You are on page 1of 16

WKNORR

PROBLEMS IN THE INTERPRETATION OFGREEK NUMBERTHEORY


EUCLID AND THEOREM THE OF FUNDAMENTAL ARITHMETIC

w o u LD appear that the historian of mathematics has a special advantage among historians of thought, in that the object of his study has a universality and an independence from contingent cultural considerations which other fields do not possess to a comparable degree. For instance, 2 + 3 = 5 independent of who makes the assertion and under what circumstances. But this advantage brings with it certain problems: (1) What can we mean in claiming that some early mathematical achievement is an equivalent of some later achievement expressed in different (e.g. more familiar modern) terms? (2) How appropriate is it to employ later standards of logic and rigor in the assessment of earlier mathematical works? I should like to take up these two issues in the context of a specific case, Euclids version of the fundamental theorem of arithmetic, which has been the subject of IT

some recent

and especially

unsophisticated

analyses.

Two authors, M. D. Hendy ( 1975) and A. A. Mullin (1965) have examined the Euclidean theorem on the assumption that to judge for ourselves the depth of Euclids propositions we need to translate his geometrical ideas into the modern language of arithmetic.2 They indicate certain defects in Euclids treatment, some of which are real, but others merely alleged, as I shall show. They correctly observe that the Euclidean theorem (Elements IX, 14) is not technically equivalent
lIf, however, we distinguish the fact that 2 + 3 = 5 from the statement 2 + 3 = 5, we may recognize that the latter is culture-dependent. The statement would have different meanings for, say, Plato and Peano-one considering it the assertion of an absolute and irreducible truth, the other viewing it as a theorem. A more detailed examination of this distinction is not necessary in the present context, but will be alluded to again briefly in our concluding remarks. 2Hendy, 190. The same assumption is implicit in Mullin and in the authorities which both Hendy and Mullin rely upon: Hardy and Wright and Buchner.
Stud. Hist. Phil. Sci. 7 (1976),

No. 3. Printed in Great Britain

353

354

Studies in History and Philosophy of Science

to

the

most which

fundamental theorem of arithmetic (FTA), to a restricted case of it. But they make three additional I maintain are invalid: (1) Euclids

complete

but at claims

proof holds only for numbers

factorable into three distinct primes, rather than into any finite number of primes; (2) this limitation is rooted in the geometric intuition underlying the Euclidean arithmetic; and (3) Euclid having gone so far could not prove the fundamental theorem itself,3-that is, technical considerations are such that his method cannot be readily adapted to the general case. In opposing each of these three claims, I shall show how they result from the authors inappropriate use of criteria based on modern number theory. Let us state the two theorems in question: (1) FTA: The factorization of any natural number into prime factors is unique, save for order. (2) Elements IX, 14: If a number be the least that is measured by prime numbers, it will not be measured by any other prime except those originally measuring it.4 The authors (Hendy and Mullin) correctly assert that the two theorems are not technically equivalent. For Euclids theorem does not cover the case of numbers which possess a square factor. The mistaken claim that the theorems are equivalent is explicit in T. L. Heaths comments on Euclid [1921, I, 401; 1926, II, 4031 and has insinuated itself into the standard secondary literature on the history of mathematics. The authors thus perform a useful service in resisting this error. Unfortunately, they propagate several other erroneous misleading claims. (1) Hendy severely circumscribes the generality of Euclids proof:
To demonstrate case again. where generously, this general proposition product generalises [IX, of 141 Euclid proves B, only C and number a number

or

the explicit

A is the

three

primes

D. Heath
of factors

but not rigourously,

this to an unspecified

The precise force of Hendys objection is not clear. Does he mean (a) that Euclid himself thought his proof valid only in the case of products of three primes? or (b) that by the standards of modern logic the validity of the proofs as given must be so restricted? We may show that
3Hendy, 191. Italics arc mine. 4Tbis and other quotations of Euclid follow Heaths translation. As the proof will be discussed later, a sketch of it will be useful. Let N be the least common multiple of primes A, B, C. If a prime p divides N, we may set N = p x M, where M is less than N. If p equals none of the primes A, B, C, then since each is a measure of N, it must measure M (by VII, 30). Ibis makes M a common multiple of A, B, C. But N was constructed as the Eeast common multiple of these numbers. Ibis contiadiction establishes that p must equal at least one of the terms measuring N. sBut one may note that many accounts avoid Heaths error. Cf. Bourbaki, 111; Bell, 49, 22Of; Taisbak, 109; and Davenport, 19.

Problems

in the Interpretation

of Greek

Number

Theory

355

(a) is just historical Without with

false, while (b) imposes interpretation doubt, of Euclidean Greek integers

an inappropriate arithmetic. suffered Euclid could matter,

conditicn the lines, handicap plane

upon the of an

arithmetic notation. (or, for that

under-developed unspecified figures)

algebraic

operate

rhetorically and solid

via a literal symbolism;

but to examine

the case of an indefinite

number of such unspecified numbers presented difficulty. Euclids solution here was to consider a definite number of terms, where this would not affect the pattern of the argument. For instance,
IX, 20: Prime numbers are more numbers than any assigned multitude of prime numbers. Let A, B, C be the assigned

.... that the multitude of

Does Euclid imagine he has only established primes is greater than three? Or, again:
VII, 33: Given as many numbers as we please, the same ratio with them. Let A, B, C be the given numbers,

to find the least of those which have as many as we please

...

In this case it is evident that Euclid assumes an explicit number of terms only for reasons of illustration. He certainly does not recognize this assumption as a restriction on the general applicability of the theorem or its proof. This same stylistic tactic appears throughout the arithmetic books.6 In the determination of the greatest common measure of given integers, Euclid develops the procedure first for two integers [VII, 21 then for three [VII, 31 . This suffices to establish the general recursive nature of the procedure; and, in fact, the ancient commentator Hero already made explicit the ability to carry through the process for any number of given integers [Heath, 1926, II, 302fl. The procedure for finding the least common multiple [VII, is also developed Implicit first for two, then for three extendability of the notion given integers 34, 361. is the indefinite

the process; and it is precisely this process which underlies of product used in the factorization theorem [IX, 141. The perfect-number theorem [IX, 361 illustrates an unusual way. Here, Euclid proves that a specified

the same tactic in construction gives

rise to perfect numbers, i.e. those equal to the sum of all of their own proper divisors. In this construction, one sets out the geometric progression 1, 2, 22, 23, . . . , 2, where n is such that the sum of the terms of the progression is a prime. For instance, 1 + 2 = 3, or 1 + 2 + 4 = 7. When this prime is multiplied by the final term in the progression, a perfect number results (e.g. 3 x 2 = 6 and 7 x 4 = 28 are perfect). But in the proof, Euclid hypothesizes as many numbers as we please, A, B, until the sum of C, D, beginning from a unit . . . in double proportion,
60ther instances are VII, 12, 14; VIII, l-4,6-10; IX, 8-13, 17, 21-23, 32, 35, 36.

356

Studies

in History

and Philosophy

ofScience

prime. Taking him literally, we are examining the terms 1, 2, 4, 8, whose sum is 15; this is not a prime, so the hypothesis of the proof does not in fact lead to the construction of a perfect number. It is thus clear through this example, as through those mentioned above, that Euclid does not view as restrictive any such assumption of a particular number of terms; the assumption is made for the convenience of the exposition of a property or process applying to ageneral class of an indefinite number of terms. One may note that this use of explicit values for an intended indefinite index appears elsewhere in the Elements. For instance, the Euclidean division algorithm for integers [VII, l-31 and for magnitudes [X, 2-41 requires an unspecified number of subtractive steps; but Euclid assumes in VII, 1 that the process terminates after three steps; a similar assumption is made in X, l-4. The Euclidean conception of proportion [V, Defn. 51 requires the introduction of arbitrary integral multiples of given ma<gnitudes. But in practice a definite multiple is taken, as in VI, 1 where Euclid assumes the doubles of Cgiven lines. Hence, if we are to adopt the critical position expressed in (b) above, we must sacrifice the major portion of the Elements as not rigorously general.7 What, then, is our judgment of Euclidean method to be? Shall we reject Euclid, concluding with Hendy and Mullin that it is reasonable to assume that formal induction either did not occur to [the Greeks], or else was considered logically unacceptable? I believe not; we do better to inspect Euclids methods and the dialectical commentary on these by his contemporaries. On the matter of formal induction, one may see that Euclid establishes results obtaining for all the natural numbers by an argument very similar to our complete-inductive proofs. For instance, as indicated before, his treatments of the procedures for finding the greatest common measure and the least common multiple fall into two parts: first Euclid verifies the procedure for two integers; then he shows how the case for three integers can be reduced to the case for two. In like fashion, a modern complete-inductive proof of a proposition P ranging over the natural numbers would first establish the validity of P (1) and then show that P (n) implies P (n+l) where n is an unspecified integer. Euclids use of the analogous recursive pattern thus behes the claim that he lacked or distrusted the principle of induction.
Modern treatments of number theory are inspired by the Disquisitiones Arithmeticae (1801) of Gauss. It is interesting that Gauss expresses general theorems in a fashion close to Euclids. For instance, in section 15 he proves that if none of the numbers a, 6, c, d, etc. can be divided by a prime p neither can their product abed etc. Shall we then judge that Gauss proofs are defective, just as Hendy and Mullin allege Euclids to be? Or are we to believe that the mere insertion of etc. is sufficient to rehabilitate the Euclidean proofs?

all becomes

Problems

in the Interpretation

of Greek

Number

Theory

357

Rather, validate

it was a technique treatment the generality of algebraic it understood

of general

quantification by implicit applies

which eluded him. n and for n = appeal to an axiom

Where the modern or theory 3, leaving

can introduce

the dummy-variable

of the argument representation, that the proof

Euclid proves his theorem of universal proof

for any other value of via general

n.8 Now, the precise

logical justification

variables has been attempted in a variety of ways and at least one logician recognizes the matter as problematic. In view of this, we should be permitted to ask whether a defence of Euclids approach can be given. Euclid could present two arguments. First, the primary objective of the Elements is pedagogical. In the context of an elementary introduction to geometry and number theory it can be quite effective to substitute specific values for general terms in the investigation of a proposition, especially when such substitution does not affect the generality of the concepts or procedures under discussion. This manoeuvre is common enough in modern teaching as not to require further remark. Second, when one proves a proposition obtaining for a set of entities which have certain defining properties in common by effecting the proof for just one member of the set, one is employing this member as a representative instance. The proof is generalizable to the extent that one has employed only those properties of the representative which it has in common with all the others in the set. Precisely this justification of mathematical generality is given by Proclus. Commenting on Elements I, 1, the construction of an equilateral triangle on a given line, he writes thus:
[geometers] suppose justified, that since pass therefore the result for to the universal to they conclusion in order instance. others that This we may procedure not is is confined figures, the particular resembling

the demonstration

use the objects

set out in the diagram of the same sort.

not as these particular

but as figures

In so stating this principle traditional Aristotelian view.

of representation, Proclus follows In Aristotles logic, most notably

the the

sThis suggests a reason for Euclids frequent choice of 3 as the particular instance examined. When a proposition or procedure involves recursive steps, the case of 2 will not suffice for showing the recursion, whereas the cases of 4 onward will introduce patent redundancies into the argument. Reduction of the case of 3 to that of 2 is both necessary and sufficient for illustrating the recursion. 9Russell writes: The notion of the variable is one of the most difficult with which logic has to deal [1937, 5-61 and disclaims the goal of providing a satisfactory theory of it in the principles of Mathematics; nevertheless, he devotes chapter VIII (esp. section 93) to the discussion of the variable and general propositions. W. van 0. Quine introduces a rule of universal generalization for the purposes of his own theory of quantification [1959, 159-1671. 1o~roclus(Friedlein), 207; Morrow, 162.

358

Studies

in History

and Philosophy

of Science

Posterior

Analytics,

problems

in the execution

of formal

proofs

are

taken up. The aspects Aristotle insists that universals, particular

of generality in proofs are of specific we can have scientific knowledge

interest. only of

not of particulars-that is, of attributes of things, not of the things themselves. We can know scientifically that all pairs even though we cannot know (in another sense)

are even, for instance,

all pairs [Post. Anal. I, 11. But attributes do not exist apart from things. How, then, can one demonstrate a proposition like A is true of B in general?-e.g. that having angles which sum to two right angles is true of being a triangle. Aristotle indicates this is done by assuming a random instance of the class in question, here a triangle chosen at random (ibid., I, 4). Care must be taken to choose the instance as representative of the widest possible class. For instance, in demonstrating the property of the sum of the angles, one should choose a triangle at random, not an isosceles triangle at random; otherwise, the proof holds for the subclass only (ibid., I, 5). Aristotles remarks thus provide guidelines for the setting out of general propositions and proofs. Euclids use of particular instances in the demonstration of general theorems is seen to conform to these Aristotelian rules. Precisely such a principle of representation answers the objections raised by Hendy against the generality of Euclids proofs. Of course, we cannot retrieve the particular thoughts of Euclid on this matter; but a historical analysis of the nature of Euclids procedure does not have such a specific objective anyway. The Elements (both those of Euclid and those, no longer extant, of his Academic predecessors), by virtue of their project of organizing mathematics systematically and their presumption of offering complete formal proofs of mathematical truths, became the focus of inquiries on the nature of proof itself. Logical problems related to the Elements were debated both before and after Euclids time, and some answers were proposed by scholars in the circles of Plato and Aristotle; moreover, Euclids presentation meets the test of many of these answers. These historian of logic and mathematics. facts must certainly interest the

11A remark on this theorem by Geminus (quoted by Eutocius) has been interpreted to indicate that the proof was originally divided into the cases of equilateral, isosceles and scalene triangles. But Heath argues this refers not to the original proof historically, but to pedagogical treatments familiar in antiquity [ 1926, I, 317-320; 1949,431. 2Related to these remarks on universal proof is another point by Aristotle: that the geometer does not err in assuming a line is straight or one-foot-long, even though the drawn representative has not these properties; for the actual straightness or the actual length of the particular line examined is not an element in his syllogism [Prior Analytics, 49, b33-7; 50 al-4; Metaphysics, 1089 a21-5; cf. Heath, (1949) 26f, 219f]. E. Beth discusses these Aristotelian passages in the course of the presentation of his own tableaufor general deduction via specific instantiation [ 1964, 190-1941.

Problems

in the Interpretation

of Greek Number

Theory

359

But to go further than this-to press the issue of rigor in accordance with more stringent modern canons-is in my view less fruitful. For instance, the objection that a representation-principle in the form given above is not adequate for the purposes of number theory is a legitimate one. For if we assume a recursive procedure terminates after three steps, say, how can we avoid appealing to the specific nature of three in the demonstration? But this objection holds equally against Aristotle and the fourth-century logicians as against Euclid. What I wish to emphasize is that Euclids procedure in this regard was well-considered and co-ordinate with the dialectical critique of mathematical methods in antiquity. By imposing strict modern canons in our assessment of his treatments one would become guilty, thus, of two major oversights: ignorance of the state of foundations studies in antiquity: and ignorance of the pedagogical context of the Elements. (2) I have suggested that the reason why Euclid chooses three as the representative instance in his arithmetic proofs follows from the recursive nature of his arithmetic processes. It has nothing to do with some specious restriction to three factors as being the maximum that [he] clearly conceived with [his] geometrically oriented notation. 3 There is a widely-circulating impression that Euclids number theory is essentially geometric in its underlying intuition. I wish now to show the untenability of this view. First, let us consider the features which have led to the adoption of such a view. (a) Euclid represents integers by line segments in the propositions in Books VII, VIII and IX. (b) Euclid introduces certain classifications of numbers, such as plane, solid, square and cube [VII, Defn. 16-191 which have an obvious geometric motivation. The linear representations in (a) are firmly established in the Elements. They have their utility for helping us keep in mind the order of the terms introduced in successive steps of the arguments. Line segments are especially well-suited for displaying the additive relations between integers. geometric mode, For example, But other than this, Euclid does nothing with the and it is this fact which we should find surprising. on the commutativity of multiplication [VII,

the theorem

161 ought to be absolutely trivial. To show A x B = B x A for given integers A and B, could we not figure each product as a rectangle? The equality would then follow from Euclids axiom of superposition and the invariance of area under translation. But no; Euclid provides an arithmetic demonstration based on numerical proportions and an asymmetrical definition of multiplication [VII, Defn. 151 involving
13Hendy [1975, 1901 owes this view to Mullin [1965, Wright [ 1938, 1821, who in turn cite S. Buchner. 2181, who lifts it from Hardy and

360

Studies in History and Philosophy of Science

repeated addition. Geometric properties and propositions utilized here, nor anywhere else in the arithmetic books. As for (b), these quasi-geometric classifications of numbers origins in the concrete

are

not

had their

already by Platos merely by analogy [cf Theaetetus 148A]. Euclids definitions are completely arithmetic. A square number is the product of two equal factors, a cube the product of three equal factors. The product [i.e. of multiplication] is in turn defined as the repeated addition of one integer according to the multitude of units in another [VII, Defn. 151. Significantly, VII, 16-18 when such plane and solid or VIII, 1 l-27, the Euclidean integers are discussed, as in manuscripts retain the linear

Pythagorean arithmetic of the fifth century. But viewed them as geometric time, mathematicians

representation. But what is the geometric sense of visualizing a square or cubic integer as a line segment? A glance at VII, 27 or VIII, l-10 and IX, 8-13 shows that Euclids concept of the power of an integer [i.e. the equivalent to our An] was not restricted to second- or third-powers, and hence a fortiori not bound Euclid obtains the higher powers by by alleged geometric limitations. means of indefinitely extended progressions in continued proportion. For instance, let us generate a progression for a given integer A according to the following recursive rule: 1, A, A x A, A x (A x A), A x (A x (A x A)), etc. Th en what we call An would be for Euclid the (n+l)st term of this progression. For reasons given earlier, Euclid might find tricky the examination of the general term; but he would certainly have no problem in discussing, say, the seventh powers, if he cared to. If my argument is accepted, one might wonder why Euclid adopted the linear representation at all. I believe the answer to this lies in the theory Book integers of incommensurable and irrational magnitudes developed in X. That theory is founded upon an arithmetic distinction: are commensurable 5-61. As Euclid if and only can construct if they have the ratio of [X, incommensurables only by

magnitudes

geometric means, the entities of Book X are of necessity geometric: line segments and rectangles. But as proportions involving integers appear throughout, it becomes convenient to operate with integers as if they were commensurable line segments. I expand upon this view elsewhere [Knorr, 1975, 241-3, 26Of, 3091. In brief, the arithmetic books VII-IX do not rely upon geometric properties or intuitions. In less than one-fourth of the theorems is even the linear representation of integers contained in the text of the propositions. i4 The adoption of this limited degree of geometrization
14The text of a demonstration points necessarily to an implied following way only: an integer AB is the sum of integers AC and

CB.

linear representation in the Such dictions appear in

Problems

in the Interpretation

of Greek Number

Theory

361

is accountable with reference itself, but to such external theory of incommensurables.

not to the Euclidean theory of number in considerations as its applications to the

(3) We may now turn to the relation of Euclids IX, 14 to the familiar modern theorems on the uniqueness of factorization. As already said, Hendy and Mullin and others have correctly observed that Euclids theorem is at most a restricted case, equivalent to the FTA for integers free of square factors. It might be objected that even this is not accurate, since Euclid does not explicitly introduce factorization in the asserting instead a condition on the divisibility of a theorem, constructed composite number by a single prime. But a careful look will show that the transition to a statement on uniqueness of factorization for such square-free integers is immediate from Euclids result. Euclid has shown that any prime which divides the least common multiple of given primes must equal one of those primes. That is, if we express the least common multiple of primes a, b, c, . . . as A, and if we discover an alternative expression for A as the least common multiple of primes a, b, c, . . ., by Euclids theorem each of the primes a, b, c, . . . appears among the terms a, b, c, . . ., and by the same token, each of the a, b, c, . . . appears the two sets are identical, save for the redundancies. The equality of these sets of the assertion of the uniqueness of square-free integer. among the a, b, c, . . . Hence, order of the terms and possible is a precise technical equivalent factorization into primes of a

We must now consider the relation of these results to the complete FTA: does Euclids omission of the complete form indicate his inability to provide a proof or even to express the complete theorem, as Hendy and Mullin charge? I argue not. Here, three points may be made: First, although textbooks today commonly assign to the FTA the status of a fundamental theorem, it should be recognized that such is hardly necessary. If one chooses to begin from the definition of prime numbers and the uniqueness of factorization into primes, developing from this the properties of relatively prime integers, then the FTA is indeed central. r5 But Euclid organizes his number theory in just the reverse order: he starts from the division algorithm for finding the greatest common measure of integers; this gives him an operative definition of relatively prime integers (i.e. those whose greatest common measure is 1). From the examination of relative-primality, he
4-11, 15, 20, 28; IX, 15, 20-27, 35. Otherwise, Euclid refers to integers via single-literal symbols: let the integers A, B, C, and D be given, for instance. There is nothing intrinsically geometric in the latter mode of reference. 1 SFor such an approach, see Gauss [section II: Congruences of the First Degree] ; Hardy and Wright [ 19881; and Davenport [ 1952, chapter I.$51.

VII, l-2,

362

Studies

in History

and Philosophy

of Science

eventually theorem loses slipped the

obtains VII, 30: of of

results

on

primes,

in particular,

the

important then it the FTA it is for the Placed in

if a prime measures its significance. IX, Far

a product from

of two integers, fundamental, theory. importance

must measure much middle

at least one of those integers. l6 In this theory, being no the end of Euclids it obviously arithmetic has

in well toward Book

verifications of the prior theorems; as it makes use of no theorems other than VII, 30 and 36, it cannot be viewed as the culmination of any major portion of the theory; nor is IX, 14 applied in any subsequent theorem. Its inclusion by Euclid was probably prompted by the familiarity of the unique factorization into primes within arithmetic practice and the suitability of demonstrating such a result in a theory of numbers. Second, we must be highly sceptical of any claim that Euclid could not prove the complete FTA. As shown above, the theorem was not particularly important for Euclids theory, so that the omission of the general FTA had no intrinsic significance. Moreover, when one perceives how readily the result is proven in modern school-text introductions to number theory-for which the only requisite is the equivalent of VII, 30, as for Euclid in IX, 14-the hypothesis of essential limitations on the Euclidean approach becomes even more unconvincing. Third, to counter Hendy and Mullins view of Euclidean incompetence, I propose to offer a proof accessible within the ancient number theory. But did Euclid even &-row the general theorem? We must come to terms with this question first. It is easy to accept a priori that Euclid must have known the more general result; for it emerges quite readily from the practical arithmetic experience of finding the prime factors of numbers. But we can find within the Elements a more explicit indication of this. The theorem on perfect numbers [IX, 361 requires the designation of all the divisors of the number Yp, where p is a prime. Euclid sets these divisors out as the numbers 1, 2, 2* : 23, . . . , 2,p, 3% 2*p, * * *, 2-p. In doing so, Euclid must be aware of the relation of the divisors of a number to a factorization of it;
t61n the proof of this theorem, Euclid does not require the full condition of primality. For he deduces from the hypothesis that the prime p measures the product A x B, butp does not measure (or equal) A, that p is relatively prime to A and, consequently, measures (or equals) B. Thus, if we consider the structure of Euclids proof of theorem VII, 30, we can perceive that this Theorem is actually the corollary to a more general assertion: that if a number (not necessarily prime) measures a product A x B and is relatively prime to A, it must measure (or equal) B. This more general form will be used later, in our construction of a Euclidean proof of the FTA. t 7Cf. Davenports alternative proof of the FTA [ 1952, 291. But he employs implicitly a non-Euclidean assumption, the order-independence of n-long products; cf. note 23 below.

Problems in the Interpretation

of Greek Number

Theory

363

moreover, when n is 2 or greater, the factored number possesses a square factor, so that we meet a case not covered in IX, 14. Now, in the course of the proof of IX, 36, Euclid demonstrates that the number 2~ has no other divisors than those designated. It is thus clear that Euclid possessed a result illustrative of the principle of the uniqueness of factorization into primes, viewed it as a result on factorization, and realized that it required a proof. a We may be satisfied that Euclid also knew the more general result treated in the FTA and its requirement of proof. On this base, we may inquire whether it was within the power of his methods.to produce such a proof. The construction of a Euclidean proof of the FTA, while presenting little difficulty, will utilize a few corollaries to Euclidean theorems, as well as some properties of the powers of numbers. The class of numbers we term powers are distinguished by the neo-Pythagoreans, who call them a special class of composite numbers, those measured by a single I, 1 l-131. Diophantus provides a prime only [cf. Nicomachus, nomenclature and a notation for powers up to the sixth order [Atithmetica, I, preface] and Archimedes denominates very large powers of 10 in the Sand-Reckoner. As said above, Euclid introduces in VII, 27 and in Books VIII and IX a construction of powers paralleling our recursive definition An =A x An- . He sets out a finite geometric progression, where the first term is unity, and examines the properties of the final term. What we denote An is thus for Euclid the (n+l)st term in such a progression. Among the results given by him are these: (i) Any term measures the final term according to the units in a term within the progression [IX, 1 l] ; that is, An = Ar x Afldr. (ii) Any prime which measures the final term measures the second also [IX, 121 ; that is, if a prime p measures An, it measures A also. (iii) The final term in a progression whose second term is prime is measured by all the numbers in the progression and by no other numbers [IX, 131; that is, for a prime p, the set 1, p, p2 . . . , p- exhausts the proper divisors of p . (iv) An additional result, not given by Euclid, follows at once from (iii): if two numbers are powers of different primes, they are relatively prime to each other.lg We state, further, two theorems which complement IX, 14: (v) If Nis the least common multiple of a set of numbers A, B, C, . . ., where each term in the set is relatively prime to each other term in the 8 Euclids proof in IX, 36 is not generalizable in this form to the general case of uniqueness of factorization into primes, however. 9Froof: if pn and qm are not relatively prime, there must exist some prime r which is a common measure; by IX, 12 (ii) r must measure both p and q; hence, r = p = q. This contradicts the choice of p and q as different primes. The special case where n = m is proved in VII, 27.

364

Studies in History and Philosophy

of Science

set, then it is impossible terms in the set. 2o

that any divisor of N be relatively a number N, as in (v), then

prime to all it measures

(vi) If a prime measures exactly one of the terms A,

B, C, ...

The FTA will develop as a corollary to the following theorem: let N be the least common multiple of a set of terms A, B, C, ...,taken as powers of different primes; let p be a prime divisor of N; let p be a divisor of N, but p n+l not be a divisor of N; then p equals exactly one of the original terms measuring N. The proof may be given as follows: Since A, B, C, ...are powers of different primes, by (iv) they are relatively prime, each to all the others. The condition of (v) and (vi) is thus satisfied; since p measures N, it measures exactly one of the terms of the set, say A. By (ii) A is a power of p. We claim A = p. For if they are unequal, one must measure the other, by (i). (1) Suppose p measures A; we set A = p" x q, where by (iii) p measures q. As A measures N, it follows that p n+l or even some higher power of p measures N. This contradicts our hypothesis that

P n is not a divisor of N. Then (2) we suppose A measures p. If p = A x q, p is a measure of q, as before. Set N = p x M. Now, B measures N and by (iv) is relatively prime to p; hence, B measures M.21 The same follows for C and all the other powers (other than A) originally measuring N. Hence, A x M is a common multiple of A, B, C, . . . But if A measures p, then A x M measures p x M = N. This contradicts the construction of N as the least common multiple of A, B, C ) . . . 22Thus, p = A.
The FTA follows as a corollary. Let N be expressed as the least common multiple of A, B, C, . . . , powers of different primes; let N be expressed as the least common multiple also of A',B',C', . . . , as an alternative set of powers of different primes. By the previous theorem, each term A',B',C', . . . equals exactly one of the terms A, B, C, . . . ; conversely, each term A, B, C, . . . exactly equals one of the terms A', B', C', . . . . Thus, the two sets are identical, save for order. The proof thus constructed establishes that the Euclidean number theory is perfectly adequate for verifying the complete form of the FTA. Although the proof given may appear intricate, even unwieldy, especially in comparison with familiar modern textbook treatments of the theorem, this cannot be supposed to indicate any inherent shortcomings in Euclids approach. In fact, modern elementary treatments commonly assume without proof certain steps which do require
zA proof may be modelled directly to appeal to VII, 24. Cf. note 4.
on that of IX, 14, where the appeal to VII, 30 is altered

2 1 This step may be justified via a corollary to VII, 30; cf. note 16. 22The steps in this argument should be compared with Euclids proof

of IX, 14; cf. note 4.

Problems

in the Interpretation

of Greek

Number

Theory

365

proof; permitted such assumptions, the above argument could considerably abridged. 23 But it is clear that there can be no question Euclids inability either to express the complete factorization

be of

theorem

or to provide an acceptable proof. This discussion addresses the second of the two issues with which we began: one cannot hope efficiently to understand or fairly to assess a mathematical system, the concepts, with like Euclids number organization and theory, logical in strict accordance standards of the

corresponding modern systems. By imposing modern standards of logic, authors like Hendy and Mullin presume to detect flaws-here, the alleged lack of generality of Euclids proofs-and thus fail to see that Euclids procedure was in accord with the studies of proof-technique of his time. In accounting for the alleged flaw, they misconstrue the geometric intuition underlying the Greek arithmetic. Finally, in drawing from this the conclusion of Euclids incompetence to prove the they miss entirely the force of more general factorization theorem, those Euclidean theorems and concepts, through which the construction of such a proof can become straightforward. Euclids omission of the theorem be This being so, shouldnt problematic? Not necessarily. As we have seen, the proof of an equivalent of the FTA would require a detailed set of extensions, albeit direct ones, of given Euclidean theorems. Such detail might well be deemed unsuitable for an introductory work such as the Elements. Moreover, the results at issue are sufficiently straightforward as not to overreach the abilities of a competent student of the Elements. Finally, as we have also noted before, the general theorem would not be fundamental within the organization chosen by Euclid. Thus, its omission does not detract in any way from the development and utility of Euclids theory.
231n particular, elementary treatments usually accept that the value of the product of n-many integers is independent of the order of multiplication A proof might require not only the commutativity and associativity of multiplication, but also a systematic inquiry into the permutations of finite sets. Such is the approach adopted by B. L. van der Waerden [1949, 11-18, 53-621. This result is possible in a Euclidean form: if N is the least common multiple of A, B, C, . . . , each term being relatively prime to all the others, and if one of those terms (say, D) be chosen; then setting N = D x M, we can prove that M is the least common multiple of the other terms originally measuring N. The proof follows immediately from the construction of the least common multiple in IX, 36. By means of this result, we may provide an alternative proof for the FTA. We choose those two terms in the two factorizations of N which by (v) are not relatively prime, say A and A; these are powers of the same prime p. If they are unequal, one measures the other: we set A = A x q, where p is a measure of Q also. If now we set N = A x M = A x M, M will be the least common multiple of B, C, . . . and M will be the least common multiple of 8, C, . . . We have also that M : M = A : A= q : 1. By VII, 20 it follows that q measures M. But 4 is relatively prime to all the terms B, C, . . . of which M is the least common measure; this is impossible by (v). Thus A = A, and so on for each other term in the expression for N.

366

Studies

in History

and Philosophy

of Science

This

brings

us to the

other

general

issue we raised:

how can the

historian avoid the risk of distortion in speaking of an ancient result as the equivalent of a modern result? In a very real sense, he cannot. Consider first that the Euclidean theorems and the modern factorization theorems presume certain concepts: the common multiple and least common multiple of specified integers, on the one hand; the n-long product, on the other. Are these concepts the same? Detectable in Euclids treatment is an element of sophistication which introa recursive conductory modern treatments lack. Euclid specifies struction for the extended product [VII, 361; he proves the commutativity of binary products [VII, 161 -this is not taken for granted; while the commutativity and associativity of the extended products are not proved, he does not apply these properties either. But implicit is a realization that such results as the order-independence of the procedure for common multiples require proofs. By contrast, modern elementary treatments of the FTA typically assume such results without proof (see notes 17 and 23). Complete treatments are deferred to the more formal introductions to higher arithmetic. These latter develop arithmetic in the context of more general algebraic structures, in particular, algebraic fields and integral ideals [Bell, 220, assumption of the order-independence of 223-51. Now, Euclids products could go wrong only in the instance of non-commutative systems; but his number theory is of course not such a system, and he the ability to generalize beyond the natural does not recognize numbers. His formal handling of the concepts in number theory thus distances him both from modern elementary accounts and modern advanced accounts. If the respective fort&i the results notions of product are not strict equivalents, then a which require them cannot be equivalents either. We

must then seek another sense of equivalence, if any such exists, which might relate the Euclidean and the modern theorems. As indicated earlier, the role played by each theorem within its respective theory of numbers will not provide us one. In modern number theory, following Gauss, the FTA is indeed fundamental; by contrast, it is the algorithm for finding the greatest common measure and the theorems on relative primality which are fundamental for Euclid. One sees that considerations based on the theory and the order and use of concepts within different systems compel us to distinguish the ancient works from the modern, rather than to formulate equivalences between them. If we had at our disposal a concept of mathematical fact, an absolute about which theorems are made and proved, as a Platonist conception might offer, then the problem of equivalence could easily be answered.

Problems

in the Interpretation

of

Greek Number do not

Theory have any such absoluteness

367 or

But

mathematical

statements character

independence ability

from cultural

considerations. of mathematics,

As modern views emphasize we must here give up the statement refers

the conventional to elements,

to appeal to the absolute. operations

For if a mathematical

and identities whose meaning and validity are accepted by postulate alone, how can we speak of the equivalence of two statements founded upon different postulates? I believe the historian can retrieve a serviceable notion of equivalence from the practice of mathematics. For instance, Euclids IX, 14 contains a partial answer to the practical question: what are the divisors of a number which is constructed as the product of specified numbers? The FTA may also be viewed as addressing this question. I maintain that to the extent the two theorems imply justifications for the same to that extent they are procedures in mathematical practice, legitimately termed equivalent. Such a rule can establish an equivalence between theorems and also between concepts, since concepts regularly have a component rooted in practice. Now, our discussion of Euclids theorems and the FTA has never claimed an equivalence between their proofs. But it sometimes happens that we do wish to speak of an equivaIence of this kind. We might term two proofs equivalent if they order equivalent concepts and equivalent theorems according to the same pattern. The historian of mathematics constantly speaks of equivalents, and this is inevitable if study of the history of mathematics is to have any relevance to what mathematicians do today. On the other hand, the philosophical description of historical equivaIence must surely be considered problematic; I should be greatly amazed if the sohrtion proposed above survives intact a close philosophical scrutiny. But it is in the clarification of such important presuppositions of historical work, I believe, that the philosopher of mathematics can abet the labors of the historian. Moreover, subjecting ancient work to the test of modern logic can point out the differences in philosophical outlook entailed by the ancient procedure. But to go beyond this-to interpret deviations as logical errors, to misconstrue concepts, or to invent essential limitations on the capacities of early mathematical systems-as the authors here criticized have attempted, serves no purpose. Their effort typifies a perversion of the logicians role in the study of the history of mathematics.

Bibliography

Bell, E. T., Development of Mathematics, New York, 1945. Beth, E., Foundations of Mathematics, Amsterdam, 1959 [revised edition 1964; reprint New York, 19661. Bourbaki, N., Eltments dhistoire des mathdmatiques, Paris, 1969. Davenport, H., The Higher Arithmetic, London, 1952. Euclid, The Thirteen Books of Euclids Elements, T. L. Heath, (ed. and trans.) 2nd edn., Cambridge, 3 vols., 1926. Gauss, C. F., Dljquisitiones Arithmeticae, Leipzig, 1801 [trans. A. A. Clarke, New Haven, Conn., 19661. Hardy, G. H. and E. M. Wright, An Introduction to the Theory of Numbers, 4th edn, Oxford, 1938. Heath, T. L., History of Greek Mathematics, Oxford, 2 vols, 1921. Heath, T. L., Mathematics in Aristotle, Oxford, 1949. Hendy, M. D., Euclid and the Fundamental Theorem of Arithmetic, Historia Mathematics, vol. 2, pp. 189-191, 1975. Knon, W. R, The Evolution of the Euclidean Elements, Dordrecht, Netherlands, 1975. Mullin, A. A. Mathematico-philosophical remarks on new theorems analogous to the fundamental theorem of arithmetic, Notre Dame J. Formal Logic, 6, pp. 218-222, 1965. Proclus, A Commentary 0% the First Book of Euclids Elements, G. R. Morrow (trans) [from the text of G. Friedlein, Leipzig, 18671, Princeton, New Jersey, 1970. Q.uine, W. van 0.. Methods of Logic, revised edn, New York, 1959. &.sei, B., Principles of Mathemu&, London, 1903 [2nd edn, 19371. Taisbak, C. M., Division and Logos, Odense, Denmark, 1971. van der Waerden, B. L, Modern Algebra, New York, 1949.

368

You might also like