Schutz B. A first course in general relativity

Подождите немного. Документ загружается.

113 5.1 On the relation of gravitation to curvature

obtain great precision in the measurement of the difference ν



− ν produced in a photon

climbing a distance h = 22.5 m. Eq. (5.1) was veriﬁed to approximately 1% precision.

With improvements in technology between 1960 and 1990, the gravitational redshift

moved from being a small exotic correction to becoming an effect that is central to society:

the GPS navigation system incorporates vital corrections for the redshift, in the absence of

which it would not remain accurate for more than a few minutes. The system uses a network

of high-precision atomic clocks in orbiting satellites, and navigation by an apparatus on

Earth is accomplished by reading the time-stamps on signals received from ﬁve or more

satellites. But, as we shall see below, the gravitational redshift implies that time itself runs

slightly faster at the higher altitude than it does on the Earth. If this were not compensated

for, the ground receiver would soon get wrong time-stamps. The successful operation of

GPS can be taken to be a very accurate veriﬁcation of the redshift. See Ashby (2003) for a

full discussion of relativity and the GPS system.

This experimental veriﬁcation of the redshift is comforting from the point of view of

energy conservation. But it is the death-blow to our chances of ﬁnding a simple, special-

relativistic theory of gravity, as we shall now show.

Nonexistence of a Lorentz frame at rest on Earth

If SR is to be valid in a gravitational ﬁeld, it is a natural ﬁrst guess to assume that the

‘laboratory’ frame at rest on Earth is a Lorentz frame. The following argument, due orig-

inally to Schild (1967), easily shows this assumption to be false. In Fig. 5.2 we draw a

spacetime diagram in this hypothetical frame, in which the one spatial dimension plotted

is the vertical one. Consider light as a wave, and look at two successive ‘crests’ of the

wave as they move upward in the Pound–Rebka–Snider experiment. The top and bottom

of the tower have vertical world lines in this diagram, since they are at rest. Light is shown

moving on a wiggly line, and it is purposely drawn curved in some arbitrary way. This is

to allow for the possibility that gravity may act on light in an unknown way, deﬂecting

Bottom Top

First crest

Second crest

Δ t

bot

Δ t

bot

Figure 5.2

In a time-independent gravitational ﬁeld, two successive ‘crests’ of an electromagnetic wave

must travel identical paths. Because of the redshift (Eq. (5.1)) the time between them at the top

is larger than at the bottom. An observer at the top therefore ‘sees’ a clock at the bottom running

slowly.

114 Prefacetocurvature

it from a null path. But no matter how light is affected by gravity the effect must be the

same on both wave crests, since the gravitational ﬁeld does not change from one time to

another. Therefore the two crests’ paths are congruent, and we conclude from the hypo-

thetical Minkowski geometry that t

top

= t

bottom

. On the other hand, the time between

two crests is simply the reciprocal of the measured frequency t = 1/ν. Since the Pound–

Rebka–Snider experiment establishes that ν

bottom

>ν

top

, we know that t

top

>t

bottom

The conclusion from Minkowski geometry is wrong, and the reference frame at rest on

Earth is not a Lorentz frame.

Is this the end, then, of SR? Not quite. We have shown that the Lorentz frame at rest on

Earth is not inertial. We have not shown that there are no inertial frames. In fact there are

certain frames which are inertial in a restricted sense, and in the next paragraph we shall

use another physical argument to ﬁnd them.

The principle of equivalence

One important property of an inertial frame is that a particle at rest in it stays at rest if no

forces act on it. In order to use this, we must have an idea of what a force is. Ordinarily,

gravity is regarded as a force. But, as Galileo demonstrated in his famous experiment at the

Leaning Tower of Pisa, gravity is distinguished from all other forces in a remarkable way:

all bodies given the same initial velocity follow the same trajectory in a gravitational ﬁeld,

regardless of their internal composition. With all other forces, some bodies are affected

and others are not: electromagnetism affects charged particles but not neutral ones, and the

trajectory of a charged particle depends on the ratio of its charge to its mass, which is not

the same for all particles. Similarly, the other two basic forces in physics – the so-called

‘strong’ and ‘weak’ interactions – affect different particles differently. With all these forces,

it would always be possible to deﬁne experimentally the trajectory of a particle unaffected

by the force, i.e. a particle that remained at rest in an inertial frame. But, with gravity, this

does not work. Attempting to deﬁne an inertial frame at rest on Earth, then, is vacuous,

since no free particle (not even a photon) could possibly be a physical ‘marker’ for it.

But there is a frame in which particles do keep a uniform velocity. This is a frame which

falls freely in the gravitational ﬁeld. Since this frame accelerates at the same rate as free

particles do (at least the low-velocity particles to which Newtonian gravitational physics

applies), it follows that all such particles will maintain a uniform velocity relative to this

frame. This frame is at least a candidate for an inertial frame. In the next section we will

show that photons are not redshifted in this frame, which makes it an even better candidate.

Einstein built GR by taking the hypothesis that these frames are inertial.

The argument we have just made, that freely falling frames are inertial, will perhaps be

more familiar to the student if it is turned around. Consider, in empty space free of grav-

ity, a uniformly accelerating rocket ship. From the point of view of an observer inside, it

appears that there is a gravitational ﬁeld in the rocket: objects dropped accelerate toward the

rear of the ship, all with the same acceleration, independent of their internal composition.

This has been tested experimentally to extremely high precision in the so-called Eötvös experiment. See Dicke

(1964).

115 5.1 On the relation of gravitation to curvature

Moreover, an object held stationary relative to the ship has ‘weight’ equal to the force

required to keep it accelerating with the ship. Just as in ‘real’ gravity, this force is pro-

portional to the mass of the object. A true inertial frame is one which falls freely toward

the rear of the ship, at the same acceleration as particles. From this it can be seen that

uniform gravitational ﬁelds are equivalent to frames that accelerate uniformly relative to

inertial frames. This is the principle of equivalence between gravity and acceleration, and

is a cornerstone of Einstein’s theory. Although Galileo and Newton would have used differ-

ent words to describe it, the equivalence principle is one of the foundations of Newtonian

gravity.

In more modern terminology, what we have described is called the weak equivalence

principle, ‘weak’ because it refers to the way bodies behave only when inﬂuenced by

gravity. Einstein realized that, in order to create a full theory of gravity, he had to extend this

to include the other laws of physics. What we now call the Einstein equivalence principle

says that we can discover how all the other forces of nature behave in a gravitational ﬁeld

by postulating that the differential equations that describe the laws of physics have the

same local form in a freely falling inertial frame as they do in SR, i.e. when there are

no gravitational ﬁelds. We shall use this stronger form of the principle of equivalence

in Ch. 7.

Before we return to the proof that freely falling frames are inertial, even for photons,

we must make two important observations. The ﬁrst is that our arguments are valid only

locally – since the gravitational ﬁeld of Earth is not uniform, particles some distance away

do not remain at uniform velocity in a particular freely falling frame. We shall discuss

this in some detail below. The second point is that there are of course an inﬁnity of freely

falling frames at any point. They differ in their velocities and in the orientation of their

spatial axes, but they all accelerate relative to Earth at the same rate.

The redshift experiment again

Let us now take a different point of view on the Pound–Rebka–Snider experiment. Let us

view it in a freely falling frame, which we have seen has at least some of the characteristics

of an inertial frame. Let us take the particular frame that is at rest when the photon begins

its upward journey and falls freely after that. Since the photon moves a distance h, it takes

time t = h to arrive at the top. In this time, the frame has acquired velocity gh downward

relative to the experimental apparatus. So the photon’s frequency relative to the freely

falling frame can be obtained by the redshift formula

ν(freely falling)



(apparatus at top)

1 + gh

√

(1 − g

)

= 1 + gh + 0(v

). (5.2)

From Eq. (5.1) we see that if we neglect terms of higher order (as we did to derive

Eq. (5.1)), then we get ν(photon emitted at bottom) = ν(in freely falling frame when pho-

ton arrives at top). So there is no redshift in a freely falling frame. This gives us a sound

basis for postulating that the freely falling frame is an inertial frame.

116 Prefacetocurvature

Local inertial frames

The above discussion suggests that the gravitational redshift experiment really does not

render SR and gravity incompatible. Perhaps we simply have to realize that the frame at

rest on Earth is not inertial and the freely falling one – in which there is no redshift and

so Fig. 5.2 leads to no contradiction – is the true inertial frame. Unfortunately, this doesn’t

completely save SR, for the simple reason that the freely falling frames on different sides

of Earth fall in different directions: there is no single global frame which is everywhere

freely falling in Earth’s gravitational ﬁeld and which is still rigid, in that the distances

between its coordinate points are constant in time. It is still impossible to construct a global

inertial frame, and so the most we can salvage is a local inertial frame, which we now

describe.

Consider a freely falling frame in Earth’s gravitational ﬁeld. An inertial frame in SR

ﬁlls all of spacetime, but this freely falling frame would not be inertial if it were extended

too far horizontally, because then it would not be falling vertically. In Fig. 5.3 the frame is

freely falling at B, but at A and C the motion is not along the trajectory of a test particle.

Moreover, since the acceleration of gravity changes with height, the frame cannot remain

inertial if extended over too large a vertical distance; if it were falling with particles at

one height, it would not be at another. Finally, the frame can have only a limited extent in

time as well, since, as it falls, both the above limitations become more severe due to the

frame’s approaching closer to Earth. All of these limitations are due to nonuniformities in

the gravitational ﬁeld. Insofar as nonuniformities can be neglected, the freely falling frame

can be regarded as inertial. Any gravitational ﬁeld can be regarded as uniform over a small

enough region of space and time, and so we can always set up local inertial frames. They

are analogous to the MCRFs of ﬂuids: in this case the frame is inertial in only a small

region for a small time. How small depends on (a) the strength of the nonuniformities of

the gravitational ﬁeld, and (b) the sensitivity of whatever experiment is being used to detect

noninertial properties of the frame. Since any nonuniformity is, in principle, detectable, a

frame can only be regarded mathematically as inertial in a vanishingly small region. But

for current technology, the freely falling frames near the surface of Earth can be regarded

Earth

Figure 5.3

A rigid frame cannot fall freely in the Earth’s ﬁeld and still remain rigid.

117 5.1 On the relation of gravitation to curvature

as inertial to a high accuracy. We will be more quantitative in a later chapter. For now, we

just emphasize the mathematical notion that any theory of gravity must admit local inertial

frames: frames that, at a point, are inertial frames of SR.

Tidal forces

Nonuniformities in gravitational ﬁelds are called tidal forces, since they are the ones that

raise tides. (If Earth were in a uniform gravitational ﬁeld, it would fall freely and have

no tides. Tides bulge due to the difference of the Moon’s and Sun’s gravitational ﬁelds

across the diameter of Earth.) We have seen that these tidal forces prevent the construction

of global inertial frames. It is therefore these forces that are regarded as the fundamental

manifestation of gravity in GR.

The role of curvature

The world lines of free particles have been our probe of the possibility of constructing

inertial frames. In SR, two such world lines which begin parallel to each other remain

parallel, no matter how far they are extended. This is exactly the property that straight lines

have in Euclidean geometry. It is natural, therefore, to discuss the geometry of spacetime

as deﬁned by the world lines of free particles. In these terms, Minkowski space is a ﬂat

space, because it obeys Euclid’s parallelism axiom. It is not a Euclidean space, however,

since its metric is different: photons travel on straight world lines of zero proper length. So

SR has a ﬂat, nonEuclidean geometry.

Now, in a nonuniform gravitational ﬁeld, the world lines of two nearby particles which

begin parallel do not generally remain parallel. Gravitational spacetime is therefore not ﬂat.

In Euclidean geometry, when we drop the parallelism axiom, we get a curved space. For

example, the surface of a sphere is curved. Locally straight lines on a sphere extend to great

circles, and two great circles always intersect. Nevertheless, sufﬁciently near to any point,

we can pretend that the geometry is ﬂat: the map of a town can be represented on a ﬂat sheet

of paper without signiﬁcant distortion, while a similar attempt for the whole globe fails

completely. The sphere is thus locally ﬂat. This is true for all so-called Riemannian

spaces:

they all are locally ﬂat, but the locally straight lines (called geodesics) do not usually remain

parallel.

Einstein’s important advance was to see the similarity between Riemannian spaces

and gravitational physics. He identiﬁed the trajectories of freely falling particles with the

geodesics of a curved geometry: they are locally straight since spacetime admits local iner-

tial frames in which those trajectories are straight lines, but globally they do not remain

parallel.

B. Riemann (1826–66) was the ﬁrst to publish a detailed study of the consequences of dropping Euclid’s

parallelism axiom.

118 Prefacetocurvature

We shall follow Einstein and look for a theory of gravity that uses a curved spacetime

to represent the effects of gravity on particles’ trajectories. To do this we shall clearly

have to study the mathematics of curvature. The simplest introduction is actually to study

curvilinear coordinate systems in a ﬂat space, where our intuition is soundest. We shall

see that this will develop nearly all the mathematical concepts we need, and the step to a

curved space will be simple. So for the rest of this chapter we will study the Euclidean

plane: no more SR (for the time being!) and no more indeﬁnite inner products. What we

are after in this chapter is parallelism, not metrics. This approach has the added bonus of

giving a more sensible derivation to such often-mysterious formulae as the expression for

∇

in polar coordinates!

5.2 Tensor algebra in polar coordinates

Consider the Euclidean plane. The usual coordinates are x and y. Sometimes polar

coordinates {r, θ} are convenient:

r = (x

+ y

)

1/2

, x = r cos θ,

θ = arctan(y/x), y = r sin θ.



(5.3)

Small increments r and θ are produced by x and y according to

r =

x +

y = cos θx +sin θy,

θ =−

x +

y =−

sin θx +

cos θy,

⎫

⎬

⎭

(5.4)

which are valid to ﬁrst order.

It is also possible to use other coordinate systems. Let us denote a general coordinate

system by ξ and η:

ξ = ξ(x, y), ξ =

∂ξ

∂x

x +

∂ξ

∂y

y,

η = η(x, y), η =

∂η

∂x

x +

∂η

∂y

y.

⎫

⎪

⎬

⎪

⎭

(5.5)

In order for (ξ , η) to be good coordinates, it is necessary that any two distinct points

, y

) and (x

, y

) be assigned different pairs (ξ

, η

) and (ξ

, η

), by Eq. (5.5). For

instance, the deﬁnitions ξ = x, η = 1 would not give good coordinates, since the distinct

points (x = 1, y = 2) and (x = 1, y = 3) both have (ξ = 1, η = 1). Mathematically, this

requires that if ξ = η = 0inEq.(5.5), then the points must be the same, or x =

y = 0. This will be true if the determinant of Eq. (5.5) is nonzero,

det



∂ξ/∂x ∂ξ/∂y

∂η/∂x ∂η/∂y



= 0. (5.6)

This determinant is called the Jacobian of the coordinate transformation, Eq. (5.5). If the

Jacobian vanishes at a point, the transformation is said to be singular there.

119 5.2 Tensor algebra in polar coordinates

Vectors and one-forms

The old way of deﬁning a vector is to say that it transforms under an arbitrary coordinate

transformation in the way that the displacement transforms. That is, a vector

→



r can be

represented

as a displacement (x, y), or in polar coordinates (r, θ), or in general

(ξ , η). Then it is clear that for small (x, y) we have (from Eq. (5.5))

⎛

⎝

ξ

η

⎞

⎠

⎛

⎝

∂ξ/∂x ∂ξ/∂y

∂η/∂x ∂η/∂y

⎞

⎠

⎛

⎝

x

y

⎞

⎠

. (5.7)

By deﬁning the matrix of transformation

(



) =



∂ξ/∂x ∂ξ/∂y

∂η/∂x ∂η/∂y



, (5.8)

we can write the transformation for an arbitrary



V in the same manner as in SR



= 



, (5.9)

where unprimed indices refer to (x, y) and primed to (ξ , η), and where indices can only

take the values 1 and 2. A vector can be deﬁned as something whose components transform

according to Eq. (5.9). There is a more sophisticated and natural way, however. This is the

modern way, which we now introduce.

Consider a scalar ﬁeld φ. Given coordinates (ξ , η) it is always possible to form the

derivatives ∂φ/∂ξ and ∂φ/∂η.Wedeﬁne the one-form

dφ to be the geometrical object

whose components are

dφ → (∂φ/∂ξ, ∂φ/∂η) (5.10)

in the (ξ, η) coordinate system. This is a general deﬁnition of an inﬁnity of one-forms, each

formed from a different scalar ﬁeld. The transformation of components is automatic from

the chain rule for partial derivatives:

∂φ

∂ξ

∂x

∂ξ

∂φ

∂x

∂y

∂ξ

∂φ

∂y

, (5.11)

and similarly for ∂φ/∂η. The most convenient way to write this in matrix notation is as a

transformation on row-vectors,

(∂φ/∂ξ ∂φ/∂η) =

(∂φ/∂x ∂φ/∂y)



∂x/∂ξ ∂x/∂η

∂y/∂ξ ∂y/∂η



, (5.12)

because then the transformation matrix for one-forms is deﬁned by analogy with Eq. (5.8)

as a set of derivatives of the (x, y)-coordinates by the (ξ , η)-coordinates:

(



) =



∂x/∂ξ ∂x/∂η

∂y/∂ξ ∂y/∂η



. (5.13)

We shall denote Euclidean vectors by arrows, and we shall use Greek letters for indices (numbered 1 and 2) to

denote the fact that the sum is over all possible (i.e. both) values.

120 Prefacetocurvature

Using this matrix the component-sum version of the transformation in Eq. (5.12)is

(

dφ)



= 



(

dφ)

. (5.14)

Note that the summation in this equation is on the ﬁrst index of the transformation matrix,

as we expect when a row-vector pre-multiplies a matrix.

It is interesting that in SR we did not have to worry about row-vectors, because the

simple Lorentz transformation matrices we used were symmetric. But if we want to go

beyond even the simplest situations, we need to see that one-form components are ele-

ments of row-vectors. However, matrix notation becomes awkward when we go beyond

tensors with two indices. In GR we need to deal with tensors with four indices, and some-

times even ﬁve. As a result, we will normally express transformation equations in their

algebraic form, as in Eq. (5.14); students will not see much matrix notation later in this

book.

What we have seen in this section is that, in the modern view, the foundation of tensor

algebra is the deﬁnition of a one-form. This is more natural than the old way, in which

a single vector (x, y) was deﬁned and others were obtained by analogy. Here a whole

class of one-forms is deﬁned in terms of derivatives, and the transformation properties of

one-forms follow automatically.

Now a vector is deﬁned as a linear function of one-forms into real numbers. The impli-

cations of this will be explored in the next paragraph. First we just note that all this is the

same as we had in SR, so that vectors do in fact obey the transformation law, Eq. (5.9). It is

of interest to see explicitly that (



) and (



) are inverses of each other. The product

of the matrices is



∂ξ/∂x ∂ξ/∂y

∂η/∂x ∂η/∂y



∂x/∂ξ ∂x/∂η

∂y/∂ξ ∂y/∂η





∂ξ

∂x

∂ξ

∂y

∂ξ

∂x

∂η

∂ξ

∂y

∂η

∂x

∂ξ

∂η

∂y

∂ξ

∂η

∂x

∂η

∂y

∂η



. (5.15)

By the chain rule this matrix is



∂ξ/∂ξ ∂ξ/∂η

∂η/∂ξ ∂η/∂η







, (5.16)

where the equality follows from the deﬁnition of a partial derivative.

Curves and vectors

The usual notion of a curve is of a connected series of points in the plane. This we shall

call a path, and reserve the word curve for a parametrized path. That is, we shall follow

modern mathematical terminology and deﬁne a curve as a mapping of an interval of the

real line into a path in the plane. What this means is that a curve is a path with a real number

associated with each point on the path. This number is called the parameter s. Each point

has coordinates which may then be expressed as a function of s:

121 5.2 Tensor algebra in polar coordinates

Curve :{ξ = f (s), η = g(s), a  s  b} (5.17)

deﬁnes a curve in the plane. If we were to change the parameter (but not the points) to



= s



(s), which is a function of the old s, then we would have

{ξ = f



), η = g



), a



 s



 b



}, (5.18)

where f



and g



are new functions, and where a



= s



(a), b



= s



(b). This is, mathematically,

a new curve, even though its image (the points of the plane that it passes through) is the

same. So there is an inﬁnite number of curves having the same path.

The derivative of a scalar ﬁeld φ along the curve is dφ/ds. This depends on s,soby

changing the parameter we change the derivative. We can write this as

dφ/ds =

dφ,



V, (5.19)

where



V is the vector whose components are (dξ/ds,dη/ds). This vector depends only on

the curve, while

dφ depends only on φ. Therefore



V is a vector characteristic of the curve,

called the tangent vector. (It clearly lies tangent to curve: see Fig. 5.4.) So a vector may be

regarded as a thing which produces dφ/ds,givenφ. This leads to the most modern view,

that the tangent vector to the curve should be called d/ds. Some relativity texts occasionally

use this notation. For our purposes, however, we shall just let



V be the tangent vector whose

components are (dξ/ds,dη/ds). Notice that a path in the plane has, at any point, an inﬁnity

of tangents, all of them parallel but differing in length. These are to be regarded as vectors

tangent to different curves, curves that have different parametrizations in a neighborhood of

that point. A curve has a unique tangent, since the path and parameter are given. Moreover,

even curves that have identical tangents at a point may not be identical elsewhere. From the

Taylor expansion ξ (s +1) ≈ ξ (s) + dξ/ds,

we see that



V(s) stretches approximately from

s to s + 1 along the curve.

Now, it is clear that under a coordinate transformation s does not change (its deﬁnition

had nothing to do with coordinates) but the components of



V will, since by the chain rule



dξ/ds

dη/ds





∂ξ/∂x ∂ξ/∂y

∂η/∂x ∂η/∂y



dx/ds

dy/ds



. (5.20)

This is the same transformation law as we had for vectors earlier, Eq. (5.7).

s = 8

= 7

= 6

= 5

= 4

= 3

= 2

= 1

→

Figure 5.4

A curve, its parametrization, and its tangent vector.

122 Prefacetocurvature

To sum up, the modern view is that a vector is a tangent to some curve, and is the

function that gives dφ/ds when it takes

dφ as an argument. Having said this, we are now

in a position to do polar coordinates more thoroughly.

Polar coordinate basis one-forms and vectors

The bases of the coordinates are clearly

e



= 



e

= 

e

+ 

e

(5.21)

∂x

∂r

e

∂y

∂r

e

= cos θ e

+ sin θ e

, (5.22)

and, similarly,

e

∂x

∂θ

e

∂y

∂θ

e

=−r sin θ e

+ r cos θ e

. (5.23)

Notice in this that we have used, among others,



∂x

∂r

. (5.24)

Similarly, to transform the other way we would need



∂r

∂x

. (5.25)

The transformation matrices are exceedingly simple: just keeping track of which index is

up and which is down gives the right derivative to use.

The basis one-forms are, analogously,

dθ =

∂θ

∂x

dx +

∂θ

∂y

dy,

=−

sin θ

dx +

cos θ

dy. (5.26)

(Notice the similarity to ordinary calculus, Eq. (5.4).) Similarly, we ﬁnd

dr = cos θ

dx +sin θ

dy. (5.27)

We can draw pictures of the bases at various points (Fig. 5.5). Drawing the basis vectors

is no problem. Drawing the basis one-forms is most easily done by drawing surfaces of

constant r and θ for

dr and

dθ. These surfaces have different orientations in different places.

There is a point of great importance to note here: the bases change from point to point.

For the vectors, the basis vectors at A in Fig. 5.5 are not parallel to those at C.Thisis

because they point in the direction of increasing coordinate, which changes from point to