SPACETIME WITHOUT REFERENCE FRAMES

SPACETIME WITHOUT
REFERENCE FRAMES
by

 MATOLCSI
TAMAS
Eotvos Lorand University
Budapest

Akademiai Kiado, Budapest, 1993

This book is the revised and enlarged version of

A Con ept of Mathemati al Physi s | Models for Spa e-Time by T. Matol si
published by Akademiai Kiado in 1984.

ISBN 963 05 6433 5

T. Matol si, 1993

All rights reserved. No part of this book may be reprodu ed by any means, or
transmitted, or translated into ma hine language without the written
permission of the publisher.
Akademiai Kiado
H{1519 P.O.Box 245.
Printed in Hungary by Akademiai Nyomda, Budapest

CONTENTS

PREFACE

::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

11

PART ONE
SPACETIME MODELS
INTRODUCTION

:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

21

1. The prin iples of ovarian e and of relativity : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 21

2. Units of measurements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22
3. What is spa etime? : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24
I. NON-RELATIVISTIC SPACETIME MODEL

:::::::::::::::::::::::::::::::

31

1. Fundamentals : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.1. Denition of the spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.2. Stru ture of world ve tors and ove tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.3. The arithmeti spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.4. Classi ation of physi al quantities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.5. Comparison of spa etime models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.6. The split spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.7. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2. World lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.1. History of a masspoint: world line : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.2. A hara terization of world lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.3. Classi ation of world lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.4. Newtonian equation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.5. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3. Observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.1. The notion of an observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.2. Splitting of spa etime due to an observer : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.3. Classi ation of observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.4. Exer ise : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4. Rigid observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.1. Inertial observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.2. Chara terization of rigid observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

31
31
34
38
40
42
44
46
47
47
49
49
50
52
52
52
54
55
57
57
57
61

4.3. About the spa es of rigid observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

4.4. Observers with origin : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.5. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5. Some spe ial observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.1. Why the inertial observers are better than the others : : : : : : : : : : : : : : : :
5.2. Uniformly a elerated observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.3. Uniformly rotating observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.4. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6. Kinemati s : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.1. The history of a masspoint is observed as a motion : : : : : : : : : : : : : : : : : :
6.2. Relative velo ities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.3. Motions relative to a rigid observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.4. Some motions relative to an inertial observer : : : : : : : : : : : : : : : : : : : : : : : :
6.5. Some motions relative to a uniformly a elerated observer : : : : : : : : : : :
6.6. Some motions relative to a uniformly rotating observer : : : : : : : : : : : : :
6.7. Exer ise : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7. Some kinds of observation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.1. Ve tors observed by inertial observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.2. Measuring rods : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8. Ve tor splittings : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.1. What is a splitting? : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.2. Splitting of ve tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.3. Splitting of ove tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.4. Ve tors and ove tors are split in a dierent way : : : : : : : : : : : : : : : : : : : :
8.5. Splitting of ve tor elds and ove tor elds a ording to inertial
observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.6. Splitting of ve tor elds and ove tor elds a ording to rigid
observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.7. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9. Tensor splittings : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.1. Splitting of tensors, otensors, et . : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.2. Splitting of antisymmetri tensors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.3. Splitting of antisymmetri otensors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.4. Splitting of otensor elds : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.5. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10. Referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.1. The notion of a referen e frame : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.2. Galilean referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.3. Subs ripts and supers ripts : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.4. Referen e systems asso iated with global rigid observers : : : : : : : : : : :
10.5. Equivalent referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.6. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11. Spa etime groups : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11.1. The three-dimensional orthogonal group : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11.2. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11.3. The Galilean group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11.4. The split Galilean group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

65
67
69
70
70
70
72
74
77
77
78
81
82
84
85
86
86
86
88
89
89
89
91
94
94
96
97
98
98
100
101
101
104
106
106
110
112
114
115
117
119
119
125
125
131

11.5.
11.6.
11.7.
11.8.
11.9.

Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
The Noether group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
The ve torial Noether group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
The split Noether group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

II. SPECIAL RELATIVISTIC SPACETIME MODELS

::::::::::::::::::::::::

1. Fundamentals : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.1. Heuristi onsiderations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.2. Denition of the spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.3. Stru ture of world ve tors and ove tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.4. The arithmeti spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.5. Classi ation of physi al quantities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.6. Comparison of spa etime models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.7. The u-split spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.8. About the two types of spa etime models : : : : : : : : : : : : : : : : : : : : : : : : : : : :
1.9. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2. World lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.1. History of a masspoint: world line : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.2. Proper time of world lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.3. World line fun tions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.4. Classi ation of world lines : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.5. World horizons : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.6. Newtonian equation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
2.7. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3. Inertial observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.1. Observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.2. The time of an inertial observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.3. The spa e of an inertial observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.4. Splitting of spa etime : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
3.5. Exer ise : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4. Kinemati s : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.1. Motions relative to an inertial observer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.2. Relative velo ities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.3. Addition of relative velo ities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.4. History regained from motion : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.5. Relative a elerations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.6. Some parti ular motions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
4.7. Light speed : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5. Some observations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.1. Physi ally equal ve tors in dierent spa es : : : : : : : : : : : : : : : : : : : : : : : : : :
5.2. Observations on erning spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.3. The Lorentz ontra tion : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.4. The tunnel paradox : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.5. No measuring rods : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.6. The time dilation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

133
134
138
140
143
149
147
147
149
151
158
160
161
163
164
164
166
166
169
172
174
176
179
181
182
182
183
186
188
189
190
190
191
193
195
196
196
198
200
200
201
204
207
208
208

5.7. The twin paradox : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

5.8. Experiments on erning time : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
5.9. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6. Some spe ial non-inertial observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.1. General observers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.2. Simultaneities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.3. Distan es in observer spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.4. A method of nding the observer spa e : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.5. Uniformly a elerated observer I : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.6. Uniformly a elerated observer II : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.7. Uniformly rotating observer I : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.8. Uniformly rotating observer II : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
6.9. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7. Ve tor splittings : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.1. Splitting of ve tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.2. Splitting of ove tors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.3. Splitting of ve tor elds : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
7.4. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8. Tensor splittings : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.1. Splitting of tensors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.2. Splitting of antisymmetri tensors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.3. Splitting of tensor elds : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
8.4. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9. Referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.1. The notion of a referen e frame : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.2. Lorentz referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.3. Equivalent referen e frames : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.4. Curve lengths al ulated in oordinates : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
9.5. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10. Spa etime groups : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.1. The Lorentz group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.2. The u-split Lorentz group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.3. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.4. The Poin are group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.5. The ve torial Poin are group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.6. The u-split Poin are group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
10.7. Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
11. Relation between the spe ial relativisti spa etime model and the nonrelativisti spa etime model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

210
211
213
213
213
214
217
219
220
224
228
232
238
240
240
244
245
246
247
247
248
249
250
251
251
253
255
256
257
259
259
264
266
266
269
270
272
272

III. FUNDAMENTAL NOTIONS OF GENERAL RELATIVISTIC SPACETIME MODELS : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 275

PART TWO
MATHEMATICAL TOOLS
IV. TENSORIAL OPERATIONS
0.
1.
2.
3.
4.
5.

::::::::::::::::::::::::::::::::::::::::::::::

283

Identi ations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Duality : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Coordinatization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Tensor produ ts : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Tensor quotients : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Tensorial operations and orientation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

283
284
288
289
302
305

V. PSEUDO-EUCLIDEAN VECTOR SPACES

1.
2.
3.
4.

::::::::::::::::::::::::::::::::

308

Pseudo-Eu lidean ve tor spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

Tensors of pseudo-Eu lidean ve tor spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Eu lidean ve tor spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Minkowskian ve tor spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

308
312
318
331

VI. AFFINE SPACES

1.
2.
3.
4.
5.
6.
7.

:::::::::::::::::::::::::::::::::::::::::::::::::::::::::

343

Fundamentals : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Ane maps : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Dierentiation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Submanifolds in ane spa es : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Coordinatization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Dierential equations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Integration on urves : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

343
347
350
355
363
371
374

VII. LIE GROUPS

1.
2.
3.
4.
5.
6.

::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

377

Groups of linear bije tions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

Groups of ane bije tions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Lie groups : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
The Lie algebra of a Lie group : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Pseudo-orthogonal groups : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
Exer ises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

377
379
381
386
388
389

SUBJECT INDEX

::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

LIST OF SYMBOLS

::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

COMMENTS AND BIBLIOGRAPHY

:::::::::::::::::::::::::::::::::::::::::

391
395
398

The star () indi ates paragraphs or se tions whi h an be skipped at rst reading
without loss in understanding the essential meaning of the book.

PREFACE

1. Mathemati s rea hed a risis at the end of the last entury when a number
of paradoxes ame to light. Mathemati ians surmounted the di ulties by
revealing the origin of the troubles: the obs ure notions, the inexa t denitions;
then the modern mathemati al exa tness was reated and all the earlier notions
and results were reappraised. After this great work nowadays mathemati s is
rmly based upon its exa tness.
Theoreti al physi s | in quantum eld theory | rea hed its own risis in
the last de ades. The reason of the troubles is the same. Earlier physi s has
treated ommon, visible and palpable phenomena, everything has been obvious.
On the other hand, modern physi s deals with phenomena of the mi roworld
where nothing is ommon, nothing is visible, nothing is obvious. Most of the
notions applied to des ribe phenomena of the mi roworld are the old ones and
in the new framework they are ne essarily onfused.
It is quite evident, that we have to follow a way similar to that followed by
mathemati ians to reate a rm theory based on mathemati al exa tness; having
mathemati al exa tness as a guiding prin iple, we must reappraise physi s, its
most ommon, most visible and most palpable notions as well. Doing so we an
hope we shall be able to over ome the di ulties.
2. A ording to a new on ept, mathemati al physi s should be a mathemati al theory of the whole physi s, a mathemati al theory based on mathemati al
exa tness, a mathemati al theory in whi h only mathemati ally dened notions
appear and in whi h all the notions used in physi s are dened in a mathemati ally exa t way.
What does the term \mathemati ally exa t" over? Sin e physi s is a natural s ien e, its riteria of truth is experiment. As a straightforward onsequen e, theoreti al physi s has be ome a mixture of mathemati al notions and
mathemati ally not formulated \ta it agreements". These agreements are organi parts of theoreti al physi s; they originate from the period when physi s
treated palpable phenomena like those in lassi al me hani s and ele trodynami s. Today's physi s deals with phenomena on very small or very large s ales.

Unfortunately, sin e the edu ation of physi ists starts with the lassi al theories
whi h are left more or less as they were at the beginning of this entury, the
a quired style of thinking is the mixture mentioned above and this is applied
further on to des ribe phenomena in regions where nothing is obvious, resulting
in onfusion and un lear thinking.
Mathemati al exa tness means that we formulate all the \ta it agreements"
in the language of mathemati s starting at the very beginning, with the most
natural, most palpable notions. Following this method, we have a good han e
of making an important step forward in modern theoreti al physi s.
At rst sight this seems to a physi ist like reating unne essary onfusion
around obvious things. Su h a feeling is quite natural; if one has never driven a
ar before, the rst few o asions are terrible. But after a while it be omes easy
and omfortable and mu h faster than walking on foot; it is worth spending a
part of our valuable time on learning to drive.

3. To build up su h a mathemati al physi s, we must start with the simplest,

most ommon notions of physi s; we annot start with quantum eld theory but
we hope that we an end up with it.
The fundamental notion of mathemati al physi s is that of models. Our aim
is to onstru t mathemati al models for physi al phenomena. The modelling
pro edure has two sides of equal importan e: the mathemati al model and the
modelled part of physi al reality. We shall sharply distinguish between these
two sides. Physi al reality is independent of our mind, it is su h as it is.
A mathemati al model depends on our mind, it is su h as it is onstru ted
by us. The onfusion of physi al reality and its models have led to heavy
misunderstandings in onne tion with quantum me hani s.
A mathemati al model is onstru ted as a result of experiments and theoreti al onsiderations; on lusions based on the model are ontrolled by experiments.
The mathemati al model is a mathemati al stru ture whi h is expe ted to re e t
some properties of the modelled part of reality. It lies outside the model to answer what and how it re e ts and to de ide in what sense it is good or bad. To
answer these questions, we have to go beyond the exa t framework of the model.
4. The whole world is an undivisible unity. However, to treat physi s, we are
for ed by our limited biologi al, mental et . apa ity to divide it into parts in
theory.
Today's physi s suggests the arrangement of physi al phenomena in three
groups; the orresponding three entities an be alled Spa etime, Matter and
Field.
The phenomena of these three entities intera t and determine ea h other
mutually. At present it is impossible to give a good des ription of the omplex
situation in whi h everything intera ts with everything, whi h an be illustrated
as follows:

5. Fortunately, a great number of phenomena allows us to negle t some

aspe ts of the intera tions. More pre isely, we an onstru t a good theory if
we an repla e intera tion by a tion, i.e. we an onsider as if the phenomena
of two of the entities above were given, xed, \sti" and only the phenomena of
the third one were \ exible", unknown and looked for. The sti phenomena of
the two entities are supposed to a t upon and even determine the phenomena of
the third one whi h do not rea t. We obtain dierent theories a ording to the
entities onsidered to be xed.
Me hani s ( lassi al and quantal), if spa etime and eld phenomena are given
to determine phenomena of matter, an be depi ted as:

In some sense ontinuum physi s and thermodynami s, too, are su h theories.

Field theory ( lassi al, i.e. ele trodynami s), if spa etime and matter phenomena are given to determine phenomena of eld, is:

Gravitation theory, if matter and eld are given to determine spa etime, is:

These theories in usual formulation are relatively simple and well appli able
to des ribe a number of phenomena: it is lear, however, that they draw roughly
simplied pi tures of the really existing physi al world.

6. Di ulties arise when we want to des ribe ompli ated situations in whi h
only one of the three entities an be regarded as known and intera tions o ur
among the phenomena of the other two entities. The following graphi ally
delineated possibilities exist:

The third one is of no physi al interest, so far. However, the other two are
very important and we are for ed to deal with them. They represent qualitatively
new problems and they annot be redu ed to the previous well-known theories,
ex ept some spe ial ases treated in the next item.
Ele tromagneti radiation of mi roparti les is, for instan e, a phenomenon,
whi h needs su h a theory. Usual quantum ele trodynami s serves as a theory
for its des ription, and in general usual quantum eld theory is destined to
des ribe the intera tion of eld and matter in a given spa etime.
As it is well known, usual quantum eld theory has failed to be ompletely
orre t and satisfa tory. One might suspe t the reason of the failure is that usual
quantum eld theory was reated in su h a way that the notions and formulae of
me hani s were mixed with those of eld theory. This way leads to nowhere: in
me hani s the eld phenomena are xed, in eld theory the matter phenomena
are xed; the orresponding notions \sti" on one side annot be fused orre tly
to produ e notions \ exible" on both sides.
The ompli ated mathemati s of quantum eld theory does not allow us to
present a simple example to illustrate the foregoings, whereas lassi al ele trodynami s oers an ex ellent example. The ele tromagneti eld of a point harge

moving on a pres ribed path is obtained by the Lienard{Wie hert potential whi h
allows us to al ulate the for e due to ele tromagneti radiation a ting upon the
harge. Then the Newtonian equation is supplemented with this radiation rea ting for e | whi h is dedu ed for a point harge moving on a given path | to
get the so- alled Lorentz{Dira equation for giving the motion of a point harge
in an ele tromagneti eld. No wonder, the result is the nonsense of \runaway
solutions".
Ele tromagneti radiation is an irreversible pro ess; in fa t every pro ess in
Nature is irreversible. The des ription of intera tions must re e t irreversibility.
Me hani s (Newtonian equation, S hrodinger equation) and ele trodynami s
(Maxwell equations) i.e. the theories dealing with a tion instead of intera tion
do not know irreversibility. Evidently, no amalgamation of these theories an
des ribe intera tion and irreversibility.

7. There is a spe ial ase in whi h intera tion an be redu ed to some

ombination of a tions yielding a good approximation. Assume that matter
phenomena an be divided into two parts, a \big" one and a \small" one. The
big one and eld (or spa etime) are onsidered to be given and supposed to
produ e spa etime (or eld) whi h in turn a ts on the small matter phenomena
to determine them. The situations an be illustrated graphi ally as follows:

An example for the appli ation of this tri k is the des ription of planetary
motion in general relativity, more losely, the advan e of the perihelion of Mer ury. Then eld is supposed to be absent, the big Sun produ es spa etime and
this spa etime determines the motion of the small Mer ury. Doing so we negle t
that spa etime is in uen ed by Mer ury and the motion of Sun is in uen ed by
spa etime as well, i.e. we negle t intera tion.
The se ond example is similar. A given spa etime and a heavy point harge
are supposed to produ e an ele tromagneti eld and this ele tromagneti eld
determines the motion of a light point harge. Doing so we negle t that the
ele tromagneti eld is in uen ed by the light point harge and the motion of
the heavy point harge is in uen ed by the ele tromagneti eld, i.e. we negle t
intera tion.

PART ONE

SPACETIME MODELS

INTRODUCTION

1. The prin iples of ovarian e and of relativity

1.1. Today the guiding prin iple for nding appropriate laws of Nature is
the prin iple of general relativity: any kind of observer should nally on lude
the same laws of Nature; the laws are independent of the way we look at them.
The usual mathemati al method of applying this prin iple is the following: in
a ertain referen e frame we have an equation that, as we suspe t, expresses
some law independent of the referen e frame. The way to he k this is referred
to as the prin iple of ovarian e: transfer the equation into another referen e
frame with an appropriate transformation (Galilean, Lorentzian, or a general
oordinate transformation), and if the form of the equation remains the same
after this pro edure, then it an be a law of some phenomenon. This method
an be illustrated in the following way:

It seems quite natural to organize the pro edure in su h a way; this is how
Galileo and Newton started it and this is how Einstein nally on luded to the
prin iple of general relativity. What ould be the next step? Very simple: sin e
the laws of Nature are the same for all observers, the theoreti al des ription does
not need the observer any longer; there should exist a way of des ribing Nature
without observers. In fa t, at that time Einstein said this in another way: \the
des ription of Nature should be oordinate-free".
This was some 70 years ago but if we take a glan e at some books on theoreti al physi s today, we stumble upon an enormous amount of indi es; thinking

starts from referen e frames and remains there; the program of oordinate-free
des ription has not yet been a omplished.
The key step (but not the only step) towards being able to des ribe Nature
without observers is the mathemati al formulation of the \ta it agreement" behind the non-mathemati al notion of observers. This formulation nally lifts
the notion of the observer from the mist and starts reorganizing the method of
des ription in a way Einstein suggested. This reorganizing results in des ribing
Nature independently of observers. If we wish to test our theory by experiments,
we have to onvert absolute quantities into relative ones orresponding to observers and then to turn them into numbers by hoosing units of time, distan e
et . arriving in this way to indi es and transformation rules. Compared with
the previous situation, this an be illustrated as follows:

1.2. The most important result of the present book is this reorganizing of
the whole method of theoreti al des ription. In this framework the prin iple of
ovarian e and the prin iple of relativity sound very simple (en ouraging us that
this might be the right way).
Prin iple of ovarian e: a ording to our present knowledge, the des ription of Nature should be done by rst hoosing one of the non-relativisti , spe ial
relativisti and general relativisti spa etime models and then using the tools of
the hosen model.
Prin iple of relativity: there must be a rule in the spa etime models that
says how an arbitrary observer derives from the absolute notions its own quantities des ribing the phenomena.
2. Units of measurements

2.1. In pra ti e, the magnitudes of a physi al quantity (observable) are always

related to some unit of measurement i.e. to a hosen and xed magnitude. We
determine, for instan e, whi h distan e is alled meter and then we express all
distan es as non-negative multiples of meter.

In general the following an be said. Let C be the set of the magnitudes of an

observable. Taking an arbitrary element of C and a non-negative real number
; we an establish whi h element of C is  times ; denoted by  : In other
words, we give a mapping, alled multipli ation by non-negative numbers,
R+
0

 C ! C; (; ) ! 
with the following properties: for all 2 C

(i) 0 is the same element, alled the zero of C and is denoted by 0 as well;
(ii) 1 =
(iii)  ( ) = () for all ;  2 R+0 and 2 C;
(iv) if 6= 0 then J : R+0 ! C;  7!  is bije tive.
In ustomary language we an say that C is a one-dimensional one.
An addition an be dened on this one-dimensional one. It is easy to see
that the mapping, alled addition,


C  C ! C; (a; b) 7! J J 1 (a) + J 1 (b) =: a + b

is independent of :
Let us introdu e the notations C := f 1g  C; := ( 1; ) ( 2 C) and
D := ( C) [ C: Then we an give a multipli ation by real numbers
R

and an addition

 D ! D;

(; d) 7! d

D  D ! D;

(d; e) 7! d + e
that are trivial extensions of the operations given on C; so that
one-dimensional real ve tor spa e. For instan e,

D be omes a

 : = jj for  < 0; 2 C;
( ) : =  for  > 0; 2 C;
( ) : = jj for  < 0; 2 C:
Furthermore, the two \halves" of this ve tor spa e have dierent importan e:
the original one ontains the physi ally meaningful elements. We express this
fa t mathemati ally by orienting D with the elements of C (see IV.5).
The pre eding onstru tion works e.g. for distan e, mass, for e magnitude,
et . In some ases | e.g. for ele tri harge | we are given originally a onedimensional real ve tor spa e of observable values.
Thus we a ept that the magnitudes of observables are represented by elements of oriented one-dimensional real ve tor spa es alled measure lines. Choosing a unit of measurement means that we pi k up a positive element of the
measure line.

2.2. In pra ti e some units of measurement are dedu ed from other ones
by multipli ation and division; for instan e, if kg; m and s are units of mass,
distan e and time period, respe tively, then kgs2m is the unit of for e. The
question arises at on e: how an we give a mathemati ally exa t meaning to
su h a symbol? A ording to what has been said, kg; m and s are elements of
one-dimensional ve tor spa es; how an we take their produ t and quotient? To
give an answer let us list the rules asso iated usually with these operations; for
instan e,
(kg)( m) = ( )(kg m)
m  m
=
s  s

(;  2 R+0 );
( 2 R+0 ;  2 R+ ):

Extending these rules to negative numbers, too, we see that the usual multipli ation is a bilinear map on the measure lines and the usual division is a
linear-quotient map, with the additional property that the produ t and quotient
of non-zero elements are not zero.
Consequently, we an state that the produ t and quotient of units of measurements are to be dened by their tensor produ t and tensor quotient, respe tively
(see IV.3 and IV.4).
Thus if D; I and G denote the measure line of distan e, time period and mass,
respe tively, m 2 D; s 2 I; kg 2 G; then kgs2m := kgs

sm 2 GI

ID :

3. What is spa etime?

Spa e and time are fundamental notions in physi s: spa e and time form the
general ba kground of phenomena in Nature.
Let us examine these notions more losely.

3.1. Sitting in a room, we on eive that a orner of the room, a spot on

the arpet are points and the table is a part of our spa e. Looking through the
window we see trees, himneys, hills that form other parts of our spa e. A ar
travelling on the road is not a part of this spa e.
On the other hand, the seats, the dashboard, et . onstitute a spa e for
someone sitting in the ar. Looking out he sees that the trees, the houses, the
hills are running, they are not parts of the spa e orresponding to the ar.

Consequently, the spa e for us in the room and the spa e for the one in the
ar are dierent. We have as ertained that spa e itself does not exist, there is
no absolute spa e, there are only spa es relative to material obje ts. A spa e is
onstituted by a material obje t.

3.2. Pro esses indi ate that time passes: we breathe, someone is speaking, a
lo k is ti king, the Sun pro eeds on the sky. In fa t this is time: the sequen e
of pro esses. Time, too, is onstituted by material obje ts.
Immediately the question arises: is time absolute or relative? In other words:
is the same time realized in the room and in the ar or not? And even: is the
same time realized in two dierent orners of the room?
There are no evident answers to these questions. Our simplest everyday
experien e suggests that time is absolute. However, some experiments ontradi t
this suggestion.
3.3. To relate the spa e of the room and that of the ar, we must involve
time, too. Spa e and time relative to a material obje t interweave to express
spa e and time relative to another material obje t. This reason suggests that a
unique spa etime exists whi h is observed by material obje ts as spa e and time.
We an think that spa e and time are something like side views of spa etime.
We try to make mathemati al models for spa etime on the basis of the
properties of spa e and time observed by material obje ts.
3.4. Our rst abstra tion in onne tion with spa e is the point. The orner
of the room, a spot on the arpet stand for points of our spa e.
Our se ond abstra tion is the straight line. A light beam, a spanned thread
stand for a segment of a straight line. We dis over that one and only one straight
line is passing through two dierent points.
Our third abstra tion is the plane. A table surfa e, a window-glass stand for
a part of a plane. We nd out that one and only one plane passes through three
points that are not on a straight line.
The notion of planes oers us the notion of parallelism: two straight lines are
parallel if have no ommon point and there is a plane ontaining them.
!
Let x and y be two distin t points of our spa e. We introdu e the ve tor xy
to be the straight line segment between x and y; oriented in su h a way that x
! = uv
! in the
and y are its initial and nal points, respe tively. We agree that xy
ase x 6= u; y 6= v if and only if the orresponding lines | i.e. the line passing
through x and y and the line passing through u and v as well as the line passing
through x and u and the line passing through y and v| are parallel.

! to be a \degenerate" segment; if we a ept

For a spa e point x we onsider xx
! and uu
! we nd that they are equal for
the pre eding rule for the equality of xx
all spa e points x and u; we all this ve tor zero.
! = xy
! + xz
!
With the aid of parallelism we introdu e the sum of two ve tors: xu
!
!
if and only if yu = xz:

A fundamental property of this addition is that for arbitrary spa e points x;

y and z
! + yz
! + zx
! = xx
! = zero ve tor:
xy
We have the well-known Eu lidean method of onstru ting the positive rational multiple of a ve tor (segment); sin e we feel the spa e is \ ontinuous", we
are onvin ed that all positive real multiples of a ve tor make sense. We a ept
that a ve tor multiplied by 1 is the same segment oriented inversely.

3.5. We have given two operations on the ve tors: addition and multipli ation
by real numbers. These operations satisfy the ne essary requirement that the
ve tors be indeed ve tors, i.e. the set of ve tors endowed with these operations
is a ve tor spa e.
We know that at most three linearly independent spa e ve tors an be found.
Moreover, we an ompare the ve tors with respe t to their length , and we
introdu e the angle between two ve tors. We nd that the sum of the lengths of
two sides of a triangle is larger than the third side, and the sum of the angles of
a triangle is the straight angle.
To sum it up, the ve tors of our spa e form a three-dimensional Eu lidean
ve tor spa e (see V.3).
Let us onsider three ve tors that are not in the same plane (linearly independent ve tors). We an order them in two manners: in right-handed way
and in left-handed way. It is an interesting question whether the right-handed

order and the left-handed order are physi ally equivalent or not: is there a phenomenon that makes dierent the two orders, i.e. the phenomenon exists but
its re e tion does not. Our simplest experien e indi ates that the two orders
are equivalent. However, some more ompli ated phenomena show that they are
not: e.g. the stru ture of mole ules of living organs, or the snail shells. Re ently
it was demonstrated that the de ay of K mesons exhibits learly that the righthanded order and the left-handed one are not physi ally equivalent. We re e t
this fa t by saying that the ve tor spa e in question is oriented (see IV.5).

3.6. It is emphasized that the points of our spa e do not form a ve tor spa e;
we asso iate a ve tor to ea h ordered pair of spa e points. This orresponden e
between spa e point pairs and ve tors has properties whi h suggest us a epting
that our spa e is an ane spa e (see Chapter VI).
Summarizing what we have found we state that our spa e is a threedimensional oriented Eu lidean ane spa e (see VI.1.6).
3.7. As on erns time, we are onvin ed that it passes \uniformly". We an
determine the period between two arbitrary time points, and time periods are
summed up as time is passing. We give sense to the real multiple of time periods.
Time is evidently oriented: past and future are not equivalent.
Thus we state that our time is a one-dimensional oriented ane spa e.
3.8. The ane stru ture of time(s) and the ane stru ture of spa es relative
to material obje ts are related to ea h other.
More losely, if an inertial material obje t observes that a body moves uniformly on a straight line then another inertial material obje t observes the same
body moving uniformly on a straight line, too. Uniform motion, involving the
ane stru ture of both time and spa e, is independent of observers.
This indi ates that spa etime itself has an ane stru ture.
3.9. After having gathered the properties of our spa e and time, and having
obtained ni e stru tures, let us hasten to pose the uneasy questions: have we
reasoned properly? have we not made some mistakes? have we not left anything
out of onsideration?
There is a serious obje tion to our reasoning: we have extrapolated our experien e gained in human size to mu h larger and mu h smaller size, too.
Let us examine rst our on ept of ontinuity of spa e and time. A ording
to our ommon experien e, i.e. from human point of view, water is a ontinuous
material. However, we already know that it is rather oarse: a mi robe does not
per eive it to be ontinuous at all. Are perhaps spa e and time oarse as well?
At present no experimental fa t supports this possibility but we annot ex lude
it in good faith.

Let us a ept the ontinuity of spa e and time. Our onvi tion that a ve tor
an be asso iated with two spa e points is based on the fa t that e.g. we an
span a thread between the orner of the room and a spot on the arpet, or we
an produ e a light beam between them. But how an we determine the ve tor
between two points whose distan e is mu h smaller than the diameter of the
thread or the light beam? If we an dene ve tors for su h near points, too, do
they obey the ustomary rules of addition and multipli ation by real numbers?
We meet a similar problem if we want to give sense of ve tors orresponding
to points very far from ea h other. A thread annot but a light beam an draw
a straight line between Earth and Moon; however, it is not evident at all that
addition and multipli ation by real numbers of su h huge ve tors make sense
with the ustomary properties.
Indeed, some experiments show that in astronomi al size the ve torial operations annot be dened for segments dened by light beams. At present we have
no similar knowledge regarding minute size.
Evidently, the same problems arise for small and large time periods.

3.10. The part of sea surfa e seen from a ship seems to be plane though it is
a part of a sphere. The domain of spa e and time observed by us seems to be a
part of an ane spa e. Are spa e and time ane?
Let us re all that we have gathered the properties of our spa e and time in
order to establish the properties of spa etime. Sin e we ould not settle exa tly
the properties of our spa e and time, we annot do that for spa etime either.
However, this does not matter. We are fa ed a kind of fundamental problem:
on the basis of some experien e we have made some abstra tions to reate
mathemati al models. Su h a model is not the reality itself; it is an image
| a ne essarily simplied and distorted image | of reality. Reality and model
should not be onfused!
A epting our experien e regarding human size as global , i.e. extrapolating
it to very small and large size, too, we make models in whi h spa etime is a
four-dimensional ane spa e.
The non-relativisti model and the spe ial relativisti model are of this kind.
The dieren e between the two models is | regarding their physi al ontent
and not their mathemati al form | that time or light spread are taken to be
absolute, respe tively.
If we admit that our experien e is only lo al, i.e. onsidering it approximately
true even in human size, we give up the ane stru ture and we make models in whi h spa etime is a four-dimensional manifold. These are the general
relativisti spa etime models.
What kind of spa etime models shall we develop? This depends on our
intention for what purpose we wish to employ it.
The non-relativisti spa etime model is suitable for the des ription of \sluggish" me hani al phenomena | when bodies move relative to ea h other with

velo ities mu h smaller than light speed | and of stati ele tromagneti phenomena.
The spe ial relativisti spa etime model is suitable for the des ription of all
me hani al and ele tromagneti phenomena, but it has a more ompli ated stru ture than the non-relativisti one, therefore it is suitable for \brisk" me hani al
phenomena and non-stati ele tromagneti phenomena.
To des ribe osmi phenomena we have to adopt general relativisti spa etime
models.

3.11. At last, let us speak about how we imagine the points of spa etime. We
have our notions of spa e points and time points (instants). Roughly speaking,
a point of spa etime is the fusion of a spa e point and a time point: a spa etime
point an be on eived as \here and now" or \there and then". We an say
more. A lamp ashes, two billiard balls ollide: \there and then" is in arnated.
Thus spa etime points an be illustrated by su h o urren es.
We all attention to the fa t that one often says event instead of o urren e
whi h auses a number of misunderstandings. Namely, the notion of an event
is well dened in probability theory and in physi s as well. An event always
happens to the obje t in question: the ash is an event of the lamp, the ollision
is an event of the balls, they are not events of spa etime. These events are
illustrations only and are not equal to a spa etime point.
That is why we prefer to say o urren e when relating these events of some
material obje ts to spa etime points.

I. NON-RELATIVISTIC SPACETIME
MODEL

1. Fundamentals

1.1. Denition of the spa etime model

1.1.1. A ording to what has been said in the Introdu tion, now we model
spa etime by a four-dimensional oriented ane spa e, denoted by M; let M be
the orresponding ve tor spa e.
The ane stru ture does not re e t ompletely our fundamental knowledge
of spa etime.
Let us a ept that absolute time exists. This will be the most important feature
of a non-relativisti spa etime model. Time is modelled by a one-dimensional
oriented ane spa e, denoted by I; let I denote the underlying ve tor spa e.
Spa etime and time are not independent; the phrase \time is absolute"
means that the time point orresponding to ea h spa etime point is determined
uniquely; in other words, there is a mapping  : M ! I: The ane stru tures of
spa etime and time are evidently related somehow. We express this relation by
supposing that  is an ane map (over the linear map  : M ! I):
Now we have to put the Eu lidean stru ture of \spa e" into the model (quotation marks are used be ause we well know that absolute spa e does not exist, we
do not want to put spa e into the model). To go on the right way, let us observe
that the Eu lidean stru ture of our spa e is established on the basis of simultaneity: the ve tor between the orner of the room and the spot on the arpet
is not dened by the orner yesterday and the spot today. Without introdu ing
spa e, we an introdu e the Eu lidean stru ture with the aid of simultaneity as
follows.
Let t be an instant, i.e. an element of I: Then
1

 (ftg) = fx 2 Mj  (x) = tg
is the set of simultaneous spa etime points to whi h the same instant t orresponds. It is a three-dimensional ane subspa e of M over the ve tor spa e (see
VI.2.2)
E : = Ker  = fx 2 Mgj 
x = 0g:

We a ept that there is a Eu lidean stru ture on E : we introdu e the measure

line of distan es, D; and a positive denite symmetri bilinear map b : E  E !
D
D; so (E; D; b) is a three-dimensional Eu lidean ve tor spa e.

1.1.2. Now we are ready to formulate a orre t denition.

Denition. A non-relativisti spa etime model is a quintuplet (M; I; ;
D; b) where
| M is an oriented four-dimensional real ane spa e (over the ve tor spa e
M);
| I is an oriented one-dimensional real ane spa e (over the ve tor spa e I);
|  : M ! I is an ane surje tion (over the linear surje tion
 : M ! I);
| D is an oriented one-dimensional real ve tor spa e,
| b : E  E ! D
D is a positive denite symmetri bilinear map where
E : = Ker  :

We shall use the following names:

M is spa etime or world ,
I is time, I is the measure line of time periods,
 is the time evaluation,
D is the measure line of distan es,
b is the Eu lidean stru ture on simultaneity.
Elements of M and I are alled world points and instants, respe tively. The
world points x and y are simultaneous if  (x) =  (y):

1.1.3. A world ve tor, i.e. an element x of M is alled spa elike or timelike

if x 2 E or x 62 E; respe tively. Evidently, x is spa elike if and only if 
x = 0:
The set of timelike elements onsists of two disjoint open subsets:
T! : = fx 2 Mj 
x > 0g;
T : = fx 2 Mj 
x < 0g:

(Re all that I is oriented, thus it makes sense to speak about its positive and
negative elements, see IV.5.3.) The ve tors in T! and in T are alled futuredire ted and past-dire ted, respe tively.
We often illustrate the world ve tors in the plane of the page:

1.1.4. If t 2 I then 1 1 (ftg) is an ane hyperplane in M; dire ted by E: The

orresponden e t 7!  (ftg) is a bije tion between I and the ane hyperplanes
dire ted by E: We make use of this orresponden e to identify the two sets, i.e. to
regard instants as ane hyperplanes dire ted by E: In this way an instant equals

the set of orresponding simultaneous world points. This tri k makes thinking
simpler and reates the possibility of omparison with relativisti models where
time an be dened only by hypersurfa es.

Sin e I is oriented and one-dimensional, a total ordering is given on it: we say

that t 2 I is later than s 2 I (or s is earlier than t) and we write s < t if t s is
a positive element of I:
Spa etime, too, will be illustrated in the plane of the page. Then verti al
lines stand for the instants (hyperplanes of simultaneous world points). A line
standing to the right of another is taken to be later.

If x is a world point, x +T! and x +T are alled the future-like and past-like
part of M; with respe t to x:

1.2. Stru ture of world ve tors and ove tors

1.2.1. There are spa etime and time in our non-relativisti spa etime model
and there is no spa e. However, there is something spa elike: the linear subspa e
E of M: Later we see what the spa elike feature of E onsists in. We nd an
important \ omplementary" onne tion between E and I: Let
i:E!M
denote the anoni al inje tion (embedding; if q 2 E; then i
q equals q regarded
as an element of M; evidently, i is linear). Then we an draw the diagram
E !i M !I ;

i is inje tive,  is surje tive, and Ran i = Ker  ; thus 
i = 0:

M; the dual of M will play an important role. Though it is also a fourdimensional oriented ve tor spa e, there is no anoni al isomorphism between
M and M; these ve tor spa es are dierent.
0:

A diagram similar to the previous one is drawn for the transposed maps:
I ! M i! E :
  is inje tive, i is surje tive (see IV.1.4) and Ran   = Ker i ; thus i
  =

It is worth mentioning that for k 2 M ; i
k = k
i is the restri tion of k

onto E:

1.2.2. Sin e   is inje tive, its range is a one-dimensional linear subspa e of


M whi h will play an important role:
Ran   = f 
ej e 2 I g = fe
 j e 2 I g = I
 :
Observe that k 2 M is in I
 if and only if i
k = k
i = 0; thus
I
 = fk 2 Mj k
q = 0 for all q 2 Eg:

If the dot denoted an inner produ t on some ve tor spa e then this would
mean that I
 is orthogonal to E; please, note, now I
 and E are in
dierent ve tor spa es, they annot be orthogonal to ea h other. We say that
I
 is the annullator
of E:
Illustrating M on the plane of the page, we draw a horizontal line for the
one-dimensional linear subspa e I
 :

As usual, the elements of M are alled ove tors. The ove tors in the linear
subspa e I
 are timelike, and the other ones are spa elike.

1.2.3. It will be often onvenient to use tensorial forms of the above linear
maps. A ording to IV.3.4 and IV.1.2 we have
2 I
M ;
  2 M
I ;


i 2 M
E ;
i 2 E
M :

1.2.4. With the aid of  ; the orientations of M and of I determine a unique

orientation of E:
Proposition. If (e1 ; e2 ; e3) is an ordered basis of E; then
(x; e1 ; e2 ; e3 ) and (y; e1 ; e2 ; e3 ) are equally oriented for all
x; y 2 T! :
Proof. Evidently, (x; e1 ; e2 ; e3) and (x; e1 ; e2 ; e3) are equally oriented if
 2 R+ ; hen e we an suppose that 
y = 
x; i.e. q := y x 2 E: Then
y ^ e1 ^ e2 ^ e3

= (x + q) ^ e1 ^ e2 ^ e3 = x ^ e1 ^ e2 ^ e3 ;

hen e the statement is true by IV.5.1.

Denition. An ordered basis (e1 ; e2 ; e3) of E is alled positively oriented
if (x; e1 ; e2 ; e3 ) is a positively oriented ordered basis of M for some (hen e for
all) x 2 T! .

1.2.5. (E; D; b) is a three-dimensional Eu lidean ve tor spa e, E and D are

oriented. An important relation is the identi ation
E  E :
D
D

We shall use the notation

N := DE

and all the results of se tion V.3.

In parti ular, we use a dot produ t notation instead of b :
q
q0

The length of q 2 E is

:= b(q; q0 ) 2 D
D

jqj := pq
q 2 D+0 ;

(q; q0 2 E):

and the angle between the non-zero elements q and q0 of E is

q
q0
arg(q; q0 ) := ar os
jqjjq0 j :
The dot produ t an be dened between spa elike ve tors of dierent types
(see later, Se tion 1.4) as well; e.g. if A and B are measure lines, for w 2 AE and
E we have
z2B
w
z

D
2D
A
B ;

arg(w; z) := ar os

jwj := pw
w 2 D
A:

w
z
jwjjzj ;

1.2.6. Do not forget that timelike ve tors (elements of M outside E) have no

length, no angles between them.
I
 is an oriented one-dimensional ve tor spa e, hen e the absolute value
of its elements makes sense; thus a length (absolute value) an be assigned to
a timelike ove tor. However, the length of spa elike ove tors (elements of M
outside I
 ) and the angle between two ove tors are not meaningful.

1.2.7. The Eu lidean stru ture of our spa e is deeply xed in our mind,
therefore we must be areful when dealing with M whi h has not a Eu lidean
stru ture; espe ially when illustrating it in the Eu lidean plane of the page. Keep
in mind that ve tors out of E have no length, do not form angles. The following
onsiderations help us to take in the situation.
Re all that the linear map  : M ! I an be applied to element of MI and
then has values in II  R (see V.2.1 ). Put
V(1) :=


M
u2
I j
u=1 :

A ording to VI.2.2, V(1) is an ane subspa e of MI over EI : It is illustrated

as follows:
Three elements of V(1) appear in the gure. Observe that it makes no sense
that
| u1 is orthogonal to EI (there are no ve tors orthogonal to EI );
| the angle between u1 and u2 is less than the angle between u1 and u3
(there is no angle between the elements of V(1); )
| u2 is longer than u1 (the elements of V(1) have no length).
We shall see in 2.1.2 that the elements of V(1) an be interpreted as velo ity
values.

1.2.8. Sin e there is no ve tor orthogonal to E; the orthogonal proje tion

of ve tors onto E makes no sense. Of ourse, we an proje t onto E in many
equivalent ways; the following proje tions will play an important role.
Let u be an element of V(1): Then u
I := futj t 2 Ig is a one- dimensional
linear subspa e of M; u
I and E are omplementary subspa es, thus every
ve tor x an be uniquely de omposed into the sum of omponents in u
I and
in E; respe tively:
x = u(
x) + (x u(
x)) :
The linear map
u : M ! E ; x ! x u(
x)
is the proje tion onto E along u: It is illustrated as follows:

V(1) is represented by a dashed line expressing that V(1) is in fa t a subset

of MI :
Observe that
u
i = idE

and in a tensorial form u 2 E

M :

1.2.9. Proposition. Let u 2 V(1): Then

hu := ( ; u ) : M ! I  E
x!
7 (
x; u
x)

is an orientation-preserving linear bije tion, and

(t 2 I; q 2 E):

hu 1 (t; q ) = ut + q

1.3.1.

1.3. The arithmeti spa etime model

Let us number the oordinates of elements of R4 from 0 to 3 :

0
1
2
( ;  ;  ;  3 ) 2 R4 : The anoni al proje tion onto the zeroth oordinate,
pr0 : R4

! R;

( 0 ;  1 ;  2 ;  3 ) 7!  0

is a linear map whose kernel is f0g R3 whi h we identify with R3 : Let B denote
3
P
the usual inner produ t on R3 : B(x; y) = xi yi : Endow R and R4 with the
i=1
standard orientation.


It is quite evident that R4 ; R; pr 0 ; R; B is a non-relativisti spa etime model
whi h we all the arithmeti non-relativisti spa etime model.
In the arithmeti spa etime model we have:
M = M = R4 ;

I = I = R;

 =  = pr0 ;
E = f0g  R3  R3 ;

D = R;
b = B:

Then

i:E!M

equals

R3

! R4 ; (x1 ; x2 ; x3 ) 7! (0; x1 ; x2 ; x3 ):

The usual identi ation yields M = (R4 )  R4 ; the ove tors are indexed
in subs ripts: (k0 ; k1 ; k2 ; k3 ) 2 (R4 ) (see IV.1.4).
In the same way, I = R  R; but here we annot make distin tion with the
aid of indi es.
Then
i : M 7! E equals (R4 ) ! (R3 ) ; (k ; k ; k ; k ) 7! (k ; k ; k )
0 1 2 3

1 2 3

and



: I ! M equals

R

! (R4 ) ; e 7! (e; 0; 0; 0):

1.3.2. It is an unpleasant feature of the arithmeti spa etime model that the
same obje t, R4 ; represents the ane spa e of world points and the ve tor spa e
of world ve tors (and even the ve tor spa e of ove tors). For a lear distin tion
we shall write Greek letters indi ating world points (ane spa e elements) and
Latin letters indi ating world ve tors or ove tors.
Moreover, the notations will be mu h simpler if we onsider R4  R  R 3 ; and
we write (; ) or ( 0 ; ) and (t; q) for its elements; similarly, (e; p) denotes an
element of (R  R3 )  R (R3 ) : Then
 : R  R3 ! R;
(; ) 7! ;
3
 : R  R ! R;
(t; q) 7! t;
3
3
i : R ! R  R ; q 7! (0; q);

i : (R  R3 ) ! (R3 ) ; (e; p) 7! p;

  : R ! (R  R3 ) ;
e 7! (e; 0):
The last formula means that I
 now equals R  f0g:
Of ourse,  and  are equal though we have written the same formula in
dierent symbols. This is a tri k similar to that of subs ripts and supers ripts:
we wish to distinguish between dierent obje ts that appear in the same form.

1.3.3. Now we have MI = RR  R4 ; and

4

V(1) = f(v0 ; v) 2 R  R3 j v0 = 1g = f1g  R3 :

V(1) has a simplest element: (1; 0) whi h is alled the basi velo ity value.
For (1; v) 2 V(1) we easily derive that
(1;v) : R  R3

! R3 ;

(t; q) 7! q

v t:

In parti ular, (1;0) is the anoni al proje tion from R  R3 onto R3 .

1.4. Classi ation of physi al quantities

1.4.1. In physi s one usually says e.g. that (relative) velo ity and a eleration
are three-dimensional ve tors and are onsidered as triplets of real numbers.
Although both are taken as elements of R3 ; they annot be added be ause they
have \dierent physi al dimensions". The framework of our spa etime model
assures a pre ise meaning of these notions.

A physi al dimension is represented by a measure line. Let A be a measure

line. Then the elements of

A are alled s alars of type A;

A
M are alled ve tors of type A;
M are alled ve tors of otype A;
A
A
(M
M) are alled tensors of type A;
M
M are alled tensors of otype A:
A
Cove tors of type A; et . are dened similarly with M instead of M:
In the ase A = R we omit the term \of type R": In parti ular, the elements
of M
M and M
M are alled tensors and otensors, respe tively; the
elements of M
M and M
M are mixed tensors.
Re all the identi ations A
M  M
A et . (see IV.3.6).

Be ause of the identi ation M
A  M
A the ve tors of otype A oin ide

with the ve tors of type A :
1.4.2. The ve tors and tensors of type A in the subspa es A
E and
A
(E
E); respe tively, are alled spa elike. 
The ove tors of type A in the subspa e A
(I
 ) are alled timelike.

A ording to our onvention (V.2.1 and V.2.2), the dot produ t of ove tors
and ve tors of dierent types makes sense; e.g.
for k 2 B
M and z 2 A
M  M
A we have k
z 2 B
A:
In parti ular,
for 

2 I
M

and z 2 A
M we have 
z 2 I
A;

similarly,

M we have 
w 2 I ;
A
A
for T 2 A
(M
M) we have 
T 2 I
A
M:
Evidently, z 2 A
M is spa elike if and only if 
z = 0:
In the same way, i : M ! E is lifted to ove tors of type A; et . i.e.
for i 2 E
E and h 2 A
M we have i
h 2 A
E et .
Evidently, h 2 A
M is timelike if and only if i
h = 0:
for w 2

1.4.3. In non-relativisti physi s one usually introdu es the notion of s alars,

three-dimensional ve tors, three-dimensional pseudo-ve tors and pseudo-s alars
as quantities having some pres ribed transformation properties. One is for ed to
adapt su h a denition be ause only oordinates are onsidered, only numbers
and triplets of numbers are used, and one must know whether a triplet of numbers
is the set of oordinates of a ve tor, or not. Of ourse, ve tors an have dierent
\physi al dimensions".
Now we formulate the orresponding notion in the framework of our nonrelativisti spa etime model. The elements of
R

E
E^E
E^E^E

are the s alars;

are the spa elike ve tors;
are the spa elike pseudo-ve tors of type
are the pseudo-s alars of type

D:

D;

The rst and the se ond names do not require explanation. The third and
fourth names are based on the fa t that we have anoni al linear bije tions
E ^ E ! E
D and E ^ E ^ E ! D
D
D (see V.3.17); the pseudo-ve tors are
\similar" to spa elike ve tors of type D; and the pseudo-s alars are \similar" to
3
s alars of type
D:
Having the notion of ve tors of type A; it is evident, how we shall dene
spa elike pseudo-ve tors and pseudo-s alars of diverse types. For the sake of
simpli ity, we onsider now \physi ally dimensionless" quantities: R; N; N ^ N;
N^N^N: Then we have the linear bije tions j : N^N ! N and jo : N^N^N !
R:
Let R : E ! E be an orthogonal map whi h is onsidered to be an orthogonal
map N ! N as well. We say that R is a rotation if it has positive determinant.
The determinant of the inversion S := idE is negative.
0
0
By denition,
R :=
S := idR; the s alars are not transformed.
Ve tors are transformed under R and S a ording to the denition of these
operations.
Pseudo-ve tors are transformed by R ^ R and S ^ S (IV.3.2.1); formulae in
V.3.16 say that

j (R ^ R) = R j;

j (S ^ S) =

j=j

whi h means that the pseudo-ve tors are transformed by rotations like ve tors
but they are not transformed by the inversion.
Similarly we have that
jo (R ^ R ^ R) = jo ;

jo (S ^ S ^ S ) = jo ;

the pseudo-s alars are not transformed by rotations and they hange sign by the
inversion.

1.5. Comparison of spa etime models

1.5.1. The spa etime model is dened as a mathemati al stru ture. It is an

interesting question both from mathemati al and from physi al points of view:
how many \dierent" non-relativisti spa etime models exist?
To answer, rst we must dene what the \dieren e" and the \similarity"
between two spa etime models mean. We pro eed as it is usual in mathemati s;
for instan e, one denes the linear stru ture (ve tor spa e) and then the linear
maps as the tool of omparison between linear stru tures; two ve tor spa es are
of the same kind if there is a linear bije tion between them, in other words, if
they are isomorphi .
Denition. The non-relativisti spa etime model (M; I; ; D; b) is isomorphi
to the non-relativisti spa etime model (M0 ; I0 ;  0 ; D0 ; b0 ) if there are
(i ) an orientation-preserving ane bije tion F : M ! M0 ;
(ii ) an orientation-preserving ane bije tion B : I ! I0 ;
(iii ) an orientation-preserving linear bije tion Z : D ! D0 su h that
(I)  0 F = B ;
(II) b0 (F  F ) = (Z
Z ) b:
The triplet (F; B; Z ) is an isomorphism between the two spa etime models.
If the two models oin ide, isomorphism is alled automorphism. An automorphism (F; B; Z ) of (M; I; ; D; b) is stri t if B = idI and
Z = idD :
In the denition and later on, B and F are the linear maps under B and F;
respe tively.
Two ommutative diagrams illustrate the isomorphism:
M

?
yB

Fy

EE

I
F

 Fy

D
D
?
yZ

Z:

! D0
D0
b0
The denition is quite natural and simple. It is worth mentioning that (I)
implies
0 F = B ;
thus for q 2 E we have  0
F
q = B

q = 0 whi h means that F maps E
into (and even onto) E0 ; hen e the requirement in (II) is meaningful.
M0

0

I0

E0  E0

b!

It is evident that (F 1 ; B 1 ; Z 1 ); the inverse of (F; B; Z ); is an isomorphism

as well. Moreover, if (F 0 ; B 0 ; Z 0 ) is an isomorphism between (M0 ; I0 ;  0 D0 ; b0 ) and
(M00 ; I00 ;  00 ; D00 ; b00 ); then (F 0 F; B 0 B; Z 0 Z ) is an isomorphism, too.

1.5.2. Proposition. The non-relativisti spa etime model (M; I; ; D; b) is

isomorphi to the arithmeti spa etime model.
Proof. Take
(i ) a positive element s of I;
(ii ) a positive element m of D;
(iii ) an element e0 of T ! su h that 
e0 = s;
(iv ) a positively oriented orthogonal basis (e1 ; e2 ; e3 ); normed to m; of E;
(v ) an element o of M:
Then u := es0 is in V(1) and it is not hard to see that
F : M ! R4 ; x 7!
B : I ! R; t 7!
Z

: D ! R; d 7!

(x o) ;  e
u
(x o) 
s
m2

t  (o)
d
m

=1;2;3

is an isomorphism.
Observe that (e0 ; e1 ; e2 ; e3 ) is a positively oriented basis in M; and F is the
ane oordinatization of M orresponding to o and that basis.
The isomorphism above has the inverse
R4

! M; ( 0 ;  1 ;  2 ;  3 ) 7! o +

! I;  7! s;
R ! D; !
7 m:

3
X
i=o

 i ei ;

1.5.3. An important onsequen e of the previous result is that two arbitrary

non-relativisti spa etime models are isomorphi , i.e. are of the same kind. The
non-relativisti spa etime model as a mathemati al stru ture is unique. This
means that there is a unique \non-relativisti physi s".
Please, note: the non-relativisti spa etime models are of the same kind,
but, in general, are not identi al. They are isomorphi , but, in general, there
is no \ anoni al" isomorphism between them, we annot identify them by a
distinguished isomorphism. It is a situation similar to that well known in the

theory of ve tor spa es: all N -dimensional ve tor spa es are isomorphi to K N
but, in general, there is no anoni al isomorphism between them.
Sin e all non-relativisti spa etime models are isomorphi , we an use an
arbitrary one for investigation and appli ation. However, an a tual model an
have additional stru tures. For instan e, in the arithmeti model, spa etime and
time are ve tor spa es, time is anoni ally embedded into spa etime as R  f0g;
V(1) has a distinguished element, (1; 0). This model tempts us to multiply world
points by real numbers (although this has no physi al meaning and that is why
it is not meaningful in the abstra t spa etime stru ture), to onsider spa etime
to be the Cartesian produ t of time and spa e (but spa e does not exist!), to
say that the distinguished element of V(1) is orthogonal to the spa e (su h an
orthogonality makes no sense in the abstra t spa etime stru ture), et .
To avoid su h onfusions, we should keep away from similar spe ially onstru ted models for investigation and general appli ation of the non-relativisti
spa etime model. However, for solving spe ial problems, for exe uting some parti ular al ulations, we an hoose a onvenient a tual model. In the same way
as in the theory of ve tor spa es where a oordinatization | i.e. the use of K N |
may help us to perform our task.

1.5.4. In present day physi s one uses ta itly the arithmeti spa etime model.
One represents time points by real numbers, spa e points by triplets of real
numbers. To arrive at su h representations, one hooses a unit of measurement
for time and an initial time point, a unit of measurement for distan e and an
initial spa e point (origin) and an orthogonal spatial basis whose elements have
unit length.
However, all the previous notions have merely a heuristi sense. Take a glan e
at the isomorphism established in 1.5.2 to re ognize that the non-relativisti
spa etime model will give these notions a mathemati ally pre ise meaning.
Evidently, s and m are the units of time period and distan e, respe tively,
fe1 ; e2 ; e3 g is the orthogonal spatial basis whose elements have unit length;  (o)
is the initial time point and o in ludes somehow the origin of spa e as well. At
present only the sense of e0 is not lear; later we shall see that it determines
the spa e in question, be ause we know that absolute spa e does not exist; e0
hara terizes an observer whi h realizes a spa e.
1.6. The split spa etime model

1.6.1. As we have said, the arithmeti spa etime model is useful for solving
parti ular problems, for exe uting pra ti al al ulations. Moreover, at present,
one usually expounds theories, too, in the frame of the arithmeti spa etime
model, so we ought to \translate" every notion in the arithmeti language.
However, the arithmeti spa etime model is a little ponderous; that is why

we introdu e an \intermediate" spa etime model between the abstra t and the
arithmeti ones, a more terse model whi h has all the essential features of the
arithmeti spa etime model.
1.6.2. Let (M; I; ; D; b) be a non-relativisti spa etime model, and use the
notations introdu ed in this hapter. Let prI : I  E ! I be the anoni al
proje tion (t; q) 7! t:
Then (I  E; I; prI ; D; b) is a non-relativisti spa etime model, alled the split
non-relativisti spa etime model orresponding to (M; I; ; D; b):
It is quite obvious that for all o 2 M and u 2 V(1);
M ! I  E;
I ! I;
D ! D;

x 7! hu
(x o)
t ! t  (o)
d!d

is an isomorphism of the two non-relativisti spa etime models where hu is

dened in 1.2.9.

1.6.3. In the split spa etime model

 : I  E ! I;
i : E ! I  E;

(t; q) 7! t;
q 7! (0; q ):

With the usual identi ation (see IV.1.3) we have that in the split spa etime
model the ove tors are elements of I  E ; orrespondingly,
: I ! I  E ;
i : I  E ! E;



e 7! (e;

0);

(e; p) 7! p:

As a onsequen e, I
 = I  f0g:

1.6.4. In the split spa etime model

E
I

V(1) = f1g  ;
so there is a simplest element (the basi velo ity value ) in it: (1; 0):
We easily derive for (1; v) 2 V(1) :
(1;v)

: I  E ! E;

(t; q) 7! q

vt:

1.7. Exer ises

1. Let fe0 ; e1 ; e2 ; e3 g be a basis in M su h that fe1 ; e2 ; e3 g is an orthogonal
basis in 
E; normed to m 2 D+ : Put s := 
e0; u := es : Then
 

0

 ; u
2ei
s
m i=1;2;3

is the dual of the basis in question.

2. (i ) Let (e0 ; e1 ; e2 ; e3 ) be a positively oriented basis in M su h that

(e1 ; e2 ; e3 ) is a positively oriented basis in E; normed to m 2 D+ : Put s := 
e0 :
Take another \primed" basis with the same properties. Then

" :=

3
^
e
i=0 i
sm3

3 0
^
e
i=0 i

4
^
M ;
= 0 03 2
sm
I

3 D

whi h is alled the Levi{Civita tensor of the non-relativisti spa etime model.
In other words, if u 2 V(1) and (n1 ; n2 ; n3 ) is a positively oriented orthonorE ; then
mal basis in N = D
3
" = u ^ ^=1n :
(ii ) Let (k0 ; k1 ; k2 ; k3 )) and (k0 0 ; k0 1 ; k0 2 ; k0 3 ) be the dual of the bases in
question (see the previous exer ise). Then

" := sm3 i=0

^3 ki = s0 m0 3 i=0
^3 k0 i 2 I

3 D
^4 M ;
whi h is alled the Levi{Civita otensor of the non-relativisti spa etime model.
In other words, if r 2 D
M and i
r ( = 1; 2; 3) form a positively
oriented orthonormal basis in N = DE ; then
3

" =  ^ ^=1r :
3

" and " an be regarded as linear maps from hI

i
D into

4
^ M and from ^4 M into I

3 D (re all that ^4 M  ^4 M ): Prove
that " is the inverse of ":
3.

4. Take the arithmeti spa etime model and the usual matrix form of linear
maps RM ! RN : Then
 = (1 0 0 0);
0
1
0
1
0 0 0
v1 1 0 0
1
0
0
i = B 0 1 0 CA ; (1;v) =  v23 0 1 0 A :
v 0 0 1
0 0 1

2. World lines

2.1. History of a masspoint: world line

2.1.1. Let us onsider a material body whi h is mu h smaller than the other
ones around it. It an be onsidered point-like in our usual spa e, and its motion
is des ribed by a fun tion that assigns spa e points (the instantaneous positions
of the body) to time points. A larger body an be onsidered point-like, too, if
we are interested only in some aspe ts of its motion; e.g. we negle t that a ball
twirls when ying, and is ompressed when boun ing over a wall.
In most of the textbooks it is emphasized, rightly, that motion is a relative
notion. The motion of a material body makes sense only relative to another
material obje t and the same body moves dierently relative to dierent material
obje ts. However, this does not imply that a body an be des ribed only with
respe t to a hosen material obje t (in a \referen e frame"). Our spa etime
model allows an absolute des ription (independent of \referen e frames"). We
have to re ognize only that the existen e (whi h is usually alled the history ) of
the body is an absolute notion and this history seems to be a motion to another
material obje t.
The history of a material point is modelled in the spa etime model by a
fun tion that assigns world points to instants; the world point assigned to an
instant gives the instantaneous spa etime position of the existen e of the material
point. Of ourse, the instant of the assigned world point must oin ide with the
instant itself.
Denition. A fun tion r : I M is alled a world line fun tion if
(i ) Dom r is an interval,
(ii ) r is pie ewise twi e ontinuously dierentiable,
(iii )  (r(t)) = t for all t 2 Dom r:
A subset C of M is a world line if it is the range of a world line fun tion.
The world line fun tion r and the world line Ran r is global if
Dom r = I:
It an be shown easily that a world line C uniquely determines the world line
fun tion r su h that C = Ran r:

2.1.2. LetMthe world line fun tion r be twi e dierentiable at t: Then r_(t) 2 MI

and r(t) 2 I
I (see VI.3.9); moreover,

(r(s) r(t))
 (r(s))  (r(t))
s t

r_ (t) = lim
=
lim
=
lim
=1
s!t
s!t
s!t s t
s t
s t
and similarly we dedu e 
r(t) = 0; in other words,

r_(t) 2 V(1);

r(t) 2

I
I:

The same is true for the right and left derivatives at instants t where r is not
twi e dierentiable.
E
The fun tions r_ : I
V(1) and r : I
I
I an be interpreted as the
(absolute) velo ity and the (absolute) a eleration of the material point whose
history is des ribed by r:
That is why we all the elements of V (1) velo ity values and the elements of
E a eleration values.
I
I

2.1.3.

Re all that V(1) is a three-dimensional ane spa e over EI : The

elements of EI will be alled relative velo ity values; later we shall see the
motivation of this name.
We know that the Eu lidean stru ture of E indu es Eu lidean stru tures on
E and on E (see 1.2.5). The magnitude of a relative velo ity value is a positive
I
I
I
element of DI ; the magnitude of an a eleration value is a positive element of
D
I
I :
D and I are the measure lines of distan es and time periods, respe tively.
Choosing a positive element in D and in I we x the unit of distan es and the
unit of time periods; for instan e, (meter=)m 2 D and (se undum=)s 2 I:
Then the units of measurements of the relative velo ity and the a eleration are
m 2 D and m2 := m 2 D ; respe tively.
s
I
s
s
s I
I
We emphasize the following important fa ts.
(i ) The velo ity values are timelike ve tors of otype I; in parti ular they
are future-dire ted. They form a three-dimensional ane spa e whi h is not a
ve tor spa e; in parti ular, there is no zero velo ity value. A velo ity value has
no magnitude, velo ity values have no angles between themselves.
(ii ) The relative velo ity values are spa elike ve tors of otype I; they form a
three-dimensional Eu lidean ve tor spa e; there is a zero relative velo ity value.
Magnitudes and angles make sense for relative velo ity values.
(iii ) The a eleration values are spa elike ve tors of otype I
I; they form a
three-dimensional Eu lidean ve tor spa e; the a eleration values have magnitudes and angles between themselves.
The absen e of magnitudes of velo ity values means that \qui kness" makes
no absolute sense; it is not meaningful that a material obje t exists more qui kly
than another. A velo ity value hara terizes somehow the tenden y of the history
of a material point. Masspoints an move slowly or qui kly relative to ea h other.

2.1.4. A3world line fun tion

in the arithmeti spa etime model is r = (r0 ; r) :
0

R
R  R su h that r (t) = t for all t 2 Dom r: In other words, a world line
fun tion is given by a fun tion r : R R3 in the form t 7! (t; r(t)):
The rst and the se ond derivative of the world line fun tion (i.e. velo ity
and a eleration) are t 7! (1; r_ (t)) and t 7! (0; r(t)); respe tively.

2.2. A hara terization of world lines

The world lines are spe ial urves in M (for the notion of urves see VI.4.3).
It is evident that if C is a world line then C \ t has at most one element for
all t 2 I (where I is identied with the ane subspa es in M; dire ted by E; see
1.1.4). We shall use the symbol
C?t
for the unique element of C \ t if this latter is not void. Then we have that the
world line fun tion r orresponding to C is given by
Dom r = ft 2 I j C \ t 6= ;g;
r(t) = C ? t
(t 2 Dom r):
It is evident as well that a twi e dierentiable urve C for whi h C \ t has at
most one element for all t 2 I need not be a world line: it an have a spa elike
tangent ve tor.
Every non-zero tangent ve tor of a world line is timelike. The onverse is true
as well.
Proposition. Let C be a onne ted twi e dierentiable urve in M whose
non-zero tangent ve tors are timelike; then C is a world line.
Proof. Let p : R M be a parameterization of C: Then 
(p_()) 6= 0 for
all  2 Dom p: The fun tion  p : R
I is dened in an interval, is twi e
ontinuously dierentiable, its derivative 
p_ is nowhere zero; hen e it is stri tly
monotone, its inverse z := ( p) 1 is twi e ontinuously dierentiable as well
and z_ (t) = 1=
p_(z (t)); as it is well known. It is obvious then that r : = p z is
a world line fun tion and Ran r = C:

2.3. Classi ation of world lines

Denition. The twi e ontinuously dierentiable world line fun tion r and
the orresponding world line are alled
(i ) inertial if r = 0;
(ii ) uniformly a elerated if r is onstant,
(iii ) twist-free if r(s) is parallel to r(t) for all t; s 2 Dom r:
Proposition. The twi e ontinuously dierentiable world line fun tion r is
(i ) inertial if and only if there are xo 2 M and uo 2 V(1) su h that
r(t) = xo + uo (t  (xo ))

(t 2 Dom r);

(ii ) uniformly a elerated if and only if there are xo

2 I
EI su h that

ao

r(t) = xo + uo (t  (xo )) +

1
a (t  (xo ))2
2 o

2 M; uo 2

V(1) and

(t 2 Dom r);

(iii ) twist-free if and only if there exist xo 2 M; uo 2 V(1); 0 6= ao 2 I

EI and
a twi e ontinuously dierentiable fun tion h : I I
I for whi h h(0) = 0;
h_ (0) = 0 and

r(t) = xo + uo(t  (xo )) + ao h(t  (xo ))

(t 2 Dom r):

Proof. The validity of the assertions omes from the theory of dierential
equations; (i ) and (ii ) are quite trivial. For (iii ) observe that r is twist-free if
and only if there is a non-zero a eleration value ao and a ontinuous fun tion
:I
R (whi h an be zero) su h that r(t) = ao (t): If xo is a point in
the range of r; we dene  : I R by (t) := ( (xo ) + t) whi h means that
(t  (xo )) = (t): Then h will be the fun tion whose se ond derivative is 
and that satises the above given initial ondition.
Observe that a twi e ontinuously dierentiable world line fun tion r is twistfree if and only if r=jrj is onstant on ea h interval where the se ond derivative
is not zero.
An inertial world line is uniformly a elerated (with zero a eleration) and a
uniformly a elerated world line is twist-free (with onstant a eleration).
A world line is inertial if and only if it is a straight line segment.

2.4. Newtonian equation

2.4.1. We shall say some words about the Newtonian equation though it does
not belong to the subje t of this volume; the Newtonian equation motivates
the notion of for e elds and potentials whi h will make us understand the
importan e of splitting of ve tors and ove tors (see Se tion 6).
First of all we have to say something about mass. One usually introdu es
the unit of mass, kg; as a unit independent of the unit of distan es, m; and of
the unit of time periods, s: This means in our language that we introdu e the
measure line G of mass as a measure line \independent" of D and I: We shall do
so in another book where we wish to treat physi al theories in a form suitable
for appli ations, so in a form whi h applies the SI physi al dimensions. However,
for the present purposes we hoose another possibility.
The results of quantum me hani s showed that Nature establishes a relation
among the measure lines D; I and G. Namely, it is dis overed, that the values

of angular momentum are integer multiples of a given quantum denoted by h=4

where h is known as the Plan k onstant. Hen e we an hoose R for the measure
line of angular momentum; a real number (more pre isely an integer) n represents
the angular momentum value nh=4: As it is known, angular momentum is the
produ t of mass, position and velo ity; thus its measure line is G
D
DI whi h
is identied with R; onsequently, G  D
I D :
In this book, for easier theoreti al onsiderations, we take D
I D as the measure
line of masses. If m is the unit distan e and s is the unit time period then ms2
is the unit mass. One nds the experimental data
m2 kg
h=4 = (1; 05:::)10 34
s

hen e if we take it equal to the real number one we arrive at the denition
kg

:= (9; 4813:::)1033 2 :
m

2.4.2. Sin e a eleration values are elements of I

EI and \the produ t of mass

and a eleration equals the for e", the for e values are elements of D
I D
I
EI 
E
E
I
D
D  I ; moreover, \a for e an depend on time, on spa e and on velo ity".
Thus we a ept that a for e eld is a dierentiable mapping
f

: M  V(1)

 EI

and the history of the material point with mass m under the a tion of the for e
eld f is given by the Newtonian equation

mx = f (x; x_ );
i.e. the world line modelling the history is a solution of this dierential equation.

2.4.3. The most important for e elds an be derived from potentials; e.g.
the gravitational eld and the ele tromagneti eld. Usually the gravitational
eld is the gradient of a s alar potential and the ele tromagneti eld is given
by the gradient of a s alar potential and the url of a ve tor potential. The
gravitational for e a ting on a material point depends only on the spa etime
position of the masspoint, the ele tromagneti for e depends on the velo ity of
the masspoint as well. To introdu e the notion of potential in the spa etime
model, we have to rely on these fa ts. Now we give the onvenient denition and
we shall show in Se tion 6 that it is suitable indeed.

A potential is a twi e dierentiable mapping

K

:M

 M

(in other words, a potential is a twi e dierentiable ove tor eld).

The eld strength orresponding to K is D ^ K : M
M ^ M (the
antisymmetri or exterior derivative of K ; see VI.3.6).
The for e eld f has a potential (is derived from a potential) if
| there is an open subset O  M su h that Dom f = O  V(1);
| there is a potential K dened on O su h that

i
F (x)
u
(x 2 O; u 2 V(1));
where F := D ^ K and i : E ! M is the embedding. Che king this formula, the
f (x; u) =

reader an seize the opportunity to pra tise using the dot produ t.

2.5. Exer ises

1. Let r1 and r2 be world lines. Chara terize the fun tion r1 r2 :
2. Another formulation of the pre eding exer ise: give ne essary and su ient
onditions for a fun tion z that r + z be a world line for all world lines r:
3. Des ribe the world lines in the split spa etime model ( f. 2.1.4).

3. Observers

3.1. The notion of an observer

3.1.1. In usual physi s the phenomena are always des ribed with respe t to
a \referen e frame" whi h means a material obje t and oordinates on it.
Our present aim is to dene the orresponding notion in the spa etime model.
To do so, we separate the material obje t and the oordinates on it; in su h a
way rst we arrive at observers and then at referen e frames.
The notion of an observer is extremely important be ause our experimental
results are always onne ted with material obje ts (experimental devi es). It
is important as well that we see learly the onne tion between absolute and
relative notions.
Our experien e about a physi al phenomenon depends on the experimental
devi es, i.e. on material obje ts of observation. This means that our experien e
and the dire t abstra tions gained from experien e are relative: they re e t
not only the properties of the observed phenomenon but some properties of the
observers and the relation between the phenomenon and the observer as well.

If we wish to separate the properties of the phenomenon we have to ompare

the experimental results of dierent observers on erning the same phenomenon;
so we an nd out what is the ore of these fa ts and we an get rid of observers.
To des ribe a phenomenon we evidently ought to use absolute notions only, i.e.
notions independent of observers. Physi al theories must be based on absolute
notions.
On the other hand, of ourse, we must lay down as well how an observer
dedu es the relative notions from the absolute ones, whi h means how the
observer sees the properties of the phenomenon. This is indispensable from
the point of view of experiments.
As a matter of fa t, this program was started by the theory of relativity at that
time, but owing to the use of inadequate mathemati al tools and the ompli ated
setting it has not yet been a omplished. Even nowadays one expresses the
absolute notions with the aid of relative notions and not vi e versa whi h would
be desirable: to dedu e the relative notions from the absolute ones.

3.1.2. An observer as a physi al reality is a material obje t or a set of material

obje ts; re all what is said in the Introdu tion: the earth, the houses on it form
an observer, the ar is another observer.
We an imagine that an observer is a olle tion of material points existing
\in lose proximity" to ea h other. The existen e of a masspoint in spa etime is
des ribed by a world line. Thus an observer would be modelled by a olle tion
of world lines that ll \ ontinuously" a domain of spa etime. How to dene
a onvenient notion of su h a ontinuity? To all points of every world line of
the observer we assign the orresponding velo ity value; in this way we dene a
velo ity eld: a fun tion dened for some world points and having values in V(1).
Conversely, given a velo ity eld (with onvenient mathemati al properties), we
an re over the world lines of the observers: world lines having everywhere the
velo ity value pres ribed by the velo ity eld. We shall see that the velo ity
eld is extremely suitable for our purposes, hen e we prefer it to the olle tion
of world lines.
Denition. An observer is a smooth map U : M
onne ted.
If Dom U = M; the observer is alled global.

 V(1) whose domain is

We emphasize that we are dealing with mathemati al models; an observer as

it is dened is a mathemati al model for a physi al obje t. To underline this fa t
we might use the term \observer model" instead of \observer" but we wish to
avoid ponderousness. If ne essary, we shall say physi al observer for the material
obje ts in question.

3.1.3. Let U be an observer. The integral urves of the dierential equation

(x : I

 M)?

x_ = U (x)

have ex lusively timelike tangent ve tors, thus they are world lines (see 2.2).
The maximal integral urves of this dierential equation will be alled U -lines;
they will play an important role.
If the world line fun tion r is a solution of the above dierential equation
| i.e. Ran r is an integral urve of U | then r_ (t) = U (r(t)) and so r(t) =
DU (r(t))
r_ (t) = DU (r(t))
U (r(t)) for all t 2 Dom r: This motivates that
AU

:M

 I
E I ;

x 7! DU (x)
U (x)

is alled the a eleration eld orresponding to the observer U :

3.1.4. Denition. An observer U is alled t if all the world line fun tions
giving the U -lines have the same domain; this uniquely determined interval of I
is the lifetime of the observer.
It may happen that the maximal integral urves of a global observer are not
global world lines (see Exer ise 3.4). A global observer U is t if and only if all
U -lines are global.
3.1.5. In3 the arithmeti
spa etime model an observer is given by a fun tion
3
1
2
3

: RR
R in the form (1; V ) = (1; V ; V ; V ): If we denote the
partial derivatives orresponding to R and R3 by 0 and r = (1 ; 2 ; 3 );
respe tively,
then the a eleration
 eld of the observer is (0; 0 V + V
rV ) =

3
P
0; (0 V i + V k k V i )i=1; 2; 3 :
V

k=1

3.2. Splitting of spa etime due to an observer

3.2.1. As it is stated in the Introdu tion, a physi al observer | a material

obje t | establishes spa e for itself. The points of its spa e are just the material
points that it onsists of. In our model these points orrespond to the maximal
integral urves of the observer. Thus the spa e of an observer U is just the
olle tion of U -lines. Now we are in position to dene the spa e of an observer
and to establish how an observer splits spa etime into time and spa e.
Denition. Let U be an observer and let EU denote the set of maximal
integral urves of U : EU is alled the spa e of the observer U ; or the U -spa e.

The elements of the U -spa e are world lines. We have to get a ustomed to
this situation, strange at rst sight, but ommon in mathemati s: the elements
of a set are sets themselves.
A maximal integral urve of U will be alled a U -line if onsidered to be a
subset of M and will be alled a U -spa e point if onsidered to be an element of
EU :
By the way, we on eive instants, too, as sets: an instant is identied with
the orresponding simultaneous hyperplane.
We measure distan es in our physi al spa e, we know what is near, what is
far. We dene limit pro edures regarding our spa e. These notions must appear
in the model.
It an be shown that, in general, the U -spa e an be endowed with a smooth
stru ture in a natural way, thus limits, dierentiability et . will make sense.
However, in this book we avoid the general theory of smooth manifolds, that is
why, in general, we do not deal with the stru ture of observer spa es. Later the
spa es of some spe ial observers, important from the point of view of appli ations,
will be treated.

3.2.2. Re all from the theory of dierential equations that dierent integral
urves of U do not interse t (VI.6.2). Let us introdu e the map CU : Dom U !
EU in su h a way that CU (x) is the (unique) U -line passing through x:
We shall say as well that CU (x) is the U -spa e point that the world point x
is in ident with.
Then the map
HU : Dom U ! I  EU ;
x 7! ( (x); CU (x))
is learly inje tive, its inverse is
(t; q) 7! q ? t
((t; q) 2 Ran HU  I  EU )
where the notation introdu ed in 2.2 is used.
In this way spa etime points in the domain of U are represented by pairs of
time points and U -spa e points. We say that the observer U splits spa etime
into time and U -spa e with the aid of HU :
Denition. HU is the splitting of spa etime a ording to U :
If EU is endowed with the smooth stru ture mentioned previously then HU
will be smooth. Its properties will be laried in spe ial ases.
3.3. Classi ation of observers

3.3.1. We have onsidered the room and the ar as examples of physi al

observers. However, mu h \worse" material obje ts an be observers as well.

For instan e, the stormy sea: the distan e of its spa e points (whi h are the
mole ules of the water) and even the dire tion of their mutual positions vary
with time. A ship on the stormy sea is a little better be ause it does not hange
its shape, it is rigid. However, it rotates, i.e. the dire tions of relative positions
of its spa e points vary with time. The slightly waving water is better than the
stormy one be ause it does not whirl. These examples show from what point of
view we should lassify observers in our spa etime model.
We mention that physi al observers, in reality, are never rigid and rotationfree; at least mole ular motion ontradi ts these properties. Besides, a physi al
observer is never global, it annot ll all the spa etime. All these notions,
as all models, are idealizations, extrapolations for a onvenient mathemati al
des ription.
Re all the notation introdu ed in 2.2.
Denition. A t observer U is alled
(i ) rigid if for all q1 ; q2 2 EU the distan e between q1 ? t and q2 ? t| in other
words jq1 ? t q2 ? tj| is the same for all t in the lifetime of U ;
(ii ) rotation-free if for all q1 ; q2 2 EU the dire tion of the ve tor q1 ? t q2 ? t
is the same for all t in the lifetime of U ;
(iii ) twist-free if all U -spa e points are twist-free;
(iv ) inertial if U is a onstant fun tion; in other words, if the U -lines are parallel
straight line segments in spa etime.
Ex ept the inertial observers, it is di ult to give a good illustration of these
types of observers. The following gure tries to show a rigid or rotation-free
observer.
Suppose q1 runs in the plane of the sheet. Letting q2 bend below the plane
of the sheet in su h a way that its points have the same distan es from the
orresponding points of q1 ; we an draw a pi ture of a rigid observer whi h is
not rotation-free.
Letting q2 bend in the plane of the sheet we an draw a pi ture of a rotationfree observer whi h is not rigid.

3.3.2. We all attention to the fa t that a t observer whose spa e points are
all inertial (i.e. straight line segments) is not ne essarily inertial: it may o ur
that its integral urves are not parallel (see Exer ise 5.4.1).
Evidently, an inertial observer is rigid, rotation-free and twist-free. The
onverse is not true: see 5.2.
A t observer U is rigid and rotation-free if and only if for all q1 ; q2 2 EU ;
q1 ? t q2 ? t is the same for all t in the lifetime of U :
3.4. Exer ise
The observer

2
( 0 ;  1 ;  2 ;  3 ) 7! (1; ( 1 ) ; 0; 0)

in the arithmeti spa etime model is global, its maximal integral urve passing
through ( 0 ;  1 ;  2 ;  3 ) is

f(t; 0;  2 ;  3 ) j t 2 Rg
f(t; t  0 1+ 1= 1 ;  2 ;  3 ) j t >  0 1= 1g
f(t; t  0 1+ 1= 1 ;  2 ;  3 ) j t <  0 1= 1g

if

 1 = 0;

if

 1 > 0;

if

 1 > 0:

Consequently, most of the maximal integral urves of the observer are not global.

4. Rigid observers

4.1. Inertial observers

4.1.1. Let us onsider a global inertial observer U and let u 2 V(1) denote
the onstant value of U :

Re all the linear map u | the proje tion onto E along u

I | dened in
1.2.8.
The observer spa e EU is the set of straight lines dire ted by u; more losely,

CU (x) = x + u
I := fx + ut j t 2 Ig:
Note that

(x + u
I) ? t = x + u(t  (x)):

As a onsequen e, U is rigid and rotation-free:

(x2 + u
I) ? t (x1 + u
I) ? t =
= (x2 + u(t  (x2 ))) (x1 + u(t  (x1 ))) =
= x2 x1 u(
(x2 x1 )) =
= u
(x2 x1 ):

4.1.2. A ording to the previous formula, if q2 and q1 are U -spa e points

then q2 ? t q1 ? t is the same ve tor in E for all t 2 I : more losely, it equals
u
(x2 x1 ) where x1 and x2 are arbitrary elements of q1 and q2 ; respe tively.
Regarding this ve tor as the dieren e of the U -spa e points, we dene an
ane stru ture on EU in a natural way.
Proposition. EU ; endowed with the subtra tion
q2

q1 := u
(x2

x1 )

(q1 ; q2 2 EU ; x1 2 q1 ; x2 2 q2 )

is an ane spa e over E:

Observe that if x1 2 q1 ; x2 2 q2 and  (x1 ) =  (x2 ); then q2

q1 = x2

x1 :

It is worth remarking that

(x + q) + u
I = (x + u
I) + q

(x 2 M; q 2 E);

whi h is not trivial be ause here the same sign + denotes dierent operations:
the rst one refers to the addition between elements of M and M; the se ond and
the third ones denote a set addition between elements of M and M; the fourth
one indi ates the addition between elements of EU and E: This formula has the
generalization
(x + x) + u
I = (x + u
I) + u
x

(x 2 M; x 2 M):

4.1.3. The spa e of any global inertial observer is a three-dimensional oriented

Eu lidean ane spa e (over E): In this way we regain our experien e regarding
our physi al spa e from the spa etime model (see the Introdu tion).
Keep in mind that the spa e of every global inertial observer is an ane
spa e over the same ve tor spa e E: Now we see why the ve tors in E are alled
spa elike.
The following assertion is proved without any di ulty.
Proposition. Let U be a global inertial observer whose onstant value is u:
Then the splitting of spa etime a ording to U ;
M ! I  EU ;

x 7! ( (x); CU (x)) = (x + E; x + u
I)

is an orientation-preserving ane bije tion having hu = ( ; u ) as its underlying

linear map.
If we onsider the elements of I as hyperplanes in M then  (x) = x + E; we
used this fa t in the previous proposition for later purposes.

4.1.4. We have to get a ustomed to the fa t that a physi al notion whi h

seems \stru tureless", \as simple as possible" (e.g. a spa e point of an observer)
is modelled by a less simple, stru tured mathemati al obje t (by a line). In
mathemati s it is ustomary that the elements of a set are themselves sets or
fun tions.
However, we have a tool to redu e some of our mathemati al obje ts to simpler
ones. This tool is the ve torization of ane spa es: hoosing an arbitrary element
(\referen e origin") in an ane spa e, we an represent every element of the ane
spa e by a ve tor.
Let U be a global inertial observer with the velo ity value u: Taking a to 2 I
and a qo 2 EU we an establish the ve torization of time and U -spa e:
Vo : I  EU

! I  E;

(t; q) ! (t to ; q

qo )

by whi h, in parti ular, we represent U -spa e points by ve tors in E that are

simpler obje ts than straight lines in M:
Observe that hoosing to and qo is equivalent to hoosing a \spa etime"
referen e origin o 2 M : o := qo ? to ; to =  (o); qo = CU (o):

Denition. An inertial observer with origin is a pair (U ; o) where U is a

global inertial observer and o is a world point.
The ve torized splitting of spa etime orresponding to (U ; o) is the map
HU ;o := Vo HU : M ! I  E; x ! ( (x)  (o); CU (x) CU (o)) =
= (
(x o); u
(x o)) :
Note that

HU ;o = hu Oo ;

where hu = ( ; u ) and Oo is the ve torization of M with origin o : M ! M;

x 7! x o:

4.1.5. Let us onsider the arithmeti spa etime model and the global inertial observer with onstant value (1; v): The spa e point of the observer that
(; ) is in ident with is the straight line (; ) + (1; v)R = f( + t;  + vt j
t 2 Rg:
As on erns the ane stru ture of the set of su h lines we have
[(; ) + (1 + v)R

[(;  ) + (1; v)R = 

v (

 ) 2 R3 :

Let the observer in question hoose (0; 0) as referen e origin. Then the observer
spa e will be represented by R3 ; the spa e point (; ) + (1; v)R will orrespond
to the dieren e of this straight line and that passing through (0; 0)| whi h is
(1; v)R| ; this dieren e is exa tly  v:
Consequently, the ve torized splitting of spa etime due to this observer is
R

 R3 ! R  R3 ;

(; ) 7! (; 

v ):

In parti ular, the splitting of spa etime a ording to the basi observer | the
one whose value is the basi velo ity value (1; 0)| with referen e origin (0; 0)is
the identity of R  R 3 : the arithmeti spa etime model is the Cartesian produ t
of ve torized time and ve torized spa e relative to the basi observer.
In other words, the observer with referen e origin makes the orresponden e
that previously has been a epted as a natural identi ation. The ve torized
splitting of spa etime is des ribed by the formula above.

4.2. Chara terization of rigid observers

4.2.1. Now we derive some mathemati al results to hara terize some properties of observers. Simple but important relations for dedu ing our results are
the following.
Re all that CU (x) denotes the U -line passing through x: Then t 7! CU (x) ?t
is the orresponding world line fun tion. So we have
CU (x) ?  (x) = x
and

d
(C (x) ? t) = U (CU (x) ? t) :
dt U

Proposition. Let U be a t global observer.

(i ) U is rigid if and only if
(U (x + q) U (x))
q = 0

(x 2 M; q 2 E):

(ii ) U is rigid and rotation-free if and only if

U (x + q )

U (x) =

(x 2 M; q 2 E);

whi h is equivalent to the existen e of a smooth map V : I

U

= V :

 V (1) su h that

Proof. Let q1; q2 2 EU :

(i ) The fun tion

t 7! jq1 ? t q2 ? tj2
is onstant if and only if its derivative
t 7! 2 (U (q1 ? t)

U (q2 ? t))
(q1 ? t

q2 ? t)

is zero.
Putting x := q2 ? t; q := q1 ? t q2 ? t in the derivative we infer that the
derivative is zero if and only if the equality in the assertion holds (every x 2 M
is of the form q2 ? t for some q2 and t and every q 2 E is of the form q1 ? t q2 ? t
for some q1 ):
(ii ) The fun tion
t 7! q1 ? t q2 ? t
is onstant if and only if its derivative

t 7! U (q1 ? t)

U (q2 ? t)

is zero.
Reasoning as previously we get the desired result.

4.2.2. Let U be a global rigid observer. For to; t 2 I let us dene

RU (t; to ) : E ! E;
q 7! CU (xo + q ) ? t
CU (xo ) ? t;
where xo is an arbitrary element of to (i.e. x 2 M and  (xo ) = to ):
Proposition. If U is a global rigid observer then RU (t; to) is a rotation in E
(a linear orthogonal map with determinant 1) for all to ; t 2 I: Moreover, RU (t; to )
is independent of xo appearing in its denition.
The global rigid observer U is rotation-free if and only if RU (t; to ) = idE for
all to ; t 2 I:
Proof. Evidently,
RU (t; to )(0) = 0:
Moreover, sin e U is rigid, for all q1 ; q2 2 E we have

j RU (t; to )(q1 ) RU (t; to )(q2 ) j =

=j (CU (xo + q1 ) ? t CU (xo ) ? t) (CU (xo + q2 ) ? t CU (xo ) ? t) j =
=j CU (xo + q1 ) ? t CU (xo + q2 ) ? t j = j (xo + q1 ) (xo + q2 ) j =
=j q1 q2 j:

As a onsequen e, RU (t; to ) is a linear orthogonal map (see V.3.7).

For xed to and xed q 2 E; the fun tion I ! E; t 7! RU (t; to )
q is smooth
sin e it is the dieren e of two solutions of the smooth dierential equation
x_ = U (x): Consequently, t 7! detRU (t; to ) is a smooth fun tion. Sin e the
determinants in question an be 1 or 1 only and

RU (to ; to ) = idE ;

all the determinants in question equal 1.

If yo is another element of to ; then with the notation qo := yo xo 2 E
we dedu e
CU (yo + q) ? t CU (yo ) ? t = CU (xo + qo + q) ? t CU (xo + qo ) ? t =
= (CU (xo + qo + q) ? t CU (xo ) ? t) (CU (xo + qo ) ? t CU (xo ? t) =
=RU (t; to )
(qo + q) RU (t; to )
qo = RU (t; to )
q;
whi h means that the denition of RU (t; to ) is independent of xo :

4.2.3. Proposition. For all to; t; s 2 I we have

(i ) RU (to ; to ) = idE ;
(ii ) RU (t; to ) 1 = RU (to ; t);
(iii ) RU (t; to ) = RU (t; s)
RU (s; to ):
Proof. (i ) is trivial.
The dening formula of RU (t; to ) an be rewritten in the following form: if
q; qo are U -spa e points then
q ? t qo ? t = RU (t; to )
(q ? to qo ? to )
(to ; t 2 I): ()
Inter hanging t and to we get
q ? to

qo ? to = RU (to ; t)
(q ? t qo ? t)

from whi h we infer (ii ).

In a similar way we obtain (iii.)
Observe that () implies that if RU (t; to ) is known for a to and for all t then
every U -spa e point q an be dedu ed from an arbitrarily hosen qo :

4.2.4. Let U be a global rigid observer. For xed to 2 I; the fun tion
I ! E
E; t 7! RU (t; to ) is smooth (be ause for all q 2 E; t 7! RU (t; to )
q is
smooth); we introdu e
dR (t; t ) E
E
(t; to 2 I);
R_ U (t; to ) := U o 2
dt
I
whi h an be regarded as a linear map
E q 7! d RU (t; to )
q
R_ U (t; to ) : E ! ;
I
dt
(VI.3.11). We dedu e from the dening formula of RU (t; to ) that
R_ U (t; to )
q = U (CU (xo + q) ? t) U (CU (xo ) ? t) =
= U (CU (xo ) ? t + RU (to ; t)
q) U (CU (xo ) ? t) =
= U (q ? t + RU (to ; t)
q) U (q ? t);

where xo is an arbitrary element of to and q is an arbitrary element of EU :

Substituting RU (t; to ) 1
q for q and introdu ing the linear map

U (t) := R_ U (t; to )
RU (t; to ) 1 : E !

for t 2 I; we obtain

E
I

(t 2 I; q 2 E):

U (t)
q = U (q ? t + q) U (q ? t)

We know that
U (t) is antisymmetri (see 11.1.10). Sin e q ? t an be an
arbitrary world point, we have proved:
Proposition. If U is a global rigid observer then
U (t) is an antisymmetri
linear map for all t 2 I; it is independent of to appearing in its denition.
Moreover,
U (x + q )

U (x) =
U ( (x))
q

(x 2 M; q 2 E): ()

The global rigid observer U is rotation-free if and only if

U (t) = 0 for all
t 2 I:
Noti e that the restri tion of U to an arbitrary simultaneous hyperplane t is
an ane map whose underlying linear map is
U (t):

U (t) an be interpreted as the angular velo ity of the observer at the instant
t (see 11.1.10).

4.2.5. For arbitrarily xed to 2 I; the fun tion t 7! RU (t; to ) denes the
fun tion t 7!
U (t) a ording to the pre eding paragraph. Conversely, if the
fun tion t 7!
U (t) is known, then t 7! RU (t; to ) is determined as the unique
solution of the dierential equation
(X : I ! E
E)?
X_ =
U
X
with the initial ondition

X (to ) = idE :

4.2.6. We see from the formula () of 4.2.4 that the rigid observer U is
ompletely determined by an arbitrarily hosen U -spa e point qo and by the
angular velo ity of the observer, i.e. by the fun tion t 7!
U (t): Indeed, putting
q := qo ?  (x) x in that formula we obtain
U (x) = U (qo ?  (x)) +
U ( (x))
(x

qo ?  (x))

(x 2 M)

and we know that the values of U on qo oin ide with the derivative of the world
line fun tion t 7! qo ? t:

4.3. About the spa es of rigid observers

4.3.1. Proposition. Let U be a t global observer. U is rigid and rotation-

free if and only if EU ; equipped with the subtra tion

q1

q2 := q1 ? t q2 ? t

(q1 ; q2 2 EU ; t 2 I)

is an ane spa e over E:

Proof. If U is rigid and rotation-free then, for all q1; q2 2 EU ,
q1 ? t q2 ? t is the same element of E for all t 2 I: It is not hard to see that the
subtra tion in the assertion satises the requirements listed in the denition of
ane spa es.
Conversely, if EU is an ane spa e over E with the given subtra tion then,
in parti ular, q1 ? t q2 ? t is independent of t for all q1 ; q2 2 EU ; hen e U is
rigid and rotation-free.

4.3.2. If U is a global rigid and rotation-free observer, then EU is an ane

spa e, thus the dierentiability of the splitting of spa etime a ording to U
makes sense.
Proposition. Let U be a global rigid and rotation-free observer. Then the
splitting
HU : M ! I  EU ;
x 7! ( (x); CU (x))
is a smooth bije tion,


DHU (x) =  ; U (x)
(x 2 M);
and the inverse of HU is smooth as well.
Proof. For x 2 M and t 2 I we have CU (x) ?t = x + U (x)(t  (x))+ordo(t
 (x)) (VI.3.3). Thus for all y; x 2 M (see Exer ise 4.5.1),
CU (y) CU (x) = y CU (x) ?  (y) =
= y x + U (x)( (y)  (x)) + ordo( (y)  (x))
and so

HU (y) HU (x) = ( (y)  (x); CU (y) CU (x)) =

= 
(y x); U (x)
(y x) + ordo( (y x)):
Hen e HU is dierentiable, its derivative is the one given in the proposition.
As a onsequen e, we see that HU is smooth; its inverse is smooth by the inverse
mapping theorem.

4.3.3. The spa e of a rigid and rotation-free global observer, endowed with a
natural subtra tion, is an ane spa e over E: The spa e of another observer is

not ane spa e with that subtra tion (in fa t that subtra tion makes no sense for
other observers). This does not mean that the spa e of other observers annot
be endowed with an ane stru ture in some other way.
Let us onsider a t global observer U : For every instant t we an dene the
instantaneous ane stru ture on EU by the subtra tion q1 q2 := q1 ?t q2 ?t: In
general, dierent instants determine dierent instantaneous ane stru tures and
all instants have the same \right" for establishing an ane stru ture on the U spa e. There is no natural way to sele t an instant and to use the orresponding
instantaneous ane stru ture as the ane stru ture of EU :
Nevertheless, we an dene a natural ane stru ture on the spa es of rigid
global observers.

4.3.4. Though the earth rotates, we experien e on it an ane stru ture

independent of time. A sti k on the earth represents a ve tor. Evidently, the sti k
rotates together with the earth. The sti k will be represented in the following
reasoning by two points (the extremities of the sti k) in the observer spa e. Now
we wish to dene that two points in the spa e of a rigid observer determine a
ve tor (rotating together with the observer).
Let U be a rigid global observer. If q1 and q2 are points in the observer spa e
EU then for all s; s0 2 I
q1 ? s q2 ? s = RU (s; s0 )
(q1 ? s0 q2 ? s0 ):
Let us introdu e

EU := f : I ! Ej

is smooth, (s) = RU (s; s0 )
(s0 ) for all s; s0 2 Ig:

It is a routine to he k that EU ; endowed with the usual pointwise addition

and pointwise multipli ation by real numbers, is a ve tor spa e; it is threedimensional, be ause EU ! E;  7! (s) is a linear bije tion for arbitrary
s 2 I (whi h means in parti ular, that the fun tion  is ompletely determined
by a single one of its values). Moreover, if  and are elements of EU ; then
(s)
(s) is the same for all instants s; thus

EU  EU ! D
D;

(; ) 7! 

: = (s)
(s)

is a positive denite symmetri bilinear map whi h turns EU into a Eu lidean

ve tor spa e.
Now it is quite evident that EU ; endowed with the subtra tion

q1

q2 := (I ! E; s 7! (q1 ? s q2 ? s))

will be an ane spa e over EU : In other words, the dieren e of two U -spa e
points is exa tly the dieren e of the orresponding world line fun tions, as the
dieren e of fun tions is dened.

If U is rotation-free, then EU onsists of the onstant fun tions from I into

E whi h an be identied with E: So we get ba k our previous result that the
spa e of a global rigid and rotation-free observer is an ane spa e over E in a
natural way.
If U is not rotation-free then EU is a three-dimensional Eu lidean ane spa e
in a natural way, but the underlying ve tor spa e is not E; in fa t the underlying
ve tor spa e EU depends on the observer itself.
4.3.5. The spa e of a global rigid observer is an ane spa e, thus the

dierentiability of the splitting of spa etime a ording to the observer makes

sense. This question, redu ed to a simpler ane stru ture, will be studied in the
next se tion.

4.4. Observers with origin

4.4.1. The ve torization of observer spa es simplies some formulae for

inertial observers and it will be a powerful tool for non-inertial rigid observers.
Let U be a global rigid and rotation-free observer. Choosing an instant to
and a U -spa e point qo ; we give the orresponding ve torization of time and
U -spa e:
Vo : I  EU ! I  E;
(t; q) 7! (t to ; q qo ):
We see that in this way U -spa e points ( urves in M) are represented by spa elike
ve tors (points in E):
Noti e that hoosing to and qo is equivalent to hoosing a \spa etime referen e
origin" o 2 M : o := qo ? to; to =  (o); qo = CU (o): That is why we have used the
symbol Vo for the ve torization whi h an be written in the following form, too:
Vo : I  EU

! I  E;

(t; q; ) 7! (t  (o); q ?  (o) o) ;

sin e q qo = q ? t qo ? t for all t 2 I; in parti ular for t :=  (o):

If U is not rotation-free, the result of a similar ve torization

Vo : I  EU

! I  EU ;

(t; q) 7! (t to ; q

qo )

is not simple enough be ause the elements of EU are fun tions. That is why we
make a further step by the linear bije tion

L o : EU

! E;

 7! (to ):

Sin e Lo
(q qo ) = (q qo )(to ) = q?to qo ?to ; we get the double ve torization

of time and U -spa e:

Wo := (idI  Lo ) Vo : I  EU

! I  E;

(t; q) 7! (t  (o); q ?  (o) o) ;

whi h oin ides formally with the ve torization of time and spa e of a rigid and
rotation-free observer.

4.4.2. Denition. An observer with referen e origin is a pair (U ; o) where

is a global rigid observer and o is a world point.

If U is rotation-free, the ve torized splitting of spa etime orresponding to
(U ; o) is the map

HU ;o := Vo HU : M ! I  E;

x 7! ( (x)  (o); CU (x) CU (o))

= ( (x)  (o); CU (x) ?  (o) o) ;

and if U is not rotation-free then the double ve torized splitting of spa etime is
the map

HU ;o := Wo HU : M ! I  E;

x 7! ( (x)  (o); CU (x) ?  (o) o) :

4.4.3. Proposition. Let(U ; o) be an observer with referen e origin.

If U is rotation-free then the ve torized splitting is a smooth bije tion whose

inverse is smooth as well and
DHU ;o(x) = ( ; U (x) )

(x 2 M):

If U is not rotation-free, the double ve torized splitting is a smooth bije tion

whose inverse is smooth as well and


DHU ;o(x) =  ; RU ( (x); to ) 1
U (x)

(x 2 M)

where to :=  (o):
Proof. For rotation-free observers the assertion is trivial be ause of 4.3.2 and
be ause the derivative of Vo is the identity of I  E:
For the double ve torization we argue as follows: the map M ! E; x 7!
CU (x) ? to o = RU ( (x); to ) 1
(x CU (o) ?  (x) ) (see formula () in 4.2.3.)
is learly dierentiable, its derivative is the linear map (see Exer ise 11.2.2)

M ! E;

x 7!

RU ( (x); to ) 1

U ( (x))
(x CU (o) ?  (x)) 
x+
+ RU ( (x); to ) 1
(x U (CU (o) ?  (x))
x) =
= RU ( (x); to ) 1
U (x)
x:

Sin e Wo is an ane bije tion, it follows that the splitting HU : M ! I  EU

is smooth and has a smooth inverse as well ( f. 4.3.5.).

4.4.4. Dealing with observers in the arithmeti spa etime model it is extremely onvenient to onsider observers with referen e origin where the referen e origin oin ides with the origin (0; 0) of R  R3 : Namely, in this ase the
(double) ve torized observer spa es are R3 and the (double) ve torized splitting is a linear map R  R3 ! R  R3 whose zeroth omponent is the zeroth
proje tion.
4.5. Exer ises
1. If EU is an ane spa e over E with the subtra tion given in 4.3.1, then
CU (x + q) =CU (x) + q;
CU (y) CU (x) =CU (y) ?  (x) x =
=y CU (x) ?  (y):
for all x; y 2 M; q 2 E.
2. Prove that

U (t) = R_ U (t; t)

(t 2 I)

(see 4.2.4).
3. We know that the derivative at a point of a double ve torization is of the
form ( ; R 1
u) : M ! I  E where u 2 V(1) and R is an orthogonal map
E ! E; i.e. R> = R 1 (see 4.4.3). Re all that theE adjointR> is identied
with the transpose R due to the identi ation D
D  E : Thus we have
R  R> = R 1 and so (i
R) = R 1
i : Prove that
( ; R 1
u ) 1 = (u; R 1
i):
4. Let U be a global rigid observer. Using Proposition 4.2.1. prove that
DU (x)jE is antisymmetri for all x 2 M (whi h is proved in 4.2.4 in another
way).
5. Let U be a t global observer. Demonstrate that U is rotation-free if and
only if there is a smooth map  : I  M  E ! R su h that
(i) CU (x + q) ? t CU (x) ? t = (t; x; q)q
(t 2 I; x 2 M; q 2 E);
(ii) ( (x); x; q) = 1
(x 2 M; q 2 E);
(iii) (t; x; 0) = 1
(t 2 I; x 2 M):
6. Using the previous result prove that if U is a global rigid and rotation-free
observer then there is a smooth map  : M ! RI su h that DU (x)jE =  (x)idE
for all x 2 M:

5. Some spe ial observers

5.1. Why the inertial observers are better than the others
5.1.1. We know that the spa e of a rigid and rotation-free global observer,
even if it is not inertial, is an ane spa e over E: However, the splitting of

spa etime a ording to non-inertial observers is not ane.

Proposition. Let U be a rigid and rotation-free global observer. The
splitting of spa etime a ording to U ;
HU : M ! I  EU ;
x 7! ( (x); CU (x))
is an ane map if and only if U is inertial.
Proof. We have seen that if U is inertial then HU is ane.
We know that HU is dierentiable, DHU (x) = ( ; U (x)) (see 4.3.2). If HU
is ane, then DHU (x) is the same for all x 2 M: This means that U (x) does
not depend on x whi h implies that U is a onstant map as well.

5.1.2. We an say that if EU is ane but U is not inertial then the ane
stru tures of M and I  EU | though they are mathemati ally isomorphi | are
not related from a physi al point of view.
If EU is ane, then (I  EU ; I; prI ; D; b) is a non-relativisti spa etime model
and so it is isomorphi to the spa etime model (M; I; ; D; b); however, the
physi ally meaningful triplet (HU ; idI ; idD ) is an isomorphism between them if
and only if U is inertial.
This shows that global inertial observers play an important role in appli ations. Let U be a global inertial observer and suppose an assertion is formulated
for some obje ts related to I  EU ; then the assertion on erns an absolute fa t
if it uses only the ane stru ture of I  EU : The assertion has not ne essarily an
absolute ontent if it uses other properties of I  EU ; for instan e, the Cartesian
produ t stru ture or the ane stru ture of EU alone.
5.2. Uniformly a elerated observer
5.2.1. The rigid global observer U is alled uniformly a elerated if its
a eleration eld is a non-zero onstant, i.e. there is a 0 6= a 2 I
EI su h that
AU (x) := DU (x)
U (x) = a:
2
Equivalently, for all U -spa e points q; ddt2 (q ? t) = a (t 2 I):

(x 2 M):

We have for all x 2 M and t 2 I that

CU (x) ? t = x + U (x) (t  (x)) +

and

1
a(t  (x))2
2

d
(C (x) ? t) = U (x) + a (t  (x)) :
dt U
Now it follows that for all x 2 M; q 2 E and t 2 I

()

U (CU (x) ? t) =

CU (x + q) ? t CU (x) ? t = q + (U (x + q)

U (x)) (t

 (x)) :

Sin e U is rigid, the length of this ve tor is independent of t; so it equals the

length of q: Then assertion (i ) in proposition 4.2.1 implies that
U (x + q )

U (x) =

(x 2 M; q 2 E)

whi h means, a ording to the quoted proposition, that U is rotation-free.

5.2.2. U is onstant on the simultaneous hyperplanes. Thus

 (y)) = U (y) for all x; y 2 M and we infer from () that
U (y ) = U (x) + a ( (y

x))

U (CU (x)?

(x; y 2 M):

As a orollary, the uniformly a elerated observer U is uniquely determined

by a single value of U at an arbitrary world point and by the onstant value of
the a eleration eld of U :
We see as well that the uniformly a elerated observer is an ane map from
M into V(1) whose underlying linear map is a
 :

5.2.3. Let the previous observer hoose a referen e origin o: Then

CU (x) CU (o) =x CU (o) ?  (x) =
=x o + U (o) ( (x)  (o))

1
a( (x)  (o))2 :
2

As a onsequen e, the ve torized splitting of spa etime is



1
a(
(x o))2 :
2

M ! I  E; x 7! 
(x o); U (o)
(x o)

5.2.4. For  > 0; the observer

( 0 ;  1 ;  2 ;  3 ) 7! (1;  0 ; 0; 0)
in the arithmeti spa etime model is uniformly a elerated. Its maximal integral
urve passing through ( 0 ;  1 ;  2 ;  3 ) is


1
t;  1 +  0 (t  0 ) + (t  0 )2 ;  2 ;  3 j t 2 R =
2



1
1 02 2 3
= t;  1 + t2
( ) ;  ;  j t 2 R :
2
2

A ordingly, if the observer hooses (0; 0) as a referen e origin then the

ve torized splitting be omes
R  R3

! R  R3 ;

( 0 ;  1 ;  2 ;  3 ) 7!  0 ;  1

1 02 2 3
( ) ;  ;  :
2

5.3. Uniformly rotating observer

5.3.1. The global observer U is alled uniformly rotating if there is a nonE );

zero antisymmetri linear map
: E ! EI (in other words,
2 N^I N ; N := D
alled the angular velo ity, su h that
U (x + q )

U (x) =

q

(x 2 M; q 2 E):

Proposition 4.2.1 (i ) implies that U is rigid. Moreover, we easily obtain that

RU (t; to ) = e(t to )

(to ; t 2 I);

be ause this is the (ne essarily unique) solution of the initial value problem given
in 4.2.5.
Consequently, 3.4.3 yields that if o; x 2 M;  (o) = to and t 2 I then

CU (x) ? t = CU (o) ? t + e(t to )

(CU (x) ? to

o) :

()

Every U -line is obtained from a given one and from

: This formula be omes
simpler if we onsider x 2 to :

CU (x) ? t = CU (o) ? t + e(t to )

(x o):
U

itself is determined by its values on a given U -line qo and by

(4.2.6):
U (x) = U (qo ?  (x)) +

(x

(x 2 M):

qo ?  (x))

5.3.2. Reformulating the previous result we an say that a uniformly rotating

observer an be given by the history of a point of the observer (by a spa e
point of the observer) and by its angular velo ity. If we deal with a uniformly
rotating observer then we are to look for its \best" spa e points to have a simple
des ription of the observer. Even if the observer is given by one of its spa e
points and by its angular velo ity, it may happen that we nd a \better" spa e
point than the given one.
Now we shall examine a uniformly rotating observer U that has an inertial
spa epoint. Then there is an o 2 M and a 2 V(1) su h that qo := o +
I is
a U -line. U equals on qo ; thus
U (x) = +


(x

(x 2 M):

o)

We see that U is an ane map, the underlying linear map is

 whose
range oin ides with the range of
whi h is a two- dimensional linear subspa e
in EI :
We know that the kernel of
is one-dimensional and orthogonal to Ran

(see V.3.9). If e 2 Ker

; then U (o + e + t) = for all t 2 I; i.e. U is onstant
on the inertial world line o + e +
I as well. Thus it is a maximal integral
urve of U ; parallel to qo : It is an easy task to show that

fx 2 M j

U (x) = g = o + Ker

+
I:

The observer has the a eleration eld

AU (x) =DU (x)
U (x) =


U (x) =

(U (x)
=




(x o)

) =

(x 2 M):

Sin e Ker (
2 ) = Ker
(Exer ise V.3.20.2), the set of a eleration-free world
points is fx 2 M j 
(x o) 2 Ker
g whi h equals o + Ker
+
I:
Thus for all e 2 Ker
; o + e +
I is an inertial U -spa e point and there
are no other inertial U -spa e points. The inertial U -spa e points orresponding
to dierent elements of Ker
are dierent. The set

fo + e +
I j e 2 Ker
g

in EU is alled the axis of rotation.

5.3.3. The axis of rotation makes sense for arbitrary uniformly rotating
observers (see Exer ise 5.4.4).
The earth an be modelled by a uniformly rotating observer. Note that the
angle between the axis of rotation and the dire tion of progression makes no
absolute sense. The dire tion of progression is the dire tion of the relative
velo ity with respe t to the Sun. The axis of rotation (Ker
; an oriented onedimensional linear subspa e in E) and a relative velo ity value (an element of EI
as we shall see in Se tion 6.2) make an angle; however, Ker
and an absolute
velo ity value ( in the former treatment) form no angle.
5.3.4. Let the previous observer hoose o as a referen e origin. Then formula
() in 5.3.1 yields that
CU (x) ? to o = e ( (x)  (o))

(x (o + ( (x)  (o))) ;
thus the double ve torized splitting of spa etime be omes
M ! I  E;

x 7!

(x o); e


(x o)



(x

o) :

5.3.5. For ! > 0; the observer

( 0 ;  1 ;  2 ;  3 ) 7! (1; ! 2; ! 1 ; 0)

in the arithmeti spa etime model is uniformly rotating. Its maximal integral
urve passing through ( 0 ;  1 ;  2 ;  3 ) is


t; 1 os !(t  0 )  2 sin !(t  0 );

 1 sin !(t  0 ) +  2 os !(t  0 );  3 j t 2 R) :

If the observer hooses (0; 0) as a referen e origin, the double ve torized

splitting will be

 R3 ! R  R3 ;


( 0 ;  1 ;  2 ;  3 ) 7!  0 ;  1 os ! 0 +  2 sin ! 0 ;  1 sin ! 0 +  2 os ! 0 ;  3 :

5.4. Exer ises

1. Let U be a global observer. Demonstrate that the following assertions are
equivalent:
(i) the a eleration eld of U is zero,
(ii) all the integral urves of U are straight lines.
Su h an observer need not be inertial. Consider the observer
( 0 ;  1 ;  2 ;  3 ) 7! (1; 0;  1 ; 0)
in the arithmeti spa etime model. Give its maximal integral urves. Show that
the observer is not rigid.
2. Let U be a global observer. Demonstrate that the following assertions are
equivalent:
(i) the a eleration eld of U is a non-zero onstant,
(ii) all the integral urves of U are uniformly a elerated with the same nonzero a eleration.
Su h an observer need not be uniformly a elerated. Consider the observer
( 0 ;  1 ;  2 ;  3 ) 7! (1; 0;  0 +  1 ; 0)
in the arithmeti spa etime model.
3. Prove that a global rigid observer whose integral urves are straight lines
is inertial.
4. Dene the axis of rotation for an arbitrary uniformly rotating observer.
5. Find the axis of rotation of the observer given in 4.3.5.
6. Sin e M and V(1) are ane spa es, it makes sense that a global observer
U : M ! V(1) is ane; let DU : M ! EI be the underlying linear map (the
derivative of U at every point equals the linear map under U ): The restri tion
of DU onto E will be denoted by
U ; it is a linear map from E into EI : Prove
that for all x 2 M the world line fun tion
1
I ! M; t 7! x+U (x)(t  (x)) + DU
U (x)(t  (x))2 +
2
1 1
X
((t  (x))
U )n 2
DU
U (x)(t  (x))2
+
n
!
n=3
gives the maximal integral urve passing through x.
7. Let U be an ane observer. Then
U := D U jE : E ! EI is a linear map.
Prove that
(i )

CU (x + q) = CU (x) + q

if and only if q 2 Ker

U ;

(x 2 M; q 2 E)

(ii ) U is rigid if and only if

U is antisymmetri .
8. Let U be a rigid ane observer. Then, a ording to the previous exer ise,

U is antisymmetri .
We distinguish four ases:
(i )
U =
6 0; DU
u 6= 0

(ii )
U =
6 0; D U
= 0

(iii )
U = 0; DU
u 6= 0
(iv )
U = 0; DU
= 0

for all u 2 V(1);

for some 2 V(1);
for all u 2 V(1);

for some 2 V(1) (i.e. DU = 0):

Demonstrate that
(iv ) is an inertial observer,
(iii ) is a uniformly a elerated observer,
(ii ) is a uniformly rotating observer having an inertial spa e point,
(i ) is a uniformly rotating observer having a uniformly a elerated spa e point.
(Hint: the kernel of
U 6= 0 is one-dimensional, U and DU are surje tions.
Hen e there is a 2 V(1) su h that a := DU
is in the kernel of
U :
Consequently, there is a world point o su h that for all world points x
U (x) = U (o) + DU

(x o) = +
U

(x o) + a
(x o)

and so the observer has the a eleration eld

AU (x) = a +
U

U

(x o)) :

9. Take an o 2 M and dene the observer

x o
U (x) :=

(x o)

(x 2 o + T ! ):

Prove that
(i ) every U -spa e point is inertial, more losely,

CU (x) ? t = o +

x o
(t  (o))

(x o)

(x 2 Dom U ; t >  (o));

(ii ) the a eleration eld orresponding to U is zero whi h follows from

DU (x) =

U (x)

(x o) ;

(iii ) U is not rigid; the distan e between two U -spa e points in reases as time
passes.

10. Take an o 2 M; a u 2 V(1); an s 2 I+ and dene the observer

U (x) := u +

u
(x
s

o)

(x 2 M):

Demonstrate that
(i ) U is an ane observer, more losely

DU (x) = u
s

for all

x 2 M;

(ii ) the a eleration eld orresponding to U is


(x o)
x 7! u 2
=
s

U (x)
s

(iii ) CU (o + q) ? t = o + u(t  (o)) + e(t  (o))=sq

(q 2 E):
(iv ) U is rotation-free and is not rigid: the distan e between two U -spa e points
in reases with time.

6. Kinemati s

6.1. The history of a masspoint is observed as a motion

6.1.1. The motion of a material point relative to an observer is des ribed by

a fun tion assigning to an instant the spa e point where the material point is at
that instant.
Now we are able to give how an observer determines the motion from the
history of a material point.
Denition. Let U be a t observer and let r be a world line fun tion,
Ran r  Dom U : Then

rU : I

 EU ;

t 7! CU (r(t))

is alled the motion relative to U ; or the U -motion, orresponding to the world

line fun tion r:

6.1.2. If U is a global rigid observer, then EU is an ane spa e thus the

dierentiability of rU makes sense and rU is pie ewise twi e dierentiable.
Given a rigid and rotation-free global observer U and a motion relative to
U ; i.e. a pie ewise twi e dierentiable fun tion m : I
EU ; we an regain the
history, i.e. the world line fun tion r for whi h rU = m holds. Indeed, for every

t; m(t) is a U -spa e point, i.e. a maximal integral urve of U ; then r(t) will be
the unique element in t \ m(t): In other words, using the splitting HU we have
r(t) = HU 1 (t; m(t)) = m(t) ? t:

Similar onsiderations an be made for a general global rigid observer.

6.1.3. Let us onsider the arithmeti spa etime model. As we know

(see
2.1.4), a world line fun tion r in it is given by a fun tion r : R
R3 in the
form r(t) = (t; r(t)): Paragraph 4.1.5 shows that r is the orresponding motion
relative to the basi observer. We see that the history is regained very simply
from the motion (in view of the previous onsiderations it is a onsequen e of
the fa t that for the basi observer (1; 0); H(1;0) is the identity of R  R3 ):
Thus if r : R R3 des ribes the motion relative to the basi observer then

(t 2 Dom r)

r(t) = (t; r(t))

is the orresponding world line fun tion.

6.2. Relative velo ities

6.2.1. Proposition. Let U be a global rigid and rotation-free observer; if

the world line fun tion r is twi e dierentiable then rU is twi e dierentiable as
well and
rU_ (t) = r_ (t)

U (r(t));

rU (t) = r(t)

AU (r(t)):

Proof. Taking into a ount the relations

CU (r(s)) CU (r(t)) =CU (r(s)) ? s CU (r(t)) ? s =
=r(s) r(t) [CU (r(t)) ? s CU (r(t)) ? t

we dedu e

CU (r(s)) CU (r(t))
rU_ (t) = slim
= r_ (t) U (r(t));
!t
s t
from whi h rU (t) = r(t) DU (r(t))
r_(t) follows immediately. Sin e now
U = V  (see 4.2.1 (ii )), we have DU = (DV  )
 and AU = DV  ;
thus the equality regarding the relative a eleration is veried.
The rst and the se ond derivative of rU is a epted as the relative velo ity
and the relative a eleration of r with respe t to the global rigid and rotation-free
observer U ; respe tively.

6.2.2. The pre eding result motivates the following denition.

Denition. Let u and u0 be elements of V(1). Then
vu0 u

:= u0

is alled the relative velo ity of u0 with respe t to u:

Proposition.E Suppose u; u0 and u00 are elements of V(1): Then
(i ) vu0 u is in I ;
(ii ) vu0 u = vuu0 ;
(iii ) vu00 u = vu00 u0 + vu0 u :
These relations are very simple and they are in a ordan e with our everyday
experien e:
(i ) the relative velo ity values form a three-dimensional Eu lidean ve tor spa e,
the length of a relative velo ity is in DI ;
(ii ) if a body moves with a given relative velo ity with respe t to another body
then the se ond body moves relative to the rst one with the opposite velo ity.
(iii ) the sum of relative velo ity values in a given order yields the resultant
relative velo ity value.

6.2.3. Let us imagine that a ar is going on a straight road and it is raining.

The raindrops hit the road and the ar at dierent angles. What is the relation
between the two angles?
Let u and u0 be two dierent elements of V(1) (the absolute velo ity values
of the road and of the ar, respe tively). If w is an element of V(1), too, w 6= u;
w 6= u0 (the absolute velo ity value of the raindrops),
v
v 0
v 0
( v 0)
(w) := ar os wu u0 u ;
0 (w) := ar os wu 0 uu0
jvwu jjvu u j

jvwu jjvuu j

are the angle formed by the relative velo ity values vwu and vu0 u and the angle
formed by the relative velo ity values vwu0 and vu0 u = vuu0 ; respe tively (the
angles at whi h the raindrops hit the road and the ar, respe tively).

A simple al ulation yields that

jv 0 j
jv 0 j
os (w) = wu os 0 (w) + u u :
jvwu j

jvwu j

We all attention to an interesting limit ase. Suppose u and u0 are xed and
w tends to innity, i.e. it varies in su h a way that jvwu j tends to innity; then
jvwu0 j tends to innity as well and the quotient of these quantities tends to the
number 1:
0
lim os (w) = wlim
w!1
!1 os  (w);
0
whi h implies wlim
!1 (w) = wlim
!1  (w):
Roughly speaking, the raindrops arriving with an \innitely big" relative
velo ity hit the road and the ar at the same angle. Repla ing \raindrops with
innitely big relative velo ity" by a \light beam" we get that non-relativisti ally
there is no aberration of light: a light beam forms the same angle with the road
and the ar moving on the road.
We have spoken intuitively; of ourse the question arises at on e: what
is the model of a light beam in the non-relativisti spa etime model? What
mathemati al obje t in the non-relativisti spa etime model will orrespond to
a light beam? We shall see that none. A light beam annot be modelled in the
present spa etime model.

6.2.4. We an obtain the results of 6.2.1 by hoosing a referen e origin o in

M; too, for the global rigid and rotation-free observer U : Let us put to :=  (o);
qo := CU (o): Evidently, the derivative of the ve torized motion
rU

:I

E;

t 7! rU (t) qo

equals the derivative of rU : Sin e

rU (t) qo =CU (r(t)) qo = CU (r(t)) ? t qo ? t =

=r(t) qo ? t;
we get immediately

rU_ (t) = r_ (t)

U (qo ? t) = r_ (t)

U (r(t));

(t 2 I);

be ause U is onstant on the simultaneous hyperplanes.

We mention, that in pra ti e it is more onvenient to use the ve torized motion
in su h a form that time is ve torized, too:

IE;

t 7! rU (to + t)

qo = r(to + t) qo ? (to + t):

6.3. Motions relative to a rigid observer

6.3.1. Re all that the spa e EU of a rigid global observer U is an ane spa e
over EU onsisting of fun tions I ! E whose values \rotate together with the
observer". Thus, in general, it is somewhat ompli ated to ontrol the ane
stru ture based on these ve tors; we an simplify the al ulations by performing
a double ve torization of the observer spa e, orresponding to a hosen referen e
origin o in M: Let to :=  (o); qo := CU (o) and let Lo : EU ! E be the linear
bije tion introdu ed in 4.4.1.
Let us take the motion rU orresponding to the world line fun tion r and let
us onsider the double ve torized motion
rU

:I

E;

t 7! Lo
(rU (t) qo ) =CU (r(t)) ? to o =

=RU (t; to ) 1
(r(t) qo ? t) :

For the sake of simpli ity, we shall use the notations r(t) := rU (t); R(t) :=
RU (t; to );
(t) :=
U (t); uo (t) := U (qo ? t); ao (t) := AU (qo ? t):
Then the previous formula an be written in the form

R(t)
r(t) = r(t) qo ? t;

dierentiating with respe t to t and then omitting t from the notation we obtain

R_
r + R
r_ = r_
yielding

R
r_ =

R
r + r_
A se ond dierentiation gives

uo

()

uo :

R_
r_ + R
r =
_
R
r

R_
r

R
r_ + r

ao

from whi h we infer

R
r = 2

R
r_



R
r
_
R
r + r

ao :

6.3.2. Let us introdu e the notation

!(t) := R(t) 1

(t)
R(t) = R(t) 1
R_ (t)

(t 2 I):

_ whi h implies
From R
! =

R we derive that R_
! + R
!_ =
_
R +

R;

R
! + R
!_ =
_
R +



R; then we an state that

!_ = R 1

_
R:

Consequently, the last formula in the pre eding paragraph an be written in the
form
r = 2!
r_ !
!
r !_
r + R 1 (
r ao ):

2!
r_ and !
!
r are alled the entripetal a eleration and the Coriolis

a eleration with respe t to the observer.

6.3.3. Re all that r : I E denotes the double ve torized motion: r(t) =

Lo
(rU (t) qo ); onsequently, the relative velo ity value at the instant t;
rU_ (t) = Lo 1
r_ (t) is in EU ; i.e. it is a fun tion from I into E whi h is uniquely
determined by an arbitrary one of its values:
rU_ (t)(s) = RU (s; to )
r_ (t)
Sin e
(t)
(r(t) qo ? t) = U (r(t))

U (qo ? t);

rU_ (t)(t) = r_ (t)

(s 2 I):

the formula () in 6.3.1 gives

U (r(t)):

The expression on the right-hand side oin ides with that for the relative
velo ity with respe t to a rotation-free observer. However, keep in mind that
now this expression is only a onvenient representative (a value) of the relative
velo ity and not the relative velo ity itself.

6.4. Some motions relative to an inertial observer

6.4.1. Suppose r is an inertial world line fun tion, use the notations of
2.3.1(iii ) and put to :=  (xo ) :
()

r(t) = xo + uo (t to ):
Let U be a global inertial observer with the onstant velo ity value u:
Then applying one of the formulae in 4.1.1 we get

rU (t) = (xo + uo (t to )) + u
I = (xo + u
I) + (u
=qxo + vuo u (t to )

uo )(t

to ) =

where qxo := xo + u
I is the U -spa e point that xo is in ident with.
This is a uniform motion along a straight line.
Conversely, suppose that we are given a uniform motion relative to the inertial
observer U ; i.e. there is a qo 2 EU ; a to 2 I and a vo 2 EI su h that

rU (t) = qo + vo (t to )

(t 2 I):

Then the orresponding history is inertial; putting xo := qo ? to ; uo := u + vo ;

we get the world line fun tion () whi h gives rise to the given motion.

6.4.2. Let r be a twist-free world line fun tion (see 2.3.1(iii )):
r(t) = xo + uo (t to ) + aoh(t to ):
If U is a global inertial observer with the velo ity value u then

rU (t) = qxo + vuo u (t to ) + ao h(t to );

where qxo := xo + u
I:
If the world line fun tion is not inertial | h 6= 0| then the motion is not
uniform. The motion is re tilinear relative to the observer if and only if vuo u is
parallel to ao :

6.4.3. Now we see that the property \re tilinear" of a motion is not absolute,
in general. The same history an appear as a re tilinear motion to an observer
and as a non-re tilinear one to another observer; ex eptions are the uniform
re tilinear motions, i.e. the inertial histories.
Re all that if U is a global inertial observer then an assertion involving U is
absolute if and only if it an be formulated ex lusively with the aid of the ane
stru ture of I  EU :
Let rU : I ! EU be a motion. Saying that the motion is re tilinear we state
that the range of rU is a straight line in the observer spa e, i.e. we involve the
ane stru ture of EU only. This is not an absolute property.
Saying the motion is re tilinear and uniform we state that f(t; rU (t) j t 2 Ig
is a straight line in I  EU ; this is an absolute property.
6.4.4. Suppose that the global inertial observer U with onstant velo ity
value u hooses a referen e origin o: Then, qo := o + u
I is the U -spa e point
that o is in ident with; hen e the ve torized motion orresponding to the world
line fun tion r be omes
I
or

 E;

t 7! r(t) (o + u(t to )) ;

I  E;

t 7! r(to + t) (o + ut);

where to :=  (o):
In parti ular, if r is the twist-free world line fun tion treated in 6.4.2. and
 (xo ) = to (whi h an be assumed without loss of generality) then the ve torized
motion is

I ! E;

t 7! qo + vuo u t + ao h(t);

where qo := xo o:
Sin e qo = qxo qo holds as well, omparing our present result with that of
6.4.2, evidently we have | as it must be by denition | that the ve torized
motion equals t 7! rU (to + t) qo : The advantage of the ve torized motion is
that it is easier to al ulate.

6.5. Some motions relative to a uniformly a elerated observer

6.5.1. Let r be the previous twist-free world line fun tion and let us examine

the orresponding motion relative to a uniformly a elerated observer U with

onstant a eleration a: We easily obtain by 5.2.1 that
1
CU (r(t)) ? s = xo + uo (t to ) + ao (t to ) + U (r(t))(s t) + a(s t)2 :
2
Then 5.2.2 helps us to transform this expression:
and so

U (r(t)) = U (xo ) + a(t

to )

1
2
a(s to ) +
2

U (xo )) (t to ) + ao h(t to )

CU (r(t)) ? s = xo +U (xo )(s to ) +

+ (uo

1
a(t to )2 :
2

Denoting by qxo the U -spa e point that xo is in ident with and putting
:= uo U (xo ); we an write:


1
2
a(t to ) :
rU (t) = qxo + vo (t to ) + ao h(t to )
2
In parti ular, it is a uniformly a elerated motion, if h = onst:; i.e. if r is
inertial or uniformly a elerated.
vo

6.5.2. Let the previous uniformly a elerated observer U hoose a referen e

origin o: Then the U -spa e point that o is in ident with is given by the world
line fun tion t 7! qo ? t := o + U (o)(t to ) + 12 a(t to )2 ; hen e the ve torized
motion orresponding to the world line fun tion r be omes
or

I  E;

 E;

t 7! r(t) qo ? t

t 7! r(to + t) qo ? (to + t):

In parti ular, the ve torized motion orresponding to the twist-free world line
fun tion r treated above is

I ! E;

t 7! qo + vo t + ao h(t)

1 2
at ;
2

where qo := xo o and vo := uo U (o) = uo U (xo ) (re all that U is onstant

on the simultaneous hyperplanes).
We see in this ase, too, that the ve torized motion is t 7! rU (to + t) qo ; as
it must be, but it is more ompli ated to sear h out the motion rU and then the
ve torized motion than to al ulate the ve torized motion dire tly.

6.6. Some motions relative to a uniformly rotating observer

6.6.1. Let the uniformly rotating observer U hoose a referen e origin o: If

qo is the U -spa e point that o is in ident with and

is the onstant angular
velo ity of the observer, then the double ve torized motion is
I ! E;

t 7! e (t to )

(r(t) qo ? t) :

In parti ular, if qo is an inertial world line, qo = o +

I; and r is an inertial
world line fun tion, r(t) = xo + uo (t to ); where we supposed without loss of
generality that  (xo ) =  (o) = to ; then the double ve torized motion be omes
I ! E;
where qo := xo

t 7! e (t to )

(qo + vuo (t to ))

o; again it is more onvenient to use ve torized time:

I ! E;

t 7! e t

(qo + vuo t):

If vuo = 0; i.e. the relative velo ity of the material point with respe t to
the axis of rotation is zero, then the motion relative to the observer is a simple
rotation around the axis. If vuo 6= 0; then the motion is the \rotation of a
uniform motion". Anyway, the observed rotation of the inertial masspoint is
opposite to the rotation of the observer (take into a ount the negative sign in
the exponent).

6.6.2. In the ase of inertial observers and uniformly a elerated observers,

the ve torized motion is dedu ed in a little simpler way than motion. On the
other hand, for uniformly rotated observers, it is signi antly simpler to get the
double ve torized motion than motion itself, as it will be seen from the following
dedu tion.

Let U and r be as in the pre eding paragraph. Then qo := CU (o) = o +

I
and so

CU (r(t)) ? s =qo ? s + e(s to )

(CU (r(t)) ? to o) =
=qo ? s + e(s to )

e (t to )

(CU (r(t)) ? t qo ? t) =
=qo ? s + e (t to )

e(s to )

(xo o + vuo (t to )) :
The fun tions

I ! E;

s 7! qo (s) := e(s to )

(xo

E
I! ;
I

o)

s 7! vo (s) := e(s to )

vuo

are in EU and in EIU (they are a ve tor and a ve tor of otype I in the observer
spa e), respe tively. Thus we have got for the motion that

rU (t) = qo + e (t to )

(qo + vo (t to ))

(t 2 I ):

Originally, the exponent of

is a linear map from E into E: Here it is regarded
as a linear map from EU into EU dened by


e (t to )

 (s) := e (t to )

(s)

( 2 EU ; s 2 I):

6.7. Exer ise

Let U be a uniformly rotating observer that has an inertial spa e point. Use
the notations of Se tion 5.3. For qo 2 E and vo 2 EI dene the world line
fun tion
t 7! o + (t to ) + e(t to )

(qo + vo (t to )) :
Prove that the orresponding motion relative to the observer U is a uniform
straight line motion.

7. Some kinds of observation

7.1. Ve tors observed by inertial observers

7.1.1. Let C1 and C2 be two world lines dened over the same time interval
J: The ve tor between C1 and C2 at the instant t 2 J is C2 ? t C1 ? t: The
distan e at t between the two world lines is jC2 ? t C1 ? tj:

The two world lines represent the history of two material points. A global
inertial observer U with onstant velo ity value u observes the two material
points des ribing their history by the orresponding motions r1;U and r2;U :
Hen e, the ve tor observed by the inertial observer between the material points
at the instant t is evidently

r2;U (t) r1;U (t) = (C2 ? t + u

I) (C1 ? t + u
I) = C2 ? t C1 ? t:
The observed ve tor oin ides with the (absolute) ve tor; onsequently, the
observed distan e, too, oin ides with the (absolute) distan e.

7.1.2. The question arises how a straight line segment in the spa e of an
inertial observer is observed by another observer. The question and the answer
are formulated orre tly as follows.
Let Uo and U be global inertial observers with onstant velo ity values uo
and u; respe tively. Let Ho be a subset (a geometri al gure) in the Uo -spa e.
The orresponding gure observed by U at the instant t| alled the tra e of
Ho at t in EU | is the set of U -spa e points that oin ide at t with the points
of Ho :
fq ? t + u
I j q 2 Ho g:
Introdu ing the mapping
Pt : EUo ! EU ;

q 7! q ? t + u
I;

we see that the tra e of Ho at t equals Pt [Ho : It is quite easy to see (re all the
denition of subtra tion in observer spa es) that

Pt (q2 ) Pt (q1 ) = q2 ? t q1 ? t = q2

q1

for all q1 ; q2 2 EUo : Thus Pt is an ane map whose underlying linear map is the
identity of E:
We an say that the observed gure and the original gure are ongruent.
Evidently, every gure in the Uo -spa e is of the form qo + Ho ; where qo 2 EUo
and Ho  E; then Pt [qo + Ho = Pt (qo ) + Ho :
In parti ular, a straight line segment in the Uo -spa e observed at an arbitrary
instant by the observer U is a straight line segment parallel to the original one.
Moreover, the original and the observed segments have the same length; the
original and the observed angle between two segments are equal as well.

7.1.3. It is an important fa t that the spa es of dierent global inertial

observers are dierent ane spa es over the same ve tor spa e E: Thus, though
the observer spa es are dierent, it makes sense that a ve tor in the spa e of an
inertial observer oin ides with a ve tor in the spa e of another inertial observer.

Evidently, the oin iden e of ve tors in dierent observer spa es is a symmetri

and transitive relation (if \your" ve tor oin ides with \my" ve tor then \mine"
oin ides with \yours"; if, moreover, \his" ve tor oin ides with \yours" then it
oin ides with \mine" as well.)
This is a trivial fa t here that does not hold in the relativisti spa etime
model.

7.2. Measuring rods

7.2.1. A physi al observer makes measurements in his spa e: measures

the distan e between two points, the length of a line, et . In pra ti e su h
measurements are based on measuring rods: one takes a rod, arries it to the
gure to be measured, puts it onse utively on onvenient pla es... One supposes
that during all this pro edure the rod is absolutely rigid: it remains a straight
line segment and its length does not hange.
We are interested in whether the non-relativisti spa etime model allows su h
measuring rods, i.e. whether we an permit in it the existen e of su h an
absolutely rigid rod.
As we shall see, the answer is positive (in ontradistin tion to the relativisti
ase).
7.2.2. The existen e of an absolutely rigid rod | if it is meaningful | an be
determined uniquely by the history of its extremities. Two world lines C0 and
C1 orrespond to the two extremities of a measuring rod if and only if they are
dened on the same interval J and their distan e at every instant is the same:
jC1 ? t C0 ? tj = d for all t 2 J:
Then for all  2 [0; 1 we an dene the world line C as follows:
C ? t := C0 ? t +  (C1 ? t C0 ? t)

(t 2 J):

It is quite evident that the set of world lines, fC j  2 [0; 1g gives an

existen e of a rigid rod: at every instant t 2 J, fC ? t j  2 [0; 1g is a straight
line segment in E; having the length d:

8. Ve tor splittings

8.1. What is a splitting?

Re all what has been said in 3.1.1: in the experien e of a physi al observer
relative to a phenomenon, and in the notions dedu ed from experien e, the
properties of the pheonomenon are mixed with those of the observer. Our aim is
to nd the absolute notions that model some properties or aspe ts of phenomena
independently of observers and then to give how the observers derive relative
notions from the absolute ones (how the absolute obje ts are observed).
We know already how spa etime is observed as spa e and time and how the
history of a mass point is observed as a motion. In the following, the splitting
of for e elds, potentials et . will be treated: su h splittings des ribe somehow
the observed form of for e elds, potentials, et . We begin with the splitting
of ve tors and ove tors a ording to velo ity values and then we dene the
splitting of ve tor elds and ove tor elds a ording to observers.

8.2. Splitting of ve tors

8.2.1. For u 2 V(1) we have already dened

u : M ! E;
x 7! x

(
x)u

and the linear bije tion

hu

:= ( ; u ) : M ! I  E;

having the inverse

(t; q) 7! ut + q

(1.2.8).
Thus
Moreover,

u

x 7! (
x; u
x)

= idM


u

= 0;

u
;
u


u

= idM

i = idE;

u:

u
u = 0:

8.2.2. Denition. 
x and u
x are alled the timelike omponent and

the u-spa elike omponent of the ve tor x: (
x; u
x) is the u-split form of
x: hu := ( ; u ) is the splitting of M orresponding to u; or the u-splitting of
M:

Note that hu
q = (0; q) for all q 2 E: In other words, E is split into f0g  E
trivially. In appli ations it is onvenient to identify f0g  E with E and to
assume that the split form of a spa elike ve tor q is itself.


8.2.3.
If A is a measure line, A
M M
is split into (A
I)  (A
E)
A


I  E by hu ; thus the timelike omponent and the u-spa elike omponent of
A A


E
;
a ve tor of type A ( otype A) are in A
I AI and in A
E A

respe tively.
In parti ular, hu splits MI into R  EI and for all u0 2 V(1)
hu
u0 = (1; u0 u) = (1; vu0 u );

the u-spa elike omponent of the velo ity value u0 is the relative velo ity of u0
with respe t to u:
Thus V(1) is split into f1g  EI ; in appli ations it is often onvenient to omit
the trivial omponent f1g; and to regard only u instead of hu as the splitting
of V(1) :
E u0 7! u0 u = vu0u:
V(1) ! ;

8.2.4. The timelike omponent of a ve tor is independent of the velo ity value
produ ing the splitting, but the u-spa elike omponents vary with u; ex ept
when the ve tor is spa elike (an element of E); then the timelike omponent is
zero and the u-spa elike omponent is the ve tor itself for all u 2 V(1):
The transformation rule that shows how the u-spa elike omponents of a
ve tor vary with u an be well seen from the following formula giving the u0spa elike omponent of the ve tor having the timelike omponent t and the
u-spa elike omponent q :
Denition. Let u; u0 2 V(1): Then
u

Hu0 u

:= hu0
hu 1 : I  E ! I  E

is alled the ve tor transformation law from u-splitting into u0 -splitting.

Proposition.

Hu0 u
(t; q ) = (t; vu0 u t + q )

(t 2 I; q 2 E):

Using the matrix form of the linear maps I  E ! I  E (see IV.3.7), we an

write


id
0
I
:
Hu0 u =
vu0 u idE

A ording to the identi ation Lin(I)  R we have idI  1: Moreover,

applying the usual onvention that the identity of a ve tor spa e is denoted
by 1 (the identity is the operation of multipli ation by 1), we obtain
Hu0 u

vu0 u

0:
1

In the lower left position of the matrix a linear map

re all that vu0 u 2 EI  Lin(I; E):

I ! E must appear;

8.2.5. Let us give the transformation rule in a form whi h is more usual in
the literature.
Let (t; q) and (t0 ; q0 ) be the u-split form and the u0-split form of the same
ve tor, respe tively. Let v denote the relative velo ity of u0 with respe t to u:
Then
t0 = t;
q 0 = q vt:
Usually one alls this formula | or, rather, a similar formula in the arithmeti spa etime model | the Galilean transformation rule and even one denes
Galilean transformations by it.
The transformation rule is a mapping from IE into IE: A (spe ial) Galilean
transformation is to be dened on spa etime ve tors, i.e. as a mapping from M
into M: Thus the transformation rule and a Galilean transformation annot be
equal. In the split spa etime model I  E stands for both spa etime ve tors and
spa etime. Thus, using the split model (or, similarly, the arithmeti spa etime
model) one an onfuse the transformation rule with a mapping dened on
spa etime ve tors or on spa etime. This indi ates very well that we must not
use the split model or the arithmeti model for the omposition of general ideas.
Of ourse, there is some onne tion between transformation rules and Galilean
transformations. We shall see (11.3.7) that there is a spe ial Galilean transformation L(u; u0 ) : M ! M su h that
Hu0 u

= hu
L(u; u0 )
hu 1 :

8.3. Splitting of ove tors

8.3.1. For u 2 V(1); M is split by the transpose of the inverse of hu :



ru := hu 1 : M ! (I  E)  I  E ;

where we used the identi ation des ribed in IV.1.3. Then for all k
(t; q) 2 I  E we have

M ;

(ru
k)
(t; q) = k
hu 1
(t; q) = k
(ut + q) = (k
u)t + k
q:

Of ourse, in the last term k an be repla ed by kjE = i
k; where i : E ! M
is the anoni al embedding. Furthermore, re all that k
u 2 RI  I and (k
u)t
stands for the tensor produ t of k
u and t: Then, in view of our onvention
regarding the duals of one-dimensional ve tor spa es (IV.3.8), we an state that
ru
k = (k
u; i
k): Re all that i
k = k
i; moreover, our dot notation
onvention allows us to inter hange the order of k and u to have the more
suitable forms
ru
k = (k
u; k
i) = (u
k; i
k)
(k 2 M ):
Denition. u
k and i
k are alled the u-timelike omponent and the
spa elike omponent of the ove tor k: (u
k; i
k) is the u-split form of the
ove tor k: ru is the splitting of M orresponding to u; or the u-splitting of
M:
Note that ru
(e ) = (e; 0) for all e 2 I : In other words, I
 is split into

I  f0g trivially. In appli ations it is onvenient to identify I  f0g with I
and to onsider that the split form of e is simply e:

8.3.2. The spa elike omponent of a ove tor is independent of the velo ity
value u establishing the splitting, but the u-timelike omponents vary with u;
ex ept when the ove tor is timelike (an element of I
 ; then the spa elike omponent is zero and the u-timelike omponent oin ides with the orresponding
element of I ): The transformation rule that shows how the u-timelike omponents of a ove tor vary with u an be well seen from the following formula giving
the u0 -timelike omponent of the ove tor having the u-timelike omponent e
and the spa elike omponent p:
Denition. Let u; u0 2 V(1): Then
Ru0 u := ru0
r 1 : I  E ! I  E
u

is alled the ove tor transformation law from u-splitting into u0 -splitting.

Proposition.

Ru0 u
(e; p) = (e + p
vu0 u ; p)

Proof. It is not hard to see that

ru 1 (e; p) = e

+ u
p

from whi h we easily obtain the desired formula.

(e 2 I ; p 2 E ):

Using the matrix form of the linear maps I  E ! I  E ; we an write



idI vu0 u
Ru0 u =
0 idE

1 vu0 u
0 1 :

In the upper right position a linear map E ! I must appear. The identi ations Lin(E ; I )  I
E  EI justify that
vu0 u stands in that position.

The denitions imply that Ru0 u = Huu10 ; whi h is re e ted in the matrix
form as well.

8.3.3. In 1.2.8 we have drawn a good pi ture how ve tors are split. Now we
give an illustration for splitting of ove tors.
Re all that for all u 2 V(1); the surje tion u : M ! E is the left inverse of
the anoni al embedding i : E ! M; i.e. u
i = idE : As a onsequen e, the
inje tion u : E ! M is the right inverse of the surje tion i : M ! E :
i
 = idE :
u

Sin e I
 = Ker i (see 1.2.1),

E
u = Ran 

is a three-dimensional linear subspa e in M ; omplementary to I
 :

Evidently, the restri tion of i is a linear bije tion from E
u onto E :
Moreover, we easily nd that
E
u = fk 2 M j k
u = 0g;

in other words, E
u is the annullator of u

I:
Then the splitting of ove tors a ording to u is illustrated as follows:

(u
k)
 is in

E
 u ;

its image by i is the spa elike omponent of k:

8.4. Ve tors and ove tors are split in a dierent way

The splitting of ve tors and the splitting of ove tors a ording to u 2 V(1)
are essentially dierent. The timelike omponent of ve tors is independent
of u; whereas the spa elike omponent of ove tors is independent of u: The
transformation laws for ve tors and ove tors are essentially dierent as well.
The reason of these dieren es lies in the fa t that there is no one-dimensional
ve tor spa e A in su h a way that M ould be anoni ally identied with M
A;
in ontradistin tion to the relativisti ase.

8.5. Splitting of ve tor elds and ove tor elds a ording to inertial
observers
8.5.1. In appli ations, ve tors and ove tors appear in two ways: rst, as

values of fun tions dened in time; se ondly, as values of fun tions dened in
spa etime. The rst ase an be redu ed to the se ond one: a fun tion dened
in time an be onsidered a fun tion dened in spa etime that is onstant on the
simultaneous hyperplanes. Thus we shall study ve tor elds and ove tor elds,
i.e. fun tions X : M M and K : M M ; respe tively.
A global inertial observer U splits ve tor elds and ove tor elds in su h a
way that at every world point x the values of the elds, X (x) and K (x); are
split a ording to the velo ity value u of the observer; thus the half U -split form
of the elds will be




: M I  E;
ru
K : M
I  E;
hu
X

x 7! (
X (x); u
X (x)) ;
x 7! (u
K (x); i
K (x)) :

However, the observer splits spa etime as well (the observer regards spa etime
as time and spa e); a ordingly, instead of world points, instants and U -spa e
points will be introdu ed to get the ompletely U -split form of the elds:
hu
X H U 1

 I  E;
 I  E;

: I  EU

ru
K HU 1 : I  EU

where q ? t := HU 1 (t; q) (see 3.2.2).

(t; q) 7! (
X (q ? t); u
X (q ? t)) ;

(t; q) 7! (u
K (q ? t); i
K (q ? t))

8.5.2. Let us examine more losely the split forms of a ove tor eld K :
( Vu ; Au ) := ru
K : M  I  E;
x 7! (u
K (x); i
K (x)) ;
( VU ; AU ) := ( Vu ; Au ) HU 1 =ru
K HU 1 : I  EU  I  E ;
(t; q) 7! (u
K (q ? t); i
K (q ? t)) :

A ove tor eld K is a potential (see 2.4.3). VU and AU are alled the
orresponding s alar potential and ve tor potential a ording to U :
If U 0 is another global inertial observer with onstant velo ity value u0 then,
in view of 8.3.2,
Au0 = Au :
Vu0 = Vu vu0 u
Au ;
As a onsequen e,
VU 0 HU 0 = (VU vu0 u
AU ) HU ;
AU 0 HU 0 =AU HU :

8.5.3. Introdu ing V := Vu , V 0 := Vu0 ; A := Au; A0 := Au0 ; v := vu0u; we

get the formulae

V0 =V

v
A;

A0

= A;

whi h are the well-known non-relativisti transformation law for s alar and
ve tor potentials in ele tromagnetism.
This supports our hoi e that (absolute) potentials are otensor elds.
The reader is asked to bear in mind the following remark. One usually
says that if an observer per eives s alar potential V and ve tor potential A;
then another observer moving with relative velo ity v per eives s alar potential
V v
A and ve tor potential A: However, an observer U per eives spa etime
as time and U -spa e, per eives the potentials to be fun tions depending on time
and U -spa e; thus, in fa t, an observer observes the ompletely split form of
the potentials. In parti ular, if U 0 6= U ; then AU 0 6= AU : the observed ve tor
potentials are dierent! Remember, usually one does not distinguish between
the half split forms and the ompletely split forms.

8.5.4. Similarly, one usually says that for e is not transformed, a for e eld is
the same for all observers. Of ourse, this is true for the half split form of for e
elds; the ompletely split forms of for e elds | whi h are a tually observed
| depend on the observers.
A for e eld
f

: M  V(1)

 I
DE
D

has ex lusively spa elike values, thus its half split form is f itself for all global
inertial observers. On the other hand, f has the ompletely split form
I  EU 

E E ;
I I
D
D

strongly depending on U :

(t; q; v) 7! f (q ? t; u + v)

8.5.5. Let the global inertial observer U hoose a referen e origin o; then
(U ; o) performs another splitting using HU ;o instead of HU : The half split form
of ve tor elds and ove tor elds a ording to (U ; o) is the same as the half
split form a ording to U ; on the other hand, the observer with referen e origin
obtains fun tions I  E I  E and I  E I  E for the ompletely split
forms of the elds.

8.6. Splitting of ve tor elds and ove tor elds a ording to

rigid observers
8.6.1. Re all that the spa e EU of a global rigid observer U is an ane spa e
over E or EU (Se tion 4.3), depending on whether U is rotation-free or not. The

orresponding splitting of spa etime, HU : M ! I  EU is a smooth bije tion

whose inverse is smooth as well.
The splitting of a ve tor eld X a ording to U is dened by the orresponding
formula of oordinatization: at every world point x; the value of the eld, X (x);
is split | i.e. is mapped from M into I  E or I  EU | by DHU (x): Similarly,
the ove tor eld K ; is split in su h a way that at every world point x the value
of the eld, K (x
) is split | i.e. is mapped from M into I  E or I  EU |
by (DHU (x)) 1 : Thus the half U -split forms of su h elds are
M
M

IE
 I  E

(or I  EU );
(or I  EU );

x 7! DHU (x)
X (x);

x 7! (DHU (x)) 1
K (x):

We get the ompletely U-split forms by substituting HU 1 (t; q) = q ? t for x in

these formulae.

8.6.2. If U is rotation-free, then, in view of 4.3.2, the half split forms of the

elds are

M
M

 I  E;
 I  E;

x 7! 
X (x); U (x)
X (x) ;

x 7! (U (x)
K (x); i
K (x)) :

The values of the elds at x are split by the orresponding value U (x) of the
observer.

8.6.3. If U is not rotation-free, EU is an uneasy obje t; that is why we let the

global rigid observer hoose a referen e origin o and use the double ve torization
HU ;o of spa etime.
Then the half split forms be ome
M

 I  E;

x 7! DHU ;o(x)
X (x) =


X (x); R(x) 1
U (x)
X (x) ;

 I  E;

x 7! (DHU ;o(x)) 1
K (x) =


= U (x)
K (x); R(x) 1
i
K (x) ;

where

R(x) := RU ( (x); to );
(see Exer ises 4.2.2 and 4.5.3).

8.7. Exer ises

1. Give the split form of ve tor elds and ove tor elds that depends only
on time; more losely, if  : I M and  : I M; onsider the splitting of
the elds X :=   and K :=  :
2. We know, it has an absolute meaning that a fun tion  dened in spa etime
depends only on time: if  is onstant on the simultaneous hyperplanes.
On the other hand, it does not have an absolute meaning that  depends only
on spa e (absolute spa e does not exist). If U is an observer, it makes sense that
 depends only on U -spa e, in other words,  is U -stati : if  is onstant in the
U -spa e points (on the U -lines), i.e. if the ompletely split form of  depends
only on the elements of EU :
Let o 2 M; 2 V(1); C : E ! M; and let U be the inertial observer with the
velo ity value u: Prove that the ve tor eld x 7! C (
(x o)) is U -stati if
and only if u = :
3. Take the arithmeti spa etime model. Give the ompletely split form of
the ve tor elds

( 0 ; ) 7! (jj; 0);
( 0 ; ) 7! ( 0 + jj;  2 +  3 ; 1 +  1 ; 0)

a ording to the inertial observer with velo ity value (1; v):
Consider the previous mappings to be ove tor elds and give their ompletely
split form.
4. Take the arithmeti spa etime model. Give the ompletely split form of
the ve tor eld


( 0 ; ) 7!  1 +  2 ; os( 0  3 ); 0; 0
a ording to the uniformly a elerated observer with referen e origin treated
in 5.2.4 and to the uniformly rotating observer with referen e origin treated
in 5.3.5. (It is easy to obtain the omposition of this ve tor eld and the
inverse of the splitting if we use dierent symbols for the variables; e.g. the
splitting
due to the uniformly
a elerated observer has the inverse ( 0 ;  ) 7!


2
1
 0 ;  1 + 2 ( 0 ) ;  2 ;  3 :

5. In the split spa etime model the splitting of ve tors a ording to the basi
velo ity value (1; 0) is the identity of I  E: The splitting a ording to (1; v) is

I  E ! I  E;

(t; q) 7! (t; q

vt);

whi h oin ides with the transformation rule from (1; 0) into (1; v): Be ause of
the spe ial stru ture of the split spa etime model a splitting and a transformation
rule | whi h are in fa t dierent obje ts | an be equal. To deal with
fundamental ideas do not use the split spa etime model or the arithmeti one.
6. In the split spa etime model the splitting of ove tors a ording to the
basi velo ity value (1; 0) is the identity of I  E and the splitting a ording
to (1; v) equals

I  E ! I  E;

(e; p) 7! (e + p
v; p):

Again we see that the splitting oin ides with the transformation rule from
(1; 0) into (1; v):

9. Tensor splittings

9.1. Splitting of tensors, otensors, et .

9.1.1. The various tensors are split a ording to u 2 V(1) by the maps
hu
hu : M
M ! (I
E)
(I  E) =
= (I
I)  (I
E)  (E
I)  (E
E);
hu
ru : M
M ! (I  E)
(I  E ) =
= (I
I )  (I
E )  (E
I )  (E
E );
ru
hu : M
M ! (I  E )
(I
E) =
= (I
I)  (I
E)  (E
I)  (E
E);
ru
ru : M
M ! (I  E )
(I  E ) =
= (I
I )  (I
E )  (E
I )  (E
E ):
Sin e we know hu and ru ; our task is only to determine the above splittings
in a perspi uous way. First re all that the elements of the Cartesian produ ts
on the right-hand sides an be well given in a matrix form (see IV.3.7). Se ond,
with the aid of the usual identi ations, onsider hu = ( ; u ) 2 (I  E)
M ;
ru = (u; i ) 2 (I  E )
M; take into a ount the identi ations I
M 
M
I;     and I
M  M
I ; u  u (see IV.3.6), and apply the dot
produ ts to have

for T

2M
M:


 
(hu
hu )(T ) = hu
T
hu = hu
T
ru 1 = 

TT

 

TT

u =
u
u
u


= T
 
uT(

T
 ) T
for L 2 M
M :


T

u
(
T )

u(
T
 )
(T
 )
u + u
u(
T
 )

(hu
ru )(L) = hu
L
ru =hu
L
hu 1 = 

LL

uu 

LL

ii =
u
u


2 M
M :


L
i
= L
u 
uL(
u

L
u) L
i u
(
L
i) ;


for P



u
P
 u
P
u

1
(ru
hu )(P ) = ru
P
hu =ru
P
ru = i
P
 i
P
 =
u

=
for F

2 M
M :

u
P


P


u
P

P

(u
P
 )u
(i
P
 )
u ;

u u
F
i :
(ru
ru )(F ) = ru
F
ru = ru
F
hu 1 = iu

F
F
u i
F
i

(To see, e.g. that u
T
 = T


u(
T
 );

take T = x
y:)

9.1.2. The splittings orresponding to dierent velo ity values u and u0 are

dierent. To ompare the dierent splittings we an dedu e transformation rules

by giving


h b
Hu0 u


Hu0 u  ;
a A

where h 2 I
I; a 2 E
I; b 2 I
E, A 2 E
E; and using similar formulae
for the other three ases as well. In general, the transformation rules are rather
ompli ated. We shall study them for antisymmetri tensors and otensors.

9.2. Splitting of antisymmetri tensors

9.2.1. If the tensor T is antisymmetri  | i.e. T 2 M ^ M| then 
T
 = 0;

T
u
= (u
T
 ) and u
T
u 2 E ^ E; whi h (of ourse) means that
the split forms of T are antisymmetri as well. Thus splittings map the elements
of M ^ M into elements of the form

0 a    0 a  ;
a

where a 2 E
I; A 2 E ^ E; a 2 I
E is the transpose of a; whi h is identied
with a in the usual identi ation I
E  E
I: We shall nd onvenient to
write
(E
I)  (E ^ E) (I  E) ^ (I  E);
(a; A)  a0


a
A

The orresponding formula in 9.1.1 gives us for T

hu
T
hu

= (T
 ; T

2M^M

(T
 ) ^ u) :

Denition. T
 and T (T
 ) ^ u are alled the timelike omponent and

the u-spa elike omponent of the antisymmetri tensor T :
9.2.2. Noti e the similarity between splittings of ve tors and splittings of
antisymmetri tensors. The timelike omponent of T is independent of u; the
u-spa elike omponent varies with u ex ept when T is spa elike, i.e. is in E ^ E;
then the timelike omponent is zero and the u-spa elike omponent is T itself
for all u:
The following transformation rule shows well how the splittings depend on
the velo ity values.
Proposition. Let u; u0 2 V(1): Then
Hu0 u
(a; A)
Hu0 u 

(a 2 E
I; A 2 E ^ E):

= (a; a ^ vu0 u + A)

Proof. Use the matrix forms:


1
00 a 1
vu0 u 1
a A
0

vu0 u

= a

a ^ vu0 u + A

9.3. Splitting of antisymmetri otensors

9.3.1. If F 2 M ^ M then u
F
u = 0; u
F
i = (i
F
u) and

i
F
i 2 E ^ E ; the split
forms of F are antisymmetri as well. Thus splitting
maps the elements of M ^ M into elements of the form

0 z    0 z   (z; Z ) 2 (E
I)  (E ^ E);
z
Z
z Z
where we used notations similar to those in 9.2.1.
The orresponding formula in 9.1.1 gives for F 2 M ^ M :
ru
F
r  = (i
F
u; i
F
i) :
u

Denition. i
F
u and i
F
i are alled the u-timelike omponent and

the spa elike omponent of the antisymmetri otensor F :

9.3.2. Noti e the similarity between splittings of ove tors and splittings of
antisymmetri otensors. The spa elike omponent of F is independent of u;
the u-timelike omponent varies with u ex ept when F is in M ^ (I
 ) :=
fk ^ (e
 ) j k 2 M ; e 2 I g; then the spa elike omponent is zero and the
u-timelike omponent is the same for all u:
The following transformation rule shows well, how the splittings depend on
the velo ity values.
Proposition. Let u; u0 2 V(1): Then
Ru0 u
(z ; Z )
Ru0 u  = (z + Z
vu0 u ; Z )
(z 2 I
E ; Z 2 E ^ E ):
Proof. Use the matrix forms:

0 z 1 0 =  0
1 vu0 u
0 1 z Z vu0u 1
z + Z
vu0 u

9.4. Splitting of otensor elds

(z + Z
vu0 u ) :
Z

9.4.1. A rotation-free rigid observer U splits various tensor elds in su h a

way that the value of the tensor eld at the world point x is split a ording to
U (x); for the sake of deniteness we shall onsider otensor elds. The half split
form of the otensor eld F : M M
M a ording to U is
M (I  E )
(I  E );
x 7! rU (x)
F (x)
rU (x) :

The ompletely split form of F a ording to U is

I  EU

 (I  E)
(I  E);

(t; q) 7! rU (q?t)
F (q ? t)
rU (q?t) ;

where q ? t = HU 1 (t; q):

In parti ular, if F is antisymmetri , then it has the half split form
M

 (E
I)  (E ^ E);

x 7! (i
F (x)
U (x); i
F (x)
i) :

9.4.2. Now let us suppose that U is a global inertial observer with the
onstant velo ity value u: Then the antisymmetri otensor eld F has the half
split form
(Eu ; Bu ) := ru
F
ru :M (E
I )  (E ^ E);
x 7! (i
F (x)
u; i
F (x)
i)

and the ompletely split form

(EU ; BU ) := (Eu ; Bu ) HU 1 =
=ru
(F HU 1 )
ru : I  EU (E
I )  (E ^ E );
(t; q) 7! (i
F (q ? t)
u; i
F (q ? t)
i) :

If U 0 is another inertial observer with the velo ity value u0; then 9.3.2 gives
(Bu is antisymmetri , hen e Bu
vu0 u = vu0u
Bu ) that
Eu0

= Eu + vu0 u
Bu ;

Bu0

= Bu :

As a onsequen e,
EU 0
BU 0

HU 0 =(EU + vu0 u
BU ) HU ;
HU 0 =BU HU :

9.4.3. Introdu ing E := Eu ; E 0 := Eu0 ; B := Bu; B0 := Bu0 ; v := vu0u;

we get the formula
E0 = E + v
B;
B0 = B
whi h is the well-known non-relativisti transformation law for the ele tri eld
E and magneti eld B : (Here B is an antisymmetri spa elike tensor of otype

4 D; an element of E ^ E = D
DE
^ED
D ; whi h an be identied with a
3
E  E (V.3.17.); with the aid of
ve tor of otype
D; an element of D
D

D
D

this identi ation magneti eld is regarded as a ve tor eld and then instead of
one has a ve torial produ t.)
This supports the idea that (absolute) ele tromagneti elds exist whose timelike and negative spa elike omponents a ording to an observer are the observed
ele tri and magneti elds, respe tively.
One usually says that if an observer per eives ele tri eld E and magneti
eld B ; then another observer moving with the velo ity v per eives ele tri eld
E + v
B and magneti eld B : However, an observer per eives spa etime as
time and U -spa e, per eives the elds as fun tions depending on time and U spa e; thus, in fa t, an observer observes the ompletely split form of the elds,
and we an repeat the remark at the end of 8.5.3.
v
B

9.4.4. Consider the ompletely split form of a potential K a ording to the

inertial observer U with velo ity value u :
( VU ; AU ) := ru
(K HU 1 ) : I  EU

 I  E:

Its derivative is
D( VU ; AU ) = ru
(DK HU 1 )
ru 
having the transpose


(D( VU ; AU )) = ru
(DK HU 1 ) 
ru  :

Consequently, for the exterior derivatives (see VI.3.6(i)) we have

D ^ ( VU ; AU ) = ru
(D ^ K ) HU 1
ru  :
Let F := D ^ K ; use the notations of the previous paragraph and let o and
r denote the partial derivations with respe t to I and EU ; respe tively. Then

(see VI.3.7(ii )) the above equality yields

o AU

rVU = EU ;

r ^ AU = BU :

9.4.5. Let us onsider the for e eld dened by the potential K :

f (x; u0 ) = i
F (x)
u0
(x 2 Dom K ; u0 2 V(1):
(F := D ^ K ):

A ording to 9.3.1, the value of the for e eld at (x; u0 ) is the u0 -timelike
omponent of the antisymmetri otensor F (x):

A masspoint at the world point x having the instantaneous velo ity value u0
\feels" only the u0 -timelike omponent of the eld; a masspoint always \feels"
the time omponent of the eld a ording to its instantaneous velo ity value.
Consider now the inertial observer with velo ity value u and use the notations
of the previous paragraphs. Then
f (x; u0 ) =i
F (x)
u + i
F (x)(u0 u) =
=Eu (x) + vu0 u
Bu(x);

a well-known formula for the Lorentz for e in ele tromagnetism.

9.4.6. If a potential K is timelike, i.e. has values in I
 ; (in fa t K is a

s alar eld: there is a fun tion V : M  I su h that K = V
 ) then D ^ K
takes values in M ^ (I
 ); onsequently the orresponding for e eld does not

depend on velo ity values; the spa elike omponent of K is zero and the half
split form of K is the same for all observers.
The possibility of (absolute) s alar potentials is a pe uliar feature of the nonrelativisti spa etime model in ontradistin tion to relativisti spa etime models.
(Newtonian gravitational elds, elasti elds are modelled by su h timelike potentials in non-relativisti physi s.)

9.4.7. Let us mention the ase of a general (rotating) global rigid observer U :
Then it is onvenient to hoose a referen e origin o for the observer and onsider
the orresponding double ve torization of spa etime.
We easily infer from the splitting of ve tor elds and ove tor elds that the
half split forms of various tensor elds a ording to (U ; o) are obtained from
the half split forms a ording to a rotation-free observer in su h a way that
RU ( (x); to ) 1
U (x) and RU ( (x); to ) 1
i are substituted for U (x) and i ;
respe tively (then i
RU ( (x); to ) is substituted for i):
For instan e, the half split form of an antisymmetri otensor eld F be omes
M

 (E
I)  (E ^ E);

x 7! R(x) 1
i
F (x)
U (x); R(x) 1
i
F (x)
i
R(x) ;

where R(x) := RU ( (x); to ):

9.5. Exer ises

1. Give the u-split form of tensors in E
E; E
M; M
E; E
M; M
E;

(I
 )
M; M
(I
 ); (I
 )
M ; M
(I
 ) and derive the transformation
0

rules between their u -splitting and u-splitting.

2. Derive the transformation rules for the splitting of arbitrary tensors.

3. A potential in the arithmeti spa etime model is a fun tion ( V; A) :
R  R3
(R  R 3 ) whi h is the ompletely split form of the potential a ording
to the basi observer.
The half split form of this potential a ording to the inertial observer with
velo ity value (1; v) is ( V + v
A; A):
Choose (0; 0) as a referen e origin for the observer and give the ompletely
split form of the potential.
4. An antisymmetri otensor eld in the arithmeti

spa etime model is a
fun tion (E ; B ) : R  R3
(R3 )  (R3 ) ^ (R3 ) ; being the ompletely
split form of the eld a ording to the basi observer:

0
B E1
(E ; B ) =  E
2
E3

E1
0
B3
B2

E2
B3
0
B1

E3 1
B2 C
B1 A :
0

The half split form of this eld a ording to the inertial observer with velo ity
value (1; v) is (E + v
B ; B ) :
Choose (0; 0) as a referen e origin for the observer and give the ompletely
split form of the eld.
5. Take the uniformly a elerated observer treated in 5.2.4.
The half split forms of the previous potential and eld a ording to this
observer are
( V + tA1 ; A)

and

((E1 ; E2

tB3 ; E3 + tB2 );

B) ;

where t is the time evaluation: R  R3 ! R; ( 0 ; ) 7!  0 :

Choose (0; 0) as a referen e origin for the observer and give the ompletely
split forms.
6. Take the uniformly rotating observer treated in 5.3.5.
Let ( V 0 ; A0 ) and (E 0 ; B 0) denote the half split forms of the previous
potential and eld, respe tively, a ording to this observer. Then

V 0 =V + !(x2 A1 x1 A2 );
A01 =A1 os !t A2 sin !t;
A02 =A1 sin !t + A2 os !t;
A03 =A3

and

E10 =(E1 + !x1 B3 ) os !t (E2 + !x2 B3 ) sin !t;

E20 =(E1 + !x1 B3 ) sin !t + (E2 + !x2 B3 ) os !t;
E30 =E3 !(x2 B2 + x1 B1 );
B10 =B1 os !t B2 sin !t;
B20 =B1 sin !t + B2 os !t;
B30 =B3 ;
where t is the time evaluation R  R 3 ! R; ( 0 ; ) 7!  0 and xi is the evaluation
of the i-th spa e oordinate: R  R3 ! R; ( 0 ; ) !  i :

10. Referen e frames

10.1. The notion of a referen e frame

10.1.1. Referen e frames are usually fundamental notions in textbooks of

physi s: the phenomena are always des ribed in referen e frames. However, this
notion is not exa tly dened there.
We have expressed our intention to give an absolute des ription of phenomena,
i.e. a des ription free of referen e frames and observers. Observers and referen e
frames | whi h must be exa tly dened in our framework | have only a pra ti al (not theoreti al) importan e: it is onvenient and suitable to use referen e
frames for solving a tual problems, for a hieving numeri al hara terization of
quantities.
It was mentioned in 3.1 that the usual notion of referen e frames involves
oordinates introdu ed with the aid of some material obje ts. The material
obje ts play a more fundamental role; that is why we reated the notion of their
model: the observer. An observer and a oordinatization of time and observer
spa e together will form a referen e system giving rise to a referen e frame.
Coordinatization models the pro edure how a physi al observer measures time
with a lo k (having a dial) and introdu es numbered referen e lines in his spa e.
Then a time point is represented by a number, and a spa e point is represented
by a triplet of numbers.
The reader is supposed to be familiar with the notion of oordinatization
whi h an be found in Se tion VI.5.
10.1.2. Re all that an observer U makes the splitting HU = (; CU ) : M 

I  EU :

Denition. A referen e system is a triplet (U ; T; SU ) where

(i ) U is an observer,
(ii ) T : I R is a stri tly monotone in reasing mapping,
(iii ) SU : EU R3 is a mapping
su h that (T  SU ) HU = (T ; SU CU ) : M R  R3 is an orientation
preserving (lo al) oordinatization of spa etime.
A ording to the denition, T  is smooth whi h implies by VI.3.5 that
T is smooth as well. Be ause of (ii ) the derivative of T | denoted by T 0| is
everywhere positive,

0 < T 0(t) 2 I 

(t 2 Dom T );
I
i.e. T is an (orientation preserving) oordinatization of time.
If U is a global rigid observer then EU is an ane spa e and CU is a smooth
map (see 4.4.3); onsequently, we an state that SU is a oordinatization of
EU : On the ontrary, sin e EU ; in general, is not an ane spa e, and we
introdu ed the notion of oordinatization only for ane spa es, we annot state
that SU is a oordinatization of U -spa e; nevertheless it will be alled the
oordinatization of U -spa e. (We mentioned that in any ase EU an be endowed
with a smooth stru ture; in the framework of smooth stru tures SU does be ome
a oordinatization.)

10.1.3. Denition. A oordinatization K : M  R  R 3 is alled a referen e

frame if there is a referen e system (U ; T; SU ) su h that K = (T  SU ) HU :

U ; T and SU are alled the observer, the time oordinatization and the U spa e oordinatization orresponding to the referen e frame.
As usual, the oordinates of R  R3 are numbered from zero to three. A ordingly, we nd onvenient to write a oordinatization of spa etime in the
form K = (0 ; ) : M
R  R3 : Using the notations pr0 : R  R3 ! R and
3
3
pr : R  R ! R for the anoni al proje tions, we have 0 = pr0 K;  = pr K:
The following important relation holds for an arbitrary oordinatization K :

D(x)
0 K 1(K (x)) = 0

(x 2 Dom K ):

Indeed, a ording to the denition of partial derivatives (VI.3.8) and the rules
of dierentiation (VI.3.4), we have


0 K 1 (K (x)) = DK 1 (K (x))
(1; 0) = DK (x) 1
(1; 0);
D(x) =pr
DK (x);

from whi h we infer the desired equality.

If K is a referen e frame then

0 = T ;

 = SU

CU

()

and

D0 (x) = T 0( (x))

from whi h we dedu e

T 0 ( (x)) =

0 K 1 (K (x))

in the following way:

pr0
DK (x) =T 0( (x)) ;
pr0 =T 0( (x))
DK (x) 1 ;
1 =T 0( (x))
DK (x) 1
(1; 0);
1 =T 0( (x))
0 K 1 (K (x)):

10.1.4. Proposition. A oordinatization K = (0; ) : M R  R3 is a

referen e frame if and only if
(i ) K is orientation-preserving,
(ii ) 0 K 1(K (x)) is a future-dire ted timelike ve tor,
(iii ) 0 (x) < 0 (y) is equivalent to  (x) <  (y) for all x; y 2 Dom K:
Then
(1)
U (x) =

(2)
(3)

0 K 1 (K (x))
=0 K 1 (K (x))
T 0( (x))
(x 2 Dom K );

0 K 1 (K (x))
T (t) =0 (x)
(t \ Dom K 6= ;; x 2 t);
SU (q) =(x)
(q 2 EU ; x 2 q)

for the orresponding referen e system (U ; T; SU ):

Proof. If K is a referen e frame, K = (T  SU ) HU ; then (i ) is trivial and
(iii ) follows from 0 = T  and the stri tly monotone hara ter of T: As on erns
(ii ), note that a world line fun tion r satises r_ (t) = U (r(t)) and takes values
in the domain of K if and only if K (r(t)) = (T (t); ); i.e. r(t) = K 1(T (t); )
for a  2 R3 and for all t 2 Dom r: As a onsequen e, we have
U (r(t))

implying

d
= K
dt
=0 K

1 (T (t);  ) =  K 1 (T (t);  ))
T 0 (t) =

0
1 (K (r(t)))
T 0 (t)

U (x) = 0 K 1 (K (x))
T 0 ( (x)) ;

(x 2 Dom K );

whi h proves (ii ), sin e T 0( (x)) > 0: It proves equality (1) as well; equalities
(2) and (3) are trivial.
Suppose now that K = (0 ; ) is a oordinatization that fullls onditions
(i ){(iii ).
Then ondition (ii ) implies that U dened by the rst equality in (1) is an
observer.
A ording to (iii ), K is onstant on the simultaneous hyperplanes, thus T is
well dened by the formula (2). Moreover, T is stri tly monotone in reasing.
If r is a world line fun tion su h that r_ (t) = U (r(t)) then a ording to () in
the pre eding paragraph
d
 K 1 (K (r(t)))
((r(t)) = D(r(t))
U (r(t)) = D(r(t))
0
= 0;
dt

0 K 1 (K (r(t)))

whi h means that  r is a onstant mapping, in other words,  is onstant on

the U -lines; hen e SU is well dened by the formula (3).
Finally, it is evident that K = (T  SU ) HU :
It is suitable to use P := K 1; the parameterization orresponding to K:
Then | putting (P ) instead of  P for any fun tion  | we an rewrite
formula (1) in the proposition:
U (P ) =

0 P
:

0 P

10.1.5. Condition (iii ) in the previous proposition an be repla ed by

(iii )' for all x 2 Dom K there is an e(x) 2 (I )+ su h that
D0 (x) = e(x) ;

i.e. the derivative of 0 in every point is a positive multiple of  :

Indeed, if K is a referen e frame, then e(x) = T 0( (x)):
Conversely, if (iii )' holds, then the restri tion
of 0 onto
every simultaneous



0

hyperplane t has zero derivative: D  t (x) = D0 (x)E = 0 (x 2 t) thus 0
is onstant on every simultaneous hyperplane whi h allows us to dene T by the
formula (1) in the previous proposition.
Moreover, Lagrange's mean value theorem implies that every x in the domain
of K has a neighbourhood su h that for all y in that neighbourhood there is a z
on the straight line segment onne ting x and y in su h a way that

0 (y) 0 (x) = D0 (z )
(y x) = e(z )
(y x);

hen e 0 (y) 0 (x) > 0 is equivalent to 
(y x) > 0 in the neighbourhood
in question. Sin e the domain of K is onne ted, this relation holds globally as
well.

10.2. Galilean referen e frames

10.2.1. Now we are interested in what kinds of ane oordinatizations of

spa etime an be referen e frames.
Let us take an ane oordinatization K of M: Then there are
| an o 2 M;
| an ordered basis (x0 ; x1 ; x2 ; x3 ) of M su h that
K (x) =

ki
(x

o) j i = 0; 1; 2; 3

(x 2 M);

where (k0 ; k1 ; k2 ; k3 ) is the dual of the basis in question.

Proposition. The ane oordinatization K is a referen e frame if and only
if
(i ) (x0 ; x1 ; x2 ; x3 ) is a positively oriented basis,
(ii ) x0 is a future-dire ted timelike ve tor,
(iii ) x1 ; x2 ; x3 are spa elike ve tors.
Then the orresponding observer is global and inertial having the onstant
value
x0
u :=
;
and

K (x) =



(x o) ; (p

(x o))
u
=1;2;3
s

X
K 1( 0 ; ) = o +  0 su +   x

=1

where

s := 
x0

(x 2 M);
( 0 ; ) 2 R  R3

and fp := k jE = k
i j ( = 1; 2; 3)g is the dual of the basis fx1; x2 ; x3 g

of E:
Proof. We show that the present onditions (i){(iii) orrespond to the
onditions (i ){(ii ) listed in Proposition 10.1.4 and ondition (iii )' in 10.1.5.
(i ) The oordinatization is orientation-preserving if and only if the orresponding basis is positively oriented;
(ii ) 0 K 1 (K (x)) = x0 ;
(iii )' D0 (x) = k0 for all x 2 M: Sin e k0
x = 0; k0 = e for some e 2 I
if and only if x -s ( = 1; 2; 3) are spa elike; then, be ause of k0
x0 = 1 > 0;
e = 
1x0 > 0:
We shall use the following names: o is the origin, (x0 ; x1 ; x2; x3 ) is the
spa etime basis of the ane referen e system; moreover, s := 
x0 2 I+ is the

time unit, u := xs0 is the velo ity value, (x1 ; x2 ; x3 ) is the spa e basis of the
ane referen e frame.

10.2.2. Let us take an ane referen e frame K: Then the restri tion of K
(the linear map under K ) onto E is a linear bije tion between E and f0g  R3 :
Let B denote the usual inner produ t on R3  f0g  R3 ; then it makes sense
that K jE = K
i : E ! f0g  R3 is b-B-orthogonal (see V.1.6).
Denition. A referen e frame K is alled Galilean if
| K is ane,
| K
i : E ! f0g  R3 is b-B-orthogonal.
Proposition. A referen e frame K is Galilean if and only if there are
an o 2 M;
an ordered basis (e0 ; e1 ; e2 ; e3 ) of M;

(i )
(ii )
|
|
|

(e0 ; e1 ; e2 ; e3 ) is positively oriented,

s := 
e0 > 0;
(e1 ; e2 ; e3 ) is a (ne essarily positively oriented) orthogonal basis in E;
normed to an m 2 D+ ; su h that

K (x) =
where

(x o) ;  e
u
(x o) 
s
m2

=1;2;3

u :=

(x 2 M);

e0
s

is the onstant value of the orresponding inertial observer.

Proof. It is quite evident that an ane referen e frame is Galilean if and
only if the spa elike elements of the orresponding basis in M are orthogonal
to ea h other and have the same length.
We know that the dual of the basis
e2 ; e3
in the identi ation E  E whi h
(e1 ; e2 ; e3 ) be omes me12 ; m
2
m2
D
D
proves the equality regarding K:
We shall use the following names for a Galilean referen e frame: o is its origin ,
(e0 ; e1 ; e2 ; e3 ) is its spa etime basis ; moreover, s := 
e0 is its time unit,
(e1 ; e2 ; e3 ) is its spa e basis, m := je j ( = 1; 2; 3) is its distan e unit, u := es0
is its velo ity value.

10.2.3. Let K be a Galilean referen e frame and use the previous notations.
Re alling 1.5.2, we see that the Galilean referen e frame establishes an isomorphism between the spa etime model (M; I; ; D; b) and the arithmeti spa etime
model. More pre isely, the oordinatization K and the mappings B : I ! R;
d onstitute an isomorphism.
t 7! t s(o) and Z : D ! R; d 7! m

This isomorphism transforms ve tors, ove tors and tensors, otensors, et .

into ve tors, ove tors, et . of the arithmeti spa etime model. In parti ular,
K

: M ! R  R3 ;

x 7!



x  e
u
x 
;
;
s
m2
=1;2;3

is the oordinatization of ve tors; note that it maps E onto f0g  R3 ;

(K 1 ) : M  R  R3 ;

k 7! (k
ei

j i = 0; 1; 2; 3) ;

is the oordinatization of ove tors; note that it maps I
 onto R  f0g:

We an generalize the oordinatization for ve tors ( ove tors)
of type or


A; M ; too, where A
otype A; i.e. for elements in M
A or M
M
A
A
M
M
is a measure line. For instan e,
elements of I or D
D are oordinatized
by the




basis esi j i = 0; 1; 2; 3 and by the basis mei2 j i = 0; 1; 2; 3 ; respe tively:

M ! R  R3 ;
I
M
3
D
D !R R ;


w  e
u
w 
;
w 7! s
;
s
m2
=1;2;3



p  e
u
p 
:
p 7! m2
;
s
m2
=1;2;3

10.3. Subs ripts and supers ripts

10.3.1. In textbooks one generally uses, without a pre ise denition, Galilean
referen e frames and the arithmeti spa etime model. Ve tors, ove tors and
tensors, otensors, et . are given by oordinates relative to a spa etime basis.
Let us survey the usual formalism from our point of view.
Let us take a Galilean referen e system and let us use the previous notations.
If (k0 ; k1 ; k2 ; k3) is the dual of the basis (e0 ; e1 ; e2 ; e3 ); then
xi := ki
x

(i = 0; 1; 2; 3)

are the oordinates of the ve tor x; we know that

x0 :=


x
;
s

e
x
x :=  2
m

( = 1; 2; 3):

The ove tor k has the oordinates

ki := k
ei

(i = 0; 1; 2; 3):

Let us a ept the onvention that the oordinates of ve tors are denoted by
supers ripts and the oordinates of ove tors are denoted by subs ripts, and we
shall not indi ate that the oordinates run from 0 to 3. Then the symbol x  xi
and k  ki will mean that the ve tor x ( ove tor k) has the oordinates xi (ki ):
We have k
x =

3
P

i=0

ki xi : A ording to the Einstein summation rule we shall

omit the symbol of summation as well: k
x = ki xi :

The various tensors are given by oordinates with respe t to the tensor
produ ts of the orresponding bases (e.g. (ei
ej j i; j = 0; 1; 2; 3) or
(ei
kj j i; j = 0; 1; 2; 3)); as the following symbols show:

2 M
M;
L 2 M
M ;
P 2 M
M;
F 2 M
M ;
T

 T ij ;
L  Li j ;
P  Pi j ;
F  Fij :
T

Applying the Einstein summation rule we an write, e.g. T
k  T ij kj ;

L
x  Li j xj ; L
T  Li j T jk ; TrL = Li i ; et .
We know that x
y makes no sense for x; y 2 M; in oordinates this means
that xi yi makes no sense. Similarly, L
k makes no sense for L 2 M
M and
k 2 M ; in oordinates this means that Li j kj makes no sense. More pre isely,
xi yi ; et . does not make an absolute sense. Of ourse, the value of this expression
an be omputed, but it depends on the referen e frame: taking the oordinates
x0 i and y0 i relative to another referen e frame and omputing x0 i y0 i we get a
dierent value.
We an see that, in general, a summation makes an absolute sense only for
equal subs ripts and supers ripts.

10.3.2.

Re all that we have the identi ation E  D

ED and under this
e j  =
identi ation the dual of the orthogonal basis (e1 ; e2 ; e3 ) be omes ( m
2

1; 2; 3): The oordinates of p 2 E are p := p
e ( = 1; 2; 3): If we onsider
E then it has the oordinates p := m2 e2
p
= p
p as an element of D

D
m
( = 1; 2; 3):
e
q ( = 1; 2; 3): If we onsider
Similarly, q 2 E has the oordinates q := m
2

q as an element of E
D
D; then its oordinates are q := m12 (q
e ) = q 
( = 1; 2; 3):
Thus dealing ex lusively with spa elike ve tors, we need not distinguish between supers ripts and subs ripts. We know that q
q makes sense for a spa elike
ve tor q; and q
q  q q = q q = q q :
We emphasize that this is true only if we use an orthogonal and normed basis
in E (see V.3.20).

10.4. Referen e systems asso iated with global rigid observers

10.4.1. We know that the spa e EU of a global rigid observer U is a three-

dimensional ane spa e. Moreover, given to 2 I and qo 2 EU ; or, equivalently,

given o 2 M| alled the origin | su h that o = qo ? to; to =  (o); qo = CU (o)|
we establish the (double) ve torization
I  EU

! I  E;

(t; q) 7! (t to ; q ? to qo ? to )
= (t  (o); q ?  (o) o) ;

whi h is an orientation-preserving ane bije tion.

Then hoosing an s 2 I+ (a positively oriented basis in I)| alled the time
unit | and a positively oriented basis (x1 ; x2 ; x3 ) in E| alled the spa e basis
| , we an establish oordinatizations of time and U -spa e:

T (t) :=

t  (o)

(t 2 I);

SU (q) := (p
(q ?  (o) o) j  = 1; 2; 3) ;
(q 2 EU );
where (p1 ; p2 ; p3 ) is the dual of the basis in question.
Evidently, T and SU are orientation-preserving ane bije tions; we know that
HU is an orientation-preserving smooth bije tion whose inverse is smooth as well
(see 4.3.2 and 4.4.3), thus (U ; T; SU ) is a referen e system. For the orresponding
referen e frame K := (T  SU ) HU we have
K (x) =



(x o) ; (p
(C (x) ?  (o) o))
(x 2 M);
U
=1;2;3
s

!
3
X



1
0
K ( ; ) = CU o +  x ?  (o) +  0 s
=1

( 0 ; ) 2 R  R3 :

T and SU are ane oordinatizations of time and U -spa e. Evidently, K =

(T  SU ) HU is an ane oordinatization of spa etime if and only if HU is an
ane map whi h holds if and only if U is a global inertial observer (see 5.1).

10.4.2. Let us take a uniformly a elerated observer U having the onstant

a eleration value a (see 5.2).
Then, a ording to 5.2.3, for the referen e frame treated in 10.4.1 we have
K (x) =

(x o) ; p

(x o)
U (o)
s
(x 2 M)

1
a(
(x o))2
2
=1;2;3

and
3

X
1

  x +  0 2 s2 a
K 1 ( 0 ; ) = o +  0 sU (o) +
2
=1

(( 0 ; ) 2 R  R3 ):

10.4.3. Let us take a uniformly rotating observer U ; and let o; and

be
the quantities introdu ed in 5.3.
Then, a ording to 5.4.3, for the referen e frame treated in 7.4.1 we have
K (x) =

(x o) ; p
e
s

(
(x o))


(x

o)

=1;2;3

3
X


0
1
0
0

s

K ( ; ) = CU (o) ?  (o) +  s + e

  x
=1

(x 2 M);

( 0 ; ) 2 R  R3 :

10.4.4. Expressing in words we an say:

Galilean referen e system = global inertial observer + measuring time with
respe t to an initial instant and a time unit + introdu ing orthogonal (Cartesian)
oordinates in the observer spa e.
Ane referen e system = global inertial observer + measuring time with
respe t to an initial instant and a time unit + introdu ing (oblique-angled)
re tilinear oordinates in the observer spa e.
Other referen e systems treated previously = global rigid observer + measuring time with respe t to an initial instant and a time unit + introdu ing
re tilinear oordinates in the observer spa e.
For the solution of some pra ti al problems we often use referen e systems in
whi h urvilinear oordinates (e.g. spheri al oordinates or ylindri al oordinates) are introdu ed in the observer spa e.
10.5. Equivalent referen e frames

10.5.1. In textbooks one usually formulates the prin iple | without a pre ise
denition | that the Galilean referen e frames are equivalent with respe t to
the des ription of phenomena. It is very important that then one takes ta itly
into onsideration Galilean referen e frames with the same time unit and the
same distan e unit.
Referen e frames as we dened them are mathemati al obje ts. The physi al
obje t modelled by them will be alled here a physi al referen e frame. When
ould we onsider two physi al referen e frames to be equivalent? The answer

is: if the experiments prepared in the same way in the referen e frames give the
same results. Let us see some illustrative examples.
Take two physi al Galilean referen e frames in whi h the time units and distan e units are dierent and perform the following experiment in both systems:
let an iron ball of unit diameter moving with unit relative velo ity hit perpendi ularly a sheet of glass of unit width. It may happen that the ball boun es
in one of the referen e frames, the glass breaks in the other. The two referen e
frames are not equivalent.
Take an ane referen e frame in whi h the rst spa e basis element is perpendi ular to the other two basis elements; take another ane referen e frame
in whi h the rst spa e basis element is not perpendi ular to the other two basis
elements. Perform the following experiment in both frames: let a ball moving
parallelly to the rst spa e axis hit a plane parallel to the other two axes. The
ball returns to its initial position in one of the referen e frames and does not in
the other. The two referen e frames are not equivalent.

10.5.2. Re all the notion of automorphisms of the spa etime model (1.5.4).
An automorphism is a transformation that leaves invariant (preserves) the stru ture of the spa etime model. Stri t automorphisms do not hange time periods
and distan es.
It is quite natural that two obje ts transformed into ea h other by a stri t
automorphism of the spa etime model are onsidered equivalent (i.e. identi al
from a physi al point of view).
In the next paragraph we shall study the Noether transformations that involve
the stri t automorphisms of the spa etime model. Now we re all the basi fa ts.
Let SO(b) denote the set of linear maps R : E ! E that preserve the
Eu lidean stru ture and the orientation of E : b (R  R) = b and detR = 1
(see 11.1.2).
Let us introdu e the notation
N +! := fL : M ! M j L is ane, 
L =  ; LjE 2 SO(b)g
and let us all the elements of N +! proper Noether transformations. It is quite
evident that (L; idI ; idD ) is a stri t automorphism of the spa etime model if and
only if L is a proper Noether transformation (11.6.4).
An ane map tiL : I ! I an be assigned to every proper Noether transformation L in su h a way that  L = (tiL)  (see 11.6.3).
10.5.3. Denition. The referen e frames K and K 0 are alled equivalent if

there is a proper Noether transformation L su h that

K 0 L = K:
Two referen e systems are equivalent if the orresponding referen e frames
are equivalent.

Proposition.

Let (U ; T; SU ) and (U 0 ; T 0; SU 0 ) be the referen e systems

orresponding to the referen e frames K and K 0 ; respe tively. If K and K 0
are equivalent, K 0 L = K; then
(i ) L 1
U 0 L = U ; in other words, L
U = U 0 L;
(ii ) T 0 (tiL) =
T; in other words, T 0 1 T = tiL:
(iii ) SU 10 SU CU = CU 0 L:
Proof. (i ) K = K 0 L; K 1 = L 1 K 0 1 and 
L =  together with 10.1.4
imply
U (x) =

0 K 1 (K (x))
=

0 K 1 (K (x))

L 1
0 K 0 1 (K 0 (L(x)))

L 1
0 K 0 1 (K 0 (L(x)))

= L 1
U 0 (L(x)):

(ii ) The equalities

T  = pr0 K = pr0 K 0 L = T 0  L = T 0 (tiL) 

yield the desired relation immediately.
(iii ) Consider the equalities

SU CU = pr K = pK 0 L = SU 0 CU 0 L:

10.5.4. Now we shall see that our denition of equivalen e of referen e frames
is in a ordan e with the intuitive notion expounded in 10.5.1.
Proposition. Two Galilean referen e frames are equivalent if and only if
they have the same time unit and distan e unit, respe tively.
Proof.0 Let the Galilean referen e frames K and K 0 be dened
by the origins
o and o and the spa etime bases (e0 ; e1 ; e2 ; e3 ) and (e00 ; e01 ; e02 ; e03 );
respe tively.
Then L := K 0 1 K : M ! M is the ane bije tion determined by
L(o) = o0 ;

L
ei

= e0i

(i = 0; 1; 2; 3):

Evidently, L is orientation-preserving. Moreover, 
L =  if and only if

= 
e00 ; and LjE 2 SO(b) if and only if je j = je0 j ( = 1; 2; 3):


e0

10.6. Exer ises

1. Referen e frames are oordinatizations, hen e we an apply all the notions
introdu ed in VI.5, e.g. the oordinatized form of ve tor elds.

Let U be the observer orresponding to the referen e frame K:

Demonstrate that the oordinatized form of U a ording to K is the onstant
mapping (1; 0): (U is a ve tor eld of otype I; hen e by denition, (DK
U )K 1
is its oordinatized form a ording to K:)
2. Take a uniformly a elerated observer U having the a eleration value
a 6= 0: Fix s 2 I+ ; m 2 D+ and dene a Galilean referen e frame K with
an arbitrary origin and with a spa etime basis su h that e0 := sU (o); e1 :=
m jaaj ; e2 and e3 are arbitrary. Demonstrate that, a ording to K; U has the
oordinatized form
( 0 ;  1 ;  2 ;  3 ) 7! (1;  0 ; 0; 0);
where  is the number for whi h jaj =  m
s2 holds.
The U -line passing through o +


t;  1 +  0 (t

3
P
 i ei

i=0

1
 0 ) + (t
2

be omes

 0 )2 ;  2 ;  3

j t2R :

3. Take a uniformly rotating observer U having the angular velo ity

and
suppose there is an inertial U -spa e point qo = o +
I: Fix s 2 I+ ; m 2 D+
and dene a Galilean referen e frame with o; e0 := sU (o); e3 positively oriented
in Ker
; je3 j = m; e1 and e2 being arbitrary. Demonstrate that, a ording to
K; U has the oordinatized form
( 0 ;  1 ;  2 ;  3 ) 7! (1; ! 2 ; ! 1 ; 0);
where ! is the number for whi h j
j = ! s1 holds.
The U -line passing through o +
n

3
P
 i ei

i=0

be omes

t;  1 os !(t  0 )  2 sin !(t  0 );

 1 sin !(t  0 ) +  2 os !(t  0 );  3 j t 2 R :

4. Prove that two ane referen e frames are equivalent if and only if they
have the same time unit and the orresponding elements of the spa e bases have
the same length and the same angles between themselves; in other words, the
ane referen e frames dened by the origins o and o0 and the spa etime bases
(x0 ; x1 ; x2 ; x3 ) and (x00 ; x01 ; x02 ; x03 ); respe tively, are equivalent if and only
if

x0 = 
x00

and

x
x

= x0
x0

(;  = 1; 2; 3):

5. Prove that two referen e frames dened for uniformly a elerated observers
in the form given in 10.4.2 are equivalent if and only if the two a eleration values
have the same magnitude, the time units are equal, the orresponding elements of
the spa e bases have the same length and the same angles between themselves,
and the a eleration values in line in the same way to the basis elements; in
other words, if a and a0 are the a eleration values, s and s0 are the time units,
(x1 ; x2 ; x3 ) and (x01 ; x02 ; x03 ) are the spa e bases, then the two referen e
systems are equivalent if and only if

jaj = ja0 j; s = s0 ;
x0
a0
x
a
= 0 0 (;  = 1; 2; 3):
x
x = x0
x0 ;
jx jjaj jx jja j
6. Prove that two referen e frames dened for uniformly rotating observers in
the form given in 10.4.3 are equivalent if and only if the angular velo ities have
the same magnitude, the time units are equal, the orresponding elements of the
spa e bases have the same length and the same angles between themselves and
the oriented kernels of the angular velo ities in line in the same way to the basis
elements.
7. Take a global inertial observer and onstru t a referen e system by spheri al
( ylindri al) oordinatization of the observer spa e. Find ne essary and su ient
onditions that two su h referen e systems be equivalent.
8. In all the treated referen e systems time is oordinatized by an ane map.
Constru t a referen e system based on a global inertial observer in whi h the
time oordinatization is not ane.

11. Spa etime groups

11.1. The three-dimensional orthogonal groups

11.1.1. (E; D; b) is a three-dimensional oriented Eu lidean ve tor spa e.

Re all the notations (see V.2.7)



A(b) := A 2 E
E j A> = A 


O(b) := R 2 E
E j R> = R 1 :

E ^ E;
D D

A(b) is a three-dimensional subspa e in E

E and O(b) is a three-dimensional
Lie group having A(b) as its Lie algebra (VII.5).

11.1.2. We know that jdetRj = 1 for R 2 O(b) (see V.2.8). We introdu e

the notations
SO(b) :=O(b)+ := fR 2 O(b) j detR = 1g ;
O(b) := fR 2 O(b) j detR = 1g :
The elements of SO(b) are alled rotations.
Sin e the determinant is a ontinuous fun tion, O(b)+ and O(b) are disjoint.
Evidently, idE 2 O(b)+ and idE 2 O(b) ; moreover, ( idE )
O(b)+ =
O(b) :
The determinant is a ontinuous fun tion, hen e both O(b)+ and O(b)
are losed. Moreover, we know that F 7! Tr(F >
F ) is an inner produ t
(real-valued positive denite bilinear form) on E
E (see V.2.10). Sin e
Tr(R>
R) = Tr(idE ) = 3 for all R 2 O(b); O(b) is a bounded set.
Thus we an state, that O(b); O(b)+ and O(b) are ompa t ( losed and
bounded) sets.
11.1.3. Let R 2 SO(b): For all x 2 E we have jR
xj = jxj: As a onsequen e,

R
x = x

implies  = 1:
Proposition. For every R 2 SO(b) there is a non-zero x 2 E su h that
R
x = x; moreover,
aR := fx 2 E j R
x = xg
is a one-dimensional linear subspa e if and only if R 6= idE :
Proof. It is trivial that aR = E for R = idE :
IV.3.18 and V.1.5 result in
det(R idE ) = det(R R>
R) = det(idE R> )detR = det(R idE ):
Consequently, det(R idE ) = 0; R idE is not inje tive, there is a non-zero
x su h that (R idE )
x = 0:
Let us suppose aR is not one-dimensional, i.e. x1 and x2 are not parallel
ve tors su h that R
x1 = x1 and R
x2 = x2 : Then for every element x in
the plane spanned by x1 and x2 we have R
x = x: This means that the plane
spanned by x1 and x2 is invariant for R and the restri tion of R onto that plane
is the identity. Let y be a non-zero ve tor orthogonal to the plane spanned by x1
and x2: Sin e R preserves orthogonality, R
y must be orthogonal to that plane,
i.e. it is parallel to y : R
y = y: R is orientation-preserving, thus R
y = y
must hold. This means that R = idE :
For R 6= idE ; aR is alled the axis of rotation of R:

11.1.4. For R 2 SO(b) the symbol a?R will stand for the orthogonal omplement of aR :
a?R := fx 2 E j x is orthogonal to aR g :

?
?
Evidently, a?
R = f0g for R = idE and aR is a plane for R 6= idE : Moreover, aR
is invariant for R:
The restri tion of R 6= idE onto a?
R is a rotation in a plane whi h \evidently"
an be hara terized by an angle of rotation. This is the ontent of the following
proposition.
Proposition. If x and y are non-zero ve tors in a?R then
x
R
x
jxj2

y
R
y
jy j2 :

Proof.

We an ex lude the trivial ases R
x = x and R
x = x for all

(note that the rst ase is R = idE ):
It will be onvenient to put n := jxxj ; k := jyyj and to onsider R to be a linear
E : Let us introdu e the notation
map on D
x 2 a?
R

SR :=

E jn
n2
D

is orthogonal to aR ;

j nj = 1 :

The proof onsists of several simple steps whose details are left to the reader.
(i ) Let n and k be elements of SR orthogonal to ea h other. Then, ex luding
the trivial ase,
n
R
k 6= 0:
Indeed,
1 = detR = (n
R
n)(k
R
k) (n
R
k)(k
R
n)
and be ause of the Cau hy inequality (apart from the trivial ase), (n
R
n)(k

R
k) < 1:
(ii ) R
n 6= R 1
n: Indeed, suppose R
n = R 1
n: Then we get from the
previous formula that
1 = (n
R 1
n)(k
R
k) (n
R
k)(k
R 1
n) =
= (n
R
n)(k
R
k) (n
R
k)(n
R
k);
whi h implies (n
R
n)(k
R
k) > 1 ontradi ting the Cau hy inequality.
(iii ) 0 6= R
n R 1
n is orthogonal to n; hen e it is parallel to k:
(iv ) R
n + R 1
n is orthogonal to R
n R 1
n; hen e it is orthogonal to
k as well. Consequently,
n
R
k+k
R
n=

0:

(v ) R
n = (n
R
n)n + (k
R
n)k; and from a similar relation for R
k we

have
0 = (R
n)
(R
k) = (n
R
n)(n
R
k) + (k
R
n)(k
R
k);

then we infer from (i ) and (iii ) that n
R
n = k
R
k:

(vi ) If m 2 SR then m = n +  k; 2 +  2 = 1 and m
R
m = n
R
n:
Now let us return to 0 6= x 2 E; orthogonal to aR : The Cau hy inequality
gives jx
R
xj  jxj2 ; thus

R := ar os

x
R
x
jxj2

2 [0; 

is meaningful, whi h is alled the angle of rotation of R:

Observe that
| R = 0 if and only if R
x = x for all x orthogonal to aR ; i.e. R = idE ;
| R =  if and only if R
x = x for all x orthogonal to aR :

11.1.5. Proposition. Let R 6= idE and R 6= : Take an arbitrary nonzero x 2 a?R : Let y 2 a?
R be orthogonal to x; jy j = jxj; and suppose (x; y ) and
(x; R
x) are equally oriented bases in a?
R : Then
R
x = ( os R )x + (sin R )y :

Proof
.


Sin e R
x = xj
xRj2
x x + yj
yRj2
x y; we easily nd that os2 R +

y
R
x 2 = 1: As a onsequen e of the equal orientation of (x; y ) and (x; R
x);
jyj2
we have yjx
Rjjy
xj > 0 whi h implies that this expression equals sin R (be ause R
is between 0 and ):

11.1.6. Let R 6= idE and R 6= : Then x and R
x are linearly independent

if x is a non-zero ve tor orthogonal to aR : It is not hard to see that if y is
another non-zero ve tor orthogonal to aR ; then the pairs (x; R
x) and (y; R
y)
are equally oriented bases in a?
R : As a onsequen e, (R
x) ^ x and (R
y ) ^ y
are positive multiples of ea h other.
Sin e j(R
x) ^ xj2 = jxj4 (x
R
x)2 = jxj4 sin2 R ; we have that for
R 6= idE ; R 6= 
log R :=

(R
x) ^ x
jxj2 sin R R 2 A(b)

is independent of x: Moreover, put

log(idE ) := 0 2 A(b):

It is easy to see that

(i ) Ker (log R) = aR ;
(ii ) j log Rj = R ;

(0 6= x 2 a?
R)

(iii ) if R 6= idE then for an arbitrary non-zero x

(x; (log R)
x) form equally oriented bases in a?
R:
In this way, assuming the notations
N := fR 2 SO(b) j R 6= g ;

2 a?R ;

(x; R
x) and

P := fA 2 A(b) j jAj < g

we dened a mapping log : N ! P; we shall show that log is a bije tion whose
inverse is the restri tion of the exponential mapping (see VII.3.7).

11.1.7. Proposition. For 0 6= A 2 A(b) putting  := jAj; Ao := jAAj ; we

have

eA = Ao 2 os  + Ao sin  + idE + Ao 2 :

Proof. Re all that A3 =

eA =

1 An
X

n=0

n!

2 A (see V.3.10); thus

=idE + A +


A2

A2

2!

A3

3!

A2

A4

4!

A2

A5

A6

A7

+ :::: =
6!
7!
2 A2 4 A2
+
::::
4!
6!

5!

= idE + 2
+

2
2!
4
2
 A  A 6 A
+
+ ::::
+A
3!
5!
7!
whi h yields the desired result by A = Ao :
Note that for A 6= 0; idE + A2o is the orthogonal proje tion onto the plane
orthogonal to the kernel of A:
As a onsequen e, if A 6= 0 then
(i ) eA
x = x for x 2 Ker A (the axis of rotation of eA is the kernel of A);
(ii ) eA
x = ( os )x + (sin )Ao
x for x orthogonal to Ker A (the angle of
rotation of eA is  := jAj);

11.1.8. Proposition. For R 2 N

elog R = R
and for A 2 P

log(eA ) = A:

Proof. Evidently, for R = idE and for A = 0 the equalities hold.

If R 6= idE and x is in aR then, obviously, eln R
x = x = R
x: If x is
orthogonal to the axis of rotation of R; then
elog R
x = ( os R )x + sin R

log R

x = R
x;
R

in view of 11.1.6 and 11.1.7.

A ording to the previous proposition, for A 6= 0; the axis of rotation of
eA is the kernel of A; the angle of rotation of eA is jAj: Thus if x 2 Ker A
then log(eA )
x = 0 = A
x: If x is orthogonal to the kernel of A; then
A x)^x
log(eA ) = j(exj2
sin
jAj and an easy al ulation based on the formula in 11.1.6
yields that log(eA )
x = A
x:

11.1.9. It is trivial that the losure of N is SO(b): It is not hard to see

that exponential mapping A(b) ! SO(b) maps the losure of P onto SO(b):
However, the exponential mapping on the losure of P is not inje tive: if jAj = 
then eA = e A :
Sin e the losure of P is onne ted and the exponential mapping is ontinuous,
SO(b) is onne ted as well.
(However, SO(b) is not simply onne ted: it is homeomorphi to a set whi h
is obtained from the losure of P by \sti king" together the diametri al points
of the boundary of P.)
The one-parameter subgroup of SO(b) orresponding to A 2 A(b) is R !
SO(b); t 7! etA : If A 6= 0; then all the elements of the one-parameter subgroup
are rotations around the same axis Ker A:
Sin e the exponential mapping is surje tive, every element of SO(b) is in a
one-parameter subgroup.
11.1.10. In physi al appli ations we meet A(Ib) instead of A(b): If
2 A(Ib) ;
then we an give a fun tion R : I ! SO(b); t 7! e(t t )
; where to is a xed
o

element of I: Then every value of su h a fun tion is a rotation around the same
axis; the angle of rotation of R(t) is (t to )j
j: Thus j
j is interpreted as the
magnitude of the angular velo ity and
itself as the angular velo ity of the
rotation.
We know that R is dierentiable, R_ =

R; from whi h we infer that

= R_
R 1 :

In general, onsider a dierentiable fun tion R : I SO(b)  E

E : Its
derivative at t; R_ (t); is a linear map from I into E
E that takes values in the
tangent spa e of SO(b) at R(t) whi h is R(t)
A(b) = fR(t)
A j A 2 A(b)g
(see VII.3.3). In other words,R_ (t) 2 R(t)
IA(b) ; i.e. R(t) 1
R_ (t) 2 A(b): Then
V.2.11(ii ) implies that R(t)
R(t) 1
R_ (t)
R(t) 1 is in A(b) as well;

(t) := R_ (t)
R(t) 1 2 A(b)

is alled the angular velo ity value at t; and the fun tion
: I
angular velo ity.

 A(b) is the

Evidently, R is the solution of the dierential equation

(X : I

 SO(b))?

X_ =

X:

11.2. Exer ises

1. Let us oordinatize SO(b) by the Euler angles as follows.

E : If R
n3
Let (n1 ; n2 ; n3 ) be a positively oriented orthonormal basis in D
n

(
R


n
)
is not parallel to n3 ; put n := jn33 (R
n33 )j and
#R := ar os(n3
R
n3 );
R :=sign(n
n2 ) ar os(n
n1 );
'R :=sign(n
R
n2 ) ar os(n
R
n1 )
where signx := jxxj if 0 6= x 2 R and sign0 := 1:
Prove that if Ri denotes the one-parameter subgroup of rotations around ni
(i = 1; 2; 3) then
R = R3 ('R )
R1 (#R )
R3 ( R ):

2. Let R : I SO(b) be a dierentiable fun tion and put R 1 : I SO(b);

t 7! R(t) 1 : Using R
R 1 = idE prove that R 1 is also dierentiable and


R 1
= R 1
R_
R 1:

3. Prove that for 0  r 2 D; fx 2 E j jxj = rg is an orbit of SO(b) and all

its orbits are of this kind.

11.3. The Galilean group

11.3.1. We shall deal with linear maps from M into M; permanently using
the identi ation Lin(M)  M
M: The restri tion of a linear map L : M ! M
(L 2 M
M) onto E equals L
i where i : E ! M (i 2 M
E) is the anoni al
embedding. The symbol L
i 2 i
O(b) means that the restri tion of L onto E
is in O(b); i.e. there is an R 2 O(b)  E
E su h that L
i = i
R:

First we dene the Galilean group and then studying it we nd its physi al

meaning.

Denition.

G := fL 2 M
M j 
L =  ; L
i 2 i
O(b)g

is alled the Galilean group ; its elements are the Galilean transformations.
If L is a Galilean transformation then
arL :=
is the arrow of L and
signL :=

+1 if
L = 
1 if
L = 

+1 ifL
i 2 i
O(b)+
1 ifL
i 2 i
O(b)

is the sign of L:
Let us put

G +! :=fL 2 G j
G + :=fL 2 G j
G ! :=fL 2 G j
G :=fL 2 G j

signL = arL = 1g;

signL = arL = 1g;
signL = arL = 1g;
signL = arL = 1g:

G +! is alled the proper Galilean group.

(i ) The ondition 
L =  implies that E is invariant for the linear map
L : M ! M:
(ii ) The ondition L
i 2 i
O(b) means that there is a (ne essarily unique) RL
in O(b) su h that
L
i = i
RL :
(iii ) The Galilean transformations are linear bije tions: if L
x = 0 then 
x = 0;

i.e. x is in E; the restri tion of L onto E is inje tive, thus x = 0:

(iv ) It is quite trivial that G is indeed a group: the produ t of its elements as
well as the inverse of its elements are Galilean transformations.

11.3.2. Proposition.

having the Lie algebra

The Galilean group is a six-dimensional Lie group

La(G ) = fH 2 M
M j 
H = 0; H
i 2 A(b)g:
Proof. A ording to the previous remark, G is a subgroup of G `(M) whi h
is sixteen-dimensional.
We have to show that the Galilean group is a six-dimensional smooth submanifold of G `(M):
Observe that if L 2 G ; then
u
L

i = RL

for all u 2 V(1); where RL is given in the previous remark.

Thus we an give


u : G `(M) ! (I
M )  S(b);
L 7! 
L; (u
L
i)>
(u
L
i)

whi h is evidently a smooth map and G is the preimage of f( ; idE )g by u

(S(b) := fS 2 E 2
E j S > = S g is a six-dimensional linear subspa e).
The derivative of u at L is the linear map
Du (L) : M
M ! (I
M )  S(b);


H 7! 
H ; (u
H
i)>
(u
L
i) + (u
L
i)>
(u
H
i)
whi h is surje tive: (h; T ) 2 (I
M )  S(b) is the image by Du (L) of
> 1
T:
u
h + (1=2)(u
L
i)
Thus, being a six-dimensional submanifold in G `(M); the Galilean group is a
Lie group; its Lie algebra is Ker Du (idM):
If Du (idM)(H ) = 0; then 
H = 0; and u
H
i is in A(b): Sin e the
rst ondition means that H 2 E
M ; we have u
H
i = H
i: Hen e the
kernel of Du (idM) is the linear subspa e given in our proposition.

11.3.3. The mappings G ! f 1; 1g; L 7! arL and G ! f 1; 1g; L 7! signL

are ontinuous group homomorphisms. As a onsequen e, the Galilean group
is dis onne ted. We shall see in 11.4.3 that the proper Galilean group G +! is
onne ted. It is quite trivial that if L 2 G + then L
G +! = G + and similar
assertions hold for G ! and G as well. Consequently, the Galilean group has
four onne ted omponents, the four subsets given in Denition 11.2.1.
From these four omponents only G +! | the proper Galilean group | is a
subgroup; nevertheless, the union of an arbitrary omponent and of the proper
Galilean group is a subgroup as well.
G ! := G +! [ G ! is alled the ortho hronous Galilean group.
If L 2 G ; then L preserves or reverses the \orientation" of timelike ve tors
a ording to whether arL = 1 or arL = 1 :
if arL = 1
then L (T! ) = T! ; L (T ) = T ;
if arL = 1
then L (T! ) = T ; L (T ) = T! :

Moreover, L preserves or reverses the orientation of E a ording to whether

signL = 1 or signL = 1:
The orientation of E given in 1.2.4 shows that the elements of G +! and G
preserve the orientation of M; whereas the elements of G + and G ! reverse
the orientation.

11.3.4. M is of even dimension, thus idM is orientation-preserving. Evidently, idM is in G ; it is alled the inversion of spa etime ve tors. We have
that G = ( idM )
G +! :

We have seen previously that the elements of G + invert in some sense the
timelike ve tors and do not invert the spa elike ve tors; the elements of G !
invert in some sense the spa elike ve tors and do not invert the timelike ve tors.
However, we annot sele t an element of G + and an element of G ! that we
ould onsider the time inversion and the spa e inversion.
For ea h u 2 V(1) we an give a u-timelike inversion and a u-spa elike
inversion as follows.
The u-timelike inversion Tu 2 G + inverts the ve tors parallel to u and leaves
invariant the spa elike ve tors:
Tu
u :=

Tu
q

and

:= q for q 2 E:

In general,
Tu
x =

u(
x) + u
x =

i.e.

(x 2 M)

2u(
x) + x

= idM 2u
 :
The u-spa elike inversion Pu 2 G ! inverts the spa elike ve tors and leaves
invariant the ve tors parallel to u :
Tu

Pu
u := u

In general,
i.e.

Pu
q

and

Pu
x = u(
x)
Pu

:= q

for

u
x = 2u(
x)

= 2u


q
x

2 E:
(x 2 M);

idM :

We easily dedu e the following equalities:

Tu 1

Pu 1
Tu = Pu ;
Tu
Pu = Pu
Tu =

= Tu;

= Pu ;
idM:

11.3.5. The three-dimensional orthogonal group is not a subgroup of the

Galilean group: O(b) annot be a subgroup of G be ause the elements of G are
linear maps dened on M whereas the elements of O(b) are linear maps dened
on E (E
E is not a subset of M
M ):
It is quite obvious that
G ! ! O(b);

L 7! RL

(where L
i = i
RL ) is a surje tive Lie group homomorphism.

For every u 2 V(1);

O(b)u := fL 2 G ! j L
u = ug ;
alled the group of u-spa elike orthogonal transformations, is a subgroup of G ! ;
the restri tion of the above Lie group homomorphism to O(b)u is a bije tion
between O(b)u and O(b):
Indeed, if L
u = u and RL = idE then L is the identity on the omplementary
subspa es u
I and E; thus L = idM : the group homomorphism from O(b)u
into O(b) is inje tive.
If R 2 O(b) then
Ru := u
 + R
u

is a Galilean transformation in O(b)u and u
Ru
i = R (re all that u
L
i =

RL for all Galilean transformations L) : the group homomorphism from O(b)u
onto O(b) is surje tive.

11.3.6. The kernel of the surje tion G ! ! O(b); i.e.

V := fL 2 G ! j

RL

= idE g = fL 2 G ! j L
i = ig

is alled the spe ial Galilean group. Observe that V is in G +! :

The spe ial Galilean group is a three-dimensional Lie group having the Lie
algebra
La(V ) = fH 2 M
M j 
H = 0; H
i = 0g :

Proposition. If L 2 V ; then there is a unique vL 2 EI su h that

L
x = vL (
x) + x

i.e.

L = idM + vL
 :

(x 2 M);

The orresponden e V ! EI ; L 7! vL is a bije tive group homomorphism

regarding the additive stru ture of EI (i.e. vL
K = vL + vK for all L; K 2 V ):
Proof. Let L be an element of V : Let us take an arbitrary u 2 V(1) and put
vL := L
u u: We laim that vL does not depend on u: Indeed, if u0 2 V(1)
then
(L
u u) (L
u0 u0 ) = L
(u u0) (u u0 ) = 0;
be ause L
(u u0 ) = u u0 : Moreover, 
(L
u u) = 0; thus vL is in EI :
This means that L
u = u + vL for all u 2 V(1):

Then we nd that for x 2 M

L
x = L
(u(
x) + u
x) = (u + vL )
x + u
x = vL (
x) + x:

This formula assures, too, that L 7! vL is a group homomorphism.

If vL = 0 then L
u = u for all u 2 V(1) implying L = idM ; thus the
orresponden e from V into EI is inje tive. Evidently, if v is in EI then idM + v

is a spe ial Galilean transformation: the orresponden e is surje tive.
In view of our result, the spe ial Galilean group is a three-dimensional ommutative group.

11.3.7. (i ) If u; u0 2 V(1); then the spe ial Galilean transformation

L(u0 ; u) := idM + (u0

u)
 ;

i.e. the one orresponding to vu0 u = u0 u is the unique one with the property
L(u0 ; u)
u = u0 :

Let us re all the splitting of M a ording to u and u0 ; then we easily nd

that
L(u0 ; u) = hu0 1
hu :
(ii ) The produ t of the u0 -timelike inversion and the u-timelike inversion is a
spe ial Galilean transformation:
Tu0
Tu

= (idM

2u0
 )
(idM

2u
 ) = idM + 2vu0u
 :

We know that Tu 1 = Tu = Pu ; then we an assert that

Tu0
Tu 1

= Pu0
Pu 1 =L(u0 ; u)2 =

=L(u + 2vu0u ; u) = L(u 2vuu0 ; u):

11.3.8. Originally the Galilean tranformations are dened to be linear maps

from M into M: In the usual way, we an onsider them to be linear maps from
M into M as we already did in the pre eding paragraphs as well.
I
I
V(1) is invariant under ortho hronous Galilean transformations. Moreover,
the restri tion of an ortho hronous Galilean transformation L onto V(1) is an
ane bije tion whose underlying linear map | whi h is the restri tion of L onto
E | preserves the Eu lidean stru ture.
I
Conversely, if F is a Eu lidean transformation of V(1)| an ane bije tion
whose underlying linear map preserves the Eu lidean stru ture | then M ! M;

x 7! F
(x=
x)
x is an ortho hronous Galilean tranformation whose restri tion

onto V(1) oin ides with F :

Thus we an state that the ortho hronous Galilean group is anoni ally isomorphi to the group of Eu lidean transformations of V(1):

11.4. The split Galilean group

11.4.1. The Galilean transformations, being elements of M

M; are split
by velo ity values a ording to 8.1.1. Sin e 
L = (arL) and u
L
i = RL

for a Galilean transformation L and for u 2 V(1); we have

hu
L
hu 1 =

arL
0 :
L
u (arL)u RL


Writing L
u (arL)u = (arL) ((arL)u u) ; we see that the following

denition des ribes the split form of Galilean transformations.
Denition. The split Galilean group is


0j

 2 f 1; 1g;

2 EI ;

b)

R 2 O(

Its elements are alled split Galilean transformations.

The split Galilean transformations an be regarded as linear maps I  E !
I  E; the one in the denition makes the orresponden e
(t; q) 7! (t; vt + R
q):

The split Galilean group is a six-dimensional Lie group having the Lie algebra


0 0j
v A

E
v2 ;
I

A 2 A(

b)

11.4.2. The splitting hu a ording to u establishes a Lie-group isomorphism

between the Galilean group and the split Galilean group. The isomorphisms
orresponding to dierent u0 and u are dierent.
The EI omponent in the split form of Galilean transformations, in general,
varies a ording to the velo ity value establishing the splitting.
The following transformation rule shows well how the splitting depends on
the velo ity values.
Let u0 ; u 2 V(1): Re all the notation
Hu0 u := hu0
hu 1 =

vu0 u

idE :

Then
Hu0 u

0 
Hu01u = 

0 :


vu0 u ) + R
vu0 u

(v

11.4.3. The splittings send the proper Galilean group into




1 0 j v 2 E ; R 2 SO(b)
v R
I
whi h is evidently a onne ted set. Sin e the splittings are Lie group isomorphisms, G +! is onne ted as well.

11.4.4. If L is a spe ial Galilean transformation and vL is the orresponding

element of EI ; then L has the split form


1 0
vL idE

for all u 2 V(1) : the splitting is independent of the velo ity value. In other
words, every u 2 V(1) makes the same bije tion between the spe ial Galilean
group V and the group



E
v2
I :

0 j
idE

Observe that for all u0 ; u 2 V(1); the ve tor transformation law is the split
form of a spe ial Galilean transformation:
Hu0 u = hu0
hu 1 = hu
L(u; u0 )
hu 1 :

11.4.5. The Lie algebra of the Galilean group, too, onsists of elements of
M
M; thus they are split by velo ity values in the same way as the Galilean
transformations; evidently, their split forms will be dierent.
If H is in the Lie algebra of the Galilean group and u 2 V(1); then
hu
H
hu 1 =

H
u H

The splitting a ording to u establishes a Lie algebra isomorphism between

the Lie algebra of the Galilean group and the Lie algebra of the split Galilean
group. The isomorphisms orresponding to dierent u0 and u are dierent:

0
0
Hu0 u


Hu0u
v H


1=

v + H
vu0 u

0 :

11.5. Exer ises

Prove that for all t 2 I; fx 2 M j 
x = tg is an orbit of the spe ial

1.
Galilean group and all orbits are of this form. The orbits of the spe ial Galilean
group and the orbits of the proper Galilean group oin ide. What are the orbits
of the (ortho hronous) Galilean group?
2. Beside the trivial linear subspa es f0g and M there is no subspa e invariant
for all the spe ial Galilean transformations.
3. The transpose of a Galilean transformation is a linear bije tion M !

M : Demonstrate that the transposed Galilean
group fL j L 2 Gg leaves I


invariant; more losely, if L 2 G and e 2 I
 ; then L
e = (arL)e:
Furthermore, if k 2 M ; and L
k is parallel to k for all Galilean transformations L; then k is in I
 :
4. The subgroup generated by fTu j u 2 V(1)g is the spe ial Galilean group.
5. Prove that
hu
Tu
hu 1 =

idE ;

hu
Pu
hu 1 =

idE :

Find hu0
Tu
hu0 1 and hu0
Pu
hu0 1 :

6. The u-splitting of the Galilean group sends the spe ial Galilean group into
the group



1 0 j v2 E
v idE
I
whose Lie algebra is

0 0 j v 2 E :
v 0

The u-splitting of spe ial Galilean transformations does not depend on u:

7. The Lie algebra of O(b)u equals

fH 2 La(G ) j

0g :

H
u=

8. The u-splitting sends the subgroup O(b)u into the group



having the Lie algebra



0j

0j

b)

R 2 O(

0 A A 2 A(b)
Find the u0 -splitting of O(b)u for u0 6= u:
0

9. Re all the notation introdu ed in 11.3.7 and prove that

| L(u0 ; u) 1 = L(u; u0);
| L(u00 u0 )
L(u0 ; u) = L(u00 ; u):
10. For all u 2 V(1) and for all Galilean transformations L we have that
R(L; u) := (arL)L(u; (arL)
u)
L = (arL)L + (u

(arL)L
u)


is in O(b)u and R (L; u)jEu = RL : In other words, given an arbitrary u 2 V(1);

every Galilean transformation L is the produ t of a spe ial Galilean transformation and a u-spa elike orthogonal transformation, multiplied by the arrow of
L:
L = (arL)L((arL(u); u)
R(L; u):

11.6. The Noether group

11.6.1. Now we shall deal with ane maps L : M ! M; as usual, the linear
map under L is denoted by L:
Denition.

N := fL : M ! M j L

is ane, L 2 Gg

is alled the Noether group ; its elements are the Noether transformations.
If L is a Noether transformation, then
arL := arL;

signL := signL:

N +! ; N + ; N

! and N
are the subsets of N onsisting of elements
whose underlying linear maps belong to G +! ; G + ; G ! and G ; respe tively.
N +! is alled the proper Noether group .
The Noether group is the ane group over the Galilean group; a ording to
VII.3.2(ii ), we an state the following.
Proposition. The Noether group is a ten-dimensional Lie group; its Lie
algebra onsists of the ane maps H : M ! M whose underlying linear map is
in the Lie algebra of the Galilean group:

La(N ) = fH 2 A(M; M) j 
H = 0;

H

i 2 A(b)g :

The proper Galilean group is a onne ted subgroup of the Galilean group. As
regards N + ; et . we an repeat what was said about the omponents of the
Galilean group.
N ! := N +! [ N ! is alled the ortho hronous Noether group.

11.6.2. We an say that the elements of N invert spa etime in some sense
but there is no element that we ould all the spa etime inversion.
For every o 2 M we an give the o- entered spa etime inversion in su h a way
that rst M is ve torized by Oo ; then the ve tors are inverted ( idM is applied),
nally the ve torization is removed :
Io := Oo 1 (idM ) Oo ;
i.e.

(x 2 M):

Io (x) := o (x o)

Similarly, we an say that in some sense the elements of N ! ontain spa elike
inversion and do not ontain timelike inversion; the elements of N + ontain
timelike inversion and do not ontain spa elike inversion. However, the spa e
inversion and the time inversion do not exist.
For every o 2 M and u 2 V(1) we an give the o- entered u-timelike inversion
and the o- entered u-spa elike inversion as follows:

Tu;o(x) := o + Tu
(x o);

Pu;o (x) := o + Pu
(x o)

(x 2 M):

11.6.3. Let L be a Noether transformation. If x and y are simultaneous then

L(x) and L(y) are simultaneous as well:
 (L(x)) L(y)) = 
L
(x y) = (arL)
(x y) = 0:

Re all that I is identied with the set of hyperplanes of M; dire ted by E:

Thus for a Noether transformation L we an dene the mapping
tiL : I ! I;
Observe that
or, in other words,

t 7! L[t:

(tiL)  =  L
(tiL)(t) =  (L(x))

(x 2 t);

from whi h we get immediately that

(tiL)(t) (tiL)(s) = (arL)(t s)
Thus tiL is an ane map over (arL)idI :

(t; s 2 I):

A ording to Exer ises VI.2.5.6{7, if arL = 1; then tiL is a translation, i.e.

there is a unique t 2 I su h that (tiL)(t) = t + t; if arL = 1; then tiL is an
inversion, i.e. there is a unique to 2 I su h that (tiL)(t) = to (t to ):

11.6.4. The Noether transformations are mappings of spa etime. They play a
fundamental role be ause the proper Noether transformations an be onsidered
to be the stri t automorphisms of the spa etime model.
Proposition. (F; B; idD ) is a stri t automorphism of the non-relativisti
spa e time model (M; I; ; D; b) if and only if F is a proper Noether transformation and B = tiF:
Proof. Let (F; B; idD) be a stri t automorphism. Then  F = B  and
B = idI imply  F =  : Moreover, b (F  F ) = b means that the restri tion
of F onto E is orthogonal. Thus F is an ortho hronous Galilean transformation
and F is an ortho hronous Noether transformation. Sin e F must be orientationpreserving, F is a proper Noether transformation.  F = B  implies that
B = tiF:
Conversely, it is evident that if F is a proper Noether transformation, then
(F; tiF; idD ) is a stri t automorphism.

11.6.5. Let us denote the translation group of I by T n(I) and onsider it as

an ane transformation group of I : t 2 I a ts as I ! I; t 7! t + t: In this respe t
0 2 I equals the identity map of I: It is quite obvious now that
N ! ! T n(I);

L 7! tiL

is a surje tive Lie group homomorphism. Its kernel,

Ni := fL 2 N ! j tiL = 0 (= idI )g = fL 2 N ! j  L =  g
is alled the instantaneous Noether group. It is a nine-dimensional Lie group
having the Lie algebra

La(Ni ) = fH 2 A(M; M) j  H = 0;

H

i 2 A(b)g :

The instantaneous Noether transformations leave every instant invariant.

T n(I) is not a subgroup of N : For every u 2 V(1);

T n(I)u := fidM + ut j t 2 Ig

is a subgroup of the ortho hronous Noether group, alled the group of u-timelike
translations. The restri tion of the homomorphism L 7! tiL onto T n(I)u is a
bije tion between T n(I)u and T n(I):

In other words, given u

transformation

2 V(1); we an assign to every t 2 I the Noether

x 7! x + ut

alled the u-timelike translation by t:

11.6.6. The Galilean group is not a subgroup of the Noether group. The
mapping N ! G ; L 7! L is a surje tive Lie group homomorphism whose kernel
is T n(M); the translation group of M;
T n(M) = fTx jx 2 Mg = fL 2 NjL = idMg :

As we know, its Lie algebra is M regarded as the set of onstant maps from
M into M (VII.3.3).
For every o 2 M;
Go := fL 2 N j L(o) = og ;
alled the group of o- entered Galilean transformations, is a subgroup of the
Noether group and even of the instantaneous Noether group; the restri tion of
the homomorphism L 7! L onto Go is a bije tion between Go and G :
In other words, given o 2 M; we an assign to every Galilean transformation
L the Noether transformation

x 7! o + L
(x o);
alled the o- entered Galilean transformation by L:
The subgroup of o- entered spe ial Galilean transformations

Vo := fL 2 No j

L 2 Vg

has a spe ial importan e.

11.6.7. The three-dimensional orthogonal group is not a subgroup of the

Noether group. The mapping N ! ! O(b); L 7! RL (where L
i = i
RL ) is a
surje tive Lie group homomorphism having the kernel
H := fL 2 N ! j L
i = ig = fL 2 N ! j L 2 Vg
is alled the spe ial Noether group. Observe that H is in N +! :
The spe ial Noether group is a seven-dimensional Lie group having the Lie
algebra

fH 2 La(N ) j

2 La(V )g = fH 2 A(M; M) j 
H = 0;

H

i = 0g :

For every u 2 V(1) and o 2 M;

O(b)u;o := fL 2 N ! j L(o) = o;

L
u = ug ;

alled the group of o- entered u-spa elike orthogonal transformations, is a subgroup of N ! and even of the instantaneous Noether group Ni : The restri tion
of the homomorphism N ! ! O(b) onto O(b)u;o is a bije tion between O(b)u;o
and O(b):
In other words, given (u; o) 2 V(1)  M; we an assign to every R 2 O(b) the
Noether transformation

x 7! o + u
(x o) + R
u
(x o);
alled the o- entered u-spa elike orthogonal transformation by R:

11.6.8. The Neumann group

C := fL 2 Ni j L
i = ig = H \ Ni
is an important subgroup of the spe ial Noether group. It is a six-dimensional
Lie group having the Lie algebra

fH 2 La(Ni ) j

2 La(V )g = fH 2 A(M; M) j  H = 0;

H

i = 0g :

Proposition. The Neumann group is a ommutative normal subgroup of the

Noether group.
Proof. Let K and L be arbitrary Neumann transformations. Sin e they
are instantaneous Noether transformations, for all world points x we have that
L(x) x and K (x) x are in E: As a onsequen e, L(x) x = K
(L(x) x) =
KL(x) K (x) and similarly, K (x) x = LK (x) L(x) from whi h we on lude
that KL(x) LK (x) = 0; i.e. KL = LK; the Neumann group is ommutative.
Now we have to show that if L is an arbitrary Neumann transformation and G
is an arbitrary Noether transformation then G 1 LG is a Neumann transformation, too. The range of G
i is in E; hen e L
G
i = G
i and so G 1
L
G
i = i
whi h ends the proof.
11.7 The ve torial Noether group

11.7.1. Re all that for an arbitrary world point o; the ve torization of M

with origin o; Oo : M ! M; x 7! x o; is an ane bije tion.

With the aid of su h a ve torization we an \ve torize" the Noether group as

well: if L is a Noether transformation then Oo L Oo 1 is an ane transformation of M; represented by the matrix (see VI.2.4(ii ) and Exer ise VI.2.5)


1
0
L(o) o L :


The Lie algebra of the Noether group onsists of ane maps H : M ! M

where M is onsidered to be a ve tor spa e (the sum of su h maps is a part of the
Lie algebra stru ture). Thus the ve torization H Oo 1 is an ane map M ! M
where the range is onsidered to be a ve tor spa e. Then it is represented by the
matrix (see VI.2.4(iii ))

0
0 :
H (o) H

11.7.2. Denition. The ve torial Noether group is




1 0  a 2 M; L 2 (G ) :

a L

The ve torial Noether group is a ten-dimensional Lie group, its Lie algebra is
the ve torization of the Lie algebra of the Noether group:


0 

a H

a2

M;

2 La(G ) :

An advantage of this matrix representation is that the ommutator of two Lie

algebra elements an be omputed by the dieren e of their produ ts in dierent
orders.

11.7.3. A ve torization of the Noether group is a Lie group isomorphism

between the Noether group and the ve torial Noether group. The following
transformation rule shows how the ve torizations depend on the world points
serving as origins of the ve torization. Let o and o0 be two world points; then
To o0 := Oo0 Oo 1 =

0
o o0 idM

and


0
To o0 a1 L

To o0 1 =

1
a + (L idM )(o0

o)

(a 2 M; L 2 G ):

As on erns the orresponding Lie algebra isomorphisms, we have



0  To

a H

1
o0 =

0
a + H (o0

o)

0

(a 2 M; H 2 La(G )):

11.8. The split Noether group

11.8.1. With the aid of the splitting orresponding to u 2 V(1); we send the
transformations of M into the transformations of I  E: Composing a ve torization and a splitting, we onvert Noether transformations into ane transformations of I  E:
For o 2 M and u 2 V(1) put
hu;o := hu Oo : M ! I  E; x 7! (
(x o);

u
(x

o)) :

Embedding the ane transformations of I  E into the linear transformations

of R  (I  E) (see VI.2.4(ii )) and using the ustomary matrix representation of
su h linear maps, we get
1
0
0
hu;o L hu;o 1 =  
(L(o) o)
arL
0 A:
u
(L(o) o) L
u (arL)u RL
1

The Lie algebra elements of the Noether group are onverted into ane maps
Then we an
represent su h maps in a matrix form as well:

I  E ! I  E where the range is regarded as a ve tor spa e.

0
0
0
hu H hu;o 1 =  
H (o)
0
0 A:
u
H (o) H
u H
i
0

11.8.2. Denition. The split Noether group is

9
1
0 0 
=
E
 0 A  2 f 1; 1g; t 2 I; q 2 E; v 2 ; R 2 O(b) :
;
I


80
< 1
t
:

The split Noether group is a ten-dimensional Lie group having the Lie algebra
80
< 0
t
:

0
0

0 1
0 A t 2 I;

E
q 2 E; v 2 ;
I

A 2 A(

9
=

b); :

Keep in mind that the group multipli ation of split Noether transformations
oin ides with the usual matrix multipli ation and the ommutator of Lie algebra
elements is the dieren e of their two produ ts.

11.8.3. Every u 2 V(1) and o 2 M establishes a Lie group isomorphism

between the Noether group and the split Noether group. Evidently, for dierent
elements of V(1)  M; the isomorphisms are dierent. The transformation
rule that shows how the isomorphism depends on (u; o) an be obtained by
a ombination of the transformation rules 11.7.3 and 11.4.2.
Though the Noether group and the split Noether group are isomorphi (they
have the same Lie group stru ture), they are not \identi al": there is no \ anoni al" isomorphism between them that we ould use to identify them.
The split Noether group is the Noether group of the split non-relativisti
spa etime model (I  E; I; prI ; D; b): The spa etime model (M; I; ; D; b) and
the orresponding split spa etime model are isomorphi , but they annot be
identied, as we pointed out in 1.5.3.
11.8.4. It is a routine to he k that the isomorphism established by an
arbitrary (u; o) 2 V(1)  M sends the subgroups of the Noether group listed
below on the left-hand side into the subgroups of the split Noether group listed
below on the right-hand side:
T n(E)
T n(M)

Ni

80
< 1

:

(spe ial
Noether group)

80
< 1
t
:

(instantaneous
Noether group)

80
< 1
t
:

0
q

80
< 1

:

0
q

0
0

1


A q

idE 

0
1

(Neumann
group)

C
H

80
< 1

:

0
0

1


A t

idE 
1

0

 q
A
1

v idE 
1

0

 t
A
1

v idE 
1

0

A q
1

v R 

0
1

0
0

0
0

0
0

9
=

2 E; ;
9
=

2 I; q 2 E; ;
9

=
2 E; v 2 EI ; ;
9
=

2 I; q 2 E; v 2 EI ; ;
9
=

2 E; v 2 EI ; R 2 O(b); ;

It is emphasized that the isomorphism established by an arbitrary (.u; o) makes

a orresponden e between the listed subgroups; of ourse, the orresponden es
due to dierent (u; o) and (u0 ; o0 ) are dierent.

Moreover, the isomorphism established by (u; o) makes orresponden es between the following subgroups, too:

T n(I)u

(u-timelike
translations)

O(b)u;o (o- entered

orthogonal
transformations)

(o- entered
Galilean
transformations)

0
0

1


A

idE 

0
1

80
< 1
0
:

u-spa elike

Go

80
< 1
t
:

0
1

0 1
0 A

t2

Vo (o- entered spe ial

Galilean transformations)

80
< 1
0
:

0
1

0 v

I; ;
9
=

b); ;

R 2 O(

0 0 
 0 A  2 f 1; 1g; v 2
0 v R 

80
< 1
0
:

9
=

0
0

1


A

idE 

E ; R 2 O(b)=
;
I
9

E= ;
v2
I;

and now it is emphasized that the isomorphism established by (u0 ; o0 ); in general,

does not make a orresponden e between the listed subgroups.

11.8.5. Corresponding to the stru ture of the split Noether group, the
following four subgroups are alled its fundamental subgroups:
80
< 1
t
:

80
< 1

:

0
0

0
1

0
0

1


A

idE 

0
1

0
0
0

1


A

idE 

t2

9
=

I; ;
9

E= ;
v2
I;

80
< 1

:

0
0

0
1

1


A

idE 

80
< 1

:

0
1

0 1
0 A

0
0 0

9
=

2 E; ;
9
=

b); :

R 2 O(

The isomorphism established by (u; o) 2 V(1)  M assigns these subgroups to

the subgroups T n(I)u ; T n(E); Vo and O(b)u;o ; respe tively.
It is worth repeating the a tual form of the orresponding Noether transformations:

T n(I)u :x 7! x + ut
T n(E) :x 7! x + q
Vo :x 7! x + v
(x o)
O(b)u;o :x 7! o + u
(x o) + R
u (x o)

(t 2 I);
(q 2 E);

E );
I
(R 2 O(b)):

(v 2

11.8.6. Taking a linear bije tion I ! R and an orthogonal linear bije tion
E ! R3 ; we an transfer the3 split Noether group into the following ane

tranformation group of R  R ;

0 0 
=
 0 A  2 f 1; 1g;  2 R;  2 R3 ;  2 R3 ;  2 O(3) ;
;
   

80
< 1

:

whi h we all the arithmeti Noether group. This is the Noether group of the
arithmeti spa etime model (O(3) denotes the orthogonal group of R3 endowed
with the usual inner produ t).
In onventional treatments one onsiders the arithmeti spa etime model
(without an expli it denition) and the arithmeti Noether group whi h is alled
there Galilean group. The spe ial form of su h transformations yields that one
speaks about the time inversion ( = 1;  = 0;  = 0;  = 0;  = 0); the
time translations ( = 1;  2 R;  = 0;  = 0;  = 0); the spa e rotations
( = 1;  = 0;  = 0;  = 0;  2 SO(3)) et ., whereas we know well that su h
Noether transformations do not exist: there are o- entered u-timelike inversions,
u-timelike translations and o- entered u-spa elike rotations, et .

11.9. Exer ises

1. Let L be a Noether transformation for whi h L = idM : Then there is a
unique o 2 M su h that L is the o- entered spa etime inversion.
2. A Noether transformation L is instantaneous, i.e. is in Ni if and only if all
the hyperplanes t 2 I are invariant for L:
3. Prove that for all o 2 M;

Oo No Oo 1 =



0  L 2 N  :

4. Find hu;o
Tu;o
hu;o 1 and hu;o
Pu;o
hu;o 1 :

5. Prove that the subgroup generated by fTu;oj u
fL 2 Nj L
i = ig:
6. For all u 2 V(1); o 2 M we have
(tiTu;o)(t) =  (o) (t  (o)) = t 2(t  (o))

2 V(1); o 2 Mg equals
(t 2 I):

7. Prove that the derived Lie algebra of the Noether group, i.e. [La(N );
La(N ) equals the Lie algebra of the instantaneous
Noether group.
8. Let L 2 T n(M): Then hu;o
L
hu;o 1 is the same for all u and o if and
only if L 2 T n(E):

9. Take a u 2 V(1) and an o 2 M: If t 2 I; q

the maps M ! M
8
>
>
>
<

(i)
(ii)
H (x) :=
>
(iii)
>
>
:
(iv)

ut
q
v
(x o)
A
u
(x

2 E; v 2 EI ; A 2 A(b); then
(x 2 M)

o)

are elements of the Lie algebra of the Noether group. Prove that
8
>
>
>
<

(i)
(ii)
eH (x) =
>
(iii)
>
>
:
(iv)

x + ut
x+q
x + v
(x o)
o + u
(x o) + eA
u
(x o)

(x 2 M):

10. Compute the produ t of two split Noether transformations:

0 0
1 0
0
t
 0 A  t0 0 0 A :
q v R
q 0 0 v 0 R0
0

10

11. Let L be a Noether transformation.

If r is a world line fun tion, then L r (tiL) 1 is a world line fun tion, too.
If C is a world line, then L[C is a world line, too; moreover, if C = Ran r;
then L[C = Ran L r (tiL) 1 :

II. SPECIAL RELATIVISTIC SPACETIME

MODELS

1. Fundamentals

1.1. Heuristi onsiderations

1.1.1. A ording to the non-relativisti spa etime model, the relative velo ity
of masspoints an be arbitrarily large: the relative velo ities form a Eu lidean
ve tor spa e. However, experien e shows that relative velo ities annot ex eed
the light speed in va uum. Experien e indi ates as well that only \sluggish" me hani al phenomena are suitably des ribed by the formulae of a non-relativisti
spa etime model, i.e. when the relative velo ities of masspoints are low ompared
to the light speed.
A simple example onvin es us that the non-relativisti spa etime model is not
right for the orre t treatment of ele tromagneti phenomena. Let us onsider
a light signal, a well-known ele tromagneti phenomenon. A ording to our
experien e, an inertial observer sees a light signal propagating along a straight
line with a uniform relative velo ity. Let us try to model a light signal in the nonrelativisti spa etime model. Evidently, the model would have to be a straight
line. There are two possibilities: the straight line is a world line or is ontained in
a simultaneous hyperplane. The rst possibility is ex luded be ause then there
would be an inertial observer relative to whi h the light signal is at rest, whi h
is in ontrast with our experien e. The se ond possibility is ex luded as well,
be ause then the light signal would propagate with an \innite" relative velo ity
(there is no time elapse during the propagation).
Thus wishing to des ribe orre tly \brisk" me hani al phenomena and ele tromagneti phenomena, we have to leave the non-relativisti spa etime model
and to onstru t a new spa etime model.
1.1.2. The re tilinear and uniform propagation of light suggests that the
ane stru ture of spa etime an be retained, i.e. spa etime will be modelled
again by a four-dimensional oriented ane spa e M (over the ve tor spa e M):
It follows then that we have to reje t absolute time. Of ourse, something
must be introdu ed instead of absolute time; we a ept absolute propagation of
light. Next we explain what the absolute propagation of light means.
1.1.3. A ording to our experien e, light | independently of its sour e |
propagates isotropi ally (with the same speed in every dire tion) relative to all
inertial observers.

Let us say so: the propagation of a light signal starting \from a given pla e at
a given instant" i.e. in a given spa etime point is independent of the sour e. This
means that a subset Z(x) of M an be assigned to every x 2 M : the set of world
points that an be rea hed from x by a light signal. Experien e attests that the
propagation of light signals starting from the same pla e relative to an observer
does not depend on time (yesterday and today the propagation is the same) and
light signals starting from dierent pla es propagate \ ongruently": a simple
translation in spa e sends dierent propagations into ea h other. A ordingly
we a ept that Z(x)-s are parallel translations of ea h other whi h implies that
there is a subset L! of M su h that Z(x) = x + L! for all x 2 M:
The light signals starting from an arbitrary world point are half lines in su h
a way that if y is a essible by a light signal starting from x; then x is not
a essible from y:
This means that
(i ) if x 2 L! and  2 R+ then x 2 L! ;
(ii ) if x 2 L! then x 62 L! :
L! is a one and does not ontain a line.

1.1.4. Absolute time in the non-relativisti spa etime model is equivalent to

assigning to every world point x the set of world points simultaneous with x;
 (x) = x + E;
where E is a three-dimensional linear subspa e of M:
We introdu e absolute light propagation in the spe ial relativisti spa etime
model by assigning to every world point x the set of world points a essible by
light signals starting from x;
Z(x) = x + L! ;
where L! is a one without being a linear subspa e. We shall see later that
L! | alled the future light one | is a three-dimensional submanifold, the
boundary of an open onvex one.

1.1.5. We want to in lude in the model that observers experien e a Eu lidean

stru ture on their spa e. In the non-relativisti ase the Eu lidean stru ture was
related to simultaneity (with respe t to absolute time). Here the Eu lidean
stru ture will be related to the isotropi propagation of light, a property that is
not re e ted in L! yet.
To get an inspiration, how to pro eed, let us take the following heuristi
onsideration. Let us a ept that the time and the spa e of an inertial observer
an be represented by R and R3 ; respe tively; let the units be hosen in su h a

way that the light speed is 1 (i.e. if s is the time unit, then the distan e unit is
the distan e overed by a light signal in 1s): Then


( 0 ; ) 2 R  R3 j jj =  0 ;  0 > 0

represents the set of spa etime points a essible by a light signal from (0; 0);
where j j denotes the usual Eu lidean norm on R3 :
The Eu lidean norm derives from an inner produ t; that is why it is suitable
to take the bilinear form
G : R4  R4

! R;

(; ) 7!  0 0 +

and to write the the above set in the form



3
X
i=1

 i i ;

 2 R4 j G(;  ) = 0;  0 > 0 :

G is a Lorentz form. The reader, having studied Se tion V.4 and being familiar
with Lorentz forms will noti e that the ondition  0 > 0 sele ts one of the arrow
lasses of f 6= 0j G(;  ) = 0g:
Now it seems natural to a ept that in our spa etime model the a essibility
by light signals is des ribed by an arrow-oriented Lorentz form. More losely, we
introdu e the measure line I of spa etime distan es and we suppose that there
is an arrow-oriented Lorentz form g : M  M ! I
I su h that L! is one of
the arrow lasses treated in V.4.13.

1.2. Denition of the spa etime model

1.2.1. Denition. A spe ial relativisti spa etime model is a triplet
(M; I; g); where
| M is an oriented four-dimensional real ane spa e (over the ve tor spa e
M);
| I is an oriented one-dimensional real ve tor spa e,
| g : M  M ! I
I is an arrow oriented Lorentz form.
We shall use the following names:
M is spa etime or world,
I is the measure line of spa etime distan es,
g is the Lorentz form.
Elements of M are alled world points. Elements of M are alled world ve tors.

1.2.2. If (M; I; g) is a spe ial relativisti spa etime model, then

(M; I; g) is an oriented and arrow-oriented Minkowskian ve tor spa e. The results and formulae of Se tion V.4 will be used all over this part. Remember

2
2
to distinguish between
q x := x
x and jxj := jx
xj; sin e I is oriented, the
pseudo-length jxj := jxj2 is meaningful. Moreover, re all

S :=fx 2 Mj x2 > 0g;

T :=fx 2 Mj x2 < 0g;
L :=fx 2 Mj x2 = 0; x 6= 0g;

the elements of S0 := S [f0g; T and L are alled spa elike, timelike and lightlike,
respe tively.
Furthermore, the arrow orientation indi ates the arrow lasses T! and L! ;
for every x 2 T! and y 2 T! [ L! we have x
y < 0: Then T := T! and
L := L! are the other arrow lasses and
T = T! [ T ;

L = L! [ L :

T! and L! are the future time one and the future light one, respe tively;
their elements are alled future-dire ted. T and L are the orresponding past
ones with past-dire ted elements.
We often illustrate the world ve tors in the plane of the page:

This illustration is based on the following: represent R  R in the plane in the

usual way by horizontal and verti al axes, alled zeroth and rst; draw the sets
S; T! ; L! ; et . orresponding to the Lorentz form

( 0 ;  1 ); (0 ; 1 )

7!  0 0 +  1 1

and to the arrow orientation determined by the ondition  0 > 0; an el the

oordinate axes.

We know that T onsists of two disjoint open subsets, the two arrow lasses
whi h an be well seen in the illustration. On the other hand, S is onne ted, in
spite of the illustration. Keep in mind this slight ina ura y of the illustration.

1.2.3. Spa etime, too, will be illustrated in the plane of the page. If x is a
world point, x + (T ! [ L! ) and x + (T [ L ) are alled the future-like and
the past-like part of M; with respe t to x:
If y 2 x + (T ! [ L! ) | or, equivalently, y x 2 (T ! [ L! ) | then we say y
is future-like with respe t to x (x is past-like with respe t to y); or y is later than
x (x is earlier than y):

We say that the world points x and y are spa elike separated, timelike separated, lightlike separated, if y x is in S; T; L; respe tively.

1.3. Stru ture of world ve tors and ove tors

1.3.1. The Eu lidean stru ture of our spa e is deeply xed in our mind,
therefore we must be areful when dealing with M whi h has not a Eu lidean
stru ture; espe ially when illustrating it in the Eu lidean plane of the page.
For instan e, keep in mind that the entre line of the one L! makes no sense
(the entre line would be the set of points that have the same distan e from
every generatrix of the one but distan e is not meaningful here). The following
onsiderations help us to take in the situation.
Put


M 
V(1) := u 2  u2 = 1; u
I+  T! :
I

We shall see in 2.3.4 that the elements of V(1) an be interpreted as velo ity
values.
A ording to our onvention, V(1) is illustrated as follows:

Three elements of V(1) appear in the Figure. Observe that it makes no sense
that
| u1 is in the entre line of T! (there is no entre line of T! );
| the angle between u1 and u2 is less than the angle between u1 and u3
(there is no angle between the elements of V(1));
| u2 is longer than u1 (the elements of V(1) have no length).
The reversed Cau hy inequality (see V.4.7) involves the following important
and frequently used relation:
u
u0

1

for u; u0 2 V(1) and equality holds if and only if u = u0:

1.3.2. For u 2 V(1) put

: M ! I;
x 7! u
x;
Eu :=Ker u = fx 2 Mj u
x = 0g ;
iu :=Eu ! M; x 7! x:
u

Sin e u is timelike, Eu is a three-dimensional linear subspa e onsisting of

spa elike ve tors. A ording to our onvention, Eu is represented by a line that
in lines to L! with the same angle as u :

We emphasize that \in lination to L! " makes no sense in the stru ture of
the spa etime model; it makes sense only in the rules of the illustration we have
hosen.
Eu and u
I are omplementary subspa es in M; thus every ve tor x an be
uniquely de omposed into the sum of omponents in u
I and in Eu ; respe tively:
x = u(u
x) + x

u(u
x)

= u( u
x) + x + u(u
x) :

The linear map

u

M ! Eu;

x 7! x + u(u
x)

is the proje tion onto Eu along u: It is illustrated as follows:

The dashed line is to express that V(1) is in fa t a subset of MI and not of M:

1.3.3. For all u 2 V(1); the restri tion bu of the Lorentz form g onto Eu  Eu
is positive denite. Thus (Eu ; I; bu ) is a three-dimensional Eu lidean ve tor
spa e.
A ordingly, the pseudo-length of ve tors in Eu is in fa t a length and the
angle between non-zero ve tors in Eu make sense; of ourse, similar notions for
ve tors in EAu an be introdu ed where A is a measure line. Moreover, all the
results obtained in I.1.2.5 an be applied.
It is trivial that every spa elike ve tor is ontained in some Eu :
S0 =

u2V(1)

Eu:

Consequently, the pseudo-length of a spa elike ve tor will be said length or

magnitude. However, we all attention to the fa t that this length satises the
triangle inequality only for two spa elike ve tors spanning a spa elike linear
subspa e (see Exer ise V.4.20.2)

1.3.4. The orientation of M and the arrow orientation of g determine a

unique orientation of Eu:
Denition. Let u 2 V(1): An ordered basis (e1 ; e2; e3) of Eu is alled
positively oriented if (ut; e1 ; e2 ; e3 ) is a positively oriented basis of M for some
(hen e for all) t 2 I+ :
1.3.5. Proposition. Let u 2 V(1): Then
hu := (u ; u ) : M ! I  Eu ; x 7!

u
x; x + u(u
x)

is an orientation-preserving linear bije tion and

hu 1 (t; q ) = ut + q

(t 2 I; q 2 Eu ):

Keep in mind that x = u( u
x) + u
x results in the following important

formula:
x2 = (u
x)2 + ju
xj2
(x 2 M):

1.3.6. Note the striking similarity between the previous formulae and the
formulae of the non-relativisti spa etime model treated in I.1.2. However,
behind the resemblan e to it there is an important dieren e: in the nonrelativisti ase a single three-dimensional subspa e E appears whereas in the
spe ial relativisti ase every u 2 V(1) indi ates its own three-dimensional
subspa e. Correspondingly, instead of a single  ; now there is a u for
all u: The range of hu is the same set in the non-relativisti ase, whereas it
depends on u in the relativisti ase.

A further very important dieren e is that M and M are dierent ve tor

spa es in the non-relativisti ase, whereas they are \nearly the same" in the
relativisti ase. More pre isely, we have the identi ation (see V.1.3).

M

I
I M ;
whi h is established by the Lorentz form g: A ording to our dot produ t notation, g does not appear in the formulae. That is why we a ept the notation
g := idM 2 M
M
as well, whi h will fa ilitate the omparison of our formulae with those of usually
employed in textbooks. Then, for instan e, we an write
u

= g + u
u:

Of ourse, we make the identi ation

Eu 
I
I  Eu;

too.
A ording to these identi ations we have
u

iu 2 M
Eu  MI

EI u ;
M ;iu 2 Eu
M  Eu
M :
u 2 M
I 
I
I
I

2 I
M  MI ;
u

Moreover,

2 Eu
M  EuI

IM :

u
u

u
u

= 0;

= idEu

and the identi ations yield the relations

u

 u 

u;

iu  u:

The reader is asked to prove the rst formula; as on erns the se ond one, see
the following equalities for x 2 M; q 2 Eu :
(iu
x)
q = x
iu
q = x
q = (u
x)
q:

1.3.7. For u; u0 2 V(1) put

vu0 u

:=

u0
u0
u

u:

We shall see later that this is the relative velo ity of u0 with respe t to u:
It is an easy task to show that jvu0 u j2 = jvuu0 j2 = 1 (u01
u)2 ; as a onsequen e
of the reversed Cau hy inequality, vu0 u = 0 if and only if u = u0: Moreover, if
q 2 Eu \ Eu0 then q
vu0 u = 0 whi h proves the following.
Proposition. Eu \ Eu0 is a two-dimensional linear subspa e if and only if
u 6= u0 and in this ase vu0 u
I (vuu0
I) is a one-dimensional linear subspa e
of Eu (Eu0 ); orthogonal to Eu \ Eu0 :
(In other words, Eu \Eu0 and vu0 u
I (vuu0
I) are orthogonal omplementary
subspa es in Eu (Eu0 )):

1.3.8.

For dierent u and u0 ; Eu and Eu0 are dierent linear subspa es;
however, we an give a distinguished bije tion between them whi h will play a
fundamental role on erning observer spa es.
Let L(u0 ; u) be the linear map from Eu onto Eu0 dened in su h a way that it
leaves invariant the elements of Eu \ Eu0 and maps the orthogonal omplements
of this subspa e into ea h other. More pre isely,
L(u0 ; u)
q

:=

q
vuu0 t

if q 2 Eu \ Eu0
:
if q = vu0 u t (t 2 I)

It is not di ult to see that L(u0 ; u) is an orientation-preserving bu bu0 orthogonal linear bije tion between Eu and Eu0 : We an extend it to a linear
bije tion M ! M by the requirement
L(u0 ; u)
u := u0

(re all that the dot produ t notation allows us to apply linear maps M ! M to
elements of MI ):
This linear bije tion an be given by a simple formula. Now we give this
formula and then hara terize its properties. Re all g := idM and for u; u0 2
V(1); u0
u 2 MI
MI  MI

IM  M
M  Lin(M; M):
Denition. Let u; u0 2 V(1): Then
L(u0 ; u) :=

g + (u +1u)
u0(
uu + u)

is alled the Lorentz boost from u to u0 :

2u0
u

Proposition. (i ) L(u0; u) is an orientation- and arrow-preserving g-orthogonal linear map from M into M;
(ii ) L(u0 ; u)
u = u0 ;
(iii ) L(u0 ; u) maps Eu onto Eu0 ; more losely,
| L(u0 ; u)
q = q if q 2 Eu \ Eu0 ;
| L(u0 ; u)
vu0 u = vuu0 ;
(iv ) L(u; u) = g;
L(u0 ; u) 1 = L(u; u0 )

and L(u0 ; u) is the unique linear map for whi h (i ){(iii ) hold.

1.3.9. Sin e the Lorentz boosts map the orresponding spa elike subspa es
onto ea h other in a \handsome" manner, we might expe t that exe uting the
Lorentz boost from u to u0 and then the Lorentz boost from u0 to u00 we should
get the Lorentz boost from u to u00 ; however, this o urs only in some spe ial
ases.
Proposition. Let u; u0; u00 be elements of V(1): Then
L(u00 ; u0 )
L(u0 ; u) = L(u00 ; u)

if and only if the three elements of V(1) are oplanar.

Proof. Suppose the equality holds. Then for all q 2 Eu \ Eu00

= L(u00 ; u0 )
L(u0 ; u)
q =

 

(u00 + u0 )
(u00 + u0)
00
u0
q + (u0 + u)u0
q =
= g+
2
u
1 u00
u0
1 u0
u

u0 + u
u00 + u0
+
+
= q + (u0
q)
00
0
1 u
u 1 u0
u

(u00 + u0 )(u00
u0 + u00
u + u0
u 1)
;
+
(1 u00
u0 )(1 u0
u)

from whi h we dedu e that

| either u0
q = 0 for all q 2 Eu \ Eu00 ; implying that u0 is in the twodimensional subspa e spanned by u and u00 ; i.e. the three elements of V(1) are
oplanar,
| or the last expression in parentheses is zero whi h implies again that the
three elements of V(1) are oplanar.
Observe that L(u00 ; u0 )
L(u0 ; u) maps Eu onto Eu00 ; as a onsequen e, if it is
a Lorentz boost, it must equal L(u00 ; u): Thus our result implies that, in general,
the produ t of Lorentz boosts is not a Lorentz boost.

1.4.1.

1.4. The arithmeti spa etime model

Let us take the Minkowskian ve tor spa e (R1+3 ; R; G) treated in

V.4.19 and endowed with the standard orientation and arrow orientation. Considering R1+3 to be an ane spa e, we easily nd that (R1+3 ; R; G) is a spe ial relativisti spa etime model whi h we all the arithmeti spe ial relativisti
spa etime model.
As in the arithmeti non-relativisti spa etime model, the same obje t,
R1+3 ; represents the ane spa e of world points and the ve tor spa e of world
ve tors (and even the ve tor spa e of ove tors). We follow our non-relativisti
onvention that the world points will be denoted by Greek letters whereas world
ve tors (and ove tors) will be denoted by Latin letters.
We nd onvenient to write the elements of the ane spa e R1+3 in the form
i
( ); the elements of the ve tor spa e R1+3 in the form (xi ) = (x0 ; x); and the
elements of (R1+3 ) in the form (ki ) = (k0 ; k):
Re all that the identi ation (R1+3 )  R1+3 established by G gives

x0 = x0 ;

x = x

( = 1; 2; 3):

Correspondingly, the dot produ t of (xi ) and (yi ) equals

xi yi ;
where the Einstein summation rule is applied: a summation is arried out from
0 to 3 for identi al subs ripts and supers ripts.

1.4.2. In the arithmeti spa etime model



V(1) = (ui ) 2 R1+3 j ui ui = 1; u0 > 0 :

The simplest element (1; 0) of V(1) is alled the basi velo ity value.
is the anoni al proje tion R1+3 ! f0g  R3 :
For an arbitrary element (ui ) of V(1) we an dene

(1;0)

u
( = 1; 2; 3);
v := (v 1 ; v 2 ; v 3 ) 2 R3 ;
v := 0
u
then with the usual norm j j on R3 we have jvj < 1 and
1
u0 = q
1 jvj2
and

(ui ) =

jvj2

(1; v):

()

We easily nd that v is exa tly the relative velo ity of (ui ) with respe t to
the basi velo ity value (see 1.3.7)

1.4.3. It is then obvious that

E(ui ) = (xi ) 2 R1+3 j x0 = x
v :
Unlike the non-relativisti ase, (ui ) for a general (ui ) in V(1) is an uneasy
obje t be ause it maps onto a three-dimensional linear subspa e in R1+3 whi h
is dierent from f0g R3 : Thus the values of (ui ) annot be given dire tly by
triplets of real numbers. However, as it is known, in textbooks one usually deals
with triplets (and quartets) of real numbers. We an a hieve this by always
referring to the spa e of the basi observer with the aid
of the orresponding
Lorentz boost, i.e. instead of (ui ) taking L (1; 0); (ui )
(ui ) ; whose range
is f0g  R3 :

1.4.4. The Lorentz boost from u0i ) to (ui ) is given by the matrix
Lik := gki +

(ui + u0 i )(uk + u0k )

1 uj u0j

2ui u0k :

If (u0 i ) is the basi velo ity value then it be omes

0

1
u1
u2
u3
12
B
u1 u2
u1 u3 C
B u1 1 + (u )
C
B
1 + u0
1 + u0
1 + u0 C
B
C
B
1 u2
2 u3 C
2 )2
B 2
C:
u
u
(
u
Bu
C
1+
0
0
0
B
1+u
1+u
1+u C
B
C

2 u3
3 )2 A
1 u3
u
(
u
u
1+
u3
1 + u0
1 + u0
1 + u0
Using the formula () in 1.4.2 and the notation

u0

 :=
we nd that

jvj2

(1; 0); (ui ) =

1
0
1
v1
v2
v3
2
1



1
1
1
2
1
3
C
Bv
 + 1+ (v )
1+ v v
1+ v v
= B
1 +  (v 2 )2
 v2 v3 C
 v1 v2
A:
 v2
1+
 1+
1+
 2 3
1
 3 2
 1 3
v3
1+ v v
1+ v v
 + 1+ (v )
L

This shows what a ompli ated form L (1; 0); (ui )
(ui ) has; later (see 7.1.4)
we give it in detail.

1.4.5. The previous matrix is the usual \Lorentz transformation". Most

frequently one onsiders the spe ial ase v2 = v3 = 0; v := v1 ; then  = p11 v2
2 2
and 1+v =  1; thus the previous matrix redu es to
0

1
Bv
0
0

v 0
0 1
1 0
0 C
0 1= 0 A :
0 0 1=

1.5. Classi ation of physi al quantities

1.5.1. We introdu e notions similar to those in the non-relativisti spa etime

model. Let A be a measure line. Then the elements of

A are alled s alars of type A;

A
M are alled ve tors of type A;
M are alled ve tors of otype A;
A
A
(M
M) are alled tensors of type A;
M
M are alled tensors of otype A:
A
Cove tors of type A; et . are dened similarly with M instead of M:
In parti ular, the elements of M
M and M
M are alled tensors and
otensors, respe tively; the elements of M
M and M
M are mixed tensors.
A very important feature of the spe ial relativisti spa etime model is that
ove tors an be identied with ve tors of otype I
I: As a onsequen e, e.g.
a ove tor of type A is identied with a ve tor of type IA

I :
1.5.2. A ording to our onvention, the dot produ t between ve torsM ( ove tors) of dierent types makes sense. For instan e, for u
for
z

2 A
M
M
w2
A

we have
we have

In parti ular, z2 2 R for z 2 MI :

2 I
A;
I
u
w 2 ;
A
u
z

2 V(1)  I

and

2 (A
A)
(I
I);
I
I
w2 2
A
A:
z2

Sin e (A
A)
(I
I)  (A
I)
(A
I) has a natural orientation, we an
speak of its positive and negative elements. Thus a ve tor z of type A is alled

spa elike if z2 > 0 or z = 0;

timelike if z2 < 0;
lightlike if z2 = 0; z 6= 0:

It an be easily shown that z is spa elike if and only if z 2 A

S0 ; et .
Moreover, a measure line A is oriented, hen e A+ makes sense. Consequently,
we dene that a timelike (lightlike) ve tor z of type A is future-dire ted if
z 2 A+
T ! (z 2 A+
L! ):

1.6. Comparison of spa etime models

1.6.1. Denition. The spe ial relativisti spa etime model (M; I; g) is
isomorphi to the spe ial relativisti spa etime model (M0 ; I0 ; g0 )if there are
(i ) an orientation- and arrow-preserving ane bije tion F : M ! M0 ;
(ii ) an orientation-preserving linear bije tion Z : I ! I0
su h that
g0 (F  F ) = (Z
Z ) g;

where F is the linear map under F: The pair (F; Z ) is an isomorphism between
the two spa etime models.
If the two models oin ide, an isomorphism is alled an automorphism. An
automorphism of the form (F; idI ) is stri t.
Three diagrams illustrate the isomorphism:
M  M g! I
I
I
M

Fy

?
yZ

M0

I0

 Fy

?
yZ

M0  M0 g!0 I0
I0:

The denition is quite natural and simple, needs no omment.

1.6.2. Proposition. The spe ial relativisti spa etime model (M; I; g) is
isomorphi to the arithmeti spa etime model.
Proof. Take
(i ) a positive element s of I;
(ii ) a positively oriented g-orthogonal basis (e0 ; e1 ; e2 ; e3 ); normed to s; of M;
for whi h e0 is future-dire ted,

(iii ) an element o of M:
Then

F : M ! R4 ;

x 7!

I ! R;

t 7!

t
s

ek
(x
e2k

o) 


k = 0; 1; 2; 3 ;

is an isomorphism.
This isomorphism has the inverse
R4

! M;

 7! o +

! I;

 7! s:

3
X
k=0

 k ek ;

1.6.3. An important onsequen e of the previous result is that any two spe ial
relativisti spa etime models are isomorphi , i.e. are of the same kind. The
spe ial relativisti spa etime model as a mathemati al stru ture is unique. This
means that there is a unique \spe ial relativisti physi s".
Note: the spe ial relativisti spa etime models are of the same kind but, in
general, are not identi al. They are isomorphi , but, in general, there is no
\ anoni al" isomorphism between them, we annot identify them by a distinguished isomorphism. The situation is the same as that we en ountered for
non-relativisti spa etime models.
Sin e all spe ial relativisti spa etime models are isomorphi , we an use an
arbitrary one for investigation and appli ation. However, an a tual model an
have additional stru tures. For instan e, in the arithmeti spa etime model,
spa etime is a ve tor spa e, V(1) has a distinguished element. This model tempts
us to multiply world points by real numbers (though this has no physi al meaning
and that is why it is not meaningful in the abstra t spa etime model), to speak
about time and spa e, onsider spa etime as the Cartesian produ t of time and
spa e (whereas neither time nor spa e exists), et .
To avoid su h onfusions, we should keep away from similar spe ially onstru ted models for investigation and general appli ation of the spe ial relativisti spa etime model. However, for solving spe ial problems, for exe uting some
parti ular al ulations, we an hoose a onvenient a tual model, like in the
non-relativisti ase.
1.6.4. Present day physi s uses ta itly the arithmeti spa etime model. One
represents time points by real numbers, spa e points by triplets of real numbers.
To obtain su h representations, one hooses a unit for time periods, an initial

time point, a distan e unit, an initial spa e point and an orthogonal spatial basis
whose elements have unit length.
However, all the previous notions in usual ir umstan es have merely a heuristi sense. The isomorphism established in 1.6.2 will give these notions a mathemati ally pre ise meaning. We shall see later that s is the time unit (and the
distan e unit), e0 hara terizes an observer whi h produ es its own time and
spa e, the spa elike ve tors e1 ; e2 ; e3 orrespond to the spatial basis, o in ludes
the initial time point and spa e point in some way.

1.7. The u-split spa etime model

1.7.1. The arithmeti spa etime model is useful for solving parti ular problems, for exe uting pra ti al al ulations. Moreover, at present, one usually expounds theories, too, in the frame of the arithmeti spa etime model, so we have
to translate every notion in the arithmeti language. As in the non-relativisti
ase, it is onvenient to introdu e an \intermediate" spa etime model between
the abstra t and the arithmeti ones.
1.7.2. Let (M; I; g) be a spe ial relativisti spa etime model and use the
notations introdu ed in this hapter. Take a u 2 V(1) and dene the Lorentz
form
gu : (I  Eu)  (I  Eu) ! I
I;

(t0 ; q0 ); (t; q) 7! t0 t + q0
q:

Put


S := (t; q)j jqj > jtj ;



T := (t; q)j jqj < jtj ;


L := (t; q)j jqj = jtj 6= 0 :

Endow I  Eu with the produ t orientation and gu with the arrow orientation
determined by


T! := (t; q) 2 T j t > 0 :
Then (I  Eu ; I; gu ) is a spe ial relativisti spa etime model, alled the u-split
spe ial relativisti spa etime model.
It is quite obvious that for all o 2 M;
M ! I  Eu;
I ! I;

x 7! hu
(x
t 7! t

o);

is an isomorphism between the two spe ial relativisti spa etime models.

1.7.3. In the u-split

model


E
V(1) = (; h) 2 R  u 
=

8
<



(1; v) v
2

v

2 + jhj2 = 1;  > 0 =
9

=
2 EIu ; jvj < 1; :

1 j j
There is a simplest element (the basi velo ity value ) in it: (1; 0):
:

1.7.4. The split non-relativisti spa etime model is simple be ause the a tual
form of the fundamental notions |  ; i; u and hu | is simple for all u 2 V(1):
This follows from the fa t that there is a single three-dimensional subspa e of
spa elike ve tors whi h appears in the split model as a Cartesian fa tor.
On the other hand, taken a uo 2 V(1); the uo -split relativisti spa etime
model is not so simple be ause the a tual form of the fundamental notions |
Eu; u and hu for u 6= uo | is rather ompli ated. This follows from the fa t
that there is not a distinguished three-dimensional subspa e of spa elike ve tors;
the spa elike subspa e orresponding to u diers from the one orresponding
to uo and only the subspa e Euo appears as a Cartesian fa tor in the uo -split
model.
We an exploit the Cartesian produ t stru ture of the uo -split model by always
referring to Euo with the aid of the Lorentz boost L(uo ; u) ( f. the arithmeti
spa etime model, 1.4.2).
1.8. About the two types of spa etime models
Let us summarize the essential features of the non-relativisti spa etime model
and the spe ial relativisti spa etime model.
The ane stru ture of spa etime is the same in both models.
In the non-relativisti model there is a  whi h gives absolute simultaneity
implying a single three-dimensional subspa e of spa elike ve tors, and then there
is a Eu lidean stru ture b on the subspa e of spa elike ve tors.
In the spe ial relativisti model there is a Lorentz stru ture g whi h gives
the absolute propagation of light and indu es the Eu lidean stru ture on the
three-dimensional spa elike subspa es.

1.9. Exer ises

1. To be later (future-like) is a transitive relation on M : if y is future-like
with respe t to x and z is future-like with respe t to y then z is future-like with
respe t to x:

2. V(1) is a three-dimensional submanifold of M; its tangent spa e at u is EIu

(see Ex .VI.4.14.3). For every u 2 V(1);

Eu ! V(1); h 7! uq1 + jhj2 + h;

I

Eu  jvj < 1 ! V(1); v 7! qu + v
v2
I
1 jvj2

and

are global parametrizations of V(1) having the inverses

u0

7! u
u0 = u0 + (u
u0 )u = q

and

u0

7!

respe tively.
3. Prove that for all u 2 V(1)

u
u0
u
u0




E
u

R n2
I  jnj = 1 ! V(1);


vu0 u

jvu0 u j2

= vu0 u ;

(; n) 7!

pu + nth2

= u h + nsh
1 th 



is a smooth map whi h is a bije tion between R+0  n 2 EIu  jnj = 1 and V(1);
having the inverse
0

u0

7! ar h(

u
u0 ); q

u
u0

(u
u0 )2

A:

4. Let u 2 V(1); n 2 EIu ; jnj = 1: Take ;  2 R and put

u0 := u h + nsh; n0 := L(u0 ; u)
n = ush + n h
u00 := u0 h + n0 sh = u h( +  ) + nsh( +  );
u000 := u h + nsh:
Prove that

L(u00 ; u0 ) = L(u000 ; u):

(Hint: L(u00 ; u0)
u = u000 and Eu \ Eu000 = Eu0 \ Eu00 :)

5. Use the notations of the pre eding exer ise and prove that
L(u000 ; u)
L(u0 ; u) = L(u00 ; u)

i.e.
L(u h + nsh; u)
L(u h + nsh; u) = L u h( +  ) + nsh( +  ); u

6. Let u 2 V(1); m; n 2 EIu ; jmj = jnj = 1; m
n = 0: Take an 0 6=  2 R

and put
u0 := u h + nsh;
u00 := u h + msh:
Then n0 := L(u0 ; u)
n = ush + n h and L(u00 ; u)
n = n: Prove that
L(u00 ; u0 )
n0 is not parallel to n:
7. Let u; u0 2 V(1): Then u0
I and Eu are omplementary subspa es. The
proje tion onto Eu along u0
I is the linear map
u
x
u0
u
: M ! M;
x 7! x + u0
:
Puu0 := g +
0
u
u
u0
u
Prove that
(i ) the restri tion of Puu0 onto Eu0 is a bije tion between Eu0 and Eu ;
(ii ) the restri tion q
of Puu0 onto Eu \ Eu0 is the identity;
(iii ) Puu0
vuu0 = 1 jvuu0 j2 vu0 u :

2. World lines

2.1. History of a masspoint: world line

2.1.1. As in the non-relativisti spa etime model, the history of a masspoint

will be des ribed by a urve in the spe ial relativisti spa etime model as well.
However, it is not obvious here, what kind of urves an be allowed.
Our heuristi onsiderations regarding the ane stru ture of spa etime imply
that the history of a free masspoint has to be des ribed by a straight line. We
an dis over simply that su h a straight line must be dire ted by a timelike
ve tor. Indeed, it annot be lightlike be ause this would mean that there is a
light signal resting with respe t to the masspoint. Suppose that the straight
line is dire ted by a spa elike ve tor, hoose two dierent points on the line and
draw the orresponding future light ones: the ones interse t ea h other. As a
onsequen e, two light signals emitted su essively by the masspoint would meet
whi h ontradi ts our experien e.

A simple generalization | in a ordan e with I.2.2 | yields that the existen e

of a masspoint must be des ribed by a urve whose tangent ve tors are timelike.
We all attention to the fa t that up to now we have spoken about light signals
and masspoint histories in a heuristi sense. The following denition gives these
notions a pre ise meaning in the spa etime model.

2.1.2. Denition. 1. A straight line segment in M; dire ted by a lightlike

ve tor, is alled a light signal.
2. A world line is a onne ted twi e dierentiable urve in M whose tangent
ve tors are timelike.
Proposition. Let C be a world line. Then y x is timelike for every x; y 2 C;
x 6= y: In other words,
C n fxg  x + T
(x 2 C):

Proof. Suppose the statement is not true: there is an x 2 C su h that C nfxg

is not ontained in x + T: Let p : R M be a parametrization of C; p(0) = x:
Then there is a 0 6=  2 Dom p su h that p() p(0) is not timelike. For the
sake of deniteness we an assume  > 0: Then

a := inf f 2 Dom pj  > 0; p() p(0) 62 Tg > 0:

Indeed, if this inmum were zero then there would be a sequen e n > 0
p(n ) p(0) 62 T for all n implying p_(0) =
(n 2 N) su h that nlim
n
!1 n = 0 and
p
(

)
p
(0)
n
lim
62 T be ause the set of timelike ve tors is open (the omplement
n
n!1


of T is losed). Be ause of the same reason, p(a) p(0) = nlim
!1 p(n ) p(0) 62 T:
Thus p() p(0) is timelike for 0 <  < a and p(a) p(0) is not timelike.
Sin e p is ontinuous,
p(a) p(0) must be in the losure of T; i.e. it is lightlike:


p(a) p(0) 2 = 0:

Lagrange's
mean value theorem, applied to the fun tion

[0; a ! I
I;  7!


2
p() p(0) assures a 20; a[ su h that 2 p( ) p(0)
p_( ) = 0: Sin e p_( )
is timelike, this means that p( ) p(0) is spa elike, a ontradi tion.

2.1.3. The previous result and the arrow orientation (whi h gives rise to the
relation to be earlier, see 1.2.3) allow us to dene an order | an orientation |
on a world line as follows.

Proposition. Let p : R M be a parametrization of the world line C: Then

one of the following two possibilities o urs:
(i )  <  if and only if p() is earlier than p( );
(ii )  <  if and only if p() is later than p( )
for all ;  2 Dom p:

Proof. p_ is a ontinuous fun tion having values in T and dened on an

interval, thus its range is onne ted whi h means that the range of p_ is ontained
either in T! or in T :
(i ) Suppose Ran p_  T! and sele t an arbitrary  from the domain of p:
Then f 2 Dom pj  <  g ! T;  7! p() p() is a ontinuous fun tion
dened on an interval, hen e its range is ontained in T! or in T : Sin e
lim p() p() = p_() 2 T! and T! is open, we on lude that p() p() 2 T! ;
 !  
whi h implies that p( ) p() is in T! ; i.e. p() is earlier than p( ) for all
 < :
(ii ) Similar onsiderations yield the desired result if Ran p_  T :
Denition. A parametrization p of a world line is alled progressive (regressive ) if  <  implies that p() is earlier (later) than p( ) for all ;  2 Dom p:
A world line is onsidered oriented by progressive parametrizations.

The reader easily veries that the orientation is orre tly dened: if p and q
are progressive parametrizations of a world line, then p 1 q : R R is stri tly
monotone in reasing.
Note that the proposition holds and the denition an be applied also for
parametrizations that are dened on an oriented one-dimensional ane spa e.

2.1.4. If x and y are dierent points of a world line then they are timelike
separated.
Conversely, if x and y are timelike separated world points then there is a
world line C su h that x; y 2 C: Indeed, the straight line passing through x and
y is su h a world line. Note the important fa t that there are many world lines
ontaining x and y:

2.2. Proper time of world lines

2.2.1. Masspoint is an abstra tion of a \small" material obje t. Imagine a

pie e of quartz os illator as a masspoint; it \feels" that time passes during its
history: the progress of time is measured by the number of os illations. Sin e
absolute time does not exist, it is evident that ea h history has its own proper
time that passes. This means physi ally that the os illations depend on the
history. Take two \small" quartz rystals resting on the table. Let one of them
ontinue to rest and seize the other, shake it for a while, then put it ba k on the
table. Count the number of os illations in ea h rystal during their separation:
the two numbers an be dierent.
2.2.2. We already know what is later and earlier on a world line. Now we
should like to measure how mu h later (or earlier) a point of a world line is than
another, i.e. we want to measure the time passed between two points of a world
line.
Our experien e indi ates that time passes \uniformly" for an inertial masspoint. A ording to the ane stru ture of spa etime, the history of an inertial
masspoint will be des ribed by a straight line segment.
Let x; y and z be points of a straight world line su h that z is later than x
and y is later than z: Then y x is parallel to z x; thus there is an  2 R+ su h
that y x = (z x): The uniform ow of time suggests that  times more time
passed between x and y than between x and z: But we do not know yet how mu h
time passed between the world points. To measure the time passed between x
and y (along the straight world line) we ought to measure somehow the \length"
of the ve tor y x: The Lorentz form g oers a possibility: we a ept that jy xj
is the time passed between x and y along the straight world line. Note that g
is not positive denite (is not a Eu lidean form), thus the pseudo-length dened
by g has strange properties (see V.4.10) whi h will be important in the sequel.
Take now a \world line" onsisting of two onse utive non-parallel straight
line segments (a ording to our present denition, su h a line is not a world line
be ause it is not dierentiable in one point, that is why we put the quotation
mark; we use su h broken world lines for our heuristi onsideration and later

we permit them by a pre ise denition, too). Let z be the breaking point, let
x be earlier than z; z earlier than y: Then we measure the time passed between
x and y along the broken world line by the sum of the time passed along the
straight line segments: jz xj + jy z j:
The generalization to a broken world line onsisting of several straight line
segments is trivial.
Let now C be an arbitrary world line, x; y 2 C; x is earlier then y: We an
approximate the time passed between x and y along C by the time passed along
broken lines approximating C:
Take a progressive parametrization p of the world line C: Then an approximation of the time passed between x and y along C has the form
n
X
k=1

whi h nearly equals

n
X
k=1

jp(k+1 ) p(k )j

jp_(k )j(k+1 k ):

We re ognize an integral approximating sum. This suggests us the following

denition (the reader is asked to study Se tion VI.7).
Denition. Let x and y be timelike separated world points or x = y: If C is
a world line passing through x and y (i.e. x; y 2 C) then
Zy

tC (x; y) := jdCj
x

is alled the time passed between x and y along C:

The time passed between x and y along a straight line is alled the inertial
time between x and y and is denoted by t(x; y):
Evidently,

earlier than y
t(x; y) = jjyy xxjj ifif xy isis earlier
than x:

2.2.3. The time passed between two world points along dierent world lines
an be dierent. The longest time passes along the inertial world line:
Proposition. Let x be a world point earlier than the world point y: If C is
a world line ontaining x and y then
tC(x; y)  t(x; y)

and equality holds if and only if C is a straight line segment between x and y:

Proof. Let z 2 C; z is earlier than y and later than x: Then the reversed
triangle inequality (V.4.10) results in t(x; z ) + t(z; y)  t(x; y); where equality
holds if and only if z is on the straight line passing through x and y: As a
onsequen e, the time passed between x and y along a broken line (dened to
be the sum of times passed along the orresponding straight line segments) is
smaller than the inertial time between x and y: The denition of tC (x; y) as an
integral involves that tC (x; y) an be obtained as the inmum of times passed
between x and y along broken lines.
2.2.4. Note that the Lorentz form g | besides the determination of the light
one and the Eu lidean stru ture on spa elike subspa es | has got a new and
important role: the determination of time passing along world lines.
We emphasize that the integral formula for the time passing along a world
line is a denition and not a statement.
2.2.5. We all attention to the fa t that in our ustomary illustration the
same time period passed along dierent inertial world lines is represented, in
general, by segments of dierent lengths.

The same length orresponds to the same time period on two inertial world
lines if and only if the two illustrating straight lines have the same in lination
to the two lines of L! :

2.3. World line fun tions

2.3.1. Denition. Let C be a world line, xo 2 C: Then the mapping

C ! I;
x 7! tC (xo ; x)
is alled the proper time of C starting from xo :
Sin e every tangent ve tor x 6= 0 of the world line C is timelike i.e. jxj 6= 0;
a ording to Proposition VI.7.5, the inverse of the proper time,

r:
dened by
and having the property

IM

r(tC (xo ; x)) = x

tC (xo ; r(t)) = t

(x 2 C)
(t 2 Dom r)

is a progressive parametrization of C; alled the proper time parametrization of

C starting from xo : We know that for all t 2 Dom r

r_(t) 2

M;
I

r_ (t) is future-dire ted timelike,

moreover, Proposition VI.7.5 implies

jr_ (t)j = 1;
all these mean that

r_ (t) 2 V(1)

(t 2 Dom r):

2.3.2. A ording to the previous onsiderations, if C is a world line then

there is a parametrization r : I M of C (i.e. r is dened on an interval, is
twi e dierentiable, its range is C) su h that r_ (t) 2 V(1) for all t 2 Dom r:
From the properties of integration on urves we derive that

tC (x; y) =

r Z1 (y)
r 1 (x)

jr_ (t)jdt = r 1 (y) r 1 (x):

As a onsequen e, if r1 and r2 are parametrizations with the above property

then there is a to 2 I su h that Dom r2 = to + Dom r1 and r2 (t) = r1 (t to )
(t 2 Dom r2 ):
Indeed, hoosing an element xo of C and putting to := r2 1 (xo ) r1 1 (xo )
we get r1 1 (x) = r2 1 (x) to whi h gives the desired result with the notation
t := r2 1 (x):

2.3.3. Our results suggest how to introdu e the notion of world line fun tions
whi h allows us to admit pie ewise dierentiability as in the non-relativisti ase.
Denition. A fun tion r : I M is alled a world line fun tion if
(i ) Dom r is an interval,
(ii ) r is pie ewise twi e ontinuously dierentiable,
(iii ) r_ (t) is in V(1) for all t 2 Dom r where r is dierentiable.
A subset C of M is a world line if it is the range of a world line fun tion.
The world line fun tion r and the world line Ran r is global if Dom r = I:

2.3.4. If r is a world line fun tion then dierentiating the onstant mapping

t 7! r_ (t)
r_ (t) =

1 dened on the dierentiable pie es of Dom r we get that

r_(t)
r(t) = 0;

i.e.

E
r(t) 2 r_(t) ;
I
I

and the same is true for right and left derivatives where r is not dierentiable.
M
The fun tions r_ : I
V(1) and r : I
I
I an be interpreted as the
(absolute) velo ity and the (absolute) a eleration of the material point whose
history is des ribed by r:
That is why we all the elements of V(1) velo ity values and the spa elike
elements in IM

I a eleration values.
2.3.5. Re all
that V(1) is a three-dimensional
smooth submanifold of MI : The


M
2

elements of v 2 I 0 < v < 1 will be alled relative velo ity values ; later we
shall see the motivation of this name.
Note the following important fa ts.

(i ) The velo ity values are timelike ve tors of otype I; in parti ular they are
future-dire ted. The velo ity values do not form either a ve tor spa e or an ane
spa e. The pseudo-length of every velo ity value is 1: There is no zero velo ity
value. Velo ity values have no angles between themselves.
(ii ) Relative velo ity values are spa elike ve tors of otype I: They do not
form a ve tor spa e. The magnitude of a relative velo ity value (see 1.3.3) is a
real number less than 1: A relative velo ity anbe smaller
there is
 than another;

a zero relative velo ity value. If u 2 V(1) then v 2 EIu  jvj < 1 is an open ball
in a three-dimensional Eu lidean ve tor spa e and onsists of relative velo ity
values. The angle between su h relative velo ities makes sense.
(iii ) A eleration values are the spa elike ve tors of otype I
I: They do not
form a ve tor spa e. The magnitude of an a eleration value is meaningful, it is
an element of RI : An a eleration value an be smaller than another; there is a zero
a eleration value. If u 2 V(1); then IE
uI is a three-dimensional Eu lidean ve tor
spa e onsisting of a eleration values. The angle between su h a eleration
values makes sense.
\Qui kness" makes no absolute sense; it is not meaningful that a material
obje t exists more qui kly than another. A velo ity value hara terizes somehow
an a tual tenden y of the history of a material point. Material obje ts an move
slowly or qui kly relative to ea h other.

2.4. Classi ation of world lines

2.4.1. We would like to lassify the world lines as we did it in the nonrelativisti ase. The notion of inertial world line is straightforward. However,
the uniformly a elerated world line and the twist-free world line give us some
trouble.
If we opied the non-relativisti denition, i.e. we required that the a eleration of a world line fun tion r be onstant, r = a; where a is a spa elike element
of IM

I ; then there is a 2 V(1) su h that r_(t) = + at (t 2 Dom r): Sin e r_

and r are g-orthogonal, + at and a are g-orthogonal:
a + jaj2 t = 0 for all
t 2 Dom r whi h implies a = 0: There would be no uniformly a elerated world
lines ex ept the inertial ones.
The problem lies in the fa t that the a tual a eleration values of a world line
belong to a subspa e g-orthogonal to the orresponding velo ity values; if the
velo ity value hanges then the orresponding subspa e hanges as well: hanging
velo ity involves hanging a eleration.
Nevertheless, we have not to give up the notion of uniform a eleration.
We established a natural mapping between two subspa es g-orthogonal to two
velo ity values: the orresponding Lorentz boost (see 1.3.8). Then we may

require that the world line fun tion r is uniformly a elerated if r(s) is mapped
into r(t) by the Lorentz boost from r_ (s) to r_ (t):
A similar requirement for jrrj leads us to twist-free world line fun tions.

2.4.2. Denition. A twi e ontinuously dierentiable world line fun tion r

and the orresponding world line is alled
(i ) inertial if r = 0;


(ii ) uniformly a elerated if L r_ (
t); r_ (s)
r(s) = r(t) for all t; s 2 Dom r;
(iii ) twist-free if jr(t)jL r_ (s); r_ (t)
r(s) = jr(s)jr(t) for all t; s 2 Dom r:
It is quite evident that a twi e ontinuously dierentiable world line fun tion
r is inertial if and only if there are an xo 2 M and a uo 2 V(1) su h that
(t 2 Dom r):

r(t) = xo + uo t

2.4.3.

put

Let r be a twi e ontinuously dierentiable world line fun tion and

u := r_

I  V(1):

If r is uniformly a elerated, then, by denition,

u_ (s)

u(t) + u(s) u(t)
u_ (s)

1 u(t)
u(s)

= u_ (t)

(t; s 2 Dom r):

Fix an s 2 Dom r; put uo := u(s) 2 V(1); ao := u_ (s)

following rst-order dierential equation for u :
u_ = ao +

2 EI
uI
o

()

to have the

(u + uo )(u
ao )
:
1 u
uo

Unfortunately, it is rather ompli ated.

Another dierential equation an be derived, too, by using u(s)
u_ (s) = 0 and
observing that jrj = ju_ j =:  is onstant (the Lorentz boosts are g-orthogonal
maps). We obtain the equality

u_ (s)

u_ (t) =

u(t) + u(s) u(t) u(s)

1 u(t)
u(s)

u_ (s)

from (); dividing it by s t and letting s tend to t we get the extremely simple
se ond-order dierential equation
 = 2 u
u

whose general solution has the form

u(t) = uo ht +

ao

(t 2 I);

sht

()

where uo 2 V(1); ao 2 EI
uIo ; jao j = :
Equality () has been derived from (): It is not hard to see that t 7! u(t)
dened by () satises (); i.e. () and () are equivalent.
Finally, a simple integration results in the following.
Proposition. The twi e ontinuously dierentiable world line fun tion r is
uniformly a elerated if and only if there are an xo 2 M; a uo 2 V(1) and an
ao 2 EI
uoI su h that

r(t) = xo + uo

shjao jt

jao j

+ ao

hjaojt 1

jao j2

(t 2 Dom r):

2.4.4. If the twi eEdierentiable

world line fun tion r is twist-free, then there
u

are uo 2 V(1); no 2 I o ; jno j = 1 su h that for u := r_ the following dierential

equation holds:


(u + uo )(u
no )
:
u_ = ju_ j no +
1 u
uo
The method applied to uniformly a elerated world line fun tions to derive
another dierential equation works here as well. The reader is asked to perform
the al ulations to have
 ju_ j2 u_ (u_
u )
uju_ j4 = u
or
!
u_
u_
2
uju_ j = g

u
ju_ j2
provided that u_ is nowhere zero (g is the identity map of M):

2.5. World horizons

2.5.1. The light signals starting from a world point !x are in x + L!; mass-

points existing in x ontinue their existen e in x + T

: every phenomenon

o urring in x an in uen e only the o urren es in x + (T! [ L! ); the futurelike part of spa etime with respe t to x:
Conversely, only the o urren es in x +(T [ L ) an in uen e an o urren e
in x:
Consider a world line C: If x +(T! [ L! ) does not meet C; then an o urren e
in x annot in uen e the masspoint whose history is des ribed by C; in other
words, the masspoint annot have information about the o urren e in x: That
is why we all


x 2 Mj C \ x + (T! [ L! ) = ; = fx 2 Mj (C x) \ (T! [ L! ) = ;g

the indierent region of spa etime with respe t to C:

It an be shown that it is a losed set (Exer ise 2.7.3) whose boundary is
alled the world horizon of the world line C:
Obviously the indierent region is void if and only if the world horizon is void.

2.5.2. Consider a world line fun tion r: Then a world point x is not indierent
to the orresponding world line if and only if there is a t 2 Dom r su h that
r(t) x 2 T! [ L! i.e.


r(t) x 2  0
and

u

for an arbitrary u 2 V(1):

r(t) x < 0

2.5.3. The world horizon of an inertial world line is empty.

Indeed, take the inertial world line xo + uo
I; an arbitrary world point x
and look for t 2 I satisfying
(xo + uo t x)2  0;
uo
(xo + uo t

Sin e (xo x)2 = juo
(xo

written in the form

x)j2

ju
(xo x)j2 jt

o

x) < 0:

juo
(xo x)j2 ; the inequalities an be

uo
(xo
t uo
(xo

x)j2  0;
x) > 0;

they are satised for every

t > uo
(xo

x) + juo
(xo

x)j:

2.5.4. The indierent region of spa etime with respe t to the uniformly
a elerated global world line des ribed by
t 7! xo + uo

is

x 2 Mj

shjao jt

jao j

+ ao

hjao jt 1

(t 2 I)

()

jao j2

jaojuo + ao
(xo x)  1 :

Indeed, a ording to 2.5.2, the world point x is not indierent if and only if
there is a t for whi h
x2

sh2 t ( ht 1)2
sht
ht 1
+
+ 2uo
x
 0;
+ 2ao
x
2
2

2
sht
<0
uo
x


where

x := xo

x;

 := jao j:

The se ond inequality holds if t is large enough.

The rst inequality an be written in the form
x2 + 2uo
x

sht ht + 1
ht 1
+ 2(ao
x + uo
x 1)

2

 0:

Sin e sht ht tends to zero and ht tends to plus innity as t tends to plus

innity, we see that if (ao
x + u
x 1) < 0 then both inequalities hold if t
is large enough, i.e. an x out of the set () is not indierent with respe t to the
world line.
Take now an x in the set () su h that uo
x = 0: Then x is a non-zero
spa elike ve tor, thus x2 > 0 and we see that the previous inequality does not

hold for any t be ause ht 1  0 : x is indierent with respe t to the world
line.
To end the proof, note that uo + to is a lightlike ve tor, hen e x is indierent
if and only if x + (uo + ao) is indierent
for some  2 I: Choose  in su h a


way that uo
xo x (uo + ao ) = 0:

2.6. Newtonian equation

2.6.1. In the spe ial relativisti spa etime model there is a single measure
line, I: Time periods and distan es are measured by the elements of I: There is no
natural way to introdu e dierent measure lines for time periods and distan es.
This re e ts the experimental fa t that light speed is a universal onstant; thus
a time unit indi ates a distan e unit as well: the distan e overed by a light
signal during the unit time period. The SI physi al dimensions are extraneous
to spe ial relativity.
The light speed in the SI units is
= (2;9979 : : : )108

m
:
s

Measuring distan es by light signals we arrive at the denition

m := (3;3356 : : : )10 9 s:

Now it is totally senseless to introdu e a measure line for masses; using the
Plan k onstant and the formulae of I.2.4.1 and the denition above we get that
I  RI is the measure line of masses and
1
kg := (8;5214 : : : )1050 :
s

2.6.2. Sin e a eleration values are elements of IM

I and \the produ t of mass
and a eleration equals the for e", the for e values are elements of I
IM

I 
M  M ; moreover, we take into a ount that the momentary a eleration
I
I
I
I
value of a masspoint is g-orthogonal to the orresponding velo ity value.
Thus we a ept that a for e eld is a dierentiable mapping
f

su h that

: M  V(1)

u
f (x; u) =

 MI 

(x; u) 2 Dom f :

The history of the material point with mass m under the a tion of the for e
eld f is des ribed by the Newtonian equation

mx = f (x; x_ )
i.e. the world line fun tion modelling the history is a solution of this dierential
equation.

2.6.3. Some of the most important for e elds in spe ial relativity, too,
an be derived from potentials; e.g. the ele tromagneti eld. However, the
gravitational eld annot be des ribed by a potential; this problem will be
dis ussed later (Chapter III).
A potential is a twi e dierentiable mapping
K

: M

 M

(in other words, a potential is a twi e dierentiable ove tor eld).

The eld strength orresponding to K is D ^ K : M
M ^ M (the
antisymmetri or exterior derivative of K ; see VI.3.6).
The for e eld f has a potential (is derived from a potential) if
| there is an open subset O  M su h that Dom f = O  V(1);
| there is a potential K dened on O su h that

f (x; u) = F (x)
u

(x 2 O; u 2 V(1))

where F := D ^ K :
It is worth mentioning: F (x) is antisymmetri , hen e u
F (x)
u = 0; as it
must be for a for e eld.

2.6.4. In the non-relativisti spa etime model a for e eld an be independent

of either of its variables, in parti ular, it an be a onstant map. In the present
ase, on the ontrary, a non-zero for e eld annot be independent of velo ity,
in parti ular, it annot be a onstant map.
We ould try to dene a onstant for e eld in su h a way that the orresponding Lorentz boosts map its values into ea h other, i.e. f would be onstant if
L(u0 ; u)
f (x; u) = f (x; u0 )

for all possible x; u and u0: However, su h a non-zero eld annot exist (Exer ise
2.7.5): there is no non-zero spe ial relativisti onstant for e eld !

2.7. Exer ises

1. Prove that the uniformly a elerated world line fun tion given in 2.4.3
satises
a
r(t) = xo + uo t + o t2 + Ordo(t3 ):
2
E
u
o
2. Let uo 2 V(1); ao 2 I
I and  : I
I a ontinuously dierentiable
fun tion dened on an interval. Demonstrate that the world line fun tion r for
whi h
p
r_ = uo  2 + 1 + ao 

holds is twist-free.
3. The indierent region of spa etime with respe t to the world line C has
the omplement
[
fz + (T [ L )g :
z2C
Using L! + T! = T! show that it equals
[

z2C

fz + T g

whi h, being a union of open sets, is open. Consequently, the indierent part of
spa etime with respe t to C is losed.
4. Let r be a global world line fun tion and put u := r:_ Prove that the world
horizon of the orresponding world line is empty if one of the following onditions
holds:
(i ) there exist tlim
!1 u(t);
(ii ) u is periodi , i.e. there is a to > 0 su h that u(t + to ) = u(t) for all t 2 I:
(Hint:
(i ) V(1) is losed, hen e the limit belongs to it. (ii ) Put zo :=
R to
u
(
t
)
d
t
;
uo := jzzoo j and onsider the inertial world line r(to ) + uo
I:)
0
5. Let  : V(1) ! M be a fun tion su h that
u
(u) =

and (u0 ) := L(u0 ; u)
(u)

u0 ; u 2 V(1)

Prove that  = 0: (Hint: L(u00 ; u0)
L(u0 ; u)
(u) = L(u00 ; u)
(u) must
hold; applying Proposition 1.3.9 nd appropriate u00 and u0 for a xed u in su h
a way that the equality fails.)

3. Inertial observers

3.1. Observers

3.1.1. We an repeat word by word what we said in I.3.11 to motivate the

following denition.
Denition. An observer is a smooth map U : M V(1) whose domain is
onne ted.
If Dom U = M; the observer is global.
The observer is alled inertial if it is a onstant map.
V(1) is a subset of MI ; the dierentiability (smoothness) of a map from M into
V(1) means the dierentiability (smoothness) of the map from M into MI :

3.1.2. Let U be an observer. The integral urves of the dierential equation

(x : I  M)?
x_ = U (x)

are evidently world lines.

As in the non-relativisti ase,
AU

:= DU
U : M

 I M
I

is the a eleration eld orresponding to U :

3.1.3. Again we an repeat the arguments onrming that the spa e of an

observer is the set of its maximal integral urves.
Denition. Let U be an observer. Then EU ; the set of maximal integral
urves of U ; is the spa e of the observer U or the U -spa e.
Again a maximal integral urve of U is alled a U -line if onsidered to be a
subset of M and is alled a U -spa e point if onsidered to be an element of EU :
CU (x) will stand for the (unique) U -line passing through x; we say that CU (x)
is the U -spa e point that x is in ident with.
3.1.4. There is no absolute time in the spe ial relativisti spa etime model.
We ould think that | on the analogy of the observer spa e | the time of an
observer an be dened in a natural way.
How do we try to introdu e the observer time? We ought to determine
somehow whi h world points have the same instant from the point of view of the
observer, i.e. whi h world points are onsidered to be simultaneous.
We know that every U -line has its proper time: in every U -spa e point the
own time of the point passes. Evidently, we expe t that simultaneity is related

to the proper time of spa e points. However, in general, \time passes dierently
in dierent spa e points" and that is why simultaneity annot be dened in a
natural way.
The exa t meaning of the above phrase in parentheses will be laried later.
Inertial observers are good ex eptions: the lines of an inertial observer are
evidently parallel straight line segments: \time passes in the same uniform way
in ea h spa e point".

3.2. The time of an inertial observer

3.2.1. Our experien e that light propagates isotropi ally with respe t to an
arbitrary inertial observer (with the same speed in all dire tions) suggests the
following method for determining simultaneity.
A lo k at a spa e point says the time, another lo k at another spa e point
says the time, too. We should like to syn hronize them: \when one of them
says 12 then (at the same moment) let the other say 12 as well". We make
su h a syn hronization by the everyday method: a radio signal (i.e. in fa t a
light signal) is emitted by the lo k that says 12 at the studio and, hearing the
signal, we set our lo k. Of ourse hearing the signal we do not set the lo k
to 12 be ause we know that some time passed between emission and re eption.
Knowing the distan e between the studio and our pla e we know the time passed
owing to the onstan y of light speed. How do we measure the distan e between
the studio and our pla e? With the aid of a radar, i.e. by means of light signals.
Let the radar be at the studio. It emits a light signal toward us, the light signal
is re e ted by us, the radar re eives the re e ted signal and measures the time
passed between emission and re eption. Knowing this time passed we know the
distan e owing to the onstan y of light speed.
We an simplify the pro edure in su h a way that the time signal and the
radar signal be the same. Let us see this simplied version.
Take two dierent spa e points of the observer. Put a sour e of light in one of
them and lo ate a mirror in the other. Emit a light signal toward the mirror and
re eive the re e ted signal. Sin e light travels the same time there and ba k, the
re e tion at the mirror is simultaneous with that time point at the sour e whi h
halves the interval between emission and re eption.
u:

3.2.2. Let U be a global inertial observer having the onstant velo ity value

We want to determine the ondition that the world point y is simultaneous

with the world point x; a ording to U ; x and y symbolize the middle point
between emission and re eption, and the re e tion at the mirror, respe tively.

The world point y is to be simultaneous with x a ording to U if there is a

I su h that y (x ut) and y (x + ut) are lightlike ve tors:

t2

y (x ut) 2 = 0;
(y x)2 + 2(y x)
ut t2 = 0;

whi h give

y x 2 Eu;

y (x + ut) 2 = 0;
(y x)2 2(y x)
ut

in other words,

t2

=0

y 2 x + Eu :

All these have been heuristi onsiderations to support the following denition.
Denition. Let U be a global inertial observer having the onstant velo ity
value u:
The set of world points simultaneous with x; a ording to U ; is x + Eu ; the
hyperplane passing through x and dire ted by Eu:
The set of hyperplanes dire ted by Eu; denoted by IU ; is alled the time of
the observer or the U -time. Its elements are the U -instants.
It is important that simultaneity with respe t to U is a symmetri relation
on M: if y is U -simultaneous with x; then x is U -simultaneous with y:
We emphasize that U -time is dened only for inertial U :
In the non-relativisti ase there is an absolute time giving absolute simultaneity and it is onvenient to identify instants with the orresponding simultaneous
hyperplanes.
Here we dene simultaneity (with respe t to an inertial observer) by hyperplanes, and then we dene an instant (with respe t to the observer in question)
to be a simultaneous hyperplane.
Evidently, dierent inertial observers determine dierent simultaneities.

3.2.3. Simultaneity with respe t to an inertial observer is in a ordan e with

the time passing in the observer spa e points: the same time passes in dierent
spa e points between simultaneous o urren es.
A hyperplane t 2 IU (a U -instant) and a line q 2 EU (a U -spa e point) meet
in a single world point whi h will be denoted by
q ? t:

Proposition. Let U be an inertial observer having the onstant velo ity

value u: Take two U -spa e points q and q0 and two U -instants t and s: Then
the time passed along q between x := q ? t and y := q ? s (the inertial time
t(0 x; y) 0between x and y) equals the
time passed along q0 between x0 := q ? t and
0
0
y := q ? s (the inertial time t(x ; y ) between x0 and y0 ):

Proof. We have that x0

x and y0 y are g-orthogonal to u and

y x = ut(x; y);
y0 x0 = ut(x0 ; y0):

Multiplying the equality

(y0 x) = (y0
by u we obtain

x0 ) + (x0

x) = (y0 y) + (y x)

t(x0 ; y0) = t(x; y):

3.2.4. The previous result oers the possibility to dene the time passed
between two U -time instants t and s as the time passed between t and s in an
arbitrary U -spa e point.

More losely, take an arbitrary world point x in the hyperplane t; nd the

unique world point y in s su h that the straight line passing through x and y is
a U -line, and then let s t := t(x; y) = u
(y x) be the time passed between
t and s:
Note that u
(y x) is the same for all y 2 s and x 2 t; thus avoiding
the notation t(x; y) we an omit the requirement that x and y be on the same
U -line.
Proposition. IU ; the U -time, endowed with the subtra tion

s t := u
(y x) =

u
(y

x)

(x 2 t; y 2 s)

is an ane spa e over I:

The proof is immediate.

Thus the time of a global inertial observer is a one-dimensional oriented ane
spa e.

3.3. The spa e of an inertial observer

3.3.1. Our heuristi notion about the ane stru ture of the spa e of a physi al
observer means that we assign a ve tor to two spa e points; the assignment
supposes simultaneity.
The simultaneity introdu ed previously oers us indeed a natural way to dene
an ane stru ture on the spa e of an inertial observer.
First we prove an analogue of Proposition 3.2.3.
Proposition. Let U be a global inertial observer with the onstant velo ity
value u: Take two U -spa e points q and q0 and two U -instants t and s: Then the

ve tor between x0 := q0 ? t and x := q ? t equals the ve tor between y0 := q0 ? s

and y := q ? s:

Proof.0 By 0Proposition 3.2.3 we have t(x; y) = t(x0 ; y0) =: t: Consequently,

y x=y

x = ut and then
(y0

y) + (y x) = (y0

gives the desired result:

y0 y = x0

x0 ) + (x0

x)

x:

3.3.2. The previous result oers the possibility to dene the ve tor between
the U -spa e points q0 and q to be the ve tor between the world points x0 and x
that are simultaneous and in ident with q0 and q; respe tively.
Note that x0 x = u
(x0 x); if x0 and x are simultaneous with respe t to
U ; and u
(x0 x) is the same for all x0 in q 0 and x in q: As a onsequen e,
using u
(x0 x); we an omit the requirement of simultaneity.
Proposition. EU ; the spa e of the global inertial observer U ; endowed with

the subtra tion

q0

q := u
(x0

is an ane spa e over Eu:

x)

(x0 2 q0 ; x 2 q)

The proof is immediate.

Re all that (Eu ; I; bu ) is a Eu lidean ve tor spa e. Thus we an say that

the spa e of a global inertial observer is a three-dimensional oriented Eu lidean
ane spa e.

3.4. Splitting of spa etime

3.4.1. Let us take a global inertial obvserver U with the onstant velo ity
value u:
The observer assigns to every world point x the U -time point U (x); the set
of world points simultaneous with x a ording to U : U (x) = x + Eu; as well as
the U -spa e point CU (x) that x is in ident with: CU (x) = x + u
I:
It is worth listing the following relations regarding the ane stru tures of IU
and of EU as well as the mappings U : M ! IU and CU : M ! EU :
(i ) (y + Eu) (x + Eu ) = u
(y x)
(x; y 2 M);
(ii ) (x + x + Eu) = (x + Eu ) u
x
(x 2 M; x 2 M);
(iii ) x + Eu = y + Eu if and only if y x is g-orthogonal to u;
and
(iv ) (x0 + u
I) (x + u
I) = u
(x0 x)
(x0 ; x 2 M);
(v ) (x + x) + u
I = (x + u
I) + u
x
(x 2 M; x 2 M);
(vi ) x + u
I = x0 + u
I if and only if x0 x is parallel to u;
moreover,



(vii ) (y + Eu) \ (x + u
I) = x + u u
(y x)
(x; y 2 M)
or, in another form,


(y + Eu) ? (x + u
I) = x + u u
(y x)
3.4.2. It is trivial by the previous formulae (i ) and (iv ) that
U : M ! IU ;
x 7! x + Eu

is an ane map over u = u and

CU : M ! EU ;
is an ane map over u :

x 7! x + u
I

Denition.

HU := (U ; CU ) : M ! IU  EU ;
x 7! (x + Eu ; x + u
I)
is the splitting of spa etime a ording to the global inertial observer U :
Proposition. The splitting HU is an orientation-preserving ane bije tion
over the linear map hu = (u ; u) ( f. 1.3.5) and
(t 2 IU ; q 2 EU ):
HU 1 (t; q) = q ? t

3.4.3. We an simplify a number of formulae and al ulations by hoosing a

U -time

point to and a U -spa e point qo and ve torizing U -time and U -spa e:

IU ! I;
t 7! t to ;
EU ! Eu ;
q 7! q qo :

Choosing to and qo is equivalent to hoosing a \spa etime referen e origin"

o 2 M : fog 2 qo \ to ; U (o) = to ; CU (o) = qo :
The pair (U ; o) is alled a global inertial observer with referen e origin. We
an establish the ve torized splitting of spa etime due to (U ; o) :

HU ;o : M ! I  Eu; x 7! U (x) U (o); CU (x) CU (o) =

= u
(x o); u
(x o) :

Thus, if Oo denotes the ve torization of M with origin o then

HU ;o = hu Oo :

3.5. Exer ise

Dene the basi observer in the arithmeti spa etime model.
Choose the zero in R1+3 to be a referen e origin for the basi observer. Then
ve torized splitting of spa etime is the identity map of R1+3 :

4. Kinemati s

4.1. Motions relative to an inertial observer

4.1.1. Let U be a global inertial observer with onstant velo ity

value u:
Take a world line fun tion r:
Then the fun tion U r : I
IU assigns U -time points to proper time
points of r: This fun tion is pie ewise twi e dierentiable and its derivative

(U r)
= u
r_ = u
r_
is everywhere positive (see 1.3.1). Consequently, U
in reasing, has a monotone in reasing inverse

zU := (U r) 1 : IU

is stri tly monotone

I

whi h gives the proper time points of r orresponding to U -time points; moreover, its derivative omes from the inverse of the derivative of U r :

z_U (t) =

u
r_ (zU (t))

(t 2 Dom zU ):

4.1.2. The history of a material point is des ribed by a world line fun tion
r: A global inertial observer U observes this history as a motion des ribed by a
fun tion rU assigning to U -time points the U -spa e points where the material
point is at that U -time point.
To establish that fun tion, sele t a U -time point t; nd the orresponding
proper time point zU (t) and the spa etime position r(zU (t)) of the material
point; look for the U -spa e point CU (r(zU (t))) that the world point in question
is in ident with.
Denition.

rU : IU

 EU ;

t 7! CU (r(zU (t))) = r(zU (t)) + u

I

is alled the motion relative to U ; or the U -motion, orresponding to the world

line fun tion r:

4.1.3. The question arises, whether the history, i.e. the world line fun tion,
an be regained from the motion. Later a positive answer will be given (Se tion
4.4).

4.1.4. Some formulae and al ulations be ome simpler if we use a ve torization of U -time and U -spa e, i.e. we introdu e a referen e origin o (see 3.4.3.).
Then U r U (o) = u
(r o) : I
I is dierentiable, its derivative
equals the derivative of U r; hen e it is stri tly monotone in reasing, its inverse

u
(r

zU ;o :=

o) 1 :

II

is monotone in reasing as well and

z_U ;o(t) =

(t 2 Dom zU ;o):

u
r_ (zU ;o (t))

The motion relative to (U ; o) is

rU ;o :

I  Eu ;

t 7! rU (t)

CU (o) = u
r(zU ;o (t)) o :

4.2. Relative velo ities

4.2.1. Proposition. Let U be a global inertial observer and let r be a

dierentiable world line fun tion; then rU is dierentiable and
r_U =

r_

u
r_

zU :

Proof. Re alling that CU : M ! EU is an ane map over u; we obtain

d
C (r(z (t))) = u
r_U (zU (t))z_U (t);
dt U U
then taking into a ount the formula in 4.1.1 for the derivative of zU ; we easily
nd the desired equality.
It is evident that, hoosing a referen e origin o; we have

r_U (t) =

r_U ;o =

r_

u
r_

zU ;o:

4.2.2. Sin e rU des ribes the motion, relative to the observer U ; of a material
point, r_U is the relative velo ity fun tion of the material point. This suggests
the following denition.
Denition. Let u and u0 be elements of V(1): Then
u0
u
vu0 u :=
u
u0

is alled the relative velo ity of u0 with respe t to u:

Proposition.EuFor all u; u0 2 V(1)
(i ) vu0 u is in I ;
(ii ) vu0 u = vuu0 if and only if u = u0;
(iii ) jvu0 uj2 = jvuu0 j2 = 1 (u
u1 0)2 < 1:
Proof. (i ) is trivial, (iii ) is demonstrated by a simple al ulation.
Suppose
u
u0
u0 ;
0 u= 0
u
u

u
u

multiply the equality by u to have

0=

u0
u

u0
u;

(u0
u)2 = 1:

A ording to the reversed Cau hy inequality (see 1.3.1) this is equivalent to

u = u0 :

Earlier we obtained that Eu and Eu0 are dierent if and only if u 6= u0 and
in this ase vu0 u
I (vuu0
I) is a one-dimensional linear subspa e in Eu (Eu0 );
orthogonal to Eu \ Eu0 (see 1.3.7) whi h oers an alternative proof of (ii ).

4.2.3. Let us take now two global inertial observers with onstant velo ity
values u and u0 : Then vu0 u and vuu0 are the relative velo ities of the observers
with respe t to ea h other. Then (iii ) of the previous proposition implies that
vu0 u = 0 if and only if u = u0 : Moreover, (ii ) says that in ontradistin tion to
the non-relativisti ase and to our habitual \eviden e", the relative velo ity of
u0 with respe t to u is not the opposite of the relative velo ity of u with respe t
to u0 ; ex ept the trivial ase u = u0 :
It is worth emphasizing this fa t be ause in most of the textbooks one takes it
for granted that vu0 u and vuu0 are equal: \if an observer moves with velo ity
v relative to another then the se ond observer moves with velo ity v relative
to the rst one".
Nevertheless, no harm omes be ause ve tors are given there by omponents
with respe t to onvenient bases and then the omponents of vu0 u and vuu0
be ome opposite to ea h other.
The reason of non-equality of vu0 u and vuu0 is that the spa es of dierent
inertial observers are ane spa es over dierent ve tor spa es.
However, we have a ni e relation between the two ve tor spa es in question:
the Lorentz boost from u to u0 maps Eu onto Eu0 in a natural way and maps
vu0 u into vuu0 :
Having the equality
L(u0 ; u)
vu0 u = vuu0

(see 1.3.8), we already know how to hoose bases in Eu and in Eu0 to get the
mentioned usual relation between the omponents of relative velo ities: take an
arbitrary
ordered basis (e1 ; e2 ; e3 ) in Eu and the basis e0i := L(u0; u)
ei j i =


1; 2; 3 in Eu0 : Now,
if

vu0 u

3
X
i=1

v i ei

then

vuu0

3
X
i=1

( vi )e0i :

4.2.4. We often shall use the equalities

u
u0

and

u0

jvu0 u j2

u + vu0 u

jvu0 u j2

deriving from 4.2.2 (iii ) and (i ) and from the denition of vu0 u:

4.2.5. The relative velo ities in the non-relativisti spa etime model form
a Eu lidean ve tor spa e. Here the relative velo ities with respe t to a xed
u 2 V(1) form the unit open ball in the Eu lidean ve tor spa e EIu :
Bu :=




E
u
2

v2
I  jv j < 1 :

The set of all relative velo ities is

u2V(1)

Bu ; a ompli ated subset of MI :

4.3. Addition of relative velo ities

4.3.1. As a onsequen e of the stru ture of relative velo ites, the \addition of
relative velo ities" is not a ve tor addition, i.e. if u; u0 ; u00 are dierent elements
of V(1) then | in ontradistin tion to the non-relativisti ase | we have
vu00 u

6= vu00 u0 + vu0 u:

The left-hand side is an element of EIu ; the right-hand side is the sum of
elements in EIu0 and in EIu whi h indi ates that they annot be, in general,
equal.

We might think that the onvenient Lorentz boost helps us; however,
vu00 u

6= L(u; u0)
vu00 u0 + vu0 u

be ause the length of the ve tor on the right-hand side an be greater than 1:

4.3.2. To nd the formula for the addition of relative velo ities | i.e. to
express vu00 u by means of vu00 u0 and vu0 u | we need some auxiliary formulae:
u0 =
u=

vu0 u

vuu0

jvu00 u j2

jvu0 u j2 + vuu0
;
jvu0 u j2

jvu0 u j2 + vu0 u ;
jvu0 u j2
1 vu00 u0
vuu0
q
:
jvu00 u0 j2 1 jvuu0 j2

Then starting from the equality

(u00 =)

u0 + vu00 u0

jvu00 u0 j2

u + vu00 u

jvu00 u j2

we arrive at the following result.

Proposition. Let u; u0 and u00 be elements of V(1): Then
vu00 u

vu00 u0

vu00 u0
vuu0 v 0
jvuu0 j2 uu

q

jvuu0 j2 +
vu00 u0
vuu0
1

vu00 u0
vuu0 v 0
uu
jvuu0 j2

4.3.3. It is important that vu00u annot be expressed as a fun tion of vu00 u0

and vu0 u or as a fun tion of vu00 u0 and vuu0 ; we need vuu0 or vu0 u as well.
We an derive other formulae, too; e.g. we an involve vu00 u0 ; vu0 u00 and vu0 u
instead of vu00 u0 ; vu0 u and vuu0 :
It is worth mentioning the spe ial ase when vu00 u0 is parallel to vuu0 :
vu00 u0 = vuu0 for some positive real number : Then
vu00 u

v 0 + ( vu0 u )
= uu
:
1 + jvu0 u j2

4.3.4. Putting v00 := vu00 u; v0 := vu00u0 ; v := vu0u;

v := vuu0 (but this is

impossible!) in the expression for vu00 u in Proposition 4.3.2, we re ognize the
formula of usual treatments for the addition of relative velo ities. In parti ular,
if vu00 u0 is parallel to vuu0 ; we get the most frequently ited Einstein formula
v + v0
v 00 =
:
1 + jv j jv 0 j

4.4. History regained from motion

4.4.1. Given a motion relative an inertial observer U ; i.e. a pie ewise twi e
dierentiable fun tion m : IU EU ; an we determine the orresponding world
line fun tion r su h that m = rU ?
Sin e rU = CU r zU and U r is the inverse of zU ; we have

(idIU ; rU ) = (U r zU ; CU r zU ) = HU r zU :
Consequently, given the motion m;

r := HU 1 (idIU ; m) zU 1
will be the orresponding world line fun tion.
Similarly, if the ve torized motion m : I

 Eu is known, then

r := HU 1;o (idI ; m) zU 1;o

is the required world line fun tion whi h an be given by a simple formula:
t 7! o + m(zU 1;o (t)) + uzU ;o (t):

4.4.2. The previous formulae are not satisfa tory yet be ause zU and zU ;o
are dened by U and (U ; o) together with the world line fun tion r to be found;
we have to determine them | or their inverse | from U ; (U ; o) and the motion
m or m:
Equalities in 4.1.1 and 4.1.4 result in


zU 1
=

jm_ j2

zU 1;o
=

1
1

jm_ j2

_ are given fun tions, hen e zU 1 and zU 1;o an be obtained by a simple

m_ and m
integration.

4.4.3. Let us onsider the basi observer with the zero as referen e origin
in the arithmeti spa etime model (see Exer ise 3.5). A motion is given by a
fun tion m : R R3 : Let h : R R be a primitive fun tion of p1 1jm_ j2 : Then


R
R  R3 ; t 7! h(t); m(h(t) is the world line fun tion regained from the
motion m:

4.5. Relative a elerations

Let r be a world line fun tion and let U be a global inertial observer with
onstant velo ity value u: Then rU is twi e dierentiable and a dierentiation
of the equality in 4.2.1 yields
1
rU =
(u
r_ )2

g + r_ u

ur_

r zU :

If a referen e origin o is hosen as well, rU ;o is given by a similar formula,

with zU ;o instead of zU :
We see that, in ontradistin tion to the non-relativisti ase, the relative
a eleration does not equal the absolute one. Of ourse, the relative a eleration
takes values in IE
uI ; the absolute a eleration takes values in EIr
_ (tI) :

4.6. Some parti ular motions

4.6.1. Take the inertial world line fun tion

(t 2 I) ():

r(t) = xo + uo t

Let U be a global inertial observer with onstant velo ity value u:

Then
1
z_U =
u
uo

(see 4.1.1) from whi h we get immediately

zU (t) =

t to

u
uo

for some to 2 IU : Consequently (see 3.4.1(v )),



rU (t) = xo + uo

t to

u
uo

= qxo + vuo u (t to )


u
+ u
I = xo + u
I + u o (t to ) =

u
uo

(t 2 IU );

where qxo := xo + u
I is the U -spa e point that xo is in ident with.
This is a uniform and re tilinear motion.
Conversely, suppose that we are given a uniform and re tilinear motion relative to the inertial observer, i.e. there are a q 2 EU ; a to 2 IU and a v 2 EIu
su h that
rU (t) = q + v(t to )
(t 2 IU ):
Then letting xo denote the unique world point in the interse tion of q and to
and putting uo := pu1 +jvvj2 ; the world line fun tion of form () gives rise to the
given motion.

4.6.2. Take the uniformly a elerated world line fun tion

r(t) = xo + uo

sht
ht 1
+ ao

2

(t 2 I)

where  := jao j; then

sht
:

Let U be the global inertial observer with onstant velo ity value u: The
formulae will be more tra table if we hoose a referen e origin o: Then

r_ (t) = uo ht + ao

zU 1;o(t) =

u
(r(t)

= u
(xo

o) =
o)

u
uo

sht


u
ao

ht 1
2

(t 2 I):

Let us onsider the spe ial ase u
ao = 0; then

zU ;o(t) =
where to := u
(xo

arsh tu
tuoo


(t 2 I);

o): Thus
q

rU ;o (t) = u
xo

t to
o + uo
u
uo

1 + 2 ((ut
utoo))2
+ ao
2

= qo + vuo u (t to ) + bo

1 +  2 (t to )2
2

1
A=

(t 2 I);

where
qo

:= u
(xo

o);

bo

:= ao 1

jvu u j2 ;
o

 2 := 2 1

jvu uj2 :
o

4.7. Light speed

4.7.1. We wish to determine the motion of a light signal with respe t to an

inertial observer. The pro edure will be similar to that in Se tions 4.1 and 4.2.
We introdu e the notation
V(0) :=



 2
M
 w = 0; w
I+  L! :
w2
I

The elements of V(0) are future-dire ted lightlike ve tors of otype I: Though
the notation is similar to V(1); observe a signi ant dieren e: if two elements
in V(1) are parallel then they are equal; on the other hand, if w is in V(0) then
w; too, is in V(0) for all  2 R+ :
Let U be a global inertial observer with onstant velo ity value u: Let us
onsider a light signal F; i.e. a straight line dire ted by a ve tor in V(0): The
motion of the light signal with respe t to the observer is des ribed by

fU : IU

! EU ;

t 7! (F ? t) + u
I

where F ? t denotes the unique element in the interse tion of the straight line F
and the hyperplane t:

If t; s 2 EU then t s = u
(F ? t F ? s) by the denition of the ane

stru ture of IU (see 3.4.2); then we easily nd that
t s
;
F?t F?s=w
u
w

hen e

fU (t) fU (s) = u
(F ? t F ? s) =

w
u
w

(t s):

Thus the light signal moves uniformly on a straight line relative to the observer.

4.7.2. Denition. Let w 2 V(0) and u 2 V(1): Then

vwu

:=

w
u
w

is the relative velo ity of w with respe t to u:

Proposition. vwu is an element of EIu and

jvwuj = 1:
Observe that given an arbitrary u; w and w have the same relative velo ities
with respe t to u:
We do not dene the relative velo ity of u with respe t to w:
A ording to the previous proposition, the magnitude of the relative velo ity
is the same number, namely 1, for every light signal and every inertial observer.
Light signals propagate isotropi ally with respe t to all inertial observers.

4.7.3. Re all that relative velo ity values have magnitude but two relative
velo ity values need not have an angle between themselves; relative velo ities
with respe t to the same element of V(1) do form an angle.
Now we look for the relation between ertain angles formed by relative velo ity
values. The physi al situation is similar to that in I.6.2.3. A ar is going on a
straight road and it is raining. The raindrops hit the road and the ar at dierent
angles relative to the dire tion of the road. What is the relation between the two
angles? Now we an treat another question, too, onsidering instead of raindrops
light beams ( ontinuous sequen es of light signals) arriving from the sun.
Let u and u0 be dierent elements of V(1) (representing the absolute velo ity
values of the road and of the ar, respe tively). If w is an element of V(1) [ V(0)
(repesenting the absolute velo ity value of the raindrops or the absolut dire tion
of the light beam), w 6= u; w 6= u0 ; then
v
v 0
(w) := ar os wu u 0u ;

jvwu j jvu u j

v 0
(
0 (w) := ar os wu

vuu0 )

jvwu0 j jvuu0 j

are the angles formed by the relative velo ity values in question. A simple
al ulation veries that
jvwu0 j os 0 (w) + jvu0 u j
jvwuj :
os (w) = jvwuj
1 + jvwu0 j jvuu0 j os 0 (w)

If w 2 V(0) then jvwuj = jvwu0 j = 1 and

os (w) =

os 0 (w) + jvuu0 j
:
1 + jvuu0 j os 0 (w)

This formula is known as the aberration of light : two dierent inertial observers see the same light beam under dierent angles with respe t to their
relative velo ities; the angles are related by the above formula.

5. Some observations

5.1. Physi ally equal ve tors in dierent spa es

5.1.1. It is an important fa t that the spa es of dierent global inertial

observers are ane spa es over dierent ve tor spa es. Thus it has no meaning,
in general, that a straight line segment (a ve tor) in the spa e of an inertial
observer oin ides with a straight line segment (with a ve tor) in the spa e of
another inertial observer.
Let us onsider two dierent inertial observers U and U 0 with onstant
velo ity values u and u0 ; respe tively. The spa es of U and U 0 are ane spa es
over Eu and Eu0 ; respe tively. We know that Eu \ Eu0 is a two-dimensional
subspa e, orthogonal to vu0 u and to vuu0 :
If a ve tor between two points in the U -spa e lies in Eu \ Eu0 then we an
nd two points in the U 0 -spa e having the same ve tor onne ting them. We
have troubles only with ve tor outside this two-dimensional subspa e.
To relate other ve tors, too, we start from the rational agreement that \if you
move with respe t to me in some dire tion in my spa e then I move with respe t
to you in the opposite dire tion in your spa e". This suggests that the ve tor
vu0 u in Eu and the ve tor vuu0 in Eu0 ould be onsidered the same for all
 2 I: More generally, every ve tor in Eu has the form
vu0 u + q
and every ve tor in Eu0 has the form

0 vuu0 + q0

( 2 I; q 2 Eu \ Eu0 )

(0 2 I; q0 2 Eu \ Eu0 ):

The observers agree that two su h ve tors are onsidered to be the same if
and only if
0 = ;
q0 = q:
We have a ni e tool to express this agreement: the Lorentz boost.

Denition. The ve tors q0 in Eu0 and q in Eu are alled physi ally equal if

and only if

L(u0 ; u)
q

= q0 :

We emphasize that the equality of ve tors in dierent observer spa es makes

no original sense, in general; we agreed to dene it onveniently.

5.1.2. To be physi ally equal in dierent observer spa es, a ording to our
onvention, is a symmetri relation, but is not a transitive relation.
Indeed, if q0 = L(u0 ; u)
q then q = L(u; u0)
q0 : the relation is symmetri .
However, if u; u0 and u00 are not oplanar, then there are q; q0 and q00 in su h
a way that
q 0 = L(u0 ; u)
q ;
q 00 = L(u00 ; u0 )
q 0 ;
q 00 is not parallel to L(u00 ; u)
q
(Exer ise 1.9.6); q0 is physi ally equal to q; q00 is physi ally equal to q0 ; but q00
is not physi ally equal to q : the relation is not transitive.
In parti ular, \if a straight line in your spa e is parallel to a straight line in
my spa e and a line in his spa e is parallel to your line then his line need not be
parallel to mine".
This is a rather embarrassing situation but there is no es ape. The truth of the
ommon sense that the relative velo ity of an observer with respe t to another
is the opposite of the other relative velo ity and the transitivity of parallelism
ex lude ea h other.
5.2. Observations on erning spa es

5.2.1. In 5.1.1 an agreement is settled what the equality | in parti ular the
parallelism | of ve tors in dierent observer spa es means.
Now the question arises whether a straight line segment in the spa e of an
inertial observer is observed by another observer to be a straight line segment
parallel to the original one. The question and the answer are formulated orre tly
as follows ( f. I.7.1.2).
Let Uo and U be global inertial observers with onstant velo ity values uo
and u; respe tively. Let Ho be a subset (a geometri al gure) in the Uo -spa e.
The orresponding gure observed by U at the U -instant t | alled the tra e of
Ho at t in EU | is the set of U -spa e points that oin ide at t with the points
of Ho :
fq ? t + u
Ij q 2 Ho g
where q ? t denotes the unique world point in the interse tion of the line q and
the hyperplane t:

Introdu ing the map

q 7! q ? t + u
I

Pt : EUo ! EU ;

we see that the tra e of Ho at t equals Pt [Ho : It is quite easy to see (re all the
denition of subtra tion in the observer spa es) that

Pt (q2 ) Pt (q1 ) = q2 ? t q1 ? t = q2
= Puuo
(q2

q1 );

q1 + uo

u
(q2 q1 )
uo
u

where Puuo is the proje tion onto Eu along uo

I (see Exer ise 1.9.7).

Sin e the restri tion of Puuo onto Euo ; denoted by Auuo ; is a linear bije tion
between Euo and Eu; Pt is an ane bije tion over Auuo :

5.2.2. We an easily nd that

Auu
q = q if q 2 Eu \ Eu
q

Auuo
vuuo

i.e. if q is orthogonal to vuuo ;

jvuu j2 vu u :
o

The linear bije tion Auuo resembles the restri tion onto Euo of the Lorentz
boost L(u; uo ); an essential dieren e is that it maps vuuo into vuo u multiplied
by a real number less than 1: Consequently, Auuo is not an orthogonal map ; it
does not preserve either lengths or angles whi h is illustrated as follows:

5.2.3. Every gure in the Uo -spa e is of the form qo + Ho; where qo 2 EU and
Ho  Eu ; then Pt [qo + Ho = Pt (qo) + Auu [Ho : Consequently, the observed
o

gure and the original one are not ongruent, in general.

If Lo is a straight line segment in the Uo -spa e, then its tra e is a straight
line segment, too. However, the observed segment and the original one are not
parallel, in general: if Lo is dire ted by the ve tor eo ; Lo = qo + Reo ; then Pt [Lo
is dire ted by Auuo
eo :
Auuo
eo is parallel to eo ; by denition, if there is a real number  su h that
Auuo
eo = L(u; uo )
eo
whi h o urs if and only if eo is in Eu \ Euo or in vuuo
I:
Thus a straight line segment Lo in the Uo -spa e is observed by U to be parallel
to Lo if and only if Lo is
| either orthogonal to vuuo
| or parallel to vuuo :

5.2.4. Let L1 and L2 be rossing straight lines in the Uo -spa e. Then U

observes at every instant that they are rossing straight lines. However, the
angle formed by L1 and L2 and the angle formed by the observed straight lines
dier, in general.
Let L1 and L2 be dire ted by e1 and e2 ; respe tively. If o denotes the angle
formed by L1 and L2 then
e
e
os o = 1 2 :
je1 j je2 j

For the angle  observed by U we have

(A
e )
(Auuo
e2 )
os  = uuo 1
=
o

where

os o

1 2

jAuu
e1 j jAuu
e2 j p1 21 p1 22 ;
e
u
e1
= vuu
1 ;
1 :=
(uo
u)je1 j
je1 j
e
u
e2
= vuu
2 :
2 :=
(uo
u)je2 j
je2 j
o

Thus  and o are equal if and only if 1 = 2 = 0; i.e. if and only if both e1
and e2 are orthogonal to the relative velo ity vuuo :

5.3. The Lorentz ontra tion

5.3.1. A straight line segment orthogonal to the relative velo ity vuu

o in the
observed by U as a straight line segment parallel to the original one
and having the same length.
A straight line segment parallel to the relative velo ity vuuo in the Uo -spa e
is observed by U as a shorter straight line segment parallel to the original one.
This is the famous Lorentz ontra tion whi h will be detailed as follows.
A straight line segment in the Uo -spa e an be represented by one of its
extremities and the ve tor between its extremities. Sin e parallel segments are
observed in a similar way, we an onsider only the ve tor eo 2 Euo between the
extremities.
The observation of eo by U yields e := Auuo
eo : A simple al ulation shows
that
2
jej2 = jeo j2 (u
eo ) 2 = jeo j2 (vuuo
eo )2 :
(u
uo )
The observed length, in general, is smaller than the proper one.
More losely, the observed length equals the original one if and only if the
segment is orthogonal to the relative velo ity; otherwise the observed length is
smaller than the original one. The observed length is the smallest if the segment
is parallel to the relative velo ity:

Uo -spa e is

jej = jeo j

jvuu j2
o

if eo is parallel to vuuo :

5.3.2. One often says that the travelling length is smaller than the proper
(or rest) length: \a moving rod is ontra ted, be omes shorter".
We emphasize that the Lorentz ontra tion formula does not state any real
physi al ontra tion at all. We an assert only that an observer moving relative
to a rod observes the rod shorter than the observer having the rod in its own
spa e.
Let us imagine two rods having the same proper length and resting in the
spa es of dierent observers: both observers will observe the other rod to be
shorter than its own one.
A number of paradoxes an arise from this situation: \I say that your rod
is shorter then mine, you say that my rod is shorter than yours; whi h of us is
right?" Keeping in mind that only illusory and no physi al ontra tions are in
question, we an a ept the orre t answer: both of us are right.

5.3.3. Suppose you do not believe that the ontra tion is illusory and you
want determine experimentally whi h of us is right. The experiment seems
extremely simple: you at h my rod (whi h is moving relative to you) and having
stopped it you put it lose to your one and then you will see whi h of them is
shorter.
We onsider an ideal ase: you seize the moving rod all at on e so that it
stops instantaneously.
Let us translate the situation into our mathemati al language. My rod is
des ribed by a line segment in the Uo -spa e:
Lo = qo + [0; 1eo
and eo is taken to be parallel to vuuo :
Then the rod has the length (the proper length)
do := jeo j with respe t to
q
Uo and the length (the travelling length) d := 1 jvuuo j2 do observed by U :
At a U -instant t the history of ea h point of the rod will be hanged into an
inertial history with the velo ity value u; then you get
L = fq ? t + u
Ij q 2 Lo g = q + [0; 1e;

where

q := qo ? t + u
I;
e := A(u; uo )
eo :
The segment (the seized rod) L is in the U -spa e and has the length jej = d;
the length of Lo observed by U :

5.3.4. You an relax: you showed that my rod is \really" shorter than yours.
But then you think that I an exe ute a similar experiment to show that your
rod is \really" shorter than mine. Again the same disturbing situation.
To solve the seeming ontradi tion, note that in your experiment your rod
ontinues to exist without any ee t on it, while my rod is ae ted by your

seizure, and in my experiment your rod is ae ted. The seizure means a physi al
hange in the rod whi h auses ontra tion.
Let us analyze the problem more thoroughly.
(i ) The rod resting in the Uo -spa e moves relative to the observer U whi h
nds that the length of the rod is d: At a U -instant t the observer U seizes the
rod, stops it, and dis overs that this rod has the same length d: A ording to U ;
the rod did not hange length in the seizure, in other words, the observer U sees
the rod as rigid and this is well understandable from its point of view be ause
the rod is stopped at an instant with respe t to U ; i.e. every point of the rod
stops simultaneously with respe t to U :
(ii ) The rod rests in the Uo -spa e. As U seizes the rod, the observer Uo
sees that the rod begins to move but not instantaneously with respe t to Uo :
the points of the rod begin the movement at dierent Uo -instants! First the
ba kward extremity (from the point of view of the relative velo ity of U with
respe t to Uo ) starts and then su essively the other points, at last the forward
extremity. Evidently, Uo sees the rod is not rigid, it ontra ts during the time
interval of seizure.
(iii ) The rod experien es that it moves relative to U whi h begins to stop it in
su h a way that rst the forward extremity (from the point of view of the relative
velo ity of the rod (i.e. of Uo ) with respe t to U ) stops and then su essively the
other points and at last the ba kward extremity. The rod experien es ontra tion
during the pro edure of seizure.
We have examined three standpoints. Two of them on ern inertial observers
and the third on erns a non-inertial obje t.

5.3.5. The ideal ase that every point of the rod hanges its velo ity abruptly
at a U -instant an be repla ed by the more realisti one that every point of the
rod hanges its velo ity from uo to u during a U -time interval, as the following
Figure shows.

5.3.6. Note that we started with the problem that \you seize the moving rod
all at on e so that it stops instantaneously", we onsidered \instantaneously"
with respe t to U without alling attention to the extremely important fa t that
\instantaneously" has no unique meaning.
The reader is asked to analyze the problem that the rod is aught by U
instantaneously with respe t to Uo (i.e. every point of the rod is stopped by U
simultaneously with respe t to Uo ):
5.4. The tunnel paradox
5.4.1. Consider a train and a tunnel. The proper length of the train is greater
than the proper length of the tunnel. The travelling train enters the tunnel.
The observer resting with respe t to the tunnel observes Lorentz ontra tion
on the train, thus it sees that, if the velo ity of the train is high enough, the
train is entirely in the tunnel during a time interval.
On the ontrary, the observer resting with respe t to the train observes
Lorentz ontra tion on the tunnel, thus it experien es that the train is never
entirely in the tunnel.
Whi h of them is right? We know that both. However, it seems to be a very
strange situation, be ause the observer resting with respe t to the tunnel says
that \when the train is entirely inside I lose both gates of the tunnel, thus I
onne the train in the tunnel, I am right and the observer in the train is wrong".
5.4.2. On the basis of our previous examination we an remove the paradox
easily.
In the assertion above \when" means that the gates be ome losed simultaneously with respe t to the tunnel.
From the point of view of the train the gates will not be losed simultaneously:
rst the forward gate loses and later the ba kward gate. When the forward gate
loses, the forepart of the train is in the tunnel (the ba k part is still outside),
when the ba kward gate loses, the ba k part of the train is in the tunnel (the
forepart is already outside).
\I onned the train in the tunnel" means that the losed gates hinder the
train from leaving the tunnel. But how do they do this? The train is moving;
it must be stopped to be onned denitely in the tunnel: some apparatus in
the tunnel brakes the train or the train hits against the front gate whi h is so
strongly losed that stops the train. In any ase, as we have seen, stopping
means a real ontra tion of the train, onsequently, it nds room in the tunnel;
however, then the train eases to be inertial in all its existen e, that is why the
assertion \I am never in the tunnel entirely" (true for an inertial train) will be
false.

5.5. No measuring rods

5.5.1. In the spe ial relativisti spa etime model the absolute rigid rod is not
a meaningful notion. We have seen in 5.3.4 that the same rod seems to be rigid
to an inertial observer and not rigid to another observer.
If C1 and C2 are world lines and U is a global inertial observer with onstant
velo ity value, then the ve tor and the distan e observed by U between C1 and
C2 at the U -instant t are
jC2 ? t C1 ? tj;
respe tively, provided neither of C2 \ t and C1 \ t is void.
C2 ? t C1 ? t

and

Absolute ve tors and absolute distan es between the world lines do not exist.
Evidently, in general, dierent observers observe dierent ve tors and distan es
between the world lines. By no means an we dene an absolutely rigid rod.

5.5.2. As a onsequen e, measuring rods are useless for determining the

distan e between two points, the length of a line et . in observer spa es: it is
questionable whether one an take a rod, arry it to the gure to be measured,
put it onse utively at onvenient pla es in su h a way that its length does not
hange during these pro edures.
Spa etime measurements in the non-relativisti ase are based on lo ks (saying the absolute time) and measuring rods (that are absolutely rigid).
Spa etime measurements in the spe ial relativisti ase are based on lo ks
(saying their proper times) and light signals.
Re all, for instan e, that dening simultaneity we need the proper times
passing in observer spa e points and the distan e between observer spa e points;
this latter is measured by light signals and proper time intervals (radar).
5.6. The time dilation

5.6.1. Let U be a global inertial observer with onstant velo ity

value u:
U -time is an ane spa e over I: Sin e I is oriented, later and earlier makes
sense between U -instants: t is earlier than s (s is later than t) if s t is positive.
A unique U -instant U (x) is assigned to every world point x: Consequently,
we an de ide whi h of two arbitrary world points is later a ording to U :
Denition. The time observed by U between the world points x and y is
tU (x; y) := U (y)

U (x) =

u
(y

x):

The world point y is later than the world point x (x is earlier than y) a ording
to U if the time observed between x and y is positive.

Neither of x and y is later a ording to U if and only if they are simultaneous

a ording to U :

5.6.2. Fix two dierent world points x and y:

If they are spa elike separated then there are global inertial observers Uo ; U1
and U2 su h that y is simultaneous with x a ording to Uo ; y is later than x
a ording to U1 and y is earlier than x a ording to U2 :
If y is future-like with respe t to x (they are lightlike or timelike separated)
then y is later than x a ording to all inertial observers.
5.6.3. Suppose y 2 x +T!: Then t(x; y) = jy xj is the inertial time between

x and y;

uo

:=

y x

jy xj 2 V(1);

and y = x + uo t(x; y):

Consequently, if U is the global inertial obvserver with the onstant velo ity
value u then
tU (x; y) = (u
uo)t(x; y) = q t(x; y) 2 :
1 jvuuo j

tU (x; y) = t(x; y) if and only if u = uo i.e. if and only if x and y are in ident
with the same U -spa e point. In any other ases tU (x; y) is greater than t(x; y):

5.6.4. Let us illustrate our result as follows.

Let us onsider an inertial (point-like) lo k. Let x and y be the o urren es
that the lo k says 11 and 12, respe tively. The observer, relative to whi h
the lo k is at rest, observes 1 hour between the o urren es. Another observer,
moving relative to the lo k, observes more than 1 hour between the o urren es.
This is the famous time dilation.
The lo k and the observer move with respe t to ea h other. In usual formulations one onsiders that the observer is at rest and the lo k is moving and one
says that \a moving lo k works more slowly than a lo k at rest".
We emphasize that the time dilation formula does not state any real physi al
dilation of time at all. We an assert only that an observer moving relative to a
lo k observes the lo k working more slowly than the observer having the lo k
in its own spa e.
In other words, an observer observes that the time of another observer passes
more slowly.
Let us take two dierent inertial observers. Then both will observe that the
other's time passes more slowly.

5.7. The twin paradox

5.7.1. Let us onsider two twins, Peter and Paul. Both are laun hed in
separate missiles. Peter says that Paul is moving relative to him, hen e Paul's
time passes more slowly and he observes that when he (Peter) is forty then his
brother (Paul) is only twenty. On the other hand, Paul says that Peter is moving
relative to him and he observes that when he (Paul) is forty then his brother
(Peter) is only twenty. Whi h of them is right?
Keeping in mind that only illusory and no physi al time dilations are in
question, we an be onvin ed that both are right. How an it be possible?
The paradox is based on our everyday on ept of absolute time, i.e. that
\when" has an absolute meaning. However, the rst \when" means simultaneity
with respe t to Peter and the se ond \when" means simultaneity with respe t
to Paul; we know well that these simultaneities are dierent.
5.7.2. Suppose the brothers do not believe that time dilation is illusory and
they want an experimental test: let them meet and then a simple inspe tion will
determine whi h of them is older.
However, the time dilation formula on erns inertial observers. It is ex usable
that both missiles are onsidered to be inertial. But if they remain inertial then
the brothers never meet. If the brothers meet then at least one of them eases
to be inertial. Anyhow, the mutually equivalent situation of the brothers breaks.
It will not be true that both are right saying \when I am forty then my brother
is only twenty".

Let the brothers meet. Both existed somehow between the two o urren es,
the departure and the arrival, and their proper times passed during their existen e. The times elapsed depend on their existen e and need not be equal. It

 an happen that Peter is older than Paul; e.g. if Peter remains inertial (the inertial time between two world points is always greater than a time passed on a
non-inertial world line, see 2.2.3). This dieren e of proper times is an absolute
fa t and has nothing to do with the illusory time dilation.
It is important to distinguish between illusory time dilation on erning two
inertial observers and really dierent times passed along two world lines between
two world points.

5.8. Experiments on erning time

5.8.1. We have an experimental proof for time dilation.

Cosmi rays produ e muons in the ionosphere. Some of those muons ome
to the earth. Dete ting the magnitude v of their velo ity with respe t to the
earth and knowing the height d of the pla e where they are produ ed, we an
al ulate the time of their travel (uniform and re tilinear motion seems a good
approximation). It turns out that the time of travel d=v ex eeds the lifetime T of
muons (muons are not stable parti les, they de ay having a well-dened average
lifetime). Thus the earth observes as if the muons lived more than their lifetime.
The muon in question exists inertially thus it \feels" the inertial time to of
its travel whi h is less than its lifetime; the earth observes a longer time (time
dilation):
d
t
= p o 2 > T > to :
v
1 v
It is interesting that
p we an give another explanation, too. The muon observes the distan e d 1 v2 between pits birth pla e and the earth (Lorentz
ontra tion), hen e it travels for to = d 1v v2 :

5.8.2. Let us suppose that, simultaneously with the muon in the ionosphere
(muon I), a muon is produ ed and remains resting on the earth (muon E).
A ording to what has been said, muon E de ays before muon I arrives at the
earth: muon E \sees" that time passes more slowly for muon I.
Of ourse, muon I \sees" as well that time passes more slowly for muon E.
Then one ould suspe t a ontradi tion (paradox): a ording to muon I, muon E
would be alive at the end of the travel of muon I.
There is no ontradi tion: simultaneously in the present ontext means simultaneously a ording to the earth, i.e. to muon E, and muon I is produ ed
simultaneously a ording to muon E. Then, a ording to muon I, the other muon
is born earlier; onsequently, muon I, though sees time passing more slowly for
muon E, will observe that the life of muon E ends before muon I meets the earth.

5.8.3. Experiments show that instable parti les revolving in an a elerator

have longer lifetime. This supports that dierent times an pass between two
world points along dierent world lines, as it will be explained.
Suppose two muons are produ ed at the same time and at the same pla e
(i.e. a single world point orresponds to their birth) and one of them (muon R)
remains resting beside the a elerator, the other (muon A) is onstrained to
revolve in the a elerator. The muons meet several times. Then muon R de ays,
but muon A ontinues to revolve and meets again the void pla e of muon R; we
observe (resting with muon R) as if muon A had a longer life time. Nevertheless,
both muons have the same proper lifetime T:
The world line of muon R is inertial while muon A has a noninertial world line;
the two world lines interse t ea h other several times. Dierent times tR and tA
pass along the dierent world lines of the muons between their two su essive
meetings. Inertial time is always greater than a non-inertial: tA < tR : That is
why there an be a natural number n su h that ntA < T < ntR ; i.e. muon R
does not last until the n-th meeting but muon A survives it.

5.9. Exer ises

1. Prove that the addition formula 4.3.2 of relative velo ities remains valid
for u; u0 2 V(1); u00 2 V(0):
2. Take the motion treated in 4.6.2. Demonstrate that
bo
tlim
!1 jr_U (t)j = 1:
tlim
!1 r_U (t) = vuuo +  ;
3. Consider the uniformly a elerated world line treated in 4.6.2. Try to des ribe the orresponding motion relative to an inertial observer with the onstant
velo ity value u whi h is not g-orthogonal to ao:
4. Let x and y be dierent world points simultaneous with respe t to an
observer (x : a plane lands in London at 12.00; y : a train leaves Paris at 12.00).
Then there is an observer whi h observes that x is later than y and there is an
observer that observes that x is earlier then y:
5. We have a lo k that an measure a proper time period of 10 8s: At whi h
relative velo ity magnitude an we measure a time dilation in a minute? (Keep
in mind that 1  (2;9979 : : : )108m=s:)
6. Let u 2 V(1); v 2 EIu ; jvj < 1: Let to 2 I+ : Consider the world line
fun tion r that passes through the world point xo and

(i )
(ii )

t =t o )
r_(t) = p1u+jvvjsin(
2
sin2 (t=to )
q

r_(t) = u 1 + jvj2 sin2 (t=to ) + v sin(t=to )

(t 2 I);
(t 2 I):

Prove that the inertial world line xo + u

I and the world line Ran r interse t
ea h other in xo + 2nto for all integers n:
Evidently, to is the time passed along Ran r between two onse utive interse tions. Estimate the time passed between two onse utive interse tions along
the inertial world line.

6. Some spe ial non-inertial observers

6.1. General observers

In most of the textbooks one says that spe ial relativity on erns only inertial
observers, non-inertial observers require the authority of general relativity. We
emphasize that this is not true.
The dieren e between spe ial relativity and general relativity does not lie in
observers whi h is evident from our point of view: spa etime models are dened
without the notion of observers; on the ontrary, observers are dened by means
of spa etime models.

Non-inertial observers are right obje ts in the spe ial relativisti spa etime
model. Inertial observers and non-inertial observers dier only in the level of
mathemati al tools they require. Inertial observers remain in the ni e and simple framework of ane spa es while the deep treatment of non-inertial observers
needs the same mathemati al tools as the treatment of general relativisti spa etime models: the theory of pseudo-Riemannian manifolds.
Fortunately, to des ribe some spe ial and important aspe ts of non-inertial
spe ial relativisti observers, we an avoid the theory of manifolds; nevertheless,
we shall meet some ompli ations.

6.2. Simultaneities

6.2.1. The general notion of observers was given in 3.1.1.

The spa e of any observer has a simple and natural meaning but time an
be dened in a satisfa tory way only for inertial observers. (In fa t we have
onsidered global inertial observers for the sake of simpli ity; globality permits
avoiding some ompli ations onne ted with domains of fun tions.)
Why the syn hronization pro edure, i.e. the method of establishing simultaneity by light signals is not ompletely satisfa tory for a non-inertial observer?
Let us take two spa e points q and q0 of a non-inertial observer U : A light
signal starting at the world point x in ident with q meets q0 at y; the re e ted
light signal meets q at x+ : Then the world point x in ident with q would be
onsidered simultaneous with y if the proper time passed between x and x
equals the proper time passed between x and x+ :

Unfortunately, the simultaneity dened by light signals starting from q0 does

not ne essarily oin ide with the simultaneity dened by light signals starting
from q (see Exer ise 6.9.12).
Let us a ept that simultaneity dened by light signals works well \innitesimally". This means that the U -line passing through the world point x; in a
neighbourhood of x; an be approximated by a straight line dire ted by U (x):

Thus we an say that in a neighbourhood of x the world points approximately

simultaneous with x a ording to U are the elements of x + EU (x): The smaller
the neighbourhood is the better the approximation we get. A lear reasoning
leads us then to the idea that world points simultaneous with ea h other a ording to U would onstitute a hypersurfa e whose tangent spa e at every x equals
EU (x): Su h a denition of simultaneity does not depend on the U -spa e point
(U -line) from whi h light signals start. However, it may happen that there is no
su h hypersurfa e at all (see 6.7.6)! And even if su h hypersurfa es exist, it may
happen that the proper times passed between two su h hypersurfa es along different U -lines are dierent (see 6.6.5), thus the simultaneity is not satisfa tory
in all respe ts.

6.2.2. In general, there is no natural simultaneity with respe t to a noninertial observer; onsequently, there is no natural time of su h an observer. Of
ourse, a non-inertial observer an hoose some sort of arti ial simultaneity
(e.g. hooses one of its spa e points and makes the syn hronization pro edure
by light signals relative to this spa e point; on the earth one makes su h a
syn hronization relative to Greenwi h).
Now we shall study what a simultaneity should mean.
First we deal with world surfa es whi h are ne essary for simultaneities.
Denition. A world surfa e is a onne ted three-dimensional smooth submanifold in M whose every tangent spa e is a spa elike linear subspa e of M:
If F is a world surfa e, then for x 2 F there is a unique u(x) 2 V(1) su h that
Tx (F) = Eu(x):
We an prove, similarly to the orresponding assertion for world lines, that if
F is a world surfa e and x 2 F; then F n fxg  S:
Consequently, if F is a world surfa e and C is a world line then C \ F is either
void or ontains a single element whi h we shall denote by C ? F:

6.2.3. Obviously, a simultaneity must be a relation between world points,

learly having the following properties:
| every world point is simultaneous with itself;
| if x is simultaneous with y; then y is simultaneous with x;
| if x is simultaneous with y and y is simultaneous with z; then x is simultaneous with z:
In other words, a simultaneity is to be an equivalen e relation.
Evidently, we require some other onditions, too. For instan e, timelike
separated world points annot be simultaneous, i.e. simultaneous world points
must be spa elike separated.
Moreover, we expe t that simultaneity is ontinuous or dierentiable in some
sense.

Denition. A simultaneity S on a onne ted open subset G of M is an

equivalen e relation on G su h that
(i ) the equivalen e lasses are world surfa es;
(ii ) S is smooth in the following sense: to every x 2 G there is a unique
US (x) 2 V(1) su h that the tangent spa e of the equivalen e lass (world
surfa e) at x equals EUS (x); then US : M V(1)  MI is required to be
smooth.
The time orresponding to the simultaneity S is the set of the equivalen e
lasses of S :

The time orresponding to S is often alled S -time and is denoted by IS ; its

elements are the S -instants and S (x) will stand for the S -instant (world surfa e)
ontaining the world point x:
Evidently, simultaneities exist: e.g. for all u 2 V(1); the simultaneity dened
by the orresponding global inertial observer: x and y are simultaneous if and
only if u
(x y) = 0:

6.2.4. If S is a simultaneity then US is an observer; in other words, a

unique observer orresponds to every simultaneity. On the other hand, there
are observers to whi h no simultaneity orresponds in a natural way.

Denition. An observer U is alled regular if there is a (ne essarily unique)

simultaneity S su h that US = U :
We mention again that there are non-regular observers (see 6.7.6).
The simultaneity due to a regular observer U is alled U -simultaneity and
the orresponding time, denoted by IU ; is alled U -time. The elements of IU
are world surfa es; they are alled U -surfa es (regarded as subsets of M) or U instants (regarded as elements of IU ): The U -surfa e ontaining the world point
x is denoted by U (x):
Let U be an arbitrary observer. A simultaneity on the domain of U whi h
does not equal U -simultaneity is alled arti ial with respe t to U and then we
say arti ial time, arti ial instants.
6.2.5. Denition. Let S be a simultaneity, t; s 2 IS : We say that s is later
than t (t is earlier than s) if there are x 2 t and y 2 s su h that y is later than
x:
let
y0
y0
y

It annot o ur that both of t and s are earlier than the other. Indeed,
s; t and x; y be as in the denition. Then for all y0 2 s; x0 2 t we have
x0 = (y0 y) + (y x) + (x x0 ): Be ause of the properties of world surfa es,
y and x x0 are spa elike ve tors. Thus, in view of Exer ise V.4.22.2, if
x 2 T! (t is earlier than s) then y0 x0 62 T (s is not earlier than t):

We easily nd that \later" is an ordering (a re exive, antisymmetri and

transitive relation) on IS : However, it need not be total: there an be t and s in
IS su h that neither of them is later than the other:

We say that the simultaneity S is well posed if the relation \later" on IS is a

total ordering.
It an be shown that every world point xo in the domain of S has a neighbourhood su h that the restri tion of S onto this neighbourhood is well posed.
If Co is the US -line passing through xo ; then fx 2 Mj Co \ S (x) 6= ;g is su h a
neighbourhood.

6.2.6. An observer U ; together with a simultaneity S on Dom U split a part

of spa etime (the domain of U ) into S -time and U -spa e:


HU ;S : M IS  EU ;
x 7! S (x); CU (x) :

6.3. Distan es in observer spa es

6.3.1. How distan es are measured in an observer spa e? Let U be an

observer and suppose a simultaneity S is given on the domain of U : We should
like to determine the distan e between two U -spa e points q and q0 at an S instant t:
First we make the following heuristi onsiderations. Let us put x := q ? t and
suppose q0 is \near" to q: A ording to the \innitesimal" simultaneity whi h is
reasonable from the point of view of the observer, y0 := q0 ? (x + EU (x)) is the
world point on q0 that is approximately simultaneous with x in a natural way.
Then d := jy0 xj is the approximate value of the distan e to be determined.
The world point x0 :=
q0 ? t is simultaneous with x a ording
to S : Sin e


0
U
(
x
)


(
x
x
)
0
0
0
0
0
y  x + U (x ) U (x)
U (x0)  x + U (x) U (x)
(x x) ; we see that d 
jU (x)
(x0 x)j:

We have got a formula for \innitesimal" distan es from whi h we an dene

the length of a urve in a natural way by an integration. The distan e between
two observer spa e points will be dened to be the least length of urves onne ting the spa e points.
Before going further, the reader is asked to study Se tion VI.7.

6.3.2. Denition. Let U be an observer. A subset L of the observer

spa e EU is alled a urve if there is a simultaneity S on Dom U su h that
Lt := fq ? tj q 2 L; q \ t 6= ;g is either void or a urve in M for all t 2 IS :
Note that in fa t Lt is ontained in the hypersurfa e t:
We say that the urve L onne ts the U -spa e points q1 and q2 if Lt onne ts
q1 ? t and q2 ? t for all t 2 IS ; provided that q1 \ t and q2 \ t are not void.
6.3.3. Denition. Let L be a urve in EU : Then
`t (L) := `U (Lt ) :=

Lt

jU (
)dLt j

is alled the length of L at the S -time point t:

The distan e between the U -spa e points q and q0 at t is
dt (q; q 0 )

:=
:= inf f`t(L)j L is a urve onne ting q and q0 g:

It is worth des ribing expli itly that if pt is a parametrization of Lt then

`t (L) =
=

Dom pt
Z
Dom pt

jU (p(a))
p_t (a)jda =

q

jp_t (a)j2 +

2

U (p(a))
p_ t (a)

da:

Note the spe ial ase when U is regular and S is the U -simultaneity; then
if x 2 t and x is a tangent ve tor of t at x: Consequently,

U (x)
x = x

`t (L) =

Lt

jdLt j =

Dom pt

jp_t (a)jda

for a regular observer.

In parti ular, if U is inertial and U -time is used then, for all t; dt (q; q0 ) equals
the distan e jq0 qj dened earlier.
Keep in mind the following important remark: suppose the S -instant t is a
hyperplane; then there is a unique uo 2 V(1) su h that t is dire ted by Euo : The
distan e between the U -spa e points at t does not equal, in general, the distan e
observed by the inertial observer with the velo ity value uo: Re all, e.g. the ase
that U is an inertial observer with the velo ity value u (see Se tion 5.3).

6.3.4. Denition. The observer U is alled rigid if there is a simultaneity

S on Dom U in su h a way that if

| L is an arbitrary urve in EU ;
| t; t0 2 IS and q \ t 6= ;; q \ t0 6= ; for all q 2 L;
then `t (L) = `t0 (L):
Note that rigidity of observers is a highly ompli ated notion in the spe ial
relativisti spa etime model, in ontradistin tion to the non-relativisti ase.

6.3.5. The following assertions an be proved by means of the tools of smooth

manifolds.
(i ) Our denition of a urve in EU involves a simultaneity; nevertheless, it
does not depend on simultaneity: if there is a simultaneity with the required
onditions then these onditions are satised for all other simultaneities as well.
(ii ) The distan e between two U -spa e points at an S -instant is dened by
an inmum; this inmum is in fa t a minimum, i.e. for ea h S -instant t there is
a urve onne ting the points whose length at t equals the distan e between the
U -spa e points at t:
(iii ) Our denition of rigidity involves a simultaneity; nevertheless, it does
not depend on simultaneity: if there is a simultaneity with respe t to whi h the
observer is rigid, then the observer is rigid with respe t to all other simultaneities
as well.
6.4. A method of nding the observer spa e

6.4.1. To nd the spa e of an observer, i.e. the U -lines, we have to nd the

solutions of the dierential equation
(x :

I  M)?

x_ = U (x):

A frequently appli able method is to transform the dierential equation by

huo ;o

: M ! I  Euo ;

x 7!

uo
(x

o); uo
(x o)

a ording to VI.6.3, where uo is a suitably hosen element of V(1):

The transformed dierential equation will have the form




(t; q)
= uo
U (o + uo t + q); uo
U (o + uo t + q) ;
(t; q) : I I  Euo ?
i.e.
(t : I I)?
t_ = uo
U (o + uo t + q );


(q : I  Eu )?
o

q_

= uo
U (o + uo t + q):

Let s 7! t(s) and s 7! q(s) denote the solutions of these dierential equations
with the initial onditions t(0) = 0; q(0) = qo ; where qo is an arbitrary element
in Euo su h that o + qo is in the domain of U : Then
s 7! o + uo t(s) + q (s)

is the world line fun tion giving the U -line passing through o + qo :
It is worth using more pre ise notations: let x be an element of (o + Euo ) \
(Dom U ); then s 7! tx (s) and s 7! qx (s) will denote the solutions of the
dierential equations with the initial onditions tx (0) = 0; qx (0) = x o: Then

I  M;

s 7! rx (s) = o + uo tx (s) + qx (s)

is the world line fun tion giving the U -spa e point that x is in ident with.

6.4.2.

Consider the global inertial observer Uo with onstant velo ity value
uo and Uo -time as an arti ial time for the observer U : Then, a ording to 4.1.1,
s 7! tx (s) gives Uo -time as a fun tion of the proper time of the U -line passing
through x; in other words, tx (s) is the Uo -time passed between to := o + uo
and tx (s) := rx (s) + uo : tx (s) = tx(s) to :

This fun tion is stri tly monotone in reasing; its inverse, denoted by I I;
and to + t passed in

t 7! sx (t); gives the proper time between the Uo -instants to

the U -spa e point that x is in ident with.

6.5. Uniformly a elerated observer I

6.5.1. In the spe ial relativisti spa etime model the denition of a uniformly
a elerated observer is not so straightforward as in the non-relativisti ase. We
know that here the a eleration of a uniformly a elerated world line fun tion is
not onstant, thus a uniformly a elerated observer will not be an observer with

 onstant a eleration eld. Anyhow, we wish to nd an observer whose lines are

uniformly a elerated.
Omitting the thorny way of sear hing, let us take an observer satisfying the
requirements and study its properties.
Let o 2 M; uo 2 V(1) and 0 6= ao 2 EI
uIo and dene the global observer
U (x) := uo

1 + jaoj2 uo
(x o) 2

ao uo
(x

o)

(x 2 M):

Note that uo = U (o):

The observer has the a eleration eld
AU (x) = ao

1 + jao j2 uo
(x o) 2

uo jao j2 uo
(x

o)

(x 2 M);

thus ao = AU (o):
It is trivial that
U (x + q ) = U (x);

AU (x + q ) = AU (x)

(x 2 M; q 2 Euo );

i.e. U and AU are onstant on the hyperplanes dire ted by Euo :

As a onsequen e, the translation of a U -line by a ve tor in Euo is a U -line,
too.

6.5.2. Transforming the dierential equation of the observer a ording to 6.4,

we get

1 + jao j2 t2 ;
q_ = ao t:
t_ =

The rst equation, with the initial value t(0) = 0; has the solution
t(s) =

shjao js

(s 2 I):

jaoj

Then the se ond equation be omes very simple and we nd its solutions in
the form
hjao js 1
q (s) = ao
+ qo
(s 2 I):
2

jao j

Hen e we obtain that the U -line passing through x 2 o + Euo is given by the
world line fun tion

rx (s) = x + uo

shjao js

jao j

+ ao

hjao js 1

jao j2

(s 2 I):

It is not hard to see that every U -line meets the hyperplane o + Euo ; hen e
every U -line an be given by su h a world line fun tion: all U -lines are uniformly
a elerated.

6.5.3. Be ause U and AU are onstant on the hyperplanes dire ted by Euo ;
all U -lines have the same velo ity and the same a eleration on the hyperplanes
dire ted by Euo :
Using the notations of 6.4, we see that
arshjao jt
sx (t) =
=: s(t)
(t 2 I)
jao j

for all x in o + Euo : Thus, given two Uo -instants, the same time passes along all
U -lines between them.
Be ause of these properties of U -lines it seems suitable to asso iate with U
the Uo -simultaneity and the Uo -time as an arti ial time.

6.5.4. Let us examine whether this observer is regular.

Evidently, o + Eu is a world surfa e g-orthogonal to U :
o

Introdu e the notation

h(x) :=
q

2

o) + ar th 1 + jao j2 uo
(x o) 2
:= ao
(x o)
for x 2 M; x 62 o + Euo and put for  2 R

1 + jaoj2 uo
(x

t :=

8
>
<

fx 2 Mj h(x) = ln ;

uo
(x

o + Euo
>
:
fx 2 Mj h(x) = ln( );

o) > 0g

uo
(x

o) < 0g

if  > 0
if  = 0
if  < 0:

Evidently, h is a dierentiable fun tion outside o + Euo and

q

1 + jao j2 uo
(x o) 2
U (x)
Dh(x) = ao +
=
:
uo
(x o)
uo
(x o)
As a onsequen e, for all  2 R; t is a three-dimensional submanifold whose
tangent spa e at x equals Ker Dh(x) = EU (x): This means that t -s are U surfa es, U is regular and
IU = ft j  2 Rg:
uo

6.5.5. If q1 and q2 are U -lines, then q2 ?t q1 ?t is the same for all Uo -instants
t: In other words, the ve tor and the distan e observed by the inertial observer
Uo between two U -spa e points is the same for all Uo -instants. We an say that
Uo observes U to be rigid and rotation-free. Is U rigid and rotation-free?
We have not dened when an observer is rotation-free, thus we an answer
only the question regarding rigidity as dened in 6.3.4.
This observer U is not rigid. Let us take a Uo -instant t: For all x 2 t we have
uo
(x o) = t to (where to := o + Euo ); thus
q

(x 2 t; t 2 IUo )
to ) + uo 1 + jao j2 (t to )2 =: ut
(U is onstant on the Uo -instants).
The formula in 6.5.3 gives us the proper time s(t) passed in every U -spa e
point between the Uo -instants (arti ial time points) to and t := to + t :
shjaojs(t)
t=
:
U (x) = ao (t

jao j

Let Lo be a urve in o + Euo ; then the set of U -spa e points that meet Lo ;
L := fq 2 EU j q \ Lo 6= ;g is a urve in the observer spa e. Indeed, if po is a
parametrization of Lo ; then

pt := po + uo t +

q

ao

1 + jaoj2 t2

jaoj2

is a parametrization of Lt (see 6.3.2) for t = to + t 2 IUo :

Then
p_t = p_o
and


U pt (a)
p_ t (a) = ao
p_ o (a)t
onsequently,

jp_t j2 +

(a 2 Dom po );

(U pt )
p_t 2 = jp_o j2 + ao
p_o 2 t2 ;

whi h shows that the length of urves depends on the arti ial time points:
the observer is not rigid.

6.5.6.

The length of urves in U -spa e, onsequently the distan e between

points, in general, de reases prior to to and in reases after to ; as Uo time passes. This is well understandable from a heuristi point of view. Though
we dened Lorentz ontra tion between two inertial observers, we an say e.g.
that after to the spa e points of U move faster and faster with respe t to Uo ;
thus their distan es seem more and more ontra ted with respe t to Uo ; that is,
their distan es must in rease ontinually in order that the distan es observed by
Uo be onstant.
U -spa e

6.6. Uniformly a elerated observer II

6.6.1. Let o 2 M; uo 2 V(1) and 0 6= ao 2 EI
uI ; put
o

B (x) := ao
(x

o) uo uo
(x o) ao
for x 2 M and dene the non-global observer by
Dom U := fx 2 Mj B (x) is future-dire ted timelikeg
B (x)
U (x) :=
(x 2 Dom U ):
jB (x)j
Note that B (x) is future-dire ted timelike if and only if

0 > B(x)
2 =
0 > uo
B(x) =

o) 2 +
ao
(x o):

ao
(x

uo
(x

o) 2 jaoj2 ;

Then we nd that a world point x for whi h x o lies in the plane generated by
if and only if x o is spa elike and ao
(x o) > 0:

uo and ao is in the domain of U

6.6.2. If q is a world ve tor g-orthogonal to both uo and ao then

Dom U + q = Dom U

and

(x 2 Dom U ; q 2 Euo ; ao
q = 0):

U (x + q ) = U (x)

The observer has the a eleration eld

AU (x) =

ao
(x

o) ao

jao j2
jB (x)j2

uo
(x

o) uo

(x 2 Dom U ):

Then we easily nd that

if and only if x o is parallel to ao ;
for all x 2 Dom U ;
if and only if x o is parallel to ao :

U (x) = uo
AU (x) 6= ao
AU (x) = ao
(axo o)

6.6.3. Let us introdu e the notation

no

:=

ao
jao j :

If  2 R then
u

u + (th)no
= uo h + no sh
:= qo
1 (th)2

is in V(1) and we easily nd that

(x 2 Dom U ; x o 2 Eu ):

U (x) = u

Thus t := o + Eu \ (Dom U ) is a U -surfa e. To every x 2 Dom U there

is su h a U -surfa e ontaining x; given by

x := arth

uo
(x
no
(x

o)
:
o)

This means that U is regular, and

IU := ft j  2 Rg:

6.6.4. To nd the U -lines we use the method outlined in 6.4.

Transforming the dierential equation x_ = U (x) by huo ;o we get

no
q

t_ = q

(no
q)2

q_

no t

(no
q)2

t2
t2

()

()

Equation () implies no
q_ = p(n
tq)2 t2 whi h, together with equation ();
o
results in
(no
q)(no
q_ ) = tt_
implying

1
= onst =: 2 :

Then dierentiating equation () we obtain
(no
q)2

t2

t = 2 t

from whi h | taking the initial values t(0) = 0; t_(0) = 1 | we infer

t(s) =

shs
:


As a onsequen e, equation () takes an extremely simple form, and we nd

its solutions easily:
hs 1
q (s) = no
+ qo :



Hen e we obtain that the U -line passing through x 2 o + Euo \ (Dom U )
is given by the world line fun tion

rx (s) = x + uo
where

shjax js

ao
ao
(x

jaxj

+ ax

hjax js 1

jaxj2

(s 2 I)

o
=
:
o) no
(x o)
It is not hard to see that every U -line meets the hyperplane o + Euo ; hen e
every U -line an be given by su h a world line fun tion; all U -lines are uniformly
a elerated.
ax

:=

6.6.5. The present observer U serves as an example to show that the observer
is regular, but dierent times pass in dierent U -spa e points (along dierent
U -lines) between two U -instants.
Let us onsider the U -line passing through x 2 o + Euo ; des ribed by the
world line fun tion rx given previously; a simple al ulation yields that rx (s) is
in the U -surfa e t if and only if s = jax j : In other words,
sx () :=

jaxj = no
(x o) 

whi h learly depends on x; is the time passed between the U -time points t0 and
t in the U -spa e point that x is in ident with.

6.6.6. Now we shall show that this observer is rigid.

Let Lo be a urve in o + Eu ; then the set of U -spa e points that meet Lo ;

L := fq 2 EU j q \ Lo 6= ;g is a urve in the observer spa e. Indeed, if po is a

parametrization of Lo ; then, a ording to the previous result on proper times,
o

pt := po + (no
po o) uo sh + no ( h 1)

is a parametrization of Lt for all t 2 IU :

Then

p_t = p_o + (no
p_o ) uo sh + no ( h 1)

and

jp_t j = jp_o j

(t 2 IU ):
Sin e U is regular and U -time is onsidered, U (pt (a))
p_t (a) = 0 for all
 2 R and a 2 Dom po = Dom pt ; this means that `t (L) = `o(L) for all

t 2 IU : It is not hard to see that every urve in the observer spa e an be

obtained from a urve in o + Euo by the previous method; onsequently, the
observer is rigid.

6.6.7. Two uniformly a elerated observers have been treated. Neither of

them possesses all the good properties of the uniformly a elerated observer in
the non-relativisti spa etime model. It is an open question whether we an nd
a spe ial relativisti observer U su h that
(i ) all U -lines are uniformly a elerated,
(ii ) U and AU are onstant on ea h instant (world surfa e) of an (arti ial)
time,
(iii ) U is rigid.
The observer in 6.4 does not satisfy (iii ); the observer in 6.5 does not satisfy
(ii ).
6.7. Uniformly rotating observer I

6.7.1. In dening the uniformly rotating observer we en ounter problems

similar to those in the previous se tion and, in the same manner, we nd two
possibilities but neither of them possesses all the good properties of the nonrelativisti uniformly rotating observer.
Let o 2 M; uo 2 V(1) and let
: Euo ! EIuo be a non-zero antisymmetri
linear map and dene the global observer
U (x) :=

uo

Note that
and

(x o) + uo
uo

1 + j

uo
(x o)j2

(x 2 M):

= U (o);
(x 2 M; q 2 Ker
):

U (x + q ) = U (x)

The observer has the a eleration eld

AU (x) =



uo

(x o)

(x 2 M):

6.7.2. To nd the U -lines we apply the well-proved method: transforming

the dierential equation x_ = U (x) by huo ;o we get
q

1 + j

qj2 ;
q_ =

q :
t_ =

The se ond equation an be solved immediately:

(s 2 I):

q (s) = es

qo

Then the rst equation be omes t_ =

with the initial value t(0) = 0 |
t(s) = s

1 + j

qo j2 having the solution |

(s 2 I):

1 + j

qo j2

Thus the U -line passing through x 2 o + Euo (the U -spa e point that x is
in ident with) is given by the world line fun tion
q

rx (s) = o + uo s 1 + j

(x o)j2 + es

(x o)

(s 2 I):

It is not hard to see that every U -line meets the hyperplane o + Euo ; thus
every U -line is of this form.

6.7.3. Note that the U -line passing through o + e; where e is in Ker

; is a
straight line dire ted by uo ; then the set of U -spa e points
fo + e + uo
Ij e 2 Ker
g

an be interpreted as the axis of rotation.

If x is in o + Euo ; then x o an be de omposed into a sum ex + qx where ex is
in Ker
and qx is orthogonal to Ker
: Then the U -line above an be written
in the form
q

rx (s) = o + ex + uo s 1 + !2 jqx j2 + es

qx;

()

where ! is the magnitude of

(see Exer ise V.3.21.1).
Hen e all the U -lines are omposed of an inertial line (with a proper time
\a elerated" relative to the proper time of the points of the axis) and a uniform
rotation.
Let Uo denote the global inertial observer that has the velo ity value uo : Let
us onsider Uo -time.
Put to := o + Euo : Then
sx (t) = q

1 + j

(x o)j2

1 + j

qx j2

time passes between the Uo -instants to and to + t in the U -spa e point that x is
in ident with.

The distan e observed by Uo at the Uo -instant to + t between the U -spa e

point that x is in ident with and o + ex + uo
I (the axis of rotation) equals

jrx (sx (t))

(o + ex + uo t)j = jqx j;

whi h is independent of t:

6.7.4.

Be ause of the term s 7! es

(x o) in () we an state that the
time period T of rotation is the same for all U -spa e points (out of the axis of
rotation), on erning their proper times: T = 2! :
On the other hand, on erning Uo -time, the time period of rotation of a
U -spa e point
p having the Uo -distan e d > 0 from the axis of rotation equals
To (d) := 2! 1 + !2 d2 ; it in reases from 2! to innity as d in reases from zero
to innity.
The following Figure illustrates the situation. Two U -line segments are
represented; the proper time passed along both segments equals 2! :

Another gure shows the plane in the Uo -spa e, orthogonal to Ker

; and
illustrates the angles of rotation of U -spa e points during a Uo -time interval 2! :

6.7.5. This observer is not rigid.

Let Lo be a urve in o + Eu ; then the set of U -spa e points that meet Lo ;

L := fq 2 EU j q \ Lo 6= ;g is a urve in the observer spa e. Indeed, if po is a

parametrization of Lo ; then
o

pt (a) := o + uo t + exp  q

1 + j

(po (a) o)j2
(a 2 Dom po )
is a parametrization of Lt for t = to + t 2 IUo : Then
0

p_t = expq

1 + j

(po

1 0

o)j2

1
A

po (a) o

po )
(

p_o )


(p o) + p_ C
o
oA:
3=2
1 + j

(po o)j2

t(

A B


We easily nd that
0

pt = exp  q

1 + j

(po

o)j2

(po o) + uo

1 + j

(po

o)j2 :

Then, using

(e

q1 )
(e

q2 ) = q1
q2

for all q1 ; q2 2 Euo and
 2 R; the reader an demonstrate without di ulty

that jp_t j2 + (U pt )
p_t 2 depends on t : the observer is not rigid.

6.7.6. This observer is not regular. It is easy to show that there is no world
surfa e g-orthogonal to U and passing through o:
Suppose su h a world surfa e F exists. Then F has Euo as its tangent spa e
at o:
For all q 2 Euo ;
f (a) := o + aq
(a 2 R)
is a fun tion su h that f (0) = o and


(U f )
f_ =

q + uo 1 + j

qj2

q = 0:

The urve (in fa t a straight line) Ran f passes through o 2 F and all of its
tangent ve tors are g-orthogonal to the orresponding values of U whi h would
imply that Ran f  F: Sin e q is arbitrary in Euo ; this means that o + Euo = F;
in parti ular, every tangent spa e of F equals Euo : However, if x 2 o + Euo = F
and x o is not in Ker
then U (x) 6= uo; thus the tangent spa e of F at x is
not g-orthogonal to U (x) : a ontradi tion.

6.8. Uniformly rotating observer II

6.8.1. Let o 2 M; uo 2 V(1) and
: Euo ! EuI be a non-zero antisymmetri
o

linear map and dene the non-global observer

n

Dom U := x 2 Mj j

uo
(x o)j2 < 1 ;
uo +

uo
(x o)
U (x) := q
1 j

uo
(x o)j2

(x 2 Dom U ):

If q is in Ker
; then
Dom U + q = Dom U
and

U (x + q ) = U (x)

The observer has the a eleration eld

(x 2 Dom U ; q 2 Ker
):

AU (x) =

6.8.2.

obtain

uo
(x o)
1 j

uo
(x o)j2

(x 2 Dom U ):

To nd the U -lines, we again use the known transformation and we

t_ = q

j

qj2

q
:
q_ = q
1 j

qj2
1

Now we apply a new tri k: \dividing" the se ond equation by the rst one we
get a very simple dierential equation whi h has the following orre t meaning.
Consider the initial onditions

0) = 0;

0) = x

t(

q(

o;

where x 2 o+Euo \(Dom U ): The formula for the derivative of inverse fun tion
results in | with the notations of 6.4 |
q
dsx (t)
()
= 1 j

q(sx (t))j2 :
dt
Then introdu ing the fun tion t 7! q(t) := q(sx (t)) we get the dierential
equation

ds (t))
dq(t)
= q_ (sx (t)) x
=

q(t)
dt
dt
whi h has the solution
q(t) = et

(x o)
(t 2 I):
Consequently j

q(sx (t))j = j

(x o)j; thus equation () be omes trivial
having the solution | with the initial ondition sx (0) = 0 |
sx (t) = t

j

(x o)j2 :

Finally we obtain
s

tx (s) = q

j

(x o)j2

qx (s) = exp  q

j

(x

;
1

o)j2

(x o)

from whi h we regain the world line fun tion giving the U -line passing through
x 2 o + Euo :
0

rx (s) = o + uo q

j

(x

+ exp  q

o)j2

(s 2 I):

j

(x

o)j2

(x o)

It is not hard to see that every U -line meets the hyperplane o + Euo ; thus
every U -line is of this form.

6.8.3. Note that the U -line passing through o + e; where e is in Ker

; is a
straight line dire ted by uo ; then the set of U -spa e points
fo + e + uo
Ij e 2 Ker
g
is interpreted as the axis of rotation.
If x is in o + Euo ; then x o an be de omposed into a sum ex + qx ; where
ex is in Ker
and qx is orthogonal to Ker
: Then the above given world line
fun tion an be written in the form
0

rx (s) = o + ex + uo q

!2jqx j2

+ exp  q

! 2 j qx j 2

A qx ;

()

where ! is the magnitude of

:
Hen e all the U -lines are omposed of an inertial line (with a proper time
\a elerated" relative to the proper time of the points of the axis) and a uniform
rotation.
Let Uo be the global inertial observer that has the velo ity value uo : Let us
onsider Uo -time putting to := o + Euo : Then
sx (t) = t

j

(x o)j2

is the time passed between the Uo -instants to and to + t in the U -spa e point
that x is in ident with.
The distan e observed by Uo at the Uo -instant to + t between the U -spa e
point that x is in ident with and o + ex + uo
I (the axis of rotation) equals

rx (sx (t))

whi h is independent of t:

(o + ex + uo t) = jqx j

6.8.4. Be ause of the term s 7! etx(s)

(x

o) in () we an state that the

time period To of rotation is the same for all U -spa e points (out of the axis of
rotation), on erning the Uo -time: To = 2! :
On the other hand, on erning the proper times of U -spa e points, the time
period of rotation of a U -spa e point having the Uo -distan e 0 < d < !1 from
the axis of rotation equals T (d) := !p12!2 d2 ; it in reases from 2! to innity as
d in reases from zero to !1 :
The following gure illustrates the situation. Two U -line segments are represented; the proper time passed along both segments equals 2! :

6.8.5. This observer is rigid.

Let Lo be a urve in o + Eu ; then the set of U -spa e points that meet Lo ;
L := fq 2 EU j q \ Lo 6= ;g is a urve in the observer spa e. Indeed, if po is a
o

parametrization of Lo then

pt := o + uo t + et

(po

o)

is a parametrization of Lt for t = to + t 2 IUo : Then

p_t = et

p_o
and we easily nd that
U

and

pt = uqo +

e
(po o2)
1 j

(po o)j

(U pt )
p_t =

(po
1

o)
p_o

j

(po o)j2

Consequently,


jp_t j2 + (U pt )
p_t 2 = jp_o j2 + p_o


(po
1 j

(po

o) 2
o)j2

is independent of t : the observer is rigid.

6.8.6. This observer furnishes a good instan e that the laws of Eu lidean
geometry do not hold ne essarily in the spa e of a non-inertial observer.
Sin e the observer

is rigid, all the lengths in U -spa e an be al ulated by
urves in o + Euo \ (Dom U ) whi h an be redu ed to urves in
Eu \(Dom U
o

o) =

= Ker
+ fq 2 Euo j q is orthogonal to Ker
; jqj <

1
g =: E
:
!

If Lo is a urve in o + Euo \ (Dom U ) then L := Lo o is a urve in E

; if
po is a parametrization of Lo then p := po o is a parametrization of L:
E
is a subset of the Eu lidean ve tor spa e Euo in whi h distan es and urve
lengths have a well-dened meaning; however, now urves in E
will represent
urves in U -spa e and their lengths will be al ulated in this sense. Thus, to
avoid misunderstanding, we shall say U -length and U -distan e, indi ating it in
notations, too.
A urve L in E
has the U -length

`U (L) =

Dom p

v
u
u
t p_ (a)

2
p_ (a)


p(a)
2
j +
da:
1 j

p(a)j2

(  )

Take arbitrary elements x and y in E

: Then we easily nd for the straight
line segment onne ting x and y; x; y[:= fx + a(y x)j 0 < a < 1g that

`U (x; y[)  jy

xj

and equality holds if and only if x

y = 0 whi h is equivalent to the fa t
that the straight line passing through x and y meets the kernel of
:
Suppose the straight line passing through x and y meets Ker
and L is a
broken line onne ting x and y; then the previous inequality implies

`U (L)  jy

xj = `U (x; y [):

As a onsequen e, the inequality above will be valid for an arbitrary L onne ting x and y be ause `(L) is obtained as the supremum of U -lengths of
broken lines approximating the urve L: Sin e the U -distan e dU (x; y) between
x and y is the inmum of urve lengths onne ting x and y we see that
dU (x; y ) = jy

xj

if the straight line passing through x and y interse ts Ker

:
Let d be an element of I; 0 < d < !1 ; and put

Cd := fq 2 E
j q is orthogonal to Ker
; jqj = dg:

Evidently, if q 2 Cd then q 2 Cd as well. Moreover, a ording to our

previous result, the U -distan e between q and q equals jq ( q)j = 2d:
This means that Cd represents a ir le of radius d in the observer spa e. Let
us al ulate the ir umferen e of this ir le.
Choosing the parametrization
p:

;  ! Cd ;

a 7! exp a
!

qo

where qo is an arbitrarily xed element of Cd ; we nd

p;
!
j

pj2 = !2 d2 ;
p_ =

jp_ j2 = d2;

2
p_


p = ! 2 d4 :

Applying formula (  ) we obtain

`U (Cd ) =

p 2d2 2 :
1 !d

The ir umferen e of the ir le of radius d is longer than 2d:

6.9. Exer ises

1. Let U be a global inertial observer with velo ity value u: Take a velo ity
value uo 6= u and the arti ial time onsisting of the hyperplanes dire ted by Euo
(i.e. IUo ): Demonstrate that the distan es in U -spa e al ulated at IUo -instants
a ording to denition 6.3.3 equal the distan es dened earlier in U -spa e.
2. Suppose the arti ial time points are hyperplanes or hyperplane se tions
and the observer U is onstant on them. Take su h an arti ial time point t;
then there is a ut 2 V(1) su h that U (x) = ut for all x in t: Suppose t is dire ted
by E: Then
jut
(q1 + q2 )j  jut
q1 j + jut
q2 j

for all q1 ; q2 2 E: As a onsequen e, straight lines realize the distan e between

the points of t; thus
dt (q1 ; q2 ) = jut (q2 ? t

q1 ? t)j

(q1 ; q2 2 EU ):

3. Let U be the uniformly a elerated observer treated in 6.5. Then

vU (x)uo

ao

( uo
(x o) )

1 + jaoj2 uo
(x o) 2

(x 2 M):

4. Let U be as before. Verify that every U -line is obtained from a hosen

one by a translation with a ve tor in Euo : In other words, EU endowed with the
subtra tion
q0 q := x0 x
(x0 2 q; x 2 q; x0 x 2 Euo )

is an ane spa e over Euo :

5. Let U be the uniformly a elerated observer treated in 6.6. Then the U line passing through x 2 o + Euo interse ts t if and only if ao
(x o) < ln jj:
6. Show that
Dom U = fo + uo +  ao + qj  > 0;  2 jao j2 > 2 ; uo
q = 0; ao
q = 0g
for the uniformly a elerated observer treated in 6.6.
7. Let U be as before. Then
vU (x)uo

ao

uo
(x

ao
(x

o)

o)

(x 2 Dom U ):

8. Show that the distan e observed by the inertial observer Uo with velo ity
value uo between the spa e points of the uniformly a elerated observer treated
in 6.6. is not onstant in Uo -time. Give an explanation similar to that in 6.5.6.

9. Verify that

1
Dom U = o + uo
I + Ker
+ q 2 Euo j q is orthogonal to Ker
; jqj <
!
for the uniformly rotating observer treated in 6.8.
10. Demonstrate that

uo
(x o)
vU (x)uo = q
1 + j

uo
(x o)j2
and

(x 2 M)

(x 2 Dom U )
=

uo
(x o)
where U is the uniformly rotating observer treated in 6.7. and 6.8, respe tively.
11. The uniformly rotating observer treated in 6.8. is not regular.
12. Let o be a world point and onsider the observer
x o
U (x) :=
(x 2 o + T! ):
jx oj
vU (x)uo

Prove that
U

EU = fo + u
I+ j u 2 V(1)g:

is regular and

fV(1)tj t 2 I+ g

is the set of U -surfa es (U -instants).

Show that if this observer dened simultaneity like an inertial observer (light
signals and mirrors, see 3.2. ) then simultaneity would depend on the U -spa e
point of the light sour e.
13. Let o 2 M; uo 2 V(1); h 2 I and dene the observer
Dom U := fx 2 Mj h2 juo
(x o)j2 < 1g;
uo + huo
(x o)
U (x) := q
1 h2 juo
(x o)j2

(x 2 Dom U ):

Applying the
method given in 6.4. nd that the U -line passing through


x 2 o + Euo \ (Dom U ) is given by the world line fun tion

rx (s) = o + uo tx (s) + ehtx (s) (x o)

where s 7! tx (s) is the solution of the dierential equation
(t :

I  I)?

t_ = q

1 h2 jx oj2 e2ht

with the initial ondition t(0) = 0:

14. Let o 2 M; uo 2 V(1); h 2 I and dene the observer
U (x) := uo

1 + h2 juo
(x o)j2 + huo (x o)

Applying the method given in 6.4 nd that the

o + Euo is given by the world line fun tion

U -line

(x 2 M):

passing through x 2

rx (s) = o + uo tx (s) + ehs (x o)

whereqs 7! tx (s) is the fun tion for whi h tx (0) = 0 holds and has the derivative
2
s 7! 1 + h2 jx oj e2hs :
15. Compare the observers of the previous two exer ises with the nonrelativisti observer in Exer ise I.5.4.9.

7. Ve tor splittings

7.1. Splitting of ve tors

7.1.1. For u 2 V(1) we have already dened

u : M ! I;
x 7! u
x
and

u

M ! Eu;

x 7! x

(u
x)u = x + (u
x)u

i.e. with the usual identi ations,

u

= u;

u

=g+u
u

(see 1.3.2) and the linear bije tion hu := (u ; u) :

inverse
(t; q) 7! ut + q
(see 1.3.5).

M ! I  Eu having the

Denition. u
x = u
x and u
x are alled the u-timelike omponent and

the u-spa elike omponent of the ve tor x: ( u
x; u
x)
is the u-split form of x: hu = (u ; u ) is the splitting of M orresponding
to u; or the u-splitting of M:
Note that
x
y = (u
x)(u
y ) + (u
x)
(u
y );

in parti ular,

= (u
x)2 + ju
xj2

for all x; y 2 M: In other words,
x2

hu
x = (t; q )

if

then

x2

= t2 + jqj2 :

7.1.2.
If A is a measure line, A
M M
is split into (A
I)  (A
Eu )
A


I  Eu by hu ; thus the u-timelike omponent and the u-spa elike omponent
A A




of a ve tor of type A ( otype A) are in A
I AI and in A
Eu EAu ;

respe tively.
In parti ular, hu splits MI into R  EIu and for all u0 2 V(1)
hu
u0

u
u0 ; u0 + (u
u0 )u

jvu0 uj2

(1; vu0 u ):

7.1.3. In ontradistin tion to the non-relativisti ase, here not only the uspa elike omponent but also the u-timelike omponent of ve tors depend on u:
The transformation rule that shows how the u- omponents of a ve tor vary with
u; is mu h more ompli ated here than in the non-relativisti ase.
Proposition. Let u; u0 2 V(1): Then for all (t; q) 2 I  Eu we have
hu0
hu 1
0

(t; q) =
1

= q

1
1

( u0
u)t u0
q; (u + (u0
u)u)t + q + (u0
q)u0

jvu0 u j2

jvu0 u j2

(t vu0 u
q);

vuu0 t

vuu0

+ vu0 u 1

jvu0 u j2

jvu0 u j2

(vu0 u
q)A + qA :

Proof. The rst equality is quite simple. The se ond one is derived
withthe

0

aid of the formulae in 4.3.2 and the relation u0
q = (u0
u)

whi h is true be ause u
q = 0:
Note that both vu0 u and vuu0 appear in that formula.

u
u0
u

q

7.1.4. The previous formula is not a good transformation rule: we want to

ompare the u0 - omponents of a ve tor with its u- omponents (t; q): However,
the u0 - omponents and the u- omponents are in dierent spa es: (t; q) is in


(t; q) is in I  Eu0 ; they annot be ompared dire tly.

To obtain a onvenient formula, we have to relate Eu0 and Eu; we have agreed
that su h a relation is established by the orresponding Lorentz boost. Thus,
leaving invariant
the rst omponent, we shall transform the se ond omponent


of hu0
hu 1
(t; q) by L(u; u0):
Denition. Let u; u0 2 V(1): Then

I  Eu and

hu0
hu 1

:= idI  L(u; u0)jEu0

Hu0 u

(hu0
hu 1 )

is alled the ve tor transformation rule from u-splitting into u0 -splitting.

Proposition. For all (t; q) 2 I  Eu we have

0

Hu0 u
(t; q ) =  q

jvu0 u j2

jvu0 u j2

 vu0 u t

(t vu0 u
q);
q

11

jvu0 u j2
(vu0 u
q)AA + qA :
jvu0 u j2
1

In onne tion with this formula we mention the following frequently useful
relation:
q
1
1 jvu0 uj2
1
q
:
=
2
jvu0 u j
1 + 1 jvu0 u j2

7.1.5. The previous formula is a bit fearsome. We an make it more apparent

de omposing q into a sum of ve tors parallel and orthogonal to vu0 u :
(t; q) = (t; qk ) + (0; q? )

where qk is parallel to vu0 u ; i.e. there is a  2 I su h that qk = vu0 u and q?

is orthogonal to vu0 u; i.e. vu0 u
q? = 0:
Then we easily nd that

0; q?) = (0; q?);

Hu0 u
(

Hu0 u
(t; qk ) = q

jvu0 u j2

vu0 u
qk ;

vu0 u t + qk

7.1.6. The last formula | in a slightly dierent form | appears in the

literature as the formula of Lorentz transformation. To get the usual form we

put v := vu0 u ; let (t; q) denote the u- omponents of a ve tor and let (t0 ; q0 )
denote its u0 - omponents mapped by the Lorentz boost L(u; u0 ) into I  Eu ;
then supposing q is parallel to v we have
t0

jv j 2

q0

(t v
q);

jv j 2

( vt + q):

This (or its equivalent in the arithmeti spa etime model) is the usual \Lorentz
transformation" formula.
We emphasize that q0 is not the u0 -spa elike omponent of the ve tor having
the u- omponents (t; q); it is the Lorentz-boosted u0-spa elike omponent.
Lorentz transformations (see Se tion 9) are transformations of ve tors, i.e.
mappings from M into M; the transformation rule is a mapping from I  Eu
into I  Eu : Transformation rules and Lorentz transformations are dierent
mathemati al obje ts. Of ourse, there is some onne tion between them. We
easily nd that
Hu0 u = hu
L(u; u0 )
h 1
u

where L(u; u0) is the Lorentz boost from u0 into u:

In the split spa etime model M and I  Eu oin ide: the spe ial stru ture
of the split spa etime model (and the arithmeti spa etime model) involves the
possibility of onfusing transformation rules with Lorentz transformations.

7.1.7. Using a matrix form of the linear maps I  Eu ! I  Eu (see IV.3.7)

we an write
Hu0 u

= (vu0 u)

vu0 u

vu0 u
D (vu0 u )

where

(v) :=

jvj2

1
(v)2
D (v ) :=
idEu +
v
v
(v)
(v) + 1
for v 2 EIu ; jvj < 1:

7.2. Splitting of ove tors

7.2.1. For u 2 V(1); M is split by the transpose of the inverse of hu :


ru := hu 1 : M ! (I  Eu )  I  Eu :
Then for all k 2 M ; (t; q) 2 Eu we have
(ru
k)
(t; q) = k
hu 1
(t; q) = k
(ut + q) = (k
u)t + k
q:
Of ourse, instead of k in k
q we an write kjEu = iu
k = k
iu 2 Eu : Then
we an state that

ru
k = (k
u; k
u ) = (u
k; u
k)

(k 2 M ):

This form is suitable for a omparison with the non-relativisti ase. However,
we an get a form more onvenient from the point of view of appli ations.
Applying the usual identi ations we have iu = u (see 1.3.6), thus
ru
k = (u
k; u
k)

Re all the identi ation

ve tor of otype I
I; too:

(k 2 M ):

M  IM
I whi h implies that k an be split as a
hu
k = ( u
k; u
k)

(k 2 M ):

The two splittings are nearly the same. In the literature (in a somewhat
dierent setting) the split form of k 2 M by ru and hu are alled the ovariant
and the ontravariant omponents of k; respe tively.
Of ourse, in view of M  I
I
M ; also the elements of M an be split by
ru : a ve tor, too, has ovariant and ontravariant omponents.
Introdu ing the notation
ju

I  Eu ! I  Eu;

(t; q) 7! ( t; q)

we have (with the usual identi ations)

ru

= ju
hu :

Note that ru 1 = hu 1
ju ; i.e.

ru 1
(e; p) =

eu + p

(e 2 I ; p 2 Eu ):

7.2.2. The ove tor transformation rule is dened to be

:= idI  L(u; u0)jEu
ru0
ru 1 :
It an be easily dedu ed from the ve tor transformation rule that, apart from
a negative sign, they are the same. Indeed,


idI  L(u; u0)jEu0 ju0 = ju
idI  L(u; u0 )jEu0 ;
thus
Ru0 u = ju
Hu0 u
ju :
Consequently, if (e; p) 2 I  Eu and p is parallel to vu0 u then


1
Ru0 u
(e; p) = q
e + vu0 u
p; vu0 u e + p :
2
1 jvu0 u j

Ru0 u

7.2.3.

It is worth mentioning that Eu an be onsidered to be a linear


subspa e of M ; sin e Eu  IE
uI  IM

I  M and
E = fk 2 Mj k
u = 0g;
u

in other words, Eu is the annullator of u

I:
In the non-relativisti ase E is not a linear subspa e of M : For all u 2 V(1)
there is a linear subspa e E
u of M ; the annullator of u
I; but it is not
the dual of any linear subspa e in M:
Observe that the spe ial relativisti ve tor transformation rule whi h is nearly
the same as the ove tor tramsformation rule resembles a ombination of the
non-relativisti ve tor and ove tor transformation rules.
We emphasize that in the spe ial relativisti ase there is no absolute spa elike
ve tor and there is no absolute timelike ove tor, in ontradistin tion to the nonrelativisti ase.

7.3. Splitting of ve tor elds

7.3.1. In appli ations ve tor elds M  M and ove tor elds M  M
appear frequently. Evidently, a ove tor eld an be onsidered a ve tor eld of
otype I
I: Their splitting a ording to global inertial observers an be treated

analogously to the non-relativisti ase (see I.8.5).

Let U be a global inertial observer with the velo ity value u and let K : M
M be a ove tor eld. At every world point x the value K (x) is split a ording
to the velo ity value u so we get the half U -split form of the eld:


( Vu ; Au ) := ru
K : M I  E ;
x 7!= u
K (x); u
K (x) :

Furthermore, the observer splits spa etime as well, thus instead of world points

U -instants and U -spa e points will be introdu ed to get the ompletely split form

of the eld:

( VU ; AU ) := ( Vu ; Au ) HU 1 = ru
K HU 1 : IU  EU I  Eu ;


(t; q) 7! u
K (q ? t); u
K (q ? t) ;

where q ? t denotes the single element in the interse tion of q and t:

7.3.2. Potentials are ove tor elds. We an introdu e the s alar potential
and the ve tor potential a ording to an observer by the previous split forms.
Regarding the transformation rule on erning s alar potentials and ve tor potentials we an repeat essentially what we said in I.8.5.3; of ourse, the transformation rule will be signi antly more ompli ated.
An important dieren e between the non-relativisti spa etime model and the
spe ial relativisti one is that here there are no absolute s alar potentials be ause
there are no absolute timelike ove tors. This fore asts that the des ription of
gravitation in the relativisti ase will dier signi antly from its des ription in
the non-relativisti ase where absolute s alar potentials are used.
7.3.3. In ontradistin tion to the non-relativisti ase, for e elds are split
dierently a ording to dierent observers.
M : Be ause of the property
Let us take a for e eld f : M  V(1)
I
0
0
0
f (x; u )
u = 0 for all (x; u ) 2 Dom f ; the u-spa elike omponent and the
u-timelike omponent of f are not independent. Using the formula in 7.1.1 we
get
0 = f (x; u0 )
u0 = u
f (x; u0 )
(u
u0) + u
f (x; u0)

(u
u0);

whi h yields

u
f (x; u0 ) = u
f (x; u0 )

vu0 u :

7.3.4. Splittings of ve tor elds a ording to rigid observers in the nonrelativisti ase an be treated in the mathemati al framework of ane spa es.
However, splittings a ording to general observers require the theory of manifolds.
In the spe ial relativisti ase splittings a ording to non-inertial observers
an be treated only in the framework of manifolds and they do not appear here.
7.4. Exer ises

1. Show that u
x = (u ^ x)
u for all u 2 V(1); x 2 M:

2. Take the arithmeti spa etime model. Give the ompletely split form of
the ve tor eld


( 0 ; ) 7!  1 +  2 ; os( 0  3 ); 0; 0
a ording to the global inertial observer with the velo ity value
p11 v2 (1; v; 0; 0):

8. Tensor splittings

8.1. Splitting of tensors

8.1.1. The various tensors | elements of M
M; M
M; et .

| are
split a ording to u 2 V(1) by the maps hu
hu ; hu
ru ; et . as in the
non-relativisti ase. However, now it su es to deal with hu
hu be ause the
identi ation M  IM

I and ru = ju
hu (see 7.2.1) allow us to derive the other

splittings from this one.
With the usual identi ations we have

M
M ! (IEu)
(IEu)  (I
I)(I
Eu)(Eu
I)(Eu
Eu);
and for T 2 M
M :

hu
hu

(hu
hu )
T = hu
T
hu = hu
T
ru 1 =
=

u
T
u
T
u u(u
T
u) T

for L 2 M
M :

2 M
M :

(ru
hu )
P = ru
P
hu = ru
P
ru 1 =
for F

2 M
M :


u
T
u

u
T
u

u
T u(u
T
u)
+ u
(u
T ) + (T
u)
u + u
u(u
T
u)

(hu
ru )
L = hu
L
ru = hu
L
hu 1 =
for P

u
T
u
u
T
u

u
L
u
u
L
u


u
L
u

u
L
 u


u
P
u
u
P
u

u
P
u u
P
u

=



 
(ru
ru )
F = ru
F
ru = ru
F
hu 1 = u

FF

uu u

FF

u :
u
u
u

8.1.2. The splittings orresponding to dierent velo ity values u and u0 are

dierent. The tensor transformation rule that shows how the splittings depend
on velo ity values is rather ompli ated, mu h more ompli ated than in the
non-relativisti ase. We shall study it only for antisymmetri tensors.

8.2. Splitting of antisymmetri tensors

8.2.1. If T is an antisymmetri
tensor, i.e. T 2 M ^ M; then u
T
u = 0;
u
T
u
= (u
T
u) and u
T
u 2 Eu ^ Eu whi h means (of ourse)
that the u-split form of T is antisymmetri as well. Thus u-splitting maps the
elements of M ^ M into elements of form

0 a   a; A

a

where a 2 Eu
I  I
Eu; A 2 Eu ^ Eu:
The orresponding formula in 8.1.1 gives for T 2 M ^ M


hu
T
h = T
u; T + (T
u) ^ u :
u

Denition. T
u and T + (T
u) ^ u are alled the u-timelike omponent

and the u-spa elike omponent of the antisymmetri tensor T :
8.2.2. The following transformation rule shows how splittings depend on
velo ity values.
Proposition. Let u; u0 2 V(1): Then


Hu0 u
a; A
H  0 =
uu

= q

jvu0 u j2

a + vu0 u

jvu0 u j2

0
 a

jvu0 uj2
(vu0 u
a) + A
vu0 u A ;
jvu0 u j2
1

A
vu0 u

jvu0 u j2 A
^ vu0 u + AA :
jvu0 u j2

Proof. Using the matrix forms we have

Hu0 u
a; A

Hu 0 u =

= (vu0 u )2

vu0 u

vu0 u
D (vu0 u )



from whi h we an get the desired formula.

a
A



vu0 u

vu0 u
D (vu0 u )

8.2.3. The previous fearsome formula be omes ni er if we write

sum of omponents parallel and orthogonal to the relative velo ity:

a = ak + a? ;

a; A

as the

A = Ak + A?

where ak is parallel to vu0 u ; a? is orthogonal to vu0 u ; and the kernel of Ak is

parallel to vu0 u; i.e. Ak
vu0 u = 0 and the kernel of A? is orthogonal to vu0 u ;
i.e. (A?
vu0 u ) ^ vu0 u = jvu0 u j2 A? (see Exer ise V.3.21.1). Then we easily
nd




Hu0 u
ak ; Ak
Hu 0 u = ak ; Ak ;




1
Hu0 u
a? ; A?
Hu 0 u = q
a? + A?
vu0 u ; a? ^ vu0 u + A? :
2
1 jvu0 u j

8.2.4. The splitting and the transformation rule of antisymmetri otensors

i.e. elements of M ^ M are the same, apart from a negative sign. The details
are left to the reader.
It is interesting that here, in ontradistin tion to the non-relativisti ase,
M ^ M makes
sense be ause of the identi ation M  IM

I : The mixed tensor


H 2 M ^ M has the u-split form
hu
H
hu 1

u
= H0
u H + (H
H
u) ^ u


whi h, as a matrix, is not antisymmetri . It need not be antisymmetri , be ause

the symmmetri or antisymmetri properties of matri es refer to these properties of linear maps regarding duals without any identi ations (see IV.1.5 and
V.4.19).

8.3. Splitting of tensor elds

8.3.1. The splitting of various tensor elds a ording to inertial observers

an be treated analogously to the non-relativisti ase.
Let U be a global inertial observer with velo ity value u: The antisymmetri
otensor eld F has the half split form a ording to U
Eu ; Bu

:= ru
F
ru : M (Eu
I )  (Eu ^ Eu);




x 7! F (x)
u; F (x) + F (x)
u ^ u

and the ompletely split form

EU ; BU

:= Eu ; Bu HU 1 =
= ru
(F HU 1 )
ru : I  EU (Eu
I)  (Eu ^ Eu );




(t; q) 7! F (q ? t)
u; F (q ? t) + F (q ? t)
u ^ u :

8.3.2. The ele tromagneti eld is des ribed by an antisymmetri otensor

eld F whi h is the exterior derivative of a potential K ; F = D ^ K : The
ele tri eld and the magneti eld relative to the inertial observer U are the
orresponding omponents of the ompletely split form of F :
The relation between the ompletely
split form ( VU ; AU ) of K and the om

pletly split form EU ; BU of F is exa tly the same as in the non-relativisti
ase:
EU = 0 AU rVU ;
BU = r ^ AU :
Sin e the for e eld dened by the potential K equals
f (x; u0 ) = F (x)
u0

(x 2 Dom K ; u0 2 V(1));

where F := D ^ K ; we an state again that a masspoint in the world point x

having the velo ity value u0 \feels" only the u0 -timelike omponent of the eld; a
masspoint always \feels" the ele tri eld a ording to its instantaneous velo ity
value.
Be ause of the more ompli ated tranformation rule in the spe ial relativisti
ase the Lorentz for e is expressed by the U -ele tri eld and the U -magneti
eld more ompli atedly than in the non-relativisti ase.

8.4. Exer ises

1. Let x 2 M; T 2 M
M and L 2 M
M : Give the u-split form of T
x

and L
x using the u-split form of x; T and L:

2. Let T and L as before. Give the u-split form of T
L and L
T using the
u-split form of T and L:
3. Re all the non-degenarate bilinear form (see V.4.15)
(M ^ M )  (M ^ M) ! R;

(F ; H ) 7! F  H :=

1
Tr F
H :
2

Express F
H using the u-timelike and the u-spa elike omponents of F and

H:

9. Referen e frames

9.1. The notion of a referen e frame

9.1.1. We an repeat word by word what we said in I. 7.1.1 with the single
ex eption that instead of (absolute) time now we have to onsider an (arti ial)
time derived from a simultaneity, to arrive at the following notion.
Re all that an observer U together with a simultaneity S S establishes the
splitting HU ;S = (S ; CU ) : M IS  EU :
Denition. A referen e system is a quartet (U ; S ; TS ; SU ) where
(i ) U is an observer,
(ii ) S is a simultaneity on the domain of U ;
(iii ) TS : IS R is a stri tly monotone in reasing mapping,
(iv ) SU : EU R3 is a mapping
su h that (TS  SU ) HU ;S = (TS S ; SU CU ) : M R  R 3 is an orientationpreserving oordinatization.
We all TS and SU the oordinatization of S -time and U -spa e, respe tively,
in spite of the fa t that we introdu ed the notion of oordinatization only for
ane spa es and, in general, neither IS nor EU is an ane spa e. (We mention
that in any ase IS and EU an be endowed with a smooth stru ture and in the
framework of smooth stru tures TS and SU do be ome a oordinatization.)
Note that ondition (iii ) involves that TS is dened on a subset of IS where
the ordering \later" is total; onsequently, the oordinatization of spa etime is
dened on a subset of Dom U where the simultaneity is well posed.




9.1.2. Denition. A oordinatization K :M  R  R3 is alled a referen e

frame if there is a referen e system (U ; S ; TS ; SU ) su h that K = (TS  SU )

HU ;S :
U ; S ; TS and SU are alled the observer, the simultaneity, the S -time oordinatization and the U -spa e oordinatization orresponding to the referen e
frame.
As usual, we number the oordinates of RR 3 from zero to three. A ordingly,
we nd onvenient to use the notation K = (0 ; ) : M
R  R3 for the
oordinatizations of spa etime. Then the equality

D(x)
0 K 1 (K (x)) = 0
well-known and used in the non-relativisti ase will hold now as well, sin e its
dedu tion rests only on the ane stru ture of M:
If K is a referen e frame then

0 = TS S ;

 = SU

CU :

9.1.3. Proposition.

A oordinatization K = (0 ; ) : M
R  R3 is a
referen e frame if and only if
(i ) K is orientation preserving,
(ii ) 0 K 1(
K (x)) is a future-dire ted timelike ve tor,
(iii ) D0 (x) is a future-dire ted timelike ve tor
for all x 2 Dom K:
Then
0 K 1(K (x))
U (x) =
(x 2 Dom K );
(1)
j0 K 1(K (x))j
is the orresponding observer and the orresponding simultaneity S is determined
as follows:

x is simultaneous with y if and only if 0 (x) = 0 (y)

(2)

moreover,

TS (t) = 0 (x)
SU (q) = (x)

(t 2 IS ; x 2 t);
(q 2 EU ; x 2 q):

(3)
(4)

Proof.
If K is a referen e frame, K = (TS  SU ) HU ;S ; then (i ) is trivial.
0

 is onstant on the S -instants. In other words, S -instants | more pre0

isely
their part


in Dom K
| have the form fx 2 Dom K j  (x) = g: Then
0
x 2 Mj D (x)
x = 0 is the tangent spa e of the orresponding
world surfa e passing through x: Sin e this tangent spa e is spa elike, D0 (x) must
be timelike. If y x 2 T! ;
then the properties of S and TS imply that
0 (y)
0 (x) > 0; then D0 (x)
(y x) + ordo(y x) > 0 results in that
D0 (x)
x > 0 for all x 2 T! ; proving (iii ).
As on erns (ii ), note that a world line fun tion r satises r_(s) = U (r
(s))
and takes values in the
domain of K if and only if K (r(s)) = 0 (r(s));  i.e.
r(s) = K 1 0 (r(s);  for a  2 R3 and for all s 2 Dom r: As a onsequen e,
we have
U (r(s))

d
K 1 0 (r(s)); ) = 0 K 1 (0 (r(s)); ))
D0 (r(s))
r_ (s)
ds

whi h, together with ondition (iii ), implies that U (x) is a positive multiple of
0 K 1 (K (x)) for all x 2 Dom K; proving (ii ) and equality (1):
Suppose now that K = (0 ; ) is a oordinatization that fulls onditions
(ii ){(iii ).
Then ondition (ii ) implies that U dened by equality (1) is an observer.
A ording to (iii ), the simultaneity S is well dened by (2) (i.e. the subsets
of form fx 2 Dom K j 0(x) = g are world surfa es and S is smooth). Consequently, TS is well dened by formula (3) and it is stri tly monotone in reasing.

If r is a world line su h that r_ (s) = U (r(s)) then

d
K
((r(s)) = D(r(s))
U (r(t)) = D(r(s))
0
ds
j 0 K

1 (K (r(s)))
1 (K (r(s)))j =

whi h means that  r is a onstant mapping, in other words,  is onstant on

the U -lines; hen e SU is well dened by formula (4):
Finally, it is evident that K = (TS  SU ) HU ;S :

9.2. Lorentz referen e frames

9.2.1. Now we are interested in what kind of ane oordinatization of

spa etime an be a referen e frame.
Let us take an ane oordinatization K : M ! R4 : Then there are
| an o 2 M;
| an ordered basis (x0 ; x1 ; x2 ; x3 ) of M
su h that
K (x) =

ki
(x

K 1 ( ) =

3
X
i=0

o)j i = 0; 1; 2; 3

 i xi

(x 2 M);
( 2 R4 );

where (k0 ; k1 ; k2 ; k3 ) is the dual of the basis in question.

Proposition. The ane oordinatization K is a referen e frame if and only
if
(i ) (x0 ; x1 ; x2 ; x3 ) is a positively oriented basis,
(ii ) x0 is a future-dire ted timelike ve tor,
(iii ) x1 ; x2 ; x3 are spa elike ve tors spanning a spa elike linear subspa e of M:
Then the orresponding observer is global and inertial, having the onstant
value
x0
u :=
;

jx0 j

and the simultaneity is given by the hyperplanes dire ted by the spa elike subspa e that x1 ; x2 ; x3 span.
Proof. We show that the present onditions (i ){(iii ) orrespond to the
onditions listed in Proposition 8.1.3.
(i ) The oordinatization is orientation-preserving if and only if the orresponding basis is positively oriented;
(ii ) 0 K 1 (K (x)) = x0 ;

(iii ) D0 (x) = k0 for all x 2 M: Sin e k0
x = 0 ( = 1; 2; 3); k0

is timelike if and only if x -s span a spa elike linear subspa e; then, sin e
k0
x0 = 1 > 0 and sin e x0 is future-dire ted timelike, k0 must be futuredire ted.

9.2.2. Let G denote the4 Lorentz form on R4 treated in V.4.19 and re all that
a linear map L : M ! R is alled g-G-orthogonal if there is an s 2 I su h that
G(L
x
L
y) = g(sx;y) for all x; y 2 M:
Denition. A referen e frame K is alled Lorentzian if
| K is ane,
| K : M ! R4 is g-G-orthogonal.
2

From the previous result we get immediately the following:

Proposition. A referen e frame K is Lorentzian if and only if there are
(i ) an o 2 M;
(ii ) a positively oriented g-orthogonal basis (e0 ; e1 ; e2 ; e3 ); normed to an s; of
M su h that e0 is future-dire ted timelike,
and



ei
(x o) 
i
=
0
;
1
;
2
;
3
(x 2 M):
K (x) =
2

ei

We shall use the following names for a Lorentz referen e frame: o is its origin,
(e0 ; e1 ; e2 ; e3 ) is its spa etime basis ; moreover, s := je0 j is its time and distan e
unit, u := es0 is its velo ity value and (e1 ; e2 ; e3 ) is its spa e basis.

9.2.3. Let K be a Lorentz referen e frame and use the previous notations.
We see from 1.6 that the Lorentz referen e frame establishes an isomorphism
between the spa etime model (M; I; g) and the arithmeti spa etime model.
More pre isely, the oordinatization K and the mapping B : I ! R; t 7! st
onstitute an isomorphism.
This isomorphism transforms ve tors, ove tors and tensors, otensors et .
into ve tors, ove tors et . of the arithmeti spa etime model.
In parti ular,
K

! R4 ;

x 7!

ei
x 
e2i 

i = 0; 1; 2; 3

is the oordinatization of ve tors and

(K 1) : M ! R4 ;
k 7! k
ei j i = 0; 1; 2; 3 ;
is the oordinatization of ove tors.
We an generalize the oordinatization for ve tors ( ove tors)
of type or


A; M ; too, where A
M
otype A; i.e. for elements in M
A or M
A
A

is a measure
line. For
instan e, elements of M
or IM

I are
oordinatized by the


I
e
e
basis si  i = 0; 1; 2; 3 and by the basis s2i  i = 0; 1; 2; 3 ; respe tively:

M ! R4 ;
I
M 4
I
I !R ;

ei
w 
w 7! s
e2 
 i 
ei
p 
p 7! s2
e2i 

i = 0; 1; 2; 3 ;


i = 0; 1; 2; 3 :

9.3. Equivalent referen e frames

9.3.1. We an repeat, a ording to the sense, what we said in I.7.5.1.

Re all the notion of automorphisms of the spa etime model (see 1.6.1). An
automorphism is a transformation that leaves invariant (preserves) the stru ture
of the spa etime model. Stri t automorphisms do not hange time periods and
distan es.
It is quite natural that two obje ts transformed into ea h other by a stri t
automorphism of the spa etime model are onsidered equivalent (i.e. the same
from a physi al point of view).
Re alling that O(g) denotes the set of g-orthogonal linear maps in M (see
V.2.7) let us introdu e the notation

P +! :=
fL : M ! Mj L is ane, L 2 O(g);

is orientation- and arrow-preserving g

and let us all the elements of P +! proper Poin are transformations. We shall
study these transformations in the next paragraph. For the moment it su es
to know the quite evident fa t that (L; idI ) is a stri t automorphism if and only
if L is a proper Poin are transformation.

9.3.2. Denition. The referen e frames K and K 0 are alled equivalent if

there is a proper Poin are transformation L su h that

K 0 L = K:
Two referen e systems are equivalent if the orresponding referen e frames
are equivalent.

Proposition. Let (U ; S ; TS ; SU ) and (U 0 ; S 0 ; TS00; SU 0 ) be the referen e sys-0

tems orresponding to the referen e frames K and K ; respe tively. If K and K

are equivalent, K 0 L = K; then
(i ) L
U = U 0 L;
(ii ) (TS 01 TS ) S = S 0 L
(iii ) (SU 10 SU ) CU = CU 0 L

Proof. For (i ) we an argue as in I.10.5.3, using (L
x)2 = x2 for all x 2 M:

As on erns (ii ) and (iii ), we an opy the reasoning of (iii ) in I.10.5.3.

9.3.3. Now we shall see that our denition of equivalen e of referen e frames
is in a ordan e with the intuitive notion expounded in I.10.5.1.
Proposition. Two Lorentz referen e frames are equivalent if and only if they
have the same unit of time (and distan e).
Proof.
Let the Lorentz referen e frames K and K 0 be dened by the origins
0
o and o and the spa etime bases (e0 ; e1 ; e2 ; e3 ) and (e00 ; e01 ; e02 ; e03 ); respe tively.
Then L := K 0 1 K : M ! M is the ane bije tion determined by
L(o) = o0 ;

L
ei

= e0i

(i = 0; 1; 2; 3):

Evidently, L is orientation-preserving. Moreover, L 2 O(g) if and only if

je0 j = je00 j and it is arrow-preserving if and only if e0 and e00 have the same

arrow.

9.4. Curve lengths al ulated in oordinates

9.4.1. In 6.3.3 we dealt with lengths of urves in the spa e of an observer U at

instants of a simultaneity S : It is an interesting question how to al ulate these
lengths in oordinates orresponding to a referen e frame K = (TS  SU ) HU ;S :
We shall use the notation P := K 1 (P is the parametrization orresponding
to the oordinatization K ):
Let L and Lt be as in 6.3.3 and let  0 be the oordinate of t 2 IS ; i.e.
0
 = S (t):
A parametrization pt of Lt has the oordinatized form




a 7! K (pt (a)) =:  0 ; (p (a)j  = 1; 2; 3) =:  0 ; p(a)

from whi h we dedu e

pt = P ( 0 ; p);
p_t =  P ( 0 ; p)p_ 

(Einstein summation).

Furthermore, we know (see 9.1.3)

U (P ) =

0 P
j0 P j :

Consequently,

jU (pt )
p_t j2 = jp_t j2 + jU (pt )
p_t j2 =

!
(0 P
 P )(0 P
 P )
( 0 ; p)p_ p_ :
=  P
 P +
j0 P j2
Let us put

gik := i P
k P (i; k = 0; 1; 2; 3):

Taking into a ount that g00 is negative, we see that
b := g g0g00g0 (;  = 1; 2; 3)

is the \metri tensor" in the U -spa e, i.e. a urve in the U -spa e parametrized by p at an S -instant oordinatized by  0 has length
Z q

b 0; p(a)
p_(a)p_ (a)da:

9.4.2. Note that gik is a fun tion from
R4 into I

I:
We know that i P ( )j i = 0; 1; 2; 3 is a basis in M (the lo al basis
at


P ( ) (see VI.5.6)). Thus, a ording to V.4.21, gik ( )j i; k = 0; 1; 2; 3 is the
oordinatized form of g orresponding to this basis. More pre isely, we get those
formulae hoosing an s 2 I+ and putting
g ()
gik ( ) := ik :
s2

9.5. Exer ises

1. Let U be the observer orresponding to the referen e frame K: Demonstrate
that the oordinatized form of U a ording to K is the onstant mapping (1; 0):
(By denition, (DK
U ) K 1 is the oordinatized form of U a ording to K;
see VI.5).
2. Take the uniformly a elerated observer U treated in 6.5. Fix s 2 I+ and
dene a Lorentz referen e frame with an arbitrary origin o and with a spa etime

basis su h that e0 := sU (o); e1 := s jaaoo j ; e2 and e3 are arbitrary. Demonstrate

that U will have the oordinatized form
( 0 ;  1 ;  2 ;  3 ) 7!

q

1 + ( 0 )2 ;  0 ; 0; 0

where  is the number for whi h jaoj =  s1 holds.

The U -line passing through o +


3
P
 i ei

i=0

be omes

1
1
shs;  1 + ( hs 1);  2 ;  3



s2R

3. Take the uniformly a elerated observer U treated in 6.6. Fix an s 2 I+ and

dene a Lorentz referen e frame with an arbitrary origin o and with a spa etime
basis su h that e0 := sU (o); e1 := s jaaoo j ; e2 and e3 are arbitrary. Demonstrate
that U will have the oordinatized form


( 0 ;  1 ;  2 ;  3 ) 2 R4 j  1 > j 0 j ! R4 ;
1
( 0 ;  1 ;  2 ;  3 ) 7! q
( 1 ;  0 ; 0; 0):
2
2
0
1
( ) + ( )
The U -line passing through o +


3
P
 i ei

i=0

be omes

1 1 1 1
sh s;  + 1 ( h 1 s 1);  2 ;  3
1


s2R :

4. Take the uniformly rotating observer U treated in 6.7. Fix an s 2 I+

and dene a Lorentz referen e frame with o; e0 := sU (o); e3 positively oriented
in Ker
; je3 j = s; e1 and e2 arbitrary. Demonstrate that U will have the
oordinatized form
( 0 ;  1 ;  2 ;  3 ) 7!

q

1 + !2 ( 1 )2 + ( 2 )2 ; ! 2 ; ! 1 ; 0

where ! is the number for whi h j

j = ! s1 holds.
The U -line passing through o +
 q

3
P
 i ei

i=0

s 1 + !2 ( 1 )2 + ( 2 )2 ;  1 os !s

be omes



2
1
2
3
 sin !s;  sin !s +  os !s;  


s2R :

5. Take the uniformly rotating observer U treated in 6.8. Fix an s 2 I+

and dene a Lorentz referen e frame with o; e0 := sU (o); e3 positively oriented
in Ker
; je3 j = s; e1 and e2 arbitrary. Demonstrate that U will have the
oordinatized form
( 0 ;  1 ;  2 ;  3 ) 2 R4 j !2 ( 1 )2 + ( 2 )2 < 1 ! R4 ;
1
( 0 ;  1 ;  2 ;  3 ) 7! q
(1; ! 2; ! 1 ; 0)
2
2
1
2
2
1 ! ( ) + ( )

where ! is the number for whi h j

j = ! s1 holds.
The U -line passing through o +


3
P
 i ei

i=0

be omes

t(s);  1 os !t(s)  2 sin !t(s);  1 sin !t(s) +  2 os !t(s);  3

where

s2R ;

s
:
1 !2 ( 1 )2 + ( 2 )2
6. Find ne essary and su ient onditions that two ane referen e frames
be equivalent.
7. Take the uniformly a elerated observer treated in 6.5, onsider Uo simultaneity and nd a onvenient referen e frame for them.
8. A referen e frame dened for a uniformly a elerated observer annot be
equivalent to a referen e frame dened for a uniformly rotating observer.
t(s) :=

10. Spa etime groups

10.1. The Lorentz group

10.1.1. We shall deal with linear maps from M into M; permanently using
the identi ation Lin(M)  M
M :
Re all the notion of g-adjoints, g-orthogonal maps, g-antisymmetri maps
(V.1.5, V.2.7).

Denition.

L := fL 2 M
M j L
L = idMg = O(g)

is alled the Lorentz group ; its elements are the Lorentz transformations.
If L is a Lorentz transformation then

+1 if L is arrow-preserving
arL :=
1 if L is arrow-reversing

is the arrow of L and

signL :=

+1
1

if LjEu is orientation-preserving
if LjEu is orientation-reversing

is the sign of L where u is an arbitrary element of V(1):

Let us put
L+! := fL 2 Lj signL = arL = 1g ;
L+ := fL 2 Lj signL = arL = 1g ;
L ! := fL 2 Lj signL = arL = 1g ;
L := fL 2 Lj signL = arL = 1g :
L+! is alled the proper Lorentz group.

10.1.2. (i ) From VII.5 we infer that the Lorentz group is a six-dimensional

Lie group having the Lie algebra
La(L) = A(g) = fH 2 M
Mj H =

Hg :

(ii ) S; T and L; the set of spa elike ve tors, the set of timelike ve tors and the
set of lightlike ve tors are invariant under Lorentz transformations. The arrow
of a Lorentz transformation L is +1 if and only if T! ; the set of future-dire ted
timelike ve tors, is invariant for L:
(iii ) The sign of Lorentz transformations is orre tly dened. Indeed, if
u 2 V(1) then L maps Eu onto E(arL)L
u ; these two linear subspa es are oriented
a ording to 1.3.4. It is not hard to see that if the restri tion of L onto Eu is
orientation-preserving for some u then it is orientation-preserving for all u:
(iv ) The mappings L ! f 1; 1g; L 7! arL and L ! f 1; 1g; L 7! signL
are ontinuous group homomorphisms. As a onsequen e, the Lorentz group
is dis onne ted. We shall see in 10.2.4 that the proper Lorentz group L+! is
onne ted. It is quite trivial that if L 2 L+ then L
L+! = L+ and similar
assertions hold for L ! and L as well. Consequently, the Lorentz group has
four onne ted omponents, the four subsets given in Denition 10.1.1.
From these four omponents only L+! | the proper Lorentz group | is a
subgroup; nevertheless, the union of an arbitrary omponent and of the proper
Lorentz group is a subgroup as well.
L! := L+! [ L ! is alled the ortho hronous Lorentz group.
(v ) The arrow of L is +1 if and only if T! ; the set of future-dire ted timelike
ve tors is invariant for L :
if arL = 1 then L[T! = T! ; L[T = T ;
if arL = 1 then L[T! = T ; L[T = T! :

Moreover, the elements of L+! and L

preserve the orientation of
whereas the elements of L+ and L ! reverse the orientation.

M;

10.1.3. M is of even dimensions, thus idM is orientation-preserving. Evidently, idM is in L ; it is alled the inversion of spa etime ve tors. We have
that L = ( idM )
L+! :
We have seen previously that the elements of L+ invert in some sense the
timelike ve tors and do not invert the spa elike ve tors; the elements of L !
invert in some sense the spa elike ve tors and do not invert the timelike ve tors.
However, we annot sele t an element of L+ and an element of L ! that we
ould onsider to be the time inversion and the spa e inversion, respe tively.
For ea h u 2 V(1) we an give a u-timelike inversion and a u-spa elike
inversion as follows.
The u-timelike inversion Tu 2 L+ inverts the ve tors parallel to u and
leaves invariant the spa elike ve tors g-orthogonal to u :
Tu
u :=

In general,

Tu
q

and

:= q

for

2 Eu :

Tu
x = u(u
x) + u
x = 2u(u
x) + x

i.e.

Tu

(x 2 M);

= g + 2u
u

where, as usual, g := idM :

The u-spa elike inversion Pu 2 L ! inverts the spa elike ve tors g-orthogonal
to u and leaves invariant the ve tors parallel to u :
Pu
u := u

and

Pu
q

:= q

for

In general,
Pu
x =

i.e.

u(u
x)

u
x =

2u(u
x) x

2 Eu :

(x 2 M);

= g 2u
u:
We easily dedu e the following equalities:
Pu

Tu 1

Pu 1 = Pu ;
Tu = Pu ;
Tu
Pu = Pu
Tu =
:

= Tu;

10.1.4. For u 2 V(1) let us onsider the Eu lidean ve tor spa e (Eu; I; bu)
where bu is the restri tion of g onto Eu  Eu: The bu -orthogonal group, O(bu );

 alled also the group of u-spa elike orthogonal transformations, an be identied

with a subgroup of the Lorentz group:
O(bu )  fL 2 L! j L
u = ug :
The Lorentz group is an analogue of the Galilean group and we have already
seen a number of their ommon properties. However, as on erns their relation
to three-dimensional orthogonal groups, they dier signi antly.
In the non-relativisti ase there is a single three-dimensional orthogonal
group in question, O(b); and it an be inje ted into the Galilean group in dierent ways a ording to dierent velo ity values. Moreover, L 7! LjE is a surje tive group homomorphism from the Galilean group onto the three-dimensional
orthogonal group.
In the relativisti ase there are a lot of three-dimensional orthogonal groups,
being subgroups of the Lorentz group; one orresponds to ea h velo ity value.
Note that, for all u; L 7! LjEu is not a surje tive group homomorphism from L
onto O(bu ); indeed, Eu is invariant for L if and only if L
u = (arL)u:
As a onsequen e, there is not either a \spe ial Lorentz group" or a \uspe ial Lorentz group" whi h would be the kernel of the group homomorphism
L 7! LjEu :

10.1.5. The problem is that, in general, Eu is not invariant for a Lorentz

transformation L; more losely, L maps Eu onto E(arL)L
u for all u 2 V(1):
Let us try to rule out this uneasiness with the aid of the orresponding Lorentz
boost L(u; (arL)L
u) whi h maps E(arL)L
u onto Eu in a \handsome" way. A
simple al ulation yields the following result.
Proposition. For all Lorentz transformations L and for all u 2 V(1);
R(L; u) := (arL)L(u; (arL)L
u)
L =

= (arL)L +

u + (arL)L
u
(arL)L) 1
u + u
1 u
(arL)L
u

2u
u

is an element of O(bu ):
This suggests the idea that an ortho hronous Lorentz transformation
L should be
onsidered \spe ial" if R(L; u)jEu = idEu ; then
L(u; (arL)L
u)
L = g and onsequently L is a Lorentz boost.
Thus Lorentz boosts an be regarded as ounterparts of spe ial Galilean
transformations. That is why we all them spe ial Lorentz transformations as
well. However, it is very important that the spe ial Lorentz transformations
(Lorentz boosts) do not form a subgroup (see 1.3.9).
Note that our result an be formulated as follows: given an arbitrary u 2 V(1);
every Lorentz transformation L an be de omposed into the produ t of a spe ial

Lorentz transformation and a u-spa elike orthogonal transformation, multiplied

by the arrow of L :
L = (arL)L((arL)L
u; u)
R(L; u):

10.1.6.

It is worth mentioning that the produ t of the u0 -timelike (u0 spa elike) inversion and the u-timelike (u-spa elike) inversion is a spe ial Lorentz
transformation. Sin e
Tu 1

we nd | be ause of u

Pu0
Pu 1

= Tu = Pu = g + 2u
u;
2(u
u0 )u0 = u

is the Lorentz boost from u to u

10.1.7.

0
p12vjuu
vuu0 j
0 :
p12vjuu
v 0j

uu

| that Tu0
Tu 1 =

(i ) Take an u 2 V(1) and a 0 6= H 2 A(g) for whi h H
u = 0

holds. Then H 3 = jH j2 H (V.4.18(i )) and we an repeat the proof of I.11.1.8
to have
!
H2
H
H2
H
sin jH j
e = g+
2 +
2 os jH j +

jH j

jH j

jH j

whi h is an element of O(bu ):

(ii ) Take an u 2 V(1) and a 0 6= H 2 A(g) whose kernel lies in Eu : Then
H 3 = jH j2 H (V.4.18(ii )) and we an prove as in I.11.1.8 that
eH =

H2

g jH j2

H2
H
hjH j +
2
jH j shjH j:
jH j

We an demonstrate this is a Lorentz boost. Re all that there is an n 2 EIu ;

jnj = 1 su h that H = u ^ n; where  := jH j: Then H 2 = 2 (n
n u
u)
and exe uting some al ulations we obtain:
Proposition. Let u 2 V(1); n 2 EIu ; jnj = 1; and  2 R: Then

exp (u ^ n) = L(u h + nsh; u):

10.1.8. Originally the Lorentz tranformations are dened to be linear maps

from M into M: In the usual way, we an onsider them to be linear maps from
M into M as we already did in the pre eding paragraphs as well.
I
I
V(1) is invariant under ortho hronous Lorentz transformations. However,
ontrary to the non-relativisti ase, here V(1) is not an ane subspa e, hen e
we annot say anything similar to those in I.11.3.8.

This, too, indi ates that the stru ture of the Lorentz group is more ompli ated than the stru ture of the Galilean group.

10.2. The u-split Lorentz group

10.2.1.

The Lorentz transformations, being elements of M

M ; are split
by velo ity values a ording to 8.1. These splittings are signi antly more
ompli ated than the splittings of Galilean transformations.
Let us start with the splittings of Lorentz boosts. The map Hu0u dened in
7.1.4 is su h a splitting:
(u; u0 2 V(1)):
Hu0 u = hu
L(u; u0 )
h 1
u

For a u 2 V(1); it is onvenient to introdu e the notations

E
Bu := v 2 u  jvj < 1


and

(v) :=

jvj2

1
(v)2
D (v ) :=
idEu +
v
v
(v)
(v) + 1

for v 2 Bu : It is worth mentioning the relation

(v) 1
(v2 )
:
=
(v) + 1
jv j 2
Applying the usual matrix forms we have
Hu0 u

= (vu0 u)

A simple al ulation yields that

vu0 u

vu0 u
D (vu0 u )


:


1
vu0 u
hu
L(u0 ; u)
hu 1 = Hu01u = (vu0 u )
:
vu0 u D (vu0 u )

10.2.2. Now taking an arbitrary Lorentz transformation L and a u 2 V(1);

we make the following manipulation:
hu
L
hu 1

= hu
L(u; (arL)L
u) 1
hu 1
hu
L(u; (arL)L
u)
L
hu 1 :

The rst fa tor on the right-hand side equals H(ar1L)L
u;u: As on erns the
se ond fa tor, we nd that

hu
L(u; (arL)L
u)
L
hu 1 (t; q ) = hu
L(u; (arL)L
u)
L (ut + q ) =

= hu
(arL)ut + R(L; u)
q = (arL)t; R(L; u)
q

for all (t; q) 2 I  Eu; i.e. the se ond fa tor has the matrix form


arL

R(L; u)

As a onsequen e, we see that the following denition des ribes the u-split
form of Lorentz transformations.
Denition. The u-split Lorentz group is


(v) v1

v
D (v )



0   2 f

1; 1g; v 2 Bu; R 2 O(bu ) :

Its elements are alled u-split Lorentz transformations.

The u-split Lorentz transformations an be regarded as linear maps I  Eu !
I  Eu; the one in the denition makes the orresponden e

(t; q) 7! (v) t + v
R
q; vt + D(v)
R
q :

The u-split Lorentz group is a six-dimensional Lie group having the Lie
algebra

0 v  v 2 Eu ; A 2 A(bu) :
v A

10.2.3. The splitting a ording to u establishes a Lie-group isomorphism

between the Lorentz group and the u-split Lorentz group. The isomorphisms
orresponding to dierent u0 and u are dierent.
The dieren e of splittings an be seen by the usual transformation rule whi h
is rather ompli ated; sin e here we need not it, we do not give the details.
10.2.4. The u-splitting sends the proper Lorentz group onto






1
v
1
0

(v) v D(v)
0 R  v 2 Bu; R 2 SO(bu)
whi h is evidently a onne ted set. Sin e the u-splitting is a Lie group isomorphisms, L+! is onne ted as well.

10.2.5. We easily verify that



(v) v1


v
 v
D (v ) 

2 Bu

is not a subgroup of the u-split Lorentz group; this re e ts the well-known fa t

that the Lorentz boosts do not form a subgroup.

10.2.6. The Lie algebra of the Lorentz group, too, onsists of elements of
M
M; thus they are split by velo ity values in the same way as the Lorentz
transformations; evidently, their split form will be dierent.
If H is in the Lie algebra of the Lorentz group | i.e. H is a g-antisymmetri

tensor | and u 2 V(1); then

hu
H
hu 1

u
= H0
u H H
:
u^H
u


The splitting a ording to u establishes a Lie algebra isomorphism between

the Lie algebra of the Lorentz group and the Lie algebra of the u-split Lorentz
group. The isomorphisms orresponding to dierent u0 and u are dierent.

10.3. Exer ises

The Lorentz group is not transitive, i.e. for all x 2 M; fL
xj

M: What are the orbits of the Lorentz group?
2. The subgroup generated by the Lorentz boosts equals the proper Lorentz
group.
3. Prove that the Lie algebra of O(bu ) equals fH 2 A(g)j H
u = 0g whi h
an be identied with A(bu ):
4. What is the subgroup generated by fTu j u 2 V(1)g?
5. Prove that


1
0
1
00
0
hu
Tu
hu = Hu;u

0 idEu  ;

1
0
hu
Pu0 hu 1 = Hu;u00

0 idEu
where u00 := u 2(vuu0 )vuu0 :
1.

L 2 Lg 6=

10.4. The Poin are group

10.4.1. Now we shall deal with ane maps L : M ! M; as usual, the linear
map under L is denoted by L:

Denition.
P := fL :

M ! Mj L is ane, L 2 Lg

is alled the Poin are group; its elements are the Poin are transformations.
If L is a Poin are transformation then
arL := arL;

signL := signL:

P +! ; P + ; P

! and P are the subsets of P onsisting of elements whose

underlying linear maps belong to L+! ; L+ ; L ! and L ; respe tively.
P +! is alled the proper Poin are group.
A ording to VII.3.2(ii ) we an state the following.
Proposition. The Poin are group is a ten-dimensional Lie group; its Lie
algebra onsists of the ane maps H : M ! M whose underlying linear map is
in the Lie algebra of the Lorentz group:

La(P ) = fH 2 A(M; M)j H 2 A(g)g :

The proper Poin are group is a onne ted subgroup of the Poin are group.
As regards P + ; et . we an repeat what we said about the omponents of the
Lorentz group.
P ! := P +! [ P ! is alled the ortho hronous Poin are group.

10.4.2. We an say that the elements of P invert spa etime in some sense
but there is no element that we ould all the spa etime inversion.
For every o 2 M we an give the o- entered spa etime inversion in the wellknown way ( f. I.11.6.2):
Io (x) := o (x o)
(x 2 M):
Similarly, we an say that in some sense the elements of P ! ontain spa elike
inversion and do not ontain timelike inversion; the elements of P + ontain
timelike inversion and do not ontain spa elike inversion. However, the spa e
inversion and the time inversion do not exist.
For every o 2 M and u 2 V(1) we an give the o- entered u-timelike inversion
and the o- entered u-spa elike inversion as follows:
Tu;o(x) := o + Tu
(x o);
Pu;o (x) := o + Pu
(x o)

(x 2 M):

10.4.3. The Poin are transformations are mappings of spa etime. They
play a fundamental role be ause the proper Poin are transformations an be

 onsidered to be the stri t automorphisms of the spa etime model. The following
statement is quite trivial.
Proposition. (F; idI) is a stri t automorphism of the spe ial relativisti spa e
time model (M; I; g) if and only if F is a proper Poin are transformation.

10.4.4. The Lorentz group is not a subgroup of the Poin are group. The
mapping P ! L; L 7! L is a surje tive Lie group homomorphism whose kernel
is T n(M); the translation group of M;
T n(M) = fTx j x 2 Mg = fL 2 Pj L = idMg :

As we know, its Lie algebra is M regarded as the set of onstant maps from
M into M (VII.3.3).
For every o 2 M;
Lo := fL 2 Pj L(o) = og ;
alled the group of o- entered Lorentz transformations, is a subgroup of the
Poin are group; the restri tion of the homomorphism L 7! L onto Lo is a
bije tion between Lo and L:
In other words, given o 2 M; we an assign to every Lorentz transformation
L the Poin are transformation

x 7! o + L
(x o);
alled the o- entered Lorentz transformation by L:

10.4.5. For every u 2 V(1) we an dene the subgroup of u-timelike transT n(I)u := fTut j t 2 Ig  T n(M)

lations

and the subgroup of u-spa elike translations

T n(Eu ) := fTq j q 2 Eug  T n(M)

10.4.6. For every u 2 V(1) and o 2 M;

O(bu )o := fL 2 P ! j L(o) = o; L
u = ug ;
alled the group of o- entered u-spa elike orthogonal transformations, is a subgroup of P ! :
In other words, given (u; o) 2 V(1)  M; we an assign to every R 2 O(bu )
the Poin are transformation

x 7! o

u u
(x

o) + R
u
(x o);

 alled the o- entered u-spa elike orthogonal transformation by R:

10.5. The ve torial Poin are group

10.5.1. Re all that for an arbitrary world point o; the ve torization of M

with origin o; Oo : M ! M; x 7! x o is an ane bije tion.
With the aid of su h a ve torization we an \ve torize" the Poin are group as
well: if L is a Poin are transformation then Oo LOo 1 is an ane transformation
of M; represented by the matrix (see VI.2.4(ii ) and Exer ise V I.2.5.2)


1
0
L(o) o L :


The Lie algebra of the Poin are group onsists of ane maps H : M ! M
where M is onsidered to be a ve tor spa e (the sum of su h maps is a part
of the Lie algebra stru ture). Thus the ve torization H Oo 1 is an ane map
M ! M where the range is regarded as a ve tor spa e. Then it is represented
by the matrix (see VI.2.4(iii ))


0
0
H (o) H :


10.5.2. Denition. The ve torial Poin are group is




1 0  a 2 M; L 2 L :
a L 
The ve torial Poin are group is a ten-dimensional Lie group, its Lie algebra
is the ve torization of the Lie algebra of the Poin are group:


0  a 2 M; H 2 La(L) :

a H

An advantage of this blo k matrix representation is that the ommutator of

two Lie algebra elements an be omputed as the dieren e of their two produ ts.

10.5.3. A ve torization of the Poin are group is a Lie group isomorphism

between the Poin are group and the ve torial Poin are group. The following
transformation rule shows how the ve torizations depend on the world points
serving as origins of the ve torization. Let o and o0 be two world points; then
Oo0 Oo 1 = To o0 =

0
(o o0 ) idM

and

To o0

1 0

a L

To 1o0 =

1
a + (L idM )
(o0

o)

(a 2 M; L 2 L):

As on erns the orresponding Lie algebra isomorphisms, we have



0 
To 1o0 = 

a H

0
a + H
(o0

o)

0

(a 2 M; H 2 La(N )):

10.6. The u-split Poin are group

10.6.1. With the aid of the splitting orresponding to u 2 V(1); we send

the transformations of M into the transformations of I  Eu : Composing a
ve torization and a splitting, we onvert Poin are transformations into ane
transformations of I  Eu :
Embedding the ane transformations of IEu into the linear transformations
of R  (I  Eu ) (see VI.2.4(ii )) and using the ustomary matrix representation
of su h linear maps, we introdu e the following notion.
Denition. The u-split Poin are group is
0
0
1 0 0 
I; q 2 Eu = :
(v)
(v)v A  0  0 A  2vf2 1B; 1;g;Rt 22 O
(bu ) ;
u
q (v )v (v )D (v )
0 0 R 

80
< 1
t
:

10

The u-split Poin are group is a ten-dimensional Lie group having the Lie
algebra
80
< 0
t
:

0 0 
=
E
0 v A t 2 I; q 2 Eu ; v 2 u ; A 2 A(bu ) :
1

Keep in mind that the group multipli ation of u-split Poin are transformations oin ides with the usual matrix multipli ation and the ommutator of Lie
algebra elements is the dieren e of their two produ ts.
For u 2 V(1) and o 2 M put

hu;o := hu Oo : M ! I  Eu :
Then L 7! hu;o L hu;o1 is a Lie group isomorphism between the Poin are
group and the u-split Poin are group. Evidently, for dierent elements of V(1) 

M; the isomorphisms are dierent. The transformation rule that shows how the
isomorphism depends on (u; o) is rather ompli ated.
Though the Poin are group and the u-split Poin are group are isomorphi
(they have the same Lie group stru ture), they are not \identi al" : there is no
\ anoni al" isomorphism between them that we ould use to identify them.
The u-split Poin are group is the Poin are group of the u-split spe ial relativisti spa etime model (I  Eu; I; gu ) (see 1.7). The spa etime model (M; I; g)
and the orresponding u-split spa etime model are isomorphi , but they annot
be identied as we pointed out in 1.6.3. Due to the additional stru tures of the
u-split spa etime model, the u-split Poin are group has a number of additional
stru tures that the Poin are group has not.

10.6.2. The u-split Poin are group has the following subgroups:
80
9
0 1 9= 8< 0 1 0 0 1
< 1 0
=
t 1
0 1
A t2I ;
A q 2 Eu ;
0
0


:
;
:
;
0 0 idEu 
q 0 idEu 
80
9
1
< 1 0 0 
=
 0 1 0 A R 2 O(bu ) :

:
;
0 0 R 
In ontradistin tion to the non-relativisti ase,
80
< 1

:

0
0


0
0
=

(v)
(v)v A v 2 Bu
;
(v)v (v)D(v) 
1

is not a subgroup of the u-split Poin are group.

The listed u-split Poin are transformations orrespond (by the isomorphism
established by (u; o) 2 V(1)  M) to the following Poin are transformations:

x 7! x + ut
x 7! x + q


x 7! o u u
(x o) + R
u
(x o)


x 7! o + L (v)(u + v); u
u
(x o)

(t 2 I);
(R 2 O(bu ));
(v 2 Bu ):

(q 2 Eu );

10.6.3. Taking a linear bije tion I ! R and an orthogonal linear bije tion
Eu ! R3 ; we an transfer
the u-split Poin are group into an ane transforma3

tion group of R  R ; alled the arithmeti Poin are group whi h is the Poin are
group of the arithmeti spa etime model.

The arithmeti Poin are transformations an be given in the same form as

the u-split Poin are transformations: R; R3 ; O(3) and the open unit ball in R3
are to be substituted for I; Eu; O(bu ) and Bu ; respe tively.
In onventional treatments one always onsiders the arithmeti Poin are group
and one speaks about the time translation subgroup, the spa e translation subgroup, the spa e rotation subgroup, the time inversion et . whi h in appli ations
an result in misunderstandings.
Sin e time and spa e do not exist and only observer time and observer spa e
make sense, the Poin are group has no su h subgroups; it ontains u-timelike
translations, u-spa elike translations, o- entered u-spa elike rotations et .

10.7. Exer ises

1. Let L be a Poin are transformation for whi h L = idM : Then there is a
unique o 2 M su h that L is the o- entered spa etime inversion.
2. Prove that for all o 2 M;

Oo Lo Oo 1 =



0  L 2 L :

3. Find hu;o
Tu;o
hu;o1 and hu;o
Pu;o
hu;o1 :

4. Prove that the subgroup generated by fTu;oj u 2 V(1); o 2 Mg equals
P +! [ P + :
5. Prove that the derived Lie algebra of the Poin are group equals the Lie
algebra of the Poin are group, i.e. [La(P ); La(P ) = La(P ):
6. Let L be a Poin are transformation. Consider the real number arL to be
a linear map I ! I; t 7! (arL)t:
If r is a world line fun tion then L r (arL) 1 is a world line fun tion, too.
If C is a world line then L
[C is a world line, too; moreover, if C = Ran r then
L[C = Ran L r (arL) 1 :

11. Relation between the spe ial relativisti spa etime

model and the non-relativisti spa etime model

11.1. One often asserts that non-relativisti physi s is the limit of spe ial
relativisti physi s as the light speed tends to innity.
Can we give su h an exa t statement on erning our spa etime models? The
answer is no.
We have two dierent mathemati al stru tures. There is no natural way of
introdu ing a onvergen e notion even on a lass of mathemati al stru tures of
the same kind (e.g. on the lass of groups and to say that a sequen e of groups

 onverges to a given one) and it is quite impossible to introdu e a onvergen e

notion on a lass onsisting of stru tures of dierent kinds (e.g. to say that a
sequen e of groups onverges to an algebra).
There is no reasonable limit pro edure in whi h a sequen e of spe ial relativisti spa etime models onverges to a non-relativisti spa etime model.

11.2. The following onsiderations show the real meaning of the usual statements.
Let us x a spe ial relativisti global inertial observer with the velo ity
value u:
Let us rename I to D; alling it the measure line of distan es. Let us introdu e
for time periods a new measure line, denoted by I: Let us hoose a positive
element of DI ; it makes the orresponden e I ! D; t 7! t:
If u0 2 V(1) then vu0 u 2 EDu and v := vu0 u 2 EIu will be onsidered the
relative velo ity with respe t to the observer. Evidently, jvj < ; thus is the
light speed in the new system of measure lines.
Substituting v for vu0 u and t for t in the formula 7.1.4 and letting tend to
innity | whi h has an exa t meaning be ause elements of nite dimensional
ve tor spa es are involved in that formula | we get the orresponding nonrelativisti transformation rule in I.8.2.4.
Similar statements hold for other formulae that on ern relative velo ities; e.g.
for the addition formula of relative velo ities, for the formula of light aberration
et .
However, su h a statement, in general, will not be valid for formulae that do
not on ern relative velo ities: e.g. the uniformly a elerated observer treated
in 6.6 has no limit as tends to innity.

III. FUNDAMENTAL NOTIONS OF

GENERAL RELATIVISTIC SPACETIME
MODELS

1. As we have already mentioned, the non-relativisti spa etime model is

suitable for des ribing \sluggish" me hani al phenomena. To des ribe \brisk"
me hani al phenomena and ele tromagneti phenomena we have to use the
spe ial relativisti spa etime model. Of ourse, the spe ial relativisti spa etime
model is good for \sluggish" me hani al phenomena, too, but their relativisti
des ription is mu h more ompli ated than the non-relativisti one and gives
pra ti ally the same results.
To avoid misunderstandings, we emphasize that the me hani al ee ts of ele tromagneti phenomena (e.g. the history of harged masspoints in a given ele tromagneti eld) an be well des ribed non-relativisti ally as well, provided
that the me hani al phenomena remain \sluggish" (the relative velo ities of
masspoints remain mu h smaller than the light speed). The non-relativisti
spa etime model is not suitable for the des ription of the ele tromagneti phenomena in va uum: how do harges produ e ele tromagneti eld, how does an
ele tromagneti radiation propagate et .
Gravitational a tions are well des ribed in the non-relativisti spa etime
model by absolute s alar potentials. Su h potentials do not exist in the spe ial relativisti spa etime model. Other potentials or for e elds do not give
onvenient (experimentally veried) models of gravitational a tions.
The problem that fa es us is that gravitational a tions in \brisk" me hani al
phenomena and ele tromagneti phenomena annot be des ribed and, of ourse,
gravitational phenomena (how do masses produ e gravitational elds) annot be
treated in the framework of the spe ial relativisti spa etime model.
There is only one way out: if we want to des ribe gravitational phenomena
as well, then we have to onstru t a new spa etime model. However, it is not
straightforward at all, how we shall do this.
2. Re all what we said about our experien e regarding the stru ture of
our spa e and time: in our spa e we nd straight lines represented by light
signals or stret hed threads. We know, however, that a thread stret hed in the
gravitational eld of the earth is not straight, it bends; if the thread is short
enough and the stret hing is strong enough then the urvature of the thread is

negligible. However, for longer threads | imagine a thread (wire) a ross a river
| the urvature an be signi ant.
It seems, that a light signal is better for realizing a straight line. Indeed,
in terrestrial distan es we do not experien e that a light signal is not straight.
However, the distan es on the earth are small for a light signal. It may happen
that light signals turn out to be urved in osmi distan es. Of ourse, to prove
or disprove this possibility we meet great di ulties. A minor problem is that
osmi distan es are hardly manageable.
To state that a line is straight or not we have to know what the straight lines
are. Straight lines in terrestrial distan es are dened in the most onvenient way
by the traje tories of light signals. Have we a better way to dene straight lines
in osmi size? Can we dene straight lines in this way? Can we dene straight
lines at all?
We have to re ognize that it makes no sense that a single line in itself is straight
or is not straight. We have to relate the traje tories of more light signals and
to test whether they satisfy the onditions we expe t the set of straight lines
have. For instan e, if two dierent light signals meet in more than one point,
the traje tories of the signals annot be straight lines. Unfortunately, it is rather
di ult to exe ute su h examinations in osmi size.
Nevertheless, we have experimental eviden e that shows that gravitation in uen es the propagation of light. The angle between two light beams arriving from
two stars have been measured in dierent ir umstan es: rst the light beams
travel \freely", far from gravitational a tion; se ond, they travel near the Sun
i.e. under a strong gravitational a tion. The angles are signi antly dierent.
Light travels along dierent traje tories in two ir umstan es. Evidently, the
traje tories annot be straight lines in both ases.

The ane stru ture of spa etime in the spe ial relativisti model has been
based on the straight propagation of light. Thus if we want to onstru t a
spa etime model suitable for the treatment of gravitational phenomena, we have
to reje t the ane stru ture.
We have to get a ustomed to the strange fa t: in general, the notion of
a straight line makes no sense. It is worth repeating why. Every notion in
our mathemati al model must have a physi al ba kground. A straight line
would be realized by a light beam: we have no better possibility. However,
in strong gravitational elds (in osmi size) the set of light beams maybe
does not satisfy the usual onditions imposed on the set of straight lines. One
usually says that gravitation \ urves" spa etime. The properties of a urved
spa etime an be illustrated as follows: it may happen that two light beams
starting simultaneously from the same sour e in dierent dire tions meet again
somewhere (this is a \spa elike urvature") or that two light beams starting from
the same sour e in the same dire tion in dierent instants meet again somewhere
(this is a \timelike urvature").

3. A ording to the idea of Einstein, spa etime models must re e t gravitational a tions, a spa etime model is to be a model of a gravitational a tion ; the
absen e of gravitation is modelled by the spe ial relativisti spa etime model.
The theory of gravitation, a deep and large theory, lies out of the s ope of this
book. That is why only the framework of general relativisti spa etime models
will be outlined.
In onstru ting a general relativisti spa etime model, we do not adhere to
the ane stru ture and we require only that spa etime is a four-dimensional
smooth manifold M:
A four-dimensional smooth manifold M is an abstra t mathemati al stru ture
similar to a four-dimensional smooth submanifold in an ane spa e; it has the
following fundamental properties: every x 2 M has a neighbourhood whi h an
be parametrized by p : R4 M; if p and q are parametrizations then q 1 p is
smooth. Then to ea h point x of M a four-dimensional ve tor spa e Tx(M); the
tangent spa e at x; is assigned; every dierentiable urve passing through x has
its tangent ve tor in Tx (M): A neighbourhood of zero of Tx (M) approximates a
neighbourhood of x in M: Smooth submanifolds of an ane spa e (thus ane
spa es themselves) are smooth manifolds.
Our experien e that gravitational a tion in small size does not ontradi t the
notion of a straight line suggests that a general relativisti spa etime model
in small size an be \similar" to a spe ial relativisti spa etime model. That
is why we a ept that there is a measure line I; and a Lorentz form gx :
Tx (M)  Tx(M) ! I
I is given for all x 2 M in su h a way that x 7! gx
is smooth in a onveniently dened sense. The assignment x 7! gx is alled a
Lorentz eld and is denoted by g: Moreover, we assume that every gx is endowed

with an arrow orientation whi h, too, depends on x in a onveniently dened

smooth way.
Denition. A general relativisti spa etime model is a triplet (M; I; g) where
| M is a four-dimensional smooth manifold ( alled spa etime or world ),
| I is a one-dimensional oriented ve tor spa e (the measure line of spa etime
lengths),
| g is an arrow-oriented Lorentz eld on M:
Evidently, a spe ial relativisti spa etime model is a general relativisti spa etime model: M is an ane spa e (then every tangent spa e equals M) and gx is
the same for all x 2 M:

4. Take a general relativisti spa etime model (M; I; g): Then Sx; Tx and Lx;
the set of spa elike tangent ve tors et . in Tx (M) are dened by gx for all world
points x and they have the following meaning:
| a world line (the history of a masspoint) is a urve in M whose tangent
ve tors are timelike (i.e. the tangent ve tor of a world line C at x is in Tx);
| a light signal is a urve in M whose tangent ve tors are lightlike.
Let us give an illustration of a general relativisti spa etime model. Let the
plane of the page represent the spa etime M; and at the same time, every tangent
spa e is represented by the plane of the page as well. Then we draw the future
light one to every world point.

Illustrating the non-relativisti and the spe ial relativisti spa etime models
we have got a ustomed to the fa t that the Eu lidean stru ture of the plane
has to be negle ted: the angles and distan es in the plane of the page do not
re e t, in general, obje ts of the spa etime model. Now we have to negle t the

ane stru ture of the plane as well: the straight lines of the plane, in general,
do not orrespond to obje ts of the spa etime.
We all attention to the fa t that in our illustration the spa etime manifold
and its tangent spa es whi h are dierent sets, are represented by the same
plane. The straight lines representing light ones in the previous gure are lines
in tangent spa es, they do not lie in the spa etime manifold.
The following gures show a world line and a light signal in the general
relativisti spa etime model.

5. As we have said, a general relativisti spa etime model is to be a model of

a gravitational a tion. The theory of gravitation has the task to expound how a
gravitational a tion is modelled by a spa etime model. We know that a spe ial
relativisti spa etime model orresponds to the la k of gravitation.
There are dierent spe ial relativisti spa etime models; however, all of them
orrespond to the same physi al situation: the la k of gravitation. This is
re e ted in the fa t that all spe ial relativisti spa etime models are isomorphi .
It may happen that two general relativisti spa etime models orrespond to
the same gravitational a tion; we expe t that they must be isomorphi . Now we
give the notion of isomorphism.
Denition. The general relativisti spa etime model (M; I; g) is isomorphi
to (M0 ; I0 ; g0 ) if there are
| a dieomorphism F : M ! M0 ;
| an orientation preserving linear bije tion Z : I ! I0
su h that
g0 DF (x)  DF (x)
= (Z
Z ) gx
(x 2 M):
F (x)

The phrase F is dieomorphism means that F is a bije tion and both F and
F 1 are smooth. The derivative of F at x; DF (x); is a linear map from Tx(M)
into TF (x) (M0 ):

6. As examples we give a ertain kind of general relativisti spa etime models

where the spa etime manifold is a submanifold of an ane spa e, hen e we an
use the well-known mathemati al tools treated in this book.
Take a spe ial relativisti spa etime model (M; I; g); sele t an open subset
MA of M; MA is an open submanifold of M and Tx (MA ) = M for all x 2 M:
Give a smooth map A : M G `(M) (i.e. A(x) is a linear bije tion M ! M
for all x 2 M): For all x 2 MA we dene the Lorentz form gxA by

gxA(x; y) := g A(x)
x; A(x)
y

(x; y 2 M):
The Lorentz form gA is endowed with an arrow orientation as follows: let T!
be the future-dire ted timelike one of g; then the future-dire ted timelike one
of gxA is dened to be A(x) 1 [T! :
Then (MA ; I; gA ) is a general relativisti spa etime model.

PART TWO

MATHEMATICAL TOOLS

IV. TENSORIAL OPERATIONS

In this se tion K denotes the eld of omplex numbers or the eld of real
numbers, and all ve tor spa es are given over K :
Tensors and operations with tensors are essential mathemati al tools in
physi s; the simplest physi al notions | e.g. meter/se undum | require tensorial operations. Those being familiar with tensors will nd no di ulty in
reading this book.

0. Identi ations
Identi ations make easy to handle tensors.
Let X and Y be ve tor spa es over the same eld. If there is a linear inje tion
i : X ! Y whi h we nd natural (" anoni al") from some point of view, we
identify x and i(x) for all x 2 X; i.e. we omit i from the notations onsidering
X to be a linear subspa e of Y: Then we write

X ! Y;
and if i is a bije tion,

X  Y;

x  (x);

x  (x):

In pra ti e, instead of x  i(x) an appropriate formula appears that allows

us to onsider i to be natural.
Of ourse, \natural" and \ anoni al" are not mathemati al notions and it
depends on us whether we a ept or reje t an identi ation. There are ommonly
a epted identi ations and there are some ases in whi h some people nd a
given identi ation onvenient and others do not.
Later, using a lot of identi ations, the reader will have the opportunity to
see their importan e.

1. Duality

1.1. Let V and U be ve tor spa es. Then Lin(V; U) denotes the ve tor spa e
of linear maps V ! U; Lin(V) := Lin(V; V):
The value of L 2 Lin(V; U) at v 2 V is denoted by L
v:
The omposition of linear maps is denoted by a dot as well: for L 2 Lin(V; U);
K 2 Lin(U; W) we write K
L:
V := Lin(V; K ) is the dual of V: The elements of V are often alled linear
fun tionals or ove tors.
The dual separates the elements of the ve tor spa e whi h means that if v 2 V;
and p
v = 0 for all p 2 V ; then v = 0 or, equivalently, if v1 and v2 are dierent
elements of V; then there is a p 2 V su h that p
v1 6= p
v2 :
If fvi j i 2 I g is a basis of V then there is a set fpi j i 2 I g in V ; alled the
dual of the basis, su h that
pi
v

if i = j
if i 6= j

1
0

(i; j 2 I ):

If V is nite dimensional, then the dual of a basis is a basis in V ; hen e

dim(V ) = dim V:
Let N denote the (nite) dimension of V: If fv1; : : : ; vN g is a basis of V and
fp1 ;


; pN g is its dual, then for all v 2 V and p 2 V we have
v

p=

N
X
i=1

N
X
i=1

(pi
v)vi ;

(p
vi )pi :

1.2. To every element v of V we an asso iate a linear mapi(v) : V ! K ;

p 7! p
v ; i.e. an element of V : The orresponden e V ! V ; v 7! i(v ) is a
linear inje tion whi h seems so natural and simple that we nd it onvenient to
identify v and i(v) for all v 2 V :
V ! V; v  i(v);
i.e.

(v 2 V; p 2 V ):
If V is nite dimensional then this orresponden e is a linear bije tion between
V and V; i.e. the whole dual of V an be identied with V :
V  V; v
p  p
v:
v
pp
v

1.3. The Cartesian produ t V  U of the ve tor spa es V and U is a ve tor

spa e with the pointwise addition and pointwise multipli ation by numbers:
(v1 ; u1) + (v2 ; u2) :=(v1 + v2 ; u1 + u2 );
(v; u) :=(v; u)
for v1 ; v2 2 V; u1 ; u2 2 U and  2 K :
We have the identi ation

V  U  (V  U) ; (p; q)
(v; u)  p
v + q
u:

((p; q) 2 V  U ;
(v; u) 2 V  U) :
1.4. The transpose of L 2 Lin(V; U) is the linear map
L : U ! V ;
f 7! f L;
i.e.

(L
f )
v = f
(L
v)

or, with the identi ation introdu ed in 1.2,

v
L
f

=f
L
v

If L; K 2 Lin(V; U);  2 K ; then

(f

2 U ; v 2 V):

(L + K ) =L + K  ;
(L) =L :
If L 2 Lin(V; U); K 2 Lin(U; W); then
(K
L) = L
K  :
If V and U are nite dimensional, then
| L is inje tive if and only if L is surje tive,
| L is surje tive if and only if L is inje tive.
Moreover, in this ase | be ause of the identi ation V  V; U  U |
we have
L = L:
If L is bije tive, then

(L 1 ) = (L ) 1 :

1.5. Let V be a nite dimensional

ve tor spa e and L 2 Lin(V; V ): Then


L is a linear map from V into V ; i.e. be ause of the identi ation V  V
we have L 2 Lin(V; V ):
The linear map L : V ! V is alled symmetri or antisymmetri if L = L
or L = L ; respe tively.
In general, the symmetri and antisymmetri parts of L 2 Lin(V; V ) are
L + L

and

L

respe tively.
Similar denitions work well for linear maps V ! V:
On the other hand, the notions of symmetri ity, symmetri part et . make no
sense for linear maps V ! V and V ! V :

1.6. K N ; the set of ordered N -tuples of numbers,N is a well-known

ve tor spa e.
M

It is known as well that the linear maps from K into K are identied with
the matri es of M rows and N olumns, in other words, Lin(K N ; K M )  K M N :
As a onsequen e, we have the identi ation
KN



= Lin(K N ; K )  K 1N = K N

p
x

N
X
i=1

pi xi

(p; x 2 K N ):

We
 adhered to the tri k used in physi al appli ations a ording to whi h
is identied with K N in su h a way that they are distinguished in notations as follows.
The omponents of the elements of K N are indexed by supers ripts:
KN

x = (x1 ; : : : ; xN ) 2 K N ;

and the omponents of the elements of

KN



 KN

p = (p1 ; : : : ; pN ) 2 K N



are indexed by subs ripts:

The identi ation in question, alled the standard identi ation,

means that


to every (x1 ; : : : ; xN ) 2 K N we assign (x1 ; : : : ; xN ) 2 K N  in su h a way that
xi = xi for all i = 1; : : : ; N:
Moreover, for the sake of simpli ity, we often shall not write that the indi es
run from 1 to N (or to M ); denoting the elements in the form (xi ) and (xi );
respe tively.

The fundamental rule is that a summation an be arried out only for indi es
in opposite positions: up and down. A ordingly, the matrix entries are indexed
orresponding to the domain and range of the matrix as a linear map:

Li k : K N ! K M ;


(Lik ) : K N ! K M ;






Li k : K N ! K M ;




Lik : K N  ! K M :
This tri k works well until a tual ve tors are not involved; this notation does
not show for
instan e whether the ordered pair of numbers (1; 2) is an element
of R2 or R2  ; and whether the matrix


1 0
2 1

maps from R2 into R2 or from R2 into R2  et .
The set of ve tors 1 := (1; 0; : : : ; 0); 2 := (0; 1; :::; 0); : : : ; N := (0;
0; : : : ; 1)
is alled the standard basis of K N : In the mentioned identi ation K N   K N
the dual of the standard basis is the standard basis itself.
A ording to this identi ation the transpose of a matrix as a linear map is
the usual matrix transpose.
The above notation shows well that
symmetri ity, symmetri part et . make
sense only for matri es (Lik ) and Lik :

1.7. The symbol Bilin(U  V; K ) stands for the ve tor spa e of bilinear maps
U  V ! K ; often alled bilinear forms.
We have that
i : Lin(V; U) ! Bilin(U  V; K )
dened by

(i(L)) (f ; v) := f
L
v
(L 2 Lin(V; U) ; f 2 U ; v 2 V)

is a linear inje tion whi h we use for the identi ation

Lin(V; U) ! Bilin(U  V; K );

f
L
v

 L(f ; v):

If the ve tor spa es U and V have nite dimension then i is a bije tion, hen e

 stands instead of ! :

The reader is asked to examine this identi ation in the ase of matri es i.e.
for Lin(K N ; K M ):

1.8. A bilinear form b : V  V ! K is alled symmetri or antisymmetri if

b(v; u) = b(u; v) or b(v; u) = b(u; v); respe tively, for all
v ; u 2 V:
Similar denitions are a epted for bilinear forms V  V ! K :
Observe that for nite dimensional V the notions introdu ed here and in 1.5
oin ide in the identi ation Lin(V; V )  Bilin(V  V ; K ):
2. Coordinatization

2.1. Let V be an N -dimensional ve tor spa e over K :

An element (v1 ; : : : ; vN ) of VN is alled an ordered basis of V if the set
fv1 ; : : : ; vN g is a basis of V:
An ordered basis of V indu es a linear bije tion K : V ! K N dened
by K
vi := i (i = 1; : : : ; N ) where (1 ; : : : N ) is the ordered standard
basis of K N : K is alled the oordinatization of V orresponding to the given
ordered basis. The inverse of the oordinatization, P := K 1 ; is alled the
parametrization of V orresponding to the given ordered basis. It is quite evident
that
P

Thus, in view of 1.1 we have

K
v

(xi ) =

N
X
i=1

(xi ) 2 K N :

xi vi

= pi
vj i = 1; : : : ; N

(v 2 V)

where (p1 ; : : : ; pN ) is the ordered dual basis of (v1 ; : : : ; vN ):

Obviously, every linear bije tion K : V ! K N is a oordinatization in the
above sense: the one orresponding to the ordered basis (v1 ; : : : ; vN ) where
vi := K 1
i (i = 1; : : : ; N ):

2.2.

A oordinatization of V determines a oordinatization of V ; that is

indu ed by the orresponding ordered dual basis. Using the previous notations
and denoting the oordinatization in question by C : V ! (K N ) we have
C
p = (p
vi j i = 1; : : : ; N )
(p 2 V ):
It is not hard to see that
C

= K 1  = P :

2.3. In the oordinatization K ; a linear map L : V ! V is represented by

the matrix
K
L
K 1

= K
L
P = pi
L
vk j i; k = 1; : : : ; N :

To dedu e this equality argue as follows:

N
X
k=1

i
K
L
K 1 k xk

= K
L
K 1
x i =

=pi
L

N
X
k=1

xk vk =

N
X
k=1

(pi
L
vk )xk :

The linear map T : V ! V is represented by the matrix


K 1 T
K 1 = P 
T
P = (v
T
v j i; k = 1; : : : ; N ) :
i

It is left to the reader to nd the matrix of linear maps V ! V and V ! V :

3. Tensor produ ts

3.1. We start with an abstra t denition of tensor produ ts that may seem
strange; the properties of tensor produ ts following from this denition will
larify its real meaning.
Denition. Let V and U be ve tor spa es (over the same eld K ): A tensor
produ t of U and V is a pair (Z; b); where
(i ) Z is a ve tor spa e,
(ii ) b : U  V ! Z is a bilinear map having the property that
| if W is a ve tor spa e and : U  V ! W is a bilinear map,
| then there exists a unique linear map L : Z ! W su h that
= L b:

Proposition. The pair (Z; b) satisfying (i ) and (ii ) is a tensor produ t of U

and V if and only if
1) Z is spanned (Z is the linear subspa e generated) by Ran b;
2) if v1 ; : : : :; vn are linearly independent elements of V and
n
P
b(ui ; vi) = 0 implies u1 =


= un = 0:
u1 ; : : : :; un are elements of U then
i=1

Proof. Ex lude the trivial ases V = 0 or U = 0:

Suppose 1) is fullled. Then every element of Z is of the form

k b(uk ; vk ): Sin e b(u; v) = b(u; v); we on lude that the elements of Z

r
P

k=1

an be written in the form

r
P

k=1

b(uk ; vk ):

Suppose 2) is fullled, too. Take a bilinear map : U  V ! W and dene

the map L : Z ! W by
L

r
X
k=1

b(uk ; vk )

:=

r
X
k=1

(uk ; vk ):

If L is well-dened, then it is linear, L b = ; and it is unique with this

property. To demonstrate that L is well-dened, we have to show that
r
X
k=1

b(uk ; vk ) = X b(xj ; yj )
j =1

implies

r
X
k=1

(uk ; vk ) = X (xj ; yj );
j =1

whi h is eqivalent to
m
X
i=1

b(ui; vi ) = 0

implies

m
X
i=1

(ui ; vi ) = 0:

Let us hoose a largest set of linearly independent ve tors from fv1 ; : : : ; vm g;

without loss of generality, we an suppose it is fv1 ; : : : ; vn g (where, of ourse,
n
n
P
P
n  m): If v = i vi then b(u; v) = b(i u; vi ) and a similar formula holds
i=1
i=1
for (u; v) as well. Consequently, a rearrangement of the terms in the previous
formulae yields that L is well-dened if
n
X
i=1

b(ui ; vi) = 0

implies

n
X
i=1

(ui ; vi) = 0

whenever v1 ; : : : :; vn are linearly independent whi h follows from ondition 2).

We have proved that onditions 1) and 2) are su ient for a tensor produ t.
Sin e L b = r an dene L only on the linear subspa e spanned by the range
of b; ondition 1) is ne essary for the uniqueness of L:
If ondition 2) is not satised then we an nd a bilinear map r su h that
L b 6= r for all linear maps L: Indeed, let the ve tors v1 ; : : : ; vn be linearly
n
P
independent, b(ui ; vi ) = 0; and at least one of the ui -s is not zero. Without
i=1
loss of generality we an assume that u1; : : : ; um (where m  n) are linearly
independent and all the other ui -s are their linear ombinations. Complete
fv1 ; : : : ; vn g to a basis in V and fu1 ; : : : ; um g to a basis in U: Dene the bilinear
map r : U  V ! K in su h a way that r(u1 ; v1 ) := 1 and r(u; v) := 0 for all
other
 nbasis elements
 u and v : nThen for all linear maps L : Z ! K we have
P
L

b(ui ; vi) = 0 6= 1 = P r(ui; vi ):
i=1

i=1

3.2. In the next item the existen e of tensor produ ts will be proved. Observe
that in the ase W = Z; = b; the identity map of Z fulls b = idZ b;
a ording to the denition of the tensor produ t this is the only possibility, i.e.
if L 2 Lin(Z) and b = L b then L = idZ :
As a onsequen e, if (Z0 ; b0 ) is another tensor produ t of U and V then there
is a unique linear bije tion L : Z ! Z0 su h that b0 = L b: This means that
the tensor produ ts of U and V are \ anoni ally isomorphi " or \essentially the
same", hen e we speak of the tensor produ t and applying a ustomary abuse
of language we all the orresponding ve tor spa e the tensor produ t (Z in the
denition) denoting it by U
V; and writing
U  V ! U
V; (u; v) 7! u
v
for the orresponding bilinear map (b in the denition); u
v is alled the tensor

produ t of u and v:
An a tual given tensor produ t is alled a realization of the tensor produ t
and the following symbols are used: U
V ! W or U
V  W denote that
the tensor produ t of U and V is realized as a subspa e of W or as the whole
ve tor spa e W; respe tively.
It is worth repeating the results of the previous paragraph in the new notations.
n
P
Every element of U
V an be written in the form ui
vi where v1 ; : : : :vn
i=1
are linearly independent ve tors in V: Moreover, if the sum is zero, then u1 =



= un = 0: In parti ular, if u 6= 0 and v 6= 0 then u
v 6= 0:

3.3. For u 2 U and v 2 V we dene the linear map

u
v : V  ! U;
p 7! (p
v )u:

Proposition. U  V ! Lin(V ; U); (u; v) 7! u

v is a bilinear map

satisfying ondition 2) of Proposition 2.1. As a onsequen e, the linear map

is the tensor produ t of u and v (that is why we used in advan e this
notation) and U
V is realized as a linear subspa e of Lin(V ; U) spanned by
su h elements.
Proof. It is trivial that (u; v) 7! u
v is bilinear.
n
P
Suppose that v1 ; : : : :; vn are linearly independent ve tors in V and ui

i=1
vi = 0: Then for arbitrary p 2 V and f 2 U we have
u
v

0=f

n
X
i=1

ui
vi

p

n
X

n
X

i=1

i=1

(f
ui )(p
vi ) = p

(f
ui )vi :

Sin e V separates the elements of V; this means that

(f
ui )vi = 0: Be ause of the linear independen e of vi -s this involves
i=1
f
ui = 0 for all i = 1; : : : ; n: Sin e U separates the elements of U; it follows that u1 = u2 =


= un = 0:
n
P

3.4. Proposition. If fvij i 2 I g is a basis in V and fuj j j 2 J g is a basis in

U then fuj
vij j 2 J; i 2 I g is a basis in U
V:
A ording to Propositions 3.3 and 1.7 we have

U
V ! Lin(V ; U) ! Bilin(U  V ; K ):

If U and V are nite dimensional then

dim(U
V) = (dim U)(dim V):

Moreover, in this ase dim(U

V) = dim (Lin(V ; U)) ; hen e the present
proposition on the bases implies that for nite dimensional ve tor spa es

U
V  Lin(V ; U)  Bilin(U  V ; K )
and be ause of V  V; U  U;
U
V  Lin(V; U)  Bilin(U  V; K );
U
V  Lin(V ; U )  Bilin(U  V ; K );
U
V  Lin(V; U )  Bilin(U  V; K ):
3.5. We have the following identi ations.
(i )
K
V  V;

v  v ;
(ii )

(U  V)
W  (U
W)  (V
W; )
(u; v)
w  (u
w; v
w);

W
(U  V)  (W
U)  (W
V);
w
(u; v )  (w
u; w
v );

(iii ) If U and V are nite dimensional then

U
V  (U
V); (f
p) : (u
v)  (f
u)(p
v);
(f 2 U ; p 2 V ; u 2 U; v 2 V)

where we found onvenient to write the symbol : for the bilinear map of duality;
we shall give an explanation later.

3.6. In mathemati al books the tensor produ t is often said to be ommutative whi h means that we have a unique linear bije tion U
V ! V
U;
u
v 7! v
u admitting an identi ation. However, we do not nd onvenient
to use this identi ation be ause of two reasons:
1) if U = V; u; v 2 V and u 6= v then, in general, u
v 6= v
u;
2) u
v 2 U
V ! Lin(V ; U); v
u 2 V
U ! Lin(U ; V) 
Lin(U ; V ); it is not hard to see that the transpose of u
v equals v
u :
(u
v) = v
u:
Hen e the unique linear bije tion between U
V and V
U that sends u
v
into v
u is the transposing map. We do not want, in general, to identify a
linear map with its transpose (e.g. a matrix with its transpose).
However, if one of the ve tor spa es is one-dimensional, we a ept the mentioned identi ation, i.e.
A
V  V
A; a
v  v
a if dim A = 1:
Moreover, in this ase we agree to omit the symbol
:
(a 2 A; v 2 V; dim A = 1):
Note that if dim A = 1 then every element of A
V has the form av:
Though, in general, A
V 6= V; it makes sense (if dim A = 1) that an element
z of A
V is parallel to an element v of V : if there is an a 2 A su h that
z = av :
av := a
v

3.7. It is well known that a linear map L :

represented in a matrix form:
L=

L11
L21

L12
L22

V1  V2 ! U1  U2 an be

where Lik 2 Lin(Vi ; Uk ) (i; k = 1; 2) and

L
(v1 ; v2 ) = (L11
v1 + L12
v2 ; L21
v1 + L22
v2 ) :
This orresponds to the nite dimensional identi ations (see in parti ular
3.5(ii))
Lin(V1  V2 ;U1  U2 ) 
(U1  U2 )
(V1  V2 )  (U1  U2 )
(V1  V2 ) 
(U1
V1 )  (U1
V2 )  (U2
V1 )  (U2
V2 ) 
Lin(V1 ; U1 )  Lin(V2 ; U1 )  Lin(V1 ; U2 )  Lin(V2 ; U2 ):

A ordingly, we nd onvenient to write



(u1 ; u2 )
(p1 ; p2 ) 

u1
p1
u2
p1

u1
p2
u2
p2

for (u1 ; u2 ) 2 (U1 ; U2 ) and (p1 ; p2 ) 2 V1  V2 :

Of ourse, a similar formula holds for other tensor produ ts, e.g. for the
elements of (U1  U2 )
(V1  V2 ) :


u1
v1 u1
v2
(u1 ; u2)
(v1 ; v2 )  u

v u
v :

It is not hard to see then ( f. 3.6) that



u1
v1
u2
v1

u1
v2
u2
v2



v1
u1
v2
u1

v1
u2
v2
u2

3.8. If A is a one-dimensional ve tor spa e then Lin(A) is identied with K :

the number  orresponds to the linear map a 7! a: As a onsequen e, we have
the following identi ation, too:
A
A  Lin(A)  K ;

ah  h
a( a
h)

(remember: ah := a
h): Indeed, by denition, ah : A ! A; b 7! (h
b)a: If
b
0 then ahb= 0 = h
a: If a 6= 0 thenb there is a unique
a 2 K for all b 2 A
b
su h that b = a a: Thus (h
b)a = (h
a a)a = (h
a) a a = (h
a)b and we see
that ah (= a
h) equals the multipli ation by h
a:
For one-dimensional ve tor spa es we prefer the symbol of (tensor) produ t
to the dot for expressing the bilinear map of duality i.e. the symbol ah to a
h:

a=

3.9.

Sin e V  V ! K ; (v; p) 7! p
v is a bilinear map, the denition of

tensor produ ts ensures the existen e of a unique linear map
Tr : V
V ! K

su h that

Tr(v
p) = p
v:

If V is nite dimensional then V

V  Lin(V); hen e TrL; alled the tra e
of L; is dened for all linear maps L : V ! V:
Sin e for u; v 2 V and p; q 2 V we have (u
p)
(v
q) = (p
v)u
q; we
easily dedu e that for all L; K 2 Lin(V) (if dim V < 1)
Tr(L
K ) = Tr(K
L):

If fvi j i = 1; : : : ; N g is a basis in V and fpi j i = 1; : : : ; N g is its dual then for

all v 2 V and p 2 V
p
v

N
X

N
X

i=1

i=1

(pi
v)(p
vi ) =

whi h gives
TrL =

N
X
i=1

pi
(v
p)
vi ;

pi
L
vi

(L 2 Lin(V)):

Note that the tra e of linear maps V ! V and V ! V makes no sense; on

the other hand, we have (for nite dimensional V)
Tr : Lin(V )  V
V ! K ;
p
v 7! p
v

and we easily see by (v

p) = p
v that
Tr(L ) = TrL

(L 2 Lin(V)):

Moreover, if Z is a nite dimensional ve tor spa e, we dene

Tr : Lin(V; Z
V)  Z
V
V ! Z;
z
v
p 7! (p
v )z :

3.10. Let V be nite dimensional.

Then, a ording to 3.5(iii ) and V  V;


0
we have V
V  (V
V ) ; (p
v0 ) : (v
p)  (p0
v)(v0
p):
It is not hard to see that in other words this reads
Lin(V )  (Lin(V)) ;
B : L  Tr(B  L);

where L 2 Lin(V); B 2 Lin(V ) and so B  2 Lin(V):

Sin e a single dot means the omposition of linear maps, we denoted the
bilinear map of duality by the symbol : to avoid misunderstandings.

3.11. In a ordan e with our results we have

KM

K N  Lin



(K N ) ; K M :

By denition, for y = (yi ) 2 K M and x = (xk ) 2 K N ;


y
x : (K N ) ! K M ;
p 7! (p
x)y ;
from whi h we dedu e that
(y
x)ik = yi xk

(i = 1; : : : ; M; k = 1; : : : ; N ):

Moreover, K N

KN



Tr

 Lin(K N ; K N ); (x
p)i k = xi pk ; and so
Li

k j i; k

= 1; : : : ; N =

N
X
i=1

Li i :

Our onvention that a summation an be arried out only for

a pair of indi es
in opposite positions shows well that the tra e of matri es Lik and (Lik ) makes
no sense.
It an be proved without di ulty that

B j i : Lk l =

N
X
i;k=1

B k i Lk i :

3.12.

Let L 2 Lin(U; X) and K 2 Lin(V; Y): Then U  V ! X

Y;
7! Lu
Kv is a bilinear map, hen e there exists a unique linear map
L
K : U
V ! X
Y su h that
(u; v)

(u 2 U; v 2 V):

(L
K )
(u
v) = Lu
Kv

It is a simple task to show that (L; K ) 7! L

K satises ondition (ii) in
3.1, hen e L
K is the tensor produ t of L and K ; in other words,
Lin(U; X)
Lin(V; Y) ! Lin(U
V; X
Y):

If the ve tor spa es are nite dimensional then  stands instead of ! :

It is not hard to show that
(L
K )
(B
A) = (L
B )
(K
A)
and if both L and K are bije tions then L
K is a bije tion and
(L
K ) 1 = L 1
K 1 :

3.13. For natural numbers n  2 the denition of n-fold tensor produ ts

of ve tor spa es is similar to denition in 3.1, only n-fold linear maps should
be taken instead of bilinear ones. We an state the existen e andn essential
uniqueness of n-fold tensor produ ts similarly. We use the notation
Vk and
n

vk

for the n-fold tensor produ t of ve tor spa es

(k = 1; : : : :; n):

k=1

Vk and

k=1
ve tors vk

2 Vk

We have the identi ations



k
k=1

vk
k=1

V  k
=1n Vk ;

k
k=m+1
n

vk
k=m+1

n
 k
=1
vk :

If the ve tor spa es are nite dimensional then

Vk is identied with the
k=1
n
n
ve tor spa e Linn (  Vk  ; K ) of n-linear maps  Vk  ! K ; alled n-linear
n

forms, su h that

k=1

k=1

vk
k=1

(p1 ; : : : ; pn ) 

n
Y
k=1

(pk
vk ):

3.14.

For natural numbers n  2; the n-fold tensor produ t of n opies of

n
1
the ve tor spa e V is denoted by
V; for onvenien e we put
V := V;

0 V := K : Then we have for all natural numbers n and m

 n

V

m V  n+
mV:

We dene the n-fold symmetri and antisymmetri tensor produ ts of elements of V as follows:

X

=1
v(k) ;
k
2Permn
X
n
n
^
vk :=
(sign)
v(k) ;
k=1
k=1
2Perm

_n v :=
k=1 k

where Permn denotes the set of permutations of f1; : : : :; ng and sign is the sign
of the permutation  : sign = 1 if  is even and sign = 1 if  is odd.
For instan e,
v1 _ v2

= v1
v2 + v2
v1 ;

v1 ^ v2

= v1
v2

v2
v1 :

The linear subspa es of

V spanned by the symmetri and antisymmetri
n
n
tensor produ ts are denoted by _ V and ^ V; respe tively.
We mention that
1 n
1 n
_
vk
and
^v
n! k=1
n! k=1 k
n

are alled the symmetri and antisymmetri part of

n vk ;
k=1
n

respe tively. It is

worth mentioning that the interse tion of ^ V and _ V is the zero subspa e;
moreover, for n = 2 the subspa e of antisymmetri tensor produ ts and that of
symmetri tensor produ ts span V
V:
n

3.15. Let V be

nite dimensional, dim V = N: Then V  V; and we have

the following identi ations:

n V  fn-linear forms on V g;

n V  fn-linear forms on Vg;

_n V  f symmetri n-linear forms on V g;
_n V  f symmetri n-linear forms on Vg;
^n V  f antisymmetri n-linear forms on V g;
^n V  f antisymmetri n-linear forms on Vg:
It is worth mentioning that


n
n
Q
(pk
vk );

vk (p1 ; : : : ; pn ) =
pk (v1 ; : : : ;vn ) =
k=1
k=1
k=1




n k
n
1
n
_ v (p ; : : : ; p ) = k_=1p (v1 ; : : : ;vn )
k=1 k
X
n
Q
=
(p(k)
vk );
2Permn k=1

^n v
k=1 k

(p1 ; : : : ; pn ) =

^n pk
k=1

(v1 ; : : : ;vn ) =
=

2Permn

sign

for all v1 ; : : : ; vn 2 V and p1 ; : : : ; pn 2 V :

If fvi j i = 1; : : : ; N g is a basis in V then


vi j
k=1 k
_n v j
k=1 ik

^n v j
k=1 ik

1  ik  N; k = 1; : : : ; n ;

1  i1  i2 


 in  N ;
1  i1 < i2 <


< in  N

n
Q
k=1

(p(k)
vk );

are bases in

n V; _n V and ^n V; respe tively. A ordingly,

dim

 n

= N n;

dim

 n

_V

 N

N +n 1
;
n

if n  N
0 if n > 0

Similar statements are true for V instead of V:
dim

 n

^V

3.16. The reader is asked to demonstrate that for n = 2 the notions of symmetri ity, symmetri part, et . oin ide with those introdu ed earlier. Moreover,
using the formulae in 3.7 we have


v2 v1
u2
(u1 ; u2) ^ (v1 ; v2 ) = u
vu1 ^ vv1
u u1
u
2
1
2
1
2 ^ v2

for v1 ; u1 2 V1 ; v2 ; u2 2 V2 ; and a similar equality holds for symmetri tensor

produ ts, too.

3.17. We have the following identi ations:


 
 

n   n 
n k
n
n k

V 
V ; k
=1p
k
=1vk  k
=1p (v1 ; : : : ; vn );
 


 
n k
n k
n
n   n 
_ V  _ V ; k_=1p
k_=1vk  k_=1p (v1 ; : : : ; vn );

 

 


n
n k
n
^n V  ^n V ; k^=1
p
^ vk  ^ pk (v1 ; : : : ; vn ):
k=1
k=1

3.18. Let V be an N -dimensional ve tor spa e. Ifn d is an n-linear (symmetri ,

antisymmetri ) form on V (i.e. d is an element of
V ) and L 2 Lin(V); then


d n L : Vn ! K ; (v1 ; : : : ; vn) 7! d(L
v1; : : : ; L
vn )
is also an n-linear (symmetri , antisymmetri ) form.
N
Sin e ^ V ; the ve tor spa e of antisymmetri N -linear forms is onedimensional, for L 2 Lin(V) there is a number (an element of K ) detL; alled
the determinant of L; su h that


N L
for all 2

^ V :

= (detL)

Proposition. For all v1; v2; : : : ; vN in V we have

N
^ L
vk
k=1

N
^v:
k=1 k

= (detL)

Proof. kN^=1L
vk is an antisymmetri N -linear form on V ; 3.15 yields that

for all p1 ; : : : ; pN 2 V


N
^ L
vk
k=1

(p1 ; : : : ; pN ) =

=(detL)

N k
^p
k=1

N k
^p
k=1

(L
v1 ; : : : ; L
vN ) =

(v1 ; : : : ; vN ) = (detL)

N
^v
k=1 k

(p1 ; : : : ; pN ):

As a onsequen e, we have for L; K 2 Lin(V) that

det(L
K ) = (detL)(detK ) = det(K
L):

3.19. Let (v1 ; : : : ; vN ) be an ordered basis of V and let

(p1 ; : : : ; pN ) be the
dual basis in V :
 orresponding

N i
We know that ^ p (v1 ; : : : ; vN ) = 1; thus if L 2 Lin(V) then
i=1


N i
^p

detL =(detL) i=1

=

2P ermN

sign

N i
^p

(v1 ; : : : ; vN ) = i=1

N
Y

i=1

(L
v1 ; : : : ; L
vN ) =

(p(i)
L
vi ):

The last formula is the determinant of the matrix representing L in the oordinatization orresponding to the given ordered basis. Thus for all oordinatizations K of V we have
detL = det(K
L
K 1):

3.20. Proposition. Let V and U be nite dimensional ve tor spa es.

Suppose A; B 2 Lin(V; U) and B is a bije tion. Then
det(A
B 1 ) = det(B 1
A):

Proof. Observe that if U = V then this equality follows from that given at
the end of 3.18. However, if U 6= V; the determinant of A and B 1 make no
sense.
V and U have the same dimension N sin e B is a bije tion between them.
Let K and L be oordinatizations of V and U; respe tively. Then
det(A
B 1 ) = det(L
A
B 1
L 1):

Sin e L
A
B 1
L 1 = L
A
K 1
K
B 1
L 1 and both L
A
K 1 and

K
B 1
L 1 are linear maps K N ! K N ; hen e their determinant is meaningful,
we an apply the formula given at the end of 3.18 to get

det(L
A
B 1
L 1) = det (L
A
K 1)
(K
B 1
L 1 ) =

=det (K
B 1
L 1 )
(L
A
K 1) = det(K
B 1
A
K 1 )
=det(B 1
A):

Our result has the following orollary: if L 2 Lin(V) and B : V

linear bije tion then
det(B
L
B 1 ) = detL:

!U

is a

3.21. For L 2 Lin(V) we dene

0 L := idK ;

n L :
n V !
n V;

n vk
k=1

n
7! k
=1
L
vk :

It is trivial that ^ V and _ V are invariant for

L; the restri tions of
L
n
n
onto these linear subspa es will be denoted by ^ L and _ L; respe tively.
n

3.22. Exer ises

1. Let fv1 ; : : : ; vN g be a basis of V and fp1; : : : ; pN g its dual. Then
N
X
i=1

vi
pi

 idV ;

N
X
i=1

pi
vi

 idV ;

where the symbols on the right-hand sides stand for the identity of V and of V ;
respe tively.
2. The linear subspa es S and T of V are omplementary if S \ T = f0g
and the linear subspa e spanned by S [ T equals V; then for every v there are

uniquely determined elements vS 2 S and vT 2 T su h that v = vS + vT : The

linear map V ! V; v 7! vS is alled the proje tion onto S along T:
Let v 2 V; p 2 V :
(i ) If p
v 6= 0 then vp

vp is the proje tion onto K v (the linear subspa e spanned
by v) along Ker p:
(ii ) If  2 K su h that p
v 6= 1 then idV v
p is a linear bije tion and

v
p:
(idV v
p) 1 = idV +
1 p
v
3. Demonstrate that
L
(v
p) = (L
v )
p;
(v
p)
L = v
L
p

for v 2 V; p 2 V and L 2 Lin(V):
4. Prove that


^n pk (v1 ; : : : ; vn ) = det pk
vi j k; i = 1; : : : ; n
k=1
for p1 ; : : : ; pn 2 V  and v1 ; : : : ; vn 2 V:
5. Prove that if V is a ve tor spa e over K then K N
V  VN ; 
v 

( 1 v; : : : :;  N v):

4. Tensor quotients

4.1. Let U; V and Z be ve tor spa es (over the same eld). A map q :
V  (U n f0g) ! Z is alled linear quotient if
(i ) v 7! q(v; u) is linear for all u 2 U n f0g;
(ii ) q(v; u) = 1 q(v; u) for all v 2 V and u 2 U n f0g;  2 K n f0g:
Denition. Let V and A be ve tor spa es, dim A = 1: A tensor quotient of
V by A is a pair (Z; q) where
(i ) Z is a ve tor spa e,
(ii ) q : V  (A n f0g) ! Z is a linear quotient map having the property that
{ if W is a ve tor spa e and r : V  (A n f0g ! W is a linear quotient
map
| then there exists a unique linear map L : Z ! W su h that
r = L q:
Proposition. The pair (Z; q) is a tensor quotient of V by A if and only if
1) Z = Ran q;
2) if v 2 V; a 2 A n f0g and q(v; a) = 0 then v = 0:
Proof. Sin e Ab is one-dimensional, for a; b 2 A; aa 6= 0 let ab denoteb the
number for whi h a a = b: Observe that if b 6= 0 then b is the inverse of a :

Condition 2) in the proposition is equivalent to the following one: if v; u 2 V

and a; b 2 A n f0g then q(v; a) = q(u; b) implies v = ab u: Conversely, it is
trivial, that if r is a linear quotient map
and v = ab u then r(v; a) = r(u; b):
Moreover, r(v; a) + r(u; b) = r ab v + u; b :
Suppose 1) is fullled. Then every element of V
A has the form q(v; a): If 2) is
valid as well and r is a linear quotient map then the formula
L
(q (v ; a)) :=

r(v; a)

denes a unique linear map L:

If 1) is not fullled, the uniqueness of linear maps L for whi h r = L q holds
fails. If 2) is not valid one an easily onstru t a linear quotient map for whi h
no linear map exists with the desired omposition property.

4.2. We shall see in the next item that tensor quotients exist. In the same
way as in the ase of tensor produ ts, we an see that the tensor quotients of V
by A are anoni ally isomorphi , that is why we speak of the tensor produ t and
applying a ustomary abuse of language we all the orresponding ve tor spa e
the tensor quotient (Z in the denition) denoting it by V
A and writing
V  (A n f0g) ! VA ;

(v; a) 7!

v
a

for the orresponding linear quotient map (q in the denition); av is alled the
tensor quotient of v by a:
We use the term realization and the symbol  in the same sense as in the
ase of tensor produ ts.
It is worth repeating the pre eding results in the new notation: every element
v
v
of V
A is of the form a and a = 0 if and only if v = 0:

4.3. For v 2 V and a 2 A n f0g we dene the linear map

v
a

: A ! V;

b 7!

where ab is the number for whi h ab a = b holds.

b
v
a

Proposition. V  A n f0g ! Lin(A; V) is a linear quotient map whi h

satises onditions (i ) and (ii ) of proposition 4.1. As a onsequen e, av is the
tensor quotient of v by a (that is why we used in advan e this notation) and
V  Lin(A; V):
A
We have Lin(A)  K where  2 K is identied with the linear map a 7! a:
A  K and b is the number for whi h
Thus, a ording to the previous result, A
a
b a = b holds, hen e our notation in 4.1 used in the present proposition as well,
a
is in a ordan e with the generally a epted notation for tensor quotients.
v
4.4. Sin e for all a 2 A n f0g the
map V ! V
A ; v 7! aV is a linear Vbije tion, if

v
fvi j i 2 I g is a basis in V then ai  i 2 I is a basis in A ; and dim A = dim V:
4.5. Let V; U; A and B be ve tor spa es, dim A = dim B = 1: We have the

following identi ations (re all 3.4, 3.5 and 3.8):

(i )

(ii )

(iii )

(iv )

V  Lin(K ; V))  V;

V  Lin(A; V)  V
A;
A
V   V ;
A A
V
B  A
B;
V

A

b   ab ;

(v )

(vi )

A  Lin(A; K ) = A ;

 1 v;

v
a

p v


h a

 v
a1 ;

v;
 pha

v

a

v
b  ab ;

V
U  V
U  V
U  et :
A B A
B A
B
v
a

u  v
u  et :

ub  v ab
ab

In parti ular,

(vii )

A
VA  A A
V  V;
V  U  V  U;
A A A

B K:
A
B A
(v; u)
a

v u

a a

Note that a ording to (v ) and (vi ) the rules of tensorial multipli ation and
division oin ide with those well known for numbers.

4.6. Let V; U; A and B be ve tor spa es, dim A = dim B = 1: IfLL
v 2

Lin(V; U) and 0 6= F 2 Lin(A; B) then V  (A n f0)) ! U
B ; (vU; a) 7! F
a is
linear quotient, hen e there exists a unique linear map FL : V
A ! B su h that
L v


F a

L
v
F
a

(v 2 V ; a 2 A n f0g):

It is not hard to see that FL is really the quotient of L by F ; in other words,

Lin(V; U)
Lin(A; B)

 Lin

V; U:
AB

5. Tensorial operations and orientation

In this se tion V denotes an N -dimensional real ve tor spa e and A denotes
a one-dimensional real ve tor spa e.

5.1. Re all that an element (v1 ; : : : ; vN ) of VN is alled an ordered basis of

V if the set fv1; : : : ; vN g is a basis in V:
Denition. Two ordered bases (v1 ; : : : ; vN ) and 0(v10 ; : : : ; vN0 ) of V are alled

equally oriented if the linear map dened by vi 7! vi (i = 1; : : : ; N ) has positive

determinant. An equivalen e lass of equally oriented bases is alled an orientation of V: V is oriented if an orientation of V is given; the bases in the hosen
equivalen e lass are alled positively oriented, the other ones are alled negatively ortiented. (More pre isely, an oriented ve tor spa e is a pair (V;o) where
V is a ve tor spa e and o is one of the equivalen e lasses of bases.)
A linear bije tion between oriented ve tor spa es is orientation-preserving or
orientation-reversing if it sends positively oriented bases into positively oriented
ones or into negatively oriented ones, respe tively.

It is worth mentioning that there are two equivalen e lasses of equally oriented bases.
Observe that the two bases in the denition are equally oriented if and only
N
N
if ^ vi0 is a positive multiple of ^ vi (see Proposition 3.18).
i=1
i=1
If V is oriented, we orient V by the dual of positively oriented bases of V:
If U and V are oriented ve tor spa es, U  V is oriented by joining positively
oriented bases; more losely, if (u1 ; : : : ; uM ) and (v1 ; : : : ; vN ) are positively oriented
bases
in
U and in V; respe tively, then
((u1 ; 0); : : : ; (uM ; 0) ; (0; v1 ); : : : ; (0; vN )) is dened to be a positively oriented
basis in U  V:
The reader is asked to verify that the orientation of the dual and the Cartesian
produ ts is welldened.

5.2. Two bases a and0 a0 of the one-dimensional ve tor spa e A are equally
a0

oriented if and only if a is a positive multiple of a; in other words, a is a

positive number.
If (v1 ; : : : ; vN ) and (v10 ; : : : ; vN0 ) are equally oriented ordered bases of V; a and
0
0 ) are
a are equally oriented bases of A; then (av1 ; : : : ; avN ) and (a0 v10 ; : : : ; a0 vN
equally oriented bases of A
V: Indeed, a ording to our onvention A
V 
N 0 0
N 0
V
A; we have i=1
^ (a vi )  (a0 )N i=1
^ vi whi h is evidently a positive multiple of
aN

^v:
i=1 i

As a onsequen e, an orientation of V and an orientation of A determine a

unique orientation of A
V; we onsider A
V to be oriented by this orientation.
We an argue similarly to show that an orientation of V and an orientation
of A determine a unique orientation of V
A ; we take this orientation of the tensor
quotient.

5.3. A non-zero element a of the oriented one-dimensional ve tor spa e A is

alled positive, in symbols 0 < a; if the orresponding basis is positively oriented.
Moreover, we write a  b if 0  b a: It is easily shown that in this way we
dened a total ordering on A for whi h
(i ) if a  b and  d then a +  b + d;
(ii ) if a  b and  2 R+ then a  b:
We introdu e the notations
A+ := fa 2 Aj 0 < ag; A+0 := A+ [ f0g:
Furthermore, the absolute value of a 2 A is
8
if a 2 A+
>
<a
j aj := > 0 if a = 0
:
a if a 2
= A+ :

5.4. Even if A is not oriented, A

A has a \ anoni al" orientation in whi h
the elements of the form a
a are positive. If A is oriented, the orientation of
A
A indu ed by the orientation of A oin ide with the anoni al one. Then

A+0 ! (A
A)+0 ; a 7! a
a
()
is a bije tion. Indeed, a
a = 0 if and only if a = 0: The elements of (A
A)+
+
has the form a
b where
p a; b2p A ; sin e b = a for some positive number

; we have a
b = a
a ; i.e. the above mapping is surje tive. If
0 6= a
a = b
b then a
a = 2 a
a whi h implies that 2 = 1; thus  = 1;
a = b : the mapping in question is inje tive.
In spite of our earlier agreement, in dedu ing the present result, we preferred
not to omit the symbol of tensorial multipli ation. However, in appli ations of
the present result we keep our agreement; in parti ular, we write
a2

:= aa (:= a
a):

The inverse of the mapping  is denoted by the symbol

square root mapping.
Note that
p 2
a = j aj

and is alled the

(a 2 A):

V. PSEUDO-EUCLIDEAN VECTOR SPACES

1. Pseudo-Eu lidean ve tor spa es

1.1. Denition. A pseudo-Eu lidean ve tor spa e is a triplet (V; B; h)

where
(i ) V is a non-zero nite dimensional real ve tor spa e,
(ii ) B is a one-dimensional real ve tor spa e,
(iii ) h : V  V ! B
B is a non-degenerate symmetri bilinear map.
Remarks. (i ) Non-degenerate means that if h(x; y) = 0 for all x 2 V then
y = 0:
(ii ) h(x; y) is often alled the h-produ t of x; y 2 V: The elements x and y
of V are alled h-orthogonal if their h-produ t is zero.
(iii ) In mathemati al literature one usually onsiders the ase B = R; i.e.
when | be ause of R
R  R| the pseudo-Eu lidean form h takes real values.
Physi al appli ations require the possibility B 6= R:
1.2. Denition. A basis feij i = 1; : : : ; N g of V is alled h-orthogonal if
h(ei ; ek ) = 0 for i 6= k:
An h-orthogonal basis fei j i = 1; : : : ; N g is normed to a 2 B if either
h(ei ; ei) = a2 or h(ei ; ei) = a2 for all i: If B = R; an h-orthogonal basis
normed to 1 is alled h-orthonormal.
Sin e B
B has a anoni al orientation, it makes sense that h(x; y) is negative
or positive for x; y 2 V:

We an argue like in the ase of real-valued bilinear forms to have the following.
Proposition. h-orthogonal bases in V exist and there is a nonnegative integer :(h) su h that for every h-orthogonal basis fei j i = 1; : : : ; N g

h(ei ; ei) < 0

h(ei ; ei) > 0

for :(h) indi es i;

for N :(h) indi es i:

An h-orthogonal basis an always be normed to an arbitrary 0 6= a 2 B:

Further on we deal with h-orthogonal bases normed to an element of B and

su h a basis will be numbered so that h takes negative values on the rst :(h)
elements, i.e.
h(ei ; ei) = (i)a2;

1 if i = 1; : : : ; :(h)
(i) =
1 if i = :(h) + 1; : : : ; N:
We say that h is positive denite if h(x; x) > 0 for all non-zero x: h is positive
denite if and only if :(h) = 0:

1.3. An important property of pseudo-Eu lidean ve tor spa es is that a

natural orresponden e exists between V and BV

B : Note that every element of

V is of the form y where y 2 V and a; b 2 B n f0g: Take su h an element
B
B
ab
of BV

B : Then
y ; x)
V ! R; x 7! h(ab
is a linear map, i.e. an element of V ; whi h we write in the form h(y;
) :
ab

Proposition. BV
B ! V; aby 7! h(aby;
) is a linear bije tion.
Proof. It is linear and inje tive be ause h is bilinear and non-degenerate, and
surje tive be ause the two ve tor spa es in question have the same dimension.

We nd this linear bije tion so natural that we use it for identifying the ve tor
spa es:
V  V; y  h(y;
) :

B
B

ab

ab

1.4. (i ) In the above identi ation the dual of an h-orthogonal basis feij i =
1; : : : ; N g be omes


ei
h(ei ; ei) j i = 1; : : : :; N
whi h equals


(i)ei
j
i
=
1
;
:
:
:
;
N
2
a

if the h-orthogonal basis is normed to a:

As a onsequen e, for all x 2 V (see IV.1.1),

h(ei ; x) ei;
h
i=1 (ei ; ei )
and x = 0 if and only if h(ei ; x) = 0 for all i = 1; : : : ; N:
x=

N
X

(ii ) If V is oriented then both V and BV

B are oriented. The above identi ation

is
orientation-preserving if
:(h) is even and is
orientation-reversing if :(h) is odd.

1.5. Let us take a linear map F : V ! V: As we know, its transpose is

a linear map F  : V ! V ; a ording to the previous identi ation we an
V
onsider it to be a linear map F  : BV

B ! B
B : Consequently, we an dene
the h-adjoint of F ;
F>

: V ! V;

7! (ab)

F

y 
:
ab

Observe that this is equivalent to

F>
y
y
= F

ab

(y 2 V; a; b 2 B n f0g):

ab

A ording to the denition of the transpose we have



>
y

F
x = F 
y
x = F
y
x;
ab

whi h means
i.e.

ab

ab

h(y; F
x) = h(F >
y; x) ;

ab

ab

h(y; F
x) = h(F >
y; x) = h(x; F >
y)

(x; y 2 V):

The denition of h-adjoints involves that the formulae in IV. 1.4 remain valid
for h-adjoints as well: if F ; G 2 Lin(V);  2 R; then
(F + G)> =F > + G> ;
(F )> =F > ;
(F
G)> =G>
F > :
Moreover,

detF > = detF :

1.6. Let (V; B; h) and (V0 ; B;0 h0) be pseudo-Eu lidean ve tor spa es. A
linear map L : V ! V0 is alled h-h0 -orthogonal if there is a linear bije tion
Z : B ! B0 su h that h0 (L  L) = (Z
Z ) h i.e.
h0(L
x; L
y) = (Z
Z )h(x; y)
(x; y 2 V):

0
Note that a ording to our identi ation, Z is an element of BB :
It is quite trivial that there is a h-h0 -orthogonal linear map between the
pseudo-Eu lidean ve tor spa es if and only if dim V = dim V0 and
:(h) = :(h0 ):
In parti ular, if fei j i = 1; : : : ; N g is an h-orthogonal basis, normed to a; of
V; and fe0ij i = 1; : : : ; N g is an h0-orthogonal basis, normed to a0; of V0 then
L
ei

:= e0 i

(i = 1; : : : ; N )

0
determine an h-h0 -orthogonal map for whi h Z = aa :

1.7. Let n and N be natural numbers, N  1; n  N: The map

Hn : RN

 RN ! R;

(x; y) 7!

n
X
i=1

xi yi +

N
X
i=n+1

xi y i =

N
X
i=1

(i)xi yi

(where (i) := 1 for i = 1; : : : ; n and (i) := 1 for i = n + 1;

: : : ; N ) is a non-degenerate symmetri bilinear map, i.e. (RN ; R; H n ) is a pseudoEu lidean ve tor spa e and :(Hn ) = n:
The standard basis of RN is Hn -orthonormal.

A ording to 1.3, we have the identi ation RN   RN ; but we must pay
attention to the fa t that if n 6= 0 it diers from the standard one mentioned in
IV.1.6.
The standard identi ation is a linear bije tion S : RN ! (RN ) ; and the
present identi ation is another one: Jn : RN ! (RN ) ; x 7! Hn (x;
): We easily
see that




(xi j i = 1; : : : :; N ) := Jn
xi j i = 1; : : : ; N = (i)xi j i = 1; : : : ; N :

The standard identi ation oin ides with J0 ; the one orresponding to H0 :
A ording to the identi ation indu ed by Hn ; the dual of the standard basis
fi j i = 1; : : : ; N g is f(i)i j i = 1; : : : ; N g:
It is useful to regard Hn as the diagonal matrix in whi h the rst n elements
in the diagonal are 1 and the others equal 1.
For the Hn -adjoint of the linear map (matrix) F we have x
Hn
F >
y =
(F
x)
Hn
y = x
F 
Hn
y for all x; y 2 RN ; where F  denotes the usual
transpose of the matrix F ; thus F 
Hn = Hn
F > or
F>

= Hn
F 
Hn :

1.8. Exer ises

1. Let e1 ; : : : ; en be pairwise h-orthogonal ve tors in the pseudoEu lidean ve tor spa e (V; B; h) su h that h(ei ; ei ) 6= 0 for all i = 1; : : : ; n:
Prove that the following statements are equivalent:
(i )
(ii )
(iii )
(iv )
(v )

n = dim V (i.e. the ve tors form a basis),

if h(ei ; x) = 0 for all i = 1; : : : ; n then
n
X
h(ei ; x) ei for all x 2 V;
x=
h
i=1 (ei ; ei )

h(ei ; x)
h(ei ; y)
h(ei; ei )
i=1
n
h(x; x) = X h(ei ;hx()e
i; ehi()ei ; x)
i=1

h(x; y) =

n
X

x=

0;

for all x; y 2 V ;
for all x 2 V:

2. Demonstrate that the set fe1 ; : : : ; en g of pairwise h-orthogonal ve tors

an be ompleted to an h-orthogonal basis if and only if h(ei ; ei ) 6= 0 for all
i = 1; : : : :; n:

2. Tensors of pseudo-Eu lidean ve tor spa es

F

2.1. Let V and A be nite dimensional ve tor spa es, dim A = 1: Suppose
: V ! V is a linear map. Then we an dene the linear maps
F A : A
V ! A
V;
av 7! a
(F
v );
FA

V ! V;
A A

v
a

7! F a
v

(F A = idA
F ; FA = idFA ; see IV.3.12 and IV.4.6).
A ording to the usual identi ations
Lin(A
V) (A
V)
(A
V)  A
V
A
V 
A
A
V
V  R
V
V  V
V 

Lin(V);

we have F A  F and similarly FA  F : Therefore we shall write F instead of

F A and FA :
for s 2A
V

for n 2

V
A

we have
we have

A
V;
V
F
n2 :
A

F
s2

2.2. Let us formulate the previous onvention in another way. V

A 
Lin(A ; V); hen e we have the omposition F
s of F 2 Lin(V)  V
V and
s 2 Lin(A ; V)  V
A:
More generally, if U and W are nite dimensional ve tor spa es, the dot
produ t of an element from U
V and an element from V
W is dened to
be an element in U
W; this dot produ t an be regarded as the omposition
of the orresponding linear maps and is hara terized by
(u
p)
(v
w) = (p
v)u
w:
The s heme is worth repeating:
U
V dot V
W

results in

U
W:

Evidently, we an have U = AK or W = AK ; thus similar formulae are valid for

tensor quotients as well.

2.3. What we have said in the previous paragraph on erns any ve tor spa es.
In the following (V; B; h) denotes a pseudo-Eu lidean ve tor spa e.
The identi ation des ribed in 1.3 and the orresponding formula suggests us
a new notation: \removing" the denominator from both sides we arrive at the
denition
x
y := h(x; y );
i.e. in the sequel we omit h; denoting the h-produ t of ve tors by a simple dot.
The dot produ t of two elements of V is an element of B
B: Then we an
extend the previous dot produ t formalism as follows:
U
V

results in
(B
B)
U
W;
(u
v0 )
(v
w) := (v0
v)u
w:

dot

V
W

2.4. A ording to the onvention introdu ed in 2.1, the h-adjoint and >the
transpose of a linear map F : V ! V an be identied, for F  = idFB
B :
However, we ontinue to distinguish between the transpose and the h-adjoint
be ause of the following reason.
In the pseudo-Eu lidean ve tor spa e (RN ; R; H n ) for n 6= 0; n 6= N a linear
map RN ! RN is represented by a matrix, and the transpose of a matrix has
a generally a epted meaning, and the Hn -adjoint of a matrix diers from its
transpose.
The h-adjoint of F 2 Lin(V) is hara terized in the new notation of dot
produ ts as follows:
y
F
x = (F >
y )
x = x
F > y
(y; x 2 V):

2.5. A ording to our onvention we have

V
n
x
is in B for
n2
B
n
m

is in

for

n; m 2

and x 2 V;

V:
B

If fei j i = 1; :::; N g is an h-orthogonal basis, normed to a 2 B; in

:= eai  i = 1; :::; N is an h-orthonormal basis of V
B:

ni

ni
nk

= (i)ik

V; then

(i; k = 1; :::; N ):

It is more onvenient to use this basis instead of the original one; for all x 2 V
we have
x=

N
X
i=1

(i)(ni
x)ni :

2.6. (i ) We have the identi ations

V
  V  V   V ; the element

B
B
B
B
B
B
V
n of V
B is identied with the linear fun tional B ! R; m 7! n V
m: V V
(ii ) In view of the identi ations Lin(V)  V
V  V
B
B  B
B ; or
in view of our dot produ t onvention, for n; m 2 V
B;
m
n : V ! V;
x 7! (n
x)m
is a linear map, and every element of Lin(V) is the sum of su h linear maps.
Evidently,
(m
n)
(m0
n0 ) = (n
m0 )m
n0
and
(m
n)> = n
m:

2.7. Denition. For the pseudo-Eu lidean ve tor spa e (V; B; h) we put
O(h) :=fL 2 Lin(V)j L> = L 1 g;
A(h) :=fA 2 Lin(V)j A> = Ag;
and the elements of O(h) and A(h) are alled h-orthogonal and h-antisymmetri ,
respe tively.

Proposition. (i ) For L 2 Lin(V) the following three statements are equivalent:

| L is in O(h);
| (L
y)
(L
x) = y
x
for all y; x 2 V;

|
(ii )
|
|
|

(L
x)
(L
x) = x
x
for all x 2 V:
For A 2 Lin(V) the following three statements are equivalent:
A is in A(h);
y
A
x = (A
y )
x = x
A
y
for all y; x 2 V;
x
A
x=0
for all x 2 V:

2.8. Proposition.

(i ) jdetLj = 1
for L 2 O(h);
(ii ) TrA = 0
for A 2 A(h):
Proof. ItVis onvenient to regard now the linear maps in question as linear
maps V
B ! B ; a ording to our identi ations des ribed in 2.1. It is not hard
to see that this does not in uen e determinants and tra es.
(i ) Let fn1 ; : : : ; nN g be an
h-orthonormal basis in VB : A ording to IV.3.15


V
and to the identi ation B  V
B we have
0 6=

N
^n
k=1 k

N
^ L
nk (L
n1 ; : : : ; L
nN ) =
k=1


N
=(detL) ^ L
nk (n1 ; : : : ; nN ) =
k=1


N
=(detL)2 ^ nk (n1 ; : : : ; nN ):
k=1

(n1 ; :::; nN ) =

(ii ) We know that the dual of the pre eding basis is f(i)ni j i = 1; : : : ; N g ;
thus in view of IV.3.9,
TrA =

N
X
i=1

(i)ni
A
ni = 0:

2.9.

A linear map S : V ! V is alled h-symmetri if S > = S or,

equivalently, x
S
y = y
S
x for all x; y 2 V: The set of h-symmetri
linear maps is denoted by S(h):
A(h) and S(h) are omplementary subspa es of Lin(V)  V
V ; i.e. their
interse tion is the zero subspa e and they span the whole spa e V
V : Indeed,
only the zero linear map is both symmetri and antisymmetri , and for any linear
map F : V ! V we have that
S

:=

+ F>
;
2

A :=

F>

are h-symmetri and h-antisymmetri , respe tively, and F = S + A:

V
V V
Taking the identi ation V
V  V
B
B we an easily see that B _ B  S(h)
V
V
and B ^ B  A(h); sin e these subspa es are omplementary, too, equalities hold
ne essarily:

V _ V := VB _ VB = S(h);

V ^ V := VB ^ VB = A(h):

As a onsequen e,
dim S(h) =
Re all that for m; n 2

N (N + 1)
;
2
V we have
B

m _ n = m
n + n
m;

2.10. Proposition.
(V
V )  (V
V ) ! R;

dim A(h) =

N (N 1)
:
2

m^n =m
n

n
m:

(F ; G) 7! F : G := Tr(F >
G)

is a non-degenerate symmetri bilinear form, whi h is positive denite if and

only if h is positive denite.
Proof. It is trivially bilinear and symmetri be ause of the properties of Tr
and h-adjoints.
Suppose that Tr(F >
G) = 0 for all F 2 V
V ; i.e.
0=

N
X
i=1

(i)ni
F >
G
ni

for all h-orthonormal bases fn1 ; : : : ; nN g of V

B : Then taking F := nj
nk for all
j; k = 1; : : : ; N; and using (nj
nk )> = nk
nj we on lude that nj
G
nk = 0
for all j; k whi h results in G = 0:
Sin e
N
X
(i)(F
ni )
(F
ni );
Tr(F >
F ) =
i=1

we see that if h is positive denite then Tr(F >
F ) > 0; if h is not positive
denite then we an easily onstru t an F su h that F : F < 0:
Remark. (i ) Compare the present bilinear form with that of the duality
treated in IV.3.10; take into a ount the identi ation V
V  V
BV

B 
V
V  V
V:
B
B

(ii ) The bilinear form is not positive denite, in general, either on the
linear subspa e of h-symmetri linear maps or on the linear subspa e of hantisymmetri linear maps.
(iii ) For k1 ; k2 ; n1 ; n2 2 V
B we have
(k1
n1 ) : (k2
n2 ) =(k1
k2 )(n1
n2 );
(k1 _ n1 ) : (k2 _ n2 ) =2 ((k1
k2 )(n1
n2 ) + (k1
n2 )(k2
n1 )) ;
(k1 ^ n1 ) : (k2 ^ n2 ) =2 ((k1
k2 )(n1
n2 ) (k1
n2 )(k2
n1 )) ;
whi h shows that sometimes it is onvenient to use the half of this bilinear form
for h-symmetri and h-antisymmetri linear maps:
F
F

 G := 12 F : G = 12 Tr(F >
G) = 21 Tr(F
G

 G := 12 F : G = 12 Tr(F >
G) =

1
Tr(F
G)
2

(F ; G 2 S(h));

(F ; G 2 A(h)):

2.11. Proposition. Let L be an h-orthogonal map. Then for all F ; G 2

Lin(V)
(i ) (L
F
L 1 ) : (L
G
L 1 ) = F : G;
(ii ) if F is h-symmetri or h-antisymmetri then L
F
L 1 is h-symmetri
or h-antisymmetri , respe tively.

2.12. Proposition. VIf (n1 ; : : : ; nN ) and (n01 ; : : : ; n0N ) are equally oriented

h-orthonormal bases in B then

N
^n
i=1 i

N 0
^n:
i=1 i

Proof.

Evidently, L
ni := n0i (i = 1; : : : ; N ) determines an h-orthogonal

map L whose determinant is positive sin e the bases are equally oriented. Then
proposition in IV.3.18 gives the desired result.
Suppose V and B are oriented; then V
B is oriented as well and the Levi{Civita
tensor of (V; B; h);
N

N ei
" := i=1
^ ni = i=1
^a

N

V
2 i=1
^ B  ^N V ;

B
N

is well dened, where (n1 ; : : : ; nN ) is a positively oriented orthonormal basis in

V ; and (e1 ; : : : ; eN ) is a positively oriented orthogonal basis in V; normed to
B
a 2 B+ :

2.13. Exer ises

1. A ording to the theory of tensor quotients, the Levi{Civita tensor an be
onsidered to be a linear map

":


N B ! N^ V;

N
N

ai 7!
ai
":
i=1
i=1

^ ei where (e1 ; : : : ; eN ) is a positively oriented hProve that

ai
" = i=1
i=1
orthogonal basis su h that jei
ei j = ja2i j (i = 1; ::; N ):
N
2. The previous linear map is a bije tion whose inverse is 1" 2 N
B ; regarded
^V
N
^
xi
N
N
N
as a linear map ^ V !
B; i=1
^ xi 7! i=1" :
N
^ xi
N
Q
P
i
=1
sign (n(i)
xi ) =: "(x1 ; : : : ; xN ):
Prove that " =
i=1
2PermN
N

3. Eu lidean ve tor spa es

3.1. A pseudo-Eu lidean ve tor spa e (V; B; h) is alled Eu lidean if h is

positive denite or, equivalently, :(h) = 0:
For a lear distin tion, in the following (E; D; b) denotes a Eu lidean ve tor
spa e.
The notations introdu ed for pseudo-Eu lidean ve tor spa es will be used, e.g.
x
y

:= b(x; y)

E then
note that if x; y 2 E and k; n 2 D
x
y

Moreover, we put

2 D
D;

n
x2

D;

(x; y 2 E);
k
n 2 R:

jxj2 := b(x; x)

(x 2 E):

Lastly, we say orthogonal, adjoint et . instead of b-orthogonal, b-adjoint et .

3.2. Re all that a anoni al order is given in D

D (IV.5.4) and so in
(D
D)
(D
D) as well. Thus the absolute value of elements in D
D and
the square root of elements in (D
D)
(D
D) make sense.
Proposition (Cau hy{S hwartz inequality). For all x; y 2 E we have
q

jx
yj  jxj2 jyj2

and equality holds if and only if x and y are parallel.

Proof. Ex lude the trivial ases x = 0 or y = 0: Then the positive deniteness of b yields
0

y
x 
y
y
y 

=jxj2
=jxj2

y
x
2
(x
y) +
y
y
(y
x)(x
y)

y
x
y
y

2

jyj2 =

jy j2

where equality holds if and only if x = yy

xy y:

In general, the right-hand side of the Cau hy inequality annot be written in
a simpler form be ause jxj and jyj make no sense, unless D is oriented.

3.3. Suppose now that D is oriented as well. Then we an dene the

magnitude or length of x 2 E as a non-negative element of D :
q

jxj := jxj2 :
The following fundamental relations hold:
(i ) jxj = 0 if and only if x = 0;
(ii ) jxj = jjjxj;
(iii ) jx + yj  jxj + jyj
for all x; y 2 E and  2 R: The third relation is alled the triangle inequality
and is proved by the Cau hy{S hwartz inequality.
Moreover, the Cau hy{S hwartz inequality allows us to dene the angle
formed by x 6= 0 and y 6= 0 :
arg(x; y) := ar os

3.4. The identi ation

x
y
jxjjyj :

E  E
D
D

(see 1.3) is a fundamental property of the Eu lidean ve tor spa e (E; D; b):
The dual of an orthogonal
basis e1 ; : : : ; eN ; normed to m 2 D; in this
e
eN
1
identi ation be omes m
;
:
:
:
;
2
m2 :
e
E whi h
A ordingly, ni := mi (i = 1; : : : ; N ) form an orthonormal
basis in D


E
E
oin ides with its dual basis in the identi ation D  D :
ni
nk = ik
(i; k = 1; : : : ; N ):
For all x 2 E we have

x=

N
X
i=1

(ni
x)ni :

E will be used frequently, so we nd it onvenient to introdu e

In the following D
a shorter notation:
N := E :

3.5. If S is a linear subspa e of E then

S? := fx 2 Ej x
y = 0 for all y 2 Sg
is alled the ortho omplement of S: It an be shown that S? is a linear subspa e,
omplementary to S; i.e. their interse tion is the zero subspa e and they span
the whole E:
Every ve tor x 2 E an be uniquely de omposed into a sum of two ve tors,
one in S and the other in S? ; alled the orthogonal proje tions of x in S and in
S?; respe tively.
Let n be a unit ve tor in N; i.e. jnj2 := n
n = 1: Then
n
D := fndj d 2 Dg
is a one-dimensional subspa e of E: Furthermore, its ortho omplement,
fx 2 Ej n
x = 0g = (n
D)?
is an (N 1)-dimensional subspa e. The orresponding proje tions of x 2 E in
n
D and in (n
D)? are
(n
x)n

and

(n
x)n;

respe tively.

3.6. Proposition. For F 2 Lin(E) we have Ker F > = (Ran F )?:

Proof. x is in Ker F > ; i.e. F >
x = 0 if and only if y
F >
x = 0 for>all
y 2 E; whi h is equivalent to x
F
y = 0 for all y 2 E; thus x is in Ker F if

and only if it is orthogonal to the range of F :

3.7.

We know that a linear map L : E ! E is orthogonal, i.e. y
x =

(L
y)
(L
x) for all x; y 2 E if and only if jL
xj2 = jxj2 for all x 2 E (see
2.7). Be ause of the Eu lidean stru ture we need not assume the linearity of L;
a ording to the following result.
Proposition. Let L : E ! E be a map su h that L(y)
L(x) = y
x for all
x; y 2 E: Then L is ne essarily linear.

Proof. First of all note that if fe1; : : : ; eN g is an orthogonal basis in E then

fL
e1 ; : : : ; L
eN g is an orthogonal basis as well. As a onsequen e, Ran L

spans E:
If L(y) = L(x) then jyj2 = jxj2 = y
x and so jy xj2 = 0; hen e L is
inje tive.
Writing x0 := L(x); y0 := L(y) and then omitting the prime, we nd that
L 1 (y )
L 1 (x) = y
x for all x; y 2 Ran L and L 1 (y )
x = y
L(x) for all
x 2 E; y 2 Ran L:
Consequently, for all y 2 Ran L and x1 ; x2 2 E we have
y
L(x1 + x2 ) =L 1 (y )
(x1 + x2 ) = L 1 (y )
x1 + L 1 (y )
x2

=y
L(x1 ) + y
L(x2 ) = y
(L(x1 ) + L(x2 )) ;

sin e y is arbitrary in Ran L whi h spans E; this means that

L(x1 + x2 ) = L(x1 ) + L(x2 )

(x1 ; x2 2 E):

A similar argument shows that

( 2 R; x 2 E):

L(x) = L(x)

This has the simple but important onsequen e that if L : E ! E is a map

su h that jL(y) L(x)j2 = jy xj2 for all x; y 2 E and L(0) = 0 then L is
ne essarily linear. The proof is left to the reader as an exer ise.

3.8. In the following, assuming that

dim E = 3;

we shall examine the stru ture of the antisymmetri linear maps of E: As we

E ^ E = N ^ N is a three-dimensional ve tor spa e
know, (see 2.9) A(b)  D
D
endowed (see 2.10) with a real-valued positive denite symmetri bilinear form
| an inner produ t | :
AB

1
1
= Tr(A>
B ) = Tr(A
B ):
2
2

The magnitude of the antisymmetri linear map A is the real number

p
jAj := A  A:
Re all that for k1 ; k2 ; n1 ; n2 2 N we have
(k1 ^ k2 )  (n1 ^ n2 ) = (k1
k2 )(n1
n2 )

(k1
n2 )(k2
n1 );

in parti ular, if k1 = k2 =: k; n1 = n2 =: n;

jk ^ nj2 = jkj2 jnj2

(k
n)2 :

3.9. If A 2 N ^ N then A> =

A; thus proposition 3.6 yields that Ker A is

the orthogonal omplement of Ran A:
Proposition. If 0 6= A 2 N ^ N then Ran A is two-dimensional, onsequently, Ker A is one-dimensional.
Proof. Sin e A 6= 0; there is a 0 6= x 2 Ran A: Then x 62 Ker A; thus
0 6= A
x 2 Ran A: x and A
x are orthogonal to ea h other be ause A is
antisymmetri . Consequently, the subspa e spanned by x and A
x is twodimensional in the range of A : Ran A is at least two-dimensional, Ker A is at
most one-dimensional. Suppose Ker A = f0g: Take a 0 6= y; orthogonal to both
x and A
x: E is three-dimensional, A
y is orthogonal to y ; so it lies in the
subspa e generated by x and A
x; i.e. A
y = x +  A
x: Multiplying by x
and using x
A
x = 0; x
A
y = y
A
x = 0; we get  = 0: As a onsequen e,
A
(y  x) = 0; the ve tor y  x is in Ker A; thus y  x = 0; y =  x; a
ontradi tion.

3.10. Let us take a non-zero A 2 N ^ N: There is an orthonormal basis

fn1 ; n2 ; n3 g in N su h that n3
D = Ker A: fn1 ^ n2 ; n3 ^ n1 ; n2 ^ n3 g is a
basis in N ^ N; thus there are real numbers 1 ; 2 ; 3 su h that
A = 3 (n1 ^ n2 ) + 2 (n3 ^ n1 ) + 1 (n2 ^ n3 ):
Sin e A
n3 = 0; we easily nd that 2 = 1 = 0 and, onsequently,
j3 j = jAj: Renaming n1 and n2 and taking their antisymmetri tensor produ t

in a onvenient order we arrive at the following result.

Proposition. If 0 6= A 2 N ^ N and n is an arbitrary unit ve tor in N;
orthogonal to the kernel of A; then there is a unit ve tor k; orthogonal to the
kernel of A and to n su h that
A = jAjk ^ n:

As a onsequen e, we have
A3

Moreover, for non-zero A and B

equivalent:
| A is a multiple of B ;

jAj2 A:
in N ^ N

the following statements are

| Ker A = Ker B ;
| Ran A = Ran B :

3.11. If A; B 2 N ^ N; their ommutator

N ^ N;
2N^N

is in

[A; B := A
B

B
A

too. Moreover, the properties of the tra e imply that for all

Proposition.

[A; B  C = [C ; A  B = [B ; C  A:

j[A; B j2 = jAj2 jB j2

(A  B )2 :

Proof. If A and B are parallel (in parti ular, if one of them is zero) then
the equality holds trivially. If A and B are not parallel, dividing the equality
by jAj2 jB j2 we redu e the problem to the ase jAj = jB j = 1: The ranges of A
and B are dierent two-dimensional subspa es, hen e their interse tion is a onedimensional subspa e (be ause E is three-dimensional). Let n be a unit ve tor
in N su h that n
D = Ran A \ Ran B : Then there are unit ve tors k and r in
N; orthogonal to n; su h that A = k ^ n; B = r ^ n: Simple al ulations yield
[A; B = k ^ r;

j[A; B j2 = 1

(k
r)2

whi h gives the desired result in view of 3.8.

3.12. A ording to the formula ited at the beginning of the previous paragraph, [A; B is orthogonal to both A and B : Consequently, if A and B are
orthogonal, jAj = jB j = 1 then A; B and [A; B form an orthonormal basis in
N ^ N:
Proposition. For all A; B; C 2 N ^ N we have
[[A; B ; C = (A  C )B

(B  C )A:

Proof. If A and B are parallel, both sides are zero. If A and B are not
parallel (in parti ular, neither of them is zero) then B = A + B0 where  is a
number and B0 6= 0 is orthogonal to A: A results in zero on both sides, hen e it
is su ient to onsider an arbitrary A; a B orthogonal to A; and three linearly
independent elements in the role of C ; they will be A; B and [A; B :
For C = [A; B the equality is trivial, both sides are zero.
For C = A; the right-hand side equals jAj2 B ; [[A; B ; A on the left-hand
side is orthogonal to both [A; B and A; hen e it is parallel to B : there is a

number  su h that [[A; B ; A = B : Take the inner produ t of both sides by

B ; apply the formula at the beginning of 3.11 to have j[A; B j2 = jB j2 whi h
implies  = jAj2 a ording to the previous result.
A similar argument is applied to C = B :

3.13. Let us ontinue to onsider the Eu lidean ve tor spa e

(E; D; b);
E
dim E = 3; and suppose that E and D are oriented. Then N = D is oriented as
3
well (see IV.5.2). A ording to 2.12, there is a well-dened " in ^ N su h that
3

" = i=1
^ ni
for an arbitrary positively oriented orthonormal basis (n1 ; n2 ; n3 ) of
alled the Levi{Civita tensor of (E; D; b):
The Levi{Civita tensor establishes a linear bije tion

j : N ^ N ! N;

N: " is

k ^ n 7! "(
; k; n):

Let us examine more losely what this is. The dual of N is identied with N;
thus " an be onsidered to be a trilinear map

N3 ! R;

(k1 ; k2 ; k3 ) 7!

2P erm3

sign

3
Y
i=1

n(i)
ki

(see IV.3.15). Thus, for given k and n; "(
; k; n) is the linear map
r 7! "(r ; k; n); i.e. it is an element of N  N:
In other words, j(k ^ n) is the element of N determined by

N ! R;

r
(k ^ n) = "(r ; k; n)

for all r 2 N:
The Levi{Civita tensor is antisymmetri , hen e j(k ^ n) is orthogonal to both
k and n:
If (n1 ; n2 ; n3 ) is a positively oriented orthonormal basis in N; then

j(n1 ^ n2 ) =

n3 ;

j(n2 ^ n3) =

n1 ;

j(n3 ^ n1 ) =

n2 :

Proposition. For all A 2 N ^ N we have

(i ) A
j(A) = 0;
(ii ) jj(A)j = jAj;
(iii ) if A 6= 0 then (j(A); n; A
n) is a positively oriented orthogonal
basis in N for arbitrary non zero n; orthogonal to Ker A:

Proof. There is a positively oriented orthonormal basis (n1 ; n2; n3) su h

that A = jAjn1 ^ n2 (and so A
n3 = 0); then j(A) = jAjn3 from whi h (i )
and (ii ) follow immediately. Moreover, we an hoose n1 := jnnj where n is an
arbitrary non-zero ve tor orthogonal to Ker A:

The kernel of a non-zero A is one-dimensional; a ording to (i ), j(A) spans

the kernel of A: The one-dimensional ve tor spa e Ker A will be oriented by
j(A):
Sin e j is linear, (ii ) is equivalent to j(A)
j(B ) = A  B for all A and B
whi h an also be proved dire tly using that A = jAjk ^ n; B = jB jr ^ n:

3.14. Denition. The map

N  N ! N;

(k; n) 7! k  n :=

j(k ^ n)

is alled the ve torial produ t.

It is evident from the properties of j that the ve torial produ t is an antisymmetri bilinear mapping, k  n = 0 if and only if k and n are parallel, k  n is
orthogonal to both k and n;

jk  nj2 = jkj2 jnj2

(k
n)2 :

If k and n are not parallel then k; n and k  n form a positively oriented

basis in N; moreover, if (n1 ; n2 ; n3 ) is a positively oriented orthonormal basis
in N then
n1  n2

Proposition.
(i )
or, equivalently,
(ii )

= n3 ;

n2  n3

= n1 ;

j([A; B) = j(A)  j(B)

j(A) ^ j(B) =
A
n = j(A)  n

n3  n1

= n2 :

(A; B 2 N ^ N)

[A; B ;
(A 2 N ^ N; n 2 N)

whi h implies

A
(B ) =

j([A; B)

(A; B 2 N ^ N):

Proof. There is an orthonormal basis fn1; n2; n3 g in N su h that A =

jAjn1 ^ n2 :
(i ) n1 ^ n2 ; n2 ^ n3 and n3 ^ n1 form a basis in N ^ N; thus it is su ient

to onsider them in the role of B ; then a simple al ulation yields the desired
result.
(ii ) Take n1 ; n2 and n3 in the role of n:
As a onsequen e of our results we have
(k  n)
r = (n  r)
k = (r  k)
n = "(r; k; n)
and
for all k; n; r 2 N:

(k  n)  r = (k
r)n (n
r)k

3.15. The Levi{Civita tensor establishes another linear bije tion as well:
3
3
jo : ^ N ! R;
^
k 7! "(k1 ; k2 ; k3 ):
i=1 i

It is quite trivial that jo 1 () = " for all  2 R:

3.16. An orthogonal map R : E ! E (also regarded as an orthogonal map

N ! N; see 2.1) sends orthogonal bases into orthogonal ones, preserves and

hanges orientation a ording to whether detR = 1 or detR = 1: In view of

IV.3.18,


3
"  R = (detR)":
Then one proves without di ulty that

j(R
k ^ R
n) = (detR)R
j(k ^ n)

(k; n 2 N):

Sin e R
k ^ R
n = R
(k ^ n)
R 1 ; the previous result an be written in

the form
j(R
A
R 1 ) = (detR)R
j(A)
(A 2 N ^ N):
Moreover,

3
3
jo ^ R
ki = (detR) jo ^ ki :
i=1
i=1

3.17. In the usual way, the linear bije tion j an be lifted

to a linear bije tion
x

E ^ E ! E
D; dened by x ^ y 7! j m ^ my m2
where m is an arbitrary non-zero element of D:
Similarly, the linear bije tion jo an be lifted to a linear bije tion


3 xi
3
^
xi 7! jo ^
m3 :
i=1 m
i=1
We have utilized here that E = N
D: Evidently, similar formulae are valid
for N
A where A is an arbitrary one-dimensional ve tor spa e.

^3 E !
D;

3.18. Let us onsider the Eu lidean ve tor spa e (R3 ; R; B) where

B(x; y) =

3
X
i=1

xi yi =: x
y

(i.e. B = H0 in the notation of
1.7).

The identi ation R3  R3  is the usual one: the fun tional orresponding
to x and represented by the usual matrix multipli ation rule oin ides with x: In
ustomary notations x onsidered to be a ve tor has the omponents (x1 ; x2 ; x3 )
and x onsidered to be a ove tor has the omponents (x1 ; x2 ; x3 ); the previous
statement says that xi = xi ; i = 1; 2; 3:
That is why in this ase one usually writes only subs ripts.
The adjoint of a 3  3 matrix (as a linear map R3 ! R3 ) oin ides with the
transpose of the matrix.
R3 and R are endowed with the usual orientations: the naturally ordered
standard bases are taken to be positively oriented.
The Levi{Civita tensor is given by a matrix of three indi es:
" = ("ijk j i; j; k = 1; 2; 3) ;

"(x; y; z) =
8
>
<

3
X

i;j;k=1

"ijk xi yj zk ;

1 if ijk is an even permutation of 123

1 if ijk is an odd permutation of 123
"ijk =
>
:
0 otherwise
Then it is an easy task to show that
0

j (Ljk j j; k = 1; 2; 3) = 
j

1 (xk j k = 1; 2; 3) =

3
X

1
" L j i = 1; 2; 3A ;
2 j;k=1 ijk jk

3
X
k=1

"ijk xk j i; j = 1; 2; 3 ;

in other notation,

1 (x1 ; x2 ; x3 ) = 

Moreover,

xy

=

3
X
j;k=1

x3
0
x1

0
x3
x2

x2
x1 A :
0
1

"ijk xj yk j i = 1; 2; 3A :

3.19. Consider the Eu lidean ve tor spa e (E; D; b); dim E = 3:

A linear oordinatization K of E is alled orthogonal if it orresponds to an
ordered orthogonal basis (e1 ; e2 ; e3 ) normed to
an m 2 D:
ei
Sin e the dual of the basis is m2 j i = 1; 2; 3 (see 3.4), we have
K
x=

i
x
j i = 1; 2; 3
m2

e

=: x1 ; x2 ; x3 :



Consider the identi ation E  D
D
E  D
ED ; then x; as an element
of the dual of D
ED ; has the oordinates


x


ei
j
i
=
1
;
2
;
3
=: (x1 ; x2 ; x3 ) :
m2

We see, in a ordan e with the previous paragraph, that xi = xi (i = 1; 2; 3)

and we an use only subs ripts.
Then all the operations regarding the Eu lidean stru ture an be represented
by the orresponding operations in (R3 ; R; B); e.g.
| the b-produ t of elements x; y of E is omputed by the inner produ t of
their oordinates in R3 :
if K
x = (x1 ; x2 ; x3 ) and K
y = (y1 ; y2 ; y3 )
i.e.

x=

3
X
i=1

xi ei ;

then x
y =

3
X
i=1

3
X
i=1
!

xi yi

yi ei

m2 ;

| the matrix in the oordinatization of an adjoint map will be the transpose

of the matrix representing the linear map in question:
then

K
L
K 1

= (Lik j i; k = 1; 2; 3)
>
1
K
L
K = (Lki j i; k = 1; 2; 3) ;

if

| if both E and D are oriented and the basis establishing the oordinatization
is positively oriented then the ve torial produ t an be omputed by the ve torial
produ t of oordinates:
0

3
X

K
(x  y ) = 
"ijk xj yk j i = 1; 2; 3A :
j;k=1

These statements fail, in general, for a non-orthogonal oordinatization.

3.20. Let (v1 ; v2; v3) be an arbitrary ordered basis in E; hose a positive
element m of D and put
bik :=

vi
vk
m2


b
(vi ; vk )
:=
2R

m2

(i; k = 1; 2; 3):

The dual of the basis an be represented by ve tors r1 ; r2 ; r3 in D

ED | usually
alled the re ipro al system of the given basis | in su h a way that
r i
vk

= ik

It is not hard to see that

r1

:=

(i; k = 1; 2; 3):

"(
; v2 ; v3 )

; et .

where
:= "(v1 ; v2 ; v3 ):
Put
bik := (ri
rk )m2 2 R
(i; k = 1; 2; 3):
Let us take the oordinatization K of E dened by the basis
(v1 ; v2 ; v3 ): Now we must distinguish between subs ripts
and supers ripts.
We


agree to write the elements of R3 in the form xi and the elements of R3  in
the form (xi ) : Then




K
x = r i
x =: xi :


Consider the identi ation E  D
D
E  D
ED ; then mvi2 j


i = 1; 2; 3 is an ordered basis in D
ED and x; as an element of the dual of
E
D
D ; has the oordinates

vi 
x
2 =: (xi ) :
m

Writing x =

3
P
xk vk

i=1

xi =

3
P
k=1

xk rk m2 we nd that

3
X
k=1

i.e., in general, xi 6= xi :
Now if

K
x=

i.e.
x=

bik xk ;

xi

3
X
i=1

xk =

i=1

bki xi ;

K
y

and

xi vi ;

3
X

3
X
i=1

= yi

y i vi

then
0

x
y

=

3
X

i;k=1

bik

xi yk A m2

3
X
k=1

xk yk m2

3
X
i=1

xi y

2
i m :

3.21. Exer ises

In the following we keep assuming that dim E = 3:
1. Let A be a non-zero element of A(b): If x is a non-zero ve tor in E;

orthogonal to Ker A; then

(i ) A2
x = jAj2 x;
(ii ) jA
xj = jAjjxj;
(iii ) A = (Aj
xxj)2^x :
2. Show that Ker (A2 ) = Ker A for A 2 N ^ N  A(b):
3. Prove that
(i )
(ii )
(iii )

3
X
k=1

3
X

j;k=1

"ijk "rst =ir js kt + is jt kr + it jr ks

it js kr is jr kt ir jt ks ;
"ijk "rsk =ir js
"ijk "rjk =2ir :

is jr ;

4. Let A and B be one-dimensional ve tor spa es. Dene the ve torial

produ t
(N
A)  (N
B) ! N
A
B:

4. Minkowskian ve tor spa es

4.1. A pseudo-Eu lidean ve tor spa e (V; B; h) is alled Minkowskian if

dim V > 1 and :(h) = 1:
For a lear distin tion, in the following (M; I; g) denotes a Minkowskian ve tor
spa e, and
dim M = 1 + N
where N  1:
We all attention to the fa t that g is usually alled a Lorentz metri ; sin e
g does not dene a metri (distan es and angles, see later) we prefer to all it a
Lorentz form.
The notations introdu ed for pseudo-Eu lidean ve tor spa es will be used, e.g.
x
y

:= g(x; y)

note that if x; y 2 M and u; v 2 MI then

x
y

Moreover, we put

2 I
I;

u
x2
x2

I;

(x; y) 2 M);
u
v

2 R:
(x 2 M):

:= x
x

In ontradistin tion to Eu lidean spa es, here we keep saying g-orthogonal,

g-adjoint et .

The elements of a g-orthogonal basis will be numbered from 0 to N; in su h

a way that for fe0 ; e1 ; : : : ; eN g we have e0 2 < 0; ei 2 > 0 if i = 1; : : : ; N:

4.2. Re all that there is a anoni al orientation on I

I; hen e it makes sense
that an element of I
I is positive or negative. Let us introdu e the notations
S :=fx 2 Mj x2 > 0g;
T :=fx 2 Mj x2 < 0g;
L :=fx 2 Mj x2 = 0; x 6= 0g

S0 := S [ f0g;
T0 := T [ f0g;
L0 := L [ f0g:

The elements of S0 ; T and L are alled spa elike, timelike and lightlike ve tors,
respe tively.
Neither of S0 ; T0 and L0 is a linear subspa e.

The bilinear map g is ontinuous (see VI.3.1). Thus S and T are open subsets,
L0 is a losed subset.

4.3. Take a non-zero element x of M: The Lorentz form g is non-degenerate,

hen e the linear map M ! I
I; y 7! x
y is a surje tion, i.e. it has an
N -dimensional kernel. In other words,
Hx := fy 2 Mj x
y = 0g
is an N -dimensional linear subspa e of M: Let gx be the restri tion of g onto
Hx  Hx; it is an I
I-valued symmetri bilinear map.
(i ) Suppose x 2 S or x 2 T: Then x is not in Hx : Rx and Hx are omplementary subspa es. As a onsequen e, gx is non-degenerate, i.e. (Hx ; I; gx) is
an N -dimensional pseudo-Eu lidean ve tor spa e. Thus there is a gx -orthogonal
basis in Hx ; su h a basis, supplemented by x; will be a g-orthogonal basis in M:
| if x is in S then x2 > 0; so one and only one element of a gx -orthogonal
basis belongs to T; the other ones belong to S: Consequently, (Hx ; I; gx ) is an
N -dimensional Minkowskian ve tor spa e.
| if x is in T then x2 < 0; so all the elements of a gx -orthogonal basis belong
to S: Consequently, Hx  S0 and (Hx ; I; gx ) is an N -dimensional Eu lidean
ve tor spa e.
(ii ) Suppose x 2 L: Then x itself is in Hx ; in other words, Rx is ontained in
Hx: One annot give naturally a subspa e omplementary to Hx: Moreover, gx
is degenerate.
Let e0 be an element of T; let s be an element of I su h that e20 = s2 : As we
have seen, He0 is ontained in S0 ; so e0
x 6= 0; and e1 := ee00

ex0 x e0 belongs
to S; e0
e1 = 0 and e1 2 = s2 : fe0 ; e1 g an be ompleted to a g-orthogonal basis
fe0 ; e1 ; : : : ; eN g; normed to s; of M: The ve tor x is a linear ombination of e0
and e1 ; thus fe2; : : : ; eN g is ontained in Hx and fx; e2; : : : ; eN g is a basis of
Hx:
This has the immediate onsequen e that every element of Hx whi h is not
parallel to x belongs to S0 :
4.4. It follows from 4.3(i ) that if x 2 T then x
y 6= 0 for all y 2 T and for
all y 2 L:
Moreover, the results in the pre eding paragraph imply that
| there are N -dimensional linear subspa es in S0 ;
| there are at most one-dimensional linear subspa es in T0 and L0 ,
| there is a one-to-one orresponden e between N -dimensional linear subspa es in S0 and one-dimensional linear subspa es in T0 in su h a way that the
subspa es in orresponden e are g-orthogonal to ea h other.
4.5. The identi ation

M  M
I
I

(see 1.3) is a fundamental property of the Minkowskian ve tor spa e

(M; I; g):
The dual of a g-orthogonal
basis fe0 ; e1 ; : : : ; eN g; normed to s 2 I; in this

identi ation be omes es02 ; es21 ; : : : ; esN2 :
A ordingly, ni := esi (i = 0; 1; : : : ; N ) form a g-orthonormal basis in MI :
n0
n0

= 1;

ni
nk

= ik

(i; k = 1; : : : ; N ):

The orresponding dual basis in the identi ation MI

n0 ; n1 ; : : : ; nN g:
For all x 2 M we have
x=

4.6.

x; y

(n0
x)n0 +

N
X
i=1

M
 is
I

(ni
x)ni :

The following relation will be a starting point of important results. If

x is not parallel to y ; z 2 T; then

2 T [ L;

2(x
y)(y
z)(z
x) < (x
z)2 y2 + (y
z)2 x2  0:

This is implied by the simple fa t that a := xy

zz x y is g-orthogonal to z;
thus a is in S : a2 > 0:

4.7. Sin e I
I is anoni ally oriented (see IV.5.4), the absolute value of its
elements makes sense.
Proposition (reversed Cau hy inequality). If x; y 2 T then
jx
yj  jx2 jjy2 j > 0
p

and equality holds if and only if x and y are parallel.

Proof. Put z := x in the previous formula, and re all that we have the
square root mapping from (I
I)
(I
I) into I
I:
In general, the right-hand side of this equality annot be written in a simpler
form be ause jxj and jyj make no sense, unless I is oriented.

4.8. Denition. The elements x and y of T have the same arrow if x
y < 0:
Proposition. Having the same arrow is an equivalen e relation on T and

there are two equivalen e lasses ( alled arrow lasses).

Proof. The relation having the same arrow is evidently re exive and symmetri . Suppose now that x and y as well as y and z have the same arrow.

Then 4.6 implies that x and z have the same arrow as well, hen e the relation
is transitive.
Let x be an element of T: It is obvious that x and x have not the same
arrow: there are at least two arrow lasses. On the other hand, sin e x
y 6= 0
for all y 2 T; y and x or y and x have the same arrow: there are at most two
arrow lasses.

4.9. Proposition. The arrow lasses of T are onvex ones, i.e. if x and y
have the same arrow then x +  y is in their arrow lass for all ;  2 R+ :
Proof. It is quite evident that (x + y)2 < 0 and (x + y)
x < 0; thus
x +  y is in T; moreover, x +  y and x have the same arrow.
The arrow lasses are open subsets of M be ause the arrow lass of x 2 T is
fy 2 Tjx
y < 0g:
4.10. Suppose now that I is oriented. Then we an take the square root of
non-negative elements of I
I; so we dene the pseudo-length of ve tors:
p

jxj := jx2 j

(x 2 M):

The length of ve tors in Eu lidean ve tor spa es has the fundamental properties listed in 3.3. Now we nd that
(i ) jxj = 0 if x = 0 but jxj = 0 does not imply x = 0;
(ii ) jxj = jjjxj for all  2 R;
(iii ) there is no denite relation between jx + yj and
jxj + jyj :
| if H  S0 is a linear subspa e then H; I; gjHH is a Eu lidean
ve tor spa e, onsequently, for x; y 2 H the triangle inequality jx + yj  jxj + jyj
holds,
| for ve tors in T a reverse relation an hold, as follows.
Proposition (reversed triangle inequality). If the elements x and y of T have
the same arrow then
jx + yj  jxj + jyj
and equality holds if and only if x and y are parallel.
Proof. A ording to the previous statement x + y belongs to T; thus we an
apply the reversed Cau hy inequality:

jx + yj2 =

(x + y)2 =jxj2

2(x
y) + jyj2 
p
jxj2 + 2 jxj2 jyj2 + jyj2 = (jxj + jyj)2 :

The triangle inequality and \non-zero ve tor has non-zero length" are indispensable properties of a length; that is why we use the name pseudo-length.

4.11. Denition. The elements x and y of L have the same arrow if x
y < 0:
Proposition. Having the same arrow is an equivalen e relation on L and

there are two equivalen e lasses ( alled arrow lasses).

Proof. Argue as in 4.8.

4.12. Now we relate the arrow lasses of L to those of T: It is evident that

the elements x and y of L have the same arrow if and only if x +  y are in T
and have the same arrow for all ;  2 R+ :
Proposition. (i ) Let x; y 2 L; z 2 T: Then x and y have the same arrow
if and only if x
z and y
z have the same sign (in the ordered one-dimensional
ve tor spa e I
I):
(ii ) Let x 2 L; y; z 2 T: Then y and z have the same arrow if and only if
x
y and x
z have the same sign.
Proof. Apply the inequality in 4.6.
As a onsequen e, the arrow lasses of T and those of L determine ea h other
uniquely. We say that the elements x of L and y of T have the same arrow if
x
y < 0: A ording to the previous proposition, if we sele t an arrow lass T!
from T then there is an arrow lass L! in L su h that all the elements of T!
and L! have the same arrow:
L! = fy 2 Lj x
y < 0; x 2 T! g:
It an be shown that L! [ f0g is the boundary of T! :

4.13. We say that (M; I; g) is arrow-oriented or an arrow orientation is

asso iated with g if we sele t one of the arrow lasses of T: More pre isely, an
arrow-oriented Minkowskian ve tor spa e is (M; I; g; T! ) where (M; I; g) is a
Minkowskian ve tor spa e and T! is one of the arrow lasses of T:
A linear isomorphism between arrow-oriented Minkowskian ve tor spa es is
alled arrow-preserving or arrow-reversing if it maps the hosen arrow lasses
into ea h other or into the opposite ones, respe tively.
4.14. In the following we assume that M and I are oriented and g is arroworiented; moreover
dim M = 4:
We introdu e the notation
V(1) :=

M j u2 =
u2
I

1; u
I

+  T! :

If u 2 V(1) then

I := futj t 2 Ig  T0;
Eu := fx 2 Mj u
x = 0g  S0
u

are omplementary subspa es. The orresponding proje tions of x 2 M in u

I
and in Eu are
(u
x)u
and
x + (u
x)u:
Let bu denote the restri tion of g onto Eu  Eu: A ording to 4.3.(i ) |
us 2 T for s 2 I and Eu = Hsu | we have that (Eu ; I; bu ) is a three-dimensional
Eu lidean ve tor spa e.

4.15. We shall examine the stru ture

of g-antisymmetri linear maps of M:
As we know (see 2.9) A(g)  MI ^ MI is a six-dimensional ve tor spa e endowed

(see 2.10) with a real-valued non-degenerate symmetri bilinear form:

1
1
Tr(H >
G) = Tr(H
G):
2
2
In parti ular, for k1 ; k2 ; n1 ; n2 2 MI we have
H  G :=

(k1 ^ k2 )  (n1 ^ n2 ) = (k1
n1 )(k2
n2 ) (k1
n2 )(k2
n1 ):

If fn0 ; n1 ; n2 ; n3 g is a g-orthonormal basis of MI then
n0 ^ n1 ;

n1 ^ n2 ;

n0 ^ n2 ;

n2 ^ n3 ;

n0 ^ n3 ;
n3 ^ n1

onstitute a basis in MI ^ MI (see IV.3.15). We an takeu := n0 2 V(1); then every

P
g-antisymmetri map an be written in the form iP=13 iu ^ ni + iP=13 k<i
ki nk ^ ni :
The ve tors n1 ; n2 ; n3 span
the three-dimensional Eu lidean ve tor spa e EIu


(more pre isely, EIu ; R;
); hen e a ording to the results of the previous hapter
there are a real number  and unit ve tors k and n; g-orthogonal to ea h other
3
3 P
P
P
ki nk ^ ni =  k ^ n: Furthermore, i u ^ ni =
and to u su h that
u^

3
P
i=1

i=1

i=1 k<i

i ni ; thus we arrive at the following result.

Proposition. Let HE be an element of MI ^ MI : Then for all u 2 V(1) there

are ;  2 R; r; k; n 2 Iu ; r2 = k2 = n2 = 1; k
n = 0 su h that
H

=  u ^ r +  k ^ n:

Observe that then

H H

= 2 +  2 :

4.16. Proposition. Take an H 2 MI ^ MI in the form given by the previous

proposition. Then Ker H = f0g if and only if  6= 0;  6= 0 and r is linearly

independent from k and n:

Proof. If r is linearly independent from k and n then u; r; k; n are linearly
independent ve tors. Furthermore, if neither of  and  is zero then for all
x2

H
x = (r
x)u

(u
x)r +  (n
x)k  (k
x)n = 0

implies r
x = u
x = n
x = k
x = 0; as a onsequen e, x = 0: This means

that Ker H = f0g:
If  = 0 then H
u = 0; if  = 0 then H
m = 0 for m 2 MI ; u
m = 0;
r
m = 0: If  6= 0 and  6= 0 but r is a linear ombination of k and n then
H
m = 0 for m 2 M
I ; u
m = k
m = n
m = 0: This means that Ker H 6= f0g:
Sin e if k0 and n0 are g-orthogonal unit ve tors in the plane spanned by k
and n (do not forget that EIu is a Eu lidean ve tor spa e) then k0 ^ n0 = k ^ n;
we an hoose n = r if Ker H 6= f0g; then for all u 2 V(1) there are ;  2 R
and k; n 2 EIu ; k2 = n2 = 1; k
n = 0 su h that
H

= (u +  k) ^ n:

4.17. Proposition. Suppose H 2 MI ^ MI ; Ker H 6= f0g and put jH j :=

jH  H j: Then
(i ) H  H > 0 if and only if there are a u 2 V(1); k; n 2 EIu ; k2 = n2 = 1;
k
n = 0 su h that
H = j H j k ^ n:

(H is the antisymmetri tensor produ t of two g-orthogonal spa elike ve tors.)

(ii ) H  H < 0 if and only if there are a u 2 V(1); an n 2 EIu ; n2 = 1 su h
that
H = j H j u ^ n:
(H is the antisymmetri tensor produ t of a timelike ve tor and a spa elike
ve tor, g-orthogonal to ea h other.)
(iii ) H  H = 0; H 6= 0 if and only if there are w; n 2 MI ; w 6= 0; w2 = 0;
n2 = 1; w
n = 0 su h that
H = w ^ n:

(H is the antisymmetri tensor produ t of a lightlike ve tor and a spa elike

ve tor, g-orthogonal to ea h other.)
Proof. Let us write the formula of the pre eding paragraph in the form
H = (u0 +  k0 ) ^ n0 ; then H  H = 2 +  2 :
 u0 + k0
u0 +  k0
Put
u :=
(i )
;
k
:=
; n := n0 :
2 +  2
2 +  2
u0 +  k0
(ii )
; n := n0 :
Put
u := 2
 2
(iii )
Put
w :=u0 +  k0 ; n := n0 :

4.18. (i ) We see that Ker H 6= f0g and H  H > 0 is equivalent to the

statement that there is a u 2 V(1) su h that H
u = 0:
Note that then H 3 = jH j2 H :
(ii ) On the ontrary, if Ker H 6= f0g and H  H < 0 then H 3 = jH j2 H :
4.19. A ording to our onvention introdu ed in 4.1, let us number the
oordinates of elements of R1+3 from 0 to 3 and let us onsider the Minkowskian
ve tor spa e (R1+3 ; R; G) where
G(x; y) = x0 y0 +

3
X
i=1

xi yi =: x
y

(i.e. G = H1 in the notation of 1.7).

Now the identi ation R1+3  R1+3  indu ed by G is des ribed in usual
notations as follows. x 2 R1+3 regarded as a ve tor has the omponents
(x0 ; x1 ; x2 ; x3 ); x regarded as a ove tor has the omponents (x0 ; x1 ; x2 ; x3 );
and the values of the linear fun tional x are omputed by the usual matrix
multipli ation:
x
y

Then we have

3
X
i=0

xi yi :

x0 = x0 ;
xi = xi
(i = 1; 2; 3):
i
As usual, we apply the symbols (x ) and (xi ) for the ve tors and ove tors
and we a ept the Einstein summation rule: a summation from 0 to 3 is to be
arried out for equal subs ripts and supers ripts.
Introdu ing
8
>
< 1 if i = k = 0
1 if i = k 2 f1; 2; 3g
gik := gik :=
>
:
0 if i 6= k

we an write that

xi = gik xk ;

xi = gik xk

(summation!).

Observe that gik = G(i ; k ) where f0 ; 1; 2 ; 3 g is the standard basis of
R1+3 :
A ording to the identi ation indu ed by G; the dual of the standard basis
f0; 1 ; 2 ; 3 g is f 0; 1 ; 2 ; 3 g:
It is useful to regard G as the diagonal matrix in whi h the rst ( \zeroth")
element in the diagonal is 1 and the other ones equal 1:
For the G-adjoint L> of the linear map (matrix) L we have
L>

= G
L
G

where L is the transpose of L (see 1.7).

A linear map L : R1+3 ! R1+3 is given by its matrix Li k ;


a linear map P : (R1+3 ) ! (R1+3 ) is given by its matrix Pi k et . see
IV.1.6.
For the transpose of L : R1+3 ! R1+3
(L )i k = Lk i ;
holds, thus for the G-adjoint we have
(L> )i k = gim Lnm gnk

(summation!).

Consequently, a G-antisymmetri linear map has the form

0
B


0
1
2
3

1
0
3
2

2
3
0
1

3 1
2 C
1 A
0

If (xi ) is in T; i.e. xi
xi < 0 then x0 6= 0: It is not hard to see that (xi )

and (yi ) have the same arrow if and only if x0 and y0 have the same sign. As a
onsequen e, an arrow lass is hara terized by the sign of the zeroth omponent
of its element. One usually takes the arrow orientation in su h a way that
T! := f(xi ) 2 R1+3 j xi xi < 0 ; x0 > 0g:

4.20. Consider the four-dimensional Minkowskian ve tor spa e (M; I; g): A

linear oordinatization K of M is alled g-orthogonal if it orresponds to an
ordered g-orthogonal basis (e0 ; e1 ; e2 ; e3 ) normed to an s 2 I: A ording to the

identi ation
thus we have

M  IM
I ; the basis in question has the dual

K
x=

;
;
;

s2
s2
s2
s2

 e x e x e x e x
1
2
3
0

=: xi

e0 ; e1 ; e2 ; e3
;
s2 s2 s2 s2

(x 2 M):



Consider the identi ation M  I
I
M  IM

I ; then x; as an element
M
of the dual of I
I ; has the oordinates


ei
x
2 j i = 0; 1; 2; 3
s

=: (xi ) :

We see, in a ordan e with the previous paragraph, that x0 = x0 and xi = xi

(i = 1; 2; 3):
Then all the operations regarding the Minkowskian stru ture an be represented by the orresponding operations in (R1+3 ; R; G); e.g.
| the g-produ t of elements x; y of M is omputed by the G-produ t of their
oordinates in R1+3 :
if
i.e.

K
x=
x=

3
X
0=1

then

xi

and

xi ei ;
x
y

K
y

3
X
0=1

= yi

y i ei

x0 y0 +

3
X
i=1

xi yi

s2 ;

| the matrix in the oordinatization of a g-adjoint map will be the G-adjoint

of the matrix representing the linear map in question:


if K
L
K 1 = Li k j i; k = 1; 2; 3


then K
L>
K 1 = Lk i j i; k = 1; 2; 3 :

4.21. Let (v0 ; v1; v2; v3 ) be an arbitrary ordered basis in M; hoose a positive
element s of I and put
gik :=

vi
vk
s2


g
(vi ; vk )
:=
2R

s2

(i; k = 0; 1; 2; 3):

The dual of the basis an be represented by ve tors r0 ; r1 ; r2; r3 in IM

I |
usually alled the re ipro al system of the given basis | in su h a way that
r i
vk

= ik

(i; k = 0; 1; 2; 3):

It is not hard to see that

r0

:=

"(
; v1 ; v2 ; v3 )

et .

where " is the Levi{Civita tensor (see V.2.12) and
:= "(v0 ; v1 ; v2 ; v3 ):

Put
gik := (ri
rk )s2 2 R
(i; k = 0; 1; 2; 3):

Let us take the oordinatization K of M dened by the basis (v0 ;

Then




K
x = r i
x =: xi :


; then sv2i j
Consider the identi ation M  I
I
M  IM

I


i = 0; 1; 2; 3 is an ordered basis in IM

I ; and x; as an element of the dual of

M ; has the oordinates
I
I

vi 
x
2 =: (xi ) :
v1 ; v2 ; v3 ):

Writing x =

3
P
i=0

xk vk =

3
P
k=0

xk rk s2 we nd that

xi = gik xk ;
Now if

K
x=

i.e.
x=

then

x
y

xi

3
X
i=0

xk = gki xi

xi vi ;

(summation!).
K
y

and
y

= yi

3
X
i=0

yi vi

= gik xi yk s2 = xk yk s2 = xi yi s2 :

4.22. Exer ises

1. Let T! and L! be the arrow lasses orresponding to ea h other a ording
to 4.12. Prove that
T! + L! =T! ;
L! + L! =T! [ L! :
2. If x is in T! and y is in S then x + y is not in T : (Hint: suppose
(x + y) 2 T! and use T! + T! = T! :)

3. If H is a linear subspa e in S0 then H; I; gjHH is a Eu lidean ve tor

spa e. Consequently, the length of ve tors and the angle between ve tors in H
makes sense. Sin e every x 2 S0 belongs to some linear subspa e in S0 (e.g. to
the linear subspa e generated by x); the length of every element in S0 makes
sense. On the other hand, if x; y 2 S0 ; the linear subspa e generated by x and y
need not be ontained in S0; as a onsequen e, the angle between two elements
of S0 may not be meaningful.
Take a g-orthogonal basis fe0; e1 ; : : : ; eN g; normed to s 2 I: Then e1 and
x := 2e1 + e0 are ve tors in S that do not satisfy the Cau hy inequality and the
triangle inequality.
4. Suppose dim M = 4; M and I are oriented. Then the Levi{Civita tensor of
(M; I; g) an be dened by "
:=
^3 ei where (e0 ; e1 ;
i=0 s
e2 ; e3 ) is a positively oriented ordered basis, normed to s 2 I:
Prove that
M ! ^3 M ; n 7! "(
;
;
; n)

and

I
I
J : MI ^ MI ! MI ^ MI ;

k ^ n 7! "(
;
; k; n)

are linear bije tions. Moreover,

J(J(H )) =

H;

H

J(G) = J(G)  G;

thus J(H )  J(G) = H  G for all H ; G 2 MI ^ MI :

5. Give the a tual form of the previous bije tions in the ase (R1+3 ; R; G):
6. Let " be the Levi{Civita tensor of the Minkowskian ve tor spa e (M; I; g);
dim M = 4: If u 2 V(1) then "(u;
;
;
) is the Levi{Civita tensor of the threedimensional Eu lidean ve tor spa e (Eu ; I; bu ) where bu is the restri tion of g
onto Eu  Eu :

VI. AFFINE SPACES

1. Fundamentals

1.1. Denition. An ane spa e is a triplet (V; V; ) where

(i ) V is a non-void set,
(ii ) V is a ve tor spa e,
(iii ) is a map from V  V into V; denoted by
(x; y) 7! x y;

having the properties

1) for every o 2 V the map Oo : V ! V; x 7! x o is bije tive,
2) (x y) + (y z ) + (z x) = 0 for all x; y; z 2 V:
Oo is often alled the ve torization of V with origin o:
As usual, we shall denote an ane spa e by a single letter; we say that V is
an ane spa e over the ve tor spa e V and we all the map subtra tion.
The dimension of an ane spa e V is, by denition, the dimension of the
underlying ve tor spa e V: V is oriented if V is oriented (in this ase V is
ne essarily a nite-dimensional real ve tor spa e).
Proposition. Let V be an ane spa e. Then
(i ) x y = 0 if and only if x = y (x; y 2 V);
(ii ) x y = (y x) (x; y 2 V);
(iii ) for a natural number n  3 and x1 ; x2 ; : : : ; xn 2 V;
(x1

x2 ) + (x2

x3 ) + : : : : + (xn

x1 ) = 0:

Proof. (i ) Put z := x; y := x in 2) of the above denition to have x x = 0:

Property 1) says then that x y 6= 0 for x 6= y:
(ii ) Put z := x in 2) and use the previous result.
(iii ) Starting with 2) we an prove by indu tion.
As a onsequen e, we an rearrange the parentheses as follows:
(x y) + (u v) = (x v) + (u y)

(x; y; u; v 2 V):

(1)

1.2. Observe that the sign in (ii ) of the previous proposition denotes two
dierent obje ts. Inside the parantheses it means the subtra tion in the ane
spa e, outside it means the subtra tion in the underlying ve tor spa e. This
ambiguity does not ause onfusion if we are areful. We even nd it onvenient
to in rease a bit the ambiguity.
For given y 2 V; the inverse of the map Oy is denoted by
V ! V;

x 7! y + x:

(2)

Hen e, by denition,
(x; y 2 V);

(3)

(x 2 V; x; y 2 V):

(4)

y + (x y) = x
and a simple reasoning shows that
(x + x) + y = x + (x + y)

Here the symbol + on the left-hand side stands twi e for the operation introdu ed
by (2), on the right-hand side rst it denotes this operation and then the addition
of ve tors.
Keep in mind the followings:
(i ) the sum and the dieren e of two ve tors, the multiple of a ve tor are
meaningful, they are ve tors;
(ii ) the dieren e of two elements of the ane spa e is meaningful, it is a
ve tor (sums and multiples make no sense);
(iii ) the sum of an ane spa e element and of a ve tor is meaningful, it is an
element of the ane spa e.
A ording to (1){(4), we an apply the usual rules of addition and subtra tion
paying always attention to that the operations be meaningful; for instan e, the
rearrangement (y + x) y in (3) makes no sense.

1.3. Linear ombinations of ane spa e elements annot be dened in general,

for multiples and sums make no sense. However, a good tri k allows us to dene
onvex ombinations.
Proposition. Let x1 ; : : : ; xn be elements of thePn ane spa e V and let
k = 1: Then there is
1 ; : : : ; n be non-negative real numbers su h that
k=1
a unique xo 2 V for whi h
n
X
k=1

k (xk

xo ) = 0:

()

Proof. Let x be an arbitrary element of V; then a simple

al ulation based
n
P

on xk xo = (xk x) + (x xo ) shows that xo := x + k (xk x) satises

k=1
equality (): Suppose yo is another element with this property. Then
n

0 = X k (xk
=

k=1
n
X
k=1

n
X

xo )

k=1

k (yo xo ) = yo

k (xk

yo ) =

n
X
k=1

k ((xk

xo ) (xk

yo )) =

xo :

Remove formally the parentheses in () to arrive at the following denition.

Denition. The element xo in the previous proposition is alled the onvex
ombination of the elements x1 ; : : : ; xn with oe ients 1 ; : : : ; n and is denoted
by
n
X

k=1

k xk :

Correspondingly we an dene onvex subsets and the onvex hull of subsets

in ane spa es as they are dened in ve tor spa es.
n
P
If 1 ; : : : ; n are non-negative real numbers,  :=
k > 0; then we an take
k=1

the onvex ombination of x1 ; : : : ; xn with the oe ients k := k (k = 1; : : : ; n)

whi h will be denoted by
n
P
k xk
k=1
:
n
P
k
k=1

1.4. (i ) A ve tor spa e V; endowed with the ve torial subtra tion, is an ane
spa e over itself.
(ii ) If M is a non-trivial linear subspa e of the ve tor spa e V and x 2 V;
x2
=M then x + M := fx + yj y 2 Mg endowed with the ve torial subtra tion is
an ane spa e but is not a ve tor spa e regarding the ve torial operations in V:
(iii ) If I is an arbitrary non-void set and Vi is an ane spa e over Vi (i 2 I )
then  Vi ; endowed with the subtra tion
i2I
(xi )i2I
is an ane spa e over

 Vi :

i2I

(yi )i2I := (xi

yi )i2I

1.5. Denition. A non-void subset S of an ane spa e V is alled an ane

subspa e if there is a linear subspa e S of V su h that fx yj x; y 2 Sg = S:
S is alled dire ted by S and the dimension of S is that of S:
One-dimensional and two-dimensional ane subspa es of a real ane spa e
are alled straight lines and planes, respe tively. Hyperplanes are ane subspa es having the dimension of V but one, in a nite dimensional ane spa e
V:
Two ane subspa es are said to be parallel if they are dire ted by the same
linear subspa e.
An ane subspa e S dire ted by S; endowed with the subtra tion inherited
from V; is an ane spa e over S:
If S is a linear subspa e of V and x 2 V then x + S := fx + sj s 2 Sg is the
unique ane subspa e ontaining x and dire ted by S:
Points of V are zero-dimensional ane subspa es.
1.6. A pseudo-Eu lidean (Eu lidean, Minkowskian) ane spa e is a triplet
(V; B; h) where V is an ane spa e over the ve tor spa e V and (V; B; h) is a
pseudo-Eu lidean (Eu lidean, Minkowskian) ve tor spa e.
1.7. Exer ises
1. Prove that the following denition of ane spa es is equivalent to that
given in 1.1.
A triplet (V; V; +) is an ane spa e if
(i ) V is a non-void set,
(ii ) V is a ve tor spa e,
(iii ) + is a map from V  V into V; denoted by
(x; x) 7! x + x

having the properties

1) (x + x) + y = x + (x + y)
(x 2 V; x; y 2 V);
2) for every x 2 V the map V ! V; x 7! x + x is bije tive.
2. Let V be an ane spa e over V: Let V=N denote the set of ane subspa es
in V; dire ted by a given linear subspa e N of V: If M is a linear subspa e
omplementary to N then V=N be omes an ane spa e over M if we dene the
subtra tion by
S T := x y

(x 2 S; y 2 T; x y 2 M):

In other words, if P denotes the proje tion onto M along N then

(x + N) (y + N) := P
(x y):

Illustrate this fa t by V := R2 ; N := f(; 0)j  2 Rg; M := f(; m)j

 2 Rg where m is a given non-zero number.
3. Prove that the interse tion of ane subspa es is an ane subspa e, thus
the ane subspa e generated by a subset of an ane spa e is meaningful.
4. Let V be a ve tor spa e over the eld K : Then f1g V is an ane subspa e
of the ve tor spa e K  V:
5. Let I be a one-dimensional oriented ane spa e over the ve tor spa e I:
Then an order an be dened on I by a < b if and only if a b < 0: Dene the
intervals of I:

2. Ane maps

2.1. Denition. Let V and U be ane spa es over V and U; respe tively.
A map L : V ! U is alled ane if there is a linear map L : V ! U su h that
L(y) L(x) = L
(y x)

(x; y; 2 V):

We say that L is an ane map over L: If L is a bije tion, V and U are oriented,
L is alled orientation-preserving or orientation-reversing if L has that property.
The formula above is equivalent to
(x 2 V; x 2 V):

L(x + x) = L(x) + L
x

It is easy to show that the linear map L in the denition is unique.

2.2. Proposition. Let L : V ! U be an ane map. Then

(i ) L is inje tive or surje tive if and only if L is inje tive or surje tive,
respe tively; if L is bije tive then L 1 is an ane bije tion over L 1 ;
(ii ) L = 0 if and only if L is a onstant map;
(iii ) Ran L is an ane subspa e of U; dire ted by Ran L;
(iv ) if Z is an ane subspa e of U; dire ted by Z; and (Ran L) \ Z 6= ; then
1
L (Z)
is
an
ane
subspa e
of
V;
dire ted
by
1
L (Z);
(v ) L preserves onvex ombinations.
Observe that a ording to (iv ), for all u
subspa e of V; dire ted by Ker L:

Ran L; L (fug) is an ane

2.3. Proposition. (i ) If L and K are ane maps su h that K L exists

then K L is ane map over K
L;

(ii ) Let I be a non-void set. If Li : Vi ! Ui (i 2 I ) are ane maps, then


Li is an ane map over  Li ;
i2I
i2I
(iii ) If Li : V ! Ui (i 2 I ) are ane maps then (Li )i2I is an ane map over

(Li )i2I ;
(iv ) If L and K are ane maps from V into U then

K
is an ane map over K
over itself).

L:V
L

! U;

x 7! K (x) L(x)

(re all that the ve tor spa e U is an ane spa e

2.4. (i ) Let V and U be ve tor spa es and onsider them to be ane spa es.
Take a linear map L : V ! U and an a 2 U; then V ! U; x 7! a + L
x is an
ane map over L:
Conversely, suppose L : V ! U is an ane map over the linear map L: Put
a := L(0): Then L(x) = a + L
x for all x 2 V:
Thus we have proved:
Proposition. We an identify the set of ane maps from V into U with
f(a; L)j a 2 U; L 2 Lin(V; U)g = U  Lin(V; U) in su h a way that
(a; L)(x) := a + L
x

(x 2 L):

Su h an ane map (a; L) : V ! U an be represented in more suitable ways,

as follows.
(ii ) V ! K  V; x 7! (1; x) (see Exer ise 1.7.4) is an ane inje tion. We often
nd onvenient to identify V; onsidered to be an ane spa e, with f1g  V:
Take an ane map (a; L) : V ! U and onsider it to be an ane map from
f1gV into f1gU: It an be uniquely extended to a linear map K V ! K U;
(; x) 7! (; a + L
x):
Representing the linear maps from K  V into K  U by a matrix (see IV.3.7),
we an write the extension of the ane map (a; L) in the form


0 :

a L

(iii ) It often o urs that the ve tor spa e V is regarded as an ane spa e
(i.e. we use only its ane stru ture, the subtra tion of ve tors) but the ve tor
spa e U is ontinued to be regarded as a ve tor spa e (i.e. we use its ve torial
stru ture, the sum of ve tors and the multiple of ve tors).
In this ase we identify V with f1g  V and U with f0g  U; and so we an
on eive that (a; L) maps from f1g V into f0g U: This map an be uniquely

extended to a linear map K  V ! K  U; (; x) 7! (0; a + L
x): Then the

ane map (a; L) in a matrix representation has the form


0 :

a L

2.5. Exer ises

1. Let o be an element of the ane spa e V: Then Oo : V ! V; x 7! x o is

an ane map over idV :
Consequently, if V is N -dimensional then there are ane bije tions V ! K N :
2. Let L : V ! V be an ane map and o an element of V: Then Oo L Oo 1
is an ane map V ! V: Using the matrix form given in the pre eding paragraph
show that


1
0
1
Oo L Oo = L(o) o L :

3. Let H : V ! V be an ane map and o an element of V: Then H Oo 1

is an ane map V ! V: Then the ve tor spa e V as the domain of this ane
map is onsidered to be an ane spa e (representing the ane spa e V); and V
as the range is onsidered to be a ve tor spa e. Using the matrix form given in
the pre eding paragraph show that

H Oo 1 =

0
0
H (o) H :


4. The matrix forms of ane maps V ! V is extremely useful for obtaining

the omposition of su h maps be ause we an apply the usual matrix multipli ation rule. Find the omposition of
(i ) (a; L) : f1g  V ! f1g  V and (b; K ) : f1g  V ! f1g  V;
(ii ) (a; L) : f1g  V ! f1g  V and (b; K ) : f1g  V ! f0g  V:
5. Let V be an ane spa e over V:
(i ) If a 2 V then Ta : V ! V; x 7! x + a is an ane map over idV :
(ii ) If L : V ! V is an ane map over idV then there is an a 2 V su h that
L = Ta :
(iii ) For all x; y 2 V we have Oy Ox 1 = Tx y :
6. If L : V ! V is an ane map over idV then there is an o 2 V su h that
L(x) = o (x o) (x 2 V):
7. Let K and L be ane maps between the same ane spa es. Show that
K = L if and only if K L is a onstant map.
8. Let K; L; A : V ! V be ane maps. Show that A K A L = A (K L):

3. Dierentiation

3.1. Let V be a ve tor spa e. A norm on V is a map

k
k : V ! R+0 ; x 7! kxk
for whi h (i ) kxk = 0 if and only if x = 0;
(ii ) kxk = jjkxk for all  2 K ; x 2 V;

kx + yk  kxk + kyk for all x; y 2 V:

The distan e of x; y 2 V is dened to be kx yk ; the map
V  V ! R; (x; y) 7! kx yk
(iii )

is alled the metri s asso iated with the norm.

The reader is supposed to be familiar with the fundamental notions of analysis
onne ted with metri s: open subsets, losed subsets, onvergen e, ontinuity,
et .
It is important that if V is nite-dimensional then all the norms on V are
equivalent, i.e. they determine the same open subsets, losed subsets, onvergent
series, ontinuous fun tions et .
As a onsequen e, in nite-dimensional ve tor spa es | e.g. in pseudoEu lidean ve tor spa es | we an speak about open subsets, losed subsets,
ontinuity et . without giving an a tual norm. Linear, bilinear, multilinear
maps between nite-dimensional ve tor spa es are ontinuous.

3.2. If V is an ane spa e over V and there is a norm on V then

V  V ! R;

(x; y) 7! kx yk

is a metri s on V: Then the open subsets, losed subsets, onvergen e et . are

dened in V:
In the following we deal with nite dimensional real ane spa es; hen e we
speak about the fundamental notions of analysis without spe ifying norms on the
underlying ve tor spa es.
As usual, if V and U are nite-dimensional ve tor spa es, ordo : V
U
denotes a fun tion su h that
(i ) it is dened in a neighbourhood of 0 2 V;
ordo(x)
(ii ) xlim
!0 kxk = 0 for some (hen e for every) norm k
k on V:

3.3. Denition. Let V and U be ane spa es. A map F : V U is

alled dierentiable at an interior point x of Dom F if there is a linear map
DF (x) : V ! U and a neighbourhood N (x)  Dom F of x su h that
F (y) F (x) = DF (x)
(y x) + ordo(y x)

(y 2 N (x)):

DF (x) is the derivative of F at x:

F is dierentiable on a subset S of Dom F if it is dierentiable at every point
of S: F is dierentiable if it is dierentiable on its domain (whi h is ne essarily
open in this ase). F is ontinuously dierentiable if it is dierentiable and
Dom F ! Lin(V; U); x 7! DF (x) is ontinuous.
If the real ane spa es V and U are oriented, a dierentiable mapping F :
V
U is alled orientation-preserving if DF (x) : V ! U is an orientationpreserving linear bije tion for all x 2 Dom F:
The dierentiability of F at x is equivalent to the following: there is a
neighbourhood N of 0 2 V su h that x + N  Dom F and

F (x + x) F (x) = DF (x)
x + ordo(x)

(x 2 N ):

This form shows immediately that DF (x) is uniquely determined.

3.4. If the ane spa es in question are a tually ve tor spa es, i.e. F is a
map between ve tor spa es then the above denition oin ides with the one in
standard analysis. Hen e in the ase of ve tor spa es we an apply the wellknown results regarding dierentiability. Moreover, for ane spa es one proves
without di ulty that
(i ) a dierentiable map is ontinuous;
(ii ) if F : V U and G : U W are dierentiable then G F is dierentiable,
too, and
D(G F )(x) = DG(F (x))
DF (x)
(x 2 Dom (G F ));
(iii ) if F; G : V U are dierentiable then F G : V U; x 7! F (x) G(x)
is dierentiable and

D(F

(x 2 Dom F \ Dom G):

G)(x) = DF (x) DG(x)

(iv ) An ane map L : V ! U is dierentiable, its derivative at every x equals

the underlying linear map:
DL(x) = L
(x 2 V):

3.5. Let V and U be ane spa es. If F : V U is dierentiable and its

derivative map V
Lin(V; U); x 7! DF (x) is dierentiable then F is alled
twi e dierentiable.
Dierentiability of higher order is dened similarly. An innitely many times
dierentiable map is alled smooth.
The se ond derivative of F at x is denoted by D2 F (x); by denition, it is an
element of Lin(V; Lin(V; U)):
The n-th derivative of F at x; Dn F (x) is an element of
Lin(V; Lin(V; : : : ; Lin(V; U) : : : )):

This rather ompli ated obje t is signi antly simplied with the aid of tensor
produ ts.
We know that Lin(V; U)  U
V : Thus DF (x) 2 U
V :
Further,
Lin(V; Lin(V; U))  Lin(V; U
V )  (U
V )
V  U
V
V ;

thus D2 F (x) 2 U
V
V :
 n

Similarly we have that Dn F (x) 2 U

V :
Moreover, a well-known
theorem states that the n-th derivative is symmetri ,
 n

n

i.e. D F (x) 2 U
_ V :

3.6. We often need the following parti ular result.

Proposition. Let V; U and Z be ane spa es, A : V ! Z an ane surje tion.

A mapping f : Z U is ktimes ( ontinuously) dierentiable if and only if f A

is k times ( ontinuously) dierentiable (k 2 N ):
Proof. The rst part of the statement is trivial.
Suppose that F := f A is k times ( ontinuously) dierentiable. We know
that there is a linear inje tion L : Z ! V su h that A
L = idZ : Then for
z 2 Dom f  Z; h in a neighbourhood of 0 2 Z we have

f (z + h) f (z ) = F (x + L
h) F (x) = DF (x)
L
h + ordo(L
h)

if A(x) = z: Sin e ordo(L
h) = ordo(h); we see that f is ( ontinuously)
dierentiable and
Df (z ) = DF (x)
L

(z 2 Dom f; x 2Afz g):

Moreover, Df : Z U
Z is a mapping su h that Df A = DF
L and
we an repeat the previous arguments to obtain that if F is twi e ( ontinuously)
dierentiable (i.e. DF is ( ontinuously) dierentiable) then f is twi e ( ontinuously) dierentiable (i.e. Df is ( ontinuously) dierentiable).
Pro eeding in this way we an demonstrate k times ( ontinuously) dierentiability.

3.7. (i ) Let C : V V be a dierentiable mapping (a ve tor eld in V):

Then DC (x) 2 V
V for all x 2 Dom C ; thus we an take its tra e:
D
C (x) := Tr (DC (x)) :

The mapping V

 R; x 7! D
C (x) is alled the divergen e of C :

If Z is a ve tor spa e, the divergen e of dierentiable mappings V Z

V
is dened similarly a ording to IV.3.9.
(ii ) Let S : V V be a dierentiable mapping (a ove tor eld in V): Then
DS (x) 2 V
V for all x 2 Dom S ; and we an take

D ^ S (x) := (DS (x))

DS (x):

The mapping V V ^ V ; x 7! D ^ S (x) is alled the url of S :

(iii ) Keep in mind that a ve tor eld has no url and a ove tor eld has no
divergen e.

3.8. (i ) Let V1 ; V2 and U be ane spa es and onsider a dierentiable

mapping F : V1  V2 U: Take an (x1 ; x2 ) 2 Dom F and x x2 : Then V1 U;
y1 7! F (y1 ; x2 ) is a dierentiable mapping; its derivative at x1 is alled the rst
partial derivative of F at (x1 ; x2 ) and is denoted by D1 F (x1 ; x2 ): By denition,
D1 F (x1 ; x2 ) is a linear map V1 ! U:
The se ond partial derivative D2 F (x1 ; x2 ) of F is dened similarly, and an
evident generalization an be made for the k-th partial derivative (k = 1; : : : ; n)
n
of a mapping  Vk U:
k=1
For a ve tor eld C : V1  V2
V1  V2 we dene the omponents
C i : V1  V 2
Vi (i = 1; 2) su h that C = (C 1; C 2): Then DC (x1 ; x2 ) is
an element of (V1  V2 )
(V1  V2 )  (V1  V2 )
(V1  V2 ): It is not hard
to see that using a matrix form orresponding to the onvention introdu ed in
IV.3.7 we have


D
C 1 D2 C 1
1
DC (x1 ; x2 ) = D C 2 D C 2 (x1 ; x2 );

where the symbol (x1 ; x2 ) after the matrix means that every entry is to be taken
at (x1 ; x2 ); shortly,


D
C 1 D2 C 1
1
DC = D C 2 D C 2 :
1

Furthermore, we easily nd that

D
C = D1
C 1 + D2
C 2 :
(ii ) Similar notations for a ove tor eld S = (S1 ; S2 ) : V1  V2

V2)  V1  V2 yield

and

1 S1 D 2 S1
DS = D
D1 S2 D 2 S2

(D1 S2 ) D2 S1 :
1
D ^ S = (D SD1)^ SD
D2 ^ S2
2 1
1 S2


 (V1 

3.9. A ve tor eld C1 : RN 

RN is given by its omponents C i : RN  R
N
N
N

(i = 1; : : : ; N ); C = (C ; : : : ; C ): Its derivative at  is a linear map R

one easily nds for its matrix entries
(DC ( ))i k = k C i ( )

!R

(i; k = 1; : : : ; N )

where k denotes the k-th partial dierentiation.

Then
D
C =
S : RN


RN

N
X
i=1

i C i :

A ove tor eld

is given by its omponents Si : RN
(i = 1; : : : ; N ); S = (S1 ; : : : ; SN ): We have

R

(DS ( ))ik = k Si ( )
and

(D ^ S )ik = i Sk

k Si

for i; k = 1; : : : ; N:

3.10. If I is a one-dimensional ane spa e, V is an ane spa e and r : I

is dierentiable, then, for t 2 Dom r; Dr(t) is an element of V
I  VI :
It is not hard to see that in this ase

V

dr(t)
r(t + t) r(t)
:= r_ (t) := Dr(t) = lim
:
t
!
0
dt
t
t2I
2
Similarly we arrive at d drt(2t) := r(t) := D2 r(t) 2 IV

I :

3.11. Let V and I as before and suppose I is real and oriented. Re all that
then I+ and I denote the sets of positive and negative elements of I; respe tively.
Then r : I V is alled dierentiable on the right at an interior point t of
Dom r if there exists

r_+ (t) := lim

t!0

t2I+

r(t + t) r(t)
t

alled the right derivative of r at t:

The dierentiability on the left and the left derivative r_ (t) are dened
similarly.
Denition. Let V and I be as before, I is oriented. A fun tion r : I V is
alled pie ewise dierentiable if it is

(i ) ontinuous,
(ii ) dierentiable with the possible ex eption of nite points where r is dierentiable both on the right and on the left.

r is alled pie ewise twi e dierentiable if

(i ) it is pie ewise dierentiable,
(ii ) it is twi e dierentiable where it is dierentiable,
(iii ) if a is a point where r is not dierentiable then there exist
lim
t!0

r_(a + t) r_+ (a)

t

t2I+

and

lim
t!0

r_ (a + t) r_ (a)
t

t2I

3.12. Re all that for a nite dimensional ve tor spa e V; Lin(V)  V

V is
a nite-dimensional ve tor spa e as well. Hen e the dierentiability of a fun tion
R : I  Lin(V) makes sense. It an be shown without di ulty that
R is



dierentiable (and then its derivative at t is R_ (t) 2 V
IV  Lin VI ; V ) if and
only if I  V; t 7! R(t)
v is dierentiable for all v 2 V and then


d
d
(R(t)
v) =
R(t)
dt
dt

Moreover, if r : I
and

v:

 V is a dierentiable fun tion then R
r is dierentiable

(R
r)
= R_
r + R
r_ :

4. Submanifolds in ane spa es

In this se tion the ane spa es are real and nite dimensional.

4.1. The inverse mapping theorem and the impli it mapping theorem are
important and well-known results of analysis. Now we formulate them for ane
spa es in a form onvenient for our appli ation.
The inverse mapping theorem. Let V and U be ane spa es, dim U =
dim V: If F : V
U is n  1 times ontinuously dierentiable, e 2 Dom F
and DF (e) : V ! U is a linear bije tion, then there is a neighbourhood N of e;
N  Dom F; su h that
(i ) F jN is inje tive,
(ii ) F [N is open in U;

(iii ) ( F jN ) 1 is n times ontinuously dierentiable.

The impli it mapping theorem. Let V and U be ane spa es, dim U <
dim V: Suppose S : V U is n  1 times ontinuously dierentiable, e 2 Dom S
and DS (e) is surje tive.
Let V1 be a linear subspa e of V su h that the restri tion of DS (e) onto V1
is a bije tion between V1 and U and suppose V0 is a subspa e omplementary
to V1 :
Then there are
| neighbourhoods N0 and N1 of the zero in V0 and in V1 ; respe tively,
e + N0 + N1  Dom S;
| a uniquely determined, n times ontinuously dierentiable mapping G :
N0 ! N1 su h that

S (e + x0 + G(x0 )) = S (e)

Observe that V0 := Ker DS (e) and a subspa e

satisfy the above requirements.

(x0 2 N0 ):

V1 omplementary to V0

4.2. Denition. Let V be an ane spa e, dim V := N  2: Let M and

n be natural numbers, 1  M  N; n  1: A subset H of V is alled an M dimensional n times dierentiable simple submanifold in V if there are
| an M -dimensional ane spa e D;
| a mapping p : D V; alled a parametrization of H; su h that
(i ) Dom p is open and onne ted, Ran p = H;
(ii ) p is n times ontinuously dierentiable and Dp( ) is inje tive for all  2
Dom p;
(iii ) p is inje tive and p 1 is ontinuous.
Re all that Dp( ) 2 Lin(D; V):
Sin e p is dierentiable, it is ontinuous.
The parametrization of H is not unique. For instan e, if E is an ane spa e
and L : E ! D is an ane bije tion then p L is a parametrization, too. In
parti ular, we an take E := RM (see Exer ise 2.5.2); as a onsequen e, D an
be repla ed by RM in the denition.
The inverse mapping theorem implies that the N -dimensional, n times dierentiable simple submanifolds are the onne ted open subsets of V:
Evidently, an M -dimensional ane subspa e of V is an M -dimensional n times
dierentiable simple manifold for all n:

4.3. Denition. Let N  2: A subset H of the N -dimensional ane spa e

V is alled an M -dimensional n times dierentiable submanifold if every x 2 H
has a neighbourhood N (x) in V su h that N (x) \H is an M -dimensional n times
dierentiable simple submanifold.

A subset whi h is an n times dierentiable submanifold for all n 2 N is a

smooth submanifold.
A submanifold means anntimes dierentiable submanifold for somen:
A submanifold whi h is a losed subset of V is alled a losed submanifold.
One-dimensional submanifolds, two-dimensional submanifolds and (N 1)dimensional submanifolds are alled urves or lines, surfa es and hypersurfa es,
respe tively.
By denition, every point of a submanifold has a neighbourhood in the submanifold that an be parametrized. A parametrization of su h a neighbourhood
is alled a lo al parametrization of the manifold.

4.4. Proposition. Let H be an M -dimensional n times dierentiable

submanifold in V; M < N; and let p : RM
V be a lo al parametrization
of H: If e 2 Ran p then there are
| a neighbourhood N of e in V;
| ontinuously n times dierentiable mappings

F :N

! RM ;

S:N

! RN

su h that
(i ) N \ H  Ran p;
(ii ) F (p( )) = ; S (p( )) = 0 for all  2 Dom p; p( ) 2 N ;
(iii ) DS (x) is surje tive for all x 2 N :
Proof. There is a unique  2 Dom p for whi h p() = e: Dp() : RM ! V is
a linear inje tion, hen e V1 := Ran Dp() is an M -dimensional linear subspa e.
Let V0 be a linear subspa e, omplementary to V1 : Evidently, dim V0 = N M:
Let P : V ! V be the proje tion onto V1 along V0 (i.e. P is linear and
P
x1 = x1 for x 2 V1 and P
x0 = 0 for x 2 V0 ): Then
P

(p e) : RM

 V1 ;

 7! P
(p( ) e)

is n times ontinuously dierentiable, its derivative at  equals

Dp(); it is a linear bije tion from RM onto V1 : Thus, a ording to the inverse mapping theorem, there is a neighbourhood
of  su h that P
(p e)j

is inje tive, its inverse is ontinuously dierentiable, (P
(p e))[

= P [p[
e
is open in V1 :
For the sake of simpli ity and without loss of generality we an suppose

= Dom p ( onsidering pj
instead of p):
P

Then the ontinuity of P involves that P (P [p[

e) is an open sub1
set of V and so e + P (P [p[
e) is an open subset of V: Sin e p 1 is

ontinuous, p[
is open in Ran p and p[
 e + P (P [p[
e);
1
thus there is an open subset N in e+ P (P [p[
e)  V su h that p[
= H\N :
Let L : V0 ! RN M be a linear bije tion and

F := (P
(p e)) 1 P
(idV

e)jN ;

S := L (idV

p F ):

N  e + P (P [p[
e) implies P [e + N  P [p[
e =
Dom (P
(p e)) 1 ; hen e both F and S are dened on N : It is left to the
reader to prove that properties (ii ) and (iii ) in the proposition hold.

4.5. Proposition. Let p : RM  V and q : RM  V be lo al parameter-

izations of the M -dimensional n times dierentiable submanifold H su h that

Ran p \ Ran q 6= ;: Then p 1 q : RM
RM is n times ontinuously dierentiable and

D(p 1 q)(q 1 (x)) = [Dp(p 1 (x)) 1
Dq(q 1 (x))

(x 2 Ran p \ Ran q):

Proof.

If M = N then the inverse mapping theorem implies that p 1 is n

times ontinuously dierentiable; as a onsequen e, p 1 q is n times ontinuously
dierentiable as well and the above formula is valid in view of the well-known
rule of dierentiation of omposite mappings.
If M < N; the dierentiability of p 1 makes no sense be ause H ontains no
open subsets in V: Nevertheless, p 1 q is ontinuously dierentiable as we shall
see below.
Let e be an arbitrary point of Ran p \ Ran q: A ording to the previous
proposition, there are a neighbourhood N of e and an n times ontinuously
dierentiable mapping F for whi h N \ H  Ran p and F p  idRM holds.
Then
:= q1 (N )  Dom (p 1 q) is open and


(p 1 q)
= (F p) (p 1 q)
= F qj
;

the mapping on the right-hand side is n times ontinuously dierentiable being

a omposition of two su h mappings. Thus we have shown that ea h point of
Dom (p 1 q) has a neighbourhood
in whi h p 1 q is n times ontinuously
dierentiable.
Let x be an element of Ran p \ Ran q;  := q 1 (x) and  := p 1 q: Then
p   q and  2 Dom (p ): Thus
Dq(q 1 (x)) = Dq( ) = Dp(( ))
D( ) = Dp(p 1 (x))
D( );
whi h gives immediately the desired equality.

()

Evidently, then q 1 p is n times ontinuously dierentiable as well. Sin e

1
q p = (p 1 q) 1 ; this means that the derivative of p 1 q at every point is

a linear bije tion RM ! RM :

As a onsequen e, the dimesion of a submanifold is uniquely determined.
Supposing that a submanifold is both M -dimensional and M 0 -dimensional we
get M = M 0 :
We have proved the statement for parametrizations from RM : Obviously, the
same is true for parametrizations with domains in ane spa es.

4.6. Proposition. Let p and q be lo al parametrizations of a submanifold

su h that Ran p \ Ran q 6= ;: If x 2 Ran p \ Ran q then




Ran Dp(p 1 (x)) = Ran Dq(q 1 (x)) :

Proof. Equality () in the pre eding paragraph involves that the range of
Dq(q) 1 (x)) is ontained in the range of Dp(p 1 (x)): A similar argument yields
that the range of Dp(p 1 (x))is ontained in the range of Dq(q) 1 (x)):
Denition. Let H be an M -dimensional submanifold, x 2 H: Then


Tx(H) := Ran Dp(p 1 (x))

is alled the tangent spa e of H at x where p is a parametrization of H su h that

x 2 Ran p: The elements of Tx (H) are alled tangent ve tors of H at x:
The pre eding proposition says that the tangent spa e, though it is dened
by a parametrization, is independent of the parametrization.
The tangent spa e is an M -dimensional linear subspa e of V: x + Tx (H) is
an ane subspa e of V whi h we all the geometri tangent spa e of H at x:

4.7. Let M < N: We have seen in Proposition 4.4. that every point e of an
M -dimensional n times dierentiable submanifold H has a neighbourhood N in
V and an n times ontinuously dierentiable mapping S : N ! RN M su h that
1
N \ H = S (f0g) and DS (x) is surje tive. Evidently, RN M and 0 2 RN M
an be repla ed by an arbitrary ane spa e U; dim U = N M; and a point
o 2 U; respe tively.
Now we prove a onverse statement.
Proposition. Let V and U be ane spa es, dim V =: N;
dim U =: N M; and S : V U an n times ontinuously dierentiable mapping.
Suppose o 2 Ran S: Then

H := fx 2 S (fog)j Ran DS (x)

is (N

M )-dimensionalg

is either void or an M -dimensionalntimes dierentiable submanifold of V:

Proof. Suppose H is not void and e belongs to it. Then V0 := Ker DS (e)
is an M -dimensional linear subspa e of V: Let V1 be a linear subspa e, omplementary to V0 : Then we an apply the impli it mapping theorem: there are
neighbourhoods N0 and N1 of the zero in V0 and in V1 ; respe tively, an n times
ontinuously dierentiable mapping G : N0 ! N1 su h that

S (e + x0 + G(x0 )) = S (e)
Let us dene

p : V0

 V;

x0

7! e + x0 + G(x0 )

(x0 2 N0 ):
(x0 2 N0 ):
1

Evidently, p is n times ontinuously dierentiable and Ran p  S (fog):

We an easily see that p is inje tive, its inverse is x 7! P
(x e) where P is
the proje tion onto V0 along V1 : Consequently, p 1 is ontinuous.
These mean that p is a parametrization of H in a neighbourhood of e:

4.8. Proposition. Let H 6= ; be the submanifold des ribed in the previous

proposition. Then

Tx (H) = Ker DS (x)

(x 2 H):

Proof. Let p be a lo al parametrization of H: Then S p = onst:; thus for

x 2 Ran p we have DS (x)
Dp(p 1 (x)) = 0 from whi h we dedu e immediately
that Tx (H) := Ran Dp(p 1 (x))  Ker DS (x): Sin e both linear subspa es on
the two sides of  are M -dimensional, equality o urs ne essarily.

4.9. Denition. Let p and q be two lo al parametrizations of a submanifold,

Dom p  RM ; Dom q  RM and Ran p \ Ran q 6= ;: Then p and q are said
to be equally oriented if the determinant of D(p 1 q)( ) is positive for all
 2 Dom (p 1 q):
A family (pi )i2I of lo al parametrizations of a submanifold H is orienting if
H = [i2I Ran pi and, in the ase Ran pi \ Ran pj 6= ;; pi and pj are equally
oriented (i; j 2 I ):
Two orienting parametrization families are alled equally orienting if their
union is orienting as well.
The submanifold is orientable if it has an orienting parametrization family.
To be equally orienting is an equivalen e relation. If the submanifold is
onne ted, there are exa tly two equivalen e lasses.
An orientable submanifold together with one of the equivalen e lasses of
the orienting lo al parametrization families is an oriented submanifold. A lo al
parametrization of an oriented submanifold is alled positively oriented if it
belongs to a family of the hosen equivalen e lass.

A simple submanifold is obviously orientable.

Conne ted N -dimensional submanifolds | i.e. onne ted open subsets | are
orientable.
4.10. Let p be a lo al parametrization of the Msubmanifold H; Dom p  RM :
If (1 ; : : : ; M ) is the standard ordered basis of R then Dp( )
i = i p( ) (i =
1; : : : ; M ) for 
2
Dom p: This means that ((1 p( );
: : : ; M p( )) is an ordered basis in Tp() (H):


In other words, 1 p(p 1 (x)); : : : ; M p(p 1 (x)) is an ordered basis in Tx (H)
(x 2 Ran p):
If q is another lo al parametrization, with domain
in RM ; and x 2 Ran p \


1
Ran q 6= ; then 1 q(q) (x)); : : : ; M q(q1 (x)) is another ordered basis in
Tx (H):
Evidently,

i q(q 1 (x)) = Dq(q 1 (x))
[Dp(p 1 (x)) 1
i p(p 1 (x))

for all i = 1; ::; M:
We know from 4.5 and IV.3.20 that




det D(p 1 q)(q 1 (x)) =det [Dp(p 1 (x)) 1
Dq(q 1 (x)) =


=det Dq(q 1 (x))
[Dp(p 1 (x)) 1 :

We have proved the following statement.

Proposition. Let p and q be lo al parametrizations of the submanifold H;
Dom p  RM ; Dom q  RM and Ran p \ Ran q 6= ;: p and q are equally oriented
if and only if the ordered bases


1 p(p 1 (x)); : : : :; M p(p 1 (x))

and

1 q(q 1 (x)); : : : :; M q(q 1 (x))

in Tx (H) are equally oriented for all x 2 Ran p \ Ran q:

4.11 Observe that in the ase M = 1; i.e. when the submanifold is a urve,
instead of partial derivatives we have a single derivative of p; denoted usually by
p:_ Then p_(p 1 (x)) spans the (one-dimensional) tangent spa e at x:
Two lo al parametrizations p and q are equally oriented if and only if one of
the following three onditions is fullled:
(i ) (p 1 q)
() > 0 for all  2 Dom (p 1 q);
(ii ) p 1 q : R R is stri tly monotone in reasing,

(iii ) p_(p 1 (x)) is a positive multiple of q_(q 1 (x)) for all x 2 Ran p\ Ran q:

4.12. The following notion on erning urves appears frequently in appli ation.
Let x and y be dierent elements of V: We say that the urve C onne ts
x and y if these points form the boundary ofC ; i.e. fx; yg = C n C where C is
the losure of C : We an on eive that x and y are the extremities of a urve
onne ting them.
4.13. Denition. Let H and F be M -dimensional and K -dimensional submanifolds of V and U; respe tively. A mapping F : H F is alled
dierentiable at x if there are lo al parametrizations q of H and p of F for
whi h x 2 Ran q; F (x) 2 Ran p; and the fun tion p 1 F q : RM
RK is
1
dierentiable at q (x):
The derivative of F at x is dened to be the linear map DF (x) : Tx (H) !
TF (x) (F ) that satises

Dp(F (x)) 1
DF (x)
Dq(q 1 (x)) = D(p 1 F q)(q 1 (x)):

F is dierentiable if it is dierentiable at ea h point of its domain.

If H and F are n times dierentiable submanifolds, we dene F to be k times
( ontinuously) dierentiable, for 0  k  n; if p 1 F q is k times ( ontinuously)
dierentiable.

4.14. Exer ises

1. Let V and U be ane spa es. The graph of an n times ontinuously
dierentiable mapping F : V
U| i.e. the set f(x; F (x)j x 2 Dom F g|
is a (dim V )-dimensional n times dierentiable submanifold in V  U: Give its
tangent spa e at an arbitrary point.
2. Prove that the mapping (F; S ) : V RM  RN M  RN des ribed in 4.4
is inje tive and its inverse is (; ) 7! p( ) + L 1 :
3. Let (V; B; h) be a pseudo-Eu lidean ve tor spa e, 0 6= a 2 B: Prove that

fx 2 Vj x
x = a2 g

and

fx 2 Vj x
x =

a2 g

are either void or hypersurfa es in V whose tangent spa e at x equals

fy 2 Vj x
y = 0g:
(The derivative of the map V ! B
B; x 7! x
x at x is 2x regarded as the
linear map V ! B
B; y 7! 2x
y:) Why is the statement not true for a = 0?

4. A linear bije tion RM ! RM has a positive determinant if and only if

it is orientation-preserving. On the basis of this remark dene that two lo al
parametrizations p : D
V and q : E
V of a submanifold are equally
oriented where D and E are oriented ane spa es.
5. Use the notations of 4.13. Prove that
(i ) p 1 F q is dierentiable for some p and q if and only if it is dierentiable
for all p and q;
(ii ) the derivative of F is uniquely dened.
(iii ) if F is the restri tion of a k times ( ontinuously) dierentiable mapping
G:V
U then F is k times ( ontinuously) dierentiable and DF (x) is the
restri tion of DG(x) onto Tx (H):

5. Coordinatization

5.1. Let V be an N -dimensional real ane spa e. Take an o 2 V and an

ordered basis (x1 ; : : : ; xN ) of V: The ane map K : V ! RN determined by
K (o + xi ) := i (i = 1; : : : ; N ) where (1 ; : : : ; N ) is the ordered standard basis
of RN is alled the oordinatization of V orresponding to o and (x1 ; : : : ; xN ):
The inverse of the oordinatization, P := K 1 ; is alled the orresponding
parametrization of V: It is quite evident that
P ( ) = o +

N
X
i=1

( 2 RN ):

 i xi

Moreover, if (p1 ; : : : ; pN ) is the dual of the basis in question, then

K (x) =

pi
(x

o)j i = 1; : : : ; N

(x 2 V):

Obviously, every ane bije tion K : V ! RN is a oordinatization in the

above sense: the one orresponding to o := K 1 (0) and (x1 ; : : : ; xN ) where
xi := K 1 (i ) o (i = 1; : : : ; N ):
Su h a parametrization maps straight lines into straight lines. More losely,
if  2 RN then P maps the straight line passing through  and parallel to i
into the straight line passing through P () and parallel to xi :

P [ + Ri = P () + Rxi

(i = 1; : : : ; N ):

This is why ane oordinatizations are generally alled re tilinear.

5.2. In appli ation we often need non-ane oordinatizations as well. Coordinatization means in general that we represent the elements of the ane spa e
by ordered N -tuples of real numbers (i.e. by elements of RN ) in a smooth way.

Denition. Let V be an N -dimensional ane spa e. A mapping K : V 

RN

is alled a lo al oordinatization of V if
(i ) K is inje tive,
(ii ) K is smooth,
(iii ) DK (x) is inje tive for all x 2 Dom K:
Evidently, DK (x) is bije tive sin e the dimensions of its domain and range
are equal; thus the inverse mapping theorem implies that also the inverse of K
has the properties (i ){(ii ){(iii ;) P := K 1 is alled a lo al parametrization of
V: We often omit the adje tive \lo al".

5.3. If  2 Ran K = Dom P then P [ + Ri is a smooth urve in V; a

parametrization of this urve is pi : R V; a 7! P ( + ai ) (i = 1; :::; N ): The
parametrization maps straight lines into urves, that is why su h oordinatizations are often alled urvilinear.
The urves orresponding to parallel straight lines do not interse t ea h other.
The urves orresponding to meeting straight lines interse t ea h other transversally, i.e. their tangent spa es at the point of interse tion do not oin ide. For
instan e, using the previous notations we have that p_i (0) = DP ()
i = i P ()
is the tangent ve tor of the urve P [ + Ri at P (); if i 6= k then p_i (0) 6= p_k (0):
If x 2 Dom K then P [K (x) + Ri is alled the i-th oordinate line passing
through x:

5.4. Re all Proposition 4.4: if H is an M -dimensional smooth submanifold

of V then for every e 2 H there is a oordinatization K := (F; S ) of V in
a neighbourhood of e su h that the rst M oordinate lines run in H: In other
words, if P is the orresponding parametrization of V then RM V;  7! P (; 0)
is a parametrization of H:

5.5. The most frequently used urvilinear oordinatizations are the polar
oordinatization, the ylindri al oordinatization and the spheri al oordinatization. We give them as oordinatizations in R2 and R3 ; omposed with ane
oordinatizations they result in urvilinear oordinatizations of two- and threedimensional ane spa es.
(i ) Polar oordinatization
K : R2 nf(x1 ; 0)j x1  0g ! R+  ; [ ;
x = (x1 ; x2 ) 7!

jxj; sign(x2 ) ar os jxx1j

its inverse is

P : R+  ; [! R2 n f(x1 ; 0)j x1  0g;

(r; ') 7! (r os '; r sin ');

for whi h

' r sin '

DP (r; ') = os
sin ' r os ' ;
det (DP (r; ')) = r:
(ii ) Cylindri al oordinatization
K :R3 n f(x1 ; 0; x3 )j x1  0; x3 2 Rg ! R+  ; [R;
!
q
x1
2
2
x =(x1 ; x2 ; x3 ) 7!
x1 + x2 ; sign(x2 ) ar os p 2 2 ; x3 ;
x1 + x2
its inverse is

P : R+  ; [R ! R3 n f(x1 ; 0; x3 ; )j x1  0; x3 2 Rg;

(; '; z ) 7! ( os ';  sin '; z );
for whi h

os '  sin ' o

DP (; '; z ) =  sin '  os ' 0 A ;
0
0
1
det (DP (; '; z )) = :
(iii ) Spheri al oordinatization

K : R3 n f(x1 ; 0; x3 )j x1  0; x3 2 Rg ! R+ 0; [ ; [;

!

jxj; ar os jxx3j ; sign(x2 ) ar os p 2x1 2 ;

x1 + x2

x = (x1 ; x2 ; x3 ) 7!
its inverse is

P : R+ 0; [[ ; [! R3 n f(x1 ; 0; x3 )j x1  0; x3 2 Rg;

(r; #; ') 7! (r sin # os '; r sin # sin '; r os #);
for whi h

sin # os ' r os # os '

DP (r; #; ') =  sin # sin ' r os # sin '
os #
r sin #
det (DP (r; #; ')) = r2 sin #:

r sin # sin '

r sin # os ' A ;
0

5.6.

Let K : V
RN be a oordinatization. Then for all x 2 Dom K
the tangent ve tors of the oordinate lines passing through x form a basis in
V: More1 losely, if P is the orresponding parametrization then i P (P 1(x)) =
DP (P (x))
i (i = 1; : : : ; N ) form a basis in V whi h is alled the lo al basis
at x orresponding to K:
Note that DK (x) : V ! RN is the linear bije tion that sends the lo al
basis into the standard basis of RN ; i.e. DK (x) is the oordinatization of V
orresponding to the lo al basis at x:
We shall often use the relation
[DK (P ( )) 1 = DP ( )

( 2 Dom P )

whi h will be written in the form

DK (P ) 1 = DP:

(i ) A ve tor eld C : V
V is oordinatized in su h a way that for
x 2 Dom C \ Dom K the ve tor C (x) is given by its oordinates with respe t
to the lo al basis at x and x is represented by the oordinatization in question;
the oordinatized form of C is the fun tion
DK (P )
C (P ) : RN

 RN ;

 7! DK (P ())
C (P ()):

(ii ) A ove tor eld S : V V is oordinatized similarly, with the aid of the
dual of the lo al bases (see IV.2.2); the oordinatized form of S is the fun tion
DP 
S (P ) : RN

! (RN ) ;

 7! DP ( )
S (P ( )) = S (P ( ))
DP ( ):

(iii ) A ordingly (see IV.2.3), the oordinatizated forms of the tensor elds
V
V  Lin(V) and F : V V
V  Lin(V; V ) are

L:V

DK (P )
L(P )
DP : RN
DP 
F (P )
DP : RN

 RN

 RN
;
  7! DK (P ())
L(P ())
DP ();
 RN
RN ;  7! DP ()
F (P ())
DP ():

5.7.

If K : V ! RN is an ane oordinatization then DK (x) = K for all

x 2 V where K is the linear map under K: Similarly, DP ( ) = P for all  2 RN :
In this ase the ve tor eld C and the ove tor eld S have the oordinatized
form

 7! K
C (P ( ));
 7! P 
S (P ( )):

(1)
(2)

 V
V; x 7! DC (x);
 V
V ; x 7! DS(x):

The derivative of C is the mixed tensor eld DC : V

and the derivative of S is the otensor eld DS : V
Now they have the oordinatized forms

 7! K
DC (P ( ))
P ;
 7! P 
DS (P ( ))
P :

(3)
(4)

A glan e at the previous formulae onvin es us that (3) and (4) are the
derivatives of (1) and (2), respe tively.
Thus in the ase of a re tilinear oordinatization the order of dierentiation
and oordinatization an be inter hanged: taking oordinates rst and then differentiating is the same as dierentiating rst and then taking oordinates.

5.8. In the ase of urvilinear oordinates, in general, the order of dierentiation and oordinatization annot be inter hanged.
To get a rule, how to ompute the oordinatized form of the derivative of a
ve tor eld or a ove tor eld from the oordinatized form of these elds, we
introdu e a new notation.
Without loss of generality we suppose that V = RN ; sin e every urvilinear
oordinatization V
RN an be obtained as the omposition of a re tilinear
oordinatization V ! RN and a urvilinear one RN RN :
For the omponents of elements in V = RN ; Latin subs ripts and supers ripts:
i; j; k; : : : ; for the omponents of the urvililinear oordinates in RN ; Greek
subs ripts and supers ripts: ; ; ; : : : are used. Moreover, we agree that all
indi es run from 1 to N and we a ept the Einstein summation rule: for equal
subs ripts and supers ripts a summation is to be taken from 1 to N:
Thus for K we write K ; for P we write P i ; moreover, for any fun tion
 : RN
R we nd it onvenient to write (P ) instead of  P: The rule of
dierentiation of omposite fun tions will be used frequently,

 ((P )) = (i )(P ) P i ;

as well as the relations

 P i j K (P ) = i j ;
j K  (P )P j =   :
The se ond one implies

i j K (P ) P i  P j + i K  (P )  P i = 0:
We put


:=   P i i K  (P ) = i j K (P ) P i  P j

()

and we all it the Christoel symbol of the oordinatization in question.

The Christoel symbol is a mapping dened
on Dom K ; for  2 Dom K; ( )


is a bilinear map from RN  RN into RN  :
( ; ) 7!

 ( )

  j  = 1; : : : ; N
:

It is usually emphasized that the Christoel symbol is not a tensor of third

order though it has three indi es. This means that in general there is no mapping
V Bilin(V  V; V ) (third order tensor eld) whose oordinatized form would
be the Christoel symbol.

5.9. The oordinatized form of

f :V
C:V
S:V
L:V
F :V

R
V
 V 
 V 
V 
V
V

is
is
is
is
is

f (P );
i K (P )C i (P ) =: C  ;
 P i Si (P ) =: S ;
i K (P )Li k (P ) P k =: L  ;
 P i Tik (P ) P k =: T :

(i ) The oordinatized form of Df : V V is  P i i f (P ) =  (f (P )) ; thus

for a real-valued fun tion the order of dierentiation and oordinatization an
be inter hanged even in the ase of urvilinear oordinatization.
(ii) The oordinatized form of DC : V V
V is

(DC )  := i K  (P )(k C i )(P ) P k ;

whereas the derivative of the oordinatized form of C reads

 i K  (P )C i (P ) = (i k K  )(P ) P k C i (P ) + i K  (P )(k C i )(P ) P k :

The se ond term equals the oordinatized form of DC ; with the aid of relation () in 5.8, the rst term is transformed into an expression ontaining the
Christoel symbol and the oordinatized form of C : In this way we get
(DC )  =  C  +   C :
(iii ) Similarly, if (DS ) denotes the oordinatized form of DS then
(DS ) =  S

 S :

5.10. Now we shall examine the oordinatizated form of two-times dierentiable fun tions I V where I is a one-dimensional ane spa e.

A useful notation will be applied: fun tions I

V and elements of V will
be denoted by the same letter. If ne essary, supplementary remarks rule out
ambiguity.
For the sake of simpli ity and without loss of generality we suppose that I = R:
Let K : V RN be a oordinatization, P := K 1 :
For x 2 V let  := K (x); then x = P ( ):
For x : I V we put  := K (x) := K x; then x = P ( ) := P :
Denoting the dierentiation by a dot we dedu e
_ = DK (x)
x;
_
x_ = DP ( )
;_
()
2
_
_

x = D P ( )(;  ) + DP ( )
;
 = D2 K (x)(x;
_ x_ ) + DK (x)
x;




from whi h we obtain

DK (x)
x = 

( )(;_ _)

(  )

where

( ) := D2 K (P ( )) (DP ( )  DP ( ))
is exa tly the Christoel symbol of the oordinatization.
In view of physi al appli ation, x; x_ and x will be alled position, velo ity and
a eleration, respe tively.
The velo ity at t 2 R; x_ (t) is in V; it is represented by its oordinates
orresponding to the lo al basis at x(t); i.e. by DK (x(t))
x_ (t): Thus () tells
us that the oordinatized form of velo ity oin ides with the derivative of the
oordinatization of position.
Similarly, DK (x(t))
x(t) gives the oordinates of a eleration in the lo al
basis at x(t): Thus () shows that the oordinatized form of a eleration does
not oin ide with the se ond derivative of the oordinatization of position.

5.11. Now we onsider the oordinatizations treated in 5.5. They are orthogonal whi h means that every lo al basis is orthogonal with respe t to the usual
inner produ t in RN (N = 2; 3); in other words, if f1; : : : ; N g is the standard
basis in RN then fDP ( )
i j i = 1; : : : ; N g is an orthogonal basis (the lo al
basis at P ( )):
Introdu ing the notation
i ( ) := jDP ( )
i j
we dene the linear map T ( ) : RN

! RN

T ( )
i := i ( )i
and then

(i = 1; : : : ; N )
by
(i = 1; : : : ; N )

DP ( ) = R( )
T ( )

where R( ) : RN ! RN is an orthogonal linear map.

In usual physi al appli ations one prefers orthonormal lo al bases, i.e. one
takes DP(i()
)i = DP ( )
T ( ) 1
i = R( )
i instead of DP ( )
i (i = 1; ::; N ):
The ve tor y 2 RN at P ( ) has the oordinates R( ) 1
y in the lo al basis
fR( )
i j i = 1; : : : ; N g:
Take x : R RN ;  := K (x); x = P ( ) as in the previous paragraph. Then
x_ = R( )
T ( )
_

from whi h we derive

x =R( )

T ( )
_ + R( )
T ( )

_ + R( )
T ( )
 =



=R( )
R( ) 1
R( )

T ( ) + T ( )

_ + T ( )
 :
A ording to the foregoings, the oordinates of velo ity in the orthonormal
lo al basis at P ( ) are
T ( )
_

and the oordinates of a eleration in the orthonormal lo al basis at P ( ) are



R( ) 1
R( )

T ( ) + T ( )

_ + T ( )
:

5.12. (i ) For polar oordinates  = (r; ');



'
sin '
R(r; ') = os
sin ' os ' =: R(')


1
0
T (r; ') = 0 r =: T (r):

Furthermore

R(')
= 'R
_ (')
R(=2);
T (r)
= T (r_);
and so velo ity and a eleration in the lo al orthonormal basis at (r; ') are
(r;_ r'_ )

and

(r r'_ 2 ; r' + 2r_'_ );

respe tively.
(ii ) For ylindri al oordinates  = (; '; z );
0
1
os '
sin ' 0
R(; '; z ) =  sin ' os ' 0 A =: R(');
0
0
1
0
1
1 0 0
T (; '; z ) =  0  0 A =: T ()
0 0 1

and we dedu e as previously that velo ity and a eleration in the lo al orthonormal basis at (; '; z ) are
(;_ ';
_ z_ )

and

( '_ 2 ; ' + 2_';

_ z);

respe tively.
(iii ) For spheri al oordinates  = (r; #; ');
0

sin # os '
R(r; #; ') =  sin # sin '
os #
0
1
T (r; #; ') =  0
0

os # os '
sin '
os ' sin ' os ' A =: R(#; ');
sin #
0
1
0
0
r
0 A =: T (r; #):
0 r sin #

The omponents of velo ity in the lo al orthonormal basis at (r; #; ') are
(r;_ r#;_ r sin # '_ ):
The omponents of a eleration are given by rather ompli ated formulae; the
ambitious reader is asked to perform the al ulations.

5.13. Exer ises

1. Give the polar oordinatized form of the linear map (ve tor eld) R2
whose matrix is


os 
sin  :
sin  os 

! R2

2. Give the ylindri al and the spheri al oordinates of the following ve tor
elds:
(i ) L : R3 ! R3 is a linear map;
(ii ) R3 ! R3 ; x 7! jxjv where v is a given non-zero element of R3 :
3. Find the oordinatized form of
(i ) the divergen e of a ve tor eld,
(ii ) the url of a ove tor eld.

6. Dierential equations

6.1. Denition. Let V be a nite-dimensional ane spa e over the ve tor

 V is a dierentiable ve tor eld, Dom C is onne ted.

spa e V:
Suppose C : V

Then a solution of the dierential equation

 V)? x_ = C (x)
is a dierentiable fun tion r : R  V su h that
(x : R

(i ) Dom r is an interval,
(ii ) Ran r  Dom C ;
(iii ) r_ (t) = C (r(t)) for t 2 Dom r:
The range of a solution is alled an integral urve of C : An integral urve is
maximal if it is not ontained properly in an integral urve.
An integral urve, in general, is not a urve in the sense of our denition in
4.3, i.e. it is not ne essarily a submanifold.

6.2. Denition. Let C be as before and let xo be an element of Dom C : A

solution of the initial value problem
(x : R

 V)?

x_ = C (x);

x(to ) = xo

()

is a solution r of the orresponding dierential equation su h that

to 2 Dom r

and

r(to ) = xo :

The range of the solution of the initial value problem is alled the integral
urve of C passing through xo :
The well-known existen e and lo al uniqueness theorem asserts that solutions
of the initial value problem exist and two solutions oin ide on the interse tion of
their domain; onsequently there is a single maximal integral urve of C passing
through xo :

6.3. Let U be another ane spa e over the ve tor spa e U ; dim U = dim V:
Suppose L : V
U is a ontinuously dierentiable inje tion whose inverse is
ontinuously dierentiable as well, and Dom C  Dom L:
Put
G:U
U; y 7! DL L 1(y)

C (L 1(y)):
Then r is a solution of the initial value problem () if and only if L r is a
solution of the initial value problem

(y : R

 U)?

y_ = G(y);

y(to ) = L(xo )

():

That is why we all () the transformation of () by L:

6.4. Proposition. Let C be a dierentiable ve tor eld in V and let H be

a submanifold in the domain of C : If C (x) 2 Tx (H) for all x 2 H and xo 2 H
then every solution r of the initial value problem () runs in H; i.e. Ran r  H:

Proof. The element xo has a neighbourhood N in V and there are ontinuously dierentiable fun tions F : N ! RM ; S : N ! RN M su h that S (x) = 0
for x 2 H \ N ; and K := (F; S ) : V RM  RN M  RN is a lo al oordinatization of V: For P := K 1 (the orresponding lo al parametrization of V)
 7! P (; 0) is a parametrization of H \ N : Thus the tangent spa e of H at
P (; 0) is Ker DS (P (; 0)) (see 4.4 and 4.8).
The oordinatized form of C be omes

(; ) : RM
where

 RN

 RM  RN

(; ) :=DF (P (; ))
C (P (; ));

(; ) :=DS (P (; ))
C (P (; )):

Then the oordinatization transforms the initial value problem () into the
following one:
(;_ _ ) = ((; ); (; )) ;

 (to ) = F (xo );

(to ) = 0:

(  )

This means that r is a solution of () if and only if (F r; S r) is a solution

of (  ); or (; ) is a solution of (  ) if and only if P (; ) is a solution of
():
Sin e C (x) 2 Tx (H) for x 2 H; C (P (; 0)) is in the kernel of DS (P (; 0)); i.e.
(; 0) = 0 for all possible  2 RM : Then if  is a solution of the initial value
problem
_ = (; 0);
 (to ) = F (xo )

then (; 0) is a solution of (): Then the uniqueness of solutions of initial value
problems implies that every solution of (  ) has the form (; 0): Consequently,
t 7! P ((t); 0); a solution of (); takes values in H:




6.5. Physi al appli ation requires dierential equations for fun tions I V
where I is a one-dimensional real ane spa e. Sin e the derivative of su h
fun tions takes values in VI ; we start with a dierentiable mapping C : V VI :
A solution of the dierential equation

 V)? x_ = C (x)
is a dierentiable fun tion r : I  V for whi h (i ){(ii ){(iii ) of denition 6.1
(x : I

holds.
Integral urves, solutions of initial value problems et . are formulated as
previously.

7. Integration on urves

7.1. Let I be an oriented one-dimensional ane spa e over the ve tor spa e
I: Suppose A is a one-dimensional ve tor spa e and f : I  A is a ontinuous

fun tion dened on an interval (see Exer ise 1.7.5). If a; b 2 Dom f; a < b; then
Zb

f (t)dt 2 A
I

is dened by some limit pro edure, in the way well-known in standard analysis
of real fun tions, using the integral approximation sums of the form
n
X

f (tk )(tk+1

k=1

tk ):

7.2. Let V be an ane spa e over the ve tor spa e V and let A be a onedimensional ve tor spa e. Suppose F : V  V ! A is a ontinuous fun tion,
positively homogeneous in the se ond variable, i.e.
(x 2 V;  2 R+0 ; x 2 V):

F (x; x) = F (x; x)

Let C be a onne ted urve in V.

Proposition. Let p; q : R V be equally oriented parametrizations of C ;
x; y 2 Ran p \ Ran q: Then

p Z1 (y)

F (p(t); p_(t)) dt =

p (x)
1

q Z1 (y)

F (q(s); q_(s)) ds:

q (x)
1

Proof. We know that  := p 1 q : R ! R is dierentiable and _ > 0 (see

_ s) and
4.11). Consequently, q = p ; q_(s) = p_((s))
(
q Z1 (y)

F (q(s); q_(s)) ds =

q 1 (x)

 1 (Zp 1 (y))

_ s)ds;
F (p((s)); p_((s))) (

 1 (p 1 (x))

whi h gives the desired result by the well-known formula of integration by substitution.

7.3. Suppose C is oriented. Then, a ording to the previous result, we

introdu e the notation

Zy

F (
; dC ) :=

p Z1 (y)

F (p(t); p_(t))dt

p 1 (x)

where p is an arbitrary positively oriented parametrization of C su h that x; y 2

Ran p:
Note that a ording to the denition we have
Zx

F (
; dC ) =

Zy

F (
; dC ):

If C is not oriented, we shall use the symbol

Z

F (
; dC ) :=

[x;y




 p Z1 (y)



F (p(t); p_(t))dt




 p 1 (x)

where p is an arbitrary parametrization.

We frequently meet the parti ular ase when F does not depend on the
elements of V; i.e. there is a positively homogeneous f : V ! A su h that
F (x; x) = f (x) for all x 2 V; x 2 V: Then we use the symbol
Zy

f (dC )

and

[x;y

for the orresponding integrals.

f (dC )

7.4. We an generalize the previous result for a parametrization r : I V

where I is an oriented one-dimensional ane spa e
over the ve tor spa e I: Then


r_ (t) is in VI and a epting the denition F x; xt := F (jx;tjx) (x 2 V; x 2 V; 0 6=
t 2 I) we have
Zy

F (
; dC ) =

r Z1 (y)

F (r(t); r_ (t)) dt

r 1 (x)

if r is positively oriented.

7.5. Let (V; B; h) be a pseudo-Eu lidean ane spa e (i.e. V is an ane

spa e over V and (V; B; h) is a pseudo-Eu lidean ve tor spa e). Supposing B
is oriented, we have the square root mapping (B
B)+0 ! B+0 and
V ! B;

x 7! jxj :=

jx
xj

is a positively homogeneous fun tion. Thus if C is an oriented urve in V; then

Zy

jdCj

is meaningful for all x; y 2 C : In the Eu lidean ase it is regarded as the signed

length of the urve segment between x and y; in the non-Eu lidean ase it is
interpreted as the pseudo-length of the urve segment.
Proposition. Suppose that jxj 6= 0 for all non-zero tangent ve tors x of C :
Then for all xo 2 C ;

C ! B;

x 7!

Zx

xo

jdCj

is a ontinuous inje tion whose inverse is a positively oriented parametrization

of C :
Proof. Let Z denote the above mapping and hoose a positively oriented
parametrization p : R V and put to := p 1(xo ): Then

(Z p)(t) =

Zt

to

jp_(s)jds

(t 2 Dom p);

onsequently, Z p : R
B is ontinuously dierentiable and (Z p)
(t) =
0 for all t 2 Dom p: Thus Z p is stri tly monotone in reasing: it
is inje tive and its inverse (Z p) 1 is ontinuously dierentiable as well, and
a ording to the well-known rule,

jp_(t)j >


(Z p) 1
=


(Z p)
(Z p)


1 > 0:

As a onsequen e, introdu ing the notation r := Z 1 ; we have that r =

p (Z p) 1 is ontinuously dierentiable, too, and

p_
r_ r 1 = p 1:
jp_j
This means that r is a parametrization of C and r 1 p (= Z p) has everywhere
positive derivative, i.e. r and p are equally oriented.
It is worth noting that jr_ j = 1:

VII. LIE GROUPS

We treat only a spe ial type of Lie groups appearing in physi s; so we avoid
the appli ation of the theory of smooth manifolds.

1. Groups of linear bije tions

1.1. Let V be an N -dimensional

real ve tor spa e, N 6= 0:
Then Lin(V) is an N 2 -dimensional real ve tor spa e.
Now the symbol of omposition between elements of Lin(V) will be omitted,
i.e. we write AB := A B for A; B 2 Lin(V):
Sin e V is nite dimensional, all norms on it are equivalent, i.e. all norms give
the same open subsets. Given a norm k k on V; a norm is dened on Lin(V)
by

kAk :=

sup kA
vk
kvk=1

for whi h kAB k  kAk kB k holds (A; B 2 Lin(V)):

We introdu e the notation

G `(V) := fF 2 Lin(V)j F
Endowed with the multipli ation (F ; G)
group whose identity (neutral element) is
I

1.2.

is bije tiveg:

7! F G ( omposition), G `(V) is a

:= idV :

One an prove without di ulty that if A

V) and
1
X
(I A) 1 =
An :

A 2 G `(

n=0

2 Lin(V); kAk < 1; then

In other words, if K 2 Lin(V); kI

K k < 1;

X
K 1=
(I
n=0

then K 2 G `(V) and

K )n :

Proposition. Let F 2 G `(V): If L 2 Lin(V) and kF Lk < kF 1 k then

L 2 G `(V):
Proof. kI F 1 Lk = kF 1(F L)k  kF 1k kF 1 Lk < 1; thus F 1
L
1

is bije tive. F is bije tive by assumption, hen e F (F L) = L is bije tive as

well.
As a orollary of this result we have that G `(V) is an open subset of Lin(V):

1.3. The proof of the following statement is elementary.

The mappings
m : G `(V)  G `(V) ! G `(V);

are smooth and

(F ; G) 7! F G;

j : G `(V) ! G `(V);

7! F

Dm(F ; G) : Lin(V)  Lin(V) ! Lin(V);

Dj(F ) : Lin(V) 7! Lin(V);

(A; B ) 7! AG + F B ;

A 7!

F 1 AF 1 :

1.4. It is a well-known fa t, too, that for A 2 Lin(V)

exp A := eA :=
is meaningful, it is an element of G `(V);

e0 = I ;

eA

1 An
X

n=0

n!

= e A:

Moreover, the exponential mapping,

Lin(V) ! G `(V);

A 7! eA

is smooth, its derivative at 0 2 Lin(V) is the identity map Lin(V) ! Lin(V):

The inverse mapping theorem implies that the exponential mapping is inje tive in a neighbourhood of 0; its inverse regarding this neighbourhood is smooth
as well.

If A; B 2 Lin( ) and AB = BA then eAeB

t
A
e esA = esA etA = e(t+s)A for t; s 2 R:

= eB eA = eA+B : In parti ular,

1.5. For A 2 Lin(V); the fun tion R ! G `(V); t 7! etA is smooth and
d tA

e = AetA = etA A:
dt
As a onsequen e, the initial value problem
(X : R

 Lin(V))?

X_

= XA;

X (0) = I

has the unique maximal solution

(t 2 R):

R(t) = etA

2. Groups of ane bije tions

2.1.

Then

Let V be an ane spa e over the N -dimensional real ve tor spa e

V:

A(V; V) := fA : V ! Vj A is aneg;
endowed with the pointwise operations, is a real ve tor spa e.
Given o 2 V; the orresponden e
A(V; V) ! V  Lin(V);

A 7! (A(o); A)

(where A is the linear map under A) is a linear bije tion; it is evidently linear and
inje tive and it is surje tive be ause the ane map V ! V; x 7! A
(x o) + a
orresponds to (a; A) 2 V  Lin(V):
As a onsequen e, A(V; V) is an (N + N 2 )-dimensional ve tor spa e.

2.2. We easily nd that

A(V) := fL : V ! Vj L is aneg;
endowed with the pointwise subtra tion (see VI.2.3(iv)), is an ane spa e over
A(V; V): Thus, a ording to the previous paragraph, A(V) is (N + N 2 )dimensional.

Two elements K and L of A(V); as well as an element A of A(V; V) and

an element L of A(V) an be omposed; the symbol of ompositions will be
ommitted, i.e. KL := K L and AL := A L:
We introdu e

G a(V) := fF 2 A(V)j F
Endowed with the multipli ation (F; G)
group whose identity (neutral element) is

is bije tiveg:

7! F G

( omposition), G a(V) is a

I := idV :

2.3. Given o 2 V; the mapping

A(V) ! V  Lin(V);

L 7! (L(o) o; L)

is an ane bije tion over the linear bije tion given in 2.1. Evidently, this bije tion
maps G a(V) onto V  G `(V): As a onsequen e, G a(V) is an open subset of
A(V):

2.4. The mappings

m : G a(V)  G a(V) ! G a(V);
j : G a(V) ! G a(V);

(F; G) 7! F G;

7! F

are smooth and

Dm(F; G) : A(V; V)  A(V; V) ! A(V; V);
Dj(F ) : A(V; V) ! A(V; V);

(A; B ) 7! AG + F B;

A 7!

F 1 AF 1 :

2.5. If P 2 G `(V) then

`P : A(V; V) ! A(V; V);

A ! PA

is a linear bije tion, (`P ) 1 = `P 1 :

If P 2 G a(V) then

`P : A(V) ! A(V);

L 7! P L

is an ane bije tion over `P ; where P is the linear map under P ; moreover,
(`P ) 1 = `P 1 :

2.6. If A 2 A(V; V) and A 2 Lin(V) is the linear map under A then

exp A := eA := I +

1 An
X

1A

n!

n=1

is meaningful, it is an element of G a(V);

e0 = I;

eA

=e A

and the linear map under eA is eA :

Moreover, the exponential mapping
A(V; V) ! G a(V);

A 7! eA

is smooth, its derivative at 0 2 A(V; V) is the identity map A(V; V) !

A(V; V):
The inverse mapping theorem implies that the exponential mapping is inje tive in a neighbourhood of 0; its inverse regarding this neighbourhood is smooth
as well.
If A; B 2 A(V; V) and AB = B A then eA eB = eB eA = eA+B : In parti ular,
tA
e esA = esA etA = e(t+s)A for t; s 2 R:

2.7. For A 2 A(V; V); the fun tion R ! G a(V); t 7! etA is smooth and
d tA
tA
e = e A:
dt
(Note that AetA makes no sense!).
As a onsequen e, the initial value problem

(X : R

 A(V; V))?

X_ = X A;

X (0) = I

(X : R Lin(V); X (t) is the linear map under X (t)) has the unique maximal
solution
R(t) = etA
(t 2 R):

3. Lie groups

3.1. Denition Let V be an N -dimensional real ane spa e. A subgroup G

of G a(V) whi h is an M -dimensional smooth submanifold of G a(V) is alled an
M -dimensional plain Lie group.

The group multipli ation G  G ! G ; (F; G) 7! F G and the inversion G ! G ;

1 are smooth mappings (see 2.4 and VI.4.13, Exer ise VI.4.14.5(iii)).
Observe that by denition 0 < M  N + N 2 : (N + N 2 )-dimensional plain Lie
groups are G a(V) and its open subgroups.
Remark. In general, a Lie group is dened to be a group endowed with a
smooth stru ture in su h a way that the group multipli ation and the inversion
are smooth mappings.
Sin e we shall deal only with plain Lie groups, we shall omit the adje tive
\plain". By the way, all the results we shall derive for plain Lie groups are valid
for arbitrary Lie groups as well.

7! F

3.2. (i) For x 2 V we dened the ane bije tion Tx : V ! V; x x + x

(VI.2.4.3), the translation by x: It is quite evident that Tx = Ty if and only if
x = y and so
T n(V) := fTxj x 2 Vg;
alled the translation group of V; is an N -dimensional Lie group. The group
multipli ation in T n(V) orresponds exa tly to the addition in V that is why
one often says that V | in parti ular RN | endowed with the addition as a
group multipli ation is an N -dimensional Lie group.
(ii) If the ve tor spa e V is onsidered to be an ane spa e then G `(V) is
a subgroup and an N 2 -dimensional submanifold of G a(V); thus G `(V) is an
N 2 -dimensional Lie group.
3.3. It is obvious that
G a(V) ! G `(V);

L 7! L

(L is the linear map under L)

is a smooth group homomorphism whose kernel is T n(V) (L = I if and only if

L 2 T n(V); see VI.2.5.6).
(i ) Take a Lie group G  G a(V): Then
under(G ) := fF

2 G `(V)j F

is under an F

2 G a(V)j F

is over an F

2 Gg;

i.e. the image of G by the above group homomorphism is a Lie group.

(ii ) Conversely, if G  G `(V) is an M -dimensional Lie group, then
over(G ) := fF

2 Gg;

the pre-image of G by the above group homomorphism, is an (M +N )-dimensional

Lie group.

3.4. Re all that the tangent spa es of G are linear subspa es of

A(V; V): Every tangent spa e of G is obtained quite simply from the tangent
spa e at I : TF (G ) is the \translation" by F of TI (G ):

Proposition. Let G  G a(V) be a Lie group. Then

TF (G ) = F [TI (G = fF Aj A 2 TI (G )g

(F

2 G ):

Proof. Let G be M -dimensional. There is a neighbourhood1 N of I in G a(V);

a smooth mapping S : N ! RN +N M su h that G \N = S (f0g) (see VI.4.4),

and TI (G ) = Ker DS (I ) (see VI.4.8).
Let F be an arbitrary element of G : Then G is invariant under the ane
bije tion `F 1 = `F 1 ; thus S `F 1 G = 0: Consequently, if P is in the domain
of S `F 1 ; i.e. `F 1 P = F 1 P is in N ; re alling that `F 1 is an ane map
over `F 1 ; hen e D`F 1 (P ) = `F 1 ; we have
2

TP (G ) =Ker D(S `F 1 )(P ) = Ker DS (`F 1 P )
D`F 1 (P ) =

=Ker DS (F 1 P )`F 1 = fA 2 A(V; V)j DS (F 1 P )F 1 A = 0g =


=fF B j B 2 Ker DS (F 1 P )g = F Ker DS (F 1 P ) :
We an take P := F to have the desired result.
The tangent spa e of G at I plays an important role; for onvenien e we
introdu e the notation

La(G ) := TI (G ):
Note that La(G a(V)) = A(V; V); La(G `(V) = Lin(V):
Moreover, La(T n(V)) = V where V is identied with the onstant maps
V ! V:
3.5. Denition. A smooth fun tion R : R ! G  G a(V) is alled a oneparameter subgroup in the Lie group G if

R(t + s) = R(t)R(s)

(t; s 2 R):

In other words, a one-parameter subgroup is a smooth group homomorphism

R : T n(R) ! G : Evidently, R(0) = I and R( t) = R(t) 1 :
There are three possibilites.
(i) There is a neighbourhood of 0 2 R su h that R(t) = I for all t in that
neighbourhood; then R is a onstant fun tion, R(t) = I for all t 2 R:
(ii) There is a T 2 R+ su h that R(T ) = I but R(t) 6= I for 0 < t < T ; then
R is periodi , R(t + T ) = R(t) for all t 2 R:
(iii) R(t) 6= I for all 0 6= t 2 R:

3.6. If R(t) denotes the linear map under R(t) then R : R

one-parameter subgroup; R(0) = I :

! under(G ) is a

Dierentiating with respe t to s in the dening equality of R and then putting

s = 0 we get
R_ (t) = R(t)R_ (0)
(t 2 R)
whi h shows that if Ran R is not a single point (if R is not onstant) then it is
a one-dimensional submanifold and a subgroup in G a(V ): Thus Ran R is either
the singleton fI g or a one-dimensional Lie group. In the ase (ii) treated in the
pre eding paragraph, the restri tion of R to an interval shorter than T is a lo al
parametrization of Ran R; in the ase (iii) R is a parametrization of Ran R:

3.7. Proposition. Every one-parameter subgroup R in G has the form

(t 2 R)

R(t) = etA

where A = R_ (0) 2 La(G ):

Conversely, if A 2 La(G )  A(V; V) then t 7! etA is a one-parameter
subgroup in G :
Proof. A ording to the previous paragraph, the one-parameter subgroup R
is the solution of the initial value problem
(X : R

 Ga(V))?

X_ = X A;

X (0) = I

where A := R_ (0): Apply 2.7 to obtain the rst statement.

Conversely, t 7! etA is a one-parameter subgroup in G a(V); we have to show
only that etA 2 G for all t 2 R whi h follows from VI.6.4.
The assertions are true for lo al one-parameter subgroups as well, i.e. for
smooth fun tions R : R ! G dened on an interval around 0 2 R su h that
R(t + s) = R(t)R(s) whenever t; s; t + s are in Dom R:

3.8.

The previous result involves that eA 2 G for A 2 La(G ); i.e. the

restri tion of the exponential mapping onto La(G ) takes values in G : Sin e the
exponential mapping is smooth and inje tive in a neighbourhood of 0; its inverse
regarding this neighbourhood is smooth as well (in parti ular ontinuous), we
an state:
Proposition. Let G be a Lie group. Then

La(G ) ! G ;

A 7! eA

is a parametrization of G in a neighbourhood of the identity I:

In parti ular, every element in a neighbourhood of I belongs to a oneparameter subgroup.

3.9. Proposition. Every element of G in a neighbourhood of the identity is

a produ t of elements taken from one-parameter subgroups orresponding to a
basis of La(G ):
Proof. Let A1; : : : AM be a basis of La(G ) and omplete it to a basis
A1 ; : : : ; AP of A(V; V) where P := N + N 2 : Then
 : RP

! G a(V);

(t1 ; t2 ; : : : ; tP ) 7! exp (t1 A1 ) exp (t2 A2 ) : : : exp (tP AP )

is a smooth map, (0; 0; : : : ; 0) = I; k (0; 0; : : : ; 0) = Ak (k = 1; : : : ; P ): We

an state on the basis of the inverse mapping theorem that  is inje tive in a
neighbourhood of (0; 0; : : : ; 0); its inverse regarding this neighbourhood is smooth
as well.
Thus the restri tion of  onto RM regarded as the subspa e of RP onsisting
of elements whose i-th omponents are zero for i = M +1; : : : ; P is a parametrization of G in a neighbourhood of I:
Note that in general
exp (t1 A1 ) exp (t2 A2 ) : : : exp (tP AP ) 6= exp

P
X
k=1

tk Ak :

3.10. If G is onne ted, every element of G is a produ t of elements in a

neighbourhood of I; hen e every element is a produ t of elements taken from
one-parameter subgroups orresponding to a basis of La(G ); sin e the following
proposition is true.
Proposition. If G is onne ted and V is a neighbourhood of the identity I
in G ; then
[
G = Vn
n2N
where V n := fF1 F2 : : : Fn j Fk 2 V ; k = 1; : : : ; ng:
Proof. Given F 2 G ; the mapping G ! G ; G 7! F G is bije tive, ontinuous,
its inverse is ontinuous as well. Thus for all F 2 G ; F V := fF Gj G 2 Vg is
open, so V 2 = [ F V is open as well. Consequently, V n is open for all n and
F 2V
thus H := [ V n is open, too. We shall show that the losure of H in G equals
n2N
H; thus H; being open and losed, equals G :
Let L be an element of the losure of H in G : Sin e LV 1 is a neighbourhood
of L; there is an F 2 H su h that F 2 LV 1 whi h implies L 2 F V ; sin e
F V  HV = H; the proof is omplete.

4. The Lie algebra of a Lie group

4.1. Re all that if G is a Lie group in G a(V) then La(G ); the tangent spa e of
G at I = idV is a linear subspa e of A(V; V): If A 2 A(V; V) then A denotes
the underlying linear map V ! V:
Proposition. Let G be a Lie group. If A; B 2 La(G ) then
AB B A 2 La(G ):

Proof. Take a1neighbourhood N of I in G a(V) and a smooth map S dened

on N su h that S (f0g) = G \N and La(G ) = Ker DS (I ) (see the proof of 3.3).
Then

and
t 7! S etB etA = 0
t 7! S etA etB = 0
for t in a neighbourhood of 0 2 R: Dierentiating the rst fun tion with respe t
to t we get




t 7! DS etA etB
etA AetB + etAetB B = 0:
Again dierentiating and then taking t = 0 we dedu e
D2 S (I ) (A + B; A + B ) + DS (I )
(AA + 2AB + B B ) = 0:
Similarly we derive from the se ond fun tion that
D2 S (I ) (B + A; B + A) + DS (I )
(B B + 2B A + AA) = 0:
Let us subtra t the equalities from ea h other to have
DS (I )
(AB

B A) = 0

whi h ends the proof.

4.2.

mapping

A ording to the previous proposition we are given the ommutator

La(G )  La(G ) ! La(G );

(A; B ) 7! AB

B A =: [A; B :

Proposition. The ommutator mapping

(i) is bilinear,
(ii) is antisymmetri ,
(iii) satises the Ja obian identity:
[[A; B ; C + [[B; C ; A + [[C; A; B = 0

(A; B; C 2 La(G )):

Denition. La(G ) endowed with the ommutator mapping is alled the Lie
algebra of G :
We dedu e without di ulty that for A; B 2 La(G )


1 d2 tA tB
e e
[A; B =
2 dt2

etB etA

t=0

4.3. The Lie algebra of G a(V) is A(V; V): We have seen that if a linear
subspa e L of A(V; V) is the tangent spa e at I of a Lie group then the
ommutator of elements from L belongs to L; too; in other words, L is a Lie
subalgebra of A(V; V):
Conversely, if L is a Lie subalgebra of A(V; V) then there is a Lie group G
su h that La(G ) = L : the subgroup generated by feA j A 2 Lg: It is not so
easy to verify that this subgroup is a submanifold.

4.4. Denition. Let G and H be Lie groups. A mapping  : G H is

alled a lo al Lie group homomorphism if
(i ) Dom  is a neighbourhood of the identity of G ;
(ii )  is smooth,
(iii ) (F G) = (F )(G) whenever F; G; F G 2 Dom:
If  is inje tive and  1 is smooth as well, then  is a lo al Lie group
isomorphism.
4.5. For a lo al Lie group homomorphism  : G  H we put
 := D(I ) 2 Lin (La(G ); La(H)) :
If A 2 La(G ); then t 7! (etA ) is a lo al one-parameter subgroup in H and


whi h implies

d
(etA )
= (A);
dt
t=0
(etA ) = et(A)

for t in a neighbourhood of 0 2 R:
Proposition.  : La(G ) ! La(H) is a Lie algebra homomorphism, i.e. it is
linear and
[(A); (B ) =  ([A; B )
(A; B 2 La(G )):

Proof. Start with

 2 
d
[(A); (B ) = 2 et(A) et(B)
dt


et(B) et(A)




t=0

d2
= 2 (etA etB ) (etB etA )
dt
t=0
and then apply the formulae in the proof of 4.1 putting  in pla e of S:

4.6. The previous proposition involves that lo ally isomorphi Lie groups
have isomorphi Lie algebras. One an prove the onverse, too, a fundamental
theorem of the theory of Lie groups: if the Lie algebras of two Lie groups are
isomorphi then the Lie groups are lo ally isomorphi .
5. Pseudo-orthogonal groups
Let (V; B; h) be a pseudo-Eu lidean ve tor spa e. Re all the notations (see
V.2.7)
O(h) :=fL 2 G `(V)j L>
L = I g;
A(h) :=fA 2 Lin(V)j A> = Ag:

Proposition. If dim V = N then O(h) is an N (N2 1) -dimensional Lie group

having A(h) as its Lie algebra.
Proof. It is evident that O(h) is a subgroup of G>`(V):
We know that A(h) and S(h) := fS 2 Lin(V)j S = S g are omplementary
subspa es, dim S(h) = N (N2+1) ; dim A(h) = N (N2 1) (see V.2.9).
Let us onsider the mapping

' : G `(V) ! S(h);

L 7! L>
L:
Sin e the h-adjun tion L 7! L> is linear and the multipli ation in Lin(V) is
bilinear, ' is smooth. Moreover, the equality
(L + H )>
(L + H ) L>
L = L>
H + H >
L + H >
H
shows that
D'(L)
H = L>
H + H >
L

(L 2 G `(V); H 2 Lin(V):

We have O(h) = fL 2 G `(V)j '(L) = I g and D'(L) is surje tive if L is

in O(h) : if S 2 S(h) then D'(L)
L2
S = S : Consequently, O(h) is a smooth
submanifold of G `(V) (see VI.4.7).

Finally, D'(I )
H = 0 if and only if H 2 A(h); hen e

La(O(h)) = Ker D'(I ) = A(h):

6. Exer ises
1. Let G be a Lie group, A; B 2 La(G ): Prove that [A; B = 0 if and only if
etA etB = etB etA for all t in an interval around 0 2 R:
Consequently, G is ommutative (Abelian) if and only if La(G ) is ommutative
(the ommutator mapping on La(G ) is zero).
2. Using the denition of exponentials (see 2.6) demonstrate that
1 tA tB
[A; B = tlim
!0 t2 e e

1 tA tB tA tB
etB etA = tlim
!0 t2 e e e e

I :

3. Let V be a nite dimensional real ve tor spa e and make the identi ation
A(V)  V  Lin(V);

A  (A(0); A);

i.e. (a; A) 2 V  Lin(V) is onsidered to be the ane map

V ! V;

x 7! Ax + a:

Then the omposition of su h ane maps is

(a; A)(b; B ) = (a + Ab; AB ):
In this way we have G a(V)  V  GL(V):
Prove that
e(a;0) = (a; I );
e(0;A) = (0; eA):

4. Let n be a positive integer. Prove that

O(n) :=fL 2 Lin(Rn )j LL = I g

SO(n) :=fL 2 O(n)j detL = 1g
are n(n2 1) -dimensional Lie groups having the same Lie algebra:
fA 2 Lin(Rn )j A = Ag
( f. Proposition in Se tion 5).
Give a lo al Lie group isomorphism between O(n) and SO(n):

5. A omplex ve tor spa e and its omplex linear maps an be onsidered to

be a real ve tor spa e and real linear maps.
Demonstrate that

S `(2; C ) := fL 2 Lin(C 2 )j detL = 1g

is a six-dimensional Lie group having

fA 2 Lin(C 2 )j TrA = 0g
as its Lie algebra.
6. Let n be a positive integer. Prove that

U (n) :=fL 2 Lin(C n )j L> L = I g;

SU (n) :=fL 2 U (n)j detL = 1g
are an n2 -dimensional and an (n2 1)-dimensional Lie group, respe tively. (The
star denotes adjoint with respe t to the usual omplex inner produ t; in other
words, if L is regarded as a matrix then L> is the onjugate of the transpose of
L:) Verify that they have the Lie algebras

fA 2 Lin(C n )j A> = Ag;

fA 2 Lin(C n )j A> = A; TrA = 0g;
respe tively.
7. Prove that

U (1) := fL 2 Lin(C )j L> L = I g  f 2 C j jj = 1g

is a one-dimensional Lie group, lo ally isomorphi but not isomorphi to

T n(R):

8. Let G  G a(V) be a Lie group. An orbit of G is a non-void subset P of V

su h that fL(x)j L 2 Gg = P for some | hen e for all | x 2 P:
Prove that distin t orbits are disjoint. V is the union of orbits of G : In other
words, the  relation on V dened by x  y if x and y are in the same orbit of
G is an equivalen e relation.
9. Find the orbits of G a(V); G `(V); T n(V); O(n); SO(n); U (n); SU (n):

SUBJECT INDEX

Aberration of light
non-relativisti I.6.2.3
relativisti II.4.7.3
absolute s alar potential I.9.4.6
a eleration
non-relativisti I.2.1.3
relativisti II.2.3.4
a eleration eld
non-relativisti I.3.1.3
relativisti II.3.1.2
addition of relative velo ities II.4.3
ane
map VI.2.1
subspa e VI.1.5
angle V.3.3
non-relativisti I.1.2.5, I.2.1.3
relativisti II.2.3.5.
angle of rotation I.11.1.5
angular velo ity I.4.2.4, I.5.3.1
arrow orientation V.4.13
arrow-preserving maps V.4.13
arti ial time II.6.2.4
automorphism of a spa etime model
non-relativisti I.1.5.1
relativisti II.1.6.1
axis of rotation
non-relativisti I.5.3.2
relativisti II.6.7.3, II.6.8.3
Basi velo ity value I.1.3.3. and I.1.6.4
Cau hy inequality V.3.2
reversed V.4.7
Centripetal a eleration I.6.3.2
Christoel symbols VI.5.8
oordinatization IV.2, VI.5
Coriolis a eleration I.6.3.2
ommutator VII.4.2

ompletely split form of : : :

non-relativisti I.8.5.1, I.9.4.1
relativisti II.7.3.1, II.8.3.1
ove tor IV.1.1
eld VI.3.7
ove tor transformation rule
non-relativisti I.8.3.2
relativisti II.7.2.2
url VI.3.7
urve VI.4.9
Derivative of a map VI.3.3
determinant IV.3.18
dierentiation VI.3.3
dierentiable map VI.3.3
dire ted, an ane subspa e VI.1.5
distan e observed by : : :
non-relativisti I.7.1.1
relativisti II.5.5.1
distan e in observer spa e II.6.3
distan e unit I.10.2.2
divergen e VI.3.7
dot produ t V.2.3
double ve torized splitting I.4.4.2
Earlier
non-relativisti I.1.1.3
relativisti II.1.2.3, II.5.6.1, II.6.2.5
equivalent referen e frames
non-relativisti I.10.5
relativisti II.9.3
Eu lidean ve tor spa e V.3
Euler angles I.11.2.1
Fit observer I.3.1.4
for e eld
non-relativisti I.2.4.2
relativisti II.2.6.2

future-dire ted
non-relativisti I.1.1.3
relativisti II.1.2.2
Galilean group I.11.3
ortho hronous I.11.3.3
spe ial I.11.3.5
Galilean referen e frame I.10.2
Half split form of : : :
non-relativisti I.8.5.1, I.9.4.1
relativisti II.7.7.1, II.8.3.1
hyperplane VI.2.5
hypersurfa e VI.4.9
Impli it mapping theorem VI.4.1
inertial time II.2.2.2
inverse mapping theorem VI.4.1
inversion
u-spa elike I.11.3.4, II.10.1.3
u-timelike I.11.3.4, II.10.1.3
isomorphism of spa etime models
non-relativisti I.1.5.1
relativisti II.1.6
Later
non-relativisti I.1.1.3
relativisti II.1.2.3, II.5.6.1, II.6.2.5
length V.3.3
non-relativisti I.1.2.5
relativisti II.6.2.3
Levi{Civita tensor V.2.12, V.3.13,
V.4.21.3
Lie
algebra of a Lie group VII.4
group VII.3
lightlike II.1.2.2
light signal II.2.1.2
Lorentz boost II.1.3.8
Lorentz ontra tion II.5.3
Lorentz group II.10.1
Lorentz referen e frame II.9.2
magnitude V.3.3
non-relativisti I.1.2.5, I.2.1.3
relativisti II.2.3.5
mass
non-relativisti I.2.4.1
relativisti II.6.1

measuring rod I.7.2, II.5.5

Minkowskian ve tor spa e V.4
Neumann group I.11.6.8
Newton equation
non-relativisti I.2.4.2
relativisti II.2.6.2
Noether group I.11.6
arithmeti I.11.8.4
instantaneous I.11.6.5
ortho hronous I.11.6.1
split I.11.8
ve torial I.11.7.2
Observer
inertial
non-relativisti I.4.1
relativisti II.3
regular II.6.2.4
rigid
non-relativisti I.3.3.1
relativisti II.6.3.4
rotation-free I.3.3.1
uniformly a elerated
non-relativisti I.5.2
relativisti II.6.5, II.6.6
uniformly rotating
non-relativisti I.5.3
relativisti II.6.7, II.6.8
with origin
non-relativisti I.4.1.4
relativisti II.3.4.1
orientation IV.5.1
orientation-preserving maps IV.5.1
origin I.10.2.2
-orthogonal
basis V.1.2
map V.2.7
Parallelism II.5.1.2
parametrization VI.4.2, VI.5.1
plane VI.1.5
Poin are group II.10.3
arithmeti II.10.5.3
u-split II.10.5
ve torial II.10.4
positive element IV.5.3
positively oriented IV.5.1

potential
non-relativisti I.2.4.3
relativisti II.2.6.3
proper time II.2.3.1
pseudo-Eu lidean
ane spa e VI.1.6
ve tor spa e V.1
pseudo-length V.4.10
referen e
frame
non-relativisti I.10.1.3
relativisti II.9.1.2
system
non-relativisti I.10.1.2
relativisti II.9.1.1
relative velo ity
non-relativisti I.6.2.2
relativisti II.4.2.2, II.4.7.2
root square (tensorial) IV.5.4
rotation I.11.1.2
Simultaneity
non-relativisti I.1.1.2
relativisti II.3.2.2, II.4.2
smooth map VI.3.5
solution of a dierential equation VI.6.1
spa elike
non-relativisti I.1.13
relativisti II.1.2.2
spa elike omponent
non-relativisti I.8.2.2, I.8.3.1, I.9.2.1,
I.9.3.1
relativisti II.7.1.1, II.8.2.1
splitting of
spa etime
non-relativisti I.3.2
relativisti II.3.4, II.6.2.6
ve tors
non-relativisti I.8.2
relativisti II.7.1
ove tors
non-relativisti I.8.3
relativisti II.7.2
tensors, otensors
non-relativisti I.9
relativisti II.8
submanifold VI.4.3

straight line VI.1.5

syn hronization II.3.2.1
Tangent ve tor, spa e VI.4 .6
tensor IV.3.21
eld VI.3.7
produ t IV.3
quotient IV.4
time dilation II.5.6
timelike
non-relativisti I.1.1.3
relativisti II.1.2.2
timelike omponent
non-relativisti I.8.2.2, I.8.3.1, I.9.2.1,
I.9.3.1
relativisti II.7.1.1, II.8.2.1
time
observed between : : : II.5.6.1
passed between : : : II.2.2.2
time unit I.10.2.2
time and distan e unit II.9.2.2
tunnel paradox II.5.4
twin paradox II.5.7

U -line
non-relativisti I.3.2.1
relativisti II. 3.1.3
U -spa e
non-relativisti I.3.2.1
relativisti II.3.1.3
U -surfa e II.6.2.4
U -time II.3.2.2
Ve tor eld VI.3.7
ve tor observed by
non-relativisti I.7.1.1
relativisti II.5.5.1
ve tor transformation rule
non-relativisti I.8.2.4
relativisti II.7.1.4
ve torized splitting I.4.1.4
velo ity
non-relativisti I.2.1.3
relativisti II.2.3.4
world horizon II.2.5
world line
non-relativisti I.2.1.1
relativisti II.2.1

world line fun tion

non-relativisti I.2.1.1
relativisti II.2.3.3
world line
inertial
non-relativisti I.2.3.1
relativisti II.2.4.2

world line
uniformly a elerated
non-relativisti I.2.3.1
relativisti II.2.4.2
twist-free
non-relativisti I.2.3.1
relativisti II.2.4.2
world surfa e II.6.2.2

LIST OF SYMBOLS

1. Basi notation

:= =:

N
R
R+

R+
0

Xn
Dom f
Ran f
f : X!Y
f:X Y

7!
f jA
f g
1

gf

 fi

i2I

n f

marks the end of a proposition, a proof or a denition, if

ne essary
dening equalities; the symbol on the side of the olon is
dened to equal the other one
the void set
the set of non-negative integers
the set of real numbers
the set of positive real numbers
the set of non-negative real numbers
the n-fold Cartesian produ t of the set X with itself (n 2 N )
the domain of the map f
the range of the map f
f is a map with Dom f = X; Ran f  Y
f is a map with Dom f  X; Ran f  Y
the symbol showing a mapping rule
the restri tion of the map f onto A \ Dom f
the map g is an extension of f; i.e.
Dom f  Dom g; gjDom f = f




the total inverse of the map f : X Y; i.e. if H  Y then

1
f (H) = fx 2 Dom f j f (x) 2 Hg
the omposition of the maps g : Y Z and f : X Y;
1
Dom (g f ) := f (Dom g) \ Dom f; x 7! g(f (x))
the
 of the maps fi : Xi 7! Yi (i 2 I) :

 produ t
 Cartesian


 Xi ! i2I Yi ; (xi )i2I 7! fi (xi ) i2I
i2I
the n-fold Cartesian produ t of f with itself
(n 2 N )

(fi )i2I
Ker L
prk
idX

the joint of the maps fi
: X ! Yi ;

X !  Yi ; x 7! fi (x) i2I
i 2I
1

:= L f0g; the kernel of the linear map L

RN ! R; the k -th oordinate proje tion
the identity map X ! X; x 7! x

2. Other notations

>

^
_
?

ar
A(h)

b
bu

Bu
CU

E
Eu
EU
EU

det
DF

HU
HU ;o

hu

marks the dual of ve tor spa es and the transpose of linear

maps, IV.1.1, IV.1.4
marks adjoints of linear maps, V.1.5
tensor produ t, IV.3.2
antisymmetri tensor produ t, IV.3.14
symmetri tensor produ t, IV.3.14
q ? t is the unique element in the interse tion of q and t; I.2.2,
II.3.4.2
the arrow of spa etime transformations, I.11.3.1, II.10.1.1
the set of h-antisymmetri maps, V.2.7
in parti ular, A(b) : V.3.8
A(g) : V.4.15
Eu lidean form, V.3.1, I.1.1.2
Eu lidean form on Eu; II.1.3.3
the set of relative velo ities with respe t to u; II.4.2.5
CU (x) is the U -line passing through x (U -spa e point that
x is in ident with), I.3.2.2, II.3.1.3
the set of spa elike ve tors, I.1.2.2
the set of ve tors g-orthogonal to u; II.1.3.2
U -spa e, I.3.2.1, II.3.1.3
the set of spa e ve tors of a rigid observer U ; I.4.3.4
determinant, IV.3.18
derivative of F; VI.3.3
Lorentz form, V.4.1, II.1.2.1
:= idM ; II.1.3.6
Galilean group, I.11.3.1
splitting a ording to U ; I.3.2.2, II.3.4.2
splitting a ording to (U ; o); I.4.1.4, I.4.4.2, II.3.4.3
ve tor splitting, I.8.2.1, II.7.1.1

hu;o
Hu0 u

i
iu

IS
IU

La( )

L(u0 ; u)

Oo
O(b)
O(bu )
u

Pu

RU (t; to )
ru
Ru0 u

sign

SO(b)


S
U
Tx (

T n( )
Tx

Tu
vu0 u

V(1)
V(0)

= HU ;o for inertial U with value u; I.11.8.1,

II.10.5.1
the ve tor transformation rule, II.8.2.4, II.7.1.4
embedding of E into M; I.1.2.1
embedding of Eu into M; II.1.3.2
time a ording to the simultaneity S ; II.6.2.3
U -time, II.3.2.2, II.6.2.4
Lorentz group, II.10.1.1
Lie algebra of a Lie group, VII.3.3, VII.4.2
Lorentz boost, II.1.3.8
spe ial Galilean transformation, I.11.3.7
Noether group, I.11.6.1
E ; I.1.2.5
:= D
ve torization with origin o; VI.1.1
group of orthogonal transformations, I.11.1.1
group of orthogonal transformations, II.10.1.4
proje tion along u; I.1.2.8, II.1.3.2
Poin are group, II.10.3.1
u-spa elike inversion I.11.3.4, II.10.1.3
rotation of a rigid observer, I.4.2.2
ove tor splitting, I.8.3.1, II.7.2.1
the ove tor transformation rule, I.8.3.2, II.7.2.2
sign of permutations, IV.3.14
sign of spa etime transformations, I.11.3.1,II.10.1.1
group of rotations, I.11.1.2
time evaluation, I.1.2.2
time evaluation of a simultaneity, II.6.2.3
U -time evaluation, II.3.4.1, II.6.2.4
tangent spa e at x; VI.4.6
translation group, VII.3.1
translation by x; VII.3.1
u-timelike inversion, I.11.3.4, II.10.1.3
relative velo ity, I.6.2.2, II.4.2.2
the set of velo ities, I.1.2.7, II.1.2.1
the set of lightlike velo ities, II.4.7.1

COMMENTS AND BIBLIOGRAPHY

The fundamental notions of spa e and time appear in all bran hes of physi s,
giving a general ba kground of phenomena. Nowadays the mathemati al way of
thinking and speaking be omes general in physi s; that is why it is indispensable
to onstru t mathemati ally exa t models of spa etime.
Sin e 17 years an edu ational and resear h programme has been in progress
at the Department of Applied Analysis, Eotvos Lorand University, Budapest, to
build up a mathemati al theory of physi s in whi h only mathemati ally dened
notions appear. In this way we an rule out ta it assumptions and the danger
of onfusions, and physi s an be put on a rm basis.
The rst results of this work were published in two books:
[1 Matol si, T.: A Con ept of Mathemati al Physi s, Models for Spa etime,
Akademiai Kiado, 1984;
[2 Matol si, T.: A Con ept of Mathemati al Physi s, Models in Me hani s,
Akademiai Kiado, 1986.
Sin e that time our tea hing experien e revealed that a mathemati al treatment of spa etime ould laim more interest than we had thought it earlier. The
notions of the spa etime models throw new light on the whole physi s, a number
of relations be ome learer, simpler and more understandable; e.g. the old problem of material obje tivity in ontinuum physi s has been ompletely solved, as
dis ussed in:
[3 Matol si, T.: On Material Frame-Indieren e, Ar hive for Rational Me hani s and Analysis, 91 (1986), 99{118.
That is why it seems ne essary that spa etime models be formulated in a way
more familiar to physi ists; so they an a quire and apply the notions and results
more easily. The present work is an enlarged and more detailed version of [1.
The notations (due to the dot produ t) be ame simpler. The amount of applied
mathemati al tools de reased (by omitting some marginal fa ts, the theory of

smooth manifolds ould be eliminated), the material, the explanations and the
number of the illustrative examples in reased.
There is only one point where the new version ontradi ts the former one be ause of the following reason. In the literature one usually distingushes between
the Lorentz group (a group of linear transformations of R4 ) and the Poin are
group, alled also the \inhomogeneous Lorentz group" (the Lorentz group together with the translations of R4 ): In our terminology, one onsiders the arithmeti Lorentz group whi h is a subgroup of the arithmeti Poin are group. However, we know that in the absolute treatment the Poin are group onsists of
transformations of the ane spa e M; whereas the Lorentz group onsists of
transformations of the ve tor spa e M; the Lorentz group is not a subgroup of
the Poin are group. Spe ial Lorentz transformations play a fundamental role in
usual treatments in onne tion with transformation rules.
The ounterpart of the Poin are group in the non-relativisti ase is usually
alled the Galilean group and one does not determine its ve torial subgroup that
orresponds to the Lorentz group. The spe ial Galilean transformations play a
fundamental role in onne tion with transformation rules. In the absolute treatment we must distinguish between the transformation group of the ane spa e
M and the transformation group of the ve tor spa e M whi h is not a subgroup
of the former group. The spe ial Galilean transformations turn to be transformations of M; that is why I found it onvenient to all the orresponding linear
transformation group the Galilean group and to introdu e the name Noether
group for the group of ane transformations.
In the former version I used these names inter hanged be ause then group
representations (applied in me hani al models) were in my mind and it es aped
my attention that from the point of view of transformation rules | whi h have
a fundamental importan e | the present names are more natural.
The present treatment of spa etime is somewhat dierent from the usual ones;
of ourse, there are works in whi h elements of the present models appear. First
of all, in
[4 Weyl, H.: Spa e{Time{Matter, Dover publ. 1922
spa etime is stated to be a four-dimensional ane spa e, the bundle stru ture
of non-relativisti spa etime (i.e. spa etime, time and time evaluation) and
the Eu lidean stru ture on a hyperplane of simultaneous world points appear
as well. However, all these are not olle ted to form a lear mathemati al
stru ture; moreover, the advantages of ane spa es are not used, immediately
oordinates and indi es are taken; thus the possibility of an absolute des ription
is not utilized.
A similar stru ture (\neo lassi al spa etime": spa etime and time elapse) is
expounded in

[5 Noll, W.: Le tures on the foundation of ontinuum me hani s and thermodynami s, Ar h.Rat.Me h. 52 (1973) 62{92.
In these works time periods and distan es are onsidered to be real numbers.
The notion of observer remains undened; even this undened notion is used to
introdu e e.g. dierentiability in the \neo lassi al spa etime".
When omparing our notions, results and formulae with those of other treatments, using the phrases \in most of the textbooks", \in onventional treatments" we refer e.g. to the following books:
[6 Fren h, A.P. Spe ial Relativity, Norton, New York, 1968
[7 Essen, L.: The Spe ial Theory of Relativity, Clarendon, Oxford, 1971
[8 Mller, C.: The Theory of Relativity, Oxford University Press, 1972
[9 Taylor, J.G.: Spe ial Relativity, Clarendon, Oxford, 1975
[10 Bergmann, P.G.: Introdu tion to the Theory of Relativity, Dover publ.,
New York, 1976
General relativity, i.e. the theory of gravitation is one of the most beautiful
and mathemati ally well elaborated area of physi s whi h is treated in a number
of ex ellent books, e.g.
[11 Misner, C.W.{Thorne, K.S.{Wheeler, J.A.: Gravitation, W.H.Freeman &
Co., 1973
[12 Adler, R.{Bazin, M.{S hier, M.: Introdu tion to General Relativity,
M Graw-Hill, 1975
[13 Ohanian, H.C.: Gravitation and Spa etime, W.W.Norton & Co., 1976
[14 Rindler, W.: Essential Relativity. Spe ial, General and Cosmologi al,
M Graw-Hill, 1977
[15 Wald, R.: Spa e, Time and Gravity, Chi ago Press, 1977
To understand the non-relativisti and spe ial relativisti spa etime models,
it is su ient to have some elementary knowledge in linear algebra and analysis. Tensors and tensorial operations are the main mathemati al tools used
throughout the present book. Those familiar with tensors will have no di ulty
in reading the book. The ne essary mathemati al tools are summarized in its
se ond part where the reader an nd a long and detailed hapter on tensors.

The book uses the basi notions and theorems of linear algebra (linear ombination, linear independen e, linear subspa e et .) without explanation. There
are many ex ellent books on linear algebra from whi h the reader an a quire
the ne essary knowledge, e.g.
[16 Halmos, P.R.: Finite Dimensional Ve tor Spa es, Springer, 1974
[17 Smith, L.: Linear Algebra, Springer, 1978
[18 Grittel, D.H.: Linear Algebra and its Appli ations, Harwood, 1989
[19 Fraleigh, J.B.{Beauregard, R.A.: Linear Algebra, Addison{Wesley, 1990
Some notions and theorems of elementary analysis (limit of fun tions, ontinuity, Lagrange's mean value theorem, impli it mapping theorem, et .) are used
without any referen e; the following books are re ommended to be onsulted
[20 Zamansky, M.: Linear Algebra and Analysis, Van Nostrand, 1969
[21 Rudin, W.: Prin iples of Mathemati al Analysis, M Graw Hill, 1976
[22 Aliprantis, C.D.{Burkinshow, O.: Prin iples of Real Analysis, Arnold, 1981
[23 Haggarty, R.: Fundamentals of Mathemati al Analysis, Addison-Wesley,
1989
[24 Adams, R.A.: Cal ulus: a Complete Course, Addison{Wesley, 1991
From the theory of dierential equations only the well-known existen e and
uniqueness theorem is used whi h an be found e.g. in
[25 Hyint-U Tyn: Ordinary Dierential Equations, North-Holland, 1978
[26 Birkho, G.{Rota, G.C.: Ordinary Dierential Equations, Wiley, 1989
The present book avoids the theory of smooth manifolds though it would be
useful for the investigation of the spa e of general observers and ne essary for
the treatment of general relativisti spa etime models. The following books are
re ommended to the reader interested in this area:
[27 Boothby, W.M.: An Introdu tion to Dierentiable Manifolds and Riemannian Geometry, A ademi Press, 1975

[28 Choquet-Bruhat, Y.{Dewitt-Morette, C.: Analysis, Manifolds and Physi s,

North-Holland, 1982
[29 Abraham, R{Marsden, J.E.{Ratiu, T.: Manifolds, Tensor Analysis, and
Appli ations, Springer, 1988
Non-relativisti and spe ial relativisti spa etime models involve some elementary fa ts about ertain Lie groups. Those who want to get more knowledge
on Lie groups an study, e.g. the following books:
[30 Warner, F.W.: Foundations of Dierentiable Manifolds and Lie Groups,
Springer, 1983
[31 Sattinger, R.H.{Weawer, O.L.: Lie groups and Lie algebras with Appli ations to Physi s, Geometry and Me hani s, Springer, 1986