;

104

Prof. A. S. Eddington.
velocity

Generalisation of

WeyVs

Hence the migration

I
n)

zzz

2M + m
.

-,

fl

Tl (I

lit)
'

dv

__

2m^/(3JcT)

v/(M)

dm

2M m 2M 2M + m
1
'

For lithium, where Mi and

M2

are 6 and 7

if

m=

18, there

should be a

difference of 1*24 per cent, in the migration velocity.

For chlorine with

If a solution

Mi

= 35

and M2

37, it should be nearly 1*7 per cent.

were

electrolysed in

some medium

like gelatine, the

head

of the

column should and

contain one isotope only, and

test this.

it is

hoped shortly

to carry out experiments to

In the above calculation,

of course, the effect of hydration

association have not been taken into account.

One
whole

other point to which

it

may

be worth while to refer

if

is

the very

small energy difference between atoms

their atomic weights

approach
chemical

numbers

closely.

Quite

apart

from

positive

rays,

evidence shows that the difference between the mass of one carbon atom

4*5

and three helium atoms 10~ 7 ergs per atom,

.

is

so small that the energy difference cannot exceed

aboufc

i.e.,

1/30

of the

energy

of the

average radio-

active

change.

If

this
it is

is

true, it

somewhat

curtails

the store of intra-

atomic energy, which

hoped

will one

day become

available.

__Egjy_-i:i-^^iju;-ide3_as

_i_a^^

Generalisation of

WeyVs Theory of the Electromagnetic and

Gravitational Fields.
A.
S.

By

Eddington, F.E.S.

(Keceived February 19, 1921.)

1.

In the non-Euclidean geometry of Eiemann, the metric

,

is

defined by
of

certain quantities, g^ v

which are

identified

by Einstein with the potentials

that,

the

gravitational

field.

H. Weyl has shown

on removing a rather
for the

artificial restriction in

Eiemann's geometry, the expression

metric
the

includes also terms which are identified with the four potentials of

electromagnetic
it
is,

field.

I believe that Weyl's geometry, far-reaching

;

though
the

yet suffers from an unnecessary and harmful restriction

still

and

it is

object of this paper to develop a

more general

theory.
its

In passing beyond Euclidean geometry, gravitation makes

in

appearance

passing

beyond Eiemannian geometry, electromagnetic force appears

to

what remains
problem

be gained by further generalisation

?
t

Clearly, the non-

Maxwellian binding-forces which hold together

of the electron

an

electron.

But

the

must be

difficult,

and

I cannot

say whether the

Theory of the Electromagnetic and Gravitational Fields.

present generalisation succeeds in providing the materials for
its

105

solution.

The present paper does not seek these unknown

solidating the

laws, but aims at con-

known

laws.

hope to show

that, in freeing

Weyl's geometry

from

its limitation,

the whole scheme becomes simplified, and

new

light is

thrown on the

origin of the

fundamental laws
is

of physics.

The

essential point of

Weyl's theoryf

that comparisons of length at

different

times and places

may

yield discordant results according to

the

route of comparison.

Accordingly, a particular standard of length should

it is;

only be used at the time and place where

and

it

is

necessary to set up

a separate unit of length (interval-length) at every point of space and time.

Such a system

of unit-standards is called a gauge-system,

and Weyl's theory

as

proceeds on the view that

coordinate system
is.

the

gauge-system

is

arbitrary, just

the

Nevertheless, he recognises that a certain

natural

gauge-system plays a very important part in physics.

lengths on the "sun and the earth
:

We
of

do compare

and, although our

method

comparison
hesitation

may
mile.

be ambiguous for the thirtieth significant figure,

we have no

in deciding

whether a given length on the sun should be called an inch or a

first
its

This natural gauge-system seems at

it

like

an excrescence on the

purely geometrical theory, diverting

necessary in order to bring
it

from

natural development, but

into touch with physics.

light

The theory here

gauge-

developed seems to throw

system.

new

on the meaning

of this natural

deductive study of world-geometry and an inductive study of obser-

vational science attack the problem of Nature from opposite ends.

The most

elementary concepts of experimental knowledge

stage in the
first,

may

not appear until a late

falls

deductive treatment.

Our work thus

into

two stages
;

the development of a pure geometry of a very general kind

second, a

physical theory based on the identification of the geometrical functions with

quantities obtained by experimental measurement.

The introduction

of the

natural gauge-system marks the transition from pure geometry to physics.

Geometrical Theory.
2.

Some

of the tensors used in the analysis possess the property of being

of gauge-system, in addition to their ordinary

unchanged by any alteration

by the prefix "in-

properties with regard to transformation of co-ordinates.

" (in -tensor, in-vector, in-invariant).

I distinguish these

In symbolic notation,

a star (*) indicates quantities generalised from the corresponding quantities

Further developments are scattered + Berlin. Sitzungsberichte,' May 30, 1918. throughout his book Kaum, Zeit, Materie.' See also his article, Nature,' February 17, I need scarcely say that I am under great obligations to these sources. 1921.
c '

t 1

*>

iU

106
in

Prof. A. S. Eddington.

Generalisation of WeyVs
wholly or

Biemannian geometry

so as to maintain this " in- " property,

partly.

displacement, dx^

is

a simple example of an in-vector.

Its

components
to

are differences of coordinates (pure numbers),

and have no reference

it

any

gauge.

Since

clx^

is

contravariant,
its

we

often write

with the notation

(dxy, which reminds us of

character.

Let A^ be an in-vector representing an infinitesimal displacement at a

point P.

Can we
?

find at

an

infinitely near, point, P',

an exactly equivalent

displacement

Clearly, the equivalent displacement will not in general be

field

represented by A^, because, for example, in a uniform

polar components vary from point to point.

of force the

There will be a spurious change

of A^, attributable to the curvilinearity of the coordinates,

in- vector is displaced
"

even when the

possible

without absolute change."

The most general

the in-vector
is

linear

expression for this spurious change,

,

when

moved
(21)

through a displacement, dxv

is

-V
where F va^
coefficients,
y

va

A(dxy,

which

is

not assumed to be a tensor, represents arbitrary

is

and the expression

summed
"

according to the usual convention.

Only

linear terms

need be taken, because

Aa and

(dx) v are both infinitesimals.

" is called

Following Levi-Civita, displacement

parallel displacement.

without absolute change

For parallel displacement, the change dA* must be

equal to (2*1); and accordingly

^ + IVA =
ox v
is

0,

(2-2)

the condition for parallel displacement.

The

definiteness of the notion of parallel displacement depends on the

recognition of some kind of equivalence between displacements at infinitely

near points not between

relations

the

analytical

expressions,
is

but

between the

which they express.

For

We

assume that there

it,

such an equivalence

and, in justification,
tion possible.

we point out

that

seems

to be the

minimum assump-

if

there were no comparability of relations, even the most

closely adjacent, the

continuum would be divested

of

even the rudiments of

else.
,

structure, and nothing in nature could resemble anything

If in (2*1)

we take

to

be a displacement PPi

(Sx)

which on parallel

displacement to P' becomes P'P/, the difference of co-ordinates of

will be

P/ and P
(2*3)

(dxy + (8xy-T vav(8x)(dx) v

Interchanging the two displacements,
i.e.,

displacing PP' along PPi,

we

shall not arrive at the same point P/, unless

IV =

r.A

(2-4)

Theory of the Eleetromagnetie and Gravitational Fields,

This
is

107

a necessary condition for

c

what

is

called qffine geometry (fully treated

in Weyl's

Kaum,

Zeit,

Materie

').

It appears to express the condition that}

it

the world

is " flat

in its smallest parts " or that

possesses a definite tangent.

I shall adopt the condition (2*4) in the present investigation, ruling out
" infinitely crinkled " world.f
3.

an

Let an in-vector
circuit.

A**

be carried by parallel displacement round a small

complete

The

difference of the final

and

initial values is

J initial

Xy

-|r,/A(^)"by(2-2)
dS",
(31)

= -i J[{Jr(r~*A)-^OVA-)'}
by Stokes's theorem.

The double
v(r

integral

is

taken over any surface bounded

tensor representing

by the
element

circuit,

and d&

is

the.

antisymmetrical
is

the

of surface.

The

factor \

needed because each component

as

of surface

appears twice in the summation,

e.g.,

dS 12 and

d$

21
.

The integrand
a*/ xx
I

is

equal to

r
rv
-*-

Va

>

x
r\

~F M + x A r
(TO.
J

l/a

m?A!_-f(Ta^ x
r\

r\

~
(2-2).

= A*(JLv n~L
Changing the

r ra")-rra"VA' + r,/VA' by
to p in the first term, tliis

dummy

suffix

from

a.

becomes

Bp^A",
where

*B pw*

Hence

= - J- IV + [A**] =

IV + T^ IV - r" IV
JJ*B pw /*APdS

(3'2)

-|

w
.

(3\3)

In Einstein's theory, even without Weyl's generalisation, the integrand

does not usually vanish and the initial and final values of A/* are unequal.

Of course, the same discrepancy appears

t

if

A^

is

displaced by different routes

this order is

enumerates the points of a continuum in particular order and of coordinates. If the world referred to an arbitrary coordinate-system turns out to be infinitely crinkled, I think this must indicate a disharmony between the coordinate-order and the structural order of the points. The remedy is to change the coordinate enumeration, not by continuous transformation, but by re-sorting the points. Evidently our displacement dx^ will not be the measure of any physical (structural) relation unless some preliminary agreement of order is postulated. The problem of determining the conditions of this preliminary agreement Theory of Time and Space.' If ever the is essentially that solved by A. A. Kobb in his theory comes to deal with essentially discontinuous phenomena (e.g., quanta) the condition (2*4) may have to be reconsidered.

A coordinate-system

unchanged by any transformation

'

"

108

Prof. A. S. Eddington.

A
is

Generalisation of WeyVs

to another point,

and the

difference

shown by

(3*3) to be proportional to
(3*3) applies only to
is

the area

of the circuit,

ceteris

paribus.

The formula

small circuits for which the square of the area can be neglected, because A?

ambiguous except on the boundary

has been defined
as [A**],
;

of the circuit
is of

where the route

of transfer

but the ambiguity of A?

the

same order

of

magnitude
[A**]

and

its
is

surface integral becomes negligible compared with

the circuit

sufficiently small.

Since then the circuit

is

when infinitesimal, we

can omit the integral

sign, obtaining

[A*]

i*B pi ^A'tfS*
*B P </'
//,

r
.

(3*4)
is

Now

[A**] is

the difference of two vectors at the same point, and

there-

fore a vector;

d$ v<T

is

a tensor; hence

is

a tensor.
to

Moreover,

it is

an

in-tensor, since there has been as yet

no reference

any gauge.
contracting,
x
a
crp

We

obtain another in-tensor by setting

#H- pv ^-X

=
o-

cr, i.e.,

*T> XJ

o-

pv(T

d
i=\

-p
-1-

o-

V
""
,

-p
-*-

t-i
-l

o-p
va

vp

r\

crp

p x

trp
era

x vp

(Q-K\ \* u /

These two in-tensors express intrinsic properties of the continuum.

as I

So far
is

can see no others (except products of these)

exist.

The

first

description as complete as possible of the structure of the continuum (the

inter-connection of relations) so far as

it

can be studied by these methods

the second
first.

is

an abbreviated summary

of the information contained in the

We

shall find later that physics has

in the

main contented
;

itself

with

studying the abridged edition of the book of nature

electron structure does
it

only in the problems of

seem necessary

to search out the

more extensive

information contained in the original

4.

In order to introduce the g^ v

defined by

*B P <A we must adopt

a definite, but at present

I

arbitrary, system of gauges.

A'"- is

The length

(interval length)

of a displacement

P
in

= g^Ar-A?,

(4T)

agreement with Einstein's fundamental equation

take g^ v to be a tensor in order to make I 2 an invariant. The tensor g^ v may be selected quite arbitrarily,! and then equation (4\L) shows how a length, Z, is to be assigned to any displacement, dx^, given in coordinateDifferent choices of giXV will give different plans of assigning length in short, different gauge-systems. Length here is still a geometrical
differences.

We

convention rather than a

physical

conception.

So long as we have no

independent means of comparing lengths, no inconsistency can possibly be

t

For convenience, however, we limit the choice

to tensors

symmetrical in

fx

and

v.

Theory of the Electromagnetic and Gravitational Fields.

detected in our scheme.

109

When we come to identify

these geometrical lengths

with physical lengths, which can be compared by experimental methods,

conditions must be introduced which will limit our choice of g^ v%
will

But that
dx 9

come

later; the present analysis applies to

any gauge-system whatever.

displacement

Move

the

displacement

A^ by

parallel

through

Differentiating (4*1)

we have

d(P)

= fe A^A'+^A" ^H+g^A* ^) dx*

= t%^ A^A^g^T^A-A^g^V^A^A-) dxv
rrr
/

by

(2-2)

__.

ffav^

(r/""^-!-

av""

A^A" \CLX)

\OX<t

by interchanging

dummy

suffixes.

In conformity with the usual rule

x
cr/u,)

for lowering suffixes,

we

write

gav
v

-*

<Tfx

>

so that

d (P)

(^ - T^,
of two

- l\ v ,^ A*A" (dxy.
an invariant.
it

(4-2)

But

<i(P),

the difference
is

invariants, is

Hence the
Thus
\*
*J

quantity in the bracket

a tensor.

We
A

denote
x

by

2K FJ<r

-"

xv /Xl>> <T

oyi.jJ'

(TUjfX,'

Similarly
9TC

effiW \vjzz--v p

3#,M

Adding these and subtracting

(4*3),

we have

OX

j/

OXn

OXfj- J

We

write

% = K^o- K^,, K w
Vitr
jcr

,^,

(4*5)
v.

where

S^

is

evidently a tensor symmetrical in

#" T
,

^ and

Then, raising the

suffixes

by multiplying through by

(4*4)

becomes
(4*6)

IV = {^,t}+S m /.
We
see that

T^/

is

a generalisation of

the Christoffel symbol, and might

tensor,
it

appropriately be denoted by *{/jlp,t\.

Though not a
to

possesses the

"in-" property, since

that

it

was introduced prior

any gauge system.

We

note

KF)(r is not an in- tensor, since

d(P) depends on the gauge.

110
It
is

Prof. A, S. Eddington.
here that

A
K
MV)(r

Generalisation of WeyVs

we make an important divergence from WeyTs geometry

by giving
to

his theory is obtained

5.

the special form

/JLV <r

g^^
terms in S^,/ are
for

From (3*2) and (4*6) we can evaluate *B dropped, we obtain immediately the usual
Christoffel tensor

p.

If the

expression

the

Riemann-

B^ [Eeport,t
{fur,

equation (23)].* The additional terms are

{vet,,

g S M</~--

S M /-f

*} S/ +

a p] S M(r

{/nv,

et} S<ra p

{a a, p] S M / + S
The
first six

S Mcr a - So-/ S^*.

terms easily reduce to

where the

final

suffix denotes

covariant differentiation according to

the

ordinary rule,

viz.,

(S^)*
Hence

3 = 5 S^ {fiv, a} S

a(r

P{(TP a}
P

8^+ {av, p} S

a
Mcr
.

*B MW ^
Setting
/o

= BMW / + (S^)
v

-(S

it

/)

+ S w P S^-S^ S MI /\

(51)

cr,

we have

*GMV

= QA + 2^-(S
2/C^

/t

/) a + S

tt

/S^-2^

Al

(5-2)

where we have written

bcrju,^,

(^'3)

and

/fy

is

the covariant derivative of ^.

Again, multiplying by ff v
(5-4)
(
/u,

*G
where

= G + 2^ + x% + S^S^+2/i;a\ V = ~"S Wfl

s 5)

'

(The distinction between

(5*3)

and

(5*5) is that in So-/"

occupies one of

the two symmetrical places, and in S

place,
ic^

"^

it

occupies the third unsymmetrical

and X^ are quite

different vectors.)
of (5*2)

The only term on the right

2*^, which may be written

which

is
-~-

not symmetrical in
Thus,
if

ju,

and

v is

*^)(/^ v + /^ + (# MV = m, + F^, *Gmv

we

write
(5*6)

where

KMV

is

symmetrical and

F^
r /u,v

antisymmetrical,

we have

fcfj.v

ftpft.

the covariant curl being the same as the ordinary curl for
t
'

a.

covariant vector.

Report on the Relativity Theory

of

Gravitation,

Physical Society (Fleetway

Press).

Theory of the Electromagnetic and Gravitational Fields.

Since

Ill

E^ is
and

half the

sum

of

*G> and

*G

vtl ,

and

FM

is

half the difference,

both

RM

F^

Similarly

we

are in-tensors.f can divide # B^ v<rp into two parts

where
is

E is antisymmetrical and F
a.)

symmetrical in
It is easy to

fi

and

p.

(The whole tensor

antisymmetrical in v and

show that

These, however, are not in-tensors, since the g^ v must be introduced in

order to lower the suffix

p.

Physical Theory.
6.

We

now

introduce the natural gauge of the world, which

is

determined

by measures

of space

and time made with material and optical appliances.

is itself

Any

apparatus used to measure the world

part of the world, so that

a natural gauge represents the world as self-gauging.

that the tensor g^ v introduced in (4*1) is not extraneous
tensor already contained in the world-geometry.

This can only

mean

and
is

arbitrary, hut is a

There

only one such tensor

available, viz.,

*GMV

Hence natural length

is

given by

= *GM A*A'\
zzz

(61)

Clearly the antisymmetrical part drops out in this equation, so that

(,

Jew p

.A.

Accordingly we must take

Xg^

IV,

(&2)

introducing a universal constant, X, for convenience, in order to remain free

to use the centimetre instead of the natural unit of length.

The

difference

between

EA

and

G^

in (5*2) consists of terms

formed out

of the tensor

KM /. We
EM

shall find later that this tensor represents electroeffects, so

magnetic and electronic

that the

more nearly does

approximate to

GM

empty the space the Hence, in an empty region, the

(6*4)
(or,

more

"

JJ

gauging-equation (6*2) becomes

G> =
This
is

Xg^.

De

Sitter's

equation for a spherical

is,

rather, a quasi-spherical)
of the

world.
If

The gauging-equation

in fact,

an alias

law

of gravitation.

approaches zero so that the dimensions of the world become infinitely

great

compared with our unit

of

measure,

we have

Einstein's

original

t It may be noticed that F r(r = ^*B pv <r p a result easily obtained from (5'1), remembering that Bp V <r p 0. It follows fairly easily that the proportionate change of volume of a four-dimensional element carried by parallel displacement round a small circuit
,

is

F v <rdS var

112

Prof. A. S. Eddington.

Generalisation of

WeyVs

gravitational equations GrM

0.

In later work he has used the more general

50).f
is

form (64), giving a naturally curved world (' Keport/ The form of the world given by these equations
because
just as
world.
it
it

only quasi-spherieal,

requires twenty conditions to determine a truly spherical world,

requires twenty conditions

(B^p

0) to determine a truly flat

But a world conditioned by

(6"4) will

have the more important

form, not from any innate

spherical properties.

It has this symmetrical

tendency, but because our natural gauge of length at different points and in
different directions is adapted to different places,

make

it

so.

electron to be spherical.

we expect them to be of the But the equilibrium shape and

accidental
variations
itself

we have two electrons at same size we also expect an

If
;

size of

an electron
it
is

must be determined by something

possibly
history).

in the world-conditions where

(apart

from

small

persisting

from

its

previous

The electron must measure

against something in order to

to be
itself

find out

how

large

it

ought to be

and in empty space there seems

it

nothing but the radii of curvature of the world for

against.

to

measure

Similarly,
size

any material object

of specified constitution

must have

determined
far as the

and shape in
of

relation to the radii of the world.

And

in so

measure

length rests on such bodies, the form of the world

is

expressed in this measure

It is only in

bound

to be quasi-spherical.
(6*4) is valid.
;

empty space that the approximate equation

Electrons are recognised to be small regions of intense local curvature

is

this

provided for in the accurate gauging equation (6*2), since the difference

between

E^

and G-^ may in certain circumstances become

large.

Intense

curvature (or mass) must always be associated with large values of

S^ and
",

we

shall find later that large electric fields are associated

with large values

of

this tensor.

Natural space and time must obey Eiemannian geometry.

It

would be

easy to devise fanciful geometries in which, for example, ds^ was a quartic
function of the coordinate-differences.

But there

is

only one in-invariant of

A**,
viz.,

a simple character! associated with a displacement

*G

flv

AA
fA

v
f

so

that

(except

for

far-fetched

combinations)

no quartic

function

of
MI/(r

the
p

coordinate-differences

could have the necessary property.

counteract the upper
physical
of
p.

Even *B
to

is

no

use, for
7.

we cannot

Having thus

identified

space-time,

we proceed
of

identify

"things."
identified
;

There are three sorts

physical
inquire

attributes

things to

be

and our plan

is

first

to

what experimental properties

total

t The difference between Einstein's and Be Sitter's theories depends on the amount of matter in the world. Both adopt (6'4) as the condition for empty space.
J

The next simplest

in-invariant

is

*B^<r p *Bx Tp

"

Am A A x A r
v

Theory of the Electromagnetic and Gravitational Fields.

they possess, and then
to

113
these

seek

geometrical

tensor

possessing

properties by virtue of mathematical identities.

The tensors need not now

is

be in-tensors, because our description of the experimental properties pre-

supposes the use of the natural gauge-system; and there

no reason

to

think
(1)

it

would hold good

for other

gauge-systems

The energy-tensor

T/

comprises the energy,

momentum, and

stress in

a region.

These must satisfy the condition

is

of conservation of

energy and

momentum, which

expressed by

T/

0.
('

(71)
Report,' 39)
(7'2)

This enables us to make the identification

-8ttT/
which
satisfies
is

- G/-i/(G-2\),
Here A might be any constant

(7'1) identically.

so far as

conservation

concerned.

We
If

consider, however, that the energy-tensor

should be something which vanishes altogether in empty space (containing

no electromagnetic

fields).

T/ =

0,

we

have, by contracting (7*2),

G = 4\
so that

= G/ g/X,

or lowering the suffix

agreeing with (6*4)

if

is

the same constant.

Thus, from the conservation of

energy and momentum, we deduce the law

(2)

of gravitation

once more.

The electromagnetic force-tensor

flt

F^

must

satisfy

two

of

Maxwell's

equations

V+

+ ^pj =

(7

3)

This will be an identity

to
is

if

FM

is

the curl of a vector.

This leads naturally

an identification with the in-tensor already called F^, which we have seen
the curl of a vector # M Hence #M will be the electromagnetic potential. The electric charge-ancl-current vector J^ must satisfy the condition of (3)
.

conservation of electric charge,

viz.,

its

divergence j%, vanishes.

This will

be an identity
variant tensor.

if

J^

is

itself

the divergence of any antisymmetrical contra-

Accordingly, we. identify

J*
Additional confirmation
is
('

= F*v

(7-4)

given by the fact that (7*4) represents Maxwell's

Eeport,' 45).
to

two remaining equations

It

would

in

any case be reasonable

make
that

these identifications

the

simplest possible

checking

them by comparison with observation

is

later on.

But

actually very little choice

possible

is

to say,

it

is

not easy to

114
find

Prof. A. S. Eddington.

Generalisation of
the

WeyVs
properties
as
1

any other geometrical tensors

Further,

having
expect
the

necessary

identities.
test,

we can

scarcely
all

since (with

one exception*)*)

much further experimental known laws of mechanics and

electromagnetism are already

satisfied.

regards alternatives, some other tensor functions of g^ v are known which satisfy (71) identically: but these do not agree with the condition for

As

empty

space, G-^

= \g^

v.

They are

extremely complicated

expressions,

involving derivatives of the fourth or higher orders.

An

alternative identifi-

cation of the electromagnetic force with the curl of X^ (instead of the curl
of Kn)

seems possible, though rather perverse

in- tensor

seeing that
be

it

would leave the

its

fundamental
8.

F^

apparently doing nothing to justify

(6*2)
can.

existence.

The gauging-equation
The
and

also

demonstrated from optical

theory.

light pulse diverging

from a point occupies a conical iocus in

satisfy

four dimensions,

must

therefore

an equation of

the form

a^dx^dxy
nates,

0.

This unique locus exists independently of gauge or coordiit.

and therefore there must be an in-invariant equation defining

Clearly a^ v
ctfi V

must be an
or

in-tensor,

which leaves as the only

possibility

="

'

vx^j/,

~K^ v dx^dx v

= =

0.

(8*1)

But

in Einstein's theory the light-cone is

g^v^dxy
so that Einstein's

0,

(8*2)

gauge

is

given by

Our gauging-equation
gated,
i.e.,

is

therefore certainly

true

wherever light

is

propais

everywhere except inside the electron.

?

Who

shall say

what
If

the

ordinary gauge inside the electron

It should, however, be noticed that this

optical treatment does not prove that

is

a universal constant.
still agree.

it

were a

varying function of position, (8*1) and (8*2) would

of

The constancy

\ depends on
If a

the arguments previously given.

is

displacement A**
circuit, its

taken by parallel displacement round a small

of length is

complete

change

81*

=8 A^Ap) - b (A^) - 2 A^SA* - A *B ^APrfS w by (3-4) = (R ^ + FpWM &*A?d&* by (5 -8) - F ^A^A^S( fJlxp
/t

pi

(8-3)

since

Bpv07X
t

is

antisymmetrical in
of

/lc

and

p.

The law

ponderomotive force

of

an electromagnetic

field, see 9.

Theory of the Electromagnetic and Gravitational Fields.

In Weyl's theory

115

F pvovx

is

of the form g PfJ,Y v(r so that (8*3) reduces to

8P

= F w tfS w

2
,

(8-4)

so that, although a vector in general alters length

on describing a
comparison

circuit,

zero-length
is

is

unaltered on Weyl's theory.

Accordingly, for him zero-length

of
;

unique and involves no specification of a route

theory zero-length
(8*3).

whereas in

my

may

not remain zero on parallel displacement accord-

ing to

This was for a long while

my

chief obstacle in generalising

his theory, since the

abandonment

of the uniqueness of zero-length

might
see

well seem to leave no unique track for the propagation of light.

that the difficulty
is

But we

surmounted

the unique light-cone

;

is

given by the only

it

simple in-invariant equation that can be formed

corresponds to zero-length
is

and the condition that

one of the factors which determine our natural

gauge- system.
9.

The ponderomotive
is

force of

an electromagnetic

field

on matter contain(91)

ing electric charges

fc^F^J",
f

Keport,' equation (46*1)).

This

is

the tensor-generalisation of the elementary

law

mg =
have to show that
(9*1)

X,

(9:2)

which follows immediately from the usual

can also be deduced

definition of electric force.

We

when

the electric force

is

defined

by the method

of identification in 7.

It will be sufficient to consider

an isolated electron.

If

we

consider the

experimental verification of (9*1) or (9*2) by the deflection of an electron

travelling in an electromagnetic field, in (9*1)
is

it is

clear that the field

is

FM

referred to

the applied external

field

field,

and no account

taken of any disturb-

ance of this

caused by the accelerated electron

itself.

We

denote this

external field by

M .

Also, for a particle,

[_ds J

ds

ds
(9*3)

=
where

mA'A'S,,
29).

A =
/x

dx^/ds

('

Keport/

If in (9*3)

we put

h^

0,

we

obtain the

usual equations for the track of an undisturbed particle,

i.e.,

a geodesic, which

can be deduced directly from the conservation

of

energy and momentum.

The equation

to be

proved

is'

thus
(9-4)

mA*AM = -F^'J".

116

Prof. A. S. Eddington.
it is

Generalisation of

WeyVs

In the ordinary text-books

proved that the potentials due to a slowly-

moving elementary charge de

are

p n
? ?

TT cb
'

d l dx

rW

dV
dt'

dz
di

\
J

These expressions are deduced solely from Maxwell's equations

(7*4),

(7'3)

and

and we need not here trouble about retarded values, since our applicato the case

tion

is

when

is

very small.

The general vector-equations are

evidently
h

= * A^,
r
is

(9'5)

where

dfe/r is

treated as an invariant, that

to

say,

its

value
is

is

to

be

calculated for Galilean axes with respect to which the electron

at rest.

momentarily

Hence the

electric force due to the electron is

ic

K =

(A

de
">

since all parts of the electron have the

same velocity
in

A**.

Now, suppose that the

the electron.

electron

moves
of

such a

way

that

its

own

field

on

the average just neutralises the external

field

E^/

in the region occupied

by

The average value

E^

for all the

elements of charge con-

stituting the electron will be given

by
deide 2

x average

F M = (A^ A^)

=
where 1/a
electron.
is

(A flv

vtl

)ei /a,

an average value

of

1/r taken for every pair of points in the

if

Also J^
is

= eA^.

Hence, multiplying by A", we shall have,

the

external field

neutralised,

IV J" = - A" (A - A m
IXV

) e

/a.

But
since

AA
v

Vf,

= AvW =
=

i (A'A,)^

0,

A?A V

Ou V --^

as

ras

1-

Hence
which agrees with
(9*4)
if

FMI /J"

= -A

A v e2 /a,
is

(9'6)

the mass of the electron

m=
It is

eP/a.
is

(9'7)

now

evident that (9*1) implies that an electron

field of

a structure which

cannot exist in a resultant

electromagnetic force.

Further than this

It

we
is

cannot expect to go

without a detailed analysis of electron structure.

a question of the adjustment of the electron to the world-conditions in

Theory of the Electromagnetic and Gravitational Fields.

which
it

117
adapt-

finds itself.

The electron evidently has a

;

certain

amount

of

ability to different world-conditions

and can always

of the

resist differential electro-

magnetic

force, just as a static electron

can resist the radial electrostatic force

laws of electron structure,,
is

due

to its

own

charge.

Knowing nothing

the existence of an isolated electron at rest in no external field

a " miracle "
;

apparently

our calculation shows that the existence of an electron having

is

the acceleration given by (9'1) in an external electromagnetic field

the

precisely

same miracle viewed from another standpoint.

is

There
tion of

a close resemblance between this

i

method

of

deducing the accelera-

an electron

n an electric field and that of a material body in a

gravitational field by the principle of equivalence.

Neither body can be in

force relative to
it
;

equilibrium

(i.e.,

exist)

if

there

is

any resultant

field of

it

would be out

of

adjustment with the surrounding state of the world, and the

Therefore the state of

hypothesis represents a self-contradictory structure.

motion must be such that no resultant

is
;

field of force exists

where the object

accelerations,

the body and the electron destroy the field by their

own

and

this criterion gives the

motion in each

case.

Both, however, are able in some

way to oppose differential forces by internal stresses. The electric mass m is not immediately a source of
so far as the foregoing discussion can show.

gravitational attraction,

But an electron has gravitational

mass

it is

by hypothesis a place where F^, and therefore

Keference to (5*4) shows that in this case

S^

",

has very large

components.

G- will differ

very

considerably from

(unless there
is

is

an unsuspected cancelling
of

of

terms),

showing that the electron

a region

abnormal world-curvature, and

therefore the source of gravitational effects.

of
electrons, the

From

the constancy of structure

(positive

gravitational

mass

of

pair of electrons

and

negative) will be in a constant ratio to their electric mass.

obligation to prove that the ratio
electrons separately;
is

We

are under no

the same for the positive and negative

is

because there

no observational evidence

for

this.

In 11, however,
gravitational mass
to positive

it

appears likely that the ratio of the electrical to the

;

if so, it may be presumed that the factor applies is 8ttX and negative electrons independently.

From

(9*1) the expression for the electromagnetic energy-tensor is

'

deduced

as explained in

Keporfc/ 46.

Further Developments.
10.

The whole energy-tensor

is

given by

-8ttT/

= MG/-i0/(G-2\)),
a

(101)

and the electromagnetic energy-tensor by

V = -lVrF- + i?/F ^,

(10-2.)

118
(

Prof; A. S. Eddington.
5

Generalisation of

WeyVs
X

Eeport, equations (50*4) and (46*4)).

It is necessary to insert the factor

in

(101) in order

to

make

the dimensions consistent.

;

Even now
but, at
i.e.,

it is

not

certain that the

two expressions are in common units

by changing the unit
identically.

any

rate, their
X.

ratio will not be altered

of length,

changing

Contracting (10*2),

E=

It is thus impossible to build

fields alone,

up

matter or electrons from electromagnetic

energy must be present.
It is convenient to write

This represents the

and some other form of non-Max wellian binding forces.

E^ = G^ + HM
so that

E = G + H,
flv }

(10-3)

from

(5*2)

and

(5*4)
*:.

BU =
Contracting (101)

+ {S a /$p ,*-2K a S H = 2K\ + X\+{8 aM Sy P + 2 Ka k}.

/c Vfl

-(S^<r)

(r

fi

>

(10-4)
(10-5)

>

But

Hence
and similarly

= x(-G + 4X). E = gr K^ = \g^ = 4X. -8ttT = X(E-G) = XH,

-8ttT
tf-*
.

(10-6)

8ttT/

= X(H/-iy/W).
S M /, as opposed
to (101),

(10-7)

We may

regard these formulae as giving the electrical aspect of material

it

energy relating

to the tensor

which shows

its

gravitational aspect.

To

fix

ideas as to the order of magnitude of terms, let us take the centimetre

as unit of length,

and choose approximately Galilean coordinates with a

centi-

metre mesh.

In these units

(the curvature of the world in

.

empty

space)
first

can scarcely be greater than 10~ 50

order, it is easily seen that S M
so
"

Calling this a small quantity of the

is

of the first order, except in electric fields

intense

as

to

alter

the

order of

magnitude

of

the local curvature.

Accordingly, the expressions for

HMV

and

break up into first-order and

second-order terms, the latter being bracketed in (10*4) and (10*5).

Evidently

the formation

of

electrons

of

definite

size

is

associated

with the non-

linearity of these equations, the product terms

electric field is sufficiently intense.

becoming important when the

For an electron at
of

rest K aa vanishes.

Although the physical interpretation

static case.

is

not known,
of

it

seems likely that X aa will also vanish in the

is

Thus the mass

an isolated electron

probably represented entirely by the

product terms in H.
11. It is natural to infer

from (101) and (10*2) that in space containing no

electrons

T/

and E/,

if

reduced to a

common

unit,

must be

equal.
is

But

this

involves an arbitrary localisation of

the energy- tensors which

scarcely

Theory of the Electromagnetic and Gravitational Fields.

justified, since the

119

connection established between them via (9*1) refers to the

integrated expressions.

The
rest.

difficulty of localisation

can be realised by considering an electron at

In this case

its relative

mass
is
;

is

equal to

its

invariant mass.

On

the

ordinary view the relative mass

attributed to the inertia of the electro-

magnetic

field

which surrounds

it

but the
equal,

field
if

contributes nothing to the

is

invariant mass.

The two masses are

not synonymous, yet one

electron.

arbitrarily located in the field

and the other in the

it is

From
found.
since

the formulae already given

possible to express

T/

and

E/ in

terms of the S or

tensor

but no obvious connection between them can be

It is possible that the formulae

may admit
some

of

some

simplification,

the gauging equation must introduce

of

relations

between the

components

",

MI/

which, however, are difficult to follow out.

The following
E/, and
it

discussion seems to arrive at a natural connection between

T/ and

appears to have a fundamental significance.

Consider the invariant

u = f *G^GrV(-^)^>
which, though not strictly an in-invariant,
is

(ii'i)

independent of the value of X,

and

is

accordingly a pure number.f

We

have

*CU*G^
owing
to the

==

(E^ + F^)(E^ + F^)

(11-2)

= EMFE^-F^F^,
symmetry
of

EMV

and antisymmetry

of

FM

Calculate the variation of

U due to small arbitrary variations of the g^

which

vanish at and near the boundary of the region considered.

(These variations

represent alterations of the state of the world expressed by the in- tensor

E^,

which lead
equation.)

to

corresponding variations of the g^ v through the gauging-

We have SCE^BVC-?)] = E*,s [V(-0)]

= \{gS [BTy/i-gK+trS [IW(-s)

but

-4X8|V(-0)],
\y/( -g) (ErSjfo, + E^
by adding,

V) = *V( -9)

(0*'$fc.

+& V) =

0>

so that,

S[E^E"V(-^)] = 2XSiEv/(-g)]-4:\^[x/(-g)] = 2\S{(G-2\) v/(-<7)} + 2\S[IV(-#)]

t

by

(10-3).

character
is

numerical invariants of a simple a distinctive property of a /^-dimensional continuum. The invariant from many points of view the most fundamental of these.
is

As Weyl has pointed

out, the existence of several

VOL.

XCIX.A.

120

Prof. A. S. Eddington.
*/{

Generalisation of

WeyVs

But (G 2\)
field,

g)

is

well known as the " action-density " of the gravitational

is

and

its

variation

reducible tof

rejecting a part

which yields a surface-integral over the boundary, since

Again,

all

the variations vanish there.

S[Fm,PV(-0)]
since the in-tensor

= *^F,Si>VV(-#)]
It is

FMV

is

not affected by the variations here considered.

readily found! that this reduces to

2 EM

s/ (

g) Sg^.
s

Further by (10*5)

SJHv/(-0)dr
The
first-order

B\{S aM

S^ + 2 Ka ^}

/(-g)dr.

terms are omitted, since

by a well-known formula
vanish.

for the divergence of a contravariant vector.

This

on integration gives a surface-integral over the boundary where

all variations

Collecting these results

we have

-JSU = J(87rV--E^)S^
(11-3)

Thus, by considering the variation of what

is

probably the most fundamental

numerical invariant, we have brought together in one formula the whole

energy-tensor and the electro-magnetic energy-tensor

a result which would

little

be

difficult to

accomplish in any other way. There can be

doubt that in
(11*4)

the combination

877-^ E^,
they are expressed in a

common
is

unit, so that the difference

the electronic or non-Maxwellian part of

the whole energy- tensor.

must represent The

remaining term on the right

also concerned
is of

with electronic terms, since

outside the electron the integrand

the second order of small quantities.

It is

Hence SU must be a
not clear whether
analysis
is

variation of " action " associated with electrons.

we

are here approaching a law of Nature or whether our

a device for bringing out hidden identities in the mathematical

;

expressions

but

I incline to the latter view.

'

Raum,

Zeit, Materie,

3rd edition, pp. 205, 253.

J Ibid., p. 197.

Theory of the Electromagnetic and Gravitational Fields,

It

121

may

be tempting to suggest a law of stationary action that

is

SU

This, however,

impossible.

If

we do not

split

up

into

G + H, we

find

-J8U = J(XVmv-E^)S^"
and our proposed law
It
of action

y/(-g)dT,

would be inconsistent with the identity

is

E = 0,

need not surprise us that the combination (11*4)

only met with when

the reduction proceeds in a particular way.

real connection
is

As

have already indicated, the

is

between the integrated energies, and the localisation

probably not such as to give the same connection in each element of volume.
I

think that the casting away of the surface-integral from

S[G\/ (#)] has

to a

redistributed the location so as to display the agreement.

The most important

unit, implied in (11*4).

result is the reduction of

TM and E^

TM

for given

masses or electric

fields is

common known in

ordinary units, whereas E^ v involves the natural units which determine the
in- tensor

F^. Thus, taking X = 10~ 50 as an upper limit, I find that the quantum is 10~ 114 natural units of action. A faint hope that the natural

unit of action involved in this theory might prove to be identical with the

quantum

is

thus disappointed.

It appears also that

fields of quite

EM

and

FM may

be of

roughly equal magnitude in electric

in the laboratory.

ordinary strength produced

12. It is a natural impression that in the light of the

more general theory

The
of the
')

Einstein's

work must be regarded

is

as

only a close approximation.

impression

wrong, for the present paper leads to the conclusion that

Einstein's postulates

and deductions are

exact.
'

The natural geometry

world (in the sense that I have defined in

the geometry of

Space, Time, and Gravitation

is

Eiemann and
have sought
is

Einstein, not "WeyFs generalised geometry or

mine.

What we

not the geometry of actual space and time,

but the geometry

of the world-structure,

which

is

the

common
The
first

basis of space

and time and

the
so-called

things.

have tried to emphasise this distinction by dividing

part
is

paper into Geometrical -and Physical theory.

the

"geometry"
of

of a general

kind of relation complex; for I think

that any conception of structure (as opposed to substance)

into

must be analysable
relations,

a complex

relations

and

relata,

the relata having no structural

of

significance

except as the meeting point

several

and

the

relations
relata.

having no significance except as connecting and

This part of the theory
is

ordering

the

structure.
i.e.,

Our method

is

an essay in assigning measure to such a based 'on an axiom of parallel displacement,

comparability of proximate relations; this seems to be the

minimum
viz.,

degree of comparability which permits of any differentiation of structure.

We

succeed in arriving at a quantitative measure of this structure,

*GMW

122

WeyVs Theory of Electromagnetic and Gravitational

parts,

Fields.

which breaks up into symmetrical and antisymmetrical

the latter being necessarily the curl of a vector.

g^ and F^,

It is at this bifurcation of

the theory that Einstein begins; he has recognised these two tensors from
their

physical manifestations,

and he expresses no view

and mechanics; from
so that his

as

to

what

lies

behind them.

From

g^ v

he builds up interval-length, and hence actual

space, time, geometry, gravitation,

F^

he builds the
is

Maxwellian theory

of electricity

and magnetism.
;

His interval

absolute,
the

for it is our in-invariant 'Rflv dx tl dx v

work stands unaffected by

ambiguity of length-comparisons in the initial stages of our theory.

material he employs
.specification.
is

The

furnished by our theory in exact accordance with his

We

have further shown that there

is

more comprehensive measure

is

of

the structure given by

*B/lvo p
.

of

which *G MV

an abbreviated summary.
is

It

seems obvious that even this extended tensor

description of all that there
is

by no means an exhaustive
it (or

in the world

but presumably

something

analogous) marks the limit of physical science, though not necessarily of the
scientific

method.
,

The curious point

is

that, for

present-day physics, the

summary *G/U suffices, and either the extended *B flvaP has no message for us Even in 11 the invariant U or we have failed to interpret the message. belongs to the summary description. It remains to be seen whether there is anything of observational importance omitted in the summary anything

which plays a part in the formation

possibly even of the quantum.

of negative

and positive

electrons,

and