Progress in Physics

Vol.3

Edited by
A. Jaffeand
D. Ruelle

Springer Science+Business Media, LLC

Yu. I. Manin
Mathematics
andPhysics
Translated by
Ann and Neal Koblitz

1981 Springer Science+Business Media, LLC

Author
Yu. 1. Hanin
Steklov Hathematical Institute
Vavi lova 42
117966, GSP-1
Hoscow
USSR

Library of Congress Cataloglng in Publication Data

Hanin, IU. 1.
Hathematics and Physics.
(Progress in Physics ; 3)
Translation of: Hatematika fizika.

1. Hathematics--1961- 2. Physics. 1. Title.

11. Series: Progress In Physics (Boston, Mass.); 3.
QA37.2.M3413 530.1'5 81-1319
AACR2

CIP- Kurztitelaufnahme der Deutschen Bibliothek

Hanin, Jurij 1.:
Hathematics and Physics 1 Yu. 1. Manin.
Translated by Ann and Neal Koblitz
Boston: Basel: Stuttgart: Birkh~user, 1981
(Progress In Physics ; 3)
ISBN 978-1~84-8
NE:GT
Orlginally published In Novoye v Zhizny, Naukye, Tekhnlke,
Seriva Hatematika, Kibernetika, Hoscow: Znaniye, No.12, 1979.

All rights reserved. No part of this publication may be

reproduced, stored in a retrieval system, or transmitted,
in any form or by any means, electronic, mechanical, photo-
copylng, recordlng or otherwise, without prior permisslon
of the copyright owner.
© Springer Science+Business Media New York 1981
Originally published by Bi rkhl:luse r Boston in 1981
Softcover re print of the hardcover 1st edition 1981
ISBN 978-1-4899-6784-8 ISBN 978-1-4899-6782-4 (eBook)
DOI 10.1007/978-1-4899-6782-4
 CONTENTS

FOREWORD ix

1. A BIRD'S EYE VIEW OF MATHEMATICS 1

Mathernatical Truth
Sets
Higher-Dimensional Space and the Idea of Linearity
Measurernents in a Linear Space
Nonlinearity and Curvature
Sorne Novelties
Sets, Forrnulas, and the Divided Brain

2. PHYSICAL QUANTITIES, DI~ffiNSIONS AND CONSTANTS:

The Source of Nurobers in Physics 35

The Spectrurn of a Scalar Quantity

Physical Law, Dimension and Sirnilarity
The Group of Dimensions of Newtonian Mechanics
Scale Invariance
Planck Units and the Problern of a Unified Physical
Theory
Classification of Physical Constants

3. A DROP OF MILK: Observer, Observation, Observable

and Unobservable 52

Isolated systerns
Principles of Quantum Description
Phase Space
Phase Curves
Observation: Ovens, Filters and Quantum Jumps-
Combination of Quantum Systems

4. SPACE-TIME AS A PHYSICAL SYSTEM 71

The N.inkowski World

Curved World
Spinors, Twistors and the Cornplex World
Space-Tirne, Gravitation and Quantum r1echanics

5. ACTION AND SYMf.ffiTRY 88

The connection between mathematics and

physics .•. Is it only that physicists

talk in the language of mathematics?

It is more. This book describes how

mathematics associates to some important

physical abstractions (modelsl its own

mental constructions, which are far

removed from the direct impressions of

experience and physical experiment.

vii
 FOREWORD

There is a story about how a certain well-known rnatherna-

tician would begin his sophornore course in logic. "Logic is

the science of laws of thought," he would declairn. "Now I

rnust tell you what science is, what law is, and what thought
is. But I will not explain what 'of' rneans."

On undertaking the task of writing the book "Mathernatics

and Physics", the author realized that its size would hardly
be sufficient to atternpt to explain what the "and" rneans in
the title. These two sciences, which were once a single
branch on the tree of knowledge, have by our time becorne
rather separate. One of the reasons for this is that in this
century they have both engaged in introspection and internal
developrnent, i.e., they have constructed their own rnodels
with their own tools. Physicists were disturbed by the
interrelation between thought and reality, while rnatherna-
ticians were disturbed by the interrelation between thought
and forrnulas. Both of these relations turned out to be rnore
cornplicated than had previously been thought, and the rnodels,
self-portraits, and self-irnages of the two disciplines have
turned out to be very dissirnilar. As a result, frorn their

earliest student days rnathernaticians and physicists are

taught to think differently. It would be wonderful to rnaster

ix
 X

both types of professional thinking just as we master the

use of both a left and a right hand. But this book is like

a melody without accompaniment.

The author, by training a mathematician, once delivered

four lectures to students under the title "How a mathema-

tician should study physics". In the lectures he said that

modern theoretical physics is a luxuriant, totally Rabelai-

sian, vigoraus world of ideas, and a mathematician can find

in it everything to satiate hirnself except the order to which

he is accustomed. Therefore a good method for attuning one-

self to the active study of physics is to pretend that you

are attempting to induce this very order in it.

In the book, which evolved from these lectures and later

reflections, I have tried to select several important

abstractions of the two sciences and make them confront one

another. At the very highest level such abstractions lose

terminological precision and are capable of becoming cultural

symbols of the time: we recall the fate of the words "evo-

lution", "relativity", or "the subconscious". Herewe go

down to a lower level and discuss words which, although not

yet symbols, have passed beyend being merely terms: "set",

"symmetry", "space-time". (Cf. M.M. Bakhtin' s attempt to

introduce this last concept into literary criticism through

the deliberately foreign term "chronotope".) Seme of these

words occur in the titles of chapters. Each reader must

have in his or her mind preliminary images of these concepts,

images which have a physical origin in the broad sense of the

word.
 xi

The author wishes to show how mathema tics associate s

new mental images with such physical abstracti ons; these

images are almost tangible to the trained mind but are far

removed from those that are given directly by life and

physical experien ce. For example, a mathema tician represen ts

the motion of the planets of the Solar System by a flow line

of an incompre ssible fluid in a 54-dimen sional phase space,

whose volume is given by the Liouvill e measure.

The reader may need an effort of will to perceive

mathemat ics as a tutor of our spatial imaginat ion. More

frequent ly one associate s mathema tics with rigorous logic and

computat ional formalisr n. But this is only disciplin e, a

ruler with which we are being taught not to die.t

The computat ional forrnalism of mathema tics is a thought

process that is external ized to such a degree that for a time

it becomes alien and is turned into a technolo gical process.

A matherna tical concept is formed when this thought process,

temporar ily removed frorn its human vessel, is transplan ted

back into a human mold. To think •.. means to calculat e with

critical awarenes s.

The "mad idea" which will lie at the basis of a future

fundamen tal physical theory will come from a realizati on

that physical meaning has sorne mathema tical form not pre-

viously associate d with reality. From this point of view

T;,.n allusion to Boris Pasterna k's poem "To Briusov" , in which

the poet says: "There was a time when you yourself each
morning/ Taught us with the strict beat of your ruler how
not to die." (Transla tor's note)
 xii

the problern of the "mad idea" is a problern of choosing, not

of generating, the right idea. One should not understand
this too literally. In the 1960's it was said (in a certain
connection) that the most important discovery of recent years
in physics was the complex numbers. The author has something
like that in mind.

I do not wish to apologize for the subjectivity of the

opinions and choice of material. Physics and mathematics
have been written about by Galileo, Maxwell, Einstein,
Poincare, Feynman, and Wigner, among others; only the hope
of saying something subjective of one's own can justify a new
attempt.
 1. A BIRD'S EYE VIEW OF MATHEMATICS

Mathematical Truth

The simplest mathematical operations are arithmetical

computations of the following sort:

0.038 .

For verisimilitude this example was not copied from a

student's workbook, but from a paper by Enrico Fermi and E.

Amaldi "On the absorption and the diffusion of slow neutrons".

Let's think a little bit about what this computation means.

a) To check this equality we can agree that it relates

to integers (by squaring, clearing the denominators, and

taking our unit to be a thousandth part of one). Our equal-

ity can then be considered as a prediction about the result

of some "physical experiment", namely the following: take

two groups of 48 objects (2 x 0. 048), repeat this 78000 times

(x 7.8 x 10 4 ), etc. In such a way first-graders make little

blocks into piles in order to learn the meaning of "counting~

"integer", "addition" and "multiplication", and also the

meaning of arithmetic identities. Thus, it is reasonable to

think of integer arithmetic as the "physics of collecting

objects into piles".

1
 2

b) Nevertheless, practical computation is, of course,

carried out differently; it consists of a series of certain

standard transformations of the left hand side of the iden-

tity. We take a group of symbols on the left, say 0.25 .;- 20,

and replace it by 0.0125 using schoolbook recipes, and so on.

All the rules, including the rule about the order of opera-

tions, can be stated beforehand. The infallibility of a com-

putation lies in its grammatical (prescriptive) correctness;

it is this which ensures the "physical truth" of the result.

(Of course, Fermi and Amaldi round off the left hand side; it

is clear even without calculations that their equality cannot

be literally true, since VI3 is an irrational number.)

c) For Fermi and Amaldi the meaning of this computation

can be summarized in the following way: "We found that the

narrowest band is group A (radiation strongly absorbed by

silver), for which Lw/w=0.04 ... the probability that a

neutron remains in the same group after one free path is

fairly low." (The 0.04 is the 0.038 on the right band side.)

Clearly we cannot immediately arrive at this conclusion, no

matter how we may imagine the meaning of an arithmetical com-

putation. Neither the laying out of78000piles with 96

objects in each, nor the division of 0.096 by 0.04 has in it-

self any relation to neutrons. Mathematical reasoning enters

into the physics article along with the act of giving it a

physical interpretation; it is this act which is the most

striking feature of modern physics.

Be that as it may, even our simple example indicates

three aspects of mathematical truth, which can conditionally

 3

be designated as set-theoretie truth, formal dedueibility or

provability, and adequaey of the physieal model.

For mathematics, as long as it is self-contained, only

the first two aspects are essential, and the difference
between these two aspects was not understood until the
twentieth century. Consider an assertion that can be stated
simply, such as the conjecture of Fermat. Although we do not
know how to prove it, or to refute it, we can be sure that it
is either true or false. This confidence is based on the
abstract possibility of performing infinitely many arithmetic
operations (or "laying out piles"), in this case compiling
all sums of powers of pairs of integers. In general, the
notion of the truth of (most) mathematical assertions relies
upon being able to imagine such infinite series of tests.
However, every mathematical proof, i.e., argument consisting
of successive applications of axioms or logical rules of
deduction, is an essentially finite procedure. In the 1930's
Kurt Gödel proved that for this reason provability is signif-
icantly narrewer than truth, even when one only talks about
the integers. In this connection it does not matter what
axioms we start from, as lang as they are true and are given
by a finite list (or by a finite nurober of rules for generating
them). This difference between truth and provability is widely
known nowadays, but it seems that its consequences are poorly
understood. Much is written discussing the problern of
reductionism: Does biology or chemistry reduce to physics?
It is clear that in this context one can speak only about
some theoretical model of physical, biological or chemical
 4

phenomena, which is, moreover, sufficiently mathematicized.

But then one must explain what is meant by reducibility--

whether it is an abstraction of a type of truth or of a type

of deducibility from axioms. Thinking through both possibil-

ities, we are struck with the impression that, when speaking

about the reducibility of laws, we simply do not understand

what we are talking about.

Sets

Modern notions of mathematical truth are connected with

the development of two major conceptions: set-theoretic

mathematics and the mathematics of formal languages. Every-

one is familiar with some mathematical formalism. The mathe-

matical baggage of the student typically may consist of the

ability to carry out arithmetic operations with numbers in

the decimal system, transform algebraic identities, diff~r­

entiate and compute some integrals. This language of calculus,

which had already been essentially developed by the time of

Euler and Lagrange, has turned out to be very convenient,

effective for the solution of problems, and accessible for

mass study. Parallel to this language there has been a

developing understanding of what this language talks about,

i.e., the meaning of such concepts as v:I, function, differ-

ential, etc. has been elucidated. Here an immense role was

played by geometric representations: complex numbers lost

their mystery only after Argand and Gauss proposed a coherent

interpretation of them by points in the Euclidean plane;

differentials were eventually interpreted using the notion of

 5

a tangent space, and so on. Set-theoretic concepts have

lain a universal foundation for defining all mathematical
constructions using some such "generalized geometric" image.
These mental constructs represent both a receptacle for the
meaning of mathematical formalisms and a means for extracting
the meaningful assertions from the vast sea of derivable
mathematical formulas.

In this book we wish to demonstrate the value of such

mental constructions as an intermediary between mathematics

and physics. Of course, the possibilities for giving a
popular presentation of these constructions are limited. In

99 pages we cannot explain their exact meaning or teach the

reader how to use them to solve problems. But perhaps some

ideas of mathematics and theoretical physics will become

clearer. The difficulty of understanding the concepts of
quantum theory or general relativity theory is in part con-
nected with the fact that attempts to explain them omit this
act of giving an intermediate set-theoretic interpretation of
the mathematical models. Even in university teaching, in-
sufficient attention is payed to this; communication between
a physicist and a mathematician is frequently complicated by
the physicist's inclination to pass from formulas directly to
their physical meaning, omitting the "mathematical meaning".
It should be noted, incidentally, that the situation has
noticeably improved in recent years.

A good physicist uses formalism as a poet uses language.

He justifies the neglect of the commands of rigor by an

eventual appeal to physical truth, as a mathematician cannot
 6

permit hirnself to do. The choice of a Lagrangian in the

unified theory of weak and electromagnetic interactions of

Salam and Weinberg, the introduction of Higgs fields in it,

the subtraction of vacuum expectation values and other sorcery,

which leads, say, to the prediction of neutral currents--all

this leaves the mathematician dumbfounded.

But we return to sets.

The most important sets for physicists are not sets of

objects, but sets of possibiZities: the configuration space of a

system is the set of its possible instantaneous states, space-

time is the set of possible events going off like "flashbulbs",

each of which gives a space-time point. A physicist is usu-

ally in a hurry to introduce coordinates in this space, i.e.,

functions with numerical values. If a set of n such func-

tions allows one to coordinatize the points of the set in a

unique way, then one may suppose that this set lies in n-

dimensional real space Rn, which consists of vectors of the

form (a 1 , ••. ,an). "Figures", i.e., subsets of this space;

measurements of distances, angles, volumes, etc. in it; and,

finally, its motions or its mappings into itself, all make up

the physicist's main arsenal of geometric forms. It is

important here that the dimension n can be as large as

desired, even infinite. In a rigorous mathematical text the

definition of infinite dimensionality must be introduced

separately, but here we shall imagine infinite dimensionality

as being "indefinitely large finite dimensionality". If the

coordinates take complex values, then our sets are in cn.

But in general it is not necessary to dwell upon coordinates.

 7

Physically they are often the vestiges of oversimplified re-

presentations of an Observation; mathematically they are a

reminder of a time when there was no language in which one

could meaningfully discuss sets other than sets of numbers or

vectors.

Set-theoretic language is good in that it does not force

one to say anything superfluous. Georg Cantor defined a set

as "any collection into a whole of definite and separate

objects of our intuition or our thought". This isthebest

way to explain a set as a concept which aids in the percep-

tion of the world.

Higher-Dimensional Space and the Idea of Linearity

If a car moves twenty meters in a second, then in two

seconds it will probably have gone forty meters. If a weak

wind deflects a light bullet by three centimeters, then a

wind that is twice as strong will deflect it by six centi-

meters. The response to small perturbations depends linearly

on the perturbation--this is a principle of natural science

which lies at the basis of a vast nurober of mathematical

models. A mathematician turns this principle into the defi-

nition of a differentiable function and into the postulate

that most processes are described by such functions most of

the time. Do Hooke's law of elasticity and Ohm's law say

something more than this principle of linear response to

small forces? The answer is yes if it turns out that the

laws still remain valid for somewhat larger perturbations.

 8

Linear space is an idealization of "arbitrarily large

small perturbations". One need not introduce Coordinates; it

is only necessary to keep in mind that the elements of a

linear space (vector space) can be added tagether and multi-

plied by numbers (real or complex--this epithet is added to

the name of the space). The archetypical geometric construct

is our "physical space" of dimension three; the spaces Rn

and cn with coordinate-wise addition and multiplication by

numbers exhaust all the finite dimensional vector spaces.

The dimension of a vector space is the number of inde-

pendent linear coordinate functions on it. There is a

theorem that says that the dimension does not depend on the

choice of these coordinate functions. For all its simplicity,

this theorem is a deep result. It establishes the first con-

nection between something continuous and something discrete:

an integer--the dimension--appears for the first time not as

the number of objects or discrete forms, but as a measure of

the size of a continuous entity.

A linear mapping or operator is the idealization of a

linear response to arbitrary perturbations. The response can

be measured by elements of the same space as the perturbation

or by elements of another space; in any case, it is a mapping

from a vector space to a vector space which takes a sum of

vectors to the sum of their images and the product of a

vector by a number to the product of the image by the same

number.

Every linear mapping of a one-dimensional space into it-

self is multiplication by some number--the "magnification
 9

coefficient". In the complex case the geometric form is a

bit more complicated: since one-dimensional complex space is

viewed as a real plane, multiplication by a complex number is

a combination of a real dilatation and a rotation. Pure

rotations, i.e., multiplication by numbers of modulus one,

play a large role in quantum mechanics: the law of evolution

of an isolated quantum system is formulated in terms of such

rotations.

An important class of linear mappings of an n-dimensional

space into itself consists of dilatations in n independent

directions with a corresponding magnification coefficient in

each direction. The set of "coefficients of dilatation" of a

linear operator is called its spectrum: the homonymy with a

term from physics reflects a deep connection.

In quantum physics the idea of linearity acquires a

fundamental physical meaning thanks to the basic postulate on

Superposition of quantum states. In classical physics and

mathematics, in addition to the original idea that "anything"

can be linearized on a small scale, a large role is played by

the Observation that functions on any set (all of them, or

only continuous ones, or differentiable functions, or Riemann

integrable functions, etc.) form a vector space, because they

can be added to each other and multiplied by numbers. Most

function spaces are not finite dimensional, but the possibil-

ity of systematically learning to apply to them our originally

developed finite dimensional (even three-dimensional) intu-

ition has turned out to be an extremely fruitful discovery.

 10

This was taught to us in the twentieth century by David

Hilbert and Stefan Banach.

Measurernents in a Linear Space

In three-dirnensional physical space there exist rigid

bodies that preserve sorne "self-identity" in large space-tirne

dornains. This is the basis of all physical rneasurernent. It
is not at all obvious beforehand which of the idealized
properties of physical rneasurernent will be rnost useful in
rnathernatical theory and in applications. In fact, the rnathe-
rnatical notions that are connected with ideas of classical
rneasurernent form a cornplex conceptual design. We can irnrne-
diately narne sorne of the concepts: length, angle, area,
scalar product, rnotion.

So that the reader does not get a false irnpression, we

rernark that the concept of a vector space by itself contains
nothing that allows us to rneasure anything in a unique way.
Vectors have no length (although proportional vectors have a
natural ratio of lengths), there is no natural rneasure of the
angle between two vectors, and so on. Therefore, for a rnathe-
rnatical forrnulation of the idea of rneasurernent we rnust intro-
duce an additional geornetric concept, or even several con-
cepts; to use rnathernatical jargon, we have to provide the
space with additional structure.

The first of these concepts is the unit sphere of the

space: the set of vectors of unit length. If any nonzero
vector lands on the unit sphere after we rnultiply it by sorne
 11

nurober a, we can write down the length of the vector, which

will be lal- 1 ; hence, a should be uniquely determined up to

multiplication by a nurober of modulus one. The distance

between two vectors x and y can be defined as the length

of their difference lx-yl. If nonzero vectors have nonzero

length, if the so-called triangle inequality holds:

lxl + IYI, and if in addition we have a condition guaranteeing

the existence of limits of Cauchy sequences, then we are led

to the concept of a Banach space. Many useful function

spaces are Banach spaces.

An arbitrary Banach space, however, is not sufficiently

symmetric to be considered the correct generalization of the

unit sphere in three-dimensional space. There are two

different methods for obtaining the necessary symmetry:

(a) require that the sphare be taken into itself by some

n(n-1)/2-dimensional continuous group of linear mappings

(where n is the dimension of the space); (b) require that

any two vectors in the space have a scalar product (x,y)

(this is a linear function with respect to each argument in

the real case and is a little more complicated in the complex

2
case) such that lxl = (x,x) for all x. The first method

is a generalization of the idea that rigid bodies can rotate,

while the second method is a generalization of the idea that

one can measure the angle between two vectors in a way that

does not vary under rotations of the pair of vectors. Both

ideas are closely connected to one another, and lead to the

concept of higher-dimensional Euclidean space (in the complex

case it is called Hilbert space) • In suitable coordinates

 12

the unit sphere in such a space is given by the usual

n 2
equation I: I x. I = l.
Rotations are linear mappings from
i=l l
this sphere into itself; they form a group which is denoted

O(n) in the real case and U(n) in the complex case. In

real Euclidean space the scalar product takes real values and

is syrnrnetric: (x, y) = (y, x) • In complex Euclidean space it

takes complex values, and we have to take the complex con-

jugate when the vectors are transposed: (x,y) = (y,x). In

both cases we have the important inequality I (x,y) 1 2 ~

lxl 2 • IYI 2 , so that the nurnber (x,yl/lxl • IYI is no greater

than one in modulus. In a real space this number is real,

and so there exists an angle 4J for which cos(jl = (x,yl/lxl·

· IYI. In the complex case this angle can be defined by:

cos(jl =I (x,y) l/lxi·IYI· The right hand side here only takes

values lying between zero and one, and we note that there is

a remarkable physical quantity with the same property--a

probability. In quantum mechanics the numbers 2

cos 4J are

interpreted as probabilities--we shall discuss this in more

detail below. In high school geometry two vectors x and y

are said to be orthogonal if the cosine of the angle between

them is equal to zero, i.e., (x,y) = 0; the same terminology

is used in the general case as well.

If we abandon some of the properties of being Euclidean,

then the concept of scalar product can lead to several new

important classes of linear geometries. For example, in R4

the formula The one minus sign leads

to many differences from the Euclidean case: for example,

 13

there are whole lines consisting of vectors of length zero.

They represent light rays in the basic space-time model for

the special theory of relativity--the famous Minkowski space.

If we reject the condition (x,y) = (y,x) and replace it

by (x,y) =- (y,x), then every vector in this type of space

will be "orthogonal to itself"! This geometry, which is

called symplectic geometry, must be studied for a long time

in order to become accustomed to it. The gyroscope that

guides a rocket is an emissary from a six-dimensional sym-

plectic world into our three-dimensional one; in its home

world its behavior looks simple and natural. Although sym-

plectic geometry was discovered in the last century, its role

in physics was underestimated for a long time, and in text-

books this role is still obscured by the old formalism.

But it is time now to return to the Euclidean world,

although in higher dimensions. The last thing we wish to

discuss is the measurement of volume. If are

mutually orthogonal vectors of unit length, then the n-

dimensional unit cube spanned by them is the set of vectors

where It is

natural to stipulate that its volume is equal to one. Trans-

lating this cube by any vector does not vary its volume; the
n
cube with edge of length a is then given volume a • After

this the volume of any n-dimensional figure can be defined by

paving it over by a large nurober of small cubes and adding

their volumes. Problems crop up around the boundary, where a

bit of free space still remains, and in attempting to pave

this over the cubes begin to go outside the boundary. But if

 14

the boundary is not very jagged, then, by dividing the cube

into still smaller ones, we can make the error as small as we

wish. This is the basic idea of integration.

We complete our construction of integration as follows.

Suppose that in our area of space there is something, a "sub-

stance" as was said in the last century, that is characterized

by its density f(x) close to the point x. The total amount

of this substance will be approximately equal to the surn of

the amounts of it in all the cubes of the paving, and the

amount in one cube will be the product of the volume of the

cube and the value of the density at some point in its inte-

rior. The whole sum is the "Riemann sum", and its limit is

the integral of the function f over the volume.

In mathematics it is difficult to find a more classical

and yet a more pertinent concept than the integral. Every

few decades brings new mathematical variants of it, and

physics requires more all the time. The definition of the

Riemann integral, which we gave above, is mathematically

reasonable only for functions f that do not vary too rapid-

ly, such as continuous ones. But almest every physical model,

as soon as it is replaced by a more detailed one, reveals

that the function f, which appeared rather smooth, is in

fact the result of averaging a more complicated "fine-

grained" picture. Charge can be measured by integrating its

density, until we reach a scale where the carriers of the

charge are electrons and ions. The density of the charge on

a point carrier is infinite, while off the carrier it is zero,

 15

and we are forced to construct an apparatus for integrating

such functions.

The remarkable path integrals of Feynman still remain a

challenge to mathematicians. These integrals have already

become a fundamental instrurnent of quantum field theory, but

they are not yet well defined as mathematical objects. Two

circumstances get in the way of understanding them: the

integration must be carried out over an infinite-dimensional

space, and, moreover, it is strongly oscillating functions

that must be integrated. Here '-ITe will say a few words about

the effects of infinite dimensionality, which we can under-

stand naively as very large finite dimensionality.

From a segment of length 1 we take out a middle part of

length 0.9. The length of the remainder is, of course, 10%

of the length of the whole segment. From a disc of diameterl

we extract a concentric disc of diameter 0.9. The area of

the annulus that remains is now 19% of the area of the disc.

From a ball of diameter 1 we remove a concentric ball of dia-

meter 0.9. The volurne of the remaining shell is 27.1% of the

volume of the ball: almost a third, as compared with one-

tenth for the segment. It is not hard to see "physically"

that the volume of an n-dimensional ball of diameter d

should be expressed by the formula c(n)dn, where c(n) is

a constant which does not depend on d. The fraction of the

volume occupied by a concentric ball of diameter 0.9d will

therefore be (0.9)n; it tends to zero as n increases.

Almost two-thirds of a twenty-dimensional watermelon with

radius 20 cm consists of rind if the rind is 1 cm thick:

 16

These calculations lead to a geometric principle: "The

volume of a multidimensional solid is almost entirely con-

centrated near its surface." (It is interesting to take a

cube instead of a ball--the same effect appears as the number

of faces increases.)

Let us imagine the simplest model of a gas: N point

atoms move in a reservoir with velocities vi, i = 1, ... ,N;

each atom has mass m. The kinetic energy E of the gas is

N
2
equal to I: mvi/2; the state of the gas described by a set
i=l
of velocities with fixed total energy E determines a point

on the (N-1)-dimensional Euclidean sphere of radius V2E/m.

For a macroscopic volume of gas under normal conditions, the

dimension of this sphere is of order 10 23 - 10 25 (determined

by Avogadro's number), i.e., it is very large. If two such

reservoirs are joined so that they can exchange energy but

not atoms, and the sum of their energies E = E1 + E 2 remains

constant, then most of the time the energies E1 and E2

will be close to the values that maximize the volume of the

space of states attained by the combined system. This volume

is equal to the product of the volumes of spheres of radii

V2E 1 /m and V2E 2 /m, the first of which grows very rapidly as

E1 increases, while the second rapidly diminishes. There-

fore, their product has a sharp peak at a point which is

easily computed; the point corresponds to the condition that

the temperatures be equal. The "concentration of the volume

of a multidimensional solid near its surface" in essence

 17

ensures the existence of temperature as a macroscopic

variable.

What space are we talking about in this example? The

space of states of a physical system, more precisely, some

quotient space of it (since we do not take into account

either the positions of the atoms or the directions of the

velocity vectors) . A point of this space is again a possibil-

ity. A typical set is not the set of chairs in a room, nor

the set of pupils in a class, nor even the set of atoms in a

reservoir, but rather the set of possible states of the atoms

in the reservoir.

The asymptotic properties of multidimensional volumes

form the geometric arsenal of statistical physics. Nature

constructs all of the variety in the world from a small

nurober of different bricks. Bricks of each type are iden-

tical, and when statistics describes the behavior of con-

glomerates of them, it uses the concept of a point wandering

in domains of an almost infinite dimensional phase space.

Macroscopic Observations allow us only roughly to indicate a

region which contains the point at any moment in time; the

!arger the volume of the region, the more probable it is that

we actually see the point there.

In infinite dimensions, where almost all the domain is

its boundary, in order to find valid rules of thinking and

computing we need some of the most refined tools of the

mathematical arsenal.
 18

Nonlinearity and Curvature

Just as the idea of linearity extrapolates small incre-

ments, the idea of curvature uses such an extrapolation to

study the deviation of a geometric object (for example, the

graph of a function f) from a linear object.

Small dimensions nourish the intuition that engenders

the geometric conception of curvature. The graph of the

2
curve y = ax in the real plane has three basic forms: a

"cup" (convex downward) for a > 0, a "dome" (convex upward)

for a < 0, and a horizontal line for a = 0. The nurober a

determines the steepness of the sides of the cup or dome, and

also the radius of curvature at the lowest (highest) point;

this radius of curvature is equal to l/(2lall. In physical

models of small oscillations a heavy ball rolling in the

bottom of a cup goes through the same oscillations as an

object on a spring, and the equivalent curvature is expressed

through the mass of the ball and the rigidity of the spring.

In higher dimensions the graph of a quadratic function takes

N 2
the form y = r aixi in a suitable coordinate system.
i=l
Among the coefficients there can be positive numbers, nega-

tive numbers and zeros; they determine the nurober of direc-

tions in which the graph goes upward, goes downward, or

remains horizontal. In modern quantum field theory, this

simple model gives the appearance of the spectrum of masses

of elementary particles and the spectrum of forces (constants)

of interaction in an astonishingly large nurober .of situations;

it serves as a first step on the long path to more refined

schemes. There quadratic functions arise in an infinite

 19

dimensional space, and horizontal directions in the graphs

appear because of the action of symmetry groups. When a

function does not change under certain motions of the space

into itself, its graph cannot be a pit, but it can possibly

have the form of a ravine (a ravine can be translated into

itself along the bottom, but a pit cannot).

Thus, the first notion of nonlinearity, which we have

briefly described, is to ask how far a multidimensional sur-

face--the graph of a function--deviates from a linear surface

near a point at which this linear surface is tangent. Imag-

ining the tangent space to be horizontal, we can distinguish

a set of mutually orthogonal directions in it, along each of

which the surface goes upward, goes downward, or remains

horizontal; the speed with which the surface goes upward or

downward is measured by the radius of curvature; there are as

many of these radii of curvature as there are dimensions of

the surface.

To describe the deviation we thus use an "external mold".

This circle of ideas is natural and useful, but Einstein's

theory of gravitation and Maxwell's theory of electromagne-

tism, and also, as we are now beginning to understand, the

theory of nuclear forces and, perhaps, all interactions in

general require finer notions of curvature. The first mathe-

matical theories of "intrinsic curvature" (as opposed to the

previously described "extrinsic curvature") were developed by

Gauss and Riemann.

The concept of intrinsic curvature is constructed first

for a domain in numerical space, in which for each pair of

 20

nearby points one is given a distance between them. The geo-

metry of our domain with the new Riemannian concept of dis-

tance should be Euclidean on an infinitesimal scale. But

various aspects of the concept of curvature show the extent

to which this geometry is different from the flat Euclidean

geometry on a large scale.

In order to explain these aspects it is convenient to

start with the analog of lines in this space ("manifold")--

these are the geodesics, the lines of least length joining

points of the space (for example, arcs of great circles on a

sphere). The length of a curve is, of course, measured by an

integral: the curve must be divided into many small segments,

and the length of each segment is approximately measured by

the distance between its endpoints.

Now we imagine a small vector in the manifold, moving

along a geodesie curve so that its angle with the geodesie

always remains constant. (On a two-dimensional surface this

recipe determines a motion uniquely, but in higher dimensions

we need more information to specify the motion precisely.)

Since the space is almost Euclidean on a small scale, it is

not difficult to give this prescription a precise meaning.

Such a motion of a vector is called a parallel translation.

We can define parallel translation along any curve: as in

the computation of length, the curve should be replaced by a

polygonal line composed of short segments of geodesics, and

then the vector should be translated in a parallel fashion

along these geodesics.

 21

We consider a small closed curve in space--an almost

flat loop. After translating a vector along this loop and

returning to the starting point, we discover that the vector

has rotated by a small angle from its original position; this

angle is proportional to the area of the loop. Moreover, the

coefficient of proportionality depends upon: (a) tne point

about which the loop is situated; (b) the direction of the

two-dimensional area spanned by the loop. This coefficient,

as a function of the point and the two-dimensional direction

at the point, is called the Riemann curvature tensor. For

flat Euclidean space the curvature tensor is identically zero.

It took a long time to see which parts of this construc-

tion are the most important, and to come to the conclusion

that the idea of parallel translation along a curve is the

fundamental one. The geometric picture of curvature that is

most relevant for understanding, for example, the theory of

Yang-Mills fields in modern physics is more general than the

picture of Riemannian curvature. To define Riemannian curva-

ture we translated a vector along a curve in space. Now

imagine this vector as a small gyroscope, and the curve as

its world-line in four-dimensional space-time (this will be

discussed in more detail below, in the chapter on space-time).

Suppose that two gyroscopes in the same initial state diverge

and are then brought together for comparison. Then a pre-

diction of how the directions of their axes differ is a

Straightforward task of physics. Meanwhile, the set of

behaviors that can be imagined for all possible gyroscopes

of this type is a mathematical concept; it is a space

 22

supplemented by rules for parallel translation of tangent

vectors along curves.

Yet one step remains before we can introduce the general

mathematical concept of a space with a connection and the

curvature of the connection. The direction of the axis of

the gyroscope is a special case of the notion that a localized

physical system can still possess internal degrees of freedom.

In classical physics this is an idealization, as a consequence

of which we summarily take into account the composite parts of

the system, their rotations, oscillations, etc.; in quantum

physics one has nonclassical degrees of freedom, such as the

spin or magnetic moment of an electron, which do not reduce

to imaginable beh?J.vior of the "parts" of an electron in space-

time. In general, suppose we are given a pair of spaces M

and E and a mapping f: E + M, where, say, M is a model of

space-time and at each point m of M there is a localized

physical system with space of internal states f- 1 (m). Then

a connection on this geometric object is a rule for trans-

lating the system along curves in M. In other words, if we

know a piece of the world-line of a system in M and the

initial internal state of the system, then, using parallel

translation according to the connection, we know its whole

history. The curvature of a connection measures the differ-

ence between final states of the system, if we go from nearby

initial points of space-time to nearby final points by

different paths, with the systems in the same state at the

beginning.
 23

These ideas connect the geornetry directly with the

physics; we are ornitting the cornplicated turns of brilliant

guesswork, error, forrnalisrn and historical accident which

accornpanied the ernergence of these new ideas and which are
constantly retold in textbooks.

A gravitational field is a connection in the space of

internal degrees of freedorn of a gyroscope; it controls the

evolution of the gyroscope in space-tirne. An electrornagnetic

field is a connection in the space of internal degrees of
freedorn of a quanturn electron; the connection controls its

evolution in space-tirne. A Yang-Mills field is a connection

in the space of colared internal degrees of freedorn of a

quark.

This geornetric picture now seems to be the rnost uni-

versal rnathernatical scherne for the classical description of

an idealized world in which a srnall nurober of basic inter-

actions is considered in turn. Matter in space-tirne is

described by a cross-section of a suitable fibre bundle
E->- M, indicating the state in which this matter is found at

each point at each rnornent. A field is described by a con-

nection on this bundle. Matter affects the connection, irn-
posing restrictions on its curvature, and the connection

affects the matter, forcing its "parallel translation" alon9

world lines. The farnaus equations of Einstein, Maxwell-Dirac

and Yang-Mills are exact expressions of these ideas.

But even without writing down the equations, we can say

a lot. The discovery that the fundamental physical fields

are connections was neither as drarnatic nor as easy to date
 24

precisely as the discevery ef these equatiens. In the theery

ef gravitation, for example, the basic concept fer Einstein

was not a cennectien, but a (pseudo-) Riemannian metric in

space-time. It was Hermann Weyl who first proposed that the

electromagnetic field is a connection, although in pre-quantum

physics he could not correctly indicate the fibre bundle on

which this connection defines parallel translation. Rather,

he guessed that the field varies the lengths of segments

passed along different paths in space-time. Einstein indi-

cated the unsuitability of this view, and the correct bundle

on which the Maxwell connection acts was discovered by Dirac.

But it teok so long to realize fully the physical peculiar-

ities of the connection field that it was only in the 1960's

that Aharanov and Bohm proposed an experiment which truly

showed the "connection-like" behavier of the r.iaxwell field.

For this, one has to divide an electron beam into two parts,

allew these twe parts te ge areund a cylindrical regien cen-

taining a magnetic flux, and then observe the interference

pattern on a screen. According to their assumption, the

picture on the screen should change after the magnetic field

is turned on and off, even though the beams always pass

through regions where the electromagnetic field strength is

zero when going areund the domain of the magnetic flux. Thus,

the beams that are divided and rejoined on the screen "feel"

the curvature ef the connection on the Dirac bundle which

comes from turning on the field in the domain they go around.

The interference pattern en the screen in fact reflects the

difference of the phase angles in the spin space of the

 25

degrees of freedom of the electron; this difference appears

because electrons arrive at a point of the screen by different

paths in going around the magnetic flux. (The experiment was

actually carried out and corroborated these predictions.)

Some Novelties

The "gallery" of important geometric shapes through

which we hastily shepherded the reader is by no means ex-

hausted by these exhibits. The number of such concepts is

constantly increasing. Among those that have attracted the

attention of physicists and mathematicians in recent years

one can name, for example, "catastrophes", "supersymmetries"

and "solitons".

The term "catastrophe" was introduced by the French

mathematician Rene Thom to convey intuitive notions connected

with mathematical schemes to describe such phenomena as dis-

continuities, jumps, corners, surfaces separating homogeneaus

phases, biological speciation, etc. Their usefulness in

models of natural phenomena remains under question at the

present time, and has even become a subject of heated debate

in the popular press. The historian of science has the good

fortune of being able to observe an attempt to establish a

new paradigm (in the sense of Thomas Kuhn), and to reflect

upon the social aspects of the process of establishing

scientific public opinion.

"Supersymmetries", studied in supergeometry, areentering

science with less fanfare, although they will perhaps have

greater impact on the development of physics and geometry.
 26

Formally speaking, supergeometry considers not only ordinary

functions on a space, say Rn, but also anticommuting func-

tions, i.e., those satisfying the condition fg = -gf, from

which it follows, in particular, that f2 = 0. Since there

are no nonzero numbers with square zero, such a function

cannot take numerical values; its natural ranges of values

are the so-called Grassmann algebras, introduced in the last

century by the remarkable mathematician and Sanskritclogist

Grassmann. In physics, Grassmann algebras appeared only

after the emergence of quantum theory and the concept of spin;

it turned out that an adequate description of a collection of

identical particles with half-integer spin, such as electron~

requires the introduction of anticommuting variables.

The notion of a soliton arose as a result of the dis-

covery of certain special solutions of equations describing

waves in various media--water, for example. The classical

wave equations are linear, i.e., the sum of solutions and the

product of a solution by a nurober are also solutions. In

other words, these equations describe waves which do not

interact with one another. If we want to take into account

such properties of real media as dispersion (a nonlinear

relation between the frequency and length of an elementary

wave) and the nonlinear dependence of the velocity of a wave

on its amplitude, we arrive at a much more complicated

picture of the interaction of wave perturbations than simple

superposition. Therefore, considerable excitement greeted

the discovery in the 1960's by a group of American physicists

and mathematicians of the effect of a nonlinear addition of

 27

certain solitary excitations described by the Korteweg-de

Vries equation (these solitary excitations were what was

first called solitons).

The height of a soliton wave is proportional to its

velocity. Hence, one can attempt to trace the fate of the

sum of two solitons that are far apart at an initial moment,

where the larger soliton moves toward the smaller one and

therefore overtakes it. The common expectation would be that

after "collision" the wave picture collapses, but machine

experiment shows that nothing of the sort occurs: after a

period of interaction the larger soliton "passes through the

smaller one", and both begin to diverge, each retaining its

form. An exact mathematical theory was constructed shortly

after this discovery--it corroborated the preservation of the

individuality of the solitons after the interaction, no matter

how many there were at the beginning.

Since that time, the number of nonlinear wave equations

found to exhibit analogaus properties has grown linearly with

time, and the number of publications devoted to them has

grown exponentially. The hope has been expressed that

soliton-like excitations provide an adequate classical image

of elementary particles: on a new intellectual level this

revitalizes the idea proposed a century ago by Rankine and

Thomson that atoms are "vortical rings of the fundamental

fluid". The point is that, in cantrast to Maxwell's

equations, the equations for Yang-Mills connections are

nonlinear.
 28

Now a small historical note on the discoverers of the

soliton, which the reader may find edifying. (Translator's

note: See also a paper by F. van der Blij, "Some details of

the history of the Korteweg-de Vries equation", Nieuw Archief
voor Wiskunde (3), 26 (1978), 54-64 for some further inter-
esting sidelights.) Diederik Johannes Korteweg was born in
Holland in 1848 and died there in 1941. He was a well-known
scientist, and several obituaries were dedicated to his
memory. None of the obituaries even mentions the now famous
paper in which the soliton was discovered. That paper was in

essence a fragment of the dissertation of Gustav de Vries,

completed under the guidance of Korteweg and defended on
December 1, 1894. de Vries was a high school teacher, and
almost nothing is known about him.

Sets, Formu1as, and the Divided Brain

What is the relation between a mathematica1 text and its

content in the wide sense of the word (the multiplicity of
its possible interpretations)? We have attempted to show
that an intermediate set-theoretic image must be constructed
between, say, Maxwe1l's equations and their meaning in terms
of physical concepts: this construct is an interpretative
middleman, functiona1ly similar to the use of an intermediate
artificial language in modern linguistic models of machine
trans1ation. In fact, a careful analysis of scientific
thought would enab1e one to discover a whole hierarchy of
language intermediaries which take part in the potential
explanation of such concepts as "number", "photon" or "time".
 29

However, most of these explanations exist in a vague,

obscure and unsteady form, frequently accessible to individual

cognition, but yielding to communication only to the extent

that the means of our natural language can be used. Natural

language plays an immense role in the discovery, discussion,

and preservation of scientific knowledge, but it is very

poorly adapted to the exact transmission of the content of

this knowledge, and especially to its processing, which forms

an important part of scientific thought. Natural language

performs different functions, and ascribes to other values.

The language of modern set-theoretic mathematics can

affect the role of such a language-intermediary thanks to its

unique capacity to form geometric, spatial and kinematic

forms, and at the same time to provide a precise formalism

for transcribing their mathematical content. Cantor's defi-

nition of a set, which was given above, has with some irony

been called "naive" by comparing it with Euclid's definition

of a point as a "place with neither length nor breadth".

This criticism is testimony to the lack of understanding of

the fact that the fundamental concepts of mathematics, which

are not reducible in a given system to more elementary con-

cepts, must necessarily be introduced in two ways: the

concrete ("naive") way and the formal way. The purpose of

the concrete definition is to create a prototypical, not yet

completely formed shape, to tune different individual

intellects to one scale, as by a tuning fork. But the formal

definition does not really introduce a concept, but rather a

term; it does not introduce the idea of a "set" into ~he

 30

structure of the rnind, but rather introduces the word "set"

into the structure of adrnissible texts about sets. These

texts are described by rules for generating thern, in rnuch the

sarne way as instructions in ALGOL describe rules for cornposing

prograrns. In the lirnit of idealization, all of rnathernatics

can be regarded as the collection of grarnrnatically correct

potential texts in a formal language.

In this irnage there is a strange, and to rnany appealing,

esthetics of ugliness. It arose in the works of thinkers who

thought deeply about how to reconcile belief in the absolute

truth of rnathernatical principles with the abstractions of

infinite sets and infinite processes of verification used to

introduce this truth. The original conjecture of David

Hilbert was that, strictly speaking, these abstractions do

not need any "alrnost physical" and thus questionable inter-

pretation, and that they can be taken to be purely linguistic

facts. "Infinity" is not a phenornenon--it is only a word

which enables us sornehow to learn truths about finite things.

We have already rnentioned that GÖdel later showed that such a

linguistic notion of "provably true" is far narrower than the

abstraction of truth that can be developed through ideas of

infinite verifications.

The external, scientific and applied (in the broad sense

of the word) aspects of rnathernatical knowledge, when analyzed

episternologically, allow us to understand sornething about

rnathernatical creation and the dialectics of its interrelation

with Gödel's prohibition. The adoption of an interpretation

of forrnalisrn which is in sorne sense physical, the belief in

 31

the adequacy of this interpretation and the knowledge of

certain features of the behavior of a physical phenomenon,

allow us to guess or to postulate mathematical truths that

arenot accessible to "pure intuition". This is the source

for extending the very basis of mathematical knowledge.

On a more special plane, it has become possible in

recent years to consider the relation between mathematical

symbolism, informal thought and the perception of nature from

a point of view provided by new data about the structure and

functions of the central nervaus system.

The brain consists of two hemispheres, a left hemisphere

and a right one, which are cross-connected with the right and

left halves of the body. The neuron connections between the

two hemispheres pass through the corpus callosum and commis-

sures. In neurosurgical practice there is a method of treat-

ment of severe epileptic seizures, in particular, which con-

sists of severing the corpus callosum and the commissures,

thereby breaking the direct connections between the hemi-

spheres. After such an operation patients are observed to

have "two perceptions". In the laconic formulation of the

American neuropsychologist K. Pribram, the results of studies

of such patients, and also patients with various injuries of

the left and right hemispheres, can be summarized in the

following way: "In right-handed persans the left hemisphere

processes information much as does the digital computer, while

the right hemisphere functions more according to the prin-

ciples of optical, holographic processing systems." In

particular, the left hemisphere contains genetically

 32

predetermined mechanisms for understanding natural language

and, more generally, syrnbolism, logic, the Latin "ratio"; the

right hemisphere controls forms, gestalt-perception, intui-

tion. Norrnally, the functioning of human consciousness

continually displays a cornbination of these two components,

one of which may be manifested more noticeably than the

other. The discovery of their physiological basis sheds light

on the nature and typology of mathematical intellects and

even schools working on the foundations of mathematics. One

might conjecture that the two great intellects standing at

the cradle of modern mathematics, Newton and Leibniz, were,

respectively, right-hemispherical and left-hemispherical

types. To Newton we owe the conception of mathematical ana-

lysis and the first fundamental results of mathematical

physics: the universal law of gravitation, and the derivation

of Kepler's laws and the theory of tides from it. On the

other hand, Leibniz invented calculus and introduced notation,

in particular f y dx, which we still use today. In the words

of the historian of mathematics D.J. Struik, he was "one of

the most fruitful inventors of mathematical syrnbols", and

even his version of mathematical analysis was invented as a

result of a search for a universal language. (It is inter-

esting that Newton, who also did not escape this craze of his

time, nevertheless did not invent the formalism of analysis

and presented his proofs geometrically.)

If we accept modern notions concerning the functional

asyrnrnetry of the brain, we can conjecture that set theory

allows the working mathematician to achieve by the shortest

 33

path a properly balanced activity of the left and right

hemispheres of his brain; it is this which explains its re-

markable effectiveness.

In conclusion, I would like to cite the words of I.A.

Sokolyansky, who devoted his life to the education of blind

and deaf children. The quote is contained in a letter to

Vyacheslav V. Ivanov, from whose book "Even and Odd, Asym-

metry of the Brain and Symbolic Systems" (Moscow, "Soviet

Radio", 1978) the following information is taken.

If those parts of the central nervous system of a blind

and deaf child that control visual perception of the outside

world are not damaged, then the child can be taught a

language, even speech, and the complete development of its

personality can be ensured. This process is accomplished in

several steps. First, the child is supported by constant

contact with the mether or a teacher, holds onto her hand or

skirt, walks areund the house, touches and feels the objects

of her actions, and on this basis manufactures a language of

gestures which imitates to some extent the actions and

properties of objects. Normally this is a function of the

right hemisphere. Sokolyansky's discovery was that in the

next stage the child can be taught to recode the language of

gestures into a finger alphabet, so that a gesture-hieroglyph

is replaced by a gesture-word. The symbol of translation is

a special gesture similar to the mathematical equal sign--two

palms extended parallel. The meaning of this recoding is

that the information is transmitted to the left hemisphere,

which since it is predisposed to be taught discrete and

 34

symbolic language (not necessarily spoken language!), begins

to develop this function with practically the same speed as

in a normal child--mastery of the language takes two to three

years. The syntax of this left-hemisphere language is

different from the "real-world" syntax imbedded in the right

hemisphere, and is more or less identical to the syntax of

verbal natural language. The semantics, however, is appar-

ently more restricted, or in any case not quite equivalent to

the semantics of someone who sees and hears. How does one

explain what "star" means to someone who has never seen a

star? Sokolyansky gives a noteworthy answer: "Verbal speech,

although it can be mastered by the dumb, cannot by itself

guarantee the deaf-blind complete intellectual development to

the extent of reflecting the external physical world as it is

accessible to a normal person. The true picture of this

world can be uneavered only by a mathematic.a1l.y developed

intellect .•• "

Even those who see stars ask "~vhat is a star?", because

to see merely with one's eyes is still very little.

 2. PHYSICAL QUANTITIES, DH1ENSIONS AND CONSTANTS:
The Source of Nurobers in Physics

"The whole purpose of physics is to find a number, with

decirnal points, etc.! Otherwise you haven't done
anything." (R. Feynrnan)

This is an overstatement. The principal aim of physical

theories is understanding. A theory's ability to find a

nurober is merely a useful criterion for a correct under-

standing.

Nurobers in physics are most often the values of physical

quantities that describe the states of physical systems.

"Quantities" is the generic name for such abstractions as

distance, time, energy, action, probability, charge, etc. A

state of a system is characterized, in turn, by the values of

a sufficiently complete collection of physical quantities;

and it is most natural to describe a system by giving the set

of its possible states. One cannot escape from this vicious

circle if one restricts oneself to purely verbal descriptions.

The circle can be broken in two ways--operationally, when we

explain how to measure the mass of the Earth or an electron,

and mathematically, when we propese a theoretical model for a

system or a class of systems and assert that the mass m is,

say, the coefficient in Newton's formula F=ma.

35
 36

Substantive, although simple, mathematics connected with

physical quantities begins with a reminder that the values of

a physical quantity (more precisely, a real scalar quantity)

can be identified with numbers only after a choice of unit of

measurement and crigin of reference (zero point). It is reasonable

to refrain from introducing this element of arbitrariness for

as long as possible. In fact, some of the most fundamental

physical laws state that certain physical quantities have

natural units. We will examine this in more detail.

The Spectrurn of a Scalar Quantity

The spectrum of a quantity is the set of all the values

that the quantity can take (on the states of a given system,

of a given class of systems, or of "all" systems--this is

made more precise as needed) . The basic mathematical postu-

late that can be considered as the definition of a scalar

quantity in theoretical models is that the spectrum is always

a subset of one-dimensional affine space over the real nurn-

bers. In other words, it lies on a line where zero and one

are not marked; if two points are specified to be zero and

one, then the spectrum becomes the set of real numbers. The

gist is that sometimes these points need not be distinguished

arbitrarily, but rather one can use the spectrum itself.

Here are some basic examples.

(a) Velocity. The smallest (relative) velocity is

naturally called "zero". The second distinguished point on

the spectrum of velocities is c, the speed of light. The

conventional assumption about the spectrum of velocities

 37

(since the creation of the special theory of relativity) is

that it fills out the interval from zero to c. Then it is

natural to declare c to be the unit of velocity and to take

all velocities in the segment [0,1); in the more usual

notation, this is true of the ratio vjc. In everyday life

we rarely encounter velocities greater than 10- 6 on this

scale (the speed of sound).

(b) Action. This is perhaps the most important quantity

of all in theoretical physics, and we shall devote a separate

chapter to it. It takes values on intervals of the history,

or paths, of a physical system, not on instantaneous states.

In classical physics it determines the dynamics, i.e., the

physically possible path intervals on which "action" takes

the smallest admissible values. The natural zero on the

spectrum of actions is the action of the "infinitesimal"

history of the system. The upper boundary of the action

spectrum is unknown. It may exist in a cosmological model,

where this boundary will be the action of the Universe in its

entire history from the Big Bang to the Big Collapse, if the

latter is predicted by the model.

Nevertheless, another distinguished point is known on

the action spectrum: this is the farnaus Planck constant h.

In human terms it is extremely small--the action of a pen

writing the word "action" is of the order of 10 29 to 10 30 h.

Pragmatically speaking, h indicates when quantum-mechanical

models must be used rather than classical ones: namely, in

cases when we are interested in details of the history of the

system where the action only varies over a few h. (However,

 38

this condition is neither necessary nor sufficient.)

Choosing h as a unit of action, we may take the action

spectrum tobe the half-line [O,oo), and the spectrum of in-

crements of action to be the whole real line.

Thus, the point h is "unseen" on the action spectrum,

in contrast, say, to the speed of light c, which is the

right endpoint of the velocity spectrum. This is very

strange. However, there are two contexts in which h appears

explicitly.

One of these is connected with the spin spectrum--the

internal angular momentum of elementary particles. Spin has

the same dimension as action, and consists of integer mul-

tiples of n/2 = h/4n. Does this not mean that spin is truly

the fundamental quantity, and action only a vestige of clas-

sical physics?

The secend context is the farnaus Reisenberg uncertainty

principle. Quantum models define a partition of the classical

system of quantities into conjugate pairs: (coordinate of

position, projection of momentum), (energy, time). A product

of conjugate quantities has the dimension of action. In a

verbal formulation the uncertainty principle asserts that

both members of a pair of conjugate quantities cannot simul-

taneously take an exact value on any state of the system.

The product of the imprecisions is bounded from below by n/2.

Applying this principle to energy and time, we formally

obtain the relation 6E • t>t > Il./2, whose Substantive meaning

has been discussed many times in the physical literature.

From our point of view this means that the notion of a
 39

classical segment of the history of the system on which the

action varies by less than n/2 is devoid of meaning. Later

on, we discuss in more detail the difficult question of the

comparative meaning of classical and quantum quantities

having the same name.

(c) Mass. According to Newton, values of inert mass

can be ascribed to stable material bodies. The smallest

objects to which the Newtonian concept of mass is still

applicable without fundamental reservations are the electron

and the proton. They give us two points on the spectrum of

masses (other than zero): and A typical mass on

the human scale can be given using Avogadro's number

23
6.02 x 10 mp. The ratio mp/me ""1840 is the first truly

fundamental number that we have encountered so far, in con-

trast to the points of the spectrum, which are not, strictly

speaking, numbers.

Other elementary particles determine other points on the

spectrum of masses; measuring them in units of or

we obtain a bunch of numbers for which theoretical explana-

tions are needed.

(d) Gravitational constant If two point masses m1 and

are found at a distance r from each other and are

attracted to each other with a force F resulting only from

their Newtonian gravitational attraction, then the quantity

Fr 2 ;m1 m2 does not depend on m1 , m2 , or r. This quantity

was discovered by Newton and is denoted G.

 40

This last example is conceptually more complicated than

the preceding ones: to introduce G we must explicitly

appeal to a "physical law". Moreover 1 we have obtained a

point of a new spectrum--the spectrum of constants of funda-

mental interactions 1 which also include electromagnetic 1

strong and weak interactions.

It is appropriate at this point to introduce the follow-

ing large group of physical abstractions.

Physical Law 1 Dimension and Similarity

For the purposes of this section "physical law" means

2
the content of such formulas as F = ma1 F =Gm 1m 2 /r (Newton) 1

E = hv (Planck) 1 E = mc 2 ( E1nste1n
. . ) 1 etc. A physical theory 1

such as Newtonian mechanics or Maxwell's electromagnetic

theory 1 includes the following data on the mathematical side:

(a) the basic quantities of the theory 1 and (b) the basic
laws connecting them. Moreover 1 on the operational side a

theory must describe: (c) the physical Situations in which

the theory can be appliedl and (d) principles for comparing

the theoretical Statements with measurements and Observations.

Here we are concerned only with the first part. Knowing

the quantities and the laws connecting them 1 we can construct

a fundamental mathematical characteristic of a theory--its

group D of dimensionst In mathematical language this is an

abelian group which can be defined by generators and

relations: the generators are the physical quantities of the

theoryl and the relations are determined by the condition

 41

that all the laws of the theory are homogeneous. The class

of a quantity in the group D is called the dimension of

this quantity. It is possible to choose basic quantities

which form an independent system of generators in the group

D; the dimensions of the remaining quantities of the theory

can be expressed in the form of formal monamials in the

basic ones. The units of the basic quantities determine the

units of the others. (We are giving a condensed presenta-

2
tion of the principles underlying such notation as ern/sec . )

We note several circumstances in which the explicit intro-

duction of the group D helps to elucidate the essence of

the theory.

The Group of Dimensions of Newtonian Mechanics

It is generated by the dimensions of length L, time T

and mass M. The law F = ma shows that force has dimension

MLT- 2 in this group, while energy (force times length) has

dimension ML 2 T- 2 , and action (energy times time) has di-

mension HL 2 T- 1 .

Progress in physics is continually accompanied by two

opposing processes: enZargement of the group D of dimen-

sions through the discovery of new kinds of quantities

(electromagnetism after Newton; new quantum numbers such as

"strangeness" and "charm" in our days), and the shrinking of

the group through the discovery of new laws giving relations

between previously independent dimensions.

 42

In order to understand this secend process we return to

the Newtonian gravitational constant G. According to our

rules, its dimension is force x (length) 2 x (mass) - 2 = M-lL 3T- 2 .

Thus, its numerical value depends on choosing units of mass,

length and time. It is a constant in the sense that, once we

choose these units, the numerical value we obtain using the

formula where vary in different

Eötvös type experiments or are calculated from astronomical

data, does not depend on the variable quantities in these

experiments: r, m1 , m2 .

After establishing this physical fact, we can use it to

construct a smaller group D' of dimensions for the theory

"Newtonian mechanics + Newtonian gravitation". Mathemati-

cally, this smaller group is the quotient group of D by the

-1 3 -2
subgroup consisting of all powers of M L T . For the
basic dimensions in D' one can take any pair (M,L), (~1,T)

or (L,T); the remaining dimensions can then be expressed in

terms of this pair and the dimension G. The number of

fundamental units for D' then decreases to two if we take

G as a unit of measurement of dimension M- 1 L 3T- 2 • From

this example we also see the physical meaning of the distin-

guished points of spectra: they are points that can be re-

produced in a series of experiments of a certain type which

isolate certain interactions, certain types of systems, and

so on.

Scale Invariance

The group D* of similarities, or scale invariances, of

a given theory can be defined from the mathematical point of

 43

view as consisting of the char>acters of the group D of dimen-

sions, i.e., the mappings x from the group D to the posi-

tive real numbers with the multiplicativity property:

X (d 1 d 2 ) =X (d 1 ) X (d 2 ) for all d 1 , d 2 E D. This group has a

direct physical meaning: it shows the proportion by which

one can increase (or decrease) the various characteristics of

a phenomenon before it leaves the range of applicability of

the theory. If we know all the laws of the theory, then D*

is trivial to calculate. The use of D* is that sometimes

it can be guessed from physical considerations before the

exact form of these laws becomes known. Then it turns out

that D* carries important information about these laws. A

well-known example of a classic discovery made in this way is

Wien' s law E: (v, T) = v 3 F (v/T) for the radiative capacitance of

an absolutely black body as a function of frequency and tem-

perature. It corresponds to the character Xa ( [v]) = a,

Xa((s]) =a 3 , Xa([T]) =a in the group D*, where [s], [v],

[Tl are the dimensions and a is an arbitrary real number.

One can also mention Galileo's arguments on the sizes of

living creatures and the numerous applications of the theory

of similarity in hydro- and aerodynamical calculations. On

the level of fundamental theories the group D* is the

simplest example of a symmetry group. In elementary particle

physics and quantum field theory such groups are more and more

frequently considered as independent physical laws on a higher

level, imposing rigid restrictions on laws on the next level,

such as Lagrangians. The most important of these groups are

noncommutative and complex, such as groups of unitary

 44

rotations in U(n), because in quantum mechanics the basic

quantities lie in higher-dimensional complex spaces, and not

in one-dimensional real ones. But that is already another

story.

Planck Units and the Problem of a Unified Physical Theory

In Newtonian physics there are no other natural units

besides G. The speed of light c can be declared a natural

unit only within a new theory which postulates its particular

role as the upper limit of the speed of propagation of mate-

rial bodies (unattainable) or signals (attainable), as an

invariant relative to a change of inertial system of coordi-

nates, etc. In a similar way, Planck's unit of action h

became the coat of arms of a new physical theory--quantum

mechanics.

In the Newtonian group D, both c and h, like G,

have completely determined dimensions: LT-l for c and

ML 2T-l for n. Having chosen G, c and h as the basic

units for the corresponding dimensions, we discover that

there is a natural scale which makes real numbers out of the

values of all classical physical quantitites expressible in

D; i.e., there arenatural units for all quantities! In fact,

the dimensions M- 1 L 3T- 2 (of G), LT-l (of c) and ML 2 T-l

of h) generate the whole group D (to be really precise,

they generate a subgroup of index two).

In particular, the natural units of length, time and

mass--the celebrated units of Planck--are:

 45

L* (fl.G/c3) 1/2 1. 616 X 10- 33 cm;

T* (nG/c5) 1/2 5. 391 X 10- 44 sec;

M* (nc/G) 11 2 2.177xl0- 5 g

The reader might ask: Why don't we ever measure in

Planck units? The pragmatic answer: because they give a

completely absurd scale. The Bohr radius is equal to

3. 9 x 10-ll cm: L* is 22 orders of magnitude less! T* is

the same nurober of orders of magnitude less than the time it

takes light to pass through a Bohr radius. On the other

hand, M* is the mass of a real macroscopic particle con-

taining about 10 19 protons. The Planck unit of density

M*/L* 3 is equal to 5 x 10 93 g/cm 3 ; we cannot imagine any-

thing having density remotely similar to this under any

circumstances.

A more substantive remark is that, in fact, we have no

unified physical theory in which c;, c and fJ. occur simul-

taneously. Even the theories which combine these constants

in pairs are considered the greatest achievements of the

twentieth century: (G,c) is general relativity theory;

(c,h) is relativistic quantum field theory, which is in re-

latively complete form only for electromagnetic interactions.

The estimates of the quantum generation of particles in

strong classical gravitational fields, in particular, close

to black holes of small mass (S. Hawking), are related to the

fragments of a future (G,c,fl.)-theory. As long as a complete

quantum (G,c,h)-theory does not exist, Planck units remain

the remote outposts of an extensive uncharted territory.

 46

There is another point of view regarding this strange

disparity of orders of magnitude of natural units with one

another and with more cornrnon units. In the hypothetical

fundamental equations of a unified theory--"the universal

theory of everything" (Stanisraw Lem)--the different terms

(now dimensionless numbers!) will, thanks to this disparity,

take values that differ very sharply in magnitude depending

on the scales of the domain of space/time, momentum/energy in

which the phenomena to be studied are located. Very small

terms can be discarded with negligible error, leading to one

of the approximate theories. This also happens at the limits

of applicability of the currently known models, when we

describe the world classically in human scales (supposing

v/c =0 and 11/S = 0, where v is a typical velocity and S

is a typical action), or when we neglect gravitation in micro-

scales, setting G=O.

In fact, this argument, though it contains a grain of

truth, is very naive. A real change of theory is not a

change of equations--it is a change of mathematical structure,

and only fragments of competing theories, often not very

important ones conceptually, admit comparison with e.ach other

within a limited range of phenomena. The "gravitational

potential" of Newton and the "curvature of the Einstein

metric" describe different worlds in different languages.

Moreover, life--perhaps the most interesting physical

phenomenon--is embroidered on a delicate quilt made up of an

interplay of instabilities, where a few quanta of action can

 47

have great informational value, and neglect of the small

terms in equations means death.

Classification of Physical Constants

We sum up what has been said. The reference book CRC

Handbook of Chemistry and Physies contains 2431 pages of text and

many millians of nurnbers. How can one understand them?

These quantities can be divided into at least four types.

(a) Natural units of measurement , or physically distinguished

points of spectra. These are not nurnbers but entities such as

G, c, h, me, e (charge of an electron). Theseare dimen-

sional characteristics of certain phenomena which are capable

of being reproduced again and again with a high degree of

accuracy. This is a reflection of the fact that nature pro-

duces elementary situations in large quantities. Contem-

plating the identical nature of each type of building block

of the universe has from time to time led to such profound

physical ideas as Bose-Einstein statistics and Fermi-Dirac

statistics. Wheeler's fantastic idea that all electrons are

identical because they are the instantaneous cross-sections

of the tangled-up world-line of a single electron, led Feynman

to an elegant simplification of the diagrarnrnatic computation

technique in quantum field theory.

(b) True, or dimensionless, constants. These are ratios of

various distinguished points on the spectrum of a scalar

quantity, for example, the ratios of masses of electrical

particles (we have already mentioned mp/me). The identifi-

cation of different dimensions because of a new law, i.e.,
 48

the reduction of the group of dirnensions, has the effect of

cornbining once-distinct spectra and giving birth to new nurn-

bers which rnust be explained.

For exarnple, the dirnensions rne' c and ft generate the

Newtonian group and therefore lead to atornic units for the

dirnensions M, L and T which are as natural as the Planck

units. Thus, their ratios to the Planck units require a

theoretical explanation. But, as we have said, this is im-

possible as lang as there is no (G,c,ft)-theory. However,

even in the (rne,c,ft)-theory--quanturn electrodynarnics--there

is a dirnensionless constant to which modern quanturn electro-

dynarnics in some sense owes its existence. We place two

electrons at a distance ft/rnec (the so-called Campton wave-

length of an electron) and rneasure the ratio of the energy of

their electrostatic repulsion to the energy rnec 2 , which is

equivalent to the rest rnass of an electron. We obtain the

nurober a = 7. 2972 x 10- 3 Rtl/137. This is the farnaus fine

structure constant.

Quantum electrodynarnics describes, in particular, pro-

cesses which do not preserve the nurober of particles: a

vacuurn generates an electron-positron pair, which are annihi-

lated • Because the energy of generation (not less than

2rnec 2 ) is 100 tirnes greater than the energy of the charac-

teristic Coulombinteraction (thanks to the value of a), one

can carry out an effective scherne of calculations in which

these radiation corrections are not cornpletely neglected, but

do not hopelessly "spoil the life" of the theoretician.

 49

There exists no theoretical explanation for the value of a.

Mathematicians have their own remarkable spectra: the

spectra of various linear Operators (for example, the genera-

tors of simple Lie groups in irreducible representations),

the volumes of fundamental domains, the dimensions of homo-

logy and cohomology spaces, etc. There is wide scope for

fantasy in identifying the spectr:a. of. mathematician,s and tlle

spectra of physicists. What we need instead is principles to

limit the choices. Let us return, however, to constants.

The next type occupies a large part of tables of

physical quantities:

(c) Coefficients for converting from one scale to another, for

example, from atomic scales to "human" scales. Among these

constants are: the Avogadro nurober mentioned before,

23
No= 6. 02 x 10 , essentially one gram expressed in "proton

masses", although the traditional definition is a bit differ-

ent; and also such things as a light year expressed in kilo-

meters. For mathematicians the most disagreeable aspect of

this is, of course, that the coefficients convert from one

set of physically meaningless units to another which is

equally meaningless: from cubits to feet or from degrees

Reaumur to degrees Fahrenheit. In human terms these are some-

times very essential numbers; as Winnie-the-Pooh remarked

philosophically: "But whatever his weight in pounds, shil-

lings and ounces, He always seems bigger because of his

bounces."

(d) "Diffuse spectra " These characterize materials (not

elements or pure compounds, but ordinary industrial brands of
 50

steel, aluminurn, copperl, astronomical data (the mass of the

Sun, the diameter of the Milky Way ••• ) and many other such

things. In centrast to the situation with electrons, nature

produces stones, planets, stars and galaxies with no concern

for creating exact similarity; but, nevertheless, their

characteristics vary only within rather definite limits. The

theoretical explanations for these "permitted zones", when

they are known, tend to be quite interesting and instructive.

V. Weisskopf gathered together a series of such explana-

tions in his excellent article "Contemporary physics in an

elementary presentation" (CERN Report No. 70-8). Here is an

example of a physical argument from this paper, in which all

of our principal heroes play their parts: "heights of the

mountains in terms of fundamental constants". Take the

following question: the Earth's highest mountain Chomolungma

(Everest) has a height of about 10 km; why are there no higher

mountains? It turns out that, even without taking into con-

sideration the geological mechanisms of erosion and weather-

ing, we find that the height of a mountain is limited by a

few tens of kilometers because of the concrete dimensions of

the Earth and the values of the fundamental constants.

Weisskopf argues as follows: a mountain of too great a

height cannot exist because of the liquefaction of its lower

part under the pressure of the upper part. A calculation of

the height at which the pressure is almest sufficient for

liquefaction yields the estimate:

1 1
y R> 40 km
Nl/3 AS/3
 51

where y = 0. 02 is the mel ting heat character istic (which can

be estimated entirely in terms of the fundamental constants);

2
a is the fine structure constant, a G =Gm p /nc·I

is the nurober of protons and neutrons in the Earth; A ~ 60

is the average atomic weight of the matter in a mountain.

Only the nurober N here is not fundamental. But even its

place on the diffuse spectrum of the planet masses is bounded

by the fundamental constants. Using very crude estimates,

Weisskopf shows that N could not exceed about 10 53 , since

otherwise the matter in the planet could not exist as un-

ionized atoms. Finally, one can estimate N from below by

requiring that the height of mountains on a planet be no

greater than the radius of the planet, i.e., that the planet

be approximately spherical, since otherwise mountains cannot

be distinguished as such! This estimate leads to the size of

large asteroids.
 3. A DROP OF MILK:
Observer, Observation, Observable and Unobservable

" .•. What would be observed (if not with one's actual
eyes at least with those of the mind) if an eagle,
carried by the force of the wind, were to drop a rock
from its talons?" (G. Galilei)

The actual eyes of my contemporaries observed how a bomb

flies when the locks of a bornher open against a smoky sky,

on the newsreel screens and in thousands of children's draw-

ings; I myself drew them. Let us try to forget this and look

at the world with the eyes of the mind, as our immortal con-

temporary Galileo Galilei taught the simple-minded Simplicio.

Isolated Systems

Among all the abstractions of classical physics one of

the main ones is the idea of an isolated, or closed, system.

This is a part of the Universe whose evolution over some

period of its existence is determined only by intrinsic laws.

The external world either does not interact with the system

at all, or in some models this interaction is incorporated

summarily, with the help of such notions as constraints, an

external field, a thermostat, etc. (thus we have used the

words "isolated" and "closed" more broadly than is usually

done; the isolation refers, rather, to the mathematical

model). There is no feedback loop, or if there is, it is

52
 53

artificially severed. The world is divided into parts,

modules and assemblies, as in industrial specifications. And

in fact this is the ideology not only of Homo Sapiens but of

Homo Faber. The screws and gears of the great machine of the

world, when their behavior is understood, can be assembled

and joined in a new order. Thus one obtains a bow, a 1oom or

an integrated circuit.

For the mathematician an isolated system consists of:

(a) its phase space, i.e., the set of possible instantaneous

states of motion of the system; (b) the set of curves in

phase space describing all possible histories of the system,

i.e., sequences of states through which the system passes in

the course of time. The first of these data is kinematics,

the secend is dynamics. It is important to distinguish a

state of the system from a state of its motion: the first is

traditionally given by Coordinates, and the secend by coordi-

nates and velocities. Knowing only the coordinates, we cannot

predict the later motion of the system, but we can if we know

the coordinates and the velocities. The assumption that a

closed system can be described by a convenient phase space

and a system of curves in it (the whole thing tagether is

often called a phase portrait of the system) is the mathe-

matical content of the classical principle of determinism.

One of the famous paradoxes of Zeno of Elea can be inter-

preted as a first approximation to understanding the role of

the phase space: a flying arrow and a stationary arrow at

each moment of time are located at the place where they are

located; what then distinguishes flight from irnrnobility?

 54

Answer: the apparent place of the arrow is only the pro-

jection onto the space of positions of its "true place" in

the space of pairs (position, velocity vector).

A classical closed system is isolated from the whole

external world; thus also from the external observer, and

from the effect which an observer can have on it. Observa-

tion is not interaction. Observation is the most important

thought-experiment which can be performed on the system, and

its first goal is to localize the system in its phase space.

The converse can also be said: the phase space is the set of

all possible results of complete instantaneous Observations.

A complete observation allows one to compute the full evolu-

tion of a classical system; the existence of complete obser-

vationsis another form of the determinism postulate. Evolu-

tion is the set of results of observations at all moments of

time. The possibility of a mental observation with no inter-

action is confirmed by considering different methods of

observations which approximate reality and for which the

interaction is part of the design, but can be made as small

as desired or completely accounted for in computations, i.e.,

it is controllable. In essence, one considers an isolated system S

tobe included as part of a larger system (S,T). An obser-

vation corresponds to an act of weak interaction between S

and T which scarcely interrupts the evolution of S (per-

haps the interaction is switched on for a short time and then

cut off). It is of principal importance that the abstraction

of conceptual observation can then be applied to the combi-

nation (S,T) without influencing the evolution of the

 55

cornbined systern. Moreover, we suppose that s can becorne

part of (S,T) for a short time without losing its individu-

ality, i.e., it can do so reversibly.

This is a very natural postulate for man, whose princi-

pal rneans of observation is vision. Electrornagnetic inter-

actions are so weak that in all scales frorn the cosrnic to the

human a glance at the systern does not act on it at all.

The "eyes of the rnind" rnust be able to see in the phase

space of rnechanics, in the space of elernentary events of

probability theory, in the curved four-dirnensional space-tirne

of general relativity, in the cornplex infinite dimensional

projective space of quanturn theory. To camprehend what is

visible to the "actual eyes", we rnust understand that it is

only the projection of an infinite dimensional world on the

retina. The irnage of Plato's cave seerns to rne the best

rnetaphor for the structure of modern scientific knowledge:

we actually see only the shadows (a shadow is the best

rnetaphor for a projection).

For a human it is psychologically very difficult to

transcend the lirnits of the usual three spatial dirnensions.

But we only rnislead ourselves if we awkwardly try to describe

the internal quanturn degrees of freedorn as the "value of the

projection of spin on the z-axis", since a spin vector lies

in a cornpletely different space frorn the z-axis. It is worth

rernernbering that even the three-dirnensionality of the world

was recognized after great effort--it was taught to us by the

artists of the Renaissance. Uccello withdrew frorn his affairs

for ten years in order to devote hirnself to the study of
 56

perspective. Among other things, modern mathematics is

rigorous training in multidimensional perspective, following

a unified program. If we believe the neuropsychologists, in

the course of this training the left and right halves of the

brain behave as the blind man and the legless guide he carries

on his back.

Classically, an observer is represented, in general, by

a system of coordinates in the fundamental spaces of the

theory. A unit of measurement determines a coordinate in the

spectrum of the quantity being measured. When these units

are chosen, the coordinate functions, i.e., the observables,

identify the space of positions, the phase space or parts of

them with subsets of the Rn and Cn of mathematicians. A

theory with observables is good in that it simultaneously

describes both the concepts and their observable 11 Shadows 11 •

A theory with observables is bad in that it may turn out to

be simpler, more instructive and more Straightforward to

explicitly separate as soon as possible the observed from the

observer and investigate their interrelation as a new object

of study.

According to Newton and Euler, color is the spectral

composition of light radiation in the band of wavelengths

araund half a micron; according to Goethe, color is what we

see. It is striking to what extent these two notions do not

lend themselves to a direct comparison--they are connected

only by a complicated physiological theory of color vision.

The humanist Goethe could not allow a denial of the observer,

since his entire system of values could not exist without

 57

the idea of human participation as a measure of things.

Much can be said for this point of view. Much can also

be said against it: often the best method to find oneself is

to turn away from oneself. The Newtonian theory of light--

and all later physical theories--were called upon to explain

what light is without any relation to the fact that it can be

seen. For Goethe the principal thing about light is that it

can be seen. And again it turns out, as always, that the

visible must be explained in terms of the invisible.

"All motions that a celestial body is observed to under-

go belong not to itself, but to the Earth" (Copernicus, 1515).

After this phrase, which moved the Earth, all subsequent

theories based only on the notion of "observable" became

archaic before they were born.

The classical observer lives in a world of human scales,

and the conception of a classical observer undergoes natural

changes when one passes to cosmological or microscopic scales.

Distance, time, energy and motion are so great for astronom-

ical phenomena that the hypothesis that the observer does not

influence events seems acceptable without further discussion.

Other problems of observation appear in the foreground. Two

of them can be summarized briefly as questions. Can the

Universe be considered as a closed system? What should be

our attitude toward a theory describing phenomena that cannot

be observed because of their destructive impact on the ob-

server or because some regions of space-time are fundamentally

isolated from observation (stellar temperatures, masses,

 58

pressures, gravitational fields of black holes, conditions of

the BigBang)?

The very principles of describing closed systems are

based on the hypothesis that such systems can be reproduced--

the phase space of a system realizes the idea of the feasi-

bility of different states and different paths of evolution.

How does one reconcile this idea with the uniqueness of the

evolution conveyed to us in Observations of the system? The

answer, of course, lies in the notion of the local inter-

action of "parts of the world" with one another, and the

essential sameness of the laws of physics that act in differ-

ent parts. In the simplest and most fundamental models of

the Universe (the Friedmann model, the Einstein-de Sitter

model), the idea of homogeneity occurs, manifesting itself in

the existence of a large group of symmetries of the mathe-

matical model.

In all models of cosmology one finds in the foreground

the idea of modeling a closed system. In describing an iso-

lated system, we not only imagine that it does not interact

with the external world; we also oversimplify its internal

structure, ignoring many "insignificant details". Now when

the World is this "isolated system", then everything even

remotely associated with our daily life shrinks into the

ranks of "insignificant details". In a cosmological model of

the World not the slightest trace of the familiar remains.

But in spite of this, or perhaps thanks to it, a paper

in "Uspekhy Fizicheskikh Nauk" may begin with the phrase:

"We would be happy if Cygnus X-1 turns out to be a black
 59

hole" (Usp. Fiz. Nauk 126 (1978), no. 3, p. 515). We know

something about the world because we are happy to perceive it.

Principles of Quantum Description

An ideal observer of the macro"Y7orld cannot change it,

but even an ideal observer of the microworld is bound to

change it. This is explained in numerous treatises on quantum

mechanics, but it seems to be poorly understood. Quantum

mechanics does not simply teach us new mathematical models of

phenomena; it furnishes a model of a new relation between

description and phenomenon. In particular, several charac-

teristics of these models are explained verbally by drawing

on the idea of "unobservability". The meaning of this word

changes its inflections like Proteus; the following are unob-

servable: the phase of the psi-function, a virtual photon,

the color of a quark, the difference between identical parti-

cles, and much more.

Let us try to look at the geometry of quantum mechanics

with the "eyes of our mind".

Phase Space

The phase space of a closed quantum system is a set of

rays (one-dimensional subspaces) in a complex vector space ~

in which a scalar product is also given. In this postulate

we express: (a) the principle of linear superposition;

(b) the principle of "unobservability of the phase". Instead

of using a whole line in ~ to describe the state of the

system, we usually consider a vector lying in this line. It
 60

is defined only up to multiplication by a complex number.

Even if we normalize the vector by requiring that its length

be equal to one, there still remains an arbitrariness in the

i8
choice of a factor e This 8 is the "unobservable phase".

Phase Curves

In order to describe them we must explain how each ray

in ~ varies with time t, as it traces the evolution of

the isolated system. The standard description is as follows.

(a) In ~ there are N mutually orthogonal rays which do

not vary at all: they correspond to the stationary states of

the system. (Here N is the dimension of ~; as in Chapter

1, for simplicity we only consider the finite dimensional

case.) (b) To each of the stationary states ~j there

corresponds a quantity E. with the dimension of energy,

J
which is called the energy level of the corresponding station-

ary state. If at time zero the system is in the state

La.~., then at time t it will be in the state ~(t)

J J E · t/i1'i.
= La.~. e J We note that E. t has the dimension of
J J J
action and is naturally measured in Planck units n. Since

eEt/in = cos (Et/11) - i sin (Et/n), each term here is periodic

in time, and essentially describes motion around a circle

with an angular velocity. Thus, their sum represents a

rotation around the N axes with different velocities. The

trajectories of two-dimensional motions of this sort are the

well-known Lissajous figures. (Another image from the ancient

history of science is the epicycles of Ptolemy, which also

lead to a sum of circular motions.) Any coordinate of the

 61

vector w(t) experiences frequent and extremely irregular

oscillations with time; the graph of even the simple func-

10
tion L cos(n 2 t) looks like a seismogram. It is convenient
n=l
in the form e -iS(t)
to write w(t) w(O), where S(t) is the

linear operator "action during the time interval [O,t]".

Observation: Ovens, Filters and Quantum Jurnps

The classical idealization of an observer capable of

fixing the instantaneous position of a system on its phase

curve is replaced by a radically new system of concepts. At

first we shall not call them by their cornrnonly accepted names,

in order not to create illusions. A highly idealized assurnp-

tion about the connection of the above-described scheme with

reality is that for each state wE .Ye one can make a physical

device (an "oven") Aw which produces the system at the

state w. Moreover, for each state X E.Yt' one can make a

device (a "filter") Bx which systems enter in the state w.

and which the system when it leaves can either be detected at

the state x or not detected at all ("the system does not

pass through the filter").

The third basic postulate of quantum mechanics (after

the superposition principle and the law of evolution) is the

following: a system which is prepared in the state w and immedi-

ately after this passes through the filter BX will arrive at the state

x with probability equaZ to the square of the cosine of the angle

between the rays w and x in .Ye. If a time t elapses between

the preparation of the system at w and its admission through

the filter BX, then the probability will be equal to the

 62

square of the cosine of the angle between e-iS(t)~ and X·

This can be interpreted as follows. As long as nothing

is done to the system, it moves along its phase curve. But

as soon as it hits a filter which produces only systems at

the state x, its state vector changes by a jump--either it

rotates by the angle between ~ and x, and the system passes

through the filter, or else the filter blocks it. A system

which has passed through the filter BX has no memory of the

state it was in before it hit the filter--x could have been

obtained from any state at all (except for those strictly

orthogonal to xl.

If ~ and x have unit length, then the "probability

of transition" from ~ to x is denoted l<x1~>1 2 , and the

scalar product <xl~> is called the transition amplitude.

Since the phases of X and ~ are not determined, neither

is the argument of the complex number <xl~>; only the

difference between the arguments of, say, <x 1 1~> and

<x 2 1~>, has a unique meaning. The square of the modulus of

the sum of two complex numbers depends not only on the num-

bers themselves, but also on the angle between them, i.e.,

the difference of their arguments. This results in "inter-

ference of amplitudes".

The interaction of a system ~ with a filter BX is a

special case of what in quantum mechanics is called an obser-

vation or a measurement. A more general scheme is obtained

if we suppose that the system ~ encounters a set of filters

B , ... ,B , where x 1 , ... ,xN are a complete set of

Xl XN
 63

orthogonal basis vectors; these filters should be irnagined

"in parallel" (in the sense of electric circuitry), so that

the systern can pass through any one of thern and end up in a

state With such a set of filters one associates a re-

presentation of sorne physical quantity B, which takes the

values b 1 , •.. ,bN in the states x 1 , ..• ,xN' respectively; we

say that the act of rneasurernent or Observation gives the

value b. for B on the state ~ if ~ has passed through

J
the filter B Mathernatically, the systern of filters (B )
xj Xj
or the quantity B is represented by a linear operator

.X + .X which takes the vector to the vector b.x. for

J J
all j. All such linear operators, which realize a dilatation

of .X with respect to the N rnutually orthogonal directions

(with real coefficients), are called observables.

As an illustration we give an idealized description of

the Stern-Gerlach experirnent on quanturn rneasurernent of the

angular rnornenturn (spin) of silver ions. The Hilbert space .X

corresponding to the spin degrees of freedorn of this systern

is two-dirnensional. Silver is vaporized in an electric stove;

the ions are collirnated (lined up) by a srnall aperture in a

screen, and the resulting bearn is sent between the poles of a

rnagnet which creates an inhomogeneaus rnagnetic field. The

ions in the initial bearn are found in all possible spin

states, but upon passing through the rnagnetic field they have

a tendency to "collapse" into one of two stationary states

and x_; these are traditionally called the states with

spin 1/2 and spin -1/2. When the bearn leaves the region of

the field, these states turn out to be spacially separated

 64

because of the inhomogeneity of the field. The beam is

divided in half; thus, the magnetic field acts as a set of

filters.

We see that a quantum "observation", in its very essence,

has nothinq in common with a classical observation, because:

{a) the act of "observing" almest inevitably knocks the

system off its phase trajectory; (bl at best, the act of

"observing" allows one to register the new position of the

system on the phase curve, but not its position up to the

moment of observation, the memory of which is lost; (c) the

new position of the system is only statistically determined

by the old one; finally, (d) among the quantum "observables"

there are physical quantities (which in reality play a funda-

mental role) that do not correspond to any classical ob-

servables.

A comparison between the values of the quantum and

classical observables can only be very indirect. For example,

to a quantum Observable B one can associate its mean value

on the state (in the sense of statistical averaging).

I t turns out to be equal to q IB ll/J> (if ll/J I = 1) • (The

reader can take this symbol simply as a new notation.) The

mean value
A
llBl/J =
I [ (B- Bl/J) Jl/J
A2 A
then measures the variance of

the values of B relative to the mean value on the

state lj!.

Let B and C be two observable quantities, and let

[B,C) = f (BC- CB). One can show that the "Pythagorean

theorem" in .X gives the inequality

 65

which is the mathematical expression of the Reisenberg un-

certainty principle. It is most often applied to pairs of

observables B, C for which [B,C] reduces to multiplication

by n. Then the inequality takes a more familiar form:

LlB·LlC;;,n/2, independently of lj!.

We note that if ~ is finite dimensional, then there

are no such pairs of observables. The uncertainty principle

is usually applied to quantum analogs of pairs of classical

observables (a coordinate and the projection of momentum on

the corresponding axis, energy and time).

Combination of Quantum Systems

When we spoke about classical Observations, we noted

that attempting a detailed description entails the inclusion

of the Observablesystem S into a larger system (S,T).

Therefore, we should expect that the unusual properties of

quantum Observations can be understood better once we examine

the principles of a quantum description of a combined system.

The postulate of quantum mechanics that relates to this

asserts that the state space ~

S,T of the combined system

is a certain subspace of the tensor product

and T are infinite dimensional, then this product must be

completed; we omit these details). What subspace of

must be taken is resolved on the basis of later postulates.

For the time being, let us consider the case

 66

Already the very formulation of the mathematical model ex-

hibits the possibility of completely nonclassical connections

between the "parts" S and T of the combined system. In

fact, it turns out that for the overwhelming majority of the

states of (S,T) it is impossible to indicate the state in

which S and T are found "separately", so that the notion

of "parts" turns out to have very limited meaning. Actually,

does contain decomposable states

When the system (S,T) is in one of these

decomposable states, we have reason to say that it consists

of S in the state Ws and T in the state wT. But even

for the state Ws 0 wT +Ws 0 wT such an assertion is untenable.

Meanwhile, the superposition principle allows us to construct

a large nurober of such states ~ wii) 0w~i). The set of de-

1.

composable state vectors has dimension m + n, and the set of

all state vectors has dimension mn, where m and n are

the dimensions of J't'8 and J't'T, respectively; thus, almost

all the states of (S,T) are indecomposable. In the over-

whelming majority of states of (S,T) the subsystems s and

T exist only "virtually".

This type of nonclassical connection between the parts

of a combined system frequently cannot be explained on the

basis of classical notions that the connection between parts

of a system is realized by an exchange of energy between them.

In fact, in the fundamental case when two identical systems

S and T are combined, the phase space of the combined

system does not, in general, contain even one decomposable

state. Let S be a fermionic elementary particle, such as

 67

an electron, and let T be another particle of the same

type. Then ~S,T is the proper subspace of

consisting of the vectors that change sign when S

and T are interchanged--these are linear combinations of

the vectors w1 0w 2 -w 2 0w 1 • I t is then easy to see that

each state of the system of two electrons is indecomposable.

In particular, there are no vectors w0 ljJ in the phase space;

in a popular presentation one says that two electrons cannot

be found in the same state. This is the reason for the

existence of stable atoms and, in the last analysis, for the

existence of the material world surrounding man. But in a

popular presentation it is almost impossible to explain the

"quasidecomposable" states w1 0 w2 - w2 0 w1 • We can say that

one of the electrons is in the state wl' while the other is

in the state w2' but it's impossible to say which of them is
in which state.

Our natural language is in an even more hopeless posi-

tion when it must explain the difference between two iden-

tical fermions and two identical bosons, where the phase

space "ls,s) consists of vectors in ~s 0 ~s that are

syrnrnetric with respect to permutation. That is, one tries to

explain that it is possible for two systems to be "different,

but indistinguishable" in two different ways (even one way

caused a lot of trouble for ancient philosophy).

This is a suitable place for a digression on "natural

language". In reality, our ideas of "classical" and "non-

classical" physics are very closely connected to our idea of

what can and what cannot be expressed adequately by simple

 68

words. Here the state of affairs is not at all trivial. Not

only a popularizer but also a working physicist often first

tries to explain a new phenomenon, law or principle in a

"rough and ready" way. But one must recognize the place of

such an explanation. It strives: (a) to name and to summen

from memory the corresponding fragment of the exact theory

with mathematical formulas, structures, etc., in much the

same way that the code for a command in ALGOL turns on the

process of fulfilling the command, which also constitutes its

meaning; (b) to switch on a process of free association,

i.e., to aid in discovering that something is similar to

something else; (c) to create in the brain a structure of

intuitive notions on a subject, whose purpese is not to

replace precise knowledge of it, but rather to form value

principles and facilitate rapid estimates--to help us see

what to seek later, in what direction to explore, what is

probable and what is improbable. (In particular, this is the

value of popularization for specialists in another discipline.)

We must again emphasize the following point: the

semantics of a verbal description of some fragment of physics

does not, in general, correspond to a complex of natural

phenomena, but rather to a fragment of the theory., .whose

semantics, in turn, is made explicit by other fragments of

the theory, operational prescriptions, etc. Nevertheless,

the compulsion to interpret verbal expressions directly can

turn out to be extremely fruitful. It was in this way that

quarks were discovered: when it turned out that the space of

certain internal degrees of freedom of a hadron splits into

 69

the tensor product of three subspaces, it becarne ternpting to

consider these three subspaces as the internal degrees of

freedorn of three new particles which rnake up the hadron.

These particles are the quarks u, d, s (they "are dis-

covered but are not detected in the free state").

Returning to the problern of quantum observations, we are

led to the conclusion that their mathematical model is non-

classical because, in the first place, it is a rough version

of a much more complicated model devised to explain the

interaction of a system with another system--a "device".

When we first discussed the meaning of the mathematical

apparatus of quantum mechanics, we particularly emphasized

that the device. is macroscopic and there is no hope for a

complete quantum theory for its process of interaction with

the system. We are thus forced to replace the device by the

linear operator of the corresponding observable. Hence, the

logic of mathematical description leads us to the following

considerations, which as a totality are almost contradictory.

To the extent that the abstraction of an isolated quantum

system is sound, we need only one "observable"--the energy

operator--to describe it. However, we should not connect

this observable with the idea of measuring energy, since the

act of "measurement" requires an extension of the system.

The localization of a system in its phase space can also be

produced by "measurement" of other observables, but they are

coarse models of a combination that is inaccessible to com-

plete description (system + device + energy operator of the

system/device). After an interaction with the device, the

 70

system may lose its individuality, and the idea that it

begins a new life at a point of a new phase curve in its

phase space may lose all meaning. Finally, since even before

its interaction with the device the system was a part of some-

thing larger, most probably at no time does it have the indi-

viduality which is needed for an adequate model. Thus, it

seems there is no closed system smaller than the whole World.

After all of this it is a miracle that our models

describe anything at all successfully. In fact, tney describe

many things well: we observe what they have predicted, and

we understand what we observe. However, this last act of

observation and understanding always eludes physical descrip-

tion.

At the end of the 1960's, optical shutters were designed

for cameras which permit exposures lasting only ten pico-

seconds. In this fantastically short time a light ray in

water travels a distance of 2.2 millimeters, and one can

obtain a photograph of a short laser impulse in a bottle,

while it moves inside of it. In the Bell Telephone Labora-

tories, where such photographs were first made, a drop of

milk was added to the water, in order to increase the

scattering of the light and make the trace of the impulse

brighter. This drop of milk is a symbol af human participa-

tion in the world, where it is impossible to be only an

observer.
 4. SPACE-TH1E AS A PHYSICAL SYSTEM

"I do not define time, space, place, and motion, as being

well known to all." (I. Newton)

" ... time and space are modes by which we think and not
conditions in which we live." (A. Einstein)

We feel ourselves localized in space and lasting in

time, and, in the last analysis, almost all schemes of modern

physics describe events that are happening in the arena of

space-time. However, since the time of the creation of

general relativity theory and quantum mechanics, there has

been an increasing tendency to consider space-time as a

physical system in its own right. The principles for de-

scribing space-time still remain classical. The difficulties

of quantum field theory apparently point to a contradiction

between these principles and universal quantum laws.

The principal mathematical image of space-time is a

four-dimensional differentiable space-time manifold, which

for brevity we call the World. A point of the World is an

idealization of a very "short" and "small" event, such as a

flash, or the radiation or absorption of a photon by an atom.

Moreover, a point of the World is only a potential event;

such a point is "prepared to take" an event, but it "exists"

with or without it. The model for an event that is

71
 72

concentrated in a small region of space but continues for

some time, such as ~he life of an observer, a star, or even a

galaxy, isa curve in space-time, called the world-line or

history of an event. It is extremely important to learn to

imagine the World of Becoming as a World of Having Become,

i.e., to imagine the whole history of the Universe oralarge

part of it as a complete four-dimensional shape, something

like the "tao" of ancient Chinese philosophy.

The introduction of temporal dynamics is the next step.

It is realized as follows. In the World a space-time distance

a 0 1 2 3
and
is defined between two nearby points x = (x ,x ,x ,x )
a a 0 0 3 3 2 aß
x +dx = (x +dx , ... ,x +dx ). Its square ds =:1: gaßdx dx

is a quadratic form in the differences of the coordinates of

the nearby points. Here xa is an arbitrary local coordinate

system. A classical observer with his small laboratory of

rulers and clocks can establish a local system of coordinates

in which x 0 = ct; x1 , x 2 , x 3 are Cartesian Coordinates in

the physical space of the observer, t is recorded an his

clock. The metric in a neighborhood of the observer is close

2 2 2 2
to the Minkowski metric dx 0 - (dx 1 + dx 2 + dx 3 ). I f we imagine

the World to be populated by observers whose coordinates act

in domains which cover the World, then the Coordinates of

events must be recalculated from one observer to another.

But the space-time interval between two nearby points will be

the same even when calculated by different observers. The

speed of light c is used to convert temporal units to

spatial ones; it is in such units that time is measured. The

world-line of an observer is his own stream of time: the

 73

observer's atomic clock counts off the values of the integral

J /ds 2 along this world-line, i.e., its length. There is no

physically meaningful "common time" of the Universe (although

it can sometimes be introduced into special models of the

World) • It is not at all surprising that two curves in the

World with the same beginning and end may have different

lengths; this is even true on the Euclidean plane. There-

fore, it is no wonder that two observers who synchronize

their clocks and go separate ways discover that their clocks

disagree when they meet again. A less familiar phenomenon is

that, if two points of space-time can be joined by a world-

line of an observer, then among such lines there is one which

is the langest, but none which is the shortest (in the

Euclidean plane just the opposite is true). This is a

special property of the Minkowski metric, connected with the

fact that it is not positive definite: the square of the

interval between distinct points can be positive, negative or

zero.

Observers with the langest world-lines, i.e., the

fastest flow of their own time, are called inertial. From

the point of view of general relativity theory, they fall

freely in a gravitational field. Their world-lines are called

timelike geodesics.

The secend important class of curves in the World is the

class of trajectories of particles flying with the speed of

light, such as neutrinos. Along such lines the space-time

interval vanishes identically--"time stops" (if the particle

is massless, as was believed until recently). The geometry
 74

of such lightlike geodesics determines what a cosmological

observer can observe and, more generally, what events in the

World can influence other events. In particular, it is in

terms of this geometry that one can give a precise formula-

tion of the postulate that no signal can be propagated faster

than light.

The Minkowski World

The simplest and most important concrete example of a

World is the flat Minkowski World. It is a good local imi-

tation of most other Worlds at most points. In this World

there is an inertial observer whose coordinate system (xa)

covers the whole World and has metric identically given by

(dx 0 ) 2 - ((dx 1 ) 2 + (dx 2 ) 2 + (dx 3 ) 2 ). We fix an origin of

reference--a point on the world-line of this observer. The

Minkowski World ult becomes a vector space with our chosen

coordinate system, and this vector space structure does not

in fact depend on the inertial observer if we allow trans-

lation of the origin of reference. Therefore, the concepts

of a line, plane, and three-dimensional subspace in ~ have

absolute meanings. The linear transformations of ~ that

preserve the Minkowski metric form the Poincare group, and

the transformations in the Poincare group that fix some

origin of coordinates form the Lorentz group. These are basic

symmetry groups in all of physics; more precisely, they are

the basic symmetry groups of physical laws: none of the

points of the Minkowski World or the coordinate systems of

different inertial observers is preferred over the others,

 75

and all coordinate formulations of the same law must be equi-

valent.

The set of points at distance zero from the origin of

reference P forms the light cone CP with equation

(x 0 ) 2 - (x 1 ) 2 - (x 2 ) 2 - (x 3 ) 2 = 0. The timelike geodesics that
pass through P are lines lying inside CP' and the light-
like geodesics are lines an CP itself, i.e., the genera-
trices of the cone. The cone consists of two halves, a

"past cone" and a "future cone". The time along a timelike

geodesie flows from the past half to the future half. The
distinction between the two halves of CP depends continu-
ously an P; this expresses the existenae of a singZe direation of

time throughout the Minkowski World, despite the absenae of a

singZe time.

The point P and a timelike unit tangent vector at P

form a model of the "instantaneous observer" in the Minkowski

World. The vector indicates the direction of the observer's

individual time. The subspace in ~ orthogonal to this
vector is a model of the three-dimensional physical space of
the instantaneous observer. Its metric (with sign reversed)

is obtained by restricting the Minkowski metric.

The physical spaces of two different instantaneous

observers are different, even if the observers are located at

the same point of ~. They intersect the world strip of,

say, a ruler at different angles. In some approximation such

an intersection is the instantaneous observable form of this

ruler. Therefore, it can have a different length for

 76

different observers--the space and time coordinates can corn-

pensate for one another.

How should one then irnagine the observation of a distant

object, say a star, in the Minkowski World? Suppose the

observer rnoves on a world-line P, and the star rnoves on its

world-line s. We irnagine the past half c+ of the light

Po
cone at Po as it rnoves along with PO E P. I t "sweeps out"
a certain part of u«, a dornain of the World which an observer

could observe. At a given point P0 the observer sees the

star at the point of intersection of s with by rneans

of the light ray joining and P0 . But we rnust still

figure out how to deterrnine the visible position of the star

in the sky of the observer. The problern is that his sky

"lies in his physical space EP ", and not in the Minkowski

0
World ~' and the position of the star is rnodeled by a ray

in EP In order to find this ray, we rnust project a ray in

0
u«--the half-line frorn passing through s n c+ --onto the
Po
physical space EP . This projection is orthogonal (of
0
course, relative to the Minkowski rnetric).

Thus, it is convenient to distinguish the "absolute sky"

at the point P 0 --the base of the past half of the light

cone--frorn the sky of an instantaneous observer at this

point, i.e., the projection of the absolute sky onto the

physical space of this observer. In classical cosrnography,

the sky can be irnagined as a crystalline sphere of indeter-

rninate radius; angular distances are defined between points

of the sky, and the geornetry of the sky is the sarne as the

geornetry of a rigid sphere. But the angular distances

 77

between stars will be different for another observer; flying

with a very large velocity in the direction of the con-

stellation Orion, we see that it becomes squeezed together,

while the opposite celestial hemisphere is stretched (astro-

nomers call this aberration). Thus, the mathematical struc-

ture of the "absolute sky" is not the same as the structure

of a Euclidean sphere: angular distances in it do not have a

meaning independent of the observer. A detailed study shows

that the natural structure for the absolute sky is the

complex Riemann sphere, i.e., the complex number plane ex-

tended by a point at infinity which is treated indistinguish-

ably from the finite points. More precisely, the Riemann

sphere is the set of one-dimensional vector subspaces in a

two-dimensional complex vector space; equivalently, it is the

complex projective line CP 1 . In particular, the natural

coordinates of a star in the sky are complex numbers. We

choose three reference stars and assign the coordinates 0, 1,

to them. Then there is a simple procedure that enables us

on the basis of Observations to associate to any fourth star

a complex number z, which will be observer-independent.

If the coordinates 0, 1, oo are given not to the refer-

ence stars, but rather to three points of the sky joined by

three orthogonal axes of the spatial coordinate system of an

observer, then, of course, the complex coordinates z and

z' of the same star can be different for different observers

at a given point of the World. But they are always connected

by a fractional-linear transformation of the form z' =

az+b/cz+d, where a,b,c,d are complex numbers independent

 78

of the star, for which ad- bc = 1. The matrix ( ~ ~) with

determinant 1 is an unusual representation of the Lorentz

transformation connecting two inertial coordinate systems at

a single point of the World. In modern physics, however,

this representation of the Lorentz group is considerably more

fundamental than the usual 4 x 4 transition matrices.

Curved World

More general models of the World are different from the

Minkowski World in several respects. First, the metric cannot

be reduced to the form even

locally in the coordinate system of an inertial observer.

Second, there may be no global coordinate system at all.

Third, the metric of the World and the history of matter and

fields in the World are not independent--the curvature of the

metric is determined by matter, and, in turn, the metric

imposes strong restrictions on the possible histories of

matter; these constraints are described by the Einstein

equations.

Geometrically, the typical ways the World curves can be

perceived by studying the possible behavior of nearby time-

like geodesics (local aspect) and light cones (global aspect).

We shall try to give a verbal description of the effects of a

very strong curvature, leading to the concept of a black hole.

We imagine a world-line S of a point mass m. With it

one associates the characteristic space-time interval length

2
2Gm/c --the so-called Schwarzschild radius. The points of
the World that lie at a timelike distance from S at most
 79

equal to the Schwarzschild radius form the Schwarzschild tube

S around S. Far away from this tube, the World is almost

flat (if we disregard the influence of other matter). But

inside the tube the World is so curved that for any point

PO ES the future half of the light cone cP lies wholly

0
within the Schwarzschild tube. The gravitational field of

the mass m does not release photons emitted inside s.

However, the past half c+ does not necessarily lie in 5--
Po
the Schwarzschild tube can absorb external radiation.

We now consider a more realistic model, when the mass m

is concentrated in a finite domain whose radius may be greater

than the Schwarzschild radius. For example, the Earth has

Schwarzschild radius of about 1 cm, and the Sun has Schwarz-

schild radius of about 3 km. In this case the Schwarzschild

tube has no particular physical meaning. But in the process

of evolution a star of sufficiently large mass can collapse

under the influence of its own gravitation, so that at some

point K0 of the world-line of its center the radius of the

star becomes comparable to the Schwarzschild radius, and then

it decreases further. The segment of the Schwarzschild tube

after the point K0 will be a domain of space-time that is

not accessible to an external observer. The four-dimensional

picture of what an external observer sees will be approxi-

mately the following. The past half of the light cone of the

observer at any Observation point intersects the world tube

of a star, but this intersection always occurs before the

point K0 • In other words, the local time along the observ-

er's end of the light rays will tend to infinity, but the
 80

local time along the world-line of the star emitting the

light rays will tend to a finite limit corresponding to K0

(we recall that local time is the length of the corresponding

world-line). According to the observer's clock, the collapse

lasts infinitely long.

We repeat once again how the four-dimensional picture

relates to the visible image in the sky of the observer. In

a curved World one more step is added to the already complex

process of translation. This step is the construction of a

flat World that is tangent to the curved World at the point

where the instantaneous observer sits. To obtain the visible

image of a star S in his sky, the observer must: (a) con-

struct in the curved four-dimensional World the incident light

geodesie joining the observer's position to the world-line of

the star; (b) construct the tangent half-line to this geo-

desic in the flat World which is tangent to the curved World

at the point where the observer sits; (c) draw the tangent

to the world-line of the observer in the same flat World;

(d) construct the instantaneous physical space of the ob-

server in this flat World; (e) project the tangent to the

light ray from the star onto this space.

Thus the "shado"YTS of ideas" move on the wall of the Cave.

Spinors, Twistcrs and the Complex World

If we adopt the idea that a point of space-time is

an idealization of the classical notion of a "smallest

possible event", then we are inevitably led to the necessity

of considering still other geometric models as our knowledge
 81

of such events increases. For exarnple, the act of absorbing

a photon is far frorn being cornpletely characterized by in-

dicating the point of the World at which it occurs--we rnust

indicate the energy and polarity of the photon. Nor does the

position of an electron on its world-line cornpletely deter-

rnine its state--the direction of its spin rnust be indicated.

Although both polarity and spin are quanturn-mechanical

internal degrees of freedorn, it is rernarkable that their geo-

rnetric description is included in a very natural way in the

geornetry of the World. Narnely, the value of polarity or spin

at a point of the World is a ray in two-dirnensional cornplex

space, or a point on the Riernann sphere CP 1 .- It turnsout

that this Riernann sphere can be naturally identified with the

absolute sky of this point, which, as we explained before, is

also a Riernann sphere. Therefore, for each World ~ we can

consider an extended World ~. a point of which is a pair (a

point of ~. a point of the absolute sky over this point of

~) . There is a natural mapping rnaking into a

bundle over ~ with fibre CP 1 . Here if we replace each

by the two-dirnensional cornplex space of which CP 1 is

the space of rays, we are led to the farnous spinor space of

Dirac.

The world-line of a spin 1/2 particle, say an electron,

is naturally represented as a line in ~. rather than in ~.

Rays of light are also lifted naturally to ~: at each

point a ray deterrnines a point on the corresponding sky,

narnely the point at which it is directed; the set of such

pairs in ~ is the irnage of this ray in ~-

 82

R. Penrose suggested that one study the remarkable

space H which is obtained if each ray in ~ is contracted

to a point. Mathematically, H is the quotient space of ~

by the equivalence relation defined by the rays in ~. In

order to better understand the construction of H, we analyze

the case when ~ is the flat world of Minkowski. Then no

point in ~ lies in the intersection of rays (in centrast

to ~), and each ray either does not intersect a given sky,

or else intersects the sky in just one point. Therefore,

each sky CP 1 is simply embedded in H without self-inter-

section, and some skies do not intersect each other. The

simplest space in which one can pack many complex lines is

the complex projective plane CP 2 . But it cannot be a

candidate for the role of H, since any two lines in cP 2

intersect. In fact, H lies in CP 3 , complex projective

three-space, the space of "projective twistors". To be sure,

the skies over the points of the Minkowski World are not all

the lines in CP 3 , but only part of them, lying in a five-

dimensional hypersurface (the whole of cP 3 is six-dimen-

sional). It is very useful to introduce additional "ideal"

points of the flat world ~. whose skies correspond to the

missing lines in CP 3 . We obtain the compact complex space-

time of Penrose, denoted C~; the coordinates of points in

it can be arbitrary complex numbers and, moreover, there is

an entire complex light cone "lying at infinity".

Can such an abstract construction have any relation to

physics? It seems that the answer must be yes. One reason

is the comparatively recent discovery of an analogy between

 83

quantum field theory and statistical physics. If the purely

imaginary coordinate ict is replaced by a real one in the

basic formulas of quantum field theory, then, roughly speak-

ing, they become the basic formulas of statistical physics

(the role of ict is played by the inverse of the tempera-

ture).

Geometrically, such a replacement corresponds to a

transition from the Minkowski World with its indefinite

metric to a four-dimensional Euclidean World with the "sum of

squares" metric. Conveniently, this World fits into C.Al;

one need only rotate the time axis in .At by 90°. All the

other points of C.Al are obtained by interpolating between

the Euclidean and Minkowski worlds, and, apparently, descrip-

tions of many important phenomena admit an analytic continua-

tion from .At to c.At or to part of c.Al. A standard

example is the interpretation of the "tunnel effect" of

quantum mechanics in terms of the classical evolution of a

system in imaginary time.

The "Penrose paradise" H = CP 3 (it is natural to use

the word "paradise" to refer to a space which contains all

skies, but in which nothing remains of space-time) has quite

recently turned outtobe very useful in the study of Maxwell's

equations and their generalizations--the Yang-Mills equations,

which, we now believe, describe gluon fields connecting the

quarks in a nucleon. There are profound physical reasons to

think that a world filled only with radiation (or with

particles traveling at velocities close to the speed of light,

almost along light cones) can be described better in terms of

 84

the geometry of H than in the real four-dimensional geometry

to which we have become accustomed. What binds us to space-

time is our rest mass, which prevents us from flying at the

speed of light, when time stops and space loses meaning. In

a world of light there are neither points nor moments of

time; beings woven from light would live "nowhere" and

"nowhen"; only poetry and mathematics are capable of speaking

meaningfully about such things. One point of CP 3 is the

whole life history of a free photon--the smallest "event"

that can happen to light.

Space-Time, Gravitation and Quantum Mechanics

The most important lesson that one learns by studying

the relations between our models of the classical World and

the quantum mechanical principles for describing matter, is

that we understand these interrelations very poorly--the

basic rules for the descriptions do not agree with one another.

Here is one example. If we try to mark a point of the

World on the world-line of a particle of mass m, we cannot

do this with error less than the Schwarzschild radius 2Gm/c 2

of this mass. Therefore, to increase the accuracy we should

make use of particles whose mass is as small as possible. On

the other hand, as noted by Landau and Peierls, the uncer-

tainty principle for (coordinate, momentum) and the fact that

all velocities are bounded by the speed of light imply that

the indeterminacy of position cannot be less than n/mc; that

is, an increase in accuracy requires the use of particles

whose mass is as large as possible!

 85

Both limits on the accuracy are equal for a mass which

is found from the equation n/mc = 2G/c 2 , i. e., precisely the

Planck mass. This common limit on the accuracy of position

measurement is the Planck length. Hence, the Planck unit of

length (recall that it is of the order of lo- 33 cm) gives

the lower bound under which the notion of a space-time domain

is automatically inapplicable. A smaller domain is incapable

of supporting an elementary event. The events studied in

modern accelerators occur in considerably larger domains, but

nevertheless the limit of localization is clearly indicated

within the framewerk of contemporary theories.

Quantum principles interfere with the notion of a point

of the World as an elementary event in many other ways as

well. A free quantum particle with fixed four-dimensional

momentum k is nowhere localized--it is "uniformly spread

over the World", since its psi-function is the de Broglie

plane wave eikx Two identical quantum particles must be

described by a psi-function depending on two points of the

World which is symmetric for bosons and antisymmetric for

fermions. But this literally means that nothing can "happen"

at a single point of the World; more precisely, a point is

inseparably connected with all the other points through the

psi-functions of the matter and fields.

In a future theory the image of the real four-dimen-

sional World with the Minkowski metric may turn out to be

something like a quasiclassical approximation to an infinite

dimensional complex quantum World. For example, in geometric

optics, which is an approximation to wave optics, there is

 86

the concept of a caustic: the set of points where the in-

tensity of radiation in this approximation is infinite. It

is tempting to think of the four-dimensional World as a kind

of caustic manifold for an infinite dimensional quantum wave

picture. Our difficulties with, for example, the infinite

density of vacuum energy in quantum electrodynamics, would be

solved nicely in this scheme.

The Lorentz group is a strange group from the real point

of view, but if it is replaced by SL(2,C), a group of complex

2 x 2 matrices, then we obtain a very natural object--the

group of symmetries of the simplest imaginable state space of

a quantum system. Does this not mean that the spin degrees

of freedom are more fundamental than the space-time degrees?

The mysterious separation of the World into space and time is

implicitly contained in the group SL(2,C), and therefore its

existence is "explained" on the basis of rules that do not

assume such a separation a pr>ior>i. Moreover, as we have seen,

a World without rest mass matter can be obtained from SL(2,C)

(or its generalization SL(4,C)) without introducing space-

time. The points of our four-dimensional World, or small

regions in it, are distinguished by events which happen with

rest mass matter. Perhaps, both mass and space-time result

from a spontaneaus violation of the symmetry of the basic

laws.

It is difficult to invent such a theory. We are still

trying to quantize the classical Universe as we did for the

hydrogen atom, rather than trying to obtain its form as the

limit of a quantum description. Perhaps the first successful

 87

quantum model of the World, say, near the Big Bang, will be

quite simple mathematically, and only the habits of an inert

mind hinder us from guessing it now. I would like to live to

see the time when such a model is proposed and accepted.

 5. ACTION AND SYMMETRY

"Physics is where Action is." (Author unknown)

A small ball rolling in a trough under the action of its

own weight is the simplest physical system. The rest state

is the simplest history it can have. The ball can rest only

at those points of the trough \'lhere the tangent is horizontal;

otherwise it rolls down an incline. We give the position of

the ball by a horizontal coordinate x, and we let V(x)

denote its potential energy (proportional to the height of

the trough) at the point x. The points x at which the

ball can rest are the solutions of the equation dV/dx = 0 or

dV = 0; the increment of the potential energy function as we

move a little away from such a point is very small compared

to the increment of the coordinate. At these points V is

stationary.

A second example is a soap film that is stretched

between two wire loops. The equilibrium form of a film at

rest is determined by requiring that for small deformations

its surface tension energy V varies by an amount which is

very small compared to some natural measure of the amount of

deformation. The surface tension energy V is proportional

to the area of the film; hence, the form of a film at rest is

88
 89

a state in which the area is stationary. The function (or

functional) V in this case is defined on a domain consisting

not of numbers x, but of all possible surfaces spanning the

given contour. Instead of the differential dV we usually

write the "first Variation" oV.

For example, suppose that circular wire loops of radii

r and R are located in parallel planes at a distance ~.

and the line joining their centers is perpendicular to these

planes. This line is an axis of symmetry of the contour, and

therefore one would expect the equilibrium form of the film

tobe a surface of revolution of a curve q(x) for which

q(O) =r, q(~) =R. We assume this tobe the case; then the

energy V is proportional to the area 2n Jf q(x) (1 +

+ (dq/dx) 2 ) l/ 2 dx; the equation ov = 0 can be rewritten:

-<ii 2 q/dx 2)+ (dq/dx) 2 +1 = 0; and then we can solve this

equation with the indicated boundary conditions.

Such a transition to formalism is both useful and

dangerous. We postulated that the symmetry of the boundary

conditions leads to symmetry of the equilibrium form. Here

is a completely analogous example where this is obviously not

the case. We load an elastic vertical rod with a compressing

force; for some critical value of the load the rod buckles.

The direction of buckling in a plane perpendicular to the rod

is in no way distinguished in the original axisymmetric

picture, but it is distinguished when the buckling occurs.

In such situations physicists speak of a spontaneaus viola-

tion of symmetry: a phenomenon subject to certain laws is

less symmetric than those laws. Besides, in our previous

 90

problern we neglected the solution consisting of two flat

films stretched over each loop separately.

This is a reminder of the need, when studying a func-

tional V on an infinite dimensional manifold such as the

space of surfaces, to try to analyze the geometric situation

beforehand. Topology is concerned with such problems. Only

in recent years have the geometric ideas furnished by topo-

logy been systematically applied in physics, for example, the

notion of "topological charge" has appeared in quantum field

theory. (Unfortunately, there is no space for us to discuss

this here.) Mathematically, it is useful to take oV = 0 as

the basic equation. The domain of definition of the function

V is perhaps not completely defined, but it may be made more

precise in the course of the solution of the problem. The

graph of V is an infinite dimensional trough, in which our

system moves, seeking rest.

Physically, this image is justified by an amazingly

universal principle, which can be stated as follows: the

deveZopment in time of fundamental eZassieaZ systems is their equi-

Zibrium in spaee-time. More precisely, the kinematics of a

system is determined by describing its set of virtual his-

tories in the appropriate space-time domain. A functional S

having the dimension of action is defined on these virtual

histories. The dynamics of the system is described by the

condition oS = 0. The passage from the dimension of energy

(V) to the dimension of action (S) is connected, of

course, with the addition of the time coordinate. Action is

primary; energy is only its time derivative. In the
 91

resulting fundamental theory action remains, while energy

becomes a quasiclassical quantity.

The virtual history or path ~ of a system in a space-

time domain U is defined classically as a cross-section of

the degree of freedom bundle of the system over this domain.

If the domain is a union of two disjoint pieces u1 and u2 ,

and the path ~ is a union of the paths ~l and ~2, then

the action of ~ is the sum of the actions of ~l and ~2•

This postulate is satisfied by almost all the action func-

tionals used in classical physics. Therefore, the complete

action of the path ~ in the domain U can be written as a

sum of actions of ~ on many small domains which cover U,

andin the limit as an integral S(~) = J0 L(~(x,y,z,t))

dx dy dz dt, where (~) is the set of intrinsic coordinates

of the system and their space and time derivatives. The

function L (or its integral over the space coordinates) is

called the Lagrangian of the system: it is the density of

action. If we place two systems with Lagrangians L1 and

L2 in a domain of space-time, then the Lagrangian of the

combined system has the form L1 + L 2 + L 12 , where the third

term is the "density of the interaction". The two systems do

not interact if this term is zero.

Space-time itself ("vacuum") contributes to the

Lagrangian a term proportional to the curvature of space-time.

Therefore, space-time can be considered on the same basis as

systems which include matter or an electromagnetic field.

 92

The role of action in quantum physics was clarified in

an extraordinary fashion by Richard Feynman, who based his

discussion on earlier werk of Dirac. In the absence of a

fundamental quantum theory we are forced to formulate his

ideas as a recipe for "quantization", i.e., transition from

the classical description of some physical system to a quantum

description. According to this recipe, one imagines that each

classical history ~ contributes to the quantum description

of the history of the system, but it does so with its complex

weight (phase factor) e -iS(~) . (The action is, of course,

measured in units of fi.)

We explain this in more detail. We fix the classical

behavior of the system on the boundary of the domain U, say

~l and ~2 at moments of time t1 and t 2. From these

conditions--"loops for soap films"--quantum theory determines

not the classical history of the development from ~l to ~2 ,

but rather a complex nurober G(~ 1 ,t 1 ,~ 2 ,t 2 )--the amplitude of

the transition probability from the state (~ 1 ,t 1 ) to the

state (~ 2 ,t 2 )--the square of whose modulus in principle is

observable or enters into other observable quantities.

Feynman's prescription says that this amplitude is

f e -iS(~) D~, where the integral is taken over the infinite

dimensional set of classical paths joining (~ 1 ,t 1 ) and

(~ 2 ,t 2 ); the use of D~ instead of d~ serves as areminder

of the infinite dimensionality: it is not a differential,

but a "volume element"!

In the prehistory of integral calculus, an important

place is occupied by the remarkable work of Kepler

 93

"Stereometry of Wine Barrels" (see Vol. 9 of his Gesammelte

Werke) • Integrals that give the volume of solids of revolu-

tion used in commerce were calculated in this work at a time

when the general definition of an integral had not yet

appeared. The mathematical theory of Feynman's magnificent

integrals, which physicists write in vast numbers, is not

really far removed from the stereometry of wine barrels.

From the viewpoint of a mathematician, each such

calculation of a Feynman integral simultaneously defines what

is calculated, i.e., it constructs a text in a formal

language whose grammar has not previously been described. In

the process of such computations a physicist calmly multiplies

and divides by infinity (more precisely, by something which,

if it were defined, would probably turn outtobe infinite);

sums infinite series of infinities, assuming here that two or

three terms of the series give a good approximation to the

whole series; and generally lives in a realm of freedom

uneneurobered by all "moral standards".

It would hardly be possible to construct a consistent

and applicable mathematical theory of Feynman integrals

without progress in the understanding of physics. The very

idea of "quantization" belongs not to physics, but to the

history and psychology of science. There can only be a

substantive meaning to "dequantization", i.e., the passage

from a quantum description to a classical one, when the

latter is meaningful; the converse, however, can never have

substantive meaning.
 94

The classical fields occurring in the Lagrangians of the

weak and strong interactions, for example, are physical

phantoms: we do not know their meaning outside of quantiza-

tion, and it is unlikely that they would describe the

virtual classical histories of anything whatsoever. (It is

believed that the situation is better with the quantization

of electromagnetism.)

Finite ·dimensional quantum models allow us to guess

which features of the Feynman formulation are essential, and

which are atavistic. As was explained in Chapter 3, the

evolution operator for a closed localized quantum system at

its local time t has the form e-iS(t), where S(t) is now

an operator with the dimension of action. If we imagine the

different world-lines of the system with different local

times, we see that quantum action is a connection in the

space of internal degrees of freedom of the system, deter-

mining physically admissible histories as parallel trans-

lations. The recipes for quantization are a primitive mani-

festation of the fact that the space of internal degrees of

freedom "at a single point" in vacuo is already infinite

dimensional because of the virtual generation of particles.

Further understanding is blocked until we relinquish the

idea of space-time as the basis for all of physics.

"Symmetry denotes that sort of concordance of several

parts by which they integrate into a whole. Beauty
is bound up with symmetry." (H. Weyl)

Other than this quote, I shall not repeat anything from

what Hermann Weyl wrote in his remarkable book "Symmetry"

 95

(Princeton, 1952). It should be read by anyone who wishes to

travel the road from perceiving symmetry as a sensory datum

(colors, ornaments, crystals) to understanding it as the

deepest of physical and mathematical ideas.

The universal mathematical structure describing symmetry

is a group G and its action on a set X, for example, the

group Sn of all permutations of the numbers (l, ... ,n).

The action is a mapping G x X-+ X which sets up a correspon-

dence between a pair (an element g of the group, a point x

of the set) and an element gx of the set (the image of x

under the action of g). Allelements of the form gx for

variable g constitute the orbit of x under the action of

the group. The group G itself is never given as a physical

object--we can imagine a rigid body as a sensory datum, but

the set of all rotations of it is an idea located on a

higher level of abstraction. The homonymy of the words

"action" in the contexts "action of a group" and "action of

a segment of a virtual history" in the basic European lan-

guages is a random consequence of the vague initial idea of

"change as a result of doing"; but in the expression

u (t) = e -iS (t) for a quantum evolution operator, this homo-

nymy unexpectedly acquires a deep significance.

One of the great achievements of mathematical thought

was the separation of the concept of an abstract group from

the concept of a group action on a set; this has turned out

to be of great importance for physics. We imagine a hydrogen

atom as an electron, moving in a central Coulomb field around

a fixed nucleus. The group 50(3) of rotations around the

 96

nucleus acts on the complex vector space of quantum states of

the electron. It turns out that the entire infinite dimen-

sional space of possible un-ionized states splits into a sum

of finite dimensional subspaces, with S0(3) acting sepa-

rately on each. These actions are irreducible linear repre-

sentations of the group, and are the possible stationary

states of the hydrogen atom. The precise mathematical

description of these representations explains the spectrum,

quantum numbers, etc. Similarly, the Poincare group--the

full group of symmetries of the Minkowski World--acts on the

space of quantum states of a solitary particle in an imagined

World containing nothing besides the particle. As Eugene

Wigner showed, the experimental classification of the ele-

mentary particles by their mass and spin is included in the

classification of (infinite dimensional) irreducible repre-

sentations of the Poincare group. This gave rise to the

joke that the twentieth century world consists not of the

four elements air, fire, earth and water, but of the irre-

ducible representations of some group. In recent decades,

theoretical physics has been occupied with an intensive

search for the symmetry groups of fundamental interactions;

their laws (Lagrangians) appear as a secondary object in the

mathematical description.

Physical accounts that stress formalism rather than

set-theoretic interpretation are sometimes obscure about what

the group of symmetries acts upon. Here are two extreme

cases, which lead to fundamentally different physics:

(a) the symmetry group G acts on the phase space of a

 97

system and takes its phase portrait into itself; (b) the

system's phase space X is itself represented as the set of

orbits of some other space Y under the action of the group

G.

Case (a) includes the description of the hydrogen atom

considered above. The discrete invariants of an irreducible

representation are the quantum numbers of the corresponding

state, but the state vector in the representation space is

itself not uniquely determined, as in the example of the

buckling of a rod, in which the syrnrnetry is spontaneously

violated. In schemes for describing fundamental interactions,

one often postulates a scheme for the violation of syrnrnetry

by means of a weaker interaction.

Violation of syrnrnetry is also an ambivalent term. It is

possible to speak of the spontaneous violation of syrnrnetry

when the phase portrait of the system is syrnrnetric, but its

individual trajectories are not. Accounting for a new inter-

action generally violates the syrnrnetry of the whole phase

portrait, since it changes the phase portrait. But if this

change is small, then its effect can be approximated by con-

sidering the new interaction as a small perturbation of the

original syrnrnetric picture.

Case (b) includes the so-called gauge theories. A

classical matter field in such a theory (an object capable of

quantization) is not a cross-section of the bundle of internal

degrees of freedom, as we stated earlier, but rather an

entire orbit of such cross-sections under the action of a

gauge group of transformations. Over a single point of

 98

space-tirne such a group is represented by certain rotations

of the space of internal degrees of freedorn, but these

rotations can vary independently and continuously as the

point varies.

The condition that the Lagrangian of the theory should

be invariant relative to such transforrnations irnposes very

stringent requirernents, which sharply lirnit the choice of

Lagrangian. In supersymmetric theories, about which we spoke

briefly in the first chapter, the gauge group can be a still

rnore general object; in particular, it can rnix up the boson

and ferrnion fields. It is the gauge and Supersymmetrie

theories which are now rekindling the basic hope for a Grand

Unification, i.e., the construction of a unified theory of

fundamental interactions.

In conclusion, I would like to say a few words about the

theory of nurnbers, a rnathernatical discipline which is highly

developed and possesses arnazing beauty, but which has as yet

found no deep applications to natural science. One of the

principal objects of study in nurober theory is prirne nurnbers

(a prirne is a positive integer that has no integer divisors

besides itself and one). Euclid's Elements already contained

a theorern to the effect that there are infinitely rnany prirne

nurnbers. (If we have a finite set p 1 , ..• ,pn of thern, we

can construct another prirne nurober as the srnallest divisor of

the nurober p 1 · • ·Pn + 1 besides one). This is a rernarkable

argurnent, for all its sirnplicity.

Here is a rnore modern exarnple of a theorern on prirne

nurnbers. Let T(n) (the nth Rarnanujan nurnber) denote the
 99

nth coefficient of the series that is obtained after a formal

expansion of the infinite product X n

rn=l
(1-xrn) 24 . If p is a

prime number, then I T (p) I < 2p 11 / 2 . It is cornpletely impossible

to prove this here; according to the author of the proof,

P. Deligne, in order to present this proof, presupposing

everything known by a beginning graduate student in rnathe-

rnatics, one would need about two thousand pages of printed

text. This theorern probably holds the record in modern

rnathernatics for the ratio of the length of its proof to the

length of its staternent. Of course, the proof has led to a

better understanding of rnany interesting things. For exarnple,

in order to prove this theorern a vast new theory was created

(".Q,-adic cohornology") and two or three old ones were used

(Lie groups, autornorphic functions, ... ).

It is rernarkable that the deepest ideas of nurober theory

reveal a far-reaching resernblance to the ideas of modern

theoretical physics. Like quanturn rnechanics, the theory of

nurnbers furnishes cornpletely non-obvious patterns of relation-

ship between the continuous and the discrete (the technique

of Dirichlet series and trigonornetric surns, p-adic nurnbers,

nonarchirnedean analysis) and ernphasizes the role of hidden

syrnrnetries (classfield theory, which describes the relation-

ship between prirne nurnbers and the Galois groups of algebraic

nurober fields). One would like to hope that this resernblance

is no accident, and that we are already hearing new words

about the l'i'orld in which we live, but we do not yet under-

stand their rneaning.