The Project Gutenberg EBook of Space--Time--Matter, by Hermann Weyl

This eBook is for the use of anyone anywhere at no cost and with
almost no restrictions whatsoever. You may copy it, give it away or
re-use it under the terms of the Project Gutenberg License included
with this eBook or online at www.gutenberg.org

Title: Space--Time--Matter

Author: Hermann Weyl

Translator: Henry L. Brose

Release Date: June 21, 2013 [EBook #43006]

Language: English

Character set encoding: ISO-8859-1

*** START OF THIS PROJECT GUTENBERG EBOOK SPACE--TIME--MATTER ***

Produced by Andrew D. Hwang, using scanned images and OCR
text generously provided by the University of Toronto
Gerstein Library through the Internet Archive.

transcriber’s note
The camera-quality files for this public-domain ebook may
be downloaded gratis at

www.gutenberg.org/ebooks/43006.

This ebook was produced using scanned images and OCR

text generously provided by the University of Toronto
Gerstein Library through the Internet Archive.

Typographical corrections, uniformization of punctuation,

and minor presentational changes have been effected
without comment.

This PDF file is optimized for screen viewing, but may be

recompiled for printing. Please consult the preamble of the
LATEX source file for instructions and other particulars.
SPACE—TIME—MATTER
BY

HERMANN WEYL

TRANSLATED FROM THE GERMAN BY

HENRY L. BROSE

WITH FIFTEEN DIAGRAMS

METHUEN & CO. LTD.

36 ESSEX STREET W.C.
LONDON
First Published in 1922
FROM THE AUTHOR’S PREFACE TO THE FIRST
EDITION
Einstein’s Theory of Relativity has advanced our ideas of the
structure of the cosmos a step further. It is as if a wall which sep-
arated us from Truth has collapsed. Wider expanses and greater
depths are now exposed to the searching eye of knowledge, regions
of which we had not even a presentiment. It has brought us much
nearer to grasping the plan that underlies all physical happening.
Although very recently a whole series of more or less popular
introductions into the general theory of relativity has appeared,
nevertheless a systematic presentation was lacking. I therefore
considered it appropriate to publish the following lectures which
I gave in the Summer Term of 1917 at the Eidgen. Technische
Hochschule in Zürich. At the same time it was my wish to present
this great subject as an illustration of the intermingling of philo-
sophical, mathematical, and physical thought, a study which is
dear to my heart. This could be done only by building up the
theory systematically from the foundations, and by restricting at-
tention throughout to the principles. But I have not been able to
satisfy these self-imposed requirements: the mathematician pre-
dominates at the expense of the philosopher.
The theoretical equipment demanded of the reader at the out-
set is a minimum. Not only is the special theory of relativity dealt
with exhaustively, but even Maxwell’s theory and analytical ge-
ometry are developed in their main essentials. This was a part
of the whole scheme. The setting up of the Tensor Calculus—by
means of which, alone, it is possible to express adequately the
physical knowledge under discussion—occupies a relatively large
amount of space. It is therefore hoped that the book will be found
suit able for making physicists better acquainted with this math-
ematical instrument, and also that it will serve as a text-book for
students and win their sympathy for the new ideas.

HERMANN WEYL
Ribbitz in Mecklenburg
Easter, 1918

PREFACE TO THE THIRD EDITION

Although this book offers fruits of knowledge in a refrac-
tory shell, yet communications that have reached me have shown
that to some it has been a source of comfort in troublous times.
To gaze up from the ruins of the oppressive present towards the
stars is to recognise the indestructible world of laws, to strengthen
faith in reason, to realise the “harmonia mundi” that transfuses
all phenomena, and that never has been, nor will be, disturbed.
My endeavour in this third edition has been to attune this har-
mony more perfectly. Whereas the second edition was a reprint
of the first, I have now undertaken a thorough revision which af-
fects Chapters II and IV above all. The discovery by Levi-Civita,
in 1917, of the conception of infinitesimal parallel displacements
suggested a renewed examination of the mathematical foundation
of Riemann’s geometry. The development of pure infinitesimal
geometry in Chapter II, in which every step follows quite natu-
rally, clearly, and necessarily, from the preceding one, is, I believe,
the final result of this investigation as far as the essentials are
concerned. Several shortcomings that were present in my first
account in the Mathematische Zeitschrift (Bd. 2, 1918) have now
been eliminated. Chapter IV, which is in the main devoted to Ein-
stein’s Theory of Gravitation has, in consideration of the various
important works that have appeared in the meanwhile, in partic-
ular those that refer to the Principle of Energy-Momentum, been
subjected to a very considerable revision. Furthermore, a new
theory by the author has been added, which draws the physical
inferences consequent on the extension of the foundations of geom-
etry beyond Riemann, as shown in Chapter II, and represents an
attempt to derive from world-geometry not only gravitational but
also electromagnetic phenomena. Even if this theory is still only
in its infant stage, I feel convinced that it contains no less truth
than Einstein’s Theory of Gravitation—whether this amount of
truth is unlimited or, what is more probable, is bounded by the
Quantum Theory.
I wish to thank Mr. Weinstein for his help in correcting the
proof-sheets.

HERMANN WEYL
Acla Pozzoli, near Samaden
August, 1919

PREFACE TO THE FOURTH EDITION

In this edition the book has on the whole preserved its general
form, but there are a number of small changes and additions, the
most important of which are: (1) A paragraph added to Chapter II
in which the problem of space is formulated in conformity with the
view of the Theory of Groups; we endeavour to arrive at an un-
derstanding of the inner necessity and uniqueness of Pythagorean
space metrics based on a quadratic differential form. (2) We show
that the reason that Einstein arrives necessarily at uniquely de-
termined gravitational equations is that the scalar of curvature is
the only invariant having a certain character in Riemann’s space.
(3) In Chapter IV the more recent experimental researches dealing
with the general theory of relativity are taken into consideration,
particularly the deflection of rays of light by the gravitational field
of the sun, as was shown during the solar eclipse of 29th May,
1919, the results of which aroused great interest in the theory on
all sides. (4) With Mie’s view of matter there is contrasted an-
other (vide particularly § 32 and § 36), according to which matter
is a limiting singularity of the field, but charges and masses are
force-fluxes in the field. This entails a new and more cautious
attitude towards the whole problem of matter.
Thanks are due to various known and unknown readers for
pointing out desirable modifications, and to Professor Nielsen (at
Breslau) for kindly reading the proof-sheets.

HERMANN WEYL
Zürich, November, 1920
 TRANSLATOR’S NOTE
In this rendering of Professor Weyl’s book into English, pains
have been taken to adhere as closely as possible to the original,
not only as regards the general text, but also in the choice of
English equivalents for technical expressions. For example, the
word affine has been retained. It is used by Möbius in his Der
Barycentrische Calcul, in which he quotes a Latin definition of the
term as given by Euler. Veblen and Young have used the word in
their Projective Geometry, so that it is not quite unfamiliar to En-
glish mathematicians. Abbildung, which signifies representation,
is generally rendered equally well by transformation, inasmuch as
it denotes a copy of certain elements of one space mapped out on,
or expressed in terms of, another space. In some cases the Ger-
man word is added in parenthesis for the sake of those who wish
to pursue the subject further in original papers. It is hoped that
the appearance of this English edition will lead to further efforts
towards extending Einstein’s ideas so as to embrace all physical
knowledge. Much has been achieved, yet much remains to be done.
The brilliant speculations of the latter chapters of this book show
how vast is the field that has been opened up by Einstein’s genius.
The work of translation has been a great pleasure, and I wish to
acknowledge here the courtesy with which suggestions concern-
ing the type and the symbols have been received and followed by
Messrs. Methuen & Co. Ltd. Acting on the advice of interested
mathematicians and physicists I have used Clarendon type for
the vector notation. My warm thanks are due to Professor G. H.
Hardy of New College and Mr. T. W. Chaundy, M.A., of Christ
Church, for valuable suggestions and help in looking through the
proofs. Great care has been taken to render the mathematical
text as perfect as possible.

HENRY L. BROSE
Christ Church, Oxford
December, 1921
 CONTENTS
PAGE
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

CHAPTER I

Euclidean Space. Its Mathematical Formulation and its

Rôle in Physics
§ 1. Deduction of the Elementary Conceptions of Space from that
of Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
§ 2. The Foundations of Affine Geometry . . . . . . . . . . . . . . . . . . . . . . . 23
§ 3. The Conception of n-dimensional Geometry. Linear Algebra.
Quadratic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
§ 4. The Foundations of Metrical Geometry . . . . . . . . . . . . . . . . . . . . 39
§ 5. Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
§ 6. Tensor Algebra. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
§ 7. Symmetrical Properties of Tensors . . . . . . . . . . . . . . . . . . . . . . . . . 79
§ 8. Tensor Analysis. Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
§ 9. Stationary Electromagnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 94
 CONTENTS x

CHAPTER II

The Metrical Continuum

PAGE
§ 10. Note on Non-Euclidean Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 113
§ 11. The Geometry of Riemann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
§ 12. Continuation. Dynamical View of Metrical Properties . . . . . . 140
§ 13. Tensors and Tensor-densities in any Arbitrary Manifold . . . . 151
§ 14. Affinely Related Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
§ 15. Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
§ 16. Metrical Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
§ 17. Observations about Riemann’s Geometry as a Special Case. 191
§ 18. Metrical Space from the Point of View of the Theory of
Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

CHAPTER III

Relativity of Space and Time

§ 19. Galilei’s Principle of Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

§ 20. The Electrodynamics of Moving Fields Lorentz’s Theorem of
Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
§ 21. Einstein’s Principle of Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
§ 22. Relativistic Geometry, Kinematics, and Optics . . . . . . . . . . . . . 267
§ 23. The Electrodynamics of Moving Bodies . . . . . . . . . . . . . . . . . . . . 281
§ 24. Mechanics according to the Principle of Relativity. . . . . . . . . . 293
§ 25. Mass and Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
§ 26. Mie’s Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
 CONTENTS xi

CHAPTER IV

The General Theory of Relativity

PAGE
§ 27. The Relativity of Motion, Metrical Fields, Gravitation . . . . . 325
§ 28. Einstein’s Fundamental Law of Gravitation . . . . . . . . . . . . . . . . 342
§ 29. The Stationary Gravitational Field—Comparison with Ex-
periment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
§ 30. Gravitational Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
§ 31. Rigorous Solution of the Problem of One Body . . . . . . . . . . . . . 377
§ 32. Additional Rigorous Solutions of the Statical Problem of
Gravitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
§ 33. Gravitational Energy. The Theorems of Conservation . . . . . . 402
§ 34. Concerning the Inter-connection of the World as a Whole . . 409
§ 35. The Metrical Structure of the World as the Origin of Elec-
tromagnetic Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
§ 36. Application of the Simplest Principle of Action. The Funda-
mental Equations of Mechanics . . . . . . . . . . . . . . . . . . . . . . . . 442
Appendix I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
Appendix II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
Bibliographical References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486

The formulæ are numbered anew for each chapter. Unless

otherwise stated, references to formulæ are to those in the
current chapter.
 SPACE—TIME—MATTER
INTRODUCTION
Space and time are commonly regarded as the forms of ex-
istence of the real world, matter as its substance. A definite
portion of matter occupies a definite part of space at a definite
moment of time. It is in the composite idea of motion that these
three fundamental conceptions enter into intimate relationship.
Descartes defined the objective of the exact sciences as consisting
in the description of all happening in terms of these three fun-
damental conceptions, thus referring them to motion. Since the
human mind first wakened from slumber, and was allowed to give
itself free rein, it has never ceased to feel the profoundly mysteri-
ous nature of time-consciousness, of the progression of the world
in time,—of Becoming. It is one of those ultimate metaphysical
problems which philosophy has striven to elucidate and unravel
at every stage of its history. The Greeks made Space the subject-
matter of a science of supreme simplicity and certainty. Out of
it grew, in the mind of classical antiquity, the idea of pure sci-
ence. Geometry became one of the most powerful expressions of
that sovereignty of the intellect that inspired the thought of those
times. At a later epoch, when the intellectual despotism of the
Church, which had been maintained through the Middle Ages,
had crumbled, and a wave of scepticism threatened to sweep away
all that had seemed most fixed, those who believed in Truth clung
to Geometry as to a rock, and it was the highest ideal of ev-
ery scientist to carry on his science “more geometrico”. Matter
was imagined to be a substance involved in every change, and it
was thought that every piece of matter could be measured as a

1
 INTRODUCTION 2

quantity, and that its characteristic expression as a “substance”

was the Law of Conservation of Matter which asserts that mat-
ter remains constant in amount throughout every change. This,
which has hitherto represented our knowledge of space and mat-
ter, and which was in many quarters claimed by philosophers as
a priori knowledge, absolutely general and necessary, stands to-
day a tottering structure. First, the physicists in the persons
of Faraday and Maxwell, proposed the “electromagnetic field” in
contra-distinction to matter, as a reality of a different category.
Then, during the last century, the mathematician, following a dif-
ferent line of thought, secretly undermined belief in the evidence
of Euclidean Geometry. And now, in our time, there has been un-
loosed a cataclysm which has swept away space, time, and matter
hitherto regarded as the firmest pillars of natural science, but only
to make place for a view of things of wider scope, and entailing a
deeper vision.
This revolution was promoted essentially by the thought of one
man, Albert Einstein. The working-out of the fundamental ideas
seems, at the present time, to have reached a certain conclusion;
yet, whether or not we are already faced with a new state of affairs,
we feel ourselves compelled to subject these new ideas to a close
analysis. Nor is any retreat possible. The development of scientific
thought may once again take us beyond the present achievement,
but a return to the old narrow and restricted scheme is out of the
question.
Philosophy, mathematics, and physics have each a share in the
problems presented here. We shall, however, be concerned above
all with the mathematical and physical aspect of these questions.
I shall only touch lightly on the philosophical implications for the
simple reason that in this direction nothing final has yet been
 INTRODUCTION 3

reached, and that for my own part I am not in a position to

give such answers to the epistemological questions involved as my
conscience would allow me to uphold. The ideas to be worked out
in this book are not the result of some speculative inquiry into the
foundations of physical knowledge, but have been developed in the
ordinary course of the handling of concrete physical problems—
problems arising in the rapid development of science which has, as
it were, burst its old shell, now become too narrow. This revision
of fundamental principles was only undertaken later, and then
only to the extent necessitated by the newly formulated ideas. As
things are to-day, there is left no alternative but that the separate
sciences should each proceed along these lines dogmatically, that is
to say, should follow in good faith the paths along which they are
led by reasonable motives proper to their own peculiar methods
and special limitations. The task of shedding philosophic light on
to these questions is none the less an important one, because it
is radically different from that which falls to the lot of individual
sciences. This is the point at which the philosopher must exercise
his discretion. If he keep in view the boundary lines determined
by the difficulties inherent in these problems, he may direct, but
must not impede, the advance of sciences whose field of inquiry is
confined to the domain of concrete objects.
Nevertheless I shall begin with a few reflections of a philosoph-
ical character. As human beings engaged in the ordinary activi-
ties of our daily lives, we find ourselves confronted in our acts of
perception by material things. We ascribe a “real” existence to
them, and we accept them in general as constituted, shaped, and
coloured in such and such a way, and so forth, as they appear to us
in our perception in “general,” that is ruling out possible illusions,
mirages, dreams, and hallucinations.
 INTRODUCTION 4

These material things are immersed in, and transfused by, a

manifold, indefinite in outline, of analogous realities which unite to
form a single ever-present world of space to which I, with my own
body, belong. Let us here consider only these bodily objects, and
not all the other things of a different category, with which we as
ordinary beings are confronted; living creatures, persons, objects
of daily use, values, such entities as state, right, language, etc.
Philosophical reflection probably begins in every one of us who
is endowed with an abstract turn of mind when he first becomes
sceptical about the world-view of naïve realism to which I have
briefly alluded.
It is easily seen that such a quality as “green” has an ex-
istence only as the correlate of the sensation “green” associated
with an object given by perception, but that it is meaningless to
attach it as a thing in itself to material things existing in them-
selves. This recognition of the subjectivity of the qualities of
sense is found in Galilei (and also in Descartes and Hobbes) in
a form closely related to the principle underlying the construc-
tive mathematical method of our modern physics which
repudiates “qualities”. According to this principle, colours are
“really” vibrations of the æther, i.e. motions. In the field of philos-
ophy Kant was the first to take the next decisive step towards the
point of view that not only the qualities revealed by the senses,
but also space and spatial characteristics have no objective signif-
icance in the absolute sense; in other words, that space, too, is
only a form of our perception. In the realm of physics it is
perhaps only the theory of relativity which has made it quite clear
that the two essences, space and time, entering into our intuition
have no place in the world constructed by mathematical physics.
Colours are thus “really” not even æther-vibrations, but merely
 INTRODUCTION 5

a series of values of mathematical functions in which occur four

independent parameters corresponding to the three dimensions of
space, and the one of time.
Expressed as a general principle, this means that the real
world, and every one of its constituents with their accompanying
characteristics, are, and can only be given as, intentional objects
of acts of consciousness. The immediate data which I receive are
the experiences of consciousness in just the form in which I receive
them. They are not composed of the mere stuff of perception, as
many Positivists assert, but we may say that in a sensation an ob-
ject, for example, is actually physically present for me—to whom
that sensation relates—in a manner known to every one, yet, since
it is characteristic, it cannot be described more fully. Following
Brentano, I shall call it the “intentional object”. In experienc-
ing perceptions I see this chair, for example. My attention is fully
directed towards it. I “have” the perception, but it is only when I
make this perception in turn the intentional object of a new inner
perception (a free act of reflection enables me to do this) that I
“know” something regarding it (and not the chair alone), and as-
certain precisely what I remarked just above. In this second act
the intentional object is immanent, i.e. like the act itself, it is a real
component of my stream of experiences, whereas in the primary
act of perception the object is transcendental, i.e. it is given in
an experience of consciousness, but is not a real component of it.
What is immanent is absolute, i.e. it is exactly what it is in the
form in which I have it, and I can reduce this, its essence, to the
axiomatic by acts of reflection. On the other hand, transcendental
objects have only a phenomenal existence; they are appearances
presenting themselves in manifold ways and in manifold “grada-
tions”. One and the same leaf seems to have such and such a size,
 INTRODUCTION 6

or to be coloured in such and such a way, according to my posi-

tion and the conditions of illumination. Neither of these modes
of appearance can claim to present the leaf just as it is “in itself”.
Furthermore, in every perception there is, without doubt, involved
the thesis of reality of the object appearing in it; the latter is,
indeed, a fixed and lasting element of the general thesis of real-
ity of the world. When, however, we pass from the natural view
to the philosophical attitude, meditating upon perception, we no
longer subscribe to this thesis. We simply affirm that something
real is “supposed” in it. The meaning of such a supposition now
becomes the problem which must be solved from the data of con-
sciousness. In addition a justifiable ground for making it must be
found. I do not by this in any way wish to imply that the view
that the events of the world are a mere play of the consciousness
produced by the ego, contains a higher degree of truth than naïve
realism; on the contrary, we are only concerned in seeing clearly
that the datum of consciousness is the starting-point at which we
must place ourselves if we are to understand the absolute meaning
as well as the right to the supposition of reality. In the field of
logic we have an analogous case. A judgment, which I pronounce,
affirms a certain set of circumstances; it takes them as true. Here,
again, the philosophical question of the meaning of, and the jus-
tification for, this thesis of truth arises; here, again, the idea of
objective truth is not denied, but becomes a problem which has
to be grasped from what is given absolutely. “Pure consciousness”
is the seat of that which is philosophically a priori. On the other
hand, a philosophic examination of the thesis of truth must and
will lead to the conclusion that none of these acts of perception,
memory, etc., which present experiences from which I seize reality,
gives us a conclusive right to ascribe to the perceived object an
 INTRODUCTION 7

existence and a constitution as perceived. This right can always

in its turn be over-ridden by rights founded on other perceptions,
etc.
It is the nature of a real thing to be inexhaustible in content;
we can get an ever deeper insight into this content by the contin-
ual addition of new experiences, partly in apparent contradiction,
by bringing them into harmony with one another. In this inter-
pretation, things of the real world are approximate ideas. From
this arises the empirical character of all our knowledge of reality.∗
Time is the primitive form of the stream of consciousness. It
is a fact, however obscure and perplexing to our minds, that the
contents of consciousness do not present themselves simply as be-
ing (such as conceptions, numbers, etc.), but as being now filling
the form of the enduring present with a varying content. So that
one does not say this is but this is now, yet now no more. If we
project ourselves outside the stream of consciousness and repre-
sent its content as an object, it becomes an event happening in
time, the separate stages of which stand to one another in the
relations of earlier and later.
Just as time is the form of the stream of consciousness, so one
may justifiably assert that space is the form of external material
reality. All characteristics of material things as they are presented
to us in the acts of external perception (e.g. colour) are endowed
with the separateness of spatial extension, but it is only when we
build up a single connected real world out of all our experiences
that the spatial extension, which is a constituent of every percep-
tion, becomes a part of one and the same all-inclusive space. Thus
space is the form of the external world. That is to say, every ma-
∗
Vide note 1.
 INTRODUCTION 8

terial thing can, without changing content, equally well occupy a

position in Space different from its present one. This immediately
gives us the property of the homogeneity of space which is the
root of the conception, Congruence.
Now, if the worlds of consciousness and of transcendental real-
ity were totally different from one another, or, rather, if only the
passive act of perception bridged the gulf between them, the state
of affairs would remain as I have just represented it, namely, on
the one hand a consciousness rolling on in the form of a lasting
present, yet spaceless; on the other, a reality spatially extended,
yet timeless, of which the former contains but a varying appear-
ance. Antecedent to all perception there is in us the experience
of effort and of opposition, of being active and being passive. For
a person leading a natural life of activity, perception serves above
all to place clearly before his consciousness the definite point of
attack of the action he wills, and the source of the opposition to
it. As the doer and endurer of actions I become a single indi-
vidual with a psychical reality attached to a body which has its
place in space among the material things of the external world,
and by which I am in communication with other similar individu-
als. Consciousness, without surrendering its immanence, becomes
a piece of reality, becomes this particular person, namely myself,
who was born and will die. Moreover, as a result of this, con-
sciousness spreads out its web, in the form of time, over reality.
Change, motion, elapse of time, becoming and ceasing to be, exist
in time itself; just as my will acts on the external world through
and beyond my body as a motive power, so the external world
is in its turn active (as the German word “Wirklichkeit,” reality,
derived from “wirken” = to act, indicates). Its phenomena are re-
lated throughout by a causal connection. In fact physics shows
 INTRODUCTION 9

that cosmic time and physical form cannot be dissociated from

one another. The new solution of the problem of amalgamating
space and time offered by the theory of relativity brings with it a
deeper insight into the harmony of action in the world.
The course of our future line of argument is thus clearly out-
lined. What remains to be said of time, treated separately, and
of grasping it mathematically and conceptually may be included
in this introduction. We shall have to deal with space at much
greater length. Chapter I will be devoted to a discussion of Eu-
clidean space and its mathematical structure. In Chapter II
will be developed those ideas which compel us to pass beyond the
Euclidean scheme; this reaches its climax in the general space-
conception of the metrical continuum (Riemann’s conception of
space). Following upon this Chapter III will discuss the prob-
lem mentioned just above of the amalgamation of Space and
Time in the world. From this point on the results of mechanics
and physics will play an important part, inasmuch as this prob-
lem by its very nature, as has already been remarked, comes into
our view of the world as an active entity. The edifice constructed
out of the ideas contained in Chapters II and III will then in the
final Chapter IV lead us to Einstein’s General Theory of Relativ-
ity, which, physically, entails a new Theory of Gravitation, and
also to an extension of the latter which embraces electromagnetic
phenomena in addition to gravitation. The revolutions which are
brought about in our notions of Space and Time will of necessity
affect the conception of matter too. Accordingly, all that has to
be said about matter will be dealt with appropriately in Chapters
III and IV.
To be able to apply mathematical conceptions to questions
of Time we must postulate that it is theoretically possible to fix
 INTRODUCTION 10

in Time, to any order of accuracy, an absolutely rigorous now

(present) as a point of Time—i.e. to be able to indicate points
of time, one of which will always be the earlier and the other the
later. The following principle will hold for this “order-relation”.
If A is earlier than B and B is earlier than C, then A is earlier
than C. Each two points of Time, A and B, of which A is the
earlier, mark off a length of time; this includes every point which
is later than A and earlier than B. The fact that Time is a form of
our stream of experience is expressed in the idea of equality: the
empirical content which fills the length of Time AB can in itself
be put into any other time without being in any way different
from what it is. The length of time which it would then occupy is
equal to the distance AB. This, with the help of the principle of
causality, gives us the following objective criterion in physics for
equal lengths of time. If an absolutely isolated physical system
(i.e. one not subject to external influences) reverts once again to
exactly the same state as that in which it was at some earlier
instant, then the same succession of states will be repeated in
time and the whole series of events will constitute a cycle. In
general such a system is called a clock. Each period of the cycle
lasts equally long.
The mathematical fixing of time by measuring it is based
upon these two relations, “earlier (or later) times” and “equal
times”. The nature of measurement may be indicated briefly as
follows: Time is homogeneous, i.e. a single point of time can only
be given by being specified individually. There is no inherent prop-
erty arising from the general nature of time which may be ascribed
to any one point but not to any other; or, every property logically
derivable from these two fundamental relations belongs either to
all points or to none. The same holds for time-lengths and point-
 INTRODUCTION 11

pairs. A property which is based on these two relations and which

holds for one point-pair must hold for every point-pair AB (in
which A is earlier than B). A difference arises, however, in the
case of three point-pairs. If any two time-points O and E are given
such that O is earlier than E, it is possible to fix conceptually fur-
ther time-points P by referring them to the unit-distance OE.
This is done by constructing logically a relation t between three
points such that for every two points O and E, of which O is
the earlier, there is one and only one point P which satisfies the
relation t between O, E and P , i.e. symbolically,

OP = t · OE

(e.g. OP = 2 · OE denotes the relation OE = EP ). Num-

bers are merely concise symbols for such relations as t, defined
logically from the primary relations. P is the “time-point with
the abscissa t in the co-ordinate system (taking OE as unit
length)”. Two different numbers t and t∗ in the same co-ordinate
system necessarily lead to two different points; for, otherwise, in
consequence of the homogeneity of the continuum of time-lengths,
the property expressed by

t · AB = t∗ · AB,

since it belongs to the time-length AB = OE, must belong to

every time-length, and hence the equations AC = t · AB, AC =
t∗ · AB would both express the same relation, i.e. t would be equal
to t∗ . Numbers enable us to single out separate time-points rel-
atively to a unit-distance OE out of the time-continuum by a
conceptual, and hence objective and precise, process. But the ob-
jectivity of things conferred by the exclusion of the ego and its
 INTRODUCTION 12

data derived directly from intuition, is not entirely satisfactory;

the co-ordinate system which can only be specified by an individ-
ual act (and then only approximately) remains as an inevitable
residuum of this elimination of the percipient.
It seems to me that by formulating the principle of measure-
ment in the above terms we see clearly how mathematics has come
to play its rôle in exact natural science. An essential feature
of measurement is the difference between the “determination” of
an object by individual specification and the determination of the
same object by some conceptual means. The latter is only possi-
ble relatively to objects which must be defined directly. That is
why a theory of relativity is perforce always involved in mea-
surement. The general problem which it proposes for an arbitrary
domain of objects takes the form: (1) What must be given such
that relatively to it (and to any desired order of precision) one can
single out conceptually a single arbitrary object P from the con-
tinuously extended domain of objects under consideration? That
which has to be given is called the co-ordinate system, the con-
ceptual definition is called the co-ordinate (or abscissa) of P
in the co-ordinate system. Two different co-ordinate systems are
completely equivalent for an objective standpoint. There is no
property, that can be fixed conceptually, which applies to one co-
ordinate system but not to the other; for in that case too much
would have been given directly. (2) What relationship exists be-
tween the co-ordinates of one and the same arbitrary object P in
two different co-ordinate systems?
In the realm of time-points, with which we are at present con-
cerned, the answer to the first question is that the co-ordinate
system consists of a time-length OE (giving the origin and the
unit of measure). The answer to the second question is that the
 INTRODUCTION 13

required relationship is expressed by the formula of transformation

t = at0 + b (a > 0)

in which a and b are constants, whilst t and t0 are the co-ordinates

of the same arbitrary point P in an “unaccented” and “accented”
system respectively. For all possible pairs of co-ordinate systems
the characteristic numbers, a and b, of the transformation may
be any real numbers with the limitation that a must always be
positive. The aggregate of transformations constitutes a group,
as their nature would imply, i.e.,
1. “identity” t = t0 is contained in it.
2. Every transformation is accompanied by its reciprocal in the
group, i.e. by the transformation which exactly cancels its effect.
0
Thus,
  inverse of the transformation (a, b), viz. t = at + b, is
the
1 b 1 b
, − , viz. t0 = t − .
a a a a
3. If two transformations of a group are given, then the one
which is produced by applying these two successively also belongs
to the group. It is at once evident that, by applying the two
transformations

t = at0 + b t0 = a0 t00 + b0

in succession, we get
t = a1 t00 + b1
where a1 = a · a0 and b1 = (ab0 ) + b; and if a and a0 are positive,
so is their product.
The theory of relativity discussed in Chapters III and IV pro-
poses the problem of relativity, not only for time-points, but for
the physical world in its entirety. We find, however, that this
 INTRODUCTION 14

problem is solved once a solution has been found for it in the case
of the two forms of this world, space and time. By choosing a co-
ordinate system for space and time, we may also fix the physically
real content of the world conceptually in all its parts by means of
numbers.
All beginnings are obscure. Inasmuch as the mathematician
operates with his conceptions along strict and formal lines, he,
above all, must be reminded from time to time that the origins
of things lie in greater depths than those to which his methods
enable him to descend. Beyond the knowledge gained from the
individual sciences, there remains the task of comprehending.
In spite of the fact that the views of philosophy sway from one
system to another, we cannot dispense with it unless we are to
convert knowledge into a meaningless chaos.
 CHAPTER I
EUCLIDEAN SPACE. ITS MATHEMATICAL
FORMULATION AND ITS RÔLE IN PHYSICS
§ 1. Deduction of the Elementary Conceptions of Space
from that of Equality
Just as we fixed the present moment (“now”) as a geometrical
point in time, so we fix an exact “here,” a point in space, as the
first element of continuous spatial extension, which, like time, is
infinitely divisible. Space is not a one-dimensional continuum like
time. The principle by which it is continuously extended cannot
be reduced to the simple relation of “earlier” or “later”. We shall
refrain from inquiring what relations enable us to grasp this conti-
nuity conceptually. On the other hand, space, like time, is a form
of phenomena. Precisely the same content, identically the same
thing, still remaining what it is, can equally well be at some place
in space other than that at which it is actually. The new portion
of Space S0 then occupied by it is equal to that portion S which
it actually occupied. S and S0 are said to be congruent. To
every point P of S there corresponds one definite homologous
point P 0 of S0 which, after the above displacement to a new posi-
tion, would be surrounded by exactly the same part of the given
content as that which surrounded P originally. We shall call this
“transformation” (in virtue of which the point P 0 corresponds to
the point P ) a congruent transformation. Provided that the
appropriate subjective conditions are satisfied the given material
thing would seem to us after the displacement exactly the same
as before. There is reasonable justification for believing that a
rigid body, when placed in two positions successively, realises this

15
 CHAPTER I 16

idea of the equality of two portions of space; by a rigid body we

mean one which, however it be moved or treated, can always be
made to appear the same to us as before, if we take up the ap-
propriate position with respect to it. I shall evolve the scheme of
geometry from the conception of equality combined with that of
continuous connection—of which the latter offers great difficulties
to analysis—and shall show in a superficial sketch how all funda-
mental conceptions of geometry may be traced back to them. My
real object in doing so will be to single out translations among
possible congruent transformations. Starting from the concep-
tion of translation I shall then develop Euclidean geometry along
strictly axiomatic lines.
First of all the straight line. Its distinguishing feature is that
it is determined by two of its points. Any other line can, even
when two of its points are kept fixed, be brought into another
position by a congruent transformation (the test of straightness).
Thus, if A and B are two different points, the straight line
g = AB includes every point which becomes transformed into it-
self by all those congruent transformations which transform AB
into themselves. (In familiar language, the straight line lies evenly
between its points.) Expressed kinematically, this is tantamount
to saying that we regard the straight line as an axis of rotation.
It is homogeneous and a linear continuum just like time. Any
arbitrary point on it divides it into two parts, two “rays”. If B lies
on one of these parts and C on the other, then A is said to be
between B and C and the points of one part lie to the right of A,
the points of the other part to the left. (The choice as to which
is right or left is determined arbitrarily.) The simplest fundamen-
tal facts which are implied by the conception “between” can be
formulated as exactly and completely as a geometry which is to
 EUCLIDEAN SPACE 17

be built up by deductive processes demands. For this reason we

endeavour to trace back all conceptions of continuity to the con-
ception “between,” i.e. to the relation “A is a point of the straight
line BC and lies between B and C” (this is the reverse of the real
intuitional relation). Suppose A0 to be a point on g to the right
of A, then A0 also divides the line g into two parts. We call that
to which A belongs the left-hand side. If, however, A0 lies to the
left of A the position is reversed. With this convention, analogous
relations hold not only for A and A0 but also for any two points
of a straight line. The points of a straight line are ordered by the
terms left and right in precisely the same way as points of time
by the terms earlier and later.
Left and right are equivalent. There is one congruent trans-
formation which leaves A fixed, but which interchanges the two
halves into which A divides the straight line. Every finite portion
of straight line AB may be superposed upon itself in such a way
that it is reversed (i.e. so that B falls on A, and A falls on B).
On the other hand, a congruent transformation which transforms
A into itself, and all points to the right of A into points to the
right of A, and all points to the left of A into points to the left
of A, leaves every point of the straight line undisturbed. The
homogeneity of the straight line is expressed in the fact that the
straight line can be placed upon itself in such a way that any
point A of it can be transformed into any other point A0 of it, and
that the half to the right of A can be transformed into the half to
the right of A0 , and likewise for the portions to the left of A and A0
respectively (this implies a mere translation of the straight line).
If we now introduce the equation AB = A0 B 0 for the points of the
straight line by interpreting it as meaning that AB is transformed
into the straight line A0 B 0 by a translation, then the same things
 CHAPTER I 18

hold for this conception as for time. These same circumstances

enable us to introduce numbers, and to establish a reversible and
single correspondence between the points of a straight line and
real numbers by using a unit of length OE.
Let us now consider the group of congruent transformations
which leaves the straight line g fixed, i.e. transforms every point
of g into a point of g again.
We have called particular attention to rotations among these
as having the property of leaving not only g as a whole, but also
every single point of g unmoved in position. How can translations
in this group be distinguished from twists?
I shall here outline a preliminary argument in which not only
the straight line, but also the plane is based on a property of
rotation.
Two rays which start from a point O form an angle. Every
angle can, when inverted, be superposed exactly upon itself, so
that one arm falls on the other, and vice versa. Every right angle
is congruent with its complementary angle. Thus, if h is a straight
line perpendicular to g at the point A, then there is one rotation
about g (“inversion”) which interchanges the two halves into which
h is divided by A. All the straight lines which are perpendicular
to g at A together form the plane E through A perpendicular to g.
Each pair of these perpendicular straight lines may be produced
from any other by a rotation about g.
If g is inverted, and placed upon itself in some way, so that A is
transformed into itself, but so that the two halves into which A
divides g are interchanged, then the plane E of necessity coincides
with itself. The plane may also be defined by taking this property
in conjunction with that of symmetry of rotation. Two congruent
tables of revolution (i.e. symmetrical with respect to rotations) are
 EUCLIDEAN SPACE 19

A′
g
A
A∗ g
E
h E∗

Fig. 1.

plane if, by means of inverting one, so that its axis is vertical in

the opposite direction, and placing it on the other, the two table-
surfaces can be made to coincide. The plane is homogeneous. The
point A on E which appears as the centre in this example is in
no way unique among the points of E. A straight line g 0 passes
through each one A0 of them in such a way that E is made up
of all straight lines through A0 perpendicular to g 0 . The straight
lines g 0 which are perpendicular to E at its points A0 respectively
form a group of parallel straight lines. The straight line g with
which we started is in no wise unique among them. The straight
lines of this group occupy the whole of space in such a way that
only one straight line of the group passes through each point of
space. This in no way depends on the point A of the straight
line g, at which the above construction was performed.
If A∗ is any point on g, then the plane which is erected normally
to g at A∗ cuts not only g perpendicularly, but also all straight
lines of the group of parallels. All such normal planes E ∗ which
are erected at all points A∗ on g form a group of parallel planes.
 CHAPTER I 20

These also fill space continuously and uniquely. We need only

take another small step to pass from the above framework of space
to the rectangular system of co-ordinates. We shall use it here,
however, to fix the conception of spatial translation.
Translation is a congruent transformation which transforms
not only g but every straight line of the group of parallels into
itself. There is one and only one translation which transfers the
arbitrary point A on g to the arbitrary point A∗ on the same
straight line.
I shall now give an alternate method of arriving at the con-
ception of translation. The chief characteristic of translation is
that all points are of equal importance in it, and that the be-
haviour of a point during translation does not allow any objective
assertion to be made about it, which could not equally well be
made of any other point (this means that the points of space for
a given translation can only be distinguished by specifying each
one singly [“that one there”], whereas in the case of rotation, for
example, the points on the axis are distinguished by the property
that they preserve their positions). By using this as a basis we
get the following definition of translation, which is quite indepen-
dent of the conception of rotation. Let the arbitrary point P be
transformed into P 0 by a congruent transformation: we shall call
P and P 0 connected points. A second congruent transformation
which has the property of again transforming every pair of con-
nected points into connected points, is to be called interchange-
able with the first transformation. A congruent transformation
is then called a translation, if it gives rise to interchangeable con-
gruent transformations, which transform the arbitrary point A
into the arbitrary point B. The statement that two congruent
transformations I and II are interchangeable signifies (as is easily
 EUCLIDEAN SPACE 21

proved from the above definition) that the congruent transforma-

tion resulting from the successive application of I and II is identical
with that which results when these two transformations are per-
formed in the reverse order. It is a fact that one translation (and,
as we shall see, only one) exists, which transforms the arbitrary
point A into the arbitrary point B. Moreover, not only is it a
fact that, if T denote a translation and A and B any two points,
there is, according to our definition, a congruent transformation,
interchangeable with T, which transforms A into B, but also that
the particular translation which transforms A into B has the
required property. A translation is therefore interchangeable with
all other translations, and a congruent transformation which is
interchangeable with all translations is also necessarily a transla-
tion. From this it follows that the congruent transformation which
results from successively performing two translations, and also the
“inverse” of a translation (i.e. that transformation which exactly
reverses or neutralises the original translation) is itself a trans-
lation. Translations possess the “group” property.∗ There is no
translation which transforms A into A except identity, in which
every point remains undisturbed. For if such a translation were
to transform P into P 0 , then, according to definition, there must
be a congruent transformation, which transforms A into P and
simultaneously A into P 0 ; P and P 0 must therefore be identical
points. Hence there cannot be two different translations both of
which transform A into another point B.
As the conception of translation has thus been defined indepen-
dently of that of rotation, the translational view of the straight
line and plane may thus be formed in contrast with the above
∗
Vide note 2.
 CHAPTER I 22

view based on rotations. Let a be a translation which transfers

the point A0 to A. This same translation will transfer A1 to a
point A2 , A2 to A3 , etc. Moreover, through it A0 will be derived
from a certain point A−1 , A−1 from A−2 , etc. This does not yet
give us the whole straight line, but only a series of equidistant
points on it. Now, if n is a natural number (integer), a trans-
a
lation exists which, when repeated n times, gives a. If, then,
n
a
starting from the point A0 we use in the same way as we just
n
now used a we shall obtain an array of points on the straight line
under construction, which will be n times as dense.
If we take all possible whole numbers as values of n this ar-
ray will become denser in proportion as n increases, and all the
points which we obtain finally fuse together into a linear contin-
uum, in which they become embedded, giving up their individual
existences (this description is founded on our intuition of continu-
ity). We may say that the straight line is derived from a point by
an infinite repetition of the same infinitesimal translation and its
inverse. A plane, however, is derived by translating one straight
line, g, along another, h. If g and h are two different straight
lines passing through the point A0 , then if we apply to g all the
translations which transform h into itself, all straight lines which
thus result from g together form the common plane of g and h.
We succeed in introducing logical order into the structure of
geometry only if we first narrow down the general conception of
congruent transformation to that of translation, and use this as
an axiomatic foundation (§§ 2 and 3). By doing this, however,
we arrive at a geometry of translation alone, viz. affine geometry
within the limits of which the general conception of congruence has
later to be re-introduced (§ 4). Since intuition has now furnished
 EUCLIDEAN SPACE 23

us with the necessary basis we shall in the next paragraph enter

into the region of deductive mathematics.

§ 2. The Foundations of Affine Geometry

For the present we shall use the term vector to denote a trans-
lation or a displacement a in the space. Later we shall have oc-
casion to attach a wider meaning to it. The statement that the
displacement a transfers the point P to the point Q (“transforms”
P into Q) may also be expressed by saying that Q is the end-point
of the vector a whose starting-point is at P . If P and Q are any
two points then there is one and only one displacement a which
transfers P to Q. We shall call it the vector defined by P and Q,
−→
and indicate it by P Q.
The translation c which arises through two successive trans-
lations a and b is called the sum of a and b, i.e. c = a + b.
The definition of summation gives us: (1) the meaning of multi-
plication (repetition) and of the division of a vector by an integer;
(2) the purport of the operation which transforms the vector a
into its inverse −a; (3) the meaning of the nil-vector 0, viz. “iden-
tity,” which leaves all points fixed, i.e. a+0 = a and a+(−a) = 0.
ma
It also tells us what is conveyed by the symbols ± = λa, in
n
which m and n are any two natural numbers (integers) and λ de-
m
notes the fraction ± . By taking account of the postulate of
n
continuity this also gives us the significance of λa, when λ is any
real number. The following system of axioms may be set up for
affine geometry:—
 CHAPTER I 24

I. Vectors
Two vectors a and b uniquely determine a vector a + b as
their sum. A number λ and a vector a uniquely define a vector λa,
which is “λ times a” (multiplication). These operations are subject
to the following laws:—
(α) Addition—
(1) a + b = b + a (Commutative Law).
(2) (a + b) + c = a + (b + c) (Associative Law).
(3) If a and c are any two vectors, then there is one and only
one value of x for which the equation a + x = c holds. It is called
the difference between c and a and signifies c − a (Possibility of
Subtraction).
(β) Multiplication—
(1) (λ + µ)a = (λa) + (µa) (First Distributive Law).
(2) λ(µa) = (λµ)a (Associative Law).
(3) 1 · a = a.
(4) λ(a + b) = (λa) + (λb) (Second Distributive Law).
For rational multipliers λ, µ, the laws (β) follow from the ax-
ioms of addition if multiplication by such factors be defined from
addition. In accordance with the principle of continuity we shall
also make use of them for any arbitrary real numbers, but we pur-
posely formulate them as separate axioms because they cannot be
derived in the general form from the axioms of addition by logical
reasoning alone. By refraining from reducing multiplication to ad-
dition we are enabled through these axioms to banish continuity,
which is so difficult to fix precisely, from the logical structure of
geometry. The law (β) 4 comprises the theorems of similarity.
(γ) The “Axiom of Dimensionality,” which occupies the next
place in the system, will be formulated later.
 EUCLIDEAN SPACE 25

II. Points and Vectors

1. Every pair of points A and B determines a vector a; ex-
−→
pressed symbolically AB = a. If A is any point and a any vector,
−→
there is one and only one point B for which AB = a.
−→ −−→ −→
2. If AB = a, BC = b, then AC = a + b.
In these axioms two fundamental categories of objects occur,
viz. points and vectors; and there are three fundamental relations,
those expressed symbolically by—
−→
a+b=c b = λa AB = a. (1)
All conceptions which may be defined from (1) by logical reasoning
alone belong to affine geometry. The doctrine of affine geometry
is composed of all theorems which can be deduced logically from
the axioms (1), and it can thus be erected deductively on the
axiomatic basis (1) and (2). The axioms are not all logically in-
dependent of one another for the axioms of addition for vectors
(Iα, 2 and 3) follow from those (II) which govern the relations be-
tween points and vectors. It was our aim, however, to make the
vector-axioms I suffice in themselves, so that we should be able to
deduce from them all those facts which involve vectors exclusively
(and not the relations between vectors and points).
From the axioms of addition Iα we may conclude that a definite
vector 0 exists which, for every vector a, satisfies the equation
−→
a + 0 = a. From the axioms II it further follows that AB is
equal to this vector 0 when, and only when, the points A and B
coincide.
If O is a point and e is a vector differing from 0, the end-
points of all vectors OP which have the form ξe (ξ being an arbi-
trary real number) form a straight line. This explanation gives
 CHAPTER I 26

the translational or affine view of straight lines the form of an

exact definition which rests solely upon the fundamental concep-
tions involved in the system of affine axioms. Those points P for
which the abscissa ξ is positive form one-half of the straight line
through O, those for which ξ is negative form the other half. If
we write e1 in place of e, and if e2 is another vector, which is not
−→
of the form ξe1 , then the end-points P of all vectors OP which
have the form ξ1 e1 + ξ2 e2 form a plane E (in this way the plane is
derived affinely by sliding one straight line along another). If we
now displace the plane E along a straight line passing through O
but not lying on E, the plane passes through all space. Accord-
ingly, if e3 is a vector not expressible in the form ξ1 e1 + ξ2 e2 ,
then every vector can be represented in one and only one way as
a linear combination of e1 , e2 , and e3 , viz.
ξ1 e1 + ξ2 e2 + ξ3 e3 .
We thus arrive at the following set of definitions:—
A finite number of vectors e1 , e2 , . . . , eh is said to be linearly
independent if
ξ1 e1 + ξ2 e2 + · · · + ξh eh (2)
only vanishes when all the co-efficients ξ vanish simultaneously.
With this assumption all vectors of the form (2) together con-
stitute a so-called h-dimensional linear vector-manifold (or
simply vector-field); in this case it is the one mapped out by the
vectors e1 , e2 , . . . , eh . An h-dimensional linear vector-manifold M
can be characterised without referring to its particular base e, as
follows:—
(1) The two fundamental operations, viz. addition of two vec-
tors and multiplication of a vector by a number do not transcend
 EUCLIDEAN SPACE 27

the manifold, i.e. the sum of two vectors belonging to M as also

the product of such a vector and any real number also lie in M.
(2) There are h linearly independent vectors in M, but every
h + 1 are linearly dependent on one another.
From the property (2) (which may be deduced from our orig-
inal definition with the help of elementary results of linear equa-
tions) it follows that h, the dimensional number, is as such charac-
teristic of the manifold, and is not dependent on the special vector
base by which we map it out. The dimensional axiom which was
omitted in the above table of axioms may now be formulated.
There are n linearly independent vectors, but every
n + 1 are linearly dependent on one another,
or: The vectors constitute an n-dimensional linear manifold. If
n = 3 we have affine geometry of space, if n = 2 plane geometry,
if n = 1 geometry of the straight line. In the deductive treat-
ment of geometry it will, however, be expedient to leave the value
of n undetermined, and to develop an “n-dimensional geometry”
in which that of the straight line, of the plane, and of space are
included as special cases. For we see (at present for affine ge-
ometry, later on for all geometry) that there is nothing in the
mathematical structure of space to prevent us from exceeding the
dimensional number 3. In the light of the mathematical unifor-
mity of space as expressed in our axioms, its special dimensional
number 3 appears to be accidental, so that a systematic deduc-
tive theory cannot be restricted by it. We shall revert to the idea
of an n-dimensional geometry, obtained in this way, in the next
paragraph.∗ We must first complete the definitions outlined.
If O is an arbitrary point, then the sum-total of all the end-
∗
Vide note 3.
 CHAPTER I 28

points P of vectors, the origin of which is at O and which belong to

an h-dimensional vector field M as represented by (2), occupy fully
an h-dimensional point-configuration. We may, as before,
say that it is mapped out by the vectors e1 , e2 , . . . , eh , which
start from O. The one-dimensional configuration of this type is
called a straight line, the two-dimensional a plane. The point O
does not play a unique part in this linear configuration. If O0 is any
−−→
other point of it, then O0 P traverses the same vector manifold M
if all possible points of the linear aggregate are substituted for P
in turn.
If we measure off all vectors of the manifold M firstly from
the point O and then from any other arbitrary point O0 the two
resulting linear point aggregates are said to be parallel to one
another. The definition of parallel planes and parallel straight
lines is contained in this. That part of the h-dimensional linear
assemblage which results when we measure off all the vectors (2)
from O, subject to the limitation

0 ≤ ξ1 ≤ 1, 0 ≤ ξ2 ≤ 1, ..., 0 ≤ ξh ≤ 1,

will be called the h-dimensional parallelepiped which has its

origin at O and is mapped out by the vectors e1 , e2 , . . . , eh .
(The one-dimensional parallelepiped is called distance, the two-
dimensional one is called parallelogram. None of these concep-
tions is limited to the case n = 3, which is presented in ordinary
experience.)
A point O in conjunction with n linear independent vectors e1 ,
e2 , . . . , eh will be called a co-ordinate system (C). Every vector x
can be presented in one and only one way in the form

x = ξ1 e1 + ξ2 e2 + · · · + ξn en . (3)
 EUCLIDEAN SPACE 29

The numbers ξi will be called its components in the co-ordinate

−→
system (C). If P is any arbitrary point and if OP is equal to
the vector (3), then the ξi are called the co-ordinates of P . All
co-ordinate systems are equivalent in affine geometry. There is no
property of this geometry which distinguishes one from another.
If
O0 | e01 , e02 , . . . , e0n
denote a second co-ordinate system, equations
n
X
e0i = αki ek (4)
k=1

will hold in which the αki form a number system which must have a
non-vanishing determinant (since the e0i are linearly independent).
If ξi are the components of a vector x in the first co-ordinate
system and ξi0 the components of the same vector in the second
co-ordinate system, then the relation
n
X
ξi = αik ξk0 (5)
k=1

holds; this is easily shown by substituting the expressions (4) in

the equation X X
ξi ei = ξi0 e0i .
i i
Let α1 , α2 , . . . , αn be the co-ordinates of O0 in the first co-ordinate
system. If xi are the co-ordinates of any arbitrary point in the first
system and x0i its co-ordinates in the second, the equations
n
X
xi = αik x0k + αi (6)
k=1
 CHAPTER I 30

hold. For xi − αi are the components of

−−0→ −→ −−→0
O P = OP − OO
−−→
in the first system; x0i are the components of O0 P in the second.
Formulæ (6) which give the transformation for the co-ordinates are
thus linear. Those (viz. 5) which transform the vector components
are easily derived from them by cancelling the terms αi which
do not involve the variables. An analytical treatment of affine
geometry is possible, in which every vector is represented by its
components and every point by its co-ordinates. The geometrical
relations between points and vectors then express themselves as
relations between their components and co-ordinates respectively
of such a kind that they are not destroyed by linear arbitrary
transformations.
Formulæ (5) and (6) may also be interpreted in another way.
They may be regarded as a mode of representing an affine trans-
formation in a definite co-ordinate system. A transformation, i.e.
a rule which assigns a vector x0 to every vector x and a point P 0
to every point P , is called linear or affine if the fundamental affine
relations (1) are not disturbed by the transformation: so that if
the relations (1) hold for the original points and vectors they also
hold for the transformed points and vectors:
−− →
a0 + b0 = c0 b0 = λa0 A0 B 0 = a0 − b0

and if in addition no vector differing from 0 transforms into the

vector 0. Expressed in other words this means that two points are
transformed into one and the same point only if they are them-
selves identical. Two figures which are formed from one another
by an affine transformation are said to be affine. From the point
 EUCLIDEAN SPACE 31

of view of affine geometry they are identical. There can be no

affine property possessed by the one which is not possessed by
the other. The conception of linear transformation thus plays
the same part in affine geometry as congruence plays in general
geometry; hence its fundamental importance. In affine transfor-
mations linearly independent vectors become transformed into lin-
early independent vectors again; likewise an h-dimensional linear
configuration into a like configuration; parallels into parallels; a
co-ordinate system O | e1 , e2 , . . . , en into a new co-ordinate sys-
tem O0 | e01 , e02 , . . . , e0n .
Let the numbers αki , αi , have the same meaning as above. The
vector (3) is changed by the affine transformation into

x0 = ξ1 e01 + ξ2 e02 + · · · + ξn e0n .

If we substitute in this the expressions for e0i and use the original
co-ordinate system O | e1 , e2 , . . . , en to picture the affine trans-
formation, then, interpreting ξi as the components of any vector
and ξi0 as the components of its transformed vector,
n
X
ξi0 = αik ξk . (50 )
k=1

−→ −−→
If P becomes P 0 , the vector OP becomes O0 P 0 , and it follows from
this that if xi are the co-ordinates of P and x0i those of P 0 , then
n
X
x0i = αik xk + αi .
k=1

In analytical geometry it is usual to characterise linear con-

figurations by linear equations connecting the co-ordinates of the
 CHAPTER I 32

“current” point (variable). This will be discussed in detail in the

next paragraph. Here we shall just add the fundamental concep-
tion of “linear forms” upon which this discussion is founded. A
function L(x), the argument x of which assumes the value of ev-
ery vector in turn, these values being real numbers only, is called
a linear form, if it has the functional properties
L(a + b) = L(a) + L(b); L(λa) = λ · L(a).
In a co-ordinate system e1 , e2 , . . . , en each of the n vector-
components ξi of x is such a linear form. If x is defined by (3),
then any arbitrary linear form L satisfies
L(x) = ξ1 L(e1 ) + ξ2 L(e2 ) + · · · + ξn L(en ).
Thus if we put L(ei ) = ai , the linear form, expressed in terms of
components, appears in the form
a1 ξ1 + a2 ξ2 + · · · + an ξn (the ai ’s are its constant co-efficients).
Conversely, every expression of this type gives a linear form. A
number of linear forms L1 , L2 , L3 , . . . , Lh are linearly indepen-
dent, if no constants λi exist, for which the identity-equation
holds:
λ1 L1 (x) + λ2 L2 (x) + . . . + λh Lh (x) = 0
except λi = 0. n + 1 linear forms are always linearly inter-
dependent.

§ 3. The Conception of n-dimensional Geometry. Linear

Algebra. Quadratic Forms
To recognise the perfect mathematical harmony underlying the
laws of space, we must discard the particular dimensional number
 EUCLIDEAN SPACE 33

n = 3. Not only in geometry, but to a still more astonishing de-

gree in physics, has it become more and more evident that as soon
as we have succeeded in unravelling fully the natural laws which
govern reality, we find them to be expressible by mathematical
relations of surpassing simplicity and architectonic perfection. It
seems to me to be one of the chief objects of mathematical in-
struction to develop the faculty of perceiving this simplicity and
harmony, which we cannot fail to observe in the theoretical physics
of the present day. It gives us deep satisfaction in our quest for
knowledge. Analytical geometry, presented in a compressed form
such as that I have used above in exposing its principles, conveys
an idea, even if inadequate, of this perfection of form. But not
only for this purpose must we go beyond the dimensional number
n = 3, but also because we shall later require four-dimensional ge-
ometry for concrete physical problems such as are introduced by
the theory of relativity, in which Time becomes added to Space
in a four-dimensional geometry.
We are by no means obliged to seek illumination from the
mystic doctrines of spiritists to obtain a clearer vision of multi-
dimensional geometry. Let us consider, for instance, a homoge-
neous mixture of the four gases, hydrogen, oxygen, nitrogen, and
carbon dioxide. An arbitrary quantum of such a mixture is spec-
ified if we know how many grams of each gas are contained in it.
If we call each such quantum a vector (we may bestow names at
will) and if we interpret addition as implying the union of two
quanta of the gases in the ordinary sense, then all the axioms I
of our system referring to vectors are fulfilled for the dimensional
number n = 4, provided we agree also to talk of negative quanta of
gas. One gram of pure hydrogen, one gram of oxygen, one gram
of nitrogen, and one gram of carbon dioxide are four “vectors,”
 CHAPTER I 34

independent of one another from which all other gas quanta may
be built up linearly; they thus form a co-ordinate system. Let
us take another example. We have five parallel horizontal bars
upon each of which a small bead slides. A definite condition of
this primitive “adding-machine” is defined if the position of each
of the five beads upon its respective rod is known. Let us call such
a condition a “point” and every simultaneous displacement of the
five beads a “vector,” then all of our axioms are satisfied for the
dimensional number n = 5. From this it is evident that construc-
tions of various types may be evolved which, by an appropriate
disposal of names, satisfy our axioms. Infinitely more important
than these somewhat frivolous examples is the following one which
shows that our axioms characterise the basis of our opera-
tions in the theory of linear equations. If αi and α are given
numbers,
α 1 x1 + α 2 x2 + . . . + α n xn = 0 (7)
is usually called a homogeneous linear equation in the un-
knowns xi , whereas
α1 x1 + α2 x2 + . . . + αn xn = α (8)
is called a non-homogeneous linear equation. In treating the
theory of linear homogeneous equations, it is found useful to have
a short name for the system of values of the variables xi ; we shall
call it “vector”. In carrying out calculations with these vectors, we
shall define the sum of the two vectors
(a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn )
to be the vector
(a1 + b1 , a2 + b2 , . . . , an + bn )
 EUCLIDEAN SPACE 35

and λ times the first vector to be

(λa1 , λa2 , . . . , λan ).

The axioms I for vectors are then fulfilled for the dimensional
number n.

e1 = (1, 0, 0, . . . , 0),
e2 = (0, 1, 0, . . . , 0),
..................
en = (0, 0, 0, . . . , 1)

form a system of independent vectors. The components of any

arbitrary vector (x1 , x2 , . . . , xn ) in this co-ordinate system are the
numbers xi themselves. The fundamental theorem in the solution
of linear homogeneous equations may now be stated thus:—

if L1 (x), L2 (x), ..., Lh (x)

are h linearly independent linear forms, the solutions x of the

equations

L1 (x) = 0, L2 (x) = 0, ..., Lh (x) = 0

form an (n − h)-dimensional linear vector manifold.

In the theory of non-homogeneous linear equations we shall
find it advantageous to denote a system of values of the variables xi
a “point”. If xi and x0i are two systems which are solutions of
equation (8), their difference

x01 − x1 , x02 − x2 , . . . , x0n − xn

 CHAPTER I 36

is a solution of the corresponding homogeneous equation (7). We

shall, therefore, call this difference of two systems of values of
the variables xi a “vector,” viz. the “vector” defined by the two
“points” (xi ) and (x0i ); we make the above conventions for the
addition and multiplication of these vectors. All the axioms
then hold. In the particular co-ordinate system composed of the
vectors ei given above, and having the “origin” O = (0, 0, . . . , 0),
the co-ordinates of a point (xi ) are the numbers xi themselves.
The fundamental theorem concerning linear equations is: those
points which satisfy h independent linear equations, form a point-
configuration of n − h dimensions.
In this way we should not only have arrived quite naturally
at our axioms without the help of geometry by using the theory
of linear equations, but we should also have reached the wider
conceptions which we have linked up with them. In some ways,
indeed, it would appear expedient (as is shown by the above for-
mulation of the theorem concerning homogeneous equations) to
build up the theory of linear equations upon an axiomatic basis
by starting from the axioms which have here been derived from
geometry. A theory developed along these lines would then hold
for any domain of operations, for which these axioms are fulfilled,
and not only for a “system of values in n variables”. It is easy to
pass from such a theory which is more conceptual, to the usual one
of a more formal character which operates from the outset with
numbers xi by taking a definite co-ordinate system as a basis, and
then using in place of vectors and points their components and
co-ordinates respectively.
It is evident from these arguments that the whole of affine
geometry merely teaches us that space is a region of three di-
mensions in linear quantities (the meaning of this statement
 EUCLIDEAN SPACE 37

will be sufficiently clear without further explanation). All the

separate facts of intuition which were mentioned in § 1 are simply
disguised forms of this one truth. Now, if on the one hand it is
very satisfactory to be able to give a common ground in the the-
ory of knowledge for the many varieties of statements concerning
space, spatial configurations, and spatial relations which, taken
together, constitute geometry, it must on the other hand be em-
phasised that this demonstrates very clearly with what little right
mathematics may claim to expose the intuitional nature of space.
Geometry contains no trace of that which makes the space of in-
tuition what it is in virtue of its own entirely distinctive qualities
which are not shared by “states of addition-machines” and “gas-
mixtures” and “systems of solutions of linear equations”. It is left
to metaphysics to make this “comprehensible” or indeed to show
why and in what sense it is incomprehensible. We as mathemati-
cians have reason to be proud of the wonderful insight into the
knowledge of space which we gain, but, at the same time, we must
recognise with humility that our conceptual theories enable us to
grasp only one aspect of the nature of space, that which, moreover,
is most formal and superficial.
To complete the transition from affine geometry to complete
metrical geometry we yet require several conceptions and facts
which occur in linear algebra and which refer to bilinear and
quadratic forms. A function Q(x, y) of two arbitrary vectors x
and y is called a bilinear form if it is a linear form in x as well
as in y. If in a certain co-ordinate system ξi are the components
of x, ηi those of y, then an equation
n
X
Q(x, y) = aik ξi ηk
i,k=1
 CHAPTER I 38

with constant co-efficients aik holds. We shall call the form “non-
degenerate” if it vanishes identically in y only when the vector
x = 0. This happens when, and only when, the homogeneous
equations
Xn
aik ξi = 0
i=1

have a single solution ξi = 0 or when the determinant |aik | = 6 0.

From the above explanation it follows that this condition, viz.
the non-vanishing of the determinant, persists for arbitrary lin-
ear transformations. The bilinear form is called symmetrical if
Q(y, x) = Q(x, y). This manifests itself in the co-efficients by the
symmetrical property aki = aik . Every bilinear form Q(x, y) gives
rise to a quadratic form which depends on only one variable
vector x n
X
Q(x) = Q(x, x) = aik ξi ξk .
i,k=1

In this way every quadratic form is derived in general from

one, and only one, symmetrical bilinear form. The quadratic
form Q(x) which we have just formed may also be produced from
the symmetrical form
1

2
Q(x, y) + Q(y, x)

by identifying x with y.
To prove that one and the same quadratic form cannot arise
from two different symmetrical bilinear forms, one need merely
show that a symmetrical bilinear form Q(x, y) which satisfies the
equation Q(x, x) identically for x, vanishes identically. This, how-
ever, immediately results from the relation which holds for every
 EUCLIDEAN SPACE 39

symmetrical bilinear form

Q(x + y, x + y) = Q(x, x) + 2Q(x, y) + Q(y, y). (9)
If Q(x) denotes any arbitrary quadratic form then Q(x, y) is al-
ways to signify the symmetrical bilinear form from which Q(x) is
derived (to avoid mentioning this in each particular case). When
we say that a quadratic form is non-degenerate we wish to convey
that the above symmetrical bilinear form is non-degenerate. A
quadratic form is positive definite if it satisfies the inequality
Q(x) > 0 for every value of the vector x 6= 0. Such a form is cer-
tainly non-degenerate, for no value of the vector x 6= 0 can make
Q(x, y) vanish identically in y, since it gives a positive result for
y = x.

§ 4. The Foundations of Metrical Geometry

To bring about the transition from affine to metrical geometry
we must once more draw from the fountain of intuition. From it
we obtain for three-dimensional space the definition of the scalar
product of two vectors a and b. After selecting a definite vector
as a unit we measure out the length of a and the length (negative
or positive as the case may be) of the perpendicular projection of b
upon a and multiply these two numbers with one another. This
means that the lengths of not only parallel straight lines may be
compared with one another (as in affine geometry) but also such
as are arbitrarily inclined to one another. The following rules hold
for scalar products:—
λa · b = λ(a · b) (a + a0 ) · b = (a · b) + (a0 · b)
and analogous expressions with reference to the second factor; in
addition, the commutative law a · b = b · a. The scalar product
 CHAPTER I 40

of a with a itself, viz. a · a = a2 , is always positive except when

a = 0, and is equal to the square of the length of a. These
laws signify that the scalar product of two arbitrary vectors, i.e.
x · y is a symmetrical bilinear form, and that the quadratic form
which arises from it is positive definite. We thus see that not the
length, but the square of the length of a vector depends in a simple
rational way on the vector itself; it is a quadratic form. This is
the real content of Pythagoras’ Theorem. The scalar product is
nothing more than the symmetrical bilinear form from which this
quadratic form has been derived. We accordingly formulate the
following:—
Metrical Axiom: If a unit vector e, differing from zero, be
chosen, every two vectors x and y uniquely determine a number
(x·y) = Q(x, y); the latter, being dependent on the two vectors, is
a symmetrical bilinear form. The quadratic form (x · x) = Q(x)
which arises from it is positive definite. Q(e) = 1.
We shall call Q the metrical groundform. We then have that
an affine transformation which, in general, transforms the vector x
into x0 is a congruent one if it leaves the metrical groundform
unchanged:—
Q(x0 ) = Q(x). (10)
Two geometrical figures which can be transformed into one another
by a congruent transformation are congruent.∗ The conception of
congruence is defined in our axiomatic scheme by these state-
ments. If we have a domain of operation in which the axioms
of § 2 are fulfilled, we can choose any arbitrary positive definite
∗
We take no notice here of the difference between direct congruence and
mirror congruence (lateral inversion). It is present even in affine transforma-
tions, in n-dimensional space as well as 3-dimensional space.
 EUCLIDEAN SPACE 41

quadratic form in it, “promote” it to the position of a fundamen-

tal metrical form, and, using it as a basis, define the conception
of congruence as was just now done. This form then endows the
affine space with metrical properties and Euclidean geometry in
its entirety now holds for it. The formulation at which we have
arrived is not limited to any special dimensional number.
It follows from (10), in virtue of relation (9) of § 3, that for a
congruent transformation the more general relation

Q(x0 , y0 ) = Q(x, y)

holds.
Since the conception of congruence is defined by the metrical
groundform it is not surprising that the latter enters into all for-
mulæ which concern the measure of geometrical quantities. Two
vectors a and a0 are congruent if, and only if,

Q(a) = Q(a0 ).

We could accordingly introduce Q(a) as a measure of the vec-

tor a. Instead of doing this, however, we shall use the positive
square root of Q(a) for this purpose and call it the length of the
vector a (this we shall adopt as our definition) so that the further
condition is fulfilled that the length of the sum of two parallel
vectors pointing in the same direction is equal to the sum of the
lengths of the two single vectors. If a, b as well as a0 , b0 are two
pairs of vectors, all of length unity, then the figure formed by the
first two is congruent with that formed by the second pair, if, and
only if, Q(a, b) = Q(a0 , b0 ).
In this case again we do not introduce the number Q(a, b)
itself as a measure of the angle, but a number θ which is related
 CHAPTER I 42

to it by the transcendental function cosine thus:—

cos θ = Q(a, b)
so as to be in agreement with the theorem that the numerical
measure of an angle composed of two angles in the same plane is
the sum of the numerical values of these angles. The angle which
is formed from any two arbitrary vectors a and b (6= 0) is then
calculated from
Q(a, b)
cos θ = p . (11)
Q(a, a) · Q(b, b)
In particular, two vectors a, b are said to be perpendicular to
one another if Q(a, b) = 0. This reminder of the simplest metrical
formulæ of analytical geometry will suffice.
The angle defined by (11) which has been formed by two vec-
tors is shown always to be real by the inequality
Q2 (a, b) ≤ Q(a) · Q(b) (12)
which holds for every quadratic form Q which is ≥ 0 for all values
of the argument. It is most simply deduced by forming
Q(λa + µb) = λ2 Q(a) + 2λµQ(a, b) + µ2 Q(b) ≥ 0.
Since this quadratic form in λ and µ cannot assume both positive
and negative values its “discriminant” Q2 (a, b)−Q(a)·Q(b) cannot
be positive.
A number, n, of independent vectors form a Cartesian co-
ordinate system if for every vector
x = x1 e1 + x2 e2 + . . . + xn en
Q(x) = x21 + x22 + . . . + x2n (13)
 EUCLIDEAN SPACE 43

holds, i.e. if (
1 (i = k),
Q(ei , ej ) =
0 (i =
6 k).
From the standpoint of metrical geometry all co-ordinate sys-
tems are of equal value. A proof (appealing directly to our geo-
metrical sense) of the theorem that such systems exist will now
be given not only for a “definite” but also for any arbitrary non-
degenerate quadratic form, inasmuch as we shall find later in the
theory of relativity that it is just the “indefinite” case that plays
the decisive rôle. We enunciate as follows:—
Corresponding to every non-degenerate quadratic form Q a co-
ordinate system ei can be introduced such that

Q(x) = 1 x21 + 2 x22 + · · · + n x2n (i = ±1). (14)

Proof.—Let us choose any arbitrary vector e1 for which

Q(e1 ) 6= 0. By multiplying it by an appropriate positive con-
stant we can arrange so that Q(e1 ) = ±1. We shall call a vector x
for which Q(e1 , x) = 0 orthogonal to e1 . If x∗ is a vector which
is orthogonal to e1 , and if x1 is any arbitrary number, then

x = x 1 e 1 + x∗ (15)

satisfies Pythagoras’ Theorem:—

Q(x) = x21 Q(e1 ) + 2x1 Q(e1 , x∗ ) + Q(x∗ ) = ±x21 + Q(x∗ ).

The vectors orthogonal to e1 constitute an (n − 1)-dimensional

linear manifold, in which Q(x) is a non-degenerate quadratic
form. Since our theorem is self-evident for the dimensional
number n = 1, we may assume that it holds for (n − 1) dimensions
 CHAPTER I 44

(proof by successive induction from the case (n − 1) to that

of n). According to this, n − 1 vectors e2 , . . . , en , orthogonal
to e1 exist, such that for

x∗ = x2 e2 + · · · + xn en

the relation
Q(x∗ ) = ±x22 ± . . . ± x2n
holds. This enables Q(x) to be expressed in the required form.
Then
Q(ei ) = i Q(ei , ek ) = 0 (i =
6 k).
These relations result in all the ei ’s being independent of one
another and in each vector x being representable in the form (13).
They give
xi = i · Q(ei , x). (16)
An important corollary is to be made in the “indefinite” case.
The numbers r and s attached to the i ’s, and having positive
and negative signs respectively, are uniquely determined by the
quadratic form: it may be said to have r positive and s negative
dimensions. (s may be called the inertial index of the quadratic
form, and the theorem just enunciated is known by the name
“Law of Inertia”. The classification of surfaces of the second or-
der depends on it.) The numbers r and s may be characterised
invariantly thus:—
There are r mutually orthogonal vectors e, for which Q(e) > 0;
but for a vector x which is orthogonal to these and not equal to 0,
it necessarily follows that Q(x) < 0. Consequently there cannot
be more than r such vectors. A corresponding theorem holds for s.
 EUCLIDEAN SPACE 45

r vectors of the required type are given by those r fundamen-

tal vectors ei of the co-ordinate system upon which the expres-
sion (14) is founded, to which the positive signs i correspond.
The corresponding components xi (i = 1, 2, 3, . . . , r) are definite
linear forms of x [cf. (16)]: xi = Li (x). If, now, ei (i = 1, 2, . . . , r)
is any system of vectors which are mutually orthogonal to one
another, and satisfy the condition Q(ei ) > 0, and if x is a vector
orthogonal to these ei , we can set up a linear combination

y = λ1 e1 + . . . + λr er + µx

in which not all the co-efficients vanish and which satisfies the
r homogeneous equations

L1 (y) = 0, ..., Lr (y) = 0.

It is then evident from the form of the expression that Q(y) must
be negative unless y = 0. In virtue of the formula

Q(y) − λ21 Q(e1 ) + · · · + λ2r Q(er ) = µ2 Q(x)



it then follows that Q(x) < 0 except in the case in which if y = 0,

λ1 = · · · = λr also = 0. But then, by hypothesis, µ must 6= 0, i.e.
x = 0.
In the theory of relativity the case of a quadratic form with one
negative and n − 1 positive dimensions becomes important. In three-
dimensional space, if we use affine co-ordinates,

−x21 + x22 + x23 = 0

is the equation of a cone having its vertex at the origin and consisting
of two sheets, as expressed by the negative sign of x21 , which are only
 CHAPTER I 46

connected with one another at the origin of co-ordinates. This division

into two sheets allows us to draw a distinction between past and future
in the theory of relativity. We shall endeavour to describe this by an
elementary analytical method here instead of using characteristics of
continuity.
Let Q be a non-degenerate quadratic form having only one nega-
tive dimension. We choose a vector, for which Q(e) = −1. We shall
call these vectors x, which are not zero and for which Q(x) ≤ 0 “neg-
ative vectors”. According to the proof just given for the Theorem of
Inertia, no negative vector can satisfy the equation Q(e, x) = 0. Neg-
ative vectors thus belong to one of two classes or “cones” according as
Q(e, x) < 0 or > 0; e itself belongs to the former class, −e to the latter.
A negative vector x lies “inside” or “on the sheet” of its cone according
as Q(x) < 0 or = 0. To show that the two cones are independent of the
choice of the vector e, one must prove that, from Q(e) = Q(e0 ) = −1,
Q(e0 , x)
and Q(x) ≤ 0, it follows that the sign of is the same as that
Q(e, x)
of −Q(e, e0 ).
Every vector x can be resolved into two summands

x = xe + x∗

such that the first is proportional and the second (x∗ ) is orthogonal
to e. One need only take x = −Q(e, x) and we then get

Q(x) = −x2 + Q(x∗ );

Q(x∗ ) is, as we know, necessarily ≥ 0. Let us denote it by Q∗ .

The equation

Q∗ = x2 + Q(x) = Q2 (e, x) + Q(x)

then shows that Q∗ is a quadratic form (degenerate), which satisfies

 EUCLIDEAN SPACE 47

the identity or inequality, Q∗ (x) ≥ 0. We now have

Q(x) = −x2 + Q∗ (x) ≤ 0, Q(e0 ) = −e02 + Q∗ (e0 ) < 0,

{x = −Q(e, x)}; {e0 = −Q(e, e)}.

From the inequality (12) which holds for Q∗ , it follows that

 ∗ 0 2
Q (e , x) ≤ Q∗ (e0 ) · Q∗ (x) < e02 x2 ;

consequently
−Q(e0 , x) = e0 x − Q∗ (e0 , x)
has the same sign as the first summand e0 x.
Let us now revert to the case of a definitely positive metrical
groundform with which we are at present concerned. If we use a
Cartesian co-ordinate system to represent a congruent transfor-
mation, the co-efficients of transformation αik in formula (50 ), § 2,
will have to be such that the equation

ξ102 + ξ202 + · · · + ξn02 = ξ12 + ξ22 + · · · + ξn2

is identically satisfied by the ξ’s. This gives the “conditions for

orthogonality” (
n
X 1 (i = j),
αri αrj = (17)
r=1
0 (i 6= j).
They signify that the transition to the inverse transformation con-
verts the co-efficients αik into αki :—
n
X
ξi = αki ξk0 .
k=1
 CHAPTER I 48

It furthermore follows that the determinant ∆ = |αik | of a congru-

ent transformation is identical with that of its inverse, and since
their product must equal 1, ∆ = ±1. The positive or the negative
sign would occur according as the congruence is real or inverted
as in a mirror (“lateral inversion”).
Two possibilities present themselves for the analytical treat-
ment of metrical geometry. Either one imposes no limitation
upon the affine co-ordinate system to be used: the problem is
then to develop a theory of invariance with respect to arbitrary
linear transformations, in which, however, in contra-distinction to
the case of affine geometry, we have a definite invariant quadratic
form, viz. the metrical groundform
n
X
Q(x) = gik ξi ξk
i,k=1

once and for all as an absolute datum. Or, we may use Carte-
sian co-ordinate systems from the outset: in this case, we are
concerned with a theory of invariance for orthogonal transforma-
tions, i.e. linear transformations, in which the co-efficients satisfy
the secondary conditions (17). We must here follow the first course
so as to be able to pass on later to generalisations which extend
beyond the limits of Euclidean geometry. This plan seems ad-
visable from the algebraic point of view, too, since it is easier to
gain a survey of those expressions which remain unchanged for all
linear transformations than of those which are only invariant for
orthogonal transformations (a class of transformations which are
subjected to secondary limitations not easy to define).
We shall here develop the Theory of Invariance as a “Tensor
Calculus” along lines which will enable us to express in a conve-
 EUCLIDEAN SPACE 49

nient mathematical form, not only geometrical laws, but also all
physical laws.

§ 5. Tensors
Two linear transformations,

αki ξ¯k ,
X
ξi = |αki | =


6 0 (18)
k
X
ᾰik η̄k , |ᾰik | = (180 )


ηi = 6 0
k

¯ η̄
in the variables ξ and η respectively, leading to the variables ξ,
are said to bePcontra-gredient to one another, if they make the
bilinear form i ηi ξ i transform into itself, i.e.

η̄i ξ¯i .
X X
ηi ξ i = (19)
i i

Contra-gredience is thus a reversible relationship. If the variables

ξ, η are transformed into ξ,¯ η̄ by one pair of contra-gredient trans-
formations A, Ă, and then ξ, ¯ η̄ into ξ, ¯ η̄¯ by a second pair B, B̆ it
follows from
η̄i ξ¯i = η̄¯i ξ¯i
X X X
ηi ξ i =
i i i

that the two transformations combined, which transform ξ directly

¯ and η into η̄¯ are likewise contra-gredient. The co-efficients
into ξ,
of two contra-gredient substitutions satisfy the conditions
(
X 1 (i = k),
αir ᾰrk = δik = (20)
r
0 (i =
6 k).
 CHAPTER I 50

If we substitute for the ξ’s in the left-hand member of (19) their

values in terms of ξ¯ obtained from (18), it becomes evident that
the equations (180 ) are derived by reduction from
X
η̄i = αik ηk . (21)
k

There is thus one and only one contra-gredient transformation

corresponding to every linear transformation. For the same reason
as (21)
ξ¯i =
X
ᾰki ξ k
k

holds. By substituting these expressions and (21) in (19), we find

that the co-efficients, in addition to satisfying the conditions (20),
satisfy X
αri ᾰkr = δki .
r
An orthogonal transformation is one which is contra-gredient to it-
self. If we subject a linear form in the variables ξi to any arbitrary
linear transformation the co-efficients become transformed contra-
grediently to the variables, or they assume a “contra-variant” re-
lationship to these, as it is sometimes expressed.
In an affine co-ordinate system O | e1 , e2 , . . . , en we have up to
the present characterised a displacement x by the uniquely defined
components ξ i given by the equation
x = ξ 1 e1 + ξ 2 e2 + · · · + ξ n en .
If we pass over into another affine co-ordinate system Ō | ē1 ,
ē2 , . . . , ēn , whereby X
ēi = αik ek ,
k
 EUCLIDEAN SPACE 51

the components of x undergo the transformation

αki ξ¯k
X
ξi =
k

as is seen from the equation

ξ¯i ēi .
X X
x= ξ i ei =
i i

These components thus transform themselves contra-grediently

to the fundamental vectors of the co-ordinate system, and are
related contra-variantly to them; they may thus be more precisely
termed the contra-variant components of the vector x. In
metrical space, however, we may also characterise a displace-
ment in relation to the co-ordinate system by the values of its
scalar product with the fundamental vectors ei of the co-ordinate
system
ξi = (x · ei ).
In passing over into another co-ordinate system these quanti-
ties transform themselves—as is immediately evident from their
definition—“co-grediently” to the fundamental vectors (just like
the latter themselves), i.e. in accordance with the equations

ξ¯i =
X
αik ξk ;
k

they behave “co-variantly”. We shall call them the co-variant

components of the displacement. The connection between co-
variant and contra-variant components is given by the formulæ
X X
ξi = (ei · ek )ξ k = gik ξ k (22)
k k
 CHAPTER I 52

or by their inverses (which are derived from them by simple reso-

lution) respectively X
ξi = g ik ξk . (220 )
k

In a Cartesian co-ordinate system the co-variant components coin-

cide with the contra-variant components. It must again be empha-
sised that the contra-variant components alone are at our disposal
in affine space, and that, consequently, wherever in the following
pages we speak of the components of a displacement without spec-
ifying them more closely, the contra-variant ones are implied.
Linear forms of one or two arbitrary displacements have al-
ready been discussed above. We can proceed from two argu-
ments to three or more. Let us take, for example, a trilinear form
A(x, y, z). If in an arbitrary co-ordinate system we represent the
two displacements x, y by their contra-variant components, z by
its co-variant components, i.e. ξ i , η i , and ζi respectively, then A is
algebraically expressed as a trilinear form of these three series of
variables with definite number-co-efficients
X
alik ξ i η k ζl . (23)
i,j,k

Let the analogous expression in a different co-ordinate system,

indicated by bars, be
ālik ξ¯i η̄ k ζ̄l .
X
(230 )
i,j,k

A connection between the two algebraic trilinear forms (23)

and (230 ) then exists, by which the one resolves into the other if
the two series of variables ξ, η are transformed contra-grediently
to the fundamental vectors, but the series ζ co-grediently to the
 EUCLIDEAN SPACE 53

latter. This relationship enables us to calculate the co-efficient ālik

of A in the second co-ordinate system if the co-efficients alik
and also the transformation co-efficient αik leading from one
co-ordinate system to the other are known. We have thus arrived
at the conception of the “r-fold co-variant, s-fold contra-variant
tensor of the (r + s)th degree”: it is not confined to metrical
geometry but only assumes the space to be affine. We shall now
give an explanation of this tensor in abstracto. To simplify our
expressions we shall take special values for the numbers r and s
as in the example quoted above: r = 2, s = l, r + s = 3. We
then enunciate:—
A trilinear form of three series of variables which is depen-
dent on the co-ordinate system is called a doubly co-variant, singly
contra-variant tensor of the third degree if the above relationship
is as follows. The expressions for the linear form in any two co-
ordinate systems, viz.:—
ālik ξ¯i η̄ k ζ̄l
X X
alik ξ i η k ζl ,

resolve into one another, if two of the series of variables (viz.

the first two ξ and η) are transformed contra-grediently to the
fundamental vectors of the co-ordinate system and the third co-
grediently to the same. The co-efficients of the linear form are
called the components of the tensor in the co-ordinate system in
question. Furthermore, they are called co-variant in the indices,
i, k, which are associated with the variables to be transformed
contra-grediently, and contra-variant in the others (here only the
one index l).
The terminology is based upon the fact that the co-efficients
of a uni-linear form behave co-variantly if the variables are trans-
formed contra-grediently, but contra-variantly if they are trans-
 CHAPTER I 54

formed co-grediently. Co-variant indices are always attached as

suffixes to the co-efficients, contra-variant ones written at the top
of the co-efficients. Variables with lowered indices are always to
be transformed co-grediently to the fundamental vectors of the co-
ordinate system, those with raised indices are to be transformed
contra-grediently to the same. A tensor is fully known if its com-
ponents in a co-ordinate system are given (assuming, of course,
that the co-ordinate system itself is given); these components may,
however, be prescribed arbitrarily. The tensor calculus is con-
cerned with setting out the properties and relations of tensors,
which are independent of the co-ordinate system. In an extended
sense a quantity in geometry and physics will be called a tensor
if it defines uniquely a Linear algebraic form depending on the
co-ordinate system in the manner described above; and conversely
the tensor is fully characterised if this form is given. For example,
a little earlier we called a function of three displacements which
depended linearly and homogeneously on each of their arguments
a tensor of the third degree—one which is twofold co-variant and
singly contra-variant. This was possible in metrical space. In
this space, indeed, we are at liberty to represent this quantity by
a “none” fold, single, twofold or threefold co-variant tensor. In
affine space, however, we should only have been able to express it
in the last form as a co-variant tensor of the third degree.
We shall illustrate this general explanation by some examples
in which we shall still adhere to the standpoint of affine geometry
alone.
1. If we represent a displacement a in an arbitrary co-ordinate
system by its (contra-variant) components ai and assign to it the
linear form
a1 ξ1 + a2 ξ2 + · · · + an ξn
 EUCLIDEAN SPACE 55

having the variables ξi in this co-ordinate system, we get a contra-

variant tensor of the first order.
From now on we shall no longer use the term “vector” as being
synonymous with “displacement” but to signify a “tensor of the
first order,” so that we shall say, displacements are contra-
variant vectors. The same applies to the velocity of a moving
point, for it is obtained by dividing the infinitely small displace-
ment which the moving point suffers during the time-element dt
by dt (in the limiting case when dt → 0). The present use of
the word vector agrees with its usual significance which includes
not only displacements but also every quantity which, after the
choice of an appropriate unit, can be represented uniquely by a
displacement.
2. It is usually claimed that force has a geometrical character
on the ground that it may be represented in this way. In op-
position, however, to this representation there is another which,
we nowadays consider, does more justice to the physical nature
of force, inasmuch as it is based on the conception of work. In
modern physics the conception work is gradually usurping the con-
ception of force, and is claiming a more decisive and fundamental
rôle. We shall define the components of a force in a co-ordinate
system O | ei to be those numbers pi which denote how much work
it performs during each of the virtual displacements ei of its point
of application. These numbers completely characterise the force.
The work performed during the arbitrary displacement

x = ξ 1 e1 + ξ 2 e2 + · · · + ξ n en

of its point of application is then = i pi ξ i . Hence it follows that

P
 CHAPTER I 56

for two definite co-ordinate systems the relation

p̄i ξ¯i
X X
pi ξ i =
i i

holds, if the variables ξ i , as signified by the upper indices, are

transformed contra-grediently with respect to the co-ordinate sys-
tem. According to this view, then, forces are co-variant vec-
tors. The connection between this representation of forces and
the usual one in which they are displacements will be discussed
when we pass from affine geometry, with which we are at present
dealing, to metrical geometry. The components of a co-variant
vector become transformed co-grediently to the fundamental vec-
tors in passing to a new co-ordinate system.
Additional Remarks.—Since the transformations of the
components ai of a co-variant vector and of the components bi
of
P a contra-variant vector are contra-gredient to one another,
i
i ai b is a definite number which is defined by these two vectors
and is independent of the co-ordinate system. This is our first
example of an invariant tensor operation. Numbers or scalars are
to be classified as tensors of zero order in the system of tensors.
It has already been explained under what conditions a bilinear
form of two series of variables is called symmetrical and what
makes a symmetrical bilinear form non-degenerate. A bilinear
form F (ξ, η) is called skew-symmetrical if the interchange of
the two sets of variables converts it into its negative, i.e. merely
changes its sign
F (η, ξ) = −F (ξ, η).
This property is expressed in the co-efficients aik by the equations
aki = −aik . These properties persist if the two sets of variables
 EUCLIDEAN SPACE 57

are subjected to the same linear transformations. The property of

being skew-symmetrical, symmetrical or (symmetrical and) non-
degenerate, possessed by co-variant or contra-variant tensors of
the second order is thus independent of the co-ordinate system.
Since the bilinear unit form resolves into itself after a contra-
gredient transformation of the two series of variables there is
among the mixed tensors of the second order (i.e. those which
are simply co-variant and simply contra-variant) one, called the
1 (i = k)
unit tensor, which has the components δik = in every
0 (i 6= k)
co-ordinate system.
3. The metrical structure underlying Euclidean space assigns
to every two displacements
X X
x= ξ i ei y= η i ei
i i

a number which is independent of the co-ordinate system and is

their scalar product
X
x·y = gik ξ i η k gik = (ei · ek ).
i,k

Hence the bilinear form on the right depends on the co-ordinate

system in such a way that a co-variant tensor of the second order
is given by it, viz. the fundamental metrical tensor. The
metrical structure is fully characterised by it. It is symmetrical
and non-degenerate.
4. A linear vector transformation makes any displace-
ment x correspond linearly to another displacement, x0 , i.e. so that
the sum x0 + y0 corresponds to the sum x + y and the product λx0
 CHAPTER I 58

to the product λx. In order to be able to refer conveniently to

such linear vector transformations, we shall call them matrices.
If the fundamental vectors ei of a co-ordinate system become
X
e0i = αik ek
k

as a result of the transformation it will in general convert the

arbitrary displacement
X X X
x= ξ i ei into x0 = ξ i e0i = αik ξ i ek . (24)
i i i,k

We may, therefore, characterise the matrix in the particular co-

ordinate system chosen by the bilinear form
X
aki ξ i ηk .
i,k

It follows from (24) that the relation

āki ξ¯i ek =
X X
aki ξ i ek (= x0 )
i,k i,k

holds between two co-ordinate systems (we have used the same
terminology as above) if

ξ¯i ēi =
X X
ξ i ei (= x);
i i

thus
āki ξ¯i η̄k =
X X
aki ξ i ηk
i,k i,k
 EUCLIDEAN SPACE 59

if the η i are transformed co-grediently to the fundamental vectors

and the ξ i are transformed contra-grediently to them (the latter
remark about the transformations of the variables is self-evident
so that in future we shall simply omit it in similar cases). In this
way matrices are represented as tensors of the second order. In
particular, the unit tensor corresponds to “identity” which assigns
to every displacement x itself.
As was shown in the examples of force and metrical space it
often happens that the representation of geometrical or physical
quantities by a tensor becomes possible only after a unit measure
has been chosen: this choice can only be made by specifying it in
each particular case. If the unit measure is altered the represen-
tative tensors must be multiplied by a universal constant, viz. the
ratio of the two units of measure.
The following criterion is manifestly equivalent to this expo-
sition of the conception tensor. A linear form in several series
of variables, which is dependent on the co-ordinate system, is
a tensor if in every case it assumes a value independent of the
co-ordinate system (a) whenever the components of an arbitrary
contra-variant vector are substituted for every contra-gredient se-
ries of variables, and (b) whenever the components of an arbitrary
co-variant vector are substituted for a co-gredient series.
If we now return from affine to metrical geometry, we see
from the arguments at the beginning of the paragraph that the
difference between co-variants and contra-variants which affects
the tensors themselves in affine geometry shrinks to a mere differ-
ence in the mode of representation.
Instead of talking of co-variant, mixed, and contra-variant ten-
sors we shall hence find it more convenient here to talk only of
the co-variant, mixed, and contra-variant components of a ten-
 CHAPTER I 60

sor. After the above remarks it is evident that the transition

from one tensor to another which has a different character of co-
variance may be formulated simply as follows. If we interpret the
contra-gredient variables in a tensor as the contra-variant compo-
nents of an arbitrary displacement, and the co-gredient variables
as co-variant components of an arbitrary displacement, the tensor
becomes transformed into a linear form of several arbitrary dis-
placements which is independent of the co-ordinate system. By
representing the arguments in terms of their co-variant or contra-
variant components in any way which suggests itself as being ap-
propriate we pass on to other representations of the same ten-
sor. From the purely algebraic point of view the conversion of a
co-variant index into a contra-variant one is performed by sub-
stituting new ξi ’s for the corresponding variables ξ i in the linear
form in accordance with (22). The invariant nature of this pro-
cess depends on the circumstance that this substitution transforms
contra-gredient variables into co-gredient ones. The converse pro-
cess is carried out according to the inverse equations(220 ). The
components themselves are changed (on account of the symmetry
of the gik ’s) from contra-variants to co-variants, i.e. the indices
are “lowered” according to the rule:
X
Substitute ai = gij aj for ai
j

irrespective of whether the numbers ai carry any other indices or

not: the raising of the index is effected by the inverse equations.
If, in particular, we apply these remarks to the fundamental
metrical tensor, we get
X X X X
gik ξ i η k = ξ i ηi = ξk η k = g ik ξi ηk .
i,k i k i,k
 EUCLIDEAN SPACE 61

Thus its mixed components are the numbers δki , its contra-variant
components are the co-efficients g ik of the equations (220 ), which
are the inverse of (22). It follows from the symmetry of the tensor
that these as well as the gik ’s satisfy the condition of symmetry
g ki = g ik .
With regard to notation we shall adopt the convention of de-
noting the co-variant, mixed, and contra-variant components of
the same tensor by similar letters, and of indicating by the posi-
tion of the index at the top or bottom respectively whether the
components are contra-variant or co-variant with respect to the
index, as is shown in the following example of a tensor of the
second order:
X X X X
aik ξ i η k = aik ξi η k = aki ξ i ηk = aik ξi ηk
i,k i,k i,k i,k

(in which the variables with lower and upper indices are connected
in pairs by (22)).
In metrical space it is clear, from what has been said, that
the difference between a co-variant and a contra-variant vector
disappears: in this case we can represent a force, which, according
to our view, is by nature a co-variant vector, as a contra-variant
vector, too, i.e. by aP displacement. For, as we represented it above
by the linear form i pi ξ i in the contra-gredient variables ξ i , we
can now transform the latterPby means of (220 ) into one having
co-gredient variables ξi , viz. i pi ξi . We then have
X X X X
pi ξi = gik pi ξ k = gik pk ξ i = pi ξ i ;
i i,k i,k i

the representative displacement p is thus defined by the fact that

the work which the force performs during an arbitrary displace-
ment is equal to the scalar product of the displacements p and x.
 CHAPTER I 62

In a Cartesian co-ordinate system in which the fundamental

tensor has the components
(
1 (i = k),
gik =
0 (i 6= k),

the connecting equations (22) are simply: ξi = ξ i . If we confine

ourselves to the use of Cartesian co-ordinate systems, the differ-
ence between co-variants and contra-variants ceases to exist, not
only for tensors but also for the tensor components. It must, how-
ever, be mentioned that the conceptions which have so far been
outlined concerning the fundamental tensor gik assume only that
it is symmetrical and non-degenerate, whereas the introduction of
a Cartesian co-ordinate system implies, in addition, that the cor-
responding quadratic form is definitely positive. This entails a dif-
ference. In the Theory of Relativity the time co-ordinate is added
as a fully equivalent term to the three-space co-ordinates, and
the measure-relation which holds in this four-dimensional man-
ifold is not based on a definite form but on an indefinite one
(Chapter III). In this manifold, therefore, we shall not be able to
introduce a Cartesian co-ordinate system if we restrict ourselves
to real co-ordinates; but the conceptions here developed which
are to be worked out in detail for the dimensional number n = 4
may be applied without alteration. Moreover, the algebraic sim-
plicity of this calculus advises us against making exclusive use of
Cartesian co-ordinate systems, as we have already mentioned at
the end of § 4. Above all, finally, it is of great importance for
later extensions which take us beyond Euclidean geometry that
the affine view should even at this stage receive full recognition
independently of the metrical one.
 EUCLIDEAN SPACE 63

Geometrical and physical quantities are scalars, vectors, and

tensors: this expresses the mathematical constitution of the space
in which these quantities exist. The mathematical symmetry
which this conditions is by no means restricted to geometry but,
on the contrary, attains its full validity in physics. As natural
phenomena take place in a metrical space this tensor calculus is
the natural mathematical instrument for expressing the unifor-
mity underlying them.

§ 6. Tensor Algebra. Examples

Addition of Tensors.—The multiplication of a linear form,
bilinear form, trilinear form . . . by a number, likewise the addition
of two linear forms, or of two bilinear forms . . . always gives rise
to a form of the same kind. Vectors and tensors may thus be
multiplied by a number (a scalar), and two or more tensors of the
same order may be added together. These operations are carried
out by multiplying the components by the number in question or
by addition, respectively. Every set of tensors of the same order
contains a unique tensor 0, of which all the components vanish,
and which, when added to any tensor of the same order, leaves
the latter unaltered. The state of a physical system is described
by specifying the values of certain scalars and tensors.
The fact that a tensor which has been derived from them by
mathematical operations and is an invariant (i.e. dependent upon
them alone and not upon the choice of the co-ordinate system) is
equal to zero is what, in general, the expression of a physical law
amounts to.
Examples.—The motion of a point is represented analytically
by giving the position of the moving-point or of its co-ordinates,
 CHAPTER I 64

dxi
respectively, as functions of the time t. The derivatives are
dt
the contra-variant components ui of the vector “velocity”. By mul-
tiplying it by the mass m of the moving-point, m being a scalar
which serves to express the inertia of matter, we get the “impulse”
(or “momentum”). By adding the impulses of several points of
mass or of those, respectively, of which one imagines a rigid body
to be composed in the mechanics of point-masses, we get the total
impulse of the point-system or of the rigid body. In the case of
continuously extended matter we must supplant these sums by
integrals. The fundamental law of motion is
dGi
= pi ; Gi = mui (25)
dt
where Gi denote the contra-variant components of the impulse of
a mass-point and pi denote those of the force.
Since, according to our view, force is primarily a co-variant
vector, this fundamental law is possible only in a metrical space,
but not in a purely affine one. The same law holds for the total
impulse of a rigid body and for the total force acting on it.
Multiplication ofPTensors.—By multiplying together two
linear forms i ai ξ , i bi η i in the variables ξ and η, we get a
i
P
bilinear form X
ai b k ξ i η k
i,k

and hence from the two vectors a and b we get a tensor c of the
second order, i.e.
ai bk = cik . (26)
Equation (26) represents an invariant relation between the vectors
a and b and the tensor c, i.e. if we pass over to a new co-ordinate
 EUCLIDEAN SPACE 65

system precisely the same equations hold for the components (dis-
tinguished by a bar) of these quantities in this new co-ordinate
system, i.e.
āi b̄k = c̄ik .
In the same way we may multiply a tensor of the first order by
one of the second order (or generally, a tensor of any order by a
tensor of any order). By multiplying
X X
ai ξ i by bki η i ζk
i i,k

in which the Greek letters denote variables which are to be trans-

formed contra-grediently or co-grediently according as the indices
are raised or lowered, we derive the trilinear form
X
ai blk ξ i η k ζl
i,k,l

and, accordingly, by multiplying the two tensors of the first and

second order, a tensor c of the third order, i.e.

ai · blk = clik .

This multiplication is performed on the components by merely

multiplying each component of one tensor by each component of
the other, as is evident above. It must be noted that the co-
variant components (with respect to the index l, for example) of
the resultant tensor of the third order, i.e. clik = ai blk , are given by:
cikl = ai bkl . It is thus immediately permissible in such multipli-
cation formulæ to transfer an index on both sides of the equation
from below to above or vice versa.
 CHAPTER I 66

Examples of Skew-symmetrical and Symmetrical Ten-

sors.—If two vectors with the contra-variant components ai , bi ,
are multiplied first in one order and then in the reverse order, and
if we then subtract the one result from the other, we get a skew-
symmetrical tensor c of the second order with the contra-variant
components
cik = ai bk − ak bi .
This tensor occurs in ordinary vector analysis as the “vectorial
product” of the two vectors a and b. By specifying a certain di-
rection of twist in three-dimensional space, it becomes possible
to establish a reversible one-to-one correspondence between these
tensors and the vectors. (This is impossible in four-dimensional
space for the obvious reason that, in it, a skew-symmetrical tensor
of the second order has six independent components, whereas a
vector has only four; similarly in the case of spaces of still higher
dimensions.) In three-dimensional space the above method of rep-
resentation is founded on the following. If we use only Cartesian
co-ordinate systems and introduce in addition to a and b an arbi-
trary displacement ξ, the determinant
 1 2 3
a a a 
 b b b  = c23 ξ 1 + c31 ξ 2 + c12 ξ 3
 1 2 3
 1 2 3
ξ ξ ξ 
becomes multiplied by the determinant of the co-efficients of trans-
formation, when we pass from one co-ordinate system to another.
In the case of orthogonal transformations this determinant = ±1.
If we confine our attention to “proper” orthogonal transformations,
i.e. such for which this determinant = +1 the above linear form
in the ξ’s remains unchanged. Accordingly, the formulæ
c23 = c∗1 c31 = c∗2 c12 = c∗3
 EUCLIDEAN SPACE 67

express a relation between the skew-symmetrical tensor c and a

vector c∗ , this relation being invariant for proper orthogonal trans-
formations. The vector c∗ is perpendicular to the two vectors
a and b, and its magnitude (according to elementary formulæ of
analytical geometry) is equal to the area of the parallelogram of
which the sides are a and b. It may be justifiable on the ground
of economy of expression to replace skew-symmetrical tensors by
vectors in ordinary vector analysis, but in some ways it hides the
essential feature; it gives rise to the well-known “swimming-rules”
in electrodynamics, which in no wise signify that there is a unique
direction of twist in the space in which electrodynamics events
occur; they become necessary only because the magnetic inten-
sity of field is regarded as a vector, whereas it is, in reality, a
skew-symmetrical tensor (like the so-called vectorial product of
two vectors). If we had been given one more space-dimension,
this error could never have occurred.
In mechanics the skew-symmetrical tensor product of two vec-
tors occurs—
1. As moment of momentum (angular momentum) about a
point O. If there is a point-mass at P and if ξ 1 , ξ 2 , ξ 3 are the
−→
components of OP and ui are the (contra-variant) components of
the velocity of the points at the moment under consideration, and
m its mass, the momentum of momentum is defined by
Lik = m(ui ξ k − uk ξ i ).
The moment of momentum of a rigid body about a point O is the
sum of the moments of momentum of each of the point-masses of
the body.
2. As the turning-moment (torque) of a force. If the lat-
ter acts at the point P and if pi are its contra-variant components,
 CHAPTER I 68

the torque is defined by

q ik = pi ξ k − pk ξ i .

The turning-moment of a system of forces is obtained by simple

addition. In addition to (25) the law
dLik
= q ik (27)
dt
holds for a point-mass as well as for a rigid body free from con-
straint. The turning-moment of a rigid body about a fixed point O
is governed by the law (27) alone.
A further example of a skew-symmetrical tensor is the rate
of rotation (angular velocity) of a rigid body about the fixed
point O. If this rotation about O brings the point P in general
−−→
to P 0 , the vector OP 0 is produced, and hence also P P 0 , by a
−→ −→
linear transformation from OP . If ξ i are the components of OP ,
δξ i those of P P 0 , vki the components of this linear transformation
(matrix), we have X
δξ i = vki ξ k . (28)
k

We shall concern ourselves here only with infinitely small rota-

tions. They are distinguished among infinitesimal matrices by the
additional property that, regarded as an identity in ξ,
X  X 
i i k
δ ξi ξ = δ gik ξ ξ = 0.
i i,k

This gives X
ξi δξ i = 0.
i
 EUCLIDEAN SPACE 69

By inserting the expressions (28) we get

X X
vki ξi ξ k = vik ξ i ξ k = 0.
i,k i,k

This must be identically true in the variables ξi , and hence

vki + vik = 0

i.e. the tensor which has vik for its co-variant components is skew-
symmetrical.
A rigid body in motion experiences an infinitely small rota-
tion during an infinitely small element of time δt. We need only
to divide by δt the infinitesimal rotation-tensor v just formed to
derive (in the limit when δt → 0) the skew-symmetrical tensor
“angular velocity,” which we shall again denote by v. If ui signify
the contra-variant components of the velocity of the point P , and
ui its co-variant components in the formulæ (28), the latter re-
solves into the fundamental formula of the kinematics of a rigid
body, viz. X
ui = vik ξ k . (29)
k

The existence of the “instantaneous axis of rotation” follows from

the circumstance that the linear equations
X
vik ξ k = 0
k

with the skew-symmetrical co-efficients vik always have solutions

in the case n = 3, which differ from the trivial one ξ 1 = ξ 2 =
ξ 3 = 0. One usually finds angular velocity, too, represented as a
vector.
 CHAPTER I 70

Finally the “moment of inertia” which presents itself in the

rotation of a body offers a simple example of a symmetrical tensor
of the second order.
If a point-mass of mass m is situated at the point P to which
−→
the vector OP starting from the centre of rotation O and with the
components ξ i leads, we call the symmetrical tensor of which the
contra-variant components are given by mξ i ξ k (multiplication!),
the “inertia of rotation” of the point-mass (with respect to the
centre of rotation O). The inertia of rotation T ik of a point-system
or body is defined as the sum of these tensors formed separately
for each of its points P . This definition is different from the usual
one, but is the correct one if the intention of regarding the velocity
of rotation as a skew-symmetrical tensor and not as a vector is to
be carried out, as we shall presently see. The tensor T ik plays
the same part with regard to a rotation about O as that of the
scalar m in translational motion.
Contraction.—IfP aki are the mixed components of a tensor
i k
of the second order, i ai is an invariant. Thus, if āi are the
mixed components of the same tensor after transformation to a
new co-ordinate system, then
X X
aii = āii .
i i

Proof.—The variables ξ i , ηi of the bilinear form

X
aki ξ i ηk
i,k

must be subjected to the contra-gredient transformations

αki ξ¯k ,
X X
ξi = ηi = ᾰik η̄k
k k
 EUCLIDEAN SPACE 71

if we wish to bring them into the form

āki ξ¯i η̄k .

X

i,k

From this it follows that

X
ārr = aki αri ᾰkr ,
i,k

and
X X X 
ārr = k
ai i r
αr ᾰk
r i,k r
X
= aii by (200 ).
i

The invariant i aii which has been formed from the compo-
P
nents aki of a matrix is called the trace (spur) of the matrix.
This theorem enables us immediately to carry out a general
operation on tensors, called “contraction,” which takes a second
place to multiplication. By making a definite upper index in the
mixed components of a tensor coincide with a definite lower one,
and summing over this index, we derive from the given tensor a
new one the order of which is two less than that of the original
one, e.g. we get from the components alm hik of a tensor of the fifth
order a tensor of the third order, thus:—
X
alr l
hir = chi . (30)
r

The connection expressed by (30) is an invariant one, i.e. it pre-

serves its form when we pass over to a new co-ordinate system,
 CHAPTER I 72

viz. X
ālr l
hir = c̄hi . (31)
r

To see this we only need the help of two arbitrary contra-variant

vectors ξ i , η i and a co-variant one ζi . By means of them we form
the components, X
alm h i m
hik ξ η ζl = fk ,
h,i,l

of a mixed tensor of the second order: to this we apply the theorem

f¯rr
X X
frr =
r r

which was just proved. We then get the formula

c̄lhi ξ¯h η̄ i ζ̄l

X X
clhi ξ h η i ζl =
h,i,l h,i,l

in which the c’s are defined by (30), the c̄’s by (31). The c̄lhi ’s are
thus, in point of fact, the components of the same tensor of the
third order in the new system, of which the components in the old
one = clih .
Examples of this process of contraction have been met with
in abundance in the above. Wherever summation took place with
respect to certain indices, the summation index appeared twice in
the general member of summation, once above and once below the
co-efficient: each such summation was an example of contraction.
For example, in formula (29): by multiplication of vik with ξ i one
can form the tensor vik ξ l of the third order; by making k coincide
with l and summing for k, we get the contracted tensor of the first
order ui . If a matrix A transforms the arbitrary displacement x
 EUCLIDEAN SPACE 73

into x0 = A(x), and if a second matrix B transforms this x0 into

x00 = −B(x0 ), a combination BA results from the two matrices,
which transforms x directly into x00 = BA(x). If A has the com-
ponents aki and B components bki , the components of the combined
matrix BA are X
cki = bri akr .
r

Here, again, we have the case of multiplication followed by con-

traction.
The process of contraction may be applied simultaneously for
several pairs of indices. From the tensors of the 1st, 2nd, 3rd, . . .
order with the co-variant components ai , aik , aikl , . . . , we thus
get, in particular, the invariants
X X X
ai ai , aik aik , aikl aikl , . . . .
i i,k i,k,l

If, as is here assumed, the quadratic form corresponding to the

fundamental metrical tensor is definitely positive, these invariants
are all positive, for, in a Cartesian co-ordinate system they disclose
themselves directly as the sums of the squares of the components.
Just as in the simplest case of a vector, the square root of these
invariants may be termed the measure or the magnitude of the
tensor of the 1st, 2nd, 3rd, . . . order.
We shall at this point make the convention, once and for all,
that if an index occurs twice (once above and once below) in a
term of a formula to which indices are attached, this is always to
signify that summation is to be carried out with respect to the
index in question, and we shall not consider it necessary to put a
summation sign in front of it.
 CHAPTER I 74

The operations of addition, multiplication, and contraction

only require affine geometry: they are not based upon a “fun-
damental metrical tensor”. The latter is only necessary for the
process of passing from co-variant to contra-variant components
and the reverse.

Euler’s Equations for a Spinning Top

As an exercise in tensor calculus, we shall deduce Euler’s equa-
tions for the motion of a rigid body under no forces about a fixed
point O. We write the fundamental equations (27) in the co-
variant form
dLik
=0
dt
and multiply them, for the sake of briefness, by the contra-variant
components wik of an arbitrary skew-symmetrical tensor which is
constant (independent of the time), and apply contraction with
respect to i and k. If we put Hik equal to the sum
X
mui ξ k
m

which is to be taken over all the points of mass, we get

1
L wik
2 ik
= Hik wik = H,
an invariant, and we can compress our equation into
dH
= 0. (32)
dt
If we introduce the expressions (29) for ui , and the tensor of iner-
tia T , then
Hik = vir Tkr . (33)
 EUCLIDEAN SPACE 75

We have hitherto assumed that a co-ordinate system which is

fixed in space has been used. The components T of inertia then
change with the distribution of matter in the course of time. If,
however, in place of this we use a co-ordinate system which is fixed
in the body, and consider the symbols so far used as referring to
the components of the corresponding tensors with respect to this
co-ordinate system, whereas we distinguish the components of the
same tensors with respect to the co-ordinate system fixed in space
by a horizontal bar, the equation (32) remains valid on account
of the invariance of H. The Tik ’s are now constants; on the other
hand, however, the wik ’s vary with the time. Our equation gives
us
dHik dwik
· wik + Hik · = 0. (34)
dt dt
dwik
To determine , we choose two arbitrary vectors fixed in
dt
the body, of which the co-variant components in the co-ordinate
system attached to the body are ξi and ηi respectively. These
quantities are thus constants, but their components ξ¯i , η̄i in the
space co-ordinate system are functions of the time. Now,

wik ξi ηk = w̄ik ξ¯i η̄k ,

and hence, differentiating with respect to the time

 ¯
dwik

dξi dη̄
· η̄k + ξ¯i ·
ik k
· ξi ηk = w̄ . (35)
dt dt dt

By formula (29)
dξ¯i
= v̄ir ξ¯r = v̄ir ξ¯r .
dt
 CHAPTER I 76

We thus get for the right-hand side of (35)

w̄ik (v̄ir ξ¯r η̄k + v̄kr ξ¯i η̄r ),
and as this is an invariant, we may remove the bars, obtaining
dwik
ξi ηk = wik (ξr ηk vir + ξi ηr vkr ).
dt
This holds identically in ξ and η; thus if the H ik are arbitrary
numbers,
dwik
Hik = wik (vir Hrk + vkr Hir ).
dt
If we take the Hik ’s to be the quantities which we denoted above
by this symbol, the second term of (34) is determined, and our
equation becomes
 
dHik
+ (vi Hrk + vk Hir ) wik = 0,
r r
dt
which is an identity in the skew-symmetrical tensor wik ; hence
" #
d(Hik − Hki ) vir Hrk + vkr Hir
+ = 0.
dt −vkr Hri + vir Hkr

We shall now substitute the expression (33) for Hik . Since, on

account of the symmetry of Tik ,
vkr Hir (= vkr vis Trs )
is also symmetrical in i and k, the two last terms of the sum
in the square brackets destroy one another. If we now put the
symmetrical tensor
vir vkr = grs vir vks = (v, v)ik
 EUCLIDEAN SPACE 77

we finally get our equations into the form

d
(vir Tkr − vkr Tir ) = (v, v)ir Tkr − (v, v)kr Tir .
dt
It is well known that we may introduce a Cartesian co-ordinate
system composed of the three principal axes of inertia, so that in
these
(
1 (i = k),
gik = and Tik = 0 (for i 6= k).
0 (i 6= k),

If we then write T1 in place of T11 , and do the same for the re-
maining indices, our equations in this co-ordinate system assume
the simple form
dvik
(Ti + Tk ) = (Tk − Ti )(v, v)ik .
dt
These are the differential equations for the components vik of the
unknown angular velocity—equations which, as is known, may
be solved in elliptic functions of t. The principal moments of
inertia Ti which occur here are connected with those, Ti∗ , given in
accordance with the usual definitions by the equations

T1∗ = T2 + T3 , T2∗ = T3 + T1 , T3∗ = T1 + T2 .

The above treatment of the problem of rotation may, in contra-

distinction to the usual method, be transposed, word for word,
from three-dimensional space to multi-dimensional spaces. This is,
indeed, irrelevant in practice. On the other hand, the fact that we
have freed ourselves from the limitation to a definite dimensional
 CHAPTER I 78

number and that we have formulated physical laws in such a way

that the dimensional number appears accidental in them, gives
us an assurance that we have succeeded fully in grasping them
mathematically.
The study of tensor-calculus∗ is, without doubt, attended by
conceptual difficulties—over and above the apprehension inspired
by indices, which must be overcome. From the formal aspect,
however, the method of reckoning used is of extreme simplicity;
it is much easier than, e.g., the apparatus of elementary vector-
calculus. There are two operations, multiplication and contrac-
tion; then putting the components of two tensors with totally
different indices alongside of one another; the identification of an
upper index with a lower one, and, finally, summation (not ex-
pressed) over this index. Various attempts have been made to
set up a standard terminology in this branch of mathematics in-
volving only the vectors themselves and not their components,
analogous to that of vectors in vector analysis. This is highly
expedient in the latter, but very cumbersome for the much more
complicated framework of the tensor calculus. In trying to avoid
continual reference to the components we are obliged to adopt an
endless profusion of names and symbols in addition to an intri-
cate set of rules for carrying out calculations, so that the balance
of advantage is considerably on the negative side. An emphatic
protest must be entered against these orgies of formalism which
are threatening the peace of even the technical scientist.
∗
Vide note 4.
 EUCLIDEAN SPACE 79

§ 7. Symmetrical Properties of Tensors

It is obvious from the examples of the preceding paragraph
that symmetrical and skew-symmetrical tensors of the second or-
der, wherever they are applied, represent entirely different kinds of
quantities. Accordingly the character of a quantity is not in gen-
eral described fully, if it is stated to be a tensor of such and such
an order, but symmetrical characteristics have to be added.
A linear form of several series of variables is called symmet-
rical if it remains unchanged after any two of these series of vari-
ables are interchanged, but is called skew-symmetrical if this
converts it into its negative, i.e. reverses its sign. A symmetrical
linear form does not change if the series of variables are subjected
to any permutations among themselves; a skew-symmetrical one
does not change if an even permutation is carried out with the
series of variables, but changes its sign if the permutation is odd.
The co-efficients aikl of a symmetrical trilinear form (we purposely
choose three again as an example) satisfy the conditions

aikl = akli = alik = akil = alki = ailk .

Of the co-efficients of a skew-symmetrical tensor only those which

have three different indices can be 6= 0 and they satisfy the equa-
tions
aikl = akli = alik = −akil = −alki = −ailk .
There can consequently be no (non-vanishing) skew-symmetri-
cal forms of more than n series of variables in a domain of n vari-
ables. Just as a symmetrical bilinear form may be entirely re-
placed by the quadratic form which is derived from it by identify-
ing the two series of variables, so a symmetrical trilinear form is
 CHAPTER I 80

uniquely determined by the cubical form of a single series of vari-

ables with the co-efficients aikl , which is derived from the trilinear
form by the same process. If in a skew-symmetrical trilinear form
X
F = aikl ξ i η k ζ l
i,k,l

we perform the 3! permutations on the series of variables ξ, η, ζ,

and prefix a positive or negative sign to each according as the
permutation is even or odd, we get the original form six times.
If they are all added together, we get the following scheme for
them:—  i k l
ξ ξ ξ 
1 X  
F = aikl η i η k η l  . (36)
3! ζ i ζ k ζ l 
In a linear form the property of being symmetrical or skew-
symmetrical is not destroyed if each series of variables is subjected
to the same linear transformation. Consequently, a meaning may
be attached to the terms symmetrical and skew-symmetrical,
co-variant or contra-variant tensors. But these expressions
have no meaning in the domain of mixed tensors. We need spend
no further time on symmetrical tensors, but must discuss skew-
symmetrical co-variant tensors in somewhat greater detail as they
have a very special significance.
The components ξ i of a displacement determine the direction
of a straight line (positive or negative) as well as its magnitude.
If ξ i and η i are any two linearly independent displacements, and
if they are marked out from any arbitrary point O, they trace out
a plane. The ratios of the quantities

ξ i η k − ξ k η i = ξ ik
 EUCLIDEAN SPACE 81

define the “position” of this plane (a “direction” of the plane) in

the same way as the ratios of the ξ i fix the position of a straight
line (its “direction”). The ξ ik are each = 0 if, and only if, the two
displacements ξ i , η i are linearly dependent; in this case they do
not map out a two-dimensional manifold. When two linearly inde-
pendent displacements ξ i and η i trace out a plane, a definite sense
of rotation is implied, viz. the sense of the rotation about O in
the plane which for a turn < 180◦ brings ξ to coincide with η; also
a definite measure (quantity), viz. the area of the parallelogram
enclosed by ξ and η. If we mark off two displacements ξ, η from
an arbitrary point O, and two ξ∗ , η∗ from an arbitrary point O∗ ,
then the position, the sense of rotation, and the magnitude of the
plane marked out are identical in each if, and only if, the ξ ik ’s of
the one pair coincide with those of the other, i.e.

ξ i η k − ξ k η i = ξ∗i η∗k − ξ∗k η∗i .

So that just as the ξ i ’s determine the direction and length of

a straight line, so the ξ ik ’s determine the sense and surface area
of a plane; the completeness of the analogy is evident.
To express this we may call the first configuration a one-
dimensional space-element, the second a two-dimensional
space-element. Just as the square of the magnitude of a one-
dimensional space-element is given by the invariant

ξi ξ i = gik ξ i ξ k = Q(ξ)

so the square of the magnitude of the two-dimensional space-

element is given, in accordance with the formulæ of analytical
geometry, by
1 ik
2
ξ ξik ;
 CHAPTER I 82

for which we may also write

ξi ηk (ξ i η k − ξ k η i ) = (ξi ξ i )(η k ηk ) − (ξi η i )(ξ k ηk )

= Q(ξ) · Q(η) − Q2 (ξ, η).

In the same sense the determinants

 i k l
ξ ξ ξ 
 
ξ ikl = η i η k η l 
ζ i ζ k ζ l 

which are derived from three independent displacements ξ, η, ζ,

are the components of a three-dimensional space-element, the
magnitude of which is given by the square root of the invariant
1 ikl
3!
ξ ξikl .

In three-dimensional space this invariant is

ξ123 ξ 123 = g1i g2k g3l ξ ikl ξ 123 ,

and since ξ ikl = ±ξ 123 , according as ikl is an even or an odd

permutation of 123, it assumes the value

g · (ξ 123 )2

where g is the determinant of the co-efficients gik of the fundamen-

tal metrical form. The volume of the parallelepiped thus becomes
 1 2 3
 ξ ξ ξ  (taking the absolute,
√  1 2 3 
= g · η η η  i.e. positive value of
ζ 1 ζ 2 ζ 3  the determinants).
 EUCLIDEAN SPACE 83

This agrees with the elementary formulæ of analytical geometry.

In a space of more than three dimensions we may similarly pass
on to four-dimensional space-elements, etc.
Just as a co-variant tensor of the first order assigns a num-
ber linearly (and independently of the co-ordinate system) to ev-
ery one-dimensional space-element (i.e. displacement), so a skew-
symmetrical co-variant tensor of the second order assigns a num-
ber to every two-dimensional space-element, a skew-symmetrical
tensor of the third order to each three-dimensional space-element,
and so on: this is immediately evident from the form in which
(36) is expressed. For this reason we consider it justifiable to call
the co-variant skew-symmetrical tensors simply linear tensors.
Among operations in the domain of linear tensors we shall mention
the two following ones:—

ai bk − ak bi = cik , (37)
ai bkl − ak bli + al bik = cikl . (38)

The former produces a linear tensor of the second order from two
linear tensors of the first order; the latter produces a linear tensor
of the third order from one of the first and one of the second.
Sometimes conditions of symmetry more complicated than
those considered heretofore occur. In the realm of quadrilinear
forms F (ξ, η, ξ 0 , η 0 ) those play a particular part which satisfy the
conditions

F (η, ξ, ξ 0 , η 0 ) = F (ξ, η, η 0 , ξ 0 ) = −F (ξ, η, ξ 0 , η 0 ), (391 )

F (ξ 0 , η 0 , ξ, η) = F (ξ, η, ξ 0 , η 0 ), (392 )
F (ξ, η, ξ 0 , η 0 ) + F (ξ, ξ 0 , η 0 , η) + F (ξ, η 0 , η, ξ 0 ) = 0. (393 )
 CHAPTER I 84

For it may be shown that for every quadratic form of an arbitrary

two-dimensional space-element
ξ ik = ξ i η k − ξ k η i
there is one and only one quadrilinear form F which satisfies these
conditions of symmetry, and from which the above quadratic form
is derived by identifying the second pair of variables ξ 0 , η 0 with
the first pair ξ, η. We must consequently use co-variant tensors
of the fourth order having the symmetrical properties (39) if we
wish to represent functions which stand in quadratic relationship
with an element of surface.
The most general form of the condition of symmetry
for a tensor F of the fifth order of which the first, second, and
fourth series of variables are contra-gredient, the third and fifth
co-gredient (we are taking a particular case) are
X
eS FS = 0
S

in which S signifies all permutations of the five series of vari-

ables in which the contra-gredient ones are interchanged among
themselves and likewise the co-gredient ones; FS denotes the form
which results from F after the permutation S; eS is a system of
definite numbers, which are assigned to the permutations S. The
summation is taken over all the permutations S. The kind of sym-
metry underlying a definite type of tensors expresses itself in one
or more of such conditions of symmetry.

§ 8. Tensor Analysis. Stresses

Quantities which describe how the state of a spatially extended
physical system varies from point to point have not a distinct value
 EUCLIDEAN SPACE 85

but only one “for each point”: in mathematical language they are
“functions of the place or point”. According as we are dealing with
a scalar, vector, or tensor, we speak of a scalar, vector, or tensor
field.
Such a field is given if a scalar, vector, or tensor of the proper
type is assigned to every point of space or to a definite region of
it. If we use a definite co-ordinate system the value of the scalar
quantities or of the components of the vector or tensor quantities
respectively, appear in the co-ordinate system as functions of the
co-ordinates of a variable point in the region under consideration.
Tensor analysis tells us how, by differentiating with respect to
the space co-ordinates, a new tensor can be derived from the old
one in a manner entirely independent of the co-ordinate system.
This method, like tensor algebra, is of extreme simplicity. Only
one operation occurs in it, viz. differentiation.
If
φ = f (x1 , x2 , . . . , xn ) = f (x)
denotes a given scalar field, the change of φ corresponding to an
infinitesimal displacement of the variable point, in which its co-
ordinates xi suffer changes dxi respectively, is given by the total
differential
∂f ∂f ∂f
df = dx1 + dx2 + · · · + dxn .
∂x1 ∂x2 ∂xn
This formula signifies that if the ∆xi are first taken as the compo-
nents of a finite displacement and the ∆f are the corresponding
changes in f , then the difference between
X ∂f
∆f and ∆xi
i
∂x i
 CHAPTER I 86

does not only decrease absolutely to zero with the components

of the displacement, but also relatively to the amount of
the displacement, the measure of which may be defined as
|∆x1 | + |∆x2 | + · · · + |∆xn |. We link up the linear form
X ∂f
ξi
i
∂xi

in the variables ξ i to this differential. If we carry out the same con-

struction in another co-ordinate system (with horizontal bars over
the co-ordinates), it is evident from the meaning of the term differ-
ential that the first linear form passes into the second, if the ξ i ’s
are subjected to the transformation which is contra-gredient to
the fundamental vectors. Accordingly
∂f ∂f ∂f
, , ...,
∂x1 ∂x2 ∂xn
are the co-variant components of a vector which arises from the
scalar field φ in a manner independent of the co-ordinate system.
In ordinary vector analysis it occurs as the gradient and is de-
noted by the symbol grad φ.
This operation may immediately be transposed from a scalar
h
to any arbitrary tensor field. If, e.g., fik (x) are components of a
tensor field of the third order, contra-variant with respect to h,
but co-variant with respect to i and k, then
h
fik ξh η i ζ k

is an invariant, if we take ξh as standing for the components of

an arbitrary but constant co-variant vector (i.e. independent of
its position), and η i , ζ i each as standing for the components of a
 EUCLIDEAN SPACE 87

similar contra-variant vector in turn. The change in this invariant

due to an infinitesimal displacement with components dxi is given
by
h
∂fik
ξh η i ζ k dxl
∂xl
hence
∂f h
h
fikl = ik
∂xl
are the components of a tensor field of the fourth order, which
arises from the given one in a manner independent of the co-
ordinate system. Just this is the process of differentiation;
as is seen, it raises the order of the tensor by 1. We have still to
remark that, on account of the circumstance that the fundamental
metrical tensor is independent of its position, one obtains the
components of the tensor just formed, for example, which are
contra-variant with respect to the index k, by transposing the
∂f hki
index k under the sign of differentiation to the top, viz. . The
∂xl
change from co-variant to contra-variant is interchangeable with
differentiation. Differentiation may be carried out purely formally
by imagining the tensor in question multiplied by a vector having
the co-variant components
∂ ∂ ∂
, , ..., (40)
∂x1 ∂x2 ∂xn
∂f
and treating the differential quotient as the symbolic product
∂xi
∂
of f and . The symbolic vector (40) is often encountered
∂xi
in mathematical literature under the mysterious name “nabla-
vector”.
 CHAPTER I 88

Examples.—The vector with the co-variant components ui

∂ui
gives rise to the tensor of the second order = uik . From this
∂xk
we form
∂ui ∂uk
− . (41)
∂xk ∂xi
These quantities are the co-variant components of a linear tensor
of the second order. In ordinary vector analysis it occurs (with
the signs reversed) as “rotation” (rot, spin or curl). On the other
hand the quantities  
1 ∂ui ∂uk
2
+
∂xk ∂xi
are the co-variant components of a symmetrical tensor of the sec-
ond order. If the vector u represents the velocity of continuously
extended moving matter as a function of its position, the vanishing
of this tensor at a point signifies that the immediate neighbour-
hood of the point moves as a rigid body; it thus merits the name
distortion tensor. Finally by contracting uik we get the scalar

∂ui
∂xi
which is known in vector analysis as “divergence” (div.).
By differentiating and contracting a tensor of the second order
having mixed components Sik we derive the vector

∂Sik
.
∂xk
If vik are the components of a linear tensor field of the second
order, then, analogously to formula (38) in which we substitute v
 EUCLIDEAN SPACE 89

or b and the symbolic vector “differentiation” for a, we get the

linear tensor of the third order with the components
∂vkl ∂vli ∂vik
+ + . (42)
∂xi ∂xk ∂xl
Tensor (41), i.e. the curl, vanishes if vi is the gradient of a scalar
field; tensor (42) vanishes if vik is the curl of a vector ui .
Stresses.—An important example of a tensor field is offered
by the stresses occurring in an elastic body; it is, indeed, from
this example that the name “tensor” has been derived. When ten-
sile or compressional forces act at the surface of an elastic body,
whilst, in addition, “volume-forces” (e.g. gravitation) act on vari-
ous portions of the matter within the body, a state of equilibrium
establishes itself, in which the forces of cohesion called up in the
matter by the distortion balance the impressed forces from with-
out. If we imagine any portion J of the matter cut out of the
body and suppose it to remain coherent after we have removed
the remaining portion, the impressed volume forces will not of
themselves keep this piece of matter in a state of equilibrium.
They are, however, balanced by the compressional forces acting
on the surface Ω of the portion J, which are exerted on it by the
portion of matter removed. We have actually, if we do not take
the atomic (granular) structure of matter into account, to imagine
that the forces of cohesion are only active in direct contact, with
the consequence that the action of the removed portion upon J
must be representable by superficial forces such as pressure: and
indeed, if S do is the pressure acting on an element of surface do
(S here denotes the pressure per unit surface), S can depend only
upon the place at which the element of surface do happens to be
and on the inward normal n of this element of surface with respect
 CHAPTER I 90

to J, which characterises the “position” of do. We shall write Sn

for S to emphasise this connection between S and n. If −n de-
notes the normal in a direction reversed to that of n, it follows
from the equilibrium of a small infinitely thin disc, that
S−n = −Sn . (43)
We shall use Cartesian co-ordinates x1 , x2 , x3 . The compres-
sional forces per unit of area at a point, which act on an element
of surface situated at the same point, the inward normals of which
coincide with the direction of the positive x1 -, x2 -, x3 -axis respec-
tively will be denoted by S1 , S2 , S3 . We now choose any three
positive numbers a1 , a2 , a3 , and a positive number , which is to
converge to the value 0 (whereas the ai remain fixed). From the
point O under consideration we mark off in the direction of the
positive co-ordinate axes the distances
OP1 = a1 , OP2 = a2 , OP3 = a3
and consider the infinitesimal tetrahedron OP1 P2 P3 having
OP2 P3 , OP3 P1 , OP1 P2 as walls and P1 P2 P3 as its “roof”. If
f is the superficial area of the roof and a1 , a2 , a3 are the direction
cosines of its inward normals n, then the areas of the walls are
−f · a1 (= 21 2 a2 a3 ), −f · a2 , −f · a3 .
The sum of the pressures on the walls and the roof becomes for
evanescent values of :

f Sn − (a1 S1 + a2 S2 + a3 S3 ) .
The magnitude of f is of the order 2 : but the volume force acting
upon the volume of the tetrahedron is only of the order of magni-
tude 3 . Hence, owing to the condition for equilibrium, we must
 EUCLIDEAN SPACE 91

have
Sn = (a1 S1 + a2 S2 + a3 S3 ).
With the help of (43) this formula may be extended immediately
to the case in which the tetrahedron is situated in any of the re-
maining 7 octants. If we call the components of Si with respect to
the co-ordinate axes Si1 , Si2 , Si3 , and if ξ i , η i are the components
of any two arbitrary displacements of length 1, then
X
Sik ξ i η k (44)
i,k

is the component, in the direction η, of the compressional force

which is exerted on an element of surface of which the inner normal
is ξ. The bilinear form (44) has thus a significance independent
of the co-ordinate system, and the Sik ’s are the components of
a “stress” tensor field. We shall continue to operate in rectangu-
lar co-ordinate systems so that we shall not have to distinguish
between co-variant and contra-variant quantities.
We form the vector S01 having components S1i , S2i , S3i . The
component of S01 in the direction of the inward normal n of an
element of surface is then equal to the x1 -component of Sn . The
x1 -component of the total pressure which acts on the surface Ω of
the detached portion of matter J is therefore equal to the surface
integral of the normal components of S01 and this, by Gauss’s
Theorem, is equal to the volume integral
Z
− div S01 · dV.
J

The same holds for the x2 and the x3 component. We have thus
 CHAPTER I 92

to form the vector p having the components

X δS k
i
pi = −
k
δxk

(this is performed, as we know, according to an invariant law).

The compressional forces S are then equivalent to a volume force
having the direction and intensity given by p per unit volume in
the sense that, for every dissociated portion of matter J,
Z Z
Sn do = p dV. (45)
Ω J

If k is the impressed force per unit volume, the first condition of

equilibrium for the piece of matter considered coherent after being
detached is Z
(p + k) dV = 0,
J
and as this must hold for every portion of matter
p + k = 0. (46)
If we choose an arbitrary origin O and if r denote the radius
vector to the variable point P , and the square bracket denote
the “vectorial” product, the second condition for equilibrium, the
equation of moments, is
Z Z
[r, Sn ] do + [r, k] dV = 0,
Ω J

and since (46) holds generally we must have, besides (45),

Z Z
[r, Sn ] do = [r, p] dV.
Ω J
 EUCLIDEAN SPACE 93

The x1 component of [r, Sn ] is equal to the component of

x2 S03 − x3 S02 in the direction of n. Hence, by Gauss’s theorem,
the x1 component of the left-hand member is
Z
− div(x2 S03 − x3 S02 ) dV.
J

Hence we get the equation

div(x2 S03 − x3 S02 ) = −(x2 p3 − x3 p2 ).

But the left-hand member

= (x2 div S03 − x3 div S02 ) + (S03 · grad x2 − S02 · grad x3 )

= −(x2 p3 − x3 p2 ) + (S23 − S32 ).

Accordingly, if we form the x2 and x3 components in addition to

the x1 component, this condition of equilibrium gives us

S23 = S32 , S31 = S13 , S12 = S21 ,

i.e. the symmetry of the stress-tensor S. For an arbitrary dis-

placement having the components ξ i ,

S ξiξk
P
P ik i k
gik ξ ξ

is the component of the pressure per unit surface for the com-
ponent in the direction ξ, which acts on an element of surface
placed at right angles to this direction. (We may here again use
any arbitrary affine co-ordinate system.) The stresses are fully
 CHAPTER I 94

equivalent to a volume force of which the density p is calcu-

lated according to the invariant formulæ

δSik
−pi = . (47)
δxk
In the case of a pressure p which is equal in all directions
δp
Sik = p · δik , pi = − .
δxi
As a result of the foregoing reasoning we have formulated in
exact terms the conception of stress alone, and have discovered
how to represent it mathematically. To set up the fundamental
laws of the theory of elasticity it is, in addition, necessary to find
out how the stresses depend on the distortion brought about in
the matter by the impressed forces. There is no occasion for us to
discuss this in greater detail.

§ 9. Stationary Electromagnetic Fields

Hitherto, whenever we have spoken of mechanical or physical
things, we have done so for the purpose of showing in what manner
their spatial nature expresses itself: namely, that its laws manifest
themselves as invariant tensor relations. This also gave us an op-
portunity of demonstrating the importance of the tensor calculus
by giving concrete examples of it. It enabled us to prepare the
ground for later discussions which will grapple with physical the-
ories in greater detail, both for the sake of the theories themselves
and for their important bearing on the problem of time. In this
connection the theory of the electromagnetic field, which is
the most perfect branch of physics at present known, will be of
 EUCLIDEAN SPACE 95

the highest importance. It will here only be considered in so far

as time does not enter into it, i.e. we shall confine our attention
to conditions which are stationary and invariable in time.
Coulomb’s Law for electrostatics may be enunciated thus. If
any charges of electricity are distributed in space with the den-
sity ρ they exert a force
K=e·E (48)
upon a point-charge e, whereby
ρ·r
Z
E=− dV. (49)
4πr3
−→
r here denotes the vector OP which leads from the “point of emer-
gence O” at which E is to be determined, to the “current point” or
source, with respect to which the integral is taken: r is its length
and dV is the element of volume. The force is thus composed of
two factors, the charge e of the small testing body, which depends
on its condition alone, and of the “intensity of field” E, which on
the contrary is determined solely by the given distribution of the
charges in space. We picture in our minds that even if we do not
observe the force acting on a testing body, an “electric field” is
called up by the charges distributed in space, this field being de-
scribed by the vector E; the action on a point-charge e expresses
itself in the force (48). We may derive E from a potential −φ in
accordance with the formulæ
ρ
Z
E = grad φ, −4πφ = dV. (50)
r
From (50) it follows (1) that E is an irrotational (and hence lamel-
lar) vector, and (2) that the flux of E through any closed surface is
 CHAPTER I 96

equal to the charges enclosed by this surface, or that the electricity

is the source of the electric field; i.e. in formulæ

curl E = 0, div E = ρ. (51)

Inversely, Coulomb’s Law arises out of these simple differential

laws if we add the condition that the field E vanish at infinite
distances. For if we put E = grad φ from the first of the equa-
tions (51), we get from the second, to determine φ, Poisson’s equa-
tion ∆φ = ρ, the solution of which is given by (50).
Coulomb’s Law deals with “action at a distance”. The in-
tensity of the field at a point is expressed by it depending on
the charges at all other points, near or far, in space. In contra-
distinction from this the far simpler formulæ (51) express laws
relating to “infinitely near” action. As a knowledge of the values
of a function in an arbitrarily small region surrounding a point
is sufficient to determine the differential quotient of the function
at the point, the values of ρ and E at a point and in its immedi-
ate neighbourhood are brought into connection with one another
by (51). We shall regard these laws of infinitely near action as
the true expression of the uniformity of action in nature, whereas
we look upon (49) merely as a mathematical result following logi-
cally from it. In the light of the laws expressed by (51) which have
such a simple intuitional significance we believe that we under-
stand the source of Coulomb’s Law. In doing this we do indeed
bow to dictates of the theory of knowledge. Even Leibniz for-
mulated the postulate of continuity, of infinitely near action, as
a general principle, and could not, for this reason, become rec-
onciled to Newton’s Law of Gravitation, which entails action at
a distance and which corresponds fully to that of Coulomb. The
mathematical clearness and the simple meaning of the laws(51)
 EUCLIDEAN SPACE 97

are additional factors to be taken into account. In building up

the theories of physics we notice repeatedly that once we have
succeeded in bringing to light the uniformity of a certain group of
phenomena it may be expressed in formulæ of perfect mathemati-
cal harmony. After all, from the physical point of view, Maxwell’s
theory in its later form bears uninterrupted testimony to the stu-
pendous fruitfulness which has resulted through passing from the
old idea of action at a distance to the modern one of infinitely
near action.
The field exerts on the charges which produce it a force of
which the density per unit volume is given by the formula

p = ρE. (52)

This is the rigorous interpretation of the equation (48).

If we bring a test charge (on a small body) into the field, it also
becomes one of the field-producing charges, and formula (48) will
lead to a correct determination of the field E existing before the
test charge was introduced, only if the test charge e is so weak that
its effect on the field is imperceptible. This is a difficulty which
permeates the whole of experimental physics, viz. that by intro-
ducing a measuring instrument the original conditions which are
to be measured become disturbed. This is, to a large extent, the
source of the errors to the elimination of which the experimenter
has to apply so much ingenuity.
The fundamental law of mechanics: mass × acceleration =
force, tells us how masses move under the influence of given forces
(the initial velocities being given). Mechanics does not, however,
teach us what is force; this we learn from physics. The fundamen-
tal law of mechanics is a blank form which acquires a concrete
content only when the conception of force occurring in it is filled
 CHAPTER I 98

in by physics. The unfortunate attempts which have been made to

develop mechanics as a branch of science distinct in itself have, in
consequence, always sought help by resorting to an explanation in
words of the fundamental law: force signifies mass×acceleration.
In the present case of electrostatics, i.e. for the particular category
of physical phenomena, we recognise what is force, and how it is
determined according to a definite law by (52) from the phase-
quantities charge and field. If we regard the charges as being
given, the field equations (51) give the relation in virtue of which
the charges determine the field which they produce. With regard
to the charges, it is known that they are bound to matter. The
modern theory of electrons has shown that this can be taken in
a perfectly rigorous sense. Matter, is composed of elementary
quanta, electrons, which have a definite invariable mass, and, in
addition, a definite invariable charge. Whenever new charges ap-
pear to spring into existence, we merely observe the separation of
positive and negative elementary charges which were previously
so close together that the “action at a distance” of the one was
fully compensated by that of the other. In such processes, ac-
cordingly, just as much positive electricity “arises” as negative.
The laws thus constitute a cycle. The distribution of the elemen-
tary quanta of matter provided with charges fixed once and for
all (and, in the case of non-stationary conditions, also their veloc-
ities) determine the field. The field exerts upon charged matter a
ponderomotive force which is given by (52). The force determines,
in accordance with the fundamental law of mechanics, the acceler-
ation, and hence the distribution and velocity of the matter at the
following moment. We require this whole network of theo-
retical considerations to arrive at an experimental means
of verification,—if we assume that what we directly observe is
 EUCLIDEAN SPACE 99

the motion of matter. (Even this can be admitted only condition-

ally.) We cannot merely test a single law detached from this the-
oretical fabric! The connection between direct experience and the
objective element behind it, which reason seeks to grasp conceptu-
ally in a theory, is not so simple that every single statement of the
theory has a meaning which may be verified by direct intuition.
We shall see more and more clearly in the sequel that Geometry,
Mechanics, and Physics form an inseparable theoretical whole in
this way. We must never lose sight of this totality when we en-
quire whether these sciences interpret rationally the reality which
proclaims itself in all subjective experiences of consciousness, and
which itself transcends consciousness: that is, truth forms a sys-
tem. For the rest, the physical world-picture here described in its
first outlines is characterised by the dualism of matter and field,
between which there is a reciprocal action. Not till the advent of
the theory of relativity was this dualism overcome, and, indeed,
in favour of a physics based solely on fields (cf. § 24).
The ponderomotive force in the electric field was traced back
to stresses even by Faraday. If we use a rectangular system of co-
ordinates x1 , x2 , x3 in which E1 , E2 , E3 are the components of the
electrical intensity of field, the xi component of the force-density
is  
∂E1 ∂E2 ∂E3
pi = ρEi = E + + .
∂x1 ∂x2 ∂x3
By a simple calculation which takes account of the irrotational
property of E we discover from this that the components pi of
the force-density are derived by the formulæ (47) from the stress
tensor, the components Sik of which are tabulated in the following
 CHAPTER I 100

quadratic scheme
1 2 
 (E2 + E32 − E12 ) −E1 E2 −E1 E3 
2 1 2 2 2

 −E2 E1 2
(E3 + E1 − E2 ) −E2 E3  (53)
1
 2 2 2 

 −E3 E1 −E3 E2 2
(E1 + E2 − E3 )

We observe that the condition of symmetry Ski = Sik is fulfilled.

It is, above all, important to notice that the components of the
stress tensor at a point depend only on the electrical intensity of
field at this point. (They, moreover, depend only on the field, and
not on the charge.) Whenever a force p can be retraced by (47) to
stresses S, which form a symmetrical tensor of the second order
only dependent on the values of the phase-quantities describing
the physical state at the point in question, we shall have to regard
these stresses as the primary factors and the actions of the forces
as their consequent. The mathematical justification for this point
of view is brought to light by the fact that the force p results from
differentiating the stress. Compared with forces, stresses are thus,
so to speak, situated on the next lower plane of differentiation, and
yet do not depend on the whole series of values traversed by the
phase-quantities, as would be the case for an arbitrary integral,
but only on its value at the point under consideration. It further
follows from the fact that the electrostatic forces which charged
bodies exert on one another can be retraced to a symmetrical
stress tensor, that the resulting total force as well as the resulting
couple vanishes (because the integral taken over the whole space
has a divergence = 0). This means that an isolated system of
charged masses which is initially at rest cannot of itself acquire a
translational or rotational motion as a whole.
The tensor (53) is, of course, independent of the choice of co-
ordinate system. If we introduce the square of the value of the
 EUCLIDEAN SPACE 101

field intensity
|E|2 = Ei E i
then we have
Sik = 21 gik |E|2 − Ei Ek .
These are the co-variant stress components not only in a Cartesian
but also in any arbitrary affine co-ordinate system, if Ei are the co-
variant components of the field intensity. The physical significance
of these stresses is extremely simple. If, for a certain point, we
use rectangular co-ordinates, the x1 axis of which points in the
direction E: then

E1 = |E|, E2 = 0, E3 = 0;

we thus find them to be composed of a tension having the intensity

1
2
|E|2 in the direction of the lines of force, and of a pressure of the
same intensity acting perpendicularly to them.
The fundamental laws of electrostatics may now be
summarised in the following invariant tensor form:—
∂Ei ∂Ek ∂φ 
(I) ∂x − = 0, or E i = respectively;
∂x ∂x


k i i 

i
∂E (54)
(II) = ρ;
∂xi




(III) 1 2 
Sik = 2 gik |E| − Ei Ek .

A system of discrete point-charges e1 , e2 , e3 , . . . has potential

energy
1 X ei ek
U=
8π i6=k rik
 CHAPTER I 102

in which rik denotes the distance between the two charges ei

and ek . This signifies that the virtual work which is performed
by the forces acting at the separate points (owing to the charges
at the remaining points) for an infinitesimal displacement of the
points is a total differential, viz. δU . For continuously distributed
charges this formula resolves into
ρ(P )ρ(P 0 )
ZZ
U= dV dV 0
8πrP P 0
in which both volume integrations with respect to P and P 0 are
to be taken over the whole space, and rP P 0 denotes the distance
between these two points. Using the potential φ we may write
Z
1
U = − 2 ρφ dV.

The integrand is φ · div E. In consequence of the equation

div(φE) = φ · div E + E grad φ

and of Gauss’s theorem, according to which the integral of div(φE)

taken over the whole space is equal to 0, we have
Z Z Z
− ρφ dV = (E grad φ) dV = |E|2 dV ;

i.e. Z
U= 1
2
|E|2 dV. (55)

This representation of the energy makes it directly evident that

the energy is a positive quantity. If we trace the forces back to
stresses, we must picture these stresses (like those in an elastic
 EUCLIDEAN SPACE 103

body) as being everywhere associated with positive potential en-

ergy of strain. The seat of the energy must hence be sought in the
field. Formula (55) gives a fully satisfactory account of this point.
It tells us that the energy associated with the strain amounts to
1
2
|E|2 per unit volume, and is thus exactly equal to the tension
and the pressure which are exerted along and perpendicularly to
the lines of force. The deciding factor which makes this view
permissible is again the circumstance that the value obtained for
the energy-density depends solely on the value, at the point in
question, of the phrase-quantity E which characterises the field.
Not only the field as a whole, but every portion
R 1 2 of the field has
a definite amount of potential energy = 2 |E| dV . In statics,
it is only the total energy which comes into consideration. Only
later, when we pass on to consider variable fields, shall we arrive
at irrefutable confirmation of the correctness of this view.
In the case of conductors in a statical field the charges collect
on the outer surface and there is no field in the interior. The
equations (51) then suffice to determine the electrical field in free
space in the “æther”. If, however, there are non-conductors, di-
electrics in the field, the phenomenon of dielectric polarisation
(displacement) must be taken into consideration. Two charges +e
and −e at the points P1 and P2 respectively, “source and sink” as
we shall call them, produce a field, which arises from the potential
 
e 1 1
−
4π r1 r2
in which r1 and r2 denote the distances of the points P1 , P2 from
−−→
the origin, O. Let the product of e and the vector P1 P2 be called
the moment m of the “source and sink” pair. If we now suppose
the two charges to approach one another in a definite direction at
 CHAPTER I 104

a point P , the charge increasing simultaneously in such a way that

the moment m remains constant, we get, in the limit, a “doublet”
of moment m, the potential of which is given by
m 1
gradP .
4π r
The result of an electric field in a dielectric is to give rise to
these doublets in the separate elements of volume: this effect is
known as polarisation. If m is the electric moment of the dou-
blets per unit volume, then, instead of (50), the following formula
holds for the potential
ρ 1
Z Z
−4πφ = dV + m · gradP dv. (56)
r r
From the point of view of the theory of electrons this circumstance
becomes immediately intelligible. Let us, for example, imagine an
atom to consist of a positively charged “nucleus” at rest, around
which an oppositely charged electron rotates in a circular path.
The mean position of the electron for the mean time of a com-
plete revolution of the electron round the nucleus will then coin-
cide with the position of the nucleus, and the atom will appear
perfectly neutral from without. But if an electric field acts, it
exerts a force on the negative electron, as a result of which its
path will lie excentrically with respect to the atomic nucleus, e.g.
will become an ellipse with the nucleus at one of its foci. In the
mean, for times which are great compared with the time of rev-
olution of the electron, the atom will act like a doublet; or if we
treat matter as being continuous we shall have to assume contin-
uously distributed doublets in it. Even before entering upon an
exact atomistic treatment of this idea we can say that, at least
 EUCLIDEAN SPACE 105

to a first approximation, the moment m per unit volume will be

proportional to the intensity E of the electric field: i.e. m = kE,
in which k denotes a constant characteristic of the matter, which
is dependent on its chemical constitution, viz. on the structure of
its atoms and molecules.
Since m 1 div m
div = m grad +
r r r
we may replace equation (56) by
ρ − div m
Z
−4πφ = dV.
r
From this we get for the field intensity E = grad φ

div E = ρ − div m.

If we now introduce the “electric displacement”

D=E+m

the fundamental equations become:

curl E = 0, div D = ρ. (57)

They correspond to equations (51); in one of them the intensity E

of field now occurs, in the other D the electric displacement. With
the above assumption m = kE we get the law of matter

D = E (58)

if we insert the constant  = 1 + k, characteristic of the matter,

called the dielectric constant.
 CHAPTER I 106

These laws are excellently confirmed by observation. The influ-

ence of the intervening medium which was experimentally proved
by Faraday, and which expresses itself in them, has been of great
importance in the development of the theory of action by contact.
We may here pass over the corresponding extension of the formulæ
for stress, energy, and force.
It is clear from the mode of derivation that (57) and (58) are
not rigorously valid laws, since they relate only to mean values
and are deduced for spaces containing a great number of atoms
and for times which are great compared with the times of revolu-
tion of the electrons round the atom. We still look upon (51)
as expressing the physical laws exactly. Our objective here
and in the sequel is above all to derive the strict physical laws.
But if we start from phenomena, such “phenomenological laws” as
(57) and (58) are necessary stages in passing from the results of
direct observation to the exact theory. In general, it is possible to
work out such a theory only by starting in this way. The validity
of the theory is then established if, with the aid of definite ideas
about the atomic structure of matter, we can again arrive at the
phenomenological laws by using mean value arguments. If the
atomic structure is known, this process must, in addition, yield
the values of the constants occurring in these laws and charac-
teristic of the matter in question (such constants do not occur in
exact physical laws). Since laws of matter such as (58), which only
take the influence of massed matter into account, certainly fail for
events in which the fine structure of matter cannot be neglected,
the range of validity of the phenomenological theory must be fur-
nished by an atomistic theory of this kind, as must also those laws
which have to be substituted in its place for the region beyond this
range. In all this the electron theory has met with great success,
 EUCLIDEAN SPACE 107

although, in view of the difficulty of the task, it is far from giving

a complete statement of the more detailed structure of the atom
and its inner mechanism.
In the first experiments with permanent magnets, magnetism
appears to be a mere repetition of electricity: here Coulomb’s Law
holds likewise! A characteristic difference, however, immediately
asserts itself in the fact that positive and negative magnetism can-
not be dissociated from one another. There are no sources, but
only doublets in the magnetic field. Magnets consist of infinitely
small elementary magnets, each of which itself contains positive
and negative magnetism. The amount of magnetism in every por-
tion of matter is de facto nil; this would appear to mean that there
is really no such thing as magnetism. The explanation of this was
furnished by Oersted’s discovery of the magnetic action of electric
currents. The exact quantitative formulation of this action as ex-
pressed by Biot and Savart’s Law leads, just like Coulomb’s Law,
to two simple laws of action by contact. If s denotes the density
of the electric current, and H the intensity of the magnetic field,
then
curl H = s, div H = 0. (59)
The second equation asserts the non-existence of sources in the
magnetic field. Equations (59) are exactly analogous to (51) if div
and curl be interchanged. These two operations of vector analysis
correspond to one another in exactly the same way as do scalar and
vectorial multiplication in vector algebra (div denotes scalar, curl
vectorial, multiplication by the symbolic vector “differentiation”).
The solution of the equations (59) vanishes for infinite distances;
for a given distribution of current it is given by
[s, r]
Z
H= dV, (60)
4πr3
 CHAPTER I 108

which is exactly analogous to (49) and is, indeed, the expression

of Biot and Savart’s Law. This solution may be derived from a
“vector potential” −f in accordance with the formulæ
s
Z
H = − curl f , −4πf = dV.
r
Finally the formula for the density of force in the magnetic field
is
p = [s, H] (61)
corresponding exactly with (52).
There is no doubt that these laws give us a true statement of
magnetism. They are not a repetition but an exact counterpart
of electrical laws, and bear the same relation to the latter as vec-
torial products to scalar products. From them it may be proved
mathematically that a small circular current acts exactly like a
small elementary magnet thrust through it perpendicularly to its
plane. Following Ampère we have thus to imagine the magnetic
action of magnetised bodies to depend on molecular currents;
according to the electron theory these are straightway given by
the electrons circulating in the atom.
The force p in the magnetic field may also be traced back
to stresses, and we find, indeed, that we get the same values for
the stress components as in the electrostatic field: we need only
replace E by H. Consequently we shall use the corresponding
value 12 H2 for the density of the potential energy contained in the
field. This step will only be properly justified when we come to
the theory of fields varying with the time.
It follows from (59) that the current distribution is free of
sources: div s = 0. The current field can therefore be entirely di-
vided into current tubes all of which again merge into themselves,
 EUCLIDEAN SPACE 109

i.e. are continuous. The same total current flows through every
cross-section of each tube. In no wise does it follow from the laws
holding in a stationary field, nor does it come into consideration
for such a field, that this current is an electric current in the or-
dinary sense, i.e. that it is composed of electricity in motion; this
is, however, without doubt the case. In view of this fact the law
div s = 0 asserts that electricity is neither created nor destroyed.
It is only because the flux of the current vector through a closed
surface is nil that the density of electricity remains everywhere
unchanged—so that electricity is neither created nor destroyed.
(We are, of course, dealing with stationary fields exclusively.) The
expression vector potential f , introduced above, also satisfies
the equation div f = 0.
Being an electric current, s is without doubt a vector in the
true sense of the word. It then follows, however, from the Law of
Biot and Savart that H is not a vector but a linear tensor of
the second order. Let its components in any co-ordinate system
(Cartesian or even merely affine) be Hik . The vector potential f
is a true vector. If φi are its co-variant components and si the
contra-variant components of the current-density (the current is
like velocity fundamentally a contra-variant vector), the following
table gives us the final form (independent of the dimensional num-
ber) of the laws which hold in the magnetic field produced
by a stationary electric current.
∂Hkl ∂Hli ∂Hik
+ + = 0, (621 )
∂xi ∂xk ∂xl
 CHAPTER I 110

∂φi ∂φk
Hik = − respectively
∂xk ∂xi
∂Hik
= si . (622 )
∂xk
The stresses are determined by:
Sik = Hir H kr − 21 δik |H|2 (623 )
in which |H| signifies the strength of the magnetic field:
|H|2 = Hik H ik .
The stress tensor is symmetrical, since
Hir Hkr = Hir Hkr = g rs Hir Hks .
The components of the force-density are
pi = Hki sk . (624 )
The energy-density = 12 |H|2 .
These are the laws that hold for the field in empty space.
We regard them as being exact physical laws which are generally
valid, as in the case of electricity. For a phenomenological theory it
is, however, necessary to take into consideration the magnetisa-
tion, a phenomenon analogous to dielectric polarisation. Just as
D occurred in conjunction with E, so the “magnetic induction” B
associates itself with the intensity of field H. The laws
curl H = s, div B = 0
hold in the field, as does the law which takes account of the mag-
netic character of the matter
B = µH. (63)
 EUCLIDEAN SPACE 111

The constant µ is called magnetic permeability. But whereas the

single atom only becomes polarised by the action of the intensity
of the electrical field (i.e. becomes a doublet), (this takes place in
the direction of the field intensity), the atom is from the outset an
elementary magnet owing to the presence of rotating electrons in
it (at least, in the case of para- and ferro-magnetic substances).
All these elementary magnets, however, neutralise one another’s
effects, as long as they are irregularly arranged and all positions
of the electronic orbits occur equally frequently on the average.
The imposed magnetic force merely fulfils the function of direct-
ing the existing doublets. It evidently is due to this fact that the
range within which (63) holds is much less than the correspond-
ing range of (63). Permanent magnets and ferro-magnetic bodies
(iron, cobalt, nickel) are, above all, not subject to it.
In the phenomenological theory there must be added to the
laws already mentioned that of Ohm:

s = σE (σ = conductivity).

It asserts that the current follows the fall of potential and is pro-
portional to it for a given conductor. Corresponding to Ohm’s
Law we have in the atomic theory the fundamental law of me-
chanics, according to which the motion of the “free” electrons is
determined by the electric and magnetic forces acting on them
which thus produce an electric current. Owing to collisions with
the molecules no permanent acceleration can come about, but
(just as in the case of a heavy body which is falling and experi-
ences the resistance of the air) a mean limiting velocity is reached,
which may, to a first approximation at least, be put proportional
to the driving electric force E. In this way Ohm’s Law acquires a
meaning.
 CHAPTER I 112

If the current is produced by a voltaic cell or an accumulator,

the chemical action which takes place maintains a constant differ-
ence of potential, the “electromotive force,” between the two ends
of the conducting wire. Since the events which occur in the con-
trivance producing the current can obviously be understood only
in the light of an atomic theory, it leads to the simplest result phe-
nomenologically to represent it by means of a cross-section taken
through the conducting circuit at each end, beyond which the
potential makes a sudden jump equal to the electromotive force.
This brief survey of Maxwell’s theory of stationary fields will
suffice for what follows. We have not the space here to enlarge
upon details and concrete applications.
 CHAPTER II
THE METRICAL CONTINUUM
§ 10. Note on Non-Euclidean Geometry∗
Doubts as to the validity of Euclidean geometry seem to have
been raised even at the time of its origin, and are not, as our
philosophers usually assume, outgrowths of the hypercritical ten-
dency of modern mathematicians. These doubts have from the
outset hovered round the fifth postulate. The substance of the
latter is that in a plane containing a given straight line g and a
point P external to the latter (but in the plane) there is only one
straight line through P which does not intersect g: it is called the
straight line parallel to P . Whereas the remaining axioms of Eu-
clid are accepted as being self-evident, even the earliest exponents
of Euclid have endeavoured to prove this theorem from the remain-
ing axioms. Nowadays, knowing that this object is unattainable,
we must look upon these reflections and efforts as the beginning
of “non-Euclidean” geometry, i.e. of the construction of a geomet-
rical system which can be developed logically by accepting all the
axioms of Euclid, except the postulate of parallels. A report of
Proclus (a.d. 5) about these attempts has been handed down to
posterity. Proclus utters an emphatic warning against the abuse
that may be practised by calling propositions self-evident. This
warning cannot be repeated too often; on the other hand, we must
not fail to emphasise the fact that, in spite of the frequency with
which this property is wrongfully used, the “self-evident” property
is the final root of all knowledge, including empirical knowledge.
Proclus insists that “asymptotic lines” may exist.
∗
Vide note 1.

113
 CHAPTER II 114

We may picture this as follows. Suppose a straight line g

be given in a plane, also a point P outside it in the plane, and
a straight line s passing through P and which may be rotated
about P . Let s be perpendicular to g initially. If we now ro-

P
′′
s s s′
g

Fig. 2.

tate s, the point of intersection of s and g glides along g, e.g. to

the right, and if we continue turning, a definite moment arrives
at which this point of intersection just vanishes to infinity; s then
occupies the position of an “asymptotic” straight line. If we con-
tinue turning, Euclid assumes that, at even this same moment,
a point of intersection already appears on the left. Proclus, on
the other hand, points out the possibility that one may perhaps
have to turn s through a further definite angle before a point of
intersection arises to the left. We should then have two “asymp-
totic” straight lines, one to the right, viz. s0 , and the other to the
left, viz. s00 . If the straight line s through P were then situated
in the angular space between s00 and s0 (during the rotation just
described) it would cut g; if it lay between s0 and s00 , it would not
intersect g. There must be at least one non-intersecting straight
line; this follows from the other axioms of Euclid. I shall recall a
familiar figure of our early studies in plane geometry, consisting
of the straight line h and two straight lines g and g 0 which inter-
sect h at A and A0 and make equal angles with it, g and g 0 are
each divided into a right and a left half by their point of intersec-
 THE METRICAL CONTINUUM 115

tion with h. Now, if g and g 0 had a common point s to the right

of h, then, since BAA0 B 0 is congruent with C 0 A0 AC (vide Fig. 3),

C′ A′ B′
g′
C A B
g

h
Fig. 3.

there would also be a point of intersection S ∗ to the left of h. But

this is impossible since there is only one straight line that passes
through two given points S and S ∗ .
Attempts to prove Euclid’s postulate were continued by Ara-
bian and western mathematicians of the Middle Ages. Passing
straight to a more recent period we shall mention the names of
only the last eminent forerunners of non-Euclidean geometry, viz.
the Jesuit father Saccheri (beginning of the eighteenth century)
and the mathematicians Lambert and Legendre. Saccheri was
aware that the question whether the postulate of parallels is valid
is equivalent to the question whether the sum of the angles of a
triangle are equal to or less than 180◦ . If they amount to 180◦
in one triangle, then they must do so in every triangle and Eu-
clidean geometry holds. If the sum is < 180◦ in one triangle then
it is < 180◦ in every triangle. That they cannot be > 180◦ is
excluded for the same reason for which we just now concluded
that not all the straight lines through P can cut the fixed straight
 CHAPTER II 116

line g. Lambert discovered that if we assume the sum of the three

angles to be < 180◦ there must be a unique length in geometry.
This is closely related to an observation which Wallis had previ-
ously made that there can be no similar figures of different sizes in
non-Euclidean geometry (just as in the case of the geometry of the
surface of a rigid sphere). Hence if there is such a thing as “form”
independent of size, Euclidean geometry is justified in its claims.
Lambert, moreover, deduced a formula for the area of a triangle,
from which it is clear that, in the case of non-Euclidean geometry,
this area cannot increase beyond all limits. It appears that the
researches of these men has gradually spread the belief in wide
circles that the postulate of parallels cannot be proved. At that
time this problem occupied many minds. D’Alembert pronounced
it a scandal of geometry that it had not yet been decisively settled.
Even the authority of Kant, whose philosophic system claims Eu-
clidean geometry as a priori knowledge representing the content
of pure space-intuition in adequate judgments, did not succeed in
settling these doubts permanently.
Gauss also set out originally to prove the axiom of parallels,
but he early gained the conviction that this was impossible and
thereupon developed the principles of a non-Euclidean geometry,
for which the axioms of parallels does not hold, to such an ex-
tent that, from it, the further development could be carried out
with the same ease as for Euclidean geometry. He did not make
his investigations known for, as he later wrote in a private letter,
he feared “the outcry of the Bœotians”; for, he said, there were
only a few people who understood what was the true essence of
these questions. Independently of Gauss, Schweikart, a professor
of jurisprudence, gained a full insight into the conditions of non-
Euclidean geometry, as is evident from a concise note addressed to
 THE METRICAL CONTINUUM 117

Gauss. Like the latter he considered it in no wise self-evident, and

established that Euclidean geometry is valid in our actual space.
His nephew Taurinus whom he encouraged to study these ques-
tions was, in contrast to him, a believer of Euclidean geometry,
but we are nevertheless indebted to Taurinus for the discovery of
the fact that the formulæ of spherical trigonometry
√ are real on a
sphere which has an imaginary radius = −1, and that through
them a geometrical system is constructed along analytical lines
which satisfies all the axioms of Euclid except the fifth postulate.
For the general public the honour of discovering and elabo-
rating non-Euclidean geometry must be shared between Nikolaj
Iwanowitsch Lobatschefskij (1793–1856), a Russian professor of
mathematics at Kasan, and Johann Bolyai (1802–1860), a Hun-
garian officer in the Austrian army. The ideas of both assumed a
tangible form in 1826. The chief manuscript of both, by which the
public were informed of their discovery and which offered an argu-
ment of the new geometry in the manner of Euclid, had its origin
in 1830–1831. The discussion by Bolyai is particularly clear, inas-
much as he carries the argument as far as possible without making
an assumption as to the validity or non-validity of the fifth postu-
late, and only afterwards derives the theorems of Euclidean and
non-Euclidean geometry from the theorems of his “absolute” ge-
ometry according to whether one decides in favour of or against
Euclid.
Although the structure was thus erected, it was by no means
definitely decided whether, in absolute geometry, the axiom of
parallels would not after all be shown to be a dependent theorem.
The strict proof that non-Euclidean geometry is absolutely
consistent in itself had yet to follow. This resulted almost of
itself in the further development of non-Euclidean geometry. As
 CHAPTER II 118

often happens, the simplest way of proving this was not discovered
at once. It was discovered by Klein as late as 1870 and depends
on the construction of a Euclidean model for non-Euclidean
geometry (vide note 2). Let us confine our attention to the plane!
In a Euclidean plane with rectangular co-ordinates x and y we
shall draw a circle U of radius unity with the origin as centre.
Introducing homogeneous co-ordinates
x1 x2
x= , y=
x3 x3
(so that the position of a point is defined by the ratio of three
numbers, i.e. x1 : x2 : x3 ), the equation to the circle becomes

−x21 − x22 + x23 = 0.

Let us denote the quadratic form on the left by Ω(x) and the
corresponding symmetrical bilinear form of two systems of value,
xi , x0i by Ω(x, x0 ). A transformation which assigns to every point x
a transformed point x0 according to the linear formulæ
3
X
x0i = αik xk (|αik | =
6 0)
k=1

is called, as we know, a collineation (affine transformations are a

special class of collineations). It transforms every straight line,
point for point, into another straight line and leaves the cross-
ratio of four points on a straight line unaltered. We shall now
set up a little dictionary by which we translate the conceptions of
Euclidean geometry into a new language, that of non-Euclidean
geometry; we use inverted commas to distinguish its words. The
vocabulary of this dictionary is composed of only three words.
 THE METRICAL CONTINUUM 119

The word “point” is applied to any point on the inside of U

(Fig. 4).
A “straight line” signifies the portion of a straight line lying
wholly in U . The collineations which transform the circle U into
itself are of two kinds; the first leaves the sense in which U is

B1

A
P

g A′ U

A′′

B2
Fig. 4.

described unaltered, whereas the second reverses it. The former

are called “congruent” transformations; two figures composed of
points are called “congruent” if they can be transformed into one
another by such a transformation. All the axioms of Euclid except
the postulate of parallels hold for these “points,” “straight lines,”
and the conception “congruence”. A whole sheaf of “straight lines”
passing through the “point” P which do not cut the one “straight
line” g is shown in Fig. 4. This suffices to prove the consistency
of non-Euclidean geometry, for things and relations are shown for
which all the theorems of Euclidean geometry are valid provided
 CHAPTER II 120

that the appropriate nomenclature be adopted. It is evident, with-

out further explanation, that Klein’s model is also applicable to
spatial geometry.
We now determine the non-Euclidean distance between two
“points” in this model, viz. between

A = (x1 : x2 : x3 ) and A0 = (x01 : x02 : x03 ).

Let the straight line AA0 cut the circle U in the two points, B1 , B2 .
The homogeneous co-ordinates yi of these two points are of the
form
yi = λxi + λ0 x0i
and the corresponding ratio of the parameters, λ : λ0 , is given by
the equation Ω(y) = 0, viz.
p
λ −Ω(x, x0 ) ± Ω2 (x, x0 ) − Ω(x)Ω(x0 )
= .
λ0 Ω(x)

Hence the cross-ratio of the four points, A, A0 , B1 , B2 is

0
p
Ω(x, x ) + Ω2 (x, x0 ) − Ω(x)Ω(x0 )
[AA0 ] = p .
Ω(x, x0 ) − Ω2 (x, x0 ) − Ω(x)Ω(x0 )

This quantity which depends on the two arbitrary “points,” A, A0 ,

is not altered by a “congruent” transformation. If A, A0 , A00 are
any three “points” lying on a “straight line” in the order written,
then
[AA00 ] = [AA0 ] · [A0 A00 ].
The quantity
1
2
log[AA0 ] = AA0 = r
 THE METRICAL CONTINUUM 121

has thus the functional property

AA0 + A0 A00 = AA00 .

As it has the same value for “congruent” distances AA0 too, we

must regard it as the non-Euclidean distance between the two
points, A, A0 . Assuming the logs to be taken to the base e, we
get an absolute determination for the unit of measure, as was
recognised by Lambert. The definition may be written in the
shorter form:
Ω(x, x0 )
cosh r = p (1)
Ω(x) · Ω(x0 )
(cosh denotes the hyperbolic cosine).

This measure-determination had already been enunciated before

Klein by Cayley∗ who referred it to an arbitrary real or imagi-
nary conic section Ω(x) = 0: he called it the “projective measure-
determination”. But it was reserved for Klein to recognise that in
the case of a real conic it leads to non-Euclidean geometry.
It must not be thought that Klein’s model shows that the non-
Euclidean plane is finite. On the contrary, using non-Euclidean
measures I can mark off the same distance on a “straight line” an
infinite number of times in succession. It is only by using Eu-
clidean measures in the Euclidean model that the distances of
these “equidistant” points becomes smaller and smaller. For non-
Euclidean geometry the bounding circle U represents unattain-
able, infinitely distant, regions.
If we use an imaginary conic, Cayley’s measure-determination
leads to ordinary spherical geometry, such as holds on the surface
∗
Vide note 3.
 CHAPTER II 122

of a sphere in Euclidean space. Great circles take the place of

straight lines in it, but every pair of points at the end of the same
diameter must be regarded as a single “point,” in order that two
“straight lines” may only intersect at one “point”. Let us project
the points on the sphere by means of (straight) rays from the
centre on to the tangential plane at a point on the surface of the
sphere, e.g. the south pole. Two diametrically opposite points will
then coincide on the tangential plane as a result of the transfor-
mation. We must, in addition, as in projective geometry, furnish
this plane with an infinitely distant straight line; this is given by
the projection of the equator. We shall now call two figures in this
plane “congruent” if their projections (through the centre) on to
the surface of the sphere are congruent in the ordinary Euclidean
sense. Provided this conception of “congruence” is used, a non-
Euclidean geometry, in which all the axioms of Euclid except the
fifth postulate are fulfilled, holds in this plane. Instead of this
postulate we have the fact that each pair of straight lines, with-
out exception, intersects, and, in accordance with this, the sum
of the angles in a triangle > 180◦ . This seems to conflict with
the Euclidean proof quoted above. The apparent contradiction
is explained by the circumstance that in the present “spherical”
geometry the straight line is closed, whereas Euclid, although he
does not explicitly state it in his axioms, tacitly assumes that it is
an open line, i.e. that each of its points divides it into two parts.
The deduction that the hypothetical point of intersection S on
the “right-hand” side is different from that S ∗ on the “left-hand”
side is rigorously true only if this “openness” be assumed.
Let us mark out in space a Cartesian co-ordinate system
x1 , x2 , x3 , having its origin at the centre of the sphere and the
line connecting the north and south poles as its x3 axis, the
 THE METRICAL CONTINUUM 123

radius of the sphere being the unit of length. If x1 , x2 , x3 are the

co-ordinates of any point on the sphere, i.e.

Ω(x) ≡ x21 + x22 + x23 = 1

x1 x2
then and are respectively the first and second co-ordinate
x3 x3
of the transformed point in our plane x3 = 1, i.e. x1 : x2 : x3 is the
ratio of the homogeneous co-ordinates of the transformed point.
Congruent transformations of the sphere are linear transforma-
tions which leave the quadratic form Ω(x) invariant. The “con-
gruent” transformations of the plane in terms of our “spherical”
geometry are thus given by such linear transformations of the ho-
mogeneous co-ordinates as convert the equation Ω(x) = 0, which
signifies an imaginary conic, into itself. This proves the statement
made above concerning the relationship between spherical geom-
etry and Cayley’s measure-relation. This agreement is expressed
in the formula for the distance r between two points A, A0 , which
is here
Ω(x, x0 )
cos r = p . (2)
Ω(x)Ω(x0 )
At the same time we have confirmed the discovery of Taurinus that
Euclidean geometry√ is identical with non-Euclidean geometry on
a sphere of radius −1.
Euclidean geometry occupies an intermediate position between
that of Bolyai-Lobatschefsky and spherical geometry. For if we
make a real conic section change to a degenerate one, and thence
to an imaginary one, we find that the plane with its corresponding
Cayley measure-relation is at first Bolyai-Lobatschefskyan, then
Euclidean, and finally spherical.
 CHAPTER II 124

§ 11. The Geometry of Riemann

The next stage in the development of non-Euclidean geometry
that concerns us chiefly is that due to Riemann. It links up with
the foundations of Differential Geometry, in particular with that
of the theory of surfaces as set out by Gauss in his Disquisitiones
circa superficies curvas.
The most fundamental property of space is that its
points form a three-dimensional manifold. What does this
convey to us? We say, for example, that ellipses form a two-
dimensional manifold (as regards their size and form, i.e. consid-
ering congruent ellipses similar, non-congruent ellipses as dissim-
ilar), because each separate ellipse may be distinguished in the
manifold by two given numbers, the lengths of the semi-major
and semi-minor axis. The difference in the conditions of equilib-
rium of an ideal gas which is given by two independent variables,
such as pressure and temperature, form a two-dimensional mani-
fold, likewise the points on a sphere, or the system of pure tones
(in terms of intensity and pitch). According to the physiological
theory which states that the sensation of colour is determined by
the combination of three chemical processes taking place on the
retina (the black-white, red-green, and the yellow-blue process,
each of which can take place in a definite direction with a def-
inite intensity), colours form a three-dimensional manifold with
respect to quality and intensity, but colour qualities form only a
two-dimensional manifold. This is confirmed by Maxwell’s famil-
iar construction of the colour triangle. The possible positions of a
rigid body form a six-dimensional manifold, the possible positions
of a mechanical system having n degrees of freedom constitute, in
general, an n-dimensional manifold. The characteristic of an
 THE METRICAL CONTINUUM 125

n-dimensional manifold is that each of the elements com-

posing it (in our examples, single points, conditions of a gas,
colours, tones) may be specified by the giving of n quan-
tities, the “co-ordinates,” which are continuous functions
within the manifold. This does not mean that the whole man-
ifold with all its elements must be represented in a single and
reversible manner by value systems of n co-ordinates (e.g. this is
impossible in the case of the sphere, for which n = 2); it signi-
fies only that if P is an arbitrary element of the manifold, then
in every case a certain domain surrounding the point P must be
representable singly and reversibly by the value system of n co-
ordinates. If xi is a system of n co-ordinates, x0i another system
of n co-ordinates, then the co-ordinate values xi , x0i of the same
element will in general be connected with one another by relations

xi = fi (x01 , x02 , . . . , x0n ) (i = 1, 2, . . . , n) (3)

which can be resolved into terms of x0i and in which the fi ’s are
continuous functions of their arguments. As long as nothing more
is known about the manifold, we cannot distinguish any one co-
ordinate system from the others. For an analytical treatment of
arbitrary continuous manifolds we thus require a theory of in-
variance with regard to arbitrary transformation of co-ordinates,
such as (3), whereas for the development of affine geometry in the
preceding chapter we used only the much more special theory of
invariance for the case of linear transformations.
Differential geometry deals with curves and surfaces in three-
dimensional Euclidean space; we shall here consider them mapped
out in Cartesian co-ordinates x, y, z. A curve is in general a
one-dimensional point-manifold; its separate points can be distin-
guished from one another by the values of a parameter u. If the
 CHAPTER II 126

point u on the curve happens to be at the point x, y, z in space,

then x, y, z will be certain continuous functions of u:

x = x(u), y = y(u), z = z(u) (4)

and (4) is called the “parametric” representation of the curve. If

we interpret u as the time, then (4) is the law of motion of a
point which traverses the given curve. The curve itself does not,
however, determine singly the parametric representation (4) of the
curve; the parameter u may, indeed, be subjected to any arbitrary
continuous transformation.
A two-dimensional point-manifold is called a surface. Its
points can be distinguished from one another by the values of two
parameters u1 , u2 . It may therefore be represented parametrically
in the form

x = x(u1 , u2 ), y = y(u1 , u2 ), z = z(u1 , u2 ). (5)

The parameters u1 , u2 may likewise undergo any arbitrary contin-

uous transformation without affecting the represented curve. We
shall assume that the functions (5) are not only continuous but
have also continuous differential co-efficients. Gauss, in his general
theory, starts from the form (5) of representing any surface; the
parameters u1 , u2 are hence called the Gaussian (or curvilinear)
co-ordinates on the surface. For example, if, as in the preceding
section, we project the points of the surface of the unit sphere in a
small region encircling the origin of the co-ordinate system on to
the tangent plane z = 1 at the south pole, and if we make x, y, z
the co-ordinates of any arbitrary point on the sphere, u1 and u2
being respectively the x and y co-ordinates of the point of projec-
 THE METRICAL CONTINUUM 127

tion in this plane, then

u1 u2 1
x= p 2 2
, y=p 2 2
, z=p . (6)
1 + u1 + u2 1 + u1 + u2 1 + u21 + u22

This is a parametric representation of the sphere. It does not, how-

ever, embrace the whole sphere, but only a certain region round
the south pole, viz. the part from the south pole to the equator,
excluding the latter. Another illustration of a parametric repre-
sentation is given by the geographical co-ordinates, latitude and
longitude.
In thermodynamics we use a graphical representation consist-
ing of a plane on which two rectangular co-ordinate axes are
drawn, and in which the state of a gas as denoted by its pressure p
and temperature θ is represented by a point having the rectangu-
lar co-ordinates p, θ. The same procedure may be adopted here.
With the point u1 , u2 on the surface, we associate a point in the
“representative” plane having the rectangular co-ordinates u1 , u2 .
The formulæ (5) do not then represent only the surface, but also
at the same time a definite continuous representation of this
surface on the u1 , u2 plane. Geographical maps are familiar in-
stances of such representations of curved portions of surface by
means of planes. A curve on a surface is given mathematically by
a parametric representation

u1 = u1 (t), u2 = u2 (t), (7)

whereas a portion of a surface is given by a “mathematical re-

gion” expressed in the variables u1 , u2 , and which must be char-
acterised by inequalities involving u1 and u2 ; i.e. graphically by
means of the representative curve or the representative region in
 CHAPTER II 128

the u1 -u2 -plane. If the representative plane be marked out with a

network of co-ordinates in the manner of squared paper, then this
becomes transposed, through the representation, to the curved
surface as a net consisting of meshes having the form of little
parallelograms, and composed of the two families of “co-ordinate
lines” u1 = const., u2 = const., respectively. If the meshes be
made sufficiently fine it becomes possible to map out any given
figure of the representative plane on the curved surface.
The distance ds between two infinitely near points of the sur-
face, namely,

(u1 , u2 ) and (u1 + du1 , u2 + du2 )

is determined by the expression

ds2 = dx2 + dy 2 + dz 2

if we set
∂x ∂x
dx = du1 + du2 (8)
∂u1 ∂u2
in it, with corresponding expressions for dy and dz. We then get
a quadratic differential form for ds2 thus:
2
X
2
ds = gik dui duk (gki = gik ) (9)
i,k=1

in which the co-efficients are

∂x ∂x ∂y ∂y ∂z ∂z
gik = + +
∂ui ∂uk ∂ui ∂uk ∂ui ∂uk
and are not, in general, functions of u1 and u2 .
 THE METRICAL CONTINUUM 129

In the case of the parametric representation of the sphere (6)

we have
(1 + u21 + u22 )(du21 + du22 ) − (u1 du1 + u2 du2 )2
ds2 = . (10)
(1 + u21 + u22 )2
Gauss was the first to recognise that the metrical groundform is
the determining factor for geometry on surfaces. The lengths of
curves, angles, and the size of given regions on the surface depend
on it alone. The geometries on two different surfaces is accordingly
identical if, for a representation in appropriate parameters, the co-
efficients gik of the metrical groundform coincide in value.
Proof.—The length of any arbitrary curve, given by (7), on the
surface is furnished by the integral
Z sX
dui duk
Z
ds = gik · dt.
i,k
dt dt

If we fix our attention on a definite point P 0 = (u01 , u02 ) on the

surface and use the relative co-ordinates

ui − u0i = dui , x − x0 = dx, y − y 0 = dy, z − z 0 = dz

for its immediate neighbourhood, then equation (8), in which the

derivatives are to be taken for the point P 0 , will hold more exactly
the smaller du1 , du2 , are taken; we say that it holds for “infinitely
small” values du1 and du2 . If we add to these the analogous equa-
tions for dy and dz, then they express that the immediate neigh-
bourhood of P 0 is a plane, and that du1 , du2 are affine co-ordinates
on it.∗ Accordingly we may apply the formulæ of affine geometry
∗
We here assume that the determinants of the second order which can be
 CHAPTER II 130

to the region immediately adjacent to P 0 . For the angle θ be-

tween two line-elements or infinitesimal displacements having the
components du1 , du2 and δu1 , δu2 respectively, we get
Q(d, δ)
cos θ = p
Q(d, d)Q(δ, δ)
in which Q(d, δ) stands for the symmetrical bilinear form
X
gik dui δuk corresponding to (9).
i,k

The area of the infinitesimal parallelogram marked out by these

two displacements is found to be
 
√ du1 du2 
g
δu1 δu2 
in which g denotes the determinant of the gik ’s. The area of a
curved portion of surface is accordingly given by the integral
√
ZZ
g du1 du2

formed from the table of co-efficients of these equations,

 ∂x ∂y ∂z 
 
 
 ∂u1 ∂u1 ∂u1 
 ∂x
 ∂y ∂z  ,
∂u2 ∂u2 ∂u2
 

do not all vanish. This condition is fulfilled for the regular points of the sur-
face, at which there is a tangent plane. The three determinants are identically
equal to 0, if, and only if, the surface degenerates to a curve, i.e. the functions
x, y, z of u1 and u2 actually depend only on one parameter, a function of
u1 and u2 .
 THE METRICAL CONTINUUM 131

taken over the corresponding part of the representative plane.

This proves Gauss’ statement. The values of the expressions ob-
tained are of course independent of the choice of parametric rep-
resentation. This invariance with respect to arbitrary transfor-
mations of the parameters can easily be confirmed analytically.
All the geometric relations holding on the surface can be studied
on the representative plane. The geometry of this plane is the
same as that of the curved surface if we agree to accept the dis-
tance ds of two infinitely near points as expressed by (9) and not
by Pythagoras’ formula

ds2 = du21 + du22 .

The geometry of the surface deals with the inner measure rela-
tions of the surface that belong to it independently of the manner
in which it is embedded in space. They are the relations that can
be determined by measurements carried out on the surface
itself. Gauss in his investigation of the theory of surfaces started
from the practical task of surveying Hanover geodetically. The
fact that the earth is not a plane can be ascertained by measuring
a sufficiently large portion of the earth’s surface. Even if each
single triangle of the network is taken too small for the deviation
from a plane to come into consideration, they cannot be put to-
gether to form a closed net on a plane in the way they do on the
earth’s surface. To show this a little more clearly let us draw a
circle C on a sphere of radius unity (the earth), having its cen-
tre P on the surface of the sphere. Let us further draw radii of
this circle, i.e. arcs of great circles of the sphere radiating from P
π
and ending at the circumference of C (let these arcs be < ). By
2
carrying out measurements on the sphere’s surface we can now
 CHAPTER II 132

ascertain that these radii starting out in all directions are the
shortest lines connecting P to the circle C, and that they are all
of the same length r; by measurement we find the closed curve C
to be of length s. If we were dealing with a plane we should infer
from this that the “radii” are straight lines and hence the curve C
would be a circle and we should expect s to be equal to 2πr. In-
stead of this, however, we find that s is less than the value given
by the above formula, for in the actual case s = 2π sin r. We thus
discover by measurements carried out on the surface of the sphere
that this surface is not a plane. If, on the other hand, we draw
figures on a sheet of paper and then roll it up, we shall find the
same values for measurements of these figures in their new condi-
tion as before, provided that no distortion has occurred through
rolling up the paper. The same geometry will hold on it now as on
the plane. It is impossible for me to ascertain that it is curved by
carrying out geodetic measurements. Thus, in general, the same
geometry holds for two surfaces that can be transformed into one
another without distortion or tearing.
The fact that plane geometry does not hold on the sphere
means analytically that it is impossible to convert the quadratic
differential form (10) by means of a transformation
u1 = u1 (u01 , u02 ) u01 = u01 (u1 , u2 )
u2 = u2 (u01 , u02 ) u02 = u02 (u1 , u2 )
into the form
(du01 )2 + (du02 )2 .
We know, indeed, that it is possible to do this for each point by a
linear transformation of the differentials, viz. by
du0i = αi1 du1 + αi2 du2 (i = 1, 2), (11)
 THE METRICAL CONTINUUM 133

but it is impossible to choose the transformation of the differ-

entials at each point so that the expressions (11) become total
differentials for du01 , du02 .
Curvilinear co-ordinates are used not only in the theory of
surfaces but also in the treatment of space problems, particularly
in mathematical physics in which it is often necessary to adapt
the co-ordinate system to the bodies presented, as is instanced in
the case of cylindrical, spherical, and elliptic co-ordinates. The
square of the distance, ds2 , between two infinitely near points in
space, is always expressed by a quadratic form
3
X
gik dxi dxk (12)
i,k=1

in which x1 , x2 , x3 are any arbitrary co-ordinates. If we uphold

Euclidean geometry, we express the belief that this quadratic form
can be brought by means of some transformation into one which
has constant co-efficients.
These introductory remarks enable us to grasp the full mean-
ing of the ideas developed fully by Riemann in his inaugural ad-
dress, “Concerning the Hypotheses which lie at the Base of Ge-
ometry”.∗ It is evident from Chapter I that Euclidean geome-
try holds for a three-dimensional linear point-configuration in a
four-dimensional Euclidean space; but curved three-dimensional
spaces, which exist in four-dimensional space just as much as
curved surfaces occur in three-dimensional space, are of a dif-
ferent type. Is it not possible that our three-dimensional space
of ordinary experience is curved? Certainly. It is not embedded
in a four-dimensional space; but it is conceivable that its inner
∗
Vide note 4.
 CHAPTER II 134

measure-relations are such as cannot occur in a “plane” space; it

is conceivable that a very careful geodetic survey of our space car-
ried out in the same way as the above-mentioned survey of the
earth’s surface might disclose that it is not plane. We shall con-
tinue to regard it as a three-dimensional manifold, and to suppose
that infinitesimal line elements may be compared with one another
in respect to length independently of their position and direction,
and that the square of their lengths, the distance between two
infinitely near points, may be expressed by a quadratic form (12),
any arbitrary co-ordinates xi being used. (There is a very good
reason for this assumption; for, since every transformation from
one co-ordinate system to another entails linear transformation-
formulæ for the co-ordinate differentials, a quadratic form must
always again pass into a quadratic form as a result of the transfor-
mation.) We no longer assume, however, that these co-ordinates
may in particular be chosen as affine co-ordinates such that they
make the co-efficients gik of the groundform become constant.
The transition from Euclidean geometry to that of Riemann
is founded in principle on the same idea as that which led from
physics based on action at a distance to physics based on infinitely
near action. We find by observation, for example, that the current
flowing along a conducting wire is proportional to the difference
of potential between the ends of the wire (Ohm’s Law). But we
are firmly convinced that this result of measurement applied to
a long wire does not represent a physical law in its most general
form; we accordingly deduce this law by reducing the measure-
ments obtained to an infinitely small portion of wire. By this
means we arrive at the expression (Chap. I, page 111) on which
Maxwell’s theory is founded. Proceeding in the reverse direction,
we derive from this differential law by mathematical processes
 THE METRICAL CONTINUUM 135

the integral law, which we observe directly, on the supposition

that conditions are everywhere similar (homogeneity). We
have the same circumstances here. The fundamental fact of Eu-
clidean geometry is that the square of the distance between two
points is a quadratic form of the relative co-ordinates of the two
points (Pythagoras’ Theorem). But if we look upon this law as
being strictly valid only for the case when these two points are in-
finitely near, we enter the domain of Riemann’s geometry. This at
the same time allows us to dispense with defining the co-ordinates
more exactly since Pythagoras’ Law expressed in this form (i.e. for
infinitesimal distances) is invariant for arbitrary transformations.
We pass from Euclidean “finite” geometry to Riemann’s “infinitesi-
mal” geometry in a manner exactly analogous to that by which we
pass from “finite” physics to “infinitesimal” (or “contact”) physics.
Riemann’s geometry is Euclidean geometry formulated to meet
the requirements of continuity, and in virtue of this formulation it
assumes a much more general character. Euclidean finite geom-
etry is the appropriate instrument for investigating the straight
line and the plane, and the treatment of these problems directed
its development. As soon as we pass over to differential geome-
try, it becomes natural and reasonable to start from the property
of infinitesimals set out by Riemann. This gives rise to no com-
plications, and excludes all speculative considerations tending to
overstep the boundaries of geometry. In Riemann’s space, too,
a surface, being a two-dimensional manifold, may be represented
parametrically in the form xi = xi (u1 , u2 ). If we substitute the
resulting differentials,
∂xi ∂xi
dxi = · du1 + · du2
∂u1 ∂u2
 CHAPTER II 136

in the metrical groundform (12) of Riemann’s space, we get for the

square of the distance between two infinitely near surface-points
a quadratic differential form in du1 , du2 (as in Euclidean space).
The measure-relations of three-dimensional Riemann space may
be applied directly to any surface existing in it, and thus converts
it into a two-dimensional Riemann space. Whereas from the Eu-
clidean standpoint space is assumed at the very outset to be of
a much simpler character than the surfaces possible in it, viz. to
be rectangular, Riemann has generalised the conception of space
just sufficiently far to overcome this discrepancy. The princi-
ple of gaining knowledge of the external world from the
behaviour of its infinitesimal parts is the mainspring of the
theory of knowledge in infinitesimal physics as in Riemann’s ge-
ometry, and, indeed, the mainspring of all the eminent work of
Riemann, in particular, that dealing with the theory of complex
functions. The question of the validity of the “fifth postulate,” on
which historical development started its attack on Euclid, seems
to us nowadays to be a somewhat accidental point of departure.
The knowledge that was necessary to take us beyond the Euclidean
view was, in our opinion, revealed by Riemann.
We have yet to convince ourselves that the geometry of Bolyai
and Lobatschefsky as well as that of Euclid and also spherical ge-
ometry (Riemann was the first to point out that the latter was a
possible case of non-Euclidean geometry) are all included as par-
ticular cases in Riemann’s geometry. We find, in fact, that if we
denote a point in the Bolyai-Lobatschefsky plane by the rectangu-
lar co-ordinates u1 , u2 of its corresponding point in Klein’s model
the distance ds between two infinitely near points is by (1)
(1 − u21 − u22 )(du21 + du22 ) + (u1 du1 + u2 du2 )2
ds2 = . (13)
(1 − u21 − u22 )2
 THE METRICAL CONTINUUM 137

By comparing this with (10) we see that the Theorem of Taurinus

is again confirmed. The metrical groundform of three-dimensional
non-Euclidean space corresponds exactly to this expression.
If we can find a curved surface in Euclidean
space for which formula (13) holds, provided ap-
propriate Gaussian co-ordinates u1 , u2 be chosen,
then the geometry of Bolyai and Lobatschefsky is
valid on it. Such surfaces can actually be con-
structed; the simplest is the surface of revolution
derived from the tractrix. The tractrix is a plane
curve of the shape shown in Fig. 5, with one vertex
and one asymptote. It is characterised geometri-
cally by the property that any tangent measured
from the point of contact to the point of inter-
section with the asymptote is of constant length.
Suppose the curve to revolve about its asymptote Fig. 5.
as axis. Non-Euclidean geometry holds on the
surface generated. This Euclidean model of striking simplicity
was first mentioned by Beltrami (vide note 5). There are certain
shortcomings in it; in the first place the form in which it is pre-
sented confines it to two-dimensional geometry; secondly, each of
the two halves of the surface of revolution into which the sharp
edge divides it represents only a part of the non-Euclidean plane.
Hilbert proved rigorously that there cannot be a surface free from
singularities in Euclidean space which pictures the whole of Lo-
batschefsky’s plane (vide note 6). Both of these weaknesses are
absent in the elementary geometrical model of Klein.
So far we have pursued a speculative train of thought and have
kept within the boundaries of mathematics. There is, however, a
difference in demonstrating the consistency of non-Euclidean ge-
 CHAPTER II 138

ometry and inquiring whether it or Euclidean geometry

holds in actual space. To decide this question Gauss long ago
measured the triangle having for its vertices Inselsberg, Brocken,
and Hoher Hagen (near Göttingen), using methods of the greatest
refinement, but the deviation of the sum of the angles from 180◦
was found to lie within the limits of errors of observation. Lo-
batschefsky concluded from the very small value of the parallaxes
of the stars that actual space could differ from Euclidean space
only by an extraordinarily small amount. Philosophers have put
forward the thesis that the validity or non-validity of Euclidean
geometry cannot be proved by empirical observations. It must
in fact be granted that in all such observations essentially physi-
cal assumptions, such as the statement that the path of a ray of
light is a straight line and other similar statements, play a promi-
nent part. This merely bears out the remark already made above
that it is only the whole composed of geometry and physics that
may be tested empirically. Conclusive experiments are thus pos-
sible only if physics in addition to geometry is worked out for
Euclidean space and generalised Riemann space. We shall soon
see that without making artificial limitations we can easily trans-
late the laws of the electromagnetic field, which were originally set
up on the basis of Euclidean geometry, into terms of Riemann’s
space. Once this has been done there is no reason why experience
should not decide whether the special view of Euclidean geometry
or the more general one of Riemann geometry is to be upheld. It
is clear that at the present stage this question is not yet ripe for
discussion.
In this concluding paragraph we shall once again present the
foundations of Riemann’s geometry in the form of a résumé, in
which we do not restrict ourselves to the dimensional number
n = 3.
 THE METRICAL CONTINUUM 139

An n-dimensional Riemann space is an n-dimensional mani-

fold, not of an arbitrary nature, but one which derives its measure-
relations from a definitely positive quadratic differential form. The
two principal laws according to which this form determines the
metrical quantities are expressed in (1) and (2) in which the xi ’s
denote any co-ordinates whatsoever.
1. If g is the determinant of the co-efficients of the groundform,
then the size of any portion of space is given by the integral
√
Z
g dx1 dx2 . . . dxn (14)

which is to be taken over the mathematical region of the vari-

ables xi , which corresponds to the portion of space in question.
2. If Q(d, δ) denote the symmetrical bilinear form, correspond-
ing to the quadratic groundform, of two line elements d and δ sit-
uated at the same point, then the angle θ between them is given
by
Q(d, δ)
cos θ = p . (15)
Q(d, d) · Q(δ, δ)
An m-dimensional manifold existing in n-dimensional space
(1 ≤ m ≤ n) is given in parametric terms by

xi = xi (u1 , u2 , . . . , um ) (i = 1, 2, . . . , n).

By substituting the differentials

∂xi ∂xi ∂xi
dxi = · du1 + · du2 + · · · + · dum
∂u1 ∂u2 ∂um
in the metrical groundform of space we get the metrical ground-
form of this m-dimensional manifold. The latter is thus itself an
 CHAPTER II 140

m-dimensional Riemann space, and the size of any portion of it

may be calculated from formula (14) in the case m = n. In this
way the lengths of segments of lines and the areas of portions of
surfaces may be determined.

§ 12. Continuation. Dynamical View of Metrical

Properties
We shall now revert to the theory of surfaces in Euclidean
space. The curvature of a plane curve may be defined in the
following way as the measure of the rate at which the normals
to the curve diverge. From a fixed point O we trace out the
vector Op, the “normal” to the curve at an arbitrary point P , and
make it of unit length. This gives us a point P , corresponding
to P , on the circle of radius unity. If P traverses a small arc ∆s of
the curve, the corresponding point p will traverse an arc ∆σ of the
circle; ∆σ is the plane angle which is the sum of the angles that the
normals erected at all points of the arc of the curve make with their
∆σ
respective neighbours. The limiting value of the quotient for
∆s
an element of arc ∆s which contracts to a point P is the curvature

∆σ
P ∆s P
O

Fig. 6.
 THE METRICAL CONTINUUM 141

at P . Gauss defined the curvature of a surface as the measure

of the rate at which its normals diverge in an exactly analogous
manner. In place of the unit circle about O, he uses the unit
sphere. Applying the same method of representation he makes a
small portion dω of this sphere correspond to a small area do of
the surface; dω is equal to the solid angle formed by the normals
dω
erected at the points of do. The ratio for the limiting case
do
when do becomes vanishingly small is the Gaussian measure of
curvature. Gauss made the important discovery that this curvature
is determined by the inner measure-relations of the surface alone,
and that it can be calculated from the co-efficients of the metrical
groundform as a differential expression of the second order. The
curvature accordingly remains unaltered if the surface be bent
without being distorted by stretching. By this geometrical means
a differential invariant of the quadratic differential forms
of two variables was discovered, that is to say, a quantity was
found, formed of the co-efficients of the differential form in such
a way that its value was the same for two differential forms that
arise from each other by a transformation (and also for parametric
pairs which correspond to one another in the transformation).
Riemann succeeded in extending the conception of curvature
to quadratic forms of three and more variables. He then found that
it was no longer a scalar but a tensor (we shall discuss this in § 15
of the present chapter). More precisely it may be stated that Rie-
mann’s space has a definite curvature at every point in the normal
direction of every surface. The characteristic of Euclidean space
is that its curvature is nil at every point and in every direction.
Both in the case of Bolyai-Lobatschefsky’s geometry and spherical
geometry the curvature has a value a independent of the place and
 CHAPTER II 142

of the surface passing through it: this value is positive in the case
of spherical geometry, negative in that of Bolyai-Lobatschefsky.
(It may therefore be put = ±1 if a suitable unit of length be cho-
sen.) If an n-dimensional space has a constant curvature a, then
if we choose appropriate co-ordinates xi , its metrical groundform
must be of the form
(1 + a i x2i ) · i dx2i − a ( i xi dxi )2
P P P
2 .
(1 + a i x2i )
P

It is thus completely defined in a single-valued manner. If space

is everywhere homogeneous in all directions, its curvature must
be constant, and consequently its metrical groundform must be
of the form just given. Such a space is necessarily either Eu-
clidean, spherical, or Lobatschefskyan. Under these circumstances
not only have the line elements an existence which is independent
of place and direction, but any arbitrary finitely extended figure
may be transferred to any arbitrary place and put in any arbitrary
direction without altering its metrical conditions, i.e. its displace-
ments are congruent. This brings us back to congruent transfor-
mations which we used as a starting-point for our reflections on
space in § 1. Of these three possible cases the Euclidean one is
characterised by the circumstance that the group of translations
having the special properties set out in § 1 are unique in the group
of congruent transformations. The facts which are summarised in
this paragraph are mentioned briefly in Riemann’s essay; they
have been discussed in greater detail by Christoffel, Lipschitz,
Helmholtz, and Sophus Lie (vide note 7).
Space is a form of phenomena, and, by being so, is necessar-
ily homogeneous. It would appear from this that out of the rich
abundance of possible geometries included in Riemann’s concep-
 THE METRICAL CONTINUUM 143

tion only the three special cases mentioned come into consider-
ation from the outset, and that all the others must be rejected
without further examination as being of no account: parturiunt
montes, nascetur ridiculus mus! Riemann held a different opin-
ion, as is evidenced by the concluding remarks of his essay. Their
full purport was not grasped by his contemporaries, and his words
died away almost unheard (with the exception of a solitary echo
in the writings of W. K. Clifford). Only now that Einstein has
removed the scales from our eyes by the magic light of his theory
of gravitation do we see what these words actually mean. To make
them quite clear I must begin by remarking that Riemann con-
trasts discrete manifolds, i.e. those composed of single isolated
elements, with continuous manifolds. The measure of every part
of such a discrete manifold is determined by the number of ele-
ments belonging to it. Hence, as Riemann expresses it, a discrete
manifold has the principle of its metrical relations in itself, a pri-
ori, as a consequence of the concept of number. In Riemann’s own
words:—
“The question of the validity of the hypotheses of geometry in
the infinitely small is bound up with the question of the ground
of the metrical relations of space. In this question, which we may
still regard as belonging to the doctrine of space, is found the ap-
plication of the remark made above; that in a discrete manifold,
the principle or character of its metric relations is already given in
the notion of the manifold, whereas in a continuous manifold this
ground has to be found elsewhere, i.e. has to come from outside.
Either, therefore, the reality which underlies space must form a
discrete manifold, or we must seek the ground of its metric rela-
tions (measure-conditions) outside it, in binding forces which act
upon it.
 CHAPTER II 144

“A decisive answer to these questions can be obtained only by

starting from the conception of phenomena which has hitherto
been justified by experience, to which Newton laid the founda-
tion, and then making in this conception the successive changes
required by facts which admit of no explanation on the old the-
ory; researches of this kind, which commence with general notions,
cannot be other than useful in preventing the work from being
hampered by too narrow views, and in keeping progress in the
knowledge of the inter-connections of things from being checked
by traditional prejudices.
“This carries us over into the sphere of another science, that
of physics, into which the character and purpose of the present
discussion will not allow us to enter.”
If we discard the first possibility, “that the reality which under-
lies space forms a discrete manifold”—although we do not by this
in any way mean to deny finally, particularly nowadays in view of
the results of the quantum-theory, that the ultimate solution of the
problem of space may after all be found in just this possibility—
we see that Riemann rejects the opinion that had prevailed up
to his own time, namely, that the metrical structure of space is
fixed and inherently independent of the physical phenomena for
which it serves as a background, and that the real content takes
possession of it as of residential flats. He asserts, on the contrary,
that space in itself is nothing more than a three-dimensional man-
ifold devoid of all form; it acquires a definite form only through
the advent of the material content filling it and determining its
metric relations. There remains the problem of ascertaining the
laws in accordance with which this is brought about. In any case,
however, the metrical groundform will alter in the course of time
just as the disposition of matter in the world changes. We recover
 THE METRICAL CONTINUUM 145

the possibility of displacing a body without altering its metric re-

lations by making the body carry along with it the “metrical field”
which it has produced (and which is represented by the metrical
groundform); just as a mass, having assumed a definite shape in
equilibrium under the influence of the field of force which it has
itself produced, would become deformed if one could keep the field
of force fixed while displacing the mass to another position in it;
whereas, in reality, it retains its shape during motion (supposed to
be sufficiently slow), since it carries the field of force, which it has
produced, along with itself. We shall illustrate in greater detail
this bold idea of Riemann concerning the metrical field produced
by matter, and we shall show that if his opinion is correct, any two
portions of space which can be transformed into one another by
a continuous deformation, must be recognised as being congruent
in the sense we have adopted, and that the same material content
can fill one portion of space just as well as the other.
To simplify this examination of the underlying principles we
assume that the material content can be described fully by scalar
phase quantities such as mass-density, density of charge, and so
forth. We fix our attention on a definite moment of time. Dur-
ing this moment the density ρ of charge, for example, will, if we
choose a certain co-ordinate system in space, be a definite func-
tion f (x1 , x2 , x3 ) of the co-ordinates x1 but will be represented
by a different function f ∗ (x∗1 , x∗2 , x∗3 ) if we use another co-ordinate
system in x∗i . A parenthetical note. Beginners are often confused
by failing to notice that in mathematical literature symbols are
used throughout to designate functions, whereas in physical lit-
erature (including the mathematical treatment of physics) they
are used exclusively to denote “magnitudes” (quantities). For
example, in thermodynamics the energy of a gas is denoted by a
 CHAPTER II 146

definite letter, say E, irrespective of whether it is a function of the

pressure p and the temperature θ or a function of the volume v
and the temperature θ. The mathematician, however, uses two
different symbols to express this:—
E = φ(p, θ) = ψ(v, θ).
∂φ ∂ψ
The partial derivatives , , which are totally different in
∂θ ∂θ
meaning, consequently occur in physics books under the common
∂E
expression . A suffix must be added (as was done by Boltz-
∂θ
mann), or it must be made clear in the text that in one case p, in
the other case v, is kept constant. The symbolism of the mathe-
matician is clear without any such addition.∗
Although the true state of things is really more complex we
shall assume the most simple system of geometrical optics, the fun-
damental law of which states that the ray of light from a point M
emitting light to an observer at P is a “geodetic” line, which is the
shortest of all the lines connecting M with P : we take no account
of the finite velocity with which light is propagated. We ascribe
to the receiving consciousness merely an optical faculty of percep-
tion and simplify this to a “point-eye” that immediately observes
the differences of direction of the impinging rays, these directions
being the values of θ given by (15); the “point-eye” thus obtains
a picture of the directions in which the surrounding objects lie
(colour factors are ignored). The Law of Continuity governs not
only the action of physical things on one another but also psycho-
physical interactions. The direction in which we observe objects
∗
This is not to be taken as a criticism of the physicist’s nomenclature
which is fully adequate to the purposes of physics, which deals with magni-
tudes.
 THE METRICAL CONTINUUM 147

is determined not by their places of occupation alone, but also by

the direction of the ray from them that strikes the retina, that
is, by the state of the optical field directly in contact with that
elusive body of reality whose essence it is to have an objective
world presented to it in the form of experiences of consciousness.
To say that a material content G is the same as the material con-
tent G0 can obviously mean no more than saying that to every
point of view P with respect to G there corresponds a point of
view P 0 with respect to G0 (and conversely) in such a way that
an observer at P 0 in G0 receives the same “direction-picture” as an
observer in G receives at P .
Let us take as a basis a definite co-ordinate system xi . The
scalar phase-quantities, such as density of electrification ρ, are
then represented by definite functions
ρ = f (x1 , x2 , x3 ).
Let the metrical groundform be
3
X
gik dxi dxk
i,k=1

in which the gik ’s likewise (in “mathematical” terminology) denote

definite functions of x1 , x2 , x3 . Furthermore, suppose any contin-
uous transformation of space into itself to be given, by which a
point P 0 corresponds to each point P respectively. Using this
co-ordinate system and the modes of expression
P = (x1 , x2 , x3 ), P 0 = (x01 , x02 , x03 ),
suppose the transformation to be represented by
x0i = φ(x1 , x2 , x3 ). (16)
 CHAPTER II 148

Suppose this transformation convert the portion S of space into S0 ,

I shall show that if Riemann’s view is correct S0 is congruent
with S in the sense defined.
I make use of a second co-ordinate system by taking as co-
ordinates of the point P the values of x0i given by (16); the ex-
pressions (16) then become the formulæ of transformation. The
mathematical region in three variables represented by S in the
co-ordinates x0 is identical with that represented by S0 in the co-
ordinates x. An arbitrary point P has the same co-ordinates in x0
as P 0 has in x. I now imagine space to be filled by matter in some
other way, namely, that represented by the formulæ

ρ = f (x01 , x02 , x03 )

at the point P , with similar formulæ for the other scalar quanti-
ties. If the metric relations of space are taken to be independent
of the contained matter, the metrical groundform will, as in the
case of the first content, be of the form
X X
0
gik dxi dxk = gik (x01 , x02 , x03 ) dx0i dx0k ,
i,k i,k

the right-hand member of which denotes the expression after

transformation to the new co-ordinate system. If, however, the
metric relations of space are determined by the matter filling
it—we assume, with Riemann, that this is actually so—then,
since the second occupation by matter expresses itself in the
co-ordinates x0 in exactly the same way as does the first in x, the
metrical groundform for the second occupation will be
X
gik (x01 , x02 , x03 ) dx0i dx0k .
i,k
 THE METRICAL CONTINUUM 149

In consequence of our underlying principle of geometrical optics

assumed above, the content in the portion S0 of space during the
first occupation will present exactly the same appearance to an
observer at P 0 as the material content in S during the second
occupation presents to an observer at P . If the older view of
“residential flats” is correct, this would of course not be the case.
The simple fact that I can squeeze a ball of modelling clay with
my hands into any irregular shape totally different from a sphere
would seem to reduce Riemann’s view to an absurdity. This, how-
ever, proves nothing. For if Riemann is right, a deformation of the
inner atomic structure of the clay is entirely different from that
which I can effect with my hands, and a rearrangement of the
masses in the universe, would be necessary to make the distorted
ball of clay appear spherical to an observer from all points of view.
The essential point is that a piece of space has no visual form at
all, but that this form depends on the material content occupying
the world, and, indeed, occupying it in such a way that by means
of an appropriate rearrangement of the mode of occupation I can
give it any visual form. By this I can also metamorphose any
two different pieces of space into the same visual form by choos-
ing an appropriate disposition of the matter. Einstein helped to
lead Riemann’s ideas to victory (although he was not directly in-
fluenced by Riemann). Looking back from the stage to which
Einstein has brought us, we now recognise that these ideas could
give rise to a valid theory only after time had been added as a
fourth dimension to the three-space dimensions in the manner set
forth in the so-called special theory of relativity. As, according to
Riemann, the conception “congruence” leads to no metrical sys-
tem at all, not even to the general metrical system of Riemann,
which is governed by a quadratic differential form, we see that
 CHAPTER II 150

“the inner ground of the metric relations” must indeed be sought

elsewhere. Einstein affirms that it is to be found in the “binding
forces” of Gravitation. In Einstein’s theory (Chapter IV) the
co-efficients gik of the metrical groundform play the same part
as does gravitational potential in Newton’s theory of gravitation.
The laws according to which space-filling matter determines the
metrical structure are the laws of gravitation. The gravitational
field affects light rays and “rigid” bodies used as measuring rods in
such a way that when we use these rods and rays in the usual man-
ner to take measurements of objects, a geometry of measurement
is found to hold which deviates very little from that of Euclid in
the regions accessible to observation. These metric relations are
not the outcome of space being a form of phenomena, but of the
physical behaviour of measuring rods and light rays as determined
by the gravitational field.
After Riemann had made known his discoveries, mathemati-
cians busied themselves with working out his system of geometri-
cal ideas formally; chief among these were Christoffel, Ricci, and
Levi-Civita (vide note 8). Riemann, in the last words of the above
quotation, clearly left the real development of his ideas in the
hands of some subsequent scientist whose genius as a physicist
could rise to equal flights with his own as a mathematician. After
a lapse of seventy years this mission has been fulfilled by Einstein.
Inspired by the weighty inferences of Einstein’s theory to exam-
ine the mathematical foundations anew the present writer made
the discovery that Riemann’s geometry goes only half-way towards
attaining the ideal of a pure infinitesimal geometry. It still remains
to eradicate the last element of geometry “at a distance,” a rem-
nant of its Euclidean past. Riemann assumes that it is possible
to compare the lengths of two line elements at different points
 THE METRICAL CONTINUUM 151

of space, too; it is not permissible to use comparisons at

a distance in an “infinitely near” geometry. One principle
alone is allowable; by this a division of length is transferable from
one point to that infinitely adjacent to it.
After these introductory remarks we now pass on to the sys-
tematic development of pure infinitesimal geometry (vide note 9),
which will be traced through three stages; from the continuum,
which eludes closer definition, by way of affinely connected
manifolds, to metrical space. This theory which, in my opin-
ion, is the climax of a wonderful sequence of logically-connected
ideas, and in which the result of these ideas has found its ultimate
shape, is a true geometry, a doctrine of space itself and not merely
like Euclid, and almost everything else that has been done under
the name of geometry, a doctrine of the configurations that are
possible in space.

§ 13. Tensors and Tensor-densities in any Arbitrary

Manifold
An n-dimensional Manifold.—Following the scheme out-
lined above we shall make the sole assumption about space that it
is an n-dimensional continuum. It may accordingly be referred to
n-co-ordinates x1 , x2 , . . . , xn , of which each has a definite numeri-
cal value at each point of the manifold; different value-systems of
the co-ordinates correspond to different points. If x̄1 , x̄2 , . . . x̄n is
a second system of co-ordinates, then there are certain relations

xi = fi (x̄1 , x̄2 , . . . x̄n ) where (i = 1, 2, . . . , n) (17)

between the x-co-ordinates and the x̄-co-ordinates; these relations

are conveyed by certain functions fi . We do not only assume that
 CHAPTER II 152

they are continuous, but also that they have continuous derivatives
∂fi
αki =
∂ x̄k
whose determinant is non-vanishing. The latter condition is neces-
sary and sufficient to make affine geometry hold in infinitely small
regions, that is, so that reversible linear relations exist between
the differentials of the co-ordinates in both systems, i.e.
X
dxi = αki dx̄k . (18)
k

We assume the existence and continuity of higher derivatives wher-

ever we find it necessary to use them in the course of our inves-
tigation. In every case, then, a meaning which is invariant and
independent of the co-ordinate system has been assigned to the
conception of continuous functions of a point which have contin-
uous first, second, third, or higher derivatives as required; the
co-ordinates themselves are such functions.
Conception of a Tensor.—The relative co-ordinates dx of a
point P 0 = (xi + dxi ) infinitely near to the point P = (xi ) are the
components of a line element at P or of an infinitesimal dis-
−−→
placement P P 0 of P . The transformation to another co-ordinate
system is effected for these components by formulæ (18), in which
αki denote the values of the respective derivatives at the point P .
The infinitesimal displacements play the same part in the devel-
opment of Tensor Calculus as do displacements in Chapter I. It
must, however, be noticed that, here, a displacement is essen-
tially bound to a point, and that there is no meaning in saying
that the infinitesimal displacements of two different points are the
equal or unequal. It might occur to us to adopt the convention
 THE METRICAL CONTINUUM 153

of calling the infinitesimal displacements of two points equal if

they have the same components; but it is obvious from the fact
that the αki ’s in (18) are not constants, that if this were the case
for one co-ordinate system it need in no wise be true for another.
Consequently we may only speak of the infinitesimal displacement
of a point and not, as in Chapter I, of the whole of space; hence
we cannot talk of a vector or tensor simply, but must talk of a
vector or tensor as being at a point P . A tensor at a point P
is a linear form, in several series of variables, which is dependent
on a co-ordinate system to which the immediate neighbourhood
of P is referred in the following way: the expressions of the lin-
ear form in any two co-ordinate systems x and x̄ pass into one
another if certain of the series of variables (with upper indices)
are transformed co-grediently, the remainder (with lower indices)
contra-grediently, to the differentials dxi , according to the scheme

αki ξ¯k and ξ¯i =

X X
ξi = αik ξk respectively. (19)
k k

By αki we mean the values of these derivatives at the point P .

The co-efficients of the linear form are called the components of
the tensor in the co-ordinate system under consideration; they are
co-variant in those indices that belong to the variables with an up-
per index, contra-variant in the remaining ones. The conception
of tensors is possible owing to the circumstance that the transition
from one co-ordinate system to another expresses itself as a lin-
ear transformation in the differentials. One here uses the exceed-
ingly fruitful mathematical device of making a problem “linear”
by reverting to infinitely small quantities. The whole of Tensor
Algebra, by whose operations only tensors at the same point
are associated, can now be taken over from Chapter I. Here,
 CHAPTER II 154

again, we shall call tensors of the first order vectors. There are
contra-variant and co-variant vectors. Whenever the word vector
is used without being defined more exactly we shall understand
it as meaning a contra-variant vector. Infinitesimal quantities of
this type are the line elements in P . Associated with every co-
ordinate system there are n “unit vectors” ei at P , namely, those
which have components
e1 1, 0, 0, . . . 0
e2 0, 1, 0, . . . 0
... .................
en 0, 0, 0, . . . 1
in the co-ordinate system. Every vector x at P may be expressed
in linear terms of these unit vectors. For if ξ i are its components,
then
x = ξ 1 e1 + ξ 2 e2 + · · · + ξ n en holds.
The unit vectors ēi of another co-ordinate system x̄ are derived
from the ei ’s according to the equations
X
ēi = αik ek .
k

The possibility of passing from co-variant to contra-variant com-

ponents of a tensor does not, of course, come into question here.
Each two linearly independent line elements having components
dxi , δxi map out a surface element whose components are

dxi δxk − dxk δxi = ∆xik .

Each three such line elements map out a three-dimensional space

element and so forth. Invariant differential forms that assign
 THE METRICAL CONTINUUM 155

a number linearly to each arbitrary line element, surface ele-

ment, etc., respectively are linear tensors (= co-variant skew-
symmetrical tensors, vide § 7). The above convention about
omitting signs of summation will be retained.
Conception of a Curve.—If to every value of a parameter s
a point P = P (s) is assigned in a continuous manner, then if we
interpret s as time, a “motion” is given. In default of a better
expression we shall apply this name in a purely mathematical
sense, even when we do not interpret s in this way. If we use a
definite co-ordinate system we may represent the motion in the
form
xi = xi (s) (20)
by means of n continuous functions xi (s), which we assume not
only to be continuous, but also continuously differentiable.∗ In
passing from the parametric value s to s + ds, the correspond-
ing point P suffers an infinitesimal displacement having compo-
nents dxi . If we divide this vector at P by ds, we get the “ve-
dxi
locity,” a vector at P having components = ui . The for-
ds
mulæ (20) is at the same time a parametric representation of the
trajectory of the motion. Two motions describe the same curve
if, and only if, the one motion arises from the other when the pa-
rameter s is subjected to a transformation s = ω(s̄), in which ω is
a continuous and continuously differentiable uniform function ω.
Not the components of velocity at a point are determinate for a
curve, but only their ratios (which characterise the direction of
the curve).
Tensor Analysis.—A tensor field of a certain kind is de-
fined in a region of space if to every point P of this region a tensor
∗
I.e. have continuous differential co-efficients.
 CHAPTER II 156

of this kind at P is assigned. Relatively to a co-ordinate system

the components of the tensor field appear as definite functions of
the co-ordinates of the variable “point of emergence” P : we as-
sume them to be continuous and to have continuous derivatives.
The Tensor Analysis worked out in Chapter I, § 8, cannot, without
alteration, be applied to any arbitrary continuum. For in defin-
ing the general process of differentiation we earlier used arbitrary
co-variant and contra-variant vectors, whose components were in-
dependent of the point in question. This condition is indeed
invariable for linear transformations, but not for any arbitrary
ones since, in these, the αki ’s are not constants. For an arbitrary
manifold we may, therefore, set up only the analysis of linear
tensor fields: this we proceed to show. Here, too, there is derived
from a scalar field f by means of differentiation, independently
of the co-ordinate system, a linear tensor field of the first order
having components
∂f
fi = . (21)
∂xi
From a linear tensor field fi of the first order we get one of the
second order
∂fi ∂fk
fik = − . (22)
∂xk ∂xi
From one of the second order, fik , we get a linear tensor field of
the third order
∂fkl ∂fli ∂fik
fikl = + + , (23)
∂xi ∂xk ∂xl
and so forth.
If φ is a given scalar field in space and if xi , x̄i denote any two
co-ordinate systems, then the scalar field will be expressed in each
 THE METRICAL CONTINUUM 157

in turn as a function of the xi ’s or x̄i ’s respectively, i.e.

φ = f (x1 , x2 , . . . , xn ) = f¯(x̄1 , x̄2 , . . . , x̄n ).

If we form the increase of φ for an infinitesimal displacement of

the current point, we get
X ∂f X ∂ f¯
dφ = dxi = dx̄i .
i
∂xi i
∂ x̄i

∂f
From this we see that the ’s are components of a co-variant
∂xi
tensor field of the first order, which is derived from the scalar
field φ in a manner independent of all co-ordinate systems. We
have here a simple illustration of the conception of vector fields.
At the same time we see that the operation “grad” is invariant
not only for linear transformations, but also for any arbitrary
transformations of the co-ordinates whatsoever, and this is what
we enunciated.
To arrive at (22) we perform the following construction. From
the point P = P00 we draw the two line elements with compo-
nents dxi and δxi , which lead to the two infinitely near points P10
and P01 . We displace (by “variation”) the line element dx in some
way so that its point of emergence describes the distance P00 P01 ;
−−−−→
suppose it to have got to P01 P11 finally. We shall call this pro-
cess the displacement δ. Let the components dxi have increased
by δdxi , so that
 
δdxi = xi (P11 ) − xi (P01 ) − xi (P10 ) − xi (P00 ) .

We now interchange d and δ. By an analogous displacement d of

the line element δx along P00 P10 , by which it finally takes up the
 CHAPTER II 158

−−−−→0
position P10 P11 , its components are increased by
0
 
dδxi = xi (P11 ) − xi (P10 ) − xi (P01 ) − xi (P00 ) .

Hence it follows that

0
δdxi − dδxi = xi (P11 ) − xi (P11 ). (24)
0
If, and only if, the two points P11 and P11 coincide, i.e. if the
two line elements dx and δx sweep out the same infinitesimal
“parallelogram” during their displacements δ and d respectively—
that is how we shall view it—then we shall have

δdxi − dδxi = 0. (25)

If, now, a co-variant vector field with components fi is given,

then we form the change in the invariant df = fi dxi owing to the
displacement δ thus:

δdf = δfi dxi + fi δdxi .

Interchanging d and δ, and then subtracting, we get

∆f = (δd − dδ)f = (δfi dxi − dfi δxi ) + fi (δdxi − dδxi )

and if both displacements pass over the same infinitesimal paral-

lelogram we get, in particular,
 
∂fi ∂fk
∆f = δfi dxi − dfi δxi = − dxi δxk . (26)
∂xk ∂xi
If one is inclined to distrust these perhaps too venturesome
operations with infinitesimal quantities the differentials may be
 THE METRICAL CONTINUUM 159

replaced by differential co-efficients. Since an infinitesimal element

of surface is only a part (or more correctly, the limiting value of
the part) of an arbitrarily small but finitely extended surface, the
argument will run as follows. Let a point (s, t) of our manifold be
assigned to every pair of values of two parameters s, t (in a certain
region encircling s = 0, t = 0). Let the functions xi = xi (s, t),
which represents this “two-dimensional motion” (extending over a
surface) in any co-ordinate system xi , have continuous first and
second differential co-efficients. For every point (s, t) there are two
dxi dxi
velocity vectors with components and . We may assign our
ds dt
parameters so that a prescribed point P = (0, 0) corresponds to
s = 0, t = 0, and that the two velocity vectors at it coincide with
two arbitrarily given vectors ui , v i (for this it is merely necessary
to make the xi ’s linear functions of s and t). Let d denote the
d d
differentiation , and δ denote . Then
ds dt
dxi ∂fi dxi dxk d2 xi
df = fi , δdf = + fi .
ds ∂xk ds dt dt ds
By interchanging d and δ, and then subtracting, we get
 
∂fi ∂fk dxi dxk
∆f = δdf − dδf = − . (27)
∂xk ∂xi ds dt
By setting s = 0 and t = 0, we get the invariant at the point P
 
∂fi ∂fk
− ui v k
∂xk ∂xi
which depends on two arbitrary vectors u, v at that point. The
connection between this view and that which uses infinitesimals
 CHAPTER II 160

consists in the fact that the latter is applied in rigorous form to the
infinitesimal parallelograms into which the surface xi = xi (s, t) is
divided by the co-ordinate lines s = const. and t = const.
Stokes’ Theorem may be recalled in this connection. The
invariant linear differential fi dxi is called integrable if its integral
along every closed curve (its “curl”) = 0. (This is true, as we know,
only for a total differential.) Let any arbitrary surface given in
a parametric form xi = xi (s, t) be spread out within the closed
curve, and be divided into infinitesimal parallelograms by the co-
ordinate lines. The curl taken around the perimeter of the whole
surface may then be traced back to the single curls around these
little surface meshes, and their values are given for every mesh
by our expression (27), after it has been multiplied by ds dt. A
differential division of the curl is produced in this way, and the
tensor (22) is a measure of the “intensity of the curl” at every
point.
In the same way we pass on to the next higher stage (23).
In place of the infinitesimal parallelogram we now use the three-
dimensional parallelepiped mapped out by the three line elements
d, δ, and ∂. We shall just indicate the steps of the argument
briefly.
∂fik
∂(fik dxi δxk ) = dxi δxk ∂xl +fik (∂dxi ·δxk +∂δxk ·dxi ). (28)
∂xl
Since fki = −fik , the second term on the right is

= fik (∂dxi · δxk − ∂δxi · dxk ). (29)

If we interchange d, δ, and ∂ cyclically, and then sum up, the six

members arising out of (29) will destroy each other in pairs on
account of the conditions of symmetry (25).
 THE METRICAL CONTINUUM 161

R
Conception of Tensor-density.—If W dx, in which
dx represents briefly the element of integration dx1 dx2 . . . dxn , is
an invariant integral, then W is a quantity dependent on the co-
ordinate system in such a way that, when transformed to another
co-ordinate system, its value become multiplied by the absolute
(numerical) value of the functional determinant. If we regard this
integral as a measure of the quantity of substance occupying the
region of integration, then W is its density. We may, therefore,
call a quantity of the kind described a scalar-density.
This is an important conception, equally as valuable as the
conception of scalars; it cannot be reduced to the latter. In an
analogous sense we may speak of tensor-densities as well as
scalar-densities. A linear form of several series of variables which
is dependent on the co-ordinate system, some of the variables car-
rying upper indices, others lower ones, is a tensor-density at a
point P , if, when the expression for this linear form is known for
a given co-ordinate system, its expression for any other arbitrary
co-ordinate system, distinguished by bars, is obtained by multi-
plying it with the absolute or numerical value of the functional
determinant

∆ = abs. |αik | i.e. the absolute value of |αik |,

and by transforming the variable according to the old scheme (19).

The words, components, co-variant, contra-variant, symmetrical,
skew-symmetrical, field, and so forth, are used exactly as in the
case of tensors. By contrasting tensors and tensor-densities, it
seems to me that we have grasped rigorously the difference be-
tween quantity and intensity, so far as this difference has a
physical meaning: tensors are the magnitudes of intensity,
tensor-densities those of quantity. The same unique part that
 CHAPTER II 162

co-variant skew-symmetrical tensors play among tensors is taken

among tensor-densities by contra-variant symmetrical tensor-
densities, which we shall term briefly linear tensor-densities.
Algebra of Tensor-densities.—As in the realm of tensors
so have here the following operations:—
1. Addition of tensor-densities of the same type; multiplication
of a tensor-density by a number.
2. Contraction.
3. Multiplication of a tensor by a tensor-density (not multi-
plication of two tensor-densities by each other). For, if two scalar-
densities, for example, were to be multiplied together, the result
would not again be a scalar-density but a quantity which, to be
transformed to another co-ordinate system, would have to be mul-
tiplied by the square of the functional determinant. Multiplying
a tensor by a tensor-density, however, always leads to a tensor-
density (whose order is equal to the sum of the orders of both
factors). Thus, for example, if a contra-variant vector with com-
ponents f i and a co-variant tensor-density with components wik
be multiplied together, we get a mixed tensor-density of the third
order with components f i wkl produced in a manner independent
of the co-ordinate system.
The analysis of tensor-densities can be established only
for linear fields in the case of an arbitrary manifold. It leads
to the following processes resembling the operation of
divergence:—

∂wi
= w, (30)
∂xi
 THE METRICAL CONTINUUM 163

∂wik
= wi , (31)
∂xk
......... .

As a result of (30) a linear tensor-density field wi of the first order

gives rise to a scalar-density field w, whereas (31) produces from
a linear field of the second order (wki = −wik ) a linear field of
the first order, and so forth. These operations are independent of
the co-ordinate system. The divergence (30) of a field wi of the
first order which has been produced from one, wik , of the second
order by means of (31) is = 0; an analogous result holds for the
higher orders. To prove that (30) is invariant, we use the following
known result of the theory of the motion of continuously extended
masses.
If ξ i is a given vector field, then

x̄i = xi + ξ i · δt (32)

expresses an infinitesimal displacement of the points of the

continuum, by which the point with the co-ordinates xi is trans-
ferred to the point with the co-ordinates x̄i . Let the constant in-
finitesimal factor δt be defined as the element of time during which
the deformation
 i takes place. The determinant of transformation
 ∂x  i
A =   differs from unity by δt ∂ξ . The displacement causes
∂xk  ∂xi
portion G of the continuum, to which, if xi ’s are used to denote
its co-ordinates, the mathematical region X in the variables xi
corresponds, to pass into the region G, from which G differs by
an infinitesimal amount. If s is a scalar-density field, which we
regard as the density of a substance occupying the medium, then
 CHAPTER II 164

the quantity of substance present in G

Z
= s(x) dx
X

whereas that which occupies G

Z Z
= s(x̄) dx̄ = s(x̄)A dx,
X

whereby the values (32) are to be inserted in the last expression

for the arguments x̄i of s. (I am here displacing the volume with
respect to the substance; instead of this, we can of course make
the substance flow through the volume; sξ i then represents the
intensity of the current.) The increase in the amount of substance
that the region G gains by the displacement is given by the inte-
gral s(x̄)A − s(x) taken with respect to the variables xi over X.
We, however, get for the integrand

∂ξ i ∂(sξ i )
 
 ∂s i
s(x̄)(A − 1) + s(x̄) − s(x) = δt s + ξ = δt · .
∂xi ∂xi ∂xi

Consequently the formula

∂(sξ i )
=w
∂xi
establishes an invariant connection between the two scalar-density
fields s and w and the contra-variant vector field with the com-
ponents ξ i . Now, since every vector-density wi is representable
in the form sξ i —for if in a definite co-ordinate system a scalar-
density s and a vector field ξ be defined by s = 1, ξ i = wi , then
 THE METRICAL CONTINUUM 165

the equation wi = sξ i holds for every co-ordinate system—the

required proof is complete.
In connection with this discussion we shall enunciate the Prin-
ciple of Partial Integration which will be of frequent use below.
If the functions wi vanish at the boundary of a region G, then
the integral
∂wi
Z
dx = 0.
G ∂xi
For this integral,
Z multiplied by δt, signifies the change that the
“volume” dx of this region suffers through an infinitesimal de-
formation whose components = δt · wi .
The invariance of the process of divergence (30) enables us
easily to advance to further stages, the next being (31). We enlist
the help of a co-variant vector field fi , which has been derived
from a potential f ; i.e.
∂f
fi = .
∂xi
We then form the linear tensor-density wik fi of the first order and
also its divergence
∂(wik fi ) ∂wik
= fi .
∂xk ∂xk
The observation that the fi ’s may assume any arbitrarily assigned
values at a point P concludes the proof. In a similar way we
proceed to the third and higher orders.

§ 14. Affinely Related Manifolds

The Conception of Affine Relationship.—We shall call a
point P of a manifold affinely related to its neighbourhood if we
 CHAPTER II 166

are given the vector P 0 into which every vector at P is transformed

by a parallel displacement from P to P 0 ; P 0 is here an arbitrary
point infinitely near P (vide note 10). No more and no less is
required of this conception than that it is endowed with all the
properties that were ascribed to it in the affine geometry of Chap-
ter I. That is, we postulate: There is a co-ordinate system (for the
immediate neighbourhood of P ) such that, in it, the components
of any vector at P are not altered by an infinitesimal parallel dis-
placement. This postulate characterises parallel displacements as
being such that they may rightly be regarded as leaving vectors
unchanged. Such co-ordinate systems are called geodetic at P .
What is the effect of this in an arbitrary co-ordinate system xi ?
Let us suppose that, in it, the point P has the co-ordinate x0i ,
P 0 the co-ordinates x0i + dxi ; let ξ i be the components of an arbi-
trary vector at P , ξ i + dξ i the components of the vector resulting
from it by parallel displacement towards P 0 . Firstly, since the par-
allel displacement from P to P 0 causes all the vectors at P to be
mapped out linearly or affinely by all the vectors at P 0 , dξ i must
be linearly dependent on ξ i , i.e.

dξ i = −dγri ξ r . (33)

Secondly, as a consequence of the postulate with which we started,

the dγri ’s must be linear forms of the differentials dxi , i.e.

dγri = Γirs dxs (330 )

in which the number co-efficients Γ, the “components of the affine

relationship,” satisfy the condition of symmetry

Γisr = Γirs . (3300 )

 THE METRICAL CONTINUUM 167

To prove this, let x̄i be a geodetic co-ordinate system at P ; the

formulæ of transformation (17) and (18) then hold. It follows
from the geodetic character of the co-ordinate system x̄i that, for
a parallel displacement,

dξ i = d(αri ξ¯r ) = dαri ξ¯r .

If we regard the ξ i ’s as components δxi of a line element at P we

must have
∂ 2 fi
−dγri δxr = δx̄r dx̄s
∂ x̄r ∂ x̄s
(in the case of the second derivatives we must of course insert their
values at P ). The statement contained in our enunciation follows
directly from this. Moreover, the symmetrical bilinear form

∂ 2 fi
−Γirs δx̄r dx̄s is derived from δx̄r dx̄s (34)
∂ x̄r ∂ x̄s
by transformation according to (18). This exhausts all the aspects
of the question. Now, if Γirs are arbitrarily given numbers that
satisfy the condition of symmetry (3300 ), and if we define the affine
relationship by (33) and (330 ), the transformation formulæ lead to

xi − x0i = x̄i − 21 Γirs x̄r x̄s ,

that is, to a geodetic co-ordinate system x̄i at P , since the equa-

tions (34) are fulfilled for them at P . In fact this transformation
at P gives us

∂ 2 fi
x̄i = 0, dx̄i = dxi (αki = δki ), = −Γirs .
∂ x̄r ∂ x̄s
 CHAPTER II 168

The formulæ according to which the components Γirs of the

affine relationship are transformed in passing from one co-ordinate
system to another may easily be obtained from the above discus-
sion; we do not, however, require them for subsequent work. The
Γ’s are certainly not components of a tensor (contra-variant in i,
co-variant in r and s) at the point P ; they have this character
with regard to linear transformations, but lose it when subjected
to arbitrary transformations. For they all vanish in a geodetic
co-ordinate system. Yet every virtual change of the affine rela-
tionship [Γirs ], whether it be finite or “infinitesimal,” is a tensor.
For
[dξ i ] = [Γirs ]ξ r dxs
is the difference of the two vectors that arise as a result of the two
parallel displacements of the vector ξ from P to P 0 .
The meaning of the parallel displacement of a co-variant
vector ξi at the point P to the infinitely near point P 0 is de-
fined uniquely by the postulate that the invariant product ξi η i of
the vector ξi and any arbitrary contra-variant vector η i remain
unchanged after the simultaneous parallel displacements, i.e.
d(ξi η i ) = (dξi · η i ) + (ξr dη r ) = (dξi − dγir ξr )η i = 0,
whence X
dξi = dγir ξr . (35)
r
We shall call a contra-variant vector field ξ i stationary at the
point P , if the vectors at the points P 0 infinitely near P arise
from the vector at P by parallel displacement, that is, if the total
differential equations
∂ξ i
dξ i + dγri ξ r = 0 (or + Γirs ξ r = 0)
∂xs
 THE METRICAL CONTINUUM 169

are satisfied at P . A vector field can obviously always be found

such that it has arbitrary given components at a point P (this
remark will be used in a construction which is to be carried out in
the sequel). The same conception may be set up for a co-variant
vector field.
From now onwards we shall occupy ourselves with affine man-
ifolds; they are such that every point of them is affinely
related to its neighbourhood. For a definite co-ordinate sys-
tem the components Γirs of the affine relationship are continuous
functions of the co-ordinates xi . By selecting the appropriate co-
ordinate system the Γirs ’s may, of course, be made to vanish at a
single point P , but it is, in general, not possible to achieve this si-
multaneously for all points of the manifold. There is no difference
in the nature of any of the affine relationships holding between
the various points of the manifold and their immediate neigh-
bourhood. The manifold is homogeneous in this sense. There are
not various types of manifolds capable of being distinguished by
the nature of the affine relationships governing each kind. The
postulate with which we set out admits of only one definite kind
of affine relationship.
Geodetic Lines.—If a point which is in motion carries a vec-
tor (which is arbitrarily variable) with it, we get for every value
of the time parameter s not only a point

P = (s) : x = xi (s)

of the manifold, but also a vector at this point with components

v i = v i (s) dependent on s. The vector remains stationary at the
moment s if
dv i dxβ
+ Γiαβ v α = 0. (36)
ds ds
 CHAPTER II 170

(This will relieve the minds of those who disapprove of operations

with differentials; they have here been converted into differential
co-efficients.) In the case of a vector being carried along according
to any arbitrary rule, the left-hand side V i of (36) consists of
the components of a vector in (s) connected invariantly with the
motion and indicating how much the vector v i changes per unit
of time at this point. For in passing from the point P = (s) to
P 0 = (s + ds), the vector v i at P becomes the vector
dv i
vi + ds
ds
at P 0 . If, however, we displace v i from P to P 0 leaving it un-
changed, we there get
v i + δv i = v i − Γiαβ v α dxβ .
Accordingly, the difference between these two vectors at P 0 , the
change in v during the time ds has components
dv i
ds − δv i = V i ds.
ds
In analytical language the invariant character of the vector V may
be recognised most readily as follows. Let us take an arbitrary
auxiliary co-variant vector ξi = (s) at P , and let us form the
change in the invariant ξi v i in its passage from (s) to (s + ds),
whereby the vector ξi is taken along unchanged. We get
d(ξi v i )
= ξi V i .
ds
If V vanishes for every value of s, the vector v glides with the
point P along the trajectory during the motion without becoming
changed.
 THE METRICAL CONTINUUM 171

dxi
Every motion is accompanied by the vector ui = of its
ds
velocity; for this particular case, V is the vector
dui d2 x dxα dxβ
Ui = + Γiαβ uα uβ = 2 + Γiαβ :
ds ds ds ds
namely, the acceleration, which is a measure of the change of
velocity per unit of time. A motion, in the course of which the
velocity remains unchanged throughout, is called a translation.
The trajectory of a translation, being a curve which preserves its
direction unchanged, is a straight or geodetic line. According
to the translational view (cf. Chapter I, § 1) this is the inherent
property of the straight line.
The analysis of tensors and tensor-densities may be de-
veloped for an affine manifold just as simply and completely as
for the linear geometry of Chapter I. For example, if fik are the
components (co-variant in i, contra-variant in k) of a tensor field
of the second order, we take two auxiliary arbitrary vectors at the
point P , of which the one, ξ, is contra-variant and the other, η, is
co-variant, and form the invariant
fik ξ i ηk
and its change for an infinitesimal displacement d of the current
point P , by which ξ and η are displaced parallel to themselves.
Now
∂fik i
d(fik ξ i ηk ) = ξ ηk dxl − frk ηk dγir ξ i + fir ξ i dγrk ηk ,
∂xl
hence
∂fik
filk = − Γril frk + Γkrl fir
∂xl
 CHAPTER II 172

are the components of a tensor field of the third order, co-variant

in i, l and contra-variant in k: this tensor field is derived from
the given one of the second order by a process independent of the
co-ordinate system. The additional terms, which the components
of the affine relationship contain, are characteristic quantities in
which, following Einstein, we shall later recognise the influence of
the gravitational field. The method outlined enables us to differ-
entiate a tensor in every conceivable case.
Just as the operation “grad” plays the fundamental part in
tensor analysis and all other operations are derivable from it, so
the operation “div” defined by (30) is the basis of the analysis of
tensor-densities. The latter leads to processes of a similar charac-
ter for tensor-densities of any order. For instance, if we wish to
find an expression for the divergence of a mixed tensor-density wik
of the second order, we make use of an auxiliary stationary vector
field ξ i wik at P and find the divergence of the tensor-density ξ i wik :
∂(ξ i wik ) ∂ξ r k k
∂wik
 
i ∂wi i r k
= w +ξ = ξ −Γik wr + .
∂xk ∂xk r ∂xk ∂xk
This quantity is a scalar-density, and since the components of a
vector field which is stationary at P may assume any values at
this point (P ), namely,
∂wik
− Γris wrs , (37)
∂xk
it is a co-variant tensor-density of the first order which has been
derived from wik in a manner independent of every co-ordinate
system.
Moreover, not only can we reduce a tensor-density to one of
the next lower order by carrying out the process of divergence,
 THE METRICAL CONTINUUM 173

but we can also transpose a tensor-density into one of the next

higher order by differentiation. Let s denote a scalar-density,
and let us again use a stationary vector field ξ i at P : we then
form the divergence of current-density, sξ i :

∂(sξ i ) ∂ξ i
 
∂s i ∂s
= ξ +s = − Γir s ξ i .
r
∂xi ∂xi ∂xi ∂xi

We thus get
∂s
− Γrir s
∂xi
as the components of a co-variant vector-density. To extend dif-
ferentiation beyond scalar tensor-densities to any tensor-densities
whatsoever, for example, to the mixed tensor-density wik of
the second order, we again proceed, as has been done repeat-
edly above, to make use of two stationary vector fields at P ,
namely, ξ i and η i , the latter being co-variant and the former
contra-variant. We differentiate the scalar-density wik ξ i ηk . If
the tensor-density that has been derived by differentiation be
contracted with respect to the symbol of differentiation and one
of the contra-variant indices, the divergence is again obtained.

§ 15. Curvature
If P and P ∗ are two points connected by a curve, and if a vector
is given at P , then this vector may be moved parallel to itself along
the curve from P to P ∗ . Equations (36), giving the unknown com-
ponents v i of the vector which is being subjected to a continuous
parallel displacement, have, for given initial values of v i , one and
only one solution. The vector transference that comes about in
this way is in general non-integrable, that is, the vector which
 CHAPTER II 174

we get at P ∗ is dependent on the path of the displacement along

which the transference is effected. Only in the particular case,
in which integrability occurs, is it allowable to speak of the same
vector at two different points P and P ∗ ; this comprises those vec-
tors that are generated from one another by parallel displacement.
Let such a manifold be called Euclidean-affine. If we subject all
points of such a manifold to an infinitesimal displacement, which
is in each case representable by an “equal” infinitesimal vector,
then the space is said to have undergone an infinitesimal total
translation. With the help of this conception, and following the
line of reasoning of Chapter I. (without entering on a rigorous
proof), we may construct “linear” co-ordinate systems which are
characterised by the fact that, in them, the same vectors have the
same components at different points of the systems. In a linear
co-ordinate system the components of the affine relationship van-
ish identically. Any two such systems are connected by linear
formulæ of transformation. The manifold is then an affine space
in the sense of Chapter I.: The integrability of the vector transfer-
ence is the infinitesimal geometrical property which distinguishes
“linear” spaces among affinely related spaces.
We must now turn our attention to the general case; it must
not be expected in this that a vector that has been taken round a
closed curve by parallel displacement finally returns to its initial
position. Just as in the proof of Stokes’s Theorem, so here we
stretch a surface over the closed curve and divide it into infinitely
small parallelograms by parametric lines. The change in any ar-
bitrary vector after it has traversed the periphery of the surface
is reduced to the change effected after it has traversed each of the
infinitesimal parallelograms marked out by two line elements dxi
and δxi at a point P . This change has now to be determined.
 THE METRICAL CONTINUUM 175

We shall adopt the convention that the amount ∆x = (∆ξ i ), by

which a vector x = ξ i increases, is derived from x by a linear
transformation, a matrix ∆F, i.e.

∆x = ∆F(x); ∆ξ α = ∆Fβα · ξ β . (38)

If ∆F = 0, then the manifold is “plane” at the point P in the

surface direction assumed by the surface element; if this is true for
all elements of a finitely extended portion of surface, then every
vector that is subjected to parallel displacement along the edge
of the surface returns finally to its initial position. ∆F is linearly
dependent on the element of surface:

∆F = Fik dxi δxk = 12 Fik ∆xik (∆xik = dxi δxk − dxk δxi ,

and
Fki = −Fik ). (39)
The differential form that occurs here characterises the curva-
ture, that is, the deviation of the manifold from plane-ness at
the point P for all possible directions of the surface; since its
co-efficients are not numbers, but matrices, we might well speak
of a “linear matrix-tensor of the second order,” and this would
undoubtedly best characterise the quantitative nature of curva-
ture. If, however, we revert from the matrices back to their
α
components—supposing Fβik to be the components of Fik or else
the co-efficients of the form

∆Fβα = Fβik
α
dxi δxk (40)

—then we arrive at the formula

α
∆x Fβik eα ξ β dxi δxk . (41)
 CHAPTER II 176

α
From this we see that the Fβik ’s are the components of a tensor
of the fourth order which is contra-variant in α and co-variant in
β, i and k. Expressed in terms of the components Γirs of the affine
relationship, it is
 α
∂Γβk ∂Γαβi

α
Fβik = − + (Γαri Γrβk − Γαrk Γrβi ). (42)
∂xi ∂xk
According to this they fulfil the conditions of “skew” and “cyclical”
symmetry, namely:—
α α α α α
Fβki = −Fβik ; Fβik + fikβ + Fkβi = 0. (43)
The vanishing of the curvature is the invariant differential law
which distinguishes Euclidean spaces among affine spaces in terms
of general infinitesimal geometry.
To prove the statements above enunciated we use the same
process of sweeping twice over an infinitesimal parallelogram as
we used on page 157 to derive the curl tensor; we use the same
notation as on that occasion. Let a vector x = x(P00 ) with com-
ponents ξ i be given at the point P00 . The vector x(P10 ) that is
derived from x(P00 ) by parallel displacement along the line ele-
ment dx is attached to the end point P10 of the same line element.
If the components of x(P10 ) are ξ i + dξ i then
dξ α = −dγβα ξ β = −Γαβi ξ β dxi .
Throughout the displacement δ to which the line element dx is
to be subjected (and which need by no means be a parallel dis-
placement) let the vector at the end point be bound always by the
specified condition to the vector at the initial point. The dξ α ’s are
then increased, owing to the displacement, by an amount
δdξ α = −δΓαβi dxi ξ β − Γαβi δdxi ξ β − dγrα δξ r .
 THE METRICAL CONTINUUM 177

If, in particular, the vector at the initial point of the line element
remains parallel to itself during the displacement, then δξ r must
be replaced in this formula by −δγβr ξ β . In the final position
−−−−→
P01 P11 of the line element we then get, at the point P01 , the vec-
tor x(P01 ), which is derived from x(P00 ) by parallel displacement
−−−−→
along P00 P01 ; at P11 we get the vector x(P11 ), into which x(P01 ) is
−−−−→
converted by parallel displacement along P01 P11 , and we have

δdξ α = ξ α (P11 ) − ξ α (P01 ) − ξ α (P10 ) − ξ α (P00 ) .

 

If the vector that is derived from x(P10 ) by parallel displacement

−−−−→
along P10 P11 is denoted by x∗ P11 , then, by interchanging d and δ,
we get an analogous expression for

dδξ α = ξ∗α (P11 ) − ξ α (P10 ) − ξ α (P01 ) − ξ α (P00 ) .

 

By subtraction we get

∆ξ α = δdξ α − dδξ α
( )
− δΓαβi dxi + dγrα δγβr − Γαβi δdxi
= ξβ .
+ dΓαβk δxk − dγrα dγβr + Γαβi dδxi

Since δdxi = dδxi the two last terms on the right destroy one
another, and we are left with

∆ξ α = ∆Fβα · ξ β

in which the ∆ξ α ’s are the components of a vector ∆x at P11 ,

which is the difference of the two vectors x and x∗ at the same
point, i.e.
−∆ξ α = ξ α (P11 ) − ξ∗α (P11 ).
 CHAPTER II 178

Since, when we proceed to the limit, P11 coincides with P = P00 ,

this proves the statements enunciated above.
The foregoing argument, based on infinitesimals, become rigor-
ous as soon as we interpret d and δ in terms of the differentiations
d d
and , as was done earlier. To trace the various stages of the
ds dt
vector x during the sequence of infinitesimal displacements, we
may well adopt the following plan. Let us ascribe to every pair of
values s, t, not only a point P = (s, t), but also a co-variant vector
at P with components fi (s, t). If ξ i is an arbitrary vector at P ,
d(fi ξ i )
then d(fi ξ i ) signifies the value that assumes if ξ i is taken
ds
along unchanged from the point (s, t) to the point (s + ds, t). And
d(fi ξ i ) is itself again an expression of the form fi ξ i excepting that
instead of fi there are now other functions fi0 of s and t. We may,
therefore, again subject it to the same process, or to the analogous
one δ. If we do the latter, and repeat the whole operation in the
reverse order, and then subtract, we get

δd(fi ξ i ) = δdfi ξ i + dfi δξ i + δfi dξ i + fi δdξ i ,

and then, since

d2 f i d2 fi
δdfi = = = dδfi ,
dt ds ds dt
we have
∆(fi ξ i ) = (δd − dδ)(fi ξ i ) = fi ∆ξ i .
In the last expression ∆ξ i is precisely the expression found above.
The invariant obtained is, for the point P = (0, 0),
α
Fβik fα ξ β ui v k .
 THE METRICAL CONTINUUM 179

It depends on an arbitrary co-variant vector with components fi

α
at this point, and on three contra-variant vectors ξ, u, v; the Fβik ’s
are accordingly the components of a tensor of the fourth order.

§ 16. Metrical Space

The Conception of Metrical Manifolds.—A manifold has
a measure-determination at the point P , if the line ele-
ments at P may be compared with respect to length; we herein
assume that the Pythagorean law (of Euclidean geometry) is valid
for infinitesimal regions. Every vector x then defines a distance
at P ; and there is a non-degenerate quadratic form x2 , such that
x and vy define the same distance if, and only if, x2 = y2 . This
postulate determines the quadratic form fully, if a factor of pro-
portionality differing from zero be prefixed. The fixing of the
latter serves to calibrate the manifold at the point P . We shall
then call x2 the measure of the vector x, or since it depends only
on the distance defined by x, we may call it the measure l of
this distance. Unequal distances have different measures; the
distances at a point P therefore constitute a one-dimensional to-
tality. If we replace this calibration by another, the new measure ¯l
is derived from the old one l by multiplying it by a constant fac-
tor λ 6= 0, independent of the distance; that is, ¯l = λl. The
relations between the measures of the distances are independent
of the calibration. So we see that just as the characterisation of
a vector at P by a system of numbers (its components) depends
on the choice of the co-ordinate system, so the fixing of a distance
by a number depends on the calibration; and just as the compo-
nents of a vector undergo a homogeneous linear transformation in
passing to another co-ordinate system, so also the measure of an
 CHAPTER II 180

arbitrary distance when the calibration is altered. We shall call

two vectors x and y (at P ), for which the symmetrical bilinear
form x · y corresponding to x2 vanishes, perpendicular to one
another; this reciprocal relation is not affected by the calibration
factor. The fact that the form x2 is definite is of no account in our
subsequent mathematical propositions, but, nevertheless, we wish
to keep this case uppermost in our minds in the sequel. If this
form has p positive and q negative dimensions (p + q = n), we say
that the manifold is (p + q)-dimensional at the point in question.
If p 6= q we fix the sign of the metrical fundamental form x2 once
and for all by the postulate that p > q; the calibration ratio λ is
then always positive. After choosing a definite co-ordinate system
and a certain calibration factor, suppose that, for every vector x
with components ξ i , we have
X
x2 = gik ξ i ξ k (gki = gik ). (44)
i,k

We now assume that our manifold has a measure-

determination at every point. Let us calibrate it everywhere,
and insert in the manifold a system of n co-ordinates xi —we must
do this so as to be able to express in numbers all quantities that
occur—then the gik ’s in (44) are perfectly definite functions of the
co-ordinates xi ; we assume that these functions are continuous
and differentiable. Since the determinant of the gik ’s vanishes at
no point, the integral numbers p and q will remain the same in
the whole domain of the manifold; we assume that p > q.
For a manifold to be a metrical space, it is not sufficient for it
to have a measure-determination at every point; in addition, every
point must be metrically related to the domain surrounding it.
The conception of metrical relationship is analogous to that of
 THE METRICAL CONTINUUM 181

affine relationship; just as the latter treats of vectors, so the

former deals with distances. A point is thus metrically related
to the domain in its immediate neighbourhood, if the distance is
known to which every distance at P gives rise when it passes by a
congruent displacement from P to any point P 0 infinitely near P .
The immediate vicinity of P may be calibrated in such a way
that the measure of any distance at P has undergone no change
after congruent displacements to infinitely near points. Such a
calibration is called geodetic at P . If, however, the manifold is
calibrated in any way, and if l is the measure of any arbitrary
distance at P , and l+dl the measure of the distance at P 0 resulting
from a congruent displacement to the infinitely near point P 0 ,
there is necessarily an equation
dl = −l dφ (45)
in which the infinitesimal factor dφ is independent of the displaced
distance, for the displacement effects a representation of the dis-
tances at P similar to that at P 0 . In (45), dφ corresponds to
the dγri ’s in the formula for vector displacements (33). If the cal-
ibration is altered at P and its neighbouring points according to
the formula ¯l = lλ (the calibration ratio λ is a positive function
of the position), we get in place of (45)
dλ
d¯l = −¯l dφ̄ in which dφ̄ = dφ − . (46)
λ
The necessary and sufficient condition that an appropriate value
of λ make dφ̄ vanish identically at P with respect to the infinites-
−−→
imal displacement P P 0 = (dxi ) is clearly that dφ must be a dif-
ferential form, that is,
dφ = φi dxi . (450 )
 CHAPTER II 182

The inferences that may be drawn from the postulate enunciated

at the outset are exhausted in (45) and (450 ). (In short, the φi ’s
are definite numbers at the point P . If P has co-ordinates P
xi = 0,
we need only assume log λ equal to the linear function φi xi
to get dφ = 0 there.) All points of the manifold are identical
as regards the measure-determinations governing each and as re-
gards their metrical relationship with their neighbouring points.
n
Yet, according as n is even or odd, there are respectively + 1
2
n+1
or different types of metrical manifolds which are distin-
2
guishable from one another by the inertial index of the metrical
groundform. One kind, with which we shall occupy ourselves par-
ticularly, is given by the case in which p = n, q = 0 (or p = 0,
q = n); other cases are p = n − 1, q = 1 (or p = 1, q = n − 1), or
p = n − 2, q = 2 (or p = 2, q = n − 2), and so forth.
We may summarise our results thus. The metrical character
of a manifold is characterised relatively to a system of reference
(= co-ordinate system + calibration) by twoPfundamental forms,
P differential form Q = i,k gik dxi dxk and a
namely, a quadratic
linear one dφ = i φi dxi . They remain invariant during transfor-
mations to new co-ordinate systems. If the calibration is changed,
the first form receives a factor λ, which is a positive function of
position with continuous derivatives, whereas the second function
becomes diminished by the differential of log λ. Accordingly all
quantities or relations that represent metrical conditions analyti-
cally must contain the functions gik φi in such a way that invari-
ance holds (1) for any transformation of co-ordinate (co-ordinate
invariance), (2) for the substitution which replaces gik and φi re-
 THE METRICAL CONTINUUM 183

spectively by
1 ∂λ
λ · gik and φi − ·
λ ∂xi
no matter, in (2), what function of the co-ordinates λ may be.
(This may be termed calibration invariance.)
In the same way as in § 15, in which we determined the change
in a vector which, remaining parallel to itself, traverses the pe-
riphery of an infinitesimal parallelogram bounded by dxi , δxi , so
here we calculate the change ∆l in the measure l of a distance
subjected to an analogous process. Making use of dl = −l dφ we
get
δdl = −δl dφ − l δdφ = l δφ dφ − l δdφ,
i.e.
∆l = δdl − dδl = −l ∆φ
where
∂φi ∂φk
∆φ = (δd − dδ)φ = fik dxi δxk and fik = − . (47)
∂xk ∂xi
Hence we may call the linear tensor of the second order with com-
ponents fik the distance curvature of metrical space as an analogy
to the vector curvature of affine space, which was derived in § 15.
Equation (46) confirms analytically that the distance curvature is
independent of the calibration; it satisfies the equations of invari-
ance
∂fkl ∂fli ∂fik
+ + = 0.
∂xi ∂xk ∂xl
Its vanishing is the necessary and sufficient condition that every
distance may be transferred from its initial position, in a manner
independent of the path, to all points of the space. This is the
 CHAPTER II 184

only case that Riemann considered. If metrical space is a Rie-

mann space, there is meaning in speaking of the same distance
at different points of space; the manifold may then be calibrated
(normal calibration) so that dφ vanishes identically. (Indeed, it
follows from fik = 0, that dφ is a total differential, namely, the
differential of a function log λ; by re-calibrating in the calibration
ratio λ, dφ may then be made equal to zero everywhere.) In nor-
mal calibration the metrical groundform Q of Riemann’s space is
determined except for an arbitrary constant factor, which may
be fixed by choosing once and for all a unit distance (no matter at
which point; the normal meter may be transported to any place).
The Affine Relationship of a Metrical Space.—We now
arrive at a fact, which may almost be called the key-note of in-
finitesimal geometry, inasmuch as it leads the logic of geometry
to a wonderfully harmonious conclusion. In a metrical space the
conception of infinitesimal parallel displacements may be given in
only one way if, in addition to our previous postulate, it is also to
satisfy the almost self-evident one: parallel displacement of a vec-
tor must leave unchanged the distance which it determines. Thus,
the principle of transference of distances or lengths which is the
basis of metrical geometry, carries with it a principle of trans-
ference of direction; in other words, an affine relationship is
inherent in metrical space.
Proof.—We take a definite system of reference. In the case
of all quantities ai which carry an upper index i (not necessarily
excluding others) we shall define the lowering of the index by
equations X
ai = gij aj
j

and the reverse process of raising the index by the corresponding

 THE METRICAL CONTINUUM 185

inverse equations. If the vector ξ i at the point P = (xi ) is to be

transformed into the vector ξ i + dξ i at P 0 = (xi + dxi ) by the
parallel displacement to P 0 which we are about to explain, then

dξ i = −dγki ξ k , dγki = Γikr dxr ,

and the equation

dl = −l dφ
must hold for the measure

l = gik ξ i ξ k

according to the postulate enunciated, and this gives

2ξi dξ i + ξ i ξ k dgik = −(gik ξ i ξ k ) dφ.

The first term on the left

= −2ξi ξ k dγki = −2ξ i ξ k dγik = −ξ i ξ k (dγik + dγki ).

Hence we get
dγik + dγki = dgik + gik dφ,
or
∂gik
Γi,kr + Γk,ir = + gik φr . (48)
∂xr
By interchanging the indices i, k, r cyclically, then adding the last
two and subtracting the first from the resultant sum, we get, bear-
ing in mind that the Γ’s must be symmetrical in their last two
indices,
 
1 ∂gir ∂gkr ∂gik
Γr,ik = 2 + − + 21 (gir φk + gkr φi − gik φr ). (49)
∂xk ∂xi ∂xr
 CHAPTER II 186

From this the Γrik are determined according to the equation

Γr,ik = grs Γsik or, explicitly, Γrik = g rs Γs,ik . (50)

These components of the affine relationship fulfil all the postulates

that have been enunciated. It is in the nature of metrical space to
be furnished with this affine relationship; in virtue of it the whole
analysis of tensors and tensor-densities with all the conceptions
worked out above, such as geodetic line, curvature, etc., may be
applied to metrical space. If the curvature vanishes identically,
the space is metrical and Euclidean in the sense of Chapter I.
In the case of vector curvature we have still to derive an
important decomposition into components, by means of which we
prove that distance curvature is an inherent constituent of the
former. This is quite to be expected since vector transference is
automatically accompanied by distance transference. If we use the
symbol ∆ = δd − dδ relating to parallel displacement as before,
then the measure l of a vector ξ i satisfies

∆l = −l ∆φ, ∆ξi ξ i = −(ξi ξ i ) ∆φ. (47)

Just as we found for the case in which fi are any functions of

position that
∆(fi ξ i ) = fi ∆ξ i
so we see that

∆(ξi ξ i ) = ∆(gik ξ i ξ k ) = gik ∆ξ i · ξ k + gik ξ i · ∆ξ k = 2ξi ∆ξ i ,

and equation (47) then leads to the following result. If for the
vector x = (ξ i ) we set

∆x = ∗∆x − x · 12 ∆φ,
 THE METRICAL CONTINUUM 187

then ∆x appears split up into a component at right angles to x

and another parallel to x, namely, ∗∆x and −x· 21 ∆φ respectively.
This is accompanied by an analogous resolution of the curvature
tensor, i.e.
α
Fβik α
= ∗Fβik − 12 δβα fik . (51)
The first component ∗F will be called “direction curvature”; it
is defined by
α
∗∆x = ∗Fβik eα ξ β dxi δxk .
The perpendicularity of ∗∆x to x is expressed by the formula
α
∗Fβik ξα ξ β dxi δxk = ∗Fαβik ξ α ξ β dxi δxk = 0.

The system of numbers ∗Fαβik is skew-symmetrical not only with

respect to i and k but also with respect to the index pair α and β.
In consequence we have also, in particular,
α
∗Fαik = 0.

Corollaries.—If the co-ordinate system and calibration

around a point P is chosen so that they are geodetic at P , then
we have, at P , φi = 0, Γrik = 0, or, according to (48) and (49), the
equivalent
∂gik
φi = 0, = 0.
∂xr
The linear form dφ vanishes at P and the co-efficients of the
quadratic groundform become stationary; in other words, those
conditions come about at P , which are obtained in Euclidean
space simultaneously for all points by a single system of reference.
This results in the following explicit definition of the parallel dis-
placement of a vector in metrical space. A geodetic system of
 CHAPTER II 188

reference at P may be recognised by the property that the φi ’s

at P vanish relatively to it and the gik ’s assume stationary values.
A vector is displaced from P parallel to itself to the infinitely near
point P 0 by leaving its components in a system of reference be-
longing to P unaltered. (There are always geodetic systems of
reference; the choice of them does not affect the conception of
parallel displacements.)
dxi
Since, in a translation xi = xi (s), the velocity vector ui =
ds
moves so that it remains parallel to itself, it satisfies

d(ui ui )
+ (ui ui )(φi ui ) = 0 in metrical geometry. (52)
ds
If at a certain moment the ui ’s have such values that ui ui = 0 (a
case that may occur if the quadratic groundform Q is indefinite),
then this equation persists throughout the whole translation: we
shall call the trajectory of such a translation a geodetic null-
line. An easy calculation shows that the geodetic null-lines do
not alter if the metric relationship of the manifold is changed in
any way, as long as the measure-determination is kept fixed at
every point.
Tensor Calculus.—It is an essential characteristic of a tensor
that its components depend only on the co-ordinate system and
not on the calibration. In a generalised sense we shall, however,
also call a linear form which depends on the co-ordinate system
and the calibration a tensor, if it is transformed in the usual way
when the co-ordinate system is changed, but becomes multiplied
by the factor λe (where λ = the calibration ratio) when the cali-
bration is changed; we say that it is of weight e. Thus the gik ’s are
components of a symmetrical co-variant tensor of the second order
 THE METRICAL CONTINUUM 189

and of weight 1. Whenever tensors are mentioned without their

weight being specified, we shall take this to mean that those of
weight 0 are being considered. The relations which were discussed
in tensor analysis are relations, which are independent of calibra-
tion and co-ordinate system, between tensors and tensor-densities
in this special sense. We regard the extended conception of a
tensor, and also the analogous one of tensor-density of weight e,
merely as an auxiliary conception, which is introduced to sim-
plify calculations. They are convenient for two reasons: (1) They
make it possible to “juggle with indices” in this extended region.
By lowering a contra-variant index in the components of a tensor
of weight e we get the components of a tensor of weight e + 1,
the components being co-variant with respect to this index. The
process may also be carried out in the reverse direction. (2) Let
g denote the determinant of the gik ’s, furnished with a plus or
minus sign according as the number g of the negative dimensions
√
is even or uneven, and let g be the positive root of this posi-
√
tive number g. Then, by multiplying any tensor by g we
n
get a tensor-density whose weight is more than that
2
n
of the tensor; from a tensor of weight − we get, in partic-
2
ular, a tensor-density in the true sense. The proof is based on
√ n
the evident fact that g is itself a scalar-density of weight .
√ 2
We shall always indicate when a quantity is multiplied by g by
changing the ordinary letter which designates the quantity into
the corresponding one printed in Clarendon type. Since, in Rie-
mann’s geometry, the quadratic groundform Q is fully determined
by normal calibration (we need not consider the arbitrary con-
stant factor), the difference in the weights of tensors disappears
 CHAPTER II 190

here: since, in this case, every quantity that may be represented

by a tensor may also be represented by the tensor-density that is
√
derived from it by multiplying it by g, the difference between
tensors and tensor-densities (as well as between co-variant and
contra-variant) is effaced. This makes it clear why for a long time
tensor-densities did not come into their right as compared with
tensors. The main use of tensor calculus in geometry is an inter-
nal one, that is, to construct fields that are derived invariantly
from the metrical structures. We shall give two examples that are
of importance for later work. Let the metrical manifold be (3+1)-
dimensional, so that −g will be the determinant of the gik ’s. In
this space, as in every other, the distance curvature with compo-
nents fik is a true linear tensor field of the second order. From it is
derived the contra-variant tensor f ik of weight −2, which, on ac-
count of its weight differing from zero, is of no actual importance;
√
multiplication by g leads to f ik , a true linear tensor-density of
the second order.
l = 41 fik f ik (53)
is
Z the simplest scalar-density that can be formed; consequently
l dx is the simplest invariant integral associated with the met-
rical basis ofZ a (3 + 1)-dimensional manifold. On the other hand,
√
the integral g dx, which occurs in Riemann’s geometry as “vol-
ume,” is meaningless in general geometry. We can derive the inten-
sity of current (vector-density) from f ik by means of the operation
divergence thus:
∂f ik
= si .
∂xk
In physics, however, we use the tensor calculus not to describe the
 THE METRICAL CONTINUUM 191

metrical condition but to describe fields expressing physical states

in metrical space—as, for example, the electromagnetic field—and
to set up the laws that hold in them. Now, we shall find at the
close of our investigations that this distinction between physics
and geometry is false, and that physics does not extend beyond
geometry. The world is a (3 + 1)-dimensional metrical manifold,
and all physical phenomena that occur in it are only modes of
expression of the metrical field. In particular, the affine relation-
ship of the world is nothing more than the gravitational field, but
its metrical character is an expression of the state of the “æther”
that fills the world; even matter itself is reduced to this kind of
geometry and loses its character as a permanent substance. Clif-
ford’s prediction, in an article of the Fortnightly Review of 1875,
becomes confirmed here with remarkable accuracy; in this he says
that “the theory of space curvature hints at a possibility of de-
scribing matter and motion in terms of extension only”.
These are, however, as yet dreams of the future. For the
present, we shall maintain our view that physical states are foreign
states in space. Now that the principles of infinitesimal geometry
have been worked out to their conclusion, we shall set out, in the
next paragraph, a number of observations about the special case
of Riemann’s space and shall give a number of formulæ which will
be of use later.

§ 17. Observations about Riemann’s Geometry as a

Special Case
General tensor analysis is of great utility even for Euclidean
geometry whenever one is obliged to make calculations, not in a
Cartesian or affine co-ordinate system, but in a curvilinear co-
 CHAPTER II 192

ordinate system, as often happens in mathematical physics. To

illustrate this application of the tensor calculus we shall here write
out the fundamental equations of the electrostatic and the mag-
netic field due to stationary currents in terms of general curvilinear
co-ordinates.
Firstly, let Ei be the components of the electric intensity of
field in a Cartesian co-ordinate system. By transforming the
quadratic and the linear differential forms

ds2 = dx21 + dx22 + dx23 E1 dx1 + E2 dx2 + E3 dx3

respectively, into terms of arbitrary curvilinear co-ordinates (again

denoted by xi ), each form being independent of the Cartesian co-
ordinate system, suppose we get

ds2 = gik dxi dxk and Ei dxi .

Then the Ei ’s are in every co-ordinate system the components of

the same co-variant vector field. From them we form a vector-
density with components
√
Ei = g · g ik Ek (g = |gik |).

We transform the potential −φ as a scalar into terms of the new

Z the density ρ of electricity as being the
co-ordinates, but we define
electric charge given by ρ dx1 dx2 dx3 contained in any portion
of space; ρ is not then a scalar but a scalar density. The laws are
 THE METRICAL CONTINUUM 193

expressed by

∂φ ∂Ei ∂Ek
Ei = − = 0

∂xi ∂xk ∂xi




∂Ei (54)
=ρ
∂xi




Ski = Ei Ek − 12 δik S,



in which S, = Ei Ei , are the components of a mixed tensor-density

of the second order, namely, the potential difference. The proof is
sufficiently indicated by the remark that these equations, in the
form we have written them, are absolutely invariant in character,
but pass into the fundamental equations, which were set up earlier,
for a Cartesian co-ordinate system.
The magnetic field produced by stationary currents was char-
acterised in Cartesian co-ordinate systems by an invariant skew-
symmetrical bilinear form Hik dxi δxk . By transforming the latter
into terms of arbitrary curvilinear co-ordinates, we get Hik , the
components of a linear tensor of the second order, namely, of the
magnetic field, these components being co-variant with respect to
arbitrary transformations of the co-ordinates. Similarly, we may
deduce the components φi of the vector potential as components
of a co-variant vector field in any curvilinear co-ordinate system.
We now introduce a linear tensor-density of the second order by
means of the equations
√
Hik = g · g iα g kβ Hαβ .
 CHAPTER II 194

The laws are then expressed by

∂φi ∂φk ∂Hkl ∂Hli ∂Hik

Hik = − or + + = 0

∂xk ∂xi ∂xi ∂xk ∂xl 



respectively,


(55)
∂Hik
= si ,



∂xk 



Ski = Hir Hkr − 12 δik S, S = 21 Hik Hik .


The si ’s are the components of a vector-density, the electric in-

tensity of current; the potential differences Ski have the same in-
variant character as in the electric field. These formulæ may be
specialised for the case of, for example, spherical and cylindrical
co-ordinates. No further calculations are required to do this, if
we have an expression for ds2 , the distance between two adjacent
points, expressed in these co-ordinates; this expression is easily
obtained from considerations of infinitesimal geometry.
It is a matter of greater fundamental importance that (54) and
(55) furnish us with the underlying laws of stationary electromag-
netic fields if unforeseen reasons should compel us to give up the
use of Euclidean geometry for physical space and replace it by
Riemann’s geometry with a new groundform. For even in the
case of such generalised geometric conditions our equations, in
virtue of their invariant character, represent statements that are
independent of all co-ordinate systems, and that express formal
relationships between charge, current, and field. In no wise can
it be doubted that they are the direct transcription of the laws
of the stationary electric field that hold in Euclidean space; it is
indeed astonishing how simply and naturally this transcription is
effected by means of the tensor calculus. The question whether
 THE METRICAL CONTINUUM 195

space is Euclidean or not is quite irrelevant for the laws of the elec-
tromagnetic field. The property of being Euclidean is expressed
in a universally invariant form by differential equations of the sec-
ond order in the gik ’s (denoting the vanishing of the curvature)
but only the gik ’s and their first derivatives appear in these laws.
It must be emphasised that a transcription of such a simple kind
is possible only for laws dealing with action at infinitesimal
distances. To derive the laws of action at a distance correspond-
ing to Coulomb’s, and Biot and Savart’s Law from these laws of
contiguous action is a purely mathematical problem that amounts
in essence to the following. In place of the usual potential equa-
tion ∆φ = 0 we get as its invariant generalisation (vide (54)) in
Riemann’s geometry the equation
 
∂ √ ik ∂φ
g·g =0
∂xi ∂xk

that is, a linear differential equation of the second order whose

co-efficients are, however, no longer constants. From this we are
to get the “standard solution,” tending to infinity, at any arbitrary
1
given point; this solution corresponds to the “standard solution”
r
of the potential equation. It presents a difficult mathematical
problem that is treated in the theory of linear partial differential
equations of the second order. The same problem is presented
when we are limited to Euclidean space if, instead of investigat-
ing events in empty space, we have to consider those taking place
in a non-homogeneous medium (for example, in a medium whose
dielectric constant varies at different places with the time). Con-
ditions are not so favourable for transcribing electromagnetic laws,
if real space should become disclosed as a metrical space of a still
 CHAPTER II 196

more general character than Riemann assumed. In that case it

would be just as inadmissible to assume the possibility of a cali-
bration that is independent of position in the case of currents and
charges as in the case of distances. Nothing is gained by pursuing
this idea. The true solution of the problem lies, as was indicated in
the concluding words of the previous paragraph, in quite another
direction.
Let us rather add a few observations about Riemann’s space
as a special case. Let the unit measure (1 centimetre) be chosen
once and for all; it must, of course, be the same at all points. The
metrical structure of the Riemann space is then described by an
invariant quadratic differential form gik dxi dxk or, what amounts
to the same thing, by a co-variant symmetrical tensor field of
the second order. The quantities φi , that are now equal to zero,
must be struck out everywhere in the formulæ of general metrical
geometry. Thus, the components of the affine relationship, which
here bear the name“Christoffel
 three-indices symbols” and are
ik
usually denoted by , are determined from
r
       
ik ∂g ir ∂g kr ∂g ik ik ik
= 12 + − , =g rs
. (56)
r ∂xk ∂xi ∂xr r s
(We give way to the usual nomenclature—although it disagrees
flagrantly with our own convention regarding rules about the po-
sition of indices—so as to conform to the usage of the text-books.)
The following formulæ are now tabulated for future reference:—
√  
1 ∂ g ir
√ − = 0, (57)
g ∂xi r
√
1 ∂( g · g ik )
 
rs rs
√ + g = 0, (570 )
g ∂xk i
 THE METRICAL CONTINUUM 197

√
1 ∂( g · g ik )
     
lr rk lr ri lr ik
√ + g + g − g = 0. (5700 )
g ∂xl i k r
√ √
These equations hold because g is a scalar and g · g ik is a tensor-
density; hence, according to the rules given by the analysis of tensor-
√
densities, the left-hand members of these equations, multiplied by g,
are
 likewise tensor-densities. If, however, we use a co-ordinate system
∂g ik

= 0, which is geodetic at P , then all terms vanish. Hence,
∂xr
in virtue of the invariant nature of these equations, they also hold in
every other co-ordinate system. Moreover,
√
dg d g
= g ik dgik , √ = 12 g ik dgik . (58)
g g

For the total differential of a determinant with n2 (independent and

variable) elements gik is equal to Gik dgik , where Gik denotes the minor
of gik . If tik (= tki ) is any symmetrical system of numbers, then we
always have
tik dgik = −tik dg ik . (59)
From
gij g jk = δik
it follows that
gij dg jk = −g jk dgij .
If these equations are multiplied by tik (this symbol cannot be misin-
terpreted since tik = gkl til = gkl tli = tik ) the required result follows. In
particular, in place of (58) we may also write
dg
= −gik dg ik . (580 )
g
The co-variant components Rαβik of curvature in Riemann’s
space, which we denote by R instead of F , satisfy the conditions of
 CHAPTER II 198

symmetry

Rαβki = −Rαβik , Rβαki = −Rαβik ,

Rαβki + Rαikβ + Rαkβi = 0,

(for the “distance curvature” vanishes). It is easy to show that, from

them, it follows that (vide note 11)

Rikαβ = Rαβik .

As the result of an observation on page 83, it follows that all those

conditions taken together enable us to characterise the curvature ten-
sor completely by means of a quadratic form that is dependent on an
arbitrary element of surface, namely,
1
4 Rαβik ∆xαβ ∆xik (∆xik = dxi δxk − dxk δxi ).

If this quadratic form is divided by the square of the magnitude of the

surface element, the quotient depends only on the ratio of the ∆xik ’s,
i.e. on the position of the surface element; Riemann calls this number
the curvature of the space at the point P in the surface direction in
question. In two-dimensional Riemann space (on a surface) there is only
one surface direction and the tensor degenerates into a scalar (Gaussian
curvature). In Einstein’s theory of gravitation the contracted tensor of
the second order
α
Riαk = Rik
which is symmetrical in Riemann’s space, becomes of importance: its
components are
         
∂ ik ∂ ir ik rs ir ks
Rik = − + − . (60)
∂xr r ∂xk r r s s r
Only in the case of the second term on the right, the symmetry with re-
spect to i and k is not immediately evident; according to (57), however,
 THE METRICAL CONTINUUM 199

it is equal to
1∂ 2 (log g)
2 .
∂xi ∂xk
Finally, by applying contraction once more we may form the scalar of
curvature
R = g ik Rik .
In general metrical space the analogously formed scalar of curvature F
is expressed in the following way (as is easily shown) by the Riemann
expression R, which is dependent only on the gik ’s and which has no
distinct meaning in that space:—
√
1 ∂( gφi ) (n − 1)(n − 2)
F = R − (n − 1) √ − (φi φi ). (61)
g ∂xi 4
F is a scalar of weight −1. Hence, in a region in which F 6= 0 we may
define a unit of length by means of the equation F = constant. This
is a remarkable result inasmuch as it contradicts in a certain sense the
original view concerning the transference of lengths in general metrical
space, according to which a direct comparison of lengths at a distance
is not possible; it must be noticed, however, that the unit of length
which arises in this way is dependent on the conditions of curvature of
the manifold. (The existence of a unique uniform calibration of this
kind is no more extraordinary than the possibility of introducing into
Riemann’s space certain unique co-ordinate systems arising out of the
metrical structure.) The “volume” that is measured by using this unit
of length is represented by the invariant integral
Z p
g · F n dx. (62)

For two vectors ξ i , η i that undergo parallel displacement we

have, in metrical space,
d(ξi η i ) + (ξi η i ) dφ = 0.
 CHAPTER II 200

In Riemann’s space, the second term is absent. From this it follows

that in Riemann’s space the parallel displacement of a contra-
variant vector ξ is expressed in exactly the same way in terms
of the quantities ξi = gik ξ k as the parallel displacement of a co-
variant vector is expressed in terms of its components ξi :
   
iα iα
dξi − dxα ξβ = 0 or dξi − dxα ξ β = 0.
β β

Accordingly, for a translation we have

 
dui 1 ∂gαβ α β i dxi k
−2 u u =0 u = , ui = gik u (63)
ds ∂xi ds

for, by equation (48),

   
iα iβ ∂gαβ
+ =
β α ∂xi

and hence for any symmetrical system of numbers tαβ :—

   
1 ∂gαβ iα αβ iα α
· tαβ = t = t . (64)
2
∂xi β β β

Since the numerical value of the velocity vector remains unchanged

during translations, we get
dxi dxk
gik = ui ui = const. (65)
ds ds
If, for the sake of simplicity, we assume the metrical groundform
to be definitely positive, then every curve xi = xi (s) [a ≤ s ≤ b]
 THE METRICAL CONTINUUM 201

has a length, which is independent of the mode of parametric

representation. This length is
Z bp  
dxi dxk
Q ds Q = gik .
a ds ds
If we use the length of arc itself as the parameter, Q becomes
equal to 1. Equation (65) states that a body in translation tra-
verses its path, the geodetic line, with constant speed, namely,
that the time-parameter is proportional to s, the length of arc. In
Riemann’s space the geodetic line possesses not only the differ-
ential property of preserving its direction unaltered, but also the
integral property that every portion of it is the shortest
line connecting its initial and its final point. This statement
must not, however, be taken literally, but must be understood in
the same sense as the statement in mechanics that, in a position
of equilibrium, the potential energy is a minimum, or when it is
said of a function f (x, y) in two variables that it has a minimum
at points where its differential
∂f ∂f
df = dx + dy
∂x ∂y
vanishes identically in dx and dy; whereas the true expression is
that it assumes a “stationary” value at that point, which may be a
minimum, a maximum, or a “point of inflexion”. The geodetic line
is not necessarily a curve of least length but is a curve of stationary
length. On the surface of a sphere, for instance, the great circles
are geodetic lines. If we take any two points, A and B, on such a
great circle, the shorter of the two arcs AB is indeed the shortest
line connecting A and B, but the other arc AB is also a geodetic
line connecting A and B; it is not of least but of stationary length.
 CHAPTER II 202

We shall seize this opportunity of expressing in a rigorous form

the principle of infinitesimal variation.
Let any arbitrary curve be represented parametrically by
xi = xi (s), (a ≤ s ≤ b).
We shall call it the “initial” curve. To compare it with neighbour-
ing curves we consider an arbitrary family of curves involving one
parameter:
xi = xi (s; ), (a ≤ s ≤ b).
The parameter  varies within an interval about  = 0; xi (s; ) are
to denote functions that resolve into xi (s) when  = 0. Since all
curves of the family are to connect the same initial point with the
same final point, xi (a; ) and xi (b; ) are independent of . The
length of such a curve is given by
Z bp
L() = Q ds.
a

Further, we assume that s denotes the length of an arc of the initial

dxi
curve, so that Q = 1 for  = 0. Let the direction components
ds
of the initial curve  = 0 be denoted by ui . We also set
 
dxi
· = ξ i (s) = δxi .
d =0
These are the components of the “infinitesimal” displacement
which makes the initial curve change into the neighbouring curve
due to the “variation” corresponding to an infinitely small value
of ; they vanish at the ends.
 
dL
 = δL
d =0
 THE METRICAL CONTINUUM 203

is the corresponding variation in the length. δL = 0 is the con-

dition that the initial curve has a stationary length as compared
with the other members of the family. If we use the symbol δQ in
the same sense, we get
Z b Z b
δQ 1
δL = √ ds = 2 δQ ds (66)
a 2 Q a

since Q = 1 in the case of the initial curve. Now

dQ ∂gαβ dxi dxα dxβ dxk d2 xi
= + 2gik
d ∂xi d ds ds ds d ds
and hence (if we interchange “variation” and “differentiation,” that
is the differentiations with respect to  and s) we get

∂gαβ α β i dξ i
δQ = u u ξ + 2gik uk .
∂xi ds
If we substitute this in (66) and rewrite the second term by apply-
ing partial integration, and note that the ξ i ’s vanish at the ends
of the interval of integration, then
Z b 
1 ∂gαβ α β dui
δL = 2
u u − ξ i ds.
a ∂x i ds
Hence the condition δL = 0 is fulfilled for any family of curves
if, and only if, (63) holds. Indeed, if, for a value s = s0 between
a and b, one of these expressions, for example the first, namely,
i = 1, differed from zero (were greater than zero), say, it would be
possible to mark off a little interval around s0 so small that, within
it, the above expression would be always > 0. If we choose a non-
negative function for ξ 1 such that it vanishes for points beyond
 CHAPTER II 204

this interval, all remaining ξ i ’s, however, being = 0, we find the

equation δL = 0 contradicted.
Moreover, it is evident from this proof that, of all the motions
that lead from the same initial point to the same final point within
R b a ≤ s ≤ b, a translation is distinguished
the same interval of time
by the property that a Q ds has a stationary value.
Although the author has aimed at lucidity of expression many
a reader will have viewed with abhorrence the flood of formulæ
and indices that encumber the fundamental ideas of infinitesimal
geometry. It is certainly regrettable that we have to enter into the
purely formal aspect in such detail and to give it so much space
but, nevertheless, it cannot be avoided. Just as anyone who wishes
to give expressions to his thoughts with ease must spend laborious
hours learning language and writing, so here too the only way that
we can lessen the burden of formulæ is to master the technique
of tensor analysis to such a degree that we can turn to the real
problems that concern us without feeling any encumbrance, our
object being to get an insight into the nature of space, time, and
matter so far as they participate in the structure of the external
world. Whoever sets out in quest of this goal must possess a
perfect mathematical equipment from the outset. Before we pass
on after these wearisome preparations and enter into the sphere
of physical knowledge along the route illumined by the genius
of Einstein, we shall seek to obtain a clearer and deeper vision
of metrical space. Our goal is to grasp the inner necessity and
uniqueness of its metrical structure as expressed in Pythagoras’
Law.
 THE METRICAL CONTINUUM 205

§ 18. Metrical Space from the Point of View of the

Theory of Groups
Whereas the character of affine relationship presents no further
difficulties—the postulate on page 184 to which we subjected the
conception of parallel displacement, and which characterises it as
a kind of unaltered transference, defines its character uniquely—
we have not yet gained a view of metrical structure that takes us
beyond experience. It was long accepted as a fact that a metrical
character could be described by means of a quadratic differential
form, but this fact was not clearly understood. Riemann many
years ago pointed out that the metrical groundform might, with
equal right essentially, be a homogeneous function of the fourth
order in the differentials, or even a function built up in some other
way, and that it need not even depend rationally on the differen-
tials. But we dare not stop even at this point. The underlying
general feature that determines the metrical structure at a point P
is the group of rotations. The metrical constitution of the man-
ifold at the point P is known if, among the linear transformations
of the vector body (i.e. the totality of vectors), those are known
that are congruent transformations of themselves. There are just
as many different kinds of measure-determinations as there are
essentially different groups of linear transformations (whereby es-
sentially different groups are such as are distinguished not merely
by the choice of co-ordinate system). In the case of Pythagorean
metrical space, which we have alone investigated hitherto, the
group of rotations consists of all linear transformations that con-
vert the quadratic groundform into itself. But the group of rota-
tions need not have an invariant at all in itself (that is, a function
which is dependent on a single arbitrary vector and which remains
 CHAPTER II 206

unaltered after any rotations).

Let us reflect upon the natural requirements that may be im-
posed on the conception of rotation. At a single point, as long
as the manifold has not yet a measure-determination, only the
n-dimensional parallelepipeds can be compared with one another
in respect to size. If ai (i = 1, 2, . . . , n) are arbitrary vectors that
are defined in terms of the initial unit vectors ei according to the
equations
ai = αik ek
then the determinant of the αik ’s which, following Grassmann, we
may conveniently denote by
[a1 , a2 , . . . , an ]
[e1 , e2 , . . . , en ]
is, according to definition, the volume of the parallelopiped
mapped out by the n vectors ai . If we choose another system of
unit vectors ēi all the volumes become multiplied by a common
constant factor, as we see from the “multiplication theorem of
determinants,” namely
[a1 , a2 , . . . , an ] [a1 , a2 , . . . , an ] [ē1 , ē2 , . . . , ēn ]
= .
[e1 , e2 , . . . , en ] [ē1 , ē2 , . . . , ēn ] [e1 , e2 , . . . , en ]
The volumes are thus determined uniquely and independently of
the co-ordinate system once the unit measure has been chosen.
Since a rotation is “not to alter” the vector body, it must obvi-
ously be a transformation that leaves the infinitesimal elements of
volume unaffected. Let the rotation that transforms the vector
x = (ξ i ) into x̄ = (ξ¯i ) be represented by the equations

ēi = αik ek or ξ i = αki ξ¯k .

 THE METRICAL CONTINUUM 207

The determinant of the rotation matrix (αki ) then becomes equal

to 1. This being the postulate that applies to a single rotation,
we must demand of the rotations as a whole that they form a
group in the sense of the definition given on page 13. Moreover,
this group has to be a continuous one, that is the rotations are
to be elements of a one-dimensional continuous manifold.
If a linear vector transformation be given by its matrix A =
(αki ) in passing from one co-ordinate system (ei ) to another (ēi )
according to the equations

U : ēi = uki ek , (67)

then A becomes changed into U AU −1 (where U −1 denotes the

inverse of U ; U U −1 and U −1 U are equal to identity E). Hence
every group that is derived from a given matrix group G by ap-
plying the operation U GU −1 on every matrix G of G (U being the
same for all G’s) may be transformed into the given matrix group
by an appropriate change of co-ordinate system. Such a group
U GU −1 will be said to be of the same kind as G (or to differ
from G only in orientation). If G is the group of rotation matri-
ces at P and if U GU −1 is identical with G (this does not mean
that G must again pass into G as a result of the operation U GU −1 ,
but all that is required is that G and U GU −1 belong to G simul-
taneously) then the expressions for the metrical structures of two
co-ordinate systems (67), that are transformed into one another
by U , are similar; U is a representation of the vector body on
itself, such that it leaves all the metrical relations unaltered. This
is the conception of similar representation. G is included in
the group G∗ of similar representations as a sub-group.
From the metrical structure at a single point we now pass on
to “metrical relationship”. The metrical relationship between
 CHAPTER II 208

the point P0 and its immediate neighbourhood is given if a linear

representation at P0 = x0i of the vector body on itself at an in-
finitely near point P = (x0i + dxi ) is a congruent transference.
Together with A every representation (or transformation) AG0 , in
which A is followed by a rotation G0 at P0 , is likewise a congru-
ent transference; thus, from one congruent transference A of the
vector body from P0 to P , we get all possible ones by making G0
traverse the group of rotations belonging to P0 . If we consider the
vector body belonging to the centre P0 for two positions congruent
to one another, they will resolve into two congruent positions at P
if subjected to the same congruent transference A; for this reason,
the group of rotations G at P is equal to AG0 A−1 . The metrical
relationship thus tells us that the group of rotations at P differs
from that at P0 only in orientation. If we pass continuously from
the point P0 to any point of the manifold, we see that the groups
of rotation are of a similar kind at all points of the manifold; thus
there is homogeneity in this respect.
The only congruent transferences that we take into consid-
eration are those in which the vector components ξ i undergo
changes dξ i that are infinitesimal and of the same order as the
displacement of the centre P0 ,
dξ i = dλik · ξ k .
If L and M are two such transferences from P0 to P , with co-
efficients dλik and dµik respectively, then the rotation M L−1 is
likewise infinitesimal: it is represented by the formula
dξ i = dαki · ξ k where dαki = dµik − dλik . (68)
The following will also be true. If an infinitesimal congruent
transference consisting in the displacement (dxi ) of the centre P0
 THE METRICAL CONTINUUM 209

is succeeded by one in which the centre is displaced by (δxi ),

we get a congruent transference that is effected by the resultant
displacement dxi + δxi of the centre (plus an error which is in-
finitesimal compared with the magnitude of the displacements).
Hence, if for the transition from P0 = (x01 , x02 , . . . , x0n ) to the point
(x01 + , x02 , . . . , x0n ), this being an infinitesimal change  in the
direction of the first co-ordinate axis,
dξ i =  · Λik ξ k
is a congruent transference, and if Λik2 , . . . , Λikn have a correspond-
ing meaning for the displacements of P0 in the direction of the 2nd
up to the nth co-ordinate in turn; then the equation
dξ i = Λikr dxr · ξ k (69)
gives a congruent transference for an arbitrary displacement hav-
ing components dxi .
Among the various kinds of metrical spaces we shall now des-
ignate by simple intrinsic relations the category to which, accord-
ing to Pythagoras’ and Riemann’s ideas, real space belongs. The
group of rotations that does not vary with position exhibits a
property that belongs to space as a form of phenomena; it char-
acterises the metrical nature of space. The metrical relationship,∗
from point to point, however, is not determined by the nature of
space, nor by the mutual orientation of the groups of rotation at
the various points of the manifold. The metrical relationship is
dependent rather on the disposition of the material content, and
is thus in itself free and capable of any “virtual” changes. We shall
formulate the fact that it is subject to no limitation as our first
axiom.
∗
Although, as will be shown later, it is everywhere of the same kind.
 CHAPTER II 210

I. The Nature of Space Imposes no Restriction on the

Metrical Relationship
It is possible to find a metrical relationship in space between
the point P0 and the points in its neighbourhood such that the
formula (69) represents a system of congruent transferences to
these neighbouring points for arbitrarily given numbers Λikr .
Corresponding to every co-ordinate system xi at P0 there is a
possible conception of parallel displacement, namely, the displace-
ment of the vectors from P0 to the infinitely near points without
the components undergoing a change in this co-ordinate system.
Such a system of parallel displacements of the vector body from P0
to all the infinitely near points is expressed, as we know, in terms
of a definite co-ordinate system, selected once and for all by the
formula
dξ i = −dγ i · ξ k
in which the differential forms dγki = Γikr dxr satisfy the condition
of symmetry
Γikr = Γirk . (70)
And, indeed, a possible conception of parallel displacement corre-
sponds to every system of symmetrical co-efficients Γ. For a given
metrical relationship the further restriction that the “parallel dis-
placements” shall simultaneously be congruent transferences must
be imposed. The second postulate is the one enunciated above as
the fundamental theorem of infinitesimal geometry; for a given
metrical relationship there is always a single system of parallel
displacements among the transferences of the vector body. We
treated affine relationship in § 15 only provisionally as a rudimen-
tary characteristic of space; the truth is, however, that parallel
 THE METRICAL CONTINUUM 211

displacements, in virtue of their inherent properties, must be ex-

cluded from congruent transferences, and that the conception of
parallel displacement is determined by the metrical relationship.
This postulate may be enunciated thus:—

II. The Affine Relationship is Uniquely Determined by

the Metrical Relationship
Before we can formulate it analytically we must deal with in-
finitesimal rotations. A continuous group G of r members is a
continuous r-dimensional manifold of matrices. If s1 , s2 , . . . , sr are
co-ordinates in this manifold, then, corresponding to every value
system of the co-ordinates there is a matrix A(s1 , s2 , . . . , sr ) of the
group which depends on the value-system continuously. There is
a definite value-system—we may assume for it that s1 = 0—to
which identity, E, corresponds. The matrices of the group that
are infinitely near E differ from E by

A1 ds1 + A2 ds2 + . . . + Ar dsr ,

 
∂A
in which Ai = . We call a matrix A an infinitesimal op-
∂si 0
eration of the group if the group contains a transformation (in-
dependent of ) that coincides with E and A to within an error
that converges more rapidly towards zero than , for decreasing
small values of . The infinitesimal operations of the group form
the linear family

g : λ1 A1 + λ2 A2 + · · · + λr Ar (λ being arbitrary numbers) (71)

g is exactly r-dimensional and the A’s are linearly independent

of one another. For if A is an arbitrary matrix of the group, the
 CHAPTER II 212

group property expresses the transformations of the group which

are infinitely near A in the formula A(E + A), in which  is an
infinitesimal factor and A traverses the group g. If g were of less
dimensions than r, the same would hold at each point of the man-
ifold; for all values of si there would be linear relations between
∂A
the derivatives , and A would in reality depend on less than r
∂si
parameters. The infinitesimal operations generate and determine
the whole group. If we carry out the infinitesimal transformation
1
E + A (n being an infinitely great number) n-times successively,
n
we get a matrix (of the group) that is finite and different from E,
namely,
n
A A2 A3

1
A = lim E + A = E + + + + ...;
n→∞ n 1! 2! 3!
and thus we get every matrix of the group (or at least every one
that may be reached continuously in the group, by starting from
identity) if we make A traverse the whole family g. Not every
arbitrarily given linear family(71) gives a group in this way, but
only those in which the A’s satisfy a certain condition of inte-
grability. The latter is obtained by a method quite analogous to
that by which, for example, the condition of integrability is ob-
tained for parallel displacement in Euclidean space. If we pass
from Identity, E(si = 0), by an infinitesimal change dsi of the
parameters, to the neighbouring matrix Ad = E + dA, and thence
by a second infinitesimal change δsi , from Aδ to Aδ Ad and then
reverse these two operations whilst preserving the same order, we
get Aδ−1 Ad−1 Aδ Ad , a matrix (of the group) differing by an infinitely
small amount from E. Let d be the change in the direction of the
first co-ordinate, and δ that in the direction of the second, then
 THE METRICAL CONTINUUM 213

we are dealing with the matrix

Ast = A−1 −1
t As At As

formed from

As = A(s, 0, 0, . . . , 0) and At = A(0, t, 0, . . . , 0).

Now, As0 = A0t = E, hence

∂ 2 Ast
 
Ast − E
lim = .
s→0,t→0 s·t ∂s ∂t s→0
t→0

Since Ast belongs to the group, this limit is an infinitesimal oper-

ation of the group. We find, however, that
∂Ast
= −A2 + A−1
s A2 As for t = 0;
∂t
leading to

∂ 2 Ast
= −A1 A2 + A2 A1 for t → 0, s → 0.
∂s ∂t
Accordingly A1 A2 − A2 A1 , or, more generally, Ai Ak − Ak Ai must
be an infinitesimal operation of the group: or, what amounts
to the same thing, if A and B are two infinitesimal operations
of the group, then AB − BA must also always be one. Sophus
Lie, to whom we are indebted for the fundamental conceptions
and facts of the theory of continuous transformation groups
(vide note 12), has shown that this condition of integrability
is not only necessary but also sufficient. Hence we may define
an r-dimensional linear family of matrices as an infinitesimal
 CHAPTER II 214

group having r members if, whenever any two matrices A and B

belong to the family, AB − BA also belongs to the family. By
introducing the infinitesimal operations of the group, the problem
of continuous transformation groups becomes a linear question.
If all the transformations of the group leave the elements of
volume unaltered, the “traces” of the infinitesimal operations =
0. For the development of the determinant of E + A in powers
of  begins with the members 1 +  · trace(A). U is a similar
transformation, if, for every G of the group of rotations, U GU −1
or, what comes to the same thing, U GU −1 G−1 , belongs to the
group of rotations G. Accordingly, A∗0 is an infinitesimal operation
of the group of similar transformations if, and only if, A∗0 A − AA∗0
also belongs to g, no matter which of the matrices A of the group
of infinitesimal rotations is used.
The infinitesimal Euclidean rotations

dξ i = vki ξ k ,

that is, the infinitesimal linear transformations that leave the unit
quadratic form

Q0 = (ξ 1 )2 + (ξ 2 )2 + · · · + (ξ n )2

invariant, were determined on page 69. The condition which char-

acterises them, namely,
1
2
dQ0 = ξ i dξ i = 0, implies that vik = −vki .

Thus it is seen that we are dealing with the infinitesimal group δ

n(n − 1)
of all skew-symmetrical matrices; it obviously has mem-
2
bers. It may be left to the reader to verify by direct calculation
 THE METRICAL CONTINUUM 215

that it possesses the group property. If Q is any quadratic form

that remains invariant during the infinitesimal Euclidean rota-
tions, i.e. dQ = 0, then Q necessarily coincides with Q0 except for
a constant factor. Indeed, if

Q = aik ξ i ξ k (aki = aik )

then for all skew-symmetrical number systems vki the equation

ark vik + ari vkr = 0 (72)

must hold. If we assume k = i and notice that the numbers

vi1 , vi2 , . . . , vin may be chosen arbitrarily for each particular i, ex-
cepting the case vii = 0, we get ari = 0 for r 6= i. If we write aii
for ai , equation (72) becomes

vik (ai − ak ) = 0

from which we immediately deduce that all ai ’s are equal. The

corresponding group δ ∗ of similar transformations is derived from δ
by “associating” the single matrix E; this here signifies dξ i = ξ i .
For if the matrix C = (cki ) belongs to δ ∗ , that is, if for every skew-
symmetrical vik , cir vkr − vri crk is also a skew-symmetrical number
system, then the quantities cik + cki = aik satisfy equation (72);
whence it follows that aik = 2a · δik ; that is, C is equal to aE plus
a skew-symmetrical matrix.
More generally, let δQ denote the infinitesimal group of lin-
ear transformations that transform an arbitrary non-degenerate
quadratic form Q into itself. δQ and δQ0 are distinguished only by
their orientation, if Q0 is generated from Q by a linear transfor-
mation. Hence there are only a finite number of different kinds of
 CHAPTER II 216

infinitesimal groups δQ that differ from one another in the iner-

tial index attached to the form Q. But even these differences are
eliminated if, instead of confining ourselves to the realm of real
quantities, we use that of complex members; in that case, every δQ
is of the same type as δ.
These preliminary remarks enable us to formulate analytically
the two postulates I and II. Let g be the group of infinitesimal
rotations at P . We take Λikr to denote every system of n3 num-
bers, Aikr to denote every system that is composed of matrices
(Aik1 ), (Aik2 ), . . . , (Aikn ) belonging to g and Γikr to denote an ar-
bitrary system of numbers that satisfies the condition of sym-
metry (70). If the group of infinitesimal rotations has N mem-
bers, these member systems form linear manifolds of n3 , nN and
n(n + 1)
n· dimensions respectively. Since, according to I, if
2
the metrical relationship runs through all possible values, any
arbitrary number systems Λik1 , Λik2 , . . . , Λikn may occur as the co-
efficients of n infinitesimal congruent transferences in the n co-
ordinate directions (cf. (69)), then, by II (cf. (68)) each Λ must
be capable of resolution in one and only one way according to the
formula
Λikr = Aikr − Γikr .
This entails two results
n(n + 1) n(n − 1)
1. n3 = nN + n · or N = ;
2 2
2. Aikr − Γikr is never equal to zero, unless all the A’s and Γ’s
vanish; or, a non-vanishing system A can never fulfil the condition
of symmetry, Aikr = Airk . To enable us to formulate this con-
dition invariantly let us define a symmetrical double matrix (an
infinitesimal double rotation) belonging to g as a law expressed
 THE METRICAL CONTINUUM 217

by
ζ i = Airs ξ r η s (Airs = Aisr ),
which produces from two arbitrary vectors, ξ and η, a vector ζ as a
bilinear symmetrical form, provided that for every fixed vector η,
the transition ξ → ζ (and hence also for every fixed vector ξ the
transition η → ζ) is an operation of g. We may then summarise
our results thus:—
The group of infinitesimal rotations has the following proper-
ties according to our axioms:
(a) The trace of every matrix = 0;
(b) No symmetrical double matrix belongs to g except zero;
(c) The dimensional number of g is the highest that is still in
n(n − 1)
agreement with postulate (b), namely, N = .
2
These properties retain their meaning for complex quantities
as well as for real ones. We shall just verify that they are true
of the infinitesimal Euclidean group of rotations δ, that is, that
n3 numbers vkl i
cannot simultaneously satisfy the conditions of
symmetry
i i
vlk = vkl , vilk = −vkl
i
,
without all of them vanishing. This is evident from the calcula-
tion which was undertaken on page 184 to determine the affine
relationship. For if we write down the three equations that we get
i
from vkl + vilk = 0 by interchanging the indices i, k, l cyclically, and
then subtract the second from the sum of the first and the third,
i
we get, as a result of the first condition of symmetry, vkl = 0.
It seems highly probable to the author that δ is the only in-
finitesimal group that satisfies the postulates (a), (b), and (c);
or, more exactly, in the case of complex quantities every such in-
finitesimal group may be made to coincide with δ by choosing the
 CHAPTER II 218

appropriate co-ordinate system. If this is true, then the group of

infinitesimal rotations must be identical with a certain group δQ ,
in which Q is a non-degenerate quadratic form. Q itself is deter-
mined by g except for a constant of proportionality. It is real if
g is real. For if we split Q (in which the variables are taken as
real) into a real and an imaginary part Q1 + iQ2 , then g leaves
both these forms Q1 and Q2 invariant. Hence we must have

Q1 = c1 Q Q2 = c2 Q.

One of these two constants is certainly different from zero, since

c1 + ic2 = 1, and hence Q must be a real form excepting for a con-
stant factor. This would link up with the line of argument followed
in the preceding paragraph and would complete the Analysis of
Space; we should then be able to claim to have made intelligible
the nature of space and the source of the validity of Pythagoras’
Theorem, by having explored the ultimate grounds accessible to
mathematical reasoning (vide note 13). If the supposed mathe-
matical proposition is not true, definite characteristics and essen-
tials of space will yet have escaped us. The author has proved that
the proposition holds actually for the lowest dimensional numbers
n = 2 and n = 3. It would lead too far to present these purely
mathematical considerations here.
In conclusion, it will be advisable to call attention to two
points. Firstly, axiom I is in no wise contradicted by the re-
sult of axiom II which states that not only the metrical structure,
but also the metrical relationship is of the same kind at every
point, namely, of the simplest type imaginable. For every point
there is a geodetic co-ordinate system such that the shifting of
all vectors at that point, which leaves its components unaltered,
to a neighbouring point is always a congruent transference. Sec-
 THE METRICAL CONTINUUM 219

ondly, the possibility of grasping the unique significance of the

metrical structure of Pythagorean space in the way here outlined
depends solely on the circumstance that the quantitative metrical
conditions admit of considerable virtual changes. This possibility
stands or falls with the dynamical view of Riemann. It is this
view, the truth of which can scarcely be doubted after the success
that has attended Einstein’s Theory of Gravitation (Chapter IV),
that opens up the road leading to the discovery of the “Rationality
of Space”.
The investigations about space that have been conducted in
Chapter II seemed to the author to offer a good example of the
kind of analysis of the modes of existence (Wesensanalyse) which
is the object of Husserl’s phenomenological philosophy, an exam-
ple that is typical of cases in which we are concerned with non-
immanent modes. The historical development of the problem of
space teaches how difficult it is for us human beings entangled
in external reality to reach a definite conclusion. A prolonged
phase of mathematical development, the great expansion of geom-
etry dating from Euclid to Riemann, the discovery of the physical
facts of nature and their underlying laws from the time of Galilei,
together with the incessant impulses imparted by new empiri-
cal data, finally the genius of individual great minds—Newton,
Gauss, Riemann, Einstein—all these factors were necessary to set
us free from the external, accidental, non-essential characteris-
tics which would otherwise have held us captive. Certainly, once
the true point of view has been adopted reason becomes flooded
with light, and it recognises and appreciates what is of itself in-
telligible to it. Nevertheless, although reason was, so to speak,
always conscious of this point of view in the whole development
of the problem, it had not the power to penetrate into it with one
 CHAPTER II 220

flash. This reproach must be directed at the impatience of those

philosophers who believe it possible to describe adequately the
mode of existence on the basis of a single act of typical presen-
tation (exemplarischer Vergegenwärtigung): in principle they are
right: yet from the point of view of human nature, how utterly
they are wrong! The problem of space is at the same time a very
instructive example of that question of phenomenology that seems
to the author to be of greatest consequence, namely, in how far
the delimitation of the essentialities perceptible in consciousness
expresses the structure peculiar to the realm of presented objects,
and in how far mere convention participates in this delimitation.
 CHAPTER III
RELATIVITY OF SPACE AND TIME
§ 19. Galilei’s Principle of Relativity
We have already discussed in the introduction how it is pos-
sible to measure time by means of a clock and how, after an ar-
bitrary initial point of time and a time-unit has been chosen, it is
possible to characterise every point of time by a number t. But
the union of space and time gives rise to difficult further prob-
lems that are treated in the theory of relativity. The solution of
these problems, which is one of the greatest feats in the history
of the human intellect, is associated above all with the names of
Copernicus and Einstein (vide note 1).
By means of a clock we fix directly the time-conditions of only
such events as occur just at the locality at which the clock hap-
pens to be situated. Inasmuch as I, as an unenlightened being,
fix, without hesitation, the things that I see into the moment of
their perception, I extend my time over the whole world. I believe
that there is an objective meaning in saying of an event which is
happening somewhere that it is happening “now” (at the moment
at which I pronounce the word!); and that there is an objective
meaning in asking which of two events that have happened at
different places has occurred earlier or later than the other. We
shall for the present accept the point of view implied
in these assumptions. Every space-time event that is strictly
localised, such as the flash of a spark that is instantaneously ex-
tinguished, occurs at a definite space-time-point or world-point,
“here-now”. As a result of the point of view enunciated above, to
every world-point there corresponds a definite time-co-ordinate t.

221
 CHAPTER III 222

We are next concerned with fixing the position of such a point-

event in space. For example, we ascribe to two point-masses a dis-
tance separating them at a definite moment. We assume that the
world-points corresponding to a definite moment t form a three-
dimensional point-manifold for which Euclidean geometry holds.
(In the present chapter we adopt the view of space set forth in
Chapter I.) We choose a definite unit of length and a rectangu-
lar co-ordinate system at the moment t (such as the corner of a
room). Every world-point whose time-co-ordinate is t then has
three definite space-co-ordinates x1 , x2 , x3 .
Let us now fix our attention on another moment t0 . We assume
that there is a definite objective meaning in stating that measure-
ments are carried out at the moment t0 with the same unit length
as that used at the moment t (by means of a “rigid” measuring
staff that exists both at the time t and at the time t0 ). In ad-
dition to the unit of time we shall adopt a unit of length fixed
once and for all (centimetre, second). We are then still free to
choose the position of the Cartesian co-ordinate system indepen-
dently of the choice of time t. Only when we believe that there
is objective meaning in stating that two point-events happening
at arbitrary moments take place at the same point of space, and
in saying that a body is at rest, are we able to fix the position
of the co-ordinate system for all times on the basis of the posi-
tion chosen arbitrarily at a certain moment, without having to
specify additional “individual objects”; that is, we accept the pos-
tulate that the co-ordinate system remains permanently at rest.
After choosing an initial point in the time-scale and a definite co-
ordinate system at this initial moment we then get four definite
co-ordinates for every world-point. To be able to represent con-
ditions graphically we suppress one space-co-ordinate, assuming
 RELATIVITY OF SPACE AND TIME 223

t t′

x2 t = const.

x1

Fig. 7.

space to be only two-dimensional, a Euclidean plane.

We construct a graphical picture by representing in a space
carrying the rectangular set of axes (x1 , x2 , t) the world-point by
a “picture”-point with co-ordinates (x1 , x2 , t). We can then trace
out graphically the “time-table” of all moving point-masses; the
motion of each is represented by a “world-line,” whose direction
has always a positive component in the direction of the t-axis.
The world-lines of point-masses that are at rest are parallels to
the t-axis. The world-line of a point-mass which is in uniform
translation is a straight line. On a section t = constant we may
read off the position of all the point-masses at the same time t. If
we choose an initial point in the time-scale and also some other
Cartesian co-ordinate system, and if (x1 , x2 , t), (x01 , x02 , t0 ) are the
co-ordinates of an arbitrary world-point in the first and second
 CHAPTER III 224

co-ordinate system respectively, the transformation formulæ

x1 = α11 x01 + α12 x02 + α1 



0 0
x2 = α21 x1 + α22 x2 + α2 (I)
t0 + a 

t=

hold; in them, the αi ’s and the a denote constants, the αik ’s, in
particular, are the co-efficients of an orthogonal transformation.
The world-co-ordinates are thus fixed except for an arbitrary
transformation of this kind in an objective manner without
individual objects or events being specified. In this we have not
yet taken into consideration the arbitrary choice of both units of
measure. If the initial point remains unchanged both in space
and in time, so that α1 = α2 = a = 0, then (x01 , x02 , t0 ) are the co-
ordinates with respect to a rectilinear system of axes whose t0 axis
coincides with the t-axis, whereas the axes x01 , x02 are derived from
x1 , x2 by a rotation in their plane t = 0.
A moment’s reflection suffices to show that one of the assump-
tions adopted is not true, namely, the one which states that the
conception of rest has an objective content.∗ When I arrange to
meet some one at the same place to-morrow as that at which we
met to-day, this means in the same material surroundings, at the
same building in the same street (which, according to Coperni-
cus, may be in a totally different part of stellar space to-morrow).
All this acquires meaning as a result of the fortunate circum-
stance that at birth we are introduced into an essentially stable
world, in which changes occur in conjunction with a comparatively
much more comprehensive set of permanent factors that preserve
∗
Even Aristotle was clear on this point, for he denotes “place” (τόπος) as
the relation of a body to the bodies in its neighbourhood.
 RELATIVITY OF SPACE AND TIME 225

their constitution (which is partly perceived directly and partly

deduced) unchanged or almost unchanged. The houses stand still;
ships travel at so and so many knots: these things are always un-
derstood in ordinary life as referring to the firm ground on which
we stand. Only the motions of bodies (point-masses) rel-
ative to one another have an objective meaning, that is,
the distances and angles that are determined from simultaneous
positions of the point-masses and their functional relation to the
time-co-ordinate. The connection between the co-ordinates of the
same world-point expressed in two different systems of this kind
is given by formulæ

x1 = α11 (t0 )x01 + α12 (t0 )x02 + α1 (t0 )



0 0 0 0 0
x2 = α21 (t )x1 + α22 (t )x2 + α2 (t ) (II)
0

t=t +a 

in which the αi ’s and αik ’s may be any continuous functions of t0 ,

and the αik ’s are the co-efficients of an orthogonal transformation
for all values of t0 . If we map out the surfaces t0 = const., as also
x01 = const. and x02 = const. by our graphical method, then the
surfaces of the first family are again planes that coincide with the
planes t = const.; on the other hand, the other two families of
surfaces are curved surfaces. The transformation formulæ are no
longer linear.
Under these circumstances we achieve an important aim, when
investigating the motion of systems of point-masses, such as plan-
ets, by choosing the co-ordinate system so that the functions
x1 (t), x2 (t) that express how the space-co-ordinates of the point-
masses depend on the time become as simple as possible or at
least satisfy laws of the greatest possible simplicity. This is the
 CHAPTER III 226

substance of the discovery of Copernicus that was afterwards elab-

orated to such an extraordinary degree by Kepler, namely, that
there is in fact a co-ordinate system for which the laws of planetary
motion assume a much simpler and more expressive form than if
they are referred to a motionless earth. The work of Copernicus
produced a revolution in the philosophic ideas about the world
inasmuch as he shattered the belief in the absolute impor-
tance of the earth. His reflections as well as those of Kepler are
purely kinematical in character. Newton crowned their work by
discovering the true ground of the kinematical laws of Kepler to lie
in the fundamental dynamical law of mechanics and in the law
of attraction. Every one knows how brilliantly the mechanics of
Newton has been confirmed both for celestial as well as for earthly
phenomena. As we are convinced that it is valid universally and
not only for planetary systems, and as its laws are by no means
invariant with respect to the transformations (II), it enables us to
fix the co-ordinate system in a manner independent of all individ-
ual specification and much more definitely than is possible on the
kinematical view to which the principle of relativity (II) leads.
Galilei’s Principle of Inertia (Newton’s First Law of Mo-
tion) forms the foundation of mechanics. It states that a point-
mass which is subject to no forces from without executes a uniform
translation. Its world-line is consequently a straight line, and the
space-co-ordinates x1 , x2 of the point-mass are linear functions of
the time t. If this principle holds for the two co-ordinate systems
connected by (II), then x1 and x2 must become linear functions
of t0 , when linear functions of t0 are substituted for x01 and x02 .
It straightway follows from this that the αik ’s must be constants,
and that α1 and α2 must be linear functions of t; that is, the one
Cartesian co-ordinate system (in space) must be moving uniformly
 RELATIVITY OF SPACE AND TIME 227

in a straight line relatively to the other co-ordinate system. Con-

versely, it is easily shown that if C1 , C2 are two such co-ordinate
systems, then if the principle of inertia and Newtonian mechanics
holds for C it will also hold for C0 . Thus, in mechanics, any two
“allowable” co-ordinate systems are connected by formulæ

x1 = α11 x01 + α12 x02 + γ1 t0 + α1 



0 0 0
x2 = α21 x1 + α22 x2 + γ2 t + α2 (III)
0

t= t +a


in which the αik ’s are constant co-efficients of an orthogonal trans-

formation, and a, αi and γi are arbitrary constants. Every trans-
formation of this kind represents a transition from one allowable
co-ordinate system to another. (This is the Principle of Rel-
ativity of Galilei and Newton.) The essential feature of this
transition is that, if we disregard the naturally arbitrary direc-
tions of the axis in space and the arbitrary initial point, there is
invariance with respect to the transformations

x1 = x01 + γ1 t0 , x2 = x02 + γ2 t0 , t = t0 . (1)

In our graphical representation (vide Fig. 7) x01 , x02 , t0 would be the

co-ordinates taken with respect to a rectilinear set of axes in which
the x01 -, x02 -axes coincide with the x1 -, x2 -axes, whereas the new
t0 -axis has some new direction. The following considerations show
that the laws of Newtonian mechanics are not altered in passing
from one co-ordinate system C to another C0 . According to the
law of attraction the gravitational force with which one point-mass
acts on another at a certain moment is a vector, in space, which is
independent of the co-ordinate system (as is also the vector that
connects the simultaneous positions of both point-masses with one
 CHAPTER III 228

another). Every force, no matter what its physical origin, must be

the same kind of magnitude; this is entailed in the assumptions of
Newtonian mechanics, which demands a physics that satisfies this
assumption in order to be able to give a content to its conception of
force. We may prove, for example, in the theory of elasticity that
the stresses (as a consequence of their relationship to deformation
quantities) are of the required kind.
Mass is a scalar that is independent of the co-ordinate sys-
tem. Finally, on account of the transformation formulæ that result
from (1) for the motion of a point-mass,

dx1 dx01 dx2 dx02 d2 x1 d2 x01 d2 x2 d2 x02

= 0 + γ1 , = 0 + γ2 ; = 02 , = 02
dt dt dt dt dt2 dt dt2 dt
not the velocity, but the acceleration is a vector (in space) inde-
pendent of the co-ordinate system. Accordingly, the fundamental
law: mass times acceleration = force, has the required invari-
ant property.
According to Newtonian mechanics the centre of inertia of ev-
ery isolated mass-system not subject to external forces moves in
a straight line. If we regard the sun and his planets as such a sys-
tem, there is no meaning in asking whether the centre of inertia of
the solar system is at rest or is moving with uniform translation.
The fact that astronomers, nevertheless, assert that the sun is
moving towards a point in the constellation of Hercules, is based
on the statistical observation that the stars in that region seem
on the average to diverge from a certain centre—just as a cluster
of trees appears to diverge as we approach them. If it is certain
that the stars are on the average at rest, that is, that the centre
of inertia of the stellar firmament is at rest, the statement about
the sun’s motion follows. It is thus merely an assertion about the
 RELATIVITY OF SPACE AND TIME 229

relative motion of the centre of inertia and of that of the stellar

firmament.
To grasp the true meaning of the principle of relativity, one
must get accustomed to thinking not in “space,” nor in “time,” but
“in the world,” that is in space-time. Only the coincidence (or
the immediate succession) of two events in space-time has a mean-
ing that is directly evident, it is just the fact that in these cases
space and time cannot be dissociated from one another absolutely
that is asserted by the principle of relativity. Following the mech-
anistic view, according to which all physical happening can be
traced back to mechanics, we shall assume that not only mechan-
ics but the whole of the physical uniformity of Nature is subject to
the principle of relativity laid down by Galilei and Newton, which
states that it is impossible to single out from the systems of refer-
ence that are equivalent for mechanics and of which each two are
correlated by the formula of transformation (III) special systems
without specifying individual objects. These formulæ condition
the geometry of the four-dimensional world in exactly the
same way as the group of transformation substitutions connecting
two Cartesian co-ordinate systems condition the Euclidean geom-
etry of three-dimensional space. A relation between world-points
has an objective meaning if, and only if, it is defined by such
arithmetical relations between the co-ordinates of the points as
are invariant with respect to the transformations (III). Space is
said to be homogeneous at all points and homogeneous in all di-
rections at every point. These assertions are, however, only parts
of the complete statement of homogeneity that all Cartesian
co-ordinate systems are equivalent. In the same way the princi-
ple of relativity determines exactly the sense in which the world
(= space-time as the “form” of phenomena, not its “accidental”
 CHAPTER III 230

non-homogeneous material content) is homogeneous.

It is indeed remarkable that two mechanical events that are
fully alike kinematically, may be different dynamically, as a com-
parison of the dynamical principle of relativity (III) with the much
more general kinematical principle of relativity (II) teaches us. A
rotating spherical mass of fluid existing all alone, or a rotating fly-
wheel, cannot in itself be distinguished from a spherical fluid mass
or a fly-wheel at rest; in spite of this the “rotating” sphere becomes
flattened, whereas the one at rest does not change its shape, and
stresses are called up in the rotating fly-wheel that cause it to
burst asunder, if the rate of rotation be sufficiently great, whereas
no such effect occurs in the case of a fly-wheel which is at rest.
The cause of this varying behaviour can be found only in the “met-
rical structure of the world,” that reveals itself in the centrifugal
forces as an active agent. This sheds light on the idea quoted from
Riemann above; if there corresponds to metrical structure (in this
case that of the world and not the fundamental metrical tensor
of space) something just as real, which acts on matter by means
of forces, as the something which corresponds to Maxwell’s stress
tensor, then we must assume that, conversely, matter also reacts
on this real something. We shall revert to this idea again later in
Chapter IV.
For the present we shall call attention only to the linear char-
acter of the transformation formulæ (III); this signifies that the
world is a four-dimensional affine space. To give a system-
atic account of its geometry we accordingly use world-vectors
or displacements in addition to world-points. A displacement of
the world is a transformation that assigns to every world-point P
a world-point P 0 , and is characterised by being expressible in an
 RELATIVITY OF SPACE AND TIME 231

allowable co-ordinate system by means of equations of the form

x0i = xi + αi (i = 0, 1, 2, 3)

in which the xi ’s denote the four space-time-co-ordinates of P

(t being represented by x0 ), and the x0i ’s are those of P 0 in this co-
ordinate system, whereas the αi ’s are constants. This conception
is independent of the allowable co-ordinate system selected. The
displacement that transforms P into P 0 (or transfers P to P 0 )
−−→
is denoted by P P 0 . The world-points and displacements satisfy
all the axioms of the affine geometry whose dimensional number
is n = 4. Galilei’s Principle of Inertia (Newton’s First Law of
Motion) is an affine law; it states what motions realise the straight
lines of our four-dimensional affine space (“world”), namely, those
executed by point-masses moving under no forces.
From the affine point of view we pass on to the metrical
one. From the graphical picture, which gave us an affine view
of the world (one co-ordinate being suppressed), we can read off
its essential metrical structure; this is quite different from that of
Euclidean space. The world is “stratified”; the planes, t = const.,
in it have an absolute meaning. After a unit of time has been cho-
sen, each two world-points A and B have a definite time-difference,
−→
the time-component of the vector AB = x; as is generally the case
with vector-components in an affine co-ordinate system, the time-
component is a linear form t(x) of the arbitrary vector x. The
vector x points into the past or the future according as t(x) is
negative or positive. Of two world-points A and B, A is earlier
than, simultaneous with, or later than B, according as
−→
t(AB) > 0, = 0, or < 0.
 CHAPTER III 232

Euclidean geometry, however, holds in each “stratum”; it is based

on a definite quadratic form, which is in this case defined only
for those world-vectors x that lie in one and the same stratum,
that is, that satisfy the equation t(x) = 0 (for there is sense only
in speaking of the distance between simultaneous positions of
two point-masses). Whereas, then, the metrical structure of
Euclidean geometry is based on a definitely positive quadratic
form, that of Galilean geometry is based on
1. A linear form t(x) of the arbitrary vector x (the “duration”
of the displacement x).
2. A definitely positive quadratic form (x, x) (the square of
the “length” of x), which is defined only for the three-dimensional
linear manifold of all the vectors x that satisfy the equation
t(x) = 0.
We cannot do without a definite space of reference, if we wish
to form a picture of physical conditions. Such a space depends
on the choice of an arbitrary displacement e in the world (within
which the time-axis falls in the picture), and is then defined by
the convention that all world-points that lie on a straight line of
direction e, meet at the same point of space. In geometrical
language, we are merely dealing with the process of parallel pro-
jection. To arrive at an appropriate formulation we shall begin
with some geometrical considerations that relate to an arbitrary
n-dimensional affine space. To enable us to form a picture of the
processes we shall confine ourselves to the case n = 3. Let us take
a family of straight lines in space all drawn parallel to the vector e
(6= 0). If we look into space along these rays, all the space-points
that lie behind one another in the direction of such a straight line
would coincide; it is in no wise necessary to specify a plane on to
which the points are projected. Hence our definition assumes the
 RELATIVITY OF SPACE AND TIME 233

following form.
Let e, a vector differing from 0, be given. If A and A0 are
−−→
two points such that AA0 is a multiple of e, we shall say that
they pass into one and the same point A of the minor space
defined by e. We may represent A by the straight line parallel
to e, on which all these coincident points A, A0 , . . . in the minor
space lie. Since every displacement x of the space transforms a
straight line parallel to e again into one parallel to e, x brings
about a definite displacement x of the minor space; but each two
displacements x and x0 become coincident in the minor space, if
their difference is a multiple of e. We shall denote the transition to
the minor space, “the projection in the direction of e,” by printing
the symbols for points and displacements in heavy oblique type.
Projection converts
−→ −→
λx, x + y, and AB into λx, x + y, AB

that is, the projection has a true affine character; this means that
in the minor space affine geometry holds, of which the dimensions
are less by one than those of the original “complete” space.
If the space is metrical in the Euclidean sense, that is, if it
is based on a non-degenerate quadratic form which is its metrical
groundform, Q(x) = (x, x),—to simplify the picture of the process
we shall keep the case for which Q is definitely positive in view, but
the line of proof is applicable generally,—then we shall obviously
ascribe to the two points of the minor space, which two straight
lines parallel to e appear to be, when we look into the space in
the direction of e, a distance equal to the perpendicular distance
between the two straight lines. Let us formulate this analytically.
The assumption is that (e, e) = e 6= 0. Every displacement x may
 CHAPTER III 234

be split up uniquely into two summands

x = ξe + x∗ , (2)

of which the first is proportional to e and the second is perpen-

dicular to it, viz.:—
1
(x∗ , e) = 0, ξ = (x, e). (3)
e
We shall call ξ the height of the displacement x (it is the differ-
−→
ence of height between A and B, if x = AB). We have

(x, x) = eξ 2 + (x∗ , x∗ ). (4)

x is characterised fully, if its height ξ and the displacement x of

the minor space produced by x are given; we write

x = ξ | x.

The “complete” space is “split up” into height and minor space,
the “position-difference” x of two points in the complete space
is split up into the difference of height ξ, and the difference of
position x in the minor space. There is a meaning not only in
saying that two points in space coincide, but also in saying that
two points in the minor space coincide or have the same height,
respectively. Every displacement x of the minor space is produced
by one and only one displacement x∗ of the complete space,
this displacement being orthogonal to e. The relation between x∗
and x is singly reversible and affine. The defining equation

(x, x) = (x∗ , x∗ )
 RELATIVITY OF SPACE AND TIME 235

endows the minor space with a metrical structure that is based

on the quadratic groundform (x, x). This converts (4) into the
fundamental equation of Pythagoras

(x, x) = eξ 2 + (x, x) (5)

which, for two displacements, may be generalised in the form

(x, y) = eξη + (x, y ). (50 )

Its symbolic form is clear.

These considerations, in so far as they concern affine space,
may be applied directly. The complete space is the four-
dimensional world: e is any vector pointing in the direction
of the future: the minor space is what we generally call space.
Each two world-points that lie on a world-line parallel to e project
into the same space-point. This space-point may be represented
graphically by the straight line parallel to e and may be indicated
permanently by a point-mass at rest, that is, one whose world-
line is just that straight line. The metrical structure, however, is,
according to the Galilean principle of relativity, of a kind different
from that we assumed just above. This necessitates the following
modifications. Every world-displacement x has a definite dura-
tion t(x) = t (this takes the place of “height” in our geometrical
argument) and produces a displacement x in the minor space; it
splits up according to the formula

x=t|x

corresponding to the resolution into space and time. In particular

every space-displacement x may be produced by one and only one
world-displacement x∗ , which satisfies the equation t(x∗ ) = 0.
 CHAPTER III 236

The quadratic form (x∗ , x∗ ) as defined for such vectors x∗ ,

impresses on space its Euclidean metrical structure

(x, x) = (x∗ , x∗ ).

The space is dependent on the direction of projection. In actual

cases the direction of projection may be fixed by any point-mass
moving with uniform translation (or by the centre of mass of a
closed isolated mass-system).
We have set forth these details with pedantic accuracy so as to
be armed at least with a set of mathematical conceptions which
have been sifted into a form that makes them immediately appli-
cable to Einstein’s principle of relativity for which our powers of
intuition are much more inadequate than for that of Galilei.
To return to the realm of physics. The discovery that light is
propagated with a finite velocity gave the death-blow to the
natural view that things exist simultaneously with their percep-
tion. As we possess no means of transmitting time-signals more
rapid than light itself (or wireless telegraphy) it is of course impos-
sible to measure the velocity of light by measuring the time that
elapses whilst a light-signal emitted from a station A travels to a
station B. In 1675 Römer calculated this velocity from the appar-
ent irregularity of the time of revolution of Jupiter’s moons, which
took place in a period which lasted exactly one year: he argued
that it would be absurd to assume a mutual action between the
earth and Jupiter’s satellites such that the period of the earth’s
revolution caused a disturbance of so considerable an amount in
the satellites. Fizeau confirmed the discovery by measurements
carried out on the earth’s surface. His method is based on the
simple idea of making the transmitting station A and the receiv-
ing station B coincide by reflecting the ray, when it reaches B,
 RELATIVITY OF SPACE AND TIME 237

back to A. According to these measurements we have to assume

that the centre of the disturbances is propagated in concentric
spheres with a constant velocity c. In our graphical picture (one
space-co-ordinate again being suppressed) the propagation of a
light-signal emitted at the world-point O is represented by the
circular cone depicted, which has the equation

c2 t2 − (x21 + x22 ) = 0. (6)

Every plane given by t = const. cuts the cone in a circle composed

of those points which the light-signal has reached at the moment t.
The equation (6) is satisfied by all and only by all those world-
points reached by the light-signal (provided that t > 0). The
question again arises on what space of reference this description
of the event is based. The aberration of the stars shows that,
relatively to this reference space, the earth moves in agreement
with Newton’s theory, that is, that it is identical with an allowable
reference space as defined by Newtonian mechanics. The propa-
gation in concentric spheres is, however, certainly not invariant
with respect to the Galilei transformations (III); for a t0 -axis that
is drawn obliquely intersects the planes t = const. at points that
are excentric to the circles of propagation. Nevertheless, this can-
not be regarded as an objection to Galilei’s principle of relativity,
if, accepting the ideas that have long held sway in physics, we as-
sume that light is transmitted by a material medium, the æther,
whose particles are movable with regard to one another. The con-
ditions that obtain in the case of light are exactly similar to those
that bring about concentric circles of waves on a surface of water
on to which a stone has been dropped. The latter phenomenon
certainly does not justify the conclusion that the equations of hy-
drodynamics are contrary to Galilei’s principle of relativity. For
 CHAPTER III 238

the medium itself, the water or the æther respectively, whose par-
ticles are at rest with respect to one another, if we neglect the
relatively small oscillations, furnishes us with the same system of
reference as that to which the statement concerning the concentric
transmission is referred.
To bring us into closer touch with this question we shall here
insert an account of optics in the theoretical guise that it has
preserved since the time of Maxwell under the name of the theory
of moving electromagnetic fields.

§ 20. The Electrodynamics of Moving Fields Lorentz’s

Theorem of Relativity
In passing from stationary electromagnetic fields to moving
electromagnetic fields (that is, to those that vary with the time)
we have learned the following:—
1. The so-called electric current is actually composed of mov-
ing electricity: a charged coil of wire in rotation produces a mag-
netic field according to the law of Biot and Savart. If ρ is the
density of charge, v the velocity, then clearly the density s of this
convection current = ρv; yet, if the Biot-Savart Law is to remain
valid in the old form, s must be measured in other units. Thus
ρv
we must set s = , in which c is a universal constant having the
c
dimensions of a velocity. The experiment carried out by Weber
and Kohlrausch, repeated later by Rowland and Eichenwald, gave
a value of c that was coincident with that obtained for the velocity
of light, within the limits of errors of observation (vide note 2).
ρ
We call = ρ0 the electromagnetic measure of the charge-density
c
and, so as to make the density of electric force = ρ0 E0 in electro-
 RELATIVITY OF SPACE AND TIME 239

magnetic units, too, we call E0 = cE the electromagnetic measure

of the field-intensity.
2. A moving magnetic field induces a current in a homogeneous
wire. It may be determined from the physical law s = σE and
Faraday’s Law of Induction; the latter asserts that the induced
electromotive force is equal to the time-decrement of the magnetic
flux through the conductor; hence we have
d
Z Z
0
E dr = − Bn do. (7)
dt
On the left there is the line-integral along a closed curve, on the
right the surface-integral of the normal components of the mag-
netic induction B, taken over a surface which fills the curve. The
flux of induction through the conducting curve is uniquely deter-
mined because
div B = 0; (80 )
that is, there is no real magnetism. By Stokes’ Theorem we get
from (7) the differential law
1 ∂B
curl E + = 0. (8)
c ∂t
The equation curl E = 0, which holds for statistical cases, is hence
1 ∂B
increased by the term on the left, which is a derivative of
c ∂t
the time. All our electro-technical sciences are based on it; thus
the necessity for introducing it is justified excellently by actual
experience.
3. On the other hand, in Maxwell’s time, the term which was
added to the fundamental equation of magnetism
curl H = s (9)
 CHAPTER III 240

was purely hypothetical. In a moving field, such as in the discharge

of a condenser, we cannot have div s = 0, but in place of it the
“equation of continuity”
1 ∂ρ
+ div s = 0 (10)
c ∂t
must hold. This gives expression to the fact that the current
consists of moving electricity. Since ρ = div D, we find that not s,
1 ∂D
but s + must be irrotational, and this immediately suggests
c ∂t
that instead of equation (9) we must write for moving fields

1 ∂D
curl H − = s. (11)
c ∂t
Besides this, we have just as before

div D = ρ. (110 )

From (11) and (110 ) we arrive conversely at the equation of conti-

1 ∂D
nuity (10). It is owing to the additional member (Maxwell’s
c ∂t
displacement current), a differential co-efficient with respect
to the time, that electromagnetic disturbances are propagated in
the æther with the finite velocity c. It is the basis of the electro-
magnetic theory of light, which interprets optical phenomena with
such wonderful success, and which is experimentally verified in the
well-known experiments of Hertz and in wireless telegraphy, one
of its technical applications. This also makes it clear that these
laws are referred to the same reference-space as that for which the
concentric propagation of light holds, namely, the “fixed” æther.
The laws involving the specific characteristics of the matter under
 RELATIVITY OF SPACE AND TIME 241

consideration have yet to be added to Maxwell’s field-equations

(8) and (80 ), (11) and (110 ).
We shall, however, here consider only the conditions in the
æther; in it
D = E and H = B,
and Maxwell’s equations are
1 ∂B
curl E + = 0, div B = 0, (12)
c ∂t
1 ∂E
curl B − = s, div E = ρ. (120 )
c ∂t
According to the atomic theory of electrons these are generally
ρv
valid exact physical laws. This theory furthermore sets s = ,
c
in which v denotes the velocity of the matter with which the
electric charge is associated.
The force which acts on the masses consists of components
arising from the electrical and the magnetic field: its density is

p = ρE + [s, B]. (13)

Since s is parallel to v, the work performed on the electrons per

unit of time and of volume is

p · v = ρE · v = c(s, E) = s · E0 .

It is used in increasing the kinetic energy of the electrons, which

is partly transferred to the neutral molecules as a result of col-
lisions. This augmented molecular motion in the interior of the
conductor expresses itself physically as the heat arising during this
phenomenon, as was pointed out by Joule. We find, in fact, ex-
perimentally that s · E0 is the quantity of heat produced per unit
 CHAPTER III 242

of time and per unit of volume by the current. The energy used
up in this way must be furnished by the instrument providing the
current. If we multiply equation (12) by −B, equation (120 ) by E
and add, we get
∂ 1 2 1 2
−c · div[E, B] − ( E + 2 B ) = c(s, E).
∂t 2
If we set
1 2
[E, B] = s, 2
E + 21 B2 = W
and integrate over any volume V , this equation becomes
d
Z Z Z
− W dV + c Sn do = c(s, E) dV.
dt V Ω V

The second member on the left is the integral, taken over the
outer surface of V1 , of the component sn of s along the inward
normal. On the right-hand side we have the work performed on
the volumeZ V per unit of time. It is compensated by the decrease
of energy W dV contained in V and by the energy that flows
into the portion of space V from without. Our equation is thus
an expression of the energy theorem. It confirms the as-
sumption which we made initially about the density W
of the field-energy, and we furthermore see that cs, familiarly
known as Poynting’s vector, represents the energy stream or
energy-flux.
The field-equations (12), (120 ) have been integrated by Lorentz
in the following way, on the assumption that the distribution of
charges and currents are known. The equation div B = 0 is satis-
fied by setting
−B = curl f (14)
 RELATIVITY OF SPACE AND TIME 243

in which −f is the vector potential. By substituting this in the

1 ∂f
first equation above we get that E − is irrotational, so that
c ∂t
we can set
1 ∂f
E− = grad φ, (15)
c ∂t
in which −φ is the scalar potential. We may make use of the arbi-
trary character yet possessed by f by making it fulfil the subsidiary
condition
1 ∂φ
+ div f = 0.
c ∂t
This is found to be expedient for our purpose (whereas for a sta-
tionary field we assumed div f = 0). If we introduce the potentials
in the two latter equations, we find by an easy calculation
1 ∂ 2φ
− + ∆φ = ρ, (16)
c2 ∂t2
1 ∂ 2f
− 2 2 + ∆f = s. (160 )
c ∂t
An equation of the form (16) denotes a wave disturbance travelling
with the velocity c. In fact, just as Poisson’s equation ∆φ = ρ has
the solution
ρ
Z
−4πφ = dV
r
so (16) has the solution

Z ρ t− r
 

−4πφ = c dV ;
r
on the left-hand side of which φ is the value at a point O at
time t; r is the distance of the source P , with respect to which we
 CHAPTER III 244

integrate, from the point of emergence O; and within the integral

r
the value of ρ is that at the point P at time t − . Similarly
c
(160 ) has the solution

Z s t− r
 

−4πf = c dV.
r
The field at a point does not depend on the distribution of charges
and currents at the same moment, but the determining r factor for
every point is the moment that lies back just as many ’s as the
c
disturbance propagating itself with the velocity c takes to travel
from the source to the point of emergence.
Just as the expression for the potential (in Cartesian co-
ordinates), namely,
∂ 2φ ∂ 2φ ∂ 2φ
∆φ = + +
∂x21 ∂x22 ∂x23
is invariant with respect to linear transformations of the variables
x1 , x2 , x3 , which are such that they convert the quadratic form

x21 + x22 + x23

into itself, so the expression which takes the place of this expres-
sion for the potential when we pass from statical to moving fields,
namely,
1 ∂ 2φ ∂ 2φ ∂ 2φ ∂ 2φ
− + 2+ 2+ 2 (retarded potentials)
c2 ∂t2 ∂x1 ∂x2 ∂x3
is an invariant for those linear transformations of the four co-
ordinates, t, x1 , x2 , x3 , the so-called Lorentz transformations, that
 RELATIVITY OF SPACE AND TIME 245

transform the indefinite form

−c2 t2 + x21 + x22 + x23 (17)
into itself. Lorentz and Einstein recognised that not only equa-
tion (16) but also the whole system of electromagnetic laws for the
æther has this property of invariance, namely, that these laws are
the expression of invariant relations between tensors which exist in
a four-dimensional affine space whose co-ordinates are t, x1 , x2 , x3
and upon which a non-definite metrical structure is impressed by
the form (17). This is the Lorentz-Einstein Theorem of Rel-
ativity.
To prove the theorem we shall choose a new unit of time by
putting ct = x0 . The co-efficients of the metrical groundform are
then
gik = 0 (i 6= k); gii = i ,
in which 0 = −1, 1 = 2 = 3 = +1; so that in passing from
components of a tensor that are co-variant with respect to an
index i to the contra-variant components of that tensor we have
only to multiply the ith component by the sign of i . The question
of continuity for electricity (10) assumes the desired invariant form
3
X ∂si
=0
i=0
∂xi

if we introduce s0 = ρ, and s1 , s2 , s3 , which are equal to the com-

ponents of s, as the four contra-variant components of a vector
in the above four-dimensional space, namely, of the “4-vector cur-
rent”. Parallel with this—as we see from (16) and (160 )—we must
combine
φ0 = φ and the components of f , namely, φ1 , φ2 , φ3 ,
 CHAPTER III 246

to make up the contra-variant components of a four-dimensional

vector, which we call the electromagnetic potential; of its co-
variant components, the 0th, i.e. φ0 = −φ, whereas the three
others φ1 , φ2 , φ3 are equal to the components of f . The equations
(14) and (15), by which the field-quantities B and E are derived
from the potentials, may then be written in the invariant form
∂φi ∂φk
− = Fik (18)
∂xk ∂xi
in which we set
E = (F10 , F20 , F30 ), B = (F23 , F31 , F12 ).
This is then how we may combine electric and magnetic intensity
of field to make up a single linear tensor of the second order F ,
the “field”. From (18) we get the invariant equations
∂Fkl ∂Fli ∂Fik
+ + = 0, (19)
∂xi ∂xk ∂xl
and this is Maxwell’s first system of equations (12). We took a
circuitous route in using Lorentz’s solution and the potentials only
so as to be led naturally to the proper combination of the three-
dimensional quantities, which converts them into four-dimensional
vectors and tensors. By passing over to contra-variant components
we get
E = (F 01 , F 02 , F 03 ), B = (F 23 , F 31 , F 12 ).
Maxwell’s second system, expressed invariantly in terms of four-
dimensional tensors, is now
X ∂F ik
= si . (20)
k
∂x k
 RELATIVITY OF SPACE AND TIME 247

If we now introduce the four-dimensional vector with the co-

variant components
pi = Fik sk (21)
(and the contra-variant components pi = F ik sk )—following our
previous practice of omitting the signs of summation—then p0 is
the “work-density,” that is, the work per unit of time and per unit
of volume: p0 = (s, E) [the unit of time is to be adapted to the
new measure of time x0 = ct], and p1 , p2 , p3 are the components
of the density of force.
This fully proves the Lorentz Theorem of Relativity. We no-
tice here that the laws that have been obtained are exactly the same
as those which hold in the stationary magnetic field (§ 9 (62)) ex-
cept that they have been transposed from three-dimensional to four-
dimensional space. There is no doubt that the real mathematical
harmony underlying these laws finds as complete an expression as
is possible in this formulation in terms of four-dimensional tensors.
Further, we learn from the above that, exactly as in the case
of three-dimensions, we may derive the “4-force” = pi from a sym-
metrical four-dimensional “stress-tensor” S, thus
∂Sik ∂S ik
−pi = or − pi = , (22)
∂xk ∂xk
Sik = Fir F kr − 12 δik |F |2 . (220 )
The square of the numerical value of the field (which is not nec-
essarily positive here) is
|F |2 = 21 Fik F ik .
We shall verify formula (22) by direct calculation. We have
∂Sik ∂F kr ∂Fir 1 kr ∂Fkr
= Fir + F kr − 2F .
∂xk ∂xk ∂xk ∂xi
 CHAPTER III 248

The first term on the right gives us

−Fir sr = −pi .

If we write the co-efficient of F kr skew-symmetrically we get for

the second term  
1 kr ∂Fir ∂Fik
2
F −
∂xk ∂xr
which, combined with the third, gives
 
1 kr ∂Fik ∂Fkr ∂Fri
−2F + + .
∂xr ∂xi ∂xk

The expression consisting of three terms in the brackets = 0,

by (19).
Now |F |2 = B2 − E2 . Let us examine what the individual
components of Sik signify, by separating the index 0 from the
others 1, 2, 3, in conformity with the partition into space and
time.
S 00 = the energy-density W = 21 (E2 + B2 ),
S 0i = the components of S = [E, B] i, k = (1, 2, 3),
S ik = the components of the Maxwell stress-tensor, which is
composed of the electrical and magnetic parts given in § 9. Ac-
cordingly the 0th equation of (22) expresses the law of energy. The
1st, 2nd, and 3rd have a fully analogous form. If, for a moment,
1
we denote the components of the vector S by G1 , G2 , G3 and
c
take t(i) to stand for the vector with the components S i1 , S i2 , S i3
we get
∂Gi
−pi = + div t(i) , (i = 1, 2, 3). (23)
∂t
 RELATIVITY OF SPACE AND TIME 249

The force which acts on the electrons enclosed in a portion of

space V produces an increase in time of momentum equal to it-
self numerically. This increase is balanced, according to (23), by
a corresponding decrease of the field-momentum distributed in
S
the field with a density , and the addition of field-momentum
c
from without. The current of the ith component of momentum
is given by t(i) , and thus the momentum-flux is nothing more
than the Maxwell stress-tensor. The Theorem of the Conservation
of Energy is only one component, the time-component, of a law
which is invariant for Lorentz transformations, the other compo-
nents being the space-components which express the conservation
of momentum. The total energy as well as the total momen-
tum remains unchanged: they merely stream from one part of
the field to another, and become transformed from field-energy
and field-momentum into kinetic-energy and kinetic-momentum
of matter, and vice versa. That is the simple physical meaning of
the formulæ (22). In accordance with it we shall in future refer
to the tensor S of the four-dimensional world as the energy-
momentum-tensor or, more briefly, as the energy-tensor. Its
1
symmetry tells us that the density of momentum = 2 times
c
the energy-flux. The field-momentum is thus very weak, but,
nevertheless, it has been possible to prove its existence by demon-
strating the pressure of light on a reflecting surface.
A Lorentz transformation is linear. Hence (again suppressing
one space co-ordinate in our graphical picture) we see that it is
tantamount to introducing a new affine co-ordinate system. Let
us consider how the fundamental vectors e00 , e01 , e02 of the new co-
ordinate system lie relatively to the original fundamental vectors
e0 , e1 , e2 , that is to the unit vectors in the direction of the x0
 CHAPTER III 250

(or t), x1 , x2 axes. Since, for

x = x0 e0 + x1 e1 + x2 e2 = x00 e00 + x01 e01 + x02 e02 ,

we must have

−x20 + x21 + x22 = −x02 02 02

 
0 + x1 + x2 = Q(x)

we get Q(e00 ) = −1. Accordingly, the vector e00 starting from O

(i.e. the t0 -axis) lies within the cone of light-propagation; the par-
allel planes t0 = const. lie so that they cut ellipses from the cone,
the middle points of which lie on the t0 -axis (see Fig. 7); the x01 -,
x02 -axis are in the direction of conjugate diameters of these ellip-
tical sections, so that the equation of each is

x02 02
1 + x2 = const.

As long as we retain the picture of a material æther, capable of

executing vibrations, we can see in Lorentz’s Theorem of Relativ-
ity only a remarkable property of mathematical transformations;
the relativity theorem of Galilei and Newton remains the truly
valid one. We are, however, confronted with the task of inter-
preting not only optical phenomena but all electrodynamics and
its laws as the result of a mechanics of the æther which satisfies
Galilei’s Theorem of Relativity. To achieve this we must bring the
field-quantities into definite relationship with the density and ve-
locity of the æther. Before the time of Maxwell’s electromagnetic
theory of light, attempts were made to do this for optical phe-
nomena; these efforts were partly, but never wholly, crowned with
success. This attempt was not carried on (vide note 3) in the case
of the more comprehensive domain into which Maxwell relegated
 RELATIVITY OF SPACE AND TIME 251

optical phenomena. On the contrary, the idea of a field exist-

ing in empty space and not requiring a medium to sustain
it gradually began to win ground. Indeed, even Faraday had ex-
pressed in unmistakable language that not the field should derive
its meaning through its association with matter, but, conversely,
rather that particles of matter are nothing more than singularities
of the field.

§ 21. Einstein’s Principle of Relativity

Let us for the present retain our conception of the æther. It
should be possible to determine the motion of a body, for example,
the earth, relative to the fixed or motionless æther. We are not
helped by aberration, for this only shows that this relative motion
changes in the course of a year. Let A1 , O, A2 be three fixed
points on the earth that share in its motion. Suppose them to
lie in a straight line along the direction of the earth’s motion and
to be equidistant, so that A1 O = OA2 = l, and let v be the
v
velocity of translation of the earth through the æther; let = q,
c
which we shall assume to be a very small quantity. A light-signal
l
emitted at O will reach A2 after a time has elapsed, and
c−v
l
A1 after a time . Unfortunately, this difference cannot be
c+v
demonstrated, as we have no signal that is more rapid than light
and that we could use to communicate the time to another place.∗
∗
It might occur to us to transmit time from one world-point to another by
carrying a clock that is marking time from one place to the other. In practice,
this process is not sufficiently accurate for our purpose. Theoretically, it is
by no means certain that this transmission is independent of the traversed
path. In fact, the theory of relativity proves that, on the contrary, they are
 CHAPTER III 252

We have recourse to Fizeau’s idea, and set up little mirrors at

A1 and A2 which reflect the light-ray back to O. If the light-
signal is emitted at the moment O, then the ray reflected from A2
will reach A after a time
l l 2lc
+ = 2
c−v c+v c − v2
whereas that reflected from A1 reaches O after a time
l l 2lc
+ = 2 .
c+v c−v c − v2
There is now no longer a difference in the times. Let us, however,
now assume a third point A which participates in the translational
motion through the æther, such that OA = l, but that OA makes
an angle θ with the direction of OA. In Fig. 8, O, O0 , O00 are the
successive positions of the point O at the time 0 at which the signal
is emitted, at the time t0 at which it is reflected from the mirror A
placed at A0 , and finally at the time t0 + t00 at which it again
reaches O, respectively. From the figure we get the proportion
OA0 : O00 A0 = OO0 : O00 O0 .
Consequently the two angles at A0 are equal to one another. The
reflecting mirror must be placed, just as when the system is at
rest, perpendicularly to the rigid connecting line OA, in order
that the light-ray may return to O. An elementary trigonometrical
calculation gives for the apparent rate of transmission in the
direction θ
2l c2 − v 2
= . (24)
t0 + t00
p
c2 − v 2 sin2 θ
dependent on one another; cf. § 22.
 RELATIVITY OF SPACE AND TIME 253

A′

ct′
ct′′

θ
′
O vt O vt′′ O ′′
′

Fig. 8.

It is thus dependent on the angle θ, which gives the direction of

transmission. Observations of the value of θ should enable us to
determine the direction and magnitude of v.
These observations were attempted in the celebrated
Michelson-Morley experiment (vide note 4). In this, two
mirrors A, A0 are rigidly fixed to O at distances l, l0 , the one
along the line of motion the other perpendicular to it. The whole
apparatus may be rotated about O. By means of a transparent
glass plate, one-half of which is silvered and which bisects the
right angle at O, a light-ray is split up into two halves, one of
which travels to A, the other to A0 . They are reflected at these
two points; and at O, owing to the partly silvered mirror, they
are again combined to a single composite ray. We take l and l0
approximately equal; then, owing to the difference in path given
by (24), namely,
2l 2l0
− ,
1 − q2
p
1 − q2
interference occurs. If the whole apparatus is now turned slowly
through 90◦ about O until A0 comes into the direction of motion,
 CHAPTER III 254

this difference of path becomes

2l 2l0
− .
1 − q2 1 − q2
p

Consequently, there is a shortening of the path by an amount

!
1 1
2(l + l0 ) 2
−p ∼ (l + l0 )q 2 .
1−q 1−q 2

This should express itself in a shift of the initial interference

A′

l′

A
l
Observer.

* Source of Light.

Fig. 9.

fringes. Although conditions were such that, numerically, even

only 1 per cent. of the displacement of the fringes expected by
Michelson could not have escaped detection, no trace of it was to
be found when the experiment was performed.
 RELATIVITY OF SPACE AND TIME 255

Lorentz (and Fitzgerald, independently) sought to explain this

strange result by the bold hypothesis that a rigid body in moving
relatively to the æther undergoes ap contraction in the direction of
the line of motion in the ratio 1 : 1 − q 2 . This would actually
account for the null result of the Michelson-Morley experiment.
p
For there, OA has in the first position the true length l 1 − q 2 ,
and OA0 the length l0 , whereas in the p second position OA has the
true length l but OA0 the length l0 · 1 − q 2 . The difference of
2(l − l0 )
path would, in each case, be p .
1 − q2
It was also found that, no matter into what direction a mirror
√ to O was turned, the same apparent velocity of trans-
rigidly fixed
mission c2 − v 2 was obtained for all directions; that is, that this
velocity did not depend on the direction θ, in the manner given
by (24). Nevertheless, theoretically, it still seemed possible to
demonstrate
√ the decrease of the velocity of transmission from c
2 2
to c − v . But if the æther shortens p the measuring rods in the
direction of motion in the ratio 1 : 1 − q 2 , it need only retard
clocks in the same ratio to hide this effect, too. In fact, not only
the Michelson-Morley experiment but a whole series of further ex-
periments designed to demonstrate that the earth’s motion has an
influence on combined mechanical and electromagnetic phenom-
ena, have led to a null result (vide note 5). Æther mechanics has
thus to account not only for Maxwell’s laws but also for this re-
markable interaction between matter and æther. It seems that
the æther has betaken itself to the land of the shades in a final
effort to elude the inquisitive search of the physicist!
The only reasonable answer that was given to the question as
to why a translation in the æther cannot be distinguished from
rest was that of Einstein, namely, that there is no æther ! (The
 CHAPTER III 256

æther has since the very beginning remained a vague hypothe-

sis and one, moreover, that has acted very poorly in the face of
facts.) The position is then this: for mechanics we get Galilei’s
Theorem of Relativity, for electrodynamics, Lorentz’s Theorem.
If this is really the case, they neutralise one another and thereby
define an absolute space of reference in which mechanical laws
have the Newtonian form, electrodynamical laws that given by
Maxwell. The difficulty of explaining the null result of the exper-
iments whose purpose was to distinguish translation from rest, is
overcome only by regarding one or other of these two principles
of relativity as being valid for all physical phenomena. That of
Galilei does not come into question for electrodynamics as this
would mean that, in Maxwell’s theory, those terms by which we
distinguish moving fields from stationary ones would not occur:
there would be no induction, no light, and no wireless telegra-
phy. On the other hand, even the contraction theory of Lorentz-
Fitzgerald suggests that Newton’s mechanics may be modified so
that it satisfies the Lorentz-Einstein Theorem  vof
2 Relativity, the
deviations that occur being only of the order ; they are then
c
easily within reach of observation for all velocities v of planets or
on the earth. The solution of Einstein (vide note 6), which at one
stroke overcomes all difficulties, is then this: the world is a four-
dimensional affine space whose metrical structure is determined
by a non-definite quadratic form

Q(x) = (x, x)

which has one negative and three positive dimensions. All physical
quantities are scalars and tensors of this four-dimensional world,
and all physical laws express invariant relations between them.
 RELATIVITY OF SPACE AND TIME 257

The simple concrete meaning of the form Q(x) is that a light-

signal which has been emitted at the world-point O arrives at all
−→
those and only those world-points A for which x = OA belongs to
the one of the two conical sheets defined by the equation Q(x) = 0
(cf. § 4). Hence that sheet (of the two cones) which “opens into
the future” namely, Q(x) ≤ 0 is distinguished objectively from
that which opens into the past. By introducing an appropriate
“normal” co-ordinate system consisting of the zero point O and
the fundamental vectors ei , we may bring Q(x) into the normal
form
−→ −→
(OA, OA) = −x20 + x21 + x22 + x23 ,
in which the xi ’s are the co-ordinates of A; in addition, the funda-
mental vector e0 is to belong to the cone opening into the future.
It is impossible to narrow down the selection from these
normal co-ordinate systems any farther: that is, none are
specially favoured; they are all equivalent. If we make use of a
particular one, then x0 must be regarded as the time; x1 , x2 , x3
as the Cartesian space co-ordinates; and all the ordinary expres-
sions referring to space and time are to be used in this system
of reference as usual. The adequate mathematical formulation of
Einstein’s discovery was first given by Minkowski (vide note 7):
to him we are indebted for the idea of four-dimensional world-
geometry, on which we based our argument from the outset.
How the null result of the Michelson-Morley experiment comes
about is now clear. For if the interactions of the cohesive forces of
matter as well as the transmission of light takes place according
to Einstein’s Principle of Relativity, measuring rods must behave
so that no difference between rest and translation can be discov-
ered by means of objective determinations. Seeing that Maxwell’s
 CHAPTER III 258

equations satisfy Einstein’s Principle of Relativity, as was recog-

nised even by Lorentz, we must indeed regard the Michelson-
Morley experiment as a proof that the mechanics of rigid bodies
must, strictly speaking, be in accordance not with that of Galilei’s
Principle of Relativity, but with that of Einstein.
It is clear that this is mathematically much simpler and more
intelligible than the former: world-geometry has been brought into
closer touch with Euclidean space-geometry through Einstein and
Minkowski. Moreover, as may easily be shown, Galilei’s princi-
ple is found to be a limiting case of Einstein’s world-geometry by
making c converge to ∞. The physical purport of this is that we
are to discard our belief in the objective meaning of simultaneity;
it was the great achievement of Einstein in the field of the theory
of knowledge that he banished this dogma from our minds, and this
is what leads us to rank his name with that of Copernicus. The
graphical picture given at the end of the preceding paragraph dis-
closes immediately that the planes x00 = const. no longer coincide
with the planes x0 = const. In consequence of the metrical struc-
ture of the world, which is based on Q(x), each plane x00 = const.
has a measure-determination such that the ellipse in which it in-
tersects the “light-cone,” is a circle, and that Euclidean geometry
holds for it. The point at which it is punctured by the x00 -axis is
the mid-point of the elliptical section. So the propagation of light
takes place in the “accented” system of reference, too, in concentric
circles.
We shall next endeavour to eradicate the difficulties that seem
to our intuition, our inner knowledge of space and time, to be
involved in the revolution caused by Einstein in the conception
of time. According to the ordinary view the following is true.
If I shoot bullets out with all possible velocities in all directions
 RELATIVITY OF SPACE AND TIME 259

from a point O, they will all reach world-points that are later
than O; I cannot shoot back into the past. Similarly, an event
which happens at O has an influence only on what happens at
later world-points, whereas “one can no longer undo” the past:
the extreme limit is reached by gravitation, acting according to
Newton’s law of attraction, as a result of which, for example, by
extending my arm, I at the identical moment produce an effect
on the planets, modifying their orbits ever so slightly. If we again
suppress a space-co-ordinate and use our graphical mode of rep-
resentation, then the absolute meaning of the plane t = 0 which
passes through O consists in the fact that it separates the “future”
world-points, which can be influenced by actions at O, from the
“past” world-points from which an effect may be conveyed to or
conferred on O. According to Einstein’s Principle of Relativity,
we get in place of the plane of separation t = 0 the light cone

x21 + x22 − c2 t2 = 0

(which degenerates to the above double plane when c = ∞). This

makes the position clear in this way. The direction of all bodies
projected from O must point into the forward-cone, opening into
the future (so also the direction of the world-line of my own body,
my “life-curve” if I happen to be at O). Events at O can influence
only happenings that occur at world-points that lie within this
forward-cone: the limits are marked out by the resulting propa-
gation of light into empty space.∗ If I happen to be at O, then
∗
The propagation of gravitational force must, of course, likewise take place
with the speed of light, according to Einstein’s Theory of Relativity. The law
for the gravitational potential must be modified in a manner analogous to
that by which electrostatic potential was modified in passing from statical to
moving fields.
 CHAPTER III 260

O divides my life-curve into past and future; no change is thereby

caused. As far as my relationship to the world is concerned, how-
ever, the forward-cone comprises all the world-points which are af-
fected by my active or passive doings at O, whereas all events that
are complete in the past, that can no longer be altered, lie exter-
nally to this cone. The sheet of the forward-cone separates
my active future from my active past. On the other hand,
the interior of the backward-cone includes all events in which I

Active future.

t=0

Passive past.

Fig. 10.

have participated (either actively or as an observer) or of which

I have received knowledge of some kind or other, for only such
events may have had an influence on me; outside this cone are
all occurrences that I may yet experience or would yet experience
if my life were everlasting and nothing were shrouded from my
 RELATIVITY OF SPACE AND TIME 261

gaze. The sheet of the backward-cone separates my pas-

sive past from my passive future. The sheet itself contains
everything on its surface that I see at this moment, or can see;
it is thus properly the picture of my external surroundings. In
the fact that we must in this way distinguish between active and
passive, present, and future, there lies the fundamental impor-
tance of Römer’s discovery of the finite velocity of light to which
Einstein’s Principle of Relativity first gave full expression. The
plane t = 0 passing through O in an allowable co-ordinate system
may be placed so that it cuts the light-cone Q(x) = 0 only at O
and thereby separates the cone of the active future from the cone
of the passive past.
For a body moving with uniform translation it is always pos-
sible to choose an allowable co-ordinate system (= normal co-
ordinate system) such that the body is at rest in it. The indi-
vidual parts of the body are then separated by definite distances
from one another, the straight lines connecting them make definite
angles with one another, and so forth, all of which may be cal-
culated by means of the formulæ of ordinary analytical geometry
from the space-co-ordinates x1 , x2 , x3 of the points under consid-
eration in the allowable co-ordinate system chosen. I shall term
them the static measures of the body (this defines, in particu-
lar, the static length of a measuring rod). If this body is a clock,
in which a periodical event occurs, there will be associated with
this period in the system of reference, in which the clock is at rest,
a definite time, determined by the increase of the co-ordinate x0
during a period; we shall call this the “proper time” of the clock.
If we push the body at one and the same moment at different
points, these points will begin to move, but as the effect can at
most be propagated with the velocity of light, the motion will only
 CHAPTER III 262

gradually be communicated to the whole body. As long as the ex-

panding spheres encircling each point of attack and travelling with
the velocity of light do not overlap, the parts surrounding these
points that are dragged along move independently of one another.
It is evident from this that, according to the theory of relativity,
there cannot be rigid bodies in the old sense; that is, no body ex-
ists which remains objectively always the same no matter to what
influences it has been subjected. How is it that in spite of this
we can use our measuring rods for carrying out measurements in
space? We shall use an analogy. If a gas that is in equilibrium in
a closed vessel is heated at various points by small flames and is
then removed adiabatically, it will at first pass through a series of
complicated stages, which will not satisfy the equilibrium laws of
thermodynamics. Finally, however, it will attain a new state of
equilibrium corresponding to the new quantity of energy it con-
tains, which is now greater owing to the heating. We require
of a rigid body that is to be used for purposes of measurement
(in particular, a linear measuring rod) that, after coming to
rest in an allowable system of reference, it shall always re-
main exactly the same as before, that is, that it shall have the
same static measures (or static length); and we require of a
clock that goes correctly that it shall always have the same
proper-time when it has come to rest (as a whole) in an
allowable system of reference. We may assume that the mea-
suring rods and clocks which we shall use satisfy this condition
to a sufficient degree of approximation. It is only when, in our
analogy, the gas is warmed sufficiently slowly (strictly speaking,
infinitely slowly) that it will pass through a series of thermody-
namic states of equilibrium; only when we move the measuring
rods and clocks steadily, without jerks, will they preserve their
 RELATIVITY OF SPACE AND TIME 263

static lengths and proper-times. The limits of acceleration within

which this assumption may be made without appreciable errors
arising are certainly very wide. Definite and exact statements
about this point can be made only when we have built up a dy-
namics based on physical and mechanical laws.
To get a clear picture of the Lorentz-Fitzgerald contraction
from the point of view of Einstein’s Theory of Relativity, we shall
imagine the following to take place in a plane. In an allowable
system of reference (co-ordinates t, x1 , x2 , one space-co-ordinate
being suppressed), to which the following space-time expressions
will be referred, there is at rest a plane sheet of paper (carrying
rectangular co-ordinates x1 , x2 marked on it), on which a closed
curve c is drawn. We have, besides, a circular plate carrying a
rigid clock-hand that rotates around its centre, so that its point
traces out the edge of the plate if it is rotated slowly, thus proving
that the edge is actually a circle. Let the plate now move along the
sheet of paper with uniform translation. If, at the same time, the
index rotates slowly, its point runs unceasingly along the edge of
the plate: in this sense the disc is circular during translation too.
Suppose the edge of the disc to coincide exactly with the curve c
at a definite moment. If we measure c by means of measuring rods
that are at rest, we find that c is not a circle but an ellipse. This
phenomenon is shown graphically in Fig. 11. We have added the
system of reference t0 , x01 , x02 with respect to which the disc is at
rest. Any plane t0 = const. intersects the light cone in this system
of reference in a circle “that exists for a single moment”. The
cylinder above it erected in the direction of the t0 -axis represents
a circle that is at rest in the accented system, and hence marks
off that part of the world which is passed over by our disc. The
section of this cylinder and the plane t = 0 is not a circle but an
 CHAPTER III 264

ellipse. The right-angled cylinder constructed on it in the direction

of the t-axis represents the constantly present curve traced on the
paper.
If we now inquire what physical laws are necessary to distin-
guish normal co-ordinate systems from all other co-ordinate sys-
tems (in Riemann’s sense), we learn that we require only Galilei’s
Principle of Relativity and the law of the propagation of light; by
means of light-signals and point-masses moving under no forces—
even if we have only small limits of velocity within which the
latter may move—we are in a position to fix a co-ordinate system
of this kind. To see this we shall next add a corollary to Galilei’s
Principle of Inertia. If a clock shares in the motion of the point-
mass moving under no forces, then its time-data are a measure of
the “proper-time” s of the motion. Galilei’s principle states that
the world-line of the point is a straight line; we elaborate this by
stating further that the moments of the motion characterised by

t t′
Light cone.

x2

O x1
Fig. 11.

s = 0, 1, 2, 3, . . . (or by any arithmetical series of values of s) rep-

 RELATIVITY OF SPACE AND TIME 265

resent equidistant points along the straight line. By introducing

the parameter of proper-time to distinguish the various stages of
the motion we get not only a line in the four-dimensional world
but also a “motion” in it (cf. the definition on page 155) and ac-
cording to Galilei this motion is a translation.
The world-points constitute a four-dimensional manifold; this
is perhaps the most certain fact of our empirical knowledge. We
shall call a system of four co-ordinates xi (i = 0, 1, 2, 3), which are
used to fix these points in a certain portion of the world, a lin-
ear co-ordinate system, if the motion of point-mass under no
forces and expressed in terms of the parameter s of the proper-time
be represented by formulæ in which the xi ’s are linear functions
of s. The fact that there are such co-ordinate systems is what
the law of inertia really asserts. After this condition of linearity,
all that is necessary to define the co-ordinate system fully is a
linear transformation. That is, if xi , x0i are the co-ordinates re-
spectively of one and the same world-point in two different linear
co-ordinate systems, then the x0i ’s a must be linear functions of
the x’s. By simultaneously interpreting the xi ’s as Cartesian co-
ordinates in a four-dimensional Euclidean space, the co-ordinate
system furnishes us with a representation of the world (or of the
portion of world in which the xi ’s exist) on a Euclidean space
of representation. We may, therefore, formulate our proposition
thus. A representation of two Euclidean spaces by one another
(or in other words a transformation from one Euclidean space to
another), such that straight lines become straight lines and a se-
ries of equidistant points become a series of equidistant points
is necessarily an affine transformation. Fig. 12 which represents
Möbius’ mesh-construction (vide note 8) may suffice to indicate
the proof to the reader. It is obvious that this mesh-system may
 CHAPTER III 266

be arranged so that the three directions of the straight lines com-

posing it may be derived from a given, arbitrarily thin, cone carry-
ing these directions on it; the above geometrical theorem remains

Fig. 12.

valid even if we only know that the straight lines whose directions
belong to this cone become straight lines again as a result of the
transformation.
Galilei’s Principle of Inertia is sufficient in itself to prove con-
clusively that the world is affine in character: it will not, how-
ever, allow us deduce any further result. The metrical ground-
form (x, x) of the world is now accounted for by the process of
light-propagation. A light-signal emitted from O arrives at the
−→
world-point A if, and only if, x = OA belongs to one of the
two conical sheets defined by (x, x) = 0. This determines the
quadratic form except for a constant factor; to fix the latter we
must choose an arbitrary unit-measure (cf. Appendix I).
 RELATIVITY OF SPACE AND TIME 267

§ 22. Relativistic Geometry, Kinematics, and Optics

We shall call a world-vector x space-like or time-like, accord-
ing as (x, x) is positive or negative. Time-like vectors are divided
into those that point into the future and those that point into
the past. We shall call the invariant
p
∆s = −(x, x) (25)

of a time-like vector x which points into the future its proper-

time. If we set
x = ∆s · e
then e, the direction of the time-like displacement, is a vector that
points into the future, and that satisfies the condition of normality
(e, e) = −1.
As in Galilean geometry, so in Einstein’s world-geometry we
must resolve the world into space and time by projection
in the direction of a time-like vector e pointing into the future
and normalised by the condition (e, e) = −1. The process of pro-
jection was discussed in detail in § 19. The fundamental formulæ
(3), (5), (50 ) that are set up must here be applied with e = −1.∗
World-points for which the vector connecting them is proportional
to e coincide at a space-point which we may mark by means of a
point-mass at rest, and which we may represent graphically by a
world-line (straight) parallel to e. The three-dimensional space Re
that is generated by the projection has a metrical character that
is Euclidean since, for every vector x∗ which is orthogonal to e,
∗
Here the units of space and time are chosen so that the velocity of
light in vacuo becomes equal to 1. To arrive at the ordinary units of the
c.g.s. systems, the equation of normality (e, e) = −1 must be replaced by
(e, e) = −c2 , and e must be taken equal to −c2 .
 CHAPTER III 268

that is, every vector x∗ that satisfies the condition (x∗ , e) = 0,

(x∗ , x∗ ) is a positive quantity (except in the case in which x∗ = 0;
cf. § 4). Every displacement x of the world may be split up ac-
cording to the formula
x = ∆t | x :
∆t is its duration (called “height” in § 19): x is the displacement
it produces in the space Re .
If e1 , e2 , e3 form a co-ordinate system in Re , then the world-
displacements e1 , e2 , e3 that are orthogonal to e = e0 , and that
produce the three given space-displacements, form in conjunction
with e0 a co-ordinate system, which belongs to Re , for the
world-points. It is normal if the three vectors ei in Re form a
Cartesian co-ordinate system. In every case the system of co-
efficients of the metrical groundform has, in it, the form
 
−1 0 0 0 
 
 0 g11 g12 g13 
 0 g21 g22 g23  .
 
 
 0 g31 g32 g33 

The proper time ∆s of a time-like vector x pointing into the

future (and for which x = ∆s · e) is equal to the duration of x
in the space of reference Re , in which x calls forth no spatial dis-
placement. In the sequel we shall have to contrast several ways of
splitting up quantities into terms of the vectors e, e0 , . . . ; e (with
or without an index) is always to denote a time-like world-vector
pointing into the future and satisfying the condition of normality
(e, e) = −1.
Let K be a body at rest in Re , K 0 a body at rest in Re0 .
K 0 moves with uniform translation in Re . If, by splitting up e0
 RELATIVITY OF SPACE AND TIME 269

into terms of e, we get in Re

e0 = h | hv (26)

then K 0 undergoes the space-displacement hv during the time (i.e.

with the duration) h in Re . Accordingly, v is the velocity of K 0
in Re or the relative velocity of K 0 with respect to K. Its
magnitude is determined by v 2 = (v , v ). By (3) we have

h = −(e0 , e); (27)

on the other hand, by (5)

1 = −(e0 , e0 ) = h2 − h2 (v , v ) = h2 (1 − v 2 ),

thus we get
1
h= √ . (28)
1 − v2
If, between two moments of K 0 ’s motion, it undergoes the world-
displacement ∆s·e0 , (26) shows that h·∆s = ∆t is the duration of
this displacement in Re . The proper time ∆s and the duration ∆t
of the displacement in Re are related by
√
∆s = ∆t 1 − v 2 . (29)

Since (27) is symmetrical in e and e0 , (28) teaches us that the

magnitude of the relative velocity of K 0 with respect to K
is equal to that of K with respect to K 0 . The vectorial
relative velocities cannot be compared with one another since
the one exists in the space Re , the other in the space Re0 .
 CHAPTER III 270

Let us consider a partition into three quantities e, e1 , e2 . Let

K1 , K2 be two bodies at rest in Re1 , Re2 respectively. Suppose we
have in Re
1
e1 = h1 | h1 v1 h1 = p ,
1 − v12
1
e2 = h2 | h2 v2 h2 = p .
1 − v22

Then 
−(e1 , e2 ) = h1 h2 1 − (v1 v2 ) .
Hence, if K1 and K2 have velocities v1 , v2 respectively in Re , with
numerical values v1 , v2 , then if these velocities v1 , v2 make an
angle θ with each other, and if v12 = v21 is the magnitude of the
velocity of K2 relatively to K1 (or vice versa), we find that the
formula
1 − v1 v2 cos θ 1
p
2
p
2
=p 2
(30)
1 − v1 1 − v2 1 − v12
holds: it shows how the relative velocity of two bodies is
determined from their given velocities. If, using hyperbolic
functions, we set v = tanh v for each of the values v of the velocity
(v being < 1), we get

cosh u1 cosh u2 − sinh u1 sinh u2 cos θ = cosh u12 .

This formula becomes the cosine theorem of spherical geome-

try if we replace the hyperbolic functions by their corresponding
trigonometrical functions; thus u12 is the side opposite the angle θ
in a triangle on the Bolyai-Lobatschefsky plane, the two remaining
 RELATIVITY OF SPACE AND TIME 271

A A

A′ O
O

O′
Fig. 13.

sides being u1 and u2 .

Analogous to the relationship (29) between time and proper-
time, there is one between length and statical-length. We shall
use Re as our space of reference. Let the individual point-masses
of the body at a definite moment be at the world-points O, A, . . . .
The space-points O, A, . . . at Re at which they are situated form
a figure in Re , on which we can confer duration, by making the
body leave behind it a copy of itself at the moment under consid-
eration in the space Re ; an example of this was presented in the
illustration given at the close of the preceding paragraph. If, on
the other hand, the world-points O, A, . . . are at the space-points
O 0 , A0 , . . . in the space Re in which K 0 is at rest, then O0 , A0 , . . .
constitute the statical shape of the body K 0 (cf. Fig. 13, in which
orthogonal world-distances are drawn perpendicularly). There is
a transformation that connects the part of Re , which receives the
imprint or copy, and the statical shape of the body in Re0 . This
 CHAPTER III 272

transformation transforms the points A, A0 into one another. It

is obviously affine (in fact, it is nothing more than an orthogonal
projection). Since the world-points O, A are simultaneous for
the partition into e, we have
−→ −→
OA = x = 0 | x in Re , and x = OA.

By formula (5)
−→2
OA = (x, x) = (x, x),
−−0→0 2
O A = (x, x) + (x, e0 )2 .

If, however, we determine (x, e0 ) in Re by (50 ) we get

(x, e0 ) = h(x, v ),

and hence
−−0→0 2 (x, v )2
O A = (x, x) + .
1 − v2
If we use a Cartesian co-ordinate system x1 , x2 , x3 in Re with O as
origin, and having its x1 -axis in the direction of the velocity v, then
if x1 , x2 , x3 are the co-ordinates of A, we have
−→2
OA = x21 + x22 + x23 ,
−−0→0 2 x21
OA = + x22 + x23 = x02 02 02
1 + x2 + x 3 ,
1 − v2

in the last term of which we have set

x1
x01 = √ , x02 = x2 , x03 = x3 . (31)
1−v 2
 RELATIVITY OF SPACE AND TIME 273

By assigning to every point in Re with co-ordinates (x1 , x2 , x3 ) the

point with co-ordinates (x01 , x02 , x03 ) as given by√(31), we effect a
dilatation of the imprinted copy in the ratio 1 : 1 − v 2 along the
direction of the body’s motion. Our formulæ assert that the copy
thereby assumes a shape congruent to that of the body when at
rest; this is the Lorentz-Fitzgerald contraction. In particular,
the volume V that the body K 0 occupies at a definite moment in
the space Re is connected to its statical volume V0 by the relation
√
V = V0 1 − v 2 .
Whenever we measure angles by optical means we determine
the angles formed by the light-rays for the system of reference in
which the (rigid) measuring instrument is at rest. Again, when our
eyes take the place of these instruments it is these angles that de-
termine the visual form of objects that lie within the field of vision.
To establish the relationship between geometry and the observa-
tion of geometrical magnitudes, we must therefore take optical
considerations into account. The solution of Maxwell’s equations
for light-rays in the æther as well as in a homogeneous medium,
which is at rest in an allowable reference system, is of a form
such that the component of the “phase” quantities (in complex
notation) are all
= const. e2πiΘ(P )
in which Θ = Θ(P ) is, with the omission of an additive constant,
the phase determined by the conditions set down; it is a function
of the world-point which here occurs as the argument. If the world
co-ordinates are transformed linearly in any way, the components
in the new co-ordinate system will again have the same form with
the same phase-function Θ. The phase is accordingly an invariant.
 CHAPTER III 274

For a plane wave it is a linear and (if we exclude absorbing media)

real function of the world-co-ordinates of P ; hence the phase-
difference at two arbitrary points Θ(B) − Θ(A) is a linear form of
−→
the arbitrary displacement x = AB, that is, a co-variant world-
vector. If we represent this by the corresponding displacement l
(we shall allude to it briefly as the light-ray l) then
Θ(B) − Θ(A) = (l, x).
If we split it up by means of the time-like vector e into space and
time and set
ν
l=ν| a (32)
q
so that the space-vector a in Re is of unit length
x = ∆t | x,
then the phase-difference is
 
(a, x)
ν − ∆t .
q
From this we see that ν signifies the frequency, q the velocity of
transmission, and a the direction of the light-ray in the space Re .
Maxwell’s equations tell us that in the æther the velocity of trans-
mission q = 1, or that
(l, l) = 0.
If we split the world up into space and time in two ways,
firstly by means of e, secondly by means of e0 , and distinguish
the magnitudes derived from the second process by accents we
immediately find as a result of the invariance of (l, l) the law
   
2 1 02 1
ν −1 =ν −1 . (33)
q2 q 02
 RELATIVITY OF SPACE AND TIME 275

If we fix our attention on two light-rays l1 , l2 with frequencies

ν1 , ν2 and velocities of transmission q1 , q2 then
 
a1 a2
(l1 , l2 ) = ν1 ν2 −1 .
q1 q2

If they make an angle ω to with one another, then

cos ω 0
   
cos ω 0 0
ν1 ν2 − 1 = ν1 ν2 −1 . (34)
q1 q2 q10 q20

For the æther, these equations become

ω ω0
q = q 0 (= 1), ν1 ν2 sin2 = ν10 ν20 sin2 . (35)
2 2
Finally, to get the relationship between the frequencies ν and ν 0
we assume a body that is at rest in Re0 ; let it have the velocity v
in the space Re , then, as before, we must set

e0 = h | hv in Re . (26)

From (26) and (32) it follows that

 
0 0 (a, v )
ν = −(l, e ) = νh 1 − .
q

Accordingly, if the direction of the light-ray in Re makes an angle θ

with the velocity of the body, then
v cos θ
0 1−
ν q
= √ . (36)
ν 1 − v2
 CHAPTER III 276

(36) is Doppler’s Principle. For example, since a sodium-molecule

which is at rest in an allowable system remains objectively the
same, this relationship (36) will exist between the frequency ν 0
of a sodium-molecule which is at rest and ν the frequency of a
sodium-molecule moving with a velocity ν, both frequencies be-
ing observed in a spectroscope which is at rest; θ is the angle
between the direction of motion of the molecule and the light-ray
which enters the spectroscope. If we substitute (36) in (33) we get
an equation between q and q 0 which enables us to calculate the
velocity of propagation q in a moving medium from the velocity of
propagation q 0 in the same medium at rest; for example, in water,
v now represents the rate of flow of the water; θ represents the
angle that the direction of flow of the water makes with the light-
rays. If we suppose these two directions to coincide, and then
neglect powers of v higher than the first (since v is in practice
very small compared with the velocity of light), we get
q = q 0 + v(1 − q 02 );
that is, not the whole of the velocity v of the medium is added
1
to the velocity of propagation, but only the fraction 1 − 2 (in
n
1
which n = 0 is the index of refraction of the medium). Fresnel’s
q
1
“convection-co-efficient” 1 − 2 was determined experimentally by
n
Fizeau long before the advent of the theory of relativity by mak-
ing two light-rays from the same source interfere, after one had
travelled through water which was at rest whilst the other had
travelled through water which was in motion (vide note 9). The
fact that the theory of relativity accounts for this remarkable re-
sult shows that it is valid for the optics and electrodynamics of
 RELATIVITY OF SPACE AND TIME 277

moving media (and also that in such cases the relativity princi-
ple, which is derived from that of Lorentz and Einstein by putting
q for c, does not hold; one might be tempted to believe this er-
roneously from the equation of wave-motion that holds in such
cases). We shall find the special form of (34) for the æther, in
which q = q 0 = 1 (cf. (35)), to be

ω (1 − v cos θ1 )(1 − v cos θ2 ) 2 ω 0

sin2 = sin .
2 1 − v2 2
If the reference-space Re happens to be the one on which the the-
ory of planets is commonly founded (and in which the centre of
mass of the solar system is at rest), and if the body in question is
the earth (on which an observing instrument is situated), v its ve-
locity in Re , ω the angle in Re that two rays which reach the solar
system from two infinitely distant stars make with one another,
θ1 , θ2 the angles which these rays make with the direction of mo-
tion of the earth in Re , then the angle ω 0 , at which the stars are
observed from the earth, is determined by the preceding equation.
We cannot, of course, measure ω, but we note the changes in ω 0
(the aberration) by taking account of the changes in θ1 and θ2
in the course of a year.
The formulæ which give the relationship between time, proper-
time, volume and statical volume are also valid in the case of non-
uniform motion. If dx is the infinitesimal displacement that a
moving point-mass experiences during an infinitesimal length of
time in the world, then

dx = ds · u, (u, u) = −1, ds > 0

give the proper-time ds and the world-direction u of this displace-

 CHAPTER III 278

ment. The integral

Z Z p
ds = −(dx, dx)

taken over a portion of the world-line is the proper-time that

elapses during this part of the motion: it is independent of the
manner in which the world has been split up into space and time
and, provided the motion is not too rapid, will be indicated by
a clock that is rigidly attached to the point-mass. If we use any
linear co-ordinates xi whatsoever in the world, and the proper-
time s as our parameters to represent our world-line analytically
(just as we use length of arc in three-dimensional geometry), then
dxi
= ui
ds
are the (contra-variant) components of u, and we get i ui ui = −1.
P
If we split up the world into space and time by means of e, we
find 
1 √ v

u= √ in Re
1 − v  1 − v2
2

in which v is the velocity of the mass-point; and we find that the

time dt that elapses during the displacement dx in Re and the
proper-time ds are connected by
√
ds = dt 1 − v 2 . (37)

If two world-points A, B are so placed with respect to one an-

−→
other that AB is a time-like vector pointing into the future, then
A and B may be connected by world-lines, whose directions all
likewise satisfy this condition: in other words, point-masses that
 RELATIVITY OF SPACE AND TIME 279

leave A may reach B. The proper-time necessary for them to do

this is dependent on the world-line; it is longest for a point-mass
that passes from A to B by uniform translation. For if we split
up the world into space and time in such a way that A and B
occupy the same point in space, this motion degenerates simply
to rest, and we derive the proposition (37) which states that the
proper-time lags behind the time t. The life-processes of mankind
may well be compared to a clock. Suppose we have two twin-
brothers who take leave from one another at a world-point A, and
suppose one remains at home (that is, permanently at rest in an
allowable reference-space), whilst the other sets out on voyages,
during which he moves with velocities (relative to “home”) that
approximate to that of light. When the wanderer returns home in
later years he will appear appreciably younger than the one who
stayed at home.
An element of mass dm (of a continuously extended body)
that moves with a velocity whose numerical value is v occupies
at a particular moment a volume dV which is connected with its
statical volume dV0 by the formula
√
dV = dV0 1 − v 2 .
dm
Accordingly, we have the relation between the density = µ
dV
dm
and the statical density = µ0 :
dV0
√
µ0 = µ 1 − v 2 .

µ0 is an invariant, and µ0 u with components µ0 ui is thus a contra-

variant vector, the “flux of matter,” which is determined by the
 CHAPTER III 280

motion of the mass independently of the co-ordinate system. It

satisfies the equation of continuity
X ∂(µ0 ui )
= 0.
i
∂xi

The same remarks apply to electricity. If it is associated with

matter so that de is the electric charge of the element of mass dm,
de
then the statical density ρ0 = is connected to the density
dV0
de
ρ= by
dV √
ρ0 = ρ 1 − v 2 ,
then
si = ρ0 ui
are the contra-variant components of the electric current (4-
vector); this corresponds exactly to the results of § 20. In
Maxwell’s phenomenological theory of electricity, the concealed
motions of the electrons are not taken into account as motions
of matter, consequently electricity is not supposed attached to
matter in his theory. The only way to explain how it is that a
piece of matter carries a certain charge is to say this charge is that
which is simultaneously in the portion of space that is occupied
by the matter at the moment under consideration. From this
we see that the charge is not, as in the theory of electrons, an
invariant determined by the portion of matter, but is dependent
on the way the world has been split up into space and time.
 RELATIVITY OF SPACE AND TIME 281

§ 23. The Electrodynamics of Moving Bodies

By splitting up the world into space and time we split up all
tensors. We shall first of all investigate purely mathematically
how this comes about, and shall then apply the results to derive
the fundamental equations of electrodynamics for moving bodies.
Let us take an n-dimensional metrical space, which we shall call
“world,” based on the metrical groundform (x, x). Let e be a
vector in it, for which (e, e) = e 6= 0. We split up the world
in the usual way into space Re and time in terms of e. Let e1 ,
e2 , . . . , en−1 be any co-ordinate system in the space Re , and let e1 ,
e2 , . . . , en−1 be the displacements of the world that are orthogonal
to e = e0 and that are produced in Re by e1 , e2 , . . . , en−1 . In the
co-ordinate system ei (i = 0, 1, 2, . . . , n − 1) “belonging to Re ” and
representing the world, the scheme of the co-variant components
of the metrical ground-tensor has the form
 
e 0 0 
 
0 g11 g12  (n = 3).
 
0 g21 g22 

As an example, we shall consider a tensor of the second order

and suppose it to have components Tik in this co-ordinate system.
Now, we assert that it splits up, in a manner dependent only on e,
according to the following scheme:

T00 T01 T02

T10 T11 T12
T20 T21 T22

that is, into a scalar, two vectors and a tensor of the second order
existing in Re , which are here characterised by their components
 CHAPTER III 282

in the co-ordinate system ei (i = 1, 2, . . . , n − 1).

For if the arbitrary world-displacement x splits up in terms
of e thus
x=ξ|x
and if, when we divide x into two factors, one of which is propor-
tional to e and the other orthogonal to e, we have

x = ξe + x∗

then, if x has components ξ i , we get

n−1
X n−1
X n−1
X
i 0 ∗ i
x= ξ ei , ξ=ξ , x = ξ ei , x= ξ i ei .
i=0 i=1 i=1

Thus, without using a co-ordinate system we may represent the

splitting up of a tensor in the following manner. If x, y are any
two arbitrary displacements of the world, and if we set

x = ξe + x∗ , y = ηe + y∗ , (38)

so that x∗ and y∗ are orthogonal to e, then the bilinear form

belonging to the tensor of the second order is

T (x, y) = ξηT (e, e) + ηT (x∗ , e) + ξT (e, y∗ ) + T (x∗ , y∗ ).

Hence, if we interpret x∗ , y∗ as the displacements of the world

orthogonal to e, which produce the two arbitrary displacements
x, y of the space, we get
1. a scalar T (e, e) = J = J,
2. two linear forms (vectors) in the space Re , defined by

L(x) = T (x∗ , e), L0 (x) = T (e, x∗ ),

 RELATIVITY OF SPACE AND TIME 283

3. a bilinear form (tensor) in the space Re , defined by

T (x, y ) = T (x∗ , y∗ ).

If x, y are arbitrary world-displacements that produce x, y , re-

spectively in Re we must replace x∗ , y∗ in this definition by x−ξe,
y − ηe in accordance with (38); in these,
1 1
ξ = (x, e), η = (y, e).
e e
If we now set

T (x, e) = L(x), T (e, x) = L0 (x),

we get
J J

0 0
L(x) = L(x) − (x, e), L (x) = L (x) − (x, e),  
e e (39)
1 1 0 J
T (x, y) = T (x, y) − (y, e)L(x) − (x, e)L (y) + 2 (x, e)(y, e).


e e e
The linear and bilinear forms (vectors and tensors) of Re on the
left may be represented by the world-vectors and world-tensors
on the right which are derived uniquely from them. In the above
representation by means of components, this amounts to the fol-
lowing: that, for example,
 
  0 0 0 
T11 T12   
T =   is represented by 0 T11 T12  .
T21 T22  
0 T21 T22 


It is immediately clear that in all calculations the tensors of space

may be replaced by the representative world-tensors. We shall,
 CHAPTER III 284

however, use this device only in the case when, if one space-tensor
is λ times another, the same is true of the representative world-
tensors.
If we base our calculations of components on an arbitrary
co-ordinate system, in which

e = (e0 , e1 , . . . , en−1 )

then the invariant is

J = Tik ei ek and e = ei ei .

But the two vectors and the tensor in Re have as their represen-
tatives in the world, according to (39), the two vectors and the
tensor with components:
J
L : Li − ei Li = Tik ek ,
e
J
L0 : L0i − ei L0i = Tki ek ;
e
ek Li + ei L0k J
T : Tik − + 2 ei ek .
e e
In the case of a skew-symmetrical tensor, J becomes = 0 and
L0 = −L; our formulæ degenerate into

L : Li = Tik ek ,
ei Lk − ek Li
T : Tik + .
e
A linear world-tensor of the second order splits up in space into a
vector and a linear space-tensor of the second order.
 RELATIVITY OF SPACE AND TIME 285

Maxwell’s field-equations for bodies at rest have been set out

in § 20. H. Hertz was the first to attempt to extend them so that
they might apply generally for moving bodies. Faraday’s Law of
Induction states that the time-decrement of the flux of induction
enclosed in a conductor is equal to the induced electromotive force,
that is
1 d
Z Z
− Bn do = E dr. (40)
c dt
The surface-integral on the left, if the conductor be in motion,
must be taken over a surface stretched out inside the conductor
and moving with it. Since Faraday’s Law of Induction has been
proved for just those cases in which the time-change of the flux of
induction within the conductor is brought about by the motion of
the conductor, Hertz did not doubt that this law was equally valid
for the case, too, when the conductor was in motion. The equation
div B = 0 remains unaffected. From vector analysis we know that,
taking this equation into consideration, the law of induction (40)
may be expressed in the differential form:
1 ∂B 1
curl E = − + curl[v, B] (41)
c ∂t c
∂B
in which denotes the differential co-efficient of B with respect
∂t
to the time for a fixed point in space, and v denotes the velocity
of the matter.
Remarkable inferences may be drawn from (41). As in Wilson’s
experiment (vide note 10), we suppose a homogeneous dielectric
between the two plates of a condenser, and assume that this di-
electric moves with a constant velocity of magnitude v between
these plates, which we shall take to be connected by means of a
 CHAPTER III 286

conducting wire. Suppose, further, that there is a homogeneous

magnetic field H parallel to the plates and perpendicular to v. We
shall imagine the dielectric separated from the plates of the con-
denser by a narrow empty space, whose thickness we shall assume
→ 0 in the limit. It then follows from (41) that, in the space be-
1
tween the plates, E − [v, B] is derivable from a potential; since
c
the latter must be zero at the plates which are connected by a
1
conducting wire it is easily seen that we must have E = [v, B].
c
µ
Hence a homogeneous electric field of intensity E = vH (in
c
which µ denotes permeability) arises which acts perpendicularly
to the plates. Consequently, a statical charge of surface-density
µ
vH ( = dielectric constant) must be called up on the plates.
c
If the dielectric is a gas, this effect should man-
ifest itself, no matter to what degree the gas
has been rarefied, since µ converges, not to-
wards 0, but towards 1, at infinite rarefaction.
This can have only one meaning if we are to
retain our belief in the æther, namely, that
the effect must occur if the æther between the
plates is moving relatively to the plates and to v
the æther outside them. To explain induction
we should, however, be compelled to assume
that the æther is dragged along by the con-
necting wire.∗ General observations, Fizeau’s Fig. 14.
experiment dealing with the propagation of
light in flowing water, and Wilson’s experiment itself, prove that
∗
In (41) v signified the velocity of the æther, not relative to the matter
but relative to what?
 RELATIVITY OF SPACE AND TIME 287

this assumption is incorrect. Just as in Fizeau’s experiment the

1
convection-co-efficient 1 − 2 appears, so in the present experi-
n
ment we observe only a change of magnitude
µ − 1
vH
c
which vanishes when µ = 1. This seems to be an inexplicable
contradiction to the phenomenon of induction in the moving con-
ductor.
The theory of relativity offers a full explanation of this. If, as
in § 20, we again set ct = x0 , and if we again build up a field F
out of E and B, and a skew-symmetrical tensor H of the second
order out of D and H, we have the field-equations

∂Fkl ∂Fli ∂Fik
+ + = 0, 

∂xi ∂xk ∂xl


X ∂H ik (42)
i 
= s .

k
∂xk 

These hold if we regard the Fik ’s as co-variant, the H ik ’s as contra-

variant components, in each case, of a tensor of the second order,
but the si ’s as the contra-variant components of a vector in the
four-dimensional world, since the latter are invariant in any arbi-
trary linear co-ordinate system. The laws of matter

D = E, B = µH, s = σE

signify, however, that if we split up the world into space and time
in such a way that matter is at rest, and if F splits up into E | B,
H into D | H, and s into ρ | s, then the above relations hold. If
 CHAPTER III 288

we now use any arbitrary co-ordinate system, and if the world-

direction of the matter has the components ui in it then, after our
explanations above, these facts assume the form

(a) Hi∗ = Fi∗ (43)

in which
Fi∗ = Fik uk and Hi∗ = Hik uk ;

Fik − (ui Fk∗ − uk Fi∗ ) = µ Hik − (ui Hk∗ − uk Hi∗ ) ;


(b) (44)

and (c) si + ui (sk uk ) = σFi∗ . (45)

This is the invariant form of these laws. For purposes of calcula-

tion it is convenient to replace (44) by the equations

Fkl ui + Fli uk + Fik ul = µ{Hkl ui + Hli uk + Hik ul } (46)

which are derived directly from them. Our manner of deriving

them makes it clear that they hold only for matter which is in
uniform translation. We may, however, consider them as being
valid also for a single body in uniform translation, if it is separated
by empty space from bodies moving with velocities differing from
its own.∗ Finally, they may also be considered to hold for matter
moving in any manner whatsoever, provided that its velocity does
∗
This is the essential point in most applications. By applying Maxwell’s
statical laws to a region composed, in each case, of a body K and the empty
space surrounding it and referred to the system of reference in which K is
at rest, we find no discrepancies occurring in empty space when we derive
results from different bodies moving relatively to one another, because the
principle of relativity holds for empty space.
 RELATIVITY OF SPACE AND TIME 289

not fluctuate too rapidly. After having obtained the invariant

form in this way, we may now split up the world in terms of any
arbitrary e. Suppose the measuring instruments that are used to
determine the ponderomotive effects of field to be at rest in Re .
We shall use a co-ordinate system belonging to Re and thus set

(F10 , F20 , F30 ) = (E1 , E2 , E3 ) = E,

(F23 , F31 , F12 ) = (B23 , B31 , B12 ) = B,
(H10 , H20 , H30 ) = (D1 , D2 , D3 ) = D,
(H23 , H31 , H12 ) = (H23 , H31 , H12 ) = H,
s0 = ρ; (s1 , s2 , s3 ) = (s 1 , s 2 , s 3 ) = s,
1 (v 1 , v 2 , v 3 ) v
u0 = √ (u1 , u2 , u3 ) = √ =√ ,
1 − v2 1−v 2 1 − v2
we hereby again arrive at Maxwell’s field-equations, which
are thus valid in a totally unchanged form, not only for
static, but also for moving matter. Does this not, however,
conflict violently with the observations of induction, which appear
to require the addition of a term as in (41)? No; for these observa-
tions do not really determine the intensity of field E, but only the
current which flows in the conductor; for moving bodies, however,
the connection between the two is given by a different equation,
namely, by (45).
If we write down those equations of (43), (45), which corre-
spond to the components with indices i = 1, 2, 3, and those of (46),
which correspond to

(i, k, l) = (2, 3, 0), (3, 1, 0), (1, 2, 0)

(the others are superfluous), the following results, as is easily seen,

 CHAPTER III 290

come about. If we set

E + [v, B] = E∗ , D+[v, H]= D∗ ,

B − [v, E] = B∗ , H−[v, D]= H∗ ,

then
D∗ = E∗ , B∗ = µH∗ .
If, in addition, we resolve s into the “convection-current” c and
the “conduction-current” s∗ , that is,

s = c + s∗ ,
ρ − (v, s)
c = ρ∗ v, ρ∗ = = ρ − (v, s∗ ),
1 − v2
then
∗ σE∗
s =√ .
1 − v2
Everything now becomes clear: the current is composed partly
of a convection-current which is due to the motion of charged
matter, and partly of a conduction-current, which is determined
by the conductivity σ of the substance. The conduction-current is
calculated from Ohm’s Law, if the electromotive force is defined
by the line-integral, not of E, but of E∗ . An equation exactly
analogous to (41) holds for E∗ , namely:
∂B
curl E∗ = − + curl[v, B] (we now always take c = 1)
∂t
or expressed in integrals, as in (40),
d
Z Z
− Bn do = E∗ dr.
dt
 RELATIVITY OF SPACE AND TIME 291

This explains fully Faraday’s phenomenon of induction in moving

conductors. For Wilson’s experiment, according to the present
theory, curl E = 0, that is, E will be zero between the plates. This
gives us the constant values of the individual vectors (of which the
electrical ones are perpendicular to the plates, whilst the magnetic
ones are directed parallel to the plates and perpendicular to the
velocity): these values are:

E ∗ = vB ∗ = vµH ∗ = µv(H + vD),

D = D∗ − vH = E ∗ − vH.

If we substitute the expression for E ∗ in the first equation, we get


D = v (µ − 1)H + µvD ,
µ − 1
D= vH.
1 − µv 2
This is the value of the superficial density of charge that is called
up on the condenser plates: it agrees with our observations since,
on account of v being very small, the denominator in our formula
differs very little from unity.
The boundary conditions at the boundary between the matter
and the æther are obtained from the consideration that the field-
magnitudes F and H must not suffer any sudden (discontinuous)
changes in moving along with the matter; but, in general, they
will undergo a sudden change, at some fixed space-point imagined
in the æther for the sake of clearness, at the instant at which
the matter passes over this point. If s is the proper-time of an
elementary particle of matter then
dFik ∂Fik l
= u
ds ∂xl
 CHAPTER III 292

must remain finite everywhere. If we set

 
∂Fik ∂Fkl ∂Fli
=− +
∂xl ∂xi ∂xk
we see that this expression
∂Fi∗ ∂Fk∗
= − .
∂xk ∂xi
Consequently, E∗ cannot have a surface-curl (and B cannot have
a surface-divergence).
The fundamental equations for moving bodies were deduced by
Lorentz from the theory of electrons in a form equivalent to the
above before the discovery of the principle of relativity. This is not
surprising, seeing that Maxwell’s fundamental laws for the æther
satisfy the principle of relativity, and that the theory of electrons
derives those governing the behaviour of matter by building up
mean values from these laws. Fizeau’s and Wilson’s experiments
and another analogous one, that of Röntgen and Eichwald (vide
note 11), prove that the electromagnetic behaviour of matter is
in accordance with the principle of relativity; the problems of the
electrodynamics of moving bodies first led Einstein to enunciate
it. We are indebted to Minkowski for recognising clearly that the
fundamental equations for moving bodies are determined uniquely
by the principle of relativity if Maxwell’s theory for matter at rest
is taken for granted. He it was, also, who formulated it in its final
form (vide note 12).
Our next aim will be to subjugate mechanics, which does
not obey the principle in its classical form, to the principle of
relativity of Einstein, and to inquire whether the modifications
that the latter demands can be made to harmonise with the facts
of experiment.
 RELATIVITY OF SPACE AND TIME 293

§ 24. Mechanics according to the Principle of Relativity

On the theory of electrons we found the mechanical effect of
the electromagnetic field to depend on a vector p whose contra-
variant components are

pi = F ik sk = ρ0 F ik uk .

It therefore satisfies the equation

pi ui = (p, u) = 0 (47)

in which u is the world-direction of the matter. If we split up p

and u in any way into space and time thus
)
u = h | hv ,
(48)
p = λ | p,

we get p as the force-density and, as we see from (47) or from


h λ − (p, v ) = 0

that λ is the work-density.

We arrive at the fundamental law of the mechanics which
agrees with Einstein’s Principle of Relativity by the same method
as that by which we obtain the fundamental equations of elec-
tromagnetics. We assume that Newton’s Law remains valid in
the system of reference in which the matter is at rest. We fix
our attention on the point-mass m, which is situated at a definite
world-point O and split up our quantities in terms of its world-
direction u into space and time. m is momentarily at rest in Ru .
Let µ0 be the density in Ru of the matter at the point O. Sup-
pose that, after an infinitesimal element of time ds has elapsed,
 CHAPTER III 294

m has the world-direction u + du. It follows from (u, u) = −1

that (u · du) = 0. Hence, splitting up with respect to u, we get

u = 1 | O, du = 0 | dv , p = 0 | p.

It follows from
u + du = 1 | dv
that dv is the relative velocity acquired by m (in Ru ) during the
time ds. Thus there can be no doubt that the fundamental law of
mechanics is
dv
µ0 = p.
ds
From this we derive at once the invariant form
du
µ0 = p, (49)
ds
which is quite independent of the manner of splitting up. In it,
µ0 is the statical density, that is, the density of the mass when at
rest; ds is the proper-time that elapses during the infinitesimal
displacement of the particle of matter, during which its world-
direction increases by du.
Resolution into terms of u is a partition which would alter
during the motion of the particle of matter. If we now split up
our quantities, however, into space and time by means of some
fixed time-like vector e that points into the future and satisfies
the condition of normality (e, e) = −1, then, by (48), (49) resolves
into  
d 1

µ0 √ = λ,

ds 1−v 2 
  (50)
d v
√

µ0 = p.

ds 1 − v2
 RELATIVITY OF SPACE AND TIME 295

If, in this partition or resolution, t denotes the time, dV the vol-

ume, and dV0 the static volume of the particle of matter at a
definite moment, its mass, however, being m = µ0 dV0 , and if

p dv = P, λ dV = L

denotes the force acting on the particle and its work, respectively,
then if we multiply our equations by dV and take into account
that
d √ d d
µ0 dV · = m 1 − v2 · =m·
ds ds dt
and that the mass m remains constant during the motion, we get
finally
 
d m
√ = L, (51)
dt 1 − v2
 
d mv
√ = P. (52)
dt 1 − v2
These are the equations for the mechanics of the point-mass. The
equation of momentum (52) differs from that of Newton only in
that the (kinetic) momentum of the point-mass is not mv but
mv
=√ . The equation of energy (51) seems strange at first: if
1 − v2
we expand it into powers of v, we get

m mv 2
√ =m+ + ...,
1 − v2 2

so that if we neglect higher powers of v and also the constant m

we find that the expression for the kinetic energy degenerates into
the one given by classical mechanics.
 CHAPTER III 296

This shows that the deviations from the mechanics of Newton

are, as we suspected, of only the second order of magnitude in
the velocity of the point-masses as compared with the velocity of
light. Consequently, in the case of the small velocities with which
we usually deal in mechanics, no difference can be demonstrated
experimentally. It will become perceptible only for velocities that
approximate to that of light; in such cases the inertial resistance
of matter against the accelerating force will increase to such an
extent that the possibility of actually reaching the velocity of light
is excluded. Cathode rays and the β-radiations emitted by ra-
dioactive substances have made us familiar with free negative elec-
trons whose velocity is comparable to that of light. Experiments
by Kaufmann, Bucherer, Ratnowsky, Hupka, and others, have
shown in actual fact that the longitudinal acceleration caused in
the electrons by an electric field or the transverse acceleration
caused by a magnetic field is just that which is demanded by the
theory of relativity. A further confirmation based on the motion
of the electrons circulating in the atom has been found recently in
the fine structure of the spectral lines emitted by the atom (vide
note 13). Only when we have added to the fundamental equa-
tions of the electron theory, which, in § 20, was brought into an
invariant form agreeing with the principle of relativity, the equa-
tion si = ρ0 ui , namely, the assertion that electricity is associated
with matter, and also the fundamental equations of mechanics,
do we get a complete cycle of connected laws, in which a state-
ment of the actual unfolding of natural phenomena is contained,
independent of all conventions of notation. Now that this final
stage has been carried out, we may at last claim to have proved
the validity of the principle of relativity for a certain region, that
of electromagnetic phenomena.
 RELATIVITY OF SPACE AND TIME 297

In the electromagnetic field the ponderomotive vector pi is

derived from a tensor Sik , dependent only on the local values of
the phase-quantities, by the formulæ:

∂Sik
pi = − .
∂xk
In accordance with the universal meaning ascribed to the con-
ception energy in physics, we must assume that this holds not
only for the electromagnetic field but for every region of physical
phenomena, and that it is expedient to regard this tensor instead
of the ponderomotive force as the primary quantity. Our pur-
pose is to discover for every region of phenomena in what manner
the energy-momentum-tensor (whose components Sik must always
satisfy the condition of symmetry) depends on the characteristic
field- or phase-quantities. The left-hand side of the mechanical
equations
dui
µ0 = pi
ds
may be reduced directly to terms of a “kinetic” energy-momentum-
tensor thus:
Uik = µ0 ui uk .
For
∂Uik ∂(µ0 uk ) ∂ui
= ui + µ0 uk .
∂xk ∂xk ∂xk
The first term on the right = 0, on account of the equation of
dui
continuity for matter; the second = µ0 because
ds
∂ui ∂ui ∂xk dui
uk = = .
∂xk ∂xk ∂s ds
 CHAPTER III 298

Accordingly, the equations of mechanics assert that the complete

energy-momentum-tensor Tik = Uik + Sik composed of the kinetic
tensor U and the potential tensor S satisfies the theorems of con-
servation
∂Tik
= 0.
∂xk
The Principle of the Conservation of Energy is here expressed in
its clearest form. But, according to the theory of relativity, it
is indissolubly connected with the principle of the conservation
of momentum and the conception momentum (or impulse)
must claim just as universal a significance as that of en-
ergy. If we express the kinetic tensor at a world-point in terms of
a normal co-ordinate system such that, relatively to it, the matter
itself is momentarily at rest, its components assume a particularly
simple form, namely, U00 = µ0 (or = c2 µ0 , if we use the c.g.s. sys-
tem, in which c is not = 1), and all the remaining components
vanish. This suggests the idea that mass is to be regarded as
concentrated potential energy that moves on through space.

§ 25. Mass and Energy

To interpret the idea expressed in the preceding sentence we
shall take up the thread by returning to the consideration of the
motion of the electron. So far, we have imagined that we have to
write for the force P in its equation of motion (52) the following:


P = e E + [v, H] (e = charge of the electron)

that is, that P is composed of the impressed electric and magnetic

fields E and H. Actually, however, the electron is subject not
only to the influence of these external fields during its motion
 RELATIVITY OF SPACE AND TIME 299

but also to the accompanying field which it itself generates. A

difficulty arises, however, in the circumstance that we do not know
the constitution of the electron, and that we do not know the
nature and laws of the cohesive pressure that keeps the electron
together against the enormous centrifugal forces of the negative
charge compressed in it. In any case the electron at rest and
its electric field (which we consider as part of it) is a physical
system, which is in a state of statical equilibrium—and that is
the essential point. Let us choose a normal co-ordinate system in
which the electron is at rest. Suppose its energy-tensor to have
components tik . The fact that the electron is at rest is expressed
by the vanishing of the energy-flux of whose components are t0i
(i = 1, 2, 3). The 0th condition of equilibrium
∂tki
=0 (53)
∂xk
then tells us that the energy-density t00 is independent of the
time x0 . On account of symmetry the components ti0 (i = 1, 2, 3)
of the momentum-density each also vanish. If t(1) is the vec-
tor whose components are t11 , t12 , t13 , the condition for equilib-
rium (53), (i = 1), gives
div t(1) = 0.
Hence we have, for example,
div(x2 t(1) ) = x2 div t(1) + t12 = t12
and since the integral of a divergence is zero (we may assume that
the t’s vanish at infinity at least as far as to the fourth order) we
get Z
t12 dx1 dx2 dx3 = 0.
 CHAPTER III 300

In the same way we find that, although Zthe tik ’s (for i, k = 1, 2, 3)

do not vanish, their volume integrals tik dV0 do so. We may
regard these circumstances as existing for every system in statical
equilibrium. The result obtained may be expressed by invariant
formulæ for the case of any arbitrary co-ordinate system thus:
Z
tik dV0 = E0 ui uk (i, k = 0, 1, 2, 3). (54)

E0 is the energy-content (measured in the space of reference for

which the electron is at rest), ui are the co-variant components
of the world-direction of the electron, and dV0 the statical volume
of an element of space (calculated on the supposition that the
whole of space participates in the motion of the electron). (54) is
rigorously true for uniform translation. We may also apply the
formula in the case of non-uniform motion if u does not change
too suddenly in space or in time. The components

i ∂tik
p̄ = −
∂xk
of the ponderomotive effect, exerted on the electron by itself, are
however, then no longer = 0.
If we assume the electron to be entirely without mass, and if
pi is the “4-force” acting from without, then equilibrium demands
that
p̄i + pi = 0. (55)
We split up u and p into space and time in terms of a fixed e,
getting
u = h | hv , p = (pi ) = λ | p
 RELATIVITY OF SPACE AND TIME 301

√
and we integrate (55) with respect to the volume dV = dV0 1 − v 2 .
Since, if we use a normal co-ordinate system corresponding to Re ,
we have
d
Z Z Z
i i
p̄ dV = p̄ dx1 dx2 dx3 = − ti0 dx1 dx2 dx3
dx0
d √ d
=− (E0 u0 ui 1 − v 2 ) = − (E0 ui )
dx0 dt
(in which x0 = t, the time), we get
   Z 
d E0
√ = L = λ dV ,
dt 1 − v2
   Z 
d E0 v
√ = P = p dV .
dt 1 − v2
These equations hold if the force P acting from without is not too
E0
great compared with , a being the radius of the electron, and
a
if its density in the neighbourhood of the electron is practically
constant. They agree exactly with the fundamental equations of
mechanics if the mass m is replaced by E. In other words, iner-
tia is a property of energy. In mechanics we ascribe to every
material body an invariable mass m which, in consequence of the
manner in which it occurs in the fundamental law of mechanics,
represents the inertia of matter, that is, its resistance to the ac-
celerating forces. Mechanics accepts this inertial mass as given
and as requiring no further explanation. We now recognise that
the potential energy contained in material bodies is the cause of
this inertia, and that the value of the mass corresponding to the
energy E0 expressed in the c.g.s. system, in which the velocity of
 CHAPTER III 302

light is not unity, is

E0
m= . (56)
c2
We have thus attained a new, purely dynamical view of mat-
ter.∗ Just as the theory of relativity has taught us to reject the
belief that we can recognise one and the same point in space at dif-
ferent times, so now we see that there is no longer a mean-
ing in speaking of the same position of matter at different
times. The electron, which was formerly regarded as a body of
foreign substance in the non-material electromagnetic field, now
no longer seems to us a very small region marked off distinctly
from the field, but to be such that, for it, the field-quantities
and the electrical densities assume enormously high values. An
“energy-knot” of this type propagates itself in empty space in a
manner no different from that in which a water-wave advances over
the surface of the sea; there is no “one and the same substance” of
which the electron is composed at all times. There is only a po-
tential; and no kinetic energy-momentum-tensor becomes added
to it. The resolution into these two, which occurs in mechanics,
is only the separation of the thinly distributed energy in the field
from that concentrated in the energy-knots, electrons and atoms;
the boundary between the two is quite indeterminate. The theory
of fields has to explain why the field is granular in structure and
why these energy-knots preserve themselves permanently from en-
ergy and momentum in their passage to and fro (although they
do not remain fully unchanged, they retain their identity to an
extraordinary degree of accuracy); therein lies the problem of
∗
Even Kant in his Metaphysischen Anfangsgründen der Naturwis-
senschaft, teaches the doctrine that matter fills space not by its mere existence
but in virtue of the repulsive forces of all its parts.
 RELATIVITY OF SPACE AND TIME 303

matter. The theory of Maxwell and Lorentz is incapable of solv-

ing it for the primary reason that the force of cohesion holding the
electron together is wanting in it. What is commonly called
matter is by its very nature atomic; for we do not usually
call diffusely distributed energy matter. Atoms and electrons
are not, of course, ultimate invariable elements, which natu-
ral forces attack from without, pushing them hither and thither,
but they are themselves distributed continuously and subject to
minute changes of a fluid character in their smallest parts. It
is not the field that requires matter as its carrier in order to be
able to exist itself, but matter is, on the contrary, an offspring
of the field. The formulæ that express the components of the
energy-tensor Tik in terms of phase-quantities of the field tell us
the laws according to which the field is associated with energy
and momentum, that is, with matter. Since there is no sharp line
of demarcation between diffuse field-energy and that of electrons
and atoms, we must broaden our conception of matter, if it is still
to retain an exact meaning. In future we shall assign the term
matter to that real thing, which is represented by the energy-
momentum-tensor. In this sense, the optical field, for example, is
also associated with matter. Just as in this way matter is merged
into the field, so mechanics is expanded into physics. For the law
of conservation of matter, the fundamental law of mechanics

∂Tik
= 0, (57)
∂xk
in which the Tik ’s are expressed in terms of the field-quantities,
represents a differential relationship between these quantities, and
must therefore follow from the field-equations. In the wide sense,
in which we now use the word, matter is that of which we take
 CHAPTER III 304

cognisance directly through our senses. If I seize hold of a piece of

ice, I experience the energy-flux flowing between the ice and my
body as warmth, and the momentum-flux as pressure. The energy-
flux of light on the surface of the epithelium of my eye determines
the optical sensations that I experience. Hidden behind the matter
thus revealed directly to our organs of sense there is, however, the
field. To discover the laws governing the latter itself and also
the laws by which it determines matter we have a first brilliant
beginning in Maxwell’s Theory, but this is not our final destination
in the quest of knowledge.∗
To account for the inertia of matter we must, according to for-
mula (56), ascribe a very considerable amount of energy-content
to it: one kilogram of water is to contain 9·1023 ergs. A small por-
tion of this energy is energy of cohesion, that keeps the molecules
or atoms associated together in the body. Another portion is the
chemical energy that binds the atoms together in the molecule and
the sudden liberation of which we observe in an explosion (in solid
bodies this chemical energy cannot be distinguished from the en-
ergy of cohesion). Changes in the chemical constitution of bodies
or in the grouping of atoms or electrons involve the energies due
to the electric forces that bind together the negatively charged
electrons and the positive nucleus; all ionisation phenomena are
included in this category. The energy of the composite atomic
nucleus, of which a part is set free during radioactive disinte-
gration, far exceeds the amounts mentioned above. The greater
part of this, again, consists of the intrinsic energy of the elements
of the atomic nucleus and of the electrons. We know of it only
through inertial effects as we have hitherto—owing to a merciful
∗
Later we shall once again modify our views of matter; the idea of the
existence of substance has, however, been finally quashed.
 RELATIVITY OF SPACE AND TIME 305

Providence—not discovered a means of bringing it to “explosion”.

Inertial mass varies with the contained energy. If a body
is heated, its inertial mass increases; if it is cooled, it decreases;
this effect is, of course, too small to be observed directly.
The foregoing treatment of systems in statical equilibrium, in
which we have in general followed Laue,∗ was applied to the elec-
tron with special assumptions concerning its constitution, even
before Einstein’s discovery of the principle of relativity. The elec-
tron was assumed to be a sphere with a uniform charge either
on its surface or distributed evenly throughout its volume, and
held together by a cohesive pressure composed of forces equal
in all directions and directed towards the centre. The resultant
E0
“electromagnetic mass” 2 agrees numerically with the results of
c
observation, if one ascribes a radius of the order of magnitude
10−13 cms. to the electron. There is no cause for surprise at the
fact that even before the advent of the theory of relativity this
interpretation of electronic inertia was possible; for, in treating
electrodynamics after the manner of Maxwell, one was already
unconsciously treading in the steps of the principle of relativity as
far as this branch of phenomena is concerned. We are indebted to
Einstein and Planck, above all, for the enunciation of the inertia
of energy (vide note 15). Planck, in his development of dynamics,
started from a “test body” which, contrary to the electron, was
fully known although it was not in the ordinary sense material,
namely, cavity-radiation in thermodynamical equilibrium, as pro-
duced according to Kirchoff’s Law, in every cavity enclosed by
walls at the same uniform temperature.
In the phenomenological theories in which the atomic structure
∗
Vide note 14.
 CHAPTER III 306

of matter is disregarded we imagine the energy that is stored up

in the electrons, atoms, etc., to be distributed uniformly over the
bodies. We need take it into consideration only by introducing the
statical density of mass µ as the density of energy in the energy-
momentum-tensor—referred to a co-ordinate system in which the
matter is at rest. Thus, if in hydrodynamics we limit ourselves to
adiabatic phenomena, we must set
 
−µ0 0 0 0
 
k
 0 p 0 0
|Ti | =  
 0 0 p 0

 0 0 0 p

in which p is the homogeneous pressure; the energy-flux is zero in

adiabatic phenomena. To enable us to write down the components
of this tensor in any arbitrary co-ordinate system, we must set
µ0 = µ∗ − p, in addition. We then get the invariant equations

Tik = µ∗ ui uk + pδik ,
or Tik = µ∗ ui uk + p · gik . (58)

The statical density of mass is

Tik ui uk = µ∗ − p = µ0

and hence we must put µ0 , and not µ∗ , equal to a constant in

the case of incompressible fluids. If no forces act on the fluid, the
hydrodynamical equations become

∂Tik
= 0.
∂xk
 RELATIVITY OF SPACE AND TIME 307

Just as is here done for hydrodynamics so we may find a form

for the theory of elasticity based on the principle of relativity
(vide note 16). There still remains the task of making the law
of gravitation, which, in Newton’s form, is entirely bound to the
principle of relativity of Newton and Galilei, conform to that of
Einstein. This, however, involves special problems of its own to
which we shall return in the last chapter.

§ 26. Mie’s Theory

The theory of Maxwell and Lorentz cannot hold for the interior
of the electron; therefore, from the point of view of the ordinary
theory of electrons we must treat the electron as something given
a priori, as a foreign body in the field. A more general theory
of electrodynamics has been proposed by Mie, by which it seems
possible to derive the matter from the field (vide note 17). We
shall sketch its outlines briefly here—as an example of a physical
theory fully conforming with the new ideas of matter, and one
that will be of good service later. It will give us an opportunity
of formulating the problem of matter a little more clearly.
We shall retain the view that the following phase-quantities are
of account: (1) the four-dimensional current-vector s, the “elec-
tricity”; (2) the linear tensor of the second order F , the “field”.
Their properties are expressed in the equations

∂si
(1) = 0,
∂xi
∂Fkl ∂Fli ∂Fik
(2) + + = 0.
∂xi ∂xk ∂xl
Equations (2) hold if F is derivable from a vector φi according to
 CHAPTER III 308

the formulæ
∂φi ∂φk
(3) Fik = − .
∂xk ∂xi
Conversely, it follows from (2) that a vector φ must exist such that
equations (3) hold. In the same way (1) is fulfilled if s is derivable
from a skew-symmetrical tensor H of the second order according
to
∂H ik
(4) si = .
∂xk
Conversely, it follows from (1) that a tensor H satisfying these
conditions must exist. Lorentz assumed generally, not only for
the æther, but also for the domain of electrons, that H = F .
Following Mie, we shall make the more general assumption that
H is not a mere number of calculation but has a real significance,
and that its components are, therefore, universal functions of the
primary phase-quantities s and F . To be logical we must then
make the same assumption about φ. The resultant scheme of
quantities
φ F
s H
contains the quantities of intensity in the first row; they are con-
nected with one another by the differential equations (3). In the
second row we have the quantities of magnitude, for which the
differential quantities (4) hold. If we perform the resolution into
space and time and use the same terms as in § 20 we arrive at the
well-known equations
dρ
(1) + div s = 0,
dt
dB
(2) + curl E = 0 (div B = 0),
dt
 RELATIVITY OF SPACE AND TIME 309

df
(3) + grad φ = E (− curl f = B),
dt
dD
(4) − curl H = −s (div D = ρ).
dt
If we know the universal functions, which express φ and H in terms
of s and F , then, excluding the equations in brackets, and counting
each component separately, we have ten “principal equations” be-
fore us, in which the derivatives of the ten phase-quantities with
respect to the time are expressed in relation to themselves and
their spatial derivatives; that is, we have physical laws in the form
that is demanded by the principle of causality. The principle
of relativity that here appears as an antithesis, in a certain sense,
to the principle of causality, demands that the principal equations
be accompanied by the bracketed “subsidiary equations,” in which
no time derivatives occur. The conflict is avoided by noticing that
the subsidiary equations are superfluous. For it follows from the
principal equations (2) and (3) that
∂
(B + curl f ) = 0,
∂t
and from (1) and (4) that
∂ρ ∂
= (div D).
∂t ∂t
It is instructive to compare Mie’s Theory with Lorentz’s funda-
mental equations of the theory of electrons. In the latter, (1), (2),
and (4) occur, whilst the law by which H is determined from the
primary phase-quantities is simply expressed by D = E , H = B.
On the other hand, in Mie’s theory, φ and f are defined in (3)
as the result of a process of calculation, and there is no law that
 CHAPTER III 310

determines how these potentials depend on the phase-quantities

of the field and on the electricity. In place of this we find the
formula giving the density of the mechanical force and the law of
mechanics, which governs the motion of electrons under the in-
fluence of this force. Since, however, according to the new view
which we have put forward, the mechanical law must follow from
the field-equations, an addendum becomes necessary; for this pur-
pose, Mie makes the assumption that φ and f acquire a physical
meaning in the sense indicated. We may, however, enunciate Mie’s
equation (3) in a form fully analogous to that of the fundamental
law of mechanics. We contrast the ponderomotive force occurring
in it with the “electrical force” E in this case. In the statical case
(3) states that
E − grad φ = 0, (59)
that is, the electric force E is counterbalanced in the æther by an
“electrical pressure” φ. In general, however, a resulting electri-
cal force arises which, by (3), now belongs to the magnitude f as
the “electrical momentum”. It inspires us with wonder to see
how, in Mie’s Theory, the fundamental equation of electrostat-
ics (59) which stands at the commencement of electrical theory,
suddenly acquires a much more vivid meaning by the appearance
of potential as an electrical pressure; this is the required cohesive
pressure that keeps the electron together.
The foregoing presents only an empty scheme that has to be
filled in by the yet unknown universal functions that connect the
quantities of magnitude with those of intensity. Up to a certain
degree they may be determined purely speculatively by means of
the postulate that the theorem of conservation (57) must hold for
the energy-momentum-tensor Tik (that is, that the principle of
energy must be valid). For this is certainly a necessary condition,
 RELATIVITY OF SPACE AND TIME 311

if we are to arrive at some relationship with experiment at all.

The energy-law must be of the form
∂W
+ div s = 0
∂t
in which W is the density of energy, and s the energy-flux. We
get at Maxwell’s Theory by multiplying (2) by H and (4) by E ,
and then adding, which gives
∂B ∂D
H +E + div[E , H] = −(E , s). (60)
∂t ∂t
In this relation (60) we have also on the right, the work, which is
used in increasing the kinetic energy of the electrons or, according
to our present view, in increasing the potential energy of the field
of electrons. Hence this term must also be composed of a term
differentiated with respect to the time, and of a divergence. If we
now treat equations (1) and (3) in the same way as we just above
treated (2) and (4), that is, multiply (1) by φ and (3) scalarly
by s, we get
∂ρ ∂f
φ +s + div(φs) = (E , s). (61)
∂t ∂t
(60) and (61) together give the energy theorem; accordingly the
energy-flux must be
S = [E , H] + φs,
and
φ δρ + s δf + H δB + E δD = δW
is the total differential of the energy-density. It is easy to see
why a term proportional to s, namely φs, has to be added to the
term (E , H) which holds in the æther. For when the electron that
 CHAPTER III 312

generates the convection-current s moves, its energy-content flows

also. In the æther the term (E , H) is overpowered by S, but in the
electron the other φs easily gains the upper hand. The quantities
ρ, f , B, D occur in the formula for the total differential of the
energy-density as independent differentiated phase-quantities. For
the sake of clearness we shall introduce φ and E as independent
variables in place of ρ and D. By this means all the quantities
of intensity are made to act as independent variables. We must
build up
L = W − ED − ρφ, (62)
and then we get

δL = (H δB − D δE ) + (s δf − ρ δφ).

If L is known as a function of the quantities of intensity, then these

equations express the quantities of magnitude as functions of the
quantities of intensity. In place of the ten unknown universal
functions we have now only one, L; this is accomplished by
the principle of energy.
Let us again return to four-dimensional notation, we then have

δL = 21 H ik δFik + si δφi . (63)

From this it follows that δL, and hence L, the “Hamiltonian

Function” is an invariant. The simplest invariants that may be
formed from a vector having components φi and a linear tensor
of the second order having components Fik are the squares of the
following expressions:

the vector φi , φi φi ,
the tensor Fik , 2L0 = 12 Fik F ik ,
 RELATIVITY OF SPACE AND TIME 313

P
the linear tensor of the fourth order with components ±Fik Flm
(the summation extends over the 24 permutations of the indices
i, k, l, m; the upper sign applies to the even permutations, the
lower ones to the odd); and finally of the vector Fik φk .
Just as in three-dimensional geometry the most important the-
orem of congruence is that a vector-pair a, b is fully characterised
in respect to congruence by means of the invariants a2 , ab, b2 ,
so it may be shown in four-dimensional geometry that the invari-
ants quoted determine fully in respect to congruence the figure
composed of a vector φ and a linear tensor of the second order F .
Every invariant, in particular the Hamiltonian Function L, must
therefore be expressible algebraically in terms of the above four
quantities. Mie’s Theory thus resolves the problem of matter into
a determination of this expression. Maxwell’s Theory of the æther
which, of course, precludes the possibility of electrons, is contained
in it as the special case L = L0 . If we also express W and the
components of S in terms of four-dimensional quantities, we see
that they are the negative (0th) row in the scheme

Tik = Fir H kr + φi sk − L · δik . (64)

The Tik ’s are thus the mixed components of the energy-momentum-

tensor, which, according to our calculations, fulfil the theorem
of conservation (57) for i = 0 and hence also for i = 1, 2, 3.
In the next chapter we shall add the proof that its co-variant
components satisfy the condition of symmetry Tki = Tik .
The laws for the field may be summarised in a very simple
principle of variation, Hamilton’s Principle. For this we regard
only the potential with components φi as an independent phase-
 CHAPTER III 314

quantity, and define the field by the equation

∂φi ∂φk
Fik = − .
∂xk ∂xi
Hamilton’s invariant function L which depends on the potential
and the field enters into these laws. We define the current-vector s
and the skew-symmetrical tensor H by means of (63). If in an
arbitrary linear co-ordinate system
√
dω = g dx0 dx1 dx2 dx3

is the four-dimensional “volume-element” of the world (−gZ is the

determinant of the metrical groundform) then the integral L dω
taken over any region of the world is an invariant. It is called the
Action contained in the region in question. Hamilton’s Principle
states that the change in the total Action for each infinitesimal
variation of the state of the field, which vanishes outside a finite
region, is zero, that is,
Z Z
δ L dω = δL dω = 0. (65)

This integral is to be taken over the whole world or, what comes
to the same thing, over a finite region beyond which the varia-
tion of the phase vanishes. This variation is represented by the
infinitesimal increments δφi of the potential-components and the
accompanying infinitesimal change of the field

∂(δφi ) ∂(δφk )
δFik = −
∂xk ∂xi
 RELATIVITY OF SPACE AND TIME 315

in which δφi are space-time functions that only differ from zero
within a finite region. If we insert for δL the expression (63), we
get
∂(δφi )
δL = si δφi + H ik .
∂xk
By the principle of partial integration (vide page 165) we get

ik ∂(δφi ) ∂H ik
Z Z
H dω = − δφi dω,
∂xk ∂xk
and, accordingly,

∂H ik
Z Z  
i
δ L dω = s − δφi dω. (66)
∂xk

Whereas (3) is given by definition, we see that Hamilton’s Prin-

ciple furnishes the field-equations (4). In point of fact, if, for
instance,
∂H ik
s− 6= 0
∂xk
but is > 0 at a certain point, then we could mark off a small region
encircling this point, such that, for it, this difference is positive
throughout. If we then choose a non-negative function for δφ1
that vanishes outside the region marked off, and if δφ2 = δφ3 =
δφ4 = 0, we arrive at a contradiction to equation (65)—(1) and (2)
follow from (3) and (4).
We find, then, that Mie’s Electrodynamics exists in a
compressed form in Hamilton’s Principle (65)—analogously
to the manner in which the development of mechanics attains its
zenith in the principle of action. Whereas in mechanics, however,
 CHAPTER III 316

a definite function L of action corresponds to every given me-

chanical system and has to be deduced from the constitution of
the system, we are here concerned with a single system, the world.
This is where the real problem of matter takes its beginning: we
have to determine the “function of action,” the world-function L,
belonging to the world. For the present it leaves us in perplexity.
If we choose an arbitrary L, we get a “possible” world governed
by this function of action, which will be perfectly intelligible to
us—more so than the actual world—provided that our mathemat-
ical analysis does not fail us. We are, of course, then concerned in
discovering the only existing world, the real world for us. Judging
from what we know of physical laws, we may expect the L which
belongs to it to be distinguished by having simple mathematical
properties. Physics, this time as a physics of fields, is again pur-
suing the object of reducing the totality of natural phenomena to
a single physical law: it was believed that this goal was almost
within reach once before when Newton’s Principia, founded on the
physics of mechanical point-masses was celebrating its triumphs.
But the treasures of knowledge are not like ripe fruits that may
be plucked from a tree.
For the present we do not yet know whether the phase-
quantities on which Mie’s Theory is founded will suffice to de-
scribe matter or whether matter is purely “electrical” in nature.
Above all, the ominous clouds of those phenomena that we are
with varying success seeking to explain by means of the quantum
of action, are throwing their shadows over the sphere of physical
knowledge, threatening no one knows what new revolution.
Let us try the following hypothesis for L:
p
L = 21 |F |2 + w( −φi φi ) (67)
 RELATIVITY OF SPACE AND TIME 317

(w is the symbol for a function of one variable); it suggests itself as

being the simplest of those that go beyond Maxwell’s Theory. We
have no grounds for assuming that the world-function has actually
this form. We shall confine ourselves to a consideration of statical
solutions, for which we have

B = H = 0, s = f = 0,
E = grad φ, div D = ρ,
D = E, ρ = −w0 (φ)

(the accent denoting the derivative). In comparison with the or-

dinary electrostatics of the æther we have here the new circum-
stance that the density ρ is a universal function of the potential,
the electrical pressure φ. We get for Poisson’s equation

∆φ + w0 (φ) = 0. (68)

If w(φ) is not an even function of φ, this equation no longer holds

after the transition from φ to −φ; this would account for the
difference between the natures of positive and negative
electricity. Yet it certainly leads to a remarkable difficulty in
the case of non-statical fields. If charges having opposite signs
are to occur in the latter, the root in (67) must have different
signs at different points of the field. Hence there must be points
in the field, for which φi φi vanishes. In the neighbourhood of
such a point φi φi must be able to assume positive and negative
values (this does not follow in the statical case, as the minimum
of the function φ20 for φ0 is zero). The solutions of our field-
equations must, therefore, become imaginary at regular distances
apart. It would be difficult to interpret a degeneration of the field
into separate portions in this way, each portion containing only
 CHAPTER III 318

charges of one sign, and separated from one another by regions in

which the field becomes imaginary.
A solution (vanishing at infinity) of equation (68) represents
a possible state of electrical equilibrium, or a possible corpuscle
capable of existing individually in the world that we now proceed
to construct. The equilibrium can be stable, only if the solution is
radially symmetrical. In this case, if r denotes the radius vector,
the equation becomes
 
1 d 2 dφ
r + w0 (φ) = 0. (69)
r2 dr dr

If (69) is to have a regular solution

e0 e1
−φ = + 2 + ... (70)
r r
at r = ∞, we find by substituting this power series for the first
term of the equation that the series for w0 (φ) begins with the
power r−4 or one with a still higher negative index, and hence
that w(x) must be a zero of at least the fifth order for x = 0.
On this assumption the equations must have a single infinity of
regular solutions at r = 0 and also a single infinity of regular so-
lutions at r = ∞. We may (in the “general” case) expect these
two one-dimensional families of solutions (included in the two-
dimensional complete family of all the solutions) to have a finite
or, at any rate, a discrete number of solutions. These would rep-
resent the various possible corpuscles. (Electrons and elements of
the atomic nucleus?) One electron or one atomic nucleus does
not, of course, exist alone in the world; but the distances between
them are so great in comparison with their own size that they do
not bring about an appreciable modification of the structure of
 RELATIVITY OF SPACE AND TIME 319

the field within the i interior of an individual electron or atomic

nucleus. If φ is a solution of (69) that represents such a corpuscle
in (70) then its total charge
Z ∞ 
0 2 2 dφ 

= 4π w (φ)r dr = −4π · r = 4πc0 ,
0 dr r=∞

but its mass is calculated as the integral of the energy-density W

that is given by (62):
Z ∞
(grad φ)2 + w(φ) − φw0 (φ) r2 dr
1
Mass = 4π 2
Z0 ∞
w(φ) − 12 φw0 (φ) r2 dr.

= 4π
0

These physical laws, then, enable us to calculate the mass

and charge of the electrons, and the atomic weights and atomic
charges of the individual existing elements whereas, hitherto, we
have always accepted these ultimate constituents of matter as
things given with their numerical properties. All this, of course, is
merely a suggested plan of action as long as the world-function L
is not known. The special hypothesis (67) from which we just
now started was assumed only to show what a deep and thorough
knowledge of matter and its constituents as based on laws would
be exposed to our gaze if we could but discover the action-function.
For the rest, the discussion of such arbitrarily chosen hypotheses
cannot lead to any proper progress; new physical knowledge and
principles will be required to show us the right way to determine
the Hamiltonian Function.
To make clear, ex contrario, the nature of pure physics of fields,
which was made feasible by Mie for the realm of electrodynamics
 CHAPTER III 320

as far as its general character furnishes hypotheses, the principle

of action (65) holding in it will be contrasted with that by which
the theory of Maxwell and Lorentz is governed; the latter theory
recognises, besides the electromagnetic field, a substance moving
in it. This substance is a three-dimensional continuum; hence its
parts may be referred in a continuous manner to the system of
values of three co-ordinates α, β, γ. Let us imagine the substance
divided up into infinitesimal elements. Every element of substance
has then a definite invariable positive mass dm and an invariable
electrical charge de. As an expression of its history there corre-
sponds to it then a world-line with a definite direction of traverse
or, in better words, an infinitely thin “world-filament”. If we again
divide this up into small portions, and if
p
ds = −gik dxi dxk
is the proper-time length of such a portion, then we may introduce
the space-time function µ0 of the statical mass-density by means
of the invariant equation
dm ds = µ0 dω. (71)
We shall call the integral
Z Z Z Z p
µ0 dω = dm ds = dm −gik dxi dxk
X

taken over a region X of the world the substance-action of

mass. In the last integral the inside integration refers to that part
of the world-line of any arbitrary element of substance of mass dm,
which belongs to the region X, the outer integral signifies summa-
tion taken for all elements of the substance. In purely mathemat-
ical language this transition from substance-proper-time integrals
 RELATIVITY OF SPACE AND TIME 321

to space-time integrals occurs as follows. We first introduce the

substance-density ν of the mass thus:
dm = ν dα dβ dγ
(ν behaves as a scalar-density for arbitrary transformations of the
substance co-ordinates α, β, γ). On each world-line of a substance-
point α, β, γ we reckon the proper-time s from a definite initial
point (which must, of course, vary continuously from substance-
point to substance-point). The co-ordinates xi of the world-point
at which the substance-point α, β, γ happens to be at the mo-
ment s of its motion (after the proper-time s has elapsed), are then
continuous functions of α, β, γ, s, whose functional determinant
∂(x0 , x1 , x2 , x3 )
∂(α, β, γ, s)
we shall suppose to have the absolute value ∆. The equation (71)
then states that
√ ν
µ0 g = .
∆
In an analogous manner we may account for the statical density ρ0
of the electrical charge. We shall set down
Z  Z 
de φi dxi

as substance-action of electricity; in it the outer integration is

again taken over all the substance-elements, but the inner one in
each case over that part of the world-line of a substance-element
carrying the charge de whose path lies in the interior of the world-
region X. We may therefore also write
Z Z Z
de ds · φu = ρ0 u φi dω = si φi dω
i
 CHAPTER III 322

dxi
if ui = are the components of the world-direction, and
ds
si = ρ0 ui are the components of the 4-current (a pure convection
current). Finally, in addition to the substance-action there is
also a field-action of electricity, for which Maxwell’s Theory
makes the simple convention
 
∂φi ∂φk
Z
1 ik
4
Fik F dω Fik = − .
∂xk ∂xi
Hamilton’s Principle, which gives a condensed statement of the
Maxwell-Lorentz Laws, may then be expressed thus:
The total action, that is, the sum of the field-action and
substance-action of electricity plus the substance-action of the
mass for any arbitrary variation (vanishing for points beyond a
finite region) of the field-phase (of the φi ’s) and for a similarly
conditioned space-time displacement of the world-lines described
by the individual substance-points undergoes no change.
This principle clearly gives us the equations
∂F ik
= si = ρ0 ui ,
∂xk
if we vary the φi ’s. If, however, we keep the φi ’s constant, and
perform variations on the world-lines of the substance-points, we
get, by interchanging differentiation and variation (as in § 17 in
determining the shortest lines), and then integrating partially:
Z Z Z
φi dxi = (δφi dxi + φi dδφi ) = (δφi dxi − δxi dφi )
Z  
∂φi ∂φk
= − δxk · dxi .
∂xk ∂xi
 RELATIVITY OF SPACE AND TIME 323

In this the δxi ’s are the components of the infinitesimal displace-

ment, which the individual points of the world-line undergo. Ac-
cordingly, we get
Z  Z  Z Z
δ de φi dxi = de ds · Fik u δxk = ρ0 Fik ui δxk · dω.
i

If we likewise perform variation on the substance-action of the

mass (this has already been done in § 17 for a more general case,
in which the gik ’s were variable), we arrive at the mechanical equa-
tions which are added to the field-equations in Maxwell’s Theory;
namely
dui
µ0 = pi pi = ρ0 Fik uk = Fik sk .
ds
This completes the cycle of laws which were mentioned on
page 297. This theory does not, of course, explain the exis-
tence of the electron, since the cohesive forces are lacking in
it.
A striking feature of the principle of action just formulated
is that a field-action does not associate itself with the substance-
action of the mass, as happens in the case of electricity. This gap
will be filled in the next chapter, in which it will be shown that the
gravitational field is what corresponds to mass in the same way
as the electromagnetic field corresponds to the electrical charge.
The great advance in our knowledge described in this chap-
ter consists in recognising that the scene of action of reality
is not a three-dimensional Euclidean space but rather a four-
dimensional world, in which space and time are linked
together indissolubly. However deep the chasm may be that
separates the intuitive nature of space from that of time in our
 CHAPTER III 324

experience, nothing of this qualitative difference enters into the

objective world which physics endeavours to crystallise out of
direct experience. It is a four-dimensional continuum, which is
neither “time” nor “space”. Only the consciousness that passes on
in one portion of this world experiences the detached piece which
comes to meet it and passes behind it, as history, that is, as a
process that is going forward in time and takes place in space.
This four-dimensional space is metrical like Euclidean space,
but the quadratic form which determines its metrical structure
is not definitely positive, but has one negative dimension. This
circumstance is certainly of no mathematical importance, but has
a deep significance for reality and the relationship of its action. It
was necessary to grasp the idea of the metrical four-dimensional
world, which is so simple from the mathematical point of view,
not only in isolated abstraction but also to pursue the weightiest
inferences that can be drawn from it towards setting up the view
of physical phenomena, so that we might arrive at a proper un-
derstanding of its content and the range of its influence: that was
what we aimed to do in a short account. It is remarkable that
the three-dimensional geometry of the statical world that was put
into a complete axiomatic system by Euclid has such a translucent
character, whereas we have been able to assume command over
the four-dimensional geometry only after a prolonged struggle and
by referring to an extensive set of physical phenomena and em-
pirical data. Only now the theory of relativity has succeeded in
enabling our knowledge of physical nature to get a full grasp of
the fact of motion, of change in the world.
 CHAPTER IV
THE GENERAL THEORY OF RELATIVITY
§ 27. The Relativity of Motion, Metrical Fields,
Gravitation∗
However successfully the Principle of Relativity of Einstein
worked out in the preceding chapter marshals the physical laws
which are derived from experience and which define the relation-
ship of action in the world, we cannot express ourselves as satisfied
from the point of view of the theory of knowledge. Let us again
revert to the beginning of the foregoing chapter. There we were in-
troduced to a “kinematical” principle of relativity; x1 , x2 , x3 , t were
the space-time co-ordinates of a world-point referred to a definite
permanent Cartesian co-ordinate system in space; x01 , x02 , x03 , t0
were the co-ordinates of the same point relative to a second such
system, that may be moving arbitrarily with respect to the first;
they are connected by the transformation formulæ (II), page 225.
It was made quite clear that two series of physical states or phases
cannot be distinguished from one another in an objective manner,
if the phase-quantities of the one are represented by the same
mathematical functions of x01 , x02 , x03 , t0 as those that describe the
first series in terms of the arguments x1 , x2 , x3 , t. Hence the
physical laws must have exactly the same form in the one system
of independent space-time arguments as in the other. It must cer-
tainly be admitted that the facts of dynamics are apparently in
direct contradiction to Einstein’s postulate, and it is just these
facts that, since the time of Newton, have forced us to attribute
an absolute meaning, not to translation, but to rotation. Yet our
∗
Vide note 1.

325
 CHAPTER IV 326

minds have never succeeded in accepting unreservedly this torso

thrust on them by reality (in spite of all the attempts that have
been made by philosophers to justify it, as, for example, Kant’s
Metaphysische Anfangsgründe der Naturwissenschaften), and the
problem of centrifugal force has always been felt to be an unsolved
enigma (vide note 2).
Where do the centrifugal and other inertial forces take their
origin? Newton’s answer was: in absolute space. The answer
given by the special theory of relativity does not differ essentially
from that of Newton. It recognises as the source of these forces
the metrical structure of the world and considers this structure as
a formal property of the world. But that which expresses itself as
force must itself be real. We can, however, recognise the metrical
structure as something real, if it is itself capable of undergoing
changes and reacts in response to matter. Hence our only way out
of the dilemma—and this way, too, was opened up by Einstein—
is to apply Riemann’s ideas, as set forth in Chapter II, to the
four-dimensional Einstein-Minkowski world which was treated in
Chapter III instead of to three-dimensional Euclidean space. In
doing this we shall not for the present make use of the most general
conception of the metrical manifold, but shall retain Riemann’s
view. According to this, we must assume the world-points to form
a four-dimensional manifold, on which a measure-determination
is impressed by a non-degenerate quadratic differential form Q
having one positive and three negative dimensions.∗ In any co-
∗
We have made a change in the notation, as compared with that of the
preceding chapter, by placing reversed signs before the metrical groundform.
The former convention was more convenient for representing the splitting up
of the world into space and time, the present one is found more expedient in
the general theory.
 THE GENERAL THEORY OF RELATIVITY 327

ordinate system xi (i = 0, 1, 2, 3), in Riemann’s sense, let

X
Q= gik dxi dxk . (1)
i,k

Physical laws will then be expressed by tensor relations that are

invariant for arbitrary continuous transformations of the argu-
ments xi . In them the co-efficients gik of the quadratic differ-
ential form (1) will occur in conjunction with the other physical
phase-quantities. Hence we shall satisfy the postulate of relativ-
ity enunciated above, without violating the facts of experience,
if we regard the gik ’s in exactly the same way as we regarded
the components φi of the electromagnetic potential (which are
formed
P by the co-efficients of an invariant linear differential form
φi dxi ), as physical phase-quantities, to which there cor-
responds something real, namely, the “metrical field”. Un-
der these circumstances invariance exists not only with respect to
the transformations mentioned (II), which have a fully arbitrary
(non-linear) character only for the time-co-ordinate, but for any
transformations whatsoever. This special distinction conferred on
the time-co-ordinate by (II), is, indeed, incompatible with the
knowledge gained from Einstein’s Principle of Relativity. By al-
lowing any arbitrary transformations in place of (II), that is, also
such as are non-linear with respect to the space-co-ordinates, we
affirm that Cartesian co-ordinate systems are in no wise more
favoured than any “curvilinear” co-ordinate system. This seals
the doom of the idea that a geometry may exist inde-
pendently of physics in the traditional sense, and it is just
because we had not emancipated ourselves from the dogma that
such a geometry existed that we arrived by logical considerations
at the relativity principle (II), and not at once at the principle
 CHAPTER IV 328

of invariance for arbitrary transformations of the four world-co-

ordinates. Actually, however, spatial measurement is based on a
physical event: the reaction of light-rays and rigid measuring rods
on our whole physical world. We have already encountered this
view in § 21, but we may, above all, take up the thread from our
discussion in § 12, for we have, indeed, here arrived at Riemann’s
“dynamical” view as a necessary consequence of the relativity of
all motion. The behaviour of light-rays and measuring rods, be-
sides being determined by their own natures, is also conditioned
by the “metrical field,” just as the behaviour of an electric charge
depends not only on it, itself, but also on the electric field. Again,
just as the electric field, for its part, depends on the charges and
is instrumental in producing a mechanical interaction between the
charges, so we must assume here that the metrical field (or, in
mathematical language, the tensor with components gik ) is re-
lated to the material content filling the world. We again
call attention to the principle of action set forth at the conclusion
of the preceding paragraph; in both of the parts which refer to
substance, the metrical field takes up the same position towards
mass as the electrical field does towards the electric charge. The
assumption, which was made in the preceding chapter, concerning
the metrical structure of the world (corresponding to that of Eu-
clidean geometry in three-dimensional space), namely, that there
are specially favoured co-ordinate systems, “linear” ones, in which
the metrical groundform has constant co-efficients, can no longer
be maintained in the face of this view.
A simple illustration will suffice to show how geometrical con-
ditions are involved when motion takes place. Let us set a plane
disc spinning uniformly. I affirm that if we consider Euclidean
geometry valid for the reference-space relative to which we speak
 THE GENERAL THEORY OF RELATIVITY 329

of uniform rotation, then it is no longer valid for the rotating disc

itself, if the latter be measured by means of measuring rods mov-
ing with it. For let us consider a circle on the disc described with
its centre at the centre of rotation. Its radius remains the same
no matter whether the measuring rods with which I measure it
are at rest or not, since its direction of motion is perpendicular
to the measuring rod when in the position required for measuring
the radius, that is, along its length. On the other hand, I get a
value greater for the circumference of the circle than that obtained
when the disc is at rest when I apply the measuring rods, owing
to the Lorentz-Fitzgerald contraction which the latter undergoes.
The Euclidean theorem which states that the circumference of the
circle = 2π times the radius thus no longer holds on the disc when
it rotates.
The falling over of glasses in a dining-car that is passing round
a sharp curve and the bursting of a fly-wheel in rapid rotation are
not, according to the view just expressed, effects of “an absolute
rotation” as Newton would state but whose existence we deny;
they are effects of the “metrical field” or rather of the affine rela-
tionship associated with it. Galilei’s principle of inertia shows that
there is a sort of “forcible guidance” which compels a body that is
projected with a definite velocity to move in a definite way which
can be altered only by external forces. This “guiding field,” which
is physically real, was called “affine relationship” above. When a
body is diverted by external forces the guidance by forces such
as centrifugal reaction asserts itself. In so far as the state of the
guiding field does not persist, and the present one has emerged
from the past ones under the influence of the masses existing in
the world, namely, the fixed stars, the phenomena cited above are
partly an effect of the fixed stars, relative to which the rotation
 CHAPTER IV 330

takes place.∗
Following Einstein by starting from the special theory of rel-
ativity described in the preceding chapter, we may arrive at the
general theory of relativity in two successive stages.
I. In conformity with the principle of continuity we take the
same step in the four-dimensional world that, in Chapter II,
brought us from Euclidean geometry to Riemann’s geometry.
This causes a quadratic differential form (1) to appear. There
is no difficulty in adapting the physical laws to this generalisa-
tion. It is expedient to represent the magnitude quantities by
tensor-densities instead of by tensors as in Chapter III; we can do
√
this by multiplying throughout by g (in which g is the negative
determinant of the gik ’s). Thus, in particular, the mass- and
charge-densities µ and ρ, instead of being given by formula (71)
of § 26, will be given by

dm ds = µ dx, de ds = ρ dx (dx = dx0 dx1 dx2 dx3 ).

The proper time ds along the world-line is determined from

ds2 = gik dxi dxk .

∗
We say “partly” because the distribution of matter in the world does not
define the “guiding field” uniquely, for both are at one moment indepen-
dent of one another and accidental (analogously to charge and electric field).
Physical laws tell us merely how, when such an initial state is given, all other
states (past and future) necessarily arise from them. At least, this is how
we must judge, if we are to maintain the standpoint of pure physics of fields.
The statement that the world in the form we perceive it taken as a whole is
stationary (i.e. at rest) can be interpreted, if it is to have a meaning at all,
as signifying that it is in statistical equilibrium. Cf. § 34.
 THE GENERAL THEORY OF RELATIVITY 331

Maxwell’s equations will be

∂φi ∂φk ∂Fik
Fik = − , = si ,
∂xk ∂xi ∂xk
in which the φi ’s are the co-efficients of an invariant linear dif-
√
ferential form φi dxi , and Fik denotes g · F ik according to our
convention above. In Lorentz’s Theory we set
 
i i i dxi
s = ρu u = .
ds
The mechanical force per unit of volume (a co-variant vector-
density in the four-dimensional world) is given by:∗
pi = −Fik sk , (2)
and the mechanical equations are in general
   
dui iβ β
µ − uα u = pi (3)
ds α
with the condition that pi ui always = 0. We may put them into
the same form as we found for them earlier by introducing, in
addition to the pi ’s, the quantities
 
iβ ∂gαβ
· µuα uβ = 21 · µuα uβ (4)
α ∂xi
(cf. § 17, equation (64)) as the density components p̄i of a “pseudo-
force” (force of reaction of the guiding field). The equations then
become
dui
µ = pi + p̄i .
ds
∗
The sign is reversed on account of the reversal of sign in the metrical
groundform.
 CHAPTER IV 332

The simplest examples of such “pseudo-forces” are centrifugal

forces and Coriolis forces. If we compare formula (4) for the
“pseudo-force” arising from the metrical field with that for the
mechanical force of the electromagnetic field, we find them fully
analogous. For just as the vector-density with the contra-variant
components si characterises electricity so, as we shall presently
see, moving matter is described by the tensor-density which has
the components Tki = µui uk . The quantities
 
α iβ
Γiβ =
α
correspond as components of the metrical field to the compo-
nents Fik of the electric field. Just as the field-components F are
derived by differentiation from the electromagnetic potential φi ,
so also the Γ’s from the gik ’s; these thus constitute the potential of
the metrical field. The force-density is the product of the electric
field and electricity on the one hand, and of the metrical field and
matter on the other, thus
pi = −Fik sk , p̄i = Γαiβ Tβα .
If we abandon the idea of a substance existing independently
of physical states, we get instead the general energy-momentum-
density Tki which is determined by the state of the field. According
to the special theory of relativity it satisfies the Law of Conserva-
tion
∂Tki
= 0.
∂xk
This equation is now to be replaced, in accordance with for-
mula (37) § 14, by the general invariant
∂Tki
− Γαiβ Tβα = 0. (5)
∂xk
 THE GENERAL THEORY OF RELATIVITY 333

If the left-hand side consisted only of the first member, T would

now again satisfy the laws of conservation. But we have, in this
case, a second term. The “real” total force
∂Tki
pi = −
∂xk
does not vanish but must be counterbalanced by the “pseudo-
force” which has its origin in the metrical field, namely
1 ∂gαβ αβ
p̄i = Γαiβ Tβα = 2
T . (6)
∂xi
These formulæ were found to be expedient in the special theory
of relativity when we used curvilinear co-ordinate systems, or such
as move curvilinearly or with acceleration. To make clear the
simple meaning of these considerations we shall use this method
to determine the centrifugal force that asserts itself in a rotating
system of reference. If we use a normal co-ordinate system for the
world, namely, t, x1 , x2 , x3 , but introduce r, z, θ, in place of the
Cartesian space co-ordinates, we get
ds2 = dt2 − (dz 2 + dr2 + r2 dθ2 ).
Using ω to denote a constant angular velocity, we make the sub-
stitution
θ = θ0 + ωt0 , t = t0
and, after the substitution, drop the accents. We then get
ds2 = dt2 (1 − r2 ω 2 ) − 2r2 ω dθ dt − (dz 2 + dr2 + r2 dθ2 ).
If we now put
t = x0 , θ = x1 , z = x2 , r = x3 ,
 CHAPTER IV 334

we get for a point-mass which is at rest in the system of reference

now used

u1 = u2 = u3 = 0; and hence (u0 )2 (1 − r2 ω 2 ) = 1.

The components of the centrifugal force satisfy formula (4)

∂g00
p̄i = 1
2
· µ(u0 )2
∂xi
and since the derivatives with respect to x0 , x1 , x2 of g00 , which
is equal to 1 − r2 ω 2 , vanish and since
∂g00 ∂g00
= = −2rω 2
∂x3 ∂r
then, if we return to the usual units, in which the velocity of light
is not unity, and if we use contra-variant components instead
of co-variant ones, and instead of the indices 0, 1, 2, 3 the more
indicative ones t, θ, z, r, we obtain

µrω 2
p̄t = p̄θ = p̄z = 0, p̄r =  rω 2 . (7)
1−
c
Two closely related circumstances characterise the “pseudo-
forces” of the metrical field. Firstly, the acceleration which they
impart to a point-mass situated at a definite space-time point (or,
more exactly, one passing through this point with a definite veloc-
ity) is independent of its mass, i.e. the force itself is proportional
to the inertial mass of the point-mass at which it acts. Secondly,
if we use an appropriate co-ordinate system, namely, a geodetic
one, at a definite space-time point, these forces vanish (cf. § 14). If
 THE GENERAL THEORY OF RELATIVITY 335

the special theory of relativity is to be maintained, this vanishing

can be effected simultaneously for all space-time points by the in-
troduction of a linear co-ordinate system, but in the general case
it is possible to make the whole 40 components Γαiβ of the affine
relationship vanish at least for each individual point by choosing
an appropriate co-ordinate system at this point.∗
Now the two related circumstances just mentioned are true, as
we know, of the force of gravitation. The fact that a given grav-
itational field imparts the same acceleration to every mass that is
brought into the field constitutes the real essence of the problem of
gravitation. In the electrostatic field a slightly charged particle is
acted on by the force e · E, the electric charge e depending only on
the particle, and E, the electric intensity of field, depending only
on the field. If no other forces are acting, this force imparts to the
particle whose inertial mass is m an acceleration which is given by
the fundamental equation of mechanics mb = eE. There is some-
thing fully analogous to this in the gravitational field. The force
that acts on the particle is equal to gG, in which g, the “gravita-
tional charge,” depends only on the particle, whereas G depends
only on the field: the acceleration is determined here again by
the equation mb = gG. The curious fact now manifests itself
that the “gravitational charge” or the “gravitational mass” g
is equal to the “inertial mass” m. Eötvös has comparatively
recently tested the accuracy of this law by actual experiments of
the greatest refinement (vide note 3). The centrifugal force im-
parted to a body at the earth’s surface by the earth’s rotation
is proportional to its inertial mass but its weight is proportional
∗
Hence we see that it is in the nature of the metrical field that it cannot
be described by a field-tensor Γ which is invariant with respect to arbitrary
transformations.
 CHAPTER IV 336

to its gravitational mass. The resultant of these two, the appar-

ent weight, would have different directions for different bodies if
gravitational and inertial mass were not proportional throughout.
The absence of this difference of direction was demonstrated by
Eötvös by means of the exceedingly sensitive instrument known
as the torsion-balance: it enables the inertial mass of a body to
be measured to the same degree of accuracy as that to which its
weight may be determined by the most sensitive balance. The
proportionality between gravitational and inertial mass holds in
cases, too, in which a diminution of mass is occasioned not by
an escape of substance in the old sense, but by an emission of
radioactive energy.
The inertial mass of a body has, according to the fundamental
law of mechanics, a universal significance. It is the inertial mass
that regulates the behaviour of the body under the influence of
any forces acting on it, of whatever physical nature they may be;
the inertial mass of the body is, however, according to the usual
view associated only with a special physical field of force, namely,
that of gravitation. From this point of view, however, the identity
between inertial and gravitational mass remains fully incompre-
hensible. Due account can be taken of it only by a mechanics
which from the outset takes into consideration gravitational as
well as inertial mass. This occurs in the case of the mechanics
given by the general theory of relativity, in which we assume that
gravitation, just like centrifugal and Coriolis forces, is in-
cluded in the “pseudo-force” which has its origin in the
metrical field. We shall find actually that the planets pursue
the courses mapped out for them by the guiding field, and that
we need not have recourse to a special “force of gravitation,” as
did Newton, to account for the influence which diverts the planets
 THE GENERAL THEORY OF RELATIVITY 337

from their paths as prescribed by Galilei’s Principle (or Newton’s

first law of motion). The gravitational forces satisfy the second
postulate also; that is, they may be made to vanish at a space-time
point if we introduce an appropriate co-ordinate system. A closed
box, such as a lift, whose suspension wire has snapped, and which
descends without friction in the gravitational field of the earth, is
a striking example of such a system of reference. All bodies that
are falling freely will appear to be at rest to an observer in the
box, and physical events will happen in the box in just the same
way as if the box were at rest and there were no gravitational
field, in spite of the fact that the gravitational force is acting.
II. The transition from the special to the general theory of
relativity, as described in I, is a purely mathematical process. By
introducing the metrical groundform (1), we may formulate phys-
ical laws so that they remain invariant for arbitrary transforma-
tions; this is a possibility that is purely mathematical in essence
and denotes no particular peculiarity of these laws. A new physical
factor appears only when it is assumed that the metrical structure
of the world is not given a priori, but that the above quadratic
form is related to matter by generally invariant laws. Only this
fact justifies us in assigning the name “general theory of relativity”
to our reasoning; we are not simply giving it to a theory which
has merely borrowed the mathematical form of relativity. The
same fact is indispensable if we wish to solve the problem of the
relativity of motion; it also enables us to complete the analogy
mentioned in I, according to which the metrical field is related to
matter in the same way as the electric field to electricity. Only if
we accept this fact does the theory briefly quoted at the end of
the previous section become possible, according to which gravi-
tation is a mode of expression of the metrical field; for we
 CHAPTER IV 338

know by experience that the gravitational field is determined (in

accordance with Newton’s law of attraction) by the distribution
of matter. This assumption, rather than the postulate of general
invariance, seems to the author to be the real pivot of the general
theory of relativity. If we adopt this standpoint we are no longer
justified in calling the forces that have their origin in the metrical
field pseudo-forces. They then have just as real a meaning as the
mechanical forces of the electromagnetic field. Coriolis or centrifu-
gal forces are real force effects, which the gravitational or guiding
field exerts on matter. Whereas, in I, we were confronted with the
easy problem of extending known physical laws (such as Maxwell’s
equations) from the special case of a constant metrical fundamen-
tal tensor to the general case, we have, in following the ideas set
out just above, to discover the invariant law of gravitation, ac-
cording to which matter determines the components Γαβi
of the gravitational field, and which replaces the Newtonian
law of attraction in Einstein’s Theory. The well-known laws of the
field do not furnish a starting-point for this. Nevertheless Einstein
succeeded in solving this problem in a convincing fashion, and in
showing that the course of planetary motions may be explained
just as well by the new law as by the old one of Newton; indeed,
that the only discrepancy which the planetary system discloses
towards Newton’s Theory, and which has hitherto remained inex-
plicable, namely, the gradual advance of Mercury’s perihelion by
4300 per century, is accounted for accurately by Einstein’s theory
of gravitation.
Thus this theory, which is one of the greatest examples of
the power of speculative thought, presents a solution not only
of the problem of the relativity of all motion (the only solution
which satisfies the demands of logic), but also of the problem of
 THE GENERAL THEORY OF RELATIVITY 339

gravitation (vide note 4). We see how cogent arguments added

to those in Chapter II bring the ideas of Riemann and Einstein
to a successful issue. It may also be asserted that their point
of view is the first to give due importance to the circumstance
that space and time, in contrast with the material content of the
world, are forms of phenomena. Only physical phase-quantities
can be measured, that is, read off from the behaviour of matter in
motion; but we cannot measure the four world-co-ordinates that
we assign a priori arbitrarily to the world-points so as to be able to
represent the phase-quantities extending throughout the world by
means of mathematical functions (of four independent variables).
Whereas the potential of the electromagnetic field is built up
from the co-efficients of an invariant linear differential form of the
world-co-ordinates φi dxi , the potential of the gravitational field is
made up of the co-efficients of an invariant quadratic differential
form. This fact, which is of fundamental importance, constitutes
the form of Pythagoras’ Theorem to which it has gradually
been transformed by the stages outlined above. It does not ac-
tually spring from the observation of gravitational phenomena in
the true sense (Newton accounted for these observations by intro-
ducing a single gravitational potential), but from geometry, from
the observations of measurement. Einstein’s theory of gravitation
is the result of the fusion of two realms of knowledge which have
hitherto been developed fully independently of one another; this
synthesis may be indicated by the scheme
Pythagoras Newton
| {z }
Einstein
To derive the values of the quantities gik from directly
observed phenomena, we use light-signals and point-masses
 CHAPTER IV 340

which are moving under no forces, as in the special theory of

relativity. Let the world-points be referred to any co-ordinates xi
in some way. The geodetic lines passing through a world-point O,
namely,
d 2 xi
 
αβ dxα dxβ
+ = 0, (8)
ds2 i ds ds
dxi dxk
gik = C = const., (9)
ds ds
split up into two classes; (a) those with a space-like direction,
(b) those with a time-like direction (C < 0 or C > 0 respec-
tively). The latter fill a “double” cone with the common vertex
at O and which, at O, separates into two simple cones, of which
one opens into the future and the other into the past. The first
comprises all world-points that belong to the “active future” of O,
the second all world-points that constitute the “passive past” of O.
The limiting sheet of the cone is formed by the geodetic null-lines
(C = 0); the “future” half of the sheet contains all the world-points
at which a light-signal emitted from O arrives, or, more generally,
the exact initial points of every effect emanating from O. The met-
rical groundform thus determines in general what world-points
are related to one another in effects. If dxi are the relative co-
ordinates of a point O0 infinitely near O, then O0 will be traversed
by a light-signal emitted from O if, and only if, gik dxi dxk = 0.
By observing the arrival of light at the points neighbouring to O
we can thus determine the ratios of the values of the gik ’s at the
point O; and, as for O, so for any other point. It is impossible,
however, to derive any further results from the phenomenon of
the propagation of light, for it follows from a remark on page 188
that the geodetic null-lines are dependent only on the ratios of
 THE GENERAL THEORY OF RELATIVITY 341

the gik ’s.

The optical “direction” picture that an observer (“point-eye” as
on page 146) receives, for instance, from the stars in the heavens,
is to be constructed as follows. From the world-point O at which
the observer is stationed those geodetic null-lines (light-lines) are
to be drawn on the backward cone which cuts the world-lines of
the stars. The direction of every light-line at O is to be resolved
into one component which lies along the direction e of the world-
line of the observer and another s which is perpendicular to it
(the meaning of perpendicular is defined by the metrical structure
of the world as given on page 180); s is the spatial direction of
the light-ray. Within the three-dimensional linear manifold of
the line-elements at O perpendicular to e, −ds2 is a definitely
positive form. The angles (that arise from it when it is taken
as the metrical groundform, and which are to be calculated from
formula (15), § 11) between the spatial directions s of the light-rays
are those that determine the positions of the stars as perceived by
the observer.
The factor of proportionality of the gik ’s which could not be
derived from the phenomenon of the transmission of light may be
determined from the motion of point-masses which carry a clock
with them. For if we assume that—at least for unaccelerated
motion under no forces—the time read off from such a clock is
the proper-time s, equation (9) clearly makes it possible to ap-
ply the unit of measure along the world-line of the motion (cf.
Appendix I).
 CHAPTER IV 342

§ 28. Einstein’s Fundamental Law of Gravitation

According to the Newtonian Theory the condition (or phase)
of matter is characterised by a scalar, the mass-density µ; and
the gravitational potential is also a scalar Φ: Poisson’s equation
holds, that is,
∆Φ = 4πkµ (10)
(∆ = div grad; k = the gravitational constant). This is the law
according to which matter determines the gravitational field. But
according to the theory of relativity matter can be described ’rig-
orously only by a symmetrical tensor of the second order Tik ,
or better still by the corresponding mixed tensor-density Tki ; in
harmony with this the potential of the gravitational field consists
of the components of a symmetrical tensor gik . Therefore, in
Einstein’s Theory we expect equation (10) to be replaced by a
system of equations of which the left side consists of differential
expressions of the second order in the gik ’s, and the right side of
components of the energy-density; this system has to be invari-
ant with respect to arbitrary transformations of the co-ordinates.
To find the law of gravitation we shall do best by taking up the
thread from Hamilton’s Principle formulated at the close of § 26.
The Action there consisted of three parts: the substance-action of
electricity, the field-action of electricity, and the substance-action
of mass or gravitation. In it there is lacking a fourth term, the
field-action of gravitation, which we have now to find. Before do-
ing this, however, we shall calculate the change in the sum of the
first three terms already known, when we leave the potentials φi
of the electromagnetic field and the world-lines of the substance-
elements unchanged but subject the gik ’s, the potentials of the
metrical field, to an infinitesimal virtual variation δ. This
 THE GENERAL THEORY OF RELATIVITY 343

is possible only from the point of view of the general theory of

relativity.
This causes no change in the substance-action of electricity,
but the change in the integrands that occur in the field-action,
namely
1
2
S = 41 Fik Fik
is √
1 √
4
gδ(Fik F ik ) + (Fik F ik )δ g .
The first summand in the curved bracket here = Frs δF rs and
hence, since
F rs = g ri g sk Fik ,
we immediately get the value
√
2 gFir Fkr δg ik .

The second summand, by (580 ) § 17,

= −Sgik δg ik .

Thus, finally, we find the variation in the field-action to be

Z Z
1 ik 1 ik
= 2
S δg dx = 2
S δgik dx (cf. (59), § 17)

if
Ski = 21 Sδik = Fir Fkr (11)
are the components of the energy-density of the electromagnetic
field.∗ It suddenly becomes clear to us now (and only now that we
∗
The signs are the reverse of those used in Chapter III on account of the
change in the sign of the metrical groundform.
 CHAPTER IV 344

have succeeded in calculating the variation of the world’s metrical

field) what is the origin of the complicated expressions (11) for
the energy-momentum density of the electromagnetic field.
We get a corresponding result for the substance-action of the
mass; for we have
p
1 dxi dxk δgik
δ gik dxi dxk = 2
= 12 ds ui uk δgik ,
ds
and hence
Z  Z p  Z
1
δ dm gik dxi dxk = 2
µui uk δgik dx.

Hence the total change in the Action so far known to us is, for
a variation of the metrical field,
Z
1 ik
2
T δgik dx (12)

in which Tki denotes the tensor-density of the total energy.

The absent fourth term of the Action, namely, the
field-action
Z of gravitation, must be an invariant integral,
G dx, of which the integrand G is composed of the poten-
 
ik
tials gik and of the field-components of the gravitational
r
field, built up from the gik ’s and their first derivatives. It would
seem to us that only under such circumstances do we obtain
differential equations of order not higher than the second for our
gravitational laws. If the total differential of this function is

δG = 21 Gik δgik + 21 Gik,r δgik,r (Gki = Gik and Gki,r = Gik,r ) (13)
 THE GENERAL THEORY OF RELATIVITY 345

we get, for an infinitesimal variation δgik which disappears for

regions beyond a finite limit, by partial integration, that
Z Z
1
δ G dx = 2
[G]ik δgik dx (14)

in which the “Lagrange derivatives” [G]ik , which are symmetrical

in i and k, are to be calculated according to the formula
∂Gik,r
[G] = Gik − .
∂xr
The gravitational equations will then actually assume the form
which was predicted, namely
[G]ki = −Tki . (15)
There is no longer any cause for surprise that it happens to be the
energy-momentum components that appear as co-efficients when
we vary the gik ’s in the first three factors of the Action in accor-
dance with (12). Unfortunately a scalar-density G,  ofthe type we
ik
wish, does not exist at all; for we can make all the ’s vanish at
r
any given point by choosing the appropriate co-ordinate system.
Yet the scalar R, the curvature defined by Riemann, has made us
familiar with an invariant which involves the second derivatives of
the gik ’s only linearly: it may even be shown that it is the only
invariant of this kind (vide Appendix II, in which the proof is
given). InZ consequence of this linearity we may use the invariant
√
integral 21 R g dx to get the derivatives of the second order by
partial integration. We then get
1 √
Z Z
2
R g dx = G dx
 CHAPTER IV 346

+ a divergence integral, that is, an integral whose integrand is of

∂wi
the form : G here depends only on the gik ’s and their first
∂xi
derivatives. Hence, for variations δgik , that vanish outside a finite
region, we get
1 √
Z Z
δ 2 R g dx = δ G dx

since, according to the principle of partial integration,

∂(δwi )
Z
dx = 0.
∂xi
Z Z
Not G dx itself is an invariant, but the variation δ G dx, and
this is the essential feature of Hamilton’sZ Principle. We need not,
therefore, have fears about introducing G dx as the Action of
the gravitational field; and this hypothesis is found to be the only
possible one. We are thus led under compulsion, as it were, to
the unique gravitational equations (15). It follows from them
that every kind of energy exerts a gravitational effect: this
is true not only of the energy concentrated in the electrons and
atoms, that is of matter in the restricted sense, but also of diffuse
field-energy (for the Tki ’s are the components of the total energy).
Before we carry out the calculations that are necessary if we
wish to be able to write down the gravitational equations explic-
itly, we must first test whether we get Z analogous results in the
case of Mie’s Theory. The Action, L dx, which occurs in it is
an invariant not only for linear, but also for arbitrary transforma-
tions. For L is composed algebraically (not as a result of tensor
analysis) of the components φi of a co-variant vector (namely, of
 THE GENERAL THEORY OF RELATIVITY 347

the electromagnetic potential), of the components Fik of a linear

tensor of the second order (namely, of the electromagnetic field),
and of the components gik of the fundamental metrical tensor. We
set the total differential δL of this function equal to
1 ik
2
T δgik + δ0 L, in which δ0 L = 21 Hik δFik + si δφi
(Tki = Tik , Hki = −Hik ). (16)

We then call the tensor-density Tki the energy or matter. By do-

ing this, we affirm once again that the metrical field (with the
potentials gik ) is related to matter (Tik ) in the same way as the
electromagnetic field (with the potentials φi ) is related to the elec-
tric current si . We are now obliged to prove that the present ex-
planation leads accurately to the expressions given in (64), § 26,
for energy and momentum. This will furnish the proof, which was
omitted above, of the symmetry of the energy-tensor. To do this
we cannot use the method of direct calculation as above in the
particular case of Maxwell’s Theory, but we must apply the fol-
lowing elegant considerations, the nucleus of which is to be found
in Lagrange, but which were discussed with due regard to formal
perfection by F. Klein (vide note 5).
We subject the world-continuum to an infinitesimal deforma-
tion, as a result of which in general the point (xi ) becomes trans-
formed into the point (x̄j )

x̄i = xi +  · ξ i (x0 , x1 , x2 , x3 ) (17)

(in which  is the constant infinitesimal parameter, all of whose

higher powers are to be struck out). We imagine the phase-
quantities to follow the deformation so that at its conclusion the
new φi ’s (we call them φ̄i ) are functions of the co-ordinates of such
 CHAPTER IV 348

a kind that, in consequence of (17), the equations

φi (x) dxi = φ̄i (x̄) dx̄i (18)
hold; and in the same sense the symmetrical and skew-symmetrical
bilinear differential form with the co-efficients gik , Fik , respec-
tively, remains unchanged. The changes φ̄i (x) − φi (x) which the
quantities φi undergo at a fixed world-point (xi ) as a result of
the deformation will be denoted by δφi ; δgik and δFik have a
corresponding meaning.
If we replace the old quantities φi in the function L by the
φ̄i arising from the deformation, we shall suppose the function
L̄ = L + δL to result; the δL in it is given by (16). Furthermore,
let X be an arbitrary region of the world which, owing to the
deformation,
Z becomes X. The deformation
Z causes the Action
L dx to undergo a change δ 0 L dx which is equal to the
X X
difference between the integral L̄ taken over X and the integral L
taken over X. The invariance of the Action is expressed by the
equation Z
δ0 L dx = 0. (19)
X
We make a natural division of this difference into two parts:
(1) the difference between the integrals of L̄ and L over X, (2) the
difference between the integral of L over X and X. Since X differs
from X only by an infinitesimal amount, we may set
Z Z
δ L dx = δL dx
X X
for the first part. On page 165 we found the second part to be
∂(Lξ i )
Z
 dx.
X ∂xi
 THE GENERAL THEORY OF RELATIVITY 349

To be able to complete the argument we must next calculate

the variations δφi , δgik , δFik . If we set φ̄i (x̄) − φi (x) = δ 0 φi for a
moment, then, owing to (18), we get

δ 0 φi · dxi + φr dξ r = 0

and hence
∂ξ i
δ 0 φi = − · φr .
∂xi
Moreover, since
∂φ r
δφi = δ 0 φi − φ̄i (x̄) − φ̄i (x) = δ 0 φi −  ·

ξ
∂xr
we get, suppressing the self-evident factor ,
∂ξ r ∂φi r
−δφi = φr + ξ . (20)
∂xi ∂xr
In the same way, we get
∂ξ r ∂ξ r ∂gik r
−δgik = gir + grk + ξ , (200 )
∂xk ∂xi ∂xr
∂ξ r ∂ξ r ∂Fik r
−δFik = Fir + Frk + ξ . (2000 )
∂xk ∂xi ∂xr
And, on account of
∂φi ∂φk ∂(δφi ) ∂(δφk )
Fik = − we have δFik = − , (21)
∂xk ∂xi ∂xk ∂xi
for since the former is an invariant relation, we get from it
∂ φ̄i (x̄) ∂ φ̄k (x̄) ∂ φ̄i (x) ∂ φ̄k (x)
F̄ik (x̄) = − , and also F̄ik (x) = − .
∂ x̄k ∂ x̄i ∂xk ∂xi
 CHAPTER IV 350

Substitution gives us
∂ξ ∂gαβ
−δL = (Tki + Hrk Fri + sk φi ) + ( 21 Tαβ + · · · +)ξ i .
∂xk ∂xi
If we remove the derivatives of ξ i by partial integration, and use
the abbreviation
Vik = Tki + Fir Hkr + φi sk − δik L,
we get a formula of the following form
∂(Vi ξ i )
Z Z Z
0
−δ L dx = dx + (ti ξ i ) dx = 0. (22)
X X ∂x k X

It follows from this that, as we know, by choosing the ξ i ’s ap-

propriately, namely, so that they vanish outside a definite region,
which we here take to be X, we must have, at every point,
ti = 0. (23)
Accordingly, the first summand of (22) is also equal to zero. The
identity which comes about in this way is valid for arbitrary quan-
tities ξ i and for any finite region of integration X. Hence, since the
integral of a continuous function taken over any and every region
can vanish only if the function itself = 0, we must have
∂(Vik ξ i ) ∂ξ i ∂Vik i
= Vik + ξ = 0.
∂xk ∂xk ∂xk
∂ξ i
Now, ξ i and may assume any values at one and the same
∂xk
point. Consequently,
∂Vik
 
k
Vi = 0 =0 .
∂xk
 THE GENERAL THEORY OF RELATIVITY 351

This gives us the desired result

Tki = Lδik − Fir Hkr − φi sk .

These considerations simultaneously give us the theorems of

conservation of energy and of momentum, which we found by cal-
culation in § 26; they are contained in equations (23). The change
in the Action of the whole world for an infinitesimal deformation
which vanishes outside a finite region of the world is found to be
Z Z Z
1 ik
δL dx = 2
T δgik dx + δ0 L dx = 0. (24)

In consequence of the equations (21) and of Hamilton’s Princi-

ple, namely Z
δ0 L dx = 0, (25)

which is here valid, the second part (in Maxwell’s equations) dis-
appears. But the first part, as we have already calculated, is
Z  i Z  k
 
k ∂ξ 1 ∂gαβ αβ i ∂Ti 1 ∂gαβ αβ
− Ti +2 T ξ dx = −2 T ξ i dx.
∂xk ∂xi ∂xk ∂xi

Thus, as a result of the laws of the electromagnetic field,

we get the mechanical equations

∂Tki 1 ∂gαβ αβ
− 2
T = 0. (26)
∂xk ∂xi
(On account of the presence of the additional term due to gravi-
tation these equations can no longer in the general theory of rel-
ativity be fitly termed theorems of conservation. The question
 CHAPTER IV 352

whether proper theorems of conservation may actually be set up

will be discussed in § 33.)
The Hamiltonian Principle which has been supplemented by
the Action of the gravitational field, namely
Z
δ (L + G) dx = 0, (27)

and in which the electromagnetic and the gravitational condition

(phase) of the field may be subjected independently of one another
to virtual infinitesimal variations gives rise to the gravitational
equations (15) in addition to the electromagnetic laws. If we apply
the process above, which ended in (26), to G instead of to L—
here, too, we have, for the variation δ caused by a deformation of
the world-continuum which vanishes outside a finite region, that
√
Z Z
δ G dx = δ 12 R g dx = 0

—we arrive at mathematical identities analogous to (26),

namely
∂[G]ki ∂gαβ
− 12 [G]αβ = 0.
∂xk ∂xi
The fact that G contains the derivatives of the gik ’s as well as
the gik ’s themselves is of no account. Accordingly, the mechanical
equations (26) are just as much a consequence of the gravitational
equations (15) as of the electromagnetic laws of the field.
The wonderful relationships, which here reveal themselves,
may be formulated in the following way independently of the ques-
tion whether Mie’s theory of electrodynamics is valid or not. The
phase (or condition) of a physical system is described relatively
 THE GENERAL THEORY OF RELATIVITY 353

to a co-ordinate system by means of certain variable space-time

phase-quantities φ (these were our φi ’s above). Besides these, we
have also to take account of the metrical field in which the sys-
tem is embedded and which is characterised by its potentials gik .
The uniformity underlying the phenomena Z occurring in the sys-
tem is expressed by an invariant integral L dx; in it, the scalar-
density L is a function of the φ’s and of their derivatives of the
first and if need be, of the second order, and also a function of
the gik ’s, but the latter quantities alone and not their derivatives
occur in L. We form the total differential of the function L by
writing down explicitly only that part which contains the differ-
entials δgik , namely,

δL = 12 Tik δgik + δ0 L.

Tki is then the tensor-density of the energy (identical with mat-

ter) associated with the physical state or phase of the system. The
determination of its components is thus reduced once and for all to
a determination of Hamilton’s Function L. The general theory of
relativity alone, which allows the process of variation to be applied
to the metrical structure of the world, leads to a true definition
of energy. The phase-laws emerge from the “partial” principle of
action in which only the phase-quantities φ are to be subjected to
variation; just as many equations arise from it as there are quanti-
ties φ. The additional ten gravitational equations (15) for the ten
potentials gik result if we enlarge the partial principle of action to
the total one (27), in which the gik ’s are also to be subjected to
variation. The mechanical equations (26) are a consequence of
the phase-laws as well as of the gravitational laws; they may, in-
deed, be termed the eliminant of the latter. Hence, in the system
 CHAPTER IV 354

of phase and gravitational laws, there are four superfluous equa-

tions. The general solution must, in fact, contain four arbitrary
functions, since the equations, in virtue of their invariant charac-
ter, leave the co-ordinate system of the xi ’s indeterminate; hence,
arbitrary continuous transformations of these co-ordinates derived
from one solution of the equations always give rise to new solu-
tions in their turn. (These solutions, however, represent the same
objective course of the world.) The old subdivision into geometry,
mechanics, and physics must be replaced in Einstein’s Theory by
the separation into physical phases and metrical or gravitational
fields.
For the sake of completeness we shall once again revert to the
Hamiltonian Principle used in the theory of Lorentz and Maxwell.
Variation applied to the φi ’s gives the electromagnetic laws, but
applied to the gik ’s the gravitational laws. Since the Action is an
invariant, the infinitesimal change which an infinitesimal deforma-
tion of the world-continuum calls up in it = 0; this deformation is
to affect the electromagnetic and the gravitational field as well as
the world-lines of the substance-elements. This change consists of
three summands, namely, of the changes which are caused in turn
by the variation of the electromagnetic field, of the gravitational
field, and of the substance-paths. The first two parts are zero as
a consequence of the electromagnetic and the gravitational laws;
hence the third part also vanishes and we see that the mechanical
equations are a result of the two groups of laws mentioned just
above. Recapitulating our former calculations we may derive this
result by taking the following steps. From the gravitational laws
there follow (26), i.e.
 k 
∂Si 1 ∂gαβ αβ
µUi + ui M = − −2 S , (28)
∂xk ∂xi
 THE GENERAL THEORY OF RELATIVITY 355

in which Ski is the tensor-density of the electromagnetic energy of

field, namely, of
dui 1 ∂gαβ α β
Ui = −2 u u ,
ds ∂xi
and M is the left-hand member of the equation of continuity for
matter, namely
∂(µui )
M= .
∂xi
As a result of Maxwell’s equations the right-hand member of (28)

= pi = −Fik sk (si = ρui ).

If we then multiply (28) by ui and sum up with respect to i, we get

M = 0; in this way we have arrived at the equation of continuity
for matter and also at the mechanical equations in their usual
form.
After having gained a full survey of how the gravitational laws
of Einstein are to be arranged into the scheme of the remaining
physical laws, we are still faced with the task of working out the
explicit expression for the [G]ki ’s (vide note 6). The virtual change
 
r ik r
δΓik = δ = γik
r

of the components of the affine relationship is, as we know

(page 168), a tensor. If we use a geodetic co-ordinate system at a
certain point, then we get directly from the formula for Rik ((60),
§ 17) that
∂γ r ∂γ r
δRik = ik − ir
∂xr ∂xk
 CHAPTER IV 356

and
r k
ik ∂γikik ir ∂γik
g δRik = g −g .
∂xr ∂xr
If we set
g ik γik
r
− g ir γik
k
= wr
we get
∂wr
g ik δRik = ,
∂xr
or, for any arbitrary co-ordinate system,
√
ik 1 ∂( gwr )
δR = Rik δg + √ .
g ∂xr

The divergence disappears in the integration and hence, since

by definition we are to have
√
Z Z Z
δ R g dx = [G] δgik dx = − [G]ik δg ik dx
ik

and since the Rik ’s are symmetrical in Riemann’s space, we get

√
[G]ik = g( 12 gik R − Rik ) = 12 gik R − Rik ,
[G]ki = 12 δik R − Rki .

Therefore the gravitational laws are

Rki − 12 δik R = Tki (29)

Here, of course (exactly as was done for the unit of charge in elec-
tromagnetic equations), the unit of mass has been suitably chosen.
If we retain the units of the c.g.s. system, a universal constant 8πκ
 THE GENERAL THEORY OF RELATIVITY 357

will have to be added as a factor to the right-hand side. It might

still appear doubtful now at the outset whether κ is positive or
negative, and whether the right-hand side of equation (29) should
not be of opposite sign. We shall find, however, in the next para-
graph that, in virtue of the fact that masses attract one another
and do not repel, κ is actually positive.
It is of mathematical importance to notice that the exact
gravitational laws are not linear; although
  they are linear in
ik
the derivatives of the field-components , they are not linear
r
in the field-components themselves. If we contract equations (29),
that is, set k = i, and sum with respect to i, we get −R = T = Tii ;
hence, in place of (29) we may also write

Rki = Tki − 12 δik T. (30)

In the first paper in which Einstein set up the gravitational

equations without following on from Hamilton’s Principle, the
term − 21 δik T was missing on the right-hand side; he recognised
only later that it is required as a result of the energy-momentum-
theorem (vide note 7). The whole series of relations here described
and which is subject to Hamilton’s Principle, has become manifest
in further works by H. A. Lorentz, Hilbert, Einstein, Klein, and
the author (vide note 8).
In the sequel we shall find it desirable to know the value of G.
To convert
√
Z Z
R g dx into 2 G dx

by means of partial integration (that is, by detaching a diver-

 CHAPTER IV 358

gence), we must set

      
√ ik ∂ ik ∂ √ ik ik ik ∂ √ ik
gg = gg − ( gg ),
∂xr r ∂xr r r ∂xr
      
√ ik ∂ ir ∂ √ ik ir ir ∂ √ ik
gg = gg − ( gg ).
∂xk r ∂xk r r ∂xk
Thus we get
   
is ∂ √ ik ik ∂ √ ik
2G = ( gg ) − ( gg )
s ∂xk r ∂xr
     
ik rs ir ks √ ik
+ − gg .
r s s r

By (570 ), (5700 ) of § 17, however, the first two terms on the right,
√
if we omit the factor g,
        
is kr kr ik rs sk ik rs ik
=− g +2 g − g
s i r i r s
         
rs ik sk ri ik rs
= − +2 − g ik
s r r s r s
     
ir ks ik rs
= 2g ik − .
s r r s
Hence we finally arrive at
     
1 1 ik ir ks ik rs
√ G = 2g − . (31)
g s r r s

This completes our development of the foundations of Einstein’s

Theory of Gravitation. We must now inquire whether observation
 THE GENERAL THEORY OF RELATIVITY 359

confirms this theory which has been built up on purely speculative

grounds, and above all, whether the motions of the planets can
be explained just as well (or better) by it as by Newton’s law of
attraction. §§ 29–32 treat of the solution of the gravitational equa-
tions. The discussion of the general theory will not be resumed
till § 33.

§ 29. The Stationary Gravitational Field—Comparison

with Experiment
To establish the relationship of Einstein’s laws with the results
of observations of the planetary system, we shall first specialise
them for the case of a stationary gravitational field (vide note 9).
The latter is characterised by the circumstance that, if we use
appropriate co-ordinates, the world resolves into space and time,
so that for the metrical form
3
X
ds2 = f 2 dt2 − dσ 2 , dσ 2 = γik dxi dxk ,
i,k=1

we get

g00 = f 2 ; g0i = gi0 = 0; gik = −γik (i, k = 1, 2, 3),

and also that the co-efficients f and γ ik occurring in it depend

only on the space-co-ordinates x1 , x2 , x3 , and not on the time
t = x0 . dσ 2 is a positive definite quadratic differential form which
determines the metrical nature of the space having co-ordinates
x1 , x2 , x3 ; f is obviously the velocity of light. The measure t of
time is fully determined (when the unit of time has been chosen)
 CHAPTER IV 360

by the postulates that have been set up, whereas the space co-
ordinates x1 , x2 , x3 are fixed only to the extent of an arbitrary con-
tinuous transformation of these co-ordinates among themselves.
In the statical case, therefore, the metrics of the world gives, be-
sides the measure-determination of the space, also a scalar field f
in space.
If we denote the Christoffel 3-indices symbol, relating to the
ternary form dσ 2 , by an appended ∗, and if the index letters i, k, l
assume only the values 1, 2, 3 in turn, then it easily follows from
definition that
   ∗
ik ik
= ,
l l
     
ik 0i 00
= 0, = 0, = 0,
0 k 0
   
i0 fi 00
= , = ff i .
0 f 0
∂f
In the above, fi = are co-variant components of the three-
∂xi
dimensional gradient, and f i = γ ik fk are the corresponding
√
contra-variant components, whereas γf i = f i are the com-
ponents of a contra-variant vector-density in space. For the
√ √
determinant γ of the γik ’s we have g = f γ. If we further set
 ∗  ∗
∂fi ik ∂ 2f ik ∂f
fik = − fr = −
∂xk r ∂xi ∂xk r ∂xr
(the summation letter r also assumes only the three values 1, 2, 3),
and if we also set
∂f √
∆f = ∆f = γ · fii ),
∂xi
 THE GENERAL THEORY OF RELATIVITY 361

we arrive by an easy calculation at the following relations between

the components Rik and Pik of the curvature tensor of the second
order which belongs to the quadratic groundform ds2 for dσ 2 ,
respectively
fik
Rik = Pik − ,
f
Ri0 = R0i = 0,
∆f
R00 = f · √ (R00 = ∆f ).
γ
For statical matter which is non-coherent (i.e. of which the parts
do not act on one another by means of stresses), T00 = µ is the only
component of the energy-density tensor that is not zero; hence
T = µ. Matter at rest produces a statical gravitational field.
Among the gravitational
0

equations (30) the only one that is of
interest to us is the 0 th: it gives us
∆f = 12 µ (32)
or, if we insert the constant factor of proportionality 8πκ, we get
∆f = 4πκµ. (320 )
If we assume that, for an appropriate choice of the space-co-
ordinates x1 , x2 , x3 , ds2 differs only by an infinitesimal amount
from
c2 dt2 − (dx21 + dx22 + dx23 ) (33)
—the masses producing the gravitational field must be infinitely
small if this is to be true—we get, by setting
Φ
f =c+ , (34)
c
 CHAPTER IV 362

that
∂ 2Φ ∂ 2Φ ∂ 2Φ
∆Φ = + + = 4πκcµ, (10)
∂x21 ∂x22 ∂x23
and µ is c-times the mass-density in the ordinary units. We find
that actually, according to all our geometric observations, this
assumption is very approximately true for the planetary system.
Since the masses of the planets are very small compared with
the mass of the sun which produces the field and is to be con-
sidered at rest, we may treat the former as “test-bodies” that are
embedded in the gravitational field of the sun. The motion of
each of them is then given by a geodetic world-line in this statical
gravitational field, if we neglect the disturbances due to the influ-
ence of the planets on one another. The motion thus satisfies the
principle of variation Z
δ ds = 0,

the ends of the portion of world-line remaining fixed. For the case
of rest, this gives us
Z p
δ f 2 − v 2 dt = 0,

in which  2 3
2 dσ X dxi dxk
v = = γik
dt i,k=1
dt dt

is the square of the velocity. This is a principle of variation of the

same form as that of classical mechanics; the “Lagrange Function”
in this case is p
L = f 2 − v2.
 THE GENERAL THEORY OF RELATIVITY 363

If we make the same approximation as just above and notice that

in an infinitely weak gravitational field the velocities that occur
will also be infinitely small (in comparison with c), we get
p √ 1
f 2 − v 2 = c2 − 2Φ − v 2 = c + (Φ − 12 v 2 ),
c
and since we may now set
3  2 X
2
X dxi
v = = ẋ2i ,
i,k=1
dt i

we arrive at Z ( X )
δ 1
2
ẋ2i − Φ dt = 0;
i

that is, the planet of mass m moves according to the laws of clas-
sical mechanics, if we assume that a force with the potential mΦ
acts in it. In this way we have linked up the theory with
that of Newton: Φ is the Newtonian potential that satisfies Pois-
son’s equation (10), and k = c2 κ is the gravitational constant of
Newton. From the well-known numerical value of the Newtonian
constant k, we get for 8πκ the numerical value
8πk
8πκ = 2
= 1.87 · 10−27 cm · gr−1 .
c
The deviation of the metrical groundform from that of Euclid (33)
is thus considerable enough to make the geodetic world-lines differ
from rectilinear uniform motion by the amount actually shown
by planetary motion—although the geometry which is valid in
space and is founded on dσ 2 differs only very little from Euclidean
 CHAPTER IV 364

geometry as far as the dimensions of the planetary system are

concerned. (The sum of the angles in a geodetic triangle of these
dimensions differs very very slightly from 180◦ .) The chief cause
of this is that the radius of the earth’s orbit amounts to about
eight light-minutes whereas the time of revolution of the world in
its orbit is a whole year!
We shall pursue the exact theory of the motion of a point-mass
and of light-rays in a statical gravitational field a little further
(vide note 10). According to § 17 the geodetic world-lines may be
characterised by the two principles of variation
dxi dxk
Z p Z
δ Q ds = 0 or δ Q ds = 0, in which Q = gik . (35)
ds ds
The second of these takes for granted that the parameter s has
been chosen suitably. The second alone is of account for the “null-
lines” which satisfy the condition Q = 0 and depict the progress
of a light-signal. The variation must be performed in such a way
that the ends of the piece of world-line under consideration remain
unchanged. If we subject only x0 = t to variation, we get in the
statical case
   
2 dx0 d 2 dx0
Z Z
δ Q ds = 2f δx0 − 2 f δx0 ds. (36)
ds ds ds
Thus we find that
dx0
f2 = const. holds.
ds
If, for the present, we keep our attention fixed on the case of the
light-ray, we can, by choosing the unit of measure of the parame-
ter s appropriately (s is standardised by the principle of variation
 THE GENERAL THEORY OF RELATIVITY 365

itself except for an arbitrary unit of measure), make the constant

which occurs on the right equal to unity. If we now carry out the
variation more generally by varying the spatial path of the ray
whilst keeping the ends fixed but dropping the subsidiary condi-
tion imposed by time, namely, that δx0 = 0 for the ends, then, as
is evident from (36), the principle becomes
Z Z
δ Q ds = 2[δt] = 2δ dt.

If the path after variation is, in particular, traversed with the

velocity of light just as the original path, then for the varied world-
line, too, we have

Q = 0, dσ = f dt,

and we get
dσ
Z Z
δ dt = δ = 0. (37)
f
This equation fixes only the spatial position of the light-ray; it is
nothing other than Fermat’s principle of the shortest path.
In the last formulation time has been eliminated entirely; it is
valid for any arbitrary portion of the path of the light-ray if the
latter alters its position by an infinitely small amount, its ends
being kept fixed.
If, for a statical field of gravitation, we use any space-co-
ordinates x1 , x2 , x3 , we may construct a graphical representation
of a Euclidean space by representing the point whose co-ordinates
are x1 , x2 , x3 by means of a point whose Cartesian co-ordinates are
x1 , x2 , x3 . If we mark the position of two stars S1 , S2 which are at
rest and also an observer B, who is at rest, in this picture-space,
 CHAPTER IV 366

then the angle at which the stars appear to the observer is not
equal to the angle between the straight lines BS1 , BS2 connect-
ing the stars with the observer; we must connect B with S1 , S2
by means of the curved lines of shortest path resulting from (37)
and then, by means of an auxiliary construction, transform the
angle which these two lines make with one another at B from Eu-
clidean measure to that of Riemann determined by the metrical
groundform dσ 2 (cf. formula (15), § 11). The angles which have
been calculated in this way are those which determine the actually
observed position of the stars to one another, and which are read
off on the divided circle of the observing instrument. Whereas
B, S1 , S2 retain their positions in space, this angle S1 BS2 may
change, if great masses happen to get into proximity of the path of
the rays. It is in this sense that we may talk of light-rays being
curved as a result of the gravitational field. But the rays are
not, as we assumed in § 12 to get at general results, geodetic lines
in space Zwith the metrical groundform dσ 2 ; they do not make the
dσ
Z
integral dσ but assume a limiting value. The bending of
f
light-rays occur, in particular, in the gravitational field of the sun.
If for our graphical representation we use co-ordinates x1 , x2 , x3 ,
for which the Euclidean formula dσ 2 = dx21 +dx22 +dx23 holds at in-
finity, then numerical calculation for the case of a light-ray passing
by close to the sun shows that it must be diverted from its path to
the extent of 1.74 seconds (vide § 31). This entails a displacement
of the positions of the stars in the apparent immediate neighbour-
hood of the sun, which should certainly be measurable. These
positions of the stars can be observed, of course, only during a
total eclipse of the sun. The stars which come into consideration
must be sufficiently bright, as numerous as possible, and suffi-
 THE GENERAL THEORY OF RELATIVITY 367

ciently close to the sun to lead to a measurable effect, and yet

sufficiently far removed to avoid being masked by the brilliance of
the corona. The most favourable day for such an observation is
the 29th May, and it was a piece of great good fortune that a total
eclipse of the sun occurred on the 29th May, 1919. Two English
expeditions were dispatched to the zone in which the total eclipse
was observable, one to Sobral in North Brazil, the other to the Is-
land of Principe in the Gulf of Guinea, for the express purpose of
ascertaining the presence or absence of the Einstein displacement.
The effect was found to be present to the amount predicted; the
final results of the measurements were 1.9800 ± 0.1200 for Sobral,
1.6100 ± 0.3000 for Principe (vide note 11).
Another optical effect which should present itself, according
to Einstein’s theory of gravitation, in the statical field and which,
under favourable conditions, may just be observable, arises from
the relationship
ds = f dt
holding between the cosmic time dt and the proper-time ds at a
fixed point in space. If two sodium atoms at rest are objectively
fully alike, then the events that give rise to the light-waves of
the D-line in each must have the same frequency, as measured in
proper-time. Hence, if f has the values f1 , f2 , respectively at
the points at which the atoms are situated, then between f1 , f2
and the frequencies ν1 , ν2 in cosmic time, there will exist the
relationship
ν1 ν2
= .
f1 f2
But the light-waves emitted by an atom will have, of course, the
same frequency, measured in cosmic time, at all points in space
 CHAPTER IV 368

(for, in a static metrical field, Maxwell’s equations have a solu-

tion in which time is represented by the factor eiνt , ν being an
arbitrary constant frequency). Consequently, if we compare the
sodium D-line produced in a spectroscope by the light sent from a
star of great mass with the same line sent by an earth-source into
the same spectroscope, there should be a slight displacement of
the former line towards the red as compared with the latter, since
f has a slightly smaller value in the neighbourhood of great masses
than at a great distance from them. The ratio in which the fre-
quency is reduced, has according to our approximate formula (34)
κm0
the value 1− at the distance r from a mass m0 . At the surface
r
of the sun this amounts to a displacement of .008 Angströms for
a line in the blue corresponding to the wave-length 4000 Å. This
effect lies just within the limits of observability. Superimposed on
this, there are the disturbances due to the Doppler effect, the un-
certainty of the means used for comparison on the earth, certain
irregular fluctuations in the sun’s lines the causes of which have
been explained only partly, and finally, the mutual disturbances
of the densely packed lines of the sun owing to the overlapping of
their intensities (which, under certain circumstances, causes two
lines to merge into one with a single maximum of intensity). If all
these factors are taken into consideration, the observations that
have so far been made, seem to confirm the displacement towards
the red to the amount stated (vide note 12). This question cannot,
however, yet be considered as having been definitely answered.
A third possibility of controlling the theory by means of ex-
periment is this. According to Einstein, Newton’s theory of the
planets is only a first approximation. The question suggests it-
self whether the divergence between Einstein’s Theory and the
 THE GENERAL THEORY OF RELATIVITY 369

latter are sufficiently great to be detected by the means at our

disposal. It is clear that the chances for this are most favourable
for the planet Mercury which is nearest the sun. In actual fact,
after Einstein had carried the approximation a step further, and
after Schwarzschild (vide note 13) had determined accurately the
radially symmetrical field of gravitation produced by a mass at
rest and also the path of a point-mass of infinitesimal mass, both
found that the elliptical orbit of Mercury should undergo
a slow rotation in the same direction as the orbit is tra-
versed (over and above the disturbances produced by the remain-
ing planets), amounting to 4300 per century. Since the time of
Leverrier an effect of this magnitude has been known among the
secular disturbances of Mercury’s perihelion, which could not be
accounted for by the usual causes of disturbance. Manifold hy-
potheses have been proposed to remove this discrepancy between
theory and observation (vide note 14). We shall revert to the
rigorous solution given by Schwarzschild in § 31.
Thus we see that, however great is the revolution produced in
our ideas of space and time by Einstein’s theory of gravitation,
the actual deviations from the old theory are exceedingly small in
our field of observation. Those which are measurable have been
confirmed up to now. The chief support of the theory is to be
found less in that lent by observation hitherto than in its inherent
logical consistency, in which it far transcends that of classical me-
chanics, and also in the fact that it solves the perplexing problem
of gravitation and of the relativity of motion at one stroke in a
manner highly satisfying to our reason.
Using the same method as for the light-ray, we may set up for
the motion of a point-mass in a statical gravitational field a “min-
imum” principle affecting only the path in space, corresponding
 CHAPTER IV 370

to Fermat’s principle of the shortest path. If s is the parameter

of proper-time, then
dt 1
Q = 1, and f 2 = const. = (38)
ds E
is the energy-integral. We now apply the first of the two principles
of variation (35) and generalise it as above by varying the spatial
path quite arbitrarily while keeping the ends, x0 = t, fixed. We
get  
1 dt
Z p Z
δ Q ds = δt = δ . (39)
E E
To eliminate the proper-time we divide the first of the equa-
tions (38) by the square of the second; the result is
(  2 )
1 2 dσ 2 2
√
f − = E dσ = f U dt, (40)
f4 dt
in which
1
U= 2
− E 2.
f
(40) is the law of velocity according to which the point-mass tra-
verses its path. If we perform the variation so that the varied path
is traversed according to the same law with the same constant E,
it follows from (39) that
s  2
dt dσ
Z Z Z Z
δ =δ 2
f − dt = δ Ef dt i.e. δ f 2 U dt = 0
2
E dt
or, finally, by expressing dt in terms of the spatial element of
arc dσ, and thus eliminating the time entirely, we get
Z √
δ U dσ = 0.
 THE GENERAL THEORY OF RELATIVITY 371

The path of the point-mass having been determined in this way,

we get as a relation giving the time of the motion in this path,
from (40), that
dσ
dt = √ .
f2 U
For E = 0, we again get the laws for the light-ray.

§ 30. Gravitational Waves

By assuming that the generating energy-field Tki is infinitely
weak, Einstein has succeeded in integrating the gravitational equa-
tions generally (vide note 15). The gik ’s will, under these circum-
0
stances, if the co-ordinates are suitably chosen, differ from the g ik ’s
by only infinitesimal amounts γik . We then regard the world as
“Euclidean,” having the metrical groundform
0
g ik dxi dxk (41)
and the γik ’s as the components of a symmetrical tensor-field of
the second order in this world. The operations that are to be per-
formed in the sequel will always be based on the metrical ground-
form (41). For the present we are again dealing with the spe-
cial theory of relativity. We shall consider the co-ordinate system
0
which is chosen to be a “normal” one, so that g ik = 0 for i 6= k
and
0 0 0
g00 = 1, g 11 = g 22 = g 33 = −1.
x0 is the time, x1 , x2 , x3 are Cartesian space-co-ordinates; the
velocity of light is taken equal to unity.
We introduce the quantities
ψik = γik − γδik (γ = 12 γii ),
 CHAPTER IV 372

and we next assert that we may without loss of generality set

∂ψik
= 0. (42)
∂xk
For, if this is not so initially, we may, by an infinitesimal change,
alter the co-ordinate system so that (42) holds. The transforma-
tion formulæ that lead to a new co-ordinate system x̄, namely,

x̄i = xi + ξ(x0 , x1 , x2 , x3 )

contain the unknown functions ξ i , which are of the same order of

infinitesimals as the γ’s. We get new co-efficients ḡik for which,
according to earlier formulæ, we must have
∂ξ r ∂ξ r ∂gik r
gik (x) − ḡik (x) = gir + gkr + ξ
∂xk ∂xi ∂xr
so that, here, we have

∂ξi ∂ξk ∂ξ i
γik (x) − γ̄ik (x) = + , γ(x) − γ̄(x) = = Ξ,
∂xk ∂xi ∂xi
and we finally get

∂γik ∂γ̄ik ∂Ξ ∂γ ∂γ̄ ∂Ξ

− = ∇ξi + , − = ,
∂xk ∂xk ∂xi ∂xi ∂xi ∂xi
in which ∇ denotes, for an arbitrary function, the differential op-
erator
∂ 2f
 2
∂ 2f ∂ 2f
  
∂ 0 ∂f ∂ f
∇f = g ik = − + + .
∂xi ∂xk ∂x20 ∂x21 ∂x22 ∂x23
 THE GENERAL THEORY OF RELATIVITY 373

The desired condition will therefore be fulfilled in the new co-

ordinate system if the ξ i ’s are determined from the equations
∂ψik
∇ξ i = ,
∂xk
which may be solved by means of retarded potentials (cf. Chap-
ter III, page 244). If the linear Lorentz transformations are dis-
carded, the co-ordinate system is defined not only to the first order
of small quantities but also to the second. It is very remarkable
that such an invariant normalisation is possible.
  the components Rik of curvature. As the
We now calculate
ik
field-quantities are infinitesimal, we get, by confining our-
r
selves to terms of the first order
   
∂ ik ∂ ir
Rik = − .
∂xr r ∂xk r
Now,    
ik 1 ∂γir ∂γkr ∂γik
=2 + − ,
r ∂xk ∂xi ∂xr
hence  r
∂γkr 0 ∂γik
  
ik 1 ∂γi
=2 + − g rs .
r ∂xk ∂xi ∂xs
Taking into account equations (42) or
∂γik ∂γ
= ,
∂xk ∂xi
we get
∂ 2γ
 
∂ ik
= − 1 ∇γik .
∂xr r ∂xi ∂xk 2
 CHAPTER IV 374

In the same way we obtain

∂ 2γ
 
∂ ir
= .
∂xk r ∂xi ∂xk
The result is
Rik = − 12 ∇γik .
Consequently, R = −∇γ and
Rik − 12 δik R = − 21 ∇ψik .
The gravitational equations are, however,
1
2
∇ψik = −Tik , (43)
and may be directly integrated with the help of retarded potentials
(cf. page 244). Using the same notation, we get
Z k
k Ti (t − r)
ψi = − dV.
2πr
Accordingly, every change in the distribution of matter produces a
gravitational effect which is propagated in space with the velocity of
light. Oscillating masses produce gravitational waves. Nowhere in
the Nature accessible to us do mass-oscillations of sufficient power
occur to allow the resulting gravitational waves to be observed.
Equations (43) correspond fully to the electromagnetic equa-
tions
∇φi = si
and, just as the potentials φi of the electric field had to satisfy the
secondary condition
∂φi
=0
∂xi
 THE GENERAL THEORY OF RELATIVITY 375

because the current si fulfils the condition

∂si
= 0,
∂xi
so we had here to introduce the secondary conditions (42) for the
system of gravitational potentials ψik , because they hold for the
matter-tensor
∂Tik
= 0.
∂xk
Plane gravitational waves may exist: they are propagated
in space free from matter: we get them by making the same sup-
position as in optics, i.e. by setting
√
−1
ψik = aki · e(α0 x0 +α1 x1 +α2 x2 +α3 x3 ) .

The aki ’s and the αi ’s are constants; the latter satisfy the condition
αi αi = 0. Moreover, α0 = ν is the frequency of the vibration and
α1 x1 + α2 x2 + α3 x3 = const. are the planes of constant phase.
The differential equations ∇ψik = 0 are satisfied identically. The
secondary conditions (42) require that

aki αk = 0. (44)

If the x1 -axis is the direction of propagation of the wave, we have

α2 = α3 = 0, −α1 = α0 = ν,
and equations (44) state that

a0i = a1i or a0i = −a1i . (45)

 CHAPTER IV 376

Accordingly, it is sufficient to specify the space part of the constant

symmetrical tensor a, namely,

a11 a12 a13

a21 a22 a23

a31 a32 a33

since the a’s with the index 0 are determined from these by (45);
the space part, however, is subject to no limitation. In its turn it
splits up into the three summands in the direction of propagation
of the waves:

a11 0 0 0 a12 a13 0 0 0

0 0 0 + a21 0 0 + 0 a22 a23 .


0 0 0 a31 0 0 0 a32 a33
The tensor-vibration may hence be resolved into three indepen-
dent components: a longitudinal-longitudinal, a longitudinal-
transverse, and a transverse-transverse wave.
H. Thirring has made two interesting applications of integra-
tion based on the method of approximation used here for the
gravitational equations (vide note 16). With its help he has in-
vestigated the influence of the rotation of a large, heavy, hollow
sphere on the motion of point-masses situated near the centre of
the sphere. He discovered, as was to be expected, a force effect
of the same kind as centrifugal force. In addition to this a second
force appears which seeks to drag the body into the equatorial
plane according to the same law as that according to which cen-
trifugal force seeks to drive it away from the axis. Secondly (in
conjunction with J. Lense), he has studied the influence of the
rotation of a central body on its planets or moons, respectively.
In the case of the fifth moon of Jupiter, the disturbance caused
 THE GENERAL THEORY OF RELATIVITY 377

attains an amount that may make it possible to compare theory

with observation.
Now that we have considered in §§ 29, 30 the approximate
integration of the gravitational equations that occur if only linear
terms are taken into account, we shall next endeavour to arrive
at rigorous solutions: our attention will, however, be confined to
statical gravitation.

§ 31. Rigorous Solution of the Problem of One Body∗

For a statical gravitational field we have

ds2 = f 2 dx20 − dσ 2

in which dσ 2 is a definitely positive quadratic form in the three-

space variables x1 , x2 , x3 ; the velocity of light f is likewise de-
pendent only on these. The field is radially symmetrical if,
for a proper choice of the space-co-ordinates, f and dσ 2 are in-
variant with respect to linear orthogonal transformations of these
co-ordinates. If this is to be the case, f must be a function of the
distance q
r = x21 + x22 + x23 ,
from the centre, but dσ 2 must have the form

λ(dx21 + dx22 + dx23 ) + l(x1 dx1 + x2 dx2 + x3 dx3 )2 (46)

in which λ and l are likewise functions of r alone. Without dis-

turbing this normal form we may subject the space-co-ordinates
to a further transformation which consists in replacing x1 , x2 , x3
∗
Vide note 17.
 CHAPTER IV 378

by τ x1 , τ x2 , τ x3 , the factor of proportionality τ being an arbitrary

function of the distance r. By choosing λ appropriately we may
clearly succeed in getting λ = 1; let us suppose this to have been
done. Then, using the notation of § 29, we have
γik = −gik = δik + l · xi xk (i, k = 1, 2, 3).
We shall next define this radially symmetrical field so that it
satisfies the homogeneous gravitational equations which hold
wherever there is no matter, that is, wherever the energy-
density Tki vanishes. These equations are all included in the
principle of variation Z
δ G dx = 0.

The gravitational field, which we are seeking, is that which is

produced by statical masses which are distributed about
the centre with radial symmetry. If the accent signify differ-
entiation with respect to r, we get
∂γik xα
= l0 xi xk + l(δiα xk + δkα xi ),
∂xα r
and hence
 
ik xα 0
− = 12 l xi xk + lδik xα (i, k, α = 1, 2, 3).
α r
Since it follows from
3
X
xα = γαβ xβ
β=1

that
1
xα = xα and h2 = 1 + lr2 ,
h2
 THE GENERAL THEORY OF RELATIVITY 379

as may be verified by direct substitution, we must have

0 k
 
ik 1 xα l xi xk + 2lrδi )
=2 .
α r h2
It is sufficient to carry out the calculation of G for the point
x1 = r, x2 = 0, x3 = 0. At this point, we get for the three-indices
symbols just calculated:
h0
     
11 22 33 lr
= and = = 2,
1 h 1 1 h
whereas the remaining ones are equal to zero. Of the three-indices
symbols containing 0, we find by § 29 that
f0 ff 0
     
10 01 00
= = and = 2,
0 0 f 1 h
whereas all the others = 0. Of the gik ’s all those situated in the
main diagonal (i = k) are equal, respectively, to
f 2, −h2 , −1, −1
whereas the lateral ones all vanish. Hence definition (31) of G
gives us
2
−√ G =
g
        
1 00 10 11 01 00
+ − 2
f 2  1   0   1   0 1   
1 11 10 11 10 10 11 11
− 2 + − −
h  1   0   1  0 0 1 1
22 10 11
−1 +
 1   0   1 
33 10 11
−1 + .
1 0 1
 CHAPTER IV 380

The terms in the first and second row taken together lead to
        
11 10 1 00 1 10
− − 2 .
1 0 f2 1 h 0
The second factor in this product, however, is equal to zero. Since,
by (57) § 17
3  
X 1i ∆0 √
= (∆ = g = hf ),
i=0
i ∆

the sum of the terms in the third and fourth row is equal to
2lr ∆0
− · .
h2 ∆
If we wish to take the world-integral G over a fixed interval with
respect to the time x0 , and over a shell enclosed by two spherical
surfaces with respect to space, then, since the element of integra-
tion is
dx = dx0 · dΩ · r2 dr (dΩ = solid angle),
the equation of variation that is to be solved is
Z
δ Gr2 dr = 0.

Hence, if we set
lr3 lr3
 
1
= = 1− 2 r = w,
h2 1 + lr2 h
we get Z
δ w∆0 dr = 0
 THE GENERAL THEORY OF RELATIVITY 381

in which ∆ and w may be regarded as the two functions that may

be varied arbitrarily.
By varying w, we get

∆0 = 0, ∆ = const.

and hence, if we choose the unit of time suitably

∆ = hf = 1.

Partial integration gives

Z Z
w∆ dr = [w, ∆] − ∆w0 dr.
0

Hence, if we vary ∆, we arrive at

w0 = 0, w = const. = 2m.

Finally, from the definition of w and ∆ = 1, we get

2m 1
f2 = 1 − , h2 =
r f2

This completes the solution of the problem. The unit of time has
been chosen so that the velocity of light at infinity = 1. For dis-
tances r, which are great compared with m, the Newtonian value
of the potential holds in the sense that the quantity m0 , intro-
duced by the equation m = κm0 occurs as the field-producing
mass in it; we call m the gravitational radius of the matter
causing the disturbance of the field. Since 4πm is the flux of the
 CHAPTER IV 382

spatial vector-density f i through an arbitrary sphere enclosing the

masses, we get, from (320 ), for discrete or non-coherent mass
Z
m0 = µ dx1 dx2 dx3 .

Since f 2 cannot become negative, it is clear from this that, if we

use the co-ordinates here introduced for the region of space devoid
of matter, r must be > 2m. Further light is shed on this by the
special case of a sphere of liquid which is to be discussed in § 32,
and for which the gravitational field inside the mass, too, will be
determined. We may apply the solution found to the gravitational
field of the sum external to itself if we neglect the effect due to the
planets and the distant stars. The gravitational radius is about
1.47 kilometres for the sun’s mass, and only 5 millimetres for the
earth.
The motion of a planet (supposed infinitesimal in comparison
with the sun’s mass) is represented by a geodetic world-line. Of
its four equations
d 2 xi
 
αβ dxα dxβ
2
+ = 0,
ds i ds ds
the one corresponding to the index i = 0 gives, for the statical
gravitational field, the energy-integral
dx0
f2 = const.
ds
as we saw above; or, since,
 2  2
dx0 dσ
f =1+ ,
ds ds
 THE GENERAL THEORY OF RELATIVITY 383

we get "  2 #
dσ
f2 1 + = const.
ds
In the case of a radially symmetrical field the equations corre-
sponding to the indices i = 1, 2, 3 give the proportion
d 2 x 1 d 2 x2 d 2 x3
: : = x1 : x2 : x3
ds2 ds2 ds2
(this is readily seen from the three-indices symbols that are written
down). And from them, there results, in the ordinary way, the
three equations which express the Law of Areas
dx2 dx1
...................., x1
− x2 = const.
ds ds
This theorem differs from the similar one derived in Newton’s
Theory, in that the differentiations are made, not according to
cosmic time, but according to the proper-time s of the planet. On
account of the Law of Areas the motion takes place in a plane that
we may choose as our co-ordinate plane x3 = 0. If we introduce
polar co-ordinates into it, namely
x1 = r cos φ, x2 = r sin φ,
the integral of the area is
dφ
r2 = const. = b. (47)
ds
The energy-integral, however, since
dx21 + dx22 = dr2 + r2 dφ2 , x1 dx1 + x2 dx2 = r dr,
dσ = (dr + r dφ ) + l(r dr)2 = h2 dr2 + r2 dφ2 ,
2 2 2 2
 CHAPTER IV 384

becomes
(  2  2 )
2 2 dr 2 dφ
f 1+h +r = const.
ds ds

Since f h = 1, we get, by substituting for f 2 its value, that

 2  2
2m dr dφ
− + + r(r − 2m) = −E = const. (48)
r ds ds
Compared with the energy-equation of Newton’s Theory this
equation differs from it only in having r − 2m in place of r in the
last term of the left-hand side.
The succeeding steps are the same as those of Newton’s The-
dφ
ory. We substitute from (47) into (48), getting
ds
 2
dr 2m b2 (r − 2m)
= −E− ,
ds r r3
1
or, using the reciprocal distance ρ = in place of r,
r
 2
dρ
2
= 2mρ − E − b2 ρ2 (1 − 2mρ).
ρ ds
To arrive at the orbit of the planet we eliminate the proper-time
by dividing this equation by the square of (47), thus
 2
dρ 2m E
= 2 ρ − 2 − ρ2 + 2mρ3 .
dφ b b
In Newton’s Theory the last term on the right is absent. Taking
into account the numerical conditions that are presented in the
 THE GENERAL THEORY OF RELATIVITY 385

case of planets, we find that the polynomial of the third degree

in ρ on the right has three positive roots ρ0 > ρ1 > ρ2 and hence
= 2m(ρ0 − ρ)(ρ1 − ρ)(ρ − ρ2 );
ρ assumes values ranging between ρ1 and ρ2 . The root ρ0 is very
great in comparison with the remaining two. As in Newton’s
Theory, we set
1 1
= a(1 − e), = a(1 + e),
ρ1 ρ2
and call a the semi-major axis and e the eccentricity. We then get
2
ρ1 + ρ2 = .
a(1 − e2 )
If we compare the co-efficients of ρ2 with one another, we find that
1
ρ0 + ρ1 + ρ2 = .
2m
φ is expressed in terms of ρ by an elliptic integral of the first kind
and hence, conversely, ρ is an elliptic function of φ. The motion
is of precisely the same type as that executed by the spherical
pendulum. To arrive at simple formulæ of approximation, we
make the same substitution as that used to determine the Kepler
orbit in the Newtonian Theory, namely
ρ1 + ρ2 ρ1 − ρ2
ρ− + cos θ.
2 2
Then
dθ
Z
φ= s  . (49)
ρ1 + ρ2 ρ1 − ρ2
2m ρ0 − − cos θ
2 2
 CHAPTER IV 386

The perihelion is characterised by the values θ = 0, 2π, . . . . The

increase of the azimuth φ after a full revolution from perihelion to
perihelion is furnished by the above integral, taken between the
limits 0 and 2π. With easily sufficient accuracy this increase may
be set
2π
=s  .
ρ1 + ρ2
2m ρ0 −
2
We find, however, that
ρ1 + ρ2 1 3
ρ0 + = (ρ0 + ρ1 + ρ2 ) − 23 (ρ1 + ρ2 ) = − .
2 2m a(1 − e2 )
Consequently the above increase (of azimuth)
 
2π 3m
=r ∼ 2π 1 + ,
6m a(1 − e2 )
1−
a(1 − e2 )
and the advance of the perihelion per revolution
6πm
= .
a(1 − e2 )
In addition, m, the gravitational radius of the sun may be ex-
pressed according to Kepler’s third law, in terms of the time of
revolution T of the planet and the semi-major axis a, thus
4π 2 a3
m= .
c2 T 2
Using the most delicate means at their disposal, astronomers have
hitherto been able to establish the existence of this advance of the
 THE GENERAL THEORY OF RELATIVITY 387

perihelion only in the case of Mercury, the planet nearest the sun
(vide note 18).
Formula (49) also gives the deflection α of the path of a ray of
π
light. If θ0 = +  is the angle θ for which ρ = 0, then the value
2
of the integral, taken between −θ0 and +θ0 = π + α. Now in the
present case
1
2m(ρ0 − ρ)(ρ1 − ρ)(ρ − ρ2 ) = − ρ2 + 2mρ3 .
b2
1
The values of ρ fluctuate between 0 and ρ2 . Moreover, = r is
ρ1
the nearest distance to which the light-ray approaches the centre
of mass O, whilst b is the distance of the two asymptotes of the
light-ray from O (for in the case of any curve, this distance is
dφ
given by the value of for ρ = 0). Now,
dρ

2m(ρ0 + ρ1 + ρ2 ) = 1
m
is accurately true. If is a small fraction, we get to a first degree
b
of approximation that
m m  m 2 m
mρ1 = −mρ2 = , (ρ1 + ρ2 ) = , = ,
b 2 b b
Z θ0
m 2m 4m
α= (1 + cos θ) dθ − π = 2 + and hence α =
−θ0 b b b

If we calculate the path of the light-ray according to Newton’s

Theory, taking into account the gravitation of light, that is, con-
sidering it as the path of a body that has the velocity c at infinity,
 CHAPTER IV 388

then if we set
1 2m
2
+ 2
ρ − ρ2 = (ρ1 − ρ)(ρ − ρ2 )
b b
in which ρ1 > 0, ρ2 < 0 and set
ρ1 + ρ2
cos θ0 = − ,
ρ1 − ρ2
we get
2m
π + α = 2θ0 , α∼ .
b
Thus Newton’s law of attraction leads to a deflection which is only
half as great as that predicted by Einstein. The observations made
at Sobral and Principe decide the question definitely in favour of
Einstein (vide note 19).

§ 32. Additional Rigorous Solutions of the Statical

Problem of Gravitation
In a Euclidean space with Cartesian co-ordinates x1 , x2 , x3 , the
equation of a surface of revolution having as its axis of rotation
the x3 -axis is
q
x3 = F (r), r = x21 + x22 .

On it, the square of the distance dσ between two infinitely near

points is

2
dσ 2 = (dx21 + dx22 ) + F 0 (r) dr2
 0 2
2 2 F (r)
= (dx1 + dx2 ) + (x1 dx1 + x2 dx2 )2 .
r
 THE GENERAL THEORY OF RELATIVITY 389

In a radially symmetrical statical gravitational field we have for a

plane (x3 = 0) passing through the centre

dσ 2 = (dx21 + dx22 ) + l(x1 dx1 + x2 dx2 )2

in which
h2 − 1 2m
l= 2
= 2 .
r r (r − 2m)
The two formulæ are identical if we set
r
0 2m p
F (r) = , F (r) = 8m(r − 2m).
r − 2m
The geometry which holds on this plane is therefore the same as
that which holds in Euclidean space on the surface of revolution
of a parabola p
z = 8m(r − 2m)
(vide note 20).
A charged sphere, besides calling up a radially symmetrical
gravitational field, calls up a similar electrostatic field. Since both
fields influence one another mutually, they may be determined
only conjointly and simultaneously (vide note 21). If we use the
ordinary units of the c.g.s. system (and not those of Heaviside
which dispose of the factor 4π in another way and which we have
generally used in the foregoing) for electricity as well as for the
other quantities, then in the region devoid of masses and charges
the integral becomes

Φ02 r2
Z  
0
w∆ − κ dr.
∆
 CHAPTER IV 390

It assumes a stationary value for the condition of equilibrium.

The notation is the same as above, Φ denoting the electrostatic
potential. The square of the numerical value of the field is used as
a basis for the function of Action of the electric field, in accordance
with the classical theory. Variation of w gives, just as in the case
of no charges,
∆0 = 0, ∆ = const. = c.
But variation of Φ leads to
d r2 Φ0
 
e0
= 0 and hence Φ = .
dr ∆ r

For the electrostatic potential we therefore get the same formula

as when gravitation is disregarded. The constant e0 is the electric
charge which excites the field. If, finally, ∆ be varied, we get

Φ02 r2
w0 − κ =0
∆2
and hence
 2
κ e2 1 f 2κm0 κ e2
w = 2m − 2 0 , = =1− + 2 02
c r h2 c r c r

in which m0 denotes the mass which produces the gravitational

field. In f 2 there occurs, as we see, in addition to the term depend-
ing on the mass, an electrical term which decreases more rapidly
as r increases. √We call m = κm0 the gravitational radius of the
κ
mass m0 , and e0 = e the gravitational radius of the charge e0 .
c
Our formula leads to a view of the structure of the elec-
tron which diverges essentially from the one commonly
 THE GENERAL THEORY OF RELATIVITY 391

accepted. A finite radius has been attributed to the electron;

this has been found to be necessary, if one is to avoid coming to
the conclusion that the electrostatic field it produces has infinite
total energy, and hence an infinitely great inertial mass. If the
inertial mass of the electron is derived from its field-energy alone,
then its radius is of the order of magnitude
e20
a= .
m0 c2
But in our formula a finite mass m0 (producing the gravitational
field) occurs quite independently of the smallness of the value of r
for which the formula is regarded as valid; how are these results to
be reconciled? According to Faraday’s view the charge enclosed
by a surface Ω is nothing more than the flux of the electrical
field through Ω. Analogously to this it will be found in the next
paragraph that the true meaning of the conception of mass, both
as field-producing mass and as inertial or gravitational mass, is
expressed by a field-flux. If we are to regard the statical solution
here given as valid for all space, the flux of the electrical field
through any sphere is 4πe0 at the centre. On the other hand the
mass which is enclosed by a sphere of radius r, assumes the value

1 e20
m0 − 2
c2 r
which is dependent on the value of r. The mass is consequently
distributed continuously. The density of mass coincides, of course,
with the density of energy. The “initial level” at the centre, from
which the mass is to be calculated, is not equal to 0 but to −∞.
Therefore the mass m0 of the electron cannot be determined from
this level at all, but signifies the “ultimate level” at an infinitely
 CHAPTER IV 392

great distance. a now signifies the radius of the sphere which

encloses the mass zero. Contrary to Mie’s view matter now ap-
pears as a real singularity of the field. In the general theory
of relativity, however, space is no longer assumed to be Euclidean,
and hence we are not compelled to ascribe to it the relationships
of Euclidean space. It is quite possible that it has other limits
besides infinity, and, in particular, that its relationships are like
those of a Euclidean space which contains punctures (cf. § 34). We
may, therefore, claim for the ideas here developed—according to
which there is no connection between the total mass of the elec-
tron and the potential of the field it produces, and in which there
is no longer a meaning in talking of a cohesive pressure holding
the electron together—equal rights as for those of Mie. An un-
satisfactory feature of the present theory is that the field is to be
entirely free of charge, whereas the mass (= energy) is to permeate
the whole of the field with a density that diminishes continuously.
√
It is to be noted that a : e = e : m or, that e = am. In
e
the case of the electron the quotient is a number of the order
m
a
of magnitude 1020 , of the order 1040 ; that is, the electric re-
m
pulsion which two electrons (separated by a great distance) exert
upon one another is 1040 times as great as that which they exert
in virtue of gravitation. The circumstance that in an electron an
integral number of this kind occurs which is of an order of mag-
nitude varying greatly from unity makes the thesis contained in
Mie’s Theory, namely, that all pure figures determined from the
measures of the electron must be derivable as mathematical con-
stants from the exact physical laws, rather doubtful: on the other
hand, we regard with equal scepticism the belief that the struc-
ture of the world is founded on certain pure figures of accidental
 THE GENERAL THEORY OF RELATIVITY 393

numerical value.
The gravitational field that is present in the interior of mas-
sive bodies is, according to Einstein’s Theory, determined only
when the dynamical constitution of the bodies are fully known;
since the mechanical conditions are included in the gravitational
equations, the conditions of equilibrium are given for the statical
case. The simplest conditions that offer themselves for consider-
ation are given when we deal with bodies that are composed of
a homogeneous incompressible fluid. The energy-tensor of a
fluid on which no volume forces are acting is given according to
§ 25, by
Tik = µ∗ ui uk − pgik
in which the ui ’s are co-variant components of the world-direction
of the matter, the scalar p denotes the pressure, and µ∗ is de-
termined from the constant density µ0 by means of the equation
µ∗ = µ0 + p. We introduce the quantities
µ∗ u i = v i
as independent variables, and set
1 p
L = √ L = µ0 − v i v i .
g

Then, if we vary only the g ik ’s, not the vi ’s,

dL = − 12 Tik δgik .
Consequently, by referring these equations to this kind of varia-
tion, we may epitomise them in the formula
Z
δ (L + G) dx = 0.
 CHAPTER IV 394

It must carefully be noted, however, that, if the vi ’s are varied as

independent variables in this principle, it does not lead to the cor-
vi
rect hydrodynamical equations (instead, we should get √ i = 0,
vi v
which leads to nowhere). But these conservation theorems of en-
ergy and momentum, are already included in the gravitational
equations.
In the statical case, v1 = v2 = v3 = 0, and all quantities are
independent of the time. We set v0 = v and apply the symbol of
variation δ just as in § 28 to denote a change that is produced by an
infinitesimal deformation (in this case a pure spatial deformation).
Then  
1 ik ∆
δL = 2 T δgik − h δv h=
f
in which δv denotes nothing more than the difference of v at two
points in space that are generated from one another as a result of
the displacement. By now arguing backwards from the conclusion
which gave us the energy-momentum theorem in § 28, we infer
from this theorem, namely
Z
Tik δgik dx = 0,

and from the equation

Z
δL dx = 0,

which expresses the invariant character of the world-integral of L,

that δv = 0. This signifies that, in a connected space filled
with fluid, v has a constant value. The theorem of energy
is true identically, and the law of momentum is expressed most
 THE GENERAL THEORY OF RELATIVITY 395

simply by this fact. A single mass of fluid in equilibrium will be

radially symmetrical in respect of the distribution of its mass and
its field. In this special case we must make the same assumption
for ds2 , involving the three unknown functions λ, l, f , as at the
beginning of § 31. If we start by setting λ = 1, we lose the equation
which is derived by varying λ. A full substitute for it is clearly
given by the equation that asserts the invariance of the Action
during an infinitesimal spatial displacement in radial directions,
that is, the theorem of momentum : v = const. The problem of
variation that has now to be solved is given by
Z
 0
∆ w + r2 µ0 ∆ − r2 vh dr = 0

δ

in which ∆ and h are to undergo variation, whereas

 
1
w = 1 − 2 r.
h

Let us begin by varying ∆; we get

µ0 3
w0 − µ0 r2 = 0 and w = r ,
3
that is
1 µ0 2
2
=1− r (50)
h 3
Let the spherical mass of fluid have a radius r = r0 . It is obvious
that r0 must remain r
3
<a= .
µ0
 CHAPTER IV 396

The energy and the mass are expressed in the rational units given
by the theory of gravitation. For a sphere of water, for example,
this upper limit of the radius works out to
r
3
= 4 · 108 km. = 22 light-minutes.
8πκ
Outside the sphere our earlier formulæ are valid, in particular
1 2m
2
=1− , ∆ = 1.
h r
The boundary conditions require that h and f have continuous
values in passing over the spherical surface, and that the pressure p
vanish at the surface. From the continuity of h we get for the
gravitational radius m of the sphere of fluid

µ0 r03
m= .
6
The inequality, which holds between r0 and µ0 , shows that the
radius r0 must be greater than 2m. Hence, if we start from infinity,
then, before we get to the singular sphere r = 2m mentioned
above, we reach the fluid, within which other laws hold. If we
now adopt the gramme as our unit, we must replace µ0 by 8πκµ0 ,
whereas m = κm0 , if m0 denotes the gravitating mass. We then
find that
4πr03
m 0 = µ0 .
3
Since
∗ µ∗ ∆
v=µ f =
h
 THE GENERAL THEORY OF RELATIVITY 397

µ0
is a constant, and assumes the value at the surface of the
h0
sphere, in which h0 denotes the value of h there as given by (50),
we see that in the whole interior
µ0
v = (µ0 + p)f = . (51)
h0
Variation of h leads to
2∆0
− + rv = 0.
h3
Since it follows from (50) that

h0 µ0
= r,
h3 3
we get immediately
3v
∆= h + const.
2µ0

Further, if we use the value of the constant v given by (51),

and calculate the value of the integration constant that occurs, by
using the boundary condition ∆ = 1 at the surface of the sphere,
then
3h − h0 3h − h0
∆= , f=
2h0 2hh0
Finally, we get from (51)

h0 − h
p = µ0 ·
3h − h0
 CHAPTER IV 398

These results determine the metrical groundform of space

(x1 dx1 + x2 dx2 + x3 dx3 )
dσ 2 = (dx21 + dx22 + dx23 ) + , (52)
a2 − r 2
the gravitational potential or the velocity of light f , and the
pressure-field p.
If we introduce a superfluous co-ordinate
√
x 4 = a2 − r 2

into space, then

x21 + x22 + x23 + x24 = a2 (53)
and hence

x1 dx1 + x2 dx2 + x3 dx3 + x4 dx4 = 0.

(52) then becomes

dσ 2 = dx21 + dx22 + dx23 + dx24 .

In the whole interior of the fluid sphere spatial spherical geometry

is valid, namely, that which is true on the “sphere” (53) in four-
dimensional Euclidean space with Cartesian co-ordinates xi . The
fluid covers a cap-shaped portion of the sphere. The pressure in
it is a linear fractional function of the “vertical height,” z = x4 on
the sphere:
p z − z0
= .
µ0 3z0 − z
Further, it is shown by this formula that, since the pressure p
may not pass, on a sphere of latitude, z = const., from positive to
negative values through infinity, 3z0 must be > a, and the upper
 THE GENERAL THEORY OF RELATIVITY 399

limit a found above for the radius

√ of the fluid sphere must be
2a 2
correspondingly reduced to .
3
These results for a sphere of fluid were first obtained by
Schwarzschild (vide note 22). After the most important cases of
radially symmetrical statical gravitational fields had been solved,
the author succeeded in solving the more general problem of
the cylindrically symmetrical statical field (vide note 23).
We shall here just mention briefly the simplest results of this
investigation. Let us consider first uncharged masses and a
gravitational field in space free from matter. It then follows from
the gravitational equations, if certain space-co-ordinates r, θ, z
(so-called canonical cylindrical co-ordinates) are used, that

r2 dθ2
ds2 = f 2 dt2 − dσ 2 , dσ 2 = h(dr2 + dz 2 ) + .
f2
θ is an angle whose modulus is 2π; that is, corresponding to
values of θ that differ by integral multiples of 2π there is only one
point. On the axis of rotation r = 0. Also, h and f are functions
of r and z. We shall plot real space in terms of a Euclidean
space, in which r, θ, z are cylindrical co-ordinates. The canonical
co-ordinate system is uniquely defined except for a displacement
in the direction of the axis of rotation z 0 = z + const. When
h = f = 1, dσ 2 is identical with the metrical groundform of the
Euclidean picture-space (used for the plotting). The gravitational
problem may be solved just as easily on this theory as on that
of Newton, if the distribution of the matter is known in terms
of canonical co-ordinates. For if we transfer these masses into
our picture-space, that is, if we make the mass contained in
a portion of each space equal to the mass contained in the
 CHAPTER IV 400

corresponding portion of the picture-space, and if ψ is then the

Newtonian potential of this mass-distribution in the Euclidean
picture-space, the simple formula
2
f = eψ/c (54)
holds. The second still unknown function h may also be deter-
mined by the solution of an ordinary Poisson equation (referring
to the meridian plane θ = 0). In the case of charged bodies,
too, the canonical co-ordinate system exists. If we assume that
the masses are negligible in comparison with the charges, that is,
that for an arbitrary portion of space the gravitational radius of
the electric charges contained in it is much greater than the grav-
itational radius of the masses contained in it, and if φ denotes
the electrostatic potential (calculated according to the classical
theory) of the transposed charges in the canonical picture-space,
then f and the electrostatic potential Φ in real space are given by
the formulæ
√ 
c κ 1
Φ = √ tan φ , f= √ . (540 )
κ c κ
cos φ
c
It is not quite easy to subordinate the radially symmetrical case
to this more general theory: it becomes necessary to carry out a
rather complicated transformation of the space-co-ordinates, into
which we shall not enter here.
Just as the laws of Mie’s electrodynamics are non-linear, so
also Einstein’s laws of gravitation. This non-linearity is not
perceptible in those measurements that are accessible to direct
observation, because, in them, the non-linear terms are quite neg-
ligible in comparison with the linear ones. It is as a result of this
 THE GENERAL THEORY OF RELATIVITY 401

that the principle of superposition is found to be confirmed

by the interplay of forces in the visible world. Only, perhaps, for
the unusual occurrences within the atom, of which we have as
yet no clear picture, does this non-linearity come into considera-
tion. Non-linear differential equations involve, in comparison with
linear equations, particularly as regards singularities, extremely
intricate, unexpected, and, at the present, quite uncontrollable
conditions. The suggestion immediately arises that these two cir-
cumstances, the remarkable behaviour of non-linear differential
equations and the peculiarities of intra-atomic occurrences, are to
be related to one another. Equations (54) and (540 ) offer a beau-
tiful and simple example of how the principle of superposition
becomes modified in the strict theory of gravitation: the field-
potentials f and Φ depend in the one case on the exponential
function of the quantity ψ, and in the other on a trigonometri-
cal function of the quantity φ, these quantities being those which
satisfy the principle of superposition. At the same time, however,
these equations demonstrate clearly that the non-linearity of the
gravitational equations will be of no assistance whatever for ex-
plaining the occurrences within the atom or the constitution of the
electron. For√ the differences between φ and Φ become appreciable
κ
only when φ assumes values that are comparable with 1. But
c
even in the interior of the electron this case arises only for spheres
whose radius corresponds to the order of gravitational radius
√
κ
e= e0 ∼ 10−33 cms.
c
for the charge e0 of the electron.
It is obvious that the statical differential equations of gravita-
tion cannot uniquely determine the solutions, but that boundary
 CHAPTER IV 402

conditions at infinity, or conditions of symmetry such as the pos-

tulate of radial symmetry must be added. The solutions which we
found were those for which the metrical groundform converges, at
spatial infinity, to

dx20 − (dx21 + dx22 + dx23 ),

the expression which is a characteristic of the special theory of

relativity.
A further series of elegant investigations into problems of stat-
ical gravitation have been initiated by Levi-Civita (vide note 24).
The Italian mathematicians have studied, besides the statical case,
also the “stationary” one, which is characterised by the circum-
stance that all the gik ’s are independent of the time-co-ordinate x0 ,
whereas the “lateral” co-efficients g01 , g02 , g03 need not vanish (vide
note 25): an example of this is given by the field that surrounds
a body which is in stationary rotation.

§ 33. Gravitational Energy. The Theorems of

Conservation
An isolated system sweeps out in the course of its history
a “world-canal”; we assume that outside this canal the stream-
density si vanishes (if not entirely, at least to such a degree that
the following argument retains its validity). It follows from the
equation of continuity
∂si
=0 (55)
∂xi
that the flux of the vector-density si has the same value e through
every three-dimensional “plane” across the canal. To fix the sign
of e, we shall agree to take for its direction that leading from the
 THE GENERAL THEORY OF RELATIVITY 403

past into the future. The invariant e is the charge of our system.
If the co-ordinate system fulfils the conditions that every “plane”
x0 = const. intersects the canal in a finite region and that these
planes, arranged according to increasing values of x0 , follow one
another in the order, past → future, then we may calculate e by
means of the equation
Z
s0 dx1 dx2 dx3 = e

in which the integration is taken over any arbitrary plane of the

family x0 = const. This integral e = e(x0 ) is accordingly inde-
pendent of the “time” x0 , as is readily seen, too, from (55) if we
integrate it with respect to the “space-co-ordinates” x1 , x2 , x3 .
What has been stated above is valid in virtue of the equation
of continuity alone; the idea of substance and the convention to
which it leads in Lorentz’s Theory, namely, si = ρui do not come
into question in this case.
Does a similar theorem of conservation hold true for en-
ergy and momentum? This can certainly not be decided from
the equation (26) of § 28, since the latter contains the additional
term which is a characteristic of the theory of gravitation. It is
possible, however, to write this addition term, too, in the form
of a divergence. We choose a definite co-ordinate system and
subject the world-continuum to an infinitesimal deformation in
the true sense, that is, we choose constants for the deformation
components ξ i in § 28. Then, of course, for any finite region X
Z
0
δ G dx = 0
X

(this is true for every function of the gik ’s and their derivatives:
it has nothing to do with properties of invariance; δ 0 denotes, as
 CHAPTER IV 404

in § 28, the variation effected by the displacement). Hence, the

displacement gives us
∂(Gξ k )
Z Z
dx + δG dx = 0.
X ∂xk X

If, as earlier, we set

δG = 12 Gαβ δgαβ + 12 Gαβ,k δgαβ,k , (13)

then partial integration gives

∂(Gαβ,k ) δgαβ,k
Z Z Z
2 δG dx = + [G]αβ δgαβ dx.
X X ∂x k X

Now, in this case, since the ξ’s are constants,

∂gαβ i
δgαβ = − ξ.
∂xi
If we introduce the quantities
∂gαβ
Gδik − 12 Gαβ,k = tki
∂xi
then, by the preceding relation, we get the equation
Z  k 
∂ti 1 αβ ∂gαβ
− 2 [G] ξ i dx = 0.
X ∂x k ∂x i

Since this holds for any arbitrary region X, the integrand must be
equal to zero. In it the ξ i ’s denote arbitrary constant numbers;
hence we get four identities:

1 ∂gαβ ∂tk
2
[G]αβ = i.
∂xi ∂xk
 THE GENERAL THEORY OF RELATIVITY 405

The left-hand side, by the gravitational equations,

∂gαβ
= − 12 Tαβ
∂xi
and, accordingly, the mechanical equations (26) become

∂Uki
= 0, where Uki = Tki + tki . (56)
∂xk

It is thus shown that if we regard the tki ’s, which are dependent
only on the potentials and the field-components of gravitation, as
the components of the energy-density of the gravitational
field, we get pure divergence equations for all energy associated
with “physical state or phase” and “gravitation” (vide note 26).
And yet, physically, it seems devoid of sense to introduce
the tki ’s as energy-components of the gravitational field, for these
quantities neither form a tensor nor are they symmetrical.
In actual fact, if we choose an appropriate co-ordinate system, we
may make all the tki ’s at one point vanish; it is only necessary to
choose a geodetic co-ordinate system. And, on the other hand, if
we use a curvilinear co-ordinate system in a “Euclidean” world to-
tally devoid of gravitation, we get tki ’s that are all different from
zero, although the existence of gravitational energy in this case
can hardly come into question. Hence, although the differential
relations (56) have no real physical meaning, we can derive from
them, by integrating over an isolated system, an invariant
theorem of conservation (vide note 27).
During motion an isolated system with its accompanying gravi-
tational field sweeps out a canal in the “world”. Beyond the canal,
in the empty surroundings of the system, we shall assume that
 CHAPTER IV 406

the tensor-density Tki and the gravitational field vanish. We may

then use co-ordinates x0 (= t), x1 , x2 , x3 , such that the metrical
groundform assumes constant co-efficients outside the canal, and
in particular assumes the form

dt2 − (dx21 + dx22 + dx23 ).

Hence, outside the canal, the co-ordinates are fixed except for a
linear (Lorentz) transformation, and the tki ’s vanish there. We as-
sume that each of the “planes” t = const. has only a finite portion
of section in common with the canal. If we integrate the equa-
tions (56) with respect to x1 , x2 , x3 over such a plane, we find
that the quantities
Z
Ji = U0i dx1 dx2 dx3

dJi
are independent of the time; that is = 0. We call J0 the
dt
energy, and J1 , J2 , J3 the momentum co-ordinates of the
system.
These quantities have a significance which is independent of
the co-ordinate system. We affirm, firstly, that they retain their
value if the co-ordinate system is changed anywhere within the
canal. Let x̄i be the new co-ordinates, identical with the old ones
for the region outside the canal. We mark out two “surfaces”

x0 = const. = a and x̄0 = const. = ā (ā 6= a)

which do not intersect in the canal (for this it suffices to choose

a and ā sufficiently different from one another). We can then
construct a third co-ordinate system x∗i which is identical with
 THE GENERAL THEORY OF RELATIVITY 407

the xi ’s in the neighbourhood of the first surface, identical with

the x̄i in that of the second system, and is identical with both
outside the canal. If we give expression to the fact that the energy-
momentum components Ji∗ in this system assume the same values
for x∗0 = a and x0 = ā, then we get the result which we enunciated,
namely, Ji = J¯i .
Consequently, the behaviour of the Ji ’s need be investigated
only in the case of linear transformations of the co-ordinates.
With respect to such, however, the conception of a tensor with
components that are constant (that is, independent of position)
is invariant. We make use of an arbitrary vector pi of this type,
and form Uk = Uki pi , and deduce from (56) that
∂Uk
= 0.
∂xk
By applying the same reasoning as was used above in the case of
the electric current, it follows from this that
Z
U0 dx1 dx2 dx3 = Ji pi

is an invariant with respect to linear transformations. Accord-

ingly, the Ji ’s are the components of a constant co-variant
vector in the “Euclidean” surroundings of the system; this
energy-momentum vector is uniquely determined by the phase
(or state) of the physical system. The direction of this vector de-
termines generally the direction in which the canal traverses the
surrounding world (a purely descriptive datum that can be ex-
pressed in an exact form accessible to mathematical analysis only
with great difficulty). The invariant
q
J02 − J12 − J22 − J32
 CHAPTER IV 408

is the mass of the system.

In the statical case J1 = J2 = J3 = 0, whereas J0 is equal to
the space-integral of R00 − ( 12 R − G). According to § 29 and § 28
(page 359), respectively,
∂f i
R00 = , and in general,
∂xi
    
1 1 ∂ √ αβ αβ iα αβ
2
R−G= 2 g g −g ,
∂xi i β
and hence, in the notation of § 29 and § 31, the mass J0 is equal
to the flux of the (spurious) spatial vector-density
    
1 √ αβ αβ iα αβ
mi = 2 f g γ −γ (i, α, β = 1, 2, 3), (57)
i β
1
which has yet to be multiplied by if we use the ordinary units.
8πκ
Since at a great distance from the system the solution of the field
laws, which was found in § 31, is always valid, and for which mi is
a radial current of intensity
1 − f2 m0
= ,
8πκr 4πr2
we get that the energy, J0 , or the inertial mass of the system, is
equal to the mass m0 , which is characteristic of the gravitational
field generated by the system (vide note 28). On the other hand
it is to be remarked parenthetically that the physics based on the
notion of substance leads to the space-integral of µ/f for the value
of the mass, whereas, in reality, for incoherent matter J0 = m0 =
the space-integral of µ; this is a definite indication of how radically
erroneous is the whole idea of substance.
 THE GENERAL THEORY OF RELATIVITY 409

§ 34. Concerning the Inter-connection of the World as a

Whole
The general theory of relativity leaves it quite undecided
whether the world-points may be represented by the values of
four co-ordinates xi in a singly reversible continuous manner or
not. It merely assumes that the neighbourhood of every world-
point admits of a singly reversible continuous representation in
a region of the four-dimensional “number-space” (whereby “point
of the four-dimensional number-space” is to signify any number-
quadruple); it makes no assumptions at the outset about the
inter-connection of the world. When, in the theory of surfaces,
we start with a parametric representation of the surface to be
investigated, we are referring only to a piece of the surface, not
to the whole surface, which in general can by no means be repre-
sented uniquely and continuously on the Euclidean plane or by a
plane region. Those properties of surfaces that persist during all
one-to-one continuous transformations form the subject-matter of
analysis situs (the analysis of position); connectivity, for example,
is a property of analysis situs. Every surface that is generated
from the sphere by continuous deformation does not, from the
point of view of analysis situs, differ from the sphere, but does
differ from an anchor-ring, for instance. For on the anchor-ring
there exist closed lines, which do not divide it into several regions,
whereas such lines are not to be found on the sphere. From the
geometry which is valid on a sphere, we derived “spherical geom-
etry” (which, following Riemann, we set up in contrast with the
geometry of Bolyai-Lobatschefsky) by identifying two diametri-
cally opposite points of the sphere. The resulting surface F is
from the point of view of analysis situs likewise different from the
 CHAPTER IV 410

sphere, in virtue of which property it is called one-sided. If we

imagine on a surface a small wheel in continual rotation in the
one direction to be moved along this surface during the rotation,
the centre of the wheel describing a closed curve, then we should
expect that when the wheel has returned to its initial position
it would rotate in the same direction as at the commencement
of its motion. If this is the case, then whatever curve the centre
of the wheel may have described on the surface, the latter is
called two-sided; in the reverse case, it is called one-sided. The
existence of one-sided surfaces was first pointed out by Möbius.
The surface F mentioned above is one-sided, whereas the sphere
is, of course, two-sided. This is obvious if the centre of the wheel
be made to describe a great circle; on the sphere the whole
circle must be traversed if this path is to be closed, whereas on F
only the half need be covered. Quite analogously to the case
of two-dimensional manifolds, four-dimensional ones may be en-
dowed with diverse properties with regard to analysis situs. But
in every four-dimensional manifold the neighbourhood of a point
may, of course, be represented in a continuous manner by four
co-ordinates in such a way that different co-ordinate quadruples
always correspond to different points of this neighbourhood. The
use of the four world-co-ordinates is to be interpreted in just this
way.
Every world-point is the origin of the double-cone of the ac-
tive future and the passive past. Whereas in the special theory
of relativity these two portions are separated by an intervening
region, it is certainly possible in the present case for the cone of
the active future to overlap with that of the passive past; so that,
in principle, it is possible to experience events now that will in
part be an effect of my future resolves and actions. Moreover, it
 THE GENERAL THEORY OF RELATIVITY 411

is not impossible for a world-line (in particular, that of my body),

although it has a time-like direction at every point, to return to
the neighbourhood of a point which it has already once passed
through. The result would be a spectral image of the world more
fearful than anything the weird fantasy of E. T. A. Hoffmann has
ever conjured up. In actual fact the very considerable fluctuations
of the gik ’s that would be necessary to produce this effect do not
occur in the region of world in which we live. Nevertheless there
is a certain amount of interest in speculating on these possibili-
ties inasmuch as they shed light on the philosophical problem of
cosmic and phenomenal time. Although paradoxes of this kind
appear, nowhere do we find any real contradiction to the facts
directly presented to us in experience.
We saw in § 26 that, apart from the consideration of gravita-
tion, the fundamental electrodynamic laws (of Mie) have a form
such as is demanded by the principle of causality. The time-
derivatives of the phase-quantities are expressed in terms of these
quantities themselves and their spatial differential co-efficients.
These facts persist when we introduce gravitation and thereby
 the table of phase-quantities φi , Fik , by the gik ’s and
increase
ik
the ’s. But on account of the general invariance of physical
r
laws we must formulate our statements so that, from the values
of the phase-quantities for one moment, all those assertions con-
cerning them, which have an invariant character, follow as
a consequence of physical laws; moreover, it must be noted that
this statement does not refer to the world as a whole but only to a
portion which can be represented by four co-ordinates. Following
Hilbert (vide note 29) we proceed thus. In the neighbourhood of
the world-point O we introduce 4 co-ordinates xi such that, at O
 CHAPTER IV 412

itself,
ds2 = dx20 − (dx21 + dx22 + dx23 ).
In the three-dimensional space x0 = 0 surrounding O we may
mark off a region R, such that, in it, −ds2 remains definitely pos-
itive. Through every point of this region we draw the geodetic
world-line which is orthogonal to that region, and which has a
time-like direction. These lines will cover singly a certain four-
dimensional neighbourhood of O. We now introduce new co-
ordinates which will coincide with the previous ones in the three-
dimensional space R, for we shall now assign the co-ordinates
x0 , x1 , x2 , x3 to the point P at which we arrive, if we go from
the point P0 = (x1 , x2 , x3 ) in R along the orthogonal geodetic
line passing through it, so far that the proper-time of the arc
traversed, P0 P , is equal to x0 . This system of co-ordinates was
introduced into the theory of surfaces by Gauss. Since ds2 = dx20
on each of the geodetic lines, we must get identically for all four
co-ordinates in this co-ordinate system:

g00 = 1. (58)

Since the lines are orthogonal to the three-dimensional space

x0 = 0, we get for x0 = 0

g01 = g02 = g03 = 0. (59)

Moreover, since the lines that are obtained when x1 , x2 , x3 are

kept constant and x0 is varied are geodetic, it follows (from the
equation of geodetic lines) that
 
00
=0 (i = 0, 1, 2, 3),
i
 THE GENERAL THEORY OF RELATIVITY 413

and hence also that  

00
= 0.
i
Taking (58) into consideration, we get from the latter

∂g0
=0 (i = 1, 2, 3)
∂x0
and, on account of (59), we have consequently not only for x0 = 0
but also identically for the four co-ordinates that

g0i = 0 (i = 1, 2, 3). (60)

The following picture presents itself to us: a family of geode-

tic lines with time-like direction which covers a certain world-
region singly and completely (without gaps); also, a similar uni-
parametric family of three-dimensional spaces x0 = const. Ac-
cording to (60) these two families are everywhere orthogonal to
one another, and all portions of arc cut off from the geodetic lines
by two of the “parallel” spaces x0 = const. have the same proper-
time. If we use this particular co-ordinate system, then
 
∂gik ik
= −2 (i, k = 1, 2, 3)
∂x0 0

and the gravitational equations enable us to express the deriva-

tives  
∂ ik
(i, k = 1, 2, 3)
∂x0 0
not only in terms of the φi ’s and their derivatives, but also in
terms of the gik ’s, their derivatives (of the first and second order)
 CHAPTER IV 414

 
ik
with respect to x1 , x2 , x3 , and the ’s themselves. Hence, by
0
regarding the twelve quantities,
 
ik
gik , (i, k = 1, 2, 3)
0

together with the electromagnetic quantities, as the unknowns,

we arrive at the required result (x0 playing the part of time).
The cone of the passive past starting from the point O0 with a
positive x0 co-ordinate will cut a certain portion R0 out of R,
which, with the sheet of the cone, will mark off a finite region of
the world G (namely, a conical cap with its vertex at O0 ). If our
assertion that the geodetic null-lines denote the initial points of
all action is rigorously true, then the values of the above twelve
quantities as well as the electromagnetic potentials φi and the
field-quantities Fik in the three-dimensional region of space R0
determine fully the values of the two latter quantities in the world-
region G. This has hitherto not been proved. In any case, we
see that the differential equations of the field contain the physical
laws of nature in their complete form, and that there cannot be a
further limitation due to boundary conditions at spatial infinity,
for example.
Einstein, arguing from cosmological considerations of the inter-
connection of the world as a whole (vide note 30) came to the con-
clusion that the world is finite in space. Just as in the Newtonian
theory of gravitation the law of contiguous action expressed in
Poisson’s equation entails the Newtonian law of attraction only if
the condition that the gravitational potential vanishes at infinity
is superimposed, so Einstein in his theory seeks to supplement the
differential equations by introducing boundary conditions at spa-
 THE GENERAL THEORY OF RELATIVITY 415

tial infinity. To overcome the difficulty of formulating conditions

of a general invariant character, which are in agreement with as-
tronomical facts, he finds himself constrained to assume that the
world is closed with respect to space; for in this case the bound-
ary conditions are absent. In consequence of the above remarks
the author cannot admit the cogency of this deduction, since the
differential equations in themselves, without boundary conditions,
contain the physical laws of nature in an unabbreviated form ex-
cluding every ambiguity. So much more weight is accordingly to
be attached to another consideration which arises from the ques-
tion: How does it come about that our stellar system with the
relative velocities of the stars, which are extraordinarily small in
comparison with that of light, persists and maintains itself and
has not, even ages ago, dispersed itself into infinite space? This
system presents exactly the same view as that which a molecule
in a gas in equilibrium offers to an observer of correspondingly
small dimensions. In a gas, too, the individual molecules are not
at rest but the small velocities, according to Maxwell’s law of dis-
tribution, occur much more often than the large ones, and the
distribution of the molecules over the volume of the gas is, on the
average, uniform, so that perceptible differences of density occur
very seldom. If this analogy is legitimate, we could interpret the
state of the stellar system and its gravitational field according to
the same statistical principles that tell us that an isolated vol-
ume of gas is almost always in equilibrium. This would, however,
be possible only if the uniform distribution of stars at rest
in a static gravitational field, as an ideal state of equilib-
rium, is reconcilable with the laws of gravitation. In a statical
field of gravitation the world-line of a point-mass at rest, that is,
a line on which x1 , x2 , x3 remain constant and x0 alone varies, is
 CHAPTER IV 416

a geodetic line if
 
00
= 0, (i = 1, 2, 3),
i

and hence  
00 ∂g00
= 0, = 0.
i ∂xi
Therefore, a distribution of mass at rest is possible only if
√
g00 = f = const. = 1.

The equation

∆f = 12 µ (µ = density of mass) (32)

then shows, however, that the ideal state of equilibrium under

consideration is incompatible with the laws of gravitation, as
hitherto assumed.
In deriving the gravitational equations in § 28, however, we
committed a sin of omission. R is not the only invariant dependent
on the gik ’s and their first and second differential co-efficients, and
which is linear in the latter; for the most general invariant of this
description has the form αR + β, in which α and β are numerical
constants. Consequently we may generalise the laws of gravitation
√
by replacing R by R+λ (and G by G+ 12 λ g), in which λ denotes
a universal constant. If it is not equal to 0, as we have hitherto
assumed, we may take it equal to 1; by this means not only has the
unit of time been reduced by the principle of relativity to the unit
of length, and the unit of mass by the law of gravitation to the
same unit, but the unit of length itself is fixed absolutely. With
 THE GENERAL THEORY OF RELATIVITY 417

these modifications the gravitational equations for statical non-

√
coherent matter (T00 = µ = µ0 g, all other components of the
tensor-density T being equal to zero) give, if we use the equation
f = 1 and the notation of § 29:
λ = µ0 [in place of (32)]
and
Pik − λγik = 0 (i, k = 1, 2, 3). (61)
Hence this ideal state of equilibrium is possible under these cir-
cumstances if the mass is distributed with the density λ. The
space must then be homogeneous metrically; and indeed the equa-
(61) are then actually satisfied for a spherical space of radius
tionsp
a = 2/λ. Thus, in space, we may introduce four co-ordinates,
connected by
x21 + x22 + x23 + x24 = a2 , (62)
for which we get
dσ 2 = dx21 + dx22 + dx23 + dx24 .
From this we conclude that space is closed and hence
finite. If this were not the case, it would scarcely be possible to
imagine how a state of statistical equilibrium could come about. If
the world is closed, spatially, it becomes possible for an observer to
see several pictures of one and the same star. These depict the star
at epochs separated by enormous intervals of time (during which
light travels once entirely round the world). We have yet to inquire
whether the points of space correspond singly and reversibly to the
value-quadruples xi which satisfy the condition (62), or whether
two value-systems
(x1 , x2 , x3 , x4 ) and (−x1 , −x2 , −x3 , −x4 )
 CHAPTER IV 418

correspond to the same point. From the point of view of analysis

situs these two possibilities are different even if both spaces are
two-sided. According as the one or the other holds, the total mass
of the world in grammes would be
πa πa
or , respectively.
2κ 4κ
Thus our interpretation demands that the total mass that happens
to be present in the world bear a definite relation to the universal
2
constant λ = 2 which occurs in the law of action; this obviously
a
makes great demands on our credulity.
The radially symmetrical solutions of the modified homoge-
neous equations of gravitation that would correspond to a world
empty of mass are derivable by means of the principle of variation
(vide § 31 for the notation)
Z
δ (2w∆0 + λ∆r2 ) dr = 0.

The variation of w gives, as earlier, ∆ = 1. On the other hand,

variation of ∆ gives
λ
w0 = r2 . (63)
2
It we demand regularity at r = 0, it follows from (63) that
λ 3
w= r
6
1 λ
and 2
= f 2 = 1 − r2 . (64)
h 6
The space may be represented congruently on a “sphere”
x21 + x22 + x23 + x24 = 3a2 (65)
 THE GENERAL THEORY OF RELATIVITY 419

√
of radius a 3 in four-dimensional Euclidean space (whereby one
of the two poles on the sphere, whose first three co-ordinates,
x1 , x2 , x3 each = 0, corresponds to the centre in our case). The
world is a cylinder erected on this sphere in the direction of a
fifth co-ordinate axis t. But since on the “greatest sphere” x4 = 0,
which may be designated as the equator or the space-horizon for
that centre, f becomes zero, and hence the metrical groundform
of the world becomes singular, we see that the possibility of a sta-
tionary empty world is contrary to the physical laws that are here
regarded as valid. There must at least be masses at the horizon.
The calculation may be performed most readily if (merely to ori-
ent ourselves on the question) we assume an incompressible fluid
to be present there. According to § 32 the problem of variation
that is to be solved is (if we use the same notation and add the
λ term)
Z    
0 λ 2 2
δ ∆ w + µ0 + r ∆ − r vh dr = 0.
2

In comparison with the earlier expression we note that the only

λ
change consists in the constant µ0 being replaced by µ0 + . As
2
earlier, it follows that
 
0 λ 2 2µ0 + λ 3
w − µ0 + r = 0, w = −2M + r ,
2 6
1 2M 2µ0 + λ 2
2
=1+ − r . (66)
h r 6

√ the two meridians x4 = const.,

If the fluid is situated between
which have a radius r0 (< a 3), then continuity of argument
 CHAPTER IV 420

with (64) demands that the constant

µ0 3
M= r .
6 0
1
To the first order 2 becomes equal to zero for a value r = b
√h
between r0 and a 3. Hence the space may still be represented
on the sphere (65), but this representation is no longer congruent
for the zone occupied by fluid. The equation for ∆ (page 397)
now yields a value of f that does not vanish at the equator. The
boundary condition of vanishing pressure gives a transcendental
relation between µ0 and r0 , from which it follows that, if the mass-
horizon is to be taken arbitrarily small, then the fluid that comes
into question must have a correspondingly great density, namely,
such that the total mass does not become less than a certain
positive limit (vide note 31).
The general solution of (63) is
1 2m λ 2
2
= f2 = 1 − − r (m = const.).
h r 6
It corresponds to the case in which a spherical mass is situated
at the centre. The world can be empty of mass only in a zone
r0 ≤ r ≤ r1 , in which this f 2 is positive; a mass-horizon is again
necessary. Similarly, if the central mass is charged electrically;
1
for in this case, too, ∆ = 1. In the expression for 2 = f 2
h
e2
the electrical term + 2 has to be added, and the electrostatic
r
e
potential = .
r
Perhaps in pursuing the above reflections we have yielded too
readily to the allurement of an imaginary flight into the region of
 THE GENERAL THEORY OF RELATIVITY 421

masslessness. Yet these considerations help to make clear what

the new views of space and time bring within the realm of pos-
sibility. The assumption on which they are based is at any rate
the simplest on which it becomes explicable that, in the world
as actually presented to us, statical conditions obtain as a whole,
so far as the electromagnetic and the gravitational field is con-
cerned, and that just those solutions of the statical equations are
valid which vanish at infinity or, respectively, converge towards
Euclidean metrics. For on the sphere these equations will have a
unique solution (boundary conditions do not enter into the ques-
tion as they are replaced by the postulate of regularity over the
whole of the closed configuration). If we make the constant λ
arbitrarily small, the spherical solution converges to that which
satisfies at infinity the boundary conditions mentioned for the in-
finite world which results when we pass to the limit.
A metrically homogeneous world is obtained most simply
if, in a five-dimensional space with the metrical groundform
ds2 = −Ω(dx), (−Ω denotes a non-degenerate quadratic form
with constant co-efficients), we examine the four-dimensional
6
“conic-section” defined by the equation Ω(x) = . Thus this
λ
basis gives us a solution of the Einstein equations of gravitation,
modified by the λ term, for the case of no mass. If, as must be
the case, the resulting metrical groundform of the world is to have
one positive and three negative dimensions, we must take for Ω a
form with four positive dimensions and one negative, thus

Ω(x) = x21 + x22 + x23 + x24 − x25 .

By means of a simple substitution this solution may easily be

transformed into the one found above for the statical case. For if
 CHAPTER IV 422

we set
x4 = z cosh t, x5 = z sinh t,
we get
6
x21 + x22 + x23 + z 2 = , −ds2 = (dx21 + dx22 + dx23 + dz 2 ) − z 2 dt2 .
λ
These “new” z, t co-ordinates, however, enable only the “wedge-
shaped” section x24 − x25 > 0 to be represented. At the “edge”
of the wedge (at which x4 = 0 simultaneously with x5 = 0),
t becomes indeterminate. This edge, which appears as a two-
dimensional configuration in the original co-ordinates is, there-
fore, three-dimensional in the new co-ordinates; it is the cylinder
erected in the direction of the t-axis over the equator z = 0 of the
sphere (65). The question arises whether it is the first or the sec-
ond co-ordinate system that serves to represent the whole world
in a regular manner. In the former case the world would not be
static as a whole, and the absence of matter in it would be in
agreement with physical laws; de Sitter argues from this assump-
tion (vide note 32). In the latter case we have a static world that
cannot exist without a mass-horizon; this assumption, which we
have treated more fully, is favoured by Einstein.

§ 35. The Metrical Structure of the World as the Origin

of Electromagnetic Phenomena∗
We now aim at a final synthesis. To be able to characterise
the physical state of the world at a certain point of it by means of
numbers we must not only refer the neighbourhood of this point to
∗
Vide note 33.
 THE GENERAL THEORY OF RELATIVITY 423

a co-ordinate system but we must also fix on certain units of mea-

sure. We wish to achieve just as fundamental a point of view with
regard to this second circumstance as is secured for the first one,
namely, the arbitrariness of the co-ordinate system, by the Ein-
stein Theory that was described in the preceding paragraph. This
idea, when applied to geometry and the conception of distance (in
Chapter II) after the step from Euclidean to Riemann geometry
had been taken, effected the final entrance into the realm of in-
finitesimal geometry. Removing every vestige of ideas of “action
at a distance,” let us assume that world-geometry is of this kind;
we then find that the metrical structure of the world, besides be-
ing dependent on the quadratic form (1), is also dependent on a
linear differential form φi dxi .
Just as the step which led from the special to the general theory
of relativity, so this extension affects immediately only the world-
geometrical foundation of physics. Newtonian mechanics, as also
the special theory of relativity, assumed that uniform translation
is a unique state of motion of a set of vector axes, and hence that
the position of the axes at one moment determines their position
in all other moments. But this is incompatible with the intuitive
principle of the relativity of motion. This principle could be
satisfied, if facts are not to be violated drastically, only by main-
taining the conception of infinitesimal parallel displacement of
a vector set of axes; but we found ourselves obliged to regard the
affine relationship, which determines this displacement, as some-
thing physically real that depends physically on the states of mat-
ter (“guiding field”). The properties of gravitation known from
experience, particularly the equality of inertial and gravitational
mass, teach us, finally, that gravitation is already contained in
the guiding field besides inertia. And thus the general theory of
 CHAPTER IV 424

relativity gained a significance which extended beyond its original

important bearing on world-geometry to a significance which
is specifically physical. The same certainty that characterises the
relativity of motion accompanies the principle of the relativity of
magnitude. We must not let our courage fail in maintaining this
principle, according to which the size of a body at one moment
does not determine its size at another, in spite of the existence
of rigid bodies.∗ But, unless we are to come into violent con-
flict with fundamental facts, this principle cannot be maintained
without retaining the conception of infinitesimal congruent trans-
formation; that is, we shall have to assign to the world besides
its measure-determination at every point also a metrical relation-
ship. Now this is not to be regarded as revealing a “geometrical”
property which belongs to the world as a form of phenomena,
but as being a phase-field having physical reality. Hence, as the
fact of the propagation of action and of the existence of rigid
bodies leads us to found the affine relationship on the metrical
character of the world which lies a grade lower, it immediately
suggests itself to us, not only to identify the co-efficients of the
quadratic groundform gik dxi dxk with the potentials of the grav-
itational field, but also to identify the co-efficients of the lin-
ear groundform φi dxi with the electromagnetic potentials.
The electromagnetic field and the electromagnetic forces are then
derived from the metrical structure of the world or the metrics,
as we may call it. No other truly essential actions of forces are,
∗
It must be recalled in this connection that the spatial direction-picture
which a point-eye with a given world-line receives at every moment from a
given region of the world, depends only on the ratios of the gik ’s, inasmuch
as this is true of the geodetic null-lines which are the determining factors in
the propagation of light.
 THE GENERAL THEORY OF RELATIVITY 425

however, known to us besides those of gravitation and electromag-

netic actions; for all the others statistical physics presents some
reasonable argument which traces them back to the above two by
the method of mean values. We thus arrive at the inference: The
world is a (3 + 1)-dimensional metrical manifold; all phys-
ical field-phenomena are expressions of the metrics of the
world. (Whereas the old view was that the four-dimensional met-
rical continuum is the scene of physical phenomena; the physical
essentialities themselves are, however, things that exist “in” this
world, and we must accept them in type and number in the form
in which experience gives us cognition of them: nothing further is
to be “comprehended” of them.) We shall use the phrase “state of
the world-æther” as synonymous with the word “metrical struc-
ture,” in order to call attention to the character of reality apper-
taining to metrical structure; but we must beware of letting this
expression tempt us to form misleading pictures. In this termi-
nology the fundamental theorem of infinitesimal geometry states
that the guiding field, and hence also gravitation, is determined
by the state of the æther. The antithesis of “physical state” and
“gravitation” which was enunciated in § 28 and was expressed in
very clear terms by the division of Hamilton’s Function into two
parts, is overcome in the new view, which is uniform and logical
in itself. Descartes’ dream of a purely geometrical physics seems
to be attaining fulfilment in a manner of which he could certainly
have had no presentiment. The quantities of intensity are sharply
distinguished from those of magnitude.
The linear groundform φi dxi is determined except for an ad-
ditive total differential, but the tensor of distance-curvature
∂φi ∂φk
fik = −
∂xk ∂xi
 CHAPTER IV 426

which is derived from it, is free of arbitrariness. According to

Maxwell’s Theory the same result obtains for the electromagnetic
potential. The electromagnetic field-tensor, which we denoted ear-
lier by Fik , is now to be identified with the distance-curvature fik .
If our view of the nature of electricity is true, then the first system
of Maxwell’s equations
∂fik ∂fkl ∂fli
+ + =0 (67)
∂xl ∂xi ∂xk
is an intrinsic law, the validity of which is wholly independent of
whatever physical laws govern the series of values that the phys-
ical phase-quantities actually run through. In a four-dimensional
metrical manifold the simplest integral invariant that exists at all
is Z Z
1
l dx = 4 fik f ik dx (68)

and it is just this one, in the form of Action, on which Maxwell’s

Theory is founded! We have accordingly a good right to claim that
the whole fund of experience which is crystallised in Maxwell’s
Theory weighs in favour of the world-metrical nature of electric-
ity. And since it is impossible to construct an integral invariant at
all of such a simple structure in manifolds of more or less than four
dimensions the new point of view does not only lead to a deeper
understanding of Maxwell’s Theory but the fact that the world
is four-dimensional, which has hitherto always been accepted as
merely “accidental,” becomes intelligible through it. In the linear
groundform φi dxi there is an arbitrary factor in the form of an
additive total differential, but there is not a factor of proportion-
ality; the quantity Action is a pure number. But this is only as it
should be, if the theory is to be in agreement with that atomistic
 THE GENERAL THEORY OF RELATIVITY 427

structure of the world which, according to the most recent results

(Quantum Theory), carries the greatest weight.
The statical case occurs when the co-ordinate system and the
calibration may be chosen so that the linear groundform becomes
equal to φ dx0 and the quadratic groundform becomes equal to

f 2 dx20 − dσ 2 ,

whereby φ and f are not dependent on the time x0 , but only on

the space-co-ordinates x1 , x2 , x3 , whilst dσ 2 is a definitely posi-
tive quadratic differential form in the three space-variables. This
particular form of the groundform (if we disregard quite particu-
lar cases) remains unaffected by a transformation of co-ordinates
and a re-calibration only if x0 undergoes a linear transformation
of its own, and if the space-co-ordinates are likewise transformed
only among themselves, whilst the calibration ratio must be a
constant. Hence, in the statical case, we have a three-dimensional
Riemann space with the groundform dσ 2 and two scalar fields
in it: the electrostatic potential φ, and the gravitational poten-
tial or the velocity of light f . The length-unit and the time-unit
(centimetre, second) are to be chosen as arbitrary units; dσ 2 has
dimensions cm2 , f has dimensions cm · sec−1 , and φ has sec−1 .
Thus, as far as one may speak of a space at all in the general
theory of relativity (namely, in the statical case), it appears as
a Riemann space, and not as one of the more general type, in
which the transference of distances is found to be non-integrable.
We have the case of the special theory of relativity again, if
the co-ordinates and the calibration may be chosen so that

ds2 = dx20 − (dx21 + dx22 + dx23 ).

 CHAPTER IV 428

If xi , x̄i denote two co-ordinate systems for which this normal

form for ds2 may be obtained, then the transition from xi to x̄i is
a conformal transformation, that is, we find

dx20 − (dx21 + dx22 + dx23 ),

except for a factor of proportionality, is equal to

dx̄20 − (dx̄21 + dx̄22 + dx̄23 ).

The conformal transformations of the four-dimensional Minkowski

world coincide with spherical transformations (vide note 34), that
is, with those transformations which convert every “sphere” of the
world again into a sphere. A sphere is represented by a linear
homogeneous equation between the homogeneous “hexaspherical”
co-ordinates
(x, x) + 1 (x, x) − 1
u0 : u1 : u2 : u3 : u4 : u5 = x0 : x1 : x2 : x3 : : ,
2 2
where
(x, x) = x20 − (x21 + x22 + x23 ).
They are bound by the condition

u20 − u21 − u22 − u23 − u24 + u25 = 0.

The spherical transformations therefore express themselves as

those linear homogeneous transformations of the ui ’s which leave
this condition, as expressed in the equation, invariant. Maxwell’s
equations of the æther, in the form in which they hold in the
special theory of relativity, are therefore invariant not only with
 THE GENERAL THEORY OF RELATIVITY 429

respect to the 10-parameter group of the linear Lorentz transfor-

mations but also indeed with respect to the more comprehensive
15-parameter group of spherical transformations (vide note 35).
To test whether the new hypothesis about the nature of the
electromagnetic field is able to account for phenomena, we must
work out its implications. We choose as our initial physical law
a Hamilton principle which states that the change in the Action
Z
W dx for every infinitely small variation of the metrical struc-
ture of the world that vanishes outside a finite region is zero. The
Action is an invariant, and hence W is a scalar-density (in the true
sense) which is derived from the metrical structure. Mie, Hilbert,
and Einstein assumed the Action to be an invariant with respect
to transformations of the co-ordinates. We have here to add the
further limitation that it must also be invariant with respect to
the process of re-calibration, in which φi , gik are replaced by
1 ∂λ
φi − and λgik , respectively, (69)
λ ∂xi
in which λ is an arbitrary positive function of position. We assume
that W is an expression of the second order, that is, built up, on
the one hand, of the gik ’s and their derivatives of the first and
second order, on the other hand, of the φi ’s and their derivatives of
the first order. The simplest example is given by Maxwell’s density
of action l. But we shall here carry out a general investigation
without binding ourselves to any particular form of W at the
beginning. According to Klein’s method, used in § 28 (and which
will only now be applied with full effect), we shall here deduce
certain mathematical identities, which are valid for every scalar-
density W which has its origin in the metrical structure.
 CHAPTER IV 430

I. If we assign to the quantities φi , gik , which describe the

metrical structure relative to a system of reference, infinitely small
increments δφi , δgik , and if X denote a finite region of the world,
then the effect of partial integration is to separate the integral
of the corresponding change δW in W over the region X into
two parts: (a) a divergence integral and (b) an integral whose
integrand is only a linear combination of δφi and δgik , thus

∂(δvk )
Z Z Z
δW dx = dx + (wi δφi + 12 Wik δgik ) dx (70)
X X ∂xk X

whereby Wki = Wik .

The wi ’s are components of a contra-variant vector-density,
but the Wik ’s are the components of a mixed tensor-density of the
second order (in the true sense). The δvk ’s are linear combinations
of  
∂gαβ
δφα , δgαβ and δgαβ,i δgαβ,i = .
∂xi
We indicate this by the formula

δvk = (k, α) δφα + (k, α, β) δgαβ + (k, i, α, β) δgαβ,i .

The δvk ’s are defined uniquely by equation (70) only if the nor-
malising condition that the co-efficients (k, i, α, β) be symmetrical
in the indices k and i is added. In the normalisation the δvk ’s are
components of a vector-density (in the true sense), if the δφi ’s
are regarded as the components of a co-variant vector of weight
zero and the δgik ’s as the components of a tensor of weight unity.
(There is, of course, no objection to applying another normalisa-
tion in place of this one, provided that it is invariant in the same
sense.)
 THE GENERAL THEORY OF RELATIVITY 431

Z
First of all, we express that W dx is a calibration invariant,
X
that is, that it does not alter when the calibration of the world is
altered infinitesimally. If the calibration ratio between the altered
and the original calibration is λ = 1 + π, π is an infinitesimal
scalar-field which characterises the event and which may be as-
signed arbitrarily. As a result of this process, the fundamental
quantities assume, according to (69), the following increments:
∂π
δgik = πgik , δφi = − . (71)
∂xi
If we substitute these values in δvk , let the following expressions
result:
∂π
sk (π) = π · sk + · hkα . (72)
∂xα
They are the components of a vector-density which depends on
the scalar-field π in a linear-differential manner. It further follows
∂π
from this, that, since the ’s are the components of a co-variant
∂xα
vector-field which is derived from the scalar-field, sk is a vector-
density, and hkα is a contra-variant tensor-density of the second
order. The variation (70) of the integral of Action must vanish on
account of its calibration invariance; that is, we have
∂sk (π)
Z  
i ∂π
Z
1 i
dx + −w + W π dx = 0.
X ∂xk X ∂xi 2 i
If we transform the first term of the second integral by means of
partial integration, we may write, instead of the preceding equa-
tion,


∂ sk (π) − πwk
Z  i 
∂w
Z
1 i
dx + π + 2 Wi dx = 0. (73)
X ∂xk X ∂xi
 CHAPTER IV 432

This immediately gives the identity

∂wi 1 i
+ 2 Wi = 0 (74)
∂xi
in the manner familiar in the calculus of variations. If the function
of position on the left were different from 0 at a point xi , say
positive, then it would be possible to mark off a neighbourhood X
of this point so small that this function would be positive at every
point within X. If we choose this region for X in (73), but choose
for π a function which vanishes for points outside X but is > 0
throughout X, then the first integral vanishes, but the second is
found to be positive—which contradicts equation (73). Now that
this has been ascertained, we see that (73) gives


∂ sk (π) − πwk
Z
dx = 0.
X ∂xk
For a given scalar-field π it holds for every finite region X, and
consequently we must have


∂ sk (π) − πwk
= 0. (75)
∂xk
If we substitute (72) in this, and observe that, for a particular
∂π ∂ 2π
point, arbitrary values may be assigned to π, , , then
∂x ∂xi ∂xk
this single formula resolves into the identities:
∂sk ∂wk ∂hαi
= ; si + = wi ; hαβ + hβα = 0. (751,2,3 )
∂xk ∂xk ∂xα
According to the third identity, hik is a linear tensor-density of
the second order. In view of the skew-symmetry of h the first is
 THE GENERAL THEORY OF RELATIVITY 433

a result of the second, since

∂ 2 hαβ
= 0.
∂xα ∂xβ
II. We subject the world-continuum to an infinitesimal de-
formation, in which each point undergoes a displacement whose
components are ξ i ; let the metrical structure accompany the de-
formation without being changed. Let δ signify the change occa-
sioned by the deformation in a quantity, if we remain at the same
space-time point, δ 0 the change in the same quantity if we share
in the displacement of the space-time point. Then, by (20), (210 ),
(71)
∂ξ r
 
∂φi r ∂π 

−δφi = φr + ξ + ,
∂xi ∂xr ∂xi 
(76)
∂ξ r ∂ξ r ∂gik r
 

−δgik = gir + gkr + ξ − πgik ,
∂xk ∂xi ∂xr
in which π denotes an infinitesimal scalar-field that has still been
left arbitrary by our conventions. The invariance of the Action
with respect to transformation of co-ordinates and change of cali-
bration is expressed in the formula which relates to this variation:
∂(Wξ k )
Z Z  
0
δ W dx = + δW dx = 0. (77)
X X ∂xk
If we wish to express the invariance with respect to the co-
ordinates alone we must make π = 0; but the resulting formulæ
of variation (76) have not then an invariant character. This
convention, in fact, signifies that the deformation is to make the
two groundforms vary in such a way that the measure l of a
 CHAPTER IV 434

line-element remains unchanged, that is, δ 0 l = 0. This equation

does not, however, express the process of congruent transference
of a distance, but indicates that
δ 0 l = −l(φi δ 0 xi ) = −l(φi ξ i ).
Accordingly, in (76) we must choose π not equal to zero but equal
to −(φi ξ i ) if we are to arrive at invariant formulæ, namely,
−δφi = fir ξ r ,


∂ξ r ∂ξ r (78)
   
∂gik
−δgik = gir + gkr + + gik φr ξ r .
∂xk ∂xi ∂xr
The change in the two groundforms which it represents is one
that makes the metrical structure appear carried along unchanged
by the deformation and every line-element to be transferred con-
gruently. The invariant character is easily recognised analytically,
too; particularly in the case of the second equation (78), if we
introduce the mixed tensor
∂ξ i
+ Γikr ξ r = ξki .
∂xk
The equation then becomes
−δgik = ξik + ξki .
Now that the calibration invariance has been applied in I, we may
in the case of (76) restrict ourselves to the choice of π, which was
discussed just above, and which we found to be alone possible
from the point of view of invariance.
For the variation (78) let
Wξ k + δvk = Sk (ξ).
 THE GENERAL THEORY OF RELATIVITY 435

Sk (ξ) is a vector-density which depends in a linear differential

manner on the arbitrary vector-field ξ i . We write in an explicit
form
kα ∂ξ
i
∂ 2ξi
Sk (ξ) = Ski ξ i + Hi + 12 Hkαβ
i
∂xα ∂xα ∂xβ
(the last co-efficient is, of course, symmetrical in the indices α, β).
The fact that Sk (ξ) is a vector-density dependent on the vector-
field ξ i expresses most simply and most fully the character of
invariance possessed by the co-efficients which occur in the ex-
pression for Sk (ξ); in particular, it follows from this that the Ski ’s
are not components of a mixed tensor-density of the second order:
we call them the components of a “pseudo-tensor-density”. If we
insert in (77) the expressions (70) and (78), we get an integral,
whose integrand is
∂Sk (ξ) ∂ξ i
   
i k 1 ∂gαβ αβ
− ξ fki w + 2 + gαβ φi W Wik .
∂xk ∂xi ∂xk
On account of
∂gαβ
+ gαβ φi = Γα,βi + Γβ,αi
∂xi
and of the symmetry of Wαβ we find
 
1 ∂gαβ
2
+ gαβ φi Wαβ = Γα,βi Wαβ = Γαβi Wαβ .
∂xi
If we apply partial integration to the last member of the integrand,
we get


∂ Sk (ξ) − Wik ξ i
Z Z
dx + [. . . ]i ξ i dx = 0.
X ∂xk X
 CHAPTER IV 436

According to the method of inference used above we get from this

the identities:
∂Wik
 
[. . . ]i , that is, α β
− Γβ Wα + fik wk = 0 (79)
∂xk

and

∂ Sk (ξ) − Wik ξ i
= 0. (80)
∂xk
The latter resolves into the following four identities:

∂Ski ∂Wik αk
∂Hi
= ;

Ski + = Wik ;


∂xk ∂xk ∂xα

(801,2,3,4 )
αβ βα ∂Hγαβ
i

(Hi + Hi ) + = 0; Hαβγ + Hβγα + Hγαβ = 0.

∂xγ i i i 

If from (4 ) we replace in (3 )
γαβ
Hi by − Hαβγ
i − Hβαγ
i ,

we get that
αβ ∂Hαβγ
Hi − i
= Hαβ
i
∂xγ
is skew-symmetrical in the indices α, β. If we introduce Hαβi
αβ
in place of Hi we see that (3 ) and (4 ) are merely statements
regarding symmetry, but (2 ) becomes

∂Hαk
i ∂ 2 Hαβk
i
Ski + + = Wik . (81)
∂xα ∂xα ∂xβ
 THE GENERAL THEORY OF RELATIVITY 437

(1 ) follows from this because, on account of the conditions of sym-

metry
∂ 2 Hαβ
i ∂ 3 Hαβγ
i
= 0, we get = 0.
∂xα ∂xβ ∂xα ∂xβ ∂xγ
Example.—In the case of Maxwell’s Action-density we have,
as is immediately obvious

δvk = f ik δφi .

Consequently

si = 0, hik = f ik ; Ski = lδik − f iα f kα , and the quantities H = 0.

Hence our identities lead to

∂f αi
i ∂wi
w = = 0, Wii = 0,
∂xα ∂xi
 k
∂f βα

k k ∂Si 1 ∂gαβ αβ
Wi = Si −2 S + fiα = 0.
∂xk ∂xi ∂xβ

We arrived at the last two formulæ by calculation earlier, the for-

mer on page 343, the latter on page 247; the latter was found to ex-
press the desired connection between Maxwell’s tensor-density Ski
of the field-energy and the ponderomotive force.
Field Laws and Theorems of Conservation.—If, in (70),
we take for δ an arbitrary variation which vanishes outside a finite
region, and for X we take the whole world or a region such that,
outside it, δ = 0, we get
Z Z
δW dx = (wi δφi + 12 Wik δgik ) dx.
 CHAPTER IV 438

Z
If W dx is the Action, we see from this that the following in-
variant laws are contained in Hamilton’s Principle:

wi = 0, Wik = 0.

Of these, we have to call the former the electromagnetic laws,

the latter the gravitational laws. Between the left-hand sides of
these equations there are five identities, which have been stated
in (74) and (79). Thus there are among the field-equations five
superfluous ones corresponding to the transition (dependent on
five arbitrary functions) from one system of reference to another.
According to (752 ) the electromagnetic laws have the following
form:
∂hik
= si [and (67)] (82)
∂xk
in full agreement with Maxwell’s Theory; si is the density of the
4-current, and the linear tensor-density of the second order hik
is the electromagnetic density of field. Without specialising the
Action at all we can read off the whole structure of Maxwell’s The-
ory from the calibration invariance alone. The particular form of
Hamilton’s function W affects only the formulæ which state that
current and field-density are determined by the phase-quantities
φi , gik of the æther. In the case of Maxwell’s Theory in the re-
stricted sense (W = l), which is valid only in empty space, we get
hik = f ik , si = 0, which is as it should be.
Just as the si ’s constitute the density of the 4-current, so the
scheme of Ski ’s is to be interpreted as the pseudo-tensor-density of
the energy. In the simplest case, W = l, this explanation becomes
identical with that of Maxwell. According to (751 ) and (801 ) the
 THE GENERAL THEORY OF RELATIVITY 439

theorems of conservation
∂si ∂Ski
= 0, =0
∂xi ∂xk
are generally valid; and, indeed, they follow in two ways from
∂si ∂wi
the field laws. For is not only identically equal to , but
∂xi ∂xi
∂Ski ∂Wik
also to − 21 Wii , and is not only identically equal to , but
∂xk ∂xk
also to Γαiβ Wαβ − fik wk . The form of the gravitational equations
is given by (81). The field laws and their accompanying laws of
conservation may, by (75) and (80), be summarised conveniently
in the two equations
∂si (π) ∂Si (ξ)
= 0, = 0.
∂xi ∂xi
Attention has already been directed above to the intimate con-
nection between the laws of conservation of the energy-momentum
and the co-ordinate-invariance. To these four laws there is to be
added the law of conservation of electricity, and, corresponding to
it, there must, logically, be a property of invariance which will in-
troduce a fifth arbitrary function; the calibration-invariance here
appears as such. Earlier we derived the law of conservation of
energy-momentum from the co-ordinate-invariance only owing to
the fact that Hamilton’s function consists of two parts, the Ac-
tion-function of the gravitational field and that of the “physical
phase”; each part had to be treated differently, and the component
results had to be combined appropriately (§ 33). If those quanti-
ties, which are derived from Wξ k + δvk by taking the variation of
the fundamental quantities from (76) for the case π = 0, instead
 CHAPTER IV 440

of from (78), are distinguished by a prefixed asterisk, then, in

consequence of the co-ordinate-invariance, the “theorems of con-
∂ ∗ Ski
servation” = 0 are generally valid. But the ∗ Ski ’s are not
∂xk
the energy-momentum components of the twofold action-function
which have been used as a basis since § 28. For the gravitational
component (W = G) we defined the energy by means of ∗ Ski
(§ 33), but for the electromagnetic component (W = L, § 28) we
introduced Wik as the energy components. This second compo-
nent L contains only the gik ’s themselves, not their derivatives;
for a quantity of this kind we have, by (802 ), Wik = Ski . Hence (if
we use the transformations which the fundamental quan-
tities undergo during an infinitesimal alteration of the
calibration), we can adapt the two different definitions of energy
to one another although we cannot reconcile them entirely. These
discrepancies are removed only here since it is the new theory
which first furnishes us with an explanation of the current si , of
the electromagnetic density of field hik , and of the energy Ski ,
which is no longer bound by the assumption that the Action is
composed of two parts, of which the one does not contain the φi ’s
and their derivatives, and the other does not contain the deriva-
tives of the g ik ’s. The virtual deformation of the world-continuum
which leads to the definition of Ski must, accordingly, carry along
the metrical structure and the line-elements “unchanged” in our
sense and not in that of Einstein. The laws of conservation of
the si ’s and the Ski ’s are then likewise not bound by an assumption
concerning the composition of the Action. Thus, after the total
energy had been introduced in § 33, we have once again passed
beyond the stand taken in § 28 to a point of view which gives a
more compact survey of the whole. What is done by Einstein’s
 THE GENERAL THEORY OF RELATIVITY 441

theory of gravitation with respect to the equality of inertial and

gravitational matter, namely, that it recognises their identity as
necessary but not as a consequence of an undiscovered law of phys-
ical nature, is accomplished by the present theory with respect to
the facts that find expression in the structure of Maxwell’s equa-
tions and the laws of conservation. Just as is the case in § 33 in
which we integrate over the cross-section of a canal of the system,
so we find here that, as a result of the laws of conservation, if
the si ’s and Ski ’s vanish outside the canal, the system has a con-
stant charge e and a constant energy-momentum J. Both may
be represented, by Maxwell’s equations (82) and the gravitational
equations (81), as the flux of a certain spatial field through a sur-
face Ω that encloses the system. If we regard this representation
as a definition, the integral theorems of conservation hold, even if
the field has a real singularity within the canal of the system. To
prove this, let us replace this field within the canal in any arbi-
trary way (preserving, of course, a continuous connection with the
region outside it) by a regular field, and let us define the si ’s and
the Ski ’s by the equations (82), (81) (in which the right-hand sides
are to be replaced by zero) in terms of the quantities h and H
belonging to the altered field. The integrals of these fictitious
quantities s0 and S0i , which are to be taken over the cross-section
of the canal (the interior of Ω), are constant; on the other hand,
they coincide with the fluxes mentioned above over the surface Ω,
since on Ω the imagined field coincides with the real one.
 CHAPTER IV 442

§ 36. Application of the Simplest Principle of Action.

The Fundamental Equations of Mechanics
We have now to show that if we uphold our new theory it is
possible to make an assumption about W which, as far as the re-
sults that have been confirmed in experience are concerned, agrees
with Einstein’s Theory. The simplest assumption∗ for purposes of
calculation (I do not insist that it is realised in nature) is:
√
W = − 41 F 2 g + αl. (83)

The quantity Action is thus to be composed of the volume, mea-

sured in terms of the radius of curvature of the world as unit of
length (cf. (62), § 17) and of Maxwell’s action of the electromag-
netic field; the positive constant α is a pure number. It follows
that
√ √
δW = − 12 F δ(F g) + 14 F 2 δ g + α δl.
We assume that −F is positive; the calibration may then be
uniquely determined by the postulate F = −1; thus
√ √
δW = the variation of 21 F g + 14 g + αl.

If we use the formula (61), § 17 for F , and omit the divergence

√
(∂ gφi )
δ
∂xi
which vanishes when we integrate over the world, and if, by means
√
of partial integration, we convert the world-integral of δ( 12 R g)
∗
Vide note 36.
 THE GENERAL THEORY OF RELATIVITY 443

into the integral of δG (§ 28), then our principle of action takes

the form
√ 
Z
δ V dx = 0, and we get V = G+αl+ 14 g 1−3(φi φi ) . (84)

This normalisation denotes that we are measuring with cosmic

measuring rods. If, in addition, we choose the co-ordinates xi so
that points of the world whose co-ordinates differ by amounts of
the order of magnitude 1, are separated by cosmic distances, then
we may assume that the gik ’s and the φi ’s are of the order of mag-
nitude 1. (It is, of course, a fact that the potentials vary percepti-
bly by amounts that are extraordinarily small in comparison with
cosmic distances.) By means of the substitution xi = x0i we intro-
duce co-ordinates of the order of magnitude in general use (that
is having dimensions comparable with those of the human body);
 is a very small constant. The gik ’s do not change during this
transformation, if we simultaneously perform the re-calibration
1
which multiplies ds2 by 2 . In the new system of reference we

then have
0
gik = gik , φ0i = φi ; F 0 = −2 .
1
is accordingly, in our ordinary measures, the radius of curvature

of the world. If gik , φi retain their old significance, but if we take
xi to represent the co-ordinates previously denoted by x0i , and if
Γrik are the components of the affine relationship corresponding to
these co-ordinates, then
2 √ 
g 1 − 3(φi φi ) ,

V = (G + αl) +
  4
ik
Γrik = + 12 2 (δir φk + δkr φi − gik φr ).
r
 CHAPTER IV 444

Thus, by neglecting the exceedingly small cosmological terms, we

arrive exactly at the classical Maxwell-Einstein theory of electric-
ity and gravitation. To make the expression correspond exactly
2
with that of § 34 we must set = λ. Hence our theory neces-
2
1 √
sarily gives us Einstein’s cosmological term λ g. The uniform
2
distribution of electrically neutral matter at rest over the whole of
(spherical) space is thus a state of equilibrium which is compatible
with our law. But, whereas in Einstein’s Theory (cf. § 34) there
must be a pre-established harmony between the universal physi-
cal constant λ that occurs in it, and the total mass of the earth
(because each of these quantities in themselves already determine
the curvature of the world), here (where λ denotes merely the
curvature), we have that the mass present in the world deter-
mines the curvature. It seems to the author that just this is
what makes Einstein’s cosmology physically possible. In the case
in which a physical field is present, Einstein’s cosmological term
3 √
must be supplemented by the further term − λ g(φi φi ); and in
2
the components Γrik of the gravitational field, too, a cosmological
term that is dependent on the electromagnetic potentials occurs.
Our theory is founded on a definite unit of electricity; let it be e in
ordinary electrostatic units. Since, in (84), if we use these units,
2κ
occurs in place of α, we have
c2
√
2e2 κ
r
α e κ 1 α
= , = :
c2 −F c  2
our
r unit is that quantity of electricity whose gravitational radius is
α
times the radius of curvature of the world. It is, therefore, like
2
 THE GENERAL THEORY OF RELATIVITY 445

the quantum of action l, of cosmic dimensions. The cosmological

factor which Einstein added to his theory later is part of ours from
the very beginning.
Variation of the φi ’s gives us Maxwell’s equations
∂f ik
= si
∂xk
and, in this case, we have simply
3λ √
si = − φi g.
α
Just as according to Maxwell the æther is the seat of energy and
mass so we obtain here an electric charge (plus current) diffused
thinly throughout the world. Variation of the gik ’s gives the grav-
itational equations
√
k R+λ g k
Ri − δi = αTki (85)
2
where
Tki = l + 12 (φr sr ) δik − fir f kr = φi sk .


The conservation of electricity is expressed in the divergence equa-

tion √
∂( gφi )
= 0. (86)
∂xi
This follows, on the one hand, from Maxwell’s equations, but
must, on the other hand, be derivable from the gravitational equa-
tions according to our general results. We actually find, by con-
tracting the latter equations with respect to i, k, that

R + 2λ = 32 (φi φi ),
 CHAPTER IV 446

and this in conjunction with −F = 2λ again gives (86). We get

for the pseudo-tensor-density of the energy-momentum, as is to
be expected
 
k k 1 √ k 1 ∂gαβ αβ,k
Si = αTi + G + 2 λ gδi − 2 G .
∂xi
Z
From the equation δ V dx = 0 for a variation δ 0 which is pro-
0

duced by the displacement in the true sense [from formula (76)

with ξ i = const., π = 0], we get
∂(∗ Ski ξ i )
= 0, (87)
∂xk
where
∗ k ∂gαβ αβ,k
1 ∂φ kr
Si = Vδik − 2
G +α f .
∂xi ∂xi
To obtain the conservation theorems, we must, according to our
earlier remarks, write Maxwell’s equations in the form
 
i ∂π ik
∂ πs + f
∂xk
= 0,
∂xi
then set π = −(φi ξ i ), and, after multiplying the resulting equation
by α, add it to (87). We then get, in fact,
∂(Ski ξ i )
= 0.
∂xk
The following terms occur in Ski : the Maxwell energy-density of
the electromagnetic field
lδik − fir f kr ,
 THE GENERAL THEORY OF RELATIVITY 447

the gravitational energy

1 ∂gαβ αβ,k
Gδik − 2
G ,
∂xi
and the supplementary cosmological terms
1 √
2
(λ g + φr sr )δik − φi sk .

The statical world is by its own nature calibrated. The ques-

tion arises whether F = const. for this calibration. The answer is
in the affirmative. For if we re-calibrate the statical world in ac-
cordance with the postulate F = −1 and distinguish the resulting
quantities by a horizontal bar, we get
Fi ∂F
φ̄i = − , where we set Fi = (i = 1, 2, 3),
F ∂xi
g ik √ √
ḡik = −F gik , that is, ḡ ik = − , g = F 2 g,
F
and equation (86) gives
3
X ∂Fi √
=0 (Fi = gF i ).
i=1
∂xi

From this, however, it follows that F = const.

From the fact that a further electrical term becomes added to
Einstein’s cosmological term, the existence of a material particle
becomes possible without a mass horizon becoming necessary. The
particle is necessarily charged electrically. If, in order to determine
the radially symmetrical solutions for the statical case, we again
 CHAPTER IV 448

use the old terms of § 31, and take φ to mean the electrostatic
potential, then the integral whose variation must vanish, is

αr2 φ02 λr2 3h2 φ2

Z Z   
2 0
Vr dr = w∆ − + ∆− dr
2∆ 2 2∆

(the accent denotes differentiation with respect to r). Variation

of w, ∆, and φ, respectively, leads to the equations
3λ 4 2
∆∆0 = h φ r,
4 
λr2 h2 φ2 α r2 φ02

w0 = 1 + 32 + ,
2 ∆2 2 ∆2
 2 0 0
r φ 3 h2 r2 φ
= .
∆ 2α ∆

As a result of the normalisations that have been performed, the

spatial co-ordinate system is fixed except for a Euclidean rotation,
and hence h2 is uniquely determined. In f and φ, as a result of
the free choice of the unit of time, a common constant factor
remains arbitrary (a circumstance that may be used to reduce
the order of the problem by 1). If the equator of the space is
reachedpwhen r = r0 , then the quantities that occur as functions
of z = r02 − r2 must exhibit the following behaviour for z = 0:
f and φ are regular, and f 6= 0; h2 is infinite to the second order,
∆ to the first order. The differential equations themselves show
that the development of h2 z 2 according to powers of z begins with
the term h20 , where
2r2
h20 = 2 0
λr0 − 2
 THE GENERAL THEORY OF RELATIVITY 449

—this proves, incidentally, that λ must be positive (the curva-

2
ture F negative) and that r02 > —whereas for the initial values
λ
f0 , φ0 , of f and φ we have
3λ 2 2
f02 = hφ.
4 0 0
If diametral points are to be identified, φ must be an even function
of z, and the solution is uniquely determined by the initial values
for z = 0, which satisfy the given conditions (vide note 37). It
cannot remain regular in the whole region 0 ≤ r ≤ r0 , but must,
if we let r decrease from r0 , have a singularity at least ultimately
when r = 0. For otherwise it would follow, by multiplying the
differential equation of φ by φ, and integrating from 0 to r0 , that
Z r0 2  
r 02 3 2 2
φ + h φ dr = 0.
0 ∆ 2α
Matter is accordingly a true singularity of the field. The fact that
the phase-quantities vary appreciably in regions whose linear di-
1
mensions are very small in comparison with √ may be explained,
l
perhaps, by the circumstance that a value must be taken for r02
1
which is enormously great in comparison with . The fact that
λ
all elementary particles of matter have the same charge and the
same mass seems to be due to the circumstance that they are all
embedded in the same world (of the same radius r0 ); this agrees
with the idea developed in § 32, according to which the charge and
the mass are determined from infinity.
In conclusion, we shall set up the mechanical equations that
govern the motion of a material particle. In actual fact we have
 CHAPTER IV 450

not yet derived these equations in a form which is admissible from

the point of view of the general theory of relativity; we shall now
endeavour to make good this omission. We shall also take this
opportunity of carrying out the intention stated in § 32, that is,
to show that in general the inertial mass is the flux of the gravi-
tational field through a surface which encloses the particle, even
when the matter has to be regarded as a singularity which limits
the field and lies, so to speak, outside it. In doing this we are, of
course, debarred from using a substance which is in motion; the
hypotheses corresponding to the latter idea, namely (§ 27):
dm ds = µ dx, Tki = µui uk
are quite impossible here, as they contradict the postulated prop-
erties of invariance. For, according to the former equation, µ is
a scalar-density of weight 21 , and, according to the latter, one of
weight 0, since Tki is a tensor-density in the true sense. And we
see that these initial conditions are impossible in the new theory
for the same reason as in Einstein’s Theory, namely, because they
lead to a false value for the mass, as was mentioned at the end
of § 33. This is obviouslyZ intimately connected with the circum-
stance that the integral dm ds has now no meaning at all, and
hence cannot be introduced as “substance-action of gravitation”.
We took the first step towards giving a real proof of the mechan-
ical equations in § 33. There we considered the special case in
which the body is completely isolated, and no external forces act
on it.
From this we see at once that we must start from the laws of
conservation
∂Ski
=0 (89)
∂xk
 THE GENERAL THEORY OF RELATIVITY 451

which hold for the total energy. Let a volume Ω, whose dimen-
sions are great compared with the actual essential nucleus of the
particle, but small compared with those dimensions of the external
field which alter appreciably, be marked off around the material
particle. In the course of the motion Ω describes a canal in the
world, in the interior of which the current filament of the material
particle flows along. Let the co-ordinate system consisting of the
“time-co-ordinate” x0 = t and the “space-co-ordinates” x1 , x2 , x3 ,
be such that the spaces x0 = const. intersect the canal (the cross-
section is the volume Ω mentioned above). The integrals
Z
S0i dx1 dx2 dx3 = Ji ,
Ω

which are to be taken in a space x0 = const. over Ω, and which

are functions of the time alone, represent the energy (i = 0) and
the momentum (i = 1, 2, 3) of the material particle. If we in-
tegrate the equation (89) in the space x0 = const. over Ω, the
dJi
first member (k = 0) gives the time-derivative ; the integral
dt
sum over the three last terms, however, becomes transformed by
Gauss’ Theorem into an integral Ki which is to be taken over the
surface of Ω. In this way we arrive at the mechanical equations
dJi
= Ki . (90)
dt
On the left side we have the components of the “inertial force,”
and on the right the components of the external “field-force”. Not
only the field-force but also the four-dimensional momentum Ji
may be represented, in accordance with a remark at the end of
§ 35, as a flux through the surface of Ω. If the interior of the canal
 CHAPTER IV 452

encloses a real singularity of the field the momentum must, in-

deed, be defined in the above manner, and then the device of the
“fictitious field,” used at the end of § 35, leads to the mechanical
equations proved above. It is of fundamental importance to notice
that in them only such quantities are brought into relationship with
one another as are determined by the course of the field outside
the particle (on the surface of Ω), and have nothing to do with the
singular states or phases in its interior. The antithesis of kinetic
and potential which receives expression in the fundamental law
of mechanics does not, indeed, depend actually on the separation
of energy-momentum into one part belonging to the external field
and another belonging to the particle (as we pictured it in § 25),
but rather on this juxtaposition, conditioned by the resolution
into space and time, of the first and the three last members of
the divergence equations which make up the laws of conservation,
that is, on the circumstance that the singularity canals of the ma-
terial particles have an infinite extension in only one dimension,
but are very limited in three other dimensions. This stand was
taken most definitely by Mie in the third part of his epoch-making
Foundations of a Theory of Matter, which deals with “Force and
Inertia” (vide note 38). Our next object is to work out the full
consequences of this view for the principle of action adopted in
this chapter.
To do this, it is necessary to ascertain exactly the meaning of
the electromagnetic and the gravitational equations. If we discuss
Maxwell’s equations first, we may disregard gravitation entirely
and take the point of view presented by the special theory of
relativity. We should be reverting to the notion of substance if we
 THE GENERAL THEORY OF RELATIVITY 453

were to interpret the Maxwell-Lorentz equation

∂f ik
= ρui
∂xk
so literally as to apply it to the volume-elements of an electron. Its
true meaning is rather this: Outside the Ω-canal, the homogeneous
equations
∂f ik
=0 (91)
∂xk
hold. The only statical radially symmetrical solution f¯ik of (91)
e
is that derived from the potential ; it gives the flux e (and not 0,
r
as it would be in the case of a solution of (91) which is free from
singularities) of the electric field through an envelope Ω enclosing
the particle. On account of the linearity of equations (91), these
properties are not lost when an arbitrary solution fik of equa-
tions (91), free from singularities, is added to f¯ik ; such a one is
given by fik = const. The field which surrounds the moving
electron must be of the type: fik + f¯ik , if we introduce at
the moment under consideration a co-ordinate system in which
the electron is at rest. This assumption concerning the consti-
tution of the field outside Ω is, of course, justified only when we
are dealing with quasi-stationary motion, that is, when the world-
line of the particle deviates by a sufficiently small amount from a
straight line. The term ρui in Lorentz’s equation is to express the
general effect of the charge-singularities for a region that contains
many electrons. But it is clear that this assumption comes into
question only for quasi-stationary motion. Nothing at all can
be asserted about what happens during rapid acceleration. The
opinion which is so generally current among physicists nowadays,
 CHAPTER IV 454

that, according to classical electrodynamics, a greatly accelerated

particle emits radiation, seems to the author quite unfounded. It
is justified only if Lorentz’s equations are interpreted in the too
literal fashion repudiated above, and if, also, it is assumed that
the constitution of the electron is not modified by the accelera-
tion. Bohr’s Theory of the Atom has led to the idea that
there are individual stationary orbits for the electrons circulat-
ing in the atom, and that they may move permanently in these
orbits without emitting radiations; only when an electron jumps
from one stationary orbit to another is the energy that is lost
by the atom emitted as electromagnetic energy of vibration (vide
note 39). If matter is to be regarded as a boundary-singularity of
the field, our field-equations make assertions only about the pos-
sible states of the field, and not about the conditioning
of the states of the field by the matter. This gap is filled
by the Quantum Theory in a manner of which the underlying
principle is not yet fully grasped. The above assumption about
the singular component f¯ of the field surrounding the particle is,
in our opinion, true for a quasi-stationary electron. We may, of
course, work out other assumptions. If, for example, the particle
is a radiating atom, the f¯ik ’s will have to be represented as the
field of an oscillating Hertzian dipole. (This is a possible state of
the field which is caused by matter in a manner which, according
to Bohr, is quite different from that imagined by Hertz.)
As far as gravitation is concerned, we shall for the present
adopt the point of view of the original Einstein Theory. In it the
(homogeneous) gravitational equations have (according to § 31) a
statical radially symmetrical solution, which depends on a single
constant m, the mass. The flux of a gravitational field through
a sufficiently great sphere described about the centre is not equal
 THE GENERAL THEORY OF RELATIVITY 455

to 0, as it should be if the solution were free from singularities,

but equal to m. We assume that this solution is characteristic of
the moving particle in the following sense: We consider the values
traversed by the gik ’s outside the canal to be extended over the
canal, by supposing the narrow deep furrow, which the path of
the material particle cuts out in the metrical picture of the world,
to be smoothed out, and by treating the stream-filament of the
particle as a line in this smoothed-out metrical field. Let ds be the
corresponding proper-time differential. For a point of the stream-
filament we may introduce a (“normal”) co-ordinate system such
that, at that point,

ds2 = dx20 − (dx21 + dx22 + dx23 ),

∂gαβ
the derivatives vanish, and the direction of the stream-
∂xi
filament is given by

dx0 : dx1 : dx2 : dx3 = 1 : 0 : 0 : 0.

In terms of these co-ordinates the field is to be expressed by

the above-mentioned statical solution (only, of course, in a cer-
tain neighbourhood of the world-point under consideration, from
which the canal of the particle is to be cut out). If we regard
the normal co-ordinates xi as Cartesian co-ordinates in a four-
dimensional Euclidean space, then the picture of the world-line
of the particle becomes a definite curve in the Euclidean space.
Our assumption is, of course, admissible again only if the motion
is quasi-stationary, that is, if this picture-curve is only slightly
curved at the point under consideration. (The transformation of
the homogeneous gravitational equations into non-homogeneous
 CHAPTER IV 456

ones, on the right side of which the tensor µui uk appears, takes
account of the singularities, due to the presence of masses, by fus-
ing them into a continuum; this assumption is legitimate only in
the quasi-stationary case.)
To return to the derivation of the mechanical equations! We
shall use, once and for all, the calibration normalised by F =
const., and we shall neglect the cosmological terms outside the
canal. The influence of the charge of the electron on the gravita-
tional field is, as we know from § 32, to be neglected in compar-
ison with the influence of the mass, provided the distance from
the particle is sufficiently great. Consequently, if we base our cal-
culations on the normal co-ordinate system, we may assume the
gravitational field to be that mentioned above. The determination
of the electromagnetic field is then, as in the gravitational case, a
linear problem; it is to have the form fik + f¯ik mentioned above
(with fik = const. on the surface of Ω). But this assumption is
compatible with the field-laws only if e = const. To prove this,
we shall deduce from a fictitious field that fills the canal regularly
and that links up with the really existing field outside, that

∂f ik
Z
=s,i
s0 dx1 dx2 dx3 = e∗
∂xk Ω

in any arbitrary co-ordinate system; e∗ is independent of the choice

of the fictitious field, inasmuch as it may be represented as a field-
flux through the surface of Ω. Since (if we neglect the cosmological
terms) the si ’s on this surface vanish, the equation of definition
∂si de∗
gives us, if = 0 is integrated, = 0; moreover, the argu-
∂xi dt
ments set out in § 33 show that e∗ is independent of the co-ordinate
system chosen. If we use the normal co-ordinate system at one
 THE GENERAL THEORY OF RELATIVITY 457

point, the representation of e∗ as a field-flux shows that e∗ = e.

Passing on from the charge to the momentum, we must notice
at once that, with regard to the representation of the energy-
momentum components by means of field-fluxes, we may not refer
to the general theory of § 35, because, by applying the process of
partial integration to arrive at (84), we sacrificed the co-ordinate
invariance of our Action. Hence we must proceed as follows. With
the help of the fictitious field which bridges the canal regularly,
we define αSki by means of
 
k 1 k k 1 ∂gαβ αβ,k
(Ri − 2 δi R) + Gδi − 2 G .
∂xi
The equation
∂Ski
=0 (92)
∂xk
is an identity for it. By integrating (92) we get (90), whereby
Z
Ji = S0i dx1 dx2 dx3 .
Ω

Ki expresses itself as the field-flux through the surface Ω. In these

expressions the fictitious field may be replaced by the real one,
and, moreover, in accordance with the gravitational equations,
we may replace
1 k 1 k
(R − 2 δi R) by lδik − fir f kr .
α i
If we use the normal co-ordinate system the part due to the gravi-
tational energy drops out; for its components depend not only lin-
∂gαβ
early but also quadratically on the (vanishing) derivatives .
∂xi
 CHAPTER IV 458

We are, therefore, left with only the electromagnetic part, which

is to be calculated along the lines of Maxwell. Since the com-
ponents of Maxwell’s energy-density depend quadratically on the
field f + f¯, each of them is composed of three terms in accordance
with the formula

(f + f¯)2 = f 2 + 2ff¯ + f¯2 .

In the case of each, the first term contributes nothing, since the
flux of a constant vector through a closed surface is 0. The last
term is to be neglected since it contains the weak field f¯ as a
square; the middle term alone remains. But this gives us

Ki = ef0i .

Concerning the momentum-quantities we see (in the same way

as in § 33, by using identities (92) and treating the cross-section
of the stream-filament as infinitely small in comparison with the
external field) (1) that, for co-ordinate transformations that are
to be regarded as linear in the cross-section of the canal, the Ji ’s
are the co-variant components of a vector which is independent
of the co-ordinate system; and (2) that if we alter the fictitious
field occupying the canal (in § 33 we were concerned, not with
this, but with a charge of the co-ordinate system in the canal)
the quantities Ji retain their values. In the normal co-ordinate
system, however, for which the gravitational field that surrounds
the particle has the form calculated in § 31, we find that, since
the fictitious field may be chosen as a statical one, according to
page 408: J1 = J2 = J3 = 0, and J0 = the flux of a spatial vector-
density through the surface of Ω, and hence = m. On account of
the property of co-variance possessed by Ji , we find that not only
 THE GENERAL THEORY OF RELATIVITY 459

at the point of the canal under consideration, but also just before
it and just after it
 
i dxi
Ji = mui u = .
ds
Hence the equations of motion of our particle expressed in the
normal co-ordinate system are
d(mui )
= ef0i . (93)
dt
dm
The 0th of these equations gives us: = 0; thus the field equa-
dt
tions require that the mass be constant. But in any arbitrary
co-ordinate system we have:
d(mui ) 1 ∂gαβ
− 2
muα uβ = e · fki uk . (94)
ds ∂xi
For the relations (94) are invariant with respect to co-ordinate
transformations, and agree with (93) in the case of the normal co-
ordinate system. Hence, according to the field-laws, a necessary
condition for a singularity canal, which is to fit into the remaining
part of the field, and in the immediate neighbourhood of which the
field has the required structure, is that the quantities e and m
that characterise the singularity at each point of the canal remain
constant along the canal, but that the world-direction of the canal
satisfy the equations
dui 1 ∂gαβ α β e
− 2
u u = · fki uk .
ds ∂xi m
In the light of these considerations, it seems to the author that
the opinion expressed in § 25 stating that mass and field-energy
 CHAPTER IV 460

are identical is a premature inference, and the whole of Mie’s

view of matter assumes a fantastic, unreal complexion. It was,
of course, a natural result of the special theory of relativity that
we should come to this conclusion. It is only when we arrive at
the general theory that we find it possible to represent the mass
as a field-flux, and to ascribe to the world relationships such as
obtain in Einstein’s Cylindrical World (§ 34), when there are cut
out of it canals of circular cross-section which stretch to infinity in
both directions. This view of m states not only that inertial and
gravitational masses are identical in nature, but also that mass
as the point of attack of the metrical field is identical in nature
with mass as the generator of the metrical field. That which is
physically important in the statement that energy has inertia still
persists in spite of this. For example, a radiating particle loses
inertial mass of exactly the same amount as the electromagnetic
energy that it emits. (In this example Einstein first recognised
the intimate relationship between energy and inertia.) This may
be proved simply and rigorously from our present point of view.
Moreover, the new standpoint in no wise signifies a relapse to
the old idea of substance, but it deprives of meaning the prob-
lem of the cohesive pressure that holds the charge of the electron
together.
With about the same reasonableness as is possessed by Ein-
stein’s Theory we may conclude from our results Z that a clock in
quasi-stationary motion indicates the proper time ds which cor-
responds to the normalisation F = const.∗ If during the motion
∗
The invariant quadratic form F · ds2 is very far from being distinguished
from all other forms of the type E · ds2 (E being a scalar of weight −1) as
is the ds2 of Einstein’s Theory, which does not contain the derivatives of the
 THE GENERAL THEORY OF RELATIVITY 461

of a clock (e.g. an atom) with infinitely small period, the world-

distance traversed by it during a period were to be transferred con-
gruently from period to period in the sense of our world-geometry,
then two clocks which set out from the same world-point A with
the same period, that is, which traverse congruent world-distances
in A during their first period will have, in general, different periods
when they meet at a later world-point B. The orbital motion of
the electrons in the atom can, therefore, certainly not take place
in the way described, independently of their previous histories,
since the atoms emit spectral lines of definite frequencies. Neither
does a measuring rod at rest in a statical field undergo a congru-
ent transference; for the measure l = dσ 2 of a measuring rod at
rest does not alter, whereas for a congruent transference it would
dl
have to satisfy the equation = −l ·φ. What is the source of this
dt
discrepancy between the conception of congruent transference and
the behaviour of measuring rods, clocks, and atoms? We may dis-
tinguish two modes of determining a quantity in nature, namely,
that of persistence and that of adjustment. This difference is
illustrated in the following example. We may prescribe to the axis
of a rotating top any arbitrary direction in space; but once this
arbitrary initial direction has been fixed the direction of the axis
of the top when left to itself is determined from it for all time by
a tendency of persistence which is active from one moment to
another; at each instant the axis experiences an infinitesimal par-
potentials at all. For this reason the inference made in our calculation of
the displacement towards the infra-red (page 367), that similar atoms
radiate the same frequency measured in the proper time ds corresponding to
the normalisation F = const., is by no means as convincing as in the theory
of Einstein: it loses its validity altogether if a principle of action other than
that here discussed holds.
 CHAPTER IV 462

allel displacement. Diametrically opposed to this is the case of a

magnet needle in the magnetic field. Its direction is determined at
every moment, independently of the state of the system at other
moments, by the fact that the system, in virtue of its constitution,
adjusts itself to the field in which it is embedded. There is no
a priori ground for supposing a pure transference, following the
tendency of persistence, to be integrable. But even if this be the
case, as, for example, for rotations of the top in Euclidean space,
nevertheless two tops which set out from the same point with axes
in the same position, and which meet after the lapse of a great
length of time, will manifest any arbitrary deviations in the posi-
tions of the axes, since they can never be fully removed from all
influences. Thus although, for example, Maxwell’s equations for
the charge e of an electron make necessary the equation of conser-
de
vation = 0, this does not explain why an electron itself after
dt
an arbitrarily long time still has the same charge, and why this
charge is the same for all electrons. This circumstance shows that
the charge is determined not by persistence but by adjustment:
there can be only one state of equilibrium of negative electric-
ity, to which the corpuscle adjusts itself afresh at every moment.
The same reason enables us to draw the same conclusion for the
spectral lines of the atoms, for what is common to atoms emitting
equal frequencies is their constitution and not the equality of their
frequencies at some moment when they were together far back in
time. In the same way, obviously, the length of a measuring rod
is determined by adjustment; for it would be impossible to give
to this rod at this point of the field any length, say two or three
times as great as the one that it now has, in the way that I can
prescribe its direction arbitrarily. The world-curvature makes it
 THE GENERAL THEORY OF RELATIVITY 463

theoretically possible to determine a length by adjustment. In

consequence of its constitution the rod assumes a length which
has such and such a value in relation to the radius of curvature of
the world. (Perhaps the time of rotation of a top gives us an ex-
ample of a time-length that is determined by persistence; if what
we assumed above is true for direction then at each moment of
the motion of the top the rotation vector would experience a par-
allel displacement.) We may briefly summarise as follows: The
affine and metrical relationship is an a priori datum telling us
how vectors and lengths alter, if they happen to follow the
tendency of persistence. But to what extent this is the case in
nature, and in what proportion persistence and adjustment mod-
ify one another, can be found only by starting from the physical
laws that hold, i.e. from the principle of action.
The subject of the above discussion is the principle of action,
compatible with the new axiom of calibration invariance, which
most nearly approaches the Maxwell-Einstein theory. We have
seen that it accounts equally well for all the phenomena which
are explained by the latter theory and, indeed, that it has decided
advantages so far as the deeper problems, such as the cosmological
problems and that of matter are concerned. Nevertheless, I doubt
whether the Hamiltonian function (83) corresponds to reality. We
√
may certainly assume that W has the form W g, in which W is
an invariant of weight −2 formed in a perfectly rational manner
from the components of curvature. Only four of these invariants
may be set up, from which every other may be built up linearly
by means of numerical co-efficients (vide note 40). One of these
is Maxwell’s:
l = 41 fik f ik ; (95)
another is the F 2 used just above. But curvature is by its nature
 CHAPTER IV 464

a linear matrix-tensor of the second order: Fik dxi δxk . According

to the same law by which (95), the square of the numerical value,
is produced from the distance-curvature fik we may form
1
F F ik
4 ik
(96)

from the total curvature. The multiplication is in this case to

be interpreted as a composition of matrices; (96) is therefore it-
self again a matrix. But its trace L is a scalar of weight −2.
The two quantities L and l seem to be invariant and of the kind
sought, and they can be formed most naturally from the curva-
ture; invariants of this natural and simple type, indeed, exist only
in a four-dimensional world at all. It seems more probable that
W is a linear combination of L and l. Maxwell’s equations be-
come then as above: (when the calibration has been normalised
√
by F = const.) si = a constant multiple of gφi , and hik = f ik .
The gravitational laws in the statical case here, too, agree to a first
approximation with Newton’s laws. Calculations by Pauli (vide
note 41) have indeed disclosed that the field determined in § 31
is not only a rigorous solution of Einstein’s equations, but also of
those favoured here, so that the amount by which the perihelion
of Mercury’s orbit advances and the amount of the deflection of
light rays owing to the proximity of the sun at least do not conflict
with these equations. But in the question of the mechanical equa-
tions and of the relationship holding between the results obtained
by measuring-rods and clocks on the one hand and the quadratic
form on the other, the connecting link with the old theory seems
to be lost; here we may expect to meet with new results.
One serious objection may be raised against the theory in its
present state: it does not account for the inequality of positive
and negative electricity (vide note 42). There seem to be two
 THE GENERAL THEORY OF RELATIVITY 465

ways out of this difficulty. Either we must introduce into the law
of action a square root or some other irrationality; in the discus-
sion on Mie’s theory, it was mentioned how the desired inequality
could be caused in this way, but it was also pointed out what ob-
stacles lie in the way of such an irrational Action. Or, secondly,
there is the following view which seems to the author to give a
truer statement of reality. We have here occupied ourselves only
with the field which satisfies certain generally invariant functional
laws. It is quite a different matter to inquire into the excitation
or cause of the field-phases that appear to be possible according
to these laws; it directs our attention to the reality lying beyond
the field. Thus in the æther there may exist convergent as well
as divergent electromagnetic waves; but only the latter event can
be brought about by an atom, situated at the centre, which emits
energy owing to the jump of an electron from one orbit to an-
other in accordance with Bohr’s hypothesis. This example shows
(what is immediately obvious from other considerations) that the
idea of causation (in contra-distinction to functional relation) is
intimately connected with the unique direction of progress
characteristic of Time, namely Past → Future. This one-
ness of sense in Time exists beyond doubt—it is, indeed, the most
fundamental fact of our perception of Time—but a priori reasons
exclude it from playing a part in physics of the field, But we saw
above (§ 33) that the sign, too, of an isolated system is fully de-
termined, as soon as a definite sense of flow, Past → Future, has
been prescribed to the world-canal swept out by the system. This
connects the inequality of positive and negative electricity with
the inequality of Past and Future; but the roots of this problem
are not in the field, but lie outside it. Examples of such regulari-
ties of structure that concern, not the field, but the causes of the
 CHAPTER IV 466

field-phases are instanced: by the existence of cylindrically shaped

boundaries of the field: by our assumptions above concerning the
constitution of the field in their immediate neighbourhood: lastly,
and above all, by the facts of the quantum theory. But the way
in which these regularities have hitherto been formulated are, of
course, merely provisional in character. Nevertheless, it seems
that the theory of statistics plays a part in it which is funda-
mentally necessary. We must here state in unmistakable language
that physics at its present stage can in no wise be regarded as
lending support to the belief that there is a causality of physical
nature which is founded on rigorously exact laws. The extended
field, “æther,” is merely the transmitter of effects and is, of itself,
powerless; it plays a part that is in no wise different from that
which space with its rigid Euclidean metrical structure plays, ac-
cording to the old view; but now the rigid motionless character
has become transformed into one which gently yields and adapts
itself. But freedom of action in the world is no more restricted by
the rigorous laws of field physics than it is by the validity of the
laws of Euclidean geometry according to the usual view.
If Mie’s view were correct, we could recognise the field as ob-
jective reality, and physics would no longer be far from the goal of
giving so complete a grasp of the nature of the physical world, of
matter, and of natural forces, that logical necessity would extract
from this insight the unique laws that underlie the occurrence of
physical events. For the present, however, we must reject these
bold hopes. The laws of the metrical field deal less with real-
ity itself than with the shadow-like extended medium that serves
as a link between material things, and with the formal consti-
tution of this medium that gives it the power of transmitting
effects. Statistical physics, through the quantum theory, has
 THE GENERAL THEORY OF RELATIVITY 467

already reached a deeper stratum of reality than is accessible to

field physics; but the problem of matter is still wrapt in deepest
gloom. But even if we recognise the limited range of field physics,
we must gratefully acknowledge the insight to which it has helped
us. Whoever looks back over the ground that has been traversed,
leading from the Euclidean metrical structure to the mobile met-
rical field which depends on matter, and which includes the field
phenomena of gravitation and electromagnetism; whoever endeav-
ours to get a complete survey of what could be represented only
successively and fitted into an articulate manifold, must be over-
whelmed by a feeling of freedom won—the mind has cast off the
fetters which have held it captive. He must feel transfused with
the conviction that reason is not only a human, a too human,
makeshift in the struggle for existence, but that, in spite of all
disappointments and errors, it is yet able to follow the intelli-
gence which has planned the world, and that the consciousness of
each one of us is the centre at which the One Light and Life of
Truth comprehends itself in Phenomena. Our ears have caught a
few of the fundamental chords from that harmony of the spheres
of which Pythagoras and Kepler once dreamed.
 APPENDIX I
(Pp. 266 and 341)

To distinguish “normal” co-ordinate systems among all others

in the special theory of relativity, and to determine the metrical
groundform in the general theory, we may dispense with not only
rigid bodies but also with clocks.
In the special theory of relativity the postulate that, for the
transformation corresponding to the co-ordinates xi of a piece
of the world to an Euclidean “picture” space, the world-lines of
points moving freely under no forces are to become straight lines
(Galilei’s and Newton’s Principle of Inertia), fixes this picture
space except for an affine transformation. For the theorem,
that affine transformations of a portion of space are the only con-

Fig. 15.

tinuous ones which transform straight lines into straight lines,

holds. This is immediately evident if, in Möbius’ mesh construc-
tion (Fig. 12), we replace infinity by a straight line intersecting
our portion of space (Fig. 15). The phenomenon of light propaga-
tion then fixes infinity and the metrical structure in our four-
dimensional projective space; for its (three dimensional) “plane at

468
 APPENDIX I 469

infinity” E is characterised by the property that the light-cones

are projections, taken from different world-points, of one and the
same two-dimensional conic section situated in E.
In the general theory of relativity these deductions are best
expressed in the following form. The four-dimensional Riemann
space, which Einstein imagines the world to be, is a particular
case of general metrical space (§ 16). If we adopt this view we
may say that the phenomenon of light propagation determines
the quadratic groundform ds2 , whereas the linear one remains
unrestricted. Two different choices of the linear groundform which
differ by dφ = φi dxi correspond to two different values of the affine
relationship. Their difference is, according to formula (49), § 16,
given by
[Γiαβ ] = 12 (δαi φβ + δβi φα − gαβ φi ).
The difference between the two vectors that are derived from a
world-vector ui at the world-point O by means of an infinitesi-
mal parallel displacement of ui in its own direction (by the same
amount dxi =  · ui ), is therefore  times
ui (φα uα ) − 21 φi , (∗)
whereby we assume gαβ uα uβ = 1. If the geodetic lines passing
through O in the direction of the vector ui coincide for the two
fields, then the above two vectors derived from ui by parallel dis-
placement must be coincident in direction; the vector (∗), and
hence φi , must have the same direction as the vector ui . If this
agreement holds for two geodetic lines passing through O in dif-
ferent directions, we get φi = 0. Hence if we know the world-lines
of two point-masses passing through O and moving only under
the influence of the guiding field, then the linear groundform, as
well as the quadratic groundform, is uniquely determined at O.
 APPENDIX II
(Page 345)

Proof of the Theorem that, in Riemann’s space, R is the sole

invariant that contains the derivatives of the gik ’s only to the sec-
ond order, and those of the second order only linearly.
According to hypothesis, the invariant J is built up of the
derivatives of the second order:
∂ 2 gik
gik,rs = ;
∂xr ∂xs
thus X
J= λik,rs gik,rs + λ.
The λ’s denote expressions in the gik ’s and their first derivatives;
they satisfy the conditions of symmetry:

λki,rs = λik,rs , λik,sr = λik,rs .

At the point O at which we are considering the invariant, we

introduce an orthogonal geodetic co-ordinate system, so that, at
that point, we have
∂gik
gik = δik , = 0.
∂xr
The λ’s become absolute constants, if these values are inserted.
The unique character of the co-ordinate system is not affected by:
(1) linear orthogonal transformations;
(2) a transformation of the type
1 i 0 0 0
xi = x0i + αkrs xk xr xs
6
470
 APPENDIX II 471

which contains no quadratic terms; the co-efficients α are sym-

metrical in k, r, and s, but are otherwise arbitrary.
Let us therefore consider in a Euclidean-Cartesian space (in
which arbitrary orthogonal linear transformations are allowable)
the biquadratic form dependent on two vectors x = (xi ), y = (yi ),
namely
G = gik,rs xi xk yr ys
with arbitrary co-efficients gik,rs that are symmetrical in i and k,
as also in r and s; then
λik,rs gik,rs (1)
must be an invariant of this form. Moreover, since as a result
of the transformation (2) above, the derivatives gik,rs transform
themselves, as may easily be calculated, according to the equation
0
gik,rs = gik,rs + 12 (αkrs
i k
+ αirs ),
we must have
i
λik,rs αkrs =0 (2)
for every system of numbers α symmetrical in the three indices
k, r, s.
Let us operate further in the Euclidean-Cartesian space;
(x, y) is to signify the scalar product x1 y1 + x2 y2 + . . . + xn yn . It
will suffice to use for G a form of the type
G = (a, x)2 (b, y)2
in which a and b denote arbitrary vectors. If we now again write
x and y for a and b, then (1) expresses the postulate that
X
Λ = Λx = λik,rs xi xk yr ys (1∗ )
 APPENDIX II 472

is an orthogonal invariant of the two vectors x, y. In (2) it is

sufficient to choose
i
αkrs = x i · yk yr ys
and then this postulate signifies that the form which is derived
from Λx by converting an x into a y, namely,
X
Λy = λik,rs xi yk yr ys (2∗ )

vanishes identically. (It is got from Λx by forming first the sym-

metrical bilinear form Λx,x0 in x, x0 (it is related quadratically
to y), which, if the series of variables x0 be identified with x, re-
solves into Λx , and by then replacing x0 by y.) I now assert that
it follows from (1∗ ) that Λ is of the form

Λ = α(x, x)(y, y) − β(x, y)2 (I)

and from (2∗ ) that

α = β. (II)
This will be the complete result, for then we shall have

J = α(gii,kk − gik,ik ) + λ

or since, in an orthogonal geodetic co-ordinate system, the Rie-

mann scalar of curvature is

R = gik,ik − gii,kk

we shall get
J = −αR + λ. (∗)
 APPENDIX II 473

Proof of I: We may introduce a Cartesian co-ordinate system

such that x coincides with the first co-ordinate axis, and y with
the (1, 2)th co-ordinate plane, thus;

x = (x1 , 0, 0, . . . , 0), y = (y1 , y2 , 0, . . . , 0),

Λ = x1 (ay1 + 2by1 y2 + cy22 ),
2 2

whereby the sense of the second co-ordinate axis may yet be chosen
arbitrarily. Since Λ may not depend on this choice, we must have
b = 0, therefore

Λ = cx21 (y12 + y22 ) + (a − c)(x1 y1 )2 = c(x, x)(y, y) + (a − c)(x, y)2 .

Proof of II: From the Λ = Λx which are given under I, we

derive the forms

Λx,x0 = α(x, x0 )(y, y) − β(x, y)(x0 , y),

Λy = (α − β)(x, y)(y, y).

If Λy is to vanish then α must equal β.

We have tacitly assumed that the metrical groundform of Rie-
mann’s space is definitely positive; in case of a different index of
inertia a slight modification is necessary in the “Proof of I”. In
order that the second derivatives be excluded from the volume
integral J by means of partial integration, it is necessary that
the λik,rs ’s depend only on the gik ’s and not on their derivatives;
we did not, however, require this fact at all in our proof. Con-
cerning the physical meaning entailed by the possibility, expressed
in (∗), of adding to a multiple of R also a universal constant λ,
we refer to § 34. Concerning the theorem here proved, cf. Vermeil,
Nachr. d. Ges. d. Wissensch. zu Göttingen, 1917, pp. 334–344.
 APPENDIX II 474

In the same way it may be proved that gik , Rgik , Rik are the
only tensors of the second order that contain derivatives of the gik ’s
only to the second order, and these, indeed, only linearly.
 BIBLIOGRAPHY
(The number of each note is followed by the number of the page on
which reference is made to it)

INTRODUCTION AND CHAPTER I

Note 1. (7) The detailed development of these ideas follows
very closely the lines of Husserl in his “Ideen zu einer reinen Phäno-
menologie und phänomenologischen Philosophie” (Jahrbuch f. Philos.
u. phänomenol. Forschung, Bd. 1, Halle, 1913).
Note 2. (21) Helmholtz in his dissertation, “Über die Tatsachen,
welche der Geometrie zugrunde liegen” (Nachr. d. K. Gesellschaft
d. Wissenschaften zu Göttingen, math.-physik. Kl., 1868), was the
first to attempt to found geometry on the properties of the group of
motions. This “Helmholtz space-problem” was defined more sharply
and solved by S. Lie (Berichte d. K. Sachs. Ges. d. Wissenschaften zu
Leipzig, math.-phys. Kl., 1890) by means of the theory of transfor-
mation groups, which was created by Lie (cf. Lie-Engel, Theorie der
Transformationsgruppen, Bd. 3, Abt. 5). Hilbert then introduced great
restrictions among the assumptions made by applying the ideas of the
theory of aggregates (Hilbert, Grundlagen der Geometrie, 3 Aufl.,
Leipzig, 1909, Anhang IV).
Note 3. (27) The systematic treatment of affine geometry not lim-
ited to the dimensional number 3 as well as of the whole subject of the
geometrical calculus is contained in the epoch-making work of Grass-
mann, Lineale Ausdehnungslehre (Leipzig, 1844). In forming the con-
ception of a manifold of more than three dimensions, Grassmann as
well as Riemann was influenced by the philosophic ideas of Herbart.
Note 4. (78) The systematic form which we have here given to
the tensor calculus is derived essentially from Ricci and Levi-Civita:
Méthodes de calcul différentiel absolu et leurs applications, Math. Ann.,

475
 BIBLIOGRAPHY 476

Bd. 54 (1901).

CHAPTER II
Note 1. (113) For more detailed information reference may be made
to Die Nicht-Euklidische Geometrie, Bonola and Liebmann, published
by Teubner.
Note 2. (118) F. Klein, Über die sogenannte Nicht-Euklidische
Geometrie, Math. Ann., Bd. 4 (1871), p. 573. Cf. also later papers in
the Math. Ann., Bd. 6 (1873), p. 112, and Bd. 37 (1890), p. 544.
Note 3. (121) Sixth Memoir upon Quantics, Philosophical Trans-
actions, t. 149 (1859).
Note 4. (133) Mathematische Werke (2 Aufl., Leipzig, 1892),
Nr. XIII, p. 272. Als besondere Schrift herausgegeben und kommen-
tiert vom Verf. (2 Aufl., Springer, 1920).
Note 5. (137) Saggio di interpretazione della geometria non eu-
clidea, Giorn. di Matem., t. 6 (1868), p. 204; Opere Matem. (Höpli,
1902), t. 1, p. 374.
Note 6. (137) Grundlagen der Geometrie (3 Aufl., Leipzig, 1909),
Anhang V.
Note 7. (142) Cf. the references in Chap. I.2 Christoffel, Über die
Transformation der homogenen Differentialausdrücke zweiten Grades,
Journ. f. d. reine und angew. Mathemathik, Bd. 70 (1869): Lipschitz,
in the same journal, Bd. 70 (1869), p. 71, and Bd. 72 (1870), p. 1.
Note 8. (150) Christoffel (l.c.7 ). Ricci and Levi-Civita, Méthodes
de calcul différentiel absolu et leurs applications, Math. Ann., Bd. 54
(1901).
Note 9. (151) The development of this geometry was strongly
influenced by the following works which were created in the light of
Einstein’s Theory of Gravitation: Levi-Civita, Nozione di parallelismo
 BIBLIOGRAPHY 477

in una varietà qualunque . . . , Rend. del Circ. Mat. di Palermo, t. 42

(1917), and Hessenberg, Vektorielle Begründung der Differentialge-
ometrie, Math. Ann., Bd. 78 (1917). It assumed a perfectly definite
form in the dissertation by Weyl, Reine Infinitesimalgeometrie, Math.
Zeitschrift, Bd. 2 (1918).
Note 10. (166) The conception of parallel displacement of a vec-
tor was set up for Riemann’s geometry in the dissertation quoted in
Note 9; to derive it, however, Levi-Civita assumed that Riemann’s
space is embedded in a Euclidean space of higher dimensions. A direct
explanation of the conception was given by Weyl in the first edition
of this book with the help of the geodetic co-ordinate system; it was
elevated to the rank of a fundamental axiomatic conception, which is
characteristic of the degree of the affine geometry, in the paper “Reine
Infinitesimalgeometrie,” mentioned in Note 9.
Note 11. (198) Hessenberg (l.c.9 ), p. 190.
Note 12. (213) Cf. the large work of Lie-Engel, Theorie der Trans-
formationsgruppen, Leipzig, 1888–93; concerning this so-called “sec-
ond fundamental theorem” and its converse, vide Bd. 1, p. 156, Bd. 3,
pp. 583, 659, and also Fr. Schur, Math. Ann., Bd. 33 (1888), p. 54.
Note 13. (218) A second view of the problem of space in the light of
the theory of groups forms the basis of the investigations of Helmholtz
and Lie quoted in Chapter I.2

CHAPTER III
Note 1. (221) All further references to the special theory of rela-
tivity will be found in Laue, Die Relativitätstheorie I (3 Aufl., Braun-
schweig, 1919).
Note 2. (238) Helmholtz, Monatsber. d. Berliner Akademie, Marz,
1876, or Ges. Abhandlungen, Bd. 1 (1882), p. 791. Eichenwald, An-
nalen der Physik, Bd. 11 (1903), p. 1.
 BIBLIOGRAPHY 478

Note 3. (250) This is true, only subject to certain limitations; vide

A. Korn, Mechanische Theorie des elektromagnetischen Feldes, Phys.
Zeitschr., Bd. 18, 19 and 20 (1917–19).
Note 4. (253) A. A. Michelson, Sill. Journ., Bd. 22 (1881), p. 120.
A. A. Michelson and E. W. Morley, idem, Bd. 34 (1887), p. 333. E. W.
Morley and D. C. Miller, Philosophical Magazine, vol. viii (1904),
p. 753, and Bd. 9 (1905), p. 680. H. A. Lorentz, Arch. Néerl., Bd. 21
(1887), p. 103, or Ges. Abhandl., Bd. 1, p. 341. Since the enunciation of
the theory of relativity by Einstein, the experiment has been discussed
repeatedly.
Note 5. (255) Cf. Trouton and Noble, Proc. Roy. Soc., vol. lxxii
(1903), p. 132. Lord Rayleigh, Phil. Mag., vol. iv (1902), p. 678. D. B.
Brace, idem (1904), p. 317, vol. x (1905), pp. 71, 591. B. Strasser,
Annal. d. Physik, Bd. 24 (1907), p. 137. Des Coudres, Wiedemanns
Annalen, Bd. 38 (1889), p. 71. Trouton and Rankine, Proc. Roy. Soc.,
vol. viii. (1908), p. 420.
Note 6. (256) Zur Elektrodynamik bewegter Körper, Annal.
d. Physik, Bd. 17 (1905), p. 891.
Note 7. (257) Minkowski, Die Grundgleichungen für die elektro-
magnetischen Vorgänge in bewegten Körpern, Nachr. d. K. Ges. d. Wis-
sensch. zu Göttingen, 1908, p. 53, or Ges. Abhandl., Bd. 2, p. 352.
Note 8. (265) Möbius, Der barycentrische Calcul (Leipzig, 1827;
or Werke, Bd. 1), Kap. 6 u. 7.
Note 9. (276) In taking account of the dispersion it is to be no-
ticed that q 0 is the velocity of propagation for the frequency ν 0 in
water at rest, and not for the frequency ν (which exists inside and
outside the water). Careful experimental confirmations of the result
have been given by Michelson and Morley, Amer. Jour. of Science,
31 (1886), p. 377, Zeeman, Versl. d. K. Akad. v. Wetensch., Amster-
dam, 23 (1914), p. 245; 24 (1915), p. 18. There is a new interference
experiment by Zeeman similar to that performed by Fizeau: Zeeman,
 BIBLIOGRAPHY 479

Versl. Akad. v. Wetensch., Amsterdam, 28 (1919), p. 1451; Zeeman

and Snethlage, idem, p. 1462. Concerning interference experiments
with rotating bodies, vide Laue, Annal. d. Physik, 62 (1920), p. 448.
Note 10. (285) Wilson, Phil. Trans. (A), vol. 204 (1904), p. 121.
Note 11. (292) Röntgen, Sitzungsber. d. Berliner Akademie, 1885,
p. 195; Wied. Annalen, Bd. 35 (1888), p. 264, and Bd. 40 (1890), p. 93.
Eichenwald, Annalen d. Physik, Bd. 11 (1903), p. 421.
Note 12. (292) Minkowski (l.c.7 ).
Note 13. (296) W. Kaufmann, Nachr. d. K. Gesellsch. d. Wis-
sensch. zu Göttingen, 1902, p. 291; Ann. d. Physik, Bd. 19 (1906),
p. 487, and Bd. 20 (1906), p. 639. A. H. Bucherer, Ann. d. Physik,
Bd. 28 (1909), p. 513, and Bd. 29 (1919), p. 1063. S. Ratnowsky,
Determination experimentale de la variation d’inertie des corpuscules
cathodiques en fonction de la vitesse, Dissertation, Geneva, 1911.
E. Hupka, Ann. d. Physik, Bd. 31 (1910), p. 169. G. Neumann, Ann.
d. Physik, Bd. 45 (1914), p. 529, mit Nachtrag von C. Schaefer, ibid.,
Bd. 49, p. 934. Concerning the atomic theory, vide K. Glitscher,
Spektroskopischer Vergleich zwischen den Theorien des starren und
des deformierbaren Elektrons, Ann. d. Physik, Bd. 52 (1917), p. 608.
Note 14. (305) Die Relativitätstheorie I (3 Aufl., 1919), p. 229.
Note 15. (305) Einstein (l.c.6 ). Planck, Bemerkungen zum
Prinzip der Aktion und Reaktion in der allgemeinen Dynamik, Physik.
Zeitschr., Bd. 9 (1908), p. 828; Zur Dynamik bewegter Systeme, Ann.
d. Physik, Bd. 26 (1908), p. 1.
Note 16. (307) Herglotz, Ann. d. Physik, Bd. 36 (1911), p. 453.
Note 17. (307) Ann. d. Physik, Bd. 37, 39, 40 (1912–13).

CHAPTER IV
Note 1. (325) Concerning this paragraph, and indeed the whole
chapter up to § 34, vide A. Einstein, Die Grundlagen der allge-
 BIBLIOGRAPHY 480

meinen Relativitätstheorie (Leipzig, Joh. Ambr. Barth, 1916); Über die

spezielle und die aligemeine Relativitätstheorie (gemeinverständlich;
Sammlung Vieweg, 10 Aufl., 1910). E. Freundlich, Die Grundlagen der
Einsteinschen Gravitationstheorie (4 Aufl., Springer, 1920). M. Schlick,
Raum und Zeit in der gegenwärtigen Physik (3 Aufl., Springer, 1920).
A. S. Eddington, Space, Time, and Gravitation (Cambridge, 1920),
an excellent, popular, and comprehensive exposition of the general
theory of relativity, including the development described in §§ 35, 36.
Eddington, Report on the Relativity Theory of Gravitation (London,
Fleetway Press, 1919). M. Born, Die Relativitätstheorie Einsteins
(Springer, 1920). E. Cassirer, Zur Einsteinschen Relativitätstheorie
(Berlin, Cassirer, 1921). E. Kretschmann, Über den physikalischen
Sinn der Relativitätspostulate, Ann. Phys., Bd. 53 (1917), p. 575.
G. Mie, Die Einsteinsche Gravitationstheorie und das Problem der
Materie, Phys. Zeitschr., Bd. 18 (1917), pp. 551–56, 574–80 and 596–
602. F. Kottler, Über die physikalischen Grundlagen der allgemeinen
Relativitätstheorie, Ann. d. Physik, Bd. 56 (1918), p. 401. Ein-
stein, Prinzipielles zur allgemeinen Relativitätstheorie, Ann. d. Physik,
Bd. 55 (1918), p. 241.
Note 2. (326) Even Newton felt this difficulty; it was stated most
clearly and emphatically by E. Mach. Cf. the detailed references in
A. Voss, Die Prinzipien der rationellen Mechanik, in der Mathematis-
chen Enzyklopädie, Bd. 4, Art. 1, Absatz 13–17 (phoronomische Grund-
begriffe).
Note 3. (335) Mathematische und naturwissenschaftliche Berichte
aus Ungarn VIII (1890).
Note 4. (339) Concerning other attempts (by Abraham, Mie, Nord-
ström) to adapt the theory of gravitation to the results arising from
the special theory of relativity, full references are given in M. Abra-
ham, Neuere Gravitationstheorien, Jahrbuch der Radioaktivität und
Elektronik, Bd. 11 (1915), p. 470.
 BIBLIOGRAPHY 481

Note 5. (347) F. Klein, Über die Differentialgesetze für die Erhal-

tung von Impuls und Energie in der Einsteinschen Gravitationstheorie,
Nachr. d. Ges. d. Wissensch. zu Göttingen, 1918. Cf., in the same
periodical, the general formulations given by E. Noether, Invariante
Variationsprobleme.
Note 6. (355) Following A. Palatini, Deduzione invariantiva delle
equazioni gravitazionali dal principio di Hamilton, Rend. del Circ.
Matem. di Palermo, t. 43 (1919), pp. 203–12.
Note 7. (357) Einstein, Zur allgemeinen Relativitätstheorie, Sitz-
ungsber. d. Preuss. Akad. d. Wissenschaften, 1915, 44, p. 778, and an
appendix on p. 799. Also Einstein, Die Feldgleichungen der Gravita-
tion, idem, 1915, p. 844.
Note 8. (357) H. A. Lorentz, Het beginsel van Hamilton in Ein-
stein’s theorie der zwaartekracht, Versl. d. Akad. v. Wetensch. te Am-
sterdam, XXIII, p. 1073: Over Einstein’s theorie der zwaartekracht
I, II, III, ibid., XXIV, pp. 1389, 1759, XXV, p. 468. Trestling, ibid.,
Nov., 1916; Fokker, ibid., Jan., 1917, p. 1067. Hilbert, Die Grundlagen
der Physik, 1 Mitteilung, Nachr. d. Gesellsch. d. Wissensch. zu Göt-
tingen, 1915, 2 Mitteilung, 1917. Einstein, Hamiltonsches Prinzip und
allgemeine Relativitätstheorie, Sitzungsber. d. Preuss. Akad. d. Wis-
sensch., 1916, 42, p. 1111. Klein, Zu Hilberts erster Note über die
Grundlagen der Physik, Nachr. d. Ges. d. Wissensch. zu Göttingen,
1918, and the paper quoted in Note 5, also Weyl, Zur Gravitationsthe-
orie, Ann. d. Physik, Bd. 54 (1917), p. 117.
Note 9. (359) Following Levi-Civita, Statica Einsteiniana, Rend.
della R. Accad. dei Linceï, 1917, vol. xxvi., ser. 5a, 1◦ sem., p. 458.
Note 10. (364) Cf. also Levi-Civita, La teoria di Einstein e il
principio di Fermat, Nuovo Cimento, ser. 6, vol. xvi. (1918), pp. 105–
14.
Note 11. (367) F. W. Dyson, A. S. Eddington, C. Davidson, A De-
termination of the Deflection of Light by the Sun’s Gravitational Field,
 BIBLIOGRAPHY 482

from Observations made at the Total Eclipse of May 29th, 1919; Phil.
Trans. of the Royal Society of London, Ser. A, vol. 220 (1920), pp. 291–
333. Cf. E. Freundlich, Die Naturwissenschaften, 1920, pp. 667–73.
Note 12. (368) Schwarzschild, Sitzungsber. d. Preuss. Akad.
d. Wissenschaften, 1914, p. 1201. Ch. E. St. John, Astrophys. Journal,
46 (1917), p. 249 (vgl. auch die dort zitierten Arbeiten von Halm und
Adams). Evershed and Royds, Kodaik. Obs. Bull., 39. L. Grebe and
A. Bachem, Verhandl. d. Deutsch. Physik. Ges., 21 (1919), p. 454;
Zeitschrift für Physik, 1 (1920), p. 51. E. Freundlich, Physik. Zeitschr.,
20 (1919), p. 561.
Note 13. (369) Einstein, Sitzungsber. d. Preuss. Akad. d. Wis-
sensch., 1915, 47, p. 831. Schwarzschild, Sitzungsber. d. Preuss. Akad.
d. Wissensch., 1916, 7, p. 189.
Note 14. (369) The following hypothesis claimed most favour.
H. Seeliger, Das Zodiakallicht und die empirischen Glieder in der Bewe-
gung der inneren Planeten, Münch. Akad., Ber. 36 (1906). Cf. E. Fre-
undlich, Astr. Nachr., Bd. 201 (June, 1915), p. 48.
Note 15. (371) Einstein, Sitzungsber. d. Preuss. Akad. d. Wis-
sensch., 1916, p. 688; and the appendix: Über Gravitationswellen,
idem, 1918, p. 154. Also Hilbert (l.c.8 ), 2 Mitteilung.
Note 16. (376) Phys. Zeitschr., Bd. 19 (1918), pp. 33 and 156.
Cf. also de Sitter, Planetary motion and the motion of the moon ac-
cording to Einstein’s theory, Amsterdam Proc., Bd. 19, 1916.
Note 17. (377) Cf. Schwarzschild (l.c.12 ); Hilbert (l.c.8 ), 2 Mitt.;
J. Droste, Versl. K. Akad. v. Wetensch., Bd. 25 (1916), p. 163.
Note 18. (387) Concerning the problem of n bodies, vide J. Droste,
Versl. K. Akad. v. Wetensch., Bd. 25 (1916), p. 460.
Note 19. (388) Cf. A. S. Eddington, Report, §§ 29, 30.
Note 20. (389) L. Flamm, Beiträge zur Einsteinschen Gravitation-
stheorie, Physik. Zeitschr., Bd. 17 (1916), p. 449.
 BIBLIOGRAPHY 483

Note 21. (389) H. Reistner, Ann. Physik, Bd. 50 (1916), pp. 106–
20. Weyl (l.c.8 ). G. Nordström, On the Energy of the Gravitation Field
in Einstein’s Theory, Versl. d. K. Akad. v. Wetensch., Amsterdam,
vol. xx., Nr. 9, 10 (Jan. 26th, 1918). C. Longo, Legge elettrostatica
elementare nella teoria di Einstein, Nuovo Cimento, ser. 6, vol. xv.
(1918). p. 191.
Note 22. (399) Sitzungsber. d. Preuss. Akad. d. Wissensch.,
1916, 18, p. 424. Also H. Bauer, Kugelsymmetrische Lösungssysteme
der Einsteinschen Feldgleichungen der Gravitation für eine ruhende,
gravitierende Flüssigkeit mit linearer Zustandsgleichung, Sitzungsber.
d. Akad. d. Wissensch. in Wien, math.-naturw. Kl., Abt. IIa, Bd. 127
(1918).
Note 23. (399) Weyl (l.c.8 ), §§ 5, 6. And a remark in Ann.
d. Physik, Bd. 59 (1919).
Note 24. (402) Levi-Civita: ds2 einsteiniani in campi newtoniani,
Rend. Accad. dei Linceï, 1917–19.
Note 25. (402) A. De-Zuani, Equilibrio relativo ed equazioni
gravitazionali di Einstein nel caso stazionario, Nuovo Cimento, ser. v,
vol. xviii. (1819), p. 5. A. Palatini, Moti Einsteiniani stazionari, Atti
del R. Instit. Veneto di scienze, lett. ed arti, t. 78 (2) (1919), p. 589.
Note 26. (405) Einstein, Grundlagen [(l.c.1 )] S. 49. The proof here
is according to Klein (l.c.5 ).
Note 27. (405) For a discussion of the physical meaning of these
equations, vide Schrödinger, Phys. Zeitschr., Bd. 19 (1918), p. 4;
H. Bauer, idem, p. 163; Einstein, idem, p. 115, and finally, Einstein,
Der Energiesatz in der allgemeinen Relativitätstheorie, in den Sitz-
ungsber. d. Preuss. Akad. d. Wissensch., 1918, p. 448, which cleared
away the difficulties, and which we have followed in the text. Cf. also
F. Klein, Über die Integralform der Erhaltungssätze und die Theorie
der räumlich geschlossenen Welt, Nachr. d. Ges. d. Wissensch. zu
Göttingen, 1918.
 BIBLIOGRAPHY 484

Note 28. (408) Cf. G. Nordström, On the mass of a material system

according to the Theory of Einstein, Akad. v. Wetensch., Amsterdam,
vol. xx., No. 7 (Dec. 29th, 1917).
Note 29. (411) Hilbert (l.c.8 ), 2 Mitt.
Note 30. (414) Einstein, Sitzungsber. d. Preuss. Akad. d. Wis-
sensch., 1917 6, p. 142.
Note 31. (420) Weyl, Physik. Zeitschr., Bd. 20 (1919), p. 31.
Note 32. (422) Cf. de Sitter’s Mitteilungen im Versl. d. Akad.
v. Wetensch. te Amsterdam, 1917, as also his series of concise articles:
On Einstein’s theory of gravitation and its astronomical consequences
(Monthly Notices of the R. Astronom. Society); also F. Klein (l.c.27 ).
Note 33. (422) The theory contained in the two following articles
were developed by Weyl in the Note “Gravitation und Elektrizität,”
Sitzungsber. d. Preuss. Akad. d. Wissensch., 1918, p. 465. Cf. also
Weyl, Eine neue Erweiterung der Relativitätstheorie, Ann. d. Physik,
Bd. 59 (1919). A similar tendency is displayed (although obscure to the
present author in essential points) in E. Reichenbächer (Grundzüge zu
einer Theorie der Elektrizität und Gravitation, Ann. d. Physik, Bd. 52
[1917], p. 135; also Ann. d. Physik, Bd. 63 [1920], pp. 93–144). Concern-
ing other attempts to derive Electricity and Gravitation from a common
root cf. the articles of Abraham quoted in Note 4; also G. Nordström,
Physik. Zeitschr., 15 (1914), p. 504; E. Wiechert, Die Gravitation als
elektrodynamische Erscheinung, Ann. d. Physik, Bd. 63 (1920), p. 301.
Note 34. (428) This theorem was proved by Liouville: Note IV
in the appendix to G. Monge, Application de l’analyse à la géométrie
(1850), p. 609.
Note 35. (429) This fact, which here appears as a self-evident
result, had been previously noted: E. Cunningham, Proc. of the London
Mathem. Society (2), vol. viii. (1910), pp. 77–98; H. Bateman, idem,
pp. 223–64.
 BIBLIOGRAPHY 485

Note 36. (442) Cf. also W. Pauli, Zur Theorie der Gravitation
und der Elektrizität von H. Weyl, Physik. Zeitschr., Bd. 20 (1919),
pp. 457–67. Einstein arrived at partly similar results by means of a
further modification of his gravitational equations in his essay: Spielen
Gravitationsfelder im Aufbau der materiellen Elementarteilchen eine
wesentliche Rolle? Sitzungsber. d. Preuss. Akad. d. Wissensch., 1919,
pp. 349–56.
Note 37. (449) Concerning such existence theorems at a point of
singularity, vide Picard, Traité d’Analyse, t. 3, p. 21.
Note 38. (452) Ann. d. Physik, Bd. 39 (1913).
Note 39. (454) As described in the book by Sommerfeld, Atombau
and Spektrallinien, Vieweg, 1919 and 1921.
Note 40. (463) This was proved by R. Weitzenböck in a letter to the
present author; his investigation will appear soon in the Sitzungsber.
d. Akad. d. Wissensch. in Wien.
Note 41. (464) W. Pauli, Merkur-Perihelbewegung und Strahlen-
ablenkung in Weyl’s Gravitationstheorie, Verhandl. d. Deutschen
physik. Ges., Bd. 21 (1919), p. 742.
Note 42. (464) Pauli (l.c.36 ).
 INDEX
Aberration, 237, 277 Atom, Bohr’s, 104, 454
Abscissa, 12 Axioms
Acceleration, 171 of affine geometry, 23
Action of metrical geometry
principle of, 315 (Euclidean), 39
Action (infinitesimal), 183
(cf. Hamilton’s Function), 314 Axis of rotation, 18
quantum of, 426
Active past and future, 259 Between, 16
Addition of tensors, 63 Bilinear form, 37
Biot and Savart’s Law, 107
of tensor-densities, 162
Bohr’s model of the atom, 104, 454
of vectors, 23
Bolyai’s geometry, 116, 117
Adjustment and persistence, 461
Æther Calibration, 179
(as a substance), 237 (geodetic), 187
(in a generalised sense), 251, 466 Canonical cylindrical co-ordinates,
Affine 399
geometry Cartesian co-ordinate systems, 42
(infinitesimal), 165 Cathode rays, 296
(linear Euclidean), 22 Causality, principle of, 308
manifold, 151 Cayley’s measure-determination,
relationship of a metrical space, 121
184 Centrifugal forces, 331, 332
transformation, 30 Charge
Allowable systems, 263 (as a substance), 319
Analysis situs, 409, 417 (generally), 403, 441
Angles Christoffel’s 3-indices symbols, 196
measurement of, 18, 42 Clocks, 10, 461
right, 18, 42 Co-gredient transformations, 59, 61
Angular Co-ordinate systems, 12
momentum, 67 Cartesian, 42
velocity, 68 normal, 257, 468
Associative law, 24 Co-ordinates, curvilinear
Asymptotic straight line, 113, 114 (or Gaussian), 126

486
 INDEX 487

(generally), 12 Coulomb’s Law, 107

(hexaspherical), 428 Curl, 88
(in a linear manifold), 23, 40 Current
Co-variant tensors, 80 conduction, 238
(generally), 152 convection, 290
Commutative law, 24 electric, 194
Components, co-variant, and Curvature
contra-variant direction, 186
of the affine relationship, 210 distance, 183
of a displacement, 51 Gaussian, 140
of a tensor, 54 (generally), 175
(generally), 152 of light rays in a gravitational
(in a linear manifold), 152 field, 365
of a vector, 29 scalar of, 198
Conduction, 290 vector, 175
Conductivity, 111 Curve, 125
Configuration, linear point, 28
Congruent, 15, 119 Definite, positive, 39
transference, 207 Density
transformations, 15, 40 (based on the notion of
Conservation, law of substance), 242, 437
of electricity, 403, 405 (of electricity and matter), 248,
of energy and momentum, 437 319, 465
Continuity, equation of (general conception), 293
of electricity, 239 Dielectric, 103
of mass, 280 constant, 105
Continuous relationship, 151, 153 Differentiation of tensors and
Continuum, 124, 125 tensor-densities, 85
Contra-gredient transformation, 50 Dimensions, 26
Contra-variant tensors, 51 (positive and negative, of a
(generally), 152 quadratic form), 45
Contraction-hypothesis of Lorentz Direction-curvature, 186
and Fitzgerald, 253 Displacement current, 240
process of, 70 dielectric, 103
Convection currents, 290 electrical, 104
Coriolis forces, 332 infinitesimal, of a point, 151
 INDEX 488

of space, 55 (possesses inertia), 304

towards red due to presence of (total energy of a system), 451
great masses, 367 Energy-density
of a vector, 163 (in the electric field), 103, 248
Distance (generally), 179 (in the magnetic field), 108
(in Euclidean geometry), 28
Energy-momentum, tensor
Distortion tensor, 88
(for the whole system, including
Distributive law, 24
gravitation), 403
Divergence (div ), 88
(general), 297
(more general), 241, 279
Doppler’s Principle, 276 (in physical events), 437
(in the electromagnetic field),
Earlier and later , 9, 259 249
Einstein’s Law of Gravitation, 351 (in the general theory of
(in its modified form), 435 relativity), 403
Electrical (kinetic and potential), 297
charge (of an incompressible fluid), 306
(as a flux of force), 441 (of the electromagnetic field), 435
(as a substance), 320 (of the gravitational field), 403
current, 194 theorem of (in the special theory
displacement, 241 of relativity), 249
intensity of field, 95, 239 Energy-momentum, tensor (cf.
momentum, 310 Energy-momentum), 249
pressure, 310 Energy-steam or energy-flux, 242
Electricity, positive and negative, Eötvös’ experiment, 335
317
Equality
Electromagnetic field, 94
of time-lengths, 10
and electrostatic units, 239
(origin in the metrics of the of vectors, 174
world), 422 Euclidean
potential, 245 geometry, 15–48
Electromotive force, 112 group of rotations, 205
Electron, 318, 389 manifolds, Chapter I (from the
Electrostatic potential, 107 point of view of infinitesimal
Energy geometry), 176
(acts gravitationally), 346, 353 Euler’s equations, 74
 INDEX 489

Faraday’s Law of Induction, 239, Galilei’s Principle of Relativity and

285 Newton’s Law of Inertia, 221
Fermat’s Principle, 365 Gaussian curvature, 140
Field action of electricity, 322 General principle of relativity, 338,
(electromagnetic), 288 351
energy, 246 Geodetic calibration, 187
forces (contrasted with inertial co-ordinate system, 165
forces), 422 line (general), 169
(general conception), 99 (in Riemann’s space, 189
of gravitation, 344 null-line, 187
(“guiding” or gravitational), 423 systems of reference, 187
intensity of electrical, 95 Geometry
of magnetic, 110 affine, 22
(metrical), 147 Euclidean, 16–48
momentum, 249 infinitesimal, 210
Finitude of space, 417 metrical, 39
Fluid, incompressible, 393 n-dimensional, 27, 35
Force, 55 non-Euclidean
(electric), 99 (Bolyai-Lobatschefsky), 116,
(field force andinertial force), 422 117
(ponderomotive, of electrical Riemann’s, 123
field), 99 spherical, 398
(ponderomotive, of on a surface, 129
electromagnetic field), 310 Gradient, 86
(ponderomotive, of gravitational (generalised), 157
field), 331 Gravitation
(ponderomotive, of magnetic Einstein’s Law of (general form),
field), 108 351
Form Einstein’s Law of (modified
bilinear, 37 form), 435
linear, 32 Newton’s Law of, 342
quadratic, 38 Gravitational
Four-current (4-current), 245 constant, 363
Four-force (4-force), 247 energy, 401
Fresnel’s convection co-efficient, 276 field, 359
Future, active and passive, 262 mass, 335
 INDEX 490

potential, 363 moment of, 70

radius of a great mass, 381 principle of (Galilei’s and
waves, 371–377 Newton’s), 226
Groundform, metrical Inertial force, 451
(in general), 207 index, 44
(of a linear manifold), 40 law of quadratic forms, 44
Groups, 13 mass, 335
infinitesimal, 213 moment, 70
of rotations, 205 Infinitesimal
of translations, 20 displacement, 163
geometry, 210
Hamilton’s group, 213
function, 312 operation of a group, 210
principle rotations, 216
(in the general theory of Integrable, 160
relativity), 438 Intensity of field, 95, 239
(according to Maxwell and quantities, 161
Lorentz), 352
(according to Mie), 312 Joule (heat-equivalent), 241
(in the special theory of
relativity), 322 Klein’s model, 117
Height of displacement, 234
Hexaspherical co-ordinates, 428 Later , 7
Homogeneity Light
of space, 135 electromagnetic theory of, 243
of the world, 229 ray, 273
Homogeneous linear equations, 34 (curved in gravitational
Homologous points, 15 field), 365
Hydrodynamics, 306, 394 Line, straight
Hydrostatic pressure, 306, 394 (in Euclidean geometry), 16
(generally), 25
Impulse (momentum), 64 geodetic, 169
Independent vectors, 27 Line-element
Induction, magnetic, 110 (in Euclidean geometry), 81
law of, 239, 285 (generally), 151
Inertia Linear equation
(as property of energy), 301 point-configuration, 28
 INDEX 491

tensor, 83, 153 stresses, 110

tensor-density, 155, 161 theory
vector manifold, 26 (derived from the world’s
transformation, 30, 31 metrics), 426
Linearly independent, 27 (general case), 239
Lobatschefsky’s geometry, 116, 117 (in the light of the general
Lorentz theory of relativity), 331
-Einstein Theorem of Relativity, (stationary case), 94
245 Measure
-Fitzgerald contraction, 255 electrostatic and
transformation, 246 electromagnetic, 238
relativity of, 423
Magnetic unit of, 59
induction, 110 Measure-index of a distance, 179
intensity of field, 110 Measurement, 262
permeability, 110 Mechanics
Magnetisation, 110 fundamental law of
Magnetism, 108 (derived from field laws),
Magnitudes, 145 434, 438
Manifold (in general theory of
affinely connected, 166 relativity), 331, 336
discrete, 143 of Newton’s, 64, 97
metrical, 151, 179 (in special theory of
Mass relativity), 293
(as energy), 305 of the principle of relativity, 33
(as a flux of force), 457 Metrical groundform, 40, 207
inertial and gravitational, 335 Metrics or metrical structure, 231
(producing a gravitational field), (general), 179, 308, 422
454, 459 Michelson-Morley experiment, 253
Matrix, 57 Mie’s Theory, 307
Matter, 99, 303, 408 Minor space, 233
flux of, 280 Molecular currents, 108
Maxwell’s Moment
application of stationary case to electrical, 310
Riemann’s space, 192 mechanical, 64, 298
density of action, 428 of momentum, 70
 INDEX 492

Momentum, 64, 298 Parallel, 19, 29

density, 249 displacement
flux, 249 co-variant vector, 170
Motion (infinitesimal, of a
(in mathematical sense), 155 contra-variant vector), 167
(under no forces), 74, 341 projection, 232
Multiplication Parallelepiped, 28
of a tensor by a number, 63 Parallelogram, 130
of a tensor-density Parallels, postulate of, 115
by a number, 161 Partial integration (principle of),
by a tensor, 162 165
of a vector by a number, 23 Passive past and future, 259
of tensors, 64 Past, active and passive, 259
Perihelion, motion of Mercury’s,
Newton’s Law of Gravitation, 342
368
Non-degenerate bilinear and
Permeability, magnetic, 110
quadratic forms, 39
Perpendicularity, 179
Non-Euclidean
(in general), 42
geometry, 113
Persistence, 461
plane
Phase, 327
(Beltrami’s model), 137
Plane, 26
(Klein’s model), 117
(Beltrami’s model), 137
(metrical groundform of),
(Klein’s model), 121
139
(in Euclidean space), 18
Non-homogeneous linear equations,
34 (metrical groundform), 139
Normal calibration of Riemann’s (non-Euclidean), 118
space, 183 Planetary motion, 382
system of co-ordinates, 256, 468 Polarisation, 104
Now , 221 Ponderomotive force
Null-lines, geodetic, 187 of the electric, magnetic and
Number, 11, 57 electromagnetic field, 98,
108, 288
Ohm’s Law, 111 of the gravitational field, 331,
One-sided surfaces, 410 332
Order of tensors, 52 Positive definite, 39
Orthogonal transformations, 50 Potential
 INDEX 493

electromagnetic, 244 metrical, 210

electrostatic, 243 of a manifold as a whole
energy-momentum tensor of, 297, (conditions of), 169
298 of the world, 409
of the gravitational field, 342 Relativity
retarded, 243, 244, 373 of magnitude, 424
vector-, 108, 243 of motion, 226, 423
Poynting’s vector, 242 principle of
Pressure, on all sides (Einstein’s special), 250
electrical, 310 (general), 338, 352
hydrostatic, 306, 394 Galilei’s, 221
Problem of one body, 379 theorem of (Lorentz-Einstein),
Product 245
scalar, 39 Resolution of tensors into space and
of a tensor and a number, 63 time of vectors, 234, 267
vectorial, 66 Rest, 222
Projection, 232 Retarded potential, 243, 244, 373
Propagation Riemann’s
of electromagnetic disturbances, curvature, 196
243 geometry, 124
of gravitational disturbances, 375 space, 196
of light, 243 Right angle, 42, 179
Proper-time, 264, 267, 294 Rotation
Pythagoras’ Theorem, 135, 339 (or curl), 88
(general), 230
Quadratic forms, 44
(in geometrical sense), 18
Quantities
(in kinematical sense), 68
intensity, 161
magnitude, 161 relativity of, 230
Quantum Theory, 426, 454 Rotations, group of, 205, 216

Radial symmetry, 377 Scalar

Reality, 317 field, 85
Red, displacement towards the, 367 product, 39
Relationship Scalar-Density, 161
affine, 165 Similar representation or
continuous, 151, 153 transformation, 207
 INDEX 494

Simultaneity, 258, 271 tensors, 63

Skew-symmetrical, 56, 80 vectors, 23
Space Surface, 126, 410
-element, 81 Symmetry, 38
Euclidean, 1–5 Systems of reference, 262
(as form of phenomena), 1, 141 geodetic, 188
-like vector, 265
metrical, 48, 54
n-dimensional, 35 Tensor
(as projection of the world), 234, density, 161
267 field, 154
Special principle of relativity, 251 (general), 84
Sphere, charged, 389 (general), 72, 152
Spherical (in linear space), 48
geometry, 122, 398 Time, 367
transformations, 428 -like vectors, 267
Static
Top, spinning, 74
density, 294
Torque of a force, 67
gravitational field, 359–371
Trace of a matrix, 71, 216
length, 261
volume, 273 Tractrix, 137
Stationary Transference, congruent, 207
field, 168, 359 Transformation or representation
orbits in the atom, 454 affine, 30
vectors, 168 congruent, 15, 40
Stokes’ Theorem, 160 linear-vector, 30, 31
Stresses similar, 207
elastic, 84, 88 Translation of a point
Maxwell’s, 110 (in the geometrical sense), 13
Substance, 320, 408 (in the kinematical sense), 170
Substance-action of electricity and Turning-moment of a force, 67
gravitation, 320 Twists, 18
(= mass), 450 Two-sided surfaces, 410
Subtraction of vectors, 24
Sum of
tensor-densities, 161 Unit vectors, 154
 INDEX 495

Vector, 23, 34 of rotation, 68

curvature, 186 Volume-element, 314
-density, 161
-manifold, linear, 26 Weight of tensors and
tensor-densities, 188
potential, 109, 242
Wilson’s experiment, 287
product, 66 World (= space-time), 281
transference, 173 -canal, 402
transformation, linear, 29, 31 -law, 317, 409, 413
Velocity, 155 -line, 221
of propagation of gravitation, 374 -point, 221
of light, 243 -vectors, 229

PRINTED IN GREAT BRITAIN AT THE UNIVERSITY PRESS, ABERDEEN.

End of the Project Gutenberg EBook of Space--Time--Matter, by Hermann Weyl

*** END OF THIS PROJECT GUTENBERG EBOOK SPACE--TIME--MATTER ***

***** This file should be named 43006-t.tex or 43006-t.zip *****

This and all associated files of various formats will be found in:
http://www.gutenberg.org/4/3/0/0/43006/

Produced by Andrew D. Hwang, using scanned images and OCR

text generously provided by the University of Toronto
Gerstein Library through the Internet Archive.

Updated editions will replace the previous one--the old editions

will be renamed.

Creating the works from public domain print editions means that no
one owns a United States copyright in these works, so the Foundation
(and you!) can copy and distribute it in the United States without
permission and without paying copyright royalties. Special rules,
set forth in the General Terms of Use part of this license, apply to
copying and distributing Project Gutenberg-tm electronic works to
protect the PROJECT GUTENBERG-tm concept and trademark. Project
Gutenberg is a registered trademark, and may not be used if you
charge for the eBooks, unless you receive specific permission. If you
do not charge anything for copies of this eBook, complying with the
rules is very easy. You may use this eBook for nearly any purpose
such as creation of derivative works, reports, performances and
research. They may be modified and printed and given away--you may do
practically ANYTHING with public domain eBooks. Redistribution is
subject to the trademark license, especially commercial
redistribution.

*** START: FULL LICENSE ***

THE FULL PROJECT GUTENBERG LICENSE

PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK

To protect the Project Gutenberg-tm mission of promoting the free

distribution of electronic works, by using or distributing this work
(or any other work associated in any way with the phrase "Project

I
 LICENSE II

Gutenberg"), you agree to comply with all the terms of the Full Project
Gutenberg-tm License available with this file or online at
www.gutenberg.org/license.

Section 1. General Terms of Use and Redistributing Project Gutenberg-tm

electronic works

1.A. By reading or using any part of this Project Gutenberg-tm

electronic work, you indicate that you have read, understand, agree to
and accept all the terms of this license and intellectual property
(trademark/copyright) agreement. If you do not agree to abide by all
the terms of this agreement, you must cease using and return or destroy
all copies of Project Gutenberg-tm electronic works in your possession.
If you paid a fee for obtaining a copy of or access to a Project
Gutenberg-tm electronic work and you do not agree to be bound by the
terms of this agreement, you may obtain a refund from the person or
entity to whom you paid the fee as set forth in paragraph 1.E.8.

1.B. "Project Gutenberg" is a registered trademark. It may only be

used on or associated in any way with an electronic work by people who
agree to be bound by the terms of this agreement. There are a few
things that you can do with most Project Gutenberg-tm electronic works
even without complying with the full terms of this agreement. See
paragraph 1.C below. There are a lot of things you can do with Project
Gutenberg-tm electronic works if you follow the terms of this agreement
and help preserve free future access to Project Gutenberg-tm electronic
works. See paragraph 1.E below.

1.C. The Project Gutenberg Literary Archive Foundation ("the Foundation"

or PGLAF), owns a compilation copyright in the collection of Project
Gutenberg-tm electronic works. Nearly all the individual works in the
collection are in the public domain in the United States. If an
individual work is in the public domain in the United States and you are
located in the United States, we do not claim a right to prevent you from
copying, distributing, performing, displaying or creating derivative
works based on the work as long as all references to Project Gutenberg
are removed. Of course, we hope that you will support the Project
Gutenberg-tm mission of promoting free access to electronic works by
freely sharing Project Gutenberg-tm works in compliance with the terms of
this agreement for keeping the Project Gutenberg-tm name associated with
the work. You can easily comply with the terms of this agreement by
keeping this work in the same format with its attached full Project
 LICENSE III

Gutenberg-tm License when you share it without charge with others.

1.D. The copyright laws of the place where you are located also govern
what you can do with this work. Copyright laws in most countries are in
a constant state of change. If you are outside the United States, check
the laws of your country in addition to the terms of this agreement
before downloading, copying, displaying, performing, distributing or
creating derivative works based on this work or any other Project
Gutenberg-tm work. The Foundation makes no representations concerning
the copyright status of any work in any country outside the United
States.

1.E. Unless you have removed all references to Project Gutenberg:

1.E.1. The following sentence, with active links to, or other immediate
access to, the full Project Gutenberg-tm License must appear prominently
whenever any copy of a Project Gutenberg-tm work (any work on which the
phrase "Project Gutenberg" appears, or with which the phrase "Project
Gutenberg" is associated) is accessed, displayed, performed, viewed,
copied or distributed:

This eBook is for the use of anyone anywhere at no cost and with
almost no restrictions whatsoever. You may copy it, give it away or
re-use it under the terms of the Project Gutenberg License included
with this eBook or online at www.gutenberg.org

1.E.2. If an individual Project Gutenberg-tm electronic work is derived

from the public domain (does not contain a notice indicating that it is
posted with permission of the copyright holder), the work can be copied
and distributed to anyone in the United States without paying any fees
or charges. If you are redistributing or providing access to a work
with the phrase "Project Gutenberg" associated with or appearing on the
work, you must comply either with the requirements of paragraphs 1.E.1
through 1.E.7 or obtain permission for the use of the work and the
Project Gutenberg-tm trademark as set forth in paragraphs 1.E.8 or
1.E.9.

1.E.3. If an individual Project Gutenberg-tm electronic work is posted

with the permission of the copyright holder, your use and distribution
must comply with both paragraphs 1.E.1 through 1.E.7 and any additional
terms imposed by the copyright holder. Additional terms will be linked
to the Project Gutenberg-tm License for all works posted with the
permission of the copyright holder found at the beginning of this work.
 LICENSE IV

1.E.4. Do not unlink or detach or remove the full Project Gutenberg-tm

License terms from this work, or any files containing a part of this
work or any other work associated with Project Gutenberg-tm.

1.E.5. Do not copy, display, perform, distribute or redistribute this

electronic work, or any part of this electronic work, without
prominently displaying the sentence set forth in paragraph 1.E.1 with
active links or immediate access to the full terms of the Project
Gutenberg-tm License.

1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form, including any
word processing or hypertext form. However, if you provide access to or
distribute copies of a Project Gutenberg-tm work in a format other than
"Plain Vanilla ASCII" or other format used in the official version
posted on the official Project Gutenberg-tm web site (www.gutenberg.org),
you must, at no additional cost, fee or expense to the user, provide a
copy, a means of exporting a copy, or a means of obtaining a copy upon
request, of the work in its original "Plain Vanilla ASCII" or other
form. Any alternate format must include the full Project Gutenberg-tm
License as specified in paragraph 1.E.1.

1.E.7. Do not charge a fee for access to, viewing, displaying,

performing, copying or distributing any Project Gutenberg-tm works
unless you comply with paragraph 1.E.8 or 1.E.9.

1.E.8. You may charge a reasonable fee for copies of or providing

access to or distributing Project Gutenberg-tm electronic works provided
that

- You pay a royalty fee of 20% of the gross profits you derive from
the use of Project Gutenberg-tm works calculated using the method
you already use to calculate your applicable taxes. The fee is
owed to the owner of the Project Gutenberg-tm trademark, but he
has agreed to donate royalties under this paragraph to the
Project Gutenberg Literary Archive Foundation. Royalty payments
must be paid within 60 days following each date on which you
prepare (or are legally required to prepare) your periodic tax
returns. Royalty payments should be clearly marked as such and
sent to the Project Gutenberg Literary Archive Foundation at the
address specified in Section 4, "Information about donations to
the Project Gutenberg Literary Archive Foundation."
 LICENSE V

- You provide a full refund of any money paid by a user who notifies
you in writing (or by e-mail) within 30 days of receipt that s/he
does not agree to the terms of the full Project Gutenberg-tm
License. You must require such a user to return or
destroy all copies of the works possessed in a physical medium
and discontinue all use of and all access to other copies of
Project Gutenberg-tm works.

- You provide, in accordance with paragraph 1.F.3, a full refund of any

money paid for a work or a replacement copy, if a defect in the
electronic work is discovered and reported to you within 90 days
of receipt of the work.

- You comply with all other terms of this agreement for free
distribution of Project Gutenberg-tm works.

1.E.9. If you wish to charge a fee or distribute a Project Gutenberg-tm

electronic work or group of works on different terms than are set
forth in this agreement, you must obtain permission in writing from
both the Project Gutenberg Literary Archive Foundation and Michael
Hart, the owner of the Project Gutenberg-tm trademark. Contact the
Foundation as set forth in Section 3 below.

1.F.

1.F.1. Project Gutenberg volunteers and employees expend considerable

effort to identify, do copyright research on, transcribe and proofread
public domain works in creating the Project Gutenberg-tm
collection. Despite these efforts, Project Gutenberg-tm electronic
works, and the medium on which they may be stored, may contain
"Defects," such as, but not limited to, incomplete, inaccurate or
corrupt data, transcription errors, a copyright or other intellectual
property infringement, a defective or damaged disk or other medium, a
computer virus, or computer codes that damage or cannot be read by
your equipment.

1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except for the "Right

of Replacement or Refund" described in paragraph 1.F.3, the Project
Gutenberg Literary Archive Foundation, the owner of the Project
Gutenberg-tm trademark, and any other party distributing a Project
Gutenberg-tm electronic work under this agreement, disclaim all
liability to you for damages, costs and expenses, including legal
 LICENSE VI

fees. YOU AGREE THAT YOU HAVE NO REMEDIES FOR NEGLIGENCE, STRICT
LIABILITY, BREACH OF WARRANTY OR BREACH OF CONTRACT EXCEPT THOSE
PROVIDED IN PARAGRAPH 1.F.3. YOU AGREE THAT THE FOUNDATION, THE
TRADEMARK OWNER, AND ANY DISTRIBUTOR UNDER THIS AGREEMENT WILL NOT BE
LIABLE TO YOU FOR ACTUAL, DIRECT, INDIRECT, CONSEQUENTIAL, PUNITIVE OR
INCIDENTAL DAMAGES EVEN IF YOU GIVE NOTICE OF THE POSSIBILITY OF SUCH
DAMAGE.

1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you discover a

defect in this electronic work within 90 days of receiving it, you can
receive a refund of the money (if any) you paid for it by sending a
written explanation to the person you received the work from. If you
received the work on a physical medium, you must return the medium with
your written explanation. The person or entity that provided you with
the defective work may elect to provide a replacement copy in lieu of a
refund. If you received the work electronically, the person or entity
providing it to you may choose to give you a second opportunity to
receive the work electronically in lieu of a refund. If the second copy
is also defective, you may demand a refund in writing without further
opportunities to fix the problem.

1.F.4. Except for the limited right of replacement or refund set forth
in paragraph 1.F.3, this work is provided to you ’AS-IS’, WITH NO OTHER
WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.

1.F.5. Some states do not allow disclaimers of certain implied

warranties or the exclusion or limitation of certain types of damages.
If any disclaimer or limitation set forth in this agreement violates the
law of the state applicable to this agreement, the agreement shall be
interpreted to make the maximum disclaimer or limitation permitted by
the applicable state law. The invalidity or unenforceability of any
provision of this agreement shall not void the remaining provisions.

1.F.6. INDEMNITY - You agree to indemnify and hold the Foundation, the
trademark owner, any agent or employee of the Foundation, anyone
providing copies of Project Gutenberg-tm electronic works in accordance
with this agreement, and any volunteers associated with the production,
promotion and distribution of Project Gutenberg-tm electronic works,
harmless from all liability, costs and expenses, including legal fees,
that arise directly or indirectly from any of the following which you do
or cause to occur: (a) distribution of this or any Project Gutenberg-tm
work, (b) alteration, modification, or additions or deletions to any
 LICENSE VII

Project Gutenberg-tm work, and (c) any Defect you cause.

Section 2. Information about the Mission of Project Gutenberg-tm

Project Gutenberg-tm is synonymous with the free distribution of

electronic works in formats readable by the widest variety of computers
including obsolete, old, middle-aged and new computers. It exists
because of the efforts of hundreds of volunteers and donations from
people in all walks of life.

Volunteers and financial support to provide volunteers with the

assistance they need are critical to reaching Project Gutenberg-tm’s
goals and ensuring that the Project Gutenberg-tm collection will
remain freely available for generations to come. In 2001, the Project
Gutenberg Literary Archive Foundation was created to provide a secure
and permanent future for Project Gutenberg-tm and future generations.
To learn more about the Project Gutenberg Literary Archive Foundation
and how your efforts and donations can help, see Sections 3 and 4
and the Foundation information page at www.gutenberg.org

Section 3. Information about the Project Gutenberg Literary Archive

Foundation

The Project Gutenberg Literary Archive Foundation is a non profit

501(c)(3) educational corporation organized under the laws of the
state of Mississippi and granted tax exempt status by the Internal
Revenue Service. The Foundation’s EIN or federal tax identification
number is 64-6221541. Contributions to the Project Gutenberg
Literary Archive Foundation are tax deductible to the full extent
permitted by U.S. federal laws and your state’s laws.

The Foundation’s principal office is located at 4557 Melan Dr. S.

Fairbanks, AK, 99712., but its volunteers and employees are scattered
throughout numerous locations. Its business office is located at 809
North 1500 West, Salt Lake City, UT 84116, (801) 596-1887. Email
contact links and up to date contact information can be found at the
Foundation’s web site and official page at www.gutenberg.org/contact

For additional contact information:

Dr. Gregory B. Newby
Chief Executive and Director
 LICENSE VIII

[email protected]

Section 4. Information about Donations to the Project Gutenberg

Literary Archive Foundation

Project Gutenberg-tm depends upon and cannot survive without wide

spread public support and donations to carry out its mission of
increasing the number of public domain and licensed works that can be
freely distributed in machine readable form accessible by the widest
array of equipment including outdated equipment. Many small donations
($1 to $5,000) are particularly important to maintaining tax exempt
status with the IRS.

The Foundation is committed to complying with the laws regulating

charities and charitable donations in all 50 states of the United
States. Compliance requirements are not uniform and it takes a
considerable effort, much paperwork and many fees to meet and keep up
with these requirements. We do not solicit donations in locations
where we have not received written confirmation of compliance. To
SEND DONATIONS or determine the status of compliance for any
particular state visit www.gutenberg.org/donate

While we cannot and do not solicit contributions from states where we

have not met the solicitation requirements, we know of no prohibition
against accepting unsolicited donations from donors in such states who
approach us with offers to donate.

International donations are gratefully accepted, but we cannot make

any statements concerning tax treatment of donations received from
outside the United States. U.S. laws alone swamp our small staff.

Please check the Project Gutenberg Web pages for current donation
methods and addresses. Donations are accepted in a number of other
ways including checks, online payments and credit card donations.
To donate, please visit: www.gutenberg.org/donate

Section 5. General Information About Project Gutenberg-tm electronic

works.

Professor Michael S. Hart was the originator of the Project Gutenberg-tm

concept of a library of electronic works that could be freely shared
with anyone. For forty years, he produced and distributed Project
 LICENSE IX

Gutenberg-tm eBooks with only a loose network of volunteer support.

Project Gutenberg-tm eBooks are often created from several printed

editions, all of which are confirmed as Public Domain in the U.S.
unless a copyright notice is included. Thus, we do not necessarily
keep eBooks in compliance with any particular paper edition.

Most people start at our Web site which has the main PG search facility:

www.gutenberg.org

This Web site includes information about Project Gutenberg-tm,

including how to make donations to the Project Gutenberg Literary
Archive Foundation, how to help produce our new eBooks, and how to
subscribe to our email newsletter to hear about new eBooks.