and the notation. Euler tends to write rather long sentences, which is rather unusual in the English language. Thus, at some instance we split such sentences into two shorter ones. Notationwise, we did not change much. Most of the modern notation we use today traces back to Euler so that we had to change only minor things. Euler simply wrote lx for the logarithm of x, in the translation we always write log x, i.e we made the change l 7→ log Figure 2.14 provides an example for Euler’s notation of the logarithm. Furthermore, Euler wrote sin.x, cos.x, tan.x instead of sin x, cos x, tan x, since he abbreviated "sinus", "cosinus", "tangent" by it. Thus, we made the changes tan. 7→ tan,

1Detailed information on [E19]: Original title: "De progressionibus transcendentibus seu quarum termini generales algebraice dari nequeunt", first published in Commentarii academiae scientiarum Petropolitanae 5 (1730/31), 1738, p. 36-57, reprint in: Opera Omnia: Series 1, Volume 14, pp. 1 - 24, Eneström-Number E19 2Detailed information on [E368]: "De curva hypergeometrica hac aequatione expressa y = 1 · 2 · 3 ··· x", first published in Novi Commentarii academiae scientiarum Petropolitanae 13, 1769, pp. 3-66, reprint in: Opera Omnia: Series 1, Volume 28, pp. 41 - 98, Eneström-Number E368 3 R f −1 m Detailed information on [E421]: Original title: "Evolutio formulae integralis x dx(log x) n integratione a valore x = 0 ad x = 1 extensa", first published Novi Commentarii academiae scientiarum Petropolitanae 16, 1772, pp. 91-139, reprint in Opera Omnia: Series 1, Volume 17, pp. 316 - 357, Eneström-Number E421

155 156 Appendix cos. 7→ cos, sin. 7→ sin. Euler’s notation is seen, e.g., in ﬁgure 2.20. Additionally, Euler did not draw the square root sign over the complete expression it is supposed to contain but just in front √ √ of it. For example, Euler would write (1 − xx) instead of the modern notation 1 − xx. See ﬁgure 2.22 for an example of Euler’s use of the square root sign. Since Euler’s notation tends to be confusing at times, this was replaced by the modern notation.

I.1.4 Why were certain Things not changed? We did not add limits of integration when Euler talked about integrals4, since Euler always described them in the text or it is is clear form the context what he meant. Furthermore, we did not write long formulas in compact form, e.g., using the summation sign to denote an inﬁnite sum, since written out matters are often more clear than in the compact notation.

I.1.5 What about the Footnotes? In the translations one will ﬁnd several footnotes. They were all included by the translator to provide additional information not given by Euler. Indeed, Euler did not make any footnotes in any of his works. At Euler’s time, mathematicians did not always provide the reader with the precise source of theorem they applied in the respective paper. Thus, some of the footnotes state the missing information, i.e. the precise source Euler actually means. In this context, we want to mention that footnotes about references Euler made to other works were also included by the respective editor in the Opera Omnia version of the Latin original.

4He did so, e.g., in [E19] and [E421]. Appendix II

Translation of E19 - On transcendental Progressions or those whose general terms cannot be given algebraically

§1 After on the occasion of the observations on series Goldbach had discussed in the Society1,I recently tried to ﬁnd a general expression, which would give all terms of this progression

1 + 1 · 2 + 1 · 2 · 3 + 1 · 2 · 3 · 4 + etc.; considering, if it is continued to inﬁnity, that it is eventually confounded with the geometric series, I discovered the following expression

1 · 2n 21−n · 3n 31−n · 4n 41−n · 5n · · · · etc., 1 + n 2 + n 3 + n 4 + n which expresses the term of order n in the before-mentioned progression. It certainly does not terminate in either case, neither if n is an integer number nor if it is a fraction; but it only yields approximations to ﬁnd the respective term, if the cases n = 0 and n = 1 are excluded, in which cases the formula is immediately seen to be 1. Set n = 2; one will then have

2 · 2 3 · 3 4 · 4 5 · 5 · · · · etc. = the second term 2 1 · 3 2 · 4 3 · 5 4 · 6 If it is n = 3, one will have

2 · 2 · 2 3 · 3 · 3 4 · 4 · 4 5 · 5 · 5 · · · · etc = the third term 6 1 · 1 · 4 2 · 2 · 5 3 · 3 · 6 4 · 4 · 7

§2 But although this expression seems to have no use for finding the terms, it can nevertheless be applied to interpolate the series or the terms, whose indices are fractional numbers. But I decided to explain nothing about this here, since more appropriate ways to achieve the same will be exhibited below. I only want to mention everything that is necessary to get to the things to 1 follow about the general term. I tried to find the term corresponding to the index n = 2 or the term which falls in the middle between the first, 1, and the preceding one, which is also 1. But 1 having put n = 2 I obtain the product 1Euler refers to Goldbach’s work on series the latter presented at the St. Petersburg Academy of Sciences. Euler spent a whole paper on Goldbach’s findings, namely [E72].

157 158 Appendix

r 2 · 4 4 · 6 6 · 8 8 · 10 · · · · etc., 3 · 3 5 · 5 7 · 7 9 · 9 which expresses the term in question. But this series, to me, seemed to be similar to the one I remembered to have seen in Wallis’s works on the area of the circle2. For, Wallis found that the area of the circle has a ratio to the diameter of the circle as of

2 · 4 · 4 · 6 · 6 · 8 · 8 · 10 · etc. to 3 · 3 · 5 · 5 · 7 · 7 · 9 · 9 · etc. Therefore, if the diameter was = 1, the area of the circle will be

2 · 4 4 · 6 6 · 8 = · · · etc. 3 · 3 5 · 5 7 · 7 Therefore, from the agreement of this series with mine it is possible to conclude that the term 1 corresponding to the index 2 is equal to the square root of the area of the circle, whose diameter is = 1.

§3 I had believed before that the general term of the series 1, 2, 6, 24 etc., if not algebraically, is at least given exponentially3. But after I had understood that certain intermediate terms depend on the quadrature of the circle, I realized that neither algebraic nor exponential quantities are sufﬁcient to express it. For, the general term of the progression must be of such a nature that one time it contains algebraic quantities but another time quantities depending on the quadrature of the circle and even on other quadratures; and no algebraic nor exponential formula fulﬁlls this condition.

§4 But after I had remembered that among differential formulas quantities exist, which in certain cases admit an integration and then yield algebraic quantities, but in other cases do not admit an integration and then exhibit quantities depending on the quadratures of curves, it came to mind that maybe formulas of this kind are appropriate to express the general terms of the mentioned progression and other similar ones. And in the following, I will call progressions requiring such general terms, which cannot be given algebraically, transcendental; as Geometers used to call everything exceeding the power of common Algebra, transcendental.

§5 Therefore, I thought about what properties differential formulas should have in order to express general terms of progressions in the best possible way. But the general term is a formula, into which both constant as well certain other non constant quantities such as n enter, which number n gives the order of the terms or the index such that, if the third term is in question, instead of n one has to put 3. But also a certain variable quantity must be contained in the differential formula. For this quantity it is not advisable to use n, since this variable is not to be integrated over, but rather, after the formula has been integrated or when it is assumed that it has been integrated, this variable n is just used to form the progression. Therefore, a certain variable quantity x must be contained in the differential formula, which after the integration must be set equal to another number concerning the progression4; and hence the term, whose index is n, results.

2Euler refers to Wallis’ book "Arithmetica inﬁntorum sive nova method inquirendi in curvilineorum quadraturam, aliaque difﬁciliora Matheseos problemata", published 1655 3Euler means as an exponential, i.e. as ab for some real numbers a and b. 4By this Euler means that the upper limit of the integral has to be chosen appropriately Translation of E19 159

§6 In order that this is understood more clearly, I say that R pdx is the general term of the progression which is to be found in the following from it; but let p denote an arbitrary function of x and of constants, one of which constants must be n. Now assume pdx to be integrated and augmented by such a constant such that having put x = 0 the whole integral vanishes; then put x equal to a certain known quantity. Having done this only quantities related to the progression will remain in the found integral, and that integral will express the term corresponding to the index = n. In other words, the integral determined in this way will be the general term. If this integration is actually possible, the differential formula is not necessary and the progression formed from this will have a general algebraic term; but things change, if the integration only succeeds for certain numbers n.

§7 Therefore, I considered and tried many differential formulas of this kind only admitting an integration, if a positive integer is substituted for n such that the principal terms become algebraic, and hence formed progressions. Therefore, their general terms were immediately clear, and it will be possible to deﬁne the quadrature describing the intermediate terms. I will certainly not go through many formulas of this kind here, but will only treat a general one which extends very far and which can be accommodated to all progressions, whose the arbitrary terms are products consisting of a certain number of terms depending on the index; the factors of these products are fractions, whose numerators and denominators proceed in an arbitrary arithmetic progression, as, e.g.,

2 2 · 4 2 · 4 · 6 2 · 4 · 6 · 8 + + + + etc. 3 3 · 5 3 · 5 · 7 3 · 5 · 7 · 9

§8 Let this formula be given Z xedx(1 − x)n describing the general term; this formula, integrated in such a way that it becomes = 0, if x = 0, and then having set x = 1, expresses the term of order n of the progression resulting from it. Therefore, let us see, which progression it actually describes. We have n n(n − 1) n(n − 1)(n − 2) (1 − x)n = 1 − x + x2 − x3 + etc. 1 1 · 2 1 · 2 · 3 and hence n n(n − 1) n(n − 1)(n − 2) xedx(1 − x)n = xedx − xe+1dx + xe+2dx − xe+3dx + etc. 1 1 · 2 1 · 2 · 3 Hence Z xe+1 nxe+2 n(n − 1)xe+3 n(n − 1)(n − 2)xe+4 xedx(1 − x)n = − + − + etc. e + 1 1 · (e + 2) 1 · 2 · (e + 3) 1 · 2 · 3 · (e + 4) Set x = 1, since the addition of the constant is not necessary; and one will have 1 n n(n − 1) n(n − 1)(n − 2) − + − + etc. e + 1 1 · (e + 2) 1 · 2 · (e + 3) 1 · 2 · 3 · (e + 4) as the general term in question of the series. The series will be of such a nature that, if n = 0, the = 1 = 1 = corresponding term results to be e+1 ; if n 1, the term (e+1)(e+2) ; if n 2, the corresponding = 1·2 = = 1·2·3·4 term is (e+1)(e+2)(e+3) , if n 3, then the corresponding term is (e+1)(e+2)(e+3)(e+4) ; the rule describing how these terms proceed, is manifest. 160 Appendix

§9 Therefore, I obtained this progression 1 1 · 2 1 · 2 · 3 + + + etc., (e + 1)(e + 2) (e + 1)(e + 2)(e + 3) (e + 1)(e + 2)(e + 3)(e + 4) whose general term is Z xedx(1 − x)n.

On the other hand, the term of order n will be expressed by this form

1 · 2 · 3 · 4 ··· n . (e + 1)(e + 2) ··· (e + n + 1) This form certainly suffices to find the terms of integer indices, but if the indices were no integers, this form cannot be used to find the corresponding terms. This following series will be helpful to find them approximately

1 n n(n − 1) n(n − 1)(n − 2) − + − + etc. e + 1 1 · (e + 2) 1 · 2 · (e + 3) 1 · 2 · 3 · (e + 4)

If R xedx(1 − x)n is multiplied by e + n + 1, one will have a progression, whose term of order n has this form

1 · 2 · 3 ··· n , (e + 1)(e + 2) ··· (e + n) whose general term will therefore be Z (e + n + 1) xedx(1 − x)n.

Here it is to be noted that the progression always becomes algebraic, whenever a positive number is assumed for e. For the sake of an example, set e = 2; then the n-th term of the progression will be 1 · 2 · 3 ··· n 1 · 2 or . 3 · 4 · 5 ··· (n + 2) (n + 1)(n + 2) The general term itself also indicates this; since this term will be Z (n + 3) xxdx(1 − x)n.

For, its integral is

(1 − x)n+1 2(1 − x)n+2 (1 − x)n+3 C − + − (n + 3); n + 1 n + 2 n + 3 in order for this to become = 0, if one has x = 0, it will be

1 2 1 C = − + . n + 1 n − 2 n + 3 Set x = 1; the general term will be

n + 3 2(n + 3) 2 − + 1 = . n + 1 n + 2 (n + 1)(n + 2) Translation of E19 161

f §10 Therefore, in order to obtain transcendental progressions, set e equal to the fraction g . The term of order n of the progression will be

1 · 2 · 3 ··· n = gn ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) or g · 2g · 3g ··· ng . ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) On the other hand, the general term will be

f + (n + 1)g Z f = x g dx(1 − x)n. g If this is divided by gn, we will obtain the general term for the progression

1 1 · 2 1 · 2 · 3 + + + etc., f + g ( f + g)( f + 2g) ( f + g)( f + 2g)( f + 3g) whose term of n-th order is

1 · 2 · 3 ··· n = . ( f + g)( f + 2g) ··· ( f + ng) Therefore, the general term of the progression will be

f + (n + 1)g Z f = x g dx(1 − x)n. gn+1

f If the fraction g is not equal to an integer number, or if f is no multiple of g, the progression will be transcendental and the intermediate terms will depend on quadratures.

§11 In order to display the general term more clearly, I want to mention a certain example. In the ﬁrst progression of the preceding paragraph, let f = 1 and g = 2; the term of order n will be

2 · 4 · 6 · 8 ··· 2n = , 3 · 5 · 7 · 9 ··· (2n + 1) the progression itself on the other hand will be this one

2 2 · 4 2 · 4 · 6 + + + etc., 3 3 · 5 3 · 5 · 7 whose general term will hence be

2n + 3 Z √ dx(1 − x)n x. 2

1 1 Let the term corresponding to the index 2 be in question; therefore, it will be n = 2 and one will ﬁnd the general term in question to be Z √ = 2 dx x − xx. 162 Appendix

Since this denotes the element of the circular area, it is perspicuous that the term in question is the area of the circle, whose diameter is = 1. Further, let this series be put forth

r r(r − 1) r(r − 1)(r − 2) 1 + + + + etc., 1 1 · 2 1 · 2 · 3 which is the coefﬁcient of the binomial raised to the power r. Therefore, the term of order n is

r(r − 1)(r − 2) ··· (r − n + 2) . 1 · 2 · 3 ··· (n − 1) In the preceding paragraph on the other hand, we found this expression

1 · 2 · 3 ··· n . ( f + g)( f + 2g) ··· ( f + ng) In order to compare these two expressions, invert the second one such that one has

( f + g)( f + 2g) ··· ( f + ng) ; 1 · 2 ··· n n now multiply it by f +ng and it will be ( f + g)( f + 2g) ··· ( f + (n − 1)g) = ; 1 · 2 ··· (n − 1) therefore, it must be f + g = r and f + 2g = r − 1, whence it will be g = −1 and f = r + 1. Treat the following general term in the same way

f + (n + 1)g Z f x g dx(1 − x)n. gn+1

For the propounded progression

r r(r − 1) 1 + + + etc. 1 1 · 2 this general term will result

n(−1)n+1 . (r − n)(r − n + 1) R x−r−1dx(1 − x)n Let r = 2; the general term of this progression

1, 2, 1, 0, 0, 0etc. will be n(−1)n+1 . (2 − n)(3 − n) R x−3dx(1 − x)n Here it must be noted that this and other cases, in which e + 1 is a negative number, cannot be deduced from the general expression; for, in these cases, the integral does not become = 0, if x = 0. But in order to treat even these cases, it is convenient to integrate Z xedx(1 − x)n Translation of E19 163 in a peculiar way; for, after the integration an inﬁnite constant is to be added. But whenever e + 1 is a positive number, as I assumed in § 8, the addition of the constant is not necessary. But having considered the progression, whose term of order n was the following

r(r − 1)(r − 2) ··· (r − n + 2) , 1 · 2 · 3 ··· (n − 1) that form of the term corresponding to the index n can be transformed into this one

r(r − 1) ··· 1 . (1 · 2 · 3 ··· (n − 1))(1 · 2 ··· (r − n + 1)) But by means of § 14, Z r(r − 1) ··· 1 = dx(− log(x))r and Z 1 · 2 · 3 ··· (n − 1) = dx(− log(x))n−1 and Z 1 · 2 ··· (r − n + 1) = dx(− log(x))r−n+1.

Therefore, the progression considered there

r r(r − 1) r(r − 1)(r − 2) 1 + + + + etc. 1 1 · 2 1 · 2 · 3 has this general term

R dx(− log(x))r − R dx(− log(x))n−1 R dx(− log(x))r−n+1 If it was r = 2, the general term will be

2 , R dx(− log(x))n−1 R dx(− log(x))3−n to which this progression corresponds

1, 2, 1, 0, 0, 0 etc.;

3 and if the term of the index 2 is in question, it will be 2 . R 1 R 3 dx(− log(x)) 2 dx(− log(x)) 2 Therefore, having called the area of the circle, whose diameter is = 1, A, since

Z 1 √ Z 3 3√ dx(− log(x)) 2 = A and dx(− log(x)) 2 = A, 2 the term falling in the middle between the ﬁrst two terms of the progression 1, 2, 1, 0, 0, 0 etc. 4 5 will be 3A , this means approximately 3 . 164 Appendix

§12 Now I proceed to the progression I talked about in the beginning,

1 + 1 · 2 + 1 · 2 · 3 + etc. and in which the term corresponding to the index n is 1 · 2 · 3 · 4 ··· n. This progression is contained in our general one, but the general term must be derived from it in a peculiar way. Until now I considered the general term, if the term of order n is

1 · 2 · 3 ··· n , ( f + g)( f + 2g) ··· ( f + ng) which, if one sets f = 1 and g = 0, goes over into 1 · 2 · 3 ··· n, which is in question; therefore, in the general term

f + (n + 1)g Z f x g dx(1 − x)n gn+1 substitute these values for f and g; the general term in question will be

1 n Z x 0 dx(1 − x) . 0n+1

And I will investigate the value of this general term as follows.

§13 Considering the condition that general terms of this kind must fulﬁll in order to be useful, it is understood that instead of x other functions can be assumed, as long as they were of such a kind that they are = 0, if x = 0, and = 1, if x = 1. For, if a function of this kind is substituted for p x, the general term will fulﬁll the same condition as before. Therefore, write x f +g instead of x and − f g g+ f as a logical consequence f +g x dx instead of dx; having done this one will have

f + (n + 1)g Z g g dx(1 − x f +g )n. gn+1 f + g Now set f = 1 and g = 0 here; one will have

Z dx(1 − x0)n . 0n But because x0 = 1, here we have a case, in which the numerator (1 − x0)n and the denominator n 5 1−x0 0 vanish. Therefore, applying the known rule let us find the value of the fraction 0 . This 1−xz will by achieved by finding the value of the fraction z for vanishing z; therefore, differentiate −xzdz log(x) the numerator and the denominator with respect to that variable z; one will find dz or −xz log(x); if now one puts z = 0, − log(x) will result. Therefore,

1 − x0 = − log(x). 0 5By this Euler means what we refer to as L’Hospital’s rule today. Euler derived the rule in book [E212] written in 1755. He most likely knew the rule from Jacob Bernoulli, his mentor in his teenage years. Jacob Bernoulli on the other hand also was a teacher of L’Hospital. Translation of E19 165

§14 Therefore, because 1 − x0 = − log(x), 0 it will be (1 − x0)n = (− log(x))n 0n R dx(1−x0)n R n and therefore the general term in question 0n is transformed into dx(− log(x)) . Its value can be expressed by means of quadratures. Therefore, the general term of this progression

1, 2, 6, 24, 120, 720 etc. is Z dx(− log(x))n, and it is to be used in the same way as it was prescribed above. That this really is the general term of the propounded progression is also seen from this, that it indeed yields the correct terms for the cases, in which the indices are positive integers. For the sake of an example, let n = 3; it will be

Z Z dx(− log(x))3 = −dx(log(x))3 = −x(log(x))3 + 3x(log(x))2 − 6x log(x) + 6x; the addition of a constant is not necessary, since for x = 0 everything vanishes; therefore, put x = 1; since log(1) = 0, all terms containing logarithms will vanish and 6 will remain, which is the third term.

§15 It is clear that this method to find the terms of this series is too laborious, even for those terms corresponding to integer indices; they are certainly obtained more easily by just continuing the progression. But this expression is nevertheless more than appropriate to find the terms corresponding to rational indices; for, it was not even possible to define these terms using the 1 R p most laborious methods. If one sets x = 2 , the corresponding term will be = dx − log(x), whose value is given by quadratures. But at the beginning [§ 11] I showed that this term is equal to the square root of the area the circle, whose diameter is 1. Hence it is certainly not possible to conclude the same because of the missing Analysis; but below a method will be explained to reduce the same intermediate terms to quadratures of algebraic curves. And comparing this method to the one already explained it will perhaps be possible to derive many results which can be used to develop the whole field of Analysis even further.

§16 The general term of the progression, whose term of order n is given by

1 · 2 · 3 ··· n , ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) is, by means of § 10, f + (n + 1)g Z f x g dx(1 − x)n. gn+1 But if the term of order n was 166 Appendix

1 · 2 · 3 ··· n, then the general term is Z dx(− log(x))n.

If this formula is substituted for 1 · 2 · 3 ··· n, one will have R dx(− log(x))n f + (n + 1)g Z f = x g dx(1 − x)n. ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) gn+1 Hence

gn+1 R dx(− log(x))n ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) = . R f ( f + (n + 1)g) x g dx(1 − x)n Therefore, this expression is the general term of this general progression

( f + g), ( f + g)( f + 2g), ( f + g)( f + 2g)( f + 3g) etc. Therefore, by means of the general term, all terms corresponding to any arbitrary index of all progressions of this kind are deﬁned. What will follow below on the reduction of R dx(− log(x))n to more familiar quadratures or quadratures of algebraic curves, will also be useful here.

§17 Let f + g = 1 and f + 2g = 3; it will be g = 2 and f = −1. Hence this particular progression will result

1, 1 · 3, 1 · 3 · 5, 1 · 3 · 5 · 7 etc. Therefore, its general term is

2n+1 R dx(− log(x))n . R − 1 (2n + 1) x 2 dx(1 − x)n Although here one exponent of x is negative, nevertheless the inconvenience addressed above 1 does not occur here, since it is greater than −1. Put n = 2 to ﬁnd the term corresponding to the 1 index 2 ; that term will be √ 3 R p R p 2 2 dx − log(x) 2 · dx − log(x) = = 1 √ R dx−xdx . R − 2 √ 2 x dx 1 − x x−xx p But from § 15 it is known that R dx − log(x) gives the square root of the circle, whose diameter 1 R p is = 1; let the circumference of that circle be p; the area will be = 4 p and hence dx − log(x) 1 √ gives 2 p. Further, Z dx − xdx Z dx √ √ = √ + x − xx, 2 x − xx 2 x − xx but R √dx gives the arc, whose sinus versus is x, of the circle. Therefore, having put x = 1 1 p 2 x−xx 2 will result. Therefore, the term in question will be Translation of E19 167

s 2 = . p

§18 Since the general term of the progression, whose term of order n is given by

( f + g)( f + 2g) ··· ( f + ng), by means of § 16 reads

gn+1 R dx(− log(x))n , R f ( f + (n + 1)g) x g dx(1 − x)n in like manner, if the term of order n was

(h + k)(h + 2k) ··· (h + nk), the general term will be

kn+1 R dx(− log(x))n . R h (h + (n + 1)k) x k dx(1 − x)n Divide the first progression by the second, namely the first term by the first, the second by the second and so forth; in this way one will arrive at a new progression, whose term of order n will be

( f + g)( f + 2g) ··· ( f + ng) . (h + k)(h + 2k) ··· (h + nk) And the general term of this progression composed of these two will be

n+1 R h n g (h + (n + 1)k) x k dx(1 − x) . R f kn+1( f + (n + 1)g) x g dx(1 − x)n This term does not contain the logarithmic integral R dx(− log(x))n.

§19 In all general terms of this kind, one has to note that not even for f , g, h, k one has to put constant numbers, but they can be assumed to depend on n arbitrarily. For, in the integration these letters are treated in the same way as n, namely as constants. Therefore, let the term of order n be this one

( f + g)( f + 2g) ··· (g + ng); nn−n set g = 1, but f = 2 . Since the progression itself is

f + g, ( f + g)( f + 2g), ( f + g)( f + 2g)( f + 3g) etc. sett 1 instead of g everywhere; the progression will be

f + 1, ( f + 1)( f + 2), ( f + 1)( f + 2)( f + 3) etc. 168 Appendix

But one has to write 0 in the ﬁrst term instead of f , 1 in the second, 3 in the third, 6 in the fourth and so forth; then this progression will result

1, 2 · 3, 4 · 5 · 6, 7 · 8 · 9 · 10 etc., whose general term is

2 R dx(− log(x))n 2 R dx(− log(x))n = . R nn−n n R n−1 n+1 (nn + n + 2) x 2 dx(1 − x) (nn + n + 2) dx(x 2 − x 2 )n

§20 Now I proceed to the progressions, which led me to the invention of the artiﬁce to deﬁne the intermediate of this progression more easily

1, 2, 6, 24, 120 etc. For, this artiﬁce extends further than only to this progression, since its general term Z dx(− log(x))n also enters the general terms of inﬁnitely many other progressions. I assume this general term

f + (n + 1)g Z f x g dx(1 − x)n, gn+1 to which this term of order n corresponds

1 · 2 · 3 ··· n . ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) Here I set f = n, g = 1; hence the general term will be Z Z (2n + 1) xndx(1 − x)n or (2n + 1) dx(x − xx)n and its form of order n will be

1 · 2 · 3 ··· n . ( f + g)( f + 2g)( f + 3g) ··· 2n The progression itself on the other hand is this one

1 1 · 2 1 · 2 · 3 , , etc. 2 3 · 4 4 · 5 · 6 or this one 1 · 1 1 · 2 · 1 · 2 1 · 2 · 3 · 1 · 2 · 3 , , . 1 · 1 1 · 2 · 3 · 4 1 · 2 · 3 · 4 · 5 · 6 In this expression, the numerators are the squares of the progression 1, 2, 6, 24 etc.; and now it is easy to ﬁnd the terms corresponding to rational indices whose denominator is 2. For, in the 1 1 progression 1, 2, 6, 24 etc. let the term corresponding to the index 2 be A; the term of order 2 of AA the propounded progression will then be = 1 . Translation of E19 169

§21 In the general term Z (2n + 1) xndx(1 − x)n

1 set n = 2 ; the term corresponding to that exponent will be Z √ AA 2 dx x − xx = , 1 whence r Z √ A = 1 · 2 dx x − xx

1 = to the term of the progression 1, 2, 6, 24 etc. corresponding to the index 2 ; hence this term, as is clear from the integral, is the square root of the area of the circle, whose diameter is 1. Now call 3 the term of order 2 of this progression A; the corresponding term in the assumed progression will be

A · A Z 3 = 4 dx(x − xx) 2 , 1 · 2 · 3 therefore, r Z 3 A = 1 · 2 · 3 · 4 dx(x − xx) 2 .

5 In like manner, the term of order 2 will be found to be r Z 5 = 1 · 2 · 3 · 4 · 5 · 6 dx(x − xx) 2 .

p From these result I conclude in general that the term of order 2 will be r Z p = 1 · 2 · 3 · 4 ··· (p + 1) dx(x − xx) 2 .

Therefore, in this way one ﬁnds all terms of the progression 1, 2, 6, 24 etc., whose indices are fractions, while the denominator is 2.

§22 Further, in the general term

f + (n + 1)g Z f x g dx(1 − x)n gn+1 I set f = 2n, while g remains = 1; then this expression results Z (3n + 1) dx(xx − x3)n as general term of the progression

1 1 · 2 1 · 2 · 3 , , etc. 3 5 · 6 7 · 8 · 9 Multiply that one by the preceding (2n + 1) R dx(x − xx)n; then this expression will result Z Z Z (2n + 1)(3n + 1) dx(x − xx)n dx(xx − x3)n dx(x3 − x4)n. 170 Appendix

This will give this progression

1 · 1 · 1 1 · 2 · 1 · 2 · 1 · 2 , etc., 1 · 2 · 3 1 · 2 · 3 · 4 · 5 · 6 where the numerators are the cubes of the corresponding terms of the progression 1, 2, 6, 24 etc. 1 Let the term of order 3 of this progression be A; the corresponding term of that progression will be

3 Z Z A 2 1 3 1 = 2 + 1 dx(x − xx) 3 dx(xx − x ) 3 , 1 3 1 therefore, the term of order 3 is r Z Z 3 5 1 1 1 · 2 · dx(x − xx) 3 dx(xx − x3) 3 ; 3 2 similarly, the term of order 3 is r Z Z 3 7 2 2 1 · 2 · 3 · dx(x − xx) 3 dx(xx − x3) 3 . 3 4 And the term of order 3 is r Z Z 3 11 4 4 1 · 2 · 3 · 4 · 5 · dx(x − xx) 3 dx(xx − x3) 3 3 p and in general the term of order 3 is r Z Z 3 2p + 3 p p 1 · 2 ··· p · · (p + 1) dx(x − xx) 3 dx(xx − x3) 3 . 3

§23 If we want to proceed further by putting f = 3n, it will be necessary to multiply the general term Z (4n + 1) dx(x3 − x4)n by the preceding ones, whence one has Z Z (2n + 1)(3n + 1)(4n + 1) dx(x − xx)n dx(x2 − x3)n, which is the general term of this series

1 · 1 · 1 · 1 1 · 2 · 1 · 2 · 1 · 2 · 1 · 2 , etc. 1 · 2 · 3 · 4 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 Using this expression, the terms of the progression 1, 2, 6, 24 etc. will be deﬁned, whose indices p are fractions having the denominator 4. For, the term, whose index is 4 , will be found to be s 2p 3p = 4 1 · 2 · 3 ··· p + 1 + 1 (p + 1) 4 4 Z Z Z p 3 p 3 4 p × dx(x − xx) 4 dx(xx − x ) 4 dx(x − x ) 4 . Translation of E19 171

p Hence it is possible to conclude in general that the term of order q is s 2p 3p 4p = q 1 · 2 · 3 ··· p + 1 + 1 + 1 ··· (p + 1) q q q

Z p Z p Z p Z p × dx(x − xx) q dx(x2 − x3) q dx(x3 − x4) q ··· dx(xq−1 − xq) q .

Therefore, from this formula the terms corresponding to any arbitrary fractional indices are found by means of a quadrature of algebraic curves; but for this 1 · 2 · 3 ··· p is required, the term, whose index is the numerator of the propounded fraction.

§24 In the same way it is possible to proceed further to higher composed progressions by assuming higher composited numbers, but I will not follow this line any further here. It is also possible to use multiple integrals so that the general term is Z Z qdx pdx; for, the integral of pdx must be multiplied by qdx and what results from the integration must be then integrated again; and the result of this second integration will just then, having put x = 1, give the general term of the series. But in each of the two integrations, in order for it to be well-deﬁned, a constant of such a kind has to be added that having put x = 0 the integral likewise becomes = 0. In like manner, general terms are to be treated, which are expressed using several integrals, i.e. Z Z Z rdx qdx pdx.

But nevertheless functions of such a kind are always to be taken for p, q, r that, as often as n is a positive integer, at least algebraic terms result.

§25 Let the general term be

Z dx Z xedx(1 − x)n; x this expression converted into a series gives

xe+1 nxe+2 n(n − 1)xe+3 − + − etc. (e + 1)2 1 · (e + 2)2 1 · 2 · (e + 3)2 Having put x = 1, one will have the term of order n expressed by this series

1 n n(n − 1) − + − etc. (e + 1)2 1 · (e + 2)2 1 · 2 · (e + 3)2 The progression, beginning from the term corresponding to the index 0, will be

1 (e + 2)2 − (e + 1)2 (e + 3)2(e + 2)2 − 2(e + 3)2(e + 1)2 + (e + 2)2(e + 1)2 , , (e + 1)2 (e + 2)2(e + 1)2 (e + 3)2(e + 2)2(e + 1)2

(e + 4)2(e + 3)2(e + 2)2 − 3(e + 4)2(e + 3)2(e + 1)2 + 3(e + 4)2(e + 2)2(e + 1)2 − (e + 3)2(e + 2)2(e + 1)2 (e + 4)2(e + 3)2(e + 2)2(e + 1)2 172 Appendix

etc.

The structure of this progression is manifest and does not require any explanation. Let e = 0; it will be

Z 1 − (1 − x)n+1 dx(1 − x)n = ; n + 1 therefore, the general term is

Z dx − dx(1 − x)n+1 , (n + 1)x the progression on the other hand will be

1 4 − 1 9 · 4 − 2 · 9 · 1 + 4 · 1 16 · 9 · 4 − 3 · 16 · 9 · 1 + 3 · 16 · 4 · 1 − 9 · 4 · 1 , , , . 1 4 · 1 9 · 4 · 1 16 · 9 · 4 · 1 The difference will constitute this progression

−1 −9 + 4 −16 · 9 + 2 · 16 · 4 − 9 · 4 , , etc. 4 · 1 9 · 4 · 1 16 · 9 · 4 · 1

§26 Therefore, in this dissertation I achieved, what I mainly intended, namely to ﬁnd the general terms of all progressions, each term of which is a product of factors proceeding in an arithmetic progression, and in which the number of factors depends on the index in an arbitrary manner. But although here the number of factors is always set equal to the index, if the number of factors is desired to depend on it in another way, this will not cause any difﬁculties. The index is denoted nn+n by the letter n; if now anyone would require that the number of factors is 2 , it is only necessary nn+n to substitute 2 for n everywhere.

§27 Instead of ending the dissertation here, I want to add something more curious than useful. It is known that by dnx the differential of order n of x is understood and dn p, if p denotes a certain function of x and dx is put constant, is proportional to dxn; and, if n is a positive integer number, the ratio of dn p to dnx can always be expressed algebraically; consider, e.g., the case n = 2 and p = x3, the ratio of d2(x3) to dx2 will be the same as 6x to 1. Now it is in question, what that ratio will be, if n is a rational number. The difﬁculty is easily understood in these cases; for, if n is a positive integer number, dn is found by iterated differentiation; but this is not possible, if n is a fractional number. But nevertheless by means of interpolations of the progressions I considered in this dissertation, it will be possible to answer this question.

n e n dn(ze) §28 Let the ratio of d (z ) to dz to be found for constant dz, or let the value of the fraction dzn be in question. First, let us see, what its values are, if n is an integer number; after this we will then proceed to the non-integer cases. If n = 1, its value will be

1 · 2 · 3 ··· e eze−1 = ze−1; 1 · 2 · 3 ··· (e − 1) I express e in this way such that later on the results we found in this paper can be applied more easily here. If n = 2, the value will be Translation of E19 173

1 · 2 · 3 ··· e e(e − 1)ze−2 = ze−2. 1 · 2 · 3 ··· (e − 2) If n = 3, one will have

1 · 2 · 3 ··· e e(e − 1)(e − 2)ze−3 = ze−3. 1 · 2 · 3 ··· (e − 3)

Hence, I generally infer, whatever n is, that it will always be

dn(ze) 1 · 2 · 3 ··· e = ze−n. dzn 1 · 2 · 3 ··· (e − n) But by means of § 14 Z Z 1 · 2 · 3 ··· e = dx(− log(x))e und 1 · 2 · 3 ··· (e − n) = dx(− log(x))e−n.

Hence one has R dn(ze) dx(− log(x))e = ze−n dzn R dx(− log(x))e−n or R dx(− log(x))e dn(ze) = ze−ndzn . R dx(− log(x))e−n Here dz is set constant and R dx(− log(x))s and R dx(− log(x))s−n must be integrated in such a way as it was prescribed above.

§29 It is not necessary to show how the true value is found; this will become clear by substituting 1 an arbitrary integer number for n. But let it be questioned what d 2 z is, if dz is constant. Therefore, 1 it will be e = 1 and n = 2 in our expression. Therefore, one will have R 1 dx(− log(x)) √ d 2 z = p zdz. R dx − log(x)

But Z dx(− log(x)) = 1 and having called the area of the circle, whose diameter is 1, A, it will be Z q √ dx − log(x) = A, whence r 1 zdz d 2 z = . A Therefore, let this equation for a curve be propounded

1 p yd 2 z = z dy, 174 Appendix

1 q 2 zdz where dz is set constant, and let it be questioned what kind of curve this is. Since d z = A , the equation will go over into this one r zdz √ y = z dz, A which, having squared it, gives

yydz = zdy, A whence one ﬁnds

1 1 log(z) = c − A y or y log(z) = cAy − A, which is the equation for the curve in question. Appendix III

Translation of E368 - On the hypergeometric curve expressed by the equation y = 1 · 2 · 3 ··· x

1. While the letter x denotes the abscissa and y the ordinate, this equation immediately indicates the quantity only of those ordinates corresponding to integer numbers; for, if one had

the abscissas x ··· 0, 1, 2, 3, 4, 5, 6 etc. one will have the ordinates y ··· 1, 1, 2, 6, 24, 120, 720 etc. such that, while the abscissas are taken according to the natural numbers, the ordinates proceed according to the Wallisian hypergeometric1 progression; therefore, it will be convenient to call also this curve hypergeometric. But even though through this equation certainly innumerable, but just a discrete set of, points of this curve are assigned, nevertheless the nature of this curve is to be considered to be determined by this equation such that to each abscissa a certain, and, via this equation, well-deﬁned ordinate corresponds2. For, the nature of this equation requires, if the ordinate y = q corresponds to a certain abscissa x = p, that the ordinate y = q(p + 1) corresponds q to the abscissa x = p + 1, but the ordinate y = p corresponds to the abscissa x = p − 1. Therefore, one cannot draw a certain curve of parabolic kind through this inﬁnitely many points arbitrarily, since all its points are determined from the equation.

2. But except for these ordinates corresponding to abscissas expressed by integer numbers, those are especially noteworthy, which fall into the middle between them from the equation; and they are all determined by the one I once showed to correspond to the abscissa x = 1 and to be equal √ 2 to 1 π. Therefore, since 2 √ π = 1.77245385090548, all these ordinates as well for the positive as for the negative abscissas will be as follows:

1Wallis called such series hypergeometric which are nowadays referred to as factorial series. Wallis reasoning for the name might was as follows: While in a geometric series the ratio of two subsequent terms is always the same, in factorial series the quotient of two subsequent terms increases and is thus more than geometric, i.e. "hypergeometric". 2From our investigations in the main text we know that this statement is incorrect, confer especially section 1.3.2.

175 176 Appendix

for the positive abscissas for the negative abscissas

x the ordinate y is x the ordinate y is

0 1 0 +1

1 1 2 0.8862269 − 2 +1.7724538 1 1 −1 ±∞

1 1 1 2 1.3293404 −1 2 −3.5449077 2 2 −2 ∓∞

1 1 2 2 3.3233509 −2 2 +2.3632718 3 6 −3 ±∞

1 1 3 2 11.6317284 −3 2 −0.9453087 4 24 −4 ∓∞

1 1 4 2 52.3427777 −4 2 +0.2700882 5 120 −5 ±∞

1 1 5 2 287.8852775 −5 2 −0.0600196 6 720 −6 ∓∞

1 1 6 2 1871.2543038 −6 2 +0.0109126 7 5040 −7 ±∞

From this I drew the curve seen in ﬁgure 1, which extends from the negative abscissa x = −1, Translation of E368 177

Figure 1 from E368. It shows the factorial function from x = −1 to x = 3. The scan was taken from the Opera Onmia Version, i.e. p. 44 of volume 28 of the first series. where the ordinate becomes infinite, to x = 3, where y = 6, and from this point on is to be understood to ascend to infinity; but the values on the left, where for each integer value of the abscissa the ordinates go over into asymptotes, I did not expresses beyond x = −1.

3. The consideration of this curve raises many rather curious questions, providing a reason to examine it more accurately; and their solution seem to be even more interesting, since the equation for our curve cannot be expressed in the usual manner. Questions of this kind first concern the determination of the remaining points of a curve in addition to those which are easily assigned. After this, in each point the tangents require their own investigation such that the behaviour of the whole curve can be defined more easily. But from the inspection of the figure it is perspicuous that between the abscissas x = 0 and x = 1 there must be a smallest ordinate somewhere: to assign its abscissa and its value will be worth of one’s while. Furthermore, between each two negative abscissas −1, −2, −3, −4, −5 etc., where the ordinates extend to infinity, there necessarily are smallest ordinates, which, the more we proceed to the left, become continuously smaller until eventually they vanish completely. Finally, also the question on the curvature radius in each point deserves our attention, and the point, where the curvature is the largest, seems to be especially remarkable, since it is manifest that in the elongation of the curve from the axis the branches come continuously closer to the straight line. Thus, I want to 178 Appendix resolve those questions.

First Question To ﬁnd a continuous equation among the abscissa x and the ordinate y for the hypergeometric curve, which equally holds, no matter whether an integer or a fractional number is taken for x.

4. Since the propounded equation y = 1 · 2 · 3 ··· x can only hold, if x is an integer number, it must be cast into another form which is not restricted by this condition; this can be achieved in multiple ways using expressions running to inﬁnity, among which at ﬁrst this one occurs:

1 2x 2 3x 3 4x 4 5x y = · · · · etc. 1 + x 1 2 + x 2 3 + x 3 4 + x 4 the factors of which must be continued to inﬁnity. The reason for the validity of this expression is obvious, since the more factors are taken the closer the true value is being obtained, and, having taken inﬁnitely many factors, the true value is obtained accurately: For, if the numbers of factors is = n, one has

1 2 3 n y = · · ··· (n + 1)x; 1 + x 2 + x 3 + x n + x if its numerator is represented this way:

1 · 2 · 3 ··· x(x + 1)(x + 2)(x + 3) ··· n, the denominator on the other hand this way

(1 + x)(2 + x)(3 + x) ··· n(n + 1)(n + 2) ··· (n + x), having cancelled the common factors, it results in

1 · 2 · 3 ··· x y = (n + 1)x. (n + 1)(n + 2)(n + 3) ··· (n + x) Hence, if n is an inﬁnite number, because of the n + 1 single factors and the total amount of x factors of the denominator, the whole denominator is cancelled by the factor (n + 1)x and the propounded equation becomes y = 1 · 2 · 3 ··· x.

5. This formula can be generalised a bit; for, since the whole task reduces to this that the factor (n + 1)x becomes equal to the last denominator

(n + 1)(n + 2)(n + 3) ··· (n + x), in the case, in which n is an infinite number, it is evident that this condition is also satisfied, if the x factor is in general set to (n + a) , where a is an arbitrary finite number; but this√ formula is most 1+x appropriate for our task, if a certain mean value of 1 and x, e.g., a = 2 or a = x, is attributed to a. Now it is necessary that this factor (n + a)x is resolved into so many factors as n contains units, which is conveniently achieved using this resolution3:

3Actually, the resolution Euler states, has n + 1 factors and not just n. But since he is interested in the case n → ∞ and the resolution is correct, this does not cause any mistakes in the following. Translation of E368 179

a + 1x a + 2x a + 3x a + n x (n + a)x = ax · · · ··· . a a + 1 a + 2 a + n − 1 Therefore, for an arbitrary abscissa x we will have the ordinate:

1 a + 1x 2 a + 2x 3 a + 3x y = ax · · · · etc. to inﬁnity, 1 + x a 2 + x a + 1 3 + x a + 2 which expression is always true, whatever number is chosen for a, but leads to the truth most 1+x quickly, if one takes a = 2 , whence it will be: 1 + x x 1 3 + x x 2 5 + x x 3 7 + x x y = · · · · etc., 2 1 + x 1 + x 2 + x 3 + x 3 + x 5 + x which expression consists of inﬁnitely many factors of the form

m a + m x m + x a − m − 1 except the ﬁrst ax, and the more terms are multiplied by each other in a given case, the closer one will get to the truth. But the initial expression results, if one takes a = 1.

6. But this expression is the more useful, the faster the factors converge to one, which happens 1+x by taking a = 2 ; indeed, then the calculation will become as much easier as smaller numbers are substituted for x; but it always sufﬁces to have investigated ordinates for abscissas x between one and zero, since from there the ordinates corresponding to x + 1, x + 2, x + 3, x + 4 etc. are α easily derived. Therefore, let x = β , while α < β, and it will be

α α α α α + β β β 3α + β β 2β 5β + α β 3β 7β + α β y = · · · · etc., 2β α + β β + α α + 2β 3β + α α + 3β 5β + α whence the power yβ of the ordinate results in this expression:

α + β α ββ(3β + α)α (2β)β(5β + α)α (3β)β(7β + α)α yβ = · · · · etc. 2β (β + α)β(β + α)α (2β + α)β(3β + α)α (3β + α)β(5β + α)α

α But for the abscissa x = − β the ordinate is hence calculated to be

2β α ββ(β − α)α (2β)β(3β − α)α (3β)β(5β − α)α yβ = · · · · etc. β − α (β − α)β(3β − α)α (2β − α)β(5β − α)α (3β − α)β(7β − α)α

1 For the sake of an example, let us take x = 2 and we will obtain: 3 2 · 2 · 7 4 · 4 · 11 6 · 6 · 15 8 · 8 · 19 y2 = · · · · · etc.; 4 3 · 3 · 3 5 · 5 · 7 7 · 7 · 11 9 · 9 · 15 since a general factor of this is

2n · 2n(4n + 3) 16n3 + 12nn 1 = = 1 + , (2n + 1)(2n + 1)(4n − 1) 16n3 + 12nn − 1 (2n + 1)2(4n − 1) 180 Appendix it is seen in general how quickly these factors converge to 1; therefore, it will be:

3 1 1 1 1 1 y2 = 1 + 1 + 1 + 1 + 1 + etc., 4 32 · 3 52 · 7 72 · 11 92 · 15 112 · 19 √ 2 π 1 where we know that y = 4 . But if we set x = − 2 , which corresponds to y = π, from the other expression it will be

2 · 2 · 1 4 · 4 · 5 6 · 6 · 9 8 · 8 · 13 π = 4 · · · · · etc. 1 · 1 · 5 3 · 3 · 9 5 · 5 · 13 7 · 7 · 17 or

1 1 1 1 π = 4 1 − 1 − 1 − 1 − etc., 12 · 5 32 · 9 52 · 13 72 · 17 hence

1 1 1 1 π = 3 1 + 1 + 1 + 1 + etc., 32 · 3 52 · 7 72 · 11 92 · 15 such that the one expression will get to the truth while it is increasing, the other while it is decreasing.

7. But the calculation is executed more conveniently, if our expression is terminated at each factor; for, then the following formulas coming continuously closer to the truth will result:

1 3 + x x y = 1 + x 2 1 2 5 + x x y = · 1 + x 2 + x 2 1 2 3 7 + x x y = · · 1 + x 2 + x 3 + x 2 1 2 3 4 9 + x x y = · · · 1 + x 2 + x 3 + x 4 + x 2 1 2 3 4 5 11 + x x y = · · · · . 1 + x 2 + x 3 + x 4 + x 5 + x 2

y Since, if one writes x − 1 instead of x, the ordinate = results, by similar formulas it will be x Translation of E368 181

− 2 + x x 1 y = 2 − 2 4 + x x 1 y = 1 + x 2 − 2 3 6 + x x 1 y = · 1 + x 2 + x 2 − 2 3 4 8 + x x 1 y = · · 1 + x 2 + x 3 + x 2 − 2 3 4 5 10 + x x 1 y = · · · . 1 + x 2 + x 3 + x 4 + x 2 √ 1 1 Hence, having set x = 2 , for the ordinate y = 2 π two series of formulas converging to it result:

r r 1√ 2 7 1√ 4 π = π = 2 3 4 2 5 r r 1√ 2 · 4 11 1√ 4 4 π = π = 2 3 · 5 4 2 3 9 r r 1√ 2 · 4 · 6 15 1√ 4 · 6 4 π = π = 2 3 · 5 · 7 4 2 3 · 5 13 r r 1√ 2 · 4 · 6 · 8 19 1√ 4 · 6 · 8 4 π = π = 2 3 · 5 · 7 · 9 4 2 3 · 5 · 7 17 r 1√ 4 · 6 · 8 · 10 4 etc. π = 2 3 · 5 · 7 · 9 21

etc.

8. But products of this kind are expanded most conveniently using logarithms; and ﬁrst from the general formula involving the arbitrary number a we obtain:

a + 1 a + 2 a + 3 a + 4 log y = x log a + x log + x log + x log + x log + etc. a a + 1 a + 2 a + 3 x x x − log(1 + x) − log 1 + − log 1 + − log 1 + − etc. 2 3 4

1+x having taken a = 2 this series is rendered most convergent: 1 + x x + 3 x + 5 x + 7 x + 9 log y = x log + x log + x log + x log + x log + etc. 2 x + 1 x + 3 x + 5 x + 7 x x x log(1 + x) − log 1 + − log 1 + − log 1 + − etc. 2 3 4 Therefore, having taken these logarithms and since in general: 182 Appendix

x + 2m + 1 2x 2x 2x 2x x log = + + + + etc. x + 2m − 1 x + 2m 3(x + 2m)3 5(x + 2m)5 7(x + 2m)7 x 2x 2x3 2x5 2x7 and log 1 + = + + + + etc. m x + 2m 3(x + 2m)3 5(x + 2m)5 7(x + 2m)7 we obtain the following formulas consisting of inﬁnitely many series:

1 + x 2 1 1 1 1 log y = x log + x(1 − xx) + + + + etc. 2 3 (x + 2)3 (x + 4)3 (x + 6)3 (x + 8)3 2 1 1 1 1 + x(1 − x4) + + + + etc. 5 (x + 2)5 (x + 4)5 (x + 6)5 (x + 8)5 2 1 1 1 1 + x(1 − x6) + + + + etc. 7 (x + 2)7 (x + 4)7 (x + 6)7 (x + 8)7 2 1 1 1 1 + x(1 − x8) + + + + etc. 9 (x + 2)9 (x + 4)9 (x + 6)9 (x + 8)9

etc.

9. Let us take a deﬁnite number of terms of the ﬁrst series, which number we want to be = n, and since the upper part is reduced to the single term x log(a + n), it will be

1 1 1 log y = x log(a + x) − log(1 + x) − log 1 + x − log 1 + x − · · · − log 1 + x , 2 3 n which expression comes the closer to the truth, the greater the number n is taken. Therefore, let n be very large, and ﬁrst we will obviously have

a aa a3 log(n + a) = log n + − + − etc., n 2n2 3n3 1+x where it will be convenient to take 2 for a; but then, for the sake of brevity, having put the fraction

0.5772156649015325 = ∆, we know the sum of the harmonic progression to be:

1 1 1 1 1 1 1 1 + + + + ··· + = ∆ + log n + − + − etc., 2 3 4 n 2n 12nn 120n4 whence, since:

1 1 1 1 1 1 1 log(n + a) = 1 + + + + ··· + − ∆ − + − + etc. 2 3 4 n 2n 12nn 120n4 a aa a3 + − + − etc., n 2nn 3n3 Translation of E368 183

1+x having taken a = 2 , we conclude 1 1 1 log y = −∆x + x + x + x + ··· + x 2 3 n 1 1 1 − log(1 + x) − log 1 + x − log 1 + x − · · · − log 1 + x 2 3 n

xx x + 6xx + 3x3 + − + etc. 2n 24nn Therefore, by increasing the number n to inﬁnity, it will actually be:

1 1 1 log y = −∆x + x + x + x + x + etc. 2 3 4 1 1 1 − log(1 + x) − log 1 + x − log 1 + x − log 1 + x − etc. 2 3 4 and, having expanded each logarithm into a series:

1 1 1 1 log y = −∆x + xx 1 + + + + etc. 2 22 32 42 1 1 1 1 − x3 1 + + + + etc. 3 23 33 43 1 1 1 1 + x4 1 + + + + etc. 4 24 34 44 1 1 1 1 − x5 1 + + + + etc. 5 25 35 45

etc.

10. But except for those formulas, in which the ordinate y corresponding to a certain abscissa x is assigned, my method to sum progressions indeﬁnitely4 provides us with an extraordinary expression accommodated to our purposes. For, since log y = log 1 + log 2 + log 3 + log 4 + ··· + log x, this progression must be summed indeﬁnitely; but introducing numerical values:

1 1 1 1 1 691 A = , B = , C = , D = , E = , F = etc., 6 90 945 9450 93555 1 · 3 · 5 ··· 15 · 315 which progression is of such a nature that

5B = 2AA, 7C = 4AB, 9D = 4AC + 2BB, 11E = 4AD + 4BC etc., I showed elsewhere5 that it will be:

4Euler refers to the Euler-Maclaurin summation formula. 5The following formula is just the Stirling formula for the factorial. Euler showed it, e.g., in [E212]. 184 Appendix

1 1 A 1 · 2B 1 · 2 · 3 · 4C 1 · 2 · 3 · 4 · 5 · 6D log y = log 2π + x + log x − x + − + − + etc., 2 2 2x 23x3 25x5 27x7 which series, compared to the ﬁrst, has the use that, the greater the abscissas x are taken, the faster it exhibits the true value of the ordinate y. Therefore, since, if the ordinate y corresponds to the abscissa x, the following ordinate

y(x + 1)(x + 2)(x + 3) ··· (x + n) to the larger abscissa x + n, we will have the following rapidly convergent series6:

1 log y = log 2π − log(x + 1) − log(x + 2) − log(x + 3) − · · · − log(x + n) 2 1 + x + n + log(x + n) − x − n 2 A 1 · 2B 1 · 2 · 3 · 4C 1 · 2 · 3 · 4 · 5 · 6D + − + − + etc. 2(x + n) 23(x + n)3 25(x + n)5 27(x + n)7 Therefore, if e denotes the number whose natural logarithm is = 1, and if, for the sake of brevity, one sets:

A 1 · 2B 1 · 2 · 3 · 4C − + − etc. = s, 2(x + n) 23(x + n)3 25(x + n)5 going back from logarithms to numbers we conclude:

p x+n 2π(x + n) x + n y = es, (x + 1)(x + 2)(x + 3) ··· (x + n) e where the integer number n is arbitrary; but the larger it is taken, the easier the true value of s can be found.

11. Finally, the ordinate y can even be exhibited by an integral formula; for, having put the abscissa x = p and having introduced the new variable u, independent of the quantity p, the ordinate will be

Z 1 p y = du log , u if the integration is extended from the value u = 0 to the value u = 1. Or, if one prefers the exponential form, it will also be Z y = e−vvpdv, extending the integration from v = 0 to v = ∞. From those formulas, if the abscissa p is an integer number, the integration indeed immediately yields

y = 1 · 2 · 3 ··· p,

6This series does, in fact, not converge and is to be understood as an asymptotic series. Confer our discussion in section 1.6.4.1. Translation of E368 185 but if p was a fractional number, hence it is at the same time understood to which class of transcendental quantities the value of y is to be referred. Indeed, I showed on another occasion, how the integral can then be expressed using quadratures of algebraic curves.

12. Therefore, lo and behold the many solutions of our ﬁrst question, in which for an arbitrary abscissa x, even though it is expressed by a non-integer number, the value of the ordinate y was sought after; it will be helpful to have listed up the principal ones, that from here in each case the one which seems to be the most useful can be chosen:

1 2 x 2 3 x 3 4 x 4 5 x I. y = · · · · etc. 1 + x 1 2 + x 2 3 + x 3 4 + x 4 1 + x x 1 3 + x x 2 5 + x x 3 7 + x x II. y = · · · · etc. 2 1 + x 2 + x 2 + x 3 + x 3 + x 5 + x 2 3 4 5 III. log y = x log + x log + x log + x log + etc. 1 2 3 4 1 1 1 − log(1 + x) − log 1 + x − log 1 + x − log 1 + x − etc. 2 3 4 1 + x x + 3 x + 5 x + 7 x + 9 IV. log y = x log + x log + x log + x log + x log + etc. 2 x + 1 x + 3 x + 5 x + 7 1 1 1 − log(1 + x) − log 1 + x − log 1 + x − log 1 + x − etc. 2 3 4 1 1 1 V. log y = −∆x + x + x + x + x + etc. 2 3 4 1 1 1 − log(1 + x) − log 1 + x − log 1 + x − log 1 + x − etc. 2 3 4   1 1 1 1 −∆x + xx 1 + + + + etc.  2 22 32 42    1 1 1 1  − x3 1 + + + + etc.  3 23 33 43   1 1 1 1 VI. log y = + x4 1 + + + + etc.  4 24 34 44    1 5 1 1 1  − x 1 + + + + etc.  5 25 35 45     + etc.

1 1 A 1 · 2B 1 · 2 · 3 · 4C 1 · 2 ··· 6D VII. log y = log 2π + x + log x − x + − + − + etc. 2 2 2x 23x3 25x5 27x7 while ∆ = 0.57721556649014225 and

1 1 1 1 1 A = , B = , C = , D = , E = etc. 6 90 945 9450 93555 Then in the three last forms one has to use natural logarithms. 186 Appendix

Second Question To deﬁne the direction of its tangent on the hypergeometric curve for each point.

13. Therefore, here we assume that for the abscissa x the value of the ordinate y has already dy been found, and since the direction of the tangent is deﬁned by the ratio of the differentials dx , by which fraction the tangent of the angle, in which the tangent at that point is inclined to the axis, is usually expressed, it is just necessary that we differentiate one of the found formulas. To this end, formula V seems especially suitable, from which we conclude:

dy 1 1 1 = −∆ + 1 + + + + etc. ydx 2 3 4 1 1 1 1 − − − − − etc. 1 + x 2 + x 3 + x 4 + x which expression is contracted into this more convenient one:

dy x x x x = −∆ + + + + + etc., ydx 1 + x 2(2 + x) 3(3 + x) 4(4 + x) whence it is clear at the same time, if x is a negative integer number, that not just the ordinate y dy but also the formula dx becomes inﬁnite such that in these points the ordinates, since they are asymptotes, become the tangents. But let us in general put the angle the tangent constitutes with the axis to be = ϕ such that

dy = tan ϕ. dx

14. Therefore, ﬁrst let us deﬁne the tangents for the abscissas x which are expressed by positive numbers, since the ordinates y are given.

I. Therefore, let x = 0 and, because of y = 1,

dy = −∆ = −0.5772156649 = tan ϕ, dx whence

the angle ϕ = −29◦5902900, where the sign − indicates that the tangent falls to the right of the axis and does not constitute an angle of 30◦ with it.

II. Let x = 1 and, because of y = 1,

dy = 1 − ∆ = 0.422784335 = tan ϕ, dx and hence the angle ϕ = 22◦550. Translation of E368 187

III. Let x = 2 and, because of y = 2,

dy 1 = 2 1 + − ∆ = 1.845568670 = tan ϕ dx 2 and hence

the angle ϕ = 61◦330.

IV. Let x = 3 and, because of y = 6,

dy 1 1 = 6 1 + + − ∆ = tan ϕ dx 2 3 or

tan ϕ = 7.536706010 and ϕ = 82◦260.

V. Let x = 4 and, because of y = 24,

dy 1 1 1 = 24 1 + + + − ∆ dx 2 3 4 and hence

tan ϕ = 36.146824040 and ϕ = 88◦250.

Therefore, in general, if the abscissa x is equal to an arbitrary integer number, because of y = 1 · 2 ··· n, it will be

dy 1 1 1 = tan ϕ = 1 · 2 · 3 ··· n 1 + + + ··· + − ∆ . dx 2 3 n

15. Hence let us also deﬁne the tangents for the intermediate points, and ﬁrst certainly for those corresponding to positive abscissas: √ 1 1 I. Let x = 2 , it will be y = 2 π and dy 2 1 2 1 2 = −∆ + 1 − + − + − + etc. ydx 3 2 5 3 7 or

dy 1 1 1 1 = −∆ + 2 − + − + etc. = −∆ + 2(1 − log 2) ydx 2 3 4 5 and hence

dy = tan ϕ = y(2(1 − log 2) − ∆) = 0.0364899739 · y. dx 188 Appendix

1 · 3√ II. Let x = 3 , it will be y = π and 2 2 · 2 dy 1 = −∆ + 2 1 + − log 2 , ydx 3 whence

dy 1 = tan ϕ = y 2 1 + − log 2 − ∆ = 0.7031566405 · y. dx 3 √ 5 1·3·5 III. Let x = 2 , it will be y = 2·2·2 π and

dy 1 1 = −∆ + 2 1 + + − log 2 ydx 3 5 hence

1 1 tan ϕ = y 2 1 + + − log 2 − ∆ = 1.1031566405 · y. 3 5

Since now

1q π · (2(1 − log 2) − ∆) = 0.0323383973, 2 for these cases it will be:

1 x = , y = 0.8862269, tan ϕ = 0.0323384, 2 3 x = , y = 1.3293404, tan ϕ = 0.9347345, 2 5 x = , y = 3.3233509, tan ϕ = 3.6661767, 2 7 x = , y = 11.6317284, tan ϕ = 16.1549694, 2 9 x = , y = 52.3427777, tan ϕ = 84.3290907, 2

etc.

16. Before I proceed, I observe, if for any abscissa it was

x = p, y = q, tan ϕ = r, that then for the following abscissa it will be

x = p + 1, y = q(p + 1) and tan ϕ = r(p + 1) + q, Translation of E368 189 but for the preceding one

q r q x = p − 1, y = and tan ϕ = − , p p pp whence we can easily continue the above values backwards:

1 x = , y = 0.8862269, tan ϕ = 0.0323384, 2 1 x = − , y = 1.7724538, tan ϕ = − 3.4802308, 2 3 x = − , y = − 3.5449077, tan ϕ = − 0.1293538, 2 5 x = − , y = + 2.3632718, tan ϕ = + 1.6617504, 2 7 x = − , y = − 0.9453087, tan ϕ = − 1.0428236, 2 9 x = − , y = + 0.2700882, tan ϕ = + 0.3751176, 2 11 x = − , y = − 0.0600196, tan ϕ = − 0.0966971, 2 13 x = − , y = + 0.0109126, tan ϕ = + 0.0195654, 2

etc.

17. The same differential equation serves for ﬁnding the point µ of the curve, where the ordinate dy is the smallest or the tangent is parallel to the axis. Therefore, having put dx = 0, the corresponding abscissa x must be found from this equation:

x x x x x ∆ = + + + + + etc., 1 + x 2(2 + x) 3(3 + x) 4(4 + x) 5(5 + x) which is expanded into this one: 190 Appendix

1 1 1 ∆ = + x 1 + + + + etc. 22 32 42 1 1 1 − x2 1 + + + + etc. 23 33 43 1 1 1 + x3 1 + + + + etc. 24 34 44 1 1 1 − x4 1 + + + + etc. 25 35 45

etc.

But having substituted the proximate sums of these series it will be

0 = + 0.5772156649 − 1.6449340668 x + 1.2020569032 x2 − 1.0823232337 x3 + 1.0369277551 x4 − 1.0173430620 x5 + 1.0083492774 x6 − 1.0040773562 x7 + 1.0020083928 x8 − 1.0009945751 x9 + 1.0004941886 x10 − 1.0002460866 x11 + 1.0001227133 x12 − 1.0000612481 x13 + 1.0000305882 x14 − 1.0000152823 x15 etc.

But if the ﬁrst two fractions are kept, the following a lot more convergent series emerges

= + − x − x 0 0.5772156649 1+x 2(2+x) + 0.0770569032x2 − 0.3949340668x + 0.0056777551x4 − 0.0198232337x3 + 0.0005367774x6 − 0.0017180620x5 + 0.0000552678x8 − 0.0001711062x7 + 0.0000059074x10 − 0.0000180126x9 + 0.0000006430x12 − 0.0000019460x11 + 0.0000000706x14 − 0.0000002130x13 + 0.0000000078x16 − 0.0000000235x15 etc. Translation of E368 191

1 Hence one ﬁnds approximately x = 2 , but this minimal ordinate will be deﬁned more easily by means of the following question.

Third Question

Given a point of the hypergeometric curve to investigate the nature of an inﬁnitesimal portion of this curve around this point.

18. Therefore, for the given abscissa x = p let the ordinate y = q have been found; and now one has to ﬁnd the ordinate, which corresponds to the abscissa p + ω differing from that one by just a small amount. Therefore, since according to formula V

1 1 1 log q = −∆p + p + p + p + p + etc. 2 3 4 1 1 1 − log(1 + p) − log 1 + p − log 1 + p − log 1 + p − etc., 2 3 4 if one writes p + ω instead of p here, instead of log q the value of log(q + ψ) will result, by which the question will be resolved. And if we set log q = P, writing p + ω instead of p, it is known to result

ωdP ω2ddP ω3d3P ω4d4P log(q + ψ) = P + + + + + etc. 1dp 1 · 2dp2 1 · 2 · 3dp3 1 · 2 · 3 · 4dp4

But on the other hand, as we have seen:

dP p p p p = −∆ + + + + + etc. dp 1 + p 2(2 + p) 3(3 + p) 4(4 + p) and hence further:

ddP 1 1 1 1 = + + + + etc. 1 · dp2 (1 + p)2 (2 + p)2 (3 + p)2 (4 + p)2 d3P 1 1 1 1 = − − − − − etc. 1 · 2dp3 (1 + p)3 (2 + p)3 (3 + p)3 (4 + p)3 d4P 1 1 1 1 = + + + + etc. 1 · 2 · 3dp4 (1 + p)4 (2 + p)4 (3 + p)4 (4 + p)4

etc. whence, because of P = log q, we conclude: 192 Appendix

ψ p p p log 1 + = −∆ω + ω + + + etc. q 1 + p 2(2 + p) 3(3 + p) 1 1 1 1 + ω2 + + + etc. 2 (1 + p)2 (2 + p)2 (3 + p)2 1 1 1 1 − ω3 + + + etc. 3 (1 + p)3 (2 + p)3 (3 + p)3 1 1 1 1 + ω4 + + + etc. 4 (1 + p)4 (2 + p)4 (3 + p)4 1 1 1 1 − ω2 + + + etc. 5 (1 + p)5 (2 + p)5 (3 + p)5

etc.

19. Here the coordinates p and q can be considered as constants, since the letters ω and ψ denote two new coordinates taken from a given point of the curve and parallel to the first set; from their relation defined here the nature of the curve around that point is easily investigated. Thus, since we have already assigned innumerable points of the curve, hence the trace of each portion of the curve between two of those conjugated points can be defined approximately. First, from that differentiated equation, as before, the inclination ϕ of the tangent to the axis is calculated and

dψ p p p = tan ϕ = q −∆ + + + + etc. . dω 1 + p 2(2 + p) 3(3 + p) Further, if for the differential equation, for the sake of brevity, we set

dψ = Adω + Bωdω + Cω2dω + etc., the curvature radius at a given point of the curve will be

3 (1 + AA) 2 1 = = B B · cos3 ϕ because of A = tan ϕ. But on the other hand

p p p B = tan ϕ −∆ + + + + etc. 1 + p 2(2 + p) 3(3 + p)

1 1 1 1 +q + + + + etc. , (1 + p)2 (2 + p)2 (3 + p)2 (4 + p)2 whence, if the curvature radius is set = r, it will be

1 sin2 ϕ cos ϕ 1 1 1 = + q + + + etc. . r q (1 + p)2 (2 + p)2 (3 + p)2 Translation of E368 193

20. But in order to extend the investigation of the direction and the curvature from the principal point deﬁned by the coordinates p and q to the points of the curve, for the sake of brevity, let us set

p p p p −∆ + + + + + etc. = P, 1 + p 2(2 + p) 3(3 + p) 4(4 + p) 1 1 1 1 + + + + etc. = Q, (1 + p)2 (2 + p)2 (3 + p)2 (4 + p)2 1 1 1 1 + + + + etc. = R, (1 + p)3 (2 + p)3 (3 + p)3 (4 + p)3 1 1 1 1 + + + + etc. = S (1 + p)4 (2 + p)4 (3 + p)4 (4 + p)4

etc., that

ψ 1 1 1 1 log 1 + = Pω + Qω2 − Rω3 + Sω4 − Tω5 + etc. q 2 3 4 5 Hence now differentiating we ﬁnd:

dψ = (q + ψ)(P + Qω − Rω2 + Sω3 − Tω4 + etc.) dω and differentiating further

ddψ = (q + ψ)(P + Qω − Rω2 + Sω3 − Tω4 + etc.)2 dω2 +(q + ψ)(Q − 2Rω + 3Sω2 − 4Tω + etc.) d3ψ = 3(q + ψ)(Q − 2Rω + 3Sω2 − 4Tω3 + etc.)(P + Qω − Rω2 + Sω3 − etc.) dω3 +(q + ψ)(P + Qω − Rω2 + Sω3 − Tω4 + etc.)3 −(q + ψ)(2R − 6Sω + 12Tω2 − etc.). Having covered these calculations, for the point of the curve corresponding to the abscissa x = p + ω and y = q + ψ the direction of the tangent will be

dψ tan ϕ = = (q + ψ)(P + Qω − Rω2 + Sω3 − Tω4 + etc.). dω But then, having put the curvature radius = r, we know that it will be:

3 dψ2 2 ddψ ddψ r = 1 + : = 1 : cos3 ϕ dω2 dω2 dω2 or

1 ddψ = cos3 ϕ, r dω2 194 Appendix whence we ﬁnd for the variability of the curvature:

dr d3ψ 3ddψ dϕ − = cos3 ϕ − · sin ϕ cos2 ϕ. rrdω dω3 dω2 dω But on the other hand

dϕ ddψ = , cos2 ϕ dω whence:

dr d3ψ ddψ 2 − = cos3 ϕ − 3 sin ϕ cos4 ϕ. rrdω dω3 dω2

Fourth Question To investigate the nature of the hypergeometric curve around its lowest point µ where the ordinate is the smallest.

21. Since this point is not far away from the point corresponding to the abscissa = 1 and the √ √ 2 1 1 1 ordinate = 2 π, let us set p = 2 so that q = 2 π, and hence ﬁrst let us ﬁnd the values of the letters P, Q, R, S etc., which will result as:

1 1 1 P = − ∆ + + + + etc. = 2(1 − log 2) − ∆ = 0.03648997397857 3 2 · 5 3 · 7 4 4 4 4 Q = + + + + etc. = 0.93480220054468 32 52 72 92 8 8 8 8 R = + + + + etc. = 0.41439832211716 33 53 73 93 16 16 16 16 S = + + + + etc. = 0.0.23484850566707 34 54 74 94 32 32 32 32 T = + + + + etc. = 0.144760040831276 35 55 75 95 64 64 64 64 V = + + + + etc. = 0.09261290502029 36 56 76 96 128 128 128 128 W = + + + + etc. = 0.06035822809843 37 57 77 97

Further, 1√ q = π = 0.88622692545274. 2

22. Hence let us especially deﬁne the point µ, where the ordinate is the smallest simple approximations shows it to correspond to the abscissa x = 0.4616, having set Translation of E368 195

1 p + ω = + ω = 0.4616, 2 one ﬁnds approximately

ω = −0.0383, dψ which value must be investigated more accurately from the equation dω = 0 or

P + Qω − Rω2 + Sω3 − Tω4 + etc. = 0. 1 1 Therefore, since approximately ω = − 26 , we set ω = − 26 − z, and after the substitution it has to be

+ 0.03595393079018 + 0.934802200z + 0.00061301526940 + 0.031876794z + 0.414398zz + 0.00001336188585 + 0.001042227z + 0.027097zz + 0.00000031677900 + 0.000032945z + 0.001285zz + 0.00000000779479 + 0.000001013z + 0.000053zz + 0.00000000019538 + 0.000000030z + 0.000002zz + 0.00000000000496 + 2z + + 13 + 0.03658063271970 + 0.967755211z + 0.442835zz 0.03648997397857 0 = 0.00009065874113 + 0.967755211z + 0.442835zz whence one ﬁnds

z = −0.00009368323 and hence

ω = −0.03836785523. Therefore, the smallest ordinate mµ corresponds to the abscissa

Om = 0.46163214477. For the ordinate mµ = q + ψ on the other hand one has to expand the equation

ψ 1 1 1 1 log 1 + = Pω + Qω2 − Rω3 + Sω4 − Tω5 + etc., q 2 3 4 5 from which one concludes

ψ log 1 + = −0.000704053 q 196 Appendix and further

ψ 1 + = 1 − 0.000703805, q so that the smallest ordinate becomes

mµ = q + ψ = 0.8856031945.

23. Now let us in general differentiate to deﬁne the value of ψ from the logarithmic equation, and, having done the calculation, we will obtain:

ψ = + 0.0364899740ω + 0.468066860ω2 q − 0.121069221ω3 + 0.16321479ω4 − 0.09360753ω5 + etc., which terms sufﬁce, if the value of ω is very small. But, for the sake of brevity, let us set

ψ = Pω + Qω2 − Rω3 + Sω4 − Tω5 q such that

P = 0.0364899740, Q = 0.468066860, R = 0.121069221, S = 0.16321479, T = 0.09360753, and hence we will have:

dψ = q P + 2Qω − 3Rω2 + 4Sω3 − 5Tω4 , dω ddψ = q(2Q − 6Rω + 12Sω2 − 20Tω3). dω2 If we now want to ﬁnd the radius of curvature at the lowest point µ, where

ω = −0.03836785523,

dψ dω2 since there we have dω = 0, that radius of curvature will be = ddψ . Set the curvature radius in this point = r, and since

1 = 2q(Q − 3Rω + 6Sω2 − 10Tω3) = 0.9669949, r for the point µ the curvature radius results as

r = 1.166893. Translation of E368 197

24. I investigated these determinations of the lowest point µ of the curve with all eagerness such that it cannot without any reason be conjectured, as this point has an extraordinary property, that the numbers exhibiting its nature contain in this way a certain elegance, and if they cannot be expressed sufﬁciently simply in terms of a rational or irrational number, that they are at least to be referred to a certain simpler kind of transcendental quantities. But against the expectation it happened that such a criterion for elegance appears neither in the abscissa

Om = 0.46163214477 nor in the ordinate

mµ = 0.8856031945 nor in the curvature radius at this point

= 1.166893; since no afﬁnity to simpler rational numbers or irrational numbers or to the quadrature of the circle or to logarithmic or exponential numbers is detected. Since, if the abscissa Om is considered as a logarithm, the number corresponding to it could seem to promise several things, I sought after this number and found

= 1.586616, in which no afﬁnity to any known quantities is recognised.

25. Before I end this speculation, it will helpful to have observed that the formula 1 · 2 · 3 ··· x can also be expressed indeﬁnitely in terms of the following series

x(x − 1) x(x − 1)(x − 2) xx − x(x − 1)x + (x − 2)x − (x − 3)x + etc., 1 · 2 1 · 2 · 3 which, as often as x is a positive integer number, immediately gives that product as 1 · 2 · 3 ··· x. This is indeed also achieved by this further extending expression:

x(x − 1) x(x − 1)(x − 2) ax − x(a − 1)x + (a − 2)x − (a − 3)x + etc., 1 · 2 1 · 2 · 3 for, if for x one successively substitutes the numbers 1, 2, 3 etc., it will be as follows:

a0 = 1 a1 − (a − 1)1 = 1 a2 − 2(a − 1)2 + (a − 2)2 = 1 · 2 a3 − 3(a − 1)3 + 3(a − 2)3 − (a − 3)3 = 1 · 2 · 3 a4 − 4(a − 1)4 + 6(a − 2)4 − 4(a − 3)4 + (a − 4)4 = 1 · 2 · 3 · 4 a5 − 5(a − 1)5 + 10(a − 2)5 − 10(a − 3)5 + 5(a − 4)5 − (a − 5)5 = 1 · 2 · 3 · 4 · 5 etc. 198 Appendix

26. These are certainly obvious from the results demonstrated about the difference of each order of algebraic progressions, but nevertheless from the nature of these series the truth is not easily uncovered; thus, the following proof seems to be in order. Since for smaller exponents x the matter is obvious, I reason as follows, i.e. that, having conceded the truth for the case x = n, I will show that it also follows for the case x = n + 1. Therefore, let

n(n − 1) I. an − n(a − 1)n + (a − 2)2 − etc. = N = 1 · 2 · 3 ··· n, 1 · 2 and since the number N does not depend on a, it will also be:

n(n − 1) II. (a − 1)n − n(a − 2)n + (a − 3)2 − etc. = N, 1 · 2 which subtracted from the ﬁrst leaves:

(n + 1) (n + 1)n (n + 1)n(n − 1) III. an − (a − 1)n + (a − 2)n − (a − 3)n + etc. = 0; 1 1 · 2 1 · 2 · 3 multiply this by a and it results

(n + 1) (n + 1)n (n + 1)n(n − 1) IV. an+1 − a(a − 1)n + a(a − 2)n − a(a − 3)n + etc. = 0; 1 1 · 2 1 · 2 · 3 to this one add equation II multiplied by n + 1, i.e.:

(n + 1)n (n + 1)n(n − 1) V. + (n + 1)1(a − 1)n − 2(a − 2)n + 3(a − 3)n − etc. = (n + 1)N 1 · 2 1 · 2 · 3 and the aggregate IV+V will give

(n + 1) (n + 1)n (n + 1)n(n − 1) VI. an+1 − (a − 1)n+1 + (a − 2)n+1 − (a − 3)n+1 + etc. = (n + 1)N, 1 1 · 2 1 · 2 · 3 where, because of N = 1 · 2 · 3 ··· N, it will be (n + 1)N = 1 · 2 · 3 ··· (n + 1). Therefore, it is proved that, if our proposition

x(x − 1) x(x − 1)(x − 2) ax − x(a − 1)x + (a − 2)x − (a − 3)x + etc. = 1 · 2 · 3 ··· x 1 · 2 1 · 2 · 3 was true in the case x = n, it will also be true in the case x = n + 1. Therefore, since it is obviously true in the case x = 1, it follows that it is also true for all positive integer numbers assumed for x.

27. But although this expression is sufficiently elegant and worth one’s complete attention, it is nevertheless less useful for our task, in which the hypergeometric curve it propounded, since for the cases, in which x is a fractional number, this series not only runs to infinity but it also, if the denominator is an even number, contains imaginary terms such that its value cannot even be 1 calculated using approximations. Therefore, having set x = 2 , this infinite series results: Translation of E368 199

√ 1√ 1 · 1√ 1 · 1 · 3√ 1 · 1 · 3 · 5√ a − a − 1 − a − 2 − a − 3 − a − 4 − etc., 2 2 · 4 2 · 4 · 6 2 · 4 · 6 · 8 √ 1 1 whose value can hardly be shown by anyone to be = 2 π. In like manner, taking x = − 2 , we already know from the above results that

√ 1 1 1 · 3 1 · 3 · 5 π = √ + √ + √ + √ + etc. a 2 a − 1 2 · 4 a − 2 2 · 4 · 6 a − 3 Nevertheless, a further investigation of this series is left to the mathematics, especially if it is extended and represented in this form:

m(m − 1) m(m − 1)(m − 2) s = xn − m(x − 1)n + (x − 2)n − (x − 3)n + etc.; 1 · 2 1 · 2 · 3 for, without much effort one soon detects extraordinary properties, whose expansion seems worth our complete attention. I will now present all extraordinary phenomena I was able to observe about it7.

Observations on the Series m(m − 1) m(m − 1)(m − 2) s = xn − m(x − 1)n + (x − 2)n − (x − 3)n + etc. 1 · 2 1 · 2 · 3 I. In the preceding, I already demonstrated, if the exponent n was = m, that the sum of this series will be

s = 1 · 2 · 3 ··· m such that in this case it does not depend on the number x. But I ﬁrst conclude, if n = m − 1, that s = 0. For, having taken n = m,

m(m − 1) ··· 1 · 2 · 3 ··· m = xm − m(x − 1)m + (x − 2)m − etc. 1 · 2 Y and, writing x − 1 instead of x and m − 1 instead of m, in like manner, we get:

(m − 1)(m − 2) ··· 1 · 2 · 3 ··· (m − 1) = (x − 1)m−1 − (m − 1)(x − 2)m−1 + (x − 3)m−1 − etc. 1 · 2 X Now represent that equation [ ] in this way: Y m(m − 1) ··· 1 · 2 · 3 ··· m = x · xm−1 − mx(x − 1)m−1 + x(x − 2)m−1 − etc. 1 · 2 ♂ m(m − 1) + (x − 1)m−1 − (x − 2)m−1 + etc.; 1 but the equation multiplied by m gives: X 7The remainder of this paper is about the last series, which Euler denoted by s. It can be read without the ﬁrst part and can be considered as a separate paper. This might also explain why Euler started the paragraphs from I. again. 200 Appendix

m(m − 1) m(m − 1)(m − 2) · · · 1 · 2 · 3 ··· m = m(x − 1)m−1 − (x − 2)m−1 + (x − 3)m−1 − etc., 1 1 · 2 which subtracted from and divided by x yields: ♂ m m(m − 1) ··· 0 = xm−1 − (x − 1)m−1 + (x − 2)m−1 − etc., 1 1 · 2 ♀ which is the propounded equation for the case n = m − 1, whose value thus is = 0.

II. In like manner, it is shown that the sum s of the propounded series also vanishes in the case n = m − 2. For, represent the series in this way: ♀

m m(m − 1) ··· 0 = x · xm−2 − x(x − 1)m−2 + x(x − 2)m−2 − etc. 1 1 · 2 ' m(m − 1) + m(x − 1)m−2 − (x − 2)m−2 + etc. 1 · 2 and if in the same series one writes x − 1 instead of x and m − 1 instead of m, but the whole series is multiplied by m,♀ it becomes

m(m − 1) ··· 0 = m(x − 1)m−2 − (x − 2)m−2 + etc. 1 % Having subtracted this one from previous one, divide the remainder by x and it will result:

m m(m − 1) 0 = xm−2 − (x − 1)m−2 + (x − 2)m−2 − etc. 1 1 · 2

And so the sum of the propounded series also vanishes in the case n = m − 2, and, in like manner, it can be shown that it also vanishes in the cases n = m − 3, n = m − 4 etc. and in general n = m − i, where i is an arbitrary positive integer number. Keep in mind that the sum of the series s is = 1 · 2 · 3 ··· m in the case n = m, but in the cases, in which the exponent n is smaller than the number m, the sum vanishes, if the numbers m and n are integers or at least n − m is a positive integer number, of course.

III. Therefore, in order to investigate the nature of the remaining cases, let us expand each term of our series and arrange them according to the powers of x, having done which we will obtain: Translation of E368 201

m(m − 1) m(m − 1)(m − 2) s = xn 1 − m + − + etc. 1 · 2 1 · 2 · 3 2m(m − 1) 3m(m − 1)(m − 2) + nxn−1 m − + − etc. 1 · 2 1 · 2 · 3 n(n − 1) 4m(m − 1) 9m(m − 1)(m − 2) − xn−2 m − + − etc. 1 · 2 1 · 2 1 · 2 · 3 n(n − 1)(n − 2) 8m(m − 1) 27m(m − 1)(m − 2) + xn−3 m − + − etc. 1 · 2 · 3 1 · 2 1 · 2 · 3

etc., the sums of which series we will ﬁnd as follows; ﬁrst, exhibit them a bit more generally, and since its sum is known:

m(m − 1) m(m − 1)(m − 2) 1 − mu + u2 − u3 + etc. = (1 − u)m, 1 · 2 1 · 2 · 3 let us differentiate it continuously and always substitute u for du again, and, having changed the signs, it will be:

2m(m − 1) 3m(m − 1)(m − 2) mu − u2 + u3 − etc. = mu(1 − u)m−1 1 · 2 1 · 2 · 3 2m(m − 1) mu − u2 +etc. = mu(1 − u)m−1 − m(m − 1)u2(1 − u)m−2 1 · 2 23m(m − 1) mu − u2 +etc. = mu(1 − u)m−1 − 3m(m − 1)u2(1 − u)m−2 1 · 2

+m(m − 1)(m − 2)u3(1 − u)m−3

24m(m − 1) mu − u2 +etc. = mu(1 − u)m−1 − 7m(m − 1)uu(1 − u)m−2 1 · 2

+6m(m − 1)(m − 2)u3(1 − u)m−3

−m(m − 1)(m − 2)(m − 3)u4(1 − u)m−4

etc.

Therefore, here one now has to write u = 1, after which all terms in each order vanish except for those, where the exponent of 1 − u becomes = 0.

IV. Now successively attribute the values 1, 2, 3, 4, 5 etc. to m and, for the sake of brevity, write n−i i+1 instead of the general coefﬁcient 202 Appendix

n(n − 1)(n − 2) ··· (n − i) 1 · 2 · 3 ··· (i + 1) and we obtain the following values:

if it will be

s n n − 1 n − 2 n − 3 n − 4 m = 1 = xn−1 − xn−2 + xn−3 − xn−4 + xn−5 − etc. 1 1 2 3 4 5 s n − 1 n − 2 n − 3 n − 4 n − 5 m = 2 = xn−2 − 3 xn−3 + 7 xn−4 − 15 xn−5 + 31 xn−6 − etc. 1 · 2 2 3 4 5 6 s n − 2 n − 3 n − 4 n − 5 n − 6 m = 3 = xn−3 − 6 xn−4 + 25 xn−5 − 90 xn−6 + 301 xn−7 − etc. 1 · 2 · 3 3 4 5 6 7 s n − 3 n − 4 n − 5 n − 6 n − 7 m = 4 = xn−4 − 10 xn−5 + 65 xn−6 − 350 xn−7 + 1701 xn−8 − etc. 1 · 2 ··· 4 4 5 6 7 8 s n − 4 n − 5 n − 6 n − 7 n − 8 m = 5 = xn−5 − 15 xn−6 + 140 xn−7 − 1050 xn−8 + 6951 xn−9 − etc. 1 · 2 ··· 5 5 6 7 8 9 s n − 5 n − 6 n − 7 n − 8 n − 9 m = 6 = xn−6 − 21 xn−7 + 266 xn−8 − 2646 xn−9 + 22827 xn−10 − etc. 1 · 2 ··· 6 6 7 8 9 10 where the formation of each numerical coefﬁcient from the preceding is obvious; hence for the last sixth series:

21 = 6 · 1 + 15, 266 = 6 · 21 + 140, 2646 = 6 · 266 + 1050 etc.

And hence it is immediately seen, if m < n, that the value of s vanishes; for, in the last series, if n < 6 and hence either 5 or 4 or 3 etc., it will be

n − 5 n − 6 = 0, = 0 etc. 6 7

But then on the other hand, if n = m, it is also evident that

s = 1, 1 · 2 ··· m for, in the lowest series:

6 − 5 6 − 6 6 − 7 6 − 8 = 1, = 0, = 0, = 0 etc. 6 7 8 9

Expansion of the Cases n = m + 1

V. Hence let us ﬁrst expand the cases in which n = m + 1; the last form yields Translation of E368 203

if these sums s m = 1, n = 2 = 2x − 1 1 s m = 2, n = 3 = 3x − 3 1 · 2 s m = 3, n = 4 = 4x − 6 1 · 2 · 3 s m = 4, n = 5 = 5x − 10 1 · 2 · 3 · 4 s m = 5, n = 6 = 6x − 15 1 · 2 ··· 6 etc.,

where the ﬁrst coefﬁcients of x are equal to n, but the absolute numbers are equal to the triangular number of n; in general, we will have

if this equation s m(m + 1) m n = m + 1 = (m + 1)x − = (m + 1) x − 1 · 2 ··· m 1 · 2 2

such that

m(m − 1) m(m − 1)(m − 2) xm+1 − m(x − 1)m+1 + (x − 2)m+1 − (x − 3)m+1 + etc. 1 · 2 1 · 2 · 3 m = 1 · 2 · 3 ··· (m + 1) x − . 2

Expansion of the Cases n = m + 2

VI. Therefore, for these cases we will have: 204 Appendix

if it was these equations

s 2 m = 1, n = 3 = 3x2 − 3 · 1 x + 1· 1 = 3 xx − x + 1 6 s 7 m = 2, n = 4 = 6x2 − 4 · 3 x + 1· 7 = 6 xx − 2x + 1 · 2 6 s 15 m = 3, n = 5 = 10x2 − 5 · 6 x + 1· 25 = 10 xx − 3x + 1 · 2 · 3 6 s 26 m = 4, n = 6 = 15x2 − 6 · 10x + 1· 65 = 15 xx − 4x + 1 · 2 · 3 · 4 6 s 40 m = 5, n = 7 = 21x2 − 7 · 15x + 1· 140 = 21 xx − 5x + 1 · 2 ··· 5 6 s 57 m = 6, n = 8 = 28x2 − 8 · 21x + 1· 266 = 28 xx − 6x + 1 · 2 ··· 6 6

etc., which forms can be represented as follows:

if it was it will be

s 2 · 3 1 · 4 m = 1, n = 3 = xx − x + 1 1 · 2 12 s 3 · 4 2 · 7 m = 2, n = 4 = xx − 2x + 1 · 2 1 · 2 12 s 4 · 5 3 · 10 m = 3, n = 5 = xx − 3x + 1 · 2 · 3 1 · 2 12 s 5 · 6 4 · 13 m = 4, n = 6 = xx − 4x + 1 · 2 · 3 · 4 1 · 2 12 s 6 · 7 5 · 16 m = 5, n = 7 = xx − 5x + 1 · 2 ··· 5 1 · 2 12 s 7 · 8 6 · 19 m = 6, n = 8 = xx − 6x + , 1 · 2 ··· 6 1 · 2 12 whence it manifestly follows, if in general n = m + 2, that it will be

s m + 1 m + 2 m(3m + 1) = · xx − mx + 1 · 2 ··· m 1 2 12 or Translation of E368 205

s m + 1 m + 2 m 2 m = · x − + . 1 · 2 ··· m 1 2 2 12 Therefore, one obtains this summation

m(m − 1) m(m − 1)(m − 2) xm+2 − m(x − 1)m+2 + (x − 2)m+2 − (x − 3)m+2 + etc. 1 · 2 1 · 2 · 3 1 m 2 m = 1 · 2 · 3 ··· (m + 2) x − + . 2 2 24

Expansion of the Cases n = m + 3 VII. For these cases we will have

if it was these equations

s m = 1, n = 4 = 4x3 − 6 · 1x2 + 4 · 1x − 1 · 1 1 s m = 2, n = 5 = 10x3 − 10 · 3x2 + 5 · 7x − 1 · 15 1 · 2 s m = 3, n = 6 = 20x3 − 15 · 6x2 + 6 · 25x − 1 · 90 1 · 2 · 3 s m = 4, n = 7 = 35x3 − 21 · 10x2 + 7 · 65x − 1 · 350 1 · 2 ··· 4 s m = 5, n = 8 = 56x3 − 28 · 15x2 + 8 · 140x − 1 · 1050 1 · 2 ··· 5 which can be represented in this way:

s 2 · 3 · 4 3 1 · 4 1 · 1 · 2 = x3 − x2 + x − 1 1 · 2 · 3 2 4 8 s 3 · 4 · 5 6 2 · 7 2 · 2 · 3 = x3 − x2 + x − 1 · 2 1 · 2 · 3 2 4 8 s 4 · 5 · 6 9 3 · 10 3 · 3 · 4 = x3 − x2 + x − 1 · 2 · 3 1 · 2 · 3 2 4 8 s 5 · 6 · 7 12 4 · 13 4 · 4 · 5 = x3 − x2 + x − 1 · 2 ··· 4 1 · 2 · 3 2 4 8 s 6 · 7 · 8 15 5 · 16 5 · 5 · 6 = x3 − x2 + x − 1 · 2 ··· 5 1 · 2 · 3 2 4 8

etc., whence in general for the cases n = m + 3 one concludes 206 Appendix

s m + 1 m + 2 m + 3 3m m(3m + 1) mm(m + 1) = · · x3 − x2 + x − 1 · 2 ··· m 1 2 3 2 4 8 m + 1 m + 2 m + 3 m 2 m m = · · x − + x − 1 2 3 2 4 2 such that we obtain:

m(m − 1) m(m − 1)(m − 2) xm+3 − m(x − 1)m+3 + (x − 2)m+3 − (x − 3)m+3 + etc. 1 · 2 1 · 2 · 3 1 m 3 m m = 1 · 2 · 3 ··· (m + 3) x − + x − . 6 2 24 2

Preparation for the following Cases VIII. Although in paragraph IV. we gave the formulas only up to the case m = 6, let us try to ﬁnd a general formula for them. To this end, let us set n = m + λ, and to abbreviate the expression

k(k − 1)(k − 2)(k − 3) ··· (k − i + 1) 1 · 2 · 3 · 4 ··· i let us write

k i such that k denotes the ﬁrst factor of the numerator, i on the other hand the last factor of the denominator. Therefore, let us put for the case

s m + λ m + λ m + λ m + λ m − 1 = xλ+1 − A xλ + B xλ−1 − C xλ−2 + etc. 1 · 2 ··· (m − 1) m − 1 m m + 1 m + 2 s m + λ m + λ m + λ m + λ m = xλ − A1 xλ−1 + B1 xλ−2 − C1 xλ−3 + etc. 1 · 2 · 3 ··· m m m + 1 m + 2 m + 3 such that A1, B1, C1, D1 etc. are the coefﬁcients which must be investigated. But from the law of these formulas we see that

A1 = m · 1 + A, B1 = mA1 + B, C1 = mB1 + C, D1 = mC1 + D etc., where it is evident that

m(m − 1) (m + 1)m A = and A1 = 1 · 2 1 · 2 or, in our notation,

m m + 1 A = and A1 = . 2 2 Now for the following operations I observe that: Translation of E368 207

m + µ + 1 m + µ m + µ − = , ν ν ν − 1 which is clear, since by expanding:

m + µ + 1 (m + µ + 1)(m + µ)(m + µ − 1) ··· (m + µ + 2 − ν) = ν 1 · 2 · 3 ··· ν m + µ (m + µ)(m + µ − 1) ··· (m + µ + 2 − ν)(m + µ + 1 − ν) = , ν 1 · 2 ··· (ν − 1)ν whence it is seen that

m + 1 m m − = = m. 2 2 1

IX. Now for it to be

m + 1 m + 1 m + 1 B1 − B = mA1 = m = 3 + , 2 3 2 let us set

m + 1 m + 1 B = α + β 4 3 and hence

m + 2 m + 2 B1 = α + β 4 3 and it will result

m + 1 m + 1 B1 − B = α + β , 3 2 whence α = 3 and β = 1 such that

m + 2 m + 2 B1 = 3 + . 4 3 But for the following operations note that in general:

m + µ m + µ m + µ m = (ν + 1) + (ν − µ) , ν ν + 1 ν m + µ which form results, if the value of expanded above is multiplied by ν m + µ − ν m = m + µ − ν − ν − µ = (ν + 1) · + (ν − µ). ν + 1 208 Appendix

X. Since, having observed these things, it must be C1 − C = mB1, because of

m + 2 m + 2 m + 2 m = 5 + 2 4 5 4 and

m + 2 m + 2 m + 2 = 4 + 1 , 3 4 3 it will be

m + 2 m + 2 m + 2 mB1 = 15 + 10 + 1 ; 5 4 3 therefore, set

m + 2 m + 2 m + 2 C = 15 + 10 + 1 6 5 4 hence

m + 3 m + 3 m + 3 C1 = 15 + 10 + 1 . 6 5 4

XI. Since, in like manner, it has to be D1 − D = mC1, since

m + 3 m + 3 m + 3 m = 7 + 3 6 7 6 m + 3 m + 3 m + 3 m = 6 + 2 5 6 5 m + 3 m + 3 m + 3 m = 5 + 1 , 4 5 4 it will be

m + 3 m + 3 m + 3 m + 3 mC1 = 105 + 105 + 25 + , 7 6 5 4 whence we conclude:

m + 4 m + 4 m + 4 m + 4 D1 = 105 + 105 + 25 + 1 . 8 7 6 5

XII. Further, because of E1 − E = mD1, since Translation of E368 209

m + 4 m + 4 m + 4 m = 9 + 4 8 9 8 m + 4 m + 4 m + 4 m = 8 + 3 7 8 7 m + 4 m + 4 m + 4 m = 7 + 2 6 7 6 m + 4 m + 4 m + 4 m = 6 + 1 , 5 6 5 we conclude

m + 4 m + 4 m + 4 m + 4 m + 4 mD1 = 945 + 1260 + 490 + 56 + 1 9 8 7 6 5 and hence

m + 5 m + 5 m + 5 m + 5 m + 5 E1 = 945 + 1260 + 490 + 56 + 1 10 9 8 7 6 and, proceeding even further,

m + 6 m + 6 m + 6 m + 6 m + 6 m + 6 F1 = 10395 + 17325 + 9450 + 1918 + 119 + . 12 11 10 9 8 7

Expansion of the case n = m + λ XIII. Therefore, for our series in the case n = m + λ

m m(m − 1) m(m − 1)(m − 2) s = xm+λ − (x − 1)m+λ + (x − 2)m+λ − (x − 3)m+λ + etc., 1 1 · 2 1 · 2 · 3 if we divide the general equation exhibited above in paragraph VIII by

m + λ (m + λ)(m + λ − 1)(m + λ − 2) ··· (λ + 1) = , m 1 · 2 · 3 ··· m we will get to this expression

s λ λ(λ − 1) = xλ − A1xλ−1 + B1xλ−2 (λ + 1)(λ + 2) ··· (λ + m) m + 1 (m + 1)(m + 2) λ(λ − 1)(λ − 2) − C1xλ−3 + etc. (m + 1)(m + 2)(m + 3) where one has to substitute the following values for the letters A1, B1, C1, D1 etc.: 210 Appendix

m + 1 (m + 1)m A1 = = 2 1 · 2 m + 2 m + 2 B1 = 3 + 4 3 m + 3 m + 3 m + 3 C1 = 15 + 10 + 6 5 4 m + 4 m + 4 m + 3 m + 4 D1 = 105 + 15 + 25 + 8 7 4 5 m + 5 m + 5 m + 5 m + 5 m + 5 E1 = 945 + 1260 + 490 + 56 + 10 9 8 7 6 m + 6 m + 6 m + 6 m + 6 m + 6 m + 6 F1 = 10395 + 17325 + 9450 + 1918 + 119 + 12 11 10 9 7 7 where

10395 = 11 · 945, 17325 = 10 · 1260 + 5 · 945 9450 = 9 · 490 + 4 · 1260 1918 = 8 · 56 + 3 · 490 119 = 7 · 1 + 2 · 56 1 = 6 · 0 + 1 · 1 hence, if for the following value one sets

m + 7 m + 7 m + 7 m + 7 m + 7 m + 7 m + 7 G1 = α + β + γ + δ + ε + ζ + η , 14 13 12 11 10 9 8 these coefﬁcients will be determined as follows:

α = 13 · 10395 ε = 9 · 119 + 3 · 1918 β = 12 · 17325 + 6 · 10395 ζ = 8 · 1 + 2 · 119 γ = 11 · 9450 + 5 · 17325 η = 7 · 0 + 1 · 1 δ = 10 · 1918 + 4 · 9450

XIV. But the same values are expressed more conveniently in this way: Translation of E368 211

m + 1 A1 = ·1 2 m + 2 m − 1 B1 = 1 + 3 · 3 4 m + 3 m − 1 m − 1 m − 2 C1 = 1 + 10 · + 15 · · 4 5 5 6 m + 4 m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 D1 = 1 + 25 · + 105 · · + 105 · · · 5 6 6 7 6 7 8 m + 5 m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 E1 = 1 + 56 · + 490 · · + 1260 · · · 6 7 7 8 7 8 9 m − 1 m − 2 m − 3 m − 4 + 945 · · · · 7 8 9 10 m + 6 m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 F1 = 1 + 119 · + 1918 · · + 9450 · · · 7 8 8 9 8 9 10 m − 1 m − 2 m − 3 m − 4 + 17325 · · · · 8 9 10 11 m − 1 m − 2 m − 3 m − 4 m − 5 + 10395 · · · · · . 8 9 10 11 12 To see the law of this progression more easily, in general let us set

m + µ − 1 m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 M1 = 1 + α · + β · · + γ · · · + etc. µ µ + 1 µ + 1 µ + 2 µ + 1 µ + 2 µ + 3 and the following one

m + µ m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 N1 = 1 + α1 · + β1 · + γ1 · · · + etc. , µ + 1 µ + 2 µ + 2 µ + 3 µ + 2 µ + 3 µ + 4 and these coefﬁcients are determined as follows by the preceding ones:

α1 = 2α + µ + 1 β1 = 3β + (µ + 2)α γ1 = 4γ + (µ + 3)β δ1 = 5δ + (µ + 4)γ ε1 = 6ε + (µ + 5)δ, whence these formulas can easily be continued arbitrarily far.

XV. Let us now substitute these values, and for the sum s of the propounded series, if n = m + λ, we will obtain the following expression: s (λ + 1)(λ + 2) ··· (λ + m) 212 Appendix

λm λ(λ − 1)m 3(m − 1) = xλ − xλ−1 + xλ−2 1 + 1 · 2 1 · 2 · 3 4 λ(λ − 1)(λ − 2)m m − 1 m − 1 m − 2 − xλ−3 1 + 10 · + 15 · · 1 · 2 · 3 · 4 5 5 6 λ(λ − 1)(λ − 2)(λ − 3)m m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 + xλ−4 1 + 25 · + 105 · · + 105 · · · 1 · 2 · 3 · 4 · 5 6 6 7 6 7 8 λ ··· (λ − 4)m m − 1 m − 1 m − 2 m − 1 m − 3 − xλ−5 1 + 56 · + 490 · · + 1260 · ··· 1 ··· 5 · 6 7 7 8 7 9 m − 1 m − 4 + 945 · ··· 7 10 λ ··· (λ − 5)m m − 1 m − 1 m − 2 m − 1 m − 3 + xλ−6 1 + 119 · + 1918 · · + 9450 · ··· 1 ··· 6 · 7 8 8 9 8 10 m − 1 m − 4 m − 1 m − 5 + 17325 · ··· + 10395 · ··· 8 8 8 12

etc. from this subtract the power

m λ λm λ(λ − 1)m2 λ(λ − 1)(λ − 2)m3 x − = xλ − xλ−1 + xλ−2 − xλ 2 2 1 · 2 · 4 1 · 2 · 3 · 8 λ ··· (λ − 3)m4 λ ··· (λ − 4)m5 λ ··· (λ − 5)m6 + xλ−4 − xλ−5 + xλ−6 − etc. 1 ··· 4 · 16 1 ··· 5 · 32 1 ··· 6 · 64

But here it conveniently happens that

15 4 105 5 945 6 10395 7 = , = , = , = , 5 · 6 8 6 · 7 · 8 16 7 · 8 · 9 · 10 32 8 · 9 · 10 · 11 · 12 64 the reason for which is obvious; therefore, the above expression, if it is expanded, takes the following form:

m λ λ(λ − 1)m 1 λ(λ − 1)(λ − 2)m m x − + xλ−2 · − xλ−3 · 2 1 · 2 · 3 4 1 · 2 · 3 · 4 2 λ ··· (λ − 3)m 5 5 1 + xλ−4 m2 + m − 1 ··· 4 · 5 8 48 24 λ ··· (λ − 4)m 5 5 1 − xλ−5 m3 + m2 − m 1 ··· 5 · 6 8 16 8 λ ··· (λ − 5)m 35 35 91 7 1 + xλ−6 m4 + m3 − m2 − + − etc. 1 ··· 6 · 7 64 64 576 96 30

m XVI. The power of x − 2 , more precisely λ(λ − 1)m m λ−2 x − 2 · 3 ··· 4 2 Translation of E368 213 is again detected to be contained in this expression; having separated this power, our expression will be:

m λ λ(λ − 1)m m λ−2 λ ··· (λ − 3)m 5 1 x − + x − + xλ−4 m − 2 2 · 3 · 4 2 1 ··· 4 · 5 48 24 λ ··· (λ − 4)m 5 1 − xλ−5 m2 − m 1 ··· 5 · 6 16 8 λ ··· (λ − 5)m 35 91 7 1 + xλ−6 m3 − m2 − m + − etc., 1 ··· 6 · 7 64 576 96 36 which still contains

λ ··· (λ − 3)m 5 1 m λ−4 m − x − 1 ··· 4 · 5 48 24 2 furthermore, there still is

λ ··· (λ − 5)m 35 7 1 xλ−6 m2 − m + , 1 ··· 6 · 7 576 96 36 whence, without any doubt, additionally this power enters:

λ ··· (λ − 5)m 35m2 − 42m + 16 m λ−6 + · x − . 1 ··· 6 · 7 576 2 Therefore, our expression will be of this nature:

s m λ λ(λ − 1) m m λ−2 = x − + · x − (λ + 1)(λ + 2) ··· (λ + m) 2 1 · 2 12 2 λ(λ − 1) ··· (λ − 3) m(5m − 2) m λ−4 λ(λ − 1) ··· (λ − 5) m(35m2 − 42m + 16) m λ−6 + · x − + · x − + etc. 1 · 2 ··· 4 240 2 1 · 2 ··· 6 4032 2

XVII. Therefore, lo and behold the extraordinary transformation of our propounded general series:

m m(m − 1) m(m − 1)(m − 2) s = xm+λ − (x − 1)m+λ + (x − 2)m+λ − (x − 3)m+λ + etc. 1 1 · 2 1 · 2 · 3 which, since found in such a long-winded way and using such intricate operations, seems so weird that a direct investigation will provide us with useful auxiliary tools for analysis: To investigate this transformation more easily, I will represent it in this way:

s m λ λ(λ − 1) m λ−2 = x − + P x − (λ + 1)(λ + 2) ··· (λ + m) 2 1 · 2 2 λ(λ − 1) ··· (λ − 3) m λ−4 λ(λ − 1) ··· (λ − 6) m λ−6 + Q x − + R x − 1 · 2 ··· 4 2 1 · 2 ··· 6 2 λ(λ − 1) ··· (λ − 7) m λ−8 λ(λ − 1) ··· (λ − 9) m λ−10 + S x − + T x − 1 · 2 ··· 7 2 1 · 2 ··· 10 2 etc., 214 Appendix for which expression up to this point I have found:

m P = 3 · 4 m(5m − 2) Q = 5 · 6 · 8 m(35mm − 42m + 16) R = 6 · 7 · 96 m(175m3 − 420m2 + 404m − 144) S = 34560 but a method to ﬁnd the values of these letters more quickly is desired.

XVIII. But here it is especially helpful to have noted that our series is transformed into another m one which is a power series in x − 2 , the exponents being λ, λ − 2, λ − 4 etc. continuously decreasing by two; but then the letters P, Q, R etc. depend only on the number m such that neither the exponent λ nor the quantity x enter it; furthermore, the preﬁxed coefﬁcients involve the number λ and follow the law of progression resulting from the expansion of the binomial. Having studied this form carefully, it becomes obvious that the values of the letters P, Q, R, S etc. can be found independently from the propounded series or from its transformed counterpart that m we gave in paragraph XV, whose law of progression is known as well, if we set x = 2 ; for, if one takes λ = 2,

s P = , (λ + 1)(λ + 2) ··· (λ + m) but, having set λ = 4,

s Q = (λ + 1)(λ + 2) ··· (λ + m) but, having set λ = 6,

s R = etc. (λ + 1)(λ + 2) ··· (λ + m)

XIX. Therefore, if we substitute the series found above in paragraph XV. for

s (λ + 1)(λ + 2) ··· (λ + m) here and, for the sake of brevity, we set: Translation of E368 215

A = 1 m − 1 B = 1 + 3 · 4 m − 1 m − 1 m − 2 C = 1 + 10 · + 15 · · 5 5 6 m − 1 m − 1 m − 2 m − 1 m − 2 m − 3 D = 1 + 25 · + 105 · · + 105 · · · 6 6 7 6 7 8 m − 1 m − 1 m − 2 m − 1 m − 3 m − 1 m − 4 E = 1 + 56 · + 490 · · + 1260 · ··· + 945 · ··· 7 7 8 7 9 7 10 m − 1 m − 1 m − 2 m − 1 m − 3 m − 1 m − 4 F = 1 + 119 · + 1918 · · + 9450 · ··· + 17325 · ··· 8 8 9 8 10 8 11 m − 1 m − 5 + 10395 · ··· 8 12 etc.

we obtain the following values

m2 m m m P = − 2A · · + B · 22 2 2 3 m4 m m3 m m2 m m m Q = − 4A · · + 6B · · − 4C · · + D · 24 2 23 3 22 4 2 5 m6 m m5 m m4 m m3 m m2 m m m R = − 6A · · + 15B · · − 20C · · + D · · − 6E · · + F · 26 2 25 3 24 4 23 5 22 6 2 7

etc.

whence it will be convenient to expand the values of those letters A, B, C, D etc., whence it results: 216 Appendix

A = 1

3 1 3 1 B = m + = m + 4 4 4 3 1 1 4 C = m2 + m = (mm + m) 2 2 8 5 5 5 1 5 1 2 D = m3 + m2 + m − = m3 + 2m2 + m − 16 8 48 24 16 3 15 3 5 5 1 6 10 5 2 E = m4 + m3 + m2 − m = m4 + m3 + m2 − m 16 8 16 8 32 3 3 3 7 35 35 91 7 1 F = m5 + m4 + m3 − m2 − m + 64 64 64 576 96 36

7 13 2 16 F = m5 + 5m4 + 5m3 − m2 − m + ; 64 9 3 63 but here, aside from the ﬁrst terms, no structure is seen.

m XX. That the transformed series is a power series in x − 2 , was based only on induction, but it can be shown to happen necessarily in this way. Since the propounded progression ends as it begins such that the last two terms will be

±m(x − m + 1)m+λ ∓ (x − m)m+λ, where the upper signs hold, if m is an odd number, the lower on the other hand, if m is even, let us assume that m is an even number (for, the same conclusion follows, if it was odd) and set m x − 2 = y, and it will be

+ + + 1 m λ m 1 m λ m(m − 1) 1 m λ 2s = + y + m − y + m − 1 + y + m − 2 − etc. 2 1 2 1 · 2 2 + + + 1 m λ m 1 m λ m(m − 1) 1 m λ + y − m − y − m − 1 + y − m − 2 − etc. 2 1 2 1 · 2 2 m and after the expansion into powers of y = x − 2 one ﬁnds: m m(m − 1) s = ym+λ 1 − + − etc. 1 1 · 2 m + λ m 2 m m 2 m(m − 1) m 2 + ym+λ−2 − − 1 + − 2 − etc. 2 2 1 2 1 · 2 2 m + λ m 4 m m 4 m(m − 1) m 4 + ym+λ−4 − − 1 + − 2 − etc. 4 2 1 2 1 · 2 2 etc. Translation of E368 217

But all these series vanish, until one gets to the one in which the exponents are m, and we know its sum to be = 1 · 2 · 3 ··· m; therefore, having omitted those, whose sum becomes zero, we will obtain:

m + λ m m m m m m(m − 1) m m s = yλ − − 1 + − 2 − etc. m 2 1 2 1 · 2 2 m + λ m m+2 m m m+2 m(m − 1) m m+2 + yλ−2 − − 1 + − 2 − etc. m + 2 2 1 2 1 · 2 2

etc. and thus it is manifest, what I tried to demonstrate, that this series descends in the powers yλ, yλ−2, yλ−4 etc.

XXI. Now let us attribute a form to the series similar to that we had in paragraph XVII., and it will be

s yλ m m m m m = − − 1 + etc. (λ + 1)(λ + 2) ··· (λ + m) 1 · 2 ··· m 2 1 2 1 · 2yλ−2 λ(λ − 1) m m+2 m m m+2 + · − − 1 + etc. 1 · 2 ··· (m + 2) 1 · 2 2 1 2 1 · 2 · 3 · 4yλ−4 λ(λ − 1) ··· (λ − 3) m m+4 + · − etc. 1 · 2 ··· (m + 4) 1 · 2 · 3 · 4 2

etc., whence the values of the letters P, Q, R etc. can be determined in a new way as follows:

1 m m+2 m m m+2 P = − − 1 + etc. 3 · 4 ··· (m + 2) 2 1 2 1 m m+4 m m m+4 Q = − − 1 + etc. 5 · 6 ··· (m + 4) 2 1 2 1 m m+6 m m m+6 R = − − 1 + etc. 7 · 8 ··· (m + 6) 2 1 2 1 m m+8 m m m+8 S = − − 1 + etc. 9 · 10 ··· (m + 8) 2 1 2

etc.

Here, certainly the summation of similar series is necessary; since these only involve the number m, our investigation is to be considered to be reduced to a simpler case. Furthermore, we realise just now that these letters depend only on the number m. 218 Appendix

XXII. But if we successively attribute here the deﬁnite values 1, 2, 3, 4, 5, 6 etc. to the letter m, we will obtain as many values for the letters P, Q, R, S etc., knowing which one can easily conclude their general forms. Thus, to ﬁnd the letter P, we will have

if m = 0, 1, 2, 3, 4, etc. 3 · 22P = 0, 1, 2, 3, 4, etc. diff 1, 1, 1, 1

· 2 = = m such that hence 3 2 P m and P 22·3 as before. Further, for the letter Q,

if m = 0, 1, 2, 3, 4, 5, 6 24 · 3 · 5Q = 0, 3, 16, 39, 72, 115, 168 diff. I. 3, 13, 23, 33, 43, 53 diff. II. 10, 10, 10, 10, 10, therefore, it will be

m(m − 1) 24 · 3 · 5Q + 3m + 10 + m(5m − 2) 1 · 2 and hence

m(5m − 2) Q = . 24 · 3 · 5 In like manner, for the letter R,

if m = 0, 1, 2, 3, 4, 5, 6 26 · 3 · 7R = 0, 3, 48, 205, 544, 1135, 2048 diff. I. 3, 45, 157, 339, 591, 913 diff. II. 42, 112, 182, 252, 322, diff. III. 70, 70, 70, 70, whence one concludes

35 26 · 3 · 7R = 3m + 21m(m − 1) + m(m − 1)(m − 2) 3 and

m(35m2 − 42m + 16) R = , 26 · 32 · 7 which same values we obtained above already; therefore, let us apply the same operation to he following letters. Translation of E368 219

XXIII. Therefore, for the letter S we will have:

if m = 0, 1, 2, 3, 4, 5, 6 28 · 5 · 9S = 0, 5, 256, 2013, 7936, 22085, 49920 diff. I. 5, 251, 1757, 5923, 14149, 27835 diff. II. 246, 1506, 4166, 8226, 13686, diff. III. 1260, 2660, 4060, 5460, diff. IV. 1400, 1400, 1400 whence

175 28 · 5 · 9S = 5m + 123m(m − 1) + 210m(m − 1)(m − 2) + m(m − 1)(m − 2)(m − 3) 3

Now further, for the letter T we will have:

if m = 0, 1, 2, 3, 4, 5, 6 210 · 3 · 11T = 0, 3, 512, 7665, 4680, 174255, 499968 diff. I. 3, 509, 7153, 38415, 128175, 325713 diff. II. 506, 6604, 31262, 89760, 197538, diff. III. 6138, 24618, 58498, 107778, diff. IV. 18480, 33880, 49280, diff. V. 15400, 15400 whence

210 · 3 · 11T = 3m + 253m(m − 1) + 1023m(m − 1)(m − 2)

+ 770m(m − 1)(m − 2)(m − 3)

385 + m(m − 1)(m − 2)(m − 3)(m − 4) 3 and

m(385m4 − 1540m3 + 2684m2 − 2288m + 768) T = . 210 · 9 · 11

XXIV. Now let us represent these values in such a way that the law of progression can be explored more easily: 220 Appendix

1m P = 12 1 · 3m 2 Q = m − 122 5 1 · 3 · 5m 6 16 R = m2 − m + 123 5 35 1 · 3 · 5 · 7m 12 404 144 S = m3 − m + m − 124 5 175 175 1 · 3 · 5 · 7 · 9m 20 244 208 768 T = m4 − m3 + m2 − m + 125 5 35 35 385 and here in the ﬁrst and second terms the law of progression is so manifest that the same can safely be assigned for all following letters, but in the remaining terms one can still not observe any law.

XXV. Therefore, to ﬁnd the value of the letter V, let us set

1 · 3 · 5 · 7 · 9 · 11m 30 V = m5 − m4 + αm3 − βm2 + γm − δ . 126 5 But from the general form

1 m m+12 m m+12 m(m − 1) m m+12 V = − − 1 + − 2 − etc. 13 · 14 ··· (m + 12) 2 2 1 · 2 2 we conclude that:

if it will be

1 5 · 7 · 11 m = 1, V = = ( − 5 + α − β + γ − δ) 212 · 13 212 · 33 1 5 · 7 · 11 m = 2, V = = ( − 64 + 8α − 4β + 2γ − δ) 7 · 13 211 · 33 597871 5 · 7 · 11 m = 3, V = = ( − 243 + 27α − 9β + 3γ − δ) 212 · 5 · 7 · 13 212 · 32 5461 5 · 7 · 11 m = 4, V = = ( − 512 + 64α − 16β + 4γ − δ) 22 · 5 · 7 · 13 210 · 33 5838647 25 · 7 · 11 m = 5, V = = ( − 625 + 125α − 25β + 5γ − δ) 212 · 7 · 13 212 · 33 63047 5 · 7 · 11 m = 6, V = = ( − 0 + 216α − 26β + 6γ − δ) 22 · 3 · 7 · 13 211 · 32 Therefore, let us form the following equations: Translation of E368 221

27 α − β + γ − δ = + 5 5 · 7 · 11 · 13 27 · 2048 8α − 4β + 2γ − δ = + 64 5 · 72 · 11 · 13 9 · 597871 27α − 9β + 3γ − δ = + 243 52 · 72 · 11 · 13 27 · 256 · 5461 64α − 16β + 4γ − δ = + 512 52 · 72 · 11 · 13 27 · 5838647 125α − 25β + 5γ − δ = + 625 52 · 72 · 11 · 13 3 · 512 · 63047 216α − 36β + 6γ − δ = + 0 5 · 72 · 11 · 13 Now the ﬁrst differences will look as follows:

27 · 157 7α − 3β + γ = + 59 5 · 72 · 11 9 · 43627 19α − 5β + γ = + 179 52 · 72 · 11 9 · 276629 37α − 7β + γ = + 269 52 · 72 · 11 27 · 341587 61α − 9β + γ = + 113 52 · 72 · 11 3 · 8373269 91α − 11β + γ = − 625 52 · 72 · 11 the second differences, divided by 2, on the other hand give

9 · 268 6α − β = + 60 52 · 7 9 · 1513 9α − β = + 45 52 · 7 9 · 4858 12α − β = − 78 52 · 7 3 · 34409 15α − β = − 369 52 · 7 Finally, the third differences, divided by 3, yield

3 · 249 3 · 669 3967 α = − 5 = − 41 = − 97, 5 · 7 5 · 7 5 · 7 which three equations give the same value 222 Appendix

572 4 · 11 · 13 α = = , 5 · 7 5 · 7

from which value the remaining ones are now deﬁned as follows:

6 · 572 9 · 268 12 · 1229 4248 8 · 9 · 59 β = − − 60 = − 60 = = 5 · 7 52 · 7 52 · 7 175 175 27 · 157 255968 γ = 3β − 7α + + 59 = 5 · 72 · 11 52 · 72 · 11 27 1061376 δ = α − β + γ − − 5 = . 5 · 7 · 11 · 13 52 · 72 · 11 · 13

XXVI. Thus, let us list up the values of the letters P, Q, R etc. found up to this point all at once

1m P = 12 1 · 3m 2 Q = m − 122 5 1 · 3 · 5m 6 16 R = m2 − m + 123 5 35 1 · 3 · 5 · 7m 12 404 144 S = m3 − m2 + m − 124 5 175 175 1 · 3 · 5 · 7 · 9m 20 244 208 768 T = m4 − m3 + m2 − m + 125 5 35 35 385 1 · 3 · 5 · 7 · 9 · 11m 30 572 4248 255968 1061376 V = m5 − m4 + m3 − m2 + m − . 126 5 35 175 13475 175175

From the ﬁrst terms I conclude that powers occur here, after the separation of which it seems the structure can be recognised more clearly: Translation of E368 223

1m P = 12 1 · 3m 21 Q = m − 122 5 1 · 3 · 5m 3 17 R = m − + 124 5 175 ! 1 · 3 · 5 · 7m 43 4 · 17 16 · 17 S = m − + m − 124 5 175 5 · 175 ! 1 · 3 · 5 · 7 · 9m 54 2 · 17 4 · 17 383 T = m − + m2 − m + 125 5 35 35 385 ! 1 · 3 · 5 · 7 · 9 · 11m 65 4 · 17 72 · 17 581296 78185568 V = m − + m3 − m2 + m − , 126 5 35 175 53 · 72 · 11 55 · 72 · 11 · 13 yes, it even seems that the second terms can be approximately contracted in this way so that it results:

1m P = · 1 12 1 · 3m 2 Q = m − 122 5 ! 1 · 3 · 5m 32 17 R = m − + 123 5 175 ! 1 · 3 · 5 · 7m 43 4 · 17 4 S = m − + m − 124 5 175 5 ! 1 · 3 · 5 · 7 · 9m 54 10 · 17 52 9 T = m − + m − + 125 5 175 5 385 ! 1 · 3 · 5 · 7 · 9 · 11m 65 20 · 17 63 15808 4672128 V = m − + m − + m − . 126 5 175 5 53 · 72 · 11 55 · 72 · 11 · 13

If we had not found the last value, it would seem that all these expressions are reduced to powers of this kind, what we now have to admit not to happen. Therefore, one must investigate the law of these letters from another source.

XXVIII. Therefore, let us rather represent each term of these formulas in this way: 224 Appendix

m P = 4 · 3 mm m Q = − 16 · 3 8 · 3 · 5 5m3 mm m R = − + 64 · 9 32 · 3 4 · 9 · 7 5 · 7m4 7m3 101m2 m S = − + − 256 · 27 64 · 9 64 · 27 · 5 16 · 3 · 5 5 · 7m5 5 · 7m4 61m3 13m2 m T = − + − + 1024 · 9 256 · 9 256 · 9 64 · 9 4 · 3 · 1 5 · 7 · 11m6 5 · 7 · 11m5 1573m4 649m3 7999m2 691m V = − + − + − , 4096 · 27 2048 · 9 1024 · 27 512 · 3 · 5 128 · 27 · 5 · 7 8 · 9 · 5 · 7 · 13 where we have already noticed the structure in the ﬁrst and second terms, but the last terms seemed to have no structure at all, until, having expanded the value of the letter V, the number 691 provided us with a criterion that the last terms contain the Bernoulli numbers. Therefore, let us denote the Bernoulli numbers by the letters α, β, γ, δ etc. such that

1 1 1 3 5 691 α = , β = , γ = , δ = , ε = , ζ = etc. 2 6 6 10 6 210 and, concerning these, let us note this law of progression:

1 α = 21 5 · 4α 2 β = − 22 · 1 · 2 · 3 23 7 · 6β 7 · 6 · 5 · 4α 3 γ = − + 22 · 1 · 2 · 3 24 · 1 · 2 ··· 5 25 9 · 8γ 9 · 8 · 7 · 6 · β 9 · 4α 4 δ = − + − 22 · 1 · 2 · 3 24 · 1 · 2 ··· 5 26 · 1 ··· 7 27 10 · 11δ 11 ··· 8γ 11 ··· 6β 11 ··· 4α 5 ε = − + − + 22 · 1 · 2 · 3 24 · 1 ··· 5 26 · 1 ··· 7 28 · 1 ··· 9 29 13 · 12ε 13 ··· 10δ 13 ··· 8γ 13 ··· 6β 13 ··· 4α 6 ζ = − + − + − 22 · 1 · 2 · 3 24 · 1 ··· 5 26 · 1 ··· 7 28 · 1 ··· 9 210 · 1 · 11 211

etc.

And the last terms of the letters P, Q, R, S etc. can represented in short form as follows

αm βm γm δm εm ζm , , , , , . 2 · 3 4 · 5 6 · 7 8 · 9 10 · 11 12 · 13 Translation of E368 225

XXIX. But to investigate how these letters P, Q, R, S etc. proceed, let us subtract a multiple of it from the preceding one such that the ﬁrst terms are cancelled, and since the letter O = 1 precedes those letters, we will have:

m P − O = 0 12 3m βm Q − P = − 12 4 · 5 5m mm γm m γm R − Q = − + = − P + 12 16 · 9 6 · 7 12 6 · 7 7m 7m3 3m2 δm S − R = − + − 12 128 · 9 16 · 5 8 · 9 9m 7m4 17m3 7m2 εm T − S = − + − + 12 128 · 9 64 · 3 · 5 8 · 9 · 5 10 · 11 11m 5 · 7 · 11m5 451m4 121m3 7159m2 ζm V − T = − + − + − . 2 2048 · 27 512 · 27 2048 · 27 · 5 128 · 27 · 5 · 7 12 · 13 Therefore, if we now consider these forms with more attention and, for the sake of brevity, set:

αm βm γm δm = α1, = β1, = γ1, = δ1 etc., 2 · 3 4 · 5 6 · 7 8 · 9 we will detect the following sufﬁciently simple law in our letters P, Q, R etc:

1 P − α1 = 0 1 3 3 · 2 · 1 Q − α1P + β1 = 0 1 1 · 2 · 3 5 5 · 4 · 3 5 · 4 · 3 · 2 · 1 R − α1Q + β1P − γ1 = 0 1 1 · 2 · 3 1 · 2 · 3 · 4 · 5 7 7 · 6 · 5 7 · 6 ··· 3 7 · 6 ··· 1 S − α1R + β1Q − γ1P + δ1 = 0 1 1 · 2 · 3 1 · 2 ··· 5 1 · 2 ··· 7 9 9 ··· 5 9 ··· 5 9 ··· 3 9 ··· 1 T − α1S + β1R − γ1Q + δ1P − ε1 = 0 1 1 ··· 3 1 ··· 5 1 ··· 7 1 ··· 9 11 11 ··· 9 11 ··· 7 11 ··· 5 11 ··· 3 11 ··· 1 V − α1T + β1S − γ1R + δ1Q − ε1P + ζ1 = 0 1 1 ··· 3 1 ··· 5 1 ··· 7 1 ··· 1 1 ··· 11

etc.

But these new letters α1, β1, γ1, δ1 etc. from the preceding ones follow this law: 226 Appendix

m α1 − = 0 22 · 3 3 · 2 m β1 − α1 + = 0 22 · 1 · 2 · 3 24 · 5 5 · 4 5 · 4 · 3 · 2 m γ1 − β1 + α1 − = 0 22 · 1 · 2 · 3 24 · 1 ··· 5 26 · 7 7 · 6 7 ··· 4 7 ··· 2 m δ1 − γ1 + β1 − α1 + = 0 22 · 1 ··· 3 24 · 1 ··· 5 26 · 1 ··· 7 28 · 9 9 · 8 9 ··· 6 9 ··· 4 9 ··· 2 m ε1 − δ1 + γ1 − β1 + α1 − = 0 22 · 1 ··· 3 24 · 1 ··· 5 26 · 1 ··· 7 281 ··· 9 210 · 11 Therefore, I now have to believe to have answered the question on that extraordinary series, that I have contemplated, completely, whence I will now present the answer in short form here.

Problem Having propounded this indeﬁnite progression:

m m(m − 1) m(m − 1)(m − 2) s = xm+λ − (x − 1)m+λ + (x − 2)m+λ − (x − 3)m+λ + etc. 1 1 · 2 1 · 2 · 3 to assign its sum, if λ was an arbitrary positive integer.

Solution Let the letters A, B, C, D etc. denote the Bernoulli numbers such that:

1 1 1 3 5 A = , B = , C = , D = , E = , 2 6 6 10 6 691 35 3617 43867 F = , G = , H = , I = , 210 2 30 42 1222277 854513 K = , L = , 110 6 1181820455 76977927 M = , N = , 546 2 23749461029 8615841276005 Q = , P = , 30 462 84802531453387 90219075042845 Q = , R = 170 6

etc.

I observed that these numbers proceed in such a way that Translation of E368 227

1 A = 2 4 A2 B = · 2 3 6 2AB C = · 2 3 8 2AC 8 · 7 · 6 B2 D = · + · 2 3 2 · 3 · 4 5 10 2AD 10 · 9 · 8 2BC E = · + · 2 3 2 · 3 · 4 5 12 2AE 12 · 11 · 10 2BD 12 · 11 · 10 · 9 · 8 CC F = · + · + · 2 3 2 · 3 · 4 5 2 · 3 · 4 · 5 · 6 7 14 2AF 14 · 13 · 12 2BE 14 · 13 · 12 · 11 · 10 2CD G = · + · + · 2 3 2 · 3 · 4 5 2 · 3 · 4 · 5 · 6 7

etc.

Hence now ﬁnd numbers P, Q, R, S etc. that

1Am P = 1 · 2 · 3 3Am 3 · 2 · 1Bm Q = P − 1 · 2 · 3 1 · 2 · 3 · 4 · 5 5Am 5 · 4 · 3 · Bm 5 · 4 ··· 1Cm R = Q − P + 1 · 2 · 3 1 · 2 · 3 · 4 · 5 1 · 2 ··· 7 7Am 7 · 6 · 5 · Bm 7 ··· 3Cm 7 ··· 1D S = R − Q + P − 1 · 2 · 3 1 · 2 ··· 5 1 · 2 ··· 7 1 · 2 ··· 9

etc.

where the law of progression also is perspicuous. Having found this series, the sum s in question will be expressed in this way: 228 Appendix

s m λ λ(λ − 1) m λ−2 = x − + P x − (λ + 1)(λ + 2) ··· (λ + m) 2 1 · 2 2 λ ··· (λ − 3) m λ−4 + Q x − 1 ··· 4 2 λ ··· (λ − 5) m λ−6 + R x − 1 ··· 6 2 λ ··· (λ − 7) m λ−8 + S x − 1 ··· 8 2

etc. where one should note, if the number m is not an integer, that the value of the product

(λ + 1)(λ + 2) ··· (λ + m) can be deﬁned through other artiﬁces explained on another occasion.

Corollary 1 If instead of the Bernoulli numbers we want to introduce the related ones, which I used for the 1 sums of the powers of the reciprocals, and denote them by the letters A, B, C, D etc. that A = 6 , 1 1 1 1 B = 90 , C = 945 , D = 9450 , E = 93555 , since these numbers depend on the ﬁrst in such a way that 1 · 2 · 3 1 ··· 5 1 ··· 7 A = A, B = B, C = C etc., 21 23 25 but are connected to each other in such a way that:

5B = 2A2, 7C = 4AB, 9D = 4AC + 2BB,

11E = 4AD + 4BC, 13F = 4AE + 4BD + 2CC etc., then from these numbers the letters P, Q, R, S etc. will be determined as follows:

1Am P = 2 3Am 3 · 2 · 1Bm Q = P − 2 23 5Am 5 · 4 · 3Bm 5 · 4 ··· 1Cm R = Q − P + 2 23 25 7Am 7 · 6 · 5Bm 7 · 6 ··· 3Cm 7 · 6 ··· 1Dm S = R − Q + P − 2 23 25 27

etc. Translation of E368 229

Corollary 2

If for the various values of the number λ we indicate the sum of the propounded progression by the sign R (λ) and now for λ successively write the numbers 0, 1, 2, 3, 4 etc., for these cases the R R R R m sums (0), (1), (2), (3) etc., for the sake of brevity having put x − 2 = y, will be expressed in the following way:

R (0) = 1 1 · 2 ··· m R (1) = y 2 · 3 ··· (m + 1) R (2) = y2 + P 3 · 4 ··· (m + 2) R (3) = y3 + 3Py 4 · 5 ··· (m + 3) R (4) = y4 + 6Py2 + Q 5 · 6 ··· (m + 4) R (5) = y5 + 10Py3 + 5Qy 6 · 7 ··· (m + 5) R (6) = y6 + 15Py4 + 15Qy2 + R 7 · 8 ··· (m + 6)

etc.

Corollary 3

Therefore, these sums can be deﬁned from the preceding ones as follows 230 Appendix

m + 1 R (1) = y R (0) 1 m + 2 (m + 2)(m + 1) R (2) = y R (1) + mA R (0) 2 2 · 2 m + 3 (m + 3)(m + 2) R (3) = y R (2) + mA R (1) 3 2 · 3 m + 4 (m + 4)(m + 3) (m + 4) ··· (m + 1) R (4) = y R (3) + mA R (2) − mB R (0) 4 2 · 4 23 · 4 m + 5 (m + 5)(m + 4) (m + 5) ··· (m + 2) R (5) = y R (4) + mA R (3) − mB R (1) 5 2 · 5 23 · 5 m + 6 (m + 6)(m + 5) (m + 6) ··· (m + 3) (m + 6) ··· (m + 1) R (6) = y R (5) + mA R (4) − mB R (2) + mC R (0) 6 2 · 6 23 · 5 25 · 6 m + 7 (m + 7)(m + 6) (m + 7) ··· (m + 4) (m + 7) ··· (m + 2) R (7) = y R (6) + mA R (5) − mB R (3) + mC R (1) 7 2 · 7 23 · 7 25 · 7 m + 8 (m + 8)(m + 7) (m + 8) ··· (m + 5) (m + 8) ··· (m + 3) R (8) = y R (7) + mA R (6) − mB R (4) + mC R (2) 8 2 · 8 23 · 8 25 · 8 (m + 8) ··· (m + 1) − mD R (0), 27 · 8 which law will become obvious soon to the attentive reader.

Conclusion Now there will be not much difﬁculty to generalise this task quite substantially such that, if ϕ : x denotes an arbitrary function of x, we can assign the sum of this series

m(m − 1) m(m − 1)(m − 2) s = ϕ : x − mϕ : (x − 1) + ϕ : (x − 2) − ϕ : (x − 3). 1 · 2 1 · 2 · 3 For, it is perspicuous that this form exhibits the difference of order m of this progression

ϕ : x, ϕ : (x − 1), ϕ : (x − 2), ϕ : (x − 3) etc. For, from that, what I covered in Institutiones Calculi Differentialis pag. 343 8, if we set ϕ : x = y, one concludes that the differences of the respective orders are:

dy ddy d3y d4y d5y ∆y = − + − + − etc. dx 2dx2 2 · 3dx3 2 · 3 · 4dx4 2 ··· 5dx5 d2y 3d3y 7d4y 15d5y 31d6y ∆2y = − + − + − etc. dx2 3dx3 3 · 4dx4 3 · 4 · 5dx5 3 ··· 6dx6 d3y 6d4y 25d5y 90d6y 301d7y ∆3y = − + − + − etc. dx3 4dx4 4 · 5dx5 4 · 5 · 6dx6 4 ··· 7dx7 d4y 10d5y 65d6y 350d7y 1701d8y ∆4y = − + − + − etc. dx4 5dx5 5 · 6dx6 5 · 6 · 7dx7 5 ··· 8dx8

etc.,

8p. 264 in the Opera Onmia Version, i.e. Volume 10 of Series 1 Translation of E368 231 since which coefﬁcients are those we had above in paragraph IV, in like manner, we will understand that the difference of order m or ∆my, i.e. the sum of the propounded series, will be

dmy A1dm+1y B1dm+2y C1dm+3y s = − + − + etc., dxm (m + 1)dxm+1 (m + 1)(m + 2)dxm+2 (m + 1) ··· (m + 3)dxm+3 which coefﬁcients A1, B1, C1 etc. I determined above in paragraph XIII. Therefore, it will be

A1 m = m + 1 2 B1 m 3m(m − 1) = + (m + 1)(m + 2) 1 · 2 · 3 1 · 2 · 3 · 4 C1 m 10m(m − 1) 15m(m − 1)(m − 2) = + + (m + 1) ··· (m + 3) 1 · 2 · 3 · 4 1 · 2 ··· 5 1 · 2 ··· 6 D1 m 25m(m − 1) 105m(m − 1)(m − 2) 105m(m − 1)(m − 2)(m − 3) = + + + (m + 1) ··· (m + 4) 1 · 2 ··· 5 1 · 2 ··· 6 1 · 2 ··· 7 1 · 2 ··· 8

etc.

m Therefore, if we set ϕ : x − = v, such that v results from y, if one writes x − m instead of y, 2 2 it will obviously be

dmv dmy mdm+1y m2dm+2y = − + − etc.; dxm dxm 2dxm+1 2 · 4dxm+2 if this equation is subtracted from that, one will have to do exactly the same calculations as above. Hence introducing the same letters P, Q, R, S etc., which we deﬁned above, we will obtain the following value of the sum s:

dmv Pdm+2v Qdm+4v Rdm+6v Sdm+8v s = + + + + + etc. dxm 1 · 2dxm+2 1 · 2 ··· 4dxm+4 1 · 2 ··· 6dxm+6 1 · 2 ··· 8dxm+8 and hence, if one takes

m m+λ y = ϕ : x = xm+λ and v = x − , 2 manifestly the same summation we found before results, and thus the whole task reduces to the letters P, Q, R, S etc., whose nature I derived from the Bernoulli numbers above. Hence it follows immediately , what has been less obvious before, that, if in the function y or v the number of dimensions was smaller than the exponent m, which number certainly has to be a positive integer, then all differentials of order m and higher vanish and the sum s will be = 0. Further, hence there is a clearer way to ﬁnd the values of the letters P,Q, R, S etc. For, since, having set

dmy αdm+1y βdm+2y γdm+3y s = − + − + etc., dxm dxm+1 dxm+2 dxm+3 we have 232 Appendix

m α = 1 · 2 m 3m(m − 1) β = + 1 · 2 · 3 1 · 2 · 3 · 4 m 10m(m − 1) 15m ··· (m − 2) γ = + + 1 ··· 4 1 ··· 5 1 ··· 6 m 25m(m − 1) 105m ··· (m − 2) 105m ··· (m − 3) δ = + + + 1 ··· 5 1 ··· 6 1 ··· 7 1 ··· 8

etc.,

m m but the function y results from the function v = ϕ : x − 2 , if in it instead of x one writes x + 2 , in general it will be

dny dnv m dn+1v m2 dn+2v m3 dn+3v = + · + · + · + etc., dxn dxn 2 dxn+1 2 · 4 dxn+2 2 · 4 · 6 dxn+3 whence, if one substitutes the differentials of v for those of y, it will be

dnv m dn+1v m2 m dn+2v m3 m2 m dn+3v s = + − α + − α + β + − α + β − γ + etc. dxn 2 dxn+1 2 · 4 2 dxn+2 2 · 4 · 6 2 · 4 2 dxn+3

and so we will have:

m − α = 0 2 m2 m P − α + β = 2 · 4 2 1 · 2 m3 m2 m − α + β − γ = 0 2 · 4 · 6 2 · 4 2 m4 m3 m2 m Q − α + β − γ + δ = 2 · 4 · 6 · 8 2 · 4 · 6 2 · 4 2 1 · 2 · 3 · 4 m5 m4 m3 m2 m − α + β − γ + δ − ε = 0 1 · 2 · 3 · 4 · 5 2 · 4 · 6 · 8 2 · 4 · 6 2 · 4 2

etc.; for, one easily sees that these expressions must vanish alternately. Appendix IV

Translation of E421 - Expansion of the R f − m integral x 1dx(log x) n having extended the integration from the value x = 0 to x = 1

Theorem 1

§1 If n denotes a positive integer and the integral Z x f −1dx(1 − xg)n is extended from the value x = 0 to x = 1, the value of the integral will be

gn 1 · 2 · 3 ··· n = · . f ( f + g)( f + 2g)( f + 3g) ··· ( f + ng)

Proof

It is known that the integral R x f −1dx(1 − xg)m can in general be reduced to this one R x f −1dx(1 − xg)m−1, since it is possible to deﬁne constant quantities A and B in such a way that Z Z x f −1dx(1 − xg)m = A x f −1dx(1 − xg)m−1 + Bx f (1 − xg)m; by differentiating, this equation results

x f −1dx(1 − xg)m = Ax f −1dx(1 − xg)m−1 + B f x f −1dx(1 − xg)m − Bmgx f +g−1dx(1 − xg)m−1, which divided by x f −1dx(1 − xg)m−1 gives

1 − xg = A + B f (1 − xg) − Bmgxg or

233 234 Appendix

1 − xg = A − Bmg + B( f + mg)(1 − xg); in order for this equation to hold, it is necessary that

1 = B( f + mg) and A = Bmg, whence we conclude

1 mg B = and A = . f + mg f + mg Therefore, we will have the following general reduction

Z mg Z 1 x f −1dx(1 − xg)m = x f −1dx(1 − xg)m−1 + x f (1 − xg)m; f + mg f + mg because it vanishes for x = 0, if f > 0, of course, the addition of a constant is not necessary. Hence having extended both integrals to x = 1, the last absolute part vanishes and for the case x = 1 it will be Z mg Z x f −1dx(1 − xg)m = x f −1dx(1 − xg)m−1. f + mg Since for m = 1

Z 1 1 x f −1dx(1 − xg)0 = x f = , f f having put x = 1, we obtain the following values for the same case x = 1

Z g 1 x f −1dx(1 − xg)1 = · , f f + g Z g2 1 2 x f −1dx(1 − xg)2 = · · , f f + g f + 2g Z g3 1 2 3 x f −1dx(1 − xg)3 = · · · f f + g f + 2g f + 3g and hence we conclude that for any positive integer n it will be

Z gn 1 2 3 n x f −1dx(1 − xg)n = · · · ··· , f f + g f + 2g f + 3g f + ng if only the numbers f and g are positive.

Corollary 1 §2 Vice versa, the value of a product of this kind, formed from an arbitrary amount of factors, can be expressed by an integral so that

1 · 2 · 3 ··· n f Z = x f −1dx(1 − xg)n ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) gn having extended this integral from the value x = 0 to x = 1. Translation of E421 235

Corollary 2 §3 Therefore, if one considers a progression of this kind

1 1 · 2 1 · 2 · 3 1 · 2 · 3 · 4 , , , etc., f + g ( f + g)( f + 2g) ( f + g)( f + 2g)( f + 3g) ( f + g)( f + 2g)( f + 3g)( f + 4g) its general term corresponding to the indeﬁnite index n is conveniently represented by this integral f R f −1 g n gn x dx(1 − x ) ; and using this formula, the progression and its terms corresponding to fractional indices can be exhibited.

Corollary 3 §4 If we write n − 1 instead of n, we will have

1 · 2 · 3 ··· (n − 1) f Z = x f −1dx(1 − xg)n−1; ( f + g)( f + 2g)( f + 3g) ··· ( f + (n − 1)g) gn−1

n multiplication by f +ng yields 1 · 2 · 3 ··· n f · ng Z = x f −1dx(1 − xg)n−1. ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) gn( f + ng)

Scholium 1 §5 It would have been possible to derive this last formula immediately from the preceding one, since we just proved that Z ng Z x f −1dx(1 − xg)n = x f −1dx(1 − xg)n−1, f + ng if both integrals are extended from the value x = 0 to x = 1; this is to be kept in mind for all the integrals in everything what follows. Furthermore, it is to be noted that the quantities f and g are positive, a condition that was used in the proof, of course. Concerning the number n, if it denotes the index of a certain term of the progression (§3), that index can also be negative, because all terms, also those corresponding to negative indices, of the progression are considered to be exhibited by the given integral formula. Nevertheless, it is to be noted that this reduction Z mg Z x f −1dx(1 − xg)m = x f −1dx(1 − xg)m−1 f + mg 1 f g m is only true, if m > 0, because otherwise the algebraic part f +mg x (1 − x ) would not vanish for x = 1.

Scholium 2 §6 I already studied series of this kind, which can be called transcendental, because the terms corresponding to fractional indices are transcendental quantities, in Comment. acad. sc.Petrop., book 5 in more detail1; therefore, I will not investigate those progressions here again but focus

1Euler refers to his paper "’De progressionibus transcendentibus seu quarum termini generales algebraice dari nequeunt."’ This is paper E19 in the Eneström-Index 236 Appendix on the remarkable comparisons of the integral formulas that can be derived from it. After I had shown that the value of the indeﬁnite product 1 · 2 · 3 ··· n is expressed by the integral formula R 1 n dx log x extended from x = 0 to x = 1, which, if n is a positive integer, is manifest by direct integration, I examined the cases, in which a fractional number is taken for n; in these cases it is indeed not obvious at all, to which kind of transcendental quantities these terms are to be referred. But, using a singular artiﬁce, I reduced the same terms to more familiar quadratures; therefore, this seems to be worth of one’s while to consider it with all eagerness.

Problem 1

§7 Since it was demonstrated that

1 · 2 · 3 ··· n f Z = x f −1dx(1 − xg)n, ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) gn having extended the integral from x = 0 to x = 1, to assign the value of the same product in the case g = 0 by means of an integral.

Solution

Having put g = 0 in the integral, the term (1 − xg)n vanishes, but at the same time also the denominator gn vanishes, whence the question reduces to the task to deﬁne the value of the (1−xg)n fraction gn in the case g = 0, in which both the numerator and the denominator vanishes. Therefore, let us consider g as an inﬁnitely small quantity, and because xg = eg log x, it will be g g n n n n 1 n x = 1 + g log x and hence (1 − x ) = g (− log x) = g log x ; hence our integral becomes R f −1 1 n f x dx log x for this case so that one now has this expression

1 · 2 · 3 ··· n Z 1 n = f x f −1dx log f n x or

Z 1 n 1 · 2 · 3 ··· n = f n+1 x f −1dx log . x

Corollary 1

R f −1 1 n §8 If n is a positive integer, the integration of the integral x dx log x succeeds and, having extended it from x = 0 to x = 1, indeed the product we found to be equal to it results. But if fractional numbers are taken for n, the same formula can be applied to interpolate this hypergeometric progression

1, 1 · 2, 1 · 2 · 3, 1 · 2 · 3 · 4, 1 · 2 · 3 · 4 · 5 etc. or

1, 2, 6, 24, 120, 720, 5040 etc. Translation of E421 237

Corollary 2 §9 If the expression just found is divided by the principal one, a product whose factors proceed in an arithmetic progression will emerge, namely

n R x f −1dx log 1 ( f + g)( f + 2g)( f + 3g) ··· ( f + ng) = f ngn x , R x f −1dx (1 − xg)n whose values can also be assigned, using the integral, if n is a fractional number.

Corollary 3 §10 Since Z ng Z x f −1dx(1 − xg)n = x f −1dx(1 − xg)n−1, f + ng in like manner, for the case g = 0 it will be

− Z 1 n n Z 1 n 1 x f −1dx log = x f −1dx log x f x and hence by those other integrals

− Z 1 n 1 1 · 2 · 3 ··· n = n f n x f −1dx log x and

n−1 R x f −1dx log 1 ( f + g)( f + 2g) ··· ( f + ng) = f n−1gn−1( f + ng) x . R x f −1dx(1 − xg)n−1

Scholium §11 Because we found that

Z 1 n 1 · 2 · 3 ··· n = f n+1 x f −1dx log , x it is plain that this integral does not depend on the value of the quantity f , which is also easily seen by putting x f = y, whence ﬁrst we ﬁnd

1 1 1 1 f x f −1dx = dy and log = − log x = − log y = log x f f y and therefore

1 n 1 n f n log = log x y such that

Z 1 n 1 · 2 · 3 ··· n = dy log , y 238 Appendix which expression results from the ﬁrst by putting f = 1. Therefore, for an interpolation of this R 1 n kind the whole task is reduced to the deﬁnition of the values of the integral dx log x for the cases, in which the exponent n is a fractional number. For example, if n = 1 , one has to assign q 2 √ R 1 1 the value of the formula dx log x , which value I already once showed to be = 2 π, while π denotes the circumference of the circle whose diameter is = 1; but for other fractional numbers I taught how to reduce its value to quadratures of algebraic curves of higher order. Because this R 1 n reduction is by no means obvious and is only valid, if the integration of the formula dx log x is extended from the value x = 0 to x = 1, it seems to be worth of one’s attention. But even though I already treated this subject once2, nevertheless, because I was led to the results in a rather non straight-forward way, I decided take on this subject here again and explain everything in more detail.

Theorem 2

§12 If the integrals are extended from the value x = 0 to x = 1 and n denotes a positive integer, it will be

R − − 1 · 2 · 3 ··· n 1 Z x f 1dx(1 − xg)n 1 = ng x f +ng−1dx(1 − xg)n−1 · , (n + 1)(n + 2)(n + 2) ··· 2n 2 R x f −1dx(1 − xg)2n−1 whatever positive numbers are taken for f and g.

Proof Because above (§4) we showed that

1 · 2 · 3 ··· n f · ng Z = x f −1dx(1 − xg)n−1, ( f + g)( f + 2g) ··· ( f + ng) gn( f + ng) if we write 2n instead of n, we will have

1 · 2 · 3 ··· 2n f · 2ng Z = x f −1dx(1 − xg)2n−1. ( f + g)( f + 2g) ··· ( f + 2ng) g2n( f + 2ng) Now divide the ﬁrst equation by the second one and this third one will result

R − − ( f + (n + 1)g)( f + (n + 2)g) ··· ( f + 2ng) gn( f + 2ng) x f 1dx(1 − xg)n 1 = · . (n + 1)(n + 2) ··· 2n 2( f + ng) R x f −1dx(1 − xg)2n−1 But if one writes f + ng instead of f in the ﬁrst equation, this fourth equation will result

1 · 2 · 3 ··· n ( f + ng)ng Z = x f +ng−1dx(1 − xg)n−1. ( f + (n + 1)g)( f + (n + 2)g) ··· ( f + 2ng) gn( f + 2ng) Multiply this fourth equation by the third and one will ﬁnd the equation to be demonstrated, namely

R − − 1 · 2 · 3 ··· n 1 Z x f 1dx(1 − xg)n 1 = ng x f +ng−1dx(1 − xg)n−1 · . (n + 1)(n + 2)(n + 3) ··· 2n 2 R x f −1dx(1 − xg)2n−1 2Euler considered this expression also in E19 mentioned already in the footnote above. Translation of E421 239

Corollary 1

§13 If one sets f = n and g = 1 in the ﬁrst equation, the same product will result, of course

1 · 2 · 3 ··· n 1 Z = n xn−1dx(1 − xg)n−1; (n + 1)(n + 2) ··· 2n 2 having compared this equation to the one mentioned above we obtain

R xn−1dx(1 − x)n−1 R x f −1dx(1 − xg)n−1 = . g R x f +ng−1dx(1 − xg)n−1 R x f −1dx(1 − xg)2n−1

Corollary 2

§14 If we write xg instead of x in that equation, it will be

1 · 2 · 3 ··· n 1 Z = ng xng−1dx(1 − xg)n−1 (n + 1)(n + 2) ··· 2n 2 such that we ﬁnd this comparison of the following integral formulas

Z Z R x f −1dx(1 − xg)n−1 xng−1dx(1 − xg)n−1 = x f +ng−1dx(1 − xg)n−1 · . R x f −1dx(1 − xg)2n−1

Corollary 3

g m m 1 m §15 If we set g = 0 in the equation of the theorem, because of (1 − x ) = g log x , the powers of g will cancel each other and this equation will result

Z n−1 R f −1 1 n−1 1 · 2 · 3 ··· n 1 f −1 1 x dx log x = n x dx log · − , (n + 1)(n + 2) ··· 2n 2 x R f −1 1 2n 1 x dx log x whence we conclude

2 R − n−1 x f 1dx log 1 Z x ng−1 g n−1 − = g x dx(1 − x ) R f −1 1 2n 1 x dx log x or, because of

− Z 1 n 1 f Z 1 n x f −1dx log = x f −1dx log , x n x this equality

2 R f −1 1 n 2 f x dx log x Z · = g xng−1dx(1 − xg)n−1. n R f −1 1 2n x dx log x 240 Appendix

Corollary 4

m §16 Let us set f = 1, g = 2 and n = 2 here so that m is a positive integer, and, because of

Z 1 m dx log = 1 · 2 · 3 ··· m, x it will be

m 2 R 2 dx log 1 Z 4 x m−1 2 m −1 · = 2 x dx(1 − x ) 2 m 1 · 2 · 3 ··· m and hence

m Z 2 r Z 1 m − m −1 dx log = 1 · 2 · 3 ··· m · xm 1dx(1 − x2) 2 x 2 and by taking m = 1, because of

Z dx π √ = , 1 − xx 2 one will have

r s Z 1 1 Z dx 1√ dx log = √ = π. x 2 1 − xx 2

Scholium §17 So lo and behold this succinct proof of the theorem I proved some time ago3, which says that q √ R 1 1 dx log x = 2 π; furthermore, note that I did not use an argument involving interpolations, which I had used back then. Here, it was deduced from this theorem I found here, which states that

2 R − n−1 x f 1dx log 1 Z x ng−1 g n−1 − = g x dx(1 − x ) . R f −1 1 2n 1 x dx log x But the principal theorem, from which this one is deduced, reads as follows

R x f −1dx(1 − xg)n−1 · R x f +ng−1dx(1 − xg)n−1 Z g = xn−1dx(1 − x)n−1; R x f −1dx(1 − xg)2n−1 for, each side, if it is actually calculated by an integration extended from x = 0 to x = 1, is equal to this product

1 · 2 · 3 ··· (n − 1) . (n + 1)(n + 2) ··· (2n − 1)

3Euler proved this theorem in [E19], Translation of E421 241

But if we want to give the one side a more general form involving a further-extending class of integrals, we can state the theorem in such a way that

R x f −1dx(1 − xg)n−1 · R x f +ng−1dx(1 − xg)n−1 Z g = k xnk−1dx(1 − xk)n−1; R x f −1dx(1 − xg)2n−1 and, if one takes g = 0 here,

2 R − n−1 x f 1dx log 1 Z x nk−1 k n−1 − = k x dx(1 − x ) . R f −1 1 2n 1 x dx log x Therefore, it has to be noted that this equality holds, whatever numbers are taken for f and g; in the case f = g, this is indeed clear, since

Z 1 − (1 − xg)n 1 xg−1dx(1 − xg)n−1 = = ; ng ng for, it will be Z Z 2g xng+g−1dx(1 − xg)n−1 = k xnk−1dx(1 − xk)n−1, and because

Z 1 Z xng+g−1dx(1 − xg)n−1 = xng−1dx(1 − xg)n−1, 2 the equality is obvious, because k can be taken arbitrarily. But in the same way we arrived at this theorem, it is possible to get to other similar ones.

Theorem 3

§18 If the following integrals are extended from the value x = 0 to x = 1 and n denotes any positive integer, it will be

R − − 1 · 2 · 3 ··· n 2 Z x f 1dx(1 − xg)2n 1 = ng x f +2ng−1dx(1 − xg)n−1 · , (2n + 1)(2n + 2) ··· 3n 3 R x f −1dx(1 − xg)3n−1 whatever positive numbers are taken for f and g.

Proof In the preceding theorem, we already saw that

1 · 2 · 3 ··· 2n f · 2ng Z = x f −1dx(1 − xg)2n−1; ( f + g)( f + 2g) ··· ( f + 2ng) g2n( f + 2ng) if, in like manner, we write 3n instead of n in the principal formula, we will have

1 · 2 · 3 ··· 3n f · 3ng Z = x f −1dx(1 − xg)3n−1; ( f + g)( f + 2g) ··· ( f + 3ng) g3n( f + 3ng) hence, dividing this equation by the ﬁrst one, we are led to 242 Appendix

R − − ( f + (2n + 1)g)( f + (2n + 2)g) ··· ( f + 3ng) 2gn( f + 3ng) x f 1dx(1 − xg)2n 1 = · . (2n + 1)(2n + 2) ··· 3n 3( f + 2ng) R x f −1dx(1 − xg)3n−1

But if we write f + 2gn instead of f in the principal equation (§4), we obtain this equation

1 · 2 · 3 ··· n ( f + 2ng)ng Z = x f +2ng−1dx(1 − xg)n−1. ( f + (2n + 1)g)( f + (2n + 2)g) ··· ( f + 3ng) gn( f + 3ng)

Now multiply this equation by the preceding and the equation to be proved will result

R − − 1 · 2 · 3 ··· n 2 Z x f 1dx(1 − xg)2n 1 = ng x f +2ng−1dx(1 − xg)n−1 · . (2n + 1)(2n + 2) ··· 3n 3 R x f −1dx(1 − xg)3n−1

Corollary 1 §19 We obtain the same value from the principal equation by putting f = 2n and g = 1 so that

1 · 2 · 3 ··· n 2 Z = n x2n−1dx(1 − x)n−1, (2n + 1)(2n + 2) ··· 3n 3 which integral formula, writing xk instead of x, is transformed into this one

2 Z nk x2nk−1dx(1 − xk)n−1 3 such that

Z R f −1 1 2n−1 Z f +2ng−1 g n−1 x dx log x 2nk−1 k n−1 g x dx(1 − x ) · − = k x dx(1 − x ) . R f −1 1 3n 1 x dx log x

Corollary 2

g 1 §20 If we set g = 0 here, because of 1 − x = g log x , we will have this equation

Z n−1 R f −1 1 2n−1 Z f −1 1 x dx log x 2nk−1 k n−1 x dx log · − = k x dx(1 − x ) ; x R f −1 1 3n 1 x dx log x because we had found before that

2 R − n−1 x f 1dx log 1 Z x nk−1 k n−1 − = k x dx(1 − x ) , R f −1 1 2n 1 x dx log x by multiplying both expressions by each other we will have this equation

3 R − n−1 x f 1dx log 1 Z Z x 2 nk−1 k n−1 2nk−1 k n−1 − = k x dx(1 − x ) · x dx(1 − x ) . R f −1 1 3n 1 x dx log x Translation of E421 243

Corollary 3

1 §21 Without any restriction one can put f = 1 here; because, then for n = 3 and k = 3 it will be

2 3 R 1 − 3 dx log x Z Z 3 − 2 3 − 2 = 9 dx(1 − x ) 3 · xdx(1 − x ) 3 R 1 0 dx log x and, because of

− 2 1 0 Z 1 3 Z 1 3 Z 1 dx log = 3 dx log and dx log = 1, x x x

1 !3 Z 3 Z Z 1 1 3 − 2 3 − 2 dx log = dx(1 − x ) 3 · xdx(1 − x ) 3 ; x 3

2 but then for n = 3 and k = 3 it will be

1 3 R 1 − 3 dx log x Z Z 3 − 1 3 3 − 1 = ( − ) 3 · ( − ) 3 R 1 9 xdx 1 x x dx 1 x dx log x or

2 !3 Z 3 Z Z 1 4 3 − 1 3 3 − 1 dx log = xdx(1 − x ) 3 · x dx(1 − x ) 3 . x 3

General Theorem

§22 If the following integrals are extended from the value x = 0 to x = 1 and n denotes a positive integer, it will be

R − − 1 · 2 · 3 ··· n λ Z x f 1dx(1 − xg)λn 1 = ng x f +λng−1dx(1 − xg)n−1 · , (λn + 1)(λn + 2) ··· (λ + 1)n λ + 1 R x f −1dx(1 − xg)(λ+1)n−1 whatever positive numbers are taken for the letters f and g.

Proof Since, as we showed above,

1 · 2 ··· n f · ng Z = x f −1dx(1 − xg)n−1, ( f + g)( f + 2g) ··· ( f + ng) gn( f + ng) if we write λn instead of n here at ﬁrst, but then (λ + 1)n instead of n, we will obtain these two equations

1 · 2 ··· λn f · λng Z = x f −1dx(1 − xg)λn−1, ( f + g)( f + 2g) ··· ( f + λng) gλn( f + λng) 244 Appendix

1 · 2 ··· (λ + 1)n f · (λ + 1)ng Z = x f −1dx(1 − xg)(λ+1)n−1; ( f + g)( f + 2g) ··· ( f + (λ + 1)ng) g(λ+1)n( f + (λ + 1)ng) dividing the ﬁrst equation by this one gives

R − − ( f + λng + g)( f + λng + 2g) ··· ( f + λng + ng) λ( f + λng + ng) x f 1dx(1 − xg)λn 1 = gn · . (λn + 1)(λn + 2) ··· (λn + n) (λ + 1)( f + λng) R x f −1dx(1 − xg)(λ+1)n−1

But if we write f + λng instead of f in the ﬁrst equation, we will obtain

1 · 2 ··· n ( f + λng)ng Z = x f +λng−1dx(1 − xg)n−1, ( f + λng + g)( f + λng + 2g) ··· ( f + λng + ng) gn( f + λng + ng) which two equations multiplied by each other produce the equation to be demonstrated

R − − 1 · 2 ··· n λng Z x f 1dx(1 − xg)λn 1 = x f +λng−1dx(1 − xg)n−1 · . (λn + 1)(λn + 2) ··· (λn + n) λ + 1 R x f −1dx(1 − xg)(λ+1)n−1

Corollary 1 §23 If we set f = λn and g = 1 in the principal equation, we will also ﬁnd

1 · 2 ··· n λn Z = xλn−1dx(1 − x)n−1, (λn + 1)(λn + 2) ··· (λn + n) λ + 1 which form writing xk instead of x changes into this one

λnk Z xλnk−1dx(1 − xk)n−1 λ + 1 such that we have this very far-extending theorem

Z R x f −1dx(1 − xg)λn−1 Z g x f +λng−1dx(1 − xg)n−1 · = k xλnk−1dx(1 − xk)n−1. R x f −1dx(1 − xg)λn+n−1

Corollary 2 §24 This theorem now holds, even if n is not an integer; because the number λ can be taken arbitrarily, let us even write m instead of λn and we will ﬁnd this theorem

R x f −1dx(1 − xg)m−1 k R xmk−1dx(1 − xk)n−1 = . R x f −1dx(1 − xg)m+n−1 g R x f +mg−1dx(1 − xg)n−1

Corollary 3

g 1 §25 If we set g = 0, because of 1 − x = g log x , that theorem will take this form

R f −1 1 m−1 R mk−1 k n−1 x dx log x k x dx(1 − x ) + − = − , R f −1 1 m n 1 R f −1 1 n 1 x dx log x x dx log x which is more conveniently represented as follows Translation of E421 245

R f −1 1 n−1 R f −1 1 m−1 Z x dx log x · x dx log x mk−1 k n−1 + − = k x dx(1 − x ) ; R f −1 1 m n 1 x dx log x here, it is evident that the numbers m and n can be permuted.

Scholium §26 Thus, we found two ways, along which many comparisons and relations of integrals formulas can be derived; the one way, found in § 24, contains integrals of this kind Z xp−1dx(1 − xg)q−1, which I already treated some time ago in my observations on the integrals4 Z p−1 n q −1 x dx(1 − x ) n , extended from the value x = 0 to x = 1; there I showed at ﬁrst that the letters p and q can be interchanged such that Z Z p−1 n q −1 q−1 n p −1 x dx(1 − x ) n = x dx(1 − x ) n , but then that

Z xp−1dx π p = ; n pπ (1 − x ) n n sin n But in particular, I demonstrated that

Z xp−1dx Z xp+q−1dx Z xp−1dx Z xp+r−1dx p · p = p · p ; n (1 − xn)n−q n (1 − xn)n−r n (1 − xn)n−r n (1 − xn)n−q the comparison found in § 24 is already contained in this equation such that nothing, which I have not already explained, can be deduced from this. Therefore, here I mainly attempt to follow the other way explained in § 25; since without any restriction one can take f = 1, our primary equation will be

n−1 m−1 R dx log 1 · R dx log 1 Z x x = k xmk−1dx(1 − xk)n−1, R 1 m+n−1 dx log x R 1 λ by means of which the values of the integral formula dx log x , if λ is not an integer, can be reduced to quadratures of algebraic curves; since, if λ is an integer, the integrals can be solved explicitly, because

Z 1 λ dx log = 1 · 2 · 3 ··· λ. x But the question of greatest importance concerns the cases, in which λ is a rational number. Therefore, I will deﬁne these here successively for some small denominators.

4 R p−1 n q −1 Euler again refers to his paper "’Observationes circa integralia formularum x dx(1 − x ) n posito post integrationem x = 1"’. This is paper E321 in the Eneström-Index. 246 Appendix

Problem 2

i R 1 2 §27 While i denotes a positive integer, to deﬁne the value of the integral dx log x , having extended the integration from x = 0 to x = 1.

Solution Let us put m = n in our general equation and it will be

2 R 1 n−1 dx log x Z = k xnk−1dx(1 − xk)n−1. R 1 2n−1 dx log x i Now let n − 1 = 2 and, because of 2n − 1 = i + 1, it will be − Z 1 2n 1 dx log = 1 · 2 · 3 ··· (i + 1); x now further take k = 2 so that nk − 1 = i + 1, and it will be

q 2 R 1 i dx log x Z i+1 2 i = 2 x dx(1 − x ) 2 1 · 2 · 3 ··· (i + 1) and hence

q i R dx log 1 r Z x i+1 2 i p = 2 x dx(1 − x ) 2 , 1 · 2 · 3 ··· (i + 1) where it is evidently sufﬁcient to take only odd numbers for i, because for the even ones the expansion is immediately obvious.

Corollary 1 §28 But all cases are easily reduced to i = 1 or even to i = −1; for, if i + 1 is not a negative number, the reduction we found holds. For this case it will therefore be s Z dx Z dx √ q = 2 √ = π, 1 1 − xx log x because of R √ dx = π . 1−xx 2

Corollary 2 §29 But having covered these principal cases, because of

− Z 1 n Z 1 n 1 dx log = n dx log , x x we will have Translation of E421 247

r 3 Z 1 1√ Z 1 2 1 · 3√ dx log = π, dx log = π, x 2 x 2 · 2 and in general

2n+1 Z 1 2 1 3 5 7 2n + 1√ dx log = · · · ··· π. x 2 2 2 2 2

Problem 3

i R 1 3 −1 §30 While i denotes a positive integer, to deﬁne the value of the integral dx log x , having extended the integration from x = 0 to x = 1, of course.

Solution Let us start from the equation of the preceding problem

2 R 1 n−1 dx log x Z = k xnk−1dx(1 − xk)n−1 R 1 2n−1 dx log x and let us set m = 2n in the general formula such that one has

n−1 2n−1 R dx log 1 · R dx log 1 Z x x = k x2nk−1dx(1 − xk)n−1, R 1 3n−1 dx log x and by multiplying these two equations we obtain

3 R 1 n−1 dx log x Z Z = kk xnk−1dx(1 − xk)n−1 · x2nk−1dx(1 − xk)n−1. R 1 3n−1 dx log x i Now just set n = 3 here such that

− Z 1 i 1 dx log = 1 · 2 · 3 ··· (i − 1), x and take k = 3 and this equation will result

q 3 R 3 1 i−3 dx log x Z q Z q = 9 xi−1dx 3 (1 − x3)i−3 · x2i−1dx 3 (1 − x3)i−3, 1 · 2 · 3 ··· (i − 1) whence we conclude q R 3 i−3 s dx log 1 Z i−1 Z 2i−1 x 3 x dx x dx p = 9 p · p . 1 · 2 · 3 ··· (i − 1) 3 (1 − x3)3−i 3 (1 − x3)3−i 248 Appendix

Corollary 1 §31 Here, two principal cases occur, on which all remaining ones depend, namely the cases i = 1 and i = 2; for these cases

s Z dx Z dx Z xdx I. = 3 9 · , q p3 p3 3 1 2 (1 − x3)2 (1 − x3)2 log x s Z dx Z dx Z x3dx = 3 √ · √ II. q 9 3 3 ; 3 1 1 − x3 1 − x3 log x the last formula, because of

Z x3dx 1 Z dx √ = √ , 3 1 − x3 3 3 1 − x3 can be transformed into this one s Z dx Z dx Z xdx = 3 √ · √ q 3 3 3 1 1 − x3 1 − x3 log x

Corollary 2 §32 If, for the sake of brevity, as in my observations mentioned before5 we set

Z xp−1dx p p = 3 (1 − x3)3−q q and, as we did it there, for this class also set

2 π = π = α, 1 3 sin 3 but then put

1 Z dx = p = A, 1 3 (1 − x3)2 it will be

s Z √ dx 3 1 2 3 I. q = 9 = 9αA, 3 1 2 1 1 log x s Z r dx 3 1 2 3 3αα II. q = 3 = . 3 1 1 2 2 A log x 5Euler refers to his paper E321 again. Translation of E421 249

Corollary 3 §33 Therefore, for the ﬁrst case we will have s Z −2 √ Z r √ 3 1 3 1 1 3 dx log = 9αA, dx 3 log = 9αA x x 3 and s Z 3n+1 √ 3 1 1 4 7 3n + 1 3 dx log = · · ··· 9αA, x 3 3 3 3 but for the other case, on the other hand, we will ﬁnd s s Z −1 r Z 2 r 3 1 3αα 3 1 2 3αα dx log = 3 , dx log = 3 x A x 3 A and s Z 3n−1 r 3 1 2 5 8 3n − 1 3αα dx log = · · ··· 3 . x 3 3 3 3 A

Problem 4

i R 1 4 −1 §34 While i denotes a positive integer, to deﬁne the value of the integral dx log x , having extended the integration from x = 0 to x = 1.

Solution In the solution of the preceding problem, we were led to this equation

n−13 R 1 − − dx log x Z xnk 1dx Z x2nk 1dx = kk · ; R 1 3n−1 (1 − xk)1−n (1 − xk)1−n dx log x but the general formula, setting m = 3n in it, yields

R n−1 R 3n−1 dx log 1 · dx log 1 Z x3nk−1dx x x = k ; R 1 4n−1 (1 − xk)1−n dx log x combining these formulas we obtain

n−14 R 1 − − − dx log x Z xnk 1dx Z x2nk 1dx Z x3nk 1dx = k3 · · . R 1 4n−1 (1 − xk)1−n (1 − xk)1−n (1 − xk)1−n dx log x i Let n = 4 and take k = 4 and it will be

i R 4 −1 s dx log 1 Z xi−1dx Z x2i−1dx Z x3i−1dx x 4 3 p = 4 p · p · p . 4 1 · 2 · 3 ··· (i − 1) 4 (1 − x4)4−i 4 (1 − x4)4−i 4 (1 − x4)4−i 250 Appendix

Corollary 1 §35 So if i = 1, we will have this equation

s s Z −3 Z dx Z xdx Z x2dx 4 1 4 3 dx log = 4 p · p · p ; x 4 (1 − x4)3 4 (1 − x4)3 4 (1 − x4)3 if this expression is denoted by the letter P, it will be in general s Z 4n−3 4 1 1 5 9 4n − 3 dx log = · · ··· P. x 4 4 4 4

Corollary 2 §36 For the other principal case, let us take i = 3 and it will be s s Z −1 Z 2 Z 5 Z 8 4 1 x dx x dx x dx dx log = 4 2 · 43 √ · √ · √ x 4 1 − x4 4 1 − x4 4 1 − x4 or, after some simpliﬁcation, s s Z −1 Z Z Z 4 1 xxdx xdx dx dx log = 4 8 √ · √ · √ ; x 4 1 − x4 4 1 − x4 4 1 − x4 if this expression is denoted by the letter Q, in general, it will be s Z 4n−1 4 1 3 7 11 4n − 1 dx log = · · ··· Q. x 4 4 4 4

Scholium

p−1 §37 If we indicate the integral formula R √x dx by the sign p , in general, the solution will 4 (1−x4)4−q q be as follows s s Z i−4 4 1 i 2i 3i dx log = 4 1 · 2 · 3 ··· (i − 1)43 x i i i and for the two cases expanded before s s 1 2 3 3 2 1 P = 4 43 and Q = 4 8 . 1 1 1 3 3 3 Now for the formulas depending on the circle, let us set

3 π 2 π = = α and = = β, 1 4 sin π 2 2π 4 4 sin 4 but for the transcendental ones of higher order let Translation of E421 251

2 Z xdx Z dx = = √ = p 2 A, 1 4 (1 − x4)3 1 − x4 on which all remaining ones depend; hence we will ﬁnd

r r αα 4 1 P = 4 43 AA and Q = 4ααβ , β AA whence it is clear that π PQ = 4α = π . sin 4 But because α = √π and β = π , it will be 2 2 4

√ r 3 4 4 π P 4A P = 32πAA and Q = and = √ . 8AA Q π

Problem 5 q R 5 1 i−5 §38 While i denotes a positive integer, to deﬁne the value of the integral dx log x , having extended the integration from x = 0 to x = 1, of course.

Solution From the preceding solutions it is already perspicuous that for this case one will obtain this formula at the end

q R 5 i−5 s dx log 1 Z xi−1dx Z x2i−1dx Z x3i−1dx Z x4i−1dx x 5 4 p = 5 p · p · p · p , 5 1 · 2 · 3 ··· (i − 1) 5 (1 − x5)5−i 5 (1 − x5)5−i 5 (1 − x5)5−i 5 (1 − x5)5−i which integral formulas belong to the fifth class introduced in my dissertation mentioned above6. p−1 Hence, if in the same way as it was done there the sign p denotes this formula R √x dx , q 5 (1−x5)5−q the value in question can be more conveniently expressed in such a way that s s Z i−5 5 1 i 2i 3i 4i dx log = 5 1 · 2 · 3 ··· (i − 1)54 ; x i i i i here it indeed suffices to have assigned values smaller than five to i; for, if the numerators exceed five just note that

5 + m m m = , i m + i i but then further

6 R p−1 n q −1 Euler again refers to his paper "’Observationes circa integralia formularum x dx(1 − x ) n posito post integrationem x = 1"’. This is paper E321 in the Eneström-Index. 252 Appendix

10 + m m m + 5 m = · , i m + i m + i + 5 i

15 + m m m + 5 m + 10 m = · · . i m + i m + i + 5 m + i + 10 i Furthermore, for this class two formulas indeed involve the quadrature of the circle; these formulas are

4 π 3 π = = α and = = β, 1 5 sin π 2 2π 5 5 sin 5 but then two contain higher quadratures, which we want to put as

3 Z xxdx Z dx 2 Z xdx = p = p = A and = p = B, 1 5 (1 − x5)4 5 (1 − x5)2 2 5 (1 − x5)3 and using these I assigned the values of all remaining formulas of this class7, namely

5 5 1 5 1 5 1 5 1 = 1, = , = = , = ; 1 2 2 3 3 4 4 5 5 4 4 β 4 β 4 α = α, = , = = ; 1 2 A 3 2B 4 3A 3 3 3 ββ = A, = β, = ; 1 2 3 αB 2 αB 2 = , = B; 1 β 2 1 αA = . 1 β

Corollary 1

Having taken the exponent i = 1, it will be

s s s Z −4 3 5 1 1 2 3 4 α dx log = 5 54 = 5 54 A2B, x 1 1 1 1 β2 whence we conclude in general, while n denotes a positive integer, that

s s Z 5n−4 3 5 1 1 6 11 5n − 4 α dx log = · · ··· 5 54 A2B. x 5 5 5 5 β

7Euler took the following list out of E321. Translation of E421 253

Corollary 2 §40 Now let i = 2, and since then this equation results s s Z −3 5 1 2 4 6 8 dx log = 5 54 , x 2 2 2 2 because of

6 1 1 1 2 8 3 3 = = and = , 2 3 2 3 1 2 3 2 the left-hand side will be

s r 2 4 2 3 BB 5 53 = 5 53αβ 2 2 1 2 A and in general s Z 5n−3 r 5 1 2 7 12 5n − 3 BB dx log = · · ··· 5 53αβ . x 5 5 5 5 A

Corollary 3 §41 Let i = 3 and the form found s s Z −2 5 1 3 6 9 12 dx log = 5 2 · 54 , x 3 3 3 3 because of

6 1 3 9 4 4 12 2 7 3 = , = , = · , 3 4 1 3 7 3 3 5 10 2 changes to s r 4 3 3 4 3 5 β A 5 2 · 52 = 52 · , 3 1 3 2 α BB whence it is concluded that in general s Z 5n−2 r 4 5 1 3 8 13 5n − 2 5 β A dx log = · · ··· 52 · . x 5 5 5 5 α BB

Corollary 4 §42 Finally, for i = 4 our equation s s Z −1 5 1 4 8 12 16 dx log = 5 6 · 54 , x 4 4 4 4 254 Appendix because of

8 3 4 12 2 7 4 16 1 6 11 4 = , = · , = · · , 4 7 3 4 6 11 2 4 5 10 15 1 will be transformed into this form

s r 4 4 4 4 5 ααββ 5 6 · 5 = 5 4 3 2 1 AAB such that in general s Z 5n−1 r 5 1 4 9 14 5n − 1 1 dx log = · · ··· 5 5ααββ . x 5 5 5 5 AAB

Scholium

R 1 λ §43 If we represent the value of the integral formula dx log x by the sign [λ], the cases expanded up to now yield

s s 4 α3 1 1 α3 − = 5 54 · A2B, + = 5 54 · A2B, 5 β2 5 5 β2 r r 3 BB 2 2 BB − = 5 53αβ · , + = 5 53αβ , 5 A 5 5 A s 4 r 4 2 β A 3 3 5 β A − = 5 52 · , + = 52 · , 5 β BB 5 5 α BB r r 1 1 4 4 1 − = 5 5α2β2 · , + = 5 5α2β2 · , 5 AAB 5 5 AAB whence by combining two, whose indices add up to 0, we conclude

1 1 π + · − = α = π , 5 5 5 sin 5 2 2 2π + · − = 2β = , 5 5 2π 5 sin 5 3 3 3π + · − = 3β = , 5 5 3π 5 sin 5 4 4 π + · − = 4α = . 5 5 4π 5 sin 5

But from the preceding problem, in like manner, we deduce: Translation of E421 255

3 r αα 1 1r αα − = P = 4 43 · AA, + = 4 43 · AA, 4 β 4 4 β r r 1 1 3 3 1 − = Q = 4 4ααβ · , + = 4 4ααβ · 4 AA 4 4 AA and hence

1 1 π + · − = α = π , 4 4 4 sin 4 3 3 3π + · − = 3α = , 4 4 3π 4 sin 4 whence, in general, we obtain this theorem that

λπ [λ] · [−λ] = ; sin λπ the reason for this can be given from the interpolation method explained some time ago8 as follows. Since

11−λ · 2λ 21−λ · 3λ 31−λ · 4λ [λ] = · · · etc., 1 + λ 2 + λ 3 + λ it will be

11+λ · 2−λ 21+λ · 3−λ 31+λ · 4−λ [−λ] = · · · etc. 1 − λ 2 − λ 3 − λ and hence

1 · 1 2 · 2 3 · 3 λπ [λ] · [−λ] = · · · etc. = , 1 − λλ 4 − λλ 9 − λλ sin λπ as I demonstrated elsewhere9.

Problem 6 - General Problem

§44 If the letters i and n denote positive integers, to deﬁne the value of the integral

i−n s − Z n Z i n 1 n 1 dx log or dx log x x having extended the integration from x = 0 to x = 1.

8Euler explains the interpolation method he talks about here also in E19. 9Euler proved this relation in his paper "’Methodus facilis computandi angulorum sinus ac tangentes tam naturales quam artiﬁciales"’. This is paper E128 in the Eneström-Index. 256 Appendix

Solution The method explained up to this point will exhibit the value in question expressed via quadratures of algebraic curves in the following way

q R n i−n s dx log 1 Z i−1 Z 2i−1 Z (n−1)i−1 x n n−1 x dx x dx x dx p = n p · p ··· p . n 1 · 2 · 3 ··· (i − 1) n (1 − xn)n−i n (1 − xn)n−i n (1 − xn)n−i

p−1 Hence, if, for the sake of brevity, we denote the integral formula R √ x dx by this character n (1−xn)n−q q p R n 1 m m m q , but on the other hand the formula dx log x by this character n such that n m denotes the value of this indeﬁnite product 1 · 2 · 3 ··· z, while z = n , the value in question will be expressed more succinctly as follows s i − n i 2i 3i ni − i = n 1 · 2 · 3 ··· (i − 1)nn−1 ··· , n i i i i whence it is also concluded that s i i i 2i 3i ni − i = n 1 · 2 · 3 ··· (i − 1)nn−1 ··· . n n i i i i Here, it will always sufﬁce to take the number i smaller than n, because it is known for larger numbers that

i + n i + n i i + 2n i + n i + 2n i = , in the same way = · etc., n n n n n n n and so the whole investigation is hence reduced to those cases in which the numerator i of the i fraction n is smaller than the denominator n. In addition, it will be helpful to have noted the following properties of the integral formulas

Z xp−1dx p p = : n (1 − xn)n−q q I. The letters p and q are interchangeable so that

p q = . q p II. If one of the two numbers p or q is equal to the exponent n, the value of the integral formula will be algebraic, namely

n p 1 n q 1 = = or = = . p n p q n q III. If the sum of the numbers p + q is equal to the exponent n, the value of the integral formula p q can be exhibited by means of the quadrature of the circle, because p n − p π q n − q π = = and = = . − pπ − qπ n p p n sin n n q q n sin n Translation of E421 257

p IV. If one of the numbers p or q is greater than the exponent n, the integral formula q can be reduced to another one whose terms are smaller than n; this is achieved using this reduction

p + n p p = . q p + q q V. There is a relation among many of these integral formulas of such a kind that

p p + q p p + r q q + r = = ; q r r q r p by means of this relation all reductions I gave in my observations on these formulas10 are found.

Corollary 1 §45 If we accommodate the formula that we found to each case in this way, by means of reduction IV, we will be able to exhibit them in the most simple way as follows. And for the case n = 2, in which no further reduction is necessary, we will have s 1 1 1 1r π 1√ 2 2 = 2 = π = π. 2 2 1 2 sin 2 2

Corollary 2 §46 For the case n = 3 we will have these reductions

s 1 1 1 2 = 3 32 3 3 1 1 s 2 2 2 1 = 3 3 · 1 . 3 3 2 2

Corollary 3 §47 For the case n = 4 one obtains these three reductions

s 1 1 1 2 3 = 4 43 , 4 4 1 1 1 s 2 s 2 2 4 2 4 1 2 = 42 · 2 = 2 4 4 4 2 2 2 2

4 1 because of 2 = 2 , s 3 3 3 2 1 = 4 · 1 · 2 ; 4 4 3 3 3

10Euler is again referring to E321. 258 Appendix

2 4−2 π because in the second equation 2 = 2 = 4 , it will, of course as before, be

2 1 1√ = = π. 4 2 2

Corollary 4

§48 Now let n = 5 and these four reductions result

s 1 1 1 2 3 4 = 5 54 , 5 5 1 1 1 1 s 2 2 2 4 1 3 = 5 53 · 1 , 5 5 2 2 2 2 s 3 3 3 1 4 2 = 5 52 · 1 · 2 , 5 5 3 3 3 3 s 4 4 4 3 2 1 = 5 5 · 1 · 2 · 3 . 5 5 4 4 4 4

Corollary 5

§49 Let n = 6 and we will have these reductions

s 1 1 1 2 3 4 5 = 6 65 , 6 6 1 1 1 1 1 s 2 2 s 2 2 6 2 4 6 1 3 4 = 64 · 2 = 3 62 , 6 6 2 2 2 3 2 2 s 3 2 s 3 3 6 3 6 1 3 = 63 · 3 · 3 = 2 6 , 6 6 3 3 2 3 s 2 2 s 4 4 8 4 2 6 2 4 2 = 62 · 2 · 4 · 2 = 3 6 · 2 , 6 6 4 4 4 3 4 4 s 5 5 5 4 3 2 1 = 6 6 · 1 · 2 · 3 · 4 . 6 6 5 5 5 5 5

Corollary 6

§50 For n = 7 the following six equations result Translation of E421 259

s 1 1 1 2 3 4 5 6 = 7 76 , 7 7 1 1 1 1 1 1 s 2 2 2 4 6 1 3 5 = 7 75 · 1 , 7 7 2 2 2 2 2 2 s 3 3 3 6 2 5 1 4 = 7 74 · 1 · 2 , 7 7 3 3 3 3 3 3 s 4 4 4 1 5 2 6 3 = 7 73 · 1 · 2 · 3 , 7 7 4 4 4 4 4 4 s 5 5 5 3 1 6 4 2 = 7 72 · 1 · 2 · 3 · 4 , 7 7 5 5 5 5 5 5 s 6 6 6 5 4 3 2 1 = 7 7 · 1 · 2 · 3 · 4 · 5 . 7 7 6 6 6 6 6 6

Corollary 7 §51 Now let n = 8 and one will get to these seven reductions

s 1 1 1 2 3 4 5 6 7 = 8 87 , 8 8 1 1 1 1 1 1 1 s 2 2 2 s 2 2 8 2 4 6 8 1 2 4 6 = 86 · 2 = 4 83 , 8 8 2 2 2 2 4 2 2 2 s 3 3 3 6 1 4 7 2 5 = 8 85 · 1 · 2 , 8 8 3 3 3 3 3 3 3 s 4 3 s 4 4 8 4 8 1 4 = 84 · 4 · 4 · 4 = 2 8 , 8 8 4 4 2 4 s 5 5 5 2 7 4 1 6 3 = 8 83 · 1 · 2 · 3 · 4 , 8 8 5 5 5 5 5 5 5 s 2 2 2 s 6 6 8 6 4 2 8 3 6 4 2 = 82 · 4 · 2 · 6 · 4 · 2 = 4 8 · 2 · 4 , 8 8 6 6 6 6 4 6 6 6 s 7 7 7 6 5 4 3 2 1 = 8 8 · 1 · 2 · 3 · 4 · 5 · 6 . 8 8 7 7 7 7 7 7 7

Scholium §52 It would be superﬂuous to expand these cases any further, because the structure of these formulas is already seen very clearly from the ones we gave. If the numbers m and n are coprime m in the propounded formula n , the rule is obvious, because 260 Appendix

s h m i m 1 2 3 n − 1 = n nn−m · 1 · 2 ··· (m − 1) ··· ; n n m m m m but if these numbers m and n have a common divisor, it will indeed be useful to reduce this fraction m n to the smallest form and extract the value in question from the preceding cases; nevertheless, the operation can also be done as follows. Because the expression in question certainly has this form h m i m p = n nn−mPQ, n n where Q is the product of the n − 1 integral formulas, P on the other hand the product of some absolute numbers, in order to ﬁnd that product Q just continue this series of formulas m 2m 3m m m m etc. until the numerator exceeds the exponent n, and instead of this numerator α write its excess over n; if this excess is set = α such that our formula is m , this numerator α will α α+m α+2m give a factor of a product P; then continue this series of formulas m m m etc. until one n+β again gets to a numerator greater than the exponent n, and the formula m emerges; instead β of this formula one then has to write m , and hence the factor β is introduced into the product and one has to continue like this until n − 1 formulas for Q will have emerged. To understand these operations more easily, let us expand the case of the formula

9 9 p = 12 123PQ 12 12 in this way; the letters P and Q are found as follows:

9 6 3 12 9 6 3 12 9 6 3 for Q ... , 9 9 9 9 9 9 9 9 9 9 9 for P . . . 6 · 3 9 · 6 · 3 9 · 6 · 3 and so one ﬁnds

93 63 33 122 Q = and P = 63 · 33 · 92. 9 9 9 9

12 1 3 3 9 3 6 3 3 3 Because 9 = 9 , PQ = 6 · 3 9 9 9 and hence s 9 3 9 6 3 = 4 12 · 6 · 3 . 12 4 9 9 9

Theorem

§53 Whatever positive numbers are indicated by the letters m and n, in the notation introduced and explained before it will always be s h m i m 1 2 3 n − 1 = n nn−m · 1 · 2 · 3 ··· (m − 1) ··· . n n m m m m Translation of E421 261

Proof For the cases in which m and n are coprime numbers, the validity of this theorem was shown in the preceding theorems; but the fact that it also holds, if those numbers m and n have a common divisor, is not evident from that theorem; but since the formula was already proved to be true in the cases in which m and n are mutually prime, it is natural to conclude that this theorem is true in general. I am completely aware that this kind to deduce something is completely unusual and must seem suspect to most people. In order to clear those doubts, because for the cases, in which the numbers m and n are composite, we obtained two expressions, it will be useful to have shown the agreement for the cases explained before. And the case m = n is already a huge conﬁrmation, in which case our formula obviously becomes = 1.

Corollary 1 §54 The ﬁrst case requiring a demonstration of the agreement is that one, in which m = 2 and n = 4, for which we found above (§47) s 2 2 2 4 2 = 42 ; 4 4 2 but now via the theorem s 2 2 1 2 3 = 4 42 · 1 , 4 4 2 2 2 where by comparison

2 1 3 = , 2 2 2 whose validity was conﬁrmed in my observations mentioned11 above.

Corollary 2 §55 If m = 2 and n = 6, using the results derived above (§49) we have s 2 2 2 2 6 2 4 = 64 ; 6 6 2 2 now on the other hand by means of the theorem s 2 2 1 2 3 4 5 = 6 64 · 1 6 6 2 2 2 2 2 and therefore it has to be

2 4 1 3 5 = , 2 2 2 2 2 whose validity is clear for the same reasons.

11 R p−1 n q −1 Euler again refers to his paper "’Observationes circa integralia formularum x dx(1 − x ) n posito post integrationem x = 1"’. This is paper E321 in the Eneström-Index. 262 Appendix

Corollary 3 §56 If m = 3 and n = 6, one arrives at this equation

32 1 2 4 5 = 1 · 2 ; 3 3 3 3 3 but if m = 4 and n = 6, in like manner,

4 2 1 3 5 22 = 1 · 2 · 3 4 4 4 4 4 or

4 2 3 1 3 5 = , 4 4 2 4 4 4 which is also found to be true.

Corollary 4 §57 The case m = 2 and n = 8 yields this equality

2 4 6 1 3 5 7 = , 2 2 2 2 2 2 2 but the case m = 4 and n = 8 this one

43 1 2 3 5 6 7 = 1 · 2 · 3 4 4 4 4 4 4 4 and ﬁnally the case m = 6 and n = 8 gives this equation

6 4 2 1 3 5 7 2 · 4 = 1 · 3 · 5 , 6 6 6 6 6 6 6 which is also true.

Scholium §58 But if in general the numbers m and n have the common factor 2 and the propounded 2m m formula is 2n = n , because s h m i m 1 2 3 n − 1 = n nn−m · 1 · 2 · 3 ··· (m − 1) ··· , n n m m m m after having reduced the same to the exponent 2n it will be s 2 2 2 2 m 2n 2 4 6 2n − 2 2n2n−2m · 22 · 42 · 62 ··· (2m − 2)2 ··· . n 2m 2m 2m 2m By the theorem, the same expression on the other hand becomes s m 1 2 3 2n − 1 2n 2n2n−2m · 1 · 2 · 3 ··· (2m − 1) ··· , n 2m 2m 2m 2m Translation of E421 263 whence for the exponent 2n it will be

2 4 6 2n − 2 2 · 4 · 6 ··· (2m − 2) ··· 2m 2m 2m 2m 1 3 5 2n − 1 = 1 · 3 · 5 ··· (2m − 1) ··· . 2m 2m 2m 2m If in the same way the common divisor is 3, for the exponent 3n one will ﬁnd

3 2 6 2 9 2 3n − 32 32 · 62 · 92 ··· (3m − 3)2 ··· 3m 3m 3m 3m 1 2 4 5 3n − 1 = 1 · 2 · 4 · 5 ··· (3m − 2)(3m − 1) ··· , 3m 3m 3m 3m 3m which equation can be more conveniently exhibited as follows

2 2 − 2 1 · 2 · 4 · 5 · 7 · 8 · 10 ··· (3m − 2)(3m − 1) 3 6 ··· 3n 3 = 3m 3m 3m . 32 · 62 · 92 ··· (3m − 3)2 1 2 4 5 7 3n−2 3n−1 3m 3m 3m 3m 3m ··· 3m 3m But if in general the common divisor is d and the exponent dn, one will have

d 2d 3d dn − d d d · 2d · 3d ··· (dm − d) ··· dm dm dm dm 1 2 3 dn − 1 = 1 · 2 · 3 · 4 ··· (dm − 1) ··· , dm dm dm dm which equation can easily be accommodated to any cases, whence the following theorem deserves to be noted.

Theorem

p §59 If α is a common divisor of the numbers m and n and the formula q denotes the value of the p−1 integral R √ x dx extended from x = 0 to x = 1, it will be n (1−xn)n−q α 2α 3α n − α α α · 2α · 3α ··· (m − α) ··· m m m m 1 2 3 n − 1 = 1 · 2 · 3 ··· (m − 1) ··· . m m m m

Proof The validity of this theorem is already seen from the preceding Scholium; while the common divisor was = d and the two propounded numbers were dm and dn there, here I just wrote m and n instead of them, but instead of their divisor d I wrote the letter α, the kind of divisor which the stated equality contains in such a way that one assumes the numbers m and n and hence also m − α and n − α to occur in the continued arithmetic progression α, 2α, 3α etc. In addition, I am forced to confess that this proof is, of course, mainly based on induction and cannot be 264 Appendix considered to be rigorous by any means; but because we are nevertheless convinced of its truth, this theorem seems to be worth of one’s greater attention; nevertheless, there is no doubt that a further expansion of integral formulas of this kind will ﬁnally lead to a complete proof; but it is an extraordinary specimen of analytical investigation that it was possible for us to see its truth before we had the complete proof.

Corollary 1 §60 Thus, if we substitute the integrals for the signs we introduced, our theorem will be as follows

Z xα−1dx Z x2α−1dx Z xn−α−1dx α · 2α · 3α ··· (m − α) p · p ··· p n (1 − xn)n−m n (1 − xn)n−m n (1 − xn)n−m s Z Z Z n−2 α dx xdx x dx = 1 · 2 · 3 ··· (m − 1) p · p ··· p . n (1 − xn)n−m n (1 − xn)n−m n (1 − xn)n−m

Corollary 2 p §61 Or if, for the sake of brevity, we set n (1 − xn)n−m = X, it will be

Z xα−1dx Z x2α−1dx Z xn−α−1dx α · 2α · 3α ··· (m − α) · ··· X X X s Z dx Z xdx Z x2dx Z xn−2dx = α 1 · 2 · 3 ··· (m − 1) · · ··· . X X X X

General Theorem

p §62 If the divisors of the two numbers m and n are α, β, γ etc. and the formula q denotes the value of p−1 the integral R √ x dx extended from x = 0 to x = 1, the following expressions consisting of integral n (1−xn)n−q formulas of this kind will be equal to each other

α 2α 3α n − α α α · 2α · 3α ··· (m − α) ··· m m m m β 2β 3β n − β β = β · 2β · 3β ··· (m − β) ··· m m m m γ 2γ 3γ n − γ γ = γ · 2γ · 3γ ··· (m − γ) ··· m m m m etc.

Proof The validity of this theorem obviously follows from the preceding theorem, because each of these expressions is equal to this one Translation of E421 265

1 2 3 n − 1 1 · 2 · 3 ··· (m − 1) ··· , m m m m which corresponds to the unity as smallest common divisor of the numbers m and n. Therefore, so many expressions of this kind, i.e. all equal to each other, can be exhibited as there were common divisors of the two numbers m and n.

Corollary 1

n 1 n §63 Because this formula m is = m and hence m m = 1, our equal expressions can be represented more succinctly as follows

α 2α 3α n α α · 2α · 3α ··· m ··· m m m m β 2β 3β n β = β · 2β · 3β ··· m ··· m m m m γ 2γ 3γ n γ = γ · 2γ · 3γ ··· m ··· . m m m m

For, even if the number of factors was increased here, the structure of these formulas is nevertheless easily seen.

Corollary 2 §64 Thus, if m = 6 and n = 12, because of the common divisors of these numbers, 6, 3, 2, 1, one will have the following four forms that are all equal to each other

6 126 3 6 9 123 = 6 = 3 · 6 6 6 6 6 6 6

2 4 6 8 10 122 = 2 · 4 · 6 6 6 6 6 6 6 1 2 3 12 = 1 · 2 · 3 · 4 · 5 · 6 ··· . 6 6 6 6

Corollary 3 §65 If the last formula is combined with the penultimate, this equation will result 1 · 3 · 5 2 4 6 8 10 12 = 6 6 6 6 6 6 , 2 · 4 · 6 1 3 5 7 9 11 6 6 6 6 6 6 but the last compared to the second yields 1 · 2 · 4 · 5 3 3 6 6 9 9 12 12 = 6 6 6 6 6 6 6 6 . 3 · 3 · 6 · 6 1 2 4 5 7 8 10 11 6 6 6 6 6 6 6 6 266 Appendix

Scholium §66 Hence inﬁnitely many relations among the integral formulas of the form

Z xp−1dx p p = n (1 − xn)n−q q follow, which are even more remarkable, because we were led to them by a completely singular method. And if anyone does not believe them to be true, he or she should consult my observations on these integral formulas12 and will then easily be convinced of their truth for any case. But even if this consideration provides some conﬁrmation, the relations found here are nevertheless of even greater importance, because a certain structure is noticed in them and they are easily generalised to all classes, whatever number was assumed for the exponent n, whereas in the ﬁrst treatment the calculation for the higher classes becomes continuously more cumbersome and intricate.

12 R p−1 n q −1 Euler again refers to his paper "Observationes circa integralia formularum x dx(1 − x ) n posito post integrationem x = 1". This is paper E321 in the Eneström-Index. Translation of E421 267

Supplement containing the Proof of the theorem propounded in §53

It is convenient to derive this proof from the results mentioned above; just take the equation given in §25, which for f = 1, having changed the letters, reads as follows

R ν−1 R µ−1 dx log 1 · dx log 1 Z xκµ−1dx x x = κ , R 1 ν+µ−1 (1 − xκ)1−ν dx log x and, using known reductions, represent it in this form

R 1 ν R 1 µ dx log · dx log κµν Z xκµ−1dx x x = . R 1 ν+µ µ + ν (1 − xκ)1−ν dx log x m λ Now set ν = n and µ = n , but then κ = n so that we have

m λ R 1 n R 1 n Z λ−1 dx log x · dx log x λm x dx + = , λ m pκ n n−m R 1 n λ + m (1 − x ) dx log x which, for the sake of brevity having used notation introduced above, is more conveniently expressed as follows m λ λm λ n n = . λ+m λ + m m n Now successively write the numbers 1, 2, 3, 4 . . . n instead of λ and multiply all these equations, whose number is = n, and the resulting equation will be

n 1 2 3 n h m i n n n ··· n n m+1 m+2 m+3 m+n n n n ··· n 1 2 3 n 1 2 3 n = mn · · ··· ··· m + 1 m + 2 m + 3 m + n m m m m 1 · 2 · 3 ··· m 1 2 3 n = mn ··· . (n + 1)(n + 2)(n + 3) ··· (m + n) m m m m But in like manner, just transform the left-hand side such that h m in 1 2 3 ··· m n n n n , n n+1 n+2 n+3 n+m n n n ··· n whose agreement with the preceding is revealed by cross multiplication. But because from the nature of these formulas n + 1 n + 1 1 n + 2 n + 2 2 n + 3 n + 3 3 = , = , = etc., n n n n 2 n n n n and since we have m of these formulas here, this left-hand side will become

h m in nm ; n (n + 1)(n + 2)(n + 3) ··· (n + m) because this one is equal to the other part exhibited before, namely 268 Appendix

1 · 2 · 3 ··· m 1 2 3 n mn ··· , (n + 1)(n + 2)(n + 3) ··· (n + m) m m m m we obtain this equation

h m in mn 1 2 3 n = 1 · 2 · 3 ··· m ··· n nn m m m m such that s h m i 1 · 2 · 3 ··· m 1 2 3 n = m n ··· ; n nm m m m m

n 1 because this equation m = m agrees with the one propounded in §53, its truth is now indeed proved from most certain principles.

Proof of the theorem propounded in §59

Also this theorem needs a more rigorous proof which I will give using the equation established before, i.e. m λ λm λ n n = , λ+m λ + m m n as follows. While α is a common divisor of the numbers m and n, successively write the numbers n α, 2α, 3α etc. up to n instead of λ, whose total amount is = α , and now multiply all equations resulting in this way such that this equation emerges

n α 2α 3α n h m i α n n n ··· n n m+α m+2α m+3α m+n n n n ··· n n α 2α 3α n α 2α 3α n = m α · · ··· ··· . m + α m + 2α m + 3α n + m m m m m Now transform the left-hand side into this one equal to it

n α 2α 3α m h m i α ··· n n n n , m n+α n+2α n+3α n+m n n n ··· n n+α n+α α which, because of n = n n and similarly for the remaining ones, is reduced to this one

n h m i α n n n n · · ··· , n n + α n + 2α n + 3α n + m In like manner, the right-hand side of the equation is transformed into this one n α 2α 3α m α 2α 3α n m α · · ··· ··· , n + α n + 2α n + 3α n + m m m m m whence this equation results

n h m i α m n α 2α 3α n n α = n α α · 2α · 3α ··· m ··· n m m m m Translation of E421 269 and hence s h m i 1 α 2α 3α n α = m n α · 2α · 3α ··· m ··· n nm m m m m which equation compared to the preceding yields this equation

α 2α 3α n α 1 2 3 n α · 2α · 3α ··· m ··· = 1 · 2 · 3 ··· m ··· , m m m m m m m m which is to be understood for all common divisors of the two numbers m and n.