Volker Schlue
Department of Mathematics, University of Toronto, 40 St George Street, Toronto, Ontario M5S 2E4, Canada E-mail address: [email protected]
Preface
I have prepared this manuscript for a course on general relativity that I taught at the University of Toronto in the winter semester 2013. It is based on notes that I took as a student at ETH Z¨urich in a lecture course that Demetrios Christodoulou gave in the fall of 2005. (Chapters that are in significant parts transcribed from these notes will be marked by a .) His teaching and understanding of general relativity made a great impression∗ on me as a student, and I hope that these lecture notes can contribute to a wider circulation of his approach. I am very grateful to Andr´eLisibach, who provided me with most of the exer- cises, as well as complementary notes from a similar course that Prof. Christodoulou gave last fall, which were useful for the preparation of the present manuscript.
Toronto, January 2014 V. Schlue
iii
Contents
Preface iii Chapter 1. The equivalence principle and its consequences ∗ 1 1. Classical theory of gravitation 2 2. Special Relativity 5 3. Geodesic correspondence 20 Chapter 2. Einstein’s field equations ∗ 47 1. Einstein’s field equations in the presence of matter 50 2. Action Principle 66 3. The material manifold 74 4. Cosmological constant 79 Chapter 3. Spherical Symmetry 81 1. Einstein’s field equations in spherical symmetry 81 2. Schwarzschild solution 86 3. General properties of the area radius and mass functions 90 4. Spherically symmetric spacetimes with a trapped surface 92 Chapter 4. Dynamical Formulation of the Einstein Equations ∗ 97 1. Decomposition relative to the level sets of a time function 97 2. Slow Motion Approximation 115 3. Gravitational Radiation 125 Bibliography 149
v
CHAPTER 1
The equivalence principle and its consequences ∗
There are two motivations for the General Theory of Relativity. (1) Extension of the principle of Special Relativity (invariance of the physical laws under change from one inertial system to another — one such system relative to another is in a state of uniform motion) to all systems of reference in any state of motion whatsoever. (2) Establish a theory of gravitation which is in accordance with Special Relativity. Einstein’s Equivalence Principle relates the two motivations. Remark. (1) is sometimes referred to as the requirement that the laws of physics should be valid in any system of spacetime coordinates. (General Covari- ance.) In fact, however, (1) refers to the requirement of invariance of the physical laws under change of physical description, from one relative to one set of observers to another relative to another set of such observers. The history of an observer is represented in relativity by a timelike curve. A family of observers is given by a family of timelike curves that do not intersect each other. Each set of curves is a foliation of a given spacetime region. See Fig. 1.
Figure 1. Two families of observers. The history of each observer is a timelike curve.
1 2 1. THE EQUIVALENCE PRINCIPLE
Notes on Special Relativity An inertial system is a system of reference relative to which any mass moves on a straight line if no external forces act upon it. In view of Newton’s Laws set in Euclidean space, any such system is equivalent to (is not distinguished from) any other system of reference in uniform relative motion, which leads to the principle of Galilean relativity. The corre- sponding transformation laws leave time unchanged (up to translations) and thus entail surfaces of absolute simultaneity (level sets of time) which separate the future from the past. See Fig. 2. Special Relativity is based on the premise that light, which prior to Ein- stein had been assumed to propagate in a medium at rest relative to a distinguished frame of reference, in fact propagates in the absence of any such medium on cocentric spheres with respect to any inertial system when emitted from a point source; in other words the speed of light is finite and the same in any inertial system. Einstein realized this obser- vation can only be reconciled if we give up the notion of absolute simultaneity in favour of a notion of simultaneity relative to the observer such that the propagation of light is independent of the system of reference, c.f. Fig. 2. This is precisely achieved in the theory of Special Relativity according to which space and time are unified in Minkowski space (R3+1, m), a 3 + 1-dimensional linear space endowed with a quadratic form m of index 1. A frame of reference corresponds to the choice of a unit timelike vector e, m(e, e) = 1, − which determines a unique spacelike hyperplane Σe = X : m(e, X) = 0 with positive- definite induced metric m 0, a 3-dimensional Euclidean{ space consisting} of all events |Σe ≥ considered simultaneous by observers on Σe at rest relative to the frame defined by e, namely observers whose “world-lines” are straight lines with tangent vector e. The set of null vectors L at a point p, m(L, L) = 0, form a “light cone” consisting of all events reached by light emitted from and received at p. The light cone takes the role of level sets of absolute time in Galilean physics in the sense that it separates the past from the future: The set of time-like vectors T at a point p, m(T,T ) < 0, the interior of the light cone, has two disconnected components referred to as future- and past-directed time-like vec- tors pointing to events that can possibly be influenced by an event at p, or have possibly influenced p, respectively. The exterior of the cone at p, the set of spacelike vectors X at p, m(X,X) > 0, is “causally disconnected” from p, given that no speed of propagation of any physical action whatsoever has ever been observed to surpass the speed of light. The “Principle of special relativity” asserts that all physical phenomena occur in one frame of reference as they do with respect to another, and in principle cannot be used to distinguish a certain frame of reference, i.e. a specific unit time-like vector e. The elements of the orthogonal group with respect to the Minkowski metric m, m(OX,OY ) = m(X,Y ), are called Lorentz transformations, and a “proper” (time-orientation preserving) subgroup mediates in particular the change of one frame of reference to another: e e0 = Oe, m(Oe, Oe) = m(e, e) = 1. Thus the validity of the principle of special relativity7→ requires the invariance of all physical− laws under Lorentz transformations.
1. Classical theory of gravitation A postulate of the classical theory (mechanics: motion of bodies in a given force field & gravitation: the gravitational force field of a given distribution of bodies) is the equality (up to a universal constant depending on the choice of physical units) of the passive gravitational mass and the inertial mass. 1. CLASSICAL THEORY OF GRAVITATION 3
Figure 2. Propagation of light in Galilean and Special Relativity.
ψ: Newtonian gravitational potential. mG ψ(t, x): Gravitational force acting on a test body of gravitational − ∇ mass mG at time t, the instantaneous position of the body at time t R 3 ∈ being x R . mG is thus a concept of the classical theory of gravitation. x = x(t): Motion∈ of a test body. If a body is subjected to a given force F when at position x and time t, then its acceleration is: d2x 1 (1.1) 2 (t) = F (x, t) dt mI
This law (Newton’s Second Law of Motion) defines the inertial mass mI . The postulate is
(1.2) mG = mI . In the classical framework there is no explanation for this equality. Set
(1.3) F = mG ψ(t, x(t)). − ∇ Then mG = mI implies that these cancel in the equations of motion: d2x (1.4) (t) = ψ(t, x(t)) dt2 −∇ That is, the motion of the test body in a given gravitational field is independent of the test body.
1.1. Tidal forces. Consider the gravitational field in a small space-time re- gion, the neighborhood of an event (t0, x0), then a0 = ψ(t0, x0) is the common −∇ acceleration of all test bodies in this region. Let x = x0(t) be the history of the test 0 mass which we take as a reference mass. In the new reference system x = x x0 the origin is translated at each time to the instantaneous position of the reference− mass. (See Fig. 3, 4.) Note that this is an accelerated reference system. 4 1. THE EQUIVALENCE PRINCIPLE
t
2 x0
x2 1 x0
x1
x = x0(t)
Figure 3. History of a reference mass.
t
x0 x x0(t) = 00
x2
x1 Figure 4. Reference coordinate system with origin at each time at the instantaneous postion of the reference mass.
Now consider the motion of another, arbitrary test body in this new reference system: x0 = x0(t).
2 0 d x 0 (t) = ψ t, x0(t) + x (t) a0(t) dt2 −∇ − (1.5) 0 = ψ t, x0(t) + x (t) + ψ t, x0(t) −∇ ∇ 2 0 0 2 = ψ t, x0(t) x (t) + ( x (t) ) −∇ · O | | In terms of rectangular components, and to linear order in x0(t) this equation reads
2 0i 3 2 d x X ∂ ψ 0j (1.6) (t) = t, x0(t) x (t) dt2 − ∂xi∂xj · j=1 and is called the tidal equation. 2. SPECIAL RELATIVITY 5
R
P 0: Periphery
Figure 5. Measurement of P 0/R in the rotating system.
Remark. This equation governs the distortion of a dust cloud in time: tidal distortions. Take here the reference mass to be the test particle occupying the center of mass of a small dust cloud. 1.2. Equivalence principle. (1) In a freely falling reference frame the gravitational force itself is elim- inated. Therefore such a system is equivalent to an inertial reference system in the absence of gravity. What remains however is the differen- tial of the gravitational field which causes tidal distortions. (2) Similarly, an accelerated reference frame in the absence of gravitational fields is equivalent to a stationary reference frame in a gravitational field. As a consequence of the equivalence principle, a theory extending the theory of special relativity must include the theory of gravitation.
2. Special Relativity We shall see that the equivalence (2) leads to Riemannian geometry.
2.1. Uniformly rotating system. Consider a uniformly rotating reference system. 2.1.1. Preliminary considerations. We can ignore the space dimension per- pendicular to the rotating plane. We have a stationary system x, and a reference system x0 obtained from x by rotation with constant angular velocity ω. Consider a circle of radius x0 = R with periphery P 0. What is P 0/R as measured in the rotating system?| Suppose| we use rods of unit length in the rotating system for measurement. As seen in the stationary system, the rods layed along the radius are also of unit length but those along the periphery experience a Lorentz-contraction so their length is only √1 v2, where v = ωR. (See Fig. 5.) However, in the stationary system the periphery− of x = x0 = R is P = 2πR. Since the unit rod of the rotating system, when laid along| | the| | periphery, has less than unit length by the factor √1 v2, it follows that − P P 0 2π (2.1) P 0 = = = > 2π √1 v2 ⇒ R √1 v2 − − 6 1. THE EQUIVALENCE PRINCIPLE
(x0, y0) t = 0
Figure 6. History of a rotating observer with respect to the ro- tating frame. in the rotating system.1 This means that the laws of Euclidean geometry do not hold in the rotating system! 2.1.2. Uniformly rotating observers in Minkowski space. Consider the situa- tion in Minkowski space with 2 space and 1 time dimensions. In rectangular coordinates (t, x, y) of the stationary inertial reference system the metric takes the form: (2.2) ds2 = dt2 + dx2 + dy2 − It is convenient to use polar coordinates (t, r, ϕ) such that x = r cos ϕ, y = r sin ϕ and thus (2.3) ds2 = dt2 + dr2 + r2dϕ2 . − Let us denote by (x0, y0), and (r0, ϕ0) the rectangular and polar spatial coor- dinates, respectively, in the uniformly rotating system.2 A uniformly rotating observer has the history
(2.4) t (t, r = r0, ϕ = ϕ0 + ωt) 7−→ with respect to the stationary system. In the rotating system the observer remains at its initial position (see Fig. 6.)
(2.5) x0 = r0 cos ϕ0 y0 = r0 sin ϕ0 , and has the history
(2.6) t (t, x0 = x0, y0 = y0) . 7−→ The arc element ds2 is a geometric invariant and can be expressed in terms of the (r0, ϕ0) coordinates; (in agreement with general covariance, compare remark on page 1). We find
(2.7) r = r0 ϕ = ϕ0 + ωt
(2.8) dr = dr0 dϕ = dϕ0 + ωdt
1This is a sign of negative curvature. 2 2 2 The domain of these coordinates is the disc ω r0 < 1, because no observer can go faster than the speed of light. 2. SPECIAL RELATIVITY 7 and thus ds2 = dt2 + dr2 + r2dϕ2 − (2.9) 2 2 2 2 = dt + dr + r (dϕ0 + ωdt) − 0 0 or 2 2 2 ωr dϕ0 2 r dϕ (2.10) ds2 = 1 ω2r2 dt 0 + dr2 + 0 0 . − − 0 − 1 ω2r2 0 1 ω2r2 − 0 − 0 Let us denote by q α = 1 ω2r2 : a real number(2.11a) − 0 2 ωr dϕ0 θ = dt 0 : a 1-form(2.11b) − 1 ω2r2 − 0 r2 dϕ2 dσ2 = dr2 + 0 0 : a Riemannian metric(2.11c) 0 1 ω2r2 − 0 then we can write (2.12) ds2 = α2θ2 + dσ2 . − We can view dσ as an arc length element in space. The rays ϕ0 = const. are geodesics of dσ and r0 is arc length along these geodesics:
(2.13) dσ = dr0 : along ϕ0 = const.
The curves r0 = const. are the geodesic circles orthogonal to these rays and r0 dϕ0 (2.14) dσ = p : along r0 = const. 1 ω2r2 − 0 0 So we find for the perimeter P of a circle of radius r0 = R that Z 2π 0 r0dϕ0 2πr0 P 2π (2.15) P 0 = = = = , p 2 p 2 2 2 0 1 ω2r 1 ω2r ⇒ R √1 ω R − 0 − 0 − which coincides with (2.1), (and shows that dσ2 is a Non-Euclidean metric). We are led to think that the coordinates (r0, ϕ0) have no physical meaning in the rotating system. However, what can be measured is the arc length element dσ in (2.12). Any such measurement of arclength in space is carried out simul- taneously as judged by the rotating observers, which brings us to the question of simultaneity in the rotating system. An inertial observer corresponds to a straight timelike line in Minkowski space. The set of events which the observer considers simultaneous with an event p of his own is the spacelike hyperplane Σ through p orthogonal (with respect to (2.2)) to the line. (See Fig. 7.) A family of observers defining an inertial system is a family of parallel timelike straight lines. The simultaneous hyperplanes Σ will be parallel and the same at each point. The family of accelerated observers defining the rotating system is (2.4) as shown in Fig. 8. The hyperplane of locally simultaneous events is different at each point. 8 1. THE EQUIVALENCE PRINCIPLE
p Σ
Figure 7. Set of simultaneous events Σ to p as judged by an inertial observer.
t > 0
Σ p
t = 0
Figure 8. History of a rotating observer.
∂ Consider the vectorfield ∂t . Its integral curves are the histories of the uniformly rotating observers, parametrized by t. In the original (stationary) coordinates (t, r, ϕ) the motion of an observer is (t, r, ϕ) (t + b, r, ϕ + ωb), while in the 7→ coordinates of the rotating system this motion is simply (t, r0, ϕ0) (t+b, r0, ϕ0). Let 7→ ∂ ∂ ∂ (2.16) V = = + ω ∂t r0,ϕ0 ∂t r,ϕ ∂ϕ t,r be the vectorfield generating the above motion (a 1-parameter group of trans- lations), and let ΣV be the hyperplane in the tangent space at a given point consisting of all vectors at that point which are orthogonal to V , namely n o (2.17) ΣV = X : g(V,X) = 0 . Here and in the following we now write g = ds2. Prop 2.1. The 1-form θ is characterized by the following two properties: (1) θ V = 1 · (2) ΣV = ker θ = X : θ X = 0 . · 2. SPECIAL RELATIVITY 9
P
C0
P0 Σ0
Figure 9. Projection map related to uniformly rotating observers.
∂ Proof. By (2.11b) and (2.16) we simply have: θ V = dt ( )r ,ϕ = 1. Now · · ∂t 0 0 dσ2(V, ) = 0 because dσ2 does not contain dt. Therefore · g(X,V ) = α2θ(X)θ(V ) = α2θ(X) = 0 − − is equivalent to θ(X) = 0, since α > 0.
Both α and ΣV have physical significance: α: the rate of a uniformly rotating clock relative to a stationary clock: p ds α = g(V,V ) = . − dt r0,ϕ0
ΣV : local simultaneous space corresponding to an observer with tangent vector V to his history.
While ΣV contains the events locally simultaneous to an event on the history of an observer with velocity V , we shall now address the question of simultaneity globally. In other words, given a set of simultaneous events as judged by the rotat- ing observers, which set of events do they correspond to as seen in the stationary system? We can rephrase this questions as follows. Let C0 be a curve in the hyperplane Σ0 = (t, x, y): t = 0 . In polar coordi- nates we have the parametric equations { }
(2.18) C0 : r0 = r0(λ) , ϕ0 = ϕ0(λ) .
Let P0 be the point on C0 with λ = 0, P0 = (0, r0(0), ϕ0(0))
Question: Find a curve C in spacetime through the point P0 such that C projects to C0 and such that the curve C is horizontal. Projection: Here the projection is a map π of spacetime onto the hyperplane Σ0 defined as follows: π(p) = p0 if p is an event belonging to a history of one of the rotating observers and p0 is the initial position of that observer. In terms of the coordinates (t, r0, ϕ0) this is simply π(t, r0, ϕ0) = (r0, ϕ0). Horizontal: This means that the tangent vector to C at each point belongs to ΣV at that point, i.e. it is orthogonal to the history of the observer through that point. This means that the curve C is locally simultaneous as judged by each observer. 10 1. THE EQUIVALENCE PRINCIPLE
C V ∆t C˙
C0 Σ0
Figure 10. A horizontal curve C that projects to C0.
Remark. The question addresses the problem of measurement in the rotating system. We can think of C0 as a curve on the rotating disc whose length we wish to measure using a ruler on the rotating disc. The curve C describes the collection of events in spacetime that are considered when the measurement of length of C0 occurs simultaneously as seen in the rotating system. The condition that C projects to C0 ensures that the curve whose length is measured is C0 in the rotating system; the condition that C is horizontal means that this measurement occurs locally simultaneous as judged by each uniformly rotating observer on C0. The conditions on C are thus simply:
(1) π(C) = C0 (2) X = C˙ ΣV ∈ By (1) we can assume that the parametric equations for C in the coordinates of the rotating system are:
(2.19) C : t = t(λ) , r0 = r0(λ) , ϕ0 = ϕ0(λ) Now C˙ is a vector with components
dt ∂ dr0 ∂ dϕ0 ∂ (2.20) C˙ = + + , dλ ∂t r0,ϕ0 dλ ∂r0 dλ ∂ϕ0 and we have shown above
(2.21) C˙ ΣV θ C˙ = 0 . ∈ ⇐⇒ · Therefore, in view of (2.11b), condition (2) implies 2 dt ωr dϕ0 (2.22) (λ) 0 (λ) = 0 , t(0) = 0 . dλ − 1 ω2r2 dλ − 0 We can integrate this equation to find Z λ 2 0 ωr (λ ) dϕ0 (2.23) t(λ) = t(0) + 0 (λ0)dλ0 . 1 ω2r2(λ0) dλ 0 − 0 In the case where C0 is a circle, r0(λ) = R, we obtain ωR2 (2.24) t(λ) = t(0) + ϕ(λ) ϕ(0) , 1 ω2R2 − − 2. SPECIAL RELATIVITY 11 so the change of t in going around a full circle is
2πωR2 (2.25) ∆t = . 1 ω2R2 − The curve C does not close! See Fig. 10. Finally, observe that dσ is in fact arc length along C. Remark. This result is contrary to our intuition of rigid space and time. It tells us that it is impossible to measure the length of a curve on the rotating disc in a globally instantaneous manner: When measuring the length of a closed curve C0 on the disc initiating from P0 — in such a way that each observer on C0 (and at rest with respect to the uniformly rotating disc) agrees that the measurement occurs instantaneously according to his local standard of simultaneity — then upon return to P0 the proper time α∆t has elapsed in the course of the measurement. This shows in particular that no global standard of simultaneity exists in the rotating system.
Note on vectorfields, 1-forms and metrics At any point p on a differentiable manifold we have a homeomorphism ϕ from a M neighborhood of p onto Rn, n = dim . M n n Consider the coordinate lines ai : R R , t (0, . . . , t, . . . , 0) in R . The coordinate → 7→ vectorfield ei : i = 1, . . . , n at p is defined to be the tangent vector to the curve Ki = 1 ϕ− a : ◦ i d ei f = f Ki for all f C∞( ) . · dt ◦ t=0 ∈ M ˜ ˜ 1 n Since f Ki = f ai, where f = f ϕ− is a function on R , and ◦ ◦ ◦ ∂f˜ ∂ ei f = 0 , we simply write: ei = . · ∂xi | ∂xi p
These vectors form a basis for the n-dimensional vectorspace Tp : the tangent space to at p. Any tangent vector X T can be expanded in thisM basis: M ∈ pM n X i ∂ X = Xp . ∂xi p i=1
The dual space of T , denoted by T∗ is the set of linear functions on T (each pM pM pM element is also called a covector). While a vectorfield X is an assignment p X(p) of a tangent vector at each point, a 1-form is an assignment p θ(p) of a covector7→ at each point. The tensor product θ2 = θ θ, or more generally θ 7→ξ of two 1-forms, is a bilinear function on T at each p :⊗ ⊗ pM ∈ M θ ξ (X ,Y ) = θ(X )ξ(Y ) . ⊗ p p p p p A special 1-form is given by the differential of function f : R: M → df X = X f . · · 12 1. THE EQUIVALENCE PRINCIPLE
If we denote by xi = ϕi local coordinates in a neighborhood of p then we find that
i ( i ∂ ∂ϕ 0 i = j dx = = δij = 6 . · ∂xj ∂xj 1 i = j
1 n The 1-forms (dx ,..., dx ) form the dual basis in Tp∗ , and any covector ξ Tp∗ can be expressed as M ∈ M n X i ξ = ξi dx p . i=1 P i P j ∂ We have that for any 1-form θ = i θidx and any vector X = j X ∂xj it holds that X θ X = θ Xi . · i i A metric g on is an assignment p g of a non-degenerate symmetric bilinear M 7→ p form gp in Tp at each p . Here symmetric means g(X,Y ) = g(Y,X), and non- degenerate: M ∈ M g (X ,Y ) = 0 for all Y T = X = 0 . p p p p ∈ pM ⇒ p A symmetric bilinear form is often called a quadratic form. Since g is symmetric and bilinear we can expand g at any point p in the basis dxi dxj: ⊗ n X g = g dxi dxj , p ij |p ⊗ |p i,j=1 where it is understood that gij = gji. A Riemannian metric is a positive definite metric. It allows us to assign a magnitude (length) to a vector X, p X = g(X,X) 0 , | | ≥ and an angle α to two non-zero vectors: g(X,Y ) = X Y cos α . | || | A Lorentzian metric is an assignment p g of a non-degenerate quadratic form of 7→ p index 1 in Tp at each point p . Recall that the index of a quadratic form is the largest dimensionM of a subspace where∈ M the quadratic form is negative definite. There is therefore a vector T T such that g (T ,T ) < 0; T is a timelike vector at p. By p ∈ pM p p p p multiplying with a suitable non-zero factor we can assume that Tp is unit: gp(Tp,Tp) = 1. Define Σ to be the the orthogonal complement of T in T . Then g is positive− p p pM p|Σp definite. Choosing an orthonormal basis (E1 p,...,En p) (here n = 3) for Σp any vector X T can be expanded as | | p ∈ pM n X X = X0T + XiE , p p i|p i=1 and we have n 2 X 2 g (X ,X ) = X0 + Xi . p p p − i=1
The set of vectors Xp for which gp(Xp,Xp) > 0 are called spacelike. A Lorentzian metric defines a cone gp(Xp,Xp) = 0 at each point, the set of null vectors at the point p. 2. SPECIAL RELATIVITY 13
ψt(K0) ψt(p0)
K0 p0
Figure 11. Condition of symmetry on the family of observers.
2.2. General accelerated system. In the general situation of a given ac- celerated reference system spacetime is invariant under a 1-parameter group ψt whose orbits are timelike curves. An orbit t ψt(p0), with p0 fixed, is precisely 7→ the history of an observer initiating at the event p0. A family of observers defining a reference system satisfies a symmetry condition: Given a spacelike curve K, the arclength of the curve Kt = ψt(K) equals that of K0 = K for all t. We shall see that the metric takes the form 2 2 2 X i (2.26) ds = α θ + γ , θ = dt + Aidx − i and that (2.25) generalises in this setting to Z i (2.27) ∆t = Ai dx . C0 General Situation. Let us now assume we are in the situation where a 1- parameter group of diffeomorphisms ψt acts on the spacetime manifold such that (1) the orbits t ψt(p) are timelike curves, 7→ (2) ψt acts by isometries. The second condition ensures that the arclength of a spacelike curve K remains unchanged under the flow of ψt; cf. Fig. 11. Moreover, (2) implies ∗ (2.28) ψt g = g and thus d ∗ (2.29) V g = ψt g = 0 , −L dt t=0 where V is the generating vectorfield of ψt, a vectorfield tangential to the orbits: the velocity of the observers. Indeed, given a curve K : λ K(λ)(λ [a, b]) with 7→ ∈ K(a) = p0, K(b) = q0 the condition that the arclength of Kt : λ ψt(K0(λ)) is independent of t reads 7→ Z b q (2.30) L[Kt] = g(K˙ t(λ), K˙ t(λ)) dλ = L[K0]; a | | ˙ ˙ ˙ ˙ ∗ ˙ ˙ Since Kt(λ) = dψt K0(λ) and g(dψt K0(λ), dψt K0(λ)) = (ψt g)(K0(λ), K0(λ)) we conclude on (2.28)· for every t, and· thus by the· definition of the Lie derivative on (2.29). (C.f. note on pull-backs on pg. 17.) 14 1. THE EQUIVALENCE PRINCIPLE
ψt(p0)
p0
H0
Figure 12. Transversal hypersurface H0 to family of observers ψt(p0).
Let H0 be a transversal hypersurface for the set of orbits of ψt (each observer intersects H0 exactly once); see Fig. 12. Here H0 is a spacelike hypersurface in 3+1-dimensional Minkowski space. Every point p can be written as p = ψt(p0) for 1 2 2 some value t where p0 H0, and we can assign to p the coordinates (t, x , x , x ) 1 2 3 ∈ where (x , x , x ) are the spatial coordinates of p0 in Minkowski space. (We can think of H0 as the space of observers. A body with fixed coordinates p0 is at rest in the accelerated system.) In these coordinates ∂ (2.31) V = (where x = (x1, x2, x3) is held fixed) ∂t x and the condition (2.29) simply becomes
∂gµν (2.32) = 0 , ∂t µ ν where gµν = g(∂/∂x , ∂/∂x )(µ, ν = 0, 1, 2, 3) are the metric components in these coordinates (here also x0 = t). Then g takes the form (2.33) g = α2θ2 + γ − where α = α(x) is a positive function, θ is a 1-form 3 X i (2.34) θ = dt + A,A(x) = Ai(x)dx , i=1 and γ = γ(x) is a Riemannian metric in space, 3 X i j (2.35) γ(x) = γij(x)dx dx . i,j=1 As above, α signifies the rate of clocks carried by the observers whose world lines are the group orbits (relative to the parameter t), and θ defines the local standard 2. SPECIAL RELATIVITY 15
C
p = ψt0 (p0)
p0 C0
H0
Figure 13. A horizontal curve C that projects to the closed curve C0 in H0. of simultaneity for the given system of observers. It is a 1-form characterized as above – c.f. Prop. on pg. 8 – by the properties that (1) θ V = 1 (normalisation conditon) · (2) θ X = 0 if and only if X belongs to the hyperplane ΣV of all vectors orthogonal· to V . Let us now turn to the question of simultaneity as in the uniformly rotating case (c.f. pg. 9): Let C0 be a curve in H0 which starts at p0, parametrized by λ such that λ = 0 corresponds to p0. We want to find a curve C through p = ψt0 (p0) which is horizontal and projects to C0. (See Fig. 13.) The parametric equations i i i i for C0 are given by x = x (λ), i = 1, 2, 3, and those for C by t = t(λ), x = x (λ), i = 1, 2, 3 because C projects to C0. The condition that C is horizontal means that 3 dt ∂ X dxi ∂ (2.36) C˙ = + ΣV , dλ ∂t dλ ∂xi ∈ i=1 which implies by the second property of θ that 3 dt X dxi (2.37) 0 = θ C˙ = + Ai(x) . · dλ dλ i=1
Consider now the case that C0 is a closed curve so that for some λ0 > 0: x(λ0) = x(0). The start and end point of C0 are thus both p0. By (2.37) we obtain an increment of 3 Z λ0 Z λ0 i dt X dx (2.38) ∆t = dλ = Ai x(λ) dλ , dλ − dλ 0 0 i=1 or Z (2.39) ∆t = A. − C0 16 1. THE EQUIVALENCE PRINCIPLE
(f(x), x) H00
(0, x) H0
Figure 14. Gauge transformations
Remark. In analogy to electromagnetic theory we may view A as the vector potential, and introduce the magnetic field (2.40) B = dA.
(B is a 2-form, the exterior derivative of a 1-form.) Consider now a surface S0 in H0 whose boundary is C0: ∂S0 = C0. Then by Stokes’ theorem Z Z (2.41) B = A. S0 C0 We may thus interpret the integral (2.39) as the magnetic flux through a surface enclosed by C0.
Remark. The result does not depend on the choice of the hypersurface H0. In 0 fact, another transversal hypersurface H0 is described as a graph t = f(x) in the coordinates constructed above (where H0 = t = 0 ); see Fig. 14. The coordinate transformation { } (2.42) t0 = t f(x) − leaves θ invariant provided the vector potential A is transformed according to (2.43) A0 = A + df , or in coordinates
0 ∂f (2.44) A (x) = Ai(x) + (x); i ∂xi then indeed (2.45) θ0 = dt0 + A0 = dt + A = θ . 2. SPECIAL RELATIVITY 17
The maps (2.43) are precisely the gauge transformations, which — as in electro- magnetic theory — leave the magnetic field unaltered: (2.46) B0 = B.
In particular, in view of (2.41), (2.38) does not depend on the choice of H0.
Finally note that γ measures arclength along C. Since C˙ ΣV we have that ∈ arclength of C w.r.t. g equals arclength of C0 relative to γ, which is
v 3 Z λ0 u i j u X dx dx (2.47) L[C] = t γij x(λ) dλ . dλ dλ 0 i,j=1
Note on pull-backs Let ψ : be a diffeomorphism, and α a 1-form on (in general the target). Recall thatM the→ M differential dψ(p) is a linear mapping of T Minto T , and we can pM ψ(p)M define the pull-back of α, ψ∗α: a 1-form on , by M (ψ∗α) X = α(ψ(p)) (dψ(p) X)(X T ) . · · · ∈ pM The metric g is a quadratic form in the tangent space at each point, and
(ψ∗g)(X,Y ) = g (dψ(p) X, dψ(p) Y )(X,Y T ). ψ(p) · · ∈ pM
Interpretation in Vector Bundles The construction of the curve C – as first discussed on page 9 – has a natural interpretation as a horizontal lift in vector bundles. In the situation (2.26) the spacetime manifold is = R (t, x), where we can think of ( , γ) as the manifold of observers. We haveB a projection× N 3π : which tells N B → N 1 us which observer q passes through a given event p: π(p) = q. The fiber π− (q) is a timelike curve in ∈: the N history of an observer q. V is the tangent vector field to these curves parametrizedB by t, and θ V = 1. Recall that the local simultaneous space at p is given by the null space of θ: · Σ = X T : θ(p) X = 0 p ∈ pB · Let now Xq be a tangent vector to at q. Observe that the projection π is surjective. Moreover dπ(p) is an isomorphism ofN Σ onto T . Therefore given any X T there p qN q ∈ qN exists a unique vector X] Σ such that p ∈ p dπ(p) X] = X ; · p q ] The vector Xp is called the horizontal lift of Xq to p; see Fig. 15. The horizontal lift is determined by the 2 conditions: (1) dπ(p) X] = X · p q (2) θ(p) X] = 0 . · p 18 1. THE EQUIVALENCE PRINCIPLE
1 ] π− (q) C0
p ] Xp
Σp
C0
q
Xq
N Figure 15. Horizontal lift of vectors and curves.
In local coordinates (t, x) we have:
X i ∂ ] ]t ∂ X ]i ∂ Xq = X Tq and X = X + X Tp . q ∂xi q∈ N p p ∂t p p ∂xi p∈ B i i ] i i ] t By (1) we have (Xp) = Xq, and (Xp) is determined from (2): X t X i θ = dt + A (x)dxi : θ X] = X] + A (x) X] = 0 i · p p i p i i Therefore, ] ∂ X i ∂ X = A(q) Xq + X . p − · ∂t p q ∂xi p i With the horizontal lift of tangent vectors known, we can define the horizontal lift of curves: Let C0 : I a curve in the manifold of observers, I R an interval con- → N 1 ⊂ ] taining 0. Let q = C0(0), and p π− (q), then there exists a unique curve C0 such that ∈ ˙] ] C (λ) = (C˙0(λ)) (λ I) . 0 C0(λ) ∈ ] Then the curve C = C0 is precisely the solution to the problem under consideration; c.f. (2.37). That is to say the horizontal lift C(λ) = (t(λ), x(λ)) is the solution to the equation dt(λ) X dxi(λ) = A (C (λ)) , t(0) = 0 . dλ − i 0 dλ i In (2.39) we have shown that the horizontal lift C of a closed curve C0 in does not close in . This is a manifestation of curvature. N B
2.3. Conclusion. We have studied accelerated systems in comoving coordi- nates (t, p0) such that the histories of the corresponing observers at rest in the accelerated sytem are given by t (t, p0). We have seen that following simul- 7→ taneous events in the accelerated system starting at (t0, p0) along a closed curve λ p(λ) with p(0) = p(λ0) = p0 will – in general – take us back to (t1, p0) at 7→ 2. SPECIAL RELATIVITY 19 advanced coordinate time t1 with ∆t = t1 t0 > 0. We conclude that coordinates do not have physical meaning beyond labeling− spacetime events. The differential dt is not the actual time increment measured by the clocks of the accelerating system, and the level sets of t are not surfaces of simultaneity. Instead the rate of clocks in the accelerating system is α, and the relevant time differential α dt. Moreover θ determines a set of locally simultaneous events, and we have shown that it is impossible in general to find a global hypersurface of simultaneity. Simi- larly we have seen that dp0 does not measure elements of arclength in the space of an accelerating system, but instead it is γ. (E.g. for the uniformly rotating system dϕ0 does not give the arc element of circles when multiplied with r0, but instead p 2 2 it is r0dϕ0/ 1 ω r0.) We are led to conclude finally: what does have physi- cal meaning (and− can be measured) are not coordinates but elements of arclength, namely the metrical ground form 3 2 X µ ν (2.48) ds = gµνdx dx . µ,ν=0 Remark. A study of the metrical ground form first appeared in 1828 in C.F. Gauss’ “General Investigations of Curved Surfaces” and stood at the be- ginning of the development of Riemannian Geometry. A. Einstein arrived at the above conclusion in 1915, and General Relativity is formulated in the context of Lorentzian Geometry.
Exercises (Receding observers in Minkowski spacetime): Consider a set of observers in uniform motion receding from a center. In other words consider a family of observers whose histories are straight lines in Minkowski space intersecting in a single point. These are rays contained in the future of the origin: 3 + 0 1 2 3 1+3 0 2 X i 2 0 I0 = (x , x , x , x ) ∈ R : −(x ) + (x ) < 0 and x > 0 . i=1 + (1) Depict the family of receding observers and the set I0 in diagram. + (2) At each p ∈ I0 construct the hyperplane of simultaneous events Σp to p as judged by the observer passing through p. + (3) Do there exist global hypersurfaces of simultaneity in I0 ? In other words, is the distribution of planes + ∆ := {Σp : p ∈ I0 } + integrable? Hint: Consider the hyperboloids Hτ given by 3 + 0 1 2 3 + 0 2 X i 2 2 Hτ := (x , x , x , x ) ∈ I0 :(x ) − (x ) = τ i=1 + + and show that for a given event p ∈ Hτ the tangent plane at p to Hτ coincides with Σp. + + (4) Consider the hyperboloid H1 . Assign to each event p ∈ H1 the polar normal coordinates (χ, θ, φ), where θ and φ are the polar and azimuthal angle of the 20 1. THE EQUIVALENCE PRINCIPLE
+ standard coordinates on the unit sphere, and χ is the geodesic distance on H1 from + (1, 0, 0, 0) to the event p. Verify that the induced metric on H1 is given by dσ2 = dχ2 + sinh2 χ(dθ2 + sin2 θ dφ2). + Show that the Gauss curvature of H1 is equal to −1. + (5) Show that the Minkowski metric can be written in I0 as ds2 = −dτ 2 + τ 2dσ2 . + + That is to say I0 can be foliated by the hyperboloids Hτ and the induced metric on + 2 2 + Hτ is precisely τ dσ . Also show that the Gauss curvature of Hτ is equal to K = −τ −2 . (6) Consider two receding observers. Suppose a light signal is sent from one to the other. Let v = tanh χ be the velocity of the observer emitting the light as measured in the rest frame of the observer receiving the signal. Show that a redshift is observed with ω e = eχ , ωr
where ωe and ωr are the frequencies of the emitted and the received light respectively.
3. Geodesic correspondence In view of the equivalence principle our conclusions for accelerated systems of reference in the absence of a gravitational field apply to stationary reference frames in the presence of a gravitational field. In particular we know that the metrical ground form (2.48) in the presence of a gravitational field is non-trivial. Since we may change to a freely falling reference frame where the gravitational force is eliminated locally we are left with the differential of the gravitational field. In the following we shall find the laws that govern the metric (2.48) in the presence of a gravitational field by an analogy of tidal motions in Newtonian gravity and geodesic motion of nearby particles on Lorentzian manifolds (with curvature).
3.1. Tidal equation. Recall our discussion in Section 1.1 of tidal forces in the framework of the Newtonian theory. We took a freely falling particle as a reference particle and translated the origin of a new coordinate system at each time to the instantaneous position of this particle.
x0(t): motion of reference particle x(t): motion of arbitrary nearby particle Then we have seen that the displacement
(3.1) y(t) = x(t) x0(t) − to a nearby freely falling particle satisfies to first order in y (small displacements) the tidal equation (1.6): 2 d y 2 (3.2) = ψ x0(t) y(t) dt2 −∇ · 3. GEODESIC CORRESPONDENCE 21
In matrix notation this simply reads
2 2 i 3 d y d y X j (3.3) = M y , i.e. (t) = Mij(t) y (t) , dt2 − dt2 − j=1 where M is the Hessian matrix of the gravitational potential: 2 ∂ ψ (3.4) Mij(t) = x0(t) . ∂xi∂xj Suppose that the nearby particle is at rest relative to the reference particle at time t = 0, i.e. dy (3.5) = 0 . dt t=0 Since the tidal equation is linear we have (3.6) y(t) = A(t) y(0) for some matrix A(t) depending on t. Clearly, A(0) = 1 (identity matrix), and the initial condition (3.5) for y translates into the analogous conditon on A: dA (3.7) (0) = 0 dt And inserting (3.6) into the tidal equation (3.3) yields of course an identical equa- tion of motion for the matrix A: d2A (3.8) (t) = M(t) A(t) dt2 − Let us introduce a matrix θ(t) by: dA (3.9) θ = A−1 dt Then, dθ d2A dA d = A−1 + A−1 dt dt2 dt dt d2A dA dA (3.10) = A−1 A−1 A−1 dt2 − dt dt d2A = A−1 θ2 , dt2 − and by the matrix tidal equation (3.8) we obtain the following equation of motion for θ: dθ (3.11) = θ2 M dt − − (3.12) θ(0) = 0 We have not yet exploited the fact that the Hessian is symmetric: (3.13) M˜ = M 22 1. THE EQUIVALENCE PRINCIPLE
(For an arbitrary matrix A we denote by A˜ the transpose of A: A˜ij = Aji.) Taking the transpose of (3.11) we obtain as well dθ˜ (3.14) = θ˜2 M dt − − We now decompose (3.15) θ = σ + ω into its symmetric part σ and its antisymmetric part ω: 1 1 (3.16) σ = (θ + θ˜) ω = (θ θ˜) 2 2 − Substracting (3.14) from (3.11) M cancels and we obtain the following equation for ω: dω 1 1 = (θ2 θ˜2) = (σ + ω)2 (σ ω)2 (3.17) dt −2 − −2 − − = (σω + ωσ) − This is a linear homogeneous equation for ω considering σ as given. It follows that if ω vanishes initially, then it vanishes for all time. But here θ(0) = 0, therefore σ(0) = 0 as well as ω(0) = 0. It follows that with our initial conditions ω vanishes identically for all time, therefore (3.18) θ = σ is symmetric. It follows that 2 X 2 2 (3.19) tr(θ ) = tr(θθ˜ ) = (θij) = θ 0 , | | ≥ i,j and taking the trace of (3.11) we obtain d (3.20) tr θ = θ 2 tr M. dt −| | − Note that 2 X ∂ ψ (3.21) tr M(t) = x0(t) = ∆ψ x0(t) , i2 i ∂x and according to the Newtonian theory of gravitation: (3.22) ∆ψ = 0 in vacuum. The meaning of (3.20) becomes evident by looking at the evolution of volume. Consider the volume of a parallelepiped spanned by linearly independent displace- ments (y1, y2, y2) of nearby particles: Ω(y1, y2, y3). (See Fig. 16.) Let us assume that three nearby particles are initially at rest and at positions (E1,E2,E3): an orthonormal (positive) basis. Then Ω(E1,E2,E3) = 1. Recall that the displace- ments at time t are then obtained using the linear map A(t): deformation matrix. 3. GEODESIC CORRESPONDENCE 23
y3(t)
E3 y2(t)
E2
y1(t) E1
Figure 16. Volume of parallelepiped.
In fact, yi(t) = A(t)Ei, i = 1, 2, 3. Therefore, we obtain for the volume of the corresponding parallelepiped at time t:
Ω(t) = Ω(y1(t), y2(t), y3(t)) = Ω(A(t)E1,A(t)E2,A(t)E3) (3.23) = det A(t) Ω(E1,E2,E3) = det A(t) Ω(0) = det A(t) This formula holds for any volume form Ω. The volume of a parallelepiped is in fact the unique volume form on a vector space with inner product; c.f. note on volume form on pg. 25. Now consider the function (3.24) µ(t) = log Ω(t) = log det A(t) . Since, by differentiation of a determinant, d dA (3.25) det A(t) = det A(t) tr A−1 dt dt we have d (3.26) µ(t) = tr θ(t) , µ(0) = 0 , dt or Z t (3.27) µ(t) = tr θ(t0)dt0 . 0 Moreover, by (3.12) θ(0) = 0 we also have dµ (3.28) (0) = 0 , dt and 2 d µ d (3.29) (0) = tr θ = tr M . dt2 dt t=0 − t=0 Remark. We have shown that tidal forces deform shapes in such a way that volumes remain the same up to third order. Note that while equality to first order is simply a consequence of the fact that all particles constituting the shape are initially at rest, the equality for volumes in second order is precisely the equation for the gravitational field in vacuum, namely the Poisson equation (3.22) for the Newtonian potential. 24 1. THE EQUIVALENCE PRINCIPLE
Elevator
7→
Earth
Figure 17. Elevator experiment.
Elevator experiment Tidal forces are easily visualised in the following experiment: Consider a distribution of masses in a freely falling elevator in the gravitational field of the Earth. See Fig. 17. From the point of view of a freely falling observer at the center of the elevator the masses accelerate towards her horizontally and away from her vertically. An initially spherical distribution of nearby masses surrounding the observer is deformed to a prolate spheroid of the same volume. The equality of the volumes is precisely the equation of the gravitational field.
Note on volume forms in vector spaces In this note we discuss the notion of a volume form Ω in a n-dimensional vector space, and show that (3.17) holds in general. Moreover, we prove that if Vn is endowed with an inner product, then Ω is in fact unique. n n Let X1,...,Xn be vectors in an n-dimensional space V . A volume form Ω in V is a n-linear form in Vn which is totally antisymmetric and nondegenerate. That is to say Ω(X1,...,Xn) depends linearly on Xi if we fix all vectors X1,...,Xn except Xi, and changes sign if we interchange two vectors: Ω(...,X ,...,X ,...) = Ω(...,X ,...,X ,...) . i j − j i n Moreover it being nondegenerate means there is a n-tuplet of vectors (X1,...,Xn) in V such that Ω(X ,...,X ) = 0 . 1 n 6 If Vn is endowed with an inner product , then this automatically defines a volume n h· ·i n form on V . For, we can then choose a basis (E1,...,En) for V which is orthonormal: E ,E = δ . h i ji ij We then set
Ω(E1,...,En) = 1 . 3. GEODESIC CORRESPONDENCE 25
However, an inner product only determines an orthonormal basis up to rotation. Let (E10 ,...,En0 ) be another such orthonormal basis. We must verify that also Ω(E10 ,...,En0 ) = 1. (Otherwise the definition does not make sense.) To be precise we shall show that a volume form is uniquely determined by the inner product and orientation. Let us first return to Ω(X1,...,Xn). We can expand any vector Xi in the basis (E1,...,En): n X j Xi = (Xi) Ej : i = 1, . . . , n . j=1 j th The coefficients (Xi) : j = 1, . . . , n form the i column of a matrix X. Then Ω(X1,...,Xn) is determined. Therefore a volume form Ω is uniquely determined by the basis (E1,...,En). Let us verify that in the case n = 2: 1 2 1 2 X1 = (X1) E1 + (X1) E2 X2 = (X2) E1 + (X2) E2 . By multilinearity 1 1 1 2 Ω(X1,X2) =(X1) (X2) Ω(E1,E1) + (X1) (X2) Ω(E1,E2) 2 1 2 2 + (X1) (X2) Ω(E2,E1) + (X1) (X2) Ω(E2,E2) =(X )1(X )2 (X )2(X )1 = det X 1 2 − 1 2 This is in fact the definition of the determinant. Given a linear transformation A of Vn there is a matrix A which represents the linear map in the basis (E ,...,E ). Since A E is a vector, it can be expanded in the basis 1 n · i (E1,...,En): n X j AEi = Ai Ej . j=1 j The coefficients Ai constitute the corresponding matrix, and thus
Ω(AE1, . . . , AEn) = (det A)Ω(X1,...,Xn) .
Given a basis (E1,...,En) we obtain another basis (E10 ,...,En0 ) by transformation with an element of the general linear group GL(n, R): n X j E0 = G E : i = 1, . . . , n where det G = 0 . i i j 6 j=1 n The set of all bases in V has two components, one obtained from (E1,...,En) with det G > 0, the other with det G < 0. Selecting one of the two classes, say the one contain- ing (E1,...,En) which one then refers to as positive bases, corresponds to introducing an orientation on Vn. Hence Ω(E10 ,...,En0 ) > 0 on all positive bases. Consider transformations O which leave the inner product invariant: OX,OY = X,Y h i h i We can represent O in a positive orthonormal basis (E1,...,En), then the columns of the matrix n X j Ei0 = Oi Ej : i = 1, . . . , n j=1 26 1. THE EQUIVALENCE PRINCIPLE are another orthonormal basis:
E0,E0 = O E ,O E = E ,E = δ h i ji h · i · ji h i ji ij This condition reads OO˜ = 1 , the defining condition of an orthogonal matrix. (Here O˜ is the transpose of O.) Therefore (det O)2 = 1, or det O = 1, with det O = +1 if the primed basis is positive. ± We defined Ω(E1,...,En) = 1. The definition makes sense because if (E10 ,...,En0 ) is another positive orthonormal basis then
Ω(E10 ,...,En0 ) = (det O)Ω(E1,...,En) = 1 , because det O = 1 . It follows that if Vn is a oriented vector space with an inner product, then Vn has a unique volume form Ω.
Exercises 1. (Ratio of volumes): Consider a freely falling reference particle in a gravitational field in vacuum. Suppose the particle is surrounded by a sphere of volume Ω which is initially at rest relative to the reference particle. Show that in fact up to fourth order the volume of the sphere does not change in time: Ω(t) = 1 + O(t4) . Ω(0) What is the obstruction to the equality of volumes to higher order in time? 2. (Convergence of particles): Consider a freely falling reference particle in vacuum as in Exercise 1. Assume now in addition that the Hessian matrix of the gravitational potential does not vanish initially in the vicinity of the reference particle. Recall that the displacement of a nearby particle y(t) is given in terms of its initial displacement y(0) (we assume that the particle is initially at rest relative to the reference particle) by y(t) = A(t)y(0) , where A is the deformation matrix. Show that there is an initial displacement y(0) 6= 0 and a time t∗ > 0 such that y(t∗) = 0. Give a physical interpretation of this result. Hints. Recall the definition of the strain matrix dA θ = A−1. dt To solve Exercise 2 one may proceed as follows: (1) Show that under the assumption that the Hessian of the gravitational field does not vanish identically initially we have 3 d tr θ < 0 . dt3 t=0 (2) Deduce from the tidal equation the following ordinary differential inequality: d 1 tr θ ≤ − tr θ2 . dt 3 (Hint: Decompose θ into its diagonal and its traceless part.) 3. GEODESIC CORRESPONDENCE 27
(3) Show that there exists a time t∗ > 0 such that tr θ → −∞ as t → t∗. (Hint: Derive an equation for u = −1/ tr θ.) (4) Show that there exists a tangent vector y(0) 6= 0 such that y(t∗) = 0. (Hint: What is the implication of (3) for A(t) as t → t∗?)
3.2. Theory of Geodesics. According to the equivalence principle the grav- itational force is eliminated in a freely falling reference frame. A reference system is given by a family of timelike curves in the spacetime. We have seen — in our discussion of accelerated reference frames in Section 2.2 — that in the presence of a gravitational field the spacetime is curved; i.e. it is not Minkowski space as in Special Relativity, but a Lorentzian manifold with curvature (as we shall discuss in some detail in the next Chapter). Which timelike curves in the spacetime are described by freely falling bodies? Hypothesis: (Einstein) A freely falling test particle in a (non-trivial) grav- itational field corresponds to a timelike geodesic in a spacetime with cur- vature. Remark. It is an important feature of the theory that this hypothesis can be recovered from the equations of motion. We shall return to this question in our discussion of the Einstein equations in the presence of matter. It is thus a consequence of the equivalence principle that the gravitational force is an aspect of the spacetime itself. The aim of this section is to relate Newton’s law for the gravitational potential to an equation for the spacetime curvature. Since in a freely falling reference frame only tidal forces can be measured it can only be the differential of the gravitational field that contributes to spacetime curvature. To derive the law — the Einstein vacuum equations — we shall look at a family of nearby timelike geodesics, corresponding to the trajectories of nearby freely falling particles in the Newtonian theory. 3.2.1. Jacobi equation. Consider a timelike reference geodesic Γ0 (a freely falling reference test particle) in a spacetime ( , g). Let Γ0 be parametrized by arc length t (proper time), and let T be theM unit future directed (parallel) tangent vector field along Γ0:
(3.30) T T = 0 . ∇ Let O be an arbitrarily chosen origin on Γ0 where we set t = 0. Consider a spacelike vector X at O orthogonal to T , and let K0 be the spacelike geodesic through O with initial tangent vector X at O. Then X extends along K0 as the p tangent vector field of K0, and the magnitude of X, X = g(X,X), is constant | | along K0. Here K0 is parametrized by λ = X s, where s is the arclength along | | K0. Let us now extend T along K0 by parallel transport; we obtain a unit timelike future directed vectorfield along K0, which is orthogonal to X:
(3.31) g(T,X) = 0 along K0 . 28 1. THE EQUIVALENCE PRINCIPLE
Γλ
Γ0(t)
Γ0
T T
O X K0 K0(λ)
Figure 18. Construction of a timelike geodesic congruence.
At each point K0(λ) along K0, we draw the timelike geodesic towards the future with initial tangent vector T . We call this geodesic Γλ, and the family of curves a timelike geodesic congruence; see Fig. 18. Extend T along Γλ to be the tangent vector field to Γλ; since the tangent vector is parallel transported along a geodsic, we have (3.32) g(T,T ) = 1 . − Remark. The timelike curves Γλ correspond to the histories of nearby freely falling bodies in the Newtonian theory. Remark. In principle, we may observe three different kinds of behaviour: the distance to the nearby geodesic remains constant; the distance decreases; or the distance increases. As we shall describe in detail these cases are related to zero, positive, or negative (sectional) curvature, respectively; see Fig. 19. We define a timelike 2-dimensional (ruled) surface [ (3.33) S = Γλ , λ and for each t 0 a curve Kt S by: ≥ ⊂ (3.34) Kt(λ) = Γλ(t) . Then extend the vectorfield X to the surface S to be the tangent field of the curves Kt. The parameters t, λ can be thought of as coordinates on S; with respect to these coordinates: ∂ ∂ (3.35) T = ,X = , ∂t ∂λ and thus (3.36) [T,X] = 0 . 3. GEODESIC CORRESPONDENCE 29
Plane Sphere Hyperboloid
Figure 19. Three characteristic cases for geodesic congruences: Constancy of distance to nearby geodesics (zero curvature); De- crease of distance (positive curvature); Increase of the distance to a nearby geodesic (negative curvature).
Kt
∂ φ ∂t
Tp ∂ (t, λ) ∂λ Xp p
K0
Γ0
Figure 20. Push-forward of coordinate vectorfields to S.
Remark. More precisely, we have a mapping
φ :(t, λ) φ(t, λ) = Γλ(t) = Kt(λ); 7→ see Fig. 20. If p = φ(t, λ) then ∂ ∂ Tp = dφ(t, λ) ,Xp = dφ(t, λ) . · ∂t (t,λ) · ∂λ (t,λ) In other words the vectorfields T and X on S are the push-forward of coordinate vectorfields in the (t, λ) plane, ∂ ∂ T = φ∗ ,X = φ∗ . ∂t ∂λ Therefore, h ∂ ∂ i h ∂ ∂ i [T,X] = φ∗ , φ∗ = φ∗ , = 0 . ∂t ∂λ ∂t ∂λ 30 1. THE EQUIVALENCE PRINCIPLE
Prop 3.1. We have that T is orthogonal to X everywhere on S: g(T,X) = 0 on S.
Proof. We know this is true along K0. Differentiating the function g(T,X) along Γλ we obtain T g(T,X) = g( T T,X) + g(T, T X) . ∇ ∇ Since the integral curves of T are geodesics the first term vanishes by (3.30). Now, by (3.36): T X X T = [T,X] = 0 ; ∇ − ∇ therefore 1 g(T, T X) = g(T, X T ) = X g(T,T ) = 0 , ∇ ∇ 2 since g(T,T ) = 1 is a constant function on S. Hence − T g(T,X) = 0 , which says that g(T,X) is constant along Γλ. Since the integral curves of T initiate on K0 where g(T,X) = 0, it follows that
g(T,X) = 0 along Γλ . Remark. Here we have used that the connection in general relativity is sym- metric3 and compatible with the metric. While having a symmetric connection means X Y Y X = [X,Y ] , ∇ − ∇ for any vectorfields X, Y , metric compatibility is equivalent to Z g(X,Y ) = g( Z X,Y ) + g(X, Z Y ) , ∇ ∇ for any vectorfields X, Y , Z.
The curves Kt for t > 0 are not in general geodesics. The arclength of Kt between the parameter values 0 and λ is: Z λ 0 0 L Kt[0, λ] = X Kt(λ )dλ . 0 | |
The magnitude X Γ0(t) thus measures the deviation of nearby geodesics. Therefore we have to study| | (3.37) X(t) = X : the displacement at time t . |Γ0(t) Remark. The vectorfield (3.37) corresponds precisely to the displacement (3.1) in the tidal equation.
T X: First rate of change of the displacement. ∇2 X: Second rate of change of the displacement. ∇T 3this is the difference to gauge theories, not the metric. 3. GEODESIC CORRESPONDENCE 31
Since T X = X T and X T = 0 along K0 (T was parallel transported along ∇ ∇ ∇ K0), we have
(3.38) T X t=0 = 0 , ∇ | which corresponds in the Newtonian theory to the condition that the nearby test particles are initially at rest relative to the reference particle. As regards the second rate of change we have 2 X = T T X = T X T. ∇T ∇ ∇ ∇ ∇ Now recall that the definition of curvature is equivalent to:4
(3.39) [ X , Y ]Z Z = R(X,Y )Z ∇ ∇ − ∇[X,Y ] Since 2 X = T X T = X T T + [ T , X ]T = [ T , X ]T, ∇T ∇ ∇ ∇ ∇ ∇ ∇ ∇ ∇ we obtain, by (3.39), in view also of the antisymmetry of the curvature,
[ T , X ]T = T + R(T,X) T = R(X,T ) T, ∇ ∇ ∇[T,X] · − · the so-called Jacobi equation:
(3.40) 2 X = R(X,T ) T ∇T − · This is a second order equation for the vectorfield X along Γ0. The displacement field X to nearby geodesics is called a Jacobi field 5. We have derived the Jacobi equation, or Geodesic deviation equation, by variation through geodesics. We express (3.40) in components relative to a comoving orthonormal frame along Γ0. First choose an orthonormal basis (E1,E2,E3) for Σ0, the hyperplane in TO consisting of all vectors orthogonal to T . In particular X(0) Σ0. In general,M for every t 0 we denote by ∈ ≥ n o (3.41) Σt = Y T : g(Y,T ) = 0 ∈ Γ0(t)M the hyperplane in the tangent space to at Γ0(t) defined by all vectors orthogonal M to T . Now recall from Prop. 3.1 that X is orthogonal to T along Γ0, and thus
(3.42) X(t) Σt (t 0) . ∈ ≥ We can propagate the vectors Ei : i = 1, 2, 3 along Γ0 by parallel transport, i.e. we define vectorfields Ei : i = 1, 2, 3 along Γ0 such that
(3.43) T Ei = 0 : i = 1, 2, 3. ∇ Since the inner products remain unchanged we have an orthonormal basis for Σt at each t 0 which we denote by (E1(t),E2(t),E3(t)). Complemented with ≥ E0 = T we obtain an orthonormal basis (Eµ : µ = 0 ..., 3) for T at each Γ0(t)M
4C.f. notes on connections and the curvature tensor on pg. 39. 5 In fact, we have restricted ourselves to the case where K0 is orthogonal to Γ0 at O; X is then called a normal Jacobi field. 32 1. THE EQUIVALENCE PRINCIPLE t 0, i.e. along Γ0. But since X(t) Σt we can expand X(t) in the basis ≥ ∈ (Ei(t): i = 1, 2, 3): 3 X i (3.44) X(t) = X (t) Ei . i=1 Remark. The quadruple (t, X1,X2,X3) defines cylindrical normal coordi- nates in a neighborhood of Γ0. We say that p = expΓ0(t)(X(t)) if p lies on the spacelike geodesic from Γ0(t) with initial tangent vector X(t) Σt at arclength X(t) . We then assign to p the coordinates (t, X1,X2,X3),∈ and p lies in the |geodesic| hypersurface n o Ht = exp X : X Σt = exp Σt . ∈ We have that τ (t, λ1τ, λ2τ, λ3τ) are spacelike geodesics for each (λ1, λ2, λ3) 3 7→ ∈ R , while τ (τ, 0, 0, 0) is a timelike geodesic; c.f. note on pg. 37 on Riemannian 7→ normal coordinates at a point. Finally note that the curves Kt constructed above do not lie in Ht in general.
Take the covariant derivative (now everything is evaluated on Γ0): 3 XdXi (3.45) T X = Ei , ∇ dt i=1 because T Ei = 0 (parallel transport), and ∇ 3 2 i 2 Xd X (3.46) X = Ei . ∇T dt2 i=1 On the other side, 3 X i (3.47) R(X,T ) T = X Ri0 E0 . · · i=1 Consider now the vector
(3.48) Yi = Ri0 E0 = R(Ei,T ) T, · · which can be expanded in the basis (Eµ : µ = 1,..., 3): 3 0 X j (3.49) Yi = Yi E0 + Yi Ej . j=1 We find the coefficients by taking inner products with the orthonormal basis vec- tors, in fact 0 j (3.50) Y = g(Yi,E0) , and Y = g(Yi,Ej) . i − i Recall the definition of the curvature tensor: (3.51) R(W, Z, X, Y ) = g(W, R(X,Y ) Z) . · 3. GEODESIC CORRESPONDENCE 33
Thus, 0 (3.52) Y = R(T,T,Ei,T ) = 0 . i − Remark. Here we have used the symmetry property (3.53) R(X,Y ) Z = R(Y,X) Z, · − · which is clear from the definition (3.39). More properties of the curvature tensor are discussed in the note on page 39. Furthermore, j (3.54) Yi = R(Ej,T,Ei,T ) . Remark. Note that the expression (3.54) is symmetric in (i, j) by the pair symmetry of the curvature tensor; c.f. Notes on page 39. The term on the right hand side of the Jacobi equation is therefore 3 3 X jX R(X,T ) T = X Ri0j0Ei . · j=1 i=1 Thus equating the two sides of (3.40) yields
2 i 3 d X (t) X j (3.55) = Ri0j0 X (t) . dt2 − j=1 3.2.2. Einstein vacuum equation. In conlusion we have shown that in a paral- lelly propagated orthonormal frame (E1,E2,E3) along the reference geodesic Γ0, P i i.e. T Ei = 0, the components of the displacement vector X(t) = i X (t)Ei(t) to a∇ nearby timelike geodesic satisfy the equation
2 i 3 d X (t) X j (3.56) = Mij(t) X (t) , dt2 − j=1 where M is the symmetric matrix
(3.57) Mij = Ri0j0 , or indeed the matrix valued function M(t) along Γ0:
(3.58) Mij(t) = R(Ei,T,Ej,T ) . Γ0(t) Thus the Jacobi equation (3.56) is identical in form to the Newtonian tidal equation (3.3). In the Newtonian theory we have (3.59) tr M = ∆ψ = 0 : in vacuum . The vacuum equations of the relativistic theory (of general relativity) should there- fore be (3.60) tr M = 0 . 34 1. THE EQUIVALENCE PRINCIPLE
3.2.3. First variation equation. We conclude our comparison of the tidal equa- tion to the Jacobi equation with a discussion of velocities. Recall that we have extended T from O along K0 by parallel transport. This corresponds to vanishing initial velocity, 3 X dXi dXi (3.61) 0 = X T O = T X O = (0) Ei = = 0 : i = 1, 2, 3 . ∇ | ∇ | dt ⇒ dt t=0 i=1
The curve K0 here lies in the geodesic spacelike hypersurface H0. This is a special case of a orthogonal hypersurface to Γ0. Let now more generally Γ0 be a timelike geodesic normal to a spacelike hypersurface M (not necessarily geodesic). The natural way to extend T from O to M in this case is to define T at every point p M to be the unit future-directed (timelike) normal to M at p. Then let Γp be the∈ timelike geodesic normal to M at p. We define the second fundamental form k of a spacelike hypersurface M with unit (future-directed timelike) normal T to be the bilinear form in TpM at each p M defined by: ∈ (3.62) k(X,Y ) = g( X T,Y ): X,Y TpM ∇ ∀ ∈ Remark. In fact, k is symmetric. So k is a quadratic form in TpM at each p M. To show this, extend X, Y to vectorfields on M tangential to M, then: ∈ k(X,Y ) = g(T, X Y ) , since g(T,Y ) = 0 . − ∇ Similarly with X and Y interchanged, so taking the difference,
k(X,Y ) k(Y,X) = g(T, X Y Y X) = g(T, [X,Y ]) = 0 , − − ∇ − ∇ − since [X,Y ] is also tangential to M.
] We define a linear transformation k in TpM at each p M by: ∈ ] (3.63) g(k X,Y ) = k(X,Y ): X,Y TpM, · ∀ ∈ in other words ] (3.64) k X = X T. · ∇ In particular at O M: ∈ ] (3.65) T X O = X T O = k X O ∇ | ∇ | · | is the initial velocity at t = 0. With respect to the orthonormal basis (Ei : i = 1, 2, 3) we define
(3.66) kij = k(Ei,Ej) . Then 3 ] X (3.67) k Ei = kijEj , · j=1 3. GEODESIC CORRESPONDENCE 35
] since g(k Ei,Ej) = kij, and we obtain
3 ] X i (3.68) k X O = kijX Ej O . · | | i,j=1 Therefore
i 3 dX X j (3.69) = kij O X t=0 . dt t=0 | | j=1
We now define – similarly to Kt above – for each t (at least in a neighborhood of O) a spacelike hypersurface Mt by n o (3.70) Mt = Γp(t): p M ∈
The vectorfield T extends to spacetime (at least a neighborhood of Γ0) to be the tangent vector field to the timelike geodesic congruence Γp. Then T remains orthogonal to each spacelike hypersurface Mt. Remark. We have already made this observation in the proposition on pg. 30, for its proof only relies on the fact that the geodesic tangent field T is orthogonal to M initially. Indeed, given any vectorfield X on M tangential to M, we denote by φt the flow generated by T (i.e. φt(p) = Γp(t) for all p M), and define ∈ (3.71) Xq = dφt Xp , where q Mt is given by q = φt(p), p M. · ∈ ∈ The vectorfield X extended this way to spacetime is tangential to each Mt and satisfies [T,X] = 0. Then 1 (3.72) T g(T,X) = g( T T,X) + g(T, T X) = X g(T,T ) = 0 , ∇ ∇ 2 which yields that if g(T,X) = 0 initially then it vanishes on each Mt, thus in spacetime.
Since T is the unit future-directed timelike normal to each Mt we have that at each t 0: ≥ i 3 dX X j (3.73) (t) = kij(t)X (t) , dt j=1 where kij(t) are the components of the second fundamental form of Mt at the point Γ0(t) with respect to the parallel transported frame (E1,E2,E3). Recall now from our discussion of the tidal equation, the dependence on the initial data being linear, we can write
3 i X j (3.74) X (t) = Aij(t) X (0) , j=1 36 1. THE EQUIVALENCE PRINCIPLE r q
Np
p
M
T pM Figure 21. Polar normal coordinates. and the deformation matrix A satisfies the same equation as in the Newtonian theory; c.f. (3.8): d2A (3.75) = MA. dt2 − Recall the definition of the rate of strain matrix dA (3.76) θ = A−1 . dt We have, from (3.74) and (3.75):
i 3 3 dX X dAij j X j (3.77) (t) = (t)X (0) = θij(t) X (t) . dt dt j=1 j=1 The comparison to (3.73) yields
(3.78) θij = kij which is also called the first variation equation.
Introductory remarks on normal coordinates and curvature Consider a (n-dimensional) Riemannian manifold ( , γ). Let us fix a point p and consider all tangent vectors N of unit length N 2M= γ(N ,N ) = 1 at p: ∈ M p | p| p p n 1 2 S − = N T : N = 1 p p ∈ pM | p| n 1 Any vector X T can written as X = r N , for some r 0 and N S − . Each p ∈ pM p p ≥ p ∈ p Xp defines a geodesic, Xp being the initial velocity at p. Recall the velocity of a geodesic is constant. We assign to a point q in a neighborhood of p the polar normal coordinates (r, Np) with origin at p if q is the point at parameter value 1 along the geodesic from p with initial velocity Xp = rNp. The point q is therefore defined only by Xp. (See also figure 21.) To obtain rectangular normal coordinates we choose an orthonormal basis (E ,...,E ) for T . Then any X T has a unique expansion 1 n pM p ∈ pM n X i Xp = x Ei . i=1 3. GEODESIC CORRESPONDENCE 37
Then we assign to q the rectangular coordinates (x1, . . . , xn). Note that q sX sX r = X = γ(X ,X ) = δ xixj = (xi)2 . | p| p p ij i,j i Hence the distance from the origin is the same expression as in Euclidean geometry. We now consider the case where is 2-dimensional and introduce the concept of M Gauss curvature. In two dimensions Np is an element of the unit circle and characterized by a single angle ϕ; the polar normal coordinates are thus simply (r, ϕ). The Lemma of Gauss says that the geodesic rays corresponding to the lines of constant ϕ are orthogonal to the geodesic circles Sr: the set of all points which are at distance r from a the point p. The metric takes the form: ds2 = dr2 + R2(r)dϕ2 .
We see that R(r)dϕ is the element of arc length along the geodesic circles Sr. The condition that the metric is locally Euclidean at p is simply R(r) ∂R 1 as r 0 or (0, ϕ) = 1 . r → → ∂r The Gauss curvature K(p) is then defined in polar normal coordinates by ∂2R + KR = 0 . ∂r2 Alternatively consider the circumference C(r) of the geodesic circle Sr: Z 2π C(r) = R(r, ϕ)dϕ . 0 Then another definition of the Gauss curvature is given by d Z Z r Z 2π 2π = K = KR drdϕ , − dr Dr 0 0 where Dr is the geodesic disc of radius r: the set of all points at distance less or equal r from p. We can expand the right hand side to leading order, Z 2 K = Kp πr + higher order terms, Dr and on the left hand side similary, a C(r) = 2π r + 3 r3 + higher order terms. 6 Therefore the Gauss curvature appears as the coefficient to the third order correction to the circumference, K = a . p − 3 Finally let us mention Gauss’ theorem for a geodesic triangle: If α, β, γ are the inner angles of a geodesic triangle enclosing the area T (see figure 22) then ⊂Z M α + β + γ π = K. − T The generalisation thereof is the Gauss-Bonnet theorem for closed surfaces S: Z K = 2πχ , S where χ = 2(1 g) is the Euler characteristic, and g the genus of S (g = 0 for a sphere, g = 1 for a torus).− 38 1. THE EQUIVALENCE PRINCIPLE
β
T γ α
Figure 22. A geodesic triangle.
Notes on connections and the curvature tensor in general relativity Recall that the definition of curvature is equivalent to R(X,Y ) Z = [ , ]Z Z. · ∇X ∇Y − ∇[X,Y ] We shall see in the next Chapter that R(X,Y ) has the cyclic property R(X,Y ) Z + R(Y,Z) X + R(Z,X) Y = 0 , · · · and satisfies the Bianchi identity ( R)(Y,Z) + ( R)(Z,X) + ( R)(X,Y ) = 0 . ∇X ∇Y ∇Z These identities come from the Jacobi identity for linear operators. The cyclic property follows by considering vectorfields as linear operators on functions, X : f X f , 7−→ · while the Bianchi identity follows by considering vectorfields as linear operators on vec- torfields: X : Y Y. 7−→ ∇X Here denotes the covariant derivative, a linear operator on vectorfields with the prop- erties:∇
(1) X (fY ) = X(f)Y + f X Y . (2) ∇ Y = f Y . ∇ ∇fX ∇X In general relativity it comes from a connection which is symmetric and compatible with the metric. The two conditions are:
(1) X Y Y X = [X,Y ] (symmetry) (2) Zg∇ (X,Y− ∇) = g( X,Y ) + g(X, Y ) (metric compatibility) ∇Z ∇Z The cyclic property and the Bianchi identity only depend on the connection being sym- metric, and do not rely on the metric property. However, metric compatibility implies the antisymmetry of the curvature tensor in the first two slots: R(W, Z, X, Y ) = R(Z, W, X, Y ) − (Note that the antisymmetry in the last two slots is an immediate consequence of the definition above.) Recall here the definition of the Riemann curvature tensor, a fully covariant tensor of fourth rank: R(W, Z, X, Y ) = g(W, R(X,Y ) Z) . · We can interpret this antisymmetry of the curvature tensor as follows: 3. GEODESIC CORRESPONDENCE 39
The curvature Rp at a point p in the spacetime is an antisymmetric bilinear form in the tangent space T to at p with values in theM space of linear transformations of pM M Tp . Let use denote in general by (U, V ) the space of linear maps from a vector space U toM a vector space V . So if X , Y L T then p p ∈ pM R (X ,Y ) (T , T ) p p p ∈ L pM pM depends linearly on Xp, Yp, and Rp(Xp,Yp) = Rp(Yp,Xp). With Xp, Yp both fixed R (X ,Y ) Z is an element of T depending− linearly on Z . The curvature R of the p p p · p pM p spacetime ( , g) is an assignment p Rp at each p . If X, Y , Z are vectorfields on , then R(MX,Y )Z is the vectorfield7→ whose value at ∈p M is R (X ,Y ) Z . M ∈ M p p p · p Consider, by comparison, a 2-form ω on , that is an assignment p ωp of an antisymmetric bilinear real valued form in T M. Then if Π is a 2-dimensional7→ plane in pM Tp , and (E1,E2) an orthonormal basis for Π, and (X1,X2) any pair of vectors in Π, weM have 1 2 Xi = Xi E1 + Xi E2 : i = 1, 2 , and hence 1 1 X1 X2 ω(X1,X2) = det[X] ω(E1,E2) , where [X] = 2 2 . X1 X2
So if (E10 ,E20 ) is another orthonormal basis with the same orientation, then 1 2 Ei0 = Oi E1 + Oi E2 i = 1, 2 , where O SO(2), and it follows that ∈ ω(E10 ,E20 ) = ω(E1,E2) . Thus ω can be thought of as assigning a real number to an oriented plane, which we may denote by ωΠ. Similarly Rp assigns a linear transformation of Tp to each oriented plane Πp (a 2- dimensional subspace of T ). This linear transformationM can be thought of as follows: pM Consider an infinitesimal closed curve C on the plane Πp, (see Fig. 23); given any vector V T we parallel transport V along the curve C, and let us denote by V 0 the vector ∈ pM obtained upon completing the contour. If the curvature at p is not zero, then V 0 does not coincide with V , and there is a difference ∆V = V 0 V ; in fact − ∆V = R V, ∆A Πp · where ∆A is the area of Πp enclosed by the curve C. The connection being metric means that magnitudes of vectors are preserved by parallel transport. So upon completing the contour C we obtain a vector V 0 which is different from V but of the same magnitude V 0 = V . | | | | Here the metric g is Lorentzian, so V 0 = g(V 0,V 0) = g(V,V ) = V means that V 0 is | | | | related to V by a Lorentz transformation; i.e. V 0 = ΛV for some Λ SO(3, 1). Moreover, ∈ since V 0 = V +(∆A) RΠ V we see that RΠp is an element of the Lie algebra of the Lorentz group. · We express this condition in an orthonormal basis (E0,E1,E2,E3) at a point p, such that gµν = g(Eµ,Eν ) is the Minkowski metric in rectangular coordinates [g ] = [η ] = diag( 1, 1, 1, 1) . µν µν − 40 1. THE EQUIVALENCE PRINCIPLE
∆V V
C
p
Πp
Figure 23. Illustration of the linear transformation RΠp .
Consider first the condition g(V,V ) = g(V 0,V 0). Expanding V , V 0 in the above basis, and ν denoting by [Λµ] the matrix representing Λ with respect to the above basis,
X µ X µ X β V = V Eµ ,V 0 = V 0 Eµ , ΛEα = ΛαEβ , µ µ β it is straight-forward to show that the condition reads:
X µ ν gµν ΛαΛβ = gαβ . µ,ν For an infinitesimal curve C the transformation Λ differs from the identity by an infini- tesimal ϑ: µ µ µ Λα = δα + ϑα . Inserting into the condition above implies
X µ ν ν µ X µ X ν gµν ϑαδβ + ϑβδα = 0 , or equivalently gµβϑα + gαν ϑβ = 0 , µ,ν µ ν P µ which is the statement that ϑαβ = µ gαµϑβ is antisymmetric in α, β:
ϑαβ + ϑβα = 0 .
µ The matrix [(RΠp ) ν ] representing RΠp with respect to the basis (Eµ : µ = 0,..., 3) therefore has the property that
X µ gλµ(RΠp ) ν = (RΠp )λν µ
µ is antisymmetric in λ, ν. [In fact, as we have sketched above, given any matric [Λ ν ] P µ such that [Λκν ] is skew symmetric, where Λκν = µ gκµΛ ν , then exp Λ is an element of the Lorentz group SO(3, 1), (and conversely).] This is equivalent to R(W, Z, X, Y ) = R(Z, W, X, Y ). For, in terms of the basis basis (E : µ = 0,..., 3), the linear transfor- − µ mation Rµν = R(Eµ,Eν ) is represented as follows: X (R ) E = (R )α E , µν · β µν β α α 3. GEODESIC CORRESPONDENCE 41 and X R = R(E ,E ,E ,E ) = g(E ,R E ) = g(E , (R )λ E ) αβµν α β µ ν α µν · β α µν β λ λ X λ X λ = (Rµν ) βg(Eα,Eλ) = gαλ(Rµν ) β . λ λ
The condition that Rµν is an element of the Lie algebra of the Lorentz group thus reads: R = R . βαµν − αβµν
3.3. Ricci curvature. We conclude this chapter with a discussion of the geometric meaning of the equations
tr M = 0 , where Mij = Ri0j0 . 3.3.1. Sectional curvature. The curvature tensor (3.79) R(W, Z, X, Y ) = g(W, R(X,Y ) Z) · is antisymmetric in (W, Z) as well as in (X,Y ). This property is a manifestation of the fact that the connection is compatible with the metric, (c.f. Note on pg. 39). Moreover we have the pair symmetry (3.80) R(X, Y, W, Z) = R(W, Z, X, Y ) . This is a consequence of the cyclic identity6: (3.81) R(X,Y ) Z + R(Y,Z) X + R(Z,X) Y = 0 . · · · Let Π be an arbitrary 2-dimensional plane in Tp at a point p , and (E1,E2) an orthonormal basis for this plane. Then M ∈ M
(3.82) KΠ = R(E1,E2,E1,E2) is independent of the choice of an orthonormal basis for Π7 and also independent 0 0 of the orientation of Π. Indeed if (E1,E2) is another orthonormal basis for Π, 0 1 2 E1 = O1E1 + O1E2 (3.83) 0 1 2 E2 = O2E1 + O2E2 , then 0 0 0 0 2 (3.84) R(E1,E2,E1,E2) = (det O) R(E1,E2,E1,E2) .
KΠ is called the sectional curvature of the plane Π.
6We will return to the cyclic identity when we discuss the Bianchi identities in the next chapter. 7 This also means that KΠ is invariant under Lorentz transformations of Π. 42 1. THE EQUIVALENCE PRINCIPLE
3.3.2. Jacobi equation. Recall the orthonormal frame (T,E1,E2,E3) that we constructed in our discussion of the Jacobi equation. If we set
X t=0 = E1 t=0 , then we obtain 2 1 d X = M11 = KΠ01 dt2 t=0 − t=0 t=0 where Π01 is the plane spanned by E0 = T , and E1. In general, v 3 u 3 X i p uX i 2 (3.85) X = X Ei , X = g(X,X) = t (X ) , | | i=1 i=1 also d2 X (3.86) | | = KΠ X , dt2 t=0 − | | where Π is the plane spanned by T , X. Remark. Review here Fig. 19 on page 29. We see that the distance to a nearby geodesic remains constant, decreases, or increases according as to whether the sectional curvature of the plane spanned by the tangent vector of the reference geodesic and the displacement vector is zero, positive or negative respectively.
Remark. Given a plane Π Tp , then consider the surface HΠ = exp Π. ⊂ M Then KΠ is in fact the Gauss curvature of the surface HΠ at the origin p. 3.3.3. Ricci curvature. Consider then 3 3 3 X X X (3.87) tr M = Mii = Ri0i0 = Ki , i=1 i=1 i=1 where Ki = KΠi0 is the sectional curvature of the plane spanned by E0 and Ei. We define the (E0,E0) component of the Ricci curvature by 3 X (3.88) Ric(E0,E0) = Ki . i=1 ˆ V ˆ ˆ Given any vector V , if V is not unit length we define V = |V | . Then Ric(V, V ) is simply defined as above taking E0 = Vˆ , and (3.89) Ric(V,V ) = V 2 Ric(V,ˆ Vˆ ) . | | This defines a symmetric bilinear form in Tp at each p by polarisation: M ∈ M 1n o (3.90) Ric(U, V ) = Ric(U + V,U + V ) Ric(U V,U V ) 4 − − − We shall from now on use the notation: (3.91) S = Ric . 3. GEODESIC CORRESPONDENCE 43
Alternatively we can express the Ricci curvature in terms of the curvature trans- formation (3.39). In an arbitrary basis we have 3 X α (3.92) Sµν = Ric(Eµ,Eν) = (Rαν) µ . α=0 We usually denote α α (3.93) (Rµν) β by R βµν . In terms of the inverse of g we have α X −1 αβ (3.94) (Rαν) µ = (g ) Rαµβν . β Since X α X −1 αβ (3.95) Sµν = R µαν = (g ) Rαµβν , α α,β the pair symmetry Rµναβ = Rαµβν implies the symmetry
(3.96) Sνµ = Sµν . The vacuum Einstein equations are therefore
(3.97) Sµν = 0 .
Exercises In the following exercises we study cylindrical normal coordinates which correspond to a freely falling laboratory. Recall our discussion of the Jacobi equation in cylindrical normal coordinates: Let Γ0 be a timelike reference geodesic with unit tangent vector field T , 0 parametrized by arclength x . We have ∇T T = 0. In the local simultaneous space Σx0 at each point along Γ0, Σ 0 = X ∈ T 0 M : g(T,X) = 0 , x Γ0(x ) we have an orthonormal basis (E1,E2,E3) which is parallel transported along Γ0: ∇T Ei = 0. 0 1 2 3 To each point p in a neighborhood of Γ0 we assign cylindrical normal coordinates (x , x , x , x ) if 1 2 2 p = exp 0 X ,X = x E + x E + x E , Γ0(x ) 1 2 3 0 namely if p lies on the spacelike geodesic with initial tangent vector X at Γ0(x ) at arclength p 1 2 2 2 3 2 distance r from Γ0: r = |X| = (x ) + (x ) + (x ) . Then on Γ0: ∂ ∂ T = ,E = : i = 1, 2, 3 . ∂x0 i ∂xi 1. (Expansion of the metric to first order): Show that to first order in r, the metric is equal to the Minkowski metric, i.e. 0 1 2 3 2 gµν (x , x , x , x ) = ηµν + O(r ). Hints: Show that as a consequence of metric compatibility ν ν ∂µgαβ = gνβ Γµα + gαν Γµβ , where Γ are the connection coefficients. The problem thus reduces to showing α Γµν = 0 along Γ0, which can be done as outlined in the following steps: 44 1. THE EQUIVALENCE PRINCIPLE
(1) Given any curve γ(τ) = (x0(τ), x1(τ), x2(τ), x3(τ)) in local coordinates, show that the geodesic equation ∇γ˙ γ˙ = 0 is equivalent to
2 µ 3 α β d x X µ dx dx + Γ = 0 : µ = 0, 1, 2, 3 . dτ 2 αβ dτ dτ α,β=0
(2) Use the fact that the curve τ 7→ (τ, 0, 0, 0) is a geodesic (namely Γ0) to deduce µ 0 0 that Γ00(x , 0, 0, 0) = 0 for all x . 0 (3) Use the fact that for any vector k = (k1, k2, k3) and any x the curve τ 7→ (x0, k1τ, k2τ, k3τ) is a geodesic, and deduce that for all x0 and τ: i j µ 0 1 2 3 k k Γij (x , k τ, k τ, k τ) = 0 . 0 (4) Use the parallel transport equation for Ei to deduce that for all x : µ 0 Γ0i(x , 0, 0, 0) = 0. 2. (Expansion of the metric to second order): Find the second order terms in the µ expansion of Exercise 1 in terms of the curvature components R νλσ. In particular, prove: 3 1 X k l 3 g = δ − R x x + O(r ): i, j = 1, 2, 3 . ij ij 3 ikjl k,l=1 Hints: Similarly to Exercise 1, show that: σ σ ∂α∂µgνλ = (∂αΓµν )gσλ + (∂αΓµλ)gνσ : along Γ0. σ µ The problem therefore reduces to expressing ∂αΓµν in terms of R νσλ, which can be done as follows: (1) Note that we have µ µ µ R νκλ = ∂κΓλν − ∂λΓκν along Γ0. Now define µ µ µ µ S κλν = ∂κΓλν + ∂λΓνκ + ∂ν Γκλ and derive µ µ µ µ 3 ∂κΓνλ = R νκλ + R λκν + S κλν : along Γ0. µ (2) Use hint (3) from Exercise 1 to show that S ijk = 0: i, j, k = 1, 2, 3 along Γ0. µ µ µ (3) Use the hints from Exercise 1 to show that ∂0Γ00 = ∂0Γ0i = ∂0Γij = 0 along Γ0. µ µ (4) Use (1) and (3) to find ∂iΓ0j and ∂iΓ00 along Γ0. 3. (Riemannian sectional curvature and Gauss curvature): Consider a Riemannian manifold M of dimension n > 2. Let Π ⊂ TpM be a 2-dimensional plane in the tangent space at a point p ∈ M. Let HΠ be the geodesic surface
HΠ = expp(Π) = {expp(X): X ∈ Π} , consisting of all geodesics starting at p with an initial tangent vector X in Π. Show that the Gauss curvature of the surface HΠ at p is equal to the Riemannian sectional curvature RΠ = R(E1,E2,E1,E2), where (E1,E2) is an orthonormal basis for Π. Hints: Introduce normal coordinates (x1, . . . , xn) with origin at p, such that 1 2 3 n 3 n HΠ = {(x , x , x , . . . , x ): x = ... = x = 0} .
Use the result of Exercise 1 to calculate the circumference Cr of a geodesic circle 1 2 1 2 2 2 2 Sr = {(x , x ):(x ) + (x ) = r } in HΠ. Finally recall from the note on pg. 37 that the Gauss curvature K appears in the formula dC r = 2π − πKr2 + O(r3) . dr CHAPTER 2
Einstein’s field equations ∗
In the previous chapter we have arrived at a unified theory of space, time and gravitation according to which spacetime is a 3 + 1 dimensional manifold endowed with a metric g of index 1 whose curvature represents tidal gravitationalM forces. The Einstein vacuum equations
(0.1) Sµν = 0 , were found in analogy to Poisson’s equation for the gravitational potential in vacuum: (0.2) ∆ψ = 0 . In this chapter we discuss Einstein’s field equations in the presence of matter, which can be thought of as the analogue of (0.3) ∆ψ = 4πµ , where µ is the mass density.1 Remark. A comparison to the theory of electromagnetism provided some guidance for Einstein from early on. The situation can be compared to electro- statics, where the electric potential φ satisfies (0.4) ∆φ = 4πρ , − and ρ is the electric charge density: the source of the electric field E = φ. It is known however that moving charges create a magnetic field B, according−∇ to Amp`ere’slaw of magnetostatics: (0.5) B = 4πJ ∇ × Only Maxwell’s theory of electrodynamics unites the sources (ρ, J), the fields (E,B), and the potentials (φ, A) as time and space components of differential forms on Minkowski space and constitutes a theory in agreement with special relativity. In the Newtonian theory the mass density µ is the source of the grav- itational field. If general relativity relates to Newtonian gravitation as Maxwell’s theory does to electrostatics, then the question is what is the source in general relativity? We shall see that the energy density ε will play the role of the source in general relativity. In special relativity ε is a component of the energy-momentum stress tensor. While this analogy is described further in the following note, we next discuss the energy-momentum tensor in general relativity.
1in units where Newton’s consant G = 1.
45 46 2. EINSTEIN’S FIELD EQUATIONS ∗
Notes on Maxwell’s theory In electrostatics the electric field E is determined from the electric charge density ρ by the system of equations (0.6) E = φ −∇ (0.7) E = 4πρ , ∆φ = 4πρ . ∇ · − These are complemented by the equations of magnetostatics for the magnetic field B, given an electric current density J (in this note we keep explicitly the speed of light c in the equations): J (0.8) B = 4π ∇ × c (0.9) B = 0 , ∇ · The last equation of course states the absence of a magnetic charge density. In a dynamical situation Gauss’ law E = 0 is replaced by Faraday’s induction law: ∇ × 1 ∂B (0.10) E + = 0 , ∇ × c ∂t and Maxwell corrected Amp`ereslaw to be: 1 ∂E J (0.11) B = 4π . ∇ × − c ∂t c Taking the divergence of this equation we obtain the continuity equation ∂ρ (0.12) + J = 0 , ∂t ∇ · which expresses the conservation of charge. This is the integrability condition of the Maxwell equations. Remark. This logic reappears in general relativity: If we introduce a source it has to satisfy an integrability condition for the field equations. We can introduce a vector potential A so that (0.13) B = A, ∇ × as a general solution to (0.9). Then in view of (0.10) we have 1 ∂A (0.14) E = φ . −∇ − c ∂t Now it turns out that Maxwell’s equations can be unified in the spacetime point of view of special relativity: Introduce the potential 1-form 3 X µ (0.15) A = Aµdx , µ=0 and consider the electromagnetic field 2-form (0.16) F = dA, where d is the exterior derivative. Then (0.17) dF = 0 . 2. EINSTEIN’S FIELD EQUATIONS ∗ 47
These are precisely the homogeneous Maxwell equations. Here, in rectangular coordinates we set x0 = ct so time intervals become length, and the metric is 3 X (0.18) g = dx0 dx0 + dxi dxi ; − ⊗ ⊗ i=1 moreover µ 1 µν (0.19) A = (g− ) Aν (0.20) A0 = φ , Ai = Ai : the components of the vector potential. Remark. Given a 1-form θ then dθ is a 2-form defined by (dθ)(X,Y ) = X(θ(Y )) Y (θ(X)) θ([X,Y ]) . − − Similarly, if ω is a 2-form then dω is a 3-form given by dω(X,Y,Z) = X(ω(Y,Z)) + Y (ω(Z,X)) + Z(ω(X,Y )) ω([X,Y ],Z) ω([Y,Z],X) ω([Z,X],Y ) − − − Note if ω = dθ then dω = 0. We have 3 X (0.21) F = F dxµ dxν µν ⊗ µ,ν=0 (0.22) F = ∂ A ∂ A , µν µ ν − ν µ ∂ ∂ (to see this simply take X = ∂xµ and Y = ∂xν in F (X,Y ) = dA(X,Y ) and note that [X,Y ] = 0 in rectangular coordinates); and moreover ∂φ 1 ∂Ai (0.23) F = ∂ A ∂ A = = E i0 i 0 − 0 i −∂xi − c ∂t i ∂Aj ∂Ai (0.24) F = ∂ A ∂ A = = ( A) = B . ij i j − j i ∂xi − ∂xj ijk ∇ × k ijk k Now, the inhomogeneous Maxwell equations are J µ (0.25) F µν = 4π ; where J 0 = cρ , J i = J i : electric charge density. ∇ν c F being antisymmetric the divergence of the left hand side vanishes identically: (0.26) F µν = 0 . ∇µ∇ν The integrability condition for Maxwell’s equations is therefore: J µ = 0(0.27) ∇µ 3 ∂J 0 X ∂J i ∂ρ (0.28) + = + J = 0 : in rectangular coordinates. ∂x0 ∂xi ∂t ∇ · i=1 In this way space and time components are united in this theory: F unites (E,B), A unites (φ, A), and J unites (ρ, J). The spacetime current J µ acts as a source for Maxwell’s equations. We compare the sources: Electrostatics: ρ Newtonian gravitation: µ Maxwell’s theory: J µ General relativity: ? 48 2. EINSTEIN’S FIELD EQUATIONS ∗
We shall see that the energy density ε will play the role of the source in general relativity; (in a quasi-Newtonian situation the leading term is µc2). In special relativity ε is the component T 00 of a contravariant tensorfield T µν , then energy momentum stress tensor: T 00 = ε : energy density T 0i = pi : momentum density T ij : stress. In fact, in Maxwell’s theory: 1 1 ε = E 2 + B 2 4π 2 | | | | 1 (0.29) pi = (E B)i 4π × 1 1 1 T ij = E E B B + δ E 2 + B 2 . 4π − i j − i j 4π ij 2 | | | |
1. Einstein’s field equations in the presence of matter 1.1. Energy-momentum-stress tensor. The energy-momentum-stress ten- ∗ sor T is a 2-contravariant symmetric tensorfield on (a quadratic form in Tx , the cotangent space at each x ) that vanishesM in vacuum. In componentsM relative to an arbitrary frame ∈ M 3 X µν µν νµ (1.1) T = T Eµ Eν ,T = T . ⊗ µ,ν=0 There is a corresponding 1-covariant-1-contravariant tensorfield with components µ X µν (1.2) T λ = T gνλ , ν X µ (1.3) so ( T )Eλ = T λEµ . − − µ + Consider the hyperboloid Hx of future-directed unit timelike vectors u at x. Such a vector u is the 4-velocity of an observer at x. Then
(1.4) T u Tx − · ∈ M is the energy-momentum density of matter relative to the observer with 4-velocity u at x. We have the following physical requirement: This energy-momentum density is a non-spacelike future-directed vector. Dominant energy condition: The linear transformation T maps the + − future hyperboloid Hx into the closure of the open future cone at x, for each x . (See Fig. 1.) ∈ M We decompose T u into a component colinear to u, εu, and a component − · orthogonal to u, p, lying in Σu, the local simultaneous space of the observer with 4-velocity u:
(1.5) T u = εu + p , p Σu = X : g(u, X) = 0 . − · ∈ { } 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 49
T u − ·
+ Hx
u
x
Figure 1. Dominant energy condition.
ε is called the energy density of matter at x relative to the observer with 4-velocity u at x, and p is called the momentum density of matter at x relative to the same observer. We have X µ λ κ X κ λ (1.6) ε = g(T u, u) = gµκT u u = Tκλu u · λ µ,κ,λ κ,λ where X µ X µν (1.7) Tκλ = gµκT λ = gµκgνλT µ µ,ν are the components of a quadratic form in Tx at each x , a 2-covariant symmetric tensorfield on . We then write: M ∈ M M (1.8) ε = T (u, u) . The dominant energy condition reads: (1.9) p ε for every u H+ . | | ≤ ∈ x In particular, ε 0. 1.1.1. Proper≥ energy density. We minimize ε with respect to u, defining the proper energy density (1.10) ρ(x) = inf T (u, u) . + u∈Hx + Note that Hx is non-compact, and consequently the infimum is not necessarily achieved. Remark. This is in contrast to its analogue in Euclidean space, the unit sphere which is compact. Indeed, consider a quadratic form S on the vector space EO of all displacements from an origin O in Euclidean space. We then consider e = inf S(u, u) , u∈SO where SO is the unit sphere in EO: SO = u EO : u = 1 . This is the lowest eigenvalue of S. Here, since the unit sphere{ is∈ compact,| | the} infimum is achieved at some u1 SO, and also at u1, i.e. at some pair of antipodal points in SO; however, uniqueness∈ may not hold.− The infimum may be achieved on a great 50 2. EINSTEIN’S FIELD EQUATIONS ∗ circle, when the first two eigenvalues coincide, or even on the whole sphere, if all three eigenvalues coincide. In the last case 2 S(u, u) = e u : for all u EO , | | ∈ i.e. S is then proportional to the Euclidean inner product. + In other words, a minimizing sequence, i.e. a sequence (un) Hx such that + ⊂ T (un, un) ρ as n , may run off to infinity in Hx . Then the vectors un would approach→ a generator→ ∞ of the future-directed null cone at x. Moreover, even if the infimum is achieved at some u H+, it may not be unique. ∈ x Example: Let K be a future-directed null vector and T = K K, ⊗ i.e. T µν = KµKν . Then ρ = 0 but the limit is not achieved. To see this + we construct a sequence of timelike vectors un H as follows: Let E0 ∈ x be a timelike vector at x, and take un to lie in the plane spanned by E0 and K. There is another null vector K such that g(K,K) = 2; then − un = an K + anK ,
1 = g(un, un) = 2 anan g(K,K) = 4 anan . − − Consider a sequence such that an . So → ∞ 2 2 2 1 T (un, un) = (g(K, un)) = ( 2an) = 4an = 2 0 , − 4an → and thus ρ(x) = 0, but it is not achieved. A material continuum2 is matter whose energy-momentum-stress tensor T has at each point x where it does not vanish the property that (i) the infimum is achieved, + (ii) it is achieved at a unique ux H . ∈ x Then the vector ux is called the material velocity. (And u is a vectorfield defined on the support of the matter.) It is in particular a critical point of T (u, u) under + the constraint that u H . So ux is an eigenvector of T : ∈ x ν ν (1.11) Tµνu = ρ gµνu , x − x where ρ is the corresponding eigenvalue. Remark. We may view this as a constraint variational problem: Minimize µ ν µ ν Tµνu u under the constraint that gµνu u = 1 (and u future-directed). By the method of Lagrange multipliers, the condition− for u to be a critical point is ν ν Tµνu = λgµνu , where λ is the Lagrange multiplier, or eigenvalue. Multiplying by uµ we obtain λ = T (u, u), and so λ = ρ. − − 2also “ponderable matter”. 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 51
Example: The electromagnetic field is not a material continuum. Since the Faraday tensor is antisymmetric,
Fµν = Fνµ , − there are two conjugate null eigenvectors L, L such that ν ν FµνL = α gµνL ν ν FµνL = α gµνL − g(L, L) = 2 . − Then F = α + β⊥ , where is the area 2-form of the timelike plane Π spanned by L, and L, and ⊥ is the area 2-form of the spacelike plane which is orthogonal to the timelike plane Π. In rectangular coordinates (t, x, y, z) at a given point, F = α dt dx + β dy dz , ∧ ∧ that is the electric and magnetic fields E, B only have x-components: ∂ ∂ E = α ,B = β . − ∂x ∂x As we shall see (c.f. note on page 2, in particular (0.29)), the energy- momentum-stress tensor in this situation takes the form: α2 + β2 0 0 0 t 1 0 α2 β2 0 0 x (1.12) T = − − 8π 0 0 α2 + β2 0 y 0 0 0 α2 + β2 z + The infimum is achieved on the curve Hx Π, and thus in particular not unique. (C.f. exercises below for the details∩ of this proof.) 1.1.2. Proper stress tensor. Given a material continuum let us now define the proper stress tensor S (not to be confused with Ricci curvature) by (1.13) S = T ρ u u . − ⊗ Then (1.14a) Sµν = T µν ρ uµuν µ µν µ −µ ν (1.14b) S λ = S gνλ = T ν ρ u uλ , uλ = u gνλ , µ− (1.14c) Sκλ = gκµS = Tκλ ρ uκuλ , λ − and we have λ λ (1.15) Sκλu = Tκλu + ρ uκ = ρuκ + ρuκ = 0 . − S is a 2-covariant symmetric tensor vanishing on u, so S is in fact a tensor on ⊥ Σx = ux : the local simultaneous space at x. The metric gx restricted to Σx is positive definite. So we have the standard eigenvalue problem for S on Σx. There 52 2. EINSTEIN’S FIELD EQUATIONS ∗ are 3 eigenvalues p1, p2, p3 called the principal pressures, and 3 corresponding eigenvalues E1, E2, E3, which form an orthonormal basis for Σx. Prop 1.1. For a material continuum, the dominant energy condition is equiv- alent to:
pi ρ for i = 1, 2, 3 . | | ≤ Proof. We show that this inequality implies the positivity condition, and we leave the converse as an exercise. Choose a basis (Eµ : µ = 0,..., 3) such that E0 = u is the material velocity and Ei : i = 1, 2, 3 are the eigenvectors of the proper stress S. Let v be a timelike future-directed vector, s 0 X i 0 X i 2 v = v E0 + v Ei , v > (v ) ; i i then we need to show that T v is a future directed causal vector. We decompose also T v with respect to− the· above basis: − · T v = εu + p , p u⊥ . − · ∈ Then ε = g(T v, u) = ρ g(u, v) = ρv0 , · − and 3 3 X i X i p = S v = v S Ei = v piEi , − · − · − i=1 i=1 and therefore 3 3 2 X i 2 2 X i2 2 02 2 p = v pi ρ v < ρ v = ε , | | ≤ i=1 i=1 because pi ρ. | | ≤ Remark. The proper stress tensor S has the following physical meaning: We know S lives on Σu, the local simultaneous space of the material (it vanishes on u). Let Π be a plane in Σu through the origin. There is a covector N of unit length whose null space is Π, i.e. Π = X Σ: N X = 0 . Π is oriented, so it + { ∈ · − } divides Σu into a postive side Σu , and a negative side Σu :
+ n o − n o Σ = X Σu : N X > 0 , Σ = X Σu : N X < 0 . u ∈ · u ∈ · The vector
S N Σu · ∈ represents the force per unit area exerted by the material across the surface with tangent plane Π. 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 53
1.1.3. Perfect fluids. A special case of particular interest is if all principal pressures are equal:
(1.16) p1 = p2 = p3 = p .
Then the material is called a perfect fluid, and the common value of pi : i = 1, 2, 3 is called the pressure p. Moreover, in that case (1.17) S = p g . Σx Σx If X, Y are arbitrary vectors at x , then ∈ M (1.18) S(X,Y ) = S(ΠX, ΠY ) = p g(ΠX, ΠY ) , where Π is the orthogonal projection to Σx. In terms of components in an arbitrary frame (Eµ : µ = 0, 1, 2, 3) we have X µ X µ (1.19a) X = X Eµ , ΠX = (ΠX) Eµ , µ µ (1.19b) ΠX = X + g(u, X)u , g(u, u) = 1 , µ µ ν µ µ µ− (1.19c) (ΠX) = Π νX , Π ν = δ ν + u uν , and it follows that (1.20) Sµν = p gµν + uµuν . Thus
(1.21) Tµν = ρ uµuν + Sµν = p gµν + (ρ + p) uµuν is the energy-momentum-stress tensor of a perfect fluid.
Exercises We study the energy-momentum-stress tensor of the electromagnetic field: 1 1 T = F F α − g F F αβ . µν 4π µα ν 4 µν αβ Recall that in a given orthonormal frame its components in terms of the electric field E and magnetic field B are given by (0.29). 1. (Proper energy density of an electromagnetic plane wave) Consider an electromagnetic plane wave in Minkowski space. We assume that in stationary coordinates (t, x, y, z) the wave propagates in the positve x-direction, and that the components of the electric field E and the magnetic field B are only a function of u = t − x (here c = 1): ∂ ∂ E = E (t − x) + E (t − x) y ∂y z ∂z ∂ ∂ B = B (t − x) + B (t − x) . y ∂y z ∂z Moreover, we assume E and B are compactly supported in u. (1) Use Maxwell’s equations to show that
By = −Ez,Bz = Ey. Sketch E, B in the y-z-plane, and their support on the t-x-plane. Calculate the magnitude of the momentum vector E × B. 54 2. EINSTEIN’S FIELD EQUATIONS ∗
(2) Recall the definition of the proper energy density (1.10). Show that ρ vanishes but that the infimum is not attained. Give a physical interpretation. + Hint: Consider a sequence (un) in Π ∩ Hp , where Π the t-x-plane (and p a point in the support of E and B).
2. (Proper energy density of the electromagnetic field) Here we shall show that the electromagnetic field is not a material continuum in general. To find a convenient frame, first consider the eigenvalue problem for the Faraday tensor F : ν ν Fµν v = λ gµν v
Here λ ∈ C is a root of the characteristic polynomial: det A(λ) = 0, where A = (Aµν ) is the matrix with components,
Aµν (λ) := Fµν − λ gµν , formed from the electromagnetic field tensor Fµν and the metric gµν at a point. Recall that 4 Fµν = −Fνµ. The vector v is the corresponding eigenvector, an element v ∈ C . (1) Show that if λ is an eigenvalue of F so is −λ and λ. There are 4 complex roots of the characteristic polynomial. Show that the following generic cases are possible: (a) Four real eigenvalues: 0 < λ1 < λ2, λ3 = −λ1, λ4 = −λ2. (b) Four imaginary eigenvalues: iµ1, iµ2, iµ3 = −iµ1, iµ4 = −iµ2, where µ1, µ2 ∈ R, 0 < µ1 < µ2. (c) Four complex eigenvalues: λ + iµ, λ − iµ, −λ − iµ, −λ + iµ, where λ, µ ∈ R. (d) Two real and two imaginary eigenvalues: λ, −λ, iµ, −iµ, where λ, µ ∈ R. (2) Show that only the case (d) is possible in the physical case of 3 + 1 dimensions. Hints: (a) Given a real eigenvalue λ the corresponding eigenvector L is also real. Show that in the case (a) of two real eigenvalues 0 < λ1 < λ2 the corresponding eigenvectors L1, L2 satisfy
g(Li,Li) = 0, g(Li,Lj ) = 0 , simply by using the antisymmetry of F :
F (Li,Li) = 0,F (Li,Lj ) = −F (Lj ,Li) . Then deduce a contradiction. (b) To a complex eigenvalue λ corresponds a complex eigenvectors M. In (b), let M1 = R1 + iS1 and M2 = R2 + iS2 be the complex eigenvectors corresponding to the imaginary eigenvalues iµ1, iµ2 (Ri, Si real vectors). Similarly to (i) show that the anti-symmetry of F implies
g(Ri,Si) = 0, g(Ri,Ri) = g(Si,Si) as well as
g(Ri,Rj ) = g(Si,Sj ) = 0, g(Ri,Sj ) = 0. Show that this is impossible in 3 + 1 dimensions. (c) For (c) use the fact that if M is a complex eigenvector corresponding to the complex eigenvalue λ it follows that M is a complex eigenvector corresponding to the complex eigenvalue λ. Then proceed similary to (a) and (b) to rule out case (c). (3) Let F be given by F = α dt ∧ dx + β dy ∧ dz, Show that this form of F corresponds to the physical case (d) above and find the eigenvalues in terms of α, β. 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 55
Verify that ∂ ∂ ∂ ∂ L = + ,L = − , ∂t ∂x ∂t ∂x ∂ ∂ ∂ ∂ M = + i , M = − i , ∂y ∂z ∂y ∂z are the corresponding eigenvectors. Conversly, show that in the physical case (d) generically we can choose the suitable Lorentz frame such that F is of the above form. (4) Finally, verify that in the above frame (ii) the energy-momentum stress tensor takes the form (1.12). Show that the infimum of T (u, u) is assumed on the curve u ∈ Π ∩ H+ where Π is the t-x-plane, and is thus in particular not uniquely achieved.
1.2. Local energy-momentum conservation laws. We now view T as a symmetric 2-covariant tensorfield. The energy-momentum-stress tensor satisfies the conservation laws: (1.22) T = 0 . ∇ · In terms of components relative to an arbitrary frame (Eµ : µ = 0, 1, 2, 3) we have αβ (1.23) E T = ( µT )Eα Eβ ; ∇ µ · ∇ ⊗ and T is a vectorfield ∇ · µν (1.24) T = ( νT )Eµ . ∇ · ∇ The energy-momentum conservation laws are therefore µν (1.25) νT = 0 , ∇ expressed in components with respect to an arbitrary frame. Remark. For a material continuum these are the equations of motion of the continuum. In particular, this encompasses Newton’s law of motion. Remark. Note that in the case of the perfect fluid (1.21) the equations of motion (1.25) are underdetermined: All in all there are 4 equations for 5 unknowns (ρ, p, uµ: 3 independent components). One possibility is to study a barotropic fluid where ρ = ρ(p) . In general, let v be the specific volume per particle, s the specific entropy, and e the energy per particle (this contains mc2 where m is the rest mass per particle). The mechanical properties of a fluid are specified once we stipulate the caloric equation of state: (1.26) e = e(v, s) According to the first law of thermodynamics we have (1.27) de = p dv + θ ds , − 56 2. EINSTEIN’S FIELD EQUATIONS ∗ where p is the pressure and θ the temperature. Let now be ρ = e/v the energy per unit volume, and n = 1/v the number of particles per unit volume, and define the particle current: (1.28) Iµ = nuµ . Then the additional equation is the differential conservation law of particle num- ber: µ (1.29) µI = 0 . ∇ 1.2.1. Quasi-Newtonian hierarchy. We analyze the components of the energy- momentum-stress tensor in a quasi-Newtonian situation. Let (E0,E1,E2,E3) be an orthonormal frame field, E0 being the 4-velocity of a background system of observers in Minkowski space. We express the components of the material velocity in conventional units, (where c denotes the speed of light): 3 3 0 X i 0 2 X i 2 (1.30) u = u E0 + u Ei , g(u, u) = (u ) + (u ) = 1 , − − i=1 i=1 i 1 v (1.31) u0 = , ui = c q 2 q 2 1 |v| 1 |v| − c2 − c2 Recall (1.14) (1.32) T µν = ρuµuν + Sµν , where the proper stress tensor S has the property (1.15) that µν (1.33) S uν = 0 . In the quasi-Newtonian limit this reads: 00 0i 00 0-component: S u0 + S ui = 0 S = 0 as c , −→ → ∞ (1.34) i0 ij i0 i-component: S u0 + S uj = 0 S = 0 as c , −→ → ∞ because
(1.35) u0 1 , and ui 0 , as c . −→ − −→ → ∞ Thus in the non-relativistic limit Sµν reduces to Sij, and the condition (1.15) reduces to (1.36) Sµ0 = 0 . Here T has the units of energy density, and ρ is dominated by the contribution of the rest mass density µ, i.e. (1.37) ρ µc2 as c . ∼ → ∞ We thus obtain the quasi-Newtonian hierarchy: T 00 = ρ(u0)2 + S00 µc2 : T 00 = (c2) ∼ O (1.38) T 0i = ρu0ui + S0i µcvi = cpi : T 0i = (c) ∼ O T ij = ρuiuj + Sij µvivj + Sij : T ij = (1) ∼ O 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 57
Here pi = µvi is the spatial momentum density, and µvivj is also referred to as kinematic stress.
Notes on the Newtonian limit of the conservation laws It might be instructive here to also write out the conservation laws (1.25) in the quasi- Newtonian situation (1.38). On Minkowski space (0.18), the components of (1.25) are µν ∂µT = 0 : 3 ∂T 00 X ∂T 0i + = 0 : µ = 0 , ∂x0 ∂xi i=1 3 ∂T i0 X ∂T ij + = 0 : µ = i . ∂x0 ∂xj j=1 However, in view of (1.38) – and noting that x0 = ct – in the limit c these become: → ∞ 3 ∂µ X ∂pi + = 0 , ∂t ∂xi i=1 3 ∂pi X ∂T ij + = 0 , where T ij = µvivj + Sij . ∂t ∂xj j=1 Consider the flow lines of the non-relativistic material velocity, dxi = vi(t, x(t)) ; dt here the velocity vectorfield v is assumed to be known. This defines a map, Ft(y) = x(t, y), where x(t, y) is the flow line with initial condition xi(0) = yi; here x = (x1, x2, x3), 1 2 3 y = (y , y , y ). Given a “portion” Ω of the material, let us denote by Ωt the same collection of particles in the material under time evolution, namely
Ωt = Ft(Ω) . Let us then consider the mass M and momentum P of the material portion: Z M(t) = µ dV Ωt Z P i(t) = pi dV Ωt In general, for any function f on spacetime, we have: d Z Z ∂f Z f dV = dV + f v dS, dt Ωt Ωt ∂t ∂Ωt · where dS is the oriented area element of the boundary ∂Ωt, i.e. dS = NdA, where N is the unit outward normal, and dA is the scalar area element. Thus by Gauss’ divergence theorem d Z Z ∂f f dV = + (fv) dV dt Ωt Ωt ∂t ∇ · 58 2. EINSTEIN’S FIELD EQUATIONS ∗
In rectangular coordinates dV = dx1dx2dx3 and (fv) = ∂ (fvi). Therefore: ∇ · i dM Z n∂µ ∂(µvi)o = + i dV = 0 , dt Ωt ∂t ∂x i.e. conservation of mass is equivalent to the first conservation law. Furthermore, dP i Z n∂pi ∂(pivj)o = + j dV, dt Ωt ∂t ∂x where pivi = µvivj is the kinematic stress. Now, by Newton’s second law, the rate of change of the momentum equals the force F which is exerted on the material by its surroundings: dP i = F i . dt Newton’s third law asserts in turn that this is equal and opposite the force exerted by the portion of the material on its surroudnings. So Z Z Z ij i ij ij ∂S F = S Nj dA = S dSj = j dV, − ∂Ωt ∂Ωt Ωt ∂x ij where S Nj is i-th component of the force per unit area across ∂Ωt, and dSj = NjdA. We conclude that the conservation of momentum corresponds to the differential conservation law, ∂pi ∂ + µvivj + Sij = 0 . ∂t ∂xj
Exercises In these exercises we study the equations of motion of a perfect fluid. These are the conservation laws (1.25) and (1.29) for the energy-momentum-stress tensor (1.21) and the particle current (1.28). 1. (Non-relativistic limit of the conservation laws.) We consider the equations of motion in Minkowski space and take the material velocity u to be of the form (1.31) in rectangular coordinates. Moreover, we assume that ρ = µc2 + h where µ is the mass density and h is the internal energy density; c.f. (1.37). (1) Derive in the non-relativistic limit c → ∞ the conservation of mass law, ∂µ + ∇ · (µv) = 0 . ∂t Hints: Express the components of the energy-momentum-stress tensor explicitly in orders of c, and derive the conservation law from the 0-component of (1.25). (2) Derive in the nonrelativistic limit c → ∞ the energy conservation law ∂ε + ∇ · f = 0 , ∂t 1 2 where ε is the total energy density given by ε = 2 µ|v| + h and f is the total energy flux given by f i = (ε + p) vi. Hints: Consider the current J µ = T 0µ − µc2uµ , and proceed similarly to (1). 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 59
2. (Adiabatic condition.) In addition to the equations of motion (1.25) and (1.28) we have the equation of state (1.26) which expresses the energy per particle e as a function of the volume per particle v and the entropy per particle s. We have ρ = e/v, n = 1/v, and according to the first law of thermodynamics the pressure p and the temperature θ are then given by (1.27). Prove the adiabatic condition
∇us = 0 , namely that the entropy is constant along the flow lines of the fluid. µν Hints: Consider the u component of (1.25), i.e. the equation uν ∇µT = 0, and use (1.29) in µ conjuction with the above relations to show that u ∇µs = 0. 3. (Pressureless fluid.) Show that if p = 0 then the integral curves of u are timelike geodesics.
1.2.2. Considerations for the field equations. Let us now return to Einstein’s vacuum equations
(1.39) Sµν = 0 , where from now on S denotes again the Ricci curvature: α (1.40) Sµν = R µαν .
The component S00 = S(E0,E0) is taken with respect to a unit future-directed 2 tangent field to a reference geodesic Γ0. Recall that Ri0j0 corresponds to ( ψ)ij ∇ in the Newtonian theory, whence we get the correspondence3:
(1.41) S00 4π T00 ∼ S00 = tr M = 4πµ : in the Newtonian theory. However, in view of the quasi-Newtonian hierarchy (1.38) there is a certain am- biguity as to what the right hand side of (1.39) in the presence of matter should be. In fact, if we set
(1.42) Sµν = a Tµν + b gµν tr T, in the presence of matter, then we obtain the Newtonian theory in the limit c for all a, b such that → ∞ (1.43) a + b = 4π . For, with respect to an orthonormal frame 3 −1 µν X (1.44) tr T = (g ) Tµν = T00 + Tii = T00 + (1) , − − O i=1 and so in the non-relativisitic limit:
(1.45) S00 = a T00 b tr T = (a + b) T00 . − This ambiguity will be broken by the requirement that the conservation laws (1.22) are a consequence of the field equations. We shall illustrate that logic first with the example of Maxwell’s theory; (c.f. note on pg. 48).
3in units where c = 1, and G = 1. 60 2. EINSTEIN’S FIELD EQUATIONS ∗
1.2.3. Conservation laws in electromagnetic theory. In Maxwell’s theory the electromagnetic field F is a 2-form on Minkowski space . The homogeneous equations are M
dF = 0(1.46) (1.47) F = dA, where A is determined from F only up to a differential of a function f, i.e. A0 is equivalent to A if
(1.48) A0 = A + df (gauge transformations).
In an arbitrary system of local coordinates (1.47) reads
(1.49) Fµν = ∂µAν ∂νAµ , − and (1.46) is an equation for the cyclic sum:
(1.50) ∂µFνλ + ∂νFλµ + ∂λFµν = 0 .
The inhomogeneous equations are
µν µ (1.51) νF = 4πJ ∇ µν −1 µκ −1 νλ (1.52) F = (g ) (g ) Fκλ , where J µ denotes the electric current density.
Remark. In macroscopic media the inhomogeneous equations are replaced by
µν µ (1.53) νG = 4πJ , ∇ where Gµν is the electromagnetic displacement. Similary to F , which decomposes into the electric, and magnetic field E, B,
k (1.54) F0i = Ei ,Fij = εijkB ,
G decomposes into the electric and magnetic displacement fields D,H:
k (1.55) G0i = Di ,Gij = εijkH .
Since
µν (1.56) µ νF = 0 : identically, ∇ ∇ the inhomogeneous equations (1.51) imply the conservation of the electric current:
µ (1.57) µJ = 0 ; ∇ namely local conservation of electric charge. 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 61
1.2.4. Conservation laws and the field equations. In general relativity the equa- tions of motion are: µν (1.58) νT = 0 . ∇ We require that these should be the consequence of the field equations of gravita- tion. In (1.42) we have seen that (1.59) Sµν = a T µν + κ (g−1)µν tr T has the correct Newtonian limit provided b (1.60) κ = , a(1 + κ) = 4π . a Taking the trace, we obtain (1.61) tr S = a(1 + 4κ) tr T, and so κ (1.62) aκ tr T = tr S. 1 + 4κ Inserting this back into (1.59) we have equally: (1.63) Sµν λ (g−1)µν tr S = a T µν , − κ where λ = 1+4κ . The requirement that (1.58) are a consequence of the field equations is therefore satisfied if and only if µν −1 µν (1.64) ν S λ(g ) tr S = 0 : identically. ∇ − In the next section we prove that this is satisfied if and only if 1 1 (1.65) λ = , i.e. κ = . 2 −2 So the Einstein equations in the presence of matter are: 1 (1.66) Sµν (g−1)µν tr S = a T µν − 2 In units where c = 1 and G = 1: a = 8π. The equations are often simply written as (1.67) Eµν = a T µν , where E is the Einstein tensor: 1 (1.68) Eµν = Sµν (g−1)µν tr S. − 2 In rationalized gravitational units, (1.69) 4πG = 1 , a = 2 . 1.3. Bianchi identities. A this Section we prove certain identities for the curvature tensor, known as the Bianchi identities, which are central to Einstein’s equations. 62 2. EINSTEIN’S FIELD EQUATIONS ∗
1.3.1. Jacobi identity. Let A, B, C be linear operators acting on some (linear) space X. Then we have the Jacobi identity: (1.70) [A, [B,C]] + [B, [C,A]] + [C, [A, B]] = 0 If X is vectorfield on it defines two kinds of linear operators: M (1) X acts on functions f by f X f. 7→ · (2) X acts on vectorfields Y by X X Y . 7→ ∇ The first kind (together with the symmetry of the connection) yields the cyclic identity of the curvature: (1.71) R(X,Y ) Z + R(Y,Z) X + R(Z,X) Y = 0 . · · · The second kind yields the Bianchi identity:
(1.72) ( X R)(Y,Z) + ( Y R)(Z,X) + ( Z R)(X,Y ) = 0 . ∇ ∇ ∇ Let us first return to (1). We have (1.73) [X, [Y,Z]]f + [Y, [Z,X]]f + [Z, [X,Y ]]f = 0 . Using the symmetry of the connection we have
(1.74) [X, [Y,Z]] = [X, Y Z Z Y ] = X Y Z X Z Y + X. ∇ − ∇ ∇ ∇ − ∇ ∇ ∇[Z,Y ] And thus,
0 = X Y Z X Z Y + X ∇ ∇ − ∇ ∇ ∇[Z,Y ] (1.75) + Y Z X Y X Z + [X,Z]Y ∇ ∇ −∇ ∇ ∇ + Z X Y Z Y X + Z ∇ ∇ − ∇ ∇ ∇[Y,X] Now recalling the definition of curvature (3.39) from Chapter 1 we see that group- ing the terms as indicated this implies (1.71). Setting X = Eµ, Y = Eν, and Z = Eβ, where (Eµ : µ = 0, 1, 2, 3) is an arbitrary frame, we can also write α α α α α (1.76) R βµν + R µνβ + R νβµ = 0 , where R βµν = (Rµν) β ; this together with the symmetries α α (1.77) R = R ,Rβαµν = Rαβµν , βνµ − βµν − implies the pair symmetry:
(1.78) Rµναβ = Rαβµν . Now consider (2). We have
(1.79) [ X , [ Y , Z ]]W = X [ Y , Z ]W [ Y , Z ] X W ∇ ∇ ∇ ∇ ∇ ∇ − ∇ ∇ ∇ = X R(Y,Z) W + W R(Y,Z) X W X W ∇ · ∇[Y,Z] − · ∇ − ∇[Y,Z]∇ = ( X R)(Y,Z) W + R( X Y,Z) W + R(Y, X Z) W ∇ · ∇ · ∇ · + R(X, [Y,Z]) W + W. · ∇[X,[Y,Z]] 1. EINSTEIN’S FIELD EQUATIONS IN THE PRESENCE OF MATTER 63
Therefore,
0 = ( X R)(W, Z) W + ( Y R)(Z,X) W + ( Z R)(X,Y ) W ∇ · ∇ · ∇ · +R( X Y,Z) W + R(Y, X Z) W ∇ · ∇ · + R( Y Z,X) W +R(Z, Y X) W (1.80) ∇ · ∇ · + R( Z X,Y ) W + R(X, Z Y ) W ∇ · ∇ · + R(X, [Y,Z]) W + R(Y, [Z,X]) W +R(Z, [X,Y ]) W · · · + W + W + W ; ∇[X,[Y,Z]] ∇[Y,[Z,X]] ∇[Z,[X,Y ]] the last line vanishes identically by the Jacobi identity (1.70), while the remaining terms cancel as indicated. This proves the Bianchi identity (1.72). With respect to arbitrary frame Eµ : µ = 0,..., 3, setting X = Eµ, Y = Eν, Z = Eλ, W = Eβ, α α (1.81) ( X R)(Y,Z) W = ( E R)νλ Eβ = (( µR)νλ) Eα = ( µR )Eα , ∇ · ∇ µ · ∇ β ∇ βνλ the Bianchi identity reads: α α α (1.82) µR + νR + λR = 0 . ∇ βνλ ∇ βλµ ∇ βµν 1.3.2. Contracted Bianchi identities. Setting α = ν and noting that X ν (1.83) R βνλ = Sβλ , ν we obtain after summation α (1.84) µSβλ + αR λSβµ = 0 , ∇ ∇ βλµ − ∇ or α (1.85) αR = λSµβ µSλβ ∇ βλµ ∇ − ∇ the first contracted Bianchi identity. Remark. Note that the right hand side is antisymmetric in (λ, µ) and the equation can expressed schematically as div R = curl S. Now multiply the contracted Bianchi identity α −1 µλ (1.86) αR = νSλµ λSνµ by (g ) . ∇ µνλ ∇ − ∇ Note on the left hand side −1 µλ α −1 αβ −1 µλ −1 αβ α (1.87) (g ) R µνλ = (g ) (g ) Rβµνλ = (g ) Sβν = S ν , while on the right hand side: −1 µλ (1.88) (g ) Sλµ = tr S −1 λµ α (1.89) (g ) λSνµ = αS . ∇ ∇ ν 64 2. EINSTEIN’S FIELD EQUATIONS ∗
We obtain the second contracted Bianchi identity:
µ 1 (1.90) µS ν tr S = 0 ∇ ν − 2∇ 1.3.3. Relation to Einstein equations. We have derived that