ANNALS OF PHYSICS: !%?

, 169491 (1963)

The Principle of Equivalence*

F. ROHRLICII

Department of Physics and Astronomy, State University of Iowa, Iowa City, Iowa
and Argonne National Laboratory, Argonne, Illinois

Various statements identified with the principle of equivalence are not

acceptable because they are either not generally valid or are simply the defini-
tion of an inertial coordinate system. The accepted statement refers, roughly
speaking, to the equivalence of inertial mass with passive and active gravita-
tional mass. The clarification of this problem is greatly eased by a detailed
study of the static homogeneous gravitational field which precedes the discus-
sion of the equivalence principle. The role of electromagnetic phenomena,
and, in particular, the presence of a charge in such a field is analyzed in detail.

I. PURPOSE OF THIS STUDY

Unfortunately, laboratory experiments are not usually performed in failing

elevators.’ They are carried out in reference frames which are not inertial, but
which are supported in a static gravitational field. Nevertheless, it is taken for
granted that the behavior of physical systems studied in this supported frame is-
apart from a simple correction due to the presence of the gravitational field, char-
acterized by the constant g-identical with that in an inertial system.
When pressed on this matter a physicist will refer to the principle of equiv-
alence. However, this principle is by itself a matter of considerable contro-
versy. It is studied experimentally2 while the theorists disagree concerning its
formulation and meaning. Two standard references (3, 4) disagree about the
equivalence of accelerated observers and gravitational fields, while a third (5)
does not seem to believe that there is such a principle at all.
The purpose of this study is a clarification of these matters (Sections VIII
and IX). To this end it will be helpful to have a detailed calculation before us
which deals with the most important gravitational field which is usually referred
to in connection with the principle of equivalence, viz. the static homogeneous
* Supported in part by the National Science Foundation.
I According to general relativity, the special theory of relativity is valid only locally
in freely falling reference frames.
* E8tviis’s classical experiment has recently been repeated by R. H. Dicke (1) and other
tests of this principle are proposed (2).
3 Throughout this paper the abbreviation SHGF will stand for “static homogeneous
gravitational field.”

169
170 ROHRLICH

gravitational field3 (SHGF). That this simplest of all fields is not completely
understood can easily be seen by raising the question of the radiation from a
uniformly accelerated charge. Does it radiate? And if it does, would this not
contradict the principle of equivalence? While the first question was answered
recently in the affirmative (6), the second question seems to be controversial;
it is closely related to a correct statement of the principle of equivalence.
Starting with an introductory section on uniform acceleration in special rela-
tivity, (Section II) which will be significant later on, the SHGF is defined in
Section III, followed by the corresponding description of free fall (Section IV)
and the associated local geodesic coordinate system (Section V) which permits
one to relate uniform acceleration in an inertial system to free fall in a noninertial
system. The relevance of conformal transformations in this respect is studied
in Section VI. The formulation of physical laws in a noninertial system (a frame
supported in an SHGF) is the problem of Section VII. It will yield the answer
to the above question on radiation. The results of this study lead to a clarifica-
tion of the principle of equivalence; its various forms can thus be evaluated
(Section VIII). The results are discussed in the last section (Section IX).
II. UNIFORM ACCELERATION

Consider an inertial frame of reference, I. In this frame the motion of a point

particle P is described by its position as a function of time r(t). At each instant
t the particle will have a velocity v(t) = dr/dt, an acceleration a(t) = dv/dt, as
well as higher derivatives of position, b(t) = da/d& etc.
At a given instant of time, to , an instantaneous inertial rest frame, II, , can be
defined by the requirement that the velocity v of P, when referred to ItO , should
vanish at t = to . After having chosen a particular to we shall denote It, simply
by I’, and we shall indicate all quantities referred to I’ by a prime.
Uniform acceleration is a special type of motion characterized by the condition

b’(t’) = 0, (2.1)
independent of the choice of to to which I’ refers. If, as is necessarily the case in
practice, uniformly accelerated motion takes place only over a finite time interval
tl < t < t2, then (2.1) is to hold for tl’ < t’ < t2’.
The defining condition (2.1) can be expressed in terms of the quantities re-
ferred to I and the relative velocity v(6) of I’ relative to I at the instant to .
To do this, it is essential to express first r’, t’, and the derivatives dr’/dt’, etc. in
terms of r, t and its derivatives by means of a Lorentz transformation with rela-
tive velocity U, and after this is done, to identify I’ with the instantaneous rest
system by u = v(to) . The condition of uniform acceleration, (2.1)) is then (7)
b + 3r2av-a = 0 (2.2)
 PRINCIPLE OF EQUIVALENCE 171

where y = ( 1 - D’)-~“. Uniformly accelerated motion takes place for all times
for which (2.2) holds. Equations (2.1) and (2.2) are equivalent.
The most general motion r(t) in 1 with uniform acceleration is given by the
general solution of (2.2). To find this solution we note first that (2.2) can be
written

$ (-y3a> = 0 (2.2)’

and can therefore be integrated immediately, (y 2 1)) to yield

T3a = g (2.3)
where g is a constant vector (independent of t) .
A simple calculation shows that Eq. (2.3) can be put into the form

d(v)
- - y(y2 - 1)v X (v X a>/2 = g.
dt
This equation can be written as an equation of motion, using

P = myv, F, = ,t,zg
(m always denotes the rest mass), provided one introduces the “pseudoforce”

Fw = -(Y + 1)~ X o (2.4)

where
o= -(y- l)vXa/v’ (2.5)
is the angular velocity vector associated with the Thomas precession. Thus, the
equation of motion of uniform acceleration is
dp/dt = F, + F, . (2.6)
The force F, is a constant force, independent of time and position. The pseudo-
force F, vanishes whenever v and a are parallel. In particular, for rectilinea?
motion (motion with initial velocity parallel to g) where both v and a are parallel
to g throughout the motion,
d(mrv) /dt = mg (rectilinear case). (2.7)
This particular case is known as hyperbolic motion, since the integral of (2.7) is
a hyperbola in Minkowski space.
When the initial velocity is not parallel to g the motion will no longer be
rectilinear and F, # 0. The instantaneous inertial rest frames It, and Itotiln will
then differ in the direction of their velocities relative to I. Since the composite
of two successive Lorentz transformations without rotation and with relative
172 ROHRLICH

velocities not parallel to each other is equivalent to one Lorentz transformation

with rotation, a Thomas precession effect necessarily arises. This effect is de-
scribed exactly by F,,, which has the structure (but not the sign) of a Coriolis
force associated with an instantaneous angular velocity o. In the nonrelativistic
limit Eq. (2.4) becomes
F,= -2mvXo. (2.4) NR
There is no term, however, analogous to a centrifugal acceleration.
Since (2.6) and (2.3) are equivalent, it is obvious that in the limit when no
force is acting, F, = 0, the equation of motion (2.6) leads to the expected so-
lution a = 0 uniquely.
The above definition of uniform acceleration can easily be put into the co-
variant language of special relativity. For charged particles the condition (2.1)
is equivalent (8) to the vanishing of the Abraham four-vector of “radiation
reaction,”
I’” = rO(du“/dr - axa%‘) = 0 (2.8)
where vp is the velocity four-vector,4 a” = dv’/dr, and r. is N of the time it takes
a light wave to travel the distance of one “classical electron radius” ( e2/mC2) .
For any other particle the corresponding charge and rest mass is to be used.
The equation

da’/dr = aha%‘, (2.9)

where we use the quasi-Euclidean metric with positive signature, can be re-
garded as the covariant definition of uniform acceleration in special relativity.
From (2.9) and the orthogonality of a’ and v’ follows at once that the in-
variant electromagnetic radiation rate (R is a constant (9),

CR = rnqaha’ = const. 2 0 (2.10)

The equation (2.9) can therefore be integrated with respect to r, yielding (10).
v’ = aLeAr + fire+ (2.11)

X = da>, apar’ = 0, P,p” = 0, 2cy,p’ = -1. (2.12)

If (R = 0, i.e., no radiation is emitted, the motion is necessarily uniform, i.e.,

the velocity is constant,
(R = 0 t) v” = a’ + p* (2.13)
4 The space components of a four-vector, W, say, are of course not to be identified with
the components of the three-vector v. The latter will never be used, so that no confusion
can arise. In the case of VP: ZP = (y, yv).
 PRINCIPLE OF EQUIVALENCE 173

If @ > 0, (2.11) can be integrated once more to yield

i = yc + A-‘( aPeAr- p”e+). (2.14)

The three constant vectors (Y’, @“,and 7” are restricted by (2.12) and are deter-
mined by the initial conditions

x’(0) = X-‘(ap - p”) + Yb

(2.15)
v”(0) = ap+ 6”.
The constant X is the magnitude of the acceleration in the instantaneous rest
system and is determined by the external force according to (2.3). The hyper-
bolic nature of the world line and the noncovariant relations discussed earlier
can easily be seenfrom the explicit solution (2.14).
All the above considerations refer to flat space (Minkowski space) and to the
absence of gravitational fields. We shall now turn to the description of a par-
ticular kind of gravitational field.

III. THE STATIC HOMOGENEOUS GRAVITATIONAL FIELD

Before we proceed with the study of this field it must be emphasized that a
very general class of gravitational fields can exist (with suitable asymptotic
conditions) which have vanishing curvature tensor in a Jinite domain, D, to a
certain approximation. Consequently, there exists a coordinate system in which
this field appears as an SHGF in D in that approximation. A measurement which
does not exceed a certain accuracy will therefore yield results which are in-
distinguishable from those in an SHGF, characterizing the paucity of information
obtained (seefootnote 2). This point should be kept in mind during the following
study of the SHGF.
Intuitively, one expects that uniformly accelerated motion takes place in
free fall in an SHGF. However, this is not generally the case. Uniform accelera-
tion was defined in special relativity and the only reasonable requirement one
can make is that the freely falling observer in an SHGF, i.e., the one for whom
(at least locally) special relativity holds, should see an object in uniform ac-
celeration when this object is supported in the SHGF. In Section V this expecta-
tion will indeed be proven. Conversely, uniform acceleration of a freely falling
object as seen by an observer supported in an SHGF will be found only for a
special choice of the coordinate system.
Definition: A static homogeneous gravitational field (SHGF) is defined by a
timelike line element, using a metric with positive signature,

dT2 = -Adx’ - Bdy’ - Cdz’+ Ddt’ (3.1)

where A, B, C, and D are functions of z only (the field is parallel to the z-axis)
174 ROHRLICH

and the curvature tensor

R xXpr= 0. (3.2)
It is now a matter of computation to determine the functions A to D from the
differential equation (3.2).
If we denote derivatives with respect to z of the functions A to D by a prime,
the only nonvanishing Christoffel symbols are
D’
I?;, = - r;, = - -A’ r,“, = - -B’ I?;, = -C’
2C’ 2C’ 2C’ 2c
(3.3)
D’ I A’ B’
rao = ri, = 20’
0
r31= ri, = 52’ r32= r;, =
2
-.
2B
Substitution of these expressions into (3.2) yields the following set of differential
equations which exhausts the 20 linearly independent components of the curva-
ture tensor:
A’B’ = 0, B’D’ = 0, D’A’ = 0 (3.4)

(E = A, B, orD). (3.5)

Equations (3.4) state that at least two of the three coefficients A, B, and D
must be constant. In order to make progress at this point we look at the non-
relativistic limit of an SHGF. The gravitational potential 4 for a constant force
in the negative z-direction is
4NR = gz such that FNR = -WLVC$NR = -mgk (3.6)
where k is the unit vector in the z-direction and g is a constant. If the gravita-
tional field is weak, so that the gPydiffer from the Minkowski values ~]r~ (signa-
ture +2) only by small terms (4 << 1)) and if (d~/dt)~ << 1, etc., (3.1) must
reduce to
-dT2 = dx2 + dy2 + dx2 - (1 + @NR) dt2. (3.7)
This shows that D(z) cannot be a constant. We are thus lead to the result that
A and B are constants, by (3.4)) and that, therefore
A=B=I (3.8)
by (3.7). Thus, a time-orthogonal metric tensor in flat space which depends
on only one Cartesian space-coordinate describes necessarily a plane-symmetric
field.5
6 One can prove that a static gravitational field with plane symmetry necessarily implies
flat space (11).
 PRINCIPLE OF EQUIVALENCE 175

The only remaining equation which needs to be satisfied is (3.5) with E = D.

This can be written
20”
--- D’ C’
(3.9)
D’ D=1;
since D’ # 0. This equation can easily be integrated:

C(z)= (+$dq.
The integration constant is determined by (3.6) and (3.7). The line element
(3.1) can therefore be written
dT* = D(x) dt* - (dB’/g) * dz* - dy* - dx*. (3.11)
We conclude that our dejkition of a static homogeneous gravitational field is
satisjied by the injinity of the line elements (S.il) where D(z) is an arbitrary real
,function, restricted only by requirements of continuity and the nonrelativistic
limit
DiqR(z) = 1 + 292. (3.12)
Among this infinity of line elements there are three which are of special in
terest. If we assume C’ = cons& i.e., C = 1 by (3.7), the line element becomes
-dT2 = dx2 + dy2 + dz” - (1 + gz)’ dt’. (3.13)
This metric is used by Moller (1.8). It means that a linear variation of clock
rate with height is sufficient to simulate a SHGF. This is known for weak fields
where the SHGF reduces to the usual nonrelativistic field but is here proven for
fields of arbitrary strength.
Another case of interest is obtained by assuming:
C proportional to D. (3.14)
This yields
-dr2 = dx* + dy’ + e2”(dz2 - dt*) . (3.15)

These coordinates satisfy the harmonic coordinate condition

r” E g”l+:@ = 0.

Finally, one can choose CD = 1 which results in the Kottler-Whittaker line

element

--dr2 = dx2 + dy2 + dz2- - (1 + 2gz)dt’.

1 + 292
176 ROHRLICH

In these coordinates the nonrelativistic form of D(z), Eq. (3.12)) is actually

exact.
IV. FREE FALL

We now turn to the motion of a test particle in the gravitational field defined
by (3.11) and (3.12). Using
u(z) = d&4, UNR = 1 + $72 (4.1)
the corresponding linear connection has as only nonvanishing components
0 U’ 3 U"
I?30 = I$, = u' r33 = -J' r:, = g2 $

The particle will fall freely according to the geodesic

(4.3)

which yields

(4.4a)

(4.4b)

(4.4c)

The last equation can immediately be solved for

where a is an arbitrary constant. Since we require dr = dt when all velocities

vanish and g = 0, it follows that (Y = 1. This choice describes particles which
are at rest (have zero velocity) at the instant at which they pass through x = 0.
By means of (4.5) the equations of motion (4.4) become

(4.6a)

(4.6b)

If we now consider one-dimensional motion’ (dx/dt = dg/dt = 0 for all t),

6 For three-dimensional motion (dx/dt and dy/dt not both vanishing initially) the form
(2.3) can never be satisfied for the equations of motion in SHGF coordinates.
 PRINCIPLE OF EQUIVALENCE 177

then the line element (3.11) combined with (4.5) yields

(4.7)

and a simple calculation verifies that this is a first integral of (4.6b). With t
as the only independent variable, (4.7) can be written

= g2u2(l - U”) (4.8)

which yields
1
- = cash g(t - to). (4.9)
u
The integration constant was so chosen that u = 1 at t = to. This equation
determines x(t) for one-dimensional motion (free fall) when u(x) is given.
In the nonrelativistic limit with gz << 1, gt << 1, it gives
2 = -$g(t - to)? (4.10)
We can choose u(x) in such a way that a freely falling test particle undergoes
hyperbolic motion.7 According to (2.3) we must then have
312
d2.z
-=
dt2
-g
[ 1
1-g

which has the solution

g(z - 20) = 1 - 2/l + g2(t - to)*, (4.11)
where to is the time at which the particle velocity 21= 0. Therefore, with (4.9)
we must have

[D(z)]-1’2 = -&- = cash d(l - gz)2 - I. (4.12)

u(z)
Here we have chosen zo = 0 in order to comply with the previously chosen initial
condition (3.12). The nonrelativistic limit of (4.12) is (4.10).
Thus, we conclude that for a given nonrelativistic limit (corresponding to our
choice (3.12) and corresponding initial conditions) the SHGF is uniquely de-
fined by the line element (3.11) and the requirement that free fall should be
hyperbolic motion; it yields (4.12). None of the line elements (3.13)) (3.15),
and (3.16) give hyperbolic motion. The difference appears in order g2; with
to = 0 we have
x = -+gt2(1 - a(gt)2 + . . .) (4.13)
7 This means that z(t) satisfies the same second order differential equation of motion
as in special relativity.
178 ROHRLICH

Various line elements differ in the value of Q: which is ss; N, and x far the
three cases mentioned, while (4.12) yields (Y = x.

V. TRANSFORMATION TO LOCAL GEODESIC COORDINATES

While we have computed the motion of free fall as seen by an observer at
rest in a static homogeneousgravitational field, we now ask how this observer
is seenwhen viewed from the freely falling test particle.
The principle of equivalence implies that the freely falling test particle de-
fines locally an inertial system. Thus, we carry out a transformation to a local
geodesic coordinate system S’ = I’ defined by

dtf2 - dz” - dy” - dx” = u2dt’ - (~‘/g)~d.z~ - dy2 - dx2. (5.1)

The corresponding coordinate transformations for free fall are found by an easy
calculation :
2’ = 2, Yl = Y
g(z’ - a,‘) = u(z) cash g(t - to) - 1 (5.2)

go - to’) = u(z) sinh g(t - to).

For the choice u(z) = 1 + gz used by Moller (Id), corresponding to the line
element (3.13)) the transformation (5.2) reduces to his well-known acceleration
transformation.
Equations (5.2) imply

[; + (x’ - d)] - (t’ - to1)2

= (t&>/g>“. (5.3)

This equation is valid everywhere in space and not only locally, becausewe are
dealing with a homogeneousfield satisfying Eq. (3.2). The physical meaning of
this result can be expressed as follows: An observer S’ freely falling (and there-
for inertial, S’ = I’) in an SHGF will seean object which is at rest in S (i.e.,
supported in the SHGF) as moving with uniform acceleration (hyperbolic
motion). This result is independent*of the choice of u(z). It provides the eventual
justification for our definition of an SHGF: a plane x = constant, at rest in the
SHGF, will be seen by an inertial observer to move parallel to itself, rigidly,
and with’constant acceleration (hyperbolic motion).
A comparison is important of the conclusions just reached and the result of
the preceding section : an observer supported in an SHGF will seea freely falling
8 The choice of u(z) determines the magnitude of the acceleration (g/u), but does not
influence the fact that the motion is hyperbolic. The reader is reminded that “acceleration”
here means “acceleration in the instantaneous rest system”; it is a constant according to
(2.1).
 PRINCIPLE OF EQUIVALENCE 179

object in general not in hyperbolic motion. This provides a simple example

dispelling the often expressed belief that in general relativity acceleration is
relative and therefore reciprocal in the sense that the motion of A relative to B
is identical (apart from a sense of direction) with the motion of B relative to A.
In fact, strangely to say, this reciprocity of acceleration is true only in non-
relativistic physics (first order in G and small velocities) where acceleration is
absolute. In general, relativity acceleration is not an intrinsic quantity (d2x”/d7’
is not a vector) and therefore has meaning only relative to a given reference
frame. Obviously, the local geodesic ones are singled out, since only in them
does d2xP/dT2 = 62x”/6~2. The latter quantity is a vector, viz. the covariant
second derivative of a vector with respect to a scalar (cf. Eq.(8.3) ) .

VI. CONFORMAL TRANSFORMATION8

The 15-parameter group of conformal transformations contains, in addition

to the lo-parameter Lorents group (which is a subgroup of it), also the dilatation
transformation x” = Xx’ and the 4-parameter acceleration transformation

(6.1)
1 - 2axxX + upu%T5”
The scalar products here refer to Minkowski space: x,?‘ = q,,x’x”. All elements
of the group satisfy the relation characterizing the general conformal group
dx,‘dx” = u(x) dx, dx’. (6.2)
The special 15-parameter group results from suitable restrictions on u(x) (13).
This conformal group is very closely linked to motion with uniform accelera-
tion: These motions are characterized by (2.2) and this conformal group is the
largest group of transformations which leaves (2.2) invariant (7). In particular,
the acceleration transformation (6.1) transforms from rest to uniform accelera-
tion. It can also be made to satisfy the nonrelativistic limit (4.10).
It is now essential, however, that there exist transformations which are not
conformal transformations (i.e., do not satisfy (6.2)), but which nevertheless
transform a particle from rest into hyperbolic motion. A case in point are the
transformations (5.2) from rest in an SHGF to hyperbolic motion in I’. Since
these are the most general transformations satisfying (5.1), it follows that
the (conjormal) acceleration transformations (6.1) do not transform from an SHGF
to an inertial system.
Physically, this result follows from the fact that the conformal transformation
(6.1) does not leave invariant the coordinate orthogonal to the acceleration, as
would be expected for the relationship between an SHGF and an inertial system,
and as is the case in (5.2). Furthermore, the covariant nature of (6.2) ties the
transformation of the spatial coordinates to the time in such a way that the
transformed line element cannot describe a static situation.
180 ROHRLICH

However, if one restricts oneself to the one-dimensional case (X = y = 0)

these objections no longer hold, as long as one regards the acceleration transfor-
mation (6.1) as connecting an inertial system with one in which there exists a
nonstatic apparent gravitational field. The weak field limit is then correctly
given by (4.10) and is not distinguishable from the static case. For larger g
the consequences of the acceleration transformation are studied elsewhere (14).
From these considerations follows that the set of all frames which are char-
acterized by a time-orthogonal line element and the property that a point at
rest in them should appear uniformly accelerated in a given direction (z-direction,
say), is a very large set indeed. It contains all the SHGF frames (3.11) as well
as all the (nonstatic) conformal frames following from (6.2). However, the
transformations T from an inertial system to this set do not form a group. The
largest subset of T which does form a group is the set of special conformal trans-
formations, CO , in the notation of ref. is.
In using conformal transformations one forces a Minkowski space onto a
noninertial observer. The subsequent distortion of clocks and yardsticks which
is necessary to accommodate that space results in the change of distances or-
thogonal to the field.
A consistent theory of conformal invariance based on (6.2) requires that masses
and the gravitational coupling constant G are not universal constants, but de-
pend on the coordinate system (i.e., on U(X)). While this and the concomitant
conformal invariance of Gm/R (where R is a length) permit a formulation of
Mach’s principle, conformal physics meets with difficulties unless it is restricted
to flat or conformally flat space. As we shall see, the principle of equivalence is a
much stronger statement when applied to curved space, SO that we shall not
pursue conformal physics further in the present paper (13).
VII. ELECTRODYNAMICS IN AN SHGF
In a Riemann space Maxwell’s equations in the absence of matter (Maxwell-
Lorentz equations) can be expressed in terms of the four-vector potential A’ by
g?,VBA” = -4uj’ (7.1)
V,A’ = 0. (7.2)
The covariant derivatives V, are defined in terms of the Christoffel symbols
I”,, which are given by (4.2) for an SHGF. The fields are then

F,,y = &A, - &A, . (7.3)

Substituting the I’“,, , one finds for the Lorentz condition (7.2)

a,# +
 PRINCIPLE OF EQUIVALENCE 181

while the four equations (7.1) become coupled linear second order differential
equations which are somewhat complex and need not be given here. While it
would be rather difficult to solve these equations, the solutions can be obtained
much easier by solving the corresponding problem in the freely falling (inertial)
system I’, and then transforming the result to the SHGF (frame X) by means
of the transformation (5.2).
To illustrate this procedure we consider the simplest possible case: a charge
e at rest at x = y = z = 0 in the SHGF. In I’ this charge undergoes hyperbolic
motion (cf. Section V) with acceleration g. If we assume that at t = 0 we have
to’ = 0, zO’ = l/g and x0’ = yO’ = 0, we have from (5.3)
12
x - tr2 = l/g2 (7.5)

for the motion of the charge. The fields produced by it were computed previously
(8). In our notation and using cylindrical coordinates p’, ‘p’, z’, they are
Eil = ( 8e/g2) p’z’/f3
E:J = - (4e/g2) ( l/g2 + t” + p12 - 2”) /&‘”
EL, = H;, = Hi, = 0 (7.6)

Hk I = (8e/g”) p’t’/03
(’ = [( l/g2 + t’2 - PI2 - 2’“) 2 + (2p’/g) 2]1’2.
The fields FNy(zr) in the SHGF are now easily obtained from FLY( r’) given by
(7.6). One only needs the transformation
FC(“( x) = u,“u~Fh,( 2’) (7.7)

where up“ follows from (5.2))

1 =
Uki = 0 i # k, a1 a22 r 0

a3 3 = ( l/g)u’ cash gt uo3 = u sinh gt (7.8)

a3” = (l/g)u’ sinh gt uo” = u cash gt.
The result of this easy calculation is
E, = H, = H, = H, = 0 (7.9)
E, = g(z’E;r - t’H& E, = (l/g)uu’E:f . (7.10)

Clearly, since the magnetic field vanishes identically and independentEy of u,

a charge at rest in any SHGF does not radiate.g However, in the inertial system
9 The specific form of the condition which assures that these fields are retarded in S
is irrelevant as far as H = 0 is concerned.
182 ROHRLICH

I’ the charge does radiate, since it is accelerated (by being supported in the
gravitational field). The derivation of (7.9) makes it obvious that these two
statements are consistent with each other. They should not come as a surprise
to anyone.
The electric field strength has no azimuthal component. It can be expressed
in terms of p, z, and t by means of

P. =P

gz’ = u(z) cash gt (7.11)

gt’ = u(n) sinh gt
and yields, after combining it into Er2 = (EP2 + g33Ez2)(go’> 2,
4eg2
E, = (7.12)
t1 - bPj2 - u”l” + (%P)2’

For u2 = 1 + 2gx one finds the weak field result E, = e/r”. The first correction
to it is given for u2 = 1 + 2gz + a(gx)2 by

The potential A, follows from ref. 6 in the same way and yields for the “Cou-
lomb” potential
1 + (gp)2 + u”
4 = -Ao = eg r<l - (gp)2 - u2)2 + (&g)2p/2
(7.14)

For small g this becomes 4 = e/r. If one chooses the Kottler-Whittaker metric
(3.16)) corresponding to u2 = 1 + 2gz, the potential (7.14) reduces to the
expression found by Whittaker (15) .I0
The method employed above for the computation of the field of a charge
supported in an SHGF, can equally well be used to compute the field of a charge
freely falling in an SHGF. One transforms the Coulomb field in the inertial
frame I’ in which such a charge is at rest, to the fields seen by S, using the
transformation (7.8). The result is that H, = H, = E, = 0 while the other
components do not vanish. The results need not be repeated here, since they
agree with those computed by Rosen (16) for the case of the M@ller SHGF.
The charge is again found to emit radiation, provided one makes sure that the
fields are retarded. The Coulomb’ field is well-known to be ambiguous in this
respect.”
10 Note that for that metric the Lorentz condition (7.4) reduces to &,A# = 0.
11 In this connection Rosen’s work (16) must be discussed, since he concludes from the
Bame fields that the charge would not radiate. His computation refers to a charge at rest
 PRINCIPLE OF EQUIVALENCE 183

VIII. THE PRINCIPLE OF EQUIVALENCE

Before proceeding to the discussion of the principle of equivalence a physically

almost obvious but conceptually essential point must be made: Any theory of
gravitation must contain in a suitable limit the formalism (though not the
philosophy) of Newton’s theory of gravitation. Similarly, if the theory is to be
valid without restriction on the particle velocities, it must contain in a suitable
limit the laws of physics as stated by the special theory of relativity. The general
theory of relativity is well known to satisfy these conditions.
Whatever the statement of the principle of equivalence (and there have been
many) its validity must extend beyond the framework of Newtonian gravitation
or special relativity. Otherwise, it would hardly be justified to be called a prin-
ciple. On the other hand, if this principle is almost trivially contained in t,he
theory of relativity, this would be no reason to ignore it once that theory is
accepted, or to bury it with appropriate honors. Thus, the principle of electric
charge conservation is fully and almost trivially contained in Maxwell’s equa-
tions, but its burial is obviously not indicated.
For Newton’s theory of gravitation the following two statements are valid:
(A) The equation of motion of a test particle in a gravitational field, i.e. of
a particle whose own gravitational field is negligible, is independent of its mass
and composition. (In Newtonian terms: the same quantity plays the role of
inertial mass and of passive gravitational mass.)
(B) Matter is acted on by gravitational fields and is itself the source of a
gravitational field. (In Newtonian terms: the same quantity plays the role of
active and passive gravitational mass.)
In Newtonian physics (A) is valid by implicit assumption, while (B) is a conse-
quence of Newton’s third law and the law of gravitation.
Either of these two statements may be elevated to a principle provided they
are postulated to hold generally, i.e., not only within Newtonian gravitation
theory. Thus, when combined with special relativity they say that the (inertial)
in I’ as seen by an observer supported in Mgller’s SHGF (3.13), since his use of a Schwarz-
schild-Nordstr8m line element is irrelevant in the limiting case of interest. As was shown
above, this system is not the one discussed by Fulton and Rohrlich (6) in which a charge
is uniformly accelerated, i.e., it is at rest in S and seen by I’. The use of the principle of
equivalence in this connection will be discussed in Section VIII. In any case, however, the
charge radiates provided the fields are retarded. The retardation condition is not auto-
matically contained in the fields which were derived from the Coulomb field. It must be
imposed explicitly and has the form z + t > 0 in Rosen’s notation or z + t > 0 in ours
(cf. p. 503 of ref.6). The time-symmetry argument whether applied on the fields or on the
energy-momentum tensor is therefore misleading; it yields no radiation in either references
(6) or (16). If the retardation condition is not imposed one has a mixture of half retarded
and half advanced fields (expressed by the identical formulas for the fields), corresponding,
however, to two charges, one being the time reverse mirror image of the other in a Minkowski
diagram. In that case there is indeed no radiation, as was explained in ref. 6.
184 ROHRLICH

mass equivalent of any form of energyI is also both active and passive gravi-
tational mass. In general relativity (A) is satisfied by the geodesic postulate
“all test particles move along geodesics.” It is also valid for a cloud of test
particles (5). In this sense it is not a local statement. (B) is satisfied by the
dual role played by the energy-momentum tensor in the gravitational field
equations. While Tpy plays the role of a field source (active gravitational mass)
in
R,ua - 3gpvR = KT~~,
it plays the role of passive matter whose motion is determined by the gravita-
tional field (passive gravitational mass) in

T,m - 3g,nT = (l/~)R,w .

Both statements, (A) and (B), are valid for true gravitational fields, i.e.,
irrespective of the value of the curvature tensor (including zero), independent
of the strength of the field.
Statement (A) is the basis for a geometrization of gravitational interaction.
Statement (B) is far from trivial as soon as Newton’s third law is abandoned
as is the case in a theory containing finite propagation velocities of interaction.
We shall later adopt (A) and (B) as the full statement of the principle of equiv-
alence. It is a generalization of the assertion of the equivalence of inertial and
gravitational mass.
An equation of motion of a test particle is an equation involving the second
(and no higher) time derivative of position of that particle: although this is
not essential we shall here assume that one can solve this equation for the second
derivative, yielding
d2X”
-=
dr2
p x,: . (8.1)
f( -> 7

If we have a set of particles of various masses and compositions, all with the
same initial conditions, it follows from (A) that f” is independent of these masses
and compositions. Therefore, over a region over which the field is sufficiently
uniform, an observer moving (“falling”) with these particles will see no effect
of the gravitational field. The field can be transformed away locally. The exist-
ence of such a comoving observer is physically obvious.
Mathematically, this means that there must exist a coordinate transformation
S + S’ such that in the new coordinate system (8.1) reads
d2x’” o
(8.2)
d72= *
12 This is the precise meaning of the vague word “matter” in (B) when this statement is
made as a local statement.
 PRINCIPLE OF EQUIVALENCE 185

The freely falling observer, characterized by the coordinate system S’, sees
no “gravitational field strength,” j” = 0. Since by assumption no other forces
are present, S’ is an inertial observer.
In this way it follows that a theory of gravitation which contains both New-
tonian gravitation theory and special relativity in suitable limits and which is
consistent with (A) must involve the following specafiation of an inertial system:
(C,) A system whose origin is freely falling and which is nonrotating in a
gravitational field is locally an inertial system and all the laws of special rela-
tivity are locally valid in it.
This statement can also be expressed in the form.
(C,) There is no physical experiment by which an observer can distinguish
locally between his own free, nonrotating fall in a gravitational field and field
free space.
A third form of the same statement is obtained by considering the inverse
transformation, i.e., the transformation leading from (8.2) to (8.1)) and using
(A). The above statements (C) then can be phrased as
(C,) An acceleration field is locally equivalent to a gravitational field.13
All three statements (C) depend on the existence of the transformation
S * S’. In general relativity this is assured by the assumption that the under-
lying space has a symmetric linear connection. Thus, in this respect a space
much more general than a Riemann space (assumed in general relativity) would
do as well. In such a space there always exists a transformation which makes
the connection I’“,, = 0 locally. It is then only necessary to make I’“,, a factor
in j“ of (8.1) to obtain the desired equation of motion. In this case it is the
geodesic
- equation

(8.3)

If one postulates covariance and a symmetric linear connection, the geodesic

postulate follows uniquely from (A)14 and the existence of S c--) S’.
General relativity also permits a precise definition of “locally” in (C) . It
means “over a space-time domain D of order (6~)~ in which the curvature
tensor vanishes everywhere in the sense that R,,,$ix”6x8 - 0,” i.e., “to the
extent that there is no true gravitational field.” For apparent gravitational
fields (R,A,, = 0) “local” means “everywhere.”
I3 This statement means that two noninertial systems are locally identical. “An inertial
frame in which there is a gravitational field present” is meaningless and a self-contradiction.
Fock (4) identifies (Ca) with the principle of equivalence. His argument that ((23) is
valid only for nonrelativistic physics to first order in G is incorrect. “Local” does not mean
“to the approximation that the field is a Meller SHGF,” but refers only to the approximate
vanishing of the curvature tensor. Even if the field approximates an SHGF it need not, have
the Mprller metric, as was shown in Section III.
I4 For a spinning particle, (A) need not be valid and neither is (8.3).
186 ROHRLICH

,It is now clear that all statements (C) are simply recipes for constructing
inertial systems in a world of gravitation. They define an inertial system and
thus give meaning to the special theory of relativity. The definition with respect
to fixed stars which is so often given is logically unsatisfactory and, being a
nonlocal definition, has no place in a theory which does not fully incorporate
Mach’s principle and which is based on differential equations.
One or the other of the forms of the statement (C) have been preposed as
statements of the principle of equivalence. According to the present point of
view this is not acceptable for two reasons:
(a) The statement (C) does not state an equivalence but is a definition (opera-
tional, in a certain sense) of an inertial system and can thus not be a principle.
This definition is used to deduce the laws of special relativity from experiments
in noninertial laboratories. The only question that could be asked is: What
assures that two differently moving inertial observers (one freely falling radially
and one circling the earth, say) would both see locally the same laws of special
relativity? This is assured by the principle of covariance (which can thereby be
tested) and has nothing to do with the principle of equivalence or the definition
of inertial systems.
(b) The statement (C) is by its nature meaningful only “to the approximation
that there is no true gravitational field (RKx,l = 0)” and therefore physically
not acceptable as principle about gravitational interactions.
On the other hand, we must keep in mind that none of the definitions (C)
would be possible without the validity of the principle of equivalence (part A)
at least for flat space. Thus, an experimental test of (C) confirms that in JEat
space the equation of motion of a test particle in a gravitational field is independ-
ent of mass. Only an experiment in which those features of the mtric enter which
assure a non-vanishing curvature, can provide a test of the full principle of equi-
valence.
In summary, then, we have two basic, fundamental, and independent as-
sertions, (A) and (B), which are to be regarded as “the principle of equivalence.”
They refer, respectively, to the equivalence of passive gravitational mass
with inertial and with active gravitational mass. They are valid for true gravi-
tational fields and are not restricted to the flat space approximation. The var-
ious statements (C) are each a definition of an inertial system and they cannot
meaningfully be regarded as principles. It must be understood, however, that no
such local definitions were possible without the principle of equivalence (state-
ment (A)) being valid at least for flat space.

IX. DISCUSSION

While the bulk of this paper is concerned with the study of a static homo-
geneous gravitational field (SHGF), the results derived from it are only auxiliary
 PRINCIPLE OF EQUIVALENCE 187

to the problem, of the precise formulation and meaning of the principle of equiv-
alence. Although an SHGF is in general a poor approximation for actual physical
systems (a region immediately above the center of a large disk-shaped galaxy
might be the best approximation realized in nature), this study greatly clarifies
the situation since an SHGF is the most important gravitational field involved
in the common statements of the principle of equivalence.
Specific calculations confirm that free fall of an object as seen by an observer
supported in an SHGF is not the same as support of this object in an SHGF
as seen by a freely falling observer. Acceleration is not reciprocal in general
relativity.
Since electromagnetic phenomena are usually not, included in discussions of
the principle of equivalence, it is important to clarify their role. As long as we
do not have a unified field theory which might well predict interference phe-
nomena between gravitational and electromagnetic interactions that are so
far not included in general relativity, the principle of equivalence is postulated
to hold also for charged test particles. l5 The electrodynamics of special relativity
is consequently valid in the local inertial frames of the statements (C). These
statements therefore imply, in particular, that an observer falling freely in an
SHGF will (a) see a similarly falling charge as purely electrostatic and non-
radiating, and (b) see a charge supported in an SHGF radiate according to the
laws of special relativity when applied to hyperbolic motion. Since these two
situations are not related by a Lorentz transformation (under which the radi-
ation rate is invariant), (a) and (b) do not contradict each other. The question
is much more subtle for the observer supported in an SHGF. Since he is non-
inertial his Maxwell equations are only formally identical with those of special
relativity. That they predict radiation from a freely falling charge and no radi-
ation from a supported charge is not obvious. That this is actually the case is
proven explicitly in Section VII. If one argues on the basis of (C,) that this
situation involves an accelerated charge which should always radiate, the
argument is erroneous, because the fact that a charge is accelerated does not
necessarily imply that it radiates, unless the acceleration takes place relative
to an inertial observer. A noninertial observer uses different clocks and yard-
sticks. Thus, even though the charge is accelerated, it follows that, because the
observer is also accelerated, the co-accelerated observer sees no radiation. Since
radiation is not a generally covariant concept the question whether the charge
really radiates is meaningless unless it is referred to a particular coordinate
system. Finally, since the Schwarzschild metric, locally, for small G, and non-
I5 By definition of “test particle” one must ignore here the effect of the particles own
field on its motion (electromagnetic as well as gravitational). But the electromagnetic
self-energy is included as part of its mass which is not supposed to enter its equation of
motion.
188 ROHRLICH

relativistically, is identical with the SHGF metric, the above conclusion also
holds for a charge at rest as seen by an observer in a Schwarzschild field. A
“local” definition of radiation (9) is here essential.
These considerations are important in that they confirm the validity of the
principle of equivalence also for charged test particles. Experimental tests which
confirm the validity of special relativity in various freely falling situations also
confirm the principle of covariance and the definition of inertial systems (C,,
say). They confirm the principle of equivalence only weakly, viz. for flat space.
Strong tests must show that part (A) of that principle also holds when space is
curved. For example, in a static field the freely falling observer must not only
see an SHGF, i.e., the world lines of objects supported in the static field must be
long enough so that they can be distinguished from a congruence of parallel
straight lines.
The experiments by Eiitvijs and by Dicke (1,Z) can provide such a test, since
the earth is rotating and the sun’s gravitational field is changing during the
experiment. A confirmation of the principle of equivalence is the simultaneous
observation of the orbiting of a nonspinning space capsule (which assures free
fall in a field of nonvanishing curvature) and the observation of force-free space
inside this capsule, irrespective of the mass, chemical composition, temperature,
polarization, etc. of the objects in it. The presence of electromagnetic forces
are to be accounted for by special relativity, to arbitrary accuracy, restricted
only by the flatness of space within the capsule. The latter can be determined
by the observer on earth. The fact that all this holds for all space capsules con-
firms that the laws of nature are independent of space and time (at least within
certain limits), i.e., it confirms the general principle of relativity as expressed
by the principle of covariance.le
Similar tests are provided by changing the bodies used in the Eotvos-Dicke
experiment not only with respect to chemical composition, but also with respect
to types of energy (kinetic energy of heat motion, potential energy of aligned
nuclei, etc.) since the equivalent of every type of energy has to obey the equiv-
alence principle. Such tests were recently proposed by Morgan and Peres
(17). However, their argument concerning the velocity dependence of the forces
in freely falling objects as seen by an observer supported in a Meller SHGF
have nothing to do with these tests. In fact, the arbitrariness of u(z) can be used
to eliminate the velocity dependence. There is no reason to prefer Mprller’s
choice over any other U(Z).
Of course, any test of the geodesic equation as the equation of motion of
freely falling test particles is a direct test of the equivalence principle. However,
such a test involves usually also other features of the theory of relativity.
The principle of equivalence (part (A)) is satisfied when light rays follow null
16 We also emphasize that this typical inertial observer is accelerated with respect to
the fixed stars.
 PRINCIPLE OF EQUIVALENCE 189

geodesics. Their bending depends on the curvature of space, i.e., on the par-
ticular type of true gravitational field (despite the fact that within the experi-
mental accuracy to which the equation can be confirmed only the first power of
G enters!) and is not predicted quantitatively (nor excluded qualitatively) by
this principle (18, 19). An analogous situation is the perihel precession of plane-
tary motion (which does measure terms in G2, however). It can also be seen
that no form of the SHGF can ever predict the same light bending as the
Schwarzschild metric, even in first approximation in G, because the SHGF ap-
proximates that metric only nonrelativistically which is insufficient for the world
line of light.
The gravitational shift of spectral lines is both the weakest and the strongest
of the three Einstein effects. The usual explanation given, which is Einstein’s
original one (20) and which precedes the general theory of relativity, is not an
explanation at all, but is at best a plausibility argument. It says that, if there
were a theory which contained (in suitable limits) Newtonian gravitation
theory, special relativity, and quantum mechanics, then the principle of equiv-
alence (part (A)) will-to first order in g-predict that shift of spectral lines.
All three theories mentioned enter into the argument, but each of these contra-
dicts the other two. Since we do not have a theory which incorporates all three
of them, this effect is strictly speaking unexplained. In this respect the red shift
effect is the strongest of the three Einstein effects.
The red shift was observed in two essentially different ways: by a comparison of
a clock (radiating atom) on a star (sun) with a terrestrial clock, and by a corn-
parison of two clocks at different terrestrial gravitational potentials (21). The
essential difference between these experiments lies in the fact that in the second
experiment (and not in the first) the gravitational field at both clocks can be
eliminated simultaneously by a suitable motion of the observer (coordinate
transformation); it involves an SHGF within the accuracy of the experiment and
is therefore a very weak confirmation of the equivalence principle. In fact, it is
correctly predicted by using the definition (C,) of inertial systems in comparing
any two clocks which are in different gravitational fields (true or apparent)
and which use coordinate systems whose go0to first order in g agree with Min-
kowski space and an SHGF respectively, i.e., which are go0 = -1 and go0=
-(l + Zgz), respectively. Since any theory which incorporates Newtonian
gravitation and special relativity will satisfy these conditions, the red shift,
experiment performed in the laboratory confirms in addition to these theories
little else than the specification of inertial systems according to (C,). It does
prove more than the red shift experiment on a rotating disk (22), since the
latter refers to flat space and an inertial clock,17 but it proves less than the
astronomical red shift observation which involves true gravitational fields.
I7 The experiment is performed in the plane orthogonal to the gravitational field of the
earth with the emitter on the axis of the rotating disk, the observer being on the periphery.
190 ROHRLICH

The comparison of a clock in a satellite and one on the earth would differ
from all the previously performed red-shift experiments in that a noninertial
observer compares a noninertial and an inertial clock in a true gravitational
field.
The principle of equivalence is fully contained in the theory of relativity.
What is not contained in it is the origin of inertia. The theory assures the equality
of inertial and gravitational mass but it does not explain why they are equal.
A possible explanation of the latter would be afforded by Mach’s principle
which is not contained in general relativity. In a theory which contains Mach’s
principle, the principle of equivalence is trivial, since inertia would have a
gravitational origin.
If we accept the equivalence principle in the form proposed above, it is quite
clear that it points to a geometrization of gravitational phenomena, as was
indeed its historic role, while a geometrization of electrodynamics is not indi-
cated; not only is the charge to mass ratio not a universal constant, but the
exact equation of motion of a test charge involves the charge in a nonlinear and
complicated manner Q’S).
Note added in proof: After completion of the above paper an article by T. C.
Bradbury (Ann. Phys. (N. Y.) 19, 323 (1962)) appeared in which the electro-
magnetic field of a charge supported in a Meller SHGF is computed for an
equally supported observer. His results are a special case of our Eqs. (7.9)
to (7.12).

RECEIVED: September 20, 1962

REFERENCES

1. R. H. DICKE, Sci. Am. 206,84 (1961); Science 129,621 (1959).

8. R. H. DICKE, Am. J. Phys. 28, 344 (1960).
8. C. M@LLER, “The Theory of Relativity.” Oxford Univ. Press, London, 1952.
4. V. FOCK, “Theory of Space-Time and Gravitation,” esp. p. 209 ff. Pergamon Press,
London, 1969.
6. J. L. SYNGE, “Relativity: The General Theory,” preface, p. IX. North-Holland Pub-
lishing Company, Amsterdam; Interscience, New York, 1960.
6. T. FULTON AND F. ROHRLICH, Ann. Phys. (N. Y.) 9, 499 (1960).
7. E. L. HILL, Phys. Rev. 72, 143 (1947).
8. F. ROHRLICH, The classical electron. In “Lectures in Theoretical Physics,” Vol. II,
W. E. Brittin and B. W. Downs, eds. Interscience, New York, 1960.
9. F. ROHRLICH, Nuovo Cimento 21, 811 (1961).
10. VACHASPATI AND BALI, Nuovo Cimento 21, 442 (1961).
11. A. LICHNEROWICZ, “Theories Relativistes de la Gravitation et de I’Electromagnetisme.”
Masson, Paris, 1955.
1.9. C. M$LLER, Kgl. Danske Videnskab. Selskab 20, No. 19 (1943).
19. See, for example, T. FULTON, F. ROHRLICH, AND L. WITTEN, Rev. Mod. Phys. 34, 442
(1962).
 PRINCIPLE OF EQUIVALENCE 191

14. T. FULTON, F. ROHRLICH, AND L. WITTEN, Nuovo Cimento 24,652 (1962).

16. E. T. WHITTAKER, Proc. Roy. Sot. (London) 116, 720 (1927).
16. N. ROSEN, Ann. Phys. (N. Y.) 17, 269 (1962).
IY. T. A. MORGAN AND A. PERES, Phys. Rev. Letters 9, 79 (1962).
18. A. S~HILD, Am. J. Phys. 28, 778 (1960).
19. R. U. SEXL, Z. Physik 167, 265 (1962).
20. A. EINSTEIN, Ann. Physik 36, 898 (1911).
21. T. F. CRANSHAW, S. P. SCHIFFER, AND A, B. WHITEHEAD, Phys. Rev. Letters 4,163 (1960);
R. V. POUND AND G. A. REBKA, Phys. Rev. Letters 4, 337, 397 (1960).
22. H. J. HAY, J. P. SCZHIFFER, T. E. CFUNSHAW, AND P. A. EGELSTAFF, Phys. Rev. Letters
4, 165 (1960).
23. F. ROHRLICH, Ann. Phys. (N. Y.) 13, 93 (1961).