Axioms of Distance, Stretches, and Magnitude of Divisibility¶
Metrical Geometry (§§ 392–399, pp. 407–414)
§ 392. Axioms of distance¶
f. §392 · p. p. 407392. Metrical Geometry is usually said to be distinguished by the introduction of quantity. It is sufficient for the characterization of metrical Geometry to observe that it introduces, between every pair of points, a relation having certain properties in virtue of which it is numerically measurable—i.e. such that numbers can be given a one-one correspondence with the various relations of the class in question. The class of relations is called distance, and will be regarded, though this is not strictly necessary, as a class of magnitudes. Some of the properties of distance are as follows.
(1) Every pair of points has one and only one distance.
(2) Distances are symmetrical relations.
(3) On a given straight line through a given point, there are two and only two points at a given distance from the given point.
(4) There is no maximum distance.
(5) The distance of a point from itself is zero [1].
(6) There is no minimum to the distance between distinct points.
(7) If d, δ be two given distances, and $A_{0}$, $A_{1}$, $A_{2}$, … $A_{n}$, … be distinct points on a straight line, whose distances one from the next are all δ, then for some value of n, $A_{0}A_{n}$ is greater than d.
(8) If $A_{0}$, $A_{n}$ be any two points, there exist n − 1 distinct points (whatever integer n may be) on the straight line $A_{0}A_{n}$, such that the distances of each from the next, of $A_{0}$ from the first, and of $A_{n}$ from the last, are all equal [2].
§ 393. Stretches¶
f. §393 · p. p. 408393. It may be observed that, if we admit the axiom that the whole is greater than the part, the properties (1), (4), (5) and (6) belong to stretches, while (2) becomes admissible by abstracting from the sense of a stretch. With regard to the remaining properties, (3), (7) and (8), there is nothing in descriptive Geometry to show whether or not they belong to stretches. Hence we may, if we choose, regard these three properties as axioms regarding stretches, and drop the word distance altogether. I believe that this represents the simplest course, and, as regards actual space, the most correct. At the same time, there is no contradiction in regarding distances as new relations distinct from stretches [3]. If we identify distance and stretch, what distinguishes metrical from descriptive Geometry is primarily the three additional axioms (3), (7) and (8), applied to a new indefinable, namely, the magnitude of divisibility of a stretch. This is not properly a notion of pure mathematics, since it cannot be derived from our original apparatus of logical notions. On the other hand, distance is not indefinable, being a class of one-one relations with certain assignable properties. On this point either course is logically permissible, but only distance can be introduced into pure mathematics in the strict sense in which the word is used in this work.
The above axioms are required for showing that all distances are numerically measurable in terms of any standard distance [4]. It is not necessary that distances should be magnitudes, or even relations; all that is essential is that distances should form a series with certain properties. If the points of a line form a continuous series, then distances do so also, in virtue of (3); thus all signless real numbers will be required for their measurement.
§ 394. Order as resulting from distance alone¶
f. §394 · p. p. 409394. Assuming that distance and stretch are distinct, it may be asked whether distances do not suffice for generating order on the straight line, without the need of any asymmetrical transitive relation of points. This represents, I think, the usual view of philosophers; but it is by no means easy to decide whether it represents a tenable view. It might perhaps be thought that (2) might be dropped, and distance regarded as an asymmetrical relation. So long as we confine our attention to one line, this view seems unobjectionable. But as soon as we consider the fact that distances on different lines may be equal, we see that the difference of sense between AB and BA is not relevant to distance, since there is no such difference between distances on different lines. Thus if CD be a distance on another line, CD may be equal both to AB and BA, and hence AB and BA must be equal, not opposite, distances. And the same thing may be made evident by considering a sphere. For this certainly consists of points at a given distance from the centre; and thus points at opposite ends of a diameter must have the same distance from the centre. Distance, then, is symmetrical; but it does not follow that the order on a line cannot be generated by distance. Let A, B be given points on a line, and let C, $C'$ be two points on AB whose distances from A are equal, and less than AB. If we now set up the axiom that either BC or $BC'$ is less than AB, while the other, $BC'$ or BC, is greater than AB, we shall, I think, after some further axioms, be able to generate order without any other relation than distance. If A, B, C be three collinear points such that the distances AC, CB are both less than AB, then we shall say that C is between A and B. If A, B, $C'$ be points such that $AC'$, AB are both less than $BC'$, then we shall say that A is between B and $C'$. If, finally, A, B, $C''$ be points such that AB, $BC''$ are both less than $AC''$, we shall say that B is between A and $C''$. It remains to see whether, as the generation of a series requires, one of these always happens. Let A, B, C be any three collinear points. First suppose, if possible, that the distances AB, BC, CA are all equal. This case is not excluded by anything hitherto assumed; we require, therefore, the further axiom that, if AB, BC be equal, AC is not equal to either of them; and I think it will be prudent to assume that AC is greater than either. Thus the case of two equal distances and one less than either is excluded. Of the three distances AB, BC, AC, therefore, one must be the greatest: let this be AC. Then in virtue of the definition, B will be between A and C. But our difficulties are not at an end. For we require further that any point between A and B shall be between A and C; and that, if A be between D and C, B shall be between D and C. With regard to the first point, if E be between A and B, AE and EB are less than AB, and therefore less than AC. But nothing assures us that EC is less than AC. For this purpose we need a new axiom, which will be just what we set out to prove, namely: If AE, EB be both less than AB, and AB, BC be both less than AC, then EC is less than AC. Finally, to prove that, if A be between D and C, and B between A and C, then B is between D and C. Here DA, AC are less than DC, and AB, BC are less than AC. Hence BC is less than DC; but nothing proves BD less than DC. For this we shall need a new axiom, and then at last our order will be definite. But the process, as is evident, is extremely complicated.
§ 395. Geometries which derive the straight line from distance¶
f. §395 · p. p. 410395. Moreover we still need a method of defining the straight line. Pieri has shown, in an admirable memoir [5], how to deduce metrical geometry by taking point and motion as the only indefinables. In §390, we objected to the introduction of motion, as usually effected, on the ground that its definition presupposes metrical properties; but Pieri escapes this objection by not defining motion at all, except through the postulates assumed concerning it. The straight line joining two points is the class of points that are unchanged by a motion which leaves the two points fixed. The sphere, the plane, perpendicularity, the order of points on a line, etc. are easily defined. This procedure is logically unimpeachable, and is probably the simplest possible for elementary geometry. But we must now return to the consideration of other suggested systems.
There is a method, invented by Leibniz [6] and revived by Frischauf [7] and Peano [8], in which distance alone is fundamental, and the straight line is defined by its means. In this method distances are given to begin with as a class of relations which are the field of a certain transitive asymmetrical relation (greater and smaller); if we assume this relation to be continuous, distances will be measurable; all distances have the same domain and the same converse domain, namely all the points of the space in question; the locus of points equidistant from two fixed points is called a plane, and the intersection of two non-coincident planes, when it is not null, is called a straight line. (The definition of the straight line given by Peano [9] is as follows: The straight line ab is the class of points x such that any point y, whose distances from a and b are respectively equal to the distances of x from a and b, must be coincident with x.) Leibniz, who invented this method, failed, according to Couturat, to prove that there are straight lines, or that a straight line is determined by any two of its points. Peano has not, so far as I am aware, succeeded in proving either of these propositions, but it is of course possible to introduce them by means of axioms. Frischauf professes to demonstrate them, but his proofs are very informal, and it is difficult to know what axioms he is assuming. In any case, however, the definitions prove that, by a sufficient use of axioms, it is possible to construct a geometry in which distance is fundamental, and the straight line derivative. The method is so complicated as to be not practically desirable; but its logical possibility is nevertheless important.
§ 396. In most spaces, magnitude of divisibility can be used instead of distance¶
f. §396 · p. p. 411396. It is thus plain that the straight line must be independent of distance, while distance may be independent of the straight line. Taking both as symmetrical relations, we can, by a very complicated series of axioms, succeed in generating order on the straight line and in explaining the addition and measurement of distances. But this complication, in most spaces [10], is logically unnecessary, and is wholly avoided by deriving distances from stretches. We now start, as in descriptive Geometry, with an asymmetrical transitive relation by which the straight line is both defined and shown to be a series. We define as the distance of two points A and B the magnitude of divisibility of the stretch from A to B or B to A—for divisibility is a signless magnitude. Divisibility being a kind of magnitude, any two distances will be equal or unequal. As with all divisibilities, the sum of the divisibilities of AB and EF is the divisibility of the logical sum of the classes AB and EF, provided these classes have no common part. If they have a common part, we substitute for EF a stretch $E'F'$ equal to it and having no part in common with AB. The difference of the distances AB, EF (supposing AB the greater) is the divisibility of a stretch CD which, added logically to EF, and having no part in common with EF, produces a stretch equal to AB. It follows at once that, if A, B, C be collinear, and B be between A and C, AB + BC = AC and AC − AB = BC. No further axiom is required for these propositions. For the proposition that, if AB = $A'B'$ , and CD = $C'D'$, then AB + CD = $A'B'$ + $C'D'$, we require only the general axiom, applicable to all divisibilities, that the sums of equals are equal. Thus by the help of the axioms (3), (7), (8) above, we have everything that is required for the numerical measurement (theoretically speaking) of all distances in terms of any given distance, and for the proof that change of unit involves multiplication throughout by a common factor.
§ 397. Meaning of magnitude of divisibility¶
f. §397 · p. p. 411397. With regard to magnitude of divisibility, in the sense in which this is relevant to metrical Geometry, it is important to realize that it is an ordinal notion, expressing a property of relations, not of their fields. We wish to say that a stretch of two inches has twice as much divisibility as a stretch of one inch, and that an area is infinitely more divisible than a stretch. Now, if we are dealing (as will be assumed in this discussion) with a continuous space, every stretch, area or volume is a class of $2^{\alpha_0}$ terms; and considered as a class, it is the field of an infinite number of relations beside that (or those) belonging to it in respect of the space we are considering. The habit of allowing the imagination to dwell upon actual space has made the order of points appear in some way intrinsic or essential, and not merely relative to one of many possible ordering relations. But this point of view is not logical: it arises, in regard to actual space, only from the fact that the generating relations of actual space have a quite peculiar connection with our perceptions, and, through the continuity of motion, with time. From the standpoint of logic, no one of the relations having a given field has any preeminence, and the points of actual space, like any other class of $2^{\alpha_0}$ terms, form, with regard to other sets of generating relations, other sorts of continuous spaces—indeed any other continuous space, having any finite number of dimensions, or even ω dimensions, can be formed of the points of a Euclidean space by attending to other generating relations.
From this it follows that magnitude of divisibility, if it is to distinguish a long stretch from a short one, or an area from a stretch, must be a property of the relations involved, not of the class of points composing the area or the stretch. It is not quite easy to define the exact property which is required; for any two stretches are ordinally similar. We require some sense for the equality or inequality of the relations whose fields are the given stretches. Where coordinates (i.e. a correlation of the points of a line with the real numbers) have been already introduced, we may define the magnitude of a stretch as the difference of the coordinates of its end-points or its limits (according as the stretch has ends or not); but if this is done, the magnitudes of stretches will depend upon the necessarily more or less arbitrary plan upon which we have introduced our coordinates. This is the course adopted in the projective theory of distance—a course which has the merit of making metrical Geometry a logical development from projective axioms alone (see next chapter). The other course that may be adopted is, to assume that the generating relations of any two stretches have either a symmetrical transitive relation (equality), or an asymmetrical transitive relation or its converse (greater or less). Certain axioms will be required, as, for example, that if the points A, B, C, D are collinear, and AC is greater than AD, then BC is greater than BD [11]. The relations of equal, greater and less may be regarded as defined by these axioms, and the common property of the generating relations of those stretches that are equal to a given stretch may be defined as the magnitude of divisibility of the said generating relations. The sense in which an area has infinitely more divisibility than a stretch is that, if n be any finite integer, and n stretches equal to a given stretch be removed from an area, there always remains an area, however great n may be. What is important to observe, in the above discussion, is that the logical parity of all the orders of which a class of terms is capable makes it necessary to regard the magnitudes with which metrical Geometry deals as belonging to relations or classes of relations, not, as is commonly supposed, to the class of points forming their fields.
§ 398. Difficulty of making distance independent of stretch¶
f. §398 · p. p. 413398. In elliptic space, where the straight line is a closed series, the attempt to make distance independent of stretch leads to still further complications. We now no longer have the axiom that, if A, B, C be collinear, we cannot have AB = BC = CA; and we have to recognize two distances between every pair of points, which, when distance is taken as fundamental, becomes extremely awkward. We may however avoid admitting two distances by refusing to regard the greater of the two as properly a distance. This will then be only a stretch. If two distances are admitted, one is always greater than the other, except in a limiting case, when both are the lower limit of the greater distances and the upper limit of the lesser distances. Further if a, b, c, d be any four distinct points, the greater of the two distances ab is always greater than the lesser of the two distances cd. Thus the whole class of greater distances may be banished, and only greater stretches be admitted.
We must now proceed as follows. Distances are a class of symmetrical relations, which are magnitudes of one kind, having a maximum, which is a one-one relation whose field is all points, and a minimum, which is the distance of any point from itself. Every point on a given line has a given distance other than the maximum or minimum from two and only two other points on the line. If a, b, c, d be four distinct points on one line, we shall say that a and c are separated by b and d in the following four cases, of which (1) and (2) and also (3) and (4) are not mutually exclusive:
(1) If ab < ac . bc < ac . ad > ac.
(2) If ab < ac . bc < ac . dc > ac.
(3) If ab > ac . ad < ac . dc < ac.
(4) If bc > ac . ad < ac . dc < ac.
We then need Vailati’s five axioms enumerated in Part IV, Chap. xxiv, in order to generate a closed series from the separation of couples so defined. Thus it is possible, though by a somewhat complicated process, to generate a closed series of points on a line by means of the symmetrical relation of distance.
I shall not work out in further detail the consequences of this hypothesis in elliptic space, but proceed at once to the hypothesis that distances are the magnitudes of stretches. When the number of dimensions exceeds two, the polar form of elliptic space is merely projective space together with the necessary metrical axioms; the antipodal form is a space in which two antipodal points together have the properties of a single projective point. Neglecting the latter, to which similar remarks will apply, I shall confine myself to the polar form. Since this is a projective space, every pair of points determines two segments on the line joining the points. The sum of these two segments, together with the two points, is the whole line, and therefore constant. It is an axiom that all complete straight lines have the same divisibility. The divisibility of either segment is a distance between the two points: when the two distances are equal, either may be called the distance; when they are unequal, it will be convenient to call the smaller the distance, except in special problems. The whole theory then proceeds as in the case of descriptive space. But it is important to observe that, in elliptic space, the quadrilateral construction and the generation of order, being prior to stretches, are prior to distances, and are presupposed in metrical Geometry.
§ 399. Theoretical meaning of measurement¶
f. §399 · p. p. 414399. So far, therefore, metrical Geometry introduces three new axioms, and one new indefinable. The stretch in every series is a quantity, and metrical Geometry merely introduces such axioms as make all stretches of points measurable. A few words may be useful as to the sense in which, in a theoretical discussion, the word measurement is to be understood. The actual application of the foot-rule is here not in question, but only those properties of pure space which are presupposed in the use of the foot-rule. A set of magnitudes is theoretically measurable when there is a one-one relation between them and some or all numbers; it is practically measurable when, given any magnitude, we can discover, with a certain margin of error, what the number is to which our magnitude has the relation in, question. But how we are to discover this is a subsequent question, presupposing that there is such a proposition to be discovered, and soluble, if at all, by empirical means to be invented in the laboratory. With practical measurement, then, we are not at all concerned in the present discussion.
Footnotes