In the Peano concept-script the judgements necessary for this are not reflected at all, and so it cannot provide a way of
checking
them.
Gottlob-Frege-Posthumous-Writings
To explain how thinking and judging take place is certainly a feasible undertaking, but it is not a logical one.
Accordingly, the logician does not have to ask what course thinking naturally takes in the human mind. What is natural to one man may well be unnatural to another. The great difference between grammars itself bears witness to this. The logician need fear nothing less than to be reproached with the fact that his statements do not accord with how we think naturally. The normal person with no training in mathematics would find it highly unnatural if he were to have the rudiments of the subject explained to him in terms of the utmost rigour, and for that very reason. A prudent teacher will therefore tend to let rigour go by the board in introducing the subject and will only seek to awaken the need for it bit by bit. Even in the history of mathematics we find that the highest degree of rigour is achieved only towards the end and that consequently it is at the farthest removed from what is natural. Hence to strive to present the process of thinking in its natural form would lead us directly away from logic. If the logician tried to take account of objections on the score that what he said was unnatural, he would be in danger of involving himself in endless disputes over what is natural-disputes which logic is quite incapable of resolving on its own grounds and which, therefore, do not belong to logic. To resolve them we should presumably have to resort to observing primitive peoples.
But above all we should be wary of the view that it is the business of logic to investigate how we actually think and judge when we are in agreement with the laws of truth. If that were so, we should have constantly to have one eye on the one thing and one eye on the other, and continue paying attention to the latter whilst taking a sidelong glance at the former, and in the process we should easily lose sight of a definite goal altogether. We should be seduced into asking questions with no clear meaning and as a result a satisfactory outcome to our investigations would be as good as impossible.
What are often called laws of thought, namely laws in accordance with which judging, at least in normal cases, takes place, can be nothing but laws for holding something to be true, not laws of truth. If a man holds something to be true-and the psychological logicians will surely hold that their own statements at least are true-he thereby acknowledges that there is such a thing as something's being true. But in that case it is surely probable that there will be laws of truth as well, and if there are, these must provide the norm for holding something to be true. And these will be the laws of logic proper. In supplement No. 26 to the 1897 Proceedings of the Allgemeinl Zeitung, T. Achelis writes in a paper entitled 'Volkerkunde und Philosophie': :But we are now clear about this, that the norms which hold in general for thinking and acting cannot be arrived at by the one-sided exercise of pure deductive abstraction alone; what is required is an empirico?
? Logic 147
critical determination of the objective principles of our psycho-physical organization which are valid at all times for the great consciousness of mankind. '
It is not quite clear whether this is about laws in accordance with which judgements are made or about laws in accordance with which they should be made. It appears to be about both. That is to say, the laws in accordance with which judgements are made are set up as a norm for how judgements are to be made. But why do we need to do this? Don't we automatically judge in accordance with these laws? No! Not automatically; normally, yes, but not always! So these are laws which have exceptions, but the exceptions will themselves be governed by further laws. So the laws that we have set up do not comprise all of them. Now what is our justification for isolating a part of the entire corpus of laws and setting it up as a norm? To do that is like wanting to present Kepler's laws of planetary motion as a norm and then being forced, alas, to recognize that the planets in their wilfulness do not behave in strict conformity with them but, like spoilt children, have disturbing effects on one another. Such behaviour would then have to be severely reprimanded.
On this view we shall have to exercise every care not to stray from the path taken by the solid majority. We shall even mistrust the greatest geniuses; for if they were normal, they would be mediocre.
With the psychological conception of logic we lose the distinction between the grounds that justify a conviction and the causes that actually produce it. This means that a justification in the proper sense is not possible; what we have in its place is an account of how the conviction was arrived at, from which it is to be inferred that everything has been caused by psychological factors. This puts a superstition on the same footing as a scientific discovery.
If we think of the laws of logic as psychological, we shall be inclined to raise the question whether they are somehow subject to change. Are they like the grammar of a language which may, of course, change with the passage of time? This is a possibility we really have to face up to if we hold that the laws of logic derive their authority from a source similar to that of the laws of grammar, if they are norms only because we seldom deviate from them, if it is normal to judge in accordance with our laws of logic as it is normal to walk upright. Just as there may have been a time when it was not normal for our ancestors to walk upright, so many modes of thinking might have been normal in the past which are not so now, and in the future something might be normal that is not so at the present time. In a language whose form is not yet fixed there are always points of grammar on which our sense of idiom is unreliable, and a similar thing would have to hold in respect of the laws of logic whenever we were in a period of transition. We might, for instance, be in two minds whether it is correct to judge that every object is identical with itself. If that were so, we should not really be entitled to speak of logical laws, but only of logical rules that specify what is
? 148
Logic
regarded as normal at a particular time. We should not be entitled to express such a rule in a form like 'Every object is identical with itself' for there is here no mention at all of the class of beings for whose judgements the rule is meant to be valid, but we should have to say something like 'At the present time it is normal for men-with the possible exception of certain primitive peoples for whom the matter has not yet been investigated-to judge that every object is identical with itself'. However, once there are laws, even if they are psychological, then, as we have seen, they must always be true, or better, they must be timelessly true if they are true at all. Therefore if we had observed that from a certain time a law ceased to hold, then we should have to say that it was altogether false. What we could do, however, is to try to find a condition that would have to be added to the law. Let us assume that for a certain period of time men make judgements in accordance with the law that every object is identical with itself, but that after this time they cease to do so. Then the cause of this might be that the phosphorus content in the cerebral cortex had changed, and we should then have to say something like 'If the amount of phosphorus present in any part of man's cerebral cortex does not exceed 4%, his judgement will always be in
accordance with the law that every object is identical with itself. '
We can at least conceive of psychological laws that refer in this way to the chemical composition of the brain or to its anatomical structure. On the other hand, such a reference would be absurd in the case of logical laws, for these are not concerned with what this or that man holds to be true, but with what is true. Whether a man holds the thought that 2 ? 2 = 4 to be true or to be false may depend on the chemical composition of his brain, but whether this thought is true cannot depend on that. Whether it is true that Julius
Caesar was assassinated by Brutus cannot depend upon the structure of Professor Mommsen's brain.
People sometimes raise the question whether the laws of logic can change with time. The laws of truth, like all thoughts, are always true if they are true at all. Nor can they contain a condition which might be satisfied at certain times but not at others, because they are concerned with the truth of thoughts and if these are true, they are true timelessly. So if at one time the truth of some thought follows from the truth of certain others, then it must always follow.
Let us summarize what we have elicited about thoughts (properly so? called).
Unlike ideas, thoughts do not belong to the individual mind (they are not subjective), but are independent of our thinking and confront each one of Ul in the same way (objectively). They are not the product of thinking, but are only grasped by thinking. In this respect they are like physical bodies. What distinguishes them from physical bodies is that they are non-spatial, and we could perhap~ really go as far as to say that they are essentially timeless-at least inasmuch as they are immune from anything that could effect a chanae in their intrinsic nature. They are like ideas in being non-spatial.
Since thoughts are not mental in nature, it follows that every
? Logic 149
psychological treatment of logic can only do harm. It is rather the task of this science to purify logic of all that is alien and hence of all that is psychological, and to free thinking from the fetters of language by pointing up the logical imperfections of language. Logic is concerned with the laws of truth, not with the laws of holding something to be true, not with the 4uestion of how men think, but with the question of how they must think if they are not to miss the truth.
Negation
1\ thought proper is either true or false. When we make a judgement about a thought, then we either accept it as true or we reject it as false. The last expression, however, can mislead us by suggesting that a thought which has been rejected ought to be consigned to oblivion as quickly as possible as being no longer of any use. On the contrary, the recognition that a thought 1s false may be just as fruitful as the recognition that a thought is true. Properly understood, there is no difference at all between the two cases. To hold one thought to be false is to hold a (different) thought to be true-a thought which we call the opposite of the first. In the German language we usually indicate that a thought is false by inserting the word 'not' into the predicate. But the assertion is still conveyed by the indicative form, and has no necessary connection with the word 'not'. The negative form can be retained although the assertion has been dropped. We can speak equally well of 'The thought that Peter did not come to Rome' as of 'The thought that Peter came to Rome'. Thus it is clear that when I assert that Peter did not come to Rome, the act of asserting and judging is no different from when I assert that Peter did come to Rome; the only difference is that we hnvc the opposite thought. So to each thought there corresponds an oppo- Nite. Here we have a symmetrical relation: If the first thought is the opposite of the second, then the second is the opposite of the first. To declare false the thought that Peter did not come to Rome is to assert that Peter came to Rome. We could declare it false by inserting a second 'not' and saying 'Peter did (not) not come to Rome' or 'It is not true that Peter did not come to Rome'. And from this it follows that two negatives cancel one another out. If we take the opposite of the opposite of something, we have what
we began with.
When it is a question of whether some thought is true, we are poised
hctwcen opposite thoughts, and the same act which recognizes one of them IIN true recognizes the other as false. Analogous relations of opposition hold In other cases too, e. g. between what is beautiful and ugly, good and bad, rtcusant and unpleasant and positive and negative in mathematics and Jlhysics. But these are different from our case in two respects. In the first ptncc there is nothing here which, like nought or a neutrally charged body, occupies a mean between opposites. We can of course say that, in relation lo positive and negative numbers, nought is its own opposite, but there is no
? 150 Logic
thought which could count as its own opposite. This is true even of fict- ions. In the second place we do not have here two classes such that one contains thoughts that are the opposite of those contained in the other, as there is a class of positive and a class of negative numbers. At any rate, I have not yet found any feature that could be used to effect such a division; for the use of the word 'not' in ordinary language is a purely external cri- terion and an unreliable one at that. We have other signs for negation like 'no', and we often use the prefix 'un' as, for example, in 'unsatisfactory'. Now in view of the fact that 'unsatisfactory' and 'bad' are very close in sense to one another, there would surely appear to be little point in wishing to assign the thoughts contained in 'This work is bad' and 'This work is satis- factory' to the first class and those contained in 'The work is not bad' and 'The work is unsatisfactory' to the second class, and it may well be the case that in some other language the word for 'unsatisfactory' is one in which a negative form is no more to be discerned that it is in 'bad'. We cannot define in what respect the first two thoughts might be supposed to be more closely related than the first and fourth. There is the further fact that negatives can occur elsewhere than in the predicate of a main clause, and that such nega- tives do not simply cancel each other out: an example would be 'Not all pieces of work are unsatisfactory'. We cannot put for this 'All pieces of work are satisfactory', and nor can we put 'Whoever has worked hard is rewarded,
for 'Whoever has not worked hard is not rewarded'. If we compare these with the sentences 'Whoever is rewarded has worked hard', 'Whoever has not worked hard goes away empty-handed', 'Whoever has been idle is not rewarded', '24 is not different from 42' and '24 is the same as 42', we shall see that we are tangling with some thorny problems here. What is more it is simply not worth while to try to extricate ourselves, and to expend a great deal of effort on finding answers to them. I at any rate know of no logical law which would take cognizance of a division of thoughts into positive and negative. We shall therefore leave the matter to look after itself until such time as it should become clear to us that such a division is somehow necessary. At which juncture we should naturally expect a criterion to emerge by which the division could be effected.
The prefix 'un', is not always used to negate. There is hardly any difference in sense between 'unhappy' and 'miserable'. Here we have a word which is the opposite of 'happy', and yet is not its negation. For this reason the sentences 'This man is not unhappy' and 'This man is happy' do not have the same sense.
Compound Thoughts
If the jury ret~rn a 'Yes' to the question 'Did the accused wilfully set fire to a pile of wood and (wilfully) start a forest fire? ' they have simultaneously asserted the two thoughts:
? Logic 151 (1) The accused wilfully set fire to a pile of wood,
(2) The accused wilfully started a forest fire.
It is true that our question contains one thought, for it can be answered by making one judgement; but this thought is composed of two thoughts-each of these being capable of being judged on its own-in such a way that by affirming the whole thought I affirm at the same time the component thoughts. Now it might seem that this is really neither here nor there and that the matter is of little importance; but it will become evident that it is closely bound up with very important logical laws. This comes out more clearly as soon as one considers the negation of such a compound thought. When will the jury have to say 'No' to the question above? Obviously the moment they hold but one of the two component thoughts to be false; for example, if they are of the opinion that whilst there is no doubt that the accused wilfully set fire to the pile of wood, he did not intend as a consequence that the forest should catch fire. '
? ? The Argument for my stricter Canons of Definition1 [1897/98 or shortly afterwards]
Since the necessity for my stricter canons of definition may not yet be sufficiently evident from what I said on the matter in the first part,2 and since, as it appears, opinion is still greatly divided on this matter, I want to try to argue for my view once again, by drawing a comparison with Peano's mathematical logic. What is at stake here is perhaps the deepest difference between the two concept-scripts. I have drawn attention to certain shortcomings which struck me in the definition of the Peano deduction-sign ':::::>',which roughly corresponds to my 't'. These shortcomings are certainly due to the fact that, whereas Peano takes a first step in my direction, he doesn't take the second. His definition of the sign ':::::>' for the case where it stands between sentences containing no indefinitely indicating letters (lettres variables) agrees in essentials with my definition of the sign 't'; but for the case where indefinitely indicating letters (lettres variables) occur on both sides of the sign ':::::>', another definition is given whose relation to the first remains unclear. In my case the combination of condition and consequence is presented as analysed into two components, of which the one is generality, signified by the use of roman letters, while the other is designated by the sign 't'. This analysis cannot be clearly discerned in Peano's case because the use of lettres variables and the sign ':::::>' are both defined simultaneously. In his most recent account, Peano has attempted to remove this shortcoming, but unfortunately, not by following me in taking my
1 The text was presumably composed in 1897/8 or shortly thereafter: right at the beginning Frege mentions having 'drawn attention to certain shortcomings in the definition of the Peano deduction-sign'. This must refer to the article of Frege: Ober die Begri. ffsschrift des Herrn Peano und meine eigene that appeared in 1897. Later on Frege talks of Peano's attempts in his most recent account to overcome these shortcomings. From the details ofwhat Frege says, this must refer to the definition of'-::;,' in Formulaire de Mathematiques' Vol. 11, p. 24. which appeared at the end of 1897. (The corresponding definition in Form. Vol. I, pp. V/VI, No. 17 is still in the form that Frege criticized in his article) (ed. ). It seems to us a natural conjecture that the present article and the next are preliminary drafts of material intended for inclusion in Volume 11 of the Grundgesetze: this both because of the overlap in content between parts of this article and early parts of Volume 11 and above all because of the way in which Grundgesetze Volume I is simply referred to as 'the first part' or 'my first volume' (trans. ).
2 As the editors point out, the reference here is clearly to the rirst part of the Grundgesetze (trans. ).
? The Argumentfor my stricter Canons ofDefinition 153
second step, but by retracting the first. He completely abandons the definition of the deduction-sign which agrees in essentials with mine, and now completely excludes this case-the one where '"::J' stands between sentences containing no indefinitely indicating letters. But there surely do occur cases where even in speech the form of a hypothetical sentence ('if . . . then . . . ')is used, without the content being general. Let us, however, waive this objection.
As an example, let us take a look now at the Peano sentence 'u, v e K. fe vfu. ]e ufv . ::J. num u = num v',
which corresponds in essentials to my ? ~nu= nv '
u'"' (v'"' ) p) v'"' (u'"' ) $. P)
Admittedly in my case the two Peano conditions u, v e K are missing, but this does not make my sentence false. Rather, my signs are so defined, that, without the truth of the claim being put in jeopardy, names of objects other than classes can also be substituted for u and v. In this formula u, v e K is meant to state that by u and v classes are to be understood. The other two conditions 'f e vfu' and '] e ufv' state thatf is a function mapping u into v and whose inverse maps v into u. According to Peano these conditions restrict the domain of what is to be understood by the letters. Now this still leaves open a certain leeway for the meanings* (significations) of the letters und that creates generality. Now how is a particular case derived from such n general sentence? Obviously by assigning, subject to the restrictions imposed by the antecedents, particular meanings to the letters in the consequent (the part of the formula to the right of the '::J'), and so by replacing the letters by signs which have these particular meanings at all times. Here in fact the letter 'f' doesn't occur in the consequent at all. Nevertheless we must still be able to specify such a meaning for this letter too. The antecedents are now dropped, because they have done their job for this case, and because the deduction-sign '::J' would otherwise stand between Ncntences that contained no indefinitely indicating letters, which according to the latest version of the Peano concept-script is forbidden. Thus, of the original general sentence there only remains the consequent, which now, however, no longer has any generality.
Let's now look at a sentence which our general sentence may be converted into, by taking as the new consequent the negation of an nntccedent, and making the negation of the original consequent into an
*By using italics I indicate that I am using this word in the sense of l'cano's 'signification'.
? 154 The Argumentfor my stricter Canons ofDefinition
antecedent. This transformation, called by English logicians 'contra- position', and otherwise known as 'the transition from modus ponens to modus tollens' is indispensable. The sense is scarcely affected by it, since the sentence gives neither more nor less information after the transformation than before. But since the conditions are not quite the same, the restrictions on the meanings of the letters are no longer the same either. If, e. g. I make the negation of the antecedent, 'u is a class', into a consequent ('then u is not a class'), then in conforming with the conditions that vis a class andfmaps u into v, and the inverse o f f v into u, and finally that the cardinality of u is different from the cardinality of v, I can now only give the letter u precisely such meanings as were previously excluded. This illustrates how, as a re~;ult of legitimate transformations, there is an alteration in the meanings that may be assigned to the letters: a fact which hardly makes for greater logical perspicuity.
As against this, my conception is as follows. First, on account ofthe basic difference between objects and concepts, it is necessary to separate function- letters from object-letters. A sentence completed with a judgement-stroke that contains roman object-letters affirms that its content is true whatever meaningful proper names you may substitute for those letters, provided the same proper name is substituted for one and the same letter throughout the sentence. Since proper names are signs which mean one individual determinate object, another way of putting this is: such a sentence affirms that its content is true, whatever objects be understood by the roman object- letters occurring in it. So here the meanings (in Peano's sense) have genuinely unrestricted scope; for that it only includes objects and not functions as well goes without saying, since in view of their fundamental dissimilarity objects and functions cannot be substituted for one another. This is how my conception contrasts with Peano's, for in his case the scope can be more or less restricted, and changes under transformations of the sentence. And so in my case antecedents don't have the function of restricting scope. If I want to derive a particular proposition from such a general sentence-from a sentence whose generality is simply due to the presence of roman letters, I simply put the same proper name for each occurrence in the sentence of the same roman object-letter. In this way the lower limbs (antecedents) stay in plac(', but can, where appropriate, be detached by certain inferences. This makes it quite unnecessary to look anxiously to see that the leeway allowed by the conditions is not exceeded.
In the Peano concept-script the judgements necessary for this are not reflected at all, and so it cannot provide a way of checking them. In my case the designation of generality is quite independent of the hypothetical form of the sentence, and the meaning of the conditional stroke is defined quite independently of generality: and this is the methodologically correct procedure to (ollow.
It follows from this that certain demands have to be made of the definitions of signs. Let's assume for the sake of simplicity that only one
? The Argumentfor my stricter Canons ofDefinition 155
roman letter-an object-letter-occurs in a sentence. It then stands in the argument-place of the designation of a function, which in this case will be a concept. And in the case where the sentence is provided with a judgement- stroke the value of this function must be the True, for every object as argument. And so the designation of this function taken in conjunction with every meaningful proper name which occupies the argument-place must have a meaning. Hence the same must also hold for every function-name which, say, helps form the designation of our function: the proper name formed from this function-name and any proper name whatever in the argument-place must always have a meaning, provided only that this last proper name means something. For the proper name thus formed out of our function-name and this proper name is a part of the proper name of the True that is formed out of the whole function-designation and that very name. But if this part has no meaning, neither can the whole mean anything, and so, in particular, it cannot mean the True. In our example this holds of the function
? PI;. This requirement is not made by Peano and hence is also rarely met, although it is scarcely less necessary for his conception of the hypothetical sentence than for mine. For if, in our example, we take as before 'then u is not a class' as consequent, we must also understand by u something that is not a class, and hence in this case too 'num u' must mean something, if it is to be possible to judge whether the condition 'if num u does not coincide with num v' is satisfied, and equally it must also be possible to judge whether a given relation maps u into v or v into u for the case where u is not a class. So the definitions of 'num u' and of mapping ('f e vfu') must be formulated accordingly. Since one cannot know from the outset in which sentences these signs will occur and what restrictions will be thereby placed on the meanings of the letters, the definitions are to be constructed in such a way that a meeting is guaranteed these combinations of signs for every meaning of the letters.
We can also argue for this requirement on the ground that the law ofthe excluded middle must hold. For that implies that every concept must have sharp boundaries, so that it is determined for every object whether it falls under the concept or not. Were this not so, there would be a third case besides just the two cases 'a falls under the concept F' and 'a does not fall under the concept F'-namely, the case where this is undecided. The fallacy known as the 'Sorites' depends on something (e. g. a heap) being treated as a concept which cannot be acknowledged as such by logic because it is not properly circumscribed.
The following consideration also gives the same result. Inference from two premises very often, if not always, depends on a concept being common to both of them. If a fallacy is to be avoided, not only must the sign for the concept be the same, it must also mean the same. It must have a meaning independent of the context and not first come to acquire one in context, which is no doubt what very often happens in speech.
What holds of functions which we have called concepts, that they must
? 156
The Argumentfor my stricter Canons ofDefinition
have a value for every argument, also holds of the others; for they can be used in the construction of concepts. Thus, e. g.
rH= ne(e=fl)
is a concept in the construction of which we use, among others, the function n<:,. If now for some or other meaningful proper name 'A', 'nA' were
meaningless, then
would also be meaningless and so could mean neither the True nor the False; i. e. we should have the case that it would not be determined for the object whether or not it fell under the concept
n~ = ne (e = fl)
What we have said about object-letters, also holds mutatis mutandis for function-letters; a roman function-letter, which is used as a mark of a first level function with one argument, must be replaceable wherever it occurs in sentences by the designation of a first level function with one argument without rendering the whole devoid of meaning. From this it follows that such a letter cannot occur without an argument place: and so Peano's designation 'f e vfu' must be rejected, because the letter 'f' occurs here without any argument place, so that it e. g. becomes impossible to put in its place the designation of the function <:, + 1. If the letter 'e' in 'f e vfu' was supposed to occupy the argument place of 'f(fj', you could substitute the designation of the function<:. + 1, thus getting
'(e + 1) vfu',
but as Peano construes the formula, no such substitution can occur. For the ? same reason many of Peano's designations in which a function-letter occurs , without an argument are to be rejected. They contradict the very essence of a function, its unsaturatedness or need of supplementation. Such designations, which belie the real situation, may indeed seem at first sight ? convenient, but in the end they always lead into a morass; for at some point or another their inadequacy will become painfully apparent. So here: the letter fis supposed to serve for the designation of generality. But try to move from it to particular cases, and it nearly always breaks down. And indeed a wide variety of designations have been introduced by mathematicians with an eye only on their immediate purpose; but the designations that promise best to survive are those which adapt themselves most readily to diverse requirements and can be applied over the most extensive domain, and this because they fit the subject-matter best. We shall not be able to discover such designations if we are merely concerned to cope with the case in hand and do not P~! rsue our reflections further, but only if we attain the deepest possible insight into the nature of the subject-matter.
? Logical Defects in Mathematics1 [1898/99 or later, probably not after 1903]
There is little cause for satisfaction with the state in which mathematics finds itself at present, if you have regard not to the outside, to the amount of it, but to the degree of perfection and clarity within. In this respect it leaves almost everything to be desired if you compare it with the ideal you may reasonably propose for this discipline, and when you consider that by its very nature it ought to be better fitted to approach its ideal than is any other discipline. If you ask what constitutes the value of mathematical knowledge, the answer must be: not so much what is known as how it is known, not so much its subject-matter as the degree to which it is intellectually perspicuous and affords insight into its logical interrelations. And it is just this which is lacking. Authors explain the commonest expressions such as 'function', 'variable' and 'equal' in totally different ways and these discrepancies are not just trivial but concern the very heart of the matter. It very often happens that the same word is used by one writer to name a sign, by another a content which he goes on to present as the meaning of that sign. So in the first case a word is understood to mean a material object with physical and chemical properties, such as colour and solubility in hydrochloric ncid-while in the second it is understood to mean an object with no such properties. But it isn't only in the case of different authors that we find such discrepancies--ones which affect the very foundations of the subject: not infrequently one and the same author uses a word in a way that conflicts with his own definition. It happens, for instance, that a mathematician so explains the term 'definite integral' that he wants us to understand by it a certain limit of a sum. But the same author isn't deterred in the slightest from using the expression as if he understood it to mean a combination of si)! . ns that contains the integral sign as a constituent. So, for instance, Ludwig Scheefer* says '. . . so that, according to the Riemannian definition
? IAllgemeine Untersuchungen iiber Rectification der Curven,] Acta mathematica V [1884], p. 49.
1 The editors date this essay between 1898/99 and 1903, largely by the way in which Volume I of the Grundgesetze is referred t o - t h e way in which Frege talks of the use of the letter '. ;' 'in the first volume'. They also mention that some of the material of this essay is developed in the essay Was ist eine Funktion? published in
1904. This argument for the dating seems cogent, but we think it possible to go l'urther and say that the way in which Grundgesetze Vol. I is referred to is only imclligible if Frege is producing here, as in the case of the preceding article, a draft of material intended for inclusion in Vol. 11 of the Grundgesetze (trans. ).
? ? 158
Logical Defects in Mathematics
the definite integral f~'yl + f'(x)2 dx acquires no meaning. ' Here the word 'integral' is used in both those senses in the same breath. For if there is talk of the meaning of an integral then all that can be meant is the meaning of a
sign, or combination of signs-in fact from the context we can only understand the expression that is here formed with the integral, root and plus signs combined with numerals and letters. But at the same time appeal is made to the Riemannian definition according to which an integral is the limit of a sum, and it would be difficult to understand a limit value in Riemann's sense to mean a sign whose meaning could be asked after. In Riemann we find the following:*
'If it** should have the property that however fJ and e may be chosen, it approaches a limit A indefinitely closely as fJ becomes indefinitely small, then this value is called f~ f(x) dx. If it does not have this property f~ f(x) dx has no meaning. Even in such a case, however, there have been a number of attempts to attach a meaning to this sign, and among these extensions of the concept of a definite integral one is accepted by all mathematicians. '
Here the sign 'f~flx)dx' is nowhere called an integral. Throughout this context I would put this sign in inverted commas to make it clearer that the sign alone is meant, for the definition itself asserts that otherwise a limit value is to be understood by it. Even I wouldn't have found this precaution necessary if experience had not taught me how far one must avoid anything that could in any way encourage the mathematical sickness of our time, of confusing the sign with what is signified. This sickness may well not have been so prevalent when Riemann wrote his article. And so it's easy to understand that he should regard inverted commas as dispensable, since the sense is made clear enough by the use of the phrases 'is called', 'has no meaning', and 'we understand by'. His usage would also correspond to that according to which you do not need to insert inverted commas after the phrase 'is called'. He doesn't seem to have used the expression 'the integral has a meaning'.
The following expressions are found in his writings (5): extension of the validity of the concept 'a function is integrable', 'we may speak of an integral of the functionf(x) between a and b', and also (6): 'the possibility of a definite integral'. None of these expressions gives any reason for suspecting Riemann to have been in the grip of the epidemic we have referred to. And so we must not hesitate to assume the opposite. If an astronomer said 'The planet 2. (. has a meaning: it means the planet of the solarsystemwiththegreatestmass. Whereastheplanet~hasasyet no meaning, but it could be that it will receive a meaning later', or if he said
? Ober die'Darstellbarkeit einer Funktion durch eine trigonometrisch? Reihe. Werke [ed. by H. Weber, Leipzig 18761: p. 225.
? ? That is, the sum.
? Logical Defects in Mathematics 159
'The planet '21-has a certain similarity with the number 4', one would perhaps usk oneself whether he had some kind of brain disorder. But if a mathematician adopts ways of speaking which appear not unworthy of a Dogberry, this happens apparently without any slur on his scientific reputation.
There is a great deal of talk about variables in the mathematical literature. But you may hardly conclude from this that there exists a general agreement nhout the sense of the word. To my mind it is much more likely that most who use it don't know exactly what sense they attach to it. Whether there is someone who has found a tenable explanation of the word, I don't know. Many do not explain the word at all, which is the most convenient thing to do, but would only be justifiable if there was a general agreement among people about it. At first one might think that you should supply the word 'number' and understand by 'variable number' a number with different properties at different times, so that e. g. at one time it was prime, at another n square. But that would obviously be quite beside the point, if only because 11 is doubtful whether there are any variable numbers in this sense at all. 11ere you might think of the fact that a number, e. g. 15 000, at one time was the number of inhabitants of Jena, at another not, and that we could l'onstrue this altered relation to the concept inhabitant of Jena as an nlteration in the number 15 000. However it wouldn't be the possibility of this alteration which we need in Analysis. The expression 'the number of Inhabitants of Jena varies' isn't strictly correct. One does not mean by it that 1he same number assumes different properties, but that in the course of time ever new numbers acquire the status of being the number of inhabitants of Jcna. It would be as if you wished to say of countries like England and llolland that the rulers were of variable sex, and as if someone wanted to infer from this that there were people who were now male, now female. The illusion arises from regarding a phrase like 'the King of England' as a proper nnme. But it isn't, and only becomes one if you supplement it with a time reference. In the same way, phrases like 'the number of inhabitants of Jena', 'lhe number of known moons of Jupiter' cannot be construed as proper nnmes of numbers. But once supplemented, the proper name designates a 4uite definite number, for which there can no longer be any talk of a vnriation that is of any use to Analysis. Fine, you say, but that is still only llll altered and, I will admit, improved way of talking: instead of saying 'a number varies', we must say 'in the course of time ever new numbers ll! lsume a certain status'; still that surely doesn't make any material ditl"crence. To me, however, the way of speaking doesn't appear to be so unimportant; a wrong idiom can easily lead to real confusion. And the di11tinction isn't all that slight either. V ariation always presupposes Momething permanent that varies. To say of a man that he has grown older I? to presuppose something permanent-which is designated by the proper nnme of the man--so that, despite the variation, you are acknowledging it I? the same man. Otherwise you would have to say 'a younger man has
? ? 160 Logical Defects in Mathematics
vanished and an older one appeared'. Without such a permanent we have nothing of which we can say that it has varied. If there are no variable numbers, one may not use proper names that designate them. In that case it is quite wrong to say 'the variable x'. What we then have is surely no mere difference of idiom. So it is not clear straight off what you ought to say instead in Analysis.
The scale of the changes required makes them none too easy to comprehend. Add to this the fact that it is only in kinematics that time comes into consideration, whereas variables also enter into other branches of mathematics. It is completely out of place to introduce time there. But a variation that isn't a variation in time isn't one at all in the ordinary sense of
the term. It is obviously a question here of a term of art which needs explaining, since the reference to ordinary usage is only misleading. The expression 'variable' gives an image or metaphor which, like most metaphors, at a certain point goes lame. Therefore we shall only be able to use this word without giving rise to objections after we have stipulated more precisely what it is to mean. In one of the most recent textbooks of higher Analysis* we find the following: 'By a real variable, we understand a number that is indeterminate at the outset, and which, depending on the problem in which it occurs, can assume indefinitely many real values. '
This gives rise to a host of questions. The author obviously distinguishes two classes of numbers: the determinate and the indeterminate. We may then ask, say, to which of these classes the primes belong, or whether maybe some primes are determinate numbers and others indeterminate. We may ask further whether in the case of indeterminate numbers we must distinguish between the rational and the irrational, or whether this distinction can only be applied to determinate numbers. How many indeterminate numbers are there? How are they distinguished from one another? Can you add two indeterminate numbers, and if so, how? How do you find the number that is to be regarded as their sum? The same questions arise for adding a determinate number to an indeterminate one. To which class does the sum belong? Or maybe it belongs to a third? Perhaps there is a seminal idea here which we could also find of value outside mathematics. Perhaps we could also divide men, and mathematicians in particular, into the determinate and the indeterminate. What is a value? How does an indeterminate number assume a value? Isn't a value just a number? In which case one number would assume another--or even perhaps itself? But then why does the author use both the words 'value' and 'number', if the same thing is meant? Was it perhaps to hide his misconceptions beneath the charitable cloak of darkness? Incidentally, according to our author, a determinate number can also possess a value-presumably, after it has assumed it-as we learn from the following quotation: 'In contrast with this,
? Emanuel Czuber, Vorlesungen iiber Differential und Integralrechnung, Vol. I, Leipzig, Teubner 1898.
? Logical Defects in Mathematics
161
we call a number, whether it be determinate or indeterminate, a constant, if by virtue of the problem it possesses a fixed (invariable) value. '
From this we may infer: there are both determinate and indeterminate numbers; not only the latter but also the former can assume values, which they thereupon possess. These values are either fixed and invariable, or variable. What fixed and what variable values are isn't explained, presumably because it is so simple that it needs no explanation. Much more difficult is the distinction between variable numbers and constants. That requires much more complicated definitions-these however offer such great difficulties to the understanding that the author himself has not yet been able to penetrate to the innermost depths of their conceptual content. You might imagine that the indeterminate numbers as such are not susceptible to definition, but the author continues:
'The totality of these values forms a value set, and has the particular name, the range or domain of the variable. The variable x counts as having been defined if given any real number it can be determined whether it belongs to the range or not. '
This seems to imply the following: that a variable is defined by its range, so that if you have the same range, you have the same variable. Now it is to be assumed, even if not from his definition, yet on other grounds, that where we have the equation of a curve of the third degree
y=x3
the author would speak of a variable x and a variable y, and would give the totality of real numbers as the range of each. So here we would have only one variable, which would however be designated by the two different signs 'x' and 'y'. In ? 3 we learn the following:
'If with every value of the real variable x that belongs to its range, is correlated a determinate number y, then y is thereby also in general defined as a variable, and is called a function of the real variable x. '
And so the determinate number y is defined as a number which is at the outset indeterminate, which, depending on the problem in which it occurs, can assume indefinitely many values. So the variation presumably consists in the fact that the number, which at the outset is indeterminate, gradually works its way through to becoming determinate. All the same, it is remarkable that the number y which has become determinate in this way is still a variable and can assume indefinitely many values. Which is the problem in which y occurs? And which the problem in which x occurs? Are they different problems or do we only have one problem? How prudent and well considered here is the restriction 'which belongs to its range'; otherwise it could easily befall the variable x, that it assumed values which, according to the problem in which it occurs, it cannot assume at all. It must be protected from this misfortune.
?
Accordingly, the logician does not have to ask what course thinking naturally takes in the human mind. What is natural to one man may well be unnatural to another. The great difference between grammars itself bears witness to this. The logician need fear nothing less than to be reproached with the fact that his statements do not accord with how we think naturally. The normal person with no training in mathematics would find it highly unnatural if he were to have the rudiments of the subject explained to him in terms of the utmost rigour, and for that very reason. A prudent teacher will therefore tend to let rigour go by the board in introducing the subject and will only seek to awaken the need for it bit by bit. Even in the history of mathematics we find that the highest degree of rigour is achieved only towards the end and that consequently it is at the farthest removed from what is natural. Hence to strive to present the process of thinking in its natural form would lead us directly away from logic. If the logician tried to take account of objections on the score that what he said was unnatural, he would be in danger of involving himself in endless disputes over what is natural-disputes which logic is quite incapable of resolving on its own grounds and which, therefore, do not belong to logic. To resolve them we should presumably have to resort to observing primitive peoples.
But above all we should be wary of the view that it is the business of logic to investigate how we actually think and judge when we are in agreement with the laws of truth. If that were so, we should have constantly to have one eye on the one thing and one eye on the other, and continue paying attention to the latter whilst taking a sidelong glance at the former, and in the process we should easily lose sight of a definite goal altogether. We should be seduced into asking questions with no clear meaning and as a result a satisfactory outcome to our investigations would be as good as impossible.
What are often called laws of thought, namely laws in accordance with which judging, at least in normal cases, takes place, can be nothing but laws for holding something to be true, not laws of truth. If a man holds something to be true-and the psychological logicians will surely hold that their own statements at least are true-he thereby acknowledges that there is such a thing as something's being true. But in that case it is surely probable that there will be laws of truth as well, and if there are, these must provide the norm for holding something to be true. And these will be the laws of logic proper. In supplement No. 26 to the 1897 Proceedings of the Allgemeinl Zeitung, T. Achelis writes in a paper entitled 'Volkerkunde und Philosophie': :But we are now clear about this, that the norms which hold in general for thinking and acting cannot be arrived at by the one-sided exercise of pure deductive abstraction alone; what is required is an empirico?
? Logic 147
critical determination of the objective principles of our psycho-physical organization which are valid at all times for the great consciousness of mankind. '
It is not quite clear whether this is about laws in accordance with which judgements are made or about laws in accordance with which they should be made. It appears to be about both. That is to say, the laws in accordance with which judgements are made are set up as a norm for how judgements are to be made. But why do we need to do this? Don't we automatically judge in accordance with these laws? No! Not automatically; normally, yes, but not always! So these are laws which have exceptions, but the exceptions will themselves be governed by further laws. So the laws that we have set up do not comprise all of them. Now what is our justification for isolating a part of the entire corpus of laws and setting it up as a norm? To do that is like wanting to present Kepler's laws of planetary motion as a norm and then being forced, alas, to recognize that the planets in their wilfulness do not behave in strict conformity with them but, like spoilt children, have disturbing effects on one another. Such behaviour would then have to be severely reprimanded.
On this view we shall have to exercise every care not to stray from the path taken by the solid majority. We shall even mistrust the greatest geniuses; for if they were normal, they would be mediocre.
With the psychological conception of logic we lose the distinction between the grounds that justify a conviction and the causes that actually produce it. This means that a justification in the proper sense is not possible; what we have in its place is an account of how the conviction was arrived at, from which it is to be inferred that everything has been caused by psychological factors. This puts a superstition on the same footing as a scientific discovery.
If we think of the laws of logic as psychological, we shall be inclined to raise the question whether they are somehow subject to change. Are they like the grammar of a language which may, of course, change with the passage of time? This is a possibility we really have to face up to if we hold that the laws of logic derive their authority from a source similar to that of the laws of grammar, if they are norms only because we seldom deviate from them, if it is normal to judge in accordance with our laws of logic as it is normal to walk upright. Just as there may have been a time when it was not normal for our ancestors to walk upright, so many modes of thinking might have been normal in the past which are not so now, and in the future something might be normal that is not so at the present time. In a language whose form is not yet fixed there are always points of grammar on which our sense of idiom is unreliable, and a similar thing would have to hold in respect of the laws of logic whenever we were in a period of transition. We might, for instance, be in two minds whether it is correct to judge that every object is identical with itself. If that were so, we should not really be entitled to speak of logical laws, but only of logical rules that specify what is
? 148
Logic
regarded as normal at a particular time. We should not be entitled to express such a rule in a form like 'Every object is identical with itself' for there is here no mention at all of the class of beings for whose judgements the rule is meant to be valid, but we should have to say something like 'At the present time it is normal for men-with the possible exception of certain primitive peoples for whom the matter has not yet been investigated-to judge that every object is identical with itself'. However, once there are laws, even if they are psychological, then, as we have seen, they must always be true, or better, they must be timelessly true if they are true at all. Therefore if we had observed that from a certain time a law ceased to hold, then we should have to say that it was altogether false. What we could do, however, is to try to find a condition that would have to be added to the law. Let us assume that for a certain period of time men make judgements in accordance with the law that every object is identical with itself, but that after this time they cease to do so. Then the cause of this might be that the phosphorus content in the cerebral cortex had changed, and we should then have to say something like 'If the amount of phosphorus present in any part of man's cerebral cortex does not exceed 4%, his judgement will always be in
accordance with the law that every object is identical with itself. '
We can at least conceive of psychological laws that refer in this way to the chemical composition of the brain or to its anatomical structure. On the other hand, such a reference would be absurd in the case of logical laws, for these are not concerned with what this or that man holds to be true, but with what is true. Whether a man holds the thought that 2 ? 2 = 4 to be true or to be false may depend on the chemical composition of his brain, but whether this thought is true cannot depend on that. Whether it is true that Julius
Caesar was assassinated by Brutus cannot depend upon the structure of Professor Mommsen's brain.
People sometimes raise the question whether the laws of logic can change with time. The laws of truth, like all thoughts, are always true if they are true at all. Nor can they contain a condition which might be satisfied at certain times but not at others, because they are concerned with the truth of thoughts and if these are true, they are true timelessly. So if at one time the truth of some thought follows from the truth of certain others, then it must always follow.
Let us summarize what we have elicited about thoughts (properly so? called).
Unlike ideas, thoughts do not belong to the individual mind (they are not subjective), but are independent of our thinking and confront each one of Ul in the same way (objectively). They are not the product of thinking, but are only grasped by thinking. In this respect they are like physical bodies. What distinguishes them from physical bodies is that they are non-spatial, and we could perhap~ really go as far as to say that they are essentially timeless-at least inasmuch as they are immune from anything that could effect a chanae in their intrinsic nature. They are like ideas in being non-spatial.
Since thoughts are not mental in nature, it follows that every
? Logic 149
psychological treatment of logic can only do harm. It is rather the task of this science to purify logic of all that is alien and hence of all that is psychological, and to free thinking from the fetters of language by pointing up the logical imperfections of language. Logic is concerned with the laws of truth, not with the laws of holding something to be true, not with the 4uestion of how men think, but with the question of how they must think if they are not to miss the truth.
Negation
1\ thought proper is either true or false. When we make a judgement about a thought, then we either accept it as true or we reject it as false. The last expression, however, can mislead us by suggesting that a thought which has been rejected ought to be consigned to oblivion as quickly as possible as being no longer of any use. On the contrary, the recognition that a thought 1s false may be just as fruitful as the recognition that a thought is true. Properly understood, there is no difference at all between the two cases. To hold one thought to be false is to hold a (different) thought to be true-a thought which we call the opposite of the first. In the German language we usually indicate that a thought is false by inserting the word 'not' into the predicate. But the assertion is still conveyed by the indicative form, and has no necessary connection with the word 'not'. The negative form can be retained although the assertion has been dropped. We can speak equally well of 'The thought that Peter did not come to Rome' as of 'The thought that Peter came to Rome'. Thus it is clear that when I assert that Peter did not come to Rome, the act of asserting and judging is no different from when I assert that Peter did come to Rome; the only difference is that we hnvc the opposite thought. So to each thought there corresponds an oppo- Nite. Here we have a symmetrical relation: If the first thought is the opposite of the second, then the second is the opposite of the first. To declare false the thought that Peter did not come to Rome is to assert that Peter came to Rome. We could declare it false by inserting a second 'not' and saying 'Peter did (not) not come to Rome' or 'It is not true that Peter did not come to Rome'. And from this it follows that two negatives cancel one another out. If we take the opposite of the opposite of something, we have what
we began with.
When it is a question of whether some thought is true, we are poised
hctwcen opposite thoughts, and the same act which recognizes one of them IIN true recognizes the other as false. Analogous relations of opposition hold In other cases too, e. g. between what is beautiful and ugly, good and bad, rtcusant and unpleasant and positive and negative in mathematics and Jlhysics. But these are different from our case in two respects. In the first ptncc there is nothing here which, like nought or a neutrally charged body, occupies a mean between opposites. We can of course say that, in relation lo positive and negative numbers, nought is its own opposite, but there is no
? 150 Logic
thought which could count as its own opposite. This is true even of fict- ions. In the second place we do not have here two classes such that one contains thoughts that are the opposite of those contained in the other, as there is a class of positive and a class of negative numbers. At any rate, I have not yet found any feature that could be used to effect such a division; for the use of the word 'not' in ordinary language is a purely external cri- terion and an unreliable one at that. We have other signs for negation like 'no', and we often use the prefix 'un' as, for example, in 'unsatisfactory'. Now in view of the fact that 'unsatisfactory' and 'bad' are very close in sense to one another, there would surely appear to be little point in wishing to assign the thoughts contained in 'This work is bad' and 'This work is satis- factory' to the first class and those contained in 'The work is not bad' and 'The work is unsatisfactory' to the second class, and it may well be the case that in some other language the word for 'unsatisfactory' is one in which a negative form is no more to be discerned that it is in 'bad'. We cannot define in what respect the first two thoughts might be supposed to be more closely related than the first and fourth. There is the further fact that negatives can occur elsewhere than in the predicate of a main clause, and that such nega- tives do not simply cancel each other out: an example would be 'Not all pieces of work are unsatisfactory'. We cannot put for this 'All pieces of work are satisfactory', and nor can we put 'Whoever has worked hard is rewarded,
for 'Whoever has not worked hard is not rewarded'. If we compare these with the sentences 'Whoever is rewarded has worked hard', 'Whoever has not worked hard goes away empty-handed', 'Whoever has been idle is not rewarded', '24 is not different from 42' and '24 is the same as 42', we shall see that we are tangling with some thorny problems here. What is more it is simply not worth while to try to extricate ourselves, and to expend a great deal of effort on finding answers to them. I at any rate know of no logical law which would take cognizance of a division of thoughts into positive and negative. We shall therefore leave the matter to look after itself until such time as it should become clear to us that such a division is somehow necessary. At which juncture we should naturally expect a criterion to emerge by which the division could be effected.
The prefix 'un', is not always used to negate. There is hardly any difference in sense between 'unhappy' and 'miserable'. Here we have a word which is the opposite of 'happy', and yet is not its negation. For this reason the sentences 'This man is not unhappy' and 'This man is happy' do not have the same sense.
Compound Thoughts
If the jury ret~rn a 'Yes' to the question 'Did the accused wilfully set fire to a pile of wood and (wilfully) start a forest fire? ' they have simultaneously asserted the two thoughts:
? Logic 151 (1) The accused wilfully set fire to a pile of wood,
(2) The accused wilfully started a forest fire.
It is true that our question contains one thought, for it can be answered by making one judgement; but this thought is composed of two thoughts-each of these being capable of being judged on its own-in such a way that by affirming the whole thought I affirm at the same time the component thoughts. Now it might seem that this is really neither here nor there and that the matter is of little importance; but it will become evident that it is closely bound up with very important logical laws. This comes out more clearly as soon as one considers the negation of such a compound thought. When will the jury have to say 'No' to the question above? Obviously the moment they hold but one of the two component thoughts to be false; for example, if they are of the opinion that whilst there is no doubt that the accused wilfully set fire to the pile of wood, he did not intend as a consequence that the forest should catch fire. '
? ? The Argument for my stricter Canons of Definition1 [1897/98 or shortly afterwards]
Since the necessity for my stricter canons of definition may not yet be sufficiently evident from what I said on the matter in the first part,2 and since, as it appears, opinion is still greatly divided on this matter, I want to try to argue for my view once again, by drawing a comparison with Peano's mathematical logic. What is at stake here is perhaps the deepest difference between the two concept-scripts. I have drawn attention to certain shortcomings which struck me in the definition of the Peano deduction-sign ':::::>',which roughly corresponds to my 't'. These shortcomings are certainly due to the fact that, whereas Peano takes a first step in my direction, he doesn't take the second. His definition of the sign ':::::>' for the case where it stands between sentences containing no indefinitely indicating letters (lettres variables) agrees in essentials with my definition of the sign 't'; but for the case where indefinitely indicating letters (lettres variables) occur on both sides of the sign ':::::>', another definition is given whose relation to the first remains unclear. In my case the combination of condition and consequence is presented as analysed into two components, of which the one is generality, signified by the use of roman letters, while the other is designated by the sign 't'. This analysis cannot be clearly discerned in Peano's case because the use of lettres variables and the sign ':::::>' are both defined simultaneously. In his most recent account, Peano has attempted to remove this shortcoming, but unfortunately, not by following me in taking my
1 The text was presumably composed in 1897/8 or shortly thereafter: right at the beginning Frege mentions having 'drawn attention to certain shortcomings in the definition of the Peano deduction-sign'. This must refer to the article of Frege: Ober die Begri. ffsschrift des Herrn Peano und meine eigene that appeared in 1897. Later on Frege talks of Peano's attempts in his most recent account to overcome these shortcomings. From the details ofwhat Frege says, this must refer to the definition of'-::;,' in Formulaire de Mathematiques' Vol. 11, p. 24. which appeared at the end of 1897. (The corresponding definition in Form. Vol. I, pp. V/VI, No. 17 is still in the form that Frege criticized in his article) (ed. ). It seems to us a natural conjecture that the present article and the next are preliminary drafts of material intended for inclusion in Volume 11 of the Grundgesetze: this both because of the overlap in content between parts of this article and early parts of Volume 11 and above all because of the way in which Grundgesetze Volume I is simply referred to as 'the first part' or 'my first volume' (trans. ).
2 As the editors point out, the reference here is clearly to the rirst part of the Grundgesetze (trans. ).
? The Argumentfor my stricter Canons ofDefinition 153
second step, but by retracting the first. He completely abandons the definition of the deduction-sign which agrees in essentials with mine, and now completely excludes this case-the one where '"::J' stands between sentences containing no indefinitely indicating letters. But there surely do occur cases where even in speech the form of a hypothetical sentence ('if . . . then . . . ')is used, without the content being general. Let us, however, waive this objection.
As an example, let us take a look now at the Peano sentence 'u, v e K. fe vfu. ]e ufv . ::J. num u = num v',
which corresponds in essentials to my ? ~nu= nv '
u'"' (v'"' ) p) v'"' (u'"' ) $. P)
Admittedly in my case the two Peano conditions u, v e K are missing, but this does not make my sentence false. Rather, my signs are so defined, that, without the truth of the claim being put in jeopardy, names of objects other than classes can also be substituted for u and v. In this formula u, v e K is meant to state that by u and v classes are to be understood. The other two conditions 'f e vfu' and '] e ufv' state thatf is a function mapping u into v and whose inverse maps v into u. According to Peano these conditions restrict the domain of what is to be understood by the letters. Now this still leaves open a certain leeway for the meanings* (significations) of the letters und that creates generality. Now how is a particular case derived from such n general sentence? Obviously by assigning, subject to the restrictions imposed by the antecedents, particular meanings to the letters in the consequent (the part of the formula to the right of the '::J'), and so by replacing the letters by signs which have these particular meanings at all times. Here in fact the letter 'f' doesn't occur in the consequent at all. Nevertheless we must still be able to specify such a meaning for this letter too. The antecedents are now dropped, because they have done their job for this case, and because the deduction-sign '::J' would otherwise stand between Ncntences that contained no indefinitely indicating letters, which according to the latest version of the Peano concept-script is forbidden. Thus, of the original general sentence there only remains the consequent, which now, however, no longer has any generality.
Let's now look at a sentence which our general sentence may be converted into, by taking as the new consequent the negation of an nntccedent, and making the negation of the original consequent into an
*By using italics I indicate that I am using this word in the sense of l'cano's 'signification'.
? 154 The Argumentfor my stricter Canons ofDefinition
antecedent. This transformation, called by English logicians 'contra- position', and otherwise known as 'the transition from modus ponens to modus tollens' is indispensable. The sense is scarcely affected by it, since the sentence gives neither more nor less information after the transformation than before. But since the conditions are not quite the same, the restrictions on the meanings of the letters are no longer the same either. If, e. g. I make the negation of the antecedent, 'u is a class', into a consequent ('then u is not a class'), then in conforming with the conditions that vis a class andfmaps u into v, and the inverse o f f v into u, and finally that the cardinality of u is different from the cardinality of v, I can now only give the letter u precisely such meanings as were previously excluded. This illustrates how, as a re~;ult of legitimate transformations, there is an alteration in the meanings that may be assigned to the letters: a fact which hardly makes for greater logical perspicuity.
As against this, my conception is as follows. First, on account ofthe basic difference between objects and concepts, it is necessary to separate function- letters from object-letters. A sentence completed with a judgement-stroke that contains roman object-letters affirms that its content is true whatever meaningful proper names you may substitute for those letters, provided the same proper name is substituted for one and the same letter throughout the sentence. Since proper names are signs which mean one individual determinate object, another way of putting this is: such a sentence affirms that its content is true, whatever objects be understood by the roman object- letters occurring in it. So here the meanings (in Peano's sense) have genuinely unrestricted scope; for that it only includes objects and not functions as well goes without saying, since in view of their fundamental dissimilarity objects and functions cannot be substituted for one another. This is how my conception contrasts with Peano's, for in his case the scope can be more or less restricted, and changes under transformations of the sentence. And so in my case antecedents don't have the function of restricting scope. If I want to derive a particular proposition from such a general sentence-from a sentence whose generality is simply due to the presence of roman letters, I simply put the same proper name for each occurrence in the sentence of the same roman object-letter. In this way the lower limbs (antecedents) stay in plac(', but can, where appropriate, be detached by certain inferences. This makes it quite unnecessary to look anxiously to see that the leeway allowed by the conditions is not exceeded.
In the Peano concept-script the judgements necessary for this are not reflected at all, and so it cannot provide a way of checking them. In my case the designation of generality is quite independent of the hypothetical form of the sentence, and the meaning of the conditional stroke is defined quite independently of generality: and this is the methodologically correct procedure to (ollow.
It follows from this that certain demands have to be made of the definitions of signs. Let's assume for the sake of simplicity that only one
? The Argumentfor my stricter Canons ofDefinition 155
roman letter-an object-letter-occurs in a sentence. It then stands in the argument-place of the designation of a function, which in this case will be a concept. And in the case where the sentence is provided with a judgement- stroke the value of this function must be the True, for every object as argument. And so the designation of this function taken in conjunction with every meaningful proper name which occupies the argument-place must have a meaning. Hence the same must also hold for every function-name which, say, helps form the designation of our function: the proper name formed from this function-name and any proper name whatever in the argument-place must always have a meaning, provided only that this last proper name means something. For the proper name thus formed out of our function-name and this proper name is a part of the proper name of the True that is formed out of the whole function-designation and that very name. But if this part has no meaning, neither can the whole mean anything, and so, in particular, it cannot mean the True. In our example this holds of the function
? PI;. This requirement is not made by Peano and hence is also rarely met, although it is scarcely less necessary for his conception of the hypothetical sentence than for mine. For if, in our example, we take as before 'then u is not a class' as consequent, we must also understand by u something that is not a class, and hence in this case too 'num u' must mean something, if it is to be possible to judge whether the condition 'if num u does not coincide with num v' is satisfied, and equally it must also be possible to judge whether a given relation maps u into v or v into u for the case where u is not a class. So the definitions of 'num u' and of mapping ('f e vfu') must be formulated accordingly. Since one cannot know from the outset in which sentences these signs will occur and what restrictions will be thereby placed on the meanings of the letters, the definitions are to be constructed in such a way that a meeting is guaranteed these combinations of signs for every meaning of the letters.
We can also argue for this requirement on the ground that the law ofthe excluded middle must hold. For that implies that every concept must have sharp boundaries, so that it is determined for every object whether it falls under the concept or not. Were this not so, there would be a third case besides just the two cases 'a falls under the concept F' and 'a does not fall under the concept F'-namely, the case where this is undecided. The fallacy known as the 'Sorites' depends on something (e. g. a heap) being treated as a concept which cannot be acknowledged as such by logic because it is not properly circumscribed.
The following consideration also gives the same result. Inference from two premises very often, if not always, depends on a concept being common to both of them. If a fallacy is to be avoided, not only must the sign for the concept be the same, it must also mean the same. It must have a meaning independent of the context and not first come to acquire one in context, which is no doubt what very often happens in speech.
What holds of functions which we have called concepts, that they must
? 156
The Argumentfor my stricter Canons ofDefinition
have a value for every argument, also holds of the others; for they can be used in the construction of concepts. Thus, e. g.
rH= ne(e=fl)
is a concept in the construction of which we use, among others, the function n<:,. If now for some or other meaningful proper name 'A', 'nA' were
meaningless, then
would also be meaningless and so could mean neither the True nor the False; i. e. we should have the case that it would not be determined for the object whether or not it fell under the concept
n~ = ne (e = fl)
What we have said about object-letters, also holds mutatis mutandis for function-letters; a roman function-letter, which is used as a mark of a first level function with one argument, must be replaceable wherever it occurs in sentences by the designation of a first level function with one argument without rendering the whole devoid of meaning. From this it follows that such a letter cannot occur without an argument place: and so Peano's designation 'f e vfu' must be rejected, because the letter 'f' occurs here without any argument place, so that it e. g. becomes impossible to put in its place the designation of the function <:, + 1. If the letter 'e' in 'f e vfu' was supposed to occupy the argument place of 'f(fj', you could substitute the designation of the function<:. + 1, thus getting
'(e + 1) vfu',
but as Peano construes the formula, no such substitution can occur. For the ? same reason many of Peano's designations in which a function-letter occurs , without an argument are to be rejected. They contradict the very essence of a function, its unsaturatedness or need of supplementation. Such designations, which belie the real situation, may indeed seem at first sight ? convenient, but in the end they always lead into a morass; for at some point or another their inadequacy will become painfully apparent. So here: the letter fis supposed to serve for the designation of generality. But try to move from it to particular cases, and it nearly always breaks down. And indeed a wide variety of designations have been introduced by mathematicians with an eye only on their immediate purpose; but the designations that promise best to survive are those which adapt themselves most readily to diverse requirements and can be applied over the most extensive domain, and this because they fit the subject-matter best. We shall not be able to discover such designations if we are merely concerned to cope with the case in hand and do not P~! rsue our reflections further, but only if we attain the deepest possible insight into the nature of the subject-matter.
? Logical Defects in Mathematics1 [1898/99 or later, probably not after 1903]
There is little cause for satisfaction with the state in which mathematics finds itself at present, if you have regard not to the outside, to the amount of it, but to the degree of perfection and clarity within. In this respect it leaves almost everything to be desired if you compare it with the ideal you may reasonably propose for this discipline, and when you consider that by its very nature it ought to be better fitted to approach its ideal than is any other discipline. If you ask what constitutes the value of mathematical knowledge, the answer must be: not so much what is known as how it is known, not so much its subject-matter as the degree to which it is intellectually perspicuous and affords insight into its logical interrelations. And it is just this which is lacking. Authors explain the commonest expressions such as 'function', 'variable' and 'equal' in totally different ways and these discrepancies are not just trivial but concern the very heart of the matter. It very often happens that the same word is used by one writer to name a sign, by another a content which he goes on to present as the meaning of that sign. So in the first case a word is understood to mean a material object with physical and chemical properties, such as colour and solubility in hydrochloric ncid-while in the second it is understood to mean an object with no such properties. But it isn't only in the case of different authors that we find such discrepancies--ones which affect the very foundations of the subject: not infrequently one and the same author uses a word in a way that conflicts with his own definition. It happens, for instance, that a mathematician so explains the term 'definite integral' that he wants us to understand by it a certain limit of a sum. But the same author isn't deterred in the slightest from using the expression as if he understood it to mean a combination of si)! . ns that contains the integral sign as a constituent. So, for instance, Ludwig Scheefer* says '. . . so that, according to the Riemannian definition
? IAllgemeine Untersuchungen iiber Rectification der Curven,] Acta mathematica V [1884], p. 49.
1 The editors date this essay between 1898/99 and 1903, largely by the way in which Volume I of the Grundgesetze is referred t o - t h e way in which Frege talks of the use of the letter '. ;' 'in the first volume'. They also mention that some of the material of this essay is developed in the essay Was ist eine Funktion? published in
1904. This argument for the dating seems cogent, but we think it possible to go l'urther and say that the way in which Grundgesetze Vol. I is referred to is only imclligible if Frege is producing here, as in the case of the preceding article, a draft of material intended for inclusion in Vol. 11 of the Grundgesetze (trans. ).
? ? 158
Logical Defects in Mathematics
the definite integral f~'yl + f'(x)2 dx acquires no meaning. ' Here the word 'integral' is used in both those senses in the same breath. For if there is talk of the meaning of an integral then all that can be meant is the meaning of a
sign, or combination of signs-in fact from the context we can only understand the expression that is here formed with the integral, root and plus signs combined with numerals and letters. But at the same time appeal is made to the Riemannian definition according to which an integral is the limit of a sum, and it would be difficult to understand a limit value in Riemann's sense to mean a sign whose meaning could be asked after. In Riemann we find the following:*
'If it** should have the property that however fJ and e may be chosen, it approaches a limit A indefinitely closely as fJ becomes indefinitely small, then this value is called f~ f(x) dx. If it does not have this property f~ f(x) dx has no meaning. Even in such a case, however, there have been a number of attempts to attach a meaning to this sign, and among these extensions of the concept of a definite integral one is accepted by all mathematicians. '
Here the sign 'f~flx)dx' is nowhere called an integral. Throughout this context I would put this sign in inverted commas to make it clearer that the sign alone is meant, for the definition itself asserts that otherwise a limit value is to be understood by it. Even I wouldn't have found this precaution necessary if experience had not taught me how far one must avoid anything that could in any way encourage the mathematical sickness of our time, of confusing the sign with what is signified. This sickness may well not have been so prevalent when Riemann wrote his article. And so it's easy to understand that he should regard inverted commas as dispensable, since the sense is made clear enough by the use of the phrases 'is called', 'has no meaning', and 'we understand by'. His usage would also correspond to that according to which you do not need to insert inverted commas after the phrase 'is called'. He doesn't seem to have used the expression 'the integral has a meaning'.
The following expressions are found in his writings (5): extension of the validity of the concept 'a function is integrable', 'we may speak of an integral of the functionf(x) between a and b', and also (6): 'the possibility of a definite integral'. None of these expressions gives any reason for suspecting Riemann to have been in the grip of the epidemic we have referred to. And so we must not hesitate to assume the opposite. If an astronomer said 'The planet 2. (. has a meaning: it means the planet of the solarsystemwiththegreatestmass. Whereastheplanet~hasasyet no meaning, but it could be that it will receive a meaning later', or if he said
? Ober die'Darstellbarkeit einer Funktion durch eine trigonometrisch? Reihe. Werke [ed. by H. Weber, Leipzig 18761: p. 225.
? ? That is, the sum.
? Logical Defects in Mathematics 159
'The planet '21-has a certain similarity with the number 4', one would perhaps usk oneself whether he had some kind of brain disorder. But if a mathematician adopts ways of speaking which appear not unworthy of a Dogberry, this happens apparently without any slur on his scientific reputation.
There is a great deal of talk about variables in the mathematical literature. But you may hardly conclude from this that there exists a general agreement nhout the sense of the word. To my mind it is much more likely that most who use it don't know exactly what sense they attach to it. Whether there is someone who has found a tenable explanation of the word, I don't know. Many do not explain the word at all, which is the most convenient thing to do, but would only be justifiable if there was a general agreement among people about it. At first one might think that you should supply the word 'number' and understand by 'variable number' a number with different properties at different times, so that e. g. at one time it was prime, at another n square. But that would obviously be quite beside the point, if only because 11 is doubtful whether there are any variable numbers in this sense at all. 11ere you might think of the fact that a number, e. g. 15 000, at one time was the number of inhabitants of Jena, at another not, and that we could l'onstrue this altered relation to the concept inhabitant of Jena as an nlteration in the number 15 000. However it wouldn't be the possibility of this alteration which we need in Analysis. The expression 'the number of Inhabitants of Jena varies' isn't strictly correct. One does not mean by it that 1he same number assumes different properties, but that in the course of time ever new numbers acquire the status of being the number of inhabitants of Jcna. It would be as if you wished to say of countries like England and llolland that the rulers were of variable sex, and as if someone wanted to infer from this that there were people who were now male, now female. The illusion arises from regarding a phrase like 'the King of England' as a proper nnme. But it isn't, and only becomes one if you supplement it with a time reference. In the same way, phrases like 'the number of inhabitants of Jena', 'lhe number of known moons of Jupiter' cannot be construed as proper nnmes of numbers. But once supplemented, the proper name designates a 4uite definite number, for which there can no longer be any talk of a vnriation that is of any use to Analysis. Fine, you say, but that is still only llll altered and, I will admit, improved way of talking: instead of saying 'a number varies', we must say 'in the course of time ever new numbers ll! lsume a certain status'; still that surely doesn't make any material ditl"crence. To me, however, the way of speaking doesn't appear to be so unimportant; a wrong idiom can easily lead to real confusion. And the di11tinction isn't all that slight either. V ariation always presupposes Momething permanent that varies. To say of a man that he has grown older I? to presuppose something permanent-which is designated by the proper nnme of the man--so that, despite the variation, you are acknowledging it I? the same man. Otherwise you would have to say 'a younger man has
? ? 160 Logical Defects in Mathematics
vanished and an older one appeared'. Without such a permanent we have nothing of which we can say that it has varied. If there are no variable numbers, one may not use proper names that designate them. In that case it is quite wrong to say 'the variable x'. What we then have is surely no mere difference of idiom. So it is not clear straight off what you ought to say instead in Analysis.
The scale of the changes required makes them none too easy to comprehend. Add to this the fact that it is only in kinematics that time comes into consideration, whereas variables also enter into other branches of mathematics. It is completely out of place to introduce time there. But a variation that isn't a variation in time isn't one at all in the ordinary sense of
the term. It is obviously a question here of a term of art which needs explaining, since the reference to ordinary usage is only misleading. The expression 'variable' gives an image or metaphor which, like most metaphors, at a certain point goes lame. Therefore we shall only be able to use this word without giving rise to objections after we have stipulated more precisely what it is to mean. In one of the most recent textbooks of higher Analysis* we find the following: 'By a real variable, we understand a number that is indeterminate at the outset, and which, depending on the problem in which it occurs, can assume indefinitely many real values. '
This gives rise to a host of questions. The author obviously distinguishes two classes of numbers: the determinate and the indeterminate. We may then ask, say, to which of these classes the primes belong, or whether maybe some primes are determinate numbers and others indeterminate. We may ask further whether in the case of indeterminate numbers we must distinguish between the rational and the irrational, or whether this distinction can only be applied to determinate numbers. How many indeterminate numbers are there? How are they distinguished from one another? Can you add two indeterminate numbers, and if so, how? How do you find the number that is to be regarded as their sum? The same questions arise for adding a determinate number to an indeterminate one. To which class does the sum belong? Or maybe it belongs to a third? Perhaps there is a seminal idea here which we could also find of value outside mathematics. Perhaps we could also divide men, and mathematicians in particular, into the determinate and the indeterminate. What is a value? How does an indeterminate number assume a value? Isn't a value just a number? In which case one number would assume another--or even perhaps itself? But then why does the author use both the words 'value' and 'number', if the same thing is meant? Was it perhaps to hide his misconceptions beneath the charitable cloak of darkness? Incidentally, according to our author, a determinate number can also possess a value-presumably, after it has assumed it-as we learn from the following quotation: 'In contrast with this,
? Emanuel Czuber, Vorlesungen iiber Differential und Integralrechnung, Vol. I, Leipzig, Teubner 1898.
? Logical Defects in Mathematics
161
we call a number, whether it be determinate or indeterminate, a constant, if by virtue of the problem it possesses a fixed (invariable) value. '
From this we may infer: there are both determinate and indeterminate numbers; not only the latter but also the former can assume values, which they thereupon possess. These values are either fixed and invariable, or variable. What fixed and what variable values are isn't explained, presumably because it is so simple that it needs no explanation. Much more difficult is the distinction between variable numbers and constants. That requires much more complicated definitions-these however offer such great difficulties to the understanding that the author himself has not yet been able to penetrate to the innermost depths of their conceptual content. You might imagine that the indeterminate numbers as such are not susceptible to definition, but the author continues:
'The totality of these values forms a value set, and has the particular name, the range or domain of the variable. The variable x counts as having been defined if given any real number it can be determined whether it belongs to the range or not. '
This seems to imply the following: that a variable is defined by its range, so that if you have the same range, you have the same variable. Now it is to be assumed, even if not from his definition, yet on other grounds, that where we have the equation of a curve of the third degree
y=x3
the author would speak of a variable x and a variable y, and would give the totality of real numbers as the range of each. So here we would have only one variable, which would however be designated by the two different signs 'x' and 'y'. In ? 3 we learn the following:
'If with every value of the real variable x that belongs to its range, is correlated a determinate number y, then y is thereby also in general defined as a variable, and is called a function of the real variable x. '
And so the determinate number y is defined as a number which is at the outset indeterminate, which, depending on the problem in which it occurs, can assume indefinitely many values. So the variation presumably consists in the fact that the number, which at the outset is indeterminate, gradually works its way through to becoming determinate. All the same, it is remarkable that the number y which has become determinate in this way is still a variable and can assume indefinitely many values. Which is the problem in which y occurs? And which the problem in which x occurs? Are they different problems or do we only have one problem? How prudent and well considered here is the restriction 'which belongs to its range'; otherwise it could easily befall the variable x, that it assumed values which, according to the problem in which it occurs, it cannot assume at all. It must be protected from this misfortune.
?
