Annals of Mathematics

The Gravitational Equations and the Problem of Motion

Author(s): A. Einstein, L. Infeld and B. Hoffmann
Reviewed work(s):
Source: Annals of Mathematics, Second Series, Vol. 39, No. 1 (Jan., 1938), pp. 65-100
Published by: Annals of Mathematics
Stable URL: [Link] .
Accessed: 27/11/2012 08:47

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
[Link]

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@[Link].

Annals of Mathematics is collaborating with JSTOR to digitize, preserve and extend access to Annals of
Mathematics.

[Link]

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 ANNALS OF MATHEMATICS
Vol. 39, No. 1, January,1938

THE GRAVITATIONAL EQUATIONS AND THE PROBLEM OF MOTION

BY A. EINSTEIN, L. INFELD, AND B. HOFFMANN

(Received June 16, 1937)

Introduction. In this paper we investigatethe fundamentallysimple ques-

tion of the extentto which the relativisticequations of gravitationdetermine
the motionof ponderablebodies.
Previous attacks on this problem'have been based upon gravitationalequa-
tions in which some specificenergy-momentum tensor for matter has been
assumed. Such energy-momentum tensors, however, must be regarded as
purelytemporaryand more or less phenomenologicaldevices for representing
the structureof matter,and theirentryinto the equations makes it impossible
to determinehow far the results obtained are independentof the particular
assumptionmade concerningthe constitutionof matter.
Actually,the only equations of gravitationwhich followwithoutambiguity
fromthe fundamentalassumptionsof the generaltheoryof relativityare the
equations foremptyspace, and it is importantto knowwhetherthey alone are
capable of determiningthe motionof bodies. The answerto thisquestion is
not at all obvious. It is possible to findexamples in classical physicsleading
to eitheranswer,yes or no. For instance,in the ordinaryMaxwell equations
foremptyspace, in whichelectricalparticlesare regardedas point singularities
of the field,the motion of these singularitiesis not determinedby the linear
fieldequations. On the otherhand, the well-knowntheoryof Helmholtzon the
motionof vorticesin a non-viscousfluidgives an instancewherethe motionof
line singularitiesis actually determinedby partial differential equations alone,
whichare therenon-linear.
We shall show in this paper that the gravitationalequations foremptyspace
are in fact sufficient to determinethe motion of matter representedas point
singularitiesof the field. The gravitational equations are non-linear,and,
because of the necessaryfreedomof choice of the coordinatesystem,are such
that fourdifferential relationsexist betweenthem so that they forman over-
determinedsystem equations. The overdetermination
of is responsibleforthe
of
existence equations of motion,and the non linear character forthe existence
of termsexpressingthe interactionof movingbodies.
Two essentialsteps lead to the determinationof the motion.

1 Droste, Ac. van Wet. Amsterdam19, 447 (1916). De Sitter,MonthlyNotices of the

R. A. S. 67, 155 (1916). Mathisson,Zeits. f. Physik,67, 270, 826 (1931), 69, 389 (1931).
Levi-Civita,Am. [Link]., lix, 3, 225 (1937).
65

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
66 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

(1) By means of a new methodof approximation,speciallysuited to the treat-

ment of quasi-stationaryfields, the gravitational field due to moving
particlesis determined.
(2) It is shown that for two-dimensionalspatial surfaces containingsingu-
larities certain surfaceintegral conditionsare valid which determinethe
motion.
In the second part of this paper we actually calculate the firsttwo non-trivial
stages of the approximation. In the firstof these the equations of motion
take the Newtonianform. In the second the equations of motion,which we
calculateonlyforthecase oftwomassiveparticles,take a morecomplicatedform
but do not involve thirdor higherderivativeswith respectto the time.
The method is, in principle,applicable for any order of approximation,the
problemreducingto specificintegrationsat each stage, but we have not proved
that highertime derivativesthan the second will not ultimatelyoccur in the
equations of motion.
In the determinationof the fieldand the equations of motionnon-Galilean
values at infinityand singularitiesof the type of dipoles, quadrupoles, and
higherpoles, must be excludedfromthe fieldin orderthat the solutionshall be
unique.
It is of significancethat our equations of motiondo not restrictthe motion
of the singularitiesmore stronglythan the Newtonianequations, but this may
be due to our simplifying assumptionthat matteris representedby singularities,
and it is possible that it would not be the case if we could representmatterin
terms of a field theoryfromwhich singularitieswere excluded. The repre-
sentationof matterby means of singularitiesdoes not enable the fieldequa-
tionsto fixthe sign of mass so that,so faras the presenttheoryis concerned,it
is only by conventionthat the interactionbetween two bodies is always an
attractionand not a repulsion. A possible clue as to why the mass must be
positive can be expected only froma theorywhich gives a representationof
matterfreefromsingularities.2
Our methodcan be applied to the case when the Maxwell energy-momentum
tensoris includedin the fieldequations and, as is shownin part II, it leads to a
derivationof the Lorentzforce.
In the Maxwell-Lorentzelectrodynamics,as also in the earlier approxima-
tionmethodforthesolutionofthe gravitationalequations,the problemofdeter-
miningthe fielddue to movingbodies is solved throughthe integrationof wave
equations by retardedpotentials. The sign of the flowof time thereplays a
decisive role since, in a certain sense, the fieldis expanded in termsof only
thosewaves whichproceedtowardsinfinity. In our theory,however,the equa-
tions to be solved at each stage of the approximationare not wave equations
but merelyspatial potential equations. Since such equations as those of the
gravitationaland of the electromagneticfield are actually invariantunder a

2 Einstein and Rosen, Phys. Rev. vol. 48, 73 (1935).

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 67

reversalof the sign of time,it would seem that the methodpresentedhere,is

the natural one for theirsolution. Our method,in which the time direction
is not distinguished,correspondsto the introductionof standingwaves in the
wave equation and cannot lead to the conclusionthat in the circularmotionof
two point masses energyis radiated to infinityin the formof waves.
I. GENERAL THEORY

1. Field Equations and CoordinateConditions. Since it is an essentialpart

of the workto make a separationbetweenspace and timewe shall, throughout
this paper, use the conventionthat Latin indices take on only the spatial
values 1, 2, 3 whileGreekindicesreferto both space and time,runningover the
values 0, 1, '2, 3.
As explainedin the introduction,we discuss only the gravitationalequations
foremptyspace, treatingthe sourcesof the fieldas singularities. If we denote
the ordinaryderivativeof a quantityby means of a line followedby the appro-
priatesuffix,as

(1, 1) ag~s.>__ ; axe axp

we may writethe fieldequations in the form

(1,2) RAY ? {z} + {x}{,:} - = 0.

Let the symbols77,.,aimbe definedby

+1 0 0 0o
o -1 0 0
(1,3)=0 = 7=
(1,3) 0 0 -1 0
,o 0 0 -1

so that theyrepresentthe metricof emptyspace-time. Then ifwe introduce

the quantities ha,, hPVby the relations
(1,4) gpv= ny;,? hghS = ?? M+hW,

the h, and h"' will representthe deviation of space-timefromthe flat case.

The hMWcan be calculated as functionsof the hy,by means of the relations
(1 ,5) gJiP
gWa
In generalthe ha,will be small relativeto unity,but we make no assumptions
here concerningtheirorderof magnitude.
By means of (1, 4), (1, 5) we can expressthe componentsof R, as functions
ofthe hy, and forreasonswhichwillbecome clear whenwe come to the method
of approximationused in the presentwork we separate the various termsso

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
68 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

obtainedinto two groupsin the followingmanner. Firstwe separate the terms

linear in the h's fromthose which are quadratic and of higherorder. At this
stage of the separationthe fieldequations are of the form
(1, 6) Roo= I-hools + 2ho8jo8
- h.1oo) + Loo= O0

(1, 7) R0n= 2{-hOnie + ho081s+ hnslo8- hs81no)+ Lln = O0

(1, 8) Rmn = { -hmnlse + hmens + hnlime - h88,mn + hmnloo- hmolno
- hnOjmO+ hoo0mnl + Lmn = O0

wherethe L,",representthe non-linearterms. We now take

(1, 9) fromRoothe termsho00os-
-hasloo
(1, 10) fromRonno terms
(1, 11) and fromRmn the terms -4homon - 'honl0m + 2hmnl00,

and add them to the non-lineargroup. Introducingthe symbolL,, to denote

the non-lineargroup L', togetherwith these added linear members,we may
writethe fieldequations in the separated from
(1, 12) Roo= -4hool,,+ Loo= O0
(1, 13) Ron = -1hon + -(hs 45nshl, + 15nhoo),o
- 1(hoo + h.,)IOn+ jhoeins + Lon = 0,

(1, 14) Rmn = jhmnJe, + 1(hm8 - 15mshl + j5mehoo)jne

1+ (hnm- 15n,,hjj + kbnehoo)lms + Lmn = O0

wherethe L's are given explicitlyby the formulas

Loo = ho0os- Jhs oo- (h"[00, a-])Ix + (h"`[fX,a]) lo

(1, 15) + {O} {,O} {0} {:}'

(1, 16) Lon =- (h" [On,a])I x + (h>"[nX,a]) Io + -

Lmn = -24mhnIOnn-hgnI +
Oim hmnI00 - (h" [mn, a]) IA
(1, 17) + (hm [ l)
In +
{a} {An} {A1}{?}

If we introducethe quantities-y,,definedby

(1, 18) 7y,, = hot, 41 7,phopy

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 69

or, in expanded form,

(1, 19) ' oo= 'ho + 'hi1,
(1, 20) Yon = hOnX

(1, 21) ymn= hmn - 4Zimnhil+ 4bmnhooy

we may writethe fieldequations (1, 12), (1, 13), (1, 14) in the form
(1, 22) Roo = - hools8 + Loo = O0

(1, 23) Ron= -1honls8

+
4'[Link]
+ i(70818 - 'Yooio)in + Lon = 0,
(1, 24) Rmn = j-2hmnis8 + 27mIen
+
-tny-sem
+ Lmn = 0.

Since thereare fouridentitiesbetweenthese fieldequations, we may impose

fourcoordinateconditions,in the formof fournon-tensorialequations involving
the gravitationalpotentials,so as to limitthe arbitrarinessof the solutionsby
limitingthe freedomof choice of the coordinatesystem. It turns out to be
simplest to use coordinate conditions which involve only quantities which
enterthe explicitlywrittenparts of the fieldequations (1, 23), (1, 24). These
equations, in fact,suggestthat we take as our coordinateconditions3
(1, 25) 7081. - 7Yo1o= 0,

(1,26) 'YmaJ = 0.

With these coordinateconditionsthe fieldequations become merely

(1, 27) hoo188= 2Loo,

(1, 28) hOnles= 2Lon

(1, 29) hmnlJs= 2Lmn.

For the furtherargumentit is necessarythat we write these equations in

such a way that the Laplacians of the it's enterinstead of the Laplacians of the
h's. We thereforereplace the above equations by the equivalent equations
(1, 30) 7yo0,8 = 2Ao00

(1, 31) 7Yon,a. = 2A~n

(1, 32) lYmn8a8 = 2AmnX

3 The choice of the coordinateconditionsis, to a large extentarbitrary,and it might

seemrathermorenaturalto use the conditions
1 'Yavii = 0
which are invariantunder a Lorentz [Link], it turns out that the
actual calculationofthefieldis simplerwhenwe use the coordinateconditionsgivenin the
textand it is forthisreasonthat we employit in the generaltheory.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
70 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

where A is related to L exactly as y is to h:

(1, 33) A0O= 'Loo + 'L11,

(1, 34) AO.= LOn X

(1, 35) Amn = Lmn - 24imnLl + 22imnLoo.

These fieldequations, (1, 33), (1, 34), (1, 35) togetherwiththe coordinatecon-
ditions (1, 25), (1, 26) will formthe basis of our furtherconsiderations.

2. Fundamental Integral Properties of the Field. Let us consider three

functionsAn; (n = 1, 2, 3). They need not be tensors. From these functions
we may build the threefurtherfunctions

(2, 1) (AnIs - A,9101.

whichcan be explicitlywrittenas

) (A112- A211)12- (A311- All 3)131, {(A213- A312)13- (A112-A21I)II},

({(A311 - A113)11 - (A213 -A312)121

These threefunctionsthus constitutethe curl of the threefunctions

(2, 3) (A213 - A312), (A311- A113), (A112- A211).

Considerany surfaceS whichdoes not pass throughsingularitiesof the field.

Since (2, 1) is the curlof (2, 3), it followsfromStokes' theoremthat the integral
of the "normal"4componentof (2, 1) over S is equal to the line integralof the
tangentialcomponentof (2, 3) taken around the rim of S. If S is a closed
surfaceits rim is of zero length so that the latter integralwill vanish. We
thereforehave the theoremthat, if S is any closed surfacewhich does not
pass throughsingularitiesof the field,then

(2, 4) f (Anj1 - A8n)18 COS (nuN) dS = 0,

where(n. N) denotesthe "angle" betweenthe directionof Xn and the "normal"

to S, and the summationconventionapplies to the n. This theoremis valid
whetherS enclosessingularitiesor not, and we shall now apply it to the present
problem.

4Words like normal, and so on are usedherein a purelyconventional

angle,sphere,
senseto designate functions
thecorresponding ofthecoordinates xmandequationswhich
are impliedby thesetermsin [Link] argument ofthisparagraphis
independent of any particularmetric,and we use the Euclideannomenclaturemerely
becauseit is apt andconvenient.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 71

Fromthe coordinateconditions(1, 25), (1, 26) and the fieldequations (1, 31),
(1, 32) we have
(2, 5) ('YOnjs - 'YOsJn)|J= 2AOn - 'Yoo|On

(2, 6) (YmnIJ - 'Yman)J= 2Amn

We see that the left-handsides of (2, 5), (2, 6) give fourquantitiesof the form
(2, 1), one comingfrom(2, 5) and threefrom(2, 6) form = 1, 2, 3. It follows
from(2, 4) that, if S is a surfacewhichdoes not pass throughsingularitiesof
the field,

(2, 7) f ('YOIOn- 2A0n) COs (n-N) dS = 0,

(2,8) f 2Amncos (n N) dS = 0.

From (2, 5), (2, 6) we see that, in those regionswhere there are no singu-
larities,

(2, 9) OI - 2AOn)in = O0
(YOO

(2, 10) (2Amn){n = 0.

ThereforeGauss' theoremshows that if we take two closed surfacesS, S' such

that no singularitylies on or betweenS and S', the integrals over S and S'
give the same result. But the validity of the integralconditionsforsurfaces
whichenclose singularities,or more generally,whichenclose regionswherethe
fieldequationsforemptyspace are not fulfilled, can onlybe shownby means of
Stokes' theorem.
We are treatingmatteras a singularityin the field. Let us assume thereare
p bodies, each representedby a point singularity. The coordinatesof each
such singularitywill be functionsof the time alone. Since (2, 7), (2, 8) are
valid forany S providedonly that it does not pass througha singularity,we
may choose p such surfaces,each enclosingonly one of the p singularities,and
thus obtain 4p distinctintegralconditions. Each of these, being now inde-
pendentof the shape of its S, will give a relationbetweenthe coordinatesof the
singularitiesand theirtime derivatives,and we shall see later that the integral
conditionsgive, in fact, the equationsof motionof the singularities. These
equations are derivedhere fromthe fieldequations and coordinateconditions
alone withoutany extraneousassumption.
If, instead of integratingaround one singularityat a time,we integrateover
a surfacewhich containsall the singularities,we obtain the laws of conserva-
tion of energyand linearmomentumforthe whole system. These laws are, of
course,merelyconsequencesof the laws of motionforthe individualparticles
but owingto many cancellationstheytake a comparativelysimpleform.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
72 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

3. The Method of Approximation. The method of approximationwhich

has been used up to now in the theoryof relativityis as follows. We consider
that in the equation
(3 1) g, = 7, + h,
the h,, depend continuouslyon a positiveparameterX in such a way that they
vanishforX = 0, so that forX = 0 space-timebecomes Galilean. We assume,
therefore, can be expanded in a powerseries"in X:
that the h,zv
00
(3,2) h,= hX
1=1 1

This expansion is introducedin the field equations which are then grouped
powersof X,takingthe form
accordingto the different

(3,3) 0 =_RR,.=
1=1 1

In orderthat a set of hp,dependingon the parameterX shall exist as a solution

of the fieldequations it is necessarythat each of the equations
(3,4) R, =O

shall be satisfied. The best knownexample of thismethodis its applicationto

the firstapproximation.
We shall now show why this methodof approximationis unsuitableforthe
treatmentof quasi-stationaryfields. If we introducean energytensorforthe
matterwhich produces the fieldwe obtain for the firstapproximation,using
imaginarytime,the well-knownequations
(3, 5) a
yuvI = -2T,
wherethe coordinatesystemis determinedby the equations
(3, 6) 'Y oro = 0.

In the simplestcase of incoherentmatter(dust) producingthe fieldwe have

T dt" dt'
(3, 7) ToV p ds ds

where da~ds are the componentsof the velocity measured in terms of the
propertime s. If we are concernedwith a quasi-static situation,de0/dsis of
the orderof magnitudeof unity while the dt'/ds are relativelysmall. Thus
in such a case we shall have
(3,8) 1Tooi >?I To.n
I A I TmnI,

5 In XIthe 1 will always be an exponent,not a contravariantindex!

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 73

and fromthe equations (3, 5) we must have correspondingly

(3, 9) zYoo>>
I TYo. I>> I -Y.n I .
The usual method of approximationdoes not take this into account since it
treats all the Y's as of the same order of magnitudealthough,in the quasi-
is verymuch largerthan the othercomponentsof-y,,. A really
static case, -Yoo
good methodof approximationforthe quasi-staticcase should make essential
use of the relations(3, 9).
We are led to our presentmethod of approximationmost simply by con-
sideringthe problem of constructinga method of approximationwhich is
suitableforthe solutionof the approximatefieldequations (3, 5) forthe quasi-
static case. It turnsout that the method of approximationto which we are
led in this way is also suitable for the solution of the rigorousgravitational
-equationseven when we are not dealing with quasi-static cases.
The firststep is to give an explicit expressionfor the fact that the time
derivativeof a fieldquantityis small relativeto the quantity itselfand to its
spatial derivatives. To do this we introducean auxiliarytime coordinate
(3, 10) r = Xx0
and assume that everyfieldquantityis a functionof (T,X Y,X22X3) ratherthan
of (X0,xX1 X2 X3). If 50is such a quantitywe now assume that so,PI,. and aco/ar
are of the same orderof magnitude,so that V1ois of the orderof Xsp.
From thiswe concludethat if Tooin (3, 7) is of the orderof magnitudeof X
then To. will be of the orderof X'+l and Tmnof the orderof X,+2
Further,it followsfromwell-knownconsiderationsconcerningthe firstap-
proximation(the conservationof energyfor the motion of a point) that yoo,
whichis the potentialenergyof a unit mass, is of the same orderof magnitude
as the square of the velocityand is thus, in our presentnotation,of the order
of X2. Hence we have the followingordersof magnitudeforthe y's:
(3, 11) z X
;-
0oo on 3 uYmn''
'-'X
If we expand the it's as power series in X we must thereforetake the lowest
powers of the expansionsto be of the ordersgiven in (3, 11). The fact that
only second derivativesof the iy'swith respectto the time enterthe equation
(3, 5) shows that the powers of X in successivetermsof the expansionsof the
'Y'smay differby two.6 We are thus led to the simple assumptionthat
27
-Y = Y0 + X -Yoo + X -oo +***X
2 4 6

(3, 12) -Yon= X3-on + X5'Yon +

3 5

'Ymn = XA
'mn + X 'Ymn +
4 6

6 The omissionof termswithX1+1 'yn and withX21

in zyoo, in 'Yom is possibleand natural,
but logicallynot strictlynecessary. The addition of the omittedtermsof (3, 12) could
be made in such a way thatit would correspondto an introductionof a retardedpotential
(outgoingwave). Such a procedurewould however,be artificialthoughit would not
influencethe equationsof motionderivedin II, as will be shownelsewhere.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
74 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

We cannotdiscussthe questionof convergencein general,but it is of interest

to show that the new method of approximationcan give convergentresults
even wherethis would not at firstbe expected. We considerthe case of the
one-dimensionalwave equation in its simplestform
(3, 13) f-X-ft=t ?-

If, in accordance with the main idea of the new methodof approximation,we
write
f + X2f + X4f +
0 2 4

(3, 14) X = f + X2f + X4f +

0 2 4

tt = 7f = X fTT +
-I f- + XfT7 +*-
0 2 4

we obtain from(3, 13) the successiveequations

(3, 15a) f1X = 0,
0

(3, 15b) f~l - fTT = 0,

2 0

(3, 15c) f - f77 = 0,

4 2

From these equations we can findthe general solution of the wave equation
(3, 13) expressedas a powerseriesin X. For simplicitywe shall consideronly
the case of a sinusoidalwave so that,out of the totalityof solutionsof (3, 15a),
(3,16) f = A(r) + xB(T),
0

we choose the particularsolution'

(3, 17a) f = sinr
0

and at each subsequentstage of the procedurewe ignoreall arbitraryfunctions

which may enter. From (3, 15b), (3, 15c), . . ., we thus find
2
(3,17b) f = - -sinT,
2 2!
4
X
(3, 17c) f = - sin T,
4 4

so that the solutiontakes the form

f= sin {1 - + (Z .._
- } COS(XX)sinT.

The inclusionofthesolutionf = x sin r also leads to sinusoidalwaves,as is easilyseen.

0

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL
EQUATIONS AND PROBLEM OF MOTION 75

On replacingr by Xtwe have

(3, 18) f = cos (Xx) sin (Xt)
whichis an exact solutionof (3, 13).

4. ExpansionPropertiesof Field Quantities. We shall show in this section

that thereis a simplegeneralrule concerningthe typesof expansionwhichwill
occur when we treat the gravitationalequations by the present method of
approximation. This rule is that
Any componenthavingan odd numberof zero suffixeswill have only odd
powersof X in its expansion,whileany componenthavingan evennumberof
suchsuffixes will involveonlyevenpowersof X in its expansion.
The fundamentalequations (3, 12) show that the -y. conformto this rule.
The relations(1, 19), (1, 20), (1, 21) between y,,and h,, have inverserelations
of preciselythe same formwith y and h interchanged,as
(4 1) hoo= yoo+ j'z ,
(4, 2) hon= 'YOn,
(4, 3) hmn= Hymn
- 26mn-11 + 26mn-Too,

and from(3, 12) it followsthat the expansionsforthe h's in powersof X are of

the form
hoo=X2hoo + X4hoo+ Xhoo+ **,
2 4 6

(4, 4) hon = X3h0o+ X5h0o+

3 6

hmn = X2hmn + X4hmn + X6hmn + ,

2 4 6

showingthat the h's also conformto the generalrule.

Further,since the i7, triviallyconformbecause 77onvanishes,it followsfrom
(1, 4) that the g,, also conform.
We may writethe relation
(4, 5) 9 =
g,&V S;

in the form
(4, 6) 9gpn9 + o0 =- M

The two groupsof termson the leftdifferby an even numberof zero suffixes
so that,since the 3' triviallyconformto the generalrule,we shall obtain enough
equations at each approximationfor findingthe expansions of the gA if we
assume that the generalrule is valid forthese componentstoo. However,the
g" are uniquelydeterminedin termsof the g,, by (4, 5) so that the expansions
accordingto the generalrule will give the only solutionand extraneouspowers
of X will necessarilyhave zero [Link] the rule is applicable to the
gMand so, also, to the h"t.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
76 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

Let us considernext the Christoffel

symbolsof both kinds. We have

(4, 7) [1,4,a] = 42(g'q1 + gazlv - 9gjsv)

and since the operation "10" introducesa factor X while the operations'
leave the order of magnitudeunchangedit is evident that the fact that the
g, obey the generalrule impliesthat the [,.w, a] do too.
The Christoffelsymbolsof the second kind are definedby

(4, 8) {X} -X[, I) a]

and since wheneverwe have a dummysuffixwe shall have eitherno extra zero
suffixesor two such suffixesenteringany termin the impliedsummation,the
fact that gXJ a] separatelyconformto the generalrule shows that this
and [,gw,
is true also of the
In the course of the above considerationswe have shown that neitherthe
entryof dummysuffixesnor the operations''Im'", "lo" disturbsthe operation
of the generalrule. It followsthat if,by the use of these operationsalone, we
form new quantities from quantities which conformto the rule these new
quantitiesmust also obey the generalrule. This has already been exemplified
by our discussionof the Christoffel symbols,and since all the quantities we
shall have to consider,such as
R(= M Ryp)A etc.
A;,,
are new quantities of this type, we see that all the quantitieswith which we
have to deal will have expansionsin powers of X whose general characteris
summedup in the statementat the head of this section.

5. AlternativeForm of the Equations When SingularitiesAre Absent. In

this section and the next we shall discuss the case where no singularitiesare
presentin the field. This case is, of course,trivialfromthe physicalpoint of
view since it correspondsto the completeabsence ofmatterand, indeed,accord-
ing to our method of approximationleads to the Galilean solution. Despite
this, the discussionof this case will not be withoutvalue, for it will serve to
exhibitthe generalmechanismof the theoryand will forma convenientintro-
duction to the later, more difficultdiscussionnecessarywhen singularitiesare
present.
Let us summarizesome of the resultswe have obtainedso far. The fieldhas
been subjected to the two restrictions
I The GravitationalField Equations, and
II The CoordinateConditions,
fromwhichwe have found
III The SurfaceIntegralConditions.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 77

That is, if we take coordinateconditions

(1, 25) 70010 - YOnin - 0,

(1, 26) 'ynmin ?0

the fieldequations take the form

(1, 30) yools8 = 2Aoo

(1, 31) yont..= 2AOn,

(1, 32) TYmnlss= 2Amn X

and fromthese two groups of equations we obtain the surfaceintegralcon-

ditions
(2, 7) f (yoc1On- 2A0n)cos (n. N) dS = 0,
(2, 8) f 2Amn,Cos (n-N) dS = 0,
and also the results
(2, 9) - 2AOn)ln= 2A0ooO- 2AOn1n= 0,
(YOOlOn
(2, 10) 2Amn,n = 0,
which are essential forthe validity of the surfaceintegralconditionsforarbi-
trarysurfaces.
We shall now show that the followingtwo sets of equations (5, 1), (5, 2) are
equivalentwhen no singularitiesare present.
(5, 1) (5, 2)

(a) yoo1.S = 2A00, (a) yooIs = 2Aoo,

(b) TOnl.. = 2A0,n (b) OnIas = 2Atn

(c) y0010- 'YOnn = 0; (c) Aoo1o - Aon In = 0,

C f(toolon- 2Aon)cos (n-N) dS

(c"t) = 0;
(d) Tmnles = 2Amn, (d) ymnss = 2Amn,

(e) ?'mnin= 0, J(e') Amnjn = 0,

{(e") f 2Amn os (n.N) dS = O.

In (5, 1) we have merelythe fieldequations and coordinateconditionsand we

show essentiallythat the coordinateconditionsmay be replaced by the surface
integralconditions8and the conditions(2, 9) (2, 10). The proofforthe present
8 When
singularitiesare absent (5, 2c'), (5, 2c") and also (5, 2e'), (5, 2e") are equivalent
equations,but we includethemall herein orderto facilitatecomparisonwiththesituation
whicharises whensingularitiesare present.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
78 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

case is trivial. For we have already shown that (5, 1) implies (5, 2) and the
conversefollowsat once fromthe followingconsiderations.
From (5, 2a) (5, 2b) and (5, 2c') we have

('Yo1o - YOnln)las = 2Aoojo- 2Aonln = 0,

and since there are no singularitiesand the -i's must be zero at infinitythis
gives
7oo1o - SYonin = 0

whichis (5, ic). The proofforTYmnis similar.

6. Splittingof the Equations When SingularitiesAre Absent. In the first

section we gave a prescriptionfor separatingthe terms of each of the field
equations into two well-definedgroups. In this section we shall discuss the
splittingof the gravitationalequations according to powers of X and shall
show why just this methodof separationis impliedby our methodof approxi-
mation.
It is necessaryfirstto introducecertainnotations. Considerthe quantity
(6, 1) hmnlOs.

When hmnis expanded in powersof X we write

(6, 2) +
hmn = X2hmn +
X4hmn *+ X2hmn +*
2 4 21

wherethe numbersunderneaththe h's on the rightserve the double purposeof

distinguishingbetween the different functionsh on the rightand of showing
with what power of X each is associated in the expansion.
Now the fundamentalassumptionof our method of approximationrequires
that [Link] a functionof (Xxo,x1, x2, X3) so that
hm
hmn~s8 =
ax
but

hmn -
8hmn
ax0- Whmn
aT

In orderto distinguishbetweenordinarydifferentiationwith respectto (x0,xl,

withrespect (r, X1, X2, x3) we shall denote
x2, X3) and ordinarydifferentiation to
the latterby a comma followedby an appropriatesuffix:

(6, 3) hmnas = a-mn- hmn,s

axa

(6, 4) hmni0= ahmn= X ahmn= X hmn,0.

ax0 ar

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 79

Thus hmn, hmn,sand hmn,oare all of the same order,but [Link] belongsto a power
of Xone higher.
With this conventionwe may writethe expansionof (6, 1) in the form
(6, 5) hmniOs = Xhmn,Os = 3hmnOs + X5hmn0s+ *-+ ?2+lhmn,0s +
2 4 21

Now, however,the numberunderneatheach h on the rightno longerindicates

directlythe power of X with which it is associated. We thereforewrite a 1
underneatheach zero suffixfollowinga comma for every h having a number
underneathso that (6, 5) becomes
(6, 6) hmni . = XhmnOs = X3hmnos+ X5hn,0Os + + x2l+lhmnOs +
2 1 4 1 21 1

Thus now the sum of the numbersunderneatheach h gives the power of X

withwhichit is associatedwhilethe firstofthesenumbersindicatesthe particu-
lar functionh we are considering. This notation is then consistentwith the
naturalnotationfora productof h's.
We considernow what happens when we introducethe power series expan-
sions forthe h's in the equations (1, 27), (1, 28), (1, 29). On equating to zero
the coefficientsof the various powersof X we shall obtain

(6, 7) hmt = 2Lm,

21 21

(6, 8) hOn 8s = 2Lo,,

21+1 21+1

(6, 9) hmns = 2L,,1.n

21 21

The lowesth's are hoo, hon, and h.n , and these will thereforebe the quantities
2 3 2
determinedin the firstapproximation. They correspondto 1 = 1 in the
scheme of (6, 7) (6, 8) (6, 9). Thus at any stage, say 1, the quantities to be
determinedare hoo, h9nX hmn, and the quantitiesalready knownfromthe solu-
21 21+1 21
tions of the previousapproximationsare the h's having lower numbersunder-
neath.
But if we look at the formsof the L's, as given in (1, 15), (1, 16), (1, 17) we
see that at the stage 1 we have eitherquadratic termsor lineartermsinvolving
differentiations with respectto x0. The quadratic termscan only involve h's
of lowerorderthan for1,and the lineartermsmay be writtenas

(6, 10) - 'has,oo

ho8,o8 in Loo,
21-1 1 21-2 11 21

(6, 11) none in Lon,

21+1

(6, 12) - - in Lmn.

11 2homOn
2hmn.00 21-1 1
'hOnOm
21-1l 21
21-2

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
80 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

These are all knownfunctionsfromthe previousapproximations. Thus the

wholeof Loo, Lo., Lmnforgiven I are knownfromthe solutionsof the previous
21 21+1 21
approximations. This is the reason for the particularmethod of separation
of the fieldequations into two parts describedin 1. When the separationis
made in this mannerand the power series expansionsare insertedfor the h's
coefficients
in (1, 27) (1, 28), (1, 29), foreach powerof Xthe corresponding auto-
maticallygroup themselvesinto those quantitieswhichenterforthe firsttime
with the approximationin question and those which are already known, at
least in principle,fromthe previousapproximations. These two groupscorre-
spond exactlyto the leftand righthand sides of (1, 27), (1, 28), (1, 29).
Before we can solve the approximationequations we must also split the
coordinate conditions (1, 25), (1, 26), and the relationsbetween the h's and
y's accordingto powersof X. It turnsout that we may take at each stage

(6 13) y00,88 = 2Aoo, 7Yon,ss

= 2A0on 706,0 - 'on,n = 0;
21 21 21+1 21+1 21 1 21+1

(6, 14) Ilmn,8s = 2Amn X 7Ymn,n = 0,

2 -4 21 21

where the A's are known because of the solutions of the previous approxi-
mations.
We may also split the alternativeequations (5, 2) and use, instead,at each
stage
7yoo,ss = 2Aoo, 70n,ss = 2A0~n Aoo, - AOnn = 0
21 21 21+1 21+1 21 1 21+1

(6, 15)
|1 (00,On- 2A0n ) cos (n *N) dS;
J\21 1 21+1/

(61 16) -Ymn,8s = 2Amn, Amn,n= 01 2Amncos (nN) dS = O.

21 21 21 21

As in the case of the unsplit equations, the surface integral conditionsare

consequences of the others because of the absence of singularities,and the
whole splittingactually presentsno fundamentaldifficulties for this case.

7. The General Theory When SingularitiesAre Present. The existenceof

singularitiesin the fieldintroducescertainfactorswhich make the theoryde-
veloped for the regularcase inadequate. For, althoughthe equations of the
fieldare undefinedat the singularities,their validity in the regularregionis
sufficientto determinethe motionof these singularities. The slightestaltera-
tion in the positionof a singularityamounts to an arbitrarilylarge alteration
fora point near enoughto the singularity,and we are thereforenot permitted
to make use of approximateexpressionsfor the equations of motion in the
developmentof our methodof approximation. This fact leads to a new diffi-
culty,in the approximationmethod,which must be discussed more fully.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 81

Let there be p particlesproducingthe field. We may representtheir posi-

tionsat any timeby means of theirspatial coordinatesi'(r), k = 1, 2, ... , p.
At thesepointsthe fieldwill be singular,but we may encloseeach of the singu-
laritieswithina small surface9 and then the regionexteriorto these p surfaces
will be regular.
Althoughthe equations (5, 1), (5, 2) are undefinedat the singularities,they
have meaningin the regularregionand we shall show that they can still be
regardedas in some sense equivalent. The discussioncan be divided into two
parts,one dealing with the (a), (b) and (c) equations,whichinvolve the suffix
zero, and the otherwith the remainingequations having only spatial suffixes.
We considerthe latter. The essentialstructureof the (d) and (e) equations is
preservedif we omit the suffixm and writeforthe total field
(7, 1) (7, 2)

(d) ynl8s= 2An (d) Ynlaa = 2An

(e)
(e) enln 0,
'Ynin = 0) ~t(e")
{ ~~~~~~(et)
Ann= 0,
f2Ancos (n N) dS = 0.

The proofthat (7, 1) implies(7, 2) has already been givenin essencein 2. To

prove the conversewe firstobtain from(7, 2d)

(7, 3) 'Ynln8s = 2An1n ,

this being valid outside the surfacesenclosingthe singularities. To solve this

we make an analytic continuationof the functionsAn into the interiorsof
these surfacesin such a way that AnIn is everywherezero. This is certainly
possiblebecause of the validityof (7, 2e"). So (7, 3) now becomes
'Ynlnrss = 0

which,being everywherevalid, has the unique solution

Tnjn = 0

which is (7, le).

Thus we have shown that if we make an analytic continuationof A. so that
(7, 2e') is valid everywhere,then (7, 1) and (7, 2) are equivalent outside the
surfacesenclosingthe singularities.
It is clear fromthe proofthat the resultwill hold forany surfacesenclosing
the singularities.
For the (a), (b) and (c) equations a similarproofcan also be given. In this
case it is necessaryto make an analytic continuationof the quantities A00and

I Throughoutthe argumentwe assume that we are dealingwiththe situationat some

definitetime r, allowingtimeto flowagain onlyafterthe argumentis concluded.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
82 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

AO,in such a way that (5, 2c') is valid everywhere,this being possiblebecause
of (5, 2c"). We omitthe details of thispart of the proofthat (5, 1), (5, 2) may
be considered as equivalent even where singularitiesare present and shall
regardthe proofas complete.
To show the difficulty broughtin by the use of our approximationmethod
let us nowconsideronlythe equations (7, 2d), (7, 2e') omittingthe surfaceinte-
gral (7, 2e"). These equations determinethe fieldin each of the approximation
steps if the motions of the singularitiesare prescribed. The motion of the
particlesis then arbitraryas, forexample,in the electrodynamicalproblemand
the fieldis determinedin each of the approximationsteps by the equations

n,88 = 2An
21 21

An,n = 0.
21

The contradictionis evident if we try to add to these equations the surface

conditionsplit accordingto our approximationmethod. We then have the
additional equation
ok
(7, 4) f 2An cos (n N) = 0
J21

where (k) on top of an integralsign means that the surfaceof integrationen-

closes only the [Link] have in (7, 4) an infiniteset of equations
k
containingthe functionst and theirtimederivatives. These equations cannot
k
be satisfiedby the arbitrarilygiven t functionscharacterisingthe motion.
This also shows how the difficulty can be avoided. We have to consider
instead of (5, 1) or (5, 2) a more generalset of conditionsgoverningthe field
which contains those equations as a particularcase. Since it is the surface
integralconditionswhich cause the troublewe remove (5, 2c"), (5, 2e") from
the set (5, 2) and considerthe significanceof what remains.
In makingthis generalisationwe have, of course, gone beyond the gravita-
tional equations to otherswhich contain them as a special case, and we must
now discusswhat changeshave been inducedin (5, 1) by thisgeneralisation.
Since the surfaceintegralsare independentof the surfaces,theirvalues will
be functionsof the time alone throughthe t's and their derivatives. There
is thereforeno loss of generalityif we denote these integralstaken over the p
surfacesenclosingthe various singularitiesby 47rco(-), 47crm(r):

4 f (y0ooI - 2AOn) cos (n .N) dS =co(T),

(7 5) 4kr

+-fkn2Amn
47r COS (nuN) dS = Cm(T).

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 83

Withthis notationwe shall now prove that the followingtwo sets of equations
(7, 6), (7, 7) are equivalent in a certain sense which will be explained in the
courseof the proof:
(7, 6) (7, 7)

(a) 700188 = 2Aoo (a) 7o1,s = 2A00

(b) 'YOn|ja = 2A0nX (b) 7yon|j = 2Aon X

(C) 70010 - 'YOnIn = -Z

k-=1
{co/r,; (c') Aoo1o- AonIn = 0;

(d) 7rnine = 2Amn (d) ymnlja= 2Amn,

(e) EYmnI = cm/} (e') Amnin = 0.

k
Here r is the "distance" fromxn to the k-singularity:

(7, 8) r = [( - x- t

We may introducethe surfacesenclosingthe singularitiesas beforeand these

equationswill certainlyhave meaningoutsidethem. The proofoftheirequiva-
lence can here too be broken up into two parts and we shall only prove the
equivalenceforthe (d) and (e) parts. Omittingthe suffixm as before,we have
(7, 9) (7, 10)

(d) Ynlss = 2An, (d) Ynlaa = 2AnX

(e) = - {k/ }i (e) Ann= 0,

withthe notation

(7, 11) - f
2Ancos(nN) dS = c

We beginby provingthat (7, 10) implies(7, 9) undercertainconditionsof con-

tinuation. It is no longer possible to make an analytic continuationof An
in such a way that (e') is everywheresatisfiedsince this would implythat the
surface integralis necessarilyzero. In fact, from (7, 11) we see, by Gauss'
theorem,that the continuationmust be such that

(7, 12) i 2AnInd = 2An cos (n N) dS = c

It is simplestfor our purposes to make the continuationin such a way that

Anand A,,,,are continuousat the surfaces,and that Aninhas a constantsign
inside each surfaceand satisfies(7, 12).

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
84 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

Such a continuationis possibleforany surfacessurroundingthe singularities

and can be made in such a way that whenthese surfacesshrinkto zero size the
functionAnngoes over to a sum of Dirac 6-functions:

(7, 13) An- 47r ) 6 - ') 6(X2 - 2) 6(XX3). -

k=l1

From (7, 10d) we have now

7n'ns=
2AnnLa

so that

(7, 14) = -)
nfIAn((X)
-Y dA',

wherethe integralis to be taken over the whole domain of Xn and r(x, x') is the
"distance" fromXn to x"':
(7, 15) r(x,x') = [(I' - X")(x - Xa)I.

Because of the validityof (7, 10e') outside the surfaceswe may write(7, 14) as

(7, 16) 1Yn(x) =

47rk=1
Z Jr(x,
k 2A ()v
x')
the integralsbeing taken only over the interiorsof the surfaces. On shrinking
these surfaceswe may regard r(x, x') as constantsover the various domains
of integrationand write

n(x) = _
Ynt t i@r(x)) f 2An1ndv,

and by (7, 12) this is

YnIn - C(Z)Irk

whichis (7, 9e).

We have thereforeshownthat withthe analyticcontinuationused above the
equations (7, 10) implythe equations (7, 9).
To prove the conversewe formfrom(7, 9) the relation

(7, 17) (YnT - 'Ynl)s = 2An+ c2

If we now formthe surfaceintegralsof the "normal" componentsof the two
sides of this equation foreach of the surfacesenclosingthe singularitiesin turn,
the lefthand side will give zero, as explained in 2, and we shall have left

cos (n.N) dS

f
f 2An cos (n.N) dS = -f {c/r}

= -c {1/r} cos (n-N) dS = 47rc(&)

which is (7.11).

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 85

That the validity of (7, 10e') forthe regularregionis containedin (7, 9) is

trivial,and the equivalence of (7, 6d, e) and (7, 7d, e') is thereforeproved.
The proofof the equivalenceforthe remainingequations of (7, 6), (7, 7) con-
tainingthe suffixzero presentsno essentiallynew problemsand will be omitted.
The whole point of our elaborate procedurein writingall the equations of
the fieldin two equivalent formsis now clear since the presentgeneralisation
from(5, 1) to (7, 6) could not be made in a convincingmannerwithoutthe
aid of the parallelismwith (5, 2) and (7, 7).
Owing to the absence of the surfaceintegralconditionsin (7, 7), thereis no
longerany objectionto the applicationof our methodof approximationto the
solution of this set of equations. The X's will cause a splittingof the equa-
tions just as before,except that the surfaceintegralconditionswill be absent.
However,at each stage we may write

f
47r
O On -
2Aon) cos N) dS =CO(r),
(7,(7 18) 21 1 21+1 21+1
18) ('O (n.
1 fk
4TJ
cos (n-N)dS
2Amn = CM(2)
21 21

and with this notationwe have the result,in preciselythe same manneras for
(7, 6), (7, 7), that for each stage of the approximationthe followingsets of
equations (7, 19), (7, 20) are equivalent:

(7, 19) (7, 20)

(a) oo, 8 = 2Aoo, (a) Yoo,8.. = 2Aoo,

21 21 21 21

(b) Yon,88 = 2A0~n (b) ,on,8 = 2AOn

21+1 21+1 21+1 21+1

(C) OO,O- 'Yon,n = -E co(T)/l}; | (c') Aooo- AOnn = 0;

211 21+1 ko1 21+1 21 1 21+1

(d) 7mn,88 = 2AmnX (d) Ymn,88 = 2AmnX

21 21 21 21

(e) 'ymn,n = -E {m(T)/r}, (et) Amn n = 0.

21 k-1 21 21

In the actual solving of the equations it is simplerto work with the sets
(7, 19) ratherthan with (7, 20). At each stage we have to solve equations of
the type y,,,= 2A and in order to make the whole solutionunambiguouswe
must impose the conditionsthat the fieldshall be Galilean at infinityand that
no harmonicfunctionsof highertype than simple poles may be added to the
partial solutions except insofaras their addition is forcedby the coordinate

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
86 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

conditions (7, 19c), (7, 19e). Let us suppose we have been able to solve all
the successive approximations. Then the quantitiescor), Cm(r) are given by
the relations

(7, 21) c ( Z) = E
1=1
C)
21+1

(7, 22) Cm(&) = Z Xcm(T)-

1=1 21

Our solution will not in general be a solution of the gravitationalequations

since (7, 6), (7, 7) are more general than those equations. However, if we
now put

(7, 23) CO(r) = 0, Cm(T) = 0

we impose such conditionson the motions of the singularitiesthat our solu-

tions will indeed become the solutions of the gravitationalequations we are
actually interestedin.
The differentialequations (7, 23) forthe I's are reallyindependentof X since
they must be expressedin terms,not of the auxiliarytime r but of the true
time x0,and when this is done the X's will be necessarilyreabsorbed.
In practice,of course,it is impossibleto carrythe computationbeyond the
firstfew stages. Let us suppose, then, that we have been able to solve the
successive approximationsup to some stage 1 = q. In this case, ifwe put
q '
(7, 24) Z X21+o(1r) =0 0 2 X Cm(T) = 0,
1=1 221+1 1=1 21

we shall obtain solutionsof the gravitationalequations correctto termsof the

order (2q + 1), and the equations (7, 24) will give the approximateequations
of motion up to this order.

8. The Zero Co6rdinateCondition. We show in this section that the solu-

tion of our equations can always be made in such a way that

(8, 1) co(T) = 0,
21+1

thus showingthat the conditions

(7, 21) co(r) = ZX2c?(T)

1=1 21+1

place no restrictionon the motionof the singularities. This resultis of signifi-

to describethe motion
cance because the conditions(7, 22) are alone sufficient
completelyand any furthercondition,if not redundant,would cause an over-
determinationof the motion.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 87

We actually use (8, 1) as normalisationconditions for each stage of the

approximationand they are essentialforthe uniquenessof the solution.
The significantequations for the presentargumentare

(7, 19a) yoo0 2Aoo=

21 21

(7, 19b) YOn,ss =2AOn

21+1 21+1

(7, 19c) Yooo- YOn,n = -E O(T)/ ,

21 1 21+1 k=1 21+1

wherethe A's are knownfromthe solutionsof the previousapproximations,and

we shall suppose that we have a solutionof these equations. If we introduce
k
the quantities F(r) by means of the equation
k k
(8, 2) ro(r) = Co(T)
21 1 21+1

we may write(7, 19c) in the form

21
- 'YOn,n
7Y00,O
1
=r-
21+1 k=1
T-'/~ r 1(irY
21 .0 21 /0

(8, 3) 1 1
P fkj\ (ck kk\
=-ZEk=1 L'/r
21 0
+ (r21 t1 /r n

wheret = d{
dT
From (7, 19a), (7, 19b) we see that yooand 'Yon are arbitraryto the extent of
21 21+1
additive harmonicfunctionsand we may thereforeadd simple poles to them
to formthe new quantities
' r k1
(8, 4) Yoo= 70 + E /r ,
21 21 k-1 21

P
1 ~k k
(8, 5) 'Yon = 'Yon - E Ir t /r}.
21+1 21+1 ke1 21,

These new y's however,whilestill satisfying(7, 19a), (7, 19b) will be such that

(8, 6) a0o.o-
1
= o.
21 21+1

Since the co's now vanish, the surfaceintegralswill also be zero and thus the
zero coordinateconditionwill not affectthe motion. This theoremand our

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
88 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

previousresultsshow us that the equations whichmust formthe basis of the

actual calculationofthefieldand theequationsofmotionofthesingularitiesare:
(8, 7a) = 2Ao00
[Link]
21 21

(8, 7b) 7On,ss = 2A0~n

21+1 21+1

(8, 7c) ooo - 'YOn,n = 0;

21 1 21+1

(8, 7d) 'Ymn,ss = 2Am,,,

21 21

and

(8, 7e) = -Z{E r(T)/l}r

21 k--1 2

f
with

(8, 8) C.m(T) = - 2AmncOs (n-N) dS.

21 47 21

The approximateequations of motionforthe stage 1 = q are given by

(8, 9) E X n
^(T) = 0.
1=1 21

II. APPLICATION OF THE GENERAL THEORY

Note. In the firstpart of this paper we developed the general theoryof a

new methodforsolvingthe equations of gravitationby successive approxima-
tion and for obtaining the equations of motion, in principleto any desired
degreeof accuracy. In the presentpart we deal withthe actual applicationof
this method,carryingthe calculationto such a stage that the main deviation
fromthe Newtonianlaws of motionis determined.
Unfortunately, as the workproceeds,the calculationsbecome moreand more
extensive involving a great amount of technical detail which can have no
intrinsicinterest. To give all these calculationsexplicitlyhere would be quite
impracticableand we are obliged to confineourselvesto stressingthe general
ideas of the work and merely announcingthe actual results. For the con-
venience of anyone who may be interestedin the details of the calculation,
however,the entirecomputationof this part of our paper has been deposited
with the InstituteforAdvanced Study so as to be available forreference.'0

9. The Approximation 1 = 1. The approximation1 = 0 is trivial,leadingto

the Galilean case, and we proceed at once to the next approximationI = 1.

10 c/o SecretaryoftheSchool ofMathematies,InstituteforAdvancedStudy, Princeton,

N. J. (U. S. A.).

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 89

Since the quantities A0O,AOn, and, Amn are all zero and, as explained in 3,
2 3 2
the Ymnare also zero, we have left fromall the equations (8.7a), . . . , (8.7e),
2
merely
(9.1) YOOs88 0,
2

(9.2) 'On,8s
3
= 0

(9.3) Yoo, - YOn,n = 0

2 1 3

The characterof our whole solution will depend essentiallyupon the choice
of the harmonicfunctionwe take as the solution of (9.1). We shall assume
that the particleswe are interestedin have sphericalsymmetryand that the
fieldis Galilean at infinity. In this case the solution of (9.1) is unique since
each singularityin yoomust now, by (9.1) be a simple pole. We therefore
2
have for Yoothe solution
2

(9.4) yoo= 2o, so = E {-2/, k = [( - - )]

2 k=l

k
where the p quantities m are independentof the spatial coordinatesx8,and
can depend at most only on the time.
From (9.2) we see that yonis also a harmonicfunction,and to determineit
3
more exactly we must use the coordinate (9.3). From (9.3), (9.4) we have

YOn,=n = E (-4m/r 0
3 2 1 k-l

4m/( )~n
r - Z r/
k=1 {}n keel

This equation canl be solved without introducingnew singularitiesonly if

m = 0. In otherwords,the quantities A,which actually measure the masses
kk

of the point singularities,are necessarilyconstants. It is now evident that,

under our general restrictingconditions,'Yon is uniquely determined:
3

(9.5) 'on= E 4m/r}

3 k:=1

In all that followswe shall limit our considerationsto the case of only two
particles. This places no essential restrictionon the resultsas far as the end
of 15, theirgeneralisationto p particlesbeing trivial,and it permitsa useful
simplificationof the ratherinconvenientnotationused forthe generalcase.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
90 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

For the case of two particleswe shall write:

(a) -2m/r = t' 2m2 =

(9.6) (b) v =+ x;
1 2
(c) 48 = S
8 =

Our results(9.4), (9.5) may thus be writtenin the form

(a) yoo= 2Vp= 2t + 2x,

(9.7) (b) 2YOn= - -

.

From (1.18) we now also have

(a) hoo= 5 = + x,
2

(9.8) (b) hon= -2 -,n- 2xt

3

(c) hmn= V
b'mn'P + x).
mn(46
2

This shows that the approximation1 = 1 has a Newtonian character but,

owing to the vanishingof cm, places no restrictionon the motion.
2

10. Calculationof the A's for 1 = 2. The firststep in the calculationof the
A's for 1 = 2 is the determinationof the ho,.
Using the methodexplainedin 4, we can calculate the expansionsof the h't
to any desireddegreeof approximation. We find,for1 = 1,
(10.1) = -hoo=
hOO ,
2 2

(10.2) hOn = hOn = YOn,

3 3 3

(10.3) h"n = _ hmn= -bmn(P.

2 2

We nexthave to calculate for1 = 2 the quantities2L, definedin (1.15), (1.16),

(1.17).
In 2 Loothe lineartermsgive
4
(P.00

Of the non-linearterms,only threecan give a contribution. They are

-2{h [0?2
0 r]} = (since ,88 = 0),

-2[00, s][OS, 2 A.a P.a,

-2[00, r][rs, s] =

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 91

where[rs, p] are Christoffel

symbols. Thus
(10.4) 2Loo = (poo + p,(p,8
4

Similar but rathermore tiresomecalculations lead to the furtherresults

(10.5) .
2L0n=p, sqho,
3
- (p,snho8 -3 ,o(p,7
5 3

(10.6) 2Lmn = -hGm,On- h0n,i,0r + Stmn(P,OO- 2(ApO,nn

- (P,m'P,n - 6tmn(P,s(P,.-
4 3 3

Therefore,by (1.30), ..., (1.35), we have

(a) yoo,ss= 2Aoo=-(P,8 , X
4 4

(10.7) (b) 'Yon,ss = 2Acn= (P,s'YOs,n - ',snYOs-3jooVn

(10.7) 5 ~~~5 3 3

(C) 'Ymn,ss = 2Amn = -7Om,On - 'YOnOm + 26mnp,o00 - 25p~A,mn- (P,m(P,n

4 4 3 3

+ Vcmn(P,a P.J-

As explained in 7, 8, these equations (10.7), togetherwith the corresponding

coordinateconditions
(a) 'YOOo- 'YOn,n = 0
4 1 5
(10.8) (22
(b) ymn, n = -
- y
4 /

are the equations which determinethe fieldin the next approximation.

11. The NewtonianEquations of Motion. We must now evaluate the sur-

face integrals
k 1 fk
(11.1) Cm() = 2Amn cOs(n N) dS, k = 1, 2.
4 4r 4

Accordingto the generaltheoryof part I, theseintegralswill be independent

of the particularshapes of the surfacesof integrationsince the divergencesof
theirintegrandsmust vanish on a consequenceof the fieldequations belonging
to the previousapproximation. We shall show here by actual calculationthat
this is the case with the 2Amngiven in (10.7c).
4
Since p and 'YOn are harmonicfunctions,we have
3

2Amn,n = -YOn,Omn + 2vp,oom

4 3

whichis zero, as can easily be seen from(9.3) and (9.7a).

In the actual calculation of the surfaceintegralswe evaluate the separate
contributionsof the differentterms in 2Amn. Since the value of a whole
4

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
92 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

integralis independentof the shape of its surfaceof integration,by takingthis

surfaceto be of finitesize and always a finitedistancefromits singularity,we
see that the whole integralcannot be infinite. Now the individual termsof
2A1mnhave not the propertythat their divergencesvanish, and so we must
4
fixthe surfacesof integrationquite definitelybeforewe begin the calculations.
It is most convenient to take definite,infinitesimally small spheres whose
centersare at the singularities,but in this case infinitiesof the types
Lim const./r', n a positiveinteger,
r -o

can occur in the values of the partial integrals. Since these must cancel, how-
ever,in the finalresult,we may merelyignorethemthroughoutthe calculation
of the surfaceintegrals.
We shall considerthe integraltaken around the firstsingularity. Owing to
the infinitesimalsize of the surfaceof integration,the only termswhich can
give resultsdifferent
fromzero or infinityare those of the orderof (1/r2).
The firsttermin 2Amnis -YOm,On , and, by (8.7b), this may be writtenas
4 3

O=- t4ts + 2)Jn1m - 2X, n~t8 + 2xn4n.

3

The only term we need consideris the second, and so we have

1
f (-omOn) cos (n.N) dS = i! f24tniimcos(n N) dS

(11.2) (4mvm) lf| {(fln - nl)(Xn -n)/r

In a similarmannerwe findthat
-(4m ) +- f { }}

(11.3) - (-0nOnm) J
cos (n.N) dS = Tm?,.
The fourthterm,(-2(P, p., n) , requiresslightlydifferent
treatment. The only
part that can be of interestis
-24,,mn X
and in orderto evaluate the correspondingcontributionto the surfaceintegral
we must expand x as a powerseriesin the neighborhoodof the firstsingularity,
writing
(11.4) x + (x -1 )X.. +
where
(11.5) X= X(X)X X,8 = X,.(Gin),
etc.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 93

Introducingthis expansion for x we see that the only term in the integrand
whichcan give a finiteresultis
(11.6) -24,mn(X - )is

The determinationof the surfaceintegralof this termdepends on the calcula-

f
tion of

(X8- 8) Onmn COS (n.N) dS.

We have
(X8- O) ,rnnCOS (n/N)

=2(x - v) {-3(x - v )(X, - n)/1 + 1a

/3} ( -
n)/

= 14
-4m(x1 - _I) ($" - 11)

Therefore

4 f (x8 -X ) ,nn COS (n N) dS = _ f(x'-" )(x3- r)/ dS

(11.7) 1
4m
= -- 6 mes

and so the surfaceintegralof the term (11.6) is

(11.8) m

whichis thus also the value of the surfaceintegralfor the whole of the term
(-Wpp,mn).
In a somewhat similar way we obtain, for the surfaceintegralsof the re-
mainingterms,the values
1

O 4M ..mn
26mmn ,
3

(11.9)
-Pon (P, n
-- 8 _
x M)

Hymn
,nP, > 2mX m.
Hence we have
(11.101 1
(11.10) Cm& = - 2Amn COS (n.N) dS = 4mji + Im}.
4 4~rJ 4

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
94 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

Let us assume for the moment that we are not going any furtherwith the
approximation. In this case our approximateequations of motion would be
of the form
(11.11) X{m + 2X,} = 0
foreach particle. It is of interestto note that this formof the equations of
motion is actually independentof the variables x8. For we have, by (11.5),
(9.6),
(11.12) X.8 = X,.8(), X = 2M/r.

For our presentargumentwe may take x as any functionofof2r. Equations

(11.12) show that to formj,., we must firstdifferentiatex with respect to x8
and then replace x8 by 18. But the resultwill be the same if we firstreplace
withrespecteitherto
x8by q8 and laterdifferentiate 8 or to ( _8). Thus

(11.13) = Ax(r) ax(r)

a
wherer denotesthe "distance" between 8 and P':
(11.14) r= [(87 - r)(X7 -r)]
We can thereforethinkof our equations of motionas involvingthe differentia-
tion of functionsdepending only on the positions of the singularities,as is
characteristicof theoriesbased on the conceptof action at a distance.
Writing(11.11) more explicitlyin vector notation as
(11.15) m= V(A /r)

we see that (11.11) gives preciselythe Newtonian law of motion."

We have thereforeobtained the Newtonian equations of motion fromthe
fieldequations alone, withoutextra assumption such as was hithertobelieved
to be necessaryand was supplied by the law of geodeticlines, or by a special
choice of an energyimpulsetensor.
From the above derivationof the Newtonianequations of motion,the general
mechanismbecomes apparent by which the Lorentz equations forthe motion
of electric particles can be obtained. In this case we have to consider the
gravitationalequations in whichthe Maxwell energy-momentum tensorappears
on the right,and also the Maxwell fieldequations, and treat the whole set of
equations by our approximationmethod. It is necessary,now, to give each
singularityan electric charge e in addition to its mass m. We may safely
ignorethe contributionarisingfromthe productsof gravitationalpotentialsin

I' Equation (11.11)and (11.15)are writtenin termsoftheauxiliarytimeand theauxiliary

masses. We shall returnto this point in 17.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 95

the new field equations. For this omission has the effectof destroyingthe
second termof (11.11), while the inclusionof the Maxwell tensorleads to the
appearance, on the rightof (11.11), of a correspondingsurfaceintegralgiving
the electrostaticforceacting on the particle. In the next approximationwe
obtainthe fullLorentzforcetogetherwiththe relativisticcorrectionto the mass.
So long as we are dealing with singularities,we have no basis within the
theoryfor excludingnegative masses; in other words, for excludinggravita-
tionalrepulsionsbetweenparticles. If, however,we decide always to take mass
positive,thenthe signwithwhichthe Maxwell energy-momentum tensorenters
the fieldequations determineswhetherlike chargesshall attract or repel each
other. This also revealsthe limitationsof any theorybased upon the existence
of singularities.

12. Normalisation of Too. The value of Y0odeterminedfrom (10.7a) is

4 4
arbitraryto withinan added harmonicfunction,and thisfunctionis to be deter-
minedfromthe relations(8.4), (8.2), togetherwith our basic requirementthat
higherharmonicfunctionsthan simplepoles are, as faras possible,to be avoided.
From (10.7a) and the fact that p is harmonic,we have at once

(12.1) Too= -34 nS + a0oo + gooX,

4

formmore
wherewe have writtenthe additive functionsof (8.4) in a different
in accordance with our present notation, aoo, Boobeing functionsof r alone
through-qand r and theirderivatives. The quantities aoo,Ioo can be deter-
minedfromthe conditionthat

(12.2) 4 f {Y00On-2Aon} cos (nuN) dS = 0.

The value of aoois foundby taking this integralover a small sphere having
its centerat the firstsingularity,and fromcalculationssimilarto those of 3, we
find,aftermakinguse of the equations of motionof the firstorder:

(12.3) aoo = {I I" + 2X}.

Similarly,by integratingover a small sphere around the second singularity,

we find
(12.4) /0o= {r8rJ+4i,

where
x= x(X),
(12.5)=

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
96 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

These resultsshowclearlythephysicalsignificance ofthe particularnormalisa-

tion requiredby the conditions(8.4), (8.2). For we now have

(12.6) X2To + X4e00= X2{(1 + IX2aoo) 2i/'+ (1 + X2) 2 3X2

2 4

and we see from(12.3), (12.4) that m(l + P!aoo) m( + 2 2,oo) involve the
firstrelativisticcorrectionsto the masses.
The calculationsup to thisstage correspondto thoseof Droste, De Sitter,and
Levi-Civita, cited in the introduction.

13. Solution of the Field Equations for 1 = 2. Since our ultimate aim is to
determinethe equations of motion up to the next approximation,we are in-
terestedonlyin thoseexpressionswhichgive a contributionto the corresponding
surfaceintegrals. We shall state dogmaticallywhat is needed forthese calcula-
tionsforthe justificationofour statementcan not be givenwithoutexposingthe
details of our actual calculation.
1. The calculation of ym. and yomin the neighborhoodof the singularities.
4 5
We do not need to care in Ymn about those termswhich do not go to infinity
4

if r -+ 0.
2. The calculation of -y, in the whole space.
4
The expression2Amnin (10, 7) can be divided into two parts,one containing
4
the lineartermstogetherwithall othertermsnot involvinginteractionsbetween
the two particles, and the other containing all the interactionterms. We
denote these two groupsof termsrespectivelyby Xmnand Ymn,. The integra-
tion ofthe equations

(13.1) 'ymnea = Xmn

4

presentsno difficulties,
but the equations

(13.2) iYmns = Ymn

4

cannot,apparently,be integratedin an elementarymannerand we are obliged

to introducea simplification. Since we need to know the values of7mn mainly
4
in orderto evaluate the surfaceintegralscmabout, say, the firstparticle,we may
6
introducepower series expansionsfor x in the neighborhoodof this point and
so obtain a solution for Ymn which is also in the formof such an expansion.
4

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 97

We find,actually,from(13.1), (13.2), thefollowingexpressionsfor-y'nand -yin

4 4

m
IYmn = gt4(Xn - ) + (Xm - n)48]- - ,

and
(13.4) y$,7 = -Im(7 - )X -
4

wherewe have included in (13.4) only those termswhich ultimatelyhave im-

portanceforthe evaluation of the surfaceintegralsCm.
8
The value of 7mn is given by
4

(13.5) Ymn =Yrnn + Ymn + atmn a}

4 4 4

where aen
ofThic is a functionof timeto be determinedfromthe coordinateconditions
vuOf eYntr onlygivneary,
In a similarway, we may calculate the values of yOn in two parts. We find,
6

on includingonly relevanttermsforthe surfaceintegrals m Xin the integrands

4

(13.6) z~m = P m
rY8 +
-.m= amn

5~~~~~~~~~~~~~~~~~
SY~n= v _ (Xm
6~~~~~~~~~~~~~~~~~~~
-3(X - -{X _-7)I

712 11 3"m

5 5 5

The value ofyYonis givenby

(13.8) 'omn = on +m'On + 4On 4d

wherea0n is a functionof time to be determinedfromthe normalisationcon~

dition.
It remainsonly to calculate -rin the whole space. From (1O.7c) we have

(13.9) 2qpXoo
Yrr,8 = + A
Is P,8,
4

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
98 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

therefore

11 2 X
(13.10) Yrr -2mr oo - 2mroo
200+,2
+ VP + a# + 3x
4

where a and # are functionsof time to be determinedin such a way that Yrr
4
in (13.10) would agree withErdeterminedfrom(13.5) near to the singularities.
4

14. Determinationof amn and ao,. In orderto findann,,,ao, fromthe condi-

tions (8.7e), (8.7c) we must make use of the values of Cm found in 3. The
4
result is up to the desired order

(14.1) amn = n2X7Y + amnXj

and

(14.2) aOn = -,)87)8, + XR - Rtn.

Finally fromour last remarkin 13 follows:

(14.3) a = 2vS" + lx; d 2tS S + j+

15. Calculation of Ann. In the calculationof Amnforour presentpurposes,

6 6
we may assume that cmis zero, as we shall now show.
4
Afterwe have evaluated the surfaceintegralscm, we may writethe approxi-
6
mate equations of motionin the form

(15.1) X4Cm + X6C m .

4 6

But thisshowsthat whenthe motionis in accordancewith (15.1) the quantities

X4cmand X6cmwill beof the same orderof magnitude. It is evident,however,
4 6
that X4Cmcan enter X6Amnonly in combination with a quantity of the type
4 6
X20. It will thereforeenter only in termswhich actually belong to the order
2
8or higher,and since we do not proposeto go beyondthe orderX6in the calcu-
lation of the equations of motion,we may neglectall termsin Amnin whichcm
6 4

appears. Even if we make use of this fact, however,the calculationsare still

types of term in the
quite tedious, and there are actually forty-onedifferent
expansionof Amn. We find:
6

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
 GRAVITATIONAL EQUATIONS AND PROBLEM OF MOTION 99

2Amn = -7Om,On - 'YOn,Om + 6mnYOO,00 + 'Ymn,0O - (PY00,mn - PY88,mn - (P,mnY700

6 5 5 4 4 4 4 4

- (P,mnY788 + (P,m87Yns + (P,nseYms - b5mn(P8,r77r - 2'p,s8mn,8 +4(PsYms,n

4 4 4 4 4 4

1 1 1 1
+ (Ps ns,m - 2';o,m7Y88,n - 2(PAnYasm - 2(pnYOO,m - 2(,mrYOO,n
4 4 4 4 4

+ 36mn'P,sorr,8 + f3mn(P,87Y00,8 - 70W70n,ms - 70YOxm,ns + 270o8e7O.,mn

4 4 3 3 3 3 3 3
(15.2)
+ 25mn7Y0s,r7Y0r,8 -ffmn'Y0s,r7Y0,r + 7O8,mYOs,n + 70m,870n8 - (P,on7Yom
3 3 3 3 3 3 3 3 3

- (,Om7YOn + 23mn(P,08O08 - (P,O7Om,n - (P,OYOn,m - (AnY0m,0 - (P,mYOn,0

3 3 3 3 3 3

+ 2(7yomon + 2VP7onom - 26mn(P(P,00 + 2 (p(p(p,mn - (P(P,m(P,n

3 3

+ 6mn(P(P,9(P,a + 16mn(PO(P,0-

The conditionthat Amn,nmust be zero affordsa valuable test of the correct-

ff
ness of the above formula. We have worked out the divergenceof the Amn
6

givenin (15.2) and have foundthat it does indeed vanish.

16. The Surface integralsforI = 3. In orderto findthe principaldeviation

fromthe Newtonianlaws of motion,all that essentiallyremainsis to calculate
the values of the surfaceintegralscm. To do this we must firstinsertin (15.2)
6
the values previouslyfoundfor700, 7mn and yonand thenit is a matterof calcu-
4 4 5
lating the contributionsof the resultingterms one by one and adding the
expressionsobtained. The general techniqueis similarto that used in 11 for
the evaluation of cmbut considerablymore complicated.
4
On makinguse of our rightto take cmto be zero, we may expressthe result
4
in the form

Cm = 2AmnCOS (n *N) dS
6 4J m

8
(16.1) = -4mm + 2? - 4 - 4 r rJ 8nr)

[4.8(gm
-
4m) + 3."m
8 -
4?8tm] 8 (r)r 2
()'I 2 anna77r
a77
+ f.
}

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions
100 A. EINSTEIN, L. INFELD, AND B. HOFFMANN

17. The Main Deviation fromthe NewtonianEquations of Motion. In order

to obtain the equations of motionbelongingto this stage of our approximation,
we must write

(17.1) X cm+ X cm= 0 k = 1,2

4 6

and then must reabsorbthe X's by goingover to the old time x0instead of the
auxiliary time r = Xx0and by introducinga correspondingchange in mass
fromm to M, whereM = X2m. There will be no confusionif we keep the old
notationforthe new quantitiesso that now = dt/dxoinstead of dt/dT,and
m is writtenfor the new mass M. And with this conventionwe may write
the equations of motion (17.1), by means of (11.10) and (16.1), in the form

M 2 8(1/r)= 2
-24a? a4r 1
r r
(17.2) 3*~(/)+1 &
- 2 )
[4~`(ff ~?f) + 3, 8- 8 t8lir 0/rm

The equations of motionforthe otherparticleare obtained by replacingm, mY

n,
Vbymymy,, 77.
These equations, giving the relativisticmotion of two massive gravitating
bodies, constitutethe main resultof our calculationsfromthe point of view of
practical application.
These equations have sincebeen integratedby H. P. Robertson,whoseresults
are given in the followingnote on "The Two Body Problem in General Rel-
ativity,"Math. Ann. 39, p. 101 (1938).

We should like to thank ProfessorRobertsonfor the very kind interesthe

took in this problemand forhis help.
THE INSTITUTE FOR ADVANCED STUDY.

This content downloaded by the authorized user from [Link] on Tue, 27 Nov 2012 08:47:03 AM
All use subject to JSTOR Terms and Conditions