Philosophy of Geometry From Riemann To Poincaré
Philosophy of Geometry From Riemann To Poincaré
Philosophy of Geometry From Riemann To Poincaré
A PALLAS PAPERBACK
ROBERTO TORRETTI
University o f Puerto Rico
P H I L O S O P H Y OF
GEOMETRY
FROM R I E M A N N
TO P O I N C A R E
(Episteme; v. 7)
Includes bibliographical references and index.
1. Geometry-Philosophy. 2. Geometry-History.
I. Title.
QA447.T67 516\001 78-12551
ISBN-13: 978-90-277-1837-2 e-ISBN-13: 978-94-009-9909-1
DOI: 10.1007/978-94-009-9909-1
PREFACE xi
C H A P T E R 1 /B A C K G R O U N D 1
1.0.1 Greek Geometry and Philosophy 1
1.0.2 Geometry in Greek Natural Science 10
1.0.3 Modern Science and the Metaphysical Idea of Space 23
1.0.4 Descartes’ Method of Coordinates 33
C H A P T E R 2 /N O N -E U C L ID E A N G E O M E T R IE S 40
2.1 PARALLELS 41
2.1.1 Euclid’s Fifth Postulate 41
2.1.2 Greek Commentators 42
2.1.3 Wallis and Saccheri 44
2.1.4 Johann Heinrich Lambert 48
2.1.5 The Discovery of Non-Euclidean Geometry 53
2.1.6 Some Results of Bolyai-Lobachevsky
Geometry 55
2.1.7 The Philosophical Outlook of the Founders of
Non-Euclidean Geometry 61
2.2 MANIFOLDS 67
2.2.1 Introduction 67
2.2.2 Curves and their Curvature 68
2.2.3 Gaussian Curvature of Surfaces 71
2.2.4 Gauss’ T h e o re m a E g reg iu m and the Intrinsic
Geometry of Surfaces 76
2.2.5 Riemann’s Problem of Space and Geometry 82
2.2.6 The Concept of a Manifold 85
2.2.7 The Tangent Space 88
2.2.8 Riemannian Manifolds, Metrics and Curvature 90
2.2.9 Riemann’s Speculations about Physical Space 103
2.2.10 Riemann and Herbart. Grassmann 107
viii T A B LE OF CO N TE N TS
C H A P T E R 3 /F O U N D A T IO N S 153
3.1 HELMHOLTZ'S PROBLEM OF SPACE 155
3.1.1 Helmholtz and Riemann 155
3.1.2 The Facts which Lie at the Foundation of Geometry 158
3.1.3 Helmholtz’s Philosophy of Geometry 162
3.1.4 Lie Groups 171
3.1.5 Lie’s Solution of Helmholtz’s Problem 176
3.1.6 Poincare and Killing on the Foundations of
Geometry 179
3.1.7 Hilbert’s Group-Theoretical Characterization of
the Euclidean Plane 185
3.2 AXIOMATICS 188
3.2.1 The Beginnings of Modern Geometrical
Axiomatics 188
3.2.2 Why are Axiomatic Theories Naturally Abstract? 191
3.2.3 Stewart, Grassmann, Pliicker 199
3.2.4 Geometrical Axiomatics before Pasch 202
3.2.5 Moritz Pasch 210
3.2.6 Giuseppe Peano 218
3.2.7 The Italian School. Pieri. Padoa 223
3.2.8 Hilbert’s G rundla g en 227
3.2.9 Geometrical Axiomatics after Hilbert 239
3.2.10 Axioms and Definitions. Frege’s Criticism of
Hilbert 249
TA B LE OF C O NTE NTS ix
C H A P T E R 4 / E M P IR IC IS M , A P R IO R IS M ,
C O N V E N T IO N A L IS M 254
4.1 EMPIRICISM IN GEOMETRY 256
4.1.1 John Stuart Mill 256
4.1.2 Friedrich Ueberweg 260
4.1.3 Benno Erdmann 264
4.1.4 Auguste Calinon 272
4.1.5 Ernst Mach 278
APPENDIX 359
1. Mappings 359
2. Algebraic Structures. Groups 359
3. Topologies 360
4. Differentiable Manifolds 361
X TABLE OF CONTENTS
NOTES 375
To Chapter 1 375
To Chapter 2 379
Part 2.1 379
Part 2.2 382
Part 2.3 386
To Chapter 3 391
Part 3.1 391
Part 3.2 395
To Chapter 4 403
Part 4.1 403
Part 4.2 407
Part 4.3 409
Part 4.4 412
REFERENCES 420
INDEX 440
PREFACE
BACKGROUND
Though this and the preceding proofs are plainly meant to be illus
trated by diagrams in which the integers under consideration are
represented by straight segments (Fig. 1), such diagrams will not in
the least aid us to visualize the force of the argument. This rests
entirely on the meaning of the words odd, even, add, subtract, and
cannot therefore ‘be seen except by thought’.8 Had they not adopted
this method of exact, forceful, yet unintuitive thinking, Greek
mathematicians could never have found out that there are incom
mensurable magnitudes, such as, for example, pairs of linear seg
ments which cannot both be integral multiples of the same unit
segment, no matter how small you choose this to be. For, as B.L. van
der Waerden pointedly observes:
When we deal with line segments which one sees and which one measures empirically,
it has no sense to ask whether they have or not a common measure; a hair’s breadth
will fit an integral number of times into every line that is drawn. The question of
commensurability makes sense only for line segments which are objects of thought.9
Fig. 1.
4 CHAPTER 1
unavoidable if, as we have just seen, one of the most basic certainties
concerning the objects of this science could only be attained by
reasoning. After the discovery of incommensurables, geometers were
bound to demand a strict proof of every statement in their field (if
they were not already inclined to do so before). Now, if a statement
can only be proved by deriving (inferring, deducing) it from other
statements, it is clear that the attempt to prove all statements must
lead to a vicious circle or to an infinite regress. It is unlikely that
Greek mathematicians had a clear perception of the inadmissibility of
circular reasoning and infinite regress before Aristotle. But they
instinctively avoided these dangers by reasoning always from
assumptions which they did not claim to be provable. In a well-known
passage of the Republic, Plato speaks disparagingly of this feature of
mathematical practice: Since geometry and the other mathematical
disciplines are thus unable to give a reason (logon didonai) for the
assumptions (hupotheseis) they take for granted, they cannot be said
to be genuine sciences (epistemai)." The name science, bestowed on
them out of habit, should be reserved for dialectic, which “does away
with assumptions and advances to the very principle (auten ten
arkhen) in order to make her ground secure” .12 Having been born too
late to benefit from Plato’s oral teaching, I find it very difficult to
determine how he conceived of the dialectician’s ascent to the unique,
transcendent principle which he claimed to be the source of all being
and all truth. His great pupil, Aristotle, succeeded in disentangling
Plato’s main epistemological insights from his mystical fancies and
built a solid, sober theory of science that has tremendously influenced
the philosophical understanding of mathematics until quite recently.
In Aristotle’s terminology, science (episteme) is equated with know
ledge by demonstration (apodeixis).13 But it is subordinated to a
different kind of knowledge, which Aristotle calls nous, a word that
literally means intellect and that has been rendered as rational in
tuition (G.R.G. Mure) and as intuitive reason (W.D. Ross).14 Nous
gives us an immediate grasp of principles (arkhai), that is, of true,
necessary, universal propositions, which cannot be demonstrated-
except, I presume, by resorting to their own consequences - but
which are self-evident and provide the ultimate foundation of all
demonstrations in their respective fields of knowledge. Intellection of
principles is attained by reflecting on perceived data, but it is
definitely not an effect of sense impressions modifying the mind. It
BA CK GR OU ND 5
Is aitema just another name for that what Aristotle called hupo-
thesisl This, as the reader will recall, is a self-evident statement of
existence concerning the subject-matter of a particular science. The
five aitemata listed above all pertain specifically to geometry. The
fifth can be read as an existential statement.24 The first three, on the
8 CHAPTER 1
We say that a has to b a greater ratio than a' has to b' (alb > a'lb')
if, for some pair of positive integers m, n, we have that ma > nb, but
m a '^ n b 'P Using these definitions, we can compare the quantitative
relations between any pair of homogeneous magnitudes with that
between two straight segments. Eudoxian ratios are linearly ordered
by the relation greater than; they can be put into a one-one order
preserving correspondence with the positive real numbers and one
can calculate with them as with the latter. The importance of
Eudoxus’ innovation for geometry is evident and demands no further
comment. It also has a direct bearing on the Pythagorean programme.
This had failed because there are relations between things which
cannot possibly correspond to relations between numbers. But even if
numbers and their ratios were inadequate for representing every
conceivable physical relation, one could still expect Eudoxian ratios
to do this job. After all, the same reasons that had initially supported
physical arithmology could be used to justify the project of a more
broadly-conceived mathematical science of nature. In fact, Eudoxus
himself, as the founder of Greek mathematical astronomy, set some
such project going at least in this branch of physical inquiry. And, in
the following centuries, Archimedes of Syracuse (287-212 B.C.) and a
few others made lasting contributions to statics and optics. But in his
pioneering search for a mathematical representation of ordinary ter
restrial phenomena, Archimedes - “suprahumanus Archimedes” , as
Galilei would call him - did not gather much of a following until the
12 CHAPTER 1
It is not easy to take in what I mean, nor yet is it very difficult or a very long business:
witness the fact that, although it is not a thing which I learnt when I was young or very
long ago, I can now, without taking much time, make it known to you: whereas, if it
had been difficult, I, at my age, should never have been able to explain it to you at
yours. [ . . . ] This view which is held about the moon, the sun, and the other stars, to the
effect that they wander and go astray (planatai), is not correct, but the fact is the very
contrary of this. For each of them traverses always the same circular path, not many
paths, but one only, though it appears to move in many paths.
The whole path and movement o f heaven and of all that is therein is by nature akin
to the movement and revolution and calculations of intelligence (nous).40
But Plato does not recant his former evaluation of nature and of the
prospects of natural science. He only concludes, in good agreement
with traditional Greek piety, that the heavens must be set apart from
the rest of the physical world. The planets would not follow those
“wonderful calculations with such exactness” if they were soulless
beings, destitute of intelligence.41 The same point is made more
explicitly in the Epinomis, a supplement to the Laws which many
20th-century scholars have attributed to Plato’s pupil, the mathema
tician Philip of Opus, but which in Antiquity was believed to be the
work of Plato himself.
It is not possible that the earth and the heaven, the stars, and the masses as a whole
which they comprise should, if they have no soul attached to each body or dwelling in
each body, nevertheless accurately describe their orbits in the way they do, year by
year, month by month, and day by day.42
(i) The sun does not traverse all four quadrants of the Zodiac in the
same time - the period from a solstice to the next equinox is not equal
with that from that equinox to the other solstice; hence, the angular
velocity of the sun about the earth cannot be constant.
(ii) The apparent luminosity of the planets and the apparent size of
the moon are subject to considerable fluctuations; hence their
distances from the earth cannot be constant.
Callippus sought to cope with fact (i) by adding two more spheres to
Eudoxus’ solar model, but fact (ii), of course, was totally in
compatible with the Aristotelian system of concentric spheres. Third-
century astronomers managed to account, on a first approximation,
for both facts by assuming that each planet moves with constant
angular speed on an eccentric, that is, on a circle whose centre is not
the centre of the earth but which contains the latter in its interior.
Then, towards the end of that century, some hundred years after the
death of Aristotle, Apollonius of Perga (2657-170 B.C.) introduced an
BA CK GR OU ND 19
astronomer need only know that if the epicycles and eccentrics did
exist, the celestial motions and the other phenomena would exist as
they do now.
The truth of the conditional is what matters, whether or not such orbits really exist
among the heavenly bodies. The assumption of such eccentrics and epicycles is
sufficient for the astronomer qua astronomer because as such he need not trouble
himself with the reason why (unde). Provided he has the means of correctly determin
ing the places and motions of the planets, he does not inquire whether or not this means
that there really are such orbits as he assumes up in the sky [. .. ]. For a consequence
can be true even when the antecedent is false.54
the former as of something that occupies a part of the latter. Since the
atoms are eternal and uncreated, there can be no question of their
‘taking up’ or ‘being received by’ the void. The void surrounds the
atoms and these move about in it: but the void is not conceived as an
underlying continuum that is partly empty and partly full. While the
atomists allowed the void to permeate the infinity of atoms, which
coalesced in infinitely many independent systems or kosmoi
(‘worlds’), the Stoic school, founded towards 300 B.C. by Zeno of
Citium, maintained that there is but one kosmos, which is finite and
tightly packed with matter, and is surrounded by a boundless void
(kenon). The union of void and kosmos they called to pan, i.e. the All.
The Stoic All can be said to contain all the points demanded by
Euclid, but there is no evidence that the Stoics had geometry in mind
when they developed their doctrine.
A likelier antecedent of the modern connotations of ‘space’ can be
found in the use of spatium by Lucretius (98-55 B.C.) in his
didactic poem De rerum natura. Here, to kenon becomes “locus ac
spatium quod inane vocamus” (the place and space that we call
void-1,426, etc.). The adjective inanis (‘empty’, ‘void’) immediately
suggests the contrast with a “locus ac spatium plenum” , a full place
and space. If this or a similar expression were used in the poem I
would not doubt that Lucretius did conceive spatium as a medium
that was partly empty and partly filled with bodies. The best examples
I have chanced upon are line 1,522, where bodies are said to hold and
fill places, not space (“[corpora] quae loca complerent quacumque
tenerent”); and lines 1,526 f., which W.H.D. Rouse translates: “There
are therefore definite bodies to mark off empty space from full.” The
last passage would satisfy my requirements if the translation were
exact. But Lucretius wrote
sunt ergo corpora certa
quae spatium pleno possint distinguere inane,
and it seems more natural to read plenum as a noun - ‘the full’ -
standing for that which bodies are said to mark off from spatium
inane, ‘the void’. On the other hand, several passages contrast, in the
best atomist tradition, body as such - not space occupied by body -
with the void, the empty void, empty space or simply space. But even
if Lucretius never meant to sing the modern idea of space, some of
his hexameters must have conjured it up in the minds of his modern
readers .72
28 CHAPTER 1
the Latin dissertation On the form and the principles of the sensible
and the intelligible world (1770). It is, in fact, the core of the
platonizing theory of the principles of human knowledge outlined in
that work. The problems raised by this theory forced its abandonment
and led Kant to his vaunted revolution in philosophy. The theory of
space and time is presented in the Transcendental Aesthetic, the first
part of the Critique o f pure reason (first edition, 1781; second,
revised, edition, 1787), almost in the same terms as in the Latin
dissertation. A consistent reading of Kant’s critical philosophy
requires however that those terms be qualified in the light of the next
two parts, the Transcendental Analytic and the Transcendental
Dialectic. Since most philosophers, outside the narrow circle of Kant
specialists, have paid scarcely any attention to this requirement, the
straightforward, precritical philosophy of space and geometry
developed by Kant in 1770 has played a much greater role in the
history of thought than the subtler, more elusive doctrine that might
be gathered from the entire Critique of 1781 and 1787 and from his
other critical writings.
The doctrine of 1770 follows a familiar metaphysical scheme. The
human mind is regarded as a substance that interacts with other
substances. The capacity of the human mind to have its state of
consciousness (status repraesentativus) modified by the active
presence of an object is called sensibility (sensualitas). The
modifications caused thereby are called sensations. The modifying
object can be the mind itself or another substance; in the latter case,
it is said to be outwardly or externally sensed .82 Sensations are a
source of knowledge of the sensed object; indeed, they are the only
source of direct knowledge of individual objects that is available to
man. Such knowledge by direct acquaintance is called by Kant
intuition (intuitus, Anschauung). Human knowledge of reality rests
therefore entirely on sense intuition. Intuitive knowledge of an object
is brought about by the combination of sensations arising from the
object’s presence into a coherent presentation of the object itself.
Such a combination of sensations is governed by a “law inherent in
the mind ” 83 of which space is a manifestation. For “space is not
something objective and real, neither a substance, nor an attribute,
nor a relation, but a subjective and ideal schema for coordinating
everything that is externally sensed in any way, which arises from the
nature of the mind according to a stable law ” .84 We shall not discuss
BA CK GRO UND 31
self-consistent, necessarily be consistent with the latter, and the laws of sensibility will
be laws of nature insofar as it can be perceived by the senses (quatenus in sensus cadere
potest ). Hence nature complies exactly (ad amussim) with the precepts of geometry
regarding all the properties of space demonstrated in this science, on the strength not of
a feigned presupposition ( hypothesis), but of one that is intuitively given as the
subjective condition of all phenomena that nature can exhibit to the senses.88
(Kl) they are both positive or both negative and the Eudoxian ratios
XY/OE and X'Y'/OE are equal, or if (K2) they are both null. These
definitions can be extended to any other Euclidean line m ’. Choose
O' and E' on m' so that the Eudoxian ratio O'E'/OE equals
OE/OE. Define linear order on m' as before. Let M be the set of all
Euclidean lines ordered in this way. Two directed segments belonging
to the same or to different lines of M are said to be congruent if they
satisfy (Kl) or (K2). The reader ought to verify that congruence of
directed segments is an equivalence. A directed linear magnitude
(dim) is an equivalence class of congruent directed segments. If XY is
any directed segment, we let [XY] denote the dim to which it belongs.
[XY] is positive, negative or null if XY is, respectively, positive,
negative or null. It will be easily seen that, for every point X on an
ordered line m, each dim has one and only one member whose first
endpoint is X and whose last endpoint lies on m. In particular, this is
true of the origin O. Consequently, as X ranges over the set of points
of m, [OX] ranges over the set of dim’s. Let X and Y be points of m.
We say that [OX] is less than [OY] (abbreviated: [OX] < [OY]) if and
only if X < Y. If [OX] < [OY] we also say that [OY] is greater than
[OX]. The relation < obviously defines a linear ordering of the set of
dim’s.
Let 2 denote the set of dim’s ‘gauged’ by the choice of a segment
OE. 2 will be endowed with a field structure. Let [OX], [OY] be two
dim’s. Let m be the line through O and X (Fig. 2). There is one and
only one member of [OY] whose first endpoint is X and whose last
endpoint lies on m. Let W be this last endpoint. We define [OW] to be
the sum ([OX] + [OY]) of [OX] and [OY]. The reader should satisfy
himself that ‘+ ’ is an operation on 2 and that (2, +) is an Abelian
group. It should be clear, in particular, that [OO] is the neutral
J ___________I_________________I
Y 0 W X
Fig. 2.
BA CK GR OU ND 37
element of the group and that [XO] is the inverse -[O X ] of [OX]. To
define the second field operation or product of two dim’s we shall use
the fact that our ordered lines are embedded in ordinary Euclidean
space. Descartes based his own construction on Euclid VI, 2: “If a
straight line be drawn parallel to one of the sides of a triangle, it will
cut the other sides [or those sides produced] proportionally.” Let m
be a line ordered as above, with respect to points O and E. Let line
m' meet m at O (Fig. 3). Choose E' on m' so that OE'/OE =
OE/OE. We now define the product [OX] • [OY] of two dim’s [OX],
[OY], as follows: Let OY' be the member of [OY] whose first
endpoint is O and whose last endpoint lies on m'; let Z be the point
where m meets the parallel to E'X through point Y'; then [OZ] =
[OX] • [OY], It can be easily verified that ‘-’ is indeed an operation
on X; that for every dim [OX], [OX] • [OE] = [OE] • [OX] = [OX]; and
that every non-null dim [OX] has a reciprocal dim [OX]-1 such that
[OX] • [OX]-1 = [OX]-1 • [OX] = [OE], The reciprocal of [OX] can be
constructed thus: Let 21 be the point where m' meets the parallel to
E'X through E; then [OZ'] = [OX]-1. The reader should verify that
the operation *•’ is commutative and associative, so that (X , +, •) is
indeed a field, with zero element [00] and unity [OE]. (X, +, •) is
ordered by the relation < . It can be shown moreover that, if line
m has the property (D), (X, +, •) is complete. Since all complete
ordered fields are structurally equivalent, they are indistinguishable
from a mathematical point of view. Any such field is usually called
the real number field and is designated by the symbol R. Its elements,
qua elements of R, are called real numbers.97
The set of all ordered triples of elements of R is denoted by R3. We
shall show how to label each point of Euclidean space with an
element of RJ. In other words, we shall define a bijective mapping of
38 CHAPTER 1
the set of Euclidean points onto R3. To do this, we first define the
directed distance from a plane to a point. Let tt be a plane, tt has two
sides, which we conventionally label the positive and the negative
side, respectively. Let P be a point not on 7r, and Q its perpendicular
projection on tt (i.e. the point where a line through P meets tt at right
angles). There is one and only one positive dim [OX], such that
PQ/OX = OX/PQ. The directed distance from tt to P is [OX] if P lies
on the positive side of tt\ -[OX] if P lies on the negative side of tt. If
P lies on 7r, its directed distance from tt is [OO]. Now, let ttu tt2, tt3
be three mutually perpendicular planes. Let /*(P) be the directed
distance from 7r, to point P (i = 1,2,3). We assign to P the ordered
triple ( f l(P), f \ P), / 3(P)). It will be easily seen that this rule defines a
bijection of the set of all Euclidean points onto R3. For the points at
directed distance /*(P) from 7n lie all on a plane on a definite side of
7T/ and at a definite distance from that plane; our ordered triple
determines therefore three mutually perpendicular planes which meet
only at P. On the other hand, for every ordered triple of dim’s ([OX!],
IQXJ, [OX3]) there is a point X whose directed distance from 7r,- is
[OX,].
We shall hereafter use the following terminology: A bijective
mapping F*-*(fl(P ),f2(P)9f \ 1P)) of Euclidean space onto R3, con
structed according to the above directions, will be called a Cartesian
mapping. Please observe that the definition of a Cartesian mapping
involves the arbitrary choice of an ordered triple of planes, of the
positive side of each of these planes and of a segment OE as a gauge
for distances. The ordered triple of planes that enter into the
definition of a Cartesian mapping is the frame, their point of inter
section, the origin of the mapping. The frame (7Ti, 7r2, 7r3) of a
Cartesian mapping is said to be right-handed if the following condi
tion is fulfilled: If I place my right hand at the origin, with the thumb
pointing toward the positive side of 7Ti, and the index finger pointing
toward the positive side of 7t2, I can bend the middle finger so that it
points toward the positive side of 7r3. The frame is left-handed if the
foregoing condition is fulfilled with the word ‘left’ substituted for
‘right’. All frames of Cartesian mappings are either left-handed or
right-handed. In this book, unless otherwise stated, we assume that
they are right-handed. The three real numbers assigned to a point P
by a Cartesian mapping are called the coordinates of P by this
mapping. If / and g are two Cartesian mappings, the composite
BA CK GR OUN D 39
N O N -E U C L ID E A N G E O M E T R IE S
2.1 PARALLELS
deny it, since that would imply that there are no similar figures of
arbitrarily different sizes - in particular, no cubes or squares, for these
figures can obviously be multiplied through mere juxtaposition.
In 1733, Girolamo Saccheri, a jesuit well versed in the literature on
the problem of parallels, published a treatise whose Book I deals with
the subject.9 He proposes to prove Postulate 5 by a method not yet
tried, that of indirect proof. Saccheri attempts to show that the denial
of Postulate 5 is incompatible with the remaining familiar assump
tions of geometry. Since he was probably aware that the proofs in
Book I of the Elements often depend on unstated premises, he
chooses to treat the first 26 propositions of that book, which, as we
know, do not depend on Postulate 5, as undemonstrated principles in
his argument. In fact, he also employs as implicit assumptions the
Archimedean postulate - if a and b are two straight segments, there
exists an integer n such that n a > b - and a principle of continuity
that may be stated as follows: If a continuously varying magnitude is
first less and then greater than a given magnitude, then at some time it
must be equal to it. Saccheri considers a certain plane figure, now
known as a Saccheri quadrilateral. To construct one, take a straight
segment AB and draw two equal perpendiculars AD and BC; the
four-sided polygon ABCD is a Saccheri quadrilateral. It is easily
proved that the angles at C and at D are equal. Saccheri proposes
three alternative hypotheses: that both angles are right angles, or that
El
B
Fig. 4.
46 CHAPTER 2
both are obtuse, or that both are acute. We shall, for brevity, speak of
Hypotheses I, II and III. (Fig. 4.) Saccheri shows that if one of them is
true in a single case, it is true in every case. Postulate 5 obtains under
Hypotheses I and II (Proposition XIII), but Hypothesis II happens to
be incompatible with some of the propositions initially admitted by
Saccheri. Postulate 5 will be proved if we manage to show that
Hypothesis III is also incompatible with that set of propositions. This
is a long and laborious enterprise that takes up most of Saccheri’s
book.
The incompatibility of Hypothesis II with the set of initial assump
tions is established in Proposition XIV. Its proof depends on Pro
position IX, which states that, in a right triangle, the sum of the two
angles adjacent to the hypotenuse is equal to, greater than or less than
the remaining angle if Hypothesis I, II or III is true. Let APX be a
right triangle (Fig. 5). If Hypothesis II is true, the angles at X and A
are together greater than a right angle. We can find therefore an acute
angle PAD such that all three angles together are equal to two right
angles. Under Hypothesis II, Postulate 5 obtains; consequently PL
and AD meet at a point H. Triangle XAH has two interior angles
adjacent to side AX which are together equal to two right angles. This
contradicts Euclid 1.17. Hypothesis II is therefore incompatible with
the set of initial assumptions.
In the course of his fight against Hypothesis III, Saccheri draws
from it several conclusions which today are well-known propositions
of BL geometry. The existence of one triangle whose three internal
angles are equal to, more than or less than ir, is sufficient to validate
Hypothesis I, II or III, respectively (Proposition XV). If AB is a
straight segment, AK a straight line normal to AB, and BD a ray on
the same side of AB as K and making with AB an acute angle on the
side of AK, it may, under Hypothesis III, very well happen that BD
does not meet AK. (Fig. 6.) Let BR be normal to AB (with R on the
N ON -E UC LI DE A N GEOMETRIES 47
_□___________________
A K
Fig. 6.
same side of AB as K). All rays BD fitting the above description fall
within angle ABR. Saccheri proves that, under Hypothesis III, some
of these rays meet AK while others share with it a common perpen
dicular. If we order the former group of rays according to the
increasing size of the angle they make with AB, and the latter group
according to the increasing size of the angle they make with BR, we
shall find that none of these two sequences of rays possesses a last
element. For reasons of continuity, Saccheri concludes that between
the two sequences there exists one and only one ray which does not
meet AK and does not share with it a common perpendicular. The
straight line comprising this ray approaches AK indefinitely. There
fore, under Hypothesis III, there exist asymptotical straight lines as
Proclus surmised.
Saccheri chooses to state these results with the help of the some
what tricky notion of an infinitely distant point. Let T be such a point
on AK, i.e. a point beyond K and beyond every other point of AK
which is at a finite distance from A. The rays of one of our sequences
meet AK at points increasingly distant from A; consequently, argues
Saccheri, the limiting ray of that sequence meets AK at T. The rays
of the other sequence are met by perpendiculars which meet AK
orthogonally at points ever more distant from A; consequently, the
limiting ray of this sequence must have a perpendicular which meets
AK orthogonally at T. But the limiting ray of both sequences is the
same ray BT. Therefore BT and AK have a common perpendicular at
one and the same point. But this is absurd, for two different straight
lines cannot both meet another line perpendicularly at one p o in t-if it
is true that all right angles are equal (Euclid, Postulate 4) and that two
different straight lines cannot have a common segment. Saccheri does
not ask himself whether everything that is true of ordinary points is
necessarily true of an infinitely distant point. It would have been
48 CHAPTER 2
The difficulties concerning Euclid’s 11th axiom [i.e. Postulate 5] have essentially to do
only with the following question: Can this axiom be derived correctly from Euclid's
,
postulates and the remaining axioms ? Or, if these premises are not sufficient can we
, ,
produce other postulates or axioms no less evident than Euclid's from which his 11th
axiom can be derived ? In dealing with the first part of this question we may wholly
ignore what I have called the representation of the subject-matter [Vorstellung der
Sache]. Since Euclid’s postulates and remaining axioms are stated in words, we can
and should demand that no appeal be made anywhere in the proof to the matter itself,
but that the proof be carried o u t - i f it is at all p o ssib le -in a thoroughly symbolic
fashion. In this respect, Euclid’s postulates are, so to speak, like so many given
algebraic equations, from which we must obtain x, y, z, etc., without ever looking back
to the matter in discussion [die Sache selbst]. Since the postulates are not quite
such formulae, we can allow the drawing of a figure as a guiding thread [Leitfaden] to
direct the proof. On the other hand, it would be preposterous to forbid consideration
and representation of the subject-matter in the second part of the question, and to
require that the new postulates and axioms be found without reflecting on their
subject-matter, off the cuff, so to speak.11
the third, which are, that the fourth angle is a right angle, that it is
obtuse, or that it is acute.12 In three separate sections, Lambert
derives consequences from each of these hypotheses. In the proofs
based on the second one, he studiously avoids using any of the
propositions in Euclid which are incompatible with it (1.16 and its
consequences): only towards the end of the section does he appeal to
one of those propositions, and this just in order to carry out the
refutation of the 2nd hypothesis. I think that this procedure ought to
be understood in the light of Lambert’s formalism which naturally
leads him to explore the possibility of a consistent deductive system
based on the 2nd hypothesis, no less than that of one based on the
3rd. On the other hand, Lambert the intuitionist knows of a
“representation of the subject-matter” which satisfies the 2nd hypo
thesis, if only we agree to give an appropriate interpretation to the
intrinsically meaningless terms of the corresponding formal system.
“It seems remarkable to me” - h e w rite s -“that the 2nd hypothesis
should hold when we consider spherical triangles instead of plane
triangles” ,13 in other words, when we understand by ‘straight lines’
the great circles on a sphere. These, as is well known, always contain
the shortest path between any two points lying on them. But they are
closed lines, and they intersect each other at more than one point; so
they do not share those properties of ordinary straight lines used in
the refutation of the 2nd hypothesis. Even more surprising are the
next two remarks of Lambert: (i) The geometry of spherical triangles
does not depend upon the solution of the problem of parallels, for it is
equally true under any of the three hypotheses; (ii) the 3rd hypo
thesis, in which the fourth angle of Lambert’s quadrilateral is
assumed to be less than a right angle, might hold true on an imaginary
sphere, i.e. on a sphere whose radius is a pure imaginary number.14
We have seen that Lambert had a formalist conception of mathema
tics which likened the premises of a deductive system to a set of
algebraic equations whose terms may denote any object satisfying the
relations expressed therein. He also discovered the modern idea of a
model, that is, of an object or domain of objects which happens to
fulfil precisely the conditions abstractly stated in the hypotheses of
the system. Such content is supplied by the “representation of the
subject-matter”, which, according to Lambert, ought to guide the
selection of hypotheses. Lambert’s last remark shows how broadly he
conceived of this kind of “representation”, for an imaginary sphere is
N ON -E U C LI D EA N GEOMETRIES 51
m - on the plus side of t equal to or less than one right angle (the plus
set) and the set of all lines making such an angle on the minus side of
t (the minus set). There is one and only one line common to both sets,
namely the perpendicular to /; let us call it n. Henceforth, we shall
consider only the plus set; whatever we learn about it applies by
symmetry, mutatis mutandis, to the minus set. For every point of m
on the plus side of t there is a line of the plus set meeting m at that
point. There is at least one line of the plus set which does not meet m,
for n is such a line. This justifies the following definition: A line s of
pencil (P, m ), making an interior angle cr on the plus side of t, is the
parallel to m in the plus direction if and only if s does not meet m but
m meets every line in (P,m) which makes an interior angle smaller
than a on the plus side of t.
Definitions of parallelism essentially identical to this were adopted
independently by Gauss, Bolyai and Lobachevsky.33 The reason for
abandoning Euclid’s definition is this: If Postulate 5 is true the new
and the old definitions are equivalent; if Postulate 5 is false, there are
two kinds of lines in (P, m) which do not meet m, namely the two
parallels, one in each direction, and an infinite set of lines between
them. These lines, called hyperparallels by some, have important
properties not shared by the two parallels; e.g. each of them has a
perpendicular which is also normal to m. Euclid’s definition, however,
makes no distinction between these two kinds of lines, for according
to it all of them are ‘parallels’.
It is clear that there is one and only one line through P which is
parallel to m in the plus direction. It can be shown that, if s is that
line, and P' is any point on s, then s is the one line through P' that is
N O N - E U C L I D E A N GEOMETRIES 57
function that takes all values between v/2 and 0 as x goes from 0 to
oo.34 Of course, if Postulate 5 is true, ri(jc) = constant = 7t/ 2 . In the
discussion to follow, we shall disregard this case.
We shall not prove Lobachevsky’s full statement but shall show
that, if II(jc) < tt/2 , 0 < x < x' implies that II(jc) > n(x'). Consider a
point Q on a line m and a line PQ perpendicular to m. Let |PQ| = x.
Produce PQ beyond P to P' and let |P'Q| = x'. Let s and s' be the
parallels to m in a given direction k that go, respectively, through P
and P'. Since s and s ’ are parallel to one another, they cannot have a
common perpendicular. Consequently, II(x) ^ II(x'). (Proof: Let H be
the midpoint of PP'; the perpendicular to s through H meets s at R, s'
at R'; if II(jc) = II(x'), triangle PHR is congruent with triangle P'HR'
and RR' is also perpendicular to s'.) If II(x) < II(x') there is a line
through P' that makes an interior angle equal to II(x) on side k of PQ
and meets m at a point G. P'G and s have a common perpendicular
(Proof: By the above construction) and P'G meets s between P' and
G. But this is impossible. Consequently, II(x)>II(x'). Q.E.D.
II is an injective or ‘one-one’ mapping of the positive real numbers
onto the open interval (0, ir/2). The inverse mapping II"1 is therefore
defined on (0, vl2) and enables us to express length in terms of
angular measure. This is the surprising feature of BL geometry which
even Gauss and Lobachevsky judged paradoxical.35 Angular measure
is absolute and involves a natural unit; while length, we are wont to
believe, is essentially relative, so that a sudden duplication of all
distances in the universe would make no difference whatsoever to the
geometrical aspect of things. This is not true of a BL world as the
following example will show. Let C be a positive real number such
that n(C) = 7r/4 . Take a point Q on a straight line m and let C be the
length of a segment PQ that meets m orthogonally at Q. Each of the
parallels to m through P makes an interior angle equal to ir/4 on the
corresponding side of PQ (Fig. 8). Consequently, the two parallels are
Q
Fig. 8.
N ON -E U C L I D E A N GEOMETRIES 59
symmetric and transitive relation. Consider now the set M of all lines
in space that are parallel to m in direction k . We call M (including m)
a family of parallels in space. The locus of the points corresponding
to Q on each of these lines is a smooth surface X which we call a
horosphere. We say that m is an axis of X. The name horosphere is
meant to remind us of the following: if horosphere cuts axis m at
Q and if P is a point of m on the concave side of X, a sphere with
centre P and radius PQ touches X but does not intersect it; as PQ
increases beyond all bounds, the sphere indefinitely approaches X -
in other words, a horosphere is the limit (horos) towards which a
sphere tends as its radius tends towards infinity. Let us now consider
any line m* belonging to set M, i.e. any line parallel to m in direction
k \ if m' intersects X at Q', X is clearly the locus of the points
corresponding to Q' on every line of M. Each line of M is therefore an
axis of Si?. Every axis of X is normal to X. All horospheres are
congruent. A plane £ through an axis of a horosphere X intersects X
on a curve h which we call a horocycle; h is clearly a locus of
corresponding points on the parallels to the given axis that lie on
plane £. On the other hand, if no axis of X lies on £ and £ meets X,
their intersection is a circle. A horocycle is an open curve; a horocy
cle lying on a horosphere divides it into two parts; all horocycles are
congruent; moreover, two arcs of horocycle are congruent if they
subtend equal chords. Horocycles, as we see, share some of the
properties of straight lines. Consider two mutually intersecting horo
cycles h, k on a horosphere X, determined by two mutually intersec
ting planes £, 17. Let w be the intersection of £ and 17. We agree to
measure the curvilinear angle formed by h and k by the rectilinear
angle made by two straight lines lying on £ and 17, respectively, and
meeting w perpendicularly at the same point. Now let a, b, c be
horocycles on a horosphere X such that c cuts a and b making
interior angles on the same side of c less than two right angles; then,
as both Bolyai and Lobachevsky proved, a and b meet. Bolyai
concluded without more ado: “From this it is evident that Euclid’s
Axiom XI [i.e. Postulate 5] and all things which are claimed in
geometry and trigonometry hold good absolutely in [horosphere X],
L-form lines [i.e. horocycles] being substituted in place of
straights” .37 Lobachevsky shows this in detail. He proves in particular
that the three interior angles of a horospherical triangle - i.e. a figure
limited by three horocycles on a horosphere - are equal to 7r and that
N O N - E U C L I D E A N GEOMETRIES 61
Horocycles satisfy the last three statements, but not the first; this
suffices to show where lies the truth about genuine straight lines if
Postulate 5 is false. The decision to use the word ‘straight’ in the
sense set by the four postulates above is of course conventional, and
if it turns out that in the world there are no straight lines in this sense,
we will probably give up some of those postulates rather than do
without such a familiar word. (This is indeed what pilots do when
they speak of flying from Sydney straight to Vancouver.) But the
point here is that Lobachevsky’s usage does not constitute a depar
ture from the established conventions governing this word, con
ventions which, as Proclus and most probably Euclid himself knew,
do not require that only such lines as fulfil Postulate 5 be called
straight - quite the contrary: it may very well happen that the lines
fulfilling this postulate are horocycles, which, by those conventions,
ought not to be called straight.
The indeterminateness of the characteristic constant of BL
geometry prompted attempts at determining it experimentally, on the
analogy of other physical constants familiar to 19th-century scientists.
The successful application of Euclidean geometry in science and in
everyday life indicated that the constant, if finite, must be very large.
Schweikart wrote that if the constant was equal to the radius of the
earth it would be practically infinite in comparison to the magnitudes
we deal with in everyday life (thus vindicating the use of Euclidean
geometry for ordinary purposes). Gauss commented that “in the light
of our astronomical experience, the constant must be enormously
larger (unermesslich grosser) than the radius of the earth”.40
Lobachevsky tried to evaluate the constant by using astronomical
data. Rather than look for an actual example of a line parallel to two
64 CHAPTER 2
To explain this idea, we assume that [. . . ] attractive forces decrease because their effect
is diffused upon a spherical surface. In ordinary geometry the area of a spherical
surface of radius r is equal to 47rr2, so that the force must be inversely proportional to
N ON -E U C LI D EA N GEOMETRIES 65
the square of the distance. I have found that in imaginary geometry the surface of a
sphere is equal to 7r(er - e~r)2; such a geometry could possibly govern molecular
forces, whose variations would then entirely depend on the very large number e.45
so that, from now on, everything else in geometry will be analysis, wherein calculations
will necessarily agree and where nothing will be able to disclose to us something new,
i.e. something that was not contained in those first equations from which all relations
between geometrical magnitudes must be derived. Therefore, if somebody unflinchingly
maintains that a subsequently emerging contradiction will force us to reject the
principles we have assumed in this new geometry, such a contradiction must already be
contained in equations (17). Let us observe however that these equations become
equations (16) of spherical trigonometry as soon as we substutute aV(-l), b V ( - 1),
cV(-l) for sides a, b, c. But in ordinary geometry and in spherical trigonometry we
only encounter relations between lines; consequently, ordinary geometry, trigonometry
and this new geometry will always stand in mutual agreement.46
66 CHAPTER 2
2.2 MANIFOLDS
2.2.1 Introduction
By 1840, a full statement of Lobachevsky’s theory had been made
available in French and in German. Contrary to Gauss’ expectation,
no uproar was heard. Most mathematicians ignored the extravagent
Russian, but some took a deep interest in the new geometry. Postu
late 5 had long been sensed by many as a mildly painful thorn in the
“supremely beutiful body of geometry” (to borrow Henry Savile’s
words).1 We thus find Bessel, in his reply to the letter where Gauss
expressed his fear of Boeotians, not unreceptive to the new ideas.
What Lambert has written and what Schweikart said have made it clear to me that our
geometry is incomplete and should be given a hypothetical correction, which vanishes
if the sum of the angles of a plane triangle equals 180 degrees. That would be the true
geometry, while Euclidean geometry would be the practical one, at least for figures on
the earth.2
£ ( X * C ( m» du. (2)
A (s) = J d u = s, (3)
o
as it ought to be.
Let c be an injective path defined on [0, A], with arc length as
70 CHAPTER 2
namely
(6)
The total curvature of c([s,, s2]) is therefore equal to the length of the
tangential image c'([si, s2]).
We have defined the curvature of a curve at a point as the
magnitude of an element of R3, i.e. as a non-negative real number. In
the case of plane curves, it is possible to define a signed curvature.
Mathematicians have not failed to use this possibility in order to
convey, through the value of the curvature, one more item of in
formation about the curve. Orientation conventions establish a posi
tive and a negative sense of rotation about a point in the plane. The
signed curvature k ( s ) at the point c(s) is equal to k (s ) if the tangent
to the curve at c(s) is constant or rotates about c(s) in a positive
sense; k ( s ) = - k (s ) if the tangent at c(s) rotates about this point in a
negative sense.
2.2.3 Gaussian Curvature o f Surfaces
As we did with smooth curves, we shall make our notion of a smooth
surface more precise by imposing certain conditions on the admissible
analytical representations of such surfaces. Let y‘ and z' denote the ith
projection functions on R2 and R3, respectively.4 Let £ be a connec
ted, open or closed region of R2 and /:£ -» R3 a differentiable function
such that the matrix [d(z‘ • f)ld y’] (i = 1 , 2, 3; / = 1, 2) everywhere has
rank 2. If x is any Cartesian mapping, x-1 • / = <I> maps £ into space. If
<t> is injective and, for any Cartesian mapping x, x • 4> everywhere
possesses partial derivatives of all orders, <!>(£) is what we would
normally call a smooth surface. Although not every smooth surface
can be wholly represented as the range of some injective mapping of
this kind, any point on such a surface has a neighbourhood which is
thus representable. The full surface can then be pieced together from
72 CHAPTER 2
To each part of a curved surface enclosed within definite limits we assign a total or
integral curvature, which is represented by the area of the figure on the sphere
corresponding to it.6 From this integral curvature must be distinguished the somewhat
more specific curvature which we shall call the measure of curvature. The latter refers
to a point of the surface, and shall denote the quotient obtained when the integral
curvature of the surface element about a point is divided by the area of the element
itself; and hence it denotes the ratio of the infinitely small areas which correspond to
one another on the curved surface and on the sphere.7
the area of figure ABHF cease being equal to the sum of the areas of
ACED and BCIJ? Certainly not, provided the paper has not been
stretched or shrunk. Moreover, every step in Euclid’s proof retains its
validity when referred to the rolled up figure, e.g. that AB = AF and
CA = AD and even, in a sense which may initially elude us, that angle
BAD is equal to angle CAF, and that therefore the area of AKGF is
equal to that of ACED. If the reader is not persuaded let him check
the proof by the method of coordinates. He may choose as axes the
top and side edges of the page. On the dunce cap both will become
curved lines, as will the perpendiculars drawn from them to any point
of the figure. But the latter will preserve their lengths and they will
continue to meet the axes orthogonally at the same points. The
Cartesian 2-mapping now maps the surface of the dunce cap into R2.
But the value assigned to each point is the same as it was, so that all
the equations used in the proof remain true. We can check by this
method any proof on the wallpaper rolls and obtain similar results.
We now see that the G-curvature function, in assigning the same
constant value 0 to the points of the plane and those of a rolled
surface, does not behave in an arbitrary, geometrically irrelevant
fashion. On the contrary, these two kinds of surface, like all surfaces
of zero G-curvature, are so closely related, that, when viewed ‘in
trinsically’ as two-dimensional expanses, apart from their relations to
the space outside, they must be regarded as geometrically equivalent
(at least locally, i.e. on a neighbourhood of each point). Curiously
78 CHAPTER 2
(a, b) = j dt
(x • c) dl
-J dt
(x • 3> • c) dt, (1 )
(2)
we obtain
s(o, ( . ) - j d s = j | ( e ( ^ ) + 2 F ^ 3 7 + g (^ 7) ) |- (5)
a a
he says, can be satisfied in several ways. The first that comes to mind
consists in supposing that the length of a line is independent of the
way how the line lies in the manifold ( Unabhangigkeit von der Lage),
so that every line can be measured by every other line. If this obtains,
the length of an arc in a manifold will be determinable as an intrinsic
property, i.e. as a property belonging to the arc as a one-dimensional
submanifold, no matter what its relation to the points outside it.31
The length of an arc in Euclidean space was traditionally conceived
as the limit of a sequence of lengths of polygonal lines inscribed in
the arc. This conception is extended quite naturally to R", where
‘straight’ segments are easily discerned.32 The length of the straight
segment joining a = (at, . . . , an) to b = (bu . . . , b„) is defined, by an
immediate generalization of the theorem of Pythagoras, as |a - b \ =
|(2"=i (a, —bi)2)m\- This method of definition is not intrinsic in the
above sense, and is not generally applicable to arcs in an arbitrary
n-dimensional manifold, since one cannot know beforehand whether
anything like a polygonal line will even exist in such a manifold. The
traditional definition of arc length can be given however a different
reading in the light of the concept of a tangent space developed in the
foregoing section. As we saw on page 69, the length of a path c was
given classically by an integral which we may note, for brevity, as
S f • c(t)dt. The ‘element of length’, / • c(t), was interpreted as the
length of an arbitrarily short straight line tangent to the path at c(t).
This notion is somewhat mysterious, for the length of an arbitrarily
short line is not a definite number at all (unless we simply equate it to
zero). The obscurity is avoided, however, if we conceive / as a
function which assigns to each point c(t) the ‘length’ of the tangent
vector cc(() defined in eqn. (2) of Section 2.2.7. This will make sense
only if that vector has been given a ‘length’. This is usually done by
defining a ‘norm’ on the vector space to which it belongs.33 Since the
concepts of tangent vector and tangent space are intrinsic, the rein
terpreted definition of arc length satisfies Riemann’s requirement and
can be extended to an arbitrary manifold. We shall now see how this
is done.
Let M be an n-dimensional differentiable manifold. To each point
PC M there is attached an n-dimensional vector space, the tangent
space Tp(M). Consider any smooth arc in M.34 We may regard it as the
range of an injective path c. At a point P € c(t), a definite element of
TP(M) is associated with path c, namely cP. If, in every tangent space
92 CHAPTER 2
( 1)
a
This definition is indeed intrinsic, for ccW spans the tangent space
Tc«)C((a, h)).35 Since we are dealing with an arbitrary manifold M, the
norm in its tangent spaces must be conceived quite broadly. It need
not even be defined in all of them in the same way. It is required only
that the norm of TP(M) does not change abruptly as P ranges over M.
This demand leaves enormous latitude of choice. Riemann imposes
two further restrictions. The first is that the norm in each tangent
space must be a positive homogeneous function of the first degree;
i.e. that for any vector v and any real number a, ||at>|| = |a| ||«||. This
requirement agrees well with our intuitive idea of length and is
ordinarily included in the general definition of a norm on vector
spaces. It ensures that the length of the arc c([a, &]) will not depend
on the choice of its parametrical representation c. Riemann’s second
restriction sounds less natural. It amounts to demanding that the
manifold M be what we now call a Riemannian manifold. Riemann is
well aware that this is not really necessary for a reasonable solution
of his problem. He observes, however, that the discussion of a more
general case would involve no essentially different principles, but
would be rather time-consuming and throw comparatively little new
light on the study of space. (Riemann, H, p. 14).
Let us say what we mean by a Riemannian manifold. A given
vector space V determines a vector space £T2(V) of bilinear functions
on V x V. Let ^ (M ) denote the union of the spaces £T2(TP(M))
determined by each tangent space TP(M) of a manifold M. 3"2(M) is
endowed with a standard differentiable structure (compare p.366). It
therefore makes sense to speak of a differentiable mapping of M into
^ 2(M). A Riemannian manifold or R-manifold is a pair (M, /i) where
M is a differentiable manifold and n is a differentiable mapping of M
into £T2(M) which assigns to each P € M a bilinear function /aP:
TP(M) x T P(M)-»R, and fulfils the following three conditions:
N O N - E U C L I D E A N GEOMETRIES 93
(1 i, j, k, h*£n ) (5)
the above manner. We shall call the integrand of (1) the line element
of the manifold. The quadratic form taken by the line element on the
manifolds given his name is used by Riemann to characterize them.
He is well aware that they are just a special kind of manifold, what he
calls the “simplest cases”. The “next simple case”, he says, would
consist of manifolds whose line element can be expressed as the
fourth root of an expression of fourth degree. “Investigation of this
more general class” , he adds, “would indeed involve essentially the
same principles, but would be rather time consuming and would
throw comparatively little new light on the study of space.”44 That is
why he restricts his research to what we call /{-manifolds. He
observes that the chart-related expression of the line element depends
on n{n + 1)/2 arbitrary functions whereas coordinate trans
formations are given by n equations. There remain therefore n(n -
l)/2 functional relations which do not depend on the choice of chart
but must be characteristic of the manifold. They should suffice to
determine metrical relations on an n-dimensional /{-manifold M. In
his lecture, Riemann approaches this question locally, showing how to
find n(n - l)/2 quantities at an arbitrary point P € M which, according
to him, determine metrical relations in a neighbourhood of P. But
before setting out to show this, he makes an important philosophical
point. The line element on M takes at each point P € M the Euclidean
form (2"=i (djc'(P)0dx‘(P))(cP, cP))U2, for a suitable chart x defined on
all M, if and only if the functions determined by x satisfy
gii = 8 ij (1 (10)
This has an important implication that fully justifies the choice of the
peculiar chart. Consider the Taylor expansion of the gy about P:
x kx h + o(|x|2), (13)
We found that n(n - l)/2 functions of position were necessary for determining the
metric relations of an [n-dimensional R-manifold]; hence, if the [G-curvature] is given
in n(n - l)/2 surface directions at each point, the metric relations of the manifold can
be determined, provided only that there are no identities among these values, and
indeed this does not, in general, occur. The metric relations of these manifolds, in
which the line element can be represented as the square root of a differential
expression of the second degree, can thus be expressed in a way completely in
dependent of the choice of coordinates.45
vectors in Tp(M):
into R" (with its standard metric; see p.95). M is what Riemann calls
a flat manifold. The second kind of manifolds considered by Riemann
includes flat manifolds as a subclass. He calls them manifolds of
constant curvature. If M is such a manifold k (a ) equals the same real
number for every two-dimensional subspace a C TP(M) at every point
P € M. Schur (1886b) subsequently proved that M is a manifold of
constant curvature if the preceding condition is fulfilled at any point
PCM . In Part III. 1, Riemann observes that only if space is a
manifold of constant curvature one may maintain that “the existence
of bodies” , and not just that of widthless lines, does not depend on
how they lie in space. In other words, only if space has a constant
curvature does it make sense to speak of rigid bodies. We shall see in
Sections 3.1.1-3.1.3 that Helmholtz regarded the existence of rigid
bodies as a conditio sine qua non for the measurement of distance in
physical space. If he is right, then physical geometry must rest on the
assumption that space is a manifold of constant curvature. This
restricts the spectrum of viable physical geometries considerably.
Helmholtz’ ‘problem of space’ consists of determining that spec
trum, under his just-mentioned assumption, by purely mathematical
means. In order to solve this problem one must give a clear mathe
matical formulation to the idea that the existence of bodies is in
dependent of how they lie in space. This can reasonably be under
stood to mean that any geometrical body placed in an arbitrary
position can be copied isometrically about any point and in any
direction. Now, one can only speak of isometrical copying with
regard to a manifold in which metrical relations are determined. If, as
Riemann contends, the latter depend wholly on the values of k(see
however, Note 50), the required copies can certainly be made in a
manifold of constant curvature, for metrical relations in such a
manifold are “exactly the same in all the directions around any one
point, as in the directions around any other, and thus the same
constructions can be effected starting from either”.51 On the other
hand, if P is a point of a manifold M and a and o ' are two-dimen
sional subspaces of the tangent space TP(M) such that k(a) ^ k(a'),
let /3 and /3' be the neighbourhoods of 0 in o and o ', respec
tively, which are diffeomorphically mapped into M by ExpP. ExpP(/3)
is then a two-dimensional submanifold of M whose tangent space at P
is o. But it is impossible to construct an isometrical copy of ExpP(/3)
with tangent space o ' at P. This can be seen as follows: ExpP(/3) is
N O N - E U C L I D E A N GEOMETRIES 101
ds = (22)
K
1 + 4
^ 2 x ‘x ‘
Riemann illustrates these ideas in a brief discussion of surfaces of
constant curvature. If the curvature is K > 0, the surface can be
mapped isometrically into a sphere of radius 1/VK. If K = 0, it can be
mapped isometrically into a Euclidean plane.
In his cursory reference to surfaces of constant curvature K < 0 ,
Riemann does not mention the fact that they can be mapped
isometrically into a BL plane, but I am convinced that he was aware
of it. After all, BL-space geometry was at that time the only known
example of a three-dimensional manifold with a non-Euclidean
metric, and it is more than likely that concern with its viability and
significance - which surely was not lacking in Gauss’ entourage-
prompted Riemann’s own revolutionary approach to the question.
His entire exposition is designed to bring out the fact that Euclidean
manifolds, i.e. manifolds of constant zero curvature, constitute only a
very peculiar species of a vast genus.52
*Riemann extended the Gaussian concept of curvature to an arbi
trary n-dimensional R-manifold M by using what we may call
sectional curvatures, i.e. the G-curvatures of two-dimensional sub
manifolds of M. The value of these sectional curvatures at a point
P € M is given by the function F defined on TP(M) x TP(M), (20). We
would possess a general conception of the curvature of M if we could
determine, once and for all, the F function attached to each point of
M. This job is performed by the celebrated Riemann tensor (in its
covariant form). Riemann himself went a long way towards its
definition in his prize-essay of 1861.53 His work was completed by
102 CHAPTER 2
the curvature of space lies within the interval (-e , e), for some real
number e > 0. This conclusion, unstated by Riemann but clearly
implied by his remarks, has considerable importance, for the
geometry of a manifold is non-Euclidean - either spherical or B L -
once its constant curvature deviates ever so slightly from zero. If
hypotheses concerning space curvature can only assign it intervals,
not fixed values, even the supposition that space curvature must be
constant appears to be ruled out. If there is no empirical means of
telling which value, within a given interval, space curvature does, in
fact, take on, the latter may just as well vary gradually within that
interval from place to place or from time to time. Towards the end of
his lecture, Riemann advances an even bolder conjecture, namely,
that space curvature may vary quite wildly within very small dis
tances, provided the total curvature over intervals of a suitable size is
approximately zero. The celebrated hypothesis on the “space-theory
of matter” put forward by W.K. Clifford (1845-1879) in 1870 is little
more than a restatement of this conjecture of Riemann’s. Clifford
wrote:
I hold in fact
(1) That small portions of space are in fact of a nature analogous to little hills on a
surface which is on the average flat; namely, that the ordinary laws of geometry are not
valid in them.
(2) That this property of being curved or distorted is continually being passed on
from one portion o f space to another after the manner of a wave.
(3) That this variation of the curvature of space is what really happens in the
phenomenon which we call the motion of matter, whether ponderable or etherial.
(4) That in the physical world nothing else takes place but this variation, subject
(possibly ) to the law of continuity.53
2.3.1 Introduction
The development of non-Euclidean geometry in Central and Eastern
Europe was half-hidden from the public owing to the obscurity of two
of its creators and the shyness of the third. In almost the same period,
the work of Jean-Victor Poncelet (1788-1867), who, in the limelight of
Paris, was laying the foundations of projective geometry, received
more attention. Partly because of its simplicity and beauty, and partly,
no doubt, because of its deceptive appearance of Euclidean
orthodoxy, the new discipline was in a short time well-known, ac
cepted and taught in the universities, often under the alluring name of
‘modern geometry’. Since there was no provocative negation
expressed in its name and since its radicalism was hidden beneath
seductive appeals to intuition, no philosopher ever raised his voice
against it. Yet, at bottom, projective geometry is much more ‘un
natural’ than, say, BL geometry, which only negates a Euclidean
postulate whose intuitive evidence had been questioned for centuries,
while, in projective geometry, the basic relations of linear order and
neighbourhood between the points of space are upset. Projective
geometry ignores distances and sizes, and thus may be regarded as
essentially non-metric.1 Nevertheless, in 1871, Felix Klein (1849-
1925), following the lead of Arthur Cayley (1821-1895), showed how
to define metric relations in projective space. Making conventional
and, from the projective point of view, seemingly inessential varia
tions in the definition of those relations, one obtained a metric
geometry satisfying the requirements of Euclid, or one satisfying
those of Bolyai and Lobachevsky, or, finally, a geometry where
triangles had an excess, as in spherical geometry, but where straight
lines would not meet at more than one point. In order to make these
results understandable to readers with no previous knowledge of
projective geometry, we shall present the main ideas of this geometry
in a more or less intuitive fashion in Section 2.3.2 which follows. A
rigorous analytical presentation of them will be given in Section 2.3.3.
Although an axiomatic characterization of projective space would
provide the best approach to such a thoroughly unintuitive entity, we
shall not give one because neither Klein nor his predecessors judged
it necessary or even useful and we wish to look at the subject as
much as possible from their point of view.
N O N - E U C L I D E A N GEOMETRIES 111
Fig. 10.
112 CHAPTER 2
Fig. 12.
plane. This does not prove, but somehow makes plausible that the
latter is also a one-sided surface.
If every plane in Euclidean space has been turned into a projective
plane, we can naturally regard the set of these planes as a new kind of
space, namely projective space. This space includes an ideal plane,
formed by all the ideal points that have been added to every ordinary
plane.
The ideal plane is determined by the pair of ideal lines where it
meets any two non-parallel ordinary planes. We cannot establish
neighbourhood relations in projective space by appealing to some
intuitively representable topological structure outside it (such as the
pencil cr(O) we used in the case of the projective plane), because
every intuitive spatial configuration is comprised in it. We might,
however, attempt to define its neighbourhood structure from within.6
But we shall go no further in the consideration of projective space
until we have made the notion of a projective plane clearer and less
problematic.
2.3.3 Projective Geometry: A Numerical Interpretation
We have introduced projective space insidiously, by a series of
natural, apparently intuitive steps. The result arrived at is, however,
ostensibly counterintuitive and we cannot be sure that it is truly
viable. A contemporary mathematician would dispel our doubts by
producing an axiom system that unambiguously determines the struc
ture we expect projective space to have and then proving its consis
tency. But before the publication of Pasch’s Lectures on Modem
Geometry (1882), even the most distinguished mathematicians had a
116 CHAPTER 2
i ) “iXi = 0 (7)
1=1
is a line in &1. This line can naturally be denoted by the set (u,) € R3.
Since the solutions of (7) are also solutions of
2
1=1k u >x ‘ = 0 (8 )
for any real number 0, the line in question can be denoted by any
member of the equivalence class [m,]. A line («,-) is incident on (or
passes through) a point (jc,) - which is then said to be incident on or to
lie on (H.-)-if and only if 2 /=1 M;*, = 0. Two or more points are
collinear if they all lie on one line; this line is their join. Two or more
lines are concurrent if they all pass through one point; this point is
their meet. It is merely a matter of algebra to prove that any two
points in 9 2 have one and only one join and that any two lines in 3‘2
have one and only one meet. If plane projective geometry concerns
the properties and relations of the points and lines of /3>1, the proofs
of its theorems will consist of equations where different points and
lines are denoted by different elements of R3. Now, if we choose, say,
the symbols (p{), (q,), (r,),. . . to denote points, while ( m,), (»,•).
N O N - EU C LI DE A N GEOMETRIES 119
(wi) , . . . denote lines, the equations in which these symbols occur will
still hold if we let (p,), (q,), (r,),. . . denote lines, while ( m,), ( i;,),
(w,),. . . denote points. Consequently, any true statement of plane
projective geometry gives rise to a ‘dual’, that is, another true
statement obtained from the former by substituting point for line,
collinear for concurrent, meet for join, and vice versa, wherever these
words occur in the former statement. This is called the principle of
duality. In our numerical interpretation of projective geometry the
principle is trivial. But Gergonne (1771-1859), who formulated it as a
general principle in 1825, did not have this interpretation at his
disposal. He discovered duality by noticing pairs of complementary
or ‘dual’ theorems, proved under the usual intuitive (or pseudo-
intuitive) conception of the projective plane.
Our numerical interpretation of the projective plane shows that
projective geometry is at least as consistent as the theory of real
numbers. Since every theorem can be stated as a relation between
number triples and every proof can be carried out through a sequence
of ordinary algebraic calculations, any contradiction arising in pro
jective geometry would show up as a contradiction in elementary
algebra. Though this result should remove the doubts expressed at the
beginning of this section, we shall now give a fully intuitive represen
tation of the projective plane for the benefit of readers who stand in
awe of numbers. Let P be a point in Euclidean space and let g 3
denote <£3-{P}. We define an equivalence F on as follows: xFy if
and only if x and y lie on the same line through P. Consider the
quotient set g 3/F. The ‘points’ of g 3/F are the lines through P. Two
lines through P define a plane through P, which we shall call their
join. Two planes through P determine a line through P, which we shall
call their meet. With these stipulations the pencil of lines and the
bundle of planes through P furnish an adequate representation of the
projective plane. They can, in fact, be identified with 0>2, our numeri
cal representation. If x is a Cartesian mapping with its origin at P, we
can represent a line m through P by the x-coordinates of any one of
the points of f? that He on m. In this way, we assign a full
equivalence class of homogeneous coordinates in R3 to each meet of
£ 3/F, thereby mapping R3/E onto &3/F. We do likewise with the joins
of &3/F, i.e. the planes through P. Call this mapping/. It is not difficult
to see that if A is the join or meet of B and C in R3/E, /(A ) is the join
or meet of /(B) and /(C) in <t/F. The existence of / shows that R3/E
120 CHAPTER 2
and % IF are isomorphic, i.e. that they both possess the same pro
jective structure. Our intuitive representation of the projective plane
makes an important result immediately obvious. All planes through P
have the same status. We cannot select one among them to play the
rdle of the ideal line, except by an arbitrary stipulation. Consequently,
from a purely projective point of view there is no essential difference
between the ideal line and every other line. This is not so clear in the
numerical representation because the ideal line has a seemingly
peculiar equation (namely p 3= 0). But it could be inferred from the
principle of duality: there is no such thing as a privileged line in 9>2,
formed by a distinguished class of ‘ideal’ points, because there is no
such thing as a privileged pencil, formed by a distinguished class of
lines.
The numerical representation of the projective plane suggests a
generalization which is assumed by Klein in his work on Non-
Euclidean geometry. Let C denote the field of complex numbers. If
C3= C3- (0, 0, 0) and if E denotes the relation of linear dependence
between two elements of C3, we denote the quotient set C3/E by &c-
We call this the complex projective plane.9 Points and lines in &c are
defined in the same terms as in 9>2. d*2 may be regarded as a proper
subset of formed by the equivalence classes [pi, p 2, Pi] one of
whose representative triples consists exclusively of complex numbers
whose imaginary part is zero. We call these points the real points of
9>\. By eliminating from 9 2 the line «3= 0 we obtain the so-called
affine plane, which is simply the Euclidean plane regarded as a proper
subset of the projective plane (and deprived, as such, of the Eucli
dean metric structure). We shall denote the affine plane by %2. For the
sake of completeness, we may mention that if R"+1 = R"+1 - {0}, Cn+1 =
C"+1 - {0} and E denotes linear dependence in one or the other of these
sets, 9 n = Rn+I/E and &c = C"+l/E are called, respectively, the real and
the complex n-dimensional projective space.
2.3.4 Projective Transformations
We shall consider two kinds of mappings defined on ‘3>1. A collinea-
tion is a continuous injective mapping of < 3>1 onto itself that matches
points with points and lines with lines, preserving incidence relations
between lines and points. Let (p,), (q,) denote points while («,), (t>,)
denote lines. The general analytic expression of the collineation
(p.)'-Hq.), («;>-»•(»;) is
N O N - E U C L I D E A N GEOMETRIES 121
3
kQ i =2 a ‘i P i ’
i31 (|ay| ^ 0; / = 1,2,3; 0). (1)
kUi = 2 aiivb
i=i
This mapping preserves incidence between lines and points since
3 j 3 3
Let Ay denote the cofactor of ay in the matrix [ay]. The inverse of (1) is
then given by
3
k 'V i = 2 A y My,
i 1 (i = 1,2,3). (3)
k 'P i =/=!
2 Ayq,-,
A correlation is a continuous injective mapping which assigns a
point to each line and a line to each point in t?2, so that collinear
points are mapped on concurrent lines, and vice versa. We obtain
the general expression of the correlation (p,-)-K!>/), (!<,>-*(<&) by
simply interchanging (t>,) and (q,) in (1):3
3
kvi = 2 a iiP b
given by
3
kvi = 2 atiP j, ( |a l7| ^ 0 , a i; = «,„ k ^ 0 , i = 1 , 2 , 3 ) . (5 )
/=■
If (pi) and (qO are conjugate points, (q,) lies on (u,); hence
3 3
2i = vtfi = 2 a n Q i P i = 0 - (6)
l ij = 1
2 a u P iP i = 0. (7)
If (7) has solutions, (8) has solutions as well. They are precisely the
tangents to the conic defined by (7). The pole of each tangent m is its
point of tangency. A conic may be regarded as a set of points, or,
dually, as a set of lines, namely, the tangents that envelop it. If we
regard it both ways, (7) and (8) represent the same conic which may
be said to be its own image under polarity (5). This property of being
a locus of self-conjugate points and lines under a fixed (hyperbolic)
polarity is normally used to define conics in (real) plane projective
geometry. The definition does not depend on the numerical inter
pretation we have made the basis ofour discussion.9 If, in eqns. (7)
and (8), the matrix of the coefficients a/, happens to be singular
(|afj| = 0), of rank 2 or rank 1, those equations are said to determine a
N ON -E U C LI D EA N GEOMETRIES 123
Pi ¥■P 4 and P 2 P 3. Let (p,), (q,) denote two points of the line on
which the points P r lie. Each point Pr (r = 1 , 2, 3, 4) is then denoted
by (krpi + m^ji) for some pair of real numbers (kr, mr), not both zero.
The cross-ratio of Pi, P2, P 3, P 4 (in that order) is then defined by the
following equation:
fc, k2 fc4 •
m i m3 m 2 m4
(P ,,P 2;P 3, P4) = (3)
k2 *3 ♦ I k t k4
m 2 mj 1m. m4
If P 3 is (p,) and P4 is (q,), (fc3, m 3) = (1, 0) and (k4, m4) = (0, 1). In that
case
ntik2
(P,, P2; P3, P4) (4)
kim 2'
Since we are always free to make that assumption, it is clear that,
given three arbitrary collinear points A, B, C,
(A, A; B, C ) = l. (5)
The cross-ratio of four collinear points depends not on the choice of
the homogeneous coordinates that represent them, but only, as we
gather from eqn. (5), on the parameters which determine the relative
positions of two of the points with respect to the other two, on the
line to which all four belong. The cross-ratio of four concurrent lines
is defined analogously. It is a matter of mere calculation to show that
the cross-ratio is preserved by projective transformations, i.e. that, if
<p is a projective transformation, then, for every four collinear points
or concurrent lines Mi, M2, M 3 , M4,
mutandis, to ^c- Let £ be a conic in @c- Let K c denote the set of all
collineations which map £ onto itself. The join of two points P, Q
meets £ at two points which we shall denote by (PQ/£)i and (PQ/£)2.
If <p € Kf, each of these points is mapped on one of the points
(<p(P)<p(Q)/£)i = <p ((? Q IO i), O' = L 2). (1)
Since the cross-ratio is a projective invariant, it follows immediately
from eqns. (3) and (6 ) of Section 2.3.5 that
(P, Q; (PQl£)u (PQlCh)
= (9 (P), <p (Q ); («p (P)<p (Q)/^) i , (<p (P)<p (Q>/^)2). (2)
Let f ( denote the complex-valued function (P, Q>-HP, Q; (PQ/£)i>
(PQ/^)2), defined on SPcx ^ c ! 15 f t is preserved by every collineation
of K{, since (2) implies that, for every <p € Kf
fc(P, Q) = /{(<P(P),<P(Q))- (3)
We now define a function dt on point-pairs of the complex projective
plane:
d((P, Q) = c • lo g /{(P, Q) (4)
(where c is an arbitrary non-zero constant and log x denotes the
principal value of the natural logarithm of x). The function d( has some
properties that make it a good choice for a (signed) distance function
on 9 \ . In the first place, d((P, P) = 0 for every point P not on £. (See
eqn. (5) of Section 2.3.5.) In the second place, if Pi, P2, P 3 are collinear
points not on £,'6
d{(P„ P2) + d;( P2, P3) = d( (P„ P3). (5)
In the third place, if P and Q are different points not on £,
d((P, Q) = - d((Q, P). (6 )
It is true that d( is undefined on a point-pair if one (or both) of its
members lies on £. But we can make sense of the statement that if Q
lies on £ then dc(P, Q) is infinite for every point P not on £. Suppose
that Q = [kpi + mqi] lies on £ and that (Q,) = ([fqp, + m,q,]) (/ = 1 , 2, 3;
j = 1 , 2 ...) is a sequence of points not on £ (but all on the same line
through Q) such that (|/c - kj\) and (|m - m^) are null sequences. Then,
if P is a point not on £ lying on QQi, |df(PQ,)| increases with j beyond
all bounds. Therefore, if we persist in regarding d{ as a distance
N ON -E U C LI D EA N GEOMETRIES 127
function we may say that the points on £ are infinitely distant from
the remaining points of SPc- Still, has a property that is rather
unusual for a distance function: d[ is complex-valued and, whatever
the value assigned to the arbitrary constant c, there will be, for every
choice of £, point-pairs (P, Q) such that d((P, Q) has a non-zero
imaginary part. One ought not to dispute about names and every
mathematician should feel free to call a complex-valued function like
d[ a ‘distance function’ on 0*c- But the ‘geometry’ thereby defined is
not what is known as a metric geometry in contemporary mathema
tics.17 However, as we shall see, d( when restricted to a well-chosen
region of does define a real-valued metric function. This is the
substance of Klein’s discovery.
The development leading to the definition of dl can be dualized by
substituting any pair of lines m, n for the points P, Q. Then (mn/£)i
and (mnl£)2 will denote, of course, the two tangents to the conic £
that pass through the meet of m and n. As a function on line-pairs, d{
seems to be a good choice for an angle-function, i.e. a function whose
value measures the size of the angle formed by its two arguments.
The choice is strongly recommended on account of the following
result due to Laguerre.18 Let m, n be two lines on the affine plane
g 2C 0>c, which meet at a point P in %2. We shall denote by m and n
the extension of m and n to &c, i.e. the sets of points in &c that
satisfy the equations characteristic of m and n. There are two lines r,
r' that join P to the two circular points of 2Pc (r, r' have each only one
real point, namely P). As Laguerre showed, the ordinary Euclidean
value of the angle made at P by m and n is equal to 1/2/ times the
natural logarithm of the cross-ratio (m, n; r, r’). The circular points
can be regarded as a degenerate line conic.19 If we let £ denote this
conic and if we take c = 1/2/ our function d( as defined on line-pairs
measures the size of Euclidean angles. This interpretation of
Laguerre’s result was given by Cayley in 1859.20 By duality he
obtained a distance function defined on point-pairs of the affine
plane. Remarkably enough, this distance function is none other than
the ordinary Euclidean metric function. Cayley’s discovery linked
angle size to segment length - a welcome achievement at a time when
projective geometry was regarded as a natural extension of ordinary
Euclidean geometry (and not as something utterly different from it, as
we regard it here). Indeed angles are the duals of segments, so that
the measure of the latter should be the dual of the measure of the
128 CHAPTER 2
ordinary metric on the straight line [i.e. the familiar system for
measuring the length of segments] belongs to the second kind because
a displacement of the straight line along itself, under the assumptions
of ordinary parabolic geometry, leaves just one point unchanged,
namely, the infinitely distant point.”31 This difference between the two
fundamental metrical systems of geometry disappears in the elliptic
and the hyperbolic cases. This is as should be expected, if the latter
indeed are more general and consequently more natural.
2.3.7 Models
Klein’s work is often linked to the construction of so-called Euclidean
models of non-Euclidean geometry. Thus, Borsuk and Szmielew, in
their well-known Foundations of Geometry, describe a Beltrami-
Klein model of BL geometry.32 We shall presently see to what extent
such a characterization of Klein’s work is justified. Strictly speaking,
a model can be conceived only in relation to an abstract axiomatic
theory. If you are given a set of sentences S which contain undefined
terms 11, . . . , tn, you can look for a model of S, that is, a domain of
entities where, through an arbitrary but consistent interpretation of
terms 11, . . . , f„, the sentences of S come true. In this strict sense, we
cannot ascribe a model-building intention to Klein who, in 1871, did
not have the notion of an abstract axiom system. But we also speak
of models in a looser sense whenever a structured collection of
objects is seen to satisfy a set of mathematical statements, given a
suitable, though usually unfamiliar, reading of its key words. We thus
say that the pencil of straight lines through a point P in space
provides a model of the projective plane if we accept the following
semantic equivalences: ‘a point’ = a line through P; ‘a line’ = a
plane through P; ‘point Q is the meet of lines m and m n = line Q is
the intersection of planes m and m'; ‘line m is the join of points Q
and Q'’ = plane m is spanned by lines Q and Q' (p.119). In this looser
sense, Klein’s theory does indeed supply models for Euclidean and
non-Euclidean geometry, but his models are projective and therefore
not Euclidean (because, as we have repeatedly observed, projective
space is not Euclidean space, Postulate 5 is false in projective
geometry, etc.). Thus parabolic plane geometry on the affine plane
^ 2C ^ c provides a somewhat peculiar model of ordinary Euclidean
plane geometry: points are points and straight lines are straight lines,
but distances between pairs of points and angles between pairs of
N O N - E U C L I D E A N GEOMETRIES 133
lines are defined with respect to a fixed entity located outside the
affine plane itself. On the other hand, hyperbolic plane geometry on
the interior of an ellipse may be viewed as a Euclidean model of
BL plane geometry if we no longer consider its domain of definition
to be a subset of &c and regard it as a region of the Euclidean plane.
Thus, we may define hyperbolic geometry in the interior of a circle
(O, r) with centre O and radius r. A BL point is any ordinary point
inside this circle; a BL line is any chord (not including the points
where it meets the circumference of the circle). Let P be a BL point
and m a BL line not through P meeting the circumference of (O, r) at
A and B. There are two parallels to m through P, namely the two
chords that join P to A and to B (Fig. 13). These parallels divide the
chords through P into two groups: those that meet m and those that
do not meet m (scil. those that do not meet the chord m in the interior
of (O, r).) In order to complete the model we must introduce pro
jective concepts. If Q and R are two points on m (inside (O, r)), the
(undirected) distance between Q and R is taken to be equal to 2|log(Q,
R; A, B)|.33 This value is preserved by all linear transformations (of
the entire projective plane) that map circle (O, r) onto itself. The
restrictions of these transformations to the interior of (O, r) play the
role of BL isometries (motions and reflections). This is, in essence,
the “Beltrami-Klein” model given by Borsuk and Szmielew. I leave it
to the reader to decide whether it is a genuine Euclidean model.
This model was found by Eugenio Beltrami (1835-1900) some time
before the publication of Klein’s paper. In his “Saggio di inter-
pretazione della geometria non euclidea” (1868), Beltrami sets out to
find a Euclidean realization of BL plane geometry and discovers it in
a surface of negative curvature. The flat model we have just
described is only used as an aid in Beltrami’s investigations. Beltrami
is aware that the new geometrical conceptions are bound to bring
about deep changes throughout classical geometry. But he is persu
aded that the introduction of new concepts in mathematics cannot
Fig. 13.
134 CHAPTER 2
upset acquired truths; it can only modify their place in the system or
their logical foundations and thereby increase or decrease their value
and utility. With this understanding, Beltrami has tried to justify to
himself (“dar ragione a noi stessi”) the results of Lobachevsky’s
theory. Following a method he believes to be “in agreement with the
best traditions of scientific research” , he has attempted “to find a real
substrate for this theory before admitting the need for a new order of
entities and concepts to support it”.34 To Beltrami’s mind, a “real
substrate” is perforce a Euclidean model. He thinks he has succeeded
in his attempt as far as BL plane geometry is concerned, but he believes
it impossible to do likewise in the case of BL space geometry.35
Beltrami reasons thus: A “real substrate” for the BL plane must be
found in a curved surface in Euclidean space, since a Euclidean plane
can provide a model only of itself, unless we tamper with the ordinary
meaning of distance, and this he seems unwilling to do. It must be a
surface of constant G-curvature, for only on such surfaces can we
apply the “fundamental criterion of proof of elementary geometry” ,
namely, the superposability of congruent figures (la sovrapponibilita
delle figure eguali). The most essential ingredient of a geometric
construction is the straight line. Its analogue on a surface of constant
curvature is the geodetic arc. The analogy breaks down on surfaces of
constant positive curvature, for there exist on them point-pairs which
do not determine a unique geodetic arc. How about surfaces of
negative curvature, or “pseudospheres” as Beltrami calls them? To
prove that every pair of points on a pseudosphere is joined by one
and only one geodetic arc, Beltrami sets up a special chart with
coordinate functions u, v. Relative to this chart, the element of length
on a pseudosphere with constant curvature equal to -1 /R 2 is given by
2 _ „ 2(a 2~ v 2) du2+ 2uv d u dt; + (a2- u 2)dv
ds = R (1)
(a2- u 2- v2)2
The main advantage of this chart is that every linear equation in u and
v represents a geodetic line and every geodetic line is represented by
a linear equation in u and v. In particular, the lines u = constant and
v = constant are geodetic. The angle 0 formed by the lines u =
constant and v = constant at (u, v) is given by
uv
cos 0 -
((a 2- u 2)(a 2- v 2))m,
(2)
a (a2- u2- v2)'12
sin 0 =
( ( f l ' - l t W - B 4))
N ON -E U C LI D EA N GEOMETRIES 135
H2+t>2=£a2. (3)
Fig. 14.
136 CHAPTER 2
If n variables x,, x2, . . . , x „ are given, the infinity to the nth value-systems we obtain if
we let the variables x independently take the real values from - » to +<*, constitute
what we shall call, in agreement with usual terminology, a manifold of n dimensions.
Each particular value-system ( x i , . . . ,x„) is called an element of the manifold.'"
It is not clear whether “the real values from —oo to +«=” are just the
values between these two extremes, i.e. all the real numbers, or
include —<* and +=c. If they exclude the latter, an n-dimensional
manifold in Klein’s sense is simply R". Now R" is not the same as an
n-dimensional manifold in Riemann’s sense (pp.86ff.), but if we
endow it with the usual differentiable structure, it is diffeomorphic to
any ‘coordinate patch’ (the domain of a chart) of such a manifold. On
the other hand, if Klein's variables may take the values —<* and +°°,
an n-dimensional manifold in his sense is a very peculiar entity whose
topology would require some further specification. In the light of
Klein’s usage in the paper we are discussing, I conclude that the truth
lies somewhere between the two alternatives: a manifold composed
of “real”-valued n-tuples turns out to be identical with 5s". As we
know, this is not homeomorphic to Rn, let alone to any arbitrary
manifold in Riemann’s sense. But it is not the same as the set
{(Xi,. . . , x„)|-°c =s=X; *£ + oo; l ss / ss n). Klein adds that in the course of
his arguments he will let the variables X \ , . . . , x n take arbitrary
complex values as well. This implies, in my opinion, that “an n-
dimensional manifold” in Klein’s paper (1873) is but another name for
the complex n-dimensional projective space This may readily be
conceived as an n-dimensional complex differentiable manifold (i.e.
one with complex-valued charts).42 But it is not diffeomorphic to
every complex n-dimensional manifold. Nor does it play a special role
among them, like that of Cn (to which every complex manifold is
diffeomorphic locally).
I have dwelt at length on such scholastic niceties, as a prelude to
the following remark. Klein will show that Riemann’s geometries of
constant curvature can be regarded as the theories of certain struc
tures defined on (or in?) However, for Riemann, those geometries
are but peculiar members of the vast family of Riemannian
geometries, dealing with R-manifolds of arbitrary curvature. Within
this family the geometries of constant curvature are, so to speak,
degenerate cases. Hence by confining his discussion to structures
definable on a particular n-dimensional manifold, Klein loses sight of
the full scope of Riemann’s conception. Geometries of constant
N O N - EU C LI DE A N GEOMETRIES 139
curvature are taken from the context in which they were originally
defined, and granted a privileged status.
However this does not mean that Klein deals with them in isolation.
They have a well-defined position in a different system which, al
though the ordinary Riemannian geometries are excluded from it, can
be extended to cover many new geometries. After his description of
n-dimensional manifolds, Klein sketches the main ideas of this
system. A more detailed exposition is given in the “Programme” he
submitted to the Faculty at Erlangen at the time of joining it.43 The
driving force behind it appears to be his desire to find a unifying
concept by means of which to comprehend and organize the wealth of
disparate discoveries in 19th-century geometry. He found it in the
concept of a group of transformations.44 We may characterize it as
follows. For any set S, a bijective mapping /: S-»S is called a
t r a n s f o r m a t i o n of S (into itself). Let T be the set of all transformations
The choice between these forms, Klein adds, should be guided by the
principle of economy of thought. In his lectures on non-Euclidean
geometry, published posthumously thirty years later, he remarks:
Let us assume that the space about us exhibits a Euclidean or a hyperbolic structure.
We can by no means infer from this that space has an infinite extent. Because, for
instance, Euclidean geometry is entirely compatible with the hypothesis that space is
finite, a fact that has been formerly overlooked. The possibility of ascribing a finite
content to space whatever its geometrical structure, is particularly welcom e because
the idea of an infinite expanse, which was originally looked upon as a substantial
progress of the human mind, gives rise to many difficulties, e.g. in connexion with the
problem of mass-distribution.67
We see thus that towards the end of his life, probably after studying
Einstein’s writings on gravitation and cosmology, Klein came to think
that cosmological considerations can furnish empirical criteria for
choosing between globally non-homeomorphic though locally
isometric space-forms.
CHAPTER 3
F O U N D A T IO N S
full impact of the new geometries did not appear until the late 1860’s,
the most important being Riemann’s lecture of 1854 (published in
1867) and three papers of Hermann von Helmholtz (1866, 1868, 1870),
who was not a mathematician, nor a professional philosopher, but a
physician - and a great physicist - with a philosophical turn of mind.
These works may be regarded as the starting-point of a series of
investigations of an increasingly mathematical character that even
tually culminated in David Hilbert’s Foundations o f Geometry (1899).
Among the publications on the subject which appeared between
Helmholtz’s and Hilbert’s the most remarkable are perhaps two
papers by Sophus Lie, “On the foundations of geometry” (1890),
reworked and expanded in Part V of the third volume of his Theory
o f Transformation Groups (1893). Lie’s approach stems directly from
Helmholtz’s. His aim is to give an exact solution of the latter’s
“problem of space” , originally conceived as a typical epistemological
question which may be stated thus: Which among the infinitely many
geometries whose mathematical viability has been shown by Rie
mann’s theory o f manifolds are compatible with the general condi
tions o f possibility o f physical measurement ? Implicit in the question
is the operationist belief that only such geometries as are compatible
with the latter conditions are suitable for the role of a physical
geometry. We shall discuss this problem in Part 3.1.
We shall see that Lie, while correcting and improving Helmholtz’s
mathematical treatment of the problem of space, lost sight of its
epistemological significance and understood it as a problem in pure
mathematics, viz. What properties are necessary and sufficient to
define Euclidean geometry and its near relatives, the geometries o f
constant non-zero curvature? Lie’s answer to this modified question
anticipates Hilbert’s achievement, insofar as it gives an exact
axiomatic characterization of Euclidean geometry. But Hilbert’s ap
proach is very different from Lie’s. Like Riemann and Helmholtz, Lie
conceives of space as a differentiable manifold, whose local topology
depends on the familiar, tacitly assumed properties of the real number
continuum (its global topology, he simply ignores). Hilbert, on the
other hand, makes no prior assumptions concerning the relation
between geometry and analysis. As a matter of fact, the first version
of his axiom system does not even ensure that space can be endowed
with a differentiable structure.3 And he makes a point of showing how
unexpectedly far one can proceed in the construction of Euclidean
FOUNDATIONS 155
3.1 H E L M H O L T Z ’S P R O B L E M O F S P A C E
The fact that (approximately) rigid bodies exist can only be known,
therefore, if we possess the idea of a (perfectly) rigid body. Bearing
this in mind, Helmholtz acknowledges that “the notion of a rigid
geometrical figure may be conceived indeed as a transcendental
notion, which has been formed independently of actual experiences,
and which will not necessarily correspond with them”.29 He adds:
168 CHAPTER 3
Taking the notion of rigidity thus as a mere ideal, a strict Kantian might certainly look
upon the geometrical axioms as propositions given a priori by transcendental intuition,
which no experience could either confirm or refute, because it must first be decided by
them whether any natural bodies can be considered as rigid. But then we should have
to maintain that the axioms of geometry are not synthetic propositions in Kant’s sense,
because they would state only what follows analytically from the notion of a rigid
geometrical figure, as it is required for measurement. Only such figures as satisfy those
axioms could be acknowledged as rigid figures.30
We have no other mark of rigidity of bodies or figures but the congruences they
continue to show whenever they are applied to one another at any time or place and after
any rotation. We cannot however decide by pure geometry [ . . . ] whether the coinciding
bodies may not both have varied in the same sense. If we judged it useful for any
FO U N D A T I O N S 169
purpose we might with perfect consistency look upon the space in which we live as the
apparent space behind a convex mirror [. . . ]; or we might consider a bounded sphere of
our space, beyond the limits of which we perceive nothing further, as infinite pseudo-
spherical space. Only then we should have to ascribe to the bodies which appear to us to
be rigid, and to our own body, corresponding dilatations and contractions and we
should have to change our system of mechanical principles entirely.32
On the face of it, this passage says that the choice between Euclidean
and BL geometry is a matter of convenience, so that, at least within
this limited spectrum of alternatives, geometry is conventional. In
terms of Helmholtz’s original question, this conclusion can be stated
saying that Euclid’s fifth postulate is not a “truth of factual
significance” but a consequence of the chosen mode of expression.
This position was held later by Henri Poincare (1854-1912).33 Did
Helmholtz anticipate him? I would say yes, insofar as he did publish
the passage quoted above, which clearly suggests the conventionalist
thesis. But it is apparent that Helmholtz did not understand his words
quite in that sense. He points out that a change in the geometry would
impose a change in the laws of mechanics. And he is not willing to
grant that the latter are, up to a point, no less conventional than the
former. In his opinion, the reference to mechanics settles the ques
tion. This implies of course that a decision concerning the truth of
Postulate 5 cannot be reached by geometrical experiments alone. But
that was to be expected after Helmholtz’s earlier assertion that the
axioms of geometry do not belong to the pure theory of space. He
now adds: “Geometric axioms do not speak about spatial relations
only, but also at the same time about the mechanical behaviour of our
most rigid bodies in motion” .34 As a consequence of it, we must
conclude that geometry - that is, physical geometry - does not provide
a groundwork for mechanics but must be built jointly with it. It might
even seem that Helmholtz expects the more elementary mechanical
principles to provide a foundation for geometry. At any rate, he does
not ask how much in these principles has a factual import, and how
much is merely a matter of linguistic preference. One thing is clear:
pure physical geometry is indeterminate; “but if to the geometrical
axioms we add propositions relating to the mechanical properties of
natural bodies, were it only the principle of inertia, or the proposition
that the mechanical and physical properties of bodies are, under
otherwise identical circumstances, independent of place, such a
system of propositions has a real import which can be confirmed or
170 CHAPTER 3
refuted by experience, but just for the same reason can also be gained
by experience”.35 This is a very powerful argument against the
Kantian philosophy of geometry and is perhaps the main reason why
the latter could not survive the discovery of non-Euclidean
geometries: a priori knowledge of physical space, devoid of physical
contents, is unable to determine its metrical structure with the pre
cision required for physical applications.
In his reply to Land (1877), Helmholtz upholds an unmitigated
empiricism. He proposes a simple experiment in order to decide the
issue between the three geometries of constant curvature. I paraph
rase: As soon as we have a method to determine whether the
distances between two-pairs are equal (“i.e. physically equivalent”) we
can also determine if three points A, B, C are so placed that no other
point D t^B satisfies the equations d(D, A) = d(B, A) and d(D, C) =
d(B, C) (where d stands for distance). We say then that A, B, C lie in
a straight line. Let us choose three points P, Q, R not in a straight
line, such that d(P, Q) = d(Q, R) = d(R, P), and two further points A,
B, such that d(A, P) = d(B, P), and P, Q, A on the one hand, and P, R,
B on the other, lie in a straight line (Fig. 17). Then, if d(A, P) =
d(A, B), the Euclidean geometry is true; but BL geometry is true if
d{A, B) < d(P, A) whenever d(P, A) < d(P, Q) and spherical geometry
is true if d(A, B) > d(P, A) whenever d(P, A) < d(P, Q). (Helmholtz,
G, p. 70). Helmholtz is right indeed, if we have a method to determine
whether the distances between two point-pairs are equal, that is,
physically equivalent. The whole issue turns therefore about this
notion, which Helmholtz defines as follows: “Physisch gleichwertig
nenne ich Raumgrossen, in denen unter gleichen Bedingungen und in
gleichen Zeitabschnitten die gleichen physikalischen Vorgange
Fig. 17.
FO U N D A T I O N S 171
(where t(g) denotes the sextuple ( t‘(g ),. . . , t 6(g)), etc.). Knowledge
of the functions /, enables us therefore to compute the coordinates of
fig, P) for every g in M which is the product of elements of U. Lie
usually represents the action of a Lie group G on the space R„ by
means of analytic functions defined like the /, above. Let us call this
the standard representation of group action. The representation is
local. However, though it is originally defined only on a neighbour
hood U of e (. G, it can be extended, in the fashion we have
explained, to the full subgroup of G generated by U. This subgroup is
176 CHAPTER 3
( 1)
Fig. 18.
FO U N D A T I O N S 185
any point p € x ( tr); by (ii) there exists a closed Jordan curve whose
interior contains p and is contained in x(7r); consequently x(7r) is
open. Choose any two points p, q in x(rr)\ by (v) there exists a closed
Jordan curve whose interior contains p and q and is contained in
jc(tt); consequently x (i r) is connected. Therefore, x( 7r) is
homeomorphic to R2. Hence, tt is homeomorphic to R2. On the other
hand, if we assume this, conditions (i)-(v) will follow.
Let x map tt homeomorphically onto R2. We paraphrase Hilbert’s
definition of motion: a motion of is a bijective continuous mapping
tt
either designator are one and the same. The reader will note that ‘= ’
is not an interpretable word. Non-interpretable words in m-English
are sometimes called logical words. Non-basic sentences are of two
kinds: truth-functional sentences and existential sentences. A
sentence S is truth-functional if its truth-value (true or false) is
uni vocally determined, according to a fixed rule, by the truth-value of
one or more sentences Si,. . . , Sn, distinct from S, called its
components. A sentence S is existential if it can be obtained from
some other sentence S' by (i) substituting a suitable variable x, which
does not occur in S', for every occurrence of a given interpretable
word in S'; (ii) prefixing the phrase ‘there is an jc such that’ (which
we shall abbreviate ( J E ) ) . This phrase is called an existential
jc
was not sufficiently precise for mathematical use. This led to the now
current practice of axiomatizing set theory, whereby the permissible
interpretations of the set-theoretical predicates are characterized by
means of axioms in which they occur as undefined terms. The
axiomatic approach to set theory in its turn raises difficulties which
have, of late, become intolerable. But we cannot deal with them
here.12
Mathematicians do not claim that their theorems are true but that
they follow from their axioms. Some authors conclude from this that
every mathematical theory is hypothetical, as they say, since its truth
depends on the truth of its axioms, and the latter, they contend, are
not held to be true, but are put forward only as suppositions or
hypotheses. To judge the merits of this opinion one should bear in
mind the following remarks. Let S be a theorem of a theory with
axioms K. Mathematicians will state then that K f= S. This statement
is a good deal stronger than what we would ordinarily call a hypo
thetical statement. Kf=S does not say merely that S will come true if
a situation described by K, in some familiar acceptation of these
sentences, is fulfilled. Kf=S says that S is true in every interpretation
of KU{S} which satisfies K. This far-reaching claim is made by
mathematicians unconditionally, when they assert that Kf=S. On the
other hand, this claim would be trifling, if K is not true in any
interpretation. Consequently, though the mathematician who states
that S is a theorem which follows from K need not hold K to be true
in a particular interpretation, he ought to make sure, lest his statement
be pointless, that there is at least one modelling of K. This require
ment is fulfilled by the more important mathematical theories if (i) the
set of natural numbers exists (ii) the conditional existential postulates
of ordinary axiomatic set theory are true, in their familiar English
meaning.13 Neither of these assumptions can be said to be beyond
every reasonable doubt. They may be viewed as the hypotheses
which lie at the foundation of mathematics. Yet it is not the truth of
mathematical theories, but rather their significance, that may be said
to rest on this hypothetical basis.
3.2.3 Stewart, Grassmann, Pliicker
The thesis that mathematical truths are hypothetical was held about a
century before the rise of modern axiomatics by the Scottish
philosopher Dugald Stewart (1753-1828). Stewart’s position was
200 CHAPTER 3
Stewart was one of the first writers to make this point so clearly.16 On
the other hand, his discussion of this matter does not show any
awareness that mathematics, thus conceived, will per force be ab
stract or formal.
In a tedious discussion of “mathematical axioms”, Stewart denies
that these are the “foundation on which the science rests”. This is
because he understands by axioms such generalities as Euclid pro
posed under the name of common notions. “From these axioms - says
S tew art-it is impossible for human ingenuity to deduce a single
inference.” He contrasts them with such genuine principles as “All
right angles are equal to one another”, or Postulate 5, “which bear no
analogy to such barren truisms as these: - ‘things that are equal to
one and the same thing are equal to one another’; - etc.” 17 In Ste
wart’s opinion, the principles of geometry are not the axioms, but the
definitions. These he understands as hypotheses, which involve the
assumption that the defined entities exist.
A conception of mathematics as the study of abstract structures or
‘forms’ freely conceived by the human intellect and devoid of in
tuitive contents was resolutely put forward by Hermann Grassmann
(1809-1877) in his Ausdehnungslehre (1844). Since geometry refers to
a given natural object, namely space, it does not belong to mathema
tics. Nevertheless, there must be a branch of mathematics “which in a
purely abstract fashion generates laws similar to those which, in
geometry, are bound to space” .18 That branch is the theory of
F O UN D AT IO N S 201
If we carry through the proof of a theorem concerning straight lines (using the letters a,
h, c , . . . to designate linear forms in two variables for representing such lines), we have,
in fact, demonstrated an untold number of theorems. For if by the letters a, b, c , . . . ,
we no longer designate linear expressions but any general function in two variables,
provided they are of the same degree, the conditional equations F(u, h , . . . , m, n , . . . ) =
0 [which formulate the relations which hold between straight lines in the initial
hypothesis], as well as all the equations derived from them, retain their meaning. [. . . ]
If we have such a proof-schema we may relate it to lines of any arbitrary order. [. . . ]
We may therefore carry over every theorem in projective geometry to curves of any
arbitrary degree.24
Every geometrical relation is to be viewed as the pictorial representation of an analytic
relation, which, irrespective of every interpretation, has its independent validity.25
there is one and only one point b which lies in a given direction and at
a given distance from a (§10). This implies that a three-point system
or triangle is uniquely determined by a point a, two directions from a
and two distances marked, respectively, along each of those direc
tions (§18). The relation between two directions D(a, jc) and D(a, y)
from the same point a is also analyzed into two factors: the angle
between D(a, x) and D(a, y) and the half-plane, determined by D (a, x)
on which D(a, y) lies (§13). These two factors are seen to correspond,
respectively, to the factors of distance and direction that determine
the relation between two points. Bolzano assumes without proof that
the angle between two directions does not depend on the order in
which they are taken (§14). Let D(a, ) be a direction and let D(a, y )
jc
which forms a given angle with D(a, ) . D(a, ) and D(a, y) are then
jc jc
said to be opposite directions (§15). This definition does not imply that
the direction opposite to a given direction is unique, for there might
be many different angles such that D(a, x ) makes each with one and
only one direction (§16). Moreover, as Bolzano boldly points out, it
does not even imply that opposite directions exist, for it is conceiv
able that every direction makes every given angle with several direc
tions at a time (§24). However, according to him, the concept of
opposite directions furnishes the basis for a satisfactory definition of
the straight line if we grant one more assumption. This can be
paraphrased as follows: If a, b and c are three points and D(a, b) is
the same as D(a, c), then either D(b, a) is the same as D(b,c) and
D(c, a) is opposite to D(c, b), or D(b, a) is opposite to D(b, c) and
D(c, a) is the same as D(c, b); but if D(a,b) is opposite to D(a,c),
D(b, a) is the same as D(b, c) and D(c, a) is the same as D(c, b) (§24).
Bolzano contends that this assumption, like the two we mentioned
earlier (§§11, 14), can be proved without using the concept of straight
line.28 He defines: a point m lies between points a and b if D(m, a) is
opposite D(m, b); a straight line between two points a and b is an
object that contains all the points lying between a and b and no other
points (§26). It follows at once that any two points will determine a
straight line between them. Bolzano ‘proves’ that if a point c lies
between two points a and b, the straight line between a and c
together with that between c and b form the straight line between a
and b (§31). He fails to mention that c must be added to the former
two lines to complete the latter.
F O UN D AT IO N S 205
In other cases, they are more complex: such are the constructions of geometry.38
In these rational and abstract sciences it is essential to distinguish the hypotheses,
considered in them selves, which are a priori essentially arbitrary and are subject only
to the condition that ‘they do not contradict each other; and the value of these
hypotheses, regarded with a view to applications. Every abstract science, founded upon
non-contradictory hypotheses and developed according to the rules of logic, is, in itself,
absolutely true.39
denote proper points. Two pairs of proper points, ab, a 'b ’, each
marked on a rigid body, are said to be congruent if we can place a on
af so that b falls on b'\ also if there is a point-pair a"b'\ marked on a
rigid body, which is congruent with both ab and a'b'. This intuitive
notion can be extended in an obvious way to figures of more than two
points. According to Pasch, the following statements are evidently
true of configurations of proper points marked on one or more rigid
bodies, when congruence is understood in the foregoing sense. They
are adopted as axioms of congruence.67
(K I) Figure ab is congruent with figure ba.
(K II) Given a figure abc, there is one, and only one, proper point
b \ distinct from a, b and c, such that ab is congruent with ab' and bf
lies within segment ac or c lies within segment ab'.
(K III) If c lies within segment ab and if figure abc is congruent
with figure a 'b 'c', then c' lies within segment a'b',
(K IV) If Ci lies within segment ab, there is an integer n > 1 and n
points c2, . . . , cn+i on line ab, such that segment acx is congruent with
segment c,cl+i (1 < / < n ) and b lies within segment acn+x. (Axiom of
Archimedes).
(K V) If segment ac is congruent with segment be, figure abc is
congruent with figure bac.
(K VI) If two figures are congruent, their homologous parts are
congruent.
(K VII) If two figures are congruent with a third figure, they are
congruent with each other.
(K VIII) Given two congruent figures, if a point is added to one,
one can always add a point to the other in such a way that the
enlarged figures are congruent.
(K IX) Given two figures ab and fgh, such that ab is congruent
with fg and h does not lie on line fg, if F is any flat surface with
contains a and b, there is a flat surface G which contains F and
exactly two points c and d, such that the figures abc and abd are
congruent with fgh. There is, moreover, a point within segment cd
which lies on line ab.
(K X) Two figures abed and abce are not congruent unless all their
points are contained on the same flat surface.
Axiom K VI introduces the new undefined term homologous parts. Its
meaning is elucidated intuitively by Pasch: they are the parts which
cover one another when two congruent figures are superposed. But
216 CHAPTER 3
Let A, B0, Bi, P be collinear points. If A and Bi are separated by B0 and P, there is a
positive integer n, such that the nth point of the harmonic set (AB0BO is identical with
P or is separated by A and P from the (n + l)th point of the net B„+i. In this last case,
B0 and Bn+i are also separated by A and P.70
We are given thus a category of objects (end) called points. These objects are not
defined. We consider a relation between three given points. This relation, noted c € ab ,
is likewise undefined. The reader may understand by the sign 1 any category of objects
whatsoever, and by c t ab any relation between three objects of that category. [. . . ]
The axioms will be satisfied or not, depending on the meaning assigned to the undefined
signs 1 and c € ab. If a particular group of axioms is verified, all propositions deduced
from them will be true as well.76
one of them and with a point collinear with the other two.) Peano proves
that if two distinct points a, b belong to a line r and to a plane p, r is
contained in p (Peano (1889), §11, p.29).
(P XV) If h is a plane, there is a point a which does not belong to h.
(P XVI) If p is a plane, a a point not belonging to p and b £ a'p,
then, if x is any point, either x £ p or the intersection of p and ax is
not empty or the intersection of p and bx is not empty.
Definition: A figure k is said to be convex if every segment deter
mined by a pair of points of k is contained in k.
(P XVII) If h is a convex figure, a and b are points, a £ h and
bj£ /i, there is a point x such that (i) either x = a or x £ ab or x = b ;
(ii) ax is contained in h \ (iii) the intersection of bx and h is empty.
(P XVII implies that if a and b are points and k is a non-empty set of
points contained in ab, there is a point x belonging to {a}Uab U{b},
such that (i) k C\xb = 0, (ii) for every point y E ax, k D yb ^ 0. To show
this, choose h = ak U{a}. P XVII postulates, therefore, the continuity
of the straight line).
A set of axioms is said to be independent if none of them is a logical
consequence of the others. If a set of axioms is not independent, you
can eliminate one or more axioms, and obtain a smaller set, which still
determines the same axiomatic theory. In his paper of 1894, after
reproducing P I-P XI, Peano remarks that the “first scientific ques
tion” regarding them is whether they are independent or not. He
adds:
The independence of some postulates from others can be proved by means of examples
( esempi). The examples for proving the independence of the postulates are obtained by
assigning arbitrary meanings (dei significati affatto qualunque) to the undefined signs. If
it is found that the basic signs, in this new meaning satisfy ( soddisfino ) a group of the
primitive propositions, but not all, it will follow that the latter are not necessary
consequences ( conseguenze necessarie) of the former. [. . . ] Hence, to prove the
independence of n postulates, it would be necessary to give n examples of inter
pretation ( essempi di interpretazione ) of the undefined signs [...], each of which
satisfies n - 1 postulates, and not the remaining one.78
“examples”. (1) If point means integer and c £ ab means a < c < b, all
axioms P I-P XI are verified, except P IV. (2) If point means a real
number o f the closed interval [0,1], and c £ ab means a < c < b, all
axioms P I-P XI are verified, except P VII. (3) Pick three lines
through a point P. Eliminate all points to the left of P. We obtain
three half-lines originating at P. Let point mean a point of any of
these half-lines. If c £ ab means that c lies on the shortest way
leading from a to b over points in the agreed sense, all axioms
P I-P XI are verified, except P X.
Peano’s axiomatic treatment of congruence depends on one more
set-theoretical notion, besides those listed on p.219: the concept of a
mapping (corrispondenza). This, like all other set-theoretical ideas, is
viewed by Peano as a part of logic. Peano writes fx for the value
assigned by the mapping / to an object x. He introduces a class of
mappings, called affinities, defined on the set of points characterized
by P I-P XVII. Let a and b be points. If / is an affinity and c £ ab,
then fc £ (fa)(fb). P III implies then that affinities are injective. It can
be easily shown that affinities map collinear points on collinear points,
coplanar points on coplanar points. If / and g are affinities, the
composite mapping g ■f is an affinity. The identity mapping x ^ x is
obviously an affinity. Let ab be a segment, / an affinity. Is f{ab)
identical with the segment (fa)(fb)? Peano declares that he does not
know the answer to this question. In other words, he does not know
whether the inverse mapping f~ l is an affinity, and he cannot say
whether affinities, in his sense of the word, form a group.
The idea of congruence is introduced through the axiomatic
characterization of a class of affinities, called motions. Two figures k,
fc', are said to be congruent if there is a motion / such that fc' = /fc.
There are eight axioms of motion. The last four can be summarized in
one, using the defined concepts of half-line and half-plane.
(M 1) The class of motions is contained in the class of affinities.
(Peano remarks that M 1 is equivalent to Pasch’s Axiom K III.)
(M 2) The identity mapping is a motion.
(M 3) If / is a motion, the inverse mapping f~ l is a motion.
(M 4) If / and g are motions, the composite mapping g • f is a
motion.
Definition: Given two points a, b, the half-line Hl(a, b) is the set
abU {b}U a'b. Given three non-collinear points a ,b ,c, let r =
FO U N D A TI O N S 223
Hl(a, b) UHl(ft, a); the half-plane Hp(ab,c) is the set {jc | x ( y'z,
y (: r, z € cr}. (Remember that cr = {x \ x € cw, r}.)
(M 5) If a, b, c are three non-collinear points and x, y, z are three
non-collinear points, there is a unique motion which maps a on x,
Hl(a, b) onto H1(jc, y), and Hp(ah, c) onto Hp(xy, z).
From these axioms, Peano derives some theorems concerning axial
symmetry and orthogonality, translations and rotations.
3.2.7 The Italian School. Pieri. Padoa
Peano’s conception of axiomatized geometry as an abstract science
was shared in Italy in the 1890’s, not only by the group of mathema
ticians who collaborated with him in the formulation of all mathema
tical theories in the artificial language, but also by others who did not
take part in this enterprise and even looked askance on it. H.
Freudenthal credits G. Fano with the first unambiguous statement of
the abstract view of geometry. In a paper of 1892 concerning the
postulates of n-dimensional linear geometry, Fano declares:
As a basis for our study we posit an arbitrary manifold of objects of any nature
whatsoever, which, for brevity, we shall call points, on the understanding, however,
that this name is independent of their own nature.79
As we saw above, Peano had said as much three years earlier (p.219,
reference 76), and it should not be too hard to discover other
statements of the same idea in contemporary Italian literature. Thus,
Giuseppe Veronese (1854-1917), in the historico-critical appendix to
his influential book Fondamenti di Geometria (1891), criticizes Pasch
for paying too much attention to the intuitive meaning of undefined
geometrical concepts. This forced him to distinguish quite un
necessarily between proper and improper objects, though both have
the same geometrical properties, and led him to restrict the scope of
his axioms, so that they did not clash with the evidence of the senses.
Pasch, observes Veronese,
rightly maintains that proofs must be independent of the intuition of the figure, or
rather, as he understands it, of the sense representation of the figure. This aim,
however, cannot be fully attained [. . . ] unless the axioms give us well-defined abstract
properties independently of intuition.®°
If you maintain that the postulates of geometry are nothing but rigorous formulations
of the intuitive concept of physical space (which merely impress stability and a seal of
rationality on the facts of spatial intuition), you ascribe, in my opinion, too much
importance to an objective representation, which you treat as a conditio sine qua non
of the very existence of geometry, whereas the latter can, in fact, very well subsist
without it. Today, geometry can exist independently of any particular interpretation of
its primitive concepts, just like arithmetic.84
Indeed, after the work of Bolyai and Lobachevsky, one can no longer
expect geometrical axioms to be intuitively evident. “How could you
account for the intuitive evidence of the postulates proper to so-
called non-Euclidean geometries, after you have found Axiom XII on
parallels evident, or vice versa?”85 It is pointless to demand that the
primitive concepts of geometry be intuitively clear, since these (“with
the exception of the logical categories, which are necessary to all
discourse and consequently cannot be described by words”) can be
given through “implicit definitions or logical descriptions [...] or as
the roots of a system of simultaneous logical equations” .86
For instance: we call, respectively, point and motion every determination of classes
II and M which have the following properties:. . . (list here the premises concerning points
and motions, denoted respectively, by II and M).87
FO UN D AT IO N S 225
Such a description conceals, in fact, a system of postulates. But since these, dressed as
definitions, amply exhibit their nature as conditional propositions concerning the primitive
concepts (i.e. their naturally arbitrary character, etc.), nobody will ask whether they are
self-evident or not. The postulates, like every conditional proposition, are neither true nor
false : they only express conditions which may or may not be verified. Thus, the equation
(x + y)2 = x2+ 2xy + y1 is true if x, y denote real numbers, false if they denote
quaternions.88
It can happen that there are many (indeed infinitely many) interpretations of a system
of undefined symbols which verify the system of unproved propositions, and,
consequently, every proposition of a theory. The system of undefined symbols can be
considered then as the abstraction of all these interpretations."
F O UN D AT IO N S 227
(III 5) If, for two triangles ABC and A 'B'C', we have that AB =
A 'B ', AC s A 'C \ ± B A C - then, always, £ A B C ^ X,A'B'C’.
V. Axioms of Continuity
(V 1) (Axiom of Measurement or Archimedean Axiom.) If AB and
CD are any segments, there is a positive integer (Anzahl) n, such
that, by successively copying CD n times from A on the ray through
B, you pass beyond B. (The meaning of this axiom will be clear to
everybody, though Hilbert employs in its formulation some expres
sions which he has not defined and are not sufficiently characterized
by the axiom itself. To copy CD successively k times (k 5= 1) from A
on the ray through B is to find the unique point A k of that ray, such
that Ak-iAk = CD, where A k~\ = A if k = 1, and is determined by the
aforesaid condition if k > 1 (on the uniqueness of A k: Hilbert, GG,
p.15). You pass beyond B by successively copying CD n times from
A on the ray through B, if B lies between A and A n.)
(V 2) (Axiom of Linear Completeness.) The system of points of a
line, with their relations of order and congruence, cannot be extended
in such a manner that the relations between the former elements, and
the fundamental properties of linear order and congruence which
follow from Axioms I—III and Axiom V 1, are all preserved.110
My translation of V 2 badly needs a paraphrase. M. Kline (MT,
p.1013) gives the following: “The points of a line form a collection of
points which, satisfying Axioms I I , 12, II, III and V 1, cannot be
extended to a larger collection which continues to satisfy these
axioms” . This sounds much better, but is not essentially clearer. What
does it mean to extend the collection of points on a line? In any given
interpretation of Hilbert’s axioms, each object called line is asso
ciated with a set of objects called points, which are said to lie on it.
To extend this set, one must change the interpretation. V 2 is thus
seen to differ substantially from the other axioms. Instead of stating
some new fact about incidence, betweenness or congruence or intro
ducing a new property of points, lines or planes, V 2 takes, so to
234 CHAPTER 3
speak, a stand outside the axiom system and says something about its
relation to the sets of objects which might conceivably satisfy it. V 2
is what nowadays one would call a metatheoretical statement; though
one of a rather peculiar sort, since the theory with which it is
concerned includes Axiom V 2 itself. To show this, let me paraphrase
V 2 once more. Let H denote Hilbert’s axiom system without V 2.
Every model of H includes many things called lines (11,18). On each
such line and the set of points lying on it, H induces a structure which
we may call line geometry. This structure is determined by Axioms I 3
(first sentence), II 1-3, III 1-3 and V 1, plus the following three
propositions which are theorems of H: (i) If A and B are points on
the line, there is a point C which lies between A and B \ (ii) any four
points on the line can be labelled A u A2, A 3 and A 4 in such way that
A2 and A3 lie between A\ and A4, A 2 lies between A x and A3 and A3
lies between A2 and A4; (iii) the point B f whose existence is postu
lated in Axiom III 1 is unique. Let L designate this axiom system,
while L* designates the system obtained by adding V 2 to L. It is not
hard to find two interpretations I and / ' such that (i) in I the set of
‘points on the line’ is a given set m, while in V it is the set m U{z},
where z is some object not belonging to m ; (ii) if w, v and w belong to
m, v ‘lies between’ u and w in I if, and only if, v ‘lies between’ u
and w in (iii) if t, u, v and w belong to m, {t, u} and {v, w} are
‘congruent’ in I if, and only if, they are ‘congruent’ in (iv) I and
I' are modellings of L. Thus, for instance, one may take m to be the
field of rational numbers and z to be tt and stipulate that in both I
and r ‘v lies between u and w’ means that u < v < w and ‘{t, u} is
congruent with {v, w}’ means that \t - u\ = \v - w|. Now, Axiom V 2
says in effect that two interpretations / and V fulfilling conditions
(i)-(iii) cannot both be modellings of L*, even if they happen to be
modellings of L. The curious thing is that V 2 does not introduce any
new determination of congruence or betweenness that might preclude
two modellings of L satisfying (i)—(iii) from simultaneously satisfying
L*. V 2 merely declares that the addition of itself to axiom system L
restricts modellings in the stated manner. There is something highly
unsatisfactory about the inclusion of a statement of this kind in an
axiom system. Richard Baldus (1930) showed, however, that the full
import of Axiom V 2 is given by the following Cantorean axiom:
There exists a segment A 0B0 with the following property: If Ah R, is a sequence of
point-pairs such that (i) for every positive integer ny An and Bn lie between A„_i and
F O UN D AT IO N S 235
Hilbert replied:
S in c e I began to think, to w rite and to lectu re ab ou t th e se m atters, I h a v e a lw a y s said
e x a c tly the contrary. If the arbitrarily p o sited a x io m s d o not co n tra d ict o n e an oth er or
an y o f their c o n se q u e n c e s, th ey are true and the things defined b y them e x ist. T hat is
for m e th e criterion o f truth and e x is t e n c e .113
set of all functions / : R-»R which fulfil one of the following condi
tions, for every t € R: (i) f(t) = t; (ii) for some g, h£Cl ( t ) , /(*) =
g(t) + h( t ) or f ( t ) = g( t )~ h(t), or f(t) = g(t)h(t), or, provided that
h(t) # 0 for all t 6 R, f(t ) = g(t )l h(t ); (iii) for some g € (l(t), f ( t ) =
Vl + (g(t))2. I f / € 0 ( 0 , / is an algebraic function on R. Consequently,
either / is identically zero, or /(f) = 0 for, at most, a finite set of
values of the argument f. In other words, unless / = 0, there is a real
number tf, such that t — tf is the largest solution of f( t ) = 0. For every
real number t > f/, /(f) is positive, in which case we shall say that / is
positive, or negative, in which case we shall say that / is negative. Let
f u fi belong to fl(t). We stipulate that f x> f 2 if f i ~ f i is positive, and
that /i < f i if f \ ~ f i is negative. Let f denote the function f •—> f ; /i, the
constant function t ^ n (n a non-negative integer). By the above
stipulation, n < f, since, for sufficiently large values of f, n - 1 < 0
always. We obtain a modelling of non-Archimedean geometry by
substituting H(f) for fl in our earlier description of Hilbert’s model
ling of H. We define as usual the length \AB\ of a segment AB by the
Pythagorean theorem. If O, X , Y are respectively the points (0,0,0),
(1,0,0) and (f,0,0), there is no positive integer n such that n \O X \^
io n
The Archimedean axiom enters into Euclid’s Elements as a
presupposition of the theory of proportions developed in Book V. We
saw on page 11 how Euclid, following Eudoxus, defined a linear
ordering on the set of ratios between magnitudes. If a and b are two
lengths, a < b implies that a /a > a lb (Euclid, V, 8). Consequently,
for any two lengths such that a < b, there must exist positive integers
m, n, such that m a > n a but m a ^ n b . Obviously m, n fulfil this
condition only if it is also fulfilled by n + 1 and n. But then (n + l)a ^
nb. Hence a ^ n(b - a). This presupposes, however, that for any pair
of lengths (areas, volumes) a and d - b - a, there exists a positive
integer n such that nd ^ a.
In Chapter III of the Grundlagen, Hilbert builds a new theory of
proportions, which can be used to compare lengths and areas (but not
volumes) in the space determined by Axioms I-IV, without assuming
the Archimedean axiom V 1. This theory rests on an algebra of
segments, which is essentially the same as developed by Descartes in
his Geometrie (see Section 1.0.4). But, while Descartes bases the
construction of the product of two segments on the theory of propor
tions, via Euclid VI, 4, Hilbert shows that the uniqueness of the product
F O UN D AT IO N S 239
mention that a sphere does not have more than one centre, but this
follows easily from his Theorems 15 (“Through a given point there is
not more than one line parallel to a given line”) and 17 (“A segment
cannot have more than one middle point”). These theorems depend
on rather strong axioms, which imply that every plane is both
Arguesian and Euclidean. With these elements, Huntington can pro
ceed to define the congruence of segments. Two segments [AB] and
[CD] are congruent if, and only if, one of the following conditions is
satisfied: (la) If line AB = line CD, either [AB] = [CD] or mid AC =
mid BD or mid AD = mid BC. (lb) If AB is parallel to CD, ABCD is a
parallelogram with diagonals [AC] and [BD]. (2a) If [AB] and [CD]
have a common midpoint, that midpoint is the centre of a sphere of
which [AB] and [CD] are chords. (2b) If [AB] and [CD] have a
common endpoint, that endpoint is the centre of a sphere of which
[AB] and [CD] are radii (a segment is a radius of a sphere S if one of
its endpoints is the centre of S and the other is the endpoint of a
chord of S). (3) There exist two segments [OX], [OY] which are
mutually congruent by (2) and are congruent by (1) with [AB] and
[CD], respectively.
Huntington classifies his axioms into existence postulates, which
demand the existence of some entity satisfying certain conditions,
and general laws, which say that if such and such entities exist, then
such and such relations will hold between them. Except for postulate
E l , which posits the existence of at least two distinct points, the
remaining existence postulates are conditional statements of the type
‘if this exists, then that exists as well’. But then, Huntington’s
general laws suffice to prove many existential theorems of this kind.
This explains perhaps why his classification has not been adopted by
other authors. By means of novel and ingenious models (which he
calls pseudogeometries), Huntington proves that the general laws are
independent of each other, while the existence postulates are in
dependent of each other and of the general laws. The system’s full
independence could be easily achieved by slight changes in wording,
but, the author observes, “such changes would tend to introduce
needless artificialities”.126 The consistency of the system is proved by
constructing a numerical model, in which the spheres are the closed
balls in R3, with radius r ^ \k\ (k € R). Huntington stresses that, since k
need not be zero, “we may speak of a perfectly rigorous geometry in
which the ‘points’, like the school-master’s chalk-marks on the
246 CHAPTER 3
blackboard, are of definite, finite size, and the ‘lines’ and ‘planes’ of
definite, finite thickness”.127 It should be noted, however, that
Huntington’s finite points behave in everything like their classical
counterparts: there are indenumerably many of them in any segment,
any two of them determine a unique line, etc. Indeed, any line can be
endowed with the structure of a complete ordered field by arbitrarily
choosing a zero point and a unit point on it. There is, thus, some
unwitting mockery in Huntington’s reference to school blackboards.
Let us finally mention that, like Veblen, Huntington makes a point of
showing that his axiom system is categorical (in the classical sense).
The reader has probably observed that Huntington’s space of
spheres is partially ordered by the relation of inclusion or contain
ment. If we agree to add to it a universal sphere U, which contains
every other sphere, and a void sphere V which is contained in every
other sphere, we can easily conceive Huntington’s space as a
lattice.128 But, though the idea of a lattice had been developed (under
a different name) by Dedekind in 1900, it went unnoticed until it was
independently rediscovered by Karl Menger and Garrett Birkhoff
some thirty years later. These authors immediately based on it a
revolutionary approach to the foundations of geometry.129 n-dimen-
sional projective and affine geometries can be entirely built in terms
of the lattice operations of joining and intersecting. Special postulates
differentiate projective from affine lattices. A similar foundation can
be provided for BL geometry, but not for Euclidean geometry. This
means that, in a definite sense, the latter is less simple than the
former.
We cannot study these matters here, but a few indications might
stimulate the reader’s curiosity.130 We consider a domain of entities
called flats. We define two associative, commutative operations which
assign to every pair of flats A, B a flat A["1B called their meet and a
flat A U B called their join. There is a unique flat U such that, for
every flat A, A n U = A and a unique flat V such that, for every flat A,
A U V = A. If AriB = A and A UB = B we say that A is a part of B
and write A C B. If A ^ B and A C B, A is a proper part of B (noted
ACB) . A flat whose only proper part is V is called a point.m Our
operations satisfy the ‘‘law of absorption” ,
A U (A n B ) = A n (A U B ), (1)
and the ‘‘law of intercalation” : if P is a point and A and B are two
FOUNDATIONS 247
4.1 E M P I R I C I S M IN G E O M E T R Y
lines exist, i.e. that given a straight line you can find another line,
which is straight like the first, and all of whose points lie at a fixed
distance from it. Mill believes, however, that the existential assump
tions of the definitions of geometry are actually false “There exist no
real things exactly conformable to the definitions. There exist no
points without magnitude; no lines without breadth, or perfectly
straight; no circles with all their radii exactly equal, nor squares with
all their angles perfectly right.”6 Mill denies that these geometric
objects are even possible: their existence would seem to be in
consistent with the physical constitution of the universe. He also
rejects the interpretation which regards them as purely mental enti
ties. Our ideas are copies of the things which we have met in our
experience. “Our idea of a point, I apprehend to be simply our idea of
the minimum visible, the smallest portion of surface which we can
see. A line, as defined by geometers, is wholly inconceivable.”7 Mill,
on the other hand, will not admit that a science like geometry might
deal with non-entities. How does he reconcile this Platonic thesis with
his former remarks about the objects described by the definitions of
geometry? Not indeed after the fashion of Plato, by claiming that
these objects, because they are the concern of a genuine science,
possess some sort of being of their own. In Mill’s opinion, the objects
characterized in the definitions are simply “such lines, angles, and
figures as really exist”8; only that in geometry we disregard all their
properties except the geometrical ones, and we even ignore the
“natural irregularities” in these. Mill says that his position is that of
Dugald Stewart, who maintained that geometry is built on hypo
theses.9 But he remarks that the term hypothesis has here a somewhat
peculiar sense, meaning, not “a supposition not proved to be true, but
surmised to be so, because if true it would account for certain facts” ,
but a proposition “known not to be literally true, while as much of [it]
as is true is not hypothetical but certain”. Indeed “the hypothetical
element in the definitions of geometry is the assumption that what is
very nearly true is exactly so. This unreal exactitude might be called a
fiction, as properly as an hypothesis”.10 On the character of these
scientific fictions, which other writers have called idealizations, Mill
observes the following:
S in ce an h y p o th e sis fram ed for the pu rp ose o f scien tific inquiry m ust relate to
so m eth in g w h ich has real e x is te n c e (for there can be no s c ie n c e resp ectin g n o n -en tities)
it fo llo w s that any h y p o th e sis w e m ake resp ectin g an o b je c t, to fa cilita te our stu d y o f
258 CHAPTER 4
it, m ust n ot in v o lv e an yth in g w h ich is d istin ctly fa lse , or repugnant to its real nature;
w e m u st not a scrib e to the thing any p rop erty w h ich it has not; our lib erty e x te n d s o n ly
to ^ slightly exaggeratin g so m e o f th o se w h ich it has (by a ssu m in g it to be co m p le te ly
w h at it is very nearly) and su p p ressin g o th ers* , under the in d isp en sa b le ob lig a tio n o f
restorin g them w h e n e v e r , and in as far a s, their p r e se n c e or a b se n c e w o u ld m ake any
m aterial d ifferen ce in the truth o f our c o n c lu sio n s. O f this nature, a cco rd in g ly , are the
first p rin cip les in v o lv e d in the d efin ition s o f g e o m e tr y .11
prove that the factually given can be explained only from these
premises.”27 This modern-sounding epistemological conception was
already held by Ueberweg when he wrote his first introduction,
though he stated it less neatly there. Even if a philosopher might
succeed in deriving geometry from the essence of space - he says - he
would not thereby discover the foundation of the general belief in its
validity. “This is, in the case of geometrical axioms, the same in fact
as in the case of physical hypotheses, namely, the uninterrupted
approximate confirmation of their consequences by experience. In
numerable propositions derived from the geometrical axioms allow
comparison with experience through factual construction. Absolute
agreement is, of course, impossible, because we cannot construct
with absolute exactness; but we find that, within the reach o f our
experience, the more exact our construction, the more exact is the
agreement.”28
But Ueberweg’s main purpose is not to determine wherein lies the
certainty of the indemonstrable principles of geometry, but, as a
preliminary contribution thereto, to exhibit a connection between the
familiar axioms, postulates and fundamental concepts in Euclid, by
deriving them from a common source. This is found by (1) analysing
our global sensory awareness, in order to obtain general concepts and
propositions; (2) idealizing the latter by ascribing them absolute,
infinite precision. If we can show that the whole of geometry can be
built upon this basis just as well or even better than upon Euclid’s
principles, we shall have paved the way “for the right opinion
concerning the logical character of the Euclidean axioms and for the
recognition of geometry as a natural science”.29
Ueberweg observes that “space is separated from the whole of
sensory intuition only through the perception of movements” .30 The
main facts revealed thereby can be stated in many ways. Ueberweg
formulates them as follows:
These obvious, familiar empirical facts are now idealized. That is, we
assume that the stated properties are true with absolute precision.
Ueberweg justifies this assumption, as we might expect, by the
empirical truth of its consequences, especially by the fact that the
propositions inferred from the idealized statements I-IV are mutually
consistent and agree with experience with increasing precision as we
carry out our constructions more exactly.32 Lie has shown that
axioms essentially equivalent to Ueberweg’s are sufficient for charac
terizing three-dimensional maximally symmetric spaces.33 But
Ueberweg thought he could characterize Euclidean space with them.
His derivations are therefore inevitably defective.
We shall only discuss his proof of the parallel postulate. Ueberweg
defines a point as “the absolutely simple space element’’; he charac
terizes it also as that element of a body which is such that any
two - but not any three - of them can be fixed without altogether
impeding the movement of the body. A movement with a fixed point
P is called a rotation about P. The set of all points occupied by a
figure F during a movement m is the path ( Weg) of F during m. A line
is the path of a moving point. A straight line is a line whose path
during rotation about two of its points coincides with itself. Given
two points P, Q, there is one and only one straight line through P and
Q.34 A straight line can also be defined as a line of constant direction.
In order to explain this, Ueberweg goes into a detailed discussion of
the concept of direction (Richtung). A moving figure changes its place
(Ort). Two places differ only in their position (Lage). If a point P is
carried to a point Q over a path absolutely determined by P and Q
“that determination of the transit of P to Q which depends on the
position of Q relatively to the other points which Q can take while
rotating about P [in other words, on the position of Q within the
sphere through Q centred at P] is called linear direction (Linien-
richtung)”.35 In the light of this definition, Ueberweg concludes that a
straight line is a line of one direction. The difference between the
directions of two straight lines meeting at a point P is called
angle. Ueberweg defines circles and arcs and shows how to use the
latter to measure angles. Two straight lines m, n which make equal
corresponding angles with a third line t are said to have the same
direction. Ueberweg defines parallels as straight lines that have the
same direction. Under this definition, two lines m, n which are
parallel relatively to a transversal t (with which they make equal
264 CHAPTER 4
dealt with them in Parts 2.2 and 3.1. They were extensively discussed
and made known to the philosophical public in The Axioms of
Geometry, a book published in 1877 by Benno Erdmann (1851-1921).
The chief purpose of this work is to show that the new geometric
theory of space, which Erdmann ascribes to Riemann and Helmholtz,
confirms the empiricist theory of spatial intuition and refutes Kant’s
philosophy of space and geometry. In order to show this, Erdmann
first presents the said theory as a successful attempt to provide a
definition of space. The definition arrived at, after a carefully
motivated exposition, is simply a restatement of Helmholtz’s axioms
for Euclidean geometry.38 Erdmann’s exposition is somewhat naive
and is marred by several mathematical misconceptions. Since the
book was very popular among philosophical readers, these features
have probably exerted a damaging influence on philosophical discus
sions about space and geometry.
According to Erdmann, Riemann’s lecture showed how to define
the concept of space by specification of the general concept of an
n-fold extended manifold. The procedure, he says, is analogous to
that used in analytical geometry for providing the concepts of in
tuitively given spatial figures. It is merely a matter of finding analytic
determinations which correspond to every essential trait of our in
tuitive representation of space. Thus, the intuitive feature usually
described by saying that space is three-dimensional “is characterized
by the fact that the position of every point is univocally determined
by its relations to three mutually independent spatial quantities, e.g.
to a system of three orthogonal coordinate axes” . This is expressed
analytically by the dependence of every point upon three independent
real variables (coordinates).39 The continuity of space is manifested
intuitively by its infinite divisibility.40 Analytically, this is expressed as
follows: as an object moves in space from a point A to a point B, the
coordinates of its position must take all real values between their
value at A and their value at B; if two coordinates change together,
while the third remains fixed, their quotient approaches a limit as their
variation tends to zero.41
Erdmann proposes the following general concept of space: Space is
a continuous quantity whose elements are univocally determined by
three mutually independent (real) variables.42 Erdmann classifies such
3-fold determined continuous quantities into two kinds: those which
have and those which do not have interchangeable coordinates.
266 CHAPTER 4
wholly depend on things, while rationalism holds that they are wholly
independent from them. Both persuasions have three varieties.
Sensualist empiricism believes that our representations agree with
things absolutely. Formalist empiricism claims only a partial
agreement, covering “the quantitative relations of space, time and
lawfulness (Gesetzlichkeit)”, while representations differ from things
in all their qualitative aspects. There is finally a brand of empiricism
which Erdmann proposes to call apriorism (Apriorismus)\ this seeks
to show that all our representations are completely different from the
constitution and the relations of things, but correspond to them in
every part. The rationalist can also maintain that representations,
though uncaused by things, entirely agree with them (preestablished
harmony); or that they agree only partially, so that, say, the forms of
thought are identical with the forms of being (formal rationalism); or,
finally, that every element in our representations is not only entirely
independent of things, but specifically different from them (nativism).
Geometric knowledge concerns the properties, specifically the
metric properties of our representation of space. The philosophy of
geometry may take any of the above forms, as applied to this
particular representation. According to Erdmann, “the mathematical
results force us to conclude that our representation of space must be
unambiguously conditioned by the actually experienced effects of
things upon our consciousness”.48 In other words, a rationalist
philosophy of geometry is incompatible with the results of geometric
research. Three arguments back this conclusion:
(i) The logical possibility of n-dimensional manifolds (n > 3) shows
that “the influence of experience, i.e. of the things affecting us from
outside” does not merely awaken but actually determines our parti
cular representation of space as a three-dimensional manifold.49 (This
argument is valueless: in Kant’s philosophy, the a priori nature of our
intuition of space is certainly compatible with the logical viability of
other spaces with a different structure.)
(ii) The foundations of geometry involve the empirical concepts of
rigid body and movement.50 (The existence of freely movable rigid
bodies is the fact which, according to Helmholtz, lies at the foun
dation of geometry. But perfectly rigid bodies do not actually exist.
To say that the concept of such a body is obtained from experience is
therefore somewhat far-fetched.)
(iii) If the representation of space were generated independently of
EMPIRICISM, APRIORI SM, C ON V E N T I ON A L I S M 269
seems contradictory. The contradiction looks even sharper when we consider that this
form, as the form of intuition of our sensations, must be, like these, the product of an
interaction, so that it cannot exist apart from this interaction, as the form of a part of
the interacting elements. The whole question depends therefore on the subjectivity of
sensations. If this is granted-and there can no longer be any doubt about this
point - the subjectivity of the binding forms, especially of our representation of space,
will follow.54
general concept, defined by property (a) alone. Let us call this the
general straight. The theory of this form has been called non-Eucli-
dean geometry. Calinon prefers to call it g e n e r a l g e o m e t r y , because
“far from being the negation of Euclidean geometry, it includes the
latter as a special case” .70 The analytic development of general
geometry supplies formulae which contain an undetermined
parameter. The familiar formulae of Euclidean geometry are obtained
by assigning a definite value to this parameter. Calinon points to a
seeming paradox. The general straight through two given points P, Q
depends on the undetermined parameter, so that we shall have
infinitely many straights through P and Q, one for each value of the
parameter. This contradicts however the definition of the general
straight.71 Calinon overcomes this difficulty by treating each value of
the parameter as the characteristic of a different three-dimensional
space. On each space there is just one straight between two given
points P and Q. This leads Calinon to an even broader conception of
general geometry: ‘‘General geometry is the study of all spaces
compatible with geometrical reasoning” , i.e. of all consistent systems
of one-, two- and three-dimensional forms. The spaces studied by
Euclid and the classical non-Euclidean geometries are only a proper
subset of such systems, which may be called i d e n t i c a l s p a c e s
( e s p a c e s i d e n t i q u e s ) , for “every figure constructed at a given point of
slabs; we can mark two points on each surface and compare their
distances by superposition, without having to employ a “very thin
string” as an instrument of comparison. Mach observes that pro
positions equivalent to Euclid’s fifth postulate can be proved ap
proximately by means of easy experiments. Thus, with a set of
congruent triangular floor-tiles we can construct the figure shown in
Fig. 20 which is possible only if equidistant lines are straight. With a
triangular piece of paper, folded as in Fig. 21, we can prove that
the three internal angles of a triangle are equal to two right angles
(when the paper is folded along EF, FG, GH, the vertices A, B, C
meet at X and the angles at A, B, C appear as the parts of a single
straight angle). The first of these two experiences was probably
familiar to men of the earliest civilizations. On the intuitive origin of
the idea of straightness Mach repeats a trite remark: “A stretched
thread furnishes the distinguishing visualization of the straight line.
The straight line is characterized by its physiological simplicity. All its
parts induce the same sensation of direction; every point evokes the
mean of the space-sensations of the neighbouring points; every part,
however small, is similar to every other part, however great”.95 This
characterization however is but of little use to geometers. It is a
Fig. 21.
EMPIRI CI SM, APRIORI SM, C O N V E N T I O N A L I S M 283
D ifferen t id eas can e x p r e ss the fa c ts w ith the sam e e x a c tn e ss in the dom ain a c c e ssib le
to o b serv a tio n . T h e facts m u st h en ce be ca refu lly d istin g u ish ed from the intellectual
co n stru cts the form ation o f w h ich th ey su g g ested . T he la t t e r - t h e c o n c e p ts - m ust be
EMPIRICISM, APRIORI SM, C ON V E N T I ON A L I S M 285
Lotze has apparently missed the point. If r denotes a straight line and
w an angle, as intuited by us, then, Lotze maintains, r and w as
elements of space determine its global configuration and internal
structure completely and unambiguously in full agreement with tradi
tional geometry. “It would be unfair to demand another proof of it
than the one provided by the actual development of science until
now.”7 That the elements r and w admit of other combinations that
are not made intuitive in our space and that such combinations are
not just verbally describable abstract possibilities, but lead to spatial
intuitions different from our own, can only be proved by actually
producing the intuitions.
Lotze believes that the Euclidean system is a perfectly satisfactory
expression of our space intuition and he will not enter into any
discussion on this matter. He knows that the theory of parallels is
regarded by many as a weak spot in the Euclidean system. That the
theory rests upon an undemonstrable postulate is, of course, no
objection to it. Geometry, as a description of intuition, is necessarily
built upon undemonstrable premises. Like other philosophers before
him, Lotze thinks that he can improve the intuitive obviousness of
Euclid’s theory by redefining parallels. He provides two new
definitions, which he apparently believes to be equivalent:
We know that these definitions are not equivalent and that none of
them is equivalent to Euclid’s. The first definition implies, of course,
that b is the only parallel to a on plane e through the meet of b and c \
but it does not imply that b is the only straight line which fulfils the
italicized condition but does not meet a. The second definition does
not guarantee that parallel lines are straight. If we include this
requirement we must prove (or postulate) that such lines exist. Lotze
probably regarded it as intuitively obvious.
Lotze was, as far as I know, the first one to make the following
important remark, which Poincare later used in support of con
ventionalism. In Euclidean geometry, the three internal angles of a
EMPIRICISM, APRIORI SM, C ON V E N T I ON A L I S M 289
triangle are equal to two right angles. This fact, Lotze claims, is not
subject to experimental verification or refutation. If astronomical
measurements of very large distances showed that the three angles of
a triangle add up to less than two right angles, we would conclude that
a hitherto unknown kind of refraction has deviated the light-rays that
form the sides of the observed triangle. In other words, we would
conclude that physical reality in space behaves in a peculiar way, but
not that space itself shows properties which contradict all our in
tuitions and are not backed by an exceptional intuition of its own.9
Lotze completes his discussion of modern philosophico-geometrical
speculations with a criticism of some of the concepts which turn up in
them, such as intrinsic geometry, fourth dimension, space curvature.
Lotze’s remarks suggest that he had not actually studied the works of
Gauss and Riemann, but tried to reconstruct their meaning proprio
Marte from a few vague hints. Thus, in his opinion, it is impossible to
define or measure a curve without presupposing the intuition of the
straight line from which that curve deviates. (Lotze, M, p.246). Lotze
is probably thinking of the classical definition of the length of a curve
as the limit of a sequence of lengths of polygonal lines inscribed in it
(see p.69). But this definition had been discarded by Riemann when he
demanded that every line should measure every other line (Riemann,
H, p.12; see p.90f.). Lotze discusses Helmholtz’s Flatland at
length. He maintains that a two-dimensional rational being living upon
a surface would develop the notion of a third dimension in order to
understand the fact that straightest lines in his world return upon
themselves. That notion would arise in him, not as a product of
immediate perception, “sondern auf Grund des unertraglichen
Widerspruchs, der in dieser sich selbst wiedererreichenden Geraden
lage” (Lotze, M, p.252). Now, I do not see why/, if we are willing to
grant Helmholtz’s fiction, we cannot also admit'that straightest lines
in spherical Flatland are ordered cyclically like projective lines. It is
difficult to make sense of Lotze’s refutation of the fourth dimension
(Lotze, M, pp.254-260). It rests upon the notion that the number of
dimensions of a space is equal to the number of mutually orthogonal
straight lines that meet at an arbitrary point of it. Lotze maintains that
this number cannot possibly exceed three. He apologizes for his in
ability to substantiate this claim with stronger arguments than the one
proposed by him (Lotze, M, p.257; Lotze’s argument is explained and
criticized in Russell, FG, p.l06f.). His discussion of space curvature
290 CHAPTER 4
argument fails to prove that the same set can be viewed indifferently
as having three or four dimensions (in Wundt’s sense). Dimension
number is defined after Brouwer (1913) as a topological property,
ascribable to a wide variety of topological spaces (Section 4.4.6). If a
set S with topology T has dimension n, it is generally possible to
define on S a topology T', which makes S into an m-dimensional
topological space (m ^ n ). From this point of view, the dimension
number of physical space can be regarded as conventional if, but only
if, we are free to endow it with several incompatible topologies,
which bestow on it different dimension numbers.
4.2.3 Charles Renouvier
Charles Renouvier (1815-1903), head of the French Neokantian
school, developed his views on the old and the new geometries in an
essay entitled “La philosophic de la regie et du compas’’.27 It aims at
“demonstrating the illogical character of non-Euclidean geometry’’.28
This aim is directly pursued in the second half of the essay, devoted
to “the sophisms of general geometry’’. Renouvier carries his ani
mosity towards the new geometries to the point of saying that
“anyone who believes that he may question the objective foundation
of the old geometry [...] cannot consistently think that the objective
foundation of morality is better safeguarded against doubt.”29 The
“illogical character” of non-Euclidean geometry is conceived rather
broadly. Renouvier grants that BL geometry is exempt from
contradiction. Indeed if BL geometry were contradictory, Euclid’s
fifth postulate would be a demonstrable truth, instead of an indemon
strable principle of geometry. The contradictory, however, is only a
species of the absurd, a much vaster genus of inadmissible notions
and untrue propositions, including, in particular, “the ideas and pro
positions which contradict the regulative principles of the under
standing”. BL geometry belongs precisely to the latter variety of the
absurd, “because it rests on the supposition that one of the principal
laws of our representation of space and figures does not express a
real relation”.30 The actual development of an absurd but noncon
tradictory geometry by Lobachevsky provides a welcome confirma
tion of Kant’s thesis that some of the principles of geometry are
synthetic judgments, in other words, that geometry must be based on
undemonstrable postulates.
For Renouvier, the ultimate foundation of geometry is intuition. He
EMPIRI CI SM, APRIORI SM, C ON V E N T I O N A L I S M 295
thus calls the “ideas of space and spatial relations insofar as they
consist of intellectual phenomena which can be analyzed, no doubt,
and from which consequences can be drawn, but which cannot be
demonstrated nor reduced to other phenomena without begging the
question.”31 In contrast with Kant, who restricted analysis to the
elucidation of concepts, and regarded spatial intuition as the source of
a priori synthetic judgments, Renouvier maintains that the contents of
intuition is expressed in analytic judgments, and treats intuitive and
analytic as synonymous. The first “fact of intuition” which lies at the
foundation of geometry is three-dimensional space itself (Vetendue
elle-meme> a trois dimensions), “in which every figure is imagined,
defined in its internal relations and placed by the mind as in a
shapeless medium (comme en un milieu lui-meme sans figure)”. This
primary fact of intuition immediately implies, according to Renouvier,
that “a figure can be transported everywhere in space, without
altering its elements or the relations of its parts”.32 This “law of the
conservation of figures” - which is tantamount to Helmholtz’s free
mobility of rigid bodies - “contains in principle every other fact of
geometrical intuition”. Does it suffice to determine the Euclidean
character of true geometry? On this point, Renouvier’s position is
ambiguous. On the one hand, he maintains that Euclidean geometry
cannot be established by analysis alone, but demands an intellectual
synthesis, which apparently operates upon intuitive data but is
somehow superimposed on them. Thus, while the statement that two
straight lines never enclose a space merely analyzes, in Renouvier’s
opinion, the intuitive notion of a straight line (defined as a line of
constant direction), the statement that the straight line is the shortest
between two points involves a synthesis which can only be due to the
understanding. On the other hand, when he discusses Riemann and
Helmholtz, he concludes that the “law of conservation of figures” is
fulfilled only in Euclidean space.
This conclusion is somewhat deviously stated at the end of p.52.
Renouvier quotes from Helmholtz (1866). He does not seem to be
aware of the big mistake in that work (corrected in Helmholtz, 1869;
see p.162 of this book). Helmholtz had ignored BL geometry, main
taining that free mobility is compatible only with Euclidean and
spherical geometry. Since the latter is automatically excluded by
Renouvier’s contention that it is analytically true that two straight
lines cannot enclose a space, Renouvier’s conclusion, though false, is
296 CHAPTER 4
Was Delboeuf aware of the full import of his words? Taken literally,
they are an invitation to physicists to discard Euclidean geometry and
to try out a space of variable curvature to represent physical space.
But Delboeuf does not pursue this idea any further. His declared aim
is to establish the absolute preeminence of the “old” over the “new”
geometry (which, as I said, he apparently knows only in the guise of
spherical and BL geometry). His lengthy argument for proving this
amounts in the end to the following: (i) Euclidean geometry is the sole
guarantee of the consistency of non-Euclidean geometries, (ii) The
geodetic arc, which, in non-Euclidean geometries, plays the same
fundamental role as the straight segment plays in ordinary geometry,
can only be defined in terms of the Euclidean straight line. Both
statements are false. The second one rests on the (mistaken) charac
terization of a geodetic arc as the shortest line joining its extremes
and on the classical definition of the length of an arc as the limit of a
sequence of lengths of straight segments. Riemann, as we saw, had
been able to discard this definition and to substitute for it another one
EMPIRICISM, APRIORISM, C ON VE N T I ON AL I S M 301
4.3 R U S S E L L ’S A P R I O R I S M O F 1897
4.3.1 The Transcendental Approach
An Essay on the Foundations o f Geometry (1897) was the first in the
long series of books published by Bertrand Russell (1873-1970). It is
based on the dissertation he submitted at the Fellowship Examination
of Trinity College, Cambridge in 1895, when he was twenty-two years
old. As it so often happens in philosophy, Russell’s ideas look very
attractive in their broad lines, but turn out to be quite disappointing
when worked out in detail. Russell very soon abandoned the philoso
phical position maintained in the book, which was not reissued until
1956, when the author, at 83, was a living classic, and everything
published under his name was rightly regarded as deserving atten
tion.1 The book reflects a much more accurate knowledge of the new
geometries than any of the writings we have discussed in Part 4.2. Its
historical Chapters I and II are still useful, and contain valuable
criticisms of the authors we have been studying.2 But our main
concern here is with Chapter III, on the axioms of projective and
metrical geometry, which, as we shall see, promises much more than
it is able to fulfil.
Like most of his contemporaries, Russell believes that the main
task of a philosophy of geometry consists in determining how much in
geometry is necessary, apodictic or a priori knowledge, i.e. knowledge
which under no circumstances can be other than it is, so that no
conceivable experience can ever clash with it. Russell characterizes a
priori knowledge in the best Kantian vein, as knowledge of the
conditions required by all experience or by a definite genus of
302 CHAPTER 4
In the light of Axiom I, each division (in the sense of Axiom II) of
space or of a part of space must consist in its partition into two or
more disjoint but otherwise indiscernible proper parts. Every part
into which division may divide a space is again a space liable to
division in exactly the same terms as any other space. In this axiom
system there are therefore no grounds for the idea that an infinite
sequence of divisions might converge to a definite result. The second
clause of Axiom II is nonsense. That clause, however, is meant to
provide the definition of the term point used in Axiom III. If we wish
to make some sense of Russell’s axioms we must take point as a
primitive term and postulate some relationship between it and the
other primitive of the system, namely, space. I suggest that we simply
regard space as the set of all points.10 The term continuous in Axiom
II cannot be viewed as primitive. Otherwise, we might just as well
substitute for it any other word or sound, since this is its only
occurrence in the system (it would thus make no difference to write,
for example, (II) Space is slithy and infinitely divisible). This term
must connect the system with other established mathematical
theories. What does it exactly mean? The following translation of the
first sentence of Axiom II into current mathematical language is no
doubt anachronistic but it is probably not too far from Russell’s
intended meaning: Space is continuous = Space is a topological space
every point o f which has a neighbourhood homeomorphic to Rn_1.
Here n (>1) is the number of points which, according to Axiom III,
suffice to “determine the whole of space” . Under this interpretation,
not every proper subset of space is a part of it in the sense of
EMPIRICISM, APRIORI SM, C O N V E N T I O N A L I S M 305
Geometry, logically indispensable to its existence, the axiom that Space must have a
finite integral number of Dimensions. For every relation required in the definition of a
point constitutes a dimension, and a fraction of a relation is meaningless. The number
of relations required must be finite, since an infinite number of dimensions would be
practically impossible to determine.24
and Q determine a unique line that joins them ((2)); all points in this
line are qualitatively equal; but “if one point be kept fixed, while the
other moves, there is obviously some change of relation” ; such
change must be a change of quantity. “If two points, therefore,
determine a unique figure, there must exist, for the distinction be
tween the various other points of this figure, a unique quantitative
relation between the two determining points. [...] This relation is
distance.” (Russell, FG, p.172). We find again the childish notion that
a figure determined by two points cannot consist of those two points
alone, but must be a line through them. It is clear that the moving
point must move along this line, otherwise its motion would introduce
a qualitative difference between it and the other points on the line
(namely, that it no longer belongs to the line). But if all points on the
line are qualitatively equal the motion of one of them along the line
cannot be defined, unless we presuppose some non-qualitative
difference between them. In Russell’s terminology, whatever is non-
qualitative is quantitative. The argument therefore begs the question:
unless we assume that the two points which determine the line sustain
a unique quantitative relation, we cannot make any sense of the
motion of one of these points against a fixed background of other
points which are qualitatively equal to it.
*1 wish to discuss finally Russell’s assertion that all the points of the
unique curve determined by P and Q are qualitatively equal. Until
now, we have understood that this curve is an arc from P to Q. On
this arc, P and Q, being the extremes, differ qualitatively from the
points which lie between them. But Russell assumes here a different
interpretation: the curve determined by P and Q is the straight line
through these points. This interpretation agrees with Projective
Axiom III, which says that two points determine a straight line.
Russell consistently identifies qualitative properties and relations in
space with projective properties and relations. The relation between P
and Q is projectively equivalent to the relation between P and any
other point R on the straight line PQ. Hence, according to Russell, Q
and R are qualitatively equal (at least as far as their relation to P is
concerned). On Russell’s assumptions, the argument is sound. But if
the unique curve determined by P and Q is the (full) straight line PQ,
we cannot claim that the length of this curve measures the distance
between P and Q (as Russell concluded in the proof of (2)).
314 CHAPTER 4
Those who assert that it is a priori evident that the sides of a triangle can be increased
in a given proportion without changing the angles, should also claim [. . . ] that it is
equally possible to change all the angles in a fixed proportion without changing the
sides. But this is, as we know, impossible in all geometries. If we admit the logically
relative nature of every magnitude, I cannot see why the argument should apply only to
linear dimensions and not to angles which are magnitudes as w ell.38
appeared obvious from the outset and probably nobody would have
chanced upon the idea of developing a non-Euclidean geometry.
4.3.7 Multidimensional Series
Soon after the publication of the Foundations of Geometry, Russell
took a very different approach to the problems of space and
geometry, based on the analysis of the “logical” ideas of series and
order. The new approach is briefly sketched toward the end of
Russell’s reply to Poincare,42 it provides the main support for the
criticism of the relationist theories of time and space which he read at
the Paris Congress of Philosophy in 1900,43 and it determines the
treatment of geometry in his great book, The Principles o f Mathema
tics (1903). In Russell’s usage, the idea of order covers both linear
and cyclical order.44 According to him, order can only arise in a set
with more than two elements. Order is generated in a set S if a
transitive antisymmetric binary relation is defined in S so that, for any
three distinct elements, jci, x2, *3 £ S, there is always a permutation a
of {1, 2, 3} such that stands in the said relation to x ^ ) and x ^ )
stands in it to x ^ . A self-sufficient simple series is an ordered set.
Russell speaks also of a simple series by correlation, which is a set
indexed by an ordered set, or, as we would rather say, the graph of a
mapping of an ordered set onto an arbitrary set. A self-sufficient
simple series is also described as “a series of one dimension”. The
ordered elements of such a series are called terms. A series of two
dimensions is a series of one dimension whose terms are series of one
dimension. Generally, a series of (n + 1) dimensions (n ^ 1) is a series
of one dimension whose terms are series of n dimensions.
Geometry - says R u ssell-m a y be considered as a pure a priori science, or as the study
of actual space. In the latter sense, I hold it to be an experimental science, to be
conducted by means of careful measurements. [. . . ] As a branch of pure mathematics,
Geometry is strictly deductive, indifferent to the choice of its premisses and to the
question whether there exist (in the strict sense) such entities as its premisses define.
Many different and even inconsistent sets of premisses lead to propositions which
would be called geometrical, but all such sets have a common element. This element is
wholly summed up by the statement that Geometry deals with series of more than one
dimension.45
foot: they are different derived units of the same metrical system. A
length stated in terms of one of them can be exactly expressed in
terms of the other just by multiplying it by a rational factor. Such is
not the case however if the yard is defined as the length of a metal
rod kept at some governmental institute, while the metre is given its
present official definition as so many wavelengths.3 The conversion
factor cannot then be expressed exactly. Moreover, it cannot be
determined to any desired degree of approximation, because there
are practical limits to the accuracy with which a wavelength can be
measured with a metal rod or vice versa. Since most lengths are
measured less accurately, you can indifferently use one or the other
unit to report them; to express, say, the height of a child, or the
distance flown by a plane from Heathrow to Kennedy. Measurements
based on an optical standard can attain, however, a greater precision
than those based on a rigid standard. As a consequence of this,
quantitative data which can be registered with instruments calibrated by
an optical standard cannot be registered with the same degree of
exactness with instruments calibrated by a rigid bar. Increase in
accuracy was indeed one of the reasons why the scientific community
discarded the original geodesic metre in 1889, adopting instead the
platinum-iridium standard kept at Breteuil, and in 1960 replaced the
latter by today’s optical metre. The newer unit was, in each case, defined
so as to make it equal to its immediate predecessor within the latter’s
range of accuracy. But its introduction opened up the possibility of
registering and reporting quantitative data which were, so to speak,
beyond the pale of the system of measurement based on the earlier
unit.
We turn now to our second example. Think of the theories of
gravitation propounded by Newton in 1689 and by Einstein in 1915.
All scientists and most philosophers will grant that the choice be
tween them is not merely a matter of convention. Though these two
conceptually very different theories agree within the bounds of
experimental error in nearly all their predictions, there are some cases
in which their discrepancy can be experimentally controlled. Thus,
for example, while gravitation, according to Newton’s theory, does
not affect the frequency of electromagnetic waves, Einstein’s theory
predicts that an electromagnetic signal sent from a point P where the
gravitational potential is lower to a point Q where it is higher will be
seen to have, upon reception at Q, a lower frequency than a signal
EMPIRI CI SM, APRIORISM, C O N V E N T I O N A L I S M 323
(I) R ela tiv e to an inertial sy ste m [determ ined b y three fr e e ly m o v in g p articles] any
additional free particle a lso m o v e s in a straight line.
EMPIRI CI SM, APRIORI SM, C O N V E N T I O N A L I S M 325
(II) Relative to an inertial time scale [determined by one freely moving particle]
every other free particle traverses in any inertial system equal distances in equal times.6
Beings with minds like ours, and having the same senses as we, but without previous
education, would receive from a suitably chosen external world impressions such that
they would be led to construct a geometry other than that of Euclid and to localize the
phenomena of that external world in a non-Euclidean space, or even in a space of four
dimensions. As for us, whose education has been accomplished by our actual world, if
we were suddenly transported into this new world, we should have no difficulty in
referring its phenomena to our Euclidean space. Conversely, if these beings were
transported into our environment, they would be led to relate our phenomena to
non-Euclidean space.14
more empirical than geometry. For are they not, considered in them
selves, just as exact? The last remark, however, suggests an inter
pretation of Poincare’s meaning which I think will remove our doubts
regarding both statements. As we saw in Section 4.4.1, Poincare
believed, like every practising scientist, that physical theories,
notwithstanding their mathematical exactness, can be compared with
and be corroborated or refuted by the inevitably imprecise data
supplied by observation and experiment. Why did he maintain that
geometry - and that means, as I take it, applied or physical geometry -
was exempt from such condition? Because geometry must mediate,
so to speak, between theories and data. The rough facts of obser
vation can be compared with the neat predictions of theory only if
they are described in terms akin to the latter. The geometrical
description of phenomena (strictly speaking, their kinematic, that is,
geodironometrical, description) provides the terms of comparison
required for the evaluation of physical theories. The translation of the
“Book of Nature” into “mathematical language” can be performed in
many different ways; as many as the different systems of geometry
which are rich enough for the purpose. The formulation of a scientific
theory must, of course, be adapted to suit the chosen system of
description, but its predictive contents will remain unaltered
throughout its “translations” . I have not been able to find a passage in
Poincare that directly bears witness to my interpretation, but I think
that his account of the manner how “geometrical space” -w hich is
none other than the space of mechanics and the rest of physics-is
constructed ratifies it indirectly.20 If my interpretation is accepted, we
at once see why physical geometry must be exact and cannot be
revised in the light of experience. Insofar as geometry itself supplies
the scheme according to which the data of experience must be
displayed if they are to make any scientific sense, it is impossible that
it should ever clash with them. Some geometries are, of course, more
manageable than others, because of their own structure and because
of the peculiar features of the empirical material which we try to
bring under their sway. Poincare believed that Euclidean geometry
was unexcelled on both counts.21
The main argument for Poincare’s rejection of empiricism was
mentioned earlier (at the end of Section 4.4.1): empirical information
has no bearing whatsoever on the structure of geometrical space. Or,
as he puts it:
332 CHAPTER 4
Experiments only teach us the relations of bodies to one another; none of them bears
or can bear on the relations of bodies with space, or on the mutual relations of the
different parts of space.22
If Lobachevsky’s geometry is true, the parallax of a very distant star will be finite; if
Riemann’s is true, it will be negative. These are results which seem within the reach of
experiment, and there have been hopes that astronomical observations might enable us
to decide between the three geometries. But in astronomy ‘straight line’ means simply
‘path of a ray of light’. If therefore negative parallaxes were found, or if it were
demonstrated that all parallaxes are superior to a certain limit, two courses would be
open to us; we might either renounce Euclidean geometry, or else modify the laws of
optics and suppose that light does not travel rigorously in a straight line.27
The sentence in italics shows that in 1891 Poincare knew very well
that physical geometry actually identifies its characteristic elements
with some reproducible physical prototypes. Had he taken
Minkowski’s standpoint he would probably have concluded that there
is no viable substitute for light-rays as a prototype of (spacetime)
straightness.28
A final anti-empiricist argument, based on the alleged possibility of
constructing bodies which “move according to the Lobachevskian
EMPIRI CI SM, APRIORI SM, C ON V E N T I ON A L I S M 335
That is why a method of measurement which would show the distance from Paris to
London to be equal to one metre would be inadequate for all practical uses and anyone
who proposed to adopt it by convention would lack common sense. But to say that the
distance from Paris to London is greater than one metre a b s o l u t e l y , independently of
every method of measurement, is neither true nor false; I find that it does not mean
anything.44
K, (i =1, 2) consists of eight thin steel rods OAJ,. . . , OA£, OB|, OB2,
permanently joined at O; it also includes some device regulating the
relative positions of the vertices Aj, BJ as K,- moves. Let P, Qi, Q2 be
three points marked, say, on a piece of wood, so that Aj, BJ and O can
be simultaneously placed upon P, Q2 and Qi, respectively (/ =
1 , . . . , 6; k = 1,2). We assume also that the point-pairs AJAj+i (1 ^ j *£
5) and AiAl can be brought into coincidence with PQ, (t = 1,2).
Poincar6 asserts that Ki “moves according to the Euclidean group or
at least that it does not move according to the Lobachevskian group” ,
while K2 “does not move according to the Euclidean group”, but might
move according to the BL group. Now K] can be built quite easily. K2on
the other hand, requires some ingenuity, but, Poincare says, any
mechanic can contrive it.45 Since Ki and K2 can exist simultaneously,
their properties cannot teach us anything about the true geometry of the
world.
There is a fallacy in the foregoing argument that it will be in
structive to expose. We must distinguish between the body K„
consisting of eight steel rods plus a regulating device, and the particle
system formed by the eight vertices Aj, B£ (i = 1 , 2; 1 «6; k =
1, 2). Let us denote the latter by KJ. Now it is only K 2, but not K2, that
can be held to move according to the BL group. The eight particles of
K 2 will preserve their BL distances because they are artfully joined
by eight or more bodies, each of which moves according to the
Euclidean group. We can imagine of course a BL polyhedron (a
double hexagonal pyramid) whose vertices are, at each moment, A),
B i and we can also make plastic models of it in its several positions.
But there is no known material which, when moulded into one of
these shapes, will take of itself the whole series of them, as it is
pushed around.
4.4.5 The Genesis of Geometry
Poincare’s deepest speculations on our subject are contained in his
long essay “On the foundations of geometry” , published in English in
The Monist of October 1898. It is a rather ambitious attempt to show
how our idea of geometrical space - as it occurs, say, in classical
mechanics - arises with experience, though certainly not from it.
Poincare’s entire construction rests upon an untenable theory of
perception, according to which all our knowledge of physical facts
can be ultimately traced to a variegated and changing aggregate of
EMPIRI CI SM, APRIORI SM, C O N V E N T I O N A L I S M 341
N ow we soon discover that any displacement whatever can always be divided into two,
three, or any number of parts whatever; I mean that we can always find another
displacement which, repeated two, three times will reproduce the given displacement.
This divisibility to infinity conducts us naturally to the notion of mathematical
continuity; yet things are not so simple as they appear at first sight.
We cannot prove this divisibility to infinity, directly. When a displacement is very
small, it is inappreciable for us. When two displacements differ very little, we cannot
EMPIRI CI SM, APRIORI SM, C ON V E N T I ON A L I S M 347
Such would be the formula of physical continuity. But such a formula is repugnant to
reason. It corresponds to none of the models which we carry about in us. We escape
the dilemma by an artifice; and for this physical continuity - or, if you prefer, for this
sensible continuity, which is presented in a form unacceptable to our m in d s-w e
substitute mathematical continuity. Severing our sensations from that something which
we call their cause, we assume that the something in question conforms to the model
which we carry about in us, and that our sensations deviate from it only in consequence
of their crudeness.61
d and beginning by the same state a, transform a into one and the
same state da. The properties of group action (p. 172) preclude 2
from acting on the space of sense awareness as it is really given. The
latter is a physical continuum. In order that 2 may act on it it must
be idealized into a proper mathematical continuum, which is not just
roughly, but exactly, homeomorphic with R". The idealized manifold
provides a basis for the realization of 2 , which however obviously
differs from the realization studied in ordinary geometry, for n is
much larger than 3. “How shall we escape the difficulty? Evidently by
replacing the group which is given us, together with its form and its
material, by another isomorphic group, the material of which is
simpler.”66 We shall not dwell on the allegedly psychological descrip
tions through which Poincare tries to show how this simplification is
suggested by human experience. They are lengthy and not altogether
clear.67 I propose instead a construction which I regard as essentially
equivalent to his. I proceed in two steps: I outline first a mathematical
argument which I deem necessary to support the mathematical
conclusions of his psychological inquiry; this will enable me to
explain then, simply and concisely, what I take to be the gist of his
conclusions. We assume, as before, that 2 is isomorphic with the
Euclidean group of motions. 2 is therefore a topological group
homeomorphic with R6 (p.174). We denote the subgroup of trans
lations by 2T. ST is a normal subgroup homeomorphic with R3. Let 2
be the set of all the rotative sub-groups of 2. Let be a given
element of 2. Then each 'M € 2 is equal, to tW0t~ \ for some t € J".
Moreover, the mapping tS€0t~l is a bijection of £T onto 2. Let
Z C 2 be open in 2 whenever f~ \Z ) is open in ST. With this topology,
2 is homeomorphic with 5", and hence with R3. Let d € 2. Let f d
denote the mapping eK ^d 'M € 2), where as on p.338, is the
left coset of by d. f d is a permutation or transformation of 2 (p.337). It
is not difficult to see that /: d *-*fd (d (. 2 ) is a realization of 2 in 2. We
see at once that in such realization every W £ 2 is its own stability
group.68 Consequently, / is similar to the standard realization of 2 in
2 1W described on p.338. / is therefore transitive. Moreover, / is faithful,
because no rotative subgroup of 2 contains a proper normal subgroup.
We have found thus a faithful, transitive and, moreover, immanent
realization of S' in a space homeomorphic with R3. 1 fear, however, that
many a reader will look down upon it as just another piece of algebraic
abracadabra, incapable of giving an insight into the origin of geometry.
EMPIRI CI SM, APRIORI SM, C O N V E N T I O N A L I S M 351
agree to regard S' as a topological space whose open sets are the
intersections of S' with the open sets of S. With this topology, S' is a
subspace of S. In particular, the empty set 0 is a subspace of every
space. We agree that 0 (and only 0) has inductive dimension -1 . A
topological space S is said to have inductive dimension n if n is the
least non-negative integer such that for each point x € S and each
open neighbourhood V of x there exists an open neighbourhood U of
x such that U C V and the boundary of U has inductive dimension
n - l.84 Having been defined exclusively by means of general topolo
gical concepts, inductive dimension is evidently a topological in
variant; in other words, any two topologically equivalent
(homeomorphic) spaces have the same inductive dimension. Inductive
dimension possesses two more features which one would naturally
expect any satisfactory concept of dimension number to share: (i) the
inductive dimension of R" and of every space locally homeomorphic
to Rn is n\ (ii) if S is an arbitrary topological space and S' is a
subspace of S, the inductive dimension of S' is less than, or equal to,
the inductive dimension of S. Unfortunately, as far as we can tell,
inductive dimension does not lend itself to the development of a rich
theory. That is probably the main reason why mathematicians have
proposed other concepts of dimension number for topological spaces.
The two most important of them -nam ely the so-called “large” in
ductive dimension defined in Note 84 and the covering dimension
characterized below - agree with inductive dimension on an important
class of spaces85 but not on all. These two concepts are, of course,
topological invariants and they both share the above mentioned
feature (i). They can also be shown to possess feature (ii) if S is
assumed to be a totally normal space.86
In 1911, commenting on Brouwer’s up to that time unpublished
proof that R" cannot be mapped homeomorphically onto Rn+P unless
Fig. 22.
358 CHAPTER 4
1. M A P P I N G S
Let S be a non-empty set. Let n denote the set of the first n natural
numbers. An ordered n-tuple or n-list of elements of S is a mapping
n-»S. We denote it by (a0, a i , a2, ■. . , where a; stands for the
value of the list at j. The collection of all such lists is designated by
S". An n-ary operation on S is a mapping S" -*■S. An algebraic
structure is an (r+ l)-list ( S , f u . . . , f r), where /, is an nr ary operation
on the set S (1 ^ r). S is the structure’s underlying set. One often
designates an algebraic structure by the name of its underlying set, or
vice versa.
359
360 APP EN D I X
the interior of W or Int W. W is open if, and only if, Int W = W. The
set {x | each neighbourhood of x contains an element of W} is called
the closure of W and is denoted by W. W is closed if, and only if,
W = W. Let Y be the complement of W (relative to S). The inter
section of W and Y is the boundary of W (and of Y).
If T and T' are two topologies on the same set S and T C T’, T is
weaker or coarser than T' (and T' is stronger or finer than T).
A topological space (S, T) is connected unless it is the union of two
non-empty disjoint open sets.
Let S be any set. A collection of sets whose union contains S is
called a cover of S. If K and K' are two covers of S and K' C K, K' is
a subcover of K. If T is a topology on S, T is obviously a cover of S.
A subcover of T is called an open cover of S. A topological space
(S, T) is compact if every open cover of S has a finite subcover.
Let (S, T) be a topological space. Let S' be any subset of S and let
T' designate the set {S' fl W | W £ T}. It can be easily verified that T' is
a topology on S', known as the topology induced on S' by T. (S', T') is
a topological subspace of (S, T).
Let (S, T) and (S', T') be topological spaces. A mapping /: S S' is
said to be open if / maps every open set of S onto an open set of S'; it
is said to be continuous if, for every open set Y of S' the set
{x | /(x) € Y} is an open set of S. An open and continuous bijective
mapping /: S-*S' is called an homeomorphism. Two topological
spaces are homeomorphic if there is a homeomorphism that maps the
underlying set of one into that of the other.
Further information on topological spaces can be gathered from
J.R. Munkres, Topology.4
4. DIFFERENTIABLE MANIFOLDS
( 1)
iLr = p . . X
dx‘ ■' ’
open interval (a, b) such that a < 0 < b. Let c(0) = P. A path fulfilling
these conditions will be called ‘a suitable path through P \ We now
define a ring of real valued functions which we shall denote by &k(P):
(i) for every neighbourhood U of P (in M), the underlying set of &k(P)
includes every C^-differentiable mapping /: U -*R; (ii) the ring
operations are given by the following rules:
(f + *)(Q) = /(Q) + S(Q),
/g(Q) = /(Q)g(Q), U
for every /, g t &k(P) and every Q € dom / ndom g. If a is any real
number, we let a denote the constant function M-»R; Q i-x* which
obviously belongs to ^*(P) for all P € M and every positive integer k.
t shall designate the identity mapping of R onto itself. The vector c(0)
tangent to path c at P = c(0) is the function ^ '(P ) -*■R which maps
each / € ^ '(P ) on the real number (d(f • c)ldt)(0). It is clear that for
every /, g € ^ '(P ) and every a, f} € R,
c(0)(a/ + fig) = ac(0)(/) + 0<*<OXg). (3)
The set of all vectors tangent at P to suitable paths through P is given
the structure of a real vector space by the following rules: For every
i>, w belonging to that set, every / € ^ ‘(P) and every a € R,
(V + w)(f) = v(f)+w(f),
(4)
(av)(f) = av(f).
Endowed with this structure, the said set is called the tangent space
of M at P and is denoted by TP(M).
Let x be a chart at P € M, where M is, as before, as n-manifold. We
denote by x' the ith ‘coordinate function’ p, • x, where p, is the ith
projection function of R" (p, assigns to each element (au a2, ■■., an)
of R" its ith term a,). Let
c,(u) = x . . . , un),
( 1 « i , / ^ n) (5)
Uj = uS] + x*(P).
(Here, 5J is the ‘Kronecker delta’ which equals 1 if i = j and equals 0
otherwise.) Equations (5) evidently define n suitable paths through P,
Ci, c2, . . . , c„. Now, for any / € ^ '(P )
( 6)
= f-<p,
(7)
(P*p(v) = v ■<pf.
Consequently, <p*P(t;), a vector tangent to M' at <p(P), maps a function
/ in ^ ‘(<p(P)) on the same real number assigned by v, a vector tangent
to M at P, to the function / • <p, which belongs to ^ ‘(P). If x is a chart
at P and y is a chart at <p(P),
(8)
(9)
366 APPENDIX
differentiable mapping.
A triple (A ,/, B), where A and B are manifolds and /: A-»B is a
surjective C°°-differentiable mapping is called a fibre bundle over B. In
particular, (A ,/, B) is a vector bundle over B if the fibre of / over
each b £ B is a vector space which is mapped onto Rn (for some fixed
positive integer n) by a linear isomorphism L b, and each b € B has an
open neighbourhood U such that U) is mapped diffeomorphically by
x*-*(f(x), L/(x)(jt)) onto U x R". It is clear that, under the stipulations of
the preceding paragraph, (TM, 7r, M) is a vector bundle over M. We call
it the tangent bundle of M. The cotangent bundle of M and the infinite
array of its tensor bundles of all types and orders are built analogously
from the diverse linear spaces attached to each point of M (p.365).
A Ck-section over B in the vector bundle (A ,/, B) is a Ck-differen-
tiable mapping g: B-»A such that f • g is the identity mapping b ^ b
of B onto itself. A Ck-section over the n-manifold M in the tangent
bundle (TM, tt, M) is called a Ck-vector field on M. A Ck-section over
M in its cotangent bundle is a Ck-covector field on M. More generally,
a Ck-section over M in one of its tensor bundles is a Ck-tensor field
APPENDIX 367
on M of the same type and order as its values. (Thus, a (j, lk)-tensor
field on M assigns to each P £ M a (/, k )-tensor on TP(M), i.e. a mixed
tensor of (j + &)th order, contravariant on the first j indices and
covariant on the last k indices, which maps suitable lists of cotangent
and tangent vectors at P into R.) A Ck-differentiable mapping of M
into R will be called a Ck-scalar field. If X is any Ck-field on M and
P € M, we usually write XP instead of X(P) for the value of X at P. If
V and F are, respectively, a C*-vector and a C*-covector field on M,
the Ck-scalar field (F . V) is given by the equation:
<F . V)P = <FP . VP), (PCM). (ID
Let M be an n-manifold. We denote by ^*(M) the ring of C^-scalar
fields on M (ring operations as in eqns (2)). We denote by ^ ( M ) the
collection of all C*-vector fields on M. T ^M ) is made into a module
over ^*(M) by the following rules: For every V, W in T*(M), every /
in &k(M) and every P in M,
(V + W)P = VP+ WP,
( 12)
(/V)P = / PVP.
Each V £ T*(M) determines a mapping V of &k(M) into &k~ \M). V
assigns to each Ck-scalar field / a C*_1-scalar field V/, whose value at
P £ M is given by
(V/)P = VP/. (13)
We usually write V for V. With this notation, if / and V are,
respectively, a scalar and a vector field on M, /V is a vector field on
M and V / is a scalar field on M. A similar notation is used for
covector and tensor fields.
Let U be an open subset of the n-manifold M. We naturally regard
U as an n-manifold on its own right, whose maximal atlas includes the
restriction to U fldomx of each chart x in the maximal atlas of M.
With this structure, U is termed an open submanifold on M. Let
/: U-*M be the canonical injection. It can be shown that for every
P £ U, /*P is a linear isomorphism of TP(U) onto TP(M). We may
therefore identify TP(U) with TP(M) for each P £ U and also TU, the
tangent bundle of the manifold U, with ir_1(U), the inverse image of
U C M by the projection 7r:TM-»M. We shall henceforth consider
vector and tensor fields defined on open submanifolds of a given
manifold. Thus, if x is a chart of M, d/dx‘ :Pi-»d/<?x‘|P is a C“-vector
368 APPENDIX
v - ? w 5?- <l4>
The scalar fields V jc' are called the components of V relative to chart
x.
Let Ui and U2 be open submanifolds of n-manifold M. If Vi is a
C“-vector field on U t and V2 is a C“-vector field on U2, the Lie
bracket [Vi, V2] of V| and V2 is the C°°-vector field on Ui flU2 defined
by
[V „V 2]/ = V1(V2/ ) - V 2(V ,/)) (15)
where / is any C°°-scalar field on Ui DU2.
A real n-parameter Lie group is an n-manifold endowed with a
group structure such that the group product is a C"-differentiable
mapping. It is often assumed that a Lie group is an analytic manifold
with analytic group product. Many important properties depend on
the further assumption that the natural topology of the underlying
manifold has a countable base; this is always true if the manifold is
connected. A complex Lie group is defined analogously. R" is a real
Lie group with the differentiable structure given on p.363 and the group
structure determined by vector addition.
Let G be a Lie group and M an n-manifold. G is said to act on M if
there is a C”-differentiable surjection F: G x M -> M such that for
every m € M and every g, h £ G, F(g, F(h, m)) = F(gh, m). F is called
the action of G on M. G acts transitively on M if for every m i,
m2 € M there is a g € G such that m2 = F(g, m i). G acts effectively on
M if F(g, m)= m for every m € M only if g is the neutral element of
G. If F is the action of G on M, there is associated with each g i G a
diffeomorphism f g: M-»M whose value at each m 6 M is given by:
fg(m) = F(g, m). (16)
Let G' be the set {fg | g € G}. G' is obviously a group with group
product given by the composition of mappings. If G acts effectively
on M, the mapping J: f g*-*g is plainly a group isomorphism of G' onto
G. If x is a chart of G, x • J is a chart of G'. With the differentiable
structure given by the collection of such charts, G' is a Lie group of
transformations of M.
An example will illustrate the foregoing notions. Let M be an
APPENDIX 369
( 21)
(24)
where the rj, are the components of the linear connection V relative
to z. Consequently, a vector field V is parallel along c if, and only if,
it is a vector field on c and Vc = 2 (- v‘ dldz‘\c, where the functions
v': R-»R are solutions of the system
(25)
The v‘ are uniquely determined for each choice of initial values v‘(t0)
(t0 € dom c). This implies that for each vector v tangent to M at c(t0)
(for each v € T c(,o)(M)) there is a unique equivalence class [V] of
vector fields on c, such that for any V € [V], VcW = v and V is parallel
along c. The mapping V c(,a)i-> V c(() (f € dom c) is a linear isomorphism
of T c(,o)(M) onto T C„)(M) called parallel transport along c from c(t0) to
c(t). If c has a tangent field which is parallel along c, c is said to be a
geodesic of V. Equation (25) implies that, if z is a chart at each point
of im c, c is a geodesic of M if, and only if,
0 1 • c) = 0, (26)
where the T}* are the components of the linear connection V relative
to jc. If T vanishes everywhere, V is said to be torsion-free. It follows
from eqns. (26) and (31) that a torsion-free linear connection is fully
determined if its geodesics are given.
Let M be an n-manifold. A semi-Riemannian C*-metric g on M is
a (0,2)-tensor field on M which is (i) symmetric and (ii) non-
degenerate. g assigns to every P € M a scalar product gP on TP(M). Let
A P P E N D IX 373
<32)
Since g is non-degenerate, the matrix of the gy is non-singular on
dom x. There exists therefore a unique (2,0)-tensor field on M, whose
components g '' relative to the chart x are given by
2 = Si (33)
C H A P T E R 1. B A C K G R O U N D
a sphere of its own that rotates about the North-South axis with the same angular
velocity as the fixed stars, and this rotation must be compensated by the rotation of an
additional sphere before it is again introduced with the next planet. Aristotle over
looked this fact and counted only 55 spheres (loc. cit., 1075all) . Seven of the 55 rotate
like the first (i.e., like the fixed stars). Since these seven rotations have none to counter
them, the Aristotelian moon must rise in the East seven times a day. Apparently
nobody noticed this before Norwood Russell Hanson (Constellations and Conjectures,
pp.66-80). Earlier scholars had pointed out that six spheres were dispensable, but none
said that, unless their motions were neutralized (by suppressing them or by adding new
spheres), the lower planets would be going too fast.
44 Aristotle, Eth . Nicom., 1094b23-25.
45 Aristotle, Metaph., 995*14-17.
46 Aristotle, Phys., 193b24f.
47 Aristotle, Phys., 193b35.
45 Aristotle, Metaphys ., 1025b2 8 - 1026*6; cf. Physica, 194*1-6.
49 This result can be very easily proved in the special case of a (fictitious) planet whose
trajectory lies on a plane through the centre of the earth. Let this plane be represented
by the com plex plane, with the centre of the earth at the origin. An arbitrary planetary
trajectory on that plane will be represented by a continuous periodic function z = f(t).
Uniform circular motion about the point designated by the com plex number k is
represented by the function z' - fc + a exp (ibt) (where a and b are real numbers
depending on the radius of the circle and the angular velocity of the motion).
Consequently, the position at time t of a body moving with nth degree uniform
geocentric epicyclical motion is given by z" = a{ exp(ibif) + • • • -I- an exp(ibnt). By a
suitable choice of n and of the 2 n constants a,, h,, the difference |z - z"| can be made to
remain less than any assigned positive real number.
50 Derek J. de S. Price (1959), p.210. Pre-Keplerian astronomy never attained the
optimal accuracy indicated on p.20; Price’s calculations show, however, that its failings
were due to a wrong choice of parameters rather than to the inadequacy of the
epicyclic models.
51 Averroes, Metaphysica , lib. 12, summae secundae cap.4, comm.45, quoted by Duhem,
SP, p.31.
52 Maimonides, The Guide of the Perplexed (transl. by Shlomo Pines), pp.325f.
33 Maimonides, GP, p.327.
54 John of Jandun, Acutissimae quaestiones in XII libros Metaphysicae, 12.20, quoted
by Duhem, SP, p.43.
55 Kepler, Epitome astronomiae copemicanae (1618); GW, Vol.VII, p.23.
56 Kepler, Ibid.; GW, Vol.VII, p.25.
37 Kepler, Letter to Johann Georg Brenger, October 4, 1607; GW, Vol.XVI, p.54.
38 Kepler, Letter to Herwart von Hohenburg, April 10, 1605; GW, Vol.XV, p.146.
Machina means here ‘structure’ or ‘edifice’; if it is rendered as ‘machine’, Kepler’s
programme sounds trivial - indeed, he ought then to prove first that the heavens are a
machine.
39 Kepler, Mysterium cosmographicum (1596); GW, Vol.I, p.26.
60 Kepler, Astronomia nova (1609); GW, Vol.III, p.241.
61 Kepler, Harmonices Mundi (1619); GW, Vol.VI, p.223.
378 NOTES TO CHAPTER 1
C H A P T E R 2. N O N - E U C L I D E A N G E O M E T R I E S
2.1 Parallels
1 H ea th , E E , V o l.I, p.155.
2 Im re T oth (1966/7) has c la sse d and a n a ly sed n u m erou s p a ssa g e s in th e C orpus
A r isto te lic u m allu sive to a p r e -E u c lid e a n d iscu ssio n o f p arallels and rela ted
m atters w h ich , h e claim s, esta b lish ed that: (i) th e e x iste n c e o f a p arallel to a
g iv en lin e through a p oin t o u tsid e it can b e p ro v ed b y c o n stru ctio n ; (ii) a lleg ed d irect
p r o o fs o f the u n iq u en ess o f the said parallel are q u estio n -b eg g in g ; (iii) if th e parallel to
a g iv e n lin e through a poin t o u tsid e it is n ot u n iq u e, th e th ree interior a n g les o f a
triangle are n ot eq u al to tr ; (iv) it can be p ro v ed b y reductio ad absurdum that th ey are
n ot greater than tt (A risto tle rep ea ted ly m en tio n s th e p r o o f o f this sta tem en t as a
fam iliar ex a m p le o f ap agogic p ro o f); (v) a ttem p ts to p ro v e that th ey are n o t le s s than tt
are in c o n c lu siv e (so m eth in g n ev er m en tio n ed b y A risto tle but w h ich sh o u ld b e o b v io u s
380 NOTES TO CHAPTER 2
from the fact that Euclid included Postulate 5 among the unproved assumptions of
geometry). If Toth is right, the debate on Postulate 5 from the 1st century B.C. to the
19th century originated in ignorance or lack of understanding of the geometrical
tradition leading up to Euclid. Although my pedestrian imagination is not always able to
follow Toth’s in its sometimes frenzied flight, I believe that his work deserves careful
study.
3 Proclus, Comm., ed. Fr., p. 191.
4 Proclus, Comm., ed. Fr., p.192.
5 Proclus, Comm., ed. Fr., pp.363-373.
6 Wallis, Operum mathematicarum volumen alterum (1693), p.678; Stackel and Engel,
TP, p.30.
7 Wallis, ibid., p.676; Stackel and Engel, TP, p.26.
8 Wallis, ibid., p.676; Stackel and Engel, TP, pp.26f.
9 Euclides ab omni naevo vindicates (Euclid freed from all flaw), Milan 1733.
10 Kliigel (1763), p.16; quoted in Stackel and Engel, TP, p.140.
11 Stackel and Engel, TP, p.162.
12 Lambert’s quadrilateral is identical with one of the two congruent parts into which a
Saccheri quadrilateral is divided by the perpendicular bisector of its base. Lambert’s
three hypotheses as numbered by him are respectively equivalent to Saccheri’s as I
have numbered them.
13 Stackel and Engel, TP, p.202.
14 Lambert writes: “Ich mochte daraus fast den Schluss machen, die dritte Hypothese
komme bei einer imaginaren Kugelflache vor.” (Stackel and Engel, TP, p.203).
15 Stackel and Engel, TP, p.200. A similar remark by Gauss is quoted below (Section 2.1.5,
P-55).
16 Dehn (1900). See Section 3.2.9.
17 Legendre’s investigations on the theory of parallels are brought together in his
“Reflexions sur diff£rentes manieres de d6 montrer la theorie des paralleles ou le
theoreme sur la somme des trois angles du triangle.’’ (1833). Two of his attempted
proofs of Postulate 5 are reported in Bonola, NEG, pp.55-60. More on Legendre
below, Section 3.2.4. Bolzano’s attempt at deriving Postulate 5 from the essential
properties of the straight line is also discussed in Section 3.2.4.
18 Stackel and Engel, TP, p.262.
19 Reprinted in Gauss, WW, Vol. 8 , pp.l80f. English translation in Bonola, NEG, p.76.
20 Gauss to Gerling, March 16, 1819. (Gauss, WW, Vol. 8 , p.181.)
21 Taurinus, Theorie der Parallellinien , p.8 6 ; Stackel and Engel, TP, p.258.
22 Norman Daniels (1972) claims that the Scottish philosopher Thomas Reid (1710-
1796) discovered a non-Euclidean geometry in the 1760’s, having published his results
in his Inquiry into the Human Mind (1764). All I find in the passages mentioned by
Daniels (see Reid, PW, Vol.I, pp. 142-153) is a discussion of the geometric structure of
the visual field, which, according to Reid, is spherical. Reid’s approach is certainly
original, but he has not made a contribution to geometry. The geometry of the sphere
which he sees embodied in his “geometry of visibles’’ (Reid, PW, Vol.I, pp.l47ff.) was
familiar to Euclid and was used profusely in ancient astronomy. There is in Reid’s book
no suggestion that the characteristic properties of the sphere (e.g. that straightest lines
meet twice) might also be realized in a three-dimensional space. Reid cannot even be
NOTES TO CHAPTER 2 381
2.2 Manifolds
1 “In pulcherrimo geometriae corpore duo sunt naevi”, viz. the theory of parallels and
the theory of proportions. (Savile, Praelectiones, p.140.)
2 Bessel to Gauss, February 10, 1829 (Gauss, WW, V ol. 8 , p.201).
3 The above definition of path presupposes, in fact, that space is Euclidean. Unlimited
differentiability is required in order to avoid qualifications that would distract the
reader from the conceptual issues which are our concern. A more general charac
terization of paths is given in the Appendix, p.363. A mapping is said to be injective if it
assigns distinct values to distinct arguments. (See p.359.)
4 The ith projection function on Rn maps each ordered n-tuple o f real numbers on its
ith member.
5 Such a convention may be set as follows: cx(t) = <1>(t, b) and c2(t ) = 4>(a, t) describe
curves on $ (£ ) meeting at <I>(u, h) = P; cJ(a) = Hi and c2(b) = H 2 are the tangential
images of these curves at P. We choose Q, as the normal image of <!>(£) at P if, and
only if, whenever the right thumb and right forefinger point, respectively, from O
toward Hi and from O toward H 2, the right middle finger can be made to point toward
Q,. This convention clearly determines our choice of Q, at each point P.
6 That is, by the area of n(£) if the “part of a curved surface” is denoted by 4>(f).
7 Gauss, GI, pp.9f. (§ 6 ).
8 A rational reconstruction of Gauss’ definition of the (local) curvature of a surface is
furnished by the theory of differential manifolds. (See Appendix, pp.36Iff.) Our surface
<!>(£) and the unit sphere can both be conceived as 2 -manifolds, n is then a differenti
able mapping of the former into the latter. Consequently, n determines a mapping n+ of
the tangent bundle over d>(f) into the tangent bundle over the unit sphere such that, if c
is any path in d>(£) and Ci denotes the path n • c, n*c - d\. n also determines a mapping
n* of each bundle of (0 , k) tensor fields on the unit sphere into the bundle of
(0, k)-tensor fields on $ (£ ), such that, for any (0, /c)-tensor field T on the former
and any fc-tuple (i?i,. . . , vk) of vector fields on the latter, n*T(vu . . . , vk) =
T(n^Vu . . . , n+vk). The area of a region U of a 2-manifold M is measured by integrating
over U a 2-differential form (i.e. an alternating (0 , 2)-tensor field) on M, called the
surface element of M. Let <p and a> be the surface elements of 4>(£) and of the unit
sphere, respectively. Then, at any point P in d>(£), <Pp and (n*a>)P both belong to the
one-dimensional vector space of alternating (0 , 2 )-tensors on the tangent space of 4>(f)
at P, so that their quotient exists and is a real number, just as in Gauss’ definition. The
mapping P ► “►<*>?/(n*<y)P is then a scalar field on 4>(f), the Gaussian curvature of the
surface. A more straightforward vindication of G auss’ definition can be achieved within
the theory of formal differential geometry currently being developed by Kock, Wraith
and Reyes. (On this theory, see, for example, A. Kock and G.E. Reyes (1977).)
9 An arc s in d>(£) joining two points A and B is said to be an arc of shortest length or a
shortest arc if its length is less than or equal to that of any other arc joining A and B which
is contained in a (suitably narrow) strip of <&(£) covering s.
10 Surfaces of constant negative curvature had been studied by Ferdinand Minding
(1839, 1840) in two papers published in the same journal (Crelle’s /. fur die reine u.
NOTES TO CHAPTER 2 383
40 The right-hand side of eqn. (4) on p.80 will read like the positive square root of the
right-hand side of eqn. (8 ) on p.95 if we make a few notational adjustments. Take
n - 2. Substitute 4 >_1 for x. (Remember that if is a surface, 4>"! is a chart on it.) c
in eqn. (4) stands for x • c in eqn. (8 ). Put E = g n • c, F = g n • c = g2i • c, G = gn • c. In
eqn. (4) the argument t was omitted.
41 A note on notation: If / and g are covector fields on a manifold M, / ® g is the
(0, 2)-tensor field which assigns to every pair (i>, w) of vector fields on M the scalar field
(J ® g)(v, w) = f(v)g(w). The ‘form’ f r^g introduced on p.99 is the alternating
(0 , 2 )-tensor field defined by / a g = ( 1/ 2 ) { f ® g - g ® /).
42 On geodesics, see Note 15 and Appendix, pp. 371-374.
43 In other words, the restriction of ExpP to the said neighbourhood of 0 in TP(M) is a
differentiable injection whose inverse is likewise differentiable.
44 Riemann, H, p.14.
45 Riemann, H, p.16.
46 That is, as differentiable mappings assigning to each point Q in M a linear function
on T q ( M ) . See Appendix, p.366.
47 H ence, if X € TP(M), dx'(X) denotes the value at X of the linear function d x‘(P).
48 Spivak, CIDG, Vol.II, proposition 4 B-4.
49 We introduce a factor of - 3 instead of Riemann’s - 3 / 4 because we divide F(X, Y)
by the area of the parallelogram formed by X and Y, not by the area of a triangle.
50 There are counterexamples to this claim, even for n - 2. But R.S. Kulkami (1970)
has shown that if n > 3, Riemann’s claim is true, except in a special family of cases. To
be more precise, let us say that two R-manifolds M and M' are isocurved if there exists
a diffeomorphism such that, for every P € M and every two-dimensional
subspace a of TP(M), k(a ) = k (/* P(a)). Kulkami’s theorem states that if M and M' are
isocurved manifolds of dimension n > 3 they are globally isometric, unless it happens
that they are diffeomorphic not globally isometric manifolds of the same constant
curvature.
51 Riemann, H, p.19. Riemann adds: “Consequently in the manifolds with constant
curvature, figures may be placed in any way we ch oose.’’
52 Riemann’s short discussion of surfaces of positive constant curvature in Section
2.2.5 plus an important remark in Section 3.1.2 regarding the finiteness of three-
dimensional manifolds of constant positive curvature are, as far as I can see, the only
justification for giving the name “Riemann geometry’’ to the intrinsic geometry of a
sphere. Riemann geometry, in this sense, should not be confused with Riemannian
geometry, or the general theory of R-manifolds. Since the latter, due to its importance
and generality, is a better candidate for preserving and honouring the name of its
creator, I recommend against the use of “Riemann geom etry’’ for conveying the
former sense. We do better just to speak of spherical geometry, even if such geometry
is true also of manifolds that are not spheres.
53 Submitted to the Paris Academy in 1861, to compete for a prize on a question
concerning heat conduction. In this work, Riemann arrives at a result which, according to
M. Spivak, “amounts to another invariant definition of the curvature tensor’’ (Spivak,
CIDG, Vol.II, p.4D-24). See Riemann, WW, pp.402f.
54 The use of the connection symbol V is explained in the Appendix, pp.369ff. [X, Y] is
the Lie bracket of X and Y (Appendix, p.368). If X and Y are vector fields on M
(Appendix, p.366) pt(X, Y) denotes the function P ^ / ap(Xp, Y P).
386 NOTES TO CHAPTER 2
55 Clifford, “On the space-theory of matter” , abstract of a paper read to the Cambridge
Philosophical Society on February 21, 1870.
56 According to the definition we have given, an n-dimensional differentiable manifold
M cannot contain a boundary. For let P be any point of M and x a chart defined on a
neighbourhood U of P. Then jc(U) is an open subset of R", so that x(P) has an open
neighbourhood V which is entirely contained in x(U ). Since P lies in x~l(V ) and x - 1(V)
is open in M, P cannot be a boundary point of M. For a definition of a manifold with
boundary see Miinkres, EDT, pp.3, 43-57.
57 Riemann, H, p.22. See Einstein (1917) in Lorentz et a i y R, pp.l30ff.
58 Riemann, H, p.22.
59 Riemann, H, p.23.
60 Riemann, WW, p.508.
61 Russell, FG, pp.62f.
62 Kant, KrV , A 76/B 102, A 100, B 134.
63 Herbart’s psychological theory of space is presented in his Psychologie als Wissen-
schaft neu gegrundet auf Erfahrung, Metaphysik und Mathematik (1824/25), Herbart,
WW, Vol.VI, pp. 114-150. See also Herbart, WW, Vol.V, pp.480-514. Space is
described by Herbart as a sort of sediment left behind by the flux of our ideas of sense
and providing a neutral background, somewhat like a river bed, wherein they flow. See
especially Herbart, WW, Vol.VI, p.134.
64 Herbart, WW, Vol.IV, p.159.
65 Herbart, WW, Vol.IV, p.171.
66 Herbart, WW, Vol.IV, p.193.
67 Compare Riemann, H, p.10, with Grassmann, WW 1.1, pp.47-49, 299.
68 Grassmann, WW, 1.1, p.297.
69 Grassmann, WW, 1.1, p.297.
70 Grassmann, WW, 1.1, p.325.
71 In an appendix to the second edition of the Ausdehnungslehre of 1844 (1878),
Grassmann explains that non-Euclidean geometries easily fit in his general theory of
extension because three-dimensional non-Euclidean spaces may be regarded as hyper
surfaces of a “region” of higher level, i.e. of a vector space of more than three
dimensions (Grassmann, WW, 1.1, pp.293f.). This is certainly true. Moreover every
connected n -dimensional differentiable manifold can be imbedded into R2,1+1 (Whitney
(1936)). H owever this does not imply that the theory of differentiable manifolds can be
absorbed by the theory of vector sp a c e s -a s Grassmann seems to b e lie v e -fo r a
manifold can always be regarded as a submanifold but is not usually a subspace of a
vector space of higher dimension.
through P that lie within the double angle a U a ' . Then the set {qa | a U a ' is a double
angle at P} is a base of a topology on q (regarded as the set of its meets). This is the
topology induced on q by a flat pencil through a point outside it. It does not depend on the
choice o f that point (P in the preceding discussion). (On topologies, see Appendix,
pp.360f.)
3 Relative to the topology defined in N ote 2.
4 Appendix, p.361.
5 Appendix, p.361.
6 Take the weakest topology that makes every collineation into a homeomorphism. (A
collineation is a bijective mapping of projective space onto itself that maps collinear
points onto collinear points.)
7 Let Ui , . . . , an be elements of a vector space over field F (or, more generally, of a
module over ring F). a i t . . . , an are said to be linearly dependent if there are elements
fci,. . . , kn in F, not all equal to zero, such that kxax+ • • • + knan = 0 . (In the text above,
regard R 3 as a vector space over R.)
8 A more satisfactory approach to complex projective geometry, which does not
depend, like ours, on numerical representations, was proposed by von Staudt in 1856.
See Staudt, BGL, pp.76ff.; Liiroth (1875). The significance of von Staudt’s proposal in
the history of mathematics is clearly brought out by Freudenthal (1974).
8a Solutions exist unless flu, its cofactor A n and the determinant \ai}\ all have the same
sign. If that condition is fulfilled the polarity is elliptic.
9 As an immediate corollary of the stated characteristic of conics let us mention that a
collineation always maps a conic onto a conic. Let / be a collineation and £ a conic
which is the locus of self-conjugate points under a polarity g ; then fgf~l is also a
polarity and its locus of self-conjugate points is f(£).
10 That is, if none of them can be represented by a set of complex homogeneous
coordinates (ay + b}i) such that b} = 0 for j = 1 ,2 , 3.
11 An interesting exception occurs if the left-hand side of the linear equation is a factor
of the left-hand side of the quadratic equation. In that case, the conic represented by
the latter is a degenerate one com posed of two lines, one of which is the one given by
the linear equation. Trivial exceptions occur if any of the left-hand sides is identically
zero.
12 But of course these “circles” in 0 >2 are not the loci of points equidistant from a given
point. We have, as yet, no concept of distance in our projective plane.
13 Lie, VCG, pp.l30f., 216f.
14 Lie, VCG, pp.l31ff. By ‘reducible’ I mean that its value can be calculated from the
value of the cross-ratio (thus, the square or cube of the cross-ratio are reducible to the
cross-ratio, etc.).
15 To ensure that fc is defined on every point-pair (P, Q) we index the points (PQ /f)i so
that (PQ l£)x* P and (P Q /f )2 * Q.
16 Since Pi, P2, P 3 are collinear their joins meet £ at the same two points, say (p,), ( qk).
We may therefore calculate fc(Ph Ph) (/, h = 1, 2, 3) using eqn. (5) of Section 2.3.5 for
the cross-ratio. Let P; = (fcyp, + m #,). Then fc{Py, Pk) = mjkhlkjmh. Consequently
k + V k 2- jc2
y —k log
X
-V F ^ 7 .
NOTES TO CHAPTER 2 389
given points or lines is presented in his Geometrie der Lage (1847), pp.43ff. The
assignment of homogeneous coordinates to space by means of this construction was
introduced in von Staudt’s Beitrdge zur Geometrie der Lage (1856), pp.26Iff.
54 Zeuthen’s proof, contained in a letter addressed to Klein after the publication of
Klein (1873), is reproduced in Klein (1874), §2. A similar proof was sent to Klein by
Liiroth. Let us recall that, if p, q, r and s are four different lines of a flat pencil, p and q
are said to separate r and s if p is contained in one of the two pairs of vertically
opposite angles formed by r and s, and q is contained in the other. Cf. p.411, n.44.
55 The lines of net ( uvw ) fall into two classes: those which have been assigned a
rational number whose absolute value is less than 7r, and those which have been
assigned a rational number whose absolute value is greater than tt. N o pair of lines in
the first class is separated by a pair o f lines in the second class. There is a unique line in
pencil X which, together with the line labelled °o, separates the remaining lines o f one class
from the remaining lines o f the other class. This line cannot be associated with a rational
number and consequently does not belong to (uvw).
56 Take, for example the last case we mentioned. Point O is mapped on [0 ,0 ,1 ]; point
Y on [s, t, 1]. Every triple of the class [s, f, 0] is a solution of the equation
*1 S 0
*2 t 0
*3 1 1
Consequently, this class lies on the join of [0 , 0 , 1 ] and [$, f, 1 ]. (See p.124.)
57 See, in particular, his Evanston lecture of 1893, “On the mathematical character of
space intuition and the relation of pure mathematics to the applied sciences” (in
English in Klein, GA, Vol.II, pp.225-231). See also the following: his 1873 lecture
“Ueber den allgemeinen Funktionsbegriff”, in Klein, GA, Vol.II, pp.214-224; Klein
(1890), pp.570-572; (1897), pp.584-587; (1902) pp.146-148; EP, p.94, etc.
58 Klein (1897), p.593. Projective intuition is mentioned in Klein (1890), p.570f.; it is
contrasted with “ordinary intuition” in Klein, EP, p.75.
59 Klein, GA, Vol.II, p.250.
60 Klein, GA, Vol.II, p.226.
61 Apparently Klein believed that intuition did more than just suggest this. He writes:
“The straight line of ordinary intuition (der gewohnlichen Anschauung) contains Qnly
one infinitely distant point. We can approach this point from either side without ever
reaching it.” (Klein (1871), p.593).
62 Clifford explained his discovery in a lecture that Klein attended in 1873. The text of
this lecture was never published but Klein reconstructed its mathematical contents in
Part I of his paper of 1890. A very clear account of Clifford’s discovery and its
implications is given in Klein, VNG, pp.233-238, 241-249, 254-270. There is a short
description in Klein (1897), pp.591-593. The matter has been further dealt with by
Killing (1891) and Hopf (1926).
63 The polar of a is the line assigned to a by the involutory correlation which defines £.
(Correlations in 3-space map points on planes, lines on lines.)
64 Let JKLM be such a quadrilateral, with JK opposite and equal to LM, JM opposite
and equal to LK. The geodesic LJ (on 2 ) divides JKLM into two congruent geodetic
NOTES TO CHAPTER 3 391
triangles, whose six angles add up to 2 it. Consequently these triangles have neither an
excess not a defect. (See p.74.)
65 W olf, Spaces of Constant Curvature, p.123. The classification of all compact 3-
dimensional Euclidean space-forms was given by W. Hantzsche and H. Wendt (1935).
66 Klein (1897), p.595.
67 Klein, VNG, p.270.
C H A P T E R 3. F O U N D A T I O N S
means all infinitesimal rotations about any arbitrarily fixed line element through that
point.
14 Helmholtz (1868), G, p.57. For a reconstruction of H elmholtz’s proof, which intro
duces clarity and precision into it, while remaining faithful to its spirit, see Weyl, MAR,
pp.29-43.
15 Helmholtz (1866), p.201.
16 Kant, A*., Vol.IV, p.312.
17 Helmholtz, G, p.71f.
18 Helmholtz, G, p.8; PSL, p.227.
19 Helmholtz, G, p.25; PSL, p.241. Cf. Poincare, FS, pp.416f.
20 Helmholtz, G, p.28; PSL, p.244.
21 Helmholtz, G, p.81.
22 Helmholtz (1878), p.213; G, p.62.
23 Schlick in Helmholtz, SE, p.162.
24 “As all our means of sense-perception extend only to space of three dimensions, and
a fourth is not merely a modification of what we have, but something perfectly new, we
find ourselves by reason of our bodily organization quite unable to represent a fourth
dim ension.’’ Helmholtz, G, p.28; PSL, p.244.
25 I cannot accept, however, Schlick’s alternative suggestion that Helmholtz’s “general
form of spatial intuition’’ exhibits space as a three-dimensional continuous manifold
wherein quantitative comparisons are possible (Schlick in Helmholtz, SE, p.161). The
possibility of quantitative comparisons is not a consequence of the general form of
extendedness, since it presupposes the existence of rigid bodies. A liquid mathemati
cian living in a liquid world cannot, according to Helmholtz, develop a geometry, but he
could very well share our general form of spatial intuition.
26 Helmholtz, G, p.29; PSL, p.244.
27 Helmholtz, G, p.17; PSL, p.234. Formerly, he had tried to make this intuitively clear
in the light of the two-dimensional case. In that context, he introduces a two-
dimensional country inhabited by two-dimensional rational beings who “have not the
power of perceiving anything outside’’ the surface they live on, but have, upon it,
“perceptions similar to ours’’! (Helmholtz, G, p.8; PSL, p.227). These
beings will supposedly develop the intrinsic geometry of their country. But if they
happen to live, say, upon an egg, they will be unable to build transportable rigid figures
and, consequently, according to Helmholtz, they will be incapable of defining a
geometry. Though Helmholtz introduces the example of an egg, he does not draw the
latter consequence; had he drawn it, it would probably have shocked him out of his
operationist bias, for he certainly knew that one can define with Gaussian methods the
intrinsic geometry of an egg-like surface. Let us remark that the country of Flatlapd is
used by Helmholtz merely as a didactic prop, and ought not to be taken too seriously.
Greater significance must be attributed to the Helmholtzian country we mentioned on
p.165 viz. Through-the-Convex-Looking-Glass.
28 Helmholtz (1866), p.197; cf. Helmholtz (1868), G, pp.32f.
29 Helmholtz (G, pp.29f.; PSL, p.245) says “transcendental in Kant’s sen se’’. But the
sense intended is certainly not Kant’s, for Helmholtz says in the same sentence that
experience “need not exactly correspond therewith’’.
30 Helmholtz, G, p.30; the English version in PSL, p.245 is very free.
NOTES TO CHAPTER 3 393
31 The expression is not really Kantian, but a shibboleth of the disreputable philoso
phies of Fichte and Schelling.
32 Helmholtz, G, p.29; PSL, pp.244f.
33 Poincar6 (1898), p.40, acknowledges that he owes much to Helmholtz.
34 Helmholtz, G, p.29; PSL, p.245 (quoted in Section 3.1.2, p.160).
35 Helmholtz, G, p.30; PSL, p.245.
36 Lie (1890), p.359 n.2.
37 That L ie’s continuous groups are connected follows from the definition in Lie, TT,
Vol.I, p.3.
38 It is clear that L* is surjective, since / is surjective. We prove that L* is injective:
Lg(m) = Lg(n)-+f(g , m) = /(g , n)-+ f(g~ \ /(g , m)) = f ( g ' \ /(g , /t))-»/(<?, m) = /(e , n)-»
m - n. (Read *-»’ as ‘implies that’.)
39 This is now the usual definition of transitive action. But Lie’s own definition, in
modern terms, would read as follows: G acts transitively on M if there is an open
subset M ' C M such that, for every pair x, y in M', there is a g € G with g x - y (see Lie,
TT, Vol.I, p.212).
40 This is said of R3 in Lie (1890), p.20. See also Lie, TT, Vol.I, p.572.
41 Lie, TT, Vol.III, pp.385, 387, etc. Observe that by fixing once and for all the nature
of the space R„ on which his groups are allowed to act, Lie was able to ignore the
complications that arise from considering the action of groups on manifolds of diverse
global topologies.
42 This statement of the problem is paraphrased from Lie, TT, Vol.III, p.397. Lie says,
however, that the sought for properties should distinguish the three groups mentioned
above from “all other possible groups of motions of a number-manifold (Zahlen-
mannigfaltigkeit)” . In the light of L ie’s actual performance I have judged it appropriate
to broaden “groups of m otions’’ to “groups of analytic transformations” , while nar
rowing down “eine Zahlen mannigfaltigkeit” to It is indeed not easy to say what Lie
(or Engel) understands by “eine Zahlenmannigfaltigkeit”. In TT, Vol.III, p.394, this
expression is introduced by saying that, in his lecture of 1854, “Riemann let all his
research be guided by the proposition that space is a Zahlenmannigfaltigkeit, so that the
points of space can be determined by coordinates.” It might seem , therefore, that a
Zahlenmannigfaltigkeit is for Lie any n-dimensional, real or com plex, presumably
analytic, differentiable manifold. This is somehow confirmed by the fact that Lie calls
any regular submanifold of R„ “eine Mannigfaltigkeit des R„” (Lie, TT, Vol.I, p.133).
On the other hand, there is no doubt that “the space R„” discussed in the subsequent
chapters on H elmholtz’s problem and the foundations of geometry is iPc, or an
open subset of either. This is very clearly brought out in Lie’s own statement of some of
his results in Lie (1890), pp.303f., 308, 312. Unfortunately, the text of TT, edited by
Engel, is not so careful on such conceptual matters.
43 The first of these, in Lie, TT, Vol.III, pp.452f., is L ie’s “translation” of H elmholtz’s
axioms into group-theoretical language. Lie criticizes Helmholtz’s own deductions from
his axioms because Helmholtz “nimmt [. . . ] stillschweigend und ohne ein Wort der
Begrlindung an, dass alle seine Axiome, die er fiber die nach Festhaltung eines Punktes
noch moglichen Bewegungen aufgestellt hat, auch auf die Punkte an wend bar bleiben,
die dem festen Punkte unendlich benachbart sind” . (Lie, TT, Vol.III, p.454). This
would imply that, if G is a group which fulfils the axioms in TT, Vol.III, pp.454f., and if
394 NOTES TO CHAPTER 3
{ d
dxl
~ * 2 ’2T
dXl
7 + C (
\
x X“ 7+
dXl
Here, c is any real number 9*0 (Lie (1890), p.290). This group is excluded by the
two-dimensional analogon of the axiom of monodromy, which is probably the reason
NOTES TO CHAPTER 3 395
why Helmholtz thought this axiom necessary also in the general n -dimensional case.
(See Helmholtz, G, p.21).
48 Poincar6 (1887).
49 Axiom D does the job Poincar6 expects it to do only if we assume that the distance
between two points P, Q is the same as the length of the segment PQ. This assumption
is implicit in Poincare’s own statement of feature (a) of the geometry defined by the
one-sheet hyperboloid.
30 Killing (1892), p.129.
51 Killing (1892), p.153. I understand he means transitive n-dimensional connected Lie
groups.
52 Killing (1892), p.131.
53 Killing (1892), p.137. Compare Killing’s treatment of dimension number in EGG,
Vol.I (1893), Absch.3, especially pp.238-265.
54 Killing (1892), p.167.
55 One-parameter groups and their generators are defined in the Appendix, p.369.
56 In other words, if there is a g in G which maps the A, on points arbitrarily near (AJ)
(i = 1, 2,3), there is an h in G which maps (A,) on (A;).
57 Essential steps in the argument show that: (i) If K is a true circle, x(K ) is a closed
Jordan curve, (ii) The rotations about a point P constitute a group isomorphic to the
group of rotations about a point in R2. (iii) If P, Q are two points of 7r, there is a unique
point M € 7T, the midpoint of PQ, such that there exists a rotation g about M with
gP = Q and g Q = P. (g is called a half-rotation.) (iv) We can choose two points O, E in
7r with the following property: let m denote the set generated from O and E by the
operations of taking midpoints and performing half-rotations; let m denote the set
{P | P is the limit of a convergent sequence of points in x(m)}; then m is the range of an
injective continuous mapping R - » R 2; if g € G, gx~\m ) is called a true line.
58 Hilbert, GG, p.l81n.
3.2 Axiomatics
1 See also Descartes, AT, Vol.VII, pp.160-170; Vol.IX, pp.124-132.
2 “Les geometres supposent toutes ces propositions sans le dire.” I have not seen a
copy of Lamy’s book; I take this quotation from P. Rossier, “Les axiomes de la
geometrie et leur histoire” (1967), pp.379f., who obtained it from the 6th edition (1734).
3 See G. Goe (1962).
4 See the passage quoted on p.49.
5 Pliicker, Neue Geometrie des Raumes (1868-69); see Nagel (1939), pp. 188-192.
6 Strictly speaking, I demand that the set of axioms be computable. For a definition of
computable set see, e.g., M. Davis, Computability and Unsolvability. Informally, we
may say that a set S is computable if an ordinary computer with unlimited memory can
be programmed to determine whether any given object belongs to S or not.
7 Let me recall that Leon Henkin, in his justly celebrated paper on “Completeness in
the theory of types” (1950), uses a different concept of interpretation (and hence, a
different notion of logical consequence). In Henkin’s semantics, object variables range
over an arbitrary set of entities, as above, but first-order n-ary predicate variables range
396 NOTES TO CHAPTER 3
over an arbitrary class of n-ary relations between the entities of that set. The latter
are freely stipulated in each interpretation. Henkin’s semantics underlies Abraham
Robinson’s “non-standard analysis’’, but is otherwise foreign to mathematical English
and its formalizations, as they are normally understood.
8 See Oswald Veblen (1911), pp.5f.
9 The informal characterization of the structural equivalence of models of first-order
theories given above can be made more precise with the aid of a few auxiliary
concepts. We consider first a theory T in which all variables are of one kind. Let T*,
/ 2, D,, D2 and / be as above. Let Tg be the set {s | S is a basic sentence and S belongs to
T*}. Let FTg be the collection of all sentence schem es (“formulae”) obtainable by
replacing in each sentence of Tg none or one or more than one of the object constants
that occur in it by free variables, in every conceivable combination. A variant of 7, is
an interpretation JJ which differs from 7, at most in the denotations assigned to one or
more object variables. For each variable x, if the variant 7J of I\ assigns to x the
denotation m € Dj, the corresponding variant of 72 (relative to bijection / ) assigns to x
the denotation /(m ). The two models of T, Ix and / 2, are structurally equivalent or
isomorphic if, and only if, for every formula (p in FTg, <p is satisfied in a variant 7 J of /,
if, and only if, it is satisfied in the corresponding variant of 72. This characterization can
be easily extended to theories with object variables of several kinds, V i , . . . , V„. 7/ will
then allow the variables of type V, to range over a domain Dj. / must be a bijection of
UyD ;i onto Uy D'2 that maps D y, onto D'2 (1 ^ j ^ n).
10 A more precise characterization will be found in Grzegorczyk, OML, pp.347ff.
11 See below, N otes 74 and 124.
12 See John Mayberry (1977).
13 The better known axiom system s for set theory derive from those propounded by
Zermelo in 1908 and by von Neumann in 1925. (In English in Heijenoort, FFG,
pp.l99ff., 393ff.) All such system s contain axioms to the effect that if some definite set
or sets exist, then another set also exists. Particularly far reaching is the so-called
Axiom of the Power Set: If a set S exists, the set of all subsets of S -c a lle d the power
set of S and denoted in this book by 0>(S) - also exists.
14 Stewart, WW, Vol.III, p.117. See also the quotation ibid., p.123, n.l.
15 Stewart, WW, Vol.III, p.114.
16 See, however, the passage from Buffon’s Histoire Naturelle (Vol.I, Paris 1749,
pp.53f.) quoted in Tonelli (1959), p.47 n.52.
17 Stewart, WW, Vol.III, pp.26, 31, 32 n.l.
18 Grassmann, WW, 1.1, p.10.
19 Grassmann, WW, 1.1, p.65.
20 See the references in N ote 67 of Part 2.2.
21 If ( eu e2, e3) is a basis of such a vector space, the scalar product of two vectors
v = 2,- w = 2, bxex is 2, axbx ( ah bx€ R). The norm assigns to each vector v the real
number ||v|| equal to the positive square root of the scalar product of v with itself. A
clear sketch of the structure of a Grassmann third level system is given in Kline, MT,
pp.782ff.
22 Veronese, GG, p.674.
23 Let V be a three-dimensional vector space with scalar product /. A basis (ely e2, e3) of
V is orthonormal if /(e„ et) = 6,, (/, j = 1,2,3).
NOTES TO CHAPTER 3 397
(abbreviated GG). Save for one small but significant change on p.120, this reproduces
the text of the 7th edition of 1930 (substantial changes have been introduced however
in the appendices and supplements). For the 7th edition, Hilbert revised the text quite
thoroughly. Important changes had been made earlier in the second edition (1903). All
changes are carefully noted in Hilbert-Rossier, FG.
104 Hilbert, GG, p.l.
105 Hilbert, GG, p.2.
106 On these matters, however, opinions are not unanimous. Thus, the great Danish
geometer J. Hjemslev, who considered Axiom 1 1 (“Two points determine a straight
line”) intuitively false, because intuitive points and lines are not widthless, counted the
Archimedean axiom as intuitively evident, apparently because he expected his “natural
geom etry” or “geometry of reality” to be displayable on a drawing-board. See
Hjelmslev (1923), p.7.
107 HUbert, GG, p.l.
108 HUbert, GG, pp.125.
109 Hubert’s definition of triangle (GG, p.9) covers the degenerate case in which the
three vertices are collinear, but he tacitly assumes that every triangle he mentions is not
degenerate. This assumption was made explicit by Bernays in a note added to the 8th
edition (GG, p.14).
110 Axiom V 2 did not appear in the first edition. Its necessity was pointed out by
HUbert in a lecture “Ueber den Zahlbegriff” delivered in 1899. (Appendix VI of GG,
7th edition; suppressed in later editions.) The French translation of the Grundlagen by
Laugel, published in 1900, includes a note, signed by HUbert, which reads, in part, as
follows: “Remarquons qu’aux cinq precedents groupes d’axiomes Ton peut encore
ajouter l’axiome suivant qui n’est pas d’une nature purement geometrique [...]: Au
systeme des points , droites et plans, il est impossible d'adjoindre d'autres etres de
maniere que le systeme ainsi generalise forme une nouvelle g&ometrie ou les axiomes des
cinq groupes (/ ) a (V) soient tous verifies; en d'autres termes ; les elements de la
geometrie fom ent un systeme d'etres qui, si Von conserve tous les axiomes, n'est
susceptible d'aucune extension. [ . . . ] La valeur de cet axiome au point de vue des
principes, tient [. . . ] a ce que l’existence de tous les points limites en est une
consequence et que, par suite, cet axiome rend possible la correspondance univoque et
reversible des points d’une droite et de tous les nombres reels.” (Hilbert-Rossier, FG,
pp.43f.). Editions 2 -6 carry the following version of V 2: “The elements of geometry,
points, lines and planes, constitute a system of objects which, if you assume the
foregoing axioms, does not admit any extension.” In the 7th edition, a statement simUar
to this is proved as Theorem 32. A discussion in depth of the axiom of completeness is
given in Bernays (1955). See also Baldus (1928).
111 I paraphrase this axiom from Baldus (1937), p.225. It implies that every segment
PoQo wUl share the property ascribed to AoB0. Copy AoB0 on a plane on which P0 and Q0
lie. Let lines AoPo and B0Qo meet at Z. Lines ZA, and ZB, meet line P0Qo at P, and Q„
respectively. The sequence ((P„ Q,)) fulfils conditions (i) and (ii). Line ZX meets P0Q0
at Y. Y lies between P„ and Q„, for every positive integer n.
1.2 Frege, KS, p.409.
1.3 Frege, KS, p.411.
114 If K is inconsistent, i.e. if K^=S and Kf="~lS, K is unsatisfiable, for every
NOTES TO CHAPTER 3 401
interpretation which satisfies K must satisfy {S, IS}, so that there is no such inter
pretation at all. On the other hand, if K is unsatisfiable, every interpretation which satisfies
K (i.e. none at all) will satisfy any arbitrary sentence. Consequently, for any sentence S,
K^=S and K ^="lS. Hence, K is inconsistent.
115 We say that an axiomatic theory T, with axioms K, is soundly formalized within a
calculus C if a sentence S can be proved from K in C only if K^= S. See p.192, Section
3.2.2.
116 Let K be formalized in a calculus C. The syntactical consistency of K in this
formalization can be established by showing that a proof from K in C cannot possibly
terminate in a particular sentence of C. This can usually be done by studying the
possible outcome of a few well-defined operations on finite strings of symbols.
117 The following remark might prevent some misunderstandings. The syntactical
consistency of the arithmetic of natural numbers in its standard formalization was
demonstrated by Gentzen in 1936. He had to resort to methods of proof that one would
normally regard as less unquestionable than those of arithmetic itself. But if Gentzen’s
result is correct, then, on the strength of Godel’s theorem (Godel (1931), Theorem XI),
that formalization cannot be complete. Consequently, the syntactical consistency of
arithmetic in that formalization does not ensure its consistency in our (semantical)
sense of the word.
118 Pascal’s theorem is a theorem in projective geometry which says that, if six distinct
points P,, P2, P3, P4, P 5 , P 6 lie on a conic, the points of intersection P 1P 2 n P 4P5,
P 2P 3 n P 5P6, and P 3P 4 n P 6P, lie on a line. The theorem used by Hilbert says that if P,,
P 3, P 5 lie on a line and P2, P4, P 6 lie on another line, then, if PjP 2 is parallel to P 4P 5, and
P 2P 3 is parallel to P 5P6, P 3P 4 is parallel to P 6Pi- This is clearly the restriction of Pascal’s
theorem to the affine plane, with a degenerate conic equal to two straight lines and the
points of intersection on the line ‘at infinity’.
119 Dehn (1900).
120 In Euclid’s and Hilbert’s sense, i.e. coplanar and not concurrent.
121 A modified version of Veblen’s axioms of order was reproduced on p.197. This
modified version is taken from the axiom system of Veblen (1911), which, following the
example of R.L. Moore (1908), includes congruence of segments among the undefined
concepts.
122 Huntington (1902), p.277.
123 Veblen (1904), p.346.
124 Like Peano and Veblen, Huntington regards the signs that stand for the set-
theoretical predicates “ jc is a set” , “x is an element of set y ” , as ‘‘symbols which are
necessary for all logical reasoning” (Huntington (1913), p.526), and treats them as
non-interpretable words.
125 Huntington (1913), p.530.
126 Huntington (1913), p.524.
127 Huntington (1913), p.540.
128 A lattice is a triple (S, U , I- !), where S is a set and U and f l are two associative and
commutative binary operations on S, such that for any u and v in S, MU(Mrit>) =
u ri(w U v) = u, u H v is usually called the meet and u U v the join of u and v. In the
case mentioned in the text above, for any two spheres A and B, let their join be the
smallest sphere which contains A and B while their meet is the largest sphere which is
402 NOTES TO CHAPTER 3
C H A P T E R 4. E M P I R I C I S M , A P R I O R I S M , C O N V E N T I O N A L I S M
distances measured along the other two. (Erdmann, AG, p.42). The distinction in the
text presupposes, therefore, the concept of distance. Erdmann apparently takes it for
granted. As an example of the second kind of 3-fold determined continuous quantity he
mentions the manifold of sounds, each of which is defined by its height, intensity and
timbre. Cf. the critical remarks in Ernst Mach, E l, pp.392ff.
44 Erdmann does not seem to be aware of the difference between the general Rieman-
nian concept of a (continuous) manifold and the concept of an R-manifold, i.e. one
endowed with a Riemannian metric. Erdmann alludes to the latter in AG, p.59.
45 Erdmann, AG, pp.58f.
46 Erdmann, AG, p.69. An additional argument for the flatness of space is given on
p.99. It is based on “the perception ( die Wahmehmung ) that the direction of a moving
body can be regarded, within the limits of our experience, as completely independent of
its position (Orr)”. Erdmann surely refers to a body moving in empty space, for the
direction of a body moving in the presence of other bodies appears to be strongly
influenced by its position relative to them, especially if they are very large. One may
wonder indeed how Erdmann managed to perceive a body moving all by itself in
absolutely empty space.
47 Erdmann, AG, pp.93f.
48 Erdmann, AG, p.116.
49 Erdmann, AG, p.146.
50 Cf. Erdmann, AG, pp.59ff., 91, 141ff.
51 Erdmann, AG, p.91; cf. pp.115, 135, 145, 152f. On Helmholtz, see Section 3.1.3.
52 Erdmann, AG, pp.l44f. Erdmann refers to the dimension number of space on pp.38,
49, 83, 95, 143; nowhere do I find an argument for the above quoted statement.
53 Erdmann, AG, p p .ll 8 f.
54 Erdmann, AG, pp.94f.; cf. p.128.
55 Erdmann extensively quotes the Erlangen Programme in AG, pp.l24f., n.2. Had he
read it? That he remained untouched by Klein’s views can be gathered from the
peculiar distinction he makes between the form and the contents of space: The form
comprises those properties which are common to all n-dimendional manifolds, while
the contents of our space consists in “the triplicity of dimensions and the flatness of
metrical relations” (Erdmann, AG, p.143). Projective geometry is not mentioned in the
whole book.
56 Erdmann, AG, p.157; cf. p.44n.2, p.50.
57 Erdmann, AG, p.158; (my italics). Cf. Helmholtz (1866), p.197; quoted above, p.167,
Section 3.1.3.
58 Erdmann, AG, p.159. I take it that the geometric concepts of construction ( die
geometrische Constructionsbegriffe ) are concepts such as straight line and circle, which
enter into the description of geometrical constructions.
59 Erdmann, AG, p.159.
60 Erdmann, AG, p.160.
61 Erdmann, AG, p. 161.
62 Erdmann, AG, p.169.
63 Erdmann, AG, p.170.
64 Erdmann, AG, p.170. Erdmann admits here that experimental results may show that
space curvature is not constant. According to the Helmholtzian stance taken by
406 NOTES TO CHAPTER 4
Erdmann in the rest of the book, such experimental discovery would imply that the
ideal of a perfectly rigid motion cannot be indefinitely approached, and that, as a
consequence of this, there is an upper bound to the accuracy of actual geometrical
measurements.
65 Calinon (1891), p.375.
66 Calinon (1891), p.375.
67 Calinon (1889), p.589. Cf. Calinon (1891), p.368; “La Geometrie generate est l’etude
de tous les groupes de formes dont les definitions premieres sont astreintes a une
condition unique, qui est de ne donner lieu a aucune contradiction, lorsqu’on les soumet
au raisonnement g^ometrique indefiniment prolong^”.
68 Calinon (1891), p.375. Cf. Klein (1890), p.571, quoted above, p.149, Section 2.3.10.
69 Calinon (1891), p.375.
70 Calinon (1889), p.590. Calinon’s “general geom etry’’ embraces Euclidean geometry,
BL geometry and the so-called geometry of Riemann, that is, spherical or elliptic
geometry. It is, in other words, a theory of maximally symmetric spaces (p.184). The
name “general geom etry’’ indicates that Calinon was less open-minded than he thought.
71 The same paradox had motivated Taurinus’ rejection of BL geometry. See p.53.
72 Calinon (1891), pp.368f. Identical spaces, in this sense, were called isogenous
( isogenes) by Delboeuf, a term often used in the French literature of this period. See
pp.206f.
73 Calinon (1889), p.594.
74 Calinon (1889), p.595.
75 Calinon (1891), p.374 (my italics).
76 Calinon (1893), p.605. The paper of 1893 is written in what the author calls an
“idealistic’’ language; that is, space is identified there with our perceptually based
representation of space. Calinon says that his views on the matters discussed by him
can also be set forth in a realistic language, a task which he proposes as an exercise to
the reader. In a realistic version of Calinon’s paper we ought perhaps to ascribe a
definite though unknown geometric structure to physical space. But it seems clear that
in Calinon’s opinion it makes no scientific sense to state hypotheses about such a
structure. The choice of a geometry in physics must be made in the light of the
requirements of the problem at hand, not with a view to determining the essence of
things in themselves.
77 Published in the Revue generate des sciences on December 15, 1891. See Part 4.4.
78 Calinon (1893), p.607. Such are also the criteria usually applied when choosing a
coordinate system. It is amazing that the great mathematician Poincare should have
overlooked this side o f the analogy drawn by him.
79 Calinon (1893), p.607.
80 Mach, E l, pp.337-448. The two longest chapters (pp.353-422) had previously been
published in English in The Monist. See References.
81 H. Poincar 6 , “Le continu mathematique” (1893); “ L’espace et la g 6 om 6 trie” (1895);
“On the foundations of geom etry’’ (1898). See Part 4.4. Federico Enriques published,
shortly before Mach, a study explaining the psychological development of geometry
from an empiricist point of view. (Enriques, “Sulla spiegazione psicologica dei postulati
della geometria’’ (1901).) I do not think it would be rewarding to discuss Enriques’
views here. They are readily accessible in English in Enriques, PS, pp. 199-231.
NOTES TO CHAPTER 4 407
much of the latter will bear translation into the former (Land (1877), p.40). Indeed,
while analytical geometry is ready to grapple with any number of dim ensions, intuition
is confined, for a mind like ours, to Euclidean 3-space. “All other ‘space’ contrived by
human ingenuity may be an aggregate with fictitious properties and a consistent
algebraical analysis of its own, but space it is called only by courtesy.’’ (Land (1877),
P-42.)
3 “Nur ein einziger grosser und zusammenhangender Irrtum’’. Lotze, M, p.234.
4 Lotze rejects Kant’s arguments for the “transcendental ideality’’ of space, but
supports it with another argument of his own: space as an aggregate of equal points can
only be held together by consciousness. (Lotze, M, pp.211f.)
5 Lotze, M, p.223.
6 “Eines Ordnungssystems leerer Platze” . (Lotze, M, p.235).
7 Lotze, M, pp.241f.
8 Lotze, M, p.247. I have changed L otze’s lettering to make the meaning clearer.
9 “Kame es aber einmal dazu, dass astronomische Messungen grosser Entfemungen
nach Ausschluss aller Beobachtungsfehler eine kleinere Winkelsumme des Dreiecks
nachwiesen, was dann? Dann wiirden wir nur glauben, eine neue sehr sonderbare Art
der Refraction entdeckt zu haben, welche die zur Bestimmung der Richtungen dienen-
den Lichtstrahlen abgelenkt habe; d.h. wir wiirden auf ein besonderes Verhalten des
physischen Realen im Raume, aber gewiss nicht auf ein Verhalten des Raumes selbst
schliessen, das alien unseren Anschauungen widersprache und durch keine eigene
exceptionelle Anschauung verbiirgt wiirde.’’ (Lotze, M, pp.248f.). See the comments in
Russell, FG, p.100. Poincare make a similar remark, without quoting Lotze, in “Les
geometries non-Euclidiennes’’ (1891), p.774 (now in Poincare, SH, pp.95f.).
10 A transcendental deduction proves the necessity of a statement by showing that it
expresses a condition of the possibility of an established branch of knowledge. See
Part. 4.3.
11 In his paper “Ueber das kosmologische Problem’’ (1877). In pp. 105-113, Wundt
discusses Zoellner’s use of “transcendent geometry’’ for constructing a temporally
eternal but spatially finite model universe. He mentions in passing what he believes are
the absurd consequences of a non-zero space curvature. In his opinion, the motion of
rigid bodies would be impossible in a curved space, even if its curvature is constant.
“In einem spharischen Raum von drei Dimensionen, z.B wiirde, da diejenigen Linien,
die den Parallelen im Euklidischen Raum entsprechen, grosste Kreise sind, ein Korper
bei der Translocation von einem bestimmten Punkte an zuerst sich ausdehnen und dann
wieder zusammenziehen.’’ (Wundt (1877), p.109.)
12 Wundt, L, Vol.I, pp.478-506; Vol.II, pp. 176-208. I quote Vol.I after the fifth
(posthumous) edition of 1924.
13 Wundt, L, Vol.I, p.489: “Die unmittelbare gegebene Ordnung unserer Wahrneh-
mungsinhalte.’’ Ibid., p.504: “ Eine Ordnung von Empfindungen’’.
14 Wundt, L, Vol.I, p.497.
15 Wundt, L, Vol.I, p.501.
16 Wundt, L, Vol.I, p.503.
17 Wundt, L, Vol.I, p.504.
18 Wundt, L, Vol.I, p.494.
19 Wundt, L, Vol.I, p.483.
NOTES TO CHAPTER 4 409
Philosophical Development say nothing on this particular point. In the latter book,
pp.39f. Russell dism isses FG as “somewhat foolish” , because Einstein’s space-time
geometry of the General Theory of Relativity “is such as I had declared to be
impossible”.
3 Russell, FG, p.136.
4 Kant, K rV y B 40.
5 Kant, K rV , B 246. This is one of the many passages where Kant might seem to favour
a psychological conception of the a priori.
6 In The Principles of Mathematics (1903), Russell will try to prove that abstract set
theory is only a branch of logic and is therefore, so to speak, trivially a priori. Later
developments have shown that this thesis can only be upheld if we broaden logic well
beyond its traditional limits in such a way that we can no longer regard its apriority as
trivial. (The triviality of pre-Russellian - or rather pre-Fregean - logic is shown by the
fact that we can encode its truths and programme a computer to recognize them; the
truths of modern logic can also be encoded, but it is demonstrably impossible to design
a general method of computation for deciding whether a formula in the appropriate
code does or does not represent a logical truth.)
7 Russell, FG, p.28.
8 Russell, FG, p. 118.
9 Russell, FG, p.132.
10 This suggestion conflicts with the thesis held by Russell in FG that the concept of a
point is self-contradictory. Russell refutes this silly notion in his Principles , Chapter 51.
But even in FG he is able to treat space as a point-set by introducing the concept of a
purely geometrical matter, whose elements or “atoms” are unextended and supply the
terms for spatial relations. See Russell, FG, pp.77ff., 190ff.; and below, p.315, Section
4.3.5.
11 Replying to a critical review by Poincare (1899), Russell writes: “Je ne pretends pas
que deux points definissent une droite. Je soutiens, au contraire, que chaque droite est
simple et inanalysable, et par suite indefinissable. Ce que je veux dire est ceci:
‘L ’ensemble complet des lignes droites etant donne, il suffit de mentionner deux points
de l’une d’elles pour la distinguer de toutes les autres, en tant qu’elle est specialement
H6 e a ces deux points’ ” . (Russell, 1899, p.696). This explanation, however, will not
increase the deductive power of Axiom III, or of the other axioms.
12 The reader is advised to read the argument in Russell, FG, pp.l36f.
13 Russell, FG, pp.l39f. Russell’s rejection of infinite-dimensional spaces is not an
instance of mathematical ignorance - such spaces had not yet been discovered - but of
a typically unmathematical narrow-mindedness which often crops up in philosophical
literature.
14 See Peano (1890).
15 Russell, FG, p.149.
16 Russell, FG, p.46.
17 Russell, FG, p.118.
18 Russell, FG, p.147.
19 Russell, FG, p.119; cf. p.30.
20 Russell, FG, p.35.
21 Russell, FG, p.164.
NOTES TO CHAPTER 4 411
(PoincarS, VS, pp.93f.; cf. p.98.) In a paper written shortly before his death, Poincare
wrote however that “it is the external world, it is experience, which determines us’’ to
choose a three-dimensional model of space (Poincare, DP, p.157; my italics).
16 Kant, KrV, B146, A557/B585, A613f./B641f.
17 In its original context (in the paper of 1891 on non-Euclidean geometries) the
argument against apriorism read as follows: Are geometrical axioms “ synthetic a priori
judgments, as Kant said? They would then impose them selves on us with such force
that we should be unable to conceive the contrary proposition or to build a theory upon
it. There would not be a non-Euclidean geom etry’’. (Poincare, SH, p.74.) In 1895, he
proposed a different argument, directed, it would seem, more specifically against Kant’s
doctrine: “If geometrical space were a frame imposed on each of our representations
individually considered it would be impossible to represent an image stripped of this
frame and we could not change anything in our geom etry’’. (Poincare, SH, p.8 8 .) Even
this second argument falls wide of the mark, insofar as Kant did not view space as a
frame imposed on each of our sense-perceptions considered individually, but on all of
them considered as a manifold. Both texts were reprinted without alterations in La
science et Vhypothese (1903).
18 Poincare, DP, p.157: “I shall conclude that we all have in us the intuition of a
continuum of an arbitrary number of dimensions, because we have the faculty of
constructing a physical and mathematical continuum; that this faculty precedes all
experience, because without it experience properly so called would be impossible and
would reduce itself to crude sensations, incapable of being organised in any way; that
this intuition is nothing but our awareness ( conscience ) of this faculty’’. This sounds
like a modified Kantianism, quite consonant with its original spirit. See, e.g., Kant, Afc.,
Vol.XVII, p.639, where the a priori intuition of space is described as awareness
(Bewusstsein ) of one’s own aptitude to perceive things according to certain relation
ships.
19 Poincare, SH, Chapter V, “L ’experience et le geom etrie”, a chapter composed of
passages pieced together from Poincare (1891), (1899), (1900), plus a short summary of
some of the ideas presented in Poincare (1898).
20 See Section 4.4.5.
21 Poincare, SH, p.76; Poincare (1898), pp.42f. See below, pp.351f., Section 4.4.5.
22 Poincare, SH, p . 1 0 1 ; cf. pp.75, 95.
23 Poincare, SH, p.99. Compare Einstein’s formulation of the general principle of
relativity : “Natural laws are only statements about spatio-temporal coincidences;
therefore their natural expression must consist in generally covariant equations’’.
(Einstein (1918), p.241.)
24 Minkowski (1908). In fact, Minkowski’s contribution concerned only the underlying
manifold of special relativity, which he showed to be a semi-Riemannian 4-manifold
(see Appendix, p.372). But he paved the way for the four-dimensional formulation of
Newtonian mechanics by Cartan (1923, 1924).
25 In 1912, he described Minkowski’s spacetime geometry as “a new convention’’
which some physicists had adopted not because they were “constrained” to do so, but
because they found it “more comfortable” . Physicists who did not share that opinion
could, of course, adhere to their earlier convention and preserve their “old habits”.
NOTES TO CHAPTER 4 415
Poincare added: “Je crois, entre nous, que c ’est ce qu’ils feront encore longtemps”.
(Poincare, DP, p.109.)
26 See p.289 and N ote 9 to Part 4.2. It is likely that Poincare reinvented this argument
independently.
27 Poincare (1891), p.774; SH, pp.95f. (my italics). On “Riemann’s geom etry” see Part
2 .2 , N ote 52.
28 Poincare foresaw the rise of “an entirely new mechanics chiefly characterized by the
fact that no velocity can surpass that of light”. (Poincar 6 , VS, pp.l38f.; the passage
belongs to a lecture given in St. Louis in 1904.)
29 Poincare, SH, p.103.
30 Poincare, Oeuvres, Vol.XI, p.90.
31 Poincare, Oeuvres, Vol.XI, p.91.
32 Poincare (1898), p.43. The normal subgroup in question is the group of translations;
see above, p.178. Normal subgroups are defined in the Appendix, p.360. Today’s
mathematicians call a group ‘simple’ if it contains no proper normal subgroups. (See,
e.g., Kurosh, TG, Vol.I, p.6 8 .) If we adopt this terminology, Poincare’s thesis must be
restated thus: the Euclidean group is simpler than the two classical non-Euclidean
groups because the latter are simple, while the former is not. This seeming paradox
bears witness to the ambiguity of simplicity.
33 Poincare, Oeuvres, Vol.XI, p.91. The “operations” of the group are not what we
would call thus (namely, the group product and the mapping which assigns to every
element its own inverse), but simply the elements of the group; the group is viewed as
“operating” through them on an underlying space.
34 A topological group is a group G endowed with a topology by virtue of which the
mapping (g,h)>-*gh~l (g,h i G) is a continuous mapping. The effective and transitive
action of a topological group on a topological space is defined like that of a Lie group
on a differentiable manifold (make the action continuous in the definition on p.172).
35 I cannot understand why Sklar (1969), p.47, stresses the fact that in the usual
Euclidean models of BL geometry the BL group is allowed to act only on a fragment of
R3 (an open ball, a half-space; see Section 2.3.7). Let B denote the interior of a sphere
in R2; B is homeomorphic with R3. Let / : B -> R 3 be an homeomorphism. Let G be the
set of all continuous transformations of B which map chords onto chords and preserve
a given orientation of B. G is a group isomorphic with the BL group (p.140).
{fgf~l | g € G} is then a group of continuous transformations of R 3 which is plainly
isomorphic with G, and hence with the BL group.
36 This group - call it G - is a compact topological group with a countable base of its
topology (Appendix, p.360). Suppose that G acts transitively on R3, choose a point
p i R 3 and let H denote the set {g | g i G and gp = p}. H is a closed subgroup of G
known as the stability (or isotropy) group of p. The quotient space G/H is compact.
Since R 3 is Hausdorff and locally compact (i.e. every point in R3 has a compact
neighbourhood), it follow s that G/H is homeomorphic with R3. But this is absurd,
because R 3 is not compact. Consequently, our supposition that G acts transitively on R 3
cannot be true. (See Matsushima, DM, p. 185.)
37 See however Glymour (1972).
38 Poincar^ (1898), (1903). See below, Section 4.4.5.
39 See pp.l72f., where this is explained in connection with Lie groups.
416 NOTES TO CHAPTER 4
and by C, since both are indistinguishable from B. This contradicts Poincare’s asser
tion.
53 In 1898, Poincare used the word displacement both in this sense and as a synonym
of our term locomotion. This creates some confusion. In 1903, he called external
locomotions changements de position ; internal locom otions, deplacements du corps en
bloc (Poincare, VS, p.79), and reserved deplacement as a name for an equivalence class
of locomotions. He lacks a word for locomotions com posed of internal and external
changes.
56 There is one noticeable difference between walking on a conveyor-belt and rowing in
a tied boat: the latter activity can be interrupted without giving rise to any external
change; on the other hand, whenever the former stops, one will be rapidly carried
backwards (unless the conveyor is arranged to stop as soon as you cease stamping your
feet on it). But according to our definitions the nature and identity of an internal change
depends exclusively on the muscular sensations involved in it.
57 N ow Chapter II of La science et Vhypothese.
58 I.e. the topology consisting of the empty set and all the unions of open intervals.
59 An n-dimensional topological manifold is a topological space every one of whose
points has a neighbourhood homeomorphic with R". A concept substantially equivalent
to this was introduced by Poincare himself in his pioneering work “Analysis situs”
(1895).
60 PoincanS, SH, p.51.
61 Poincare (1898), pp.l4f. Poincare’s argument implies that every displacement is
joined to 0 by a simple physical continuum; it follows easily that any two displace
ments are also joined thus.
62 The italicized assumption is made in Poincare (1903), but not in Poincar 6 (1898).
Instead of it, the earlier work assumes the following: If locom otions A, A' are cancelled
by X, and B, B' are cancelled by Y, then A + B and A' -I- B' are cancelled by Y + X. This
also conflicts with the physical continuity of the group of displacements. For let A and
B be instances of kd, A' and B', instances of (k + 1)d, where k and d are as above; then
A + B and A' + B' are instances of 2 kd and (2k + 2)d, respectively, and they will not be
cancelled by the same locomotion.
63 Poincar 6 fails to make this distinction (in 1898, he even calls locom otions and
displacements by the same name). But it is unavoidable, because locomotions do not
form a group (not every locomotion X can be added to a given locomotion A to
make A + X, but only those whose initial state is indistinguishable from the final state
of A).
64 Poincar 6 (1898), p.21. Poincare’s assertion depends on L ie’s second main solution of
the Lie-H elm holtz space problem (Lie, TT, Vol.III, pp.506f.; see Part 3.1, N ote 43, ad
finem). This presupposes that the group in question is a Lie group. Can this be said of
3)2 It has been observed lately that “it is difficult to estimate to which degree [Lie’s]
proofs [in this second main solution] can withstand modern demands of rigour”.
Freudenthal (1965), p.152.) Lie’s first main solution is generally acknowledged to be
rigorous, but it depends on the assumption of free mobility in the infinitesimal (p.178)
which Poincar 6 has not proved for 3).
65 Poincar 6 puts this as follows: “Displacements, we have seen, correspond to changes
418 NOTES TO CHAPTER 4
in our sensations, and if w e distinguish in the present group between form and material,
the material can be nothing else than that which the displacements cause to change,
viz., our sensations. Even if we suppose that what we have above called sensible space
has already been elaborated, the material would then be represented by as many
continuous variables as there are nerve-fibres.” (Poincare (1898), pp.22f.)
66 Poincare (1898), p.23. On the form and material of a group, see above, pp.336f.
67 They will be found in Poincar 6 (1898), pp.23-32 and in a much improved version in
Poincar 6 , VS, pp.80-93.
68 See N ote 36.
69 Poincar 6 (1898), p.5; cf. Poincar 6 (1895), now in SH, p.82. By drawing our attention
to the fact that Euclidean space is not a whit more imaginable than the non-Euclidean
spaces Poincar 6 has finally disposed of a staple Boeotian argument.
70 Poincar 6 (1898), p.42.
71 Poincar 6 , SH, p.76.
72 Poincar 6 (1898), p.42.
73 His main contribution is contained in his long paper “Analysis situs” (1895) and its
five supplements, published between 1899 and 1904. But the “qualitative theory” of
differential equations developed in his four papers “Sur les courbes d6 finies par une
Equation difTSrentielle” (1881-1886) also follows an essentially topological approach.
See Kline, MT, pp.732ff., 1170ff.
74 Poincar*, VS, pp.59f.; DP, pp.l32ff.
75 Poincar 6 , DP, p.134. We say, today, that two figures F, G (belonging to the same or
to different topological spaces) are topologically equivalent if there is a continuous
bijection / : F -* G whose inverse f~l is also continuous. This concept is wider than the
one proposed by Poincar 6 in DP. It should be noted, however, that the name
homeomorphism given today to such bijections as / was introduced by Poincare. See
Georg Feigl (1928), p.274.
76 Poincar 6 , VS, p.59; cf. VS, p.55; FS, p.417; DP, pp.100, 136.
77 Poincard, DP, p.135; VS, p.60.
78 “L ’espace et ses trois dim ensions” , now in Poincar 6 , VS, Chapters III and IV;
“Pourquoi l’espace a trois dim ensions” , now in DP, pp. 133-157. Both articles appeared
originally in the Revue de metaphysique et de morale.
79 Poincar^ (1893); reprinted in Poincar 6 , SH, pp.58-60.
80 Euclid, Book I, Definitions 3 and 6 ; Book XI, Definition 2. (Heath, EE, Vol.I, pp.165,
171; vol.III, p.263.) Compare Menger (1926), p p .ll 8 f.
81 Poincar^, SH, pp.58-60; VS, pp.61-64.
82 Poincar 6 , SH, p.60; VS, pp.64f.
83 A subset U of S is open if and only if U is the intersection of S with an open set of
Euclidean 3-space. (I thank D . M. Johnson for pointing out to me, in private cor
respondence, a bad mistake in the text preceding reference number 83 on page 356 of the
original edition of this book.)
84 Cf. Hurewicz and Wallman, DT, p.24. Inductive dimension as defined above is called
by recent authors “small inductive dimension” or ind, and is distinguished from “large
inductive dimension” or Ind, which is defined as follows: lnd(0) = - 1 ; Ind(S) = n if n
is the least non-negative integer such that for each closed set E and each open set G of
NOTES TO CHAPTER 4 419
S such that E C G there is an open set F such that E C F C G and the large inductive
dimension of the boundary of F is n - 1. (See Pears, DT, p.155.)
85 Nam ely, separable metrizable spaces. A topological space S is separable if it
contains a countable subset C whose closure C is identical with S. S is metrizable if
there exists a distance function d: S x S -» R such that for every x in S and every r in R,
the ‘open ball’ {y | y 6 S and d( x, y ) < r} is open in S. Distance functions were defined in
Part 2.2, Note 36 (p.384).
86 A topological space S is normal if (i) S is T r separable (i.e. if for every pair of points
xy y 6 S there is a neighbourhood X of x and a neighbourhood Y of y such that y / X
and x / Y ) ; (ii) for each pair of disjoint closed sets A and B of S there exist disjoint
open sets U and V of S such that A C U and B C V. A normal space S is totally normal
if every subspace of S is normally situated in S, that is, if every A C S satisfies the
following requirements: for every open set U of S which contains A there exists an
open set G such that A C G C U ; G is the union of a family (G ,)l€/ of open sets of S,
each of which is the union of a countable family of closed sets of S, and each x 6 G has
a neighbourhood which intersects only a finite number of members of the family
< G , ) , « f.
87 L ebesgue’s comments are contained in a letter to O. Blumenthal and were published
in Mathematische Annalen , Vol.70, right after Brouwer’s proof. Their main purpose
was to show that Brouwer’s result could be demonstrated in a simpler way. Lebesgue’s
argument, however, is unsatisfactory.
Books: Abbreviations referring to books are obtained from the book titles and are
given after the respective author’s name(s).
In both the case of books and journals, when more than one edition, or issue, is listed
under a particular entry, quotations given refer to the one mentioned first.
Ahrens, J.: 1959, ‘Begriindung der absoluten Geometrie des Raumes aus dem
SpiegelungsbegrifT, MZ 71, 154-185.
Alexander, H.G. (LCC), The Leibniz-Clarke Correspondence. Manchester: Manchester
University Press, 1956.
Angell, R.B.: 1974, ‘The geometry of visibles’, Nous 8 87-117.
Aristotelis Opera edidit Academia Regia Borussica. Aristoteles graece ex recognitione
Immanuelis Bekkeri. Berlin: G. Reimer, 1831. 2 Vols.
Aristotle: Physics. A revised text with introduction and commentary by W.D. Ross.
Oxford: Clarendon Press, 1936.
Bachmann, Friedrich: 1936, ‘Eine Begriindung der absoluten Geometrie in der Ebene’,
MA 113 424—451.
Bachmann, Friedrich: 1951, ‘Zur Begriindung der Geometrie aus dem Spiegelungs-
begriff, MA 123 341-344.
Bachmann, Friedrich: Aufbau der Geometrie aus dem Spiegelungsbegriff. Zweite
erganzte Auflage. Berlin: Springer, 1973.
Baldus, Richard: 1928, ‘Zur Axiomatik der Geometrie. I. Ueber Hilberts Vollstandig-
keitsaxiom’. MA 100 321-333.
Baldus, Richard: 1930, ‘Zur Axiomatik der Geometrie. III. Ueber das Archimedische
und das Cantorsche Axiom ” . Sitzungsberichte der Heidelberger Akademie der
Wissenschaften , Math.-Naturwiss. KL, 1930, 5. Abhandlung.
420
R EFE R EN C ES 421
Baldus, Richard: 1937, ‘Zur Axiomatik der Geometric. V. Vereinfachung des Hil-
bertschen Axiom ensystem s der Euklidischen Geometrie.’ Sitzungsberichte der
Bayerischen Akademie der Wissenschaften, Mathem.-Naturwiss. Abteilung, 1937,
189-228.
Becker, Oskar: 1934, ‘Die Lehre vom Geraden und Ungeraden im Neunten Buch der
Euklidischen Elemente. (Versuch einer Wiederherstellung in der urspriinglichen
Gestalt).’ In Becker, GGM pp. 125-145. (Originally published in Quellen und Studien
zur Geschichte der Mathematik, Astronomie und Physik, Abt. B, Bd. 3, S. 533-553.)
Becker, Oskar: (GGM), Zur Geschichte der griechischen Mathematik. Darmstadt:
W issenschaftliche Buchgesellschaft, 1965.
Behnke, H., et al.: (FM ), Fundamentals of Mathematics. Translated by S.H. Gould,
Cambridge, Mass.: The MIT Press, 1974. 3 Vols.
Beltrami, Eugenio: ( OM ), Opere matematiche. Milano: Ulrico Hoepli, 1902, Vol.I.
Beltrami, Eugenio: 1868, ‘Saggio di interpretazione della geometria non-euclidea’,
Giomale di matematiche, 6 284-312. (Now in Beltrami, OM Vol.I, pp.374-405.)
Beltrami, Eugenio: 1868/9. ‘Teoria fondamentale degli spazii di curvatura costante’,
Annali di Matematica Pura ed Applicata , serie II, tomo II, 232-255. (Now in
Beltrami, OM Vol.I, pp.406-429.)
Bernays, Paul: 1955, ‘Betrachtungen iiber das Vollstandigkeitsaxiom und verwandte
A xiom e.’ MZ 63 219-229.
Bernays, Paul: 1959, ‘Die Mannigfaltigkeit der Direktiven fur die Gestaltung
geometrischer Axiom ensystem e.’ In Henkin et al., AM 1-15.
Birkhoff, Garrett: 1935, ‘Combinatorial relations in projective geom etry.’ Annals of
Mathematics, 36 743-748.
Birkhoff, Garrett: (LT), Lattice Theory. Providence: American Mathematical Society,
1967.
Black, Max: 1942, ‘Conventionalism in geometry and the interpretation of necessary
statements’, PS, 9 335-349.
Blumenthal, L.M. and Menger, K.: Studies in Geometry. San Francisco: W.H.
Freeman, 1970.
Bolyai, Farkas: Tentamen juventutem studiosam in elementa matheseos purae elemen-
taris ac sublimioris, methodo intuitiva, evidentiaque huic propria, introducendi.
Tomus Primus. Maros Vasarhely: J. et S. Kali, 1832.
Bolyai, Janos: (SA S), The Science Absolute of Space. Translated by G.B. Halsted.
Printed as a Supplement to Bonola, NEG.
Bolzano, Bernard: (O), Oeuvres editees par la Societe Royale d es.L ettres et des
Sciences de Boheme. Tome 5. Memoires Geometriques, publiees et commentees
par Dr. Jan Vojtfcch. Praha: Kralovska Ceska Spolecnost Nauk, 1948.
Bolzano, Bernard: (BGE), Betrachtungen iiber einige Gegenstande der Elementar-
geometrie. Prag: Karl Barth, 1804. (Now in Bolzano, O, Vol.5, pp.9-49.)
Bonola, Roberto: (NEG), Non-Euclidean Geometry. A critical and historical study of
its development. English translation with additional appendices by H.S. Carslaw.
N ew York: Dover, 1955.
Borsuk, Karol and Szmielew, Wanda: (FG), Foundations of Geometry. Euclidean and
Bolyai-Lobachevskian Geometry. Projective Geometry. Translated from Polish by
Erwin Marquit. Amsterdam: North-Holland, 1960.
422 R EFE R EN C ES
Gauss, Carl Friedrich: Briefwechsel zwischen C.F. Gauss und H.C. Schumacher,
herausgegeben von C.A.F. Peters. Altona: G. Esch, 1860-1865. 6 Vols.
Gentzen, Gerhard: 1936, ‘Die Widerspruchsfreiheit der reinen Zahlentheorie’, MA 112
493-565.
Gergonne, Joseph-Diez: 1818, ‘Essai sur la theorie des definitions’, Annales des
Mathematiques, 9 1-35.
Glymour, Clark; 1972, ‘Topology, Cosmology and Convention’, Synthese, 24 195-218.
Godel, Kurt: 1931, ‘Ueber formal unentscheidbare Satze der Principia Mathematica
und verwandter Systeme. I.’ Monatshefte fur Mathematik und Physik, 38 173-198.
Goe, George: 1962, ‘Kastner, forerunner of Gauss, Pasch and Hilbert’, Proceedings of
the Tenth International Congress of the History of Science, Ithaca 1962. Paris:
Hermann, 1964, pp.659-661.
Gomez-Lobo, Alfonso: 1977, ‘Aristotle’s Hypotheses and the Euclidean Postulates’,
Review of Metaphysics, 30 430-439.
Gonseth, Ferdinand: ( GPE ), La Geometrie et le Probleme de VEspace, Neuchatel:
Griffon, 1945-1955. 6 fasc.
Grassmann, Hermann: (WW), Gesammelte mathematische und physikalische Werke.
Ersten Bandes erster Theil. Die Ausdehnungslehre von 1844 und die Geometrische
Analyse. Unter der Mitwirkung von E. Study hrsg. von F. Engel. New York:
Chelsea, 1969. (Reprint of Leipzig edition of 1894.)
Grassmann, Hermann: Die Wissenschaft der extensiven Grosse oder die Ausdeh
nungslehre, eine neue mathematische Disciplin. Erster Theil, die lineale Ausdeh
nungslehre enthaltend. Leipzig: Otto Wigand, 1844.
Griinbaum, Adolf: (PPST ), Philosophical Problems of Space and Time. Second,
enlarged edition. Dordrecht: D. Reidel, 1973.
Grzegorczyk, A.: (OML), An Outline of Mathematical Logic. Fundamental Results
and Notions Explained with All Details. Dordrecht: D. Reidel, 1974.
Hanson, Norwood Russell: (CC), Constellations and Conjectures, edited by W.C.
Humphreys. Dordrecht: D. Reidel, 1973.
Hantzsche, W. and Wendt, H.: 1935, ‘Dreidimensionale euklidische Raumformen’, MA
110 593-611.
Hausdorff, Felix: 1904, ‘Das Raumproblem’, Annalen der Naturphilosophie, 3 1-23.
Heath, Thomas L.: (EE), The Thirteen Books of Euclid's Elements, translated from the
text of Heiberg with introduction and commentary. Second edition, revised with
additions. N ew York: Dover, 1956. 3 Vols.
Heath, Thomas L.: ( GA ), Greek Astronomy. London: J.M. Dent, 1932.
Hegel, Georg Wilhelm Friedrich: (Werke), Samtliche Werke. Jubilaumsausgabe in 20
Banden, herausgegeben von H. Glockner. Stuttgart: Frommann-Holzboog, 1958.
Heijenoort, Jean van: (FFG), From Frege to Godel. A Source-Book in Mathematical
Logic, 1879-1931. Cambridge, Mass.: Harvard University Press, 1967.
Heilbronner, Johann Christoph: Versuch einer mathematischen Historie. Erster Theil.
Frankfurt u. Leipzig: S. Wohler, 1739.
Helmholtz, Hermann von: (WA), Wissenschaftliche Abhandlungen. Leipzig: J.A.
Barth, 1882-1895. 3 Vols.
Helmholtz, Hermann von: (SE), Schriften zur Erkenntnistheorie. Herausgegeben und
erlautert von Paul Hertz und Moritz Schlick. Berlin: Springer, 1921.
426 RE FE R EN C ES
Lambert, Johann Heinrich: Theorie der Parallellinien, in Stackel and Engel, TP,
pp. 152-207. (Originally published in Magazin fur die reine und angewandte Mat he-
matik,, Nr.2, pp. 137-164 and Nr.3, pp.325-358 (1786).)
Lamy, Bernard: Les Elements de Geometrie; ou, De la mesure du corps. Qui
comprennent tout ce qu'Euclide en a enseigne; les plus belles propositions
d'Archimede et Vanalise. Paris: A. Pralard, 1685.
Land, J.P.N.: 1877, ‘Kant’s space and modern mathematics’, Mind, 2 38^16.
Lange, Ludwig: 1885, ‘Ueber das Beharrungsgesetz’, Berichte iiber die Verhandlungen
der K. Sachsischen Gesellschaft der Wissenschaften zu Leipzig. Math.-Phys. Klasse,
37 333-351.
Lebesgue, H.: 1911, ‘Sur la non-applicabilite de deux domaines appartenant a des
espaces de n et n + p dimensions. (Extrait d’une lettre a M.O. Blumenthal.)’ MA 70
166-168.
Lechalas, Georges: 1889, ‘La geometrie generate’, Critique Philosophique, n.s., 5 II,
217-231.
Lechalas, Georges: 1890a, ‘La geometrie generale et les jugements synthetiques a
priori’, RPh 30 157-169.
Lechalas, Georges: 1890b, ‘Les bases experimentales de la geom etrie’, RPh 30 639-
641.
Lechalas, Georges: 1890c, ‘La geometrie generale et l’intuition’, Annales de Philoso-
phie Chretienne, n.s., 23 57-74.
Lechalas, Georges: 1893, ‘La geometrie non-euclidienne et le principe de similitude’,
RMM 1 199-201.
Lechalas, Georges: 1894, ‘M. Delboeuf et le probleme des mondes sem blables’, RPh 37
73-78. (Followed by ‘Reponse de M. D elboeuf’, Ibid., pp.78-84.)
Lechalas, Georges: 18%, ‘Note sur la courbure et la distance en geometrie generale’,
RMM 4 194-202.
Lechalas, Georges: (ET), Etude sur I'Espace et le Temps. Paris: Alcan, 1896.
Lechalas, Georges: 1898, ‘L’axiome de libre mobilite d’apres M. R ussell’, RMM 6
746-758.
Lechalas, Georges: 1898b, ‘Les fondements de la geometrie d’apres M. Russell’,
Annales de Philosophie Chretienne, n.s., 38 646-660; 39 75-93, 179-197, 317-334.
Lechalas, Georges: 1901a, ‘Un paradoxe geometrique’, RMM 9 361-367.
Lechalas, Georges: 1901b, ‘Les fondements de la geometrie. A propos d’un livre
recent’, Revue Neoscolastique, 8 (1901), 338-354; 9 (1902), 19-34.
Lechalas, Georges: (JGG), Introduction a la geometrie generale. Paris: Gauthier-
Villars, 1904.
Legendre, Adrien-Marie: Elements de geometrie, avec des notes. Paris: F. Didot, 1794.
Legendre, Adrien-Marie: 1833, ‘Reflexions sur les differentes manieres de demontrer la
theorie des paralleles’, Memoires de VAcademie des Sciences, Paris, 12 367-410.
Levi-Civita, Tullio: 1917, ‘Nozione di parallelismo in una varieta qualunque’, Rendi-
conti del Circolo Matematico di Palermo, 42 173-205.
Lie, Sophus: 1886, ‘Bemerkungen zu v. Helmholtzs Arbeit: Ueber die Tatsachen, die
der Geometrie zu Grunde liegen’, in Lie, Gesammelte Abhandlungen, II. Band, 1 .
Teil, Leipzig: B.G. Teubner, 1935, pp.374-379. (Originally published in Berichte
iiber die Abhandlungen der Kgl. Sachsischen Gesellschaft der Wissenschaften in
Leipzig, Math.-Phys. Classe, Supplement, abgeliefert am 21.2.1887, pp.337-342.)
430 RE FER ENC ES
Lie, Sophus: 1890, Ueber die Grundlagen der Geometrie, Darmstadt: W issenschaft-
liche Buchgesellschaft, 1967. (Originally published in Berichte iiber die Abhand-
lungen der Kg. Sachsischen Gesellschaft der Wissenschaften in Leipzig, Math.-Phys.
Classe, 42 (1890), pp.284-321, 335-418.)
Lie, Sophus: (7 T ), Theorie der Transformationsgruppen. Unter Mitwirkung von Dr.
Friedrich Engel. New York: Chelsea, 1970. 3 Vols. (Reprint of Leipzig edition of
1888-1893).
Lie, Sophus: (VCG), Vorlesungen iiber continuierliche Gruppen, mit geometrischen und
anderen Anwendungen, bearbeitet und herausgegeben von Dr. Georg Scheffers.
New York: Chelsea, 1971. (Reprint of Leipzig edition of 1893.)
Lipschitz, R.: 1870, ‘Fortgesetzte Untersuchungen in Betreff der ganzen homogenen
Functionen von n D if f e r e n t ia l’, JraM 72 1-56.
Lobachevsky, Nikolai Ivanovich: ‘O nachalakh geometrii’, Kasanski Vestnik, Feb.-
March 1829, pp. 178-187; April 1829, pp.228-241; N ov.-D ee. 1829, pp.227-
243; March-April 1830, pp.251-283; July-Aug. 1830, pp.571-636.
Lobachevsky, Nikolai Ivanovich: ‘N ovye nachala geometrii’, Uchonia sapiski Kasan-
skava Universiteta, 1835, III, pp.3-48; 1836, II, pp.3-98; III, pp.3-50; 1837, I,
pp.3-97, 1838, I, pp.3-124; III, pp.3-65.
Lobachevsky, Nikolai Ivanovich: (ZGA ), Zwei geometrische Abhandlungen, aus dem
Russischen ubersetzt, mit Anmerkungen und mit einer Biographie des Verfassers
von Friedrich Engel. Leipzig: B.G. Teubner, 1898-1899. 2 Vols. (German annotated
translation of the two preceding items.)
Lobachevsky, Nikolai Ivanovich: 1837, ‘Geometrie Imaginaire’, JraM 17 295-320.
Lobachevsky, Nikolai Ivanovich: Geometrische Untersuchungen zur Theorie der
Parallellinien, Berlin: F. Fincke, 1840.
Lobachevsky, Nikolai Ivanovich: (GRTP), Geometrical Researches on the Theory of
Parallels, translated by G.B. Halsted. Printed as a Supplement to Bonola, NEG.
(English translation of the preceding German monograph.)
Lobachevsky, Nikolai Ivanovich: Pangeometrie ou precis de geometrie fondee sur une
theorie generate et rigoureuse des paralleles. Kazan: Universitet, 1856.
Lorentz, H.A., Einstein, A., Minkowski, H.: (R), Das Relativitatsprinzip. Eine Samm-
lung von Abhandlungen. Darmstadt: W issenschaftliche Buchgesellschaft, 1958.
(Reprint of 1923 edition.)
Lotze, Hermann: (M ), Metaphysik. Drei Bucher der Ontologie, Kosmologie und Psy-
chologie. Leipzig: S. Hirzel, 1879.
Lucretius: De rerum natura, with an English translation by W .H.D. Rouse. London:
William Heinemann, 1924. (The Loeb Classical Library.)
Liiroth, J.: 1875, ‘Das Imaginare in der Geometrie und das Rechnen mit Wurfen.
Darstellung und Erweiterung der v. Staudt’schen Theorie’, MA 8 145-214.
Liiroth, J.: 1877, ‘Das Imaginare in der Geometrie (Zweite Abhandlung)’, MA 1 1
84-110.
Mach, Ernst: (BAE)y Beitrage zur Analyse der Empfindungen. Jena: Gustav Fischer,
1886.
Mach, Ernst: 1901/2, ‘On the psychology and natural development of geom etry’, The
Monist, 12 481-515.
Mach, Ernst: 1903/4, ‘Space and geometry from the point of view of physical inquiry’,
The Monist, 14 1-32.
R EF ER ENC ES 431
Mach, Ernst: (El), Erkenntnis und Irrtum. Skizzen zur Psychologie der Forschung.
Darmstadt: W issenschaftliche Buchgesellschaft, 1968. (Reprint of 5th edition, pub
lished in Leipzig in 1925; the first edition appeared in 1905.)
MacLane, Saunders and Birkhoff, Garrett: Algebra. N ew York: Macmillan, 1967.
Maimonides, Moses: ( GP ), The Guide of the Perplexed. Translated with an intro
duction and notes by Shlomo Pines. Chicago: The University of Chicago Press,
1963.
Malliavin, Paul: (GDI), Geometrie differentielle intrinseque. Paris: Hermann, 1972.
Mannheim, Jerome H.: ( GPST ), The Genesis of Point-Set Topology. London: Pergamon,
1964.
Matsushima, Yozo: (DM), Differentiable Manifolds. Translated by E.T. Kobayashi.
N ew York: Marcel Dekker, 1972.
Mayberry, John: 1977, ‘On the Consistency Problem for Set Theory: An Essay on the
Cantorian Foundations of Classical Mathematics’, BJPS 28 1-34, 137-170.
Menger, Karl: 1923, 1924, ‘Ueber die D im ensionality von Punktmengen’, Monatshefte
fur Mathematik und Physik, 33 148-160; 34 137-161.
Menger, Karl: 1926, ‘Bericht iiber die Dimensionstheorie’, JDMV 35 113-150.
Menger, Karl: 1928, ‘Axiomatik der endlichen Mengen und der elementargeometrischen
Verknupfungsbeziehungen’, JDMV 37 309-325.
Menger, Karl: 1938, ‘Non-Euclidean Geometry of Joining and Intersecting’, American
Mathematical Society Bulletin, 44 821-824.
Menger, Karl: 1940, Three Lectures on Mathematical Subjects - Delivered at the Rice
Institute, December 11, 12 and 13, 1939. The Rice Institute Pamphlet, XXVII, No. 1.
Menger, Karl, Alt, F. and Schreiber, O.: 1936, ‘N ew foundations of projective and
affine geom etry’, Annals of Mathematics, 27 456-482.
M£ray, Charles: 1869, ‘Remarques sur la nature des quantites defines par la condition
de servir de limites a des variables donnees’, Revue des Societes Savantes, 4 (1869),
280-289.
Meray, Charles: (NEG), Nouveaux elements de geometrie, Paris: F. Savy, 1874.
Mill, John Stuart: ( SL), A System of Logic Ratiocinative and Inductive, being a
Connected View of the Principles of Evidence and the Methods of Scientific
Investigation. Edited by J.M. Robson. Toronto: University of Toronto Press,
1973-1974. 2 Vols.
Miller, Arthur I.: 1972, ‘The myth of Gauss’ experiment on the Euclidean nature of
physical space’, Isis, 63 345-348.
Miller, Arthur I. et al.: (1974, ‘Comments on Miller’s “The myth of G auss’ experiment
on the Euclidean nature of physical space” ,’ Isis, 65 83-87. (Comment I by George
Goe; Comment II by B. L. van der Waerden; Reply by Arthur I. Miller.)
Milnor, J.: 1956, ‘On manifolds homeomorphic to the 7-sphere’, Annals of Mathema
tics, 64 399-405.
Minding, Ferdinand: 1837, ‘Ueber die Biegung gewisser Flachen’, JraM 17 295-320.
Minding, Ferdinand: 1839, ‘Wie sich entscheiden lasst, ob zwei gegebene krumme
Flachen aufeinander abwickelbar sind oder nicht, nebst Bemerkungen iiber die
Flachen von unveranderlichen Kriimmungsmasse’, JraM 19 370-387.
Minding, Ferdinand: 1840, ‘Beitrage zur Theorie der kiirzesten Linien auf krummen
Flachen’, JraM 20 323-327.
Minkowski, Hermann: 1908, ‘Raum und Zeit’, in Lorentz et al., R, pp.54-66. (Lecture
432 RE FE R EN C ES
read before the 80. Versammlung Deutscher Naturforscher und Aerzte in Cologne
on September 21, 1908.)
Montucla, Jean Etienne: Historie des mathematiques, dans laquelle on rend compte de
leurs progres depuis leur origine jusqu's nos jours. Paris: C.A. Jombert, 1758. 2 Vols.
Mooij, J.J.A.: La philosophic des mathematiques de Henri Poincare, Paris: Gauthier-
Villars, 1966.
Moore, Eliakim Hastings: 1902, ‘On the projective axioms of geom etry’, American
Mathematical Society Transactions, 3 142-158.
Moore, G.E.: 1899, Review of Russell, An Essay on the Foundations of Geometry.
Mind, n.s., 8 397-405.
Moore, Robert L.: 1908, ‘Sets of metrical hypotheses for geom etry’, American
Mathematical Society Transactions, 9 487-512.
More, Henry: (OO), Opera Omnia. Hildesheim, Olms, 1967. 2 Vols. in 3. (Reprint of
the London edition of 1679.)
Mouret, G.: 1892, ‘Sur les geometries non-euclidiennes’, Revue generale des Sciences
pures et appliquees, 3 39-40.
Munkres, James R.: ( EDT ), Elementary Differential Topology. Princeton: Princeton
University Press, 1963.
Munkres, James R.: Topology. A First Course. Englewood Cliffs: Prentice-Hall, 1975.
Nagel, Ernest: 1939, ‘The formation of modern conceptions of formal logic in the
development of geom etry’, Osiris, 7 142-224.
Neumann, Carl Gottfried: (PO N T ), Ueber die Principien der Galilei-Newton'schen
Theorie, Leipzig: B.G. Teubner, 1870.
Nomizu, K.: 1954, ‘Invariant affine connections on homogeneous spaces’, American
Journal of Mathematics, 76 33-65.
Padoa, Alessandro: 1900, ‘Essai d’une theorie algebrique des nombres entiers, pr£c6 de
d’une introduction logique a une theorie deductive quelconque’, in Bibliotheque du
Congres International de Philosophic, Paris 1900, Paris: Armand Colin, 1901, Vol.
3, pp.309-265.
Padoa, Alessandro: 1900b, ‘Un nouveau system e irreductible de postulats pour l’al-
gebre’, Compte rendu du Deuxieme Congres International des Mathematiciens tenu a
Paris du 6 au 12 aout 1900. Paris: Gauthier-Villars, 1902, pp.249-256.
Pasch, Moritz: ( VNG)> Vorlesungen iiber neuere Geometrie. Leipzig: B.G. Teubner,
1882.
Pasch, Moritz: ( VNG 2), Vorlesungen iiber neuere Geometrie. Zweite Auflage. Berlin:
Springer, 1926.
Pasch, Moritz: 1917, ‘Grundfragen der Geometrie’, JraM 147 184-190.
Peano, Giuseppe: 1889, I Principii di Geometria logicamente esposti. Torino: Fratelli
Bocca, 1889.
Peano, Giuseppe: 1890, ‘Sur une courbe, qui remplit toute une aire plane’, MA 36
157-160.
Peano, Giuseppe: 1894, ‘Sui fondamenti della geometria’, Rivista di Matematica, 4
51-90.
Peano, Giuseppe: 1900, ‘Les definitions mathematiques’, Bibliotheque du Congres
International de Philosophie, Paris 1900. Paris: A. Colin, 1901, V ol.3, pp.279-288.
Pears, A.R.: (D T ), Dimension Theory of General Spaces. Cambridge: Cambridge
University Press, 1975.
R EFE R EN C ES 433
Russell, Bertrand: 1896b, T h e logic of geom etry’, Mind, n.s., 5 1-23. (This article is
almost identical with the preceding one; their text is reproduced with minor changes
in Russell, FG, Chapter III. B, Sections I and II.)
Russell, Bertrand: (FG ), An Essay on the Foundations of Geometry. N ew York:
Dover, 1956. (Reprint of the first edition, published by Cambridge University Press
in 1897.)
Russell, Bertrand: 1898, ‘Les axiomes propres a Euclide sont-ils empiriques?’, RMM 6
759-776. (Reply to Couturat, 1898.)
Russell, Bertrand: 1899, ‘Sur les axiomes de la geometrie’, RMM 7 684-707. (Reply to
Poincar 6 , 1899.)
Russell, Bertrand: 1900, ‘L ’id6 e d’ordre et la position absolue dans l’espace et le tem ps’.
Bibliotheque du Congres International de Philosophie, Paris 1900, Paris: A. Colin,
1901, Vol.3, pp.241-277.
Russell, Bertrand: 1901, ‘Is position in time and space absolute or relative?’ Mind, n.s.,
10 293-317.
Russell, Bertrand: 1901b, ‘On the notion of order’, Mind, n.s., 1 0 30-51.
Russell, Bertrand: (Principles), The Principles of Mathematics. Cambridge: Cambridge
University Press, 1903.
Russell, Bertrand: ( K E W ), Our Knowledge of the External World as a Field for
Scientific Method in Philosophy. Chicago: Open Court, 1915.
Russell, Bertrand: (A M ), The Analysis of Matter. London: Allen and Unwin, 1927.
Russell, Bertrand: My Philosophical Development. London: Allen and Unwin, 1959.
Russell, Bertrand: The Autobiography of Bertrand Russell, 1872-1914. Boston: Little,
Brown, 1967.
Russo, Francois: 1969, Groupes et geometrie: La genese du Programme d'Erlangen de
Felix Klein. Paris: Palais de la Decouverte, 1969.
Saccheri, Girolamo: Euclides ab omni naevo vindicatus sive conatus geometricus quo
stabiliuntur prima ipsa universae geometriae principia. Mediolani: Ex Typographia
Pauli Antonii Montani, 1733.
Saccheri, Girolamo: Euclides vindicatus, edited and translated by G.B. Hals ted. Chi
cago: Open Court, 1920.
Savile, Henry: Praelectiones tresdecim in principium Elementorum Euclidis. Oxonii:
Iohannes Lichfield & Iacobus Short, 1621.
Scholz, Heinrich: 1930, ‘Die Axiomatik der Alten’, Blatter fur deutsche Philosophie, 4
259-278.
Schopenhauer, Arthur: (WW), Samtliche Werkey nach der ersten, von Julius Frauen-
stadt besorgten Gesamtausgabe neu bearbeitet und herausgegeben von Arthur Hub-
scher. Wiesbaden: E. Brockhaus, 1948-1950. 7 Vols.
Schouten, J.A.: 1926, ‘Erlanger Programm und Uebertragungslehre. N eue Gesichts-
punkte zur Grundlegung der Geometrie’, Rendiconti del Circolo Matematico di
Palermo, 50 142-169.
Schur, Friedrich: 1886a, ‘Ueber die Deformation der Raume constanten Riemann-
schen Krummungsmaasses’, MA 27 163-176.
Schur, Friedrich: 1886b, ‘Ueber den Zusammenhang der Raume constanten Riemann-
schen Krummungsmaasses mit den projectiven Raumen’, MA 27 537-567.
Schur, Friedrich: 1891, ‘Ueber die Einfuhrung der sogenannten idealen Elemente in die
projective Geometrie’, MA 39 113-124.
REF E R E NC E S 437
Taurinus, Franz Adolf: Theorie der Parallellinien. Koln am Rhein: J.P. Bachem, 1825.
Taurinus, Franz Adolf: Geometriae Prima Elementa. Colonia Agrippinae: Typis J.P.
Bachemii, 1826.
Tits, J.: 1952, "Etude de certains espaces metriques’, Bulletin de la Societe Mathema-
tique de Belgique, 4 pp.44-52. (‘Addenda’, Ibid., 5 (1953) 124-125.)
Tobias, Wilhelm: Grenzen der Philosophies constatirt gegen Riemann und Helmholtz,
vertheidigt gegen von Hartmann und Lasker. Berlin, 1875.
Tonelli, Giorgio: 1959, ‘Der Streit iiber die mathematische Methode in der Philosophic
der ersten Halfte des 18. Jahrhunderts und die Enstehung von Kants Schrift iiber
die “Deutlichkeit’V Archiv fur Philosophies 9 37-66.
Torretti, Roberto: 1970, ‘On the Subjectivity of Objective Space’, Proceedings of the
Third International Kant Congress, Rochester 1970t Dordrecht: D. Reidel, 1972,
pp.569-573.
Toth, Imre: 1966/7, ‘Das Parallellenproblem im Corpus Aristotelicum’, Archive for the
History of Exact Sciences, 3 249-422.
Ueberweg, Friedrich: (PG ), Die Prinzipien der Geometrie wissenschaftlich dargestellt.
In Die Welt- und Lebensanschauung Friedrich Ueberwegst hrsg. v. Moritz Brasch.
Leipzig: Gustav Engel, 1889, pp.263-316. (Originally published in Archiv fur
Philologie und Padagogiks 17 (1851).)
Urysohn, P.S.: 1922, ‘Les m ultiplicity cantoriennes’, Comptes Rendus de VAcadtmie
des Sciences (Paris), Ser. A -B y 157 440-442.
Urysohn, P.S.: 1925, 1926, ‘M6moire sur les m ultiplicity cantoriennes’, Fundamenta
Mathematicae, 7 30-137; 8 225-359.
Van der Waerden, B.L.: 1947/9, ‘Die Arithmetik der Pythagoreer’, MA 120 127-153,
676-700. (Also in Becker, GGM, pp.203-254.)
Van der Waerden, B.L.: (SA), Science Awakening. English translation by Arnold
Dresden with additions of the author. N ew York: Oxford University Press, 1961.
Veblen, Oswald: 1904, ‘A system of axioms for geom etry’, American Mathematical
Society Transactions 5 343-384.
Veblen, Oswald: 1911, ‘The foundations of geometry’, in J.W.A. Young, Monographs
on Topics in Modem Mathematics relevant to the Elementary Fields N ew York:
Dover, 1955, pp.3-51. (This is a reprint of the 1911 edition of Young’s book, in
which Veblen’s article was originally published.)
Veronese, Giuseppe: 1891, Fondamenti di Geometria a piu dimensioni e a piu specie di
unitd rettilinees esposti in forma elementare. Padova: Tipografia del Seminario, 1891.
Veronese, Giuseppe: (GG), Grundziige der Geometrie von mehreren Dimensionen und
mehreren Arten gradliniger Einheiten in elementarer Form entwickelt. Mit Geneh-
migung des Verfassers nach einer neuen Bearbeitung des Originals iibersetzt von
Adolf Schepp. Leipzig: B.G. Teubner, 1894. (Pp.631-684, Anhang: Historisch-
kritische Untersuchungen liber die Principien der Geometrie.)
Vuillemin, Jules: 1972, ‘Poincare’s philosophy of space’, Synthese, 24 161-179.
Wallis, John: Opera mathematica. Oxonii: e theatro Sheldoniano, 1695-1699. 3 Vols.
Waltershausen, W. Sartorius von: Gauss zum G e d a c h tn iS s Leipzig, 1856.
Wehrli, Fritz: Die Schule des Aristoteles. Texte und Kommentar. Straton von Lamp -
sakos. Basel: Schwabe and Co., 1950.
W eyl, Hermann: 1918, ‘Reine Infinitesimalgeometrie’, MZ 2 384-411.
R E F E R E NC E S 439
41-42, 43,48, 54, 55, 56, 57, 60, Formal theories need not be
61,62, 63,66,67, 130, 143, 153, formalized 192
169, 205, 207, 216, 224, 228, Formal differential geometry 98,
233, 237, 239, 258, 264, 273, 382 n.8
282, 294, 296, 297, 318, 380, Formalism 196; Hilbert 236;
389, 399 Lambert 49
Euclid’s Postulates 7, 376; see Formalization 192; complete 402
also Aitema n. 135; decidable 402 n.135;
E u d o x u s o f C n i d u s 9, 10, 11, sound 401 n. 115
12, 13, 14, 18, 238 Form of externality (Russell) 302,
Eudoxian ratios 11 305, 306, 307, 310, 411 n.36
E u l e r , Leonhard 72, 74 Foundations of geometry, facts
Existential claims in mathematics or hypotheses? (Helmholtz vs.
199, 235f. Riemann) 84, 156f., 161, 167,
Existential quantifier 194 169
Experience and geometry 209, Fourth harmonic 125, 144, 389
224, 225, 254, 257, 272, 273, Free mobility 157, 295, 296, 310,
274, 278, 284, 292, 318, 325, 312, 314-318, 411 n.36; see also
329, 331, 332, 333, 351; see Movement and geometry, Rigid
also Empiricism, Geometrical body
experiments Free mobility in the infinitesimal
ExpP 95 178, 394 n.47
Extensive relations (Ausdeh- F r e g e , Gottlob 226, 235, 249,
nungsverhaltnisse - Riemann) 251, 252, 253, 397, 399,
104 400, 402, 424
F r e u d e n t h a l , Hans xii, 188,
Fano, G. 223, 399, 424 223, 387, 397, 399, 402, 417, 424
F e ig l Georg 418, 424
, Fringe 354
Fibre 359; bundle 366 F r i t z , Kurt von 5, 375, 424
F i c h t e , Johann Gottlieb 393
Figure 212, 220, 230, 312, 313
Finite geometries 225, 399 n.93 gib g“ 94
Flat 246 G Galileo 11, 12, 15, 24,
a l il e i,
Flat space 99f. 189, 378, 424
Flatland (Helmholtz) 392 n.27 G a u s s , Carl Friedrich 40, 52, 53,
Forces as the foundation of 54, 55, 56, 58, 62, 63, 64, 65, 67,
space structure: Kant 29; 68, 72, 73, 74, 75, 78, 79, 82, 84,
Lobachevsky 64f.; Riemann 88, 95, 101, 107, 109, 163, 189,
383 n.21 254, 255, 256, 285, 289, 317,
I NDEX 447
394 n.45; simple 415 n.32; see tions 176-179, 394 n.43, 404
also Lie, One-parameter, n.33
Stability, Topological and H e n k i n , Leon 395, 426
Transformation group H e r b a r t , Johann Friedrich 84,
Group invariant 139, 176f. 107-108, 386, 426
Group realization 337; basis 337; H e r z , Paul 394, 425
faithful 337; immanent 337f., H e s s e n b e r g , Gerhard 426
350f.; transitive 337 H i l b e r t , David xi, 141, 154, 155,
G r O n b a u m , Adolf 275, 352, 383, 185-188, 189, 190, 193, 198,
425 202, 227-239, 243, 249, 250,
G r z e g o r c z y k , Andrzej 396, 425 251, 252, 330, 389, 395, 397,
399, 400, 401, 402, 426
Half-plane 223 H i n t i k k a , Jaakko 378, 426
H a n k e l , Hermann 109 H i p p a r c h u s 19
H a n s o n , Norwood Russell 377, H i p p a s u s o f M e t a p o n t u m 10
425 H j e l m s l e v , J. 248, 427
H a n t z s c h e , W. 391, 425 H o b b e s , Thomas 323
Haptic space 280, 407 H o h e n b u r g , Herwart von 377
Harmonic net 144, 145, 390 n.55 Homeomorphism 139, 361, 418
Harmonic points and lines 125 n.75
H a u s d o r f f , Felix 362, 383, 425 Homogeneity 203, 206f., 274, 299,
H e a t h , Thomas L. 375, 376, 379, 316, 378, 411 n.36
402, 418, 425 Homogeneous coordinates 116,
H e g e l , Georg Wilhelm Friedrich 118, 146, 217, 397 n.69
153, 391,425 Homologous parts (Pasch) 215f.
H e i j e n o o r t , Jean van 396, 425 H o p f , Heinz 390, 427
H e i l b r o n n e r , Johann Christoph Horocycle 60, 63
48, 425 Horosphere 60
H e l m h o l t z , Hermann von 100, H o u e l , J. 189, 207-210, 397, 427
153, 155-171, 172, 176, 177, H u m e , David 254, 329
179, 185, 190, 206, 226, 254, H u n t i n g t o n , Edward V. 240,
255, 261, 264, 265, 266, 268, 243-246, 401, 427
269, 281, 289, 295, 296, 314, H u r e w i c z , Witold 418, 427
318, 337, 391, 392, 393, 394, H u s s e r l , Edmund 229
395, 404, 405, 407, 425, 426 Hyperbolic geometry 128, 129,
Helmholtz-Lie problem of space 130
100, 154, 155, 162, 171f., 188; Hyperbolic transformation of the
Helmholtz’s error according to line 131
Lie 161, 393 n.43; Lie’s solu Hyperplane 247
I NDEX 449