c Springer-Verlag 2000

Arch. Hist. Exact Sci. 55 (2000) 5776.

Plato: Parmenides 149a7-c3.

A Proof by Complete Induction?
F. Acerbi
Communicated by A. Jones
1. Introduction
It is a generally accepted opinion that the first instance of conscious use of Complete Induction (henceforth CI) as a proof method is contained in the Traite du triangle
arithmetique by Blaise Pascal (16231662):
Quoique cette proposition ait une infinite de cas, jen donnerai une demonstration
bien courte, en supposant 2 lemmes.
Le 1, qui est e vident de soi-meme, que cette proportion se rencontre dans la seconde
base; car il est bien visible, que est a` comme 1 a` 1.
Le 2, que si cette proportion se trouve dans une base quelconque, elle se trouvera
necessairement dans la base suivante.
Do`u il se voit quelle est necessairement dans toutes les bases: car elle est dans la
seconde base par le premier lemme; donc par le second elle est dans la troisi`eme base,
donc dans la quatri`eme, et a` linfini.1

Pascal then proceeds to prove lemma 2, i.e. the inductive step.

A very different question is to establish whether, before Pascal, convincing examples of use of CI as a proof technique are attested, disconnected from the perception of
the fact that it happened to be a particular instance of a general demonstrative scheme.
Scholars have found proofs fitting in the scheme of CI in almost every geographical
ambit of pre-Pascalian mathematics.2 Some of these proposals, especially those concerning the ancient mathematical corpus, are untenable, even surprising in view of the

1
Pascal 1954, p. 103. The passage is contained in the Consequence Douzi`eme. The treatise
was composed towards the end of 1654 and published posthumously, together with earlier drafts
and related treatises, in 1665, with the title Traite du Triangle arithmetique, avec quelques autres
petits traites sur la meme mati`ere. We can find in it four proofs in the scheme of CI, with almost
identical wording and pattern of proof: see also pp. 113114, 122 and 150151 in Pascal 1954.
2
See Rashed 197273 for Arab mathematics (XXII century), Rabinovitch 1970 for the
works of Levi Ben Gerson (12881344). A pivotal point in the scholarly debate on this subject is the ascription of proofs by CI to Francesco Maurolico (14941575): see Vacca 1909
for the first proposal, and the subsequent discussions in Bussey 1917 and, in a deeper way, in
Freudenthal 1953.

58

F. Acerbi

serious misunderstandings they contain.3 Recently, the polemical tone of the debate has
attained a local maximum: in Fowler 1994 versus Unguru 1994 preconceptions of
historiographical method have strongly influenced the evaluation of the relevant instance
of CI. In this sense, both Fowlers and Ungurus analyses, though interesting in several respects, seem to me to have missed their target. Moreover, a careful survey of the
literature shows that every single scholar sets up his own reading of the principle of
CI, and on this basis he is able to affirm or to deny that specific proofs constitute well
formed examples of it.
My aim is to increase the confusion on this subject. In fact, I suggest to regard a
Platonic passage, i.e. Parmenides 149a7-c3, as a full-fledged example of proof by CI.
A comparison with an extremely similar text by Aristotle (An. pr. 42b6-16) will allow
me to put forward a conjecture about the sources of the Platonic text.
The paper is organized as follows: in Sect. 2 I discuss the meaning and structure of a
proof by CI, extracting from its ritual structure the characteristics I regard as essential.
In Sect. 3 the Platonic passage is presented and analysed. On the basis of the discussion
of Sect. 2, I propose to regard it as the only extant example of proof by CI in the ancient
mathematical corpus.4 In Sect. 4 I briefly discuss other alleged examples of CI in ancient
mathematical and philosophical works.
2. Principles of complete induction and proofs by complete induction
In order to clarify the meaning of the title of this section, compare the following two
deductive patterns (P(.) is a property of natural numbers):
P (1)
n N, P (n) P (n + 1)
a.
n N, P (n)
b. If some collection of natural numbers satisfies: i. the unit belongs to the collection,
and ii. the successor of any number in the collection also belongs to the collection,
then this collection will comprise all of the natural numbers.5
We could be tempted to regard them as equivalent formulations of CI, but the equivalence
involves assumptions or problems like:
3
See for instance the lapse by Stamatis in Euclides, vol. I, p. XXXV. I find unacceptable,
if only because lacking in a supporting discussion, statements like the following: Cependant on
peut trouver quelques demonstrations par recurrence ou induction compl`ete. On ne retrouvera
jamais le leitmotiv moderne, [. . .] et ceux qui ne voient linduction compl`ete quaccompagnee
de sa rengaine auront le droit de dire quon ne la trouve pas dans les Elements. Pour nous, nous
la voyons dans les prop. 3, 27 et 36, VII, 2, 4 et 13, VIII, 8 et 9, IX. [. . .] Dans tous les cas le
raisonnement est mene jusquau point o`u sa repetition mecanique assurera sa generalite. (Itard
1961, pp. 7374).
4
I prefer not to use the term ancient Greek mathematics. I intend to refer to a very concrete,
tangible object, not to a conceptual framework or to a specific way of doing mathematics. I agree
with O. Neugebauer that a concept as Greek mathematics [. . .] seems to me more misleading
than helpful. (Neugebauer 1969, p. 190.)
5
This is the fifth Peano axiom as given in Fowler 1994, p. 263.

Plato: Parmenides 149a7-c3. Complete Induction?

59

1. Every property defines a set, and the converse.

2. The concept of collection of all natural numbers does not give rise to problems
and can be freely used instead of quantification over every single natural number.
3. The difference between formalized and informal statements is not a substantial issue.
Moreover, it is crucial to remark that neither a. nor b. represent a formalization of the
Principle of CI (henceforth PCI): to this purpose it is necessary to premise a statement
specifying that the property P(.), or the collection we are dealing with, is generic.
If employed in connection with the ancient mathematical corpus, any of these definitions of CI is misleading: a consideration of the above very partial list of modern
problems they carry with themselves will suffice. It is thus necessary to refer to a definition formulated in a language which is the most informal possible, in particular with
respect to the problem of expressing the different levels of generality. To this end it is
useful to start with the claim by Stamatis: inductio completa primum ab Aristotele
=


< o oo ` o
commemorata est: `o oo `

o
, o

(Anal. Poster. 73b32).6 This is a serious and surprising mistake. In fact,

the statement is transparent, even considering the extreme density of the Aristotelian
style; we could translate it [a predicate] applies universally whenever it is shown [to
apply] in a generic case and in a primary way, where primary means that the relation
between subject and predicate must be regarded as immediate with respect to the properties which define the essence of the subject.7 In a proof formalized in a modern way,
the generality alluded to by the definition in An. post. 73b32 is secured by the presence
of universal quantifiers acting upon the free variables; in an informal context, generality
is guaranteed by recognizing that a proof which does not use some specific property
of a mathematical object holds also for every object of the same genus not sharing that
particular property.8 Explicit awareness of this fact,9 and its massive use in technical
contexts, are a fundamental achievement of Greek thought.
In the particular case of CI, its character of generality follows from the remark that,
both in the inductive step and in the conclusion, n acts as a free variable. Only in
this sense CI instantiates the Aristotelian prescription (which then assumes the status
of statement belonging to what is nowadays termed proof theory)10 : very much as
every general proof does, CI proceeds by instantiation and subsequent generalization
over the free variables (which, in other contexts, may even be geometrical figures, as
in the examples Aristotle provides during his proof). But it is completely erroneous
to maintain the converse, i.e. that An. post. 73b32 formalizes PCI.

Euclides, vol. I, p. XXXV.

The example provided in 73b33-39 to clarify the meaning of this statement is completely
unambiguous.
8
Problems could arise at the moment of specifying the most general genus of objects the
proof may be applied to (see also the discussion in Sect. 4).
9
See for instance Proclus 1873, 207.4-25. A recent analysis of the shaping of generality
in Greek mathematics is carried through in Netz 1999, chap. 6. I do not agree with some of his
conclusions, and I am currently working on this subject.
10
For Aristotle as proof theorist, see for instance Smith 1984, Smith 1986, or Smiley
1994 and references therein.
7

60

F. Acerbi

On this basis, I think the following sequence could correctly describe the series of
steps which are necessary to achieve a proof by CI:
A. To set the basis of induction.
B. To enunciate the inductive step and to explicitly recognize its generality ( universal
quantification An. post. 73b32).11
C. To prove the inductive step and to explicitly recognize the generality of the proof (
universal quantification An. post. 73b32).
D. To recognize the generality of the conclusion ( universal quantification An. post.
73b32).
Step A is often trivial; the real proof is developed in B and C. D is matter of formal
awareness, restricted to the context of one specific proof in the scheme of CI; it may be
expressed in an informal way: but it is crucial that its statement be explicit and clearly
verbalized.12 The sequence AD is not, a priori, beyond the possibilities of the ancient
mathematical corpus. But, as we have seen, an additional move is needed to recognize it
as a valid demonstrative scheme (i.e. to establish a sort of PCI): actually, a further application of An. post. 73b32 is enough, where the generalization is made on that specific property, whose applicability to every natural number had been proved in AD ( universal
quantification on the variable P(.) in version a. of CI). This step (of a metamathematical
character) is never explicitly made in the ancient mathematical corpus.13

11

In the following, I shall denote in this way the implicit or explicit ascription of generality
to a statement, whenever it results from a (possible) application of an argument of the kind of
An. post. 73b32. I do not claim the usage is always explicit (it is almost never explicit). It is a
convenient shorthand, and there is no anachronism to invoke it in dealing with Plato.
12
In the case of informal arguments, things go even worse: indeed, essendo lobiettivo quello
di riconoscere una regola logica in sequenze discorsive che non hanno mai la scansione netta dei
passaggi delle dimostrazioni matematiche, e` inevitabile ricorrere a ricostruzioni e precisazioni di
significati che sono atti interpretativi, e quindi in certa misura soggettivi (see p. 177 in Bellissima-Pagli 1996). The history of the logical rule called Consequentia Mirabilis ((A A) A;
it appears in Elem. IX.12, 36) reconstructed in that book has several points of contact with that of
CI.
13
Someone could (misleadingly) venture that the situation was not so clear even to Pascal:
indeed, in the passage we have quoted, he speaks of cette proposition and clearly does not
quantify on it i.e. he does not establish CI as a principle; moreover, a little earlier (Pascal
1954, p. 101; Consequence Huiti`eme) he proves a proposition using Incomplete Induction (Et
ainsi a` linfini); the same thing happens in Lemme IV, which immediately precedes Proposition I
(p. 113). Finally, it is very relevant that Pascal proves the validity of his demonstrative scheme
by CI appealing to an argument by Incomplete Induction: see the last paragraph of our quotation.
Equally relevant is the fact that he never proves his second lemme the inductive step for
a generic n; rather he deals with a particular case (for instance, from n = 4 to n = 5 in the proofs
at pp. 103 and 113114), of which he soon afterwards declares the generality: On le montrera
de meme dans tout le reste, puisque cette preuve nest fondee que sur ce que cette proportion se
trouve dans la base precedente, et que chaque cellule est e gale a` sa precedente, plus a` sa superieure,
ce qui est vrai partout. which is a paradigm instance of application of An. post. 73b32. Hence,
Pascal does not recognize independence or priority even intuitive to the argument by CI,
regarding it as a shorthand and a convenient form of formalization of the argument by Incomplete

Plato: Parmenides 149a7-c3. Complete Induction?

61

Very strong objections against the mere possibility that ancient Greek mathematics could develop single proofs by CI (even in absence of an explicit recognition of its
validity as a demonstrative scheme) have been raised in Unguru 1991.
The first argument, constituting in my opinion the core (pp. 280284) of his paper,
is summarized by the author as follows: without a number that can serve as an independent variable, it is impossible to formulate a true proof by mathematical induction,
in which the claim requiring proof is a function of the natural numbers (p. 284). This
happens because of the wide, and historically unbridgeable, gap between the Greek and
the modern conceptualization of number (p. 281). The latter is indeterminate, general,
abstract whilst the former is always a number of (p. 284). I think that, as it stands,
Ungurus argument must be rejected:
whenever it is supported only by mathematical sources, it is based on an undue application of inductive reasoning: from the fact that a well determined set is empty
(the sources analysed by Unguru in which CI is extant) it is inferred that every set
containing it is empty. As a consequence,
the author presents a series of meta-historical, meta-theorical, meta-textual, metacontextual arguments, providing an aprioristic direction to the analysis of every
possible textual evidence.14 These supporting arguments are based on a handful
of quotations from primary sources,15 whereas a detailed exegesis is devoted to
the relevant passages of Klein 1968, with further references to Wittgenstein and
W. C. Booth. This way, an easy historiographic hypothesis results, which the empirical basis provided by the ancient mathematical corpus will never be able to falsify.
The latter seems to be the main flaw of the quasi-intentional approach advocated
by Unguru (p. 289).
Even accepting Ungurus thesis as well founded, the obstruction arising from the
fact that the claim requiring proof is a function of the natural numbers still allows
for a way out: this happens when the function (i.e. the property P(.)) is very simple,
for example if it maps every number into itself, that is P (n) = n; or it is a slight
variation of the latter mapping, P (n) = n 1 for instance; or it is a combination of
both, such as P (n) = P (n) P (n) = 1. The arithmetical and meta-arithmetical
level are not easy to disentangle in this case: in order to use a function which maps
every integer n into itself it is not necessary to conceive the numbers in an abstract

Induction: compare the statements at p. 122 Quoique cette proposition ait une infinite de cas, je
la demontrerai neanmoins en peu de mots par le moyen de deux lemmes and at p. 150 Quamvis
infiniti sint hujus propositionis casus, [. . .], breviter tamen demonstrabo, positis duobus assumptis. (italics mine). These considerations are worthy of further elaboration, but here this hint will
suffice. See also the following note 21 and the well-balanced remarks made in Rashed 197273,
section IV.
14
It is not a coincidence that textual considerations, which I regard as crucial in the present
case, are almost absent from Ungurus paper.
15
Most (provided it is meaningful to make use of this term with such a poor sample) of them are
philosophical sources: Aristotle: Meth. 1020a13 and 1057a2-5; Alexander of Aphrodisias:
In Meth. 86, 56 (Hayduck). Luckily also Euclides: Elem. VII def. 2 is included.

62

F. Acerbi

way, nor to have any concept of function at all.16 In other words, Unguru fails
to distinguish between PCI and CI, and applies arguments to the latter which should
properly have been applied to the former.
The second objection (p. 278) is well summarized by the statement Its [of a CI]
conclusions [. . .] are valid for all natural numbers. Needless to emphasize that the concept of all natural numbers is foreign to Greek mathematics. The last claim is correct,
but its relevance to our problem is not clear: an easy way out is to make use of a wording
which does not refer to the set of all natural numbers, and to state that the property under
analysis is true for every single integer:17 a possibility that ancient Greek language easily
allowed for, see for instance Elem. IX.8.18
On the other hand, in Fowler 1994, which is an explicit confutation of Unguru
1991, reconstructions in the scheme of CI using techniques which are well rooted in
the ancient mathematical corpus are presented (pp. 254257), both of a proposition
which is not attested19 and of the well known relationship between side numbers sn and
diagonal numbers dn : (dn )2 2(sn )2 = 1.
Fowler supports his reconstructions by observing (pp. 261264) that: i. in the Elements use is made of the Principle of Descent (PD) (Elem. VII.31) and of the well-ordering principle (Elem. VII.31 A
)20 ; ii. these two principles are now generally,
though not universally, perceived as being equivalent to each other and to mathematical
induction. Remarking that before the nineteenth century, we find a more natural approach to the natural numbers, he concludes: It is in this sense that I believe that the
Greeks did use principles equivalent to mathematical induction, and that they could have
used mathematical induction itself had the situation presented itself. Moreover, such a

16

The arguments we can find in Mueller 1981 appear better balanced, although in many
respects close to those advocated by Unguru. Referring only to the Elements, he says that [. . .]
numbers are not characterized as generated from units in a serial order. They are simply finite ag<
< o

gregates of units. (paraphrase of Elem. VII def. 2:

o
o`
` o`

o) Of course, there are important relations between these aggregates, but the relation of
successor is not one which plays an important role in the Elements. Thus it can be said that the
integers themselves are not conceived in the structural way conducive to the use of induction, but
that there is inductive reasoning about collections or sequences of positive integers. This difference
might be compared to the difference in formal theories between the use of induction in the object
language and its use in the metalanguage (p. 69). In this way, Mueller denies that Euclides
could consciously use the principle of CI, but does not assert that could be impossible for him to
present proofs in the scheme of CI. In fact, not even this latter case does occur in the Elements (see
Sect. 4), and Mueller rightly points out that where Euclid proceeds in this quasi-inductive way,
the implicit induction is on the number of terms involved in a construction or assertion and not on
the integers themselves (p. 69). As we shall see, where Plato proceeds in a fully inductive way,
the explicit induction will be precisely of this kind.
17
The same difference exists between version a. and b. of PCI.
18
<

o


o

[scil.
o`] in Euclides II, pp. 194.17, 195.20.
19
If as many numbers as we please beginning from a unit be set out, their sum will be a
triangular number (p. 254).
20
This alternative proof is deemed interpolated by Heiberg, since it appears only in the Theonine manuscripts: ante prop. add. Theon (Euclides II, p. 237).

Plato: Parmenides 149a7-c3. Complete Induction?

63

situation did present itself, in the properties of figured numbers and side-and-diagonal
numbers [. . .] I believe it very probable that some earlier, now lost, treatment of some of
these topics, in particular the side-and-diagonal numbers, did in fact prove the described
properties [. . .] and that these proofs could very probably have been by some form
of mathematical induction.
I think that Fowlers argument must be rejected:
an equivalence between PCI and PD may be established only after the validity of both
of them as patterns of proof is accepted, and after an outline is provided of a proof
(even informal) of their equivalence. To require this, even before the 19th century, is
not a historical solecism but a necessary condition for the arguments consistency.
If this does not occur, so much the worse for the argument.21
Conjecturing is often necessary but the risk of losing contact with the extant textual
data is always present. As we have said, Fowler sets up (pp. 254255) the reconstruction, in an inductive frame, of a proposition not existing in the ancient mathematical
corpus but modelled [. . .] on Heaths translation of Elementa IX.36. Concluding
his analysis, he confesses that these kinds of arguments are not explicitly attested in
Greek mathematics, but the later commentators [. . .] give abundant though unsophisticated testimony to an earlier pebble (psephoi) arithmetic using these ingredients.
It is quite the contrary: the evidence that proofs of the kind he develops in his reconstruction are not attested is a strong argument against the presence of CI as a
widespread proof technique.

21

I think instead that the counter-objection in Unguru 1994, p. 271 cannot be regarded as
valid. He maintains that PD and PCI are not conceptually the same procedure, since PD deals
with a finite number of cases, while PCI covers an infinity of cases. In fact, even a proof by CI
explicitly deals with only a finite number of cases (indeed only one, in the proof of the inductive
step). Precisely the awareness of the quasi-general character of this proof introduces the infinitary
component in CI. But the same happens in PD, since this latter is a valid step in a deductive chain
exactly because of the character of generality granted to it by its being true for every decreasing
sequence of integers. In both kinds of proof one passes from instantiation to the general case
precisely via application of An. post. 73b32. Rather, I believe that PCI and PD cannot be regarded
as equivalent (I consider it obvious that they cannot be historically equivalent) for a different
series of reasons: they are not so 1) from the point of view of intuitive evidence which, when one
has to choose a proof technique instead of another, has a decisive weight; 2) since they involve
very different logico-mathematical concepts. We cannot presume that different concepts can be
grasped with equal ease (for instance the concept of decreasing sequence in PD or the notion
of property defined on an infinite collection of objects in PCI; or, as Mueller (p. 78) rightly
points out, unlike the principle of induction, the least number principle or the denial of infinitely
descending chains does not seem to depend upon a genuinely structural conception of the positive
integers. For these principles can be understood solely in terms of the Euclidean conception of
numbers as finite concatenations of units); 3) since every distinct technique employed to prove
the same theorem gives it a different meaning, (if nothing else) because it changes the net of
propositions needed to achieve the proof. Moreover, 4) to speak of equivalence is meaningful only
after PCI and PD have been recognised as such, and after logical techniques have been developed
to give sense to this equivalence. Even in this case, an equivalence proof can be higly nontrivial
(even worse if we want it formalized: one should take a look at 40 in Kleene 1952).

64

F. Acerbi

Summarizing, the conception of number we may extract from the ancient mathematical corpus seems to rule out the possibility that a principle of CI could be used.22 Indeed,
the latter is never attested, not even in a rudimentary stage.23 Proofs in the scheme of CI,
working on the number of terms involved in a statement and not on the integers as such
even proofs such that the steps from A to D above outlined are verbalized and formalized
in a way which is coherent with what we think to be the ontology underlying ancient
Greek mathematics seem to be well within the range of possibilities.24 Surprisingly,
the only full-fledged example, though very simple, does not appear in a mathematical
work, but in Platos Parmenides.

3. Analysis of Plato, Parmenides, 149a7-c3

The Platonic passage bracketed numerals are inserted for reference is the following:

< o
<,
<

=

(1) o

`o o
` `

<. 
. (2) E
=
 ,
< ` `

,
=

o o oo
o o

 `

o.

=
<
`
 o oo
o `
=
N. (3) K` o
`
`
o
, (4)
=
<
< o


`

o o

<. (5)
`

<
=
<

< `o

< `o `
` o
o >
`
=
<
=


,
(6)
` o

`
o

.

o
=
=

<
=
=

(7)
` `o o`o

.

o
`

< ` o
<
< 

=
< o

O. (8) =O

`
`o `
o,
`

< . <A.25

<`

(1) Then they must be two, at least, if there is to be contact. They must. (2) And if
to the two terms a third be added in immediate succession, they will be three, while the
contacts [will be] two. Yes. (3) And thus, one [term] being continually added, one
contact also is added, (4) and it follows that the contacts are one less than the number
of terms. (6) For the whole successive number [of terms] exceeds the number of all the
contacts as much (5) as the first two exceeded the contacts, for being greater in number
than the contacts: (7) for afterwards, when an additional term is added, also one contact
to the contacts [is added]. Right. (8) Then whatever the number of terms, the contacts
are always one less. True.26

22
If applied to the principle of CI, some of Ungurus arguments seem to be well-founded. But
he intended to apply them to the very different question of the existence of proofs in the scheme
of CI, and I think that the passage by Plato I shall presently discuss confutes them.
23
But the statement of the principle of CI must obviously have been preceded by more or less
explicit applications of CI, as is natural in the historical development of a mathematical object or
of a proof technique.
24
I completely agree with Muellers remarks quoted in note 16.
25
Greek text from Plato, Parmenides (Burnet).
26
I have reworked the translation in Plato 1926, p. 275.

65

Plato: Parmenides 149a7-c3. Complete Induction?

Remarks
a. The text is embedded in the discussion of the consequences of the second hypothesis Parmenides envisages in his inquiry about the nature of the one: if the one
exists (142b3).27 In particular, it is the main argument supporting the conclusion: the
one touches and does not touch itself and the others (149d6).
b. The reasoning has the following structure (the congruences with the scheme AD
outlined in Sect. 2 are conspicuous):
A. The basis of induction is established in (1). In (5) this very fact is explicitly
recognized (` o);
moreover,
` plays the role of pointing out that (1) or

(2) are crucial steps in the proof. As will become clear in (4), the property to be regarded
as relevant is the following: (number of terms) (number of contacts) = 1.
B. The validity of the inductive step is shown as evident in two particular cases

(n = 2 and n = 3) in (1) and (2) and stated in general in (5)(6), where o

.
. .
has the function to make generality explicit, even if, by variatio with respect to
(3) (see below), the attributive form substitutes the adverbial form. The transition from
(1)(2) to (5)(6) is very subtle: it constitutes the logical center of the argument and it
is verbalized in a slightly compressed form: it is worth a detailed discussion.
<


o

`
o presents two difficulties: the role of

and of .
The former is conveniently rendered as successive, but it may refer both to a generic successive integer, and to a successive in the sense of a repetition of the scheme
employed in (1)(2)28 and synthesized in (5), to which (6) is explicitly set in parallel:
in the same way in which n = 3 is obtained by adding in succession (o

 ) a third term to n = 2, o

<


`
o is what one obtains as successive from
the preceding case by adding a single unit. Such an interpretation is strengthened by the

explicit correlation ` . . . o

,
which links both the results and the proofs,
and which qualifies the number under consideration as a second term in a logico-temporal sequence whose paradigm instance is displayed in (2). The way is employed
provides further clues. Indeed, its use in conjunction with the article o, its postponement
and the position, correlated to the subsequent , give strong support to the choice
of translation the whole successive number (i.e. the whole aggregate of units which
constitutes it). In this way, (6) refers to the transition to the whole successive number
as far as it is obtained, as a whole, adding a further unit, in the way explained in (2) as a
particular case. To assume, instead, the meaning every for would make conclusion (8), already contained in (6) as a particular instance, completely useless. A further
element of generalization is present in (5)(6): it is not declared that in the first two


<


(142b3). It is not relevant to my purposes to discuss the question whether the

above translation provides the correct meaning of the Platonic expression.

28
In fact, the reference to the first two in (5) carries out a twofold function: first because
both cases verify the requested property, but also in the way in which this fact is proved to occur:
the case n = 2 is asserted to be true, while n = 3 is proved following the scheme of (2). And in
this scheme one passes from an integer to the next. I find unlikely that the first two can refer to
the first two terms, i.e. to the state of affairs described in (1): it would be difficult, for instance,
=
to explain the presence in (5) of

in the plural, since only one contact is present.

27

66

F. Acerbi

cases the difference must be equal to the unit; the property of invariance of the difference
is proved whichever the initial difference may be: in fact, the proof that follows does not
employ the fact that the initial difference is a well determined number (a further instance
of application of An. post. 73b32). This remark gives weight to my contention that the
first two in (5) is a reference to the logical and argumentative connection established
in (1)(2), rather than to a pleonastic duplication of the basis of the induction.
C. The proof of the inductive step and the statement of its generality are provided
< combines the idea of extratemporal validity with that of indefinitein (3), where
`
ly repeated operation; and in (7). The latter is explicitly connected with (5) and (6),

qua their proof, by the use of ,
` whereas .
. . `o o`o is employed in a sense
which abstracts from the primary (temporal) meaning of the words to assume a logical
function:29 `o o`o refers to a series of demonstrative steps as (2).
D. The generality of the conclusion is expressed in (8), which enlarges the field of
validity of (4) and slightly paraphrases it. In (8), the role of quantifier is played by the
< . . . . The particle

completely unambiguous phrase, = O. . .
`
establishes that (8)
is a logical consequence of what has been said before. We see then that several phrases
<
can work as quantifiers; among them, the more explicit ones employ the words
`
and , which will hold this technical meaning even in more formalized contexts.

I propose a concise interlinear, formalized in a modern way, of the Platonic passage. As should be clear, this rephrasing tendentiously separates the variable labelling
the property from the objects the latter applies to; as it stands, it strains the reading of
the Greek text, in which an induction on the number of terms in contact is employed,
not an induction on the natural numbers. Nevertheless, I think it worth giving since it
conveniently synthesizes the logical structure of the proof.
0. One needs to show that P (n) = P (n) 1, where P (n) = number of n terms
in contact, and P (n) = number of contacts of n consecutive terms (these symbols are
shorthands of the corresponding wording).
A. (1) If n = 2, P (2) = 2 and P (2) = 1.
=
 ),
B. (2) If n = 2 + 1 (E
` `
o o oo
o o

< ` `

)
=

P (2 + 1) = 3 (
and P (2 + 1) = 2 ( `

o).

(5)

. . o

and
Following the same scheme as in (2) (the correlatives ` o.
 o

), [in the more general case] in which P (2) P (2) = m (

 . . .

 ) and

hence P (2 + 1) P (2 + 1) = m, (6) [one obtains] also (`) P (n + 1) (o

<
=
 o
`
o ) P (n + 1) = (

) m (
).

=
< + participle/.

C. (3/7) Indeed (K` o
/
` ),
` for every n (
`
. . `o o
 o oo
o/
.
= . .
<
`o), P (n+1) = P (n)+1 (
`

o
) if
=
. .
. . . `) P (n+1) =
and only if (genitive absolute with temporal shade. . . `/.
=
=
=
P (n) + 1 (. . .
o
/

).

D. (4/8) It follows (`
/),
for every n (in explicit form it is absent,
but notice the use of the determinate article in the subsequent subordinate/ =O . . .


We could say a tempore transfertur
ad causarum vel ratiocinandi seriem et conse ).
quentiam (Bonitz 1870 sub voce
29

67

Plato: Parmenides 149a7-c3. Complete Induction?


< ` o
<
< ), P (n) = P (n) 1 (
=

`
`o `
o,
`
`

o o

<

< o
=
<
< ).
<

<

/

o

c. It is clear that this proof does not invoke a particular instance regarded as generic
(with a tacit reference to something like An. post. 73b32; most quasi-inductive proofs
contained in the Elements are framed in this way see below), or after which phrases
like and so on are called on. Rather, there are two proofs nested the one in the other:
(1)(4) set out a proof by CI of (4) (we have to include also this latter since in this case
the
plays the role of an absent o); (5)(7) explicitly state (5),
(6) and prove (7) the inductive step in full generality, and this requires in any case a
separated argument this fact is made apparent by the use of ,
` which introduces in
(5) the sub-proof. Moreover, notice that in (1)(4) the inductive step is verbalized (3) by
< ).
reporting a sketch of its proof (generalized through the use of
`
d. It is crucial, on the basis of our discussion in Sect. 2, that the numerical variable
labelling the property coincides with the concrete object (`o o) representing the
extension of the concept identified by the property itself, so that the induction is on
the number of terms in contact. This sort of merging is reflected in the terminological
=

ambiguity connected with the transition from the grammatical subjects o,
oo
(or a
30 Notice

<
gender neuter with implied subject) to o`
(in 67) and then back to o.
<
<

. We are faced with an
also the passage from o o
31 to o`
oscillation of meaning from item or term in a series to amount, sum, which makes
the presence of o unnecessary.32 Relevant in this sense is the use of a different
= versus ) to denote the one which is to be added to the terms or to the
gender (

contacts.

e. The whole argument does not imply the conclusion Plato draws from it in 149c4:
But if only the one exists, and not the dyad, there can be no contact. The latter is an
immediate consequence of (1), but cannot be deduced from (8), which holds for every
successive number. Yet, the passage employs a lexicon, a syntaxis and a logical linking
of the arguments revealing a good degree of formalization. This suggests that Plato
does not really master the proof he is elaborating, and that the latter is a (reworked)
excerptum from some other source. The following remarks point to the same end:
i. The proof is very simple and is expounded in a pedagogical way (the interlocutor
is a boy): the conclusion is drawn in (4) and repeated almost verbatim in (8). The only

30
Remark here the extraordinary writing technique: the transition occurs in (5), where the role
<
of o`
o` is ambiguous, since it can be construed both as an accusative of relation and
in this case the subject should be still ` o , and as the subject of the subordinate
<
o`

very much as in (6)(7).

31
<
I follow the lectio of the best manuscripts. Heindorfs emendation o`
o` does not
change the substance of the argument.
32
<
See LSJ sub voce
o` and recall also Elem. VII def. 2.

68

F. Acerbi

difficulties are of a logical order, and this is natural with a kind of proof probably
regarded as unusual.33 Recall, for the sake of comparison, that the property of ratios
established in Parmenides 154d1-7, very important for a reconstruction of pre-Euclidean
proportion theory,34 is stated without proof.
ii. The wording has several peculiarities: the form
+ infinitive is employed in the technical meaning of denoting a logical consequence. It is not an unicum
in the Platonic corpus,35 but in Parmenides there are no further occurrences of this con<
struction, a surprising fact if one envisages the character of the dialogue.36 The form
`
+ participle in 149b2 is well attested in the technical literature as standard wording
for an indefinitely repeated operation. It is important to notice that this wording is present
also in other relics of pre-Euclidean formalized mathematics.37 The impersonal use, and
without any moral hue, of
o

has no analogue in the Platonic dialogues, nor

is it attested in early technical works.38 I regard it as unlikely that this verb could have
been contained in the source; maybe it has been introduced by Plato as a variatio.39
Also the use of oo in this meaning is not attested as common in the technical literature, standardized on o. I suggest that in this case we do not have
a reworking by Plato, rather a testimony of a terminological field not yet well settled
(see further comments in the following point f).40
iii. A fraction of the text (5)(7) has a more abstract character: we can observe
<
a transition, in (6)(7), to a coherent use of `
o in order to denote the objects in contact; moreover, (5)(7) prove the truth of statement (4) as a particular case
<
 o
(

. . .
. . .
o

) and do not make use of the fact

 . . .

o

that the difference is equal to one. This suggests the existence of a source in which the
argument had been discussed at a greater degree of generality.

33

This fact explains its length, and the careful setting out of the relevant steps.
See Fowler 1987, pp. 4244, 64, 320.
35
See for instance Phaedo 74a2-4.
36
The verb occurs at several places in Parmenides. However, its presence is restricted to the short summaries Parmenides provides in order to pinpoint the development of the
discussion (see Parmenides 136ac, 137b4, 142bc, 160b5, 163c1; in 143d4 the verb assumes the
meaning to occur), and in them only the participle ` o, or an (indirect) interrogative phrase (such as ` `
) are employed. The relevant fact is that only in this
passage a relation of consequence is expressed through
+ infinitive; elsewhere, the
< o,
<


particles o
or the form
+ infinitive are used.
37
See Appendix IV (pp. 235244) in Knorr 1978, in the context of a detailed comparison
of the several forms of wording of the bisection principle. Knorrs analysis suggests that the
following texts can be ascribed to pre-Euclidean formulations: those in Elementa XII, most of
those extant in the Archimedean corpus and the Lemma to Sphaerica III. 9 of Theodosius. It
is more difficult to decide whether a direct quotation from Antiphon is present in the relevant
passage in Simplicius, In Phys. 55 (Diels).
38
For use in later authors, see e.g. Iamblichus, Theologoumena arithmeticae, 83.14 (De
Falco).
39
<
An indication in this direction is the presence in (5) of the accusative o`
o` , whose
function is to clarify in which sense the terms exceed the contacts.
40
See Aristoteles, Phys. 245a13 and 260a32 for a very similar usage of oo
in a non-mathematical context.
34

69

Plato: Parmenides 149a7-c3. Complete Induction?

=
in 149a8 is attested in the best manuscripts41 and provides us with
f. The term oo
a clue to set out a sensible conjecture about the sources of the Platonic passage. In fact,
let us read the following Aristotelian extract, in which bracketed numerals have been
inserted in order to make a comparison with the corresponding propositions in the text
of Parmenides easier:42

=





[. . . ] `o `
o o

` o

(>
`

. <
=

o
`
(4 )

`
>
< `o
o

o
oo
.
=
<
 o )

< `
 `
o

o,
=
<
<


o<
o <
`  `

o
o  `

o,
,
o
`

<

.
=
=
=


o

o` o  oo,
o
< o  oo
o,

`  o

= o
o,

=
> oo
o
=
(7 )

`
o
o oo,
o
<  `

<
<

= < (5 )

o  `

o` ,

(7 )
(6 )
< o

o
. 43

. . . the number of the terms will exceed that of the premisses, as before, by one (for
each further term which is introduced will be placed either externally or intermediately;
but in either case (4 ) it follows that the intervals are one fewer than the terms, and there
are as many premisses as intervals); the former will not, however, always be even and
the latter odd, but alternately when the premisses [are] even the terms [will be] odd, and
when the terms [are] even the premisses [will be] odd ((7 ) for wherever a term is added
one premiss is added as well together with the term), so that (5 ) since the premisses were
even and the terms odd, (6 ) they must change accordingly (7 ) when the same addition
is made to both.44

i. Aristotle is briefly discussing the relationship between number of terms, premisses and conclusions in the case of a complex syllogism. The logical structure of the
argument is incomplete, very compressed in the crucial points and excessively expanded
on trivial questions. The feeling is strong that Aristotle is reworking and mixing two
arguments, leaving unexpressed the most important parts of them. The simplest among
the arguments is reduced to the remark that when the premisses [are] even the terms
[will be] odd, and when the terms [are] even the premisses [will be] odd. A proof, very
short, is provided in (7 )(7 ); in the steps which are made explicit the proof mimics
the one contained in the Platonic text (see below). The second argument hinted at by
Aristotle refers to the more refined combinatoric analysis45 needed to explicitly determine the relationship between number of terms, premisses and conclusions in a syllogism
which contains n terms. Aristotle says: thus there will be many more conclusions

41
See the critical apparatus in Plato, ad loc. The lemma has been variously emended, if only
because its gender does not fit for the neuter of the subject o. This variatio can be easily
justified on stylistic grounds, but see the subsequent analysis.
42
The connection between the two passages had already been pointed out in Einarson 1936
(see note 56 on pp. 163164).
43
Aristoteles, An. pr. 42b6-16.
44
I have slightly reworked the translation in Aristotle 1938, p. 331.
45
There is no explicit trace of it in the Platonic text. But recall that the milieu of combinatoric
analysis has always been an ideal one for proofs in the scheme of CI, from Levi Ben Gerson to
Pascal.

70

F. Acerbi

than either terms or premisses.46 But the sketchy setting up of the problem, which is
developed in the text just quoted and in a few subsequent lines, shows an awareness of
some subtle aspects47 which is not counterbalanced by the triviality of the conclusion.48
ii. There are clear points of consonance with the Platonic passage. The demonstrative
pattern, though less cogent, is similar: compare the final sentences (7 ) and (7 ) with the
<
corresponding ones in Parmenides. The imperfect tense verb
which denotes the first
pair of even-odd quantities has a parallel in the unique instance of secondary sequence
<
verb

which is present in the Platonic passage. Syntactical and lexical

choices in (4 )/(4) and in (7 )/(7) are strictly related: notice the presence of
=
and ,

which implies the alterand the interchange of subject between oo

<

< o
<
<
nance

, parallel to

/
o

. Moreover,
=
in addition to the already commented on term oo,
the correspondence between `o
=
<
and o o

is remarkable. (7 ) and (7) are
o o
almost the same sentence, as written by Plato and by Aristotle.
iii. The dissonances are interesting as well. It is clear that (7 )(7 ) constitute in no
way a correct proof by CI. But we should recall that in the Aristotelian corpus almost
no proof is formalized in a convenient way, and could not have been, given the character
of the extant writings. On the other hand, Aristotle extensively employs arguments
which shorten and rework more refined mathematical proofs. The importance of the
above text, if read in strict parallel with the Platonic one, lies exactly in revealing the existence of mathematical argumentative patterns otherwise destined to be submerged. In
this perspective, it is relevant to remark the change of subject /o


displayed by the clause  `
o

o .

It is not easy to understand this fact without the assumption that Aristotle is in fact making reference
to a text or to a representation of the syllogisms as intervals and their terms: a text or
representation which should have been conveniently formalized, as the density of technical terms (see below) shows, and whose content Plato slightly reworks, in view of
his philosophical-stylistic aims. In my opinion a derivation of this kind is emphasized
=
by the transition from one term of the contraposition /

to the other. This

transition exploits the ambiguity of the first word: it is an interval separating the very
terms it links. Hence, it is not surprising that in Aristotle we can find a more standardized technical lexicon,49 whereas Plato preserves almost unchanged the logical

46

Aristoteles, An. pr. 42b25.

Notice for instance the careful separation just to declare the separation immaterial of
the cases in which the intervening term is added externally or in an intermediate position. This
Aristotelian strategy is not new: see, in a different context, Phys. VI.2, 233b2-11 (compare the
discussion in Knorr 1982, note 19 at p. 120).
48
This fact had already been recognized by Waitz: Hoc quidem certum est, sed miror, quod
in numerum conclusionum non amplius inquisivit. [. . .] Sit numerus propositionum = n, numerus
terminorum = n + 1, erit numerus conclusionum n(n 1)/2. (Waitz, Organon, I, ad 42b25,
p. 441). Now, this proposition can find a convenient formalization just in the way proposed in
Fowler 1994 (see note 19 above). I think this could be a good (albeit very tenuous) textual link
missing in Fowlers reconstruction.
49

The presence of the standard verbs

and o, instead of the highly

unusual
o

and oo, should be enough. In my opinion, this lexical nor47

Plato: Parmenides 149a7-c3. Complete Induction?

71

structure of the argument. Among the conceptual points Aristotle alludes to (relationship even-odd and possible relevance of the position of the interpolated term), Plato
does not deal with the second point, which could have been relevant for him (the first
is not relevant). Perhaps such an accessory discussion could have slowed down rhythm
and cogency of Parmenides reasoning.
g. Collecting the remarks made on both texts, the individuation of the environment
in which to place a possible Ur-text, as well as the contents of such an archetype, is
almost forced: a Pythagorean source is very likely. As a further point, I only remark the
coherent and systematic use, in the passage from the Analytica, of terms like ,

50
, 51 o, and the introduction, resulting again from Aris=
oo,
totle, in a context dealing with the properties of odd and even. A thorough discussion
of such a question would take us outside the scope of the present paper, for instance
to completely rethink the connection between Aristotelian syllogistic and Pythagorean
arithmetic, as well as to reconsider the alleged narrowness of interest and absence of
nontrivial results elaborated by ancient Greek mathematics in the field of combinatoric
analysis.52

4. Complete induction and the sorites

A certain number of propositions contained in ancient mathematical works have
been alleged as instances of proof by CI. I list here some of them:
Elem. VII. 3, 14, 17, 35; VIII. 13; IX. 8, 9, 20 (Stamatis)53 .

malization reflects a closer agreement with the original text, whose range of terminological possibilities easily included all the relevant terms (see next remark), rather than a process of refinement
of the technical lexicon which had occurred in the comparatively few years which separate
these works of Plato and Aristotle.
50
The importance of these two terms in the field of the musical theory inspired by Pythagorean
ideas and of the researches on the properties of ratios between integers is evident. It is enough to
read fragment B2 DK of Archytas, or the Sectio canonis. A classical discussion of the terminology can be found in Szabo` 1978, pp. 103119. For the peculiar lexicon introduced by Aristotle
in his theory of syllogism see for instance Einarson 1936 or Smith 1978, but the whole subject
is worth a more detailed analysis, which I plan to undertake elsewhere. The insertion of the Sectio
canonis in the Pythagorean tradition is a well-established fact. See for instance Barker 1981, or
Fowler 1987, pp. 143153.
51 <

is the technical term used when referring to the interpolation of a mean proportional. It is employed, for instance, in this sense in Elem. VIII. 810, 1921, 2427. It is present
also in prop. 3 an earlier version of which is ascribed to Archytas by Boethius of Sectio
canonis; and Aristotle makes use of it with this meaning in An. post. 84b12. For a full discussion
of Boethius translation of Archytas result and proof see Knorr 1975, pp. 212225.
52
But see Stanley 1997 for a first, strong indication in an exactly opposite direction. This
paper analyses Plutarch, Table-Talk VIII. 9, 732f. Contrary views are expressed in Biggs 1979,
while in Rome 1930 is said that an attentive depouillement des mathematici graeci minores
is needed.
53
See note 3.

72

F. Acerbi

Elem. VII. 3, 27, 36; VIII. 2, 4, 13; IX. 8, 9 (Itard)54 .

Proclus, In Platonis Rem Publicam, II. 27.129.4 (Kroll)55 . (Freudenthal)56 .
I do not want to embark on a detailed analysis of these texts: Freudenthal 1953 and
Unguru 1991 will suffice to this end. The answer is always the same: they are not
well formed examples of proofs by CI, not even in the very relaxed meaning of this
expression advocated in Fowler 1994.57 Typically, the proofs are framed in a quasigeneral way, with instantiation of a few cases and subsequent (implicit) appeal to the
generality of the scheme of proof.58 In several instances, notably some among those
connected with the property of side and diagonal numbers, the proof itself is altogether
lacking.
I shall focus my attention, instead, on a different issue. I take as a datum, on the basis
of the analysis carried on in the preceding sections, that proofs in the scheme of CI had
actually been developed in a pre-Euclidean arithmetical context, and that their demonstrative structure could have represented (as is attested by the care Plato employs in
stressing the relevant steps) a sort of logical challenge. I assume as a further datum that
proofs by CI are absent from later mathematical works, and that the absence of CI, even
in contexts in which we should regard as natural its application, is a strong indication
that the generality of this proof technique had not been understood. I leave to others
the task of explaining this fact.59 Rather, I shall now briefly discuss a piece of evidence
supporting my contention that proofs by CI actually existed but that their potentialities
were misunderstood. Diogenes Laertius and Sextus Empiricus report the following
forms of the paradox known as the Sorites:60
It cannot be that if two is few, three is not so likewise, nor that if two or three are few,
four is not so; and so on up to ten. But two is few, therefore so also is ten.61
[T]he Sage [. . .] will assent to Fifty-one is few; for there is nothing between this representation and that of Fifty is few. But as Fifty is few was the apprehensive repre-

54

See note 3.
Or Theon of Smyrna, Expositio Rerum ad Legendum Platonem Utilium, 42.1045.8 (Hiller), or Iamblichus, In Nicomachi Arithmeticam Introductionem Liber, 91.393.6 (Pistelli-Klein).
These passages discuss the construction of successive side and diagonal numbers.
56
See note 2.
57
But fragments like Archytas A24 DK and Zenos B1 DK could be worth a more detailed
discussion.
58
This step of the proof is usually verbalized through the repetition of the relevant part of the
o , preceded by the adverb o o. A paradigm instance of this way of proceeding is
=
< ` o
represented by Elem. IX. 8. Proclus concludes his quasi-general proof with `

(II. 29.4).
59
I only remark that textual tradition has been strongly selective in this respect, having almost
completely erased information on what we could call techniques of combinatoric analysis, even
developed in a very simple form (but see note 52 for textual support to the fact that techniques of
this kind should have existed).
60
I am indebted to Lucio Russo for illuminating discussions on this point. For a thorough analysis of the Sorites in the Hellenistic philosophical tradition see Barnes 1982 and Burnyeat
1982.
61
English translation from Diogenes Laertius, II, p. 191 (VII. 82).
55

Plato: Parmenides 149a7-c3. Complete Induction?

73

sentation placed last in order, Fifty-one is few is the first non-apprehensive one. The
Good Man, therefore, will assent to the non-apprehensive representation Fifty-one is
few. And if he will assent to this as being in no wise different from Fifty is few, he will
assent also to the non-apprehensive Ten thousand is few; for every non-apprehensive
representation is equal to every other non-apprehensive representation.62

Remarks
1. In the above examples the numerical variable labelling the property coincides
with the objects the property directly applies to. This fact links the object language
with the metatheorical level in the same way as we have observed in Plato. Notice also
that in this case the numbers involved are never numbers of.
2. The property under consideration (very often a variation on P (n) n is few) is
what we could call nowadays a vague concept.63 Clearly, CI does not work for vague
concepts. The difficulties connected with such a kind of predicates, and an inability to
clearly circumscribe the most general genus of predicates for which CI does work could
have resulted in (explicitly or implicitly) deeming this type of reasoning unreliable in
the context of a proof. The coherent use of quasi-general proofs in books VII-IX of the
Elements is, very likely, an editorial choice which entailed a standardization of later
works. Moreover, the paradoxes themselves could have been built, and can be construed
as, an explicit critique to inductive proofs: they display, exploiting the weak point of the
argument by unduly stretching the set of allowed predicates, the typical features of an
argument ad hominem.64
3. Plato sets up a correct proof by CI but is wrong in its application. He somehow
reverses the ascending direction of the argument, referring the property just proved to
an integer which cannot be reached by the induction. Some versions of the Sorites
make appeal to a heap (the Greek name has this origin), from which a unit is subtracted
in succession.65 Here the vague concept is that of heap, and the inductive chain is

62

English translation (slightly modified) from Sextus Empiricus, p. 225 (Adv. Math. VII.
418419). Other loci in Sextus are Adv. Math. IX. 182190, Hyp. Pyrrh. II. 253, III. 80. See also
Cicero, Acad. II. 93, Galenus On Medical Experience 17.117.3 (extant only in Arabic; cited
in Long-Sedley 1990, vol. 1, pp. 222223). For a complete list see Barnes 1982.
63
The Sorites itself often has a new name: it is called Wangs paradox, the only difference
being that it is now stated with an explicit use of CI, and so the conclusion n is few follows
for every integer. In my opinion, this is an instance of the modern tendency to attach a content to
formal manipulations: Wangs paradox clearly displays an undue application of CI. The Sorites,
insofar as it does not make use of CI, only allows for a comparison of a pair of well determinate
integers. For a discussion of Wangs paradox, see for instance Dummett 1978. See also The
Monist, LXXXI (1998), dedicated to Vagueness.
64
Recall that Eubulides of Miletus is credited to have put forward arguments in dialectical
fashion, among which the Sorites is included. Eubulides was contemporary of Aristotle and
pupil of Euclides, the founder of the Megarian school (see Diogenes Laertius, I, p. 236 (II.
108109)).
65
See Horatius, Epistulae II. 1.34 ff., Galenus On Medical Experience 20.3.

74

F. Acerbi

descending. Clearly, a descending inductive chain is nothing more than a shorthand,

standing for a finite number of identical steps. Failure of distinguishing between ascending and descending inductions amounts to a weakening of the logical independence
of CI, and reduces it to a notational device, eventually superseded by the method of
quasi-general proofs.
The preceding remarks strongly suggest that the Sorites paradoxes originate from
a (tendentiously) wrong use of a mathematical demonstrative technique, which disappeared early from the available supply of proof schemes: CI, in the form we have seen
at work in the Parmenides, is a good candidate for it.

Acknowledgments. I thank Lucio Russo for his comments on an earlier version of this work,
Giuseppe Frappa for the collaboration. I am also grateful to Alexander Jones for the suggestions.

Sources and works cited

Archimedes, Opera mathematica. Ed. I. L. Heiberg. 2 ed. 3 voll. Leipzig, B. G. Teubner, 191015.
(Reprinted 1972)
Aristoteles, Analytica priora et posteriora. Recensuit brevique adnotatione critica instruxit
W. D. Ross, praefatione et appendice auxit L. Minio-Paluello. Oxford, Oxford University
Press, 1964.
Aristoteles, Physica. Recensuit brevique adnotatione critica instruxit W. D. Ross. Oxford, Oxford University Press, 1950.
Aristotle 1938, Prior Analytics. Edited and translated by H. Tredennick. Cambridge (Mass.),
Harvard University Press.
Barker A. 1981, Methods and aims in the Euclidean Sectio Canonis. Journal of Hellenic Studies,
101, pp. 116.
Barnes J. 1982, Medicine, Experience and Logic. In Science and Speculation. Edited by
J. Barnes, J. Brunschwig, M. F. Burnyeat, M. Schofield. Cambridge, Cambridge University Press, pp. 2468.
Bellissima F. - Pagli P. 1996, Consequentia mirabilis. Firenze, Leo S. Olschki.
Biggs N. L. 1979, The roots of combinatorics. Historia Mathematica, 6, pp. 109136.
Bonitz H. 1870, Index Aristotelicus. Ed. H. Bonitz. Vol. V in Aristotelis Opera. Ed. I. Bekker.
Berlin (Academia litterarum regia borussica), G. Reimer. (Reprinted: Berlin, De Gruyter,
1961)
Burnyeat M. F. 1982, Gods and heaps. In Language and Logos. Edited by M. Schofield, M. C.
Nussbaum. Cambridge, Cambridge University Press, pp. 315338.
Bussey W. H. 1917, The Origin of Mathematical Induction. American Mathematical Monthly,
24, pp. 199207.
Cajori F. 1918, Origin of the name Mathematical Induction. American Mathematical Monthly,
25, pp. 197201.
Diogenes Laertius, Lives of Eminent Philosophers. With an English translation by R. D. Hicks.
Cambridge (Mass.), Harvard University Press, 1931.
Dummett M. 1978, Wangs paradox. In Truth and Other Enigmas. London, Duckworth,
pp. 248268.
Einarson B. 1936, On Certain Mathematical Terms in Aristotles Logic. The American Journal
of Philology, LVII, pp. 3354; 151172.

Plato: Parmenides 149a7-c3. Complete Induction?

75

Euclides, Elementa. Post I. L. Heiberg edidit E. S. Stamatis. 5 voll. Leipzig-Stuttgart, B. G.

Teubner, 19691977.
Fowler D. 1987, The Mathematics of Platos Academy. Oxford, Clarendon Press.
Fowler D. 1994, Could the Greeks Have Used Mathematical Induction? Did They Use It? Physis,
XXXI, pp. 253265.
Freudenthal H. 1953, Zur Geschichte der vollstandigen Induktion. Archives Internationales
dHistoire des Sciences, 6, pp. 1737.
Itard J. 1961, Les livres arithmetiques dEuclide. Paris, Hermann.
Klein J. 1968, Greek Mathematical Thought and the Origin of Algebra. Cambridge (Mass.), MIT
Press.
Kleene S. C. 1952, Introduction to Metamathematics. Amsterdam, North-Holland; New York,
Van Nostrand.
Knorr W. R. 1975, The Evolution of the Euclidean Elements. Dordrecht-Boston, D. Reidel.
Knorr W. R. 1978, Archimedes and the Pre-Euclidean Proportion Theory. Archives Internationales dHistoire des Sciences, 28, pp. 183244.
Knorr W. R. 1982, Infinity and Continuity: The Interaction of Mathematics and Philosophy in Antiquity. In Infinity and Continuity in Ancient and Medieval Thought. Edited by N. Kretzmann.
Ithaca, Cornell University Press, pp. 112145.
Long A. A. - Sedley D. N. 1990, The Hellenistic Philosophers. Cambridge, Cambridge University Press.
LSJ H. G. Liddell, R. Scott, H. S. Jones, A Greek-English Lexicon. Oxford, Clarendon Press
1983.
Mueller I. 1981, Philosophy of Mathematics and Deductive Structure in Euclids Elements.
Cambridge (Mass.), MIT Press.
Netz R. 1999, The Shaping of Deduction in Greek Mathematics. Cambridge, Cambridge University Press.
Neugebauer O. 1969, The Exact Sciences in Antiquity. New York, Dover.

Pascal B. 1954, uvres Compl`etes. Edition

e tablie et annotee par J. Chevalier. Paris, Gallimard.
Plato 1906, Parmenides, in Platonis Opera, vol. II, recognovit brevique adnotatione critica instruxit J. Burnet. Oxford, Oxford University Press.
Plato 1926, Cratylus, Parmenides, Greater Hippias, Lesser Hippias. With an English translation
by H. N. Fowler. Cambridge (Mass.), Harvard University Press.
Proclus 1873, In Primum Euclidis Elementorum Librum Commentarii. Ex recognitione G. Friedlein. Leipzig, B. G. Teubner. (Reprinted: Hildesheim, Georg Olms Verlag 1992)
Rabinovitch N. L. 1970, Rabbi Levi Ben Gerson and the Origins of Mathematical Induction.
Archive for History of Exact Sciences, 6, pp. 237248.
Rashed R. 197273, Linduction mathematique: al-Karaji, as-Samawal. Archive for History of
Exact Sciences, 9, pp. 121.
Rome A. 1930, Procedes anciens de calcul des combinaisons. Annales de la Societe scientifique
de Bruxelles, A 50, pp. 97104.
Sextus Empiricus, Against the Logicians. With an English translation by R. G. Bury. Cambridge
(Mass.), Harvard University Press, 1935.
Smiley T. 1994, Aristotles Completeness Proof. Ancient Philosophy, 14, pp. 2538.
Smith R. 1978, The Mathematical Origins of Aristotles syllogistic. Archive for History of Exact
Sciences, 19, pp. 201209.
Smith R. 1984, Aristotle as Proof Theorist. Philosophia Naturalis, 27, pp. 590597.
Smith R. 1986, Immediate Propositions and Aristotles Proof Theory. Ancient Philosophy, 6,
pp. 4768.
Stanley R. P. 1997, Hipparchus, Plutarch, Schroder, and Hough. American Mathematical
Monthly, 104, pp. 344350.

76

F. Acerbi

` A. 1978, The Beginnings of Greek Mathematics. Dordrecht-Boston, D. Reidel.

Szabo
Unguru S. 1991, Greek Mathematics and Mathematical Induction. Physis, XXVIII, pp. 273289.
Unguru S. 1994, Fowling after Induction. Physis, XXXI, pp. 267272.
Vacca G. 1909, Maurolycus, the first discoverer of the principle of mathematical induction.
Bulletin of the American Mathematical Society, XVI, pp. 7073.
Waitz Th., Organon graece, novis codicum auxiliis adiutus recognovit, scholiis ineditis et commentario instruxit Th. Waitz, 2 vols., Leipzig 184446. (Reprinted: Aalen 1965)
Dipartimento di Matematica
Universit`a di Roma Tor Vergata
via della Ricerca Scientifica
00133 Roma
Italy
[email protected]
(Received March 13, 2000)