Arxiv:0911.0334V2 [Gr-Qc] 4 Jul 2020

Home , Antisymmetric tensor, Einstein tensor, Nonmetricity tensor, Pseudotensor, Tensor density

Nikodem Poplawski

Department of Mathematics and Physics, University of New Haven, CT, USA

Preface

We present a self-contained introduction to the classical theory of spacetime and fields. This exposition is based on the most general principles: the principle of general covariance (relativity) and the principle of least action. The order of the exposition is: 1. Spacetime (principle of general covariance and tensors, affine connection, curvature, metric, tetrad and spin connection, Lorentz group, spinors); 2. Fields (principle of least action, action for gravitational field, matter, symmetries and conservation laws, gravitational field equations, spinor fields, electromagnetic field, action for particles). In this order, a particle is a special case of a field existing in spacetime, and classical mechanics can be derived from field theory.

I dedicate this book to my Parents: Bo˙zennaPop lawska and Janusz Poplawski. I am also grateful to Chris Cox for inspiring this book.

The Laws of Physics are simple, beautiful, and universal. arXiv:0911.0334v2 [gr-qc] 4 Jul 2020

1 Contents

1 Spacetime 5 1.1 Principle of general covariance and tensors ...... 5 1.1.1 Vectors ...... 5 1.1.2 Tensors ...... 6 1.1.3 Densities ...... 7 1.1.4 Contraction ...... 7 1.1.5 Kronecker and Levi-Civita symbols ...... 8 1.1.6 Dual densities ...... 8 1.1.7 Covariant integrals ...... 9 1.1.8 Antisymmetric derivatives ...... 9 1.2 Affine connection ...... 10 1.2.1 Covariant differentiation of tensors ...... 10 1.2.2 Parallel transport ...... 11 1.2.3 Torsion tensor ...... 11 1.2.4 Covariant differentiation of densities ...... 12 1.2.5 Antisymmetric covariant derivatives ...... 13 1.2.6 Partial integration ...... 14 1.2.7 Geodesic frame of reference ...... 14 1.2.8 Affine geodesics and four-velocity ...... 15 1.2.9 Infinitesimal coordinate transformations ...... 16 1.2.10 Killing vectors ...... 17 1.3 Curvature ...... 18 1.3.1 Curvature tensor ...... 18 1.3.2 Integrability of connection ...... 19 1.3.3 Parallel transport along closed curve ...... 20 1.3.4 Bianchi identities ...... 20 1.3.5 Ricci tensor ...... 21 1.3.6 Geodesic deviation ...... 21 1.4 Metric ...... 22 1.4.1 Metric tensor ...... 22 1.4.2 Christoffel symbols ...... 24 1.4.3 Riemann tensor ...... 27 1.4.4 Properties of Riemann tensor ...... 28 1.4.5 Metric geodesics ...... 29 1.4.6 Galilean frame of reference and Minkowski tensor ...... 30 1.4.7 Riemann normal coordinates ...... 31 1.4.8 Intervals, proper time, and distances ...... 32 1.4.9 Spatial vectors ...... 35 1.4.10 Embedded hypersurfaces ...... 37 1.4.11 Event horizon ...... 43 1.5 Tetrad and spin connection ...... 43 1.5.1 Tetrad ...... 43 1.5.2 Lorentz transformation ...... 44 1.5.3 Tetrad transport ...... 44 1.5.4 Spin connection ...... 45 1.5.5 Tetrad representation of curvature tensor ...... 46 1.6 Lorentz group ...... 47 1.6.1 Subgroups of Lorentz group and Einstein principle of relativity ...... 47 1.6.2 Infinitesimal Lorentz transformations ...... 48 1.6.3 Generators and Lie algebra of Lorentz group ...... 48 1.6.4 Rotations and boosts ...... 49 1.6.5 Poincarégroup ...... 51

2 1.6.6 Invariants of Lorentz and Poincar´egroup ...... 53 1.6.7 Relativistic kinematics ...... 54 1.6.8 Four-acceleration ...... 57 1.7 Spinors ...... 59 1.7.1 Spinor representation of Lorentz group ...... 59 1.7.2 Spinor connection ...... 60 1.7.3 Curvature spinor ...... 62

2 Fields 63 2.1 Principle of least action ...... 63 2.2 Action for gravitational field ...... 64 2.3 Matter ...... 65 2.3.1 Metric energy-momentum tensor ...... 65 2.3.2 Tetrad energy-momentum tensor ...... 66 2.3.3 Canonical energy-momentum density ...... 66 2.3.4 Spin tensor ...... 67 2.3.5 Belinfante-Rosenfeld relation ...... 68 2.4 Symmetries and conservation laws ...... 69 2.4.1 Noether theorem ...... 69 2.4.2 Conservation of spin ...... 69 2.4.3 Conservation of metric energy-momentum ...... 70 2.4.4 Conservation of tetrad energy-momentum ...... 72 2.4.5 Conservation laws for Lorentz group ...... 73 2.4.6 Momentum four-vector ...... 74 2.4.7 Mass ...... 76 2.4.8 Angular momentum four-tensor ...... 76 2.4.9 Energy-momentum tensor for particles ...... 79 2.4.10 Spin tensor for particles ...... 83 2.4.11 Relativistic ideal fluids ...... 85 2.4.12 Multipole expansion of spin tensor ...... 88 2.4.13 Multipole expansion of energy-momentum tensor ...... 88 2.4.14 Mathisson-Papapetrou-Dixon equations ...... 90 2.5 Gravitational field equations ...... 94 2.5.1 Einstein-Cartan action and equations ...... 94 2.5.2 Sciama-Kibble action ...... 96 2.5.3 Einstein-Hilbert action and Einstein equations ...... 97 2.5.4 Utiyama action ...... 99 2.5.5 Einstein pseudotensor and principle of equivalence ...... 99 2.5.6 Møller pseudotensor ...... 102 2.5.7 Landau-Lifshitz energy-momentum pseudotensor ...... 104 2.5.8 Palatini variation ...... 106 2.5.9 Gravitational potential ...... 107 2.5.10 Raychaudhuri equation ...... 108 2.5.11 Relativistic spin fluids ...... 109 2.6 Spinor fields ...... 111 2.6.1 Dirac matrices ...... 111 2.6.2 Lagrangian density and spin tensor for spinor field ...... 114 2.6.3 Dirac equation ...... 115 2.6.4 Energy-momentum tensor for spinor field ...... 117 2.6.5 Discrete symmetries of spinors ...... 118 2.7 Electromagnetic field ...... 119 2.7.1 Gauge invariance and electromagnetic potential ...... 119 2.7.2 Electromagnetic field tensor ...... 121 2.7.3 Lagrangian density for electromagnetic field ...... 123

3 2.7.4 Electromagnetic current and electric charge ...... 124 2.7.5 Maxwell equations ...... 126 2.7.6 Energy-momentum tensor for electromagnetic ﬁeld ...... 128 2.7.7 Lorentz force ...... 129 2.8 Action for particles ...... 131

4 1 Spacetime 1.1 Principle of general covariance and tensors Physical processes are described in coordinate systems in four-dimensional spacetime, called systems of reference or frames of reference. The principle of general covariance or Einstein’s general principle of relativity states that physical laws do not change their form (are covariant) under arbitrary, diﬀerentiable (and thereby continuous) coordinate transformations. Equivalently, physical laws have the same form in all admissible frames of reference.

1.1.1 Vectors Let us consider a coordinate transformation from old (unprimed) to new (primed) coordinates in a four-dimensional manifold: xi → x0j(xi), (1.1.1) where x0j are differentiable and nondegenerate functions of xi and the index i (and the other Latin ∂x0j indices) can be 0,1,2, or 3. The corresponding transformation matrix ∂xi , which is four-dimensional ∂x0j i and square (4 × 4), has a nonzero determinant | ∂xi | 6= 0, thereby that x are differentiable and 0j ∂xi ∂x0j nondegenerate functions of x . The matrix ∂x0j is inverse to ∂xi : X ∂x0i ∂xk = δk, (1.1.2) ∂xj ∂x0i j i where 1 i = k δi = . (1.1.3) k 0 i 6= k A scalar (invariant) is defined as a quantity that does not change under coordinate transformations:

φ0 = φ. (1.1.4)

Accordingly, the diﬀerential of a scalar is also a scalar:

dφ0 = dφ. (1.1.5)

If φ(xi) is a scalar function of the coordinates xi, then its diﬀerential can be expressed as X ∂φ dφ(xi) = dxi. (1.1.6) ∂xi i

i ∂ Coordinate differentials dx and partial derivatives ∂i = ∂xi transform according to X ∂x0j dx0j = dxi, (1.1.7) ∂xi i ∂ X ∂xi ∂ X ∂xi = , ∂0 = ∂ . (1.1.8) ∂x0j ∂x0j ∂xi j ∂x0j i i i A contravariant vector is defined as a set of quantities that transform like coordinate differentials:

X ∂x0j A0j = Ai. (1.1.9) ∂xi i These quantities are referred to as the components of the contravariant vector. A covariant vector is deﬁned as a set of quantities that transform like partial derivatives of a scalar:

X ∂xi B0 = B . (1.1.10) j ∂x0j i i

5 These quantities are referred to as the components of the covariant vector. Therefore, coordinate diﬀerentials form a contravariant vector and partial derivatives of a scalar form a covariant vector. The coordinates xi do not form a vector. A linear combination of two scalars is a scalar. A linear combination aC + bD of two contravariant vectors C and D, where a and b are scalars, is a contravariant vector E whose components are Ei = aCi + bDi. A linear combination aC + bD of two covariant vectors C and D is a covariant vector E whose components are Ei = aCi + bDi. An upper index (in a contravariant vector) is called contravariant, and a lower index is called covariant. The derivative with respect to a quantity with a contravariant index i is a quantity with a covariant index i. Conversely, the derivative with respect to a quantity with a covariant index i is a quantity with a contravariant index i. Henceforth, we adopt the following Einstein’s summation convention. If the same coordinate index i appears in a given expression twice, as a contravariant index and a covariant index, and we apply the summation P in this expression, then we do not need P i to write the summation sign i. Accordingly, we can omit the summation signs in the formulae of this section.

1.1.2 Tensors A product of several vectors transforms under diﬀerentiable coordinate transformations such that each coordinate index transforms separately:

∂x0i ∂x0j ∂xp ∂xq A0iB0j ...C0 D0 ··· = AmBn ...C D .... (1.1.11) k l ∂xm ∂xn ∂x0k ∂x0l p q A tensor is deﬁned as a set of quantities that transform like products of the components of vectors:

∂x0i ∂x0j ∂xp ∂xq T 0ij... = T mn... . (1.1.12) kl... ∂xm ∂xn ∂x0k ∂x0l pq... These quantities are referred to as the components of the tensor. A tensor is of rank (k, l) if it has k contravariant and l covariant indices. A scalar is a tensor of rank (0,0), a contravariant vector is a tensor of rank (1,0), and a covariant vector is a tensor of rank (0,1). A linear combination of two tensors of rank (k, l) is a tensor of rank (k, l) such that its components are the same linear combinations of the corresponding components of the tensors. The product of two tensors of ranks (k1, l1) and (k2, l2) is a tensor of rank (k1 + k2, l1 + l2). Tensor indices (all contravariant or all covariant) can be symmetrized:

1 X T = T , (1.1.13) (ij...k) n! {ij...k} permutations or antisymmetrized: 1 X T = T (−1)N , (1.1.14) [ij...k] n! {ij...k} permutations where n is the number of symmetrized or antisymmetrized indices and N is the number of per- 1 mutations that bring Tij...k to T{ij...k}. For example, for two indices: T(ik) = 2 (Tik + Tki) and 1 1 T[ik] = 2 (Tik − Tki), and for three indices: T[ijk] = 3 (Tijk + Tjki + Tkij). If n > 4 then T[ij...k] = 0. Symmetrized and antisymmetrized tensors or rank (k, l) are tensors of rank (k, l). Symmetrization of an antisymmetric tensor or antisymmetrization of a symmetric tensor bring these tensors to zero. Any tensor of rank (0,2) is the sum of its symmetric and antisymmetric part,

T(ik) + T[ik] = Tik. (1.1.15)

The number 0 can be regarded as a tensor of arbitrary rank. Therefore, all covariant equations of ij... classical physics must be represented in the tensor form: T kl... = 0.

6 1.1.3 Densities The element of volume in four-dimensional spacetime transforms according to 0i 4 0 ∂x 4 d x = d x. (1.1.16) ∂xk A scalar density is defined as a quantity that transforms such that its product with the element of volume is a scalar, s0d4x0 = sd4x: i 0 ∂x s = s. (1.1.17) ∂x0k A tensor density, which includes a contravariant and covariant vector density, is defined as a set of quantities that transform like products of the components of a tensor and a scalar density: i 0i 0j p q 0ij... ∂x ∂x ∂x ∂x ∂x mn... T = T . (1.1.18) kl... ∂x0k ∂xm ∂xn ∂x0k ∂x0l pq... These quantities are referred to as the components of the tensor density. A tensor density is of rank (k, l) if it has k contravariant and l covariant indices. For example, the square root of the determinant of a tensor of rank (0, 2) is a scalar density of weight 1: s s q l m j 2 j 0 ∂x ∂x ∂x ∂x p |T | = Tlm = |Tik| = |Tik|. (1.1.19) ik ∂x0i ∂x0k ∂x0n ∂x0n The above densities are said to be of weight 1. One can generalize this definition of densities by introducing densitites of weight w, which transform according to i w 0 ∂x s = s. (1.1.20) ∂x0k For example, d4x is a scalar density of weight -1. A linear combination of two densities of rank (k, l) and weight w is a density of rank (k, l) and weight w such that its components are the same linear combinations of the corresponding components of the densities. The product of two densities of weights w1 and w2 is a density of weight w1 + w2. Symmetrized and antisymmetrized densities of weight w are densities of weight w. Densities of weight 1 are simply referred to as densities. Tensors are densities of weight 0.

1.1.4 Contraction Einstein’s summation convention also applies within the same tensor or tensor density, if a given coordinate index i appears twice (as a contravariant and covariant index). Such a tensor or density is said to be contracted over index i. A contracted tensor of rank (k, l) transforms like a tensor of rank (k − 1, l − 1): ∂x0i ∂x0j ∂xp ∂xq ∂x0j ∂xq ∂x0j ∂xq T 0ij... = T mn... = δp T mn... = T mn... . (1.1.21) il... ∂xm ∂xn ∂x0i ∂x0l pq... ∂xn ∂x0l m pq... ∂xn ∂x0l mq... i For example, the contraction of a contravariant and covariant vector A Bi is a scalar (scalar product). A contracted tensor density of rank (k, l) and weight w transforms like a tensor density of rank (k − 1, l − 1) and weight w: i w 0i 0j p q i w 0j q 0ij... ∂x ∂x ∂x ∂x ∂x mn... ∂x ∂x ∂x p mn... T = T = δ T il... ∂x0k ∂xm ∂xn ∂x0i ∂x0l pq... ∂x0k ∂xn ∂x0l m pq... i w 0j q ∂x ∂x ∂x mn... = T . (1.1.22) ∂x0k ∂xn ∂x0l mq... Contraction of a symmetric tensor with an antisymmetric tensor (over indices with respect to which these tensors are symmetric or antisymmetric) gives zero. If contraction of two tensors gives zero, these tensors are said to be orthogonal. Two orthogonal vectors (one contravariant and one covariant) are said to be perpendicular.

7 1.1.5 Kronecker and Levi-Civita symbols

i The Kronecker symbol δk (1.1.3) is a tensor with constant components: ∂x0i ∂xl ∂x0i ∂xj δ0i = δj = = δi . (1.1.23) k ∂xj ∂x0k l ∂xj ∂x0k k A completely antisymmetric tensor of rank (4, 0), T ijkl = T [ijkl] has 1 independent component T : T ijkl = T ijkl, where ijkl is the completely antisymmetric, contravariant Levi-Civita permutation symbol: 0123 = 1, ijkl = [ijkl] = (−1)N , (1.1.24) and N is the number of permutations that bring ijkl to 0123. The determinant of a square matrix i i i Sk, det(Sk) = |Sk|, is deﬁned through the permutation symbol: r ijkl i j k l mnpq |Ss | = SmSnSp Sq . (1.1.25)

i ∂x0i Taking Sk = ∂xk gives r 0i 0j 0k 0l ijkl ∂x ∂x ∂x ∂x ∂x mnpq = . (1.1.26) ∂x0s ∂xm ∂xn ∂xp ∂xq This equation looks like a transformation law for a tensor density (of weight 1) with constant components: 0ijkl = ijkl. Accordingly, T is a scalar density of weight -1. We also introduce the covariant Levi-Civita symbol εijkl through: i i i i δm δn δp δq j j j j ijkl δm δn δp δq εmnpq = − k k k k . (1.1.27) δm δn δp δq l l l l δm δn δp δq Therefore, the covariant Levi-Civita symbol is a tensor density of weight -1 and its product with a scalar density is a tensor. The covariant Levi-Civita symbol is given by N ε0123 = −1, εijkl = ε[ijkl] = (−1) , (1.1.28) where N is the number of permutations that bring εijkl to ε0123, and satisﬁes r m n p q |Ss |εijkl = Si Sj Sk Sl εmnpq. (1.1.29) Contracting (1.1.27) gives the following relations: i i i δm δn δp ijkl j j j εmnpl = − δm δn δp , k k k δm δn δp ijkl i j i j εmnkl = −2(δmδn − δnδm), ijkl i εmjkl = −6δm, ijkl εijkl = −24. (1.1.30)

1.1.6 Dual densities A contracted product of a covariant tensor and the contravariant Levi-Civita symbol gives a dual contravariant tensor density of weight 1: iklm ikl iklm ik iklm i Am = A , Blm = B , Cklm = C . (1.1.31) A contracted product of a contravariant tensor and the covariant Levi-Civita symbol gives a dual covariant tensor density of weight -1: m lm klm εiklmA = Aikl, εiklmB = Bik, εiklmC = Ci. (1.1.32) Therefore, there exists an algebraic correspondence between covariant tensors and contravariant densities of weight 1, and between contravariant tensors and covariant densities of weight -1.

8 1.1.7 Covariant integrals

i R j... i A covariant line integral is an integral of a tensor contracted with the line diﬀerential dx : T i...dx . A covariant surface integral is an integral of a tensor contracted with the surface diﬀerential df ik = dxidx0k −dxkdx0i (which can be geometrically represented as a parallelogram spanned by the vectors i 0i R j... ik dx and dx ): T ik...df . A covariant hypersurface (volume) integral is an integral of a tensor

dxi dx0i dx“i

contracted with the volume diﬀerential dSikl = dxk dx0k dx“k (which can be geometrically

dxl dx0l dx“l i 0i “i R j... ikl represented as a parallelepiped spanned by the vectors dx , dx ) and dx : T ikl...dS . A covariant four-volume integral is an integral of a tensor contracted with the four-volume differential dSijkl, defined analogously to dSikl. The dual density corresponding to the surface element is given by 1 df ? = ε df lm, (1.1.33) ik 2 iklm which gives 1 df lm = − lmikdf ? , df ikdf ? = 0. (1.1.34) 2 ik ik The dual density corresponding to the hypersurface element is given by 1 dS = − ε dSklm, dSklm = −klmidS . (1.1.35) i 6 iklm i The dual density corresponding to the four-volume element is given by 1 dΩ = ε dSiklm = dx0dx1dx2dx3. (1.1.36) 24 iklm Covariant integrands that include the above dual densities of weight -1 must be multiplied by a scalar density, for example, by the square root of the determinant of a tensor of rank (0, 2). According to Gauß’ and Stokes’ theorems, there exists relations between integrals over different elements: ∂ dxi ↔ df ki , (1.1.37) ∂xk ∂ ∂ df ? ↔ dS − dS , (1.1.38) ik i ∂xk k ∂xi ∂ dS ↔ dΩ . (1.1.39) i ∂xi

1.1.8 Antisymmetric derivatives A derivative of a covariant vector does not transform like a tensor: ∂A0 ∂ ∂xm ∂xm ∂A ∂2xm ∂xl ∂xm ∂A ∂2xm k = A = m + A = m + A , (1.1.40) ∂x0i ∂x0i ∂x0k m ∂x0k ∂x0i ∂x0i∂x0k m ∂x0i ∂x0k ∂xl ∂x0i∂x0k m

i because of the second term which is linear and homogeneous in Ai, unless x are linear functions of 0j ∂Ak x . This term is symmetric in the indices i, k, thereby the antisymmetric part of ∂xi with respect to these indices is a tensor: ∂xl ∂xm ∂xl ∂xm ∂0 A0 = ∂ A = ∂ A . (1.1.41) [i k] ∂x0[i ∂x0k] l m ∂x0i ∂x0k [l m]

The curl of a covariant vector Ai is deﬁned as twice the antisymmetric part of ∂iAk: ∂iAk − ∂kAi, ∂ i and is a tensor. We will also use ,i = ∂xi to denote a partial derivative with respect to x . Similarly, completely antisymmetrized derivatives of tensors of rank (0, 2) and (0, 3), ∂[iBkl] and ∂[iCklm], are tensors. If Bkl = A[k,l] then ∂[iBkl] = 0, or conversely, if ∂[iBkl] = 0 then there exists a vector Ai such that Bkl = A[k,l]. The divergence of a tensor (or density) is a contracted derivative of this tensor

9 il... (density): ∂iT jk.... Because of the correspondence between tensors and dual densities, divergences of (completely antisymmetric if more than 1 index) contravariant densities are densities, dual to completely antisymmetrized derivatives of tensors:

i iklm ik iklm ikl iklm ∂iC = ∂[iCklm], ∂kB = ∂[kBlm], ∂lA = ∂[lAm]. (1.1.42)

ik k For example, the equations F[ik,l] = 0 and F ,i = j , that describe Maxwell’s electrodynamics (confer (2.7.32) and (2.7.82)), are tensorial. References: [1, 2].

1.2 Aﬃne connection 1.2.1 Covariant diﬀerentiation of tensors

An ordinary derivative of a covariant vector Ai is not a tensor, because its coordinate transformation law (1.1.40) contains an additional noncovariant term, linear and homogeneous in Ai. Such a term vanishes only if xi are linear functions of x0j, that is, if the coordinate transformation from old to new coordinates (1.1.1) is linear. Let us consider the expression

l Ai;k = Ai,k − Γi kAl, (1.2.1)

l where the quantity Γi k (in the second term which is linear and homogeneous in Ai) transforms such that Ai;k is a tensor:

∂xl ∂xm ∂xl ∂xm A0 = A = (A − Γ n A ). (1.2.2) i;k ∂x0i ∂x0k l;m ∂x0i ∂x0k l,m l m n Also, (1.1.40) gives

∂xm ∂xl ∂2xn ∂xn A0 = A0 − Γ0 l A0 = A + A − Γ0 l A , (1.2.3) i;k i,k i k l ∂x0k ∂x0i l,m ∂x0k∂x0i n ∂x0l i k n so we obtain ∂xn ∂xl ∂xm ∂2xn Γ0 l = Γ n + . (1.2.4) ∂x0l i k ∂x0i ∂x0k l m ∂x0k∂x0i ∂x0j l Multiplying this equation by ∂xn gives the transformation law for Γi k: ∂x0j ∂xl ∂xm ∂x0j ∂2xn Γ0 j = Γ n + . (1.2.5) i k ∂xn ∂x0i ∂x0k l m ∂xn ∂x0k∂x0i l The algebraic object Γi k, which equips spacetime in order to covariantize a derivative of a vector, is referred to as the affine connection, affinity or simply connection. The connection has generally 64 independent components. The tensor Ai;k is the covariant derivative of a vector Ai with respect i to x . We will also use ∇i = ;i to denote the covariant derivative. The contracted affine connection transforms according to ∂xm ∂x0i ∂2xn Γ0 i = Γ l + . (1.2.6) i k ∂x0k l m ∂xn ∂x0k∂x0i The affine connection is not a tensor because of the second term on the right-hand side of (1.2.5). A derivative of a scalar is a covariant vector. Therefore, the covariant derivative of a scalar is equal to an ordinary derivative: φ;i = φ,i. (1.2.7) If we also assume that the covariant derivative of the product of two tensors obeys the same chain rule as an ordinary derivative: (TU);i = T;iU + TU;i, (1.2.8) then

k k k k k k k k l k Ak,iB + AkB ,i = (AkB ),i = (AkB );i = Ak;iB + AkB ;i = Ak,iB − Γl iAkB + AkB ;i. (1.2.9)

10 Therefore, we obtain the covariant derivative of a contravariant vector:

k k k l B ;i = B ,i + Γl iB . (1.2.10)

The chain rule (1.2.8) also infers that the covariant derivative of a tensor is equal to the sum of the corresponding ordinary derivative of this tensor and terms with the aﬃne connection that covariantize each index:

ij... ij... i nj... j in... n ij... n ij... T kl...;m = T kl...,m + Γn mT kl... + Γn mT kl... + · · · − Γk mT nl... − Γl mT kn... − .... (1.2.11) the covariant derivative of the Kronecker symbol vanishes:

k k j j k δl;i = Γj iδl − Γl iδj = 0. (1.2.12) The second term on the right of (1.2.5) does not depend on the affine connection, but only on the coordinate transformation. Therefore, the difference between two different connections transforms j like a tensor of rank (1,2). Consequently, the variation δΓi k, which is an infinitesimal difference between two connections, is a tensor of rank (1,2).

1.2.2 Parallel transport Let us consider two infinitesimally separated points in spacetime, P (xi) and Q(xi + dxi), and a k k k k k i vector field A which takes the value A at P and A + dA at Q. Because dA = A ,idx and k k k k k A ,i is not a tensor, the difference dA between the vectors A + dA and A is not a vector. The differential dAk is not a vector because it arises from subtracting two vectors which are located at two points with different coordinate transformation laws. The transformation law for dAk follows from (1.1.40):

∂xm ∂xm ∂xm ∂xm ∂2xm dA0 = d A = dA + d A = dA + A dx0i. (1.2.13) k ∂x0k m ∂x0k m ∂x0k m ∂x0k m ∂x0i∂x0k m

In order to calculate the covariant difference between two vectors at two different points, we must bring these vectors to the same point. Instead of subtracting from the vector Ak + dAk at Q the vector Ak at P , we must subtract a vector Ak + δAk at Q that corresponds to Ak at P , thereby that the resulting difference (covariant differential) DAk = dAk − δAk is a vector. The vector Ak + δAk is the parallel-transported or parallel-translated Ak from P to Q. A parallel-transported linear combination of vectors must be equal to the same linear combination of parallel-transported vectors. Therefore, δAk is a linear and homogeneous function of Ak. It is also on the order of a differential, thus a linear and homogeneous function of dxi. The most general form of δAk is

k k l i δA = −Γl iA dx , (1.2.14) so k k k l i k i DA = dA + Γl iA dx = A ;idx . (1.2.15) k k k k Because δA is not a vector, Γl i is not a tensor. Because DA is a vector, A ;i is a tensor. The expressions for covariant derivatives of a covariant vector and tensors result from

δφ = 0, δ(TU) = δT U + T δU. (1.2.16)

1.2.3 Torsion tensor The second term on the right-hand side of (1.2.5) is symmetric in the indices i, k. Antisymmetrizing (1.2.5) with respect to these indices gives

∂x0j ∂xl ∂xm S0j = Sn , (1.2.17) ik ∂xn ∂x0i ∂x0k lm

11 where j j S ik = Γ[i k] (1.2.18) is the antisymmetric (in the covariant indices) part of the aﬃne connection. Equation (1.2.17) is a transformation formula for a tensor, thereby (1.2.18) is a tensor, called the Cartan torsion tensor. The torsion tensor has generally 24 independent components. The contracted torsion tensor,

k S ik = Si, (1.2.19) is called the torsion trace or torsion vector.

1.2.4 Covariant diﬀerentiation of densities The diﬀerential of the determinant S of a square matrix S is given by

k i dS = s idS k, (1.2.20)

k i where s i is the minor corresponding to the component S k of the matrix. The components of the matrix S−1 inverse to S, i −1 j −1 i j i S j(S ) k = (S ) jS k = δk, (1.2.21) are related to the minors of S by sk (S−1)k = i . (1.2.22) i S The diﬀerential dS is therefore equal to

−1 k i k −1 i dS = S(S ) idS k = −SS id(S ) k, (1.2.23) which is equivalent to −1 k i k −1 i ∂lS = S(S ) i∂lS k = −SS i∂l(S ) k. (1.2.24) i ∂xi Taking S k = ∂x0k gives r r 0n m ∂x ∂x ∂x ∂ ∂x ∂l = . (1.2.25) ∂x0s ∂x0s ∂xm ∂xl ∂x0n A derivative of a scalar density s of weight w does not transform like a covariant vector density:

l j w l j w l j w−1 r 0 0 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂ s = ∂l s = ∂ls + w ∂l s i ∂x0i ∂x0k ∂x0i ∂x0k ∂x0i ∂x0k ∂x0s l j w l j w−1 r 0n m ∂x ∂x ∂x ∂x ∂x ∂x ∂ ∂x = ∂ls + w s ∂x0i ∂x0k ∂x0i ∂x0k ∂x0s ∂xm ∂xl ∂x0n l j w j w 0n 2 m ∂x ∂x ∂x ∂x ∂ x = ∂ls + w s. (1.2.26) ∂x0i ∂x0k ∂x0k ∂xm ∂x0n∂x0i Let us consider the expression s;i = s,i − wΓis, (1.2.27) where the quantity Γi transforms such that s;i is a vector density of weight w:

l j w l j w 0 ∂x ∂x ∂x ∂x s = s;l = (s,l − wΓls). (1.2.28) ;i ∂x0i ∂x0k ∂x0i ∂x0k

Also, (1.2.26) gives

l j w j w 0n 2 m j w 0 0 0 0 ∂x ∂x ∂x ∂x ∂ x ∂x 0 s = s − wΓ s = ∂ls + w s − w Γ s, (1.2.29) ;i ,i i ∂x0i ∂x0k ∂x0k ∂xm ∂x0n∂x0i ∂x0k i

12 so we obtain the transformation law for Γi:

∂xl ∂x0n ∂2xm Γ0 = Γ + , (1.2.30) i ∂x0i l ∂xm ∂x0n∂x0i

k k which is the same as the transformation law for Γk i (1.2.6). Therefore, the diﬀerence Γi − Γk i is some covariant vector Vi. If we assume that parallel transport of the product of a scalar density of any weight and a tensor obeys the chain rule: δ(sT ) = δsT + sδT, (1.2.31) so the covariant derivative of such product behaves like an ordinary derivative:

(sT );i = s;iT + sT;i, (1.2.32)

then the covariant derivative of a tensor density of weight w is equal to the sum of the corresponding ordinary derivative of this tensor, terms with the aﬃne connection that covariantize each index, and the term with Γi:

ij... ij... i nj... j in... T kl...;m = T kl...,m + Γn mT kl... + Γn mT kl... + ... n ij... n ij... ij... −Γk mT nl... − Γl mT kn... − · · · − wΓmT kl.... (1.2.33) the covariant derivative of the contravariant Levi-Civita density is

ijkl i njkl j inkl k ijnl l ijkn ijkl ;m = Γn m + Γn m + Γn m + Γn m − Γm . (1.2.34) In the summations over n only one term does not vanish for each term on the right-hand side of (1.2.34), thereby

The Levi-Civita symbol is a tensor density with constant components, thereby it does not change under a parallel transport, δ = 0. Therefore, we have

ijkl ;m = 0. (1.2.36) By means of (1.1.27), we also have εijkl;m = 0. (1.2.37)

Consequently, we obtain Vi = 0 and k Γi = Γk i. (1.2.38)

1.2.5 Antisymmetric covariant derivatives

Completely antisymmetrized ordinary derivatives of tensors, A[i,k], B[ik,l] and C[ikl,m], are tensors because of their antisymmetry. Completely antisymmetrized covariant derivatives of tensors are tensors because ∇i is a covariant operation, and are given by direct calculation using the deﬁnition of the covariant derivative:

l m A[i;k] = A[i,k] − S ikAl,B[ik;l] = B[ik,l] − 2S [ikBl]m. (1.2.39)

i ik Divergences of (completely antisymmetric if more than 1 index) contravariant densities, C ,i, B ,i ikl and A ,i, are densities because of the correspondence between tensors and dual densities. Covariant divergences of contravariant densities are densities, and are given by direct calculation:

i i i ik ik k il ik C ;i = C ,i + 2SiC , B ;i = B ,i − S ilB + 2SiB . (1.2.40)

13 1.2.6 Partial integration If the product of two quantities (tensors or densities) TU is a contravariant density Ck then Z Z Z Z Z Z TU;kdΩ = (TU);kdΩ − T;kUdΩ = (TU),kdΩ + 2 SkT UdΩ − T;kUdΩ. (1.2.41)

R The ﬁrst term on the right-hand side can be transformed into a hypersurface integral T UdSk. If the region of integration extends to inﬁnity and Ck corresponds to some physical quantity then the R boundary integral T UdSk vanishes, giving Z Z Z TU;kdΩ = 2 SkT UdΩ − T;kUdΩ. (1.2.42)

If T = δk, then U = Ci and i Z Z i i C ;idΩ = 2 SiC dΩ. (1.2.43)

Equation (1.2.43) can be written as Z ∗ i ∇i C dΩ = 0, (1.2.44) where ∗ ∇i = ∇i − 2Si (1.2.45) is the modiﬁed covariant derivative.

1.2.7 Geodesic frame of reference Let us consider a coordinate transformation 1 xk = x0k + ak x0lx0m, (1.2.46) 2 lm

k where a lm is symmetric in the indices l, m. Substituting this transformation to (1.2.5) and calcu- lating it at xk = x0k = 0 gives ∂xi = δi (1.2.47) ∂x0k k and 0 j j j Γi k = Γi k + a ik. (1.2.48) Putting j j a ik = −Γ(i k)|xl=0 (1.2.49) gives 0 j Γ(i k) = 0. (1.2.50) Therefore, there always exists a coordinate frame of reference in which the symmetric part of the connection vanishes locally (at one point). If the aﬃne connection is symmetric in the covariant j j indices, Γi k = Γk i (the torsion tensor vanishes), then (1.2.50) gives

0 j Γi k = 0. (1.2.51) The coordinate frame of reference in which the torsionless part of the connection vanishes (locally) is referred to as geodesic.

14 1.2.8 Affine geodesics and four-velocity Let us consider a point in spacetime P (xk) and a vector dxk at this point. We construct a point P 0(xk + dxk) and find the vector d0xk which is the parallel-transported dxk from P to P 0. Then construct a point P 00(xk +dxk +d0xk) and find the vector d00xk which is the parallel-transported d0xk from P 0 to P 00. The next point is P 000(xk + dxk + d0xk + d00xk) etc. Repeating this step constructs k dxk a polygonal line which in the limit dx → 0 becomes a curve such that the vector dλ (where λ is a parameter along the curve) tangent to it at any point, when parallely translated to another point on this curve, coincides with the tangent vector there. Such curve is referred to as an autoparallel curve or affine geodesic. Affine geodesics can be attributed with the concept of length, which, for the polygonal curve, is proportional to the number of parallel-transport steps described above. The condition that parallel transport of a tangent vector be a tangent vector is dxi dxi dxi dxk dxi d2xi + δ = − Γ i dxl = M + dλ , (1.2.52) dλ dλ dλ k l dλ dλ dλ2 where the proportionality factor M is some function of λ, or d2xi dxk dxl 1 − M dxi M + Γ i = , (1.2.53) dλ2 k l dλ dλ dλ dλ from which it follows that M must differ from 1 by the order of dλ. In the first term on the left-hand side of (1.2.53) we can therefore put M = 1, and we denote 1 − M by φ(λ)dλ, thereby d2xi dxk dxl dxi + Γ i = φ(λ) . (1.2.54) dλ2 k l dλ dλ dλ If we replace λ by a new variable s(λ) then (1.2.54) becomes d2xi dxk dxl φs0 − s00 dxi + Γ i = , (1.2.55) ds2 k l ds ds s02 ds where the prime denotes differentiation with respect to λ. Requiring φs0 − s00 = 0, which has a general solution s = R λ dλ exp[− R λ φ(x)dx], brings (1.2.55) to

d2xi dxk dxl + Γ i = 0, (1.2.56) ds2 k l ds ds where the scalar variable s is called the affine parameter. The autoparallel equation (1.2.56) is invariant under linear transformations s → as + b since the two lower limits of integration in the expression for s(λ) are arbitrary. We define the four-velocity vector: dxi ui = . (1.2.57) ds This definition brings (1.2.15) to DAk dAk = Ak ui, = Ak ui, (1.2.58) ds ;i ds ,i thereby Dui dui = + Γ i ukul = ui uj = 0. (1.2.59) ds ds k l ;j The relations (1.2.58) can be generalized to any tensor density T : DT dT = T ui, = T ui, (1.2.60) ds ;i ds ,i

15 i Only the symmetric part Γ(k l) of the connection enters the autoparallel equation (1.2.56) because dxk dxl of the symmetry of ds ds with respect to the indices k, l; aﬃne geodesics do not depend on torsion. At any point, a coordinate transformation to the geodesic frame (1.2.46) brings all the components i dui Γ(k l) to zero, thereby the autoparallel equation becomes ds = 0. The autoparallel equation is also invariant under a projective transformation

i i i Γk l → Γk l + δkAl, (1.2.61) where Ai is an arbitrary vector. Substituting this transformation to (1.2.59) gives

dui + Γ i ukul = −uiukA . (1.2.62) ds k l k If we replace s by a new variables ˜(s) then (1.2.62) becomes

dU i ukA s˜0 +s ˜00 dxi + Γ i U kU l = − k , (1.2.63) ds˜ k l s˜02 ds˜ where dxi U i = (1.2.64) ds˜ k 0 00 and the prime denotes diﬀerentiation with respect to s. Requiring u Aks˜ +s ˜ = 0, which has a R s R s k general solutions ˜ = − ds exp[ Aku (x)dx], brings (1.2.63) to

dU i + Γ i U kU l = 0. (1.2.65) ds˜ k l

1.2.9 Inﬁnitesimal coordinate transformations Let us consider a coordinate transformation

xi → x0i = xi + ξi, (1.2.66)

where ξi = δxi is an inﬁnitesimal vector (a variation of xi). For a tensor or density T deﬁne

δT = T 0(x0i) − T (xi), (1.2.67) ¯ 0 i i k δT = T (x ) − T (x ) = δT − ξ T,k. (1.2.68)

For a scalar we ﬁnd ¯ k δφ = 0, δφ = −ξ φ,k. (1.2.69) For a covariant vector

k ∂x k δAi = Ak − Ai ≈ −ξ Ak, (1.2.70) ∂x0i ,i δA¯ ≈ −ξk A − ξkA . (1.2.71) i ,i k i,k The variation (1.2.70) is not a tensor, but (1.2.71) is:

δA¯ = −ξk A − ξkA − 2Sj ξkA . (1.2.72) i ;i k i;k ik j

¯ i We refer to −δT as the Lie derivative of T along the vector ξ , LξT . For a contravariant vector

∂x0i δBi = Bk − Bi ≈ ξi Bk, (1.2.73) ∂xk ,k δB¯ i ≈ ξi Bk − ξkBi = ξi Bk − ξkBi + 2Si ξkBj. (1.2.74) ,k ,k ;k ;k jk

16 For a scalar density

i ∂x i δs = − 1 s ≈ −ξ s, (1.2.75) ∂x0i ,i ¯ i k i k i δs ≈ −ξ ,is − ξ s,k = −ξ ;is − ξ s;k + 2Siξ s. (1.2.76)

The chain rule for δ infers that, for a tensor density of weight w (which includes tensors as densities of weight 0), we have

ij... i mj... j im... m ij... m ij... δT kl... ≈ ξ ,mT kl... + ξ ,mT kl... + · · · − ξ ,kT ml... − ξ ,lT km... − ... m ij... −wξ ,mT kl..., (1.2.77) ¯ ij... i mj... j im... m ij... m ij... δT kl... ≈ ξ ;mT kl... + ξ ;mT kl... + · · · − ξ ;kT ml... − ξ ;lT km... − ... m ij... m ij... i m nj... j m in... −wξ ;mT kl... − ξ T kl...;m + 2S nmξ T kl... + 2S nmξ T kl... + ... n m ij... n m ij... m ij... −2S kmξ T nl... − 2S lmξ T kn... − · · · + 2wSmξ T kl.... (1.2.78) A Lie derivative of a tensor density of rank (k, l) and weight w is a tensor density of rank (k, l) and weight w. The formula for the covariant derivative of T can be written as

j ˆi T;k = T,k + Γi kCjT, (1.2.79)

where Cˆ is an operator acting on tensor densities:

î î i î k k i î i Cjφ = 0, CjAk = −δkAj, CjB = δj B , Cjs = −δjs, (1.2.80)

or generally

ˆm ij... i mj... j im... m ij... m ij... m ij... Cn T kl... = δnT kl... + δnT kl... + · · · − δk T nl... − δl T kn... − · · · − wδn T kl.... (1.2.81) Such deﬁned operator also enters the formula for δT :

ˆk i δT = Ci T ξ ,k. (1.2.82)

1.2.10 Killing vectors

A covariant vector ζi that satisﬁes ζ(i;k) = 0 (1.2.83) is referred to as a Killing vector. Along an aﬃne geodesic, D (uiζ ) = uk(uiζ ) = uiukζ + ζ ukui = 0. (1.2.84) ds i i ;k i;k i ;k

The first term in the sum in (1.2.84) vanishes because of the definition of ζi and the second term vanishes because of the affine geodesic equation. Therefore, to each Killing vector ζi there corresponds i a quantity u ζi which does not change along the affine geodesic:

i u ζi = const. (1.2.85)

References: [1, 2, 3].

17 1.3 Curvature 1.3.1 Curvature tensor We deﬁne the commutator [A, B] of two operators A and B as

[A, B] = AB − BA = −[B,A]. (1.3.1)

The commutator of covariant derivatives is thus

[∇i, ∇k] = 2∇[i∇k]. (1.3.2)

The commutator of covariant derivatives of a contravariant vector is a tensor:

i i i l i i l [∇j, ∇k]B = 2∇[j∇k]B = 2∂[j∇k]B − 2Γ[k j]∇lB + 2Γl [j∇k]B i m l i i l i l m = 2∂[j(Γ|m| k]B ) + 2S jk∇lB + 2Γl [j∂k]B + 2Γl [jΓ|m| k]B i i l m l i i m l i = 2(∂[jΓ|m| k] + Γl [jΓ|m| k])B + 2S jk∇lB = R mjkB + 2S jk∇lB , (1.3.3) where || embraces indices which are excluded from symmetrization or antisymmetrization. Therefore, i R mjk, deﬁned as i i i l i l i R mjk = ∂jΓm k − ∂kΓm j + Γm kΓl j − Γm jΓl k, (1.3.4) i is a tensor, referred to as the curvature tensor. The curvature tensor R mjk is antisymmetric in the indices j, k and has generally 96 independent components. The commutator of covariant derivatives of a covariant vector is m l [∇j, ∇k]Ai = −R ijkAm + 2S jk∇lAi, (1.3.5) and the commutator of covariant derivatives of a tensor is

in... i mn... n im... m in... m in... [∇j, ∇k]T lp... = R mjkT lp... + R mjkT lp... + · · · − R ljkT mp... − R pjkT lm... l in... − · · · + 2S jk∇lT lp.... (1.3.6)

A change in the connection, ˜ i i i Γj k = Γj k + T jk, (1.3.7) i where T jk is a tensor, results in the following change of the curvature tensor:

˜i ˜ i ˜ i ˜ j ˜ i ˜ j ˜ i i i j i j i R klm = Γk m,l − Γk l,m + Γk mΓj l − Γk lΓj m = Γk m,l − Γk l,m + Γk mΓj l − Γk lΓj m i i j i j i i j i j j i j i +T km,l − T kl,m + Γk mT jl − Γk lT jm + Γj lT km − Γj mT kl + T kmT jl − T klT jm i i i j i j i = R klm + T km;l − T kl;m + T kmT jl − T klT jm. (1.3.8)

i i For a projective transformation (1.2.61), T jk = δjAk, the curvature tensor changes according to

˜i i i R klm = R klm + δk(Am;l − Al;m). (1.3.9)

The variation of the curvature tensor is

i i i i j i j i j i j δR klm = (δΓk m),l − (δΓk l),m + δΓj lΓk m + Γj lδΓk m − δΓj mΓk l − Γj mδΓk l i i j j i j i i i j j i = (δΓk m);l − Γj lδΓk m + Γk lδΓj m + Γm lδΓk j − (δΓk l);m + Γj mδΓk l − Γk mδΓj l j i i j i j i j i j −Γl mδΓk j + δΓj lΓk m + Γj lδΓk m − δΓj mΓk l − Γj mδΓk l i i n i = (δΓk m);l − (δΓk l);m − 2S lmδΓk n. (1.3.10)

18 1.3.2 Integrability of connection The aﬃne connection is integrable if parallel transport of a vector from point P to point Q is independent of a path along which this vector is parallelly translated, or equivalently, parallel transport of a vector around a closed curve does not change this vector. For an integrable connection, we can uniquely translate parallelly a given vector hi at point P to all points in spacetime:

δhi = dhi, (1.3.11)

or i i j h ,k = −Γj kh . (1.3.12) Therefore, we have

i j i j i j i j m i j i j m i j (Γj kh ),l − (Γj lh ),k = Γj k,lh − Γj kΓm lh − Γj l,kh + Γj lΓm kh = R jlkh = 0, (1.3.13) so, because hi is arbitrary, i R klm = 0. (1.3.14) i Spacetime with a vanishing curvature tensor R klm = 0 is ﬂat. Let us consider 4 linearly independent i i vectors ha, where a is 1,2,3,4, and vectors inverse to ha:

X i i hahka = δk. (1.3.15) a

If the aﬃne connection is integrable then (1.3.12) becomes

i i l ha,k = −Γl kha. (1.3.16)

Multiplying (1.3.16) by hja gives

i i i Γj k = −hjaha,k = hja,kha. (1.3.17)

An integrable connection has thus 16 independent components. If the connection is also symmetric, i S jk = 0, then hja,k − hka,j = 0, (1.3.18) which is the condition for the independence of the coordinates

Z Q i ya = hiadx (1.3.19) P

of the path of integration PQ. Adopting ya as the new coordinates (with point P = (0, 0, 0, 0) in the center) gives i ∂ya ∂x i i = hia, = ha, (1.3.20) ∂x ∂ya so (1.3.17) becomes i 2 i i ∂x ∂ ya Γj k(x ) = k j . (1.3.21) ∂ya ∂x ∂x 0j The transformation law for the connection (1.2.5) gives (with ya corresponding to x )

i Γj k(ya) = 0. (1.3.22)

A torsionless integrable connection can be thus transformed to zero; one can always ﬁnd a system of coordinates which is geodesic everywhere. If a connection is symmetric but nonintegrable then a geodesic frame of reference can be constructed only at a given point (or along a given world line).

19 1.3.3 Parallel transport along closed curve Let us consider parallel transport of a covariant vector around an inﬁnitesimal closed curve. Such a transport changes this vector, according to Stokes’ theorem (1.1.37) by

I I 1 Z ∂(Γ i A ) ∂(Γ i A ) ∆A = δA = Γ i A dxl = k m i − k l i df lm. (1.3.23) k k k l i 2 ∂xl ∂xm

i l i Along the curve, we have dAk = Γk lAidx , which gives Ak,l = Γk lAi. The last relation is approximately valid, to terms of ﬁrst order in ∆f lm = R df lm, inside this curve:

1 Z ∂Γ i ∂Γ i 1 ∆A ≈ k m − k l A + (Γ i Γ n − Γ i Γ n )A df lm ≈ Ri A ∆f lm. (1.3.24) k 2 ∂xl ∂xm i k m i l k l i m n 2 klm i

The change of a contravariant vector in parallel transport around an inﬁnitesimal closed curve results k from ∆(AkB ) = 0: 1 ∆Bk ≈ − Rk Bi∆f lm, (1.3.25) 2 ilm and the corresponding change of a tensor results from the chain rule for parallel transport: 1 ∆T ik... ≈ − (Ri T jk... +Rk T ij... +· · ·−Rj T ik... −Rj T ik... −... )∆f lm. (1.3.26) np... 2 jlm np... jlm np... nlm jp... plm nj...

1.3.4 Bianchi identities Let us consider 1 ∇ ∇ ∇ Bi = ∇ (Ri Bm) + ∇ (Sm ∇ Bi) (1.3.27) j [k l] 2 j mkl j kl m and 1 1 1 ∇ ∇ ∇ Bi = − Rm ∇ Bi + Ri ∇ Bm + Sm ∇ ∇ Bi = − Rm ∇ Bi [j k] l 2 ljk m 2 mjk l jk m l 2 ljk m 1 + Ri ∇ Bm + Sm ∇ ∇ Bi + Sm Ri Bn + 2Sm Sn ∇ Bi. (1.3.28) 2 mjk l jk l m jk nml jk ml n Total antisymmetrization of the indices j, k, l in (1.3.27) and (1.3.28) gives 1 1 ∇ ∇ ∇ Bi = ∇ Ri Bm + Ri ∇ Bm + ∇ Sm ∇ Bi + Sm ∇ ∇ Bi (1.3.29) [j k l] 2 [j |m|kl] 2 m[kl] j] [j kl] m kl j] m and 1 1 ∇ ∇ ∇ Bi = − Rm ∇ Bi + Ri ∇ Bm + Sm ∇ ∇ Bi [j k l] 2 [ljk] m 2 m[jk l] [jk l] m m i n m n i +S [jkR |nm|l]B + 2S [jkS |m|l]∇nB , (1.3.30) so 1 1 ∇ Ri Bm + ∇ Sm ∇ Bi = − Rm ∇ Bi + Sm Ri Bn 2 [j |m|kl] [j kl] m 2 [ljk] m [jk |nm|l] m n i +2S [jkS |m|l]∇nB . (1.3.31)

Comparing terms in (1.3.31) with Bi gives the second Bianchi identity or the Bianchi identity:

i i m R n[jk;l] = 2R nm[jS kl], (1.3.32)

i while comparing terms with ∇kB gives the ﬁrst Bianchi identity or the Ricci cyclic identity:

m m m n R [jkl] = −2S [jk;l] + 4S n[jS kl]. (1.3.33)

20 Contracting (1.3.32) and (1.3.33) with respect to one contravariant and one covariant index gives

i i m R n[ik;l] = 2R nm[iS kl], (1.3.34) k k k n R [jkl] = −2S [jk;l] + 4S n[jS kl]. (1.3.35)

i For a symmetric connection, S jk = 0, the Bianchi identity and the cyclic identity reduce to

i R n[jk;l] = 0, (1.3.36) m R [jkl] = 0. (1.3.37) The cyclic identity (1.3.37) imposes 16 constraints on the curvature tensor, thereby the curvature tensor with a vanishing torsion has 80 independent components.

1.3.5 Ricci tensor Contraction of the curvature tensor with respect to the contravariant index and the second covariant index gives the Ricci tensor:

j j j l j l j Rik = R ijk = Γi k,j − Γi j,k + Γi kΓl j − Γi jΓl k. (1.3.38) Contraction of the curvature tensor with respect to the contravariant index and the third covariant index gives the Ricci tensor with the opposite sign because of the antisymmetry of the curvature tensor with respect to its last indices. Contraction of the curvature tensor with respect to the contravariant index and the ﬁrst covariant index gives the homothetic or segmental curvature tensor:

j j j Qik = R jik = Γj k,i − Γj i,k, (1.3.39) which is a curl. A change in the connection (1.3.7) results in the following changes of the Ricci tensor and segmental curvature tensor:

l l j l j l Rik → Rik + T ik;l − T il;k + T ikT jl − T ilT jk, (1.3.40) j j Qik → Qik + T jk,i − T ji,k. (1.3.41) For a projective transformation (1.2.61)

Rik → Rik + Ak;i − Ai;k, (1.3.42)

Qik → Qik + 4(Ak,i − Ai,k). (1.3.43) Therefore, the symmetric part of the Ricci tensor is invariant under projective transformations. The variation of the Ricci tensor is

l l j l δRik = (δΓi k);l − (δΓi l);k − 2S lkδΓi j, (1.3.44) while the variation of the segmental curvature tensor is

j j δQik = (δΓj k),i − (δΓj i),k. (1.3.45)

1.3.6 Geodesic deviation Let us consider a family of affine geodesics characterized by the affine parameter s, measured along each curve from its point of intersection with a given hypersurface, and distinguished by a scalar parameter t: xi = xi(s, t). We define ∂xi vi = , (1.3.46) ∂t which gives dui dvi vi uk − ui vk = vi uk − ui vk − 2Si ukvl = − − 2Si ukvl = −2Si ukvl, (1.3.47) ;k ;k ,k ,k kl dt ds kl kl

21 i ∂xi where u = ∂s is the four-velocity along each curve. We therefore have D2vi = (vi uj) uk = (ui vj) uk − 2(Si ukvl) uj ds2 ;j ;k ;j ;k kl ;j i j k i j k i k l j = u ;jkv u + u ;jv ;ku − 2(S klu v );ju i j k i l j k l i j k i j k i k l j = u ;kjv u − R ljku v u − 2S jku ;lv u + u ;jv ;ku − 2(S klu v );ju i j k i l j k l i j k i j k j k l = u ;kjv u − R ljku v u − 2S jku ;lv u + u ;j(u ;kv − 2S klu v ) i k l j i k j i j k l i k l j −2(S klu v );ju = (u ;ku );jv + R jklu u v − 2(S klu v );ju D = Ri ujukvl − 2 (Si ukvl), (1.3.48) jkl ds kl which can be written as D Dvi + 2Si ukvl = Ri ujukvl. (1.3.49) ds ds kl jkl This is the equation of geodesic deviation. If we replace affine geodesics by arbitrary curves then i k u ;ku 6= 0 and (1.3.49) becomes D Dvi + 2Si ukvl = Ri ujukvl + (ui uk) vj. (1.3.50) ds ds kl jkl ;k ;j The separation vector ξi = vidt (1.3.51) connects points on two infinitely close affine geodesics with t and t + dt for the same s. Multiplying (1.3.48) by dt gives another form of the equation of geodesic deviation,

D2ξi D = Ri ujukξl − 2 (Si ukξl). (1.3.52) ds2 jkl ds kl

References: [1, 2, 3, 4].

1.4 Metric 1.4.1 Metric tensor The affine parameter s is a measure of the length only along an affine geodesic. In order to extend the concept of length to all points in spacetime, we equip spacetime with an algebraic object gik, referred to as the covariant metric tensor and defined as

2 i k ds = gikdx dx . (1.4.1) The quantity ds in (1.4.1) is called the line element. The metric tensor is a symmetric tensor of rank (0,2): gik = gki. (1.4.2) The affine parameter s, whose differential is given by (1.4.1), is referred to as the interval. Because ds does not change under parallel transport along an affine geodesic from point P (xi) to point i i i i j Q(x + dx ), ds|Q = ds|P , and dx |Q is a parallel translation of dx |P , gik|Q = gik|P + gik,jdx is a parallel translation of gik|P : gik|Q = gik|P + δgik, (1.4.3) so j j Dgik = gik;jdx = dgik − δgik = gik,jdx − δgik = 0. (1.4.4) Therefore, the covariant derivative of the covariant metric tensor vanishes:

gik;j = 0. (1.4.5)

22 This relation is equivalent to l l gik,j − Γi jglk − Γk jgil = 0. (1.4.6) ik ki The symmetric contravariant metric tensor g = g is deﬁned as the tensor inverse to gik:

ik k gijg = δj . (1.4.7)

Since the contravariant metric tensor is a function of the covariant metric tensor only, its covariant derivative also vanishes: ik g ;j = 0. (1.4.8) The metric tensor allows to associate covariant and contravariant vectors:

i ik A = g Ak, (1.4.9) k Bi = gikB , (1.4.10)

because such association works for the covariant diﬀerentials of these vectors which are vectors:

i ik ik k k DA = D(g Ak) = g DAi,DBi = D(gikB ) = gikDB (1.4.11)

(raising and lowering of coordinate indices commutes with covariant diﬀerentiation with respect to ρ Γµ ν ). For covariant and contravariant indices of tensors and densities this association is

ij... j... gimT kl... = Tm kl..., (1.4.12) km ij... ijm... g T kl... = T l.... (1.4.13) The contravariant and covariant components of a two-dimensional vector are shown in Figure 1. The four-velocity vector (1.2.57) is normalized because of (1.4.1):

Figure 1: Contravariant and covariant components of a vector.

g dxidxk uiu = g uiuk = ik = 1. (1.4.14) i ik ds2 This vector thus has 3 independent components. Let us consider the determinant of the matrix composed from the components of the covariant metric tensor gik, g = |gik|. (1.4.15) The square root of the absolute value of this determinant, p|g|, is a scalar density of weight 1. We can use it to construct from the Levi-Civita symbols a quantity which behaves like a tensor with respect to continuous coordinate transformations: p eiklm = |g|εiklm, (1.4.16)

iklm 1 iklm in kp lq mr e = = g g g g enpqr. (1.4.17) p|g|

23 If we change the sign of 1 or 3 of the coordinates, then the components of eiklm do not change iklm because and εiklm have the same components in all coordinate systems, whereas some of the components of a tensor change sign. The components (1.4.16) and (1.4.17) are thus referred to as those of the completely antisymmetric unit pseudotensor. The relations (1.1.30) are also valid if we replace and ε by e. The diﬀerential and derivatives of the determinant of the metric tensor are given, following (1.2.23) and (1.2.24), by

ik ik dg = gg dgik = −ggikdg , (1.4.18) ik ik g,l = gg gik,l = −ggikg ,l. (1.4.19) The variation of the determinant of the metric tensor is thus

ik ik δg = gg δgik = −ggikδg . (1.4.20)

The covariant derivative of the determinant of the metric tensor vanishes:

g;j = 0. (1.4.21)

The relations (1.2.36) and (1.2.37) give thus

ijkl e ;m = 0, eijkl;m = 0. (1.4.22) A Lie derivative of the metric tensor is

ik (i;k) (ik) l Lξg = −2ξ − 4S lξ , (1.4.23)

;i ik where =;k g . The covariant derivative of the covariant metric tensor deﬁnes the nonmetricity tensor:

Njik = −gik;j. (1.4.24)

The commutator of covariant derivatives (1.3.6) of the metric tensor gives

(ij) ij m ij ij R kl = −N[k ;l] − S klNm = −N[k ,l], (1.4.25)

so the segmental curvature tensor (1.3.39) is

ij Qkl = −N[k ,l]gij. (1.4.26)

The nonmetricity tensor vanishes because of (1.4.5). Consequently, the curvature tensor is antisymmetric in its ﬁrst two indices: Rijkl = −Rjikl. (1.4.27) Therefore, the segmental curvature tensor also vanishes, and

jl Rijklg = Rik. (1.4.28)

Consequently, there is only one independent way to contract the curvature tensor, which gives the Ricci tensor up to a sign.

1.4.2 Christoﬀel symbols The condition (1.4.5) is referred to as metricity or metric compatibility of the aﬃne connection, and imposes 40 constraints on the connection:

l l l l l gik;j + gkj;i − gji;k = gik,j − Γi jglk − Γk jgil + gkj,i − Γk iglj − Γj igkl − gji,k + Γj kgli l l l l +Γi kgjl = gik,j + gkj,i − gji,k − 2Γ(i j)gkl − 2S kjgil − 2S kigjl = 0. (1.4.29)

24 Multiplying (1.4.29) by gkm gives

m m m Γ(i j) = {i j } + 2S(ij) , (1.4.30) where 1 { m} = gmk(g + g − g ) (1.4.31) i j 2 kj,i ki,j ij,k are the Christoﬀel symbols. Using (1.4.7), they can be written as 1 { m} = − (g gmk + g gmk − gmkg g gln ). (1.4.32) i j 2 kj ,i ki ,j il jn ,k The Christoﬀel symbols are symmetric in their covariant indices:

k k {i j} = {j i}. (1.4.33)

k k k Because Γi j = Γ(i j) + S ij, the metric-compatible aﬃne connection equals

k k k Γi j = {i j} + C ij, (1.4.34) where k k k C ij = 2S(ij) + S ij (1.4.35) is the contortion tensor, antisymmetric in its ﬁrst two indices:

Cijk = −Cjik. (1.4.36) The inverse relation between the torsion and contortion tensor is

i i S jk = C [jk]. (1.4.37) The Christoffel symbols are the torsionless part of the connection. The difference between two affine connections is a tensor, thereby the sum of a connection and a tensor of rank (1,2) is a connection. Therefore, the Christoffel symbols form a connection, referred to as the Levi-Civita connection. We define the covariant derivative with respect to the Levi-Civita k k {} connection analogously to (1.2.11), with Γi j replaced by {i j}, and denote it :i instead of ;i, or ∇i instead of ∇i. The covariant derivative with respect to the Levi-Civita connection of the metric tensor vanishes, as for that with respect to any connection:

l l gik:j = gik,j − {i j}glk − {k j}gil = 0. (1.4.38) This equation agrees with (1.4.31) and gives the relation between ordinary derivatives of the metric tensor and the Christoﬀel symbols:

l l gik,j = {i j}glk + {k j}gil. (1.4.39) Similarly, we have ik ik i lk k il g :j = g ,j + {l j}g + {l j}g = 0. (1.4.40) The variation of the Levi-Civita connection is, as for any connection, a tensor: 1 1 δ{ k } = gkl(δg ) + (δg ) − (δg ) + δgkl(g + g − g ) i j 2 lj ,i li ,j ij ,l 2 lj,i li,j ij,l 1 1 = gkl(δg ) + (δg ) − (δg ) + gkl({ m}δg + { m}δg + { m}δg + { m}δg 2 lj :i li :j ij :l 2 l i mj j i lm l j mi i j lm 1 −{ m}δg − { m}δg ) + δgkl{ m}g = gkl(δg ) + (δg ) − (δg ) i l mj j l im i j lm 2 lj :i li :j ij :l 1 +gkl{ m}δg + δgkl{ m}g = gkl(δg ) + (δg ) − (δg ) + { m}δδk i j lm i j lm 2 lj :i li :j ij :l i j m 1 = gkl(δg ) + (δg ) − (δg ) , (1.4.41) 2 lj :i li :j ij :l

25 where we used (1.4.7). The covariant derivative over s of a tensor density with respect to the Levi-Civita connection is, analogously to (1.2.60),

D{}T = T ui. (1.4.42) ds :i The following formulae are satisﬁed: 1 1 1 g, i { k } = gjkg = − g gjk = = (lnp|g|) , (1.4.43) k i 2 jk,i 2 jk ,i 2 g ,i

k ij 1 p ik { }g = − ( |g|g ),i, (1.4.44) i j p|g|

i 1 p i B = ( |g|B ),i, (1.4.45) :i p|g|

ik 1 p ik F = ( |g|F ),i, (1.4.46) :i p|g|

Ai:k − Ak:i = Ai,k − Ak,i, (1.4.47) I Z ip i p B |g|dSi = B :i |g|dΩ, (1.4.48)

ik ki k where F = −F . The Christoﬀel symbols satisfy all formulae that are satisﬁed by Γi j in which i S jk = 0. Because the Levi-Civita connection is a symmetric connection, it can be brought to zero by transforming the coordinates to a geodesic frame of reference. In a geodesic frame, the covariant {} derivative with respect to the Levi-Civita connection, ∇i , coincides with the ordinary derivative ∂i. Since the covariant derivatives of the Levi-Civita symbols are equal to zero, according to (1.2.36) and (1.2.37), their covariant derivatives with respect to the Levi-Civita connection vanish:

ijkl :m = 0, εijkl:m = 0. (1.4.49) The following covariant derivatives with respect to the Levi-Civita connection also vanish:

ijkl g:j = 0, e :m = 0, eijkl:m = 0. (1.4.50)

The Lie derivative of the metric tensor (1.4.23) along a vector ξi can be written as

ik (i:k) Lξg = −2ξ , Lξgik = 2ξ(i:k), (1.4.51)

:i ik where =:k g . A Killing vector (1.2.83) for the Levi-Civita connection satisﬁes

ζ(i:k) = 0. (1.4.52)

It thus becomes a generator of a transformation

x0i = xi + ζi, (1.4.53) where is an inﬁnitesimal scalar, which coincides with (1.2.66) for

ξi = ζi. (1.4.54)

Such transformations are isometries: they do not change the metric tensor. If the nonmetricity tensor does not vanish, the general formula for the aﬃne connection (1.4.34) is 1 Γ k = { k } + Ck − N k + N k . (1.4.55) i j i j ij 2 ij (i j)

26 1.4.3 Riemann tensor The commutator of covariant derivatives with respect to the Levi-Civita connection of a covariant vector is {} {} m [∇j , ∇k ]Ai = −P ijkAm, (1.4.56) analogously to (1.3.5) and without the torsion tensor of this connection that vanishes. The curvature tensor constructed from the Levi-Civita connection is referred to as the Riemannian curvature tensor or the Riemann tensor:

i i i i l i l P mjk = ∂j{m k} − ∂k{m j} + {l j}{m k} − {l k}{m j}. (1.4.57) Similarly, the commutators of covariant derivatives of a contravariant vector and of a tensor are i i i respectively given by (1.3.3) and (1.3.6), in which R jkl is replaced with P jkl and S jk = 0. The commutator of covariant derivatives of the metric tensor vanishes:

{} {} m m [∇j , ∇k ]glp = −P ljkgmp − P pjkglm = 0, (1.4.58) so the covariant Riemann tensor Pimjk is also antisymmetric in the indices i, m. Substituting (1.4.31) in (1.4.57) gives 1 P = (g + g − g − g ) + g ({ j }{ n } − { j }{ n }), (1.4.59) iklm 2 im,kl kl,im il,km km,il jn i m k l i l k m which explicitly shows the following symmetry and antisymmetry properties:

Piklm = −Pikml, (1.4.60)

Piklm = −Pkilm, (1.4.61)

Piklm = Plmik. (1.4.62)

Accordingly, the Riemannian Ricci tensor is symmetric:

j Pik = P ijk = Pki. (1.4.63) Substituting (1.4.34) in (1.3.7) and (1.3.8) gives the relation between the curvature and Riemann tensors: i i i i j i j i R klm = P klm + C km:l − C kl:m + C kmC jl − C klC jm. (1.4.64) Contracting (1.4.64) with respect to in the indices i, l gives

i i j i j i Rkm = Pkm + C km:i − C ki:m + C kmC ji − C kiC jm. (1.4.65) Consequently, the Ricci scalar or the curvature scalar,

ik R = Rikg , (1.4.66)

is given by ik l j l l m R = P − g (2C il:k + C ijC kl − C imC kl), (1.4.67) where P is the Riemannian curvature scalar or the Riemann scalar:

ik P = Pikg . (1.4.68)

The variation of the Riemann tensor is, analogously to (1.3.10),

i i i δP klm = (δ{k m}):l − (δ{k l}):m, (1.4.69) and the variation of the Riemannian Ricci tensor is

l l δPik = (δ{i k}):l − (δ{i l}):k. (1.4.70)

27 Contracting (1.3.34) and (1.3.35) with the metric tensor gives

i m i m i m Rnk;l − Rnl;k + R nkl;i = −2RnmS kl − 2R nmkS il + 2R nmlS ik (1.4.71) and the contracted cyclic identity:

k n Rjl − Rlj = −2Sj;l + 2Sl;j − 2S lj;k + 4SnS lj. (1.4.72) Further contraction of (1.4.71) with the metric tensor gives the contracted Bianchi identity: 1 Ri − R = 2R Smk − Rik Sm . (1.4.73) l;i 2 ;l km l ml ik The Bianchi identity (1.3.36) and the cyclic identity (1.3.37) for the Riemann tensor are

i P n[jk:l] = 0, (1.4.74) m P [jkl] = 0. (1.4.75) Contracting these equations with the metric tensor gives

i Pnk:l + P nkl:i − Pnl:k = 0, (1.4.76)

Pjl − Plj = 0, (1.4.77) in agreement with (1.4.63). Further contraction of (1.4.76) with the metric tensor gives the contracted Bianchi identity: i G k:i = 0, (1.4.78) for the symmetric Einstein tensor, deﬁned as 1 G = P − P g = G . (1.4.79) ik ik 2 ik ki This identity is a covariant conservation of the Einstein tensor.

1.4.4 Properties of Riemann tensor

In two dimensions there is only 1 independent component of the Riemann tensor, P1212. The Riemann scalar is 2P P = 1212 , (1.4.80) s

where s is the determinant of the two-dimensional metric tensor γik:

2 s = |γik| = γ11γ22 − γ12. (1.4.81) A surface near point x = 0, y = 0 is given by

x2 y2 z = + , (1.4.82) 2ρ1 2ρ2

where ρ1 and ρ2 are the radii of curvature. Substituting (1.4.82) to

2 2 2 2 i k dl = dx + dy + dz = γikdx dx (1.4.83)

gives γik(x, y), which then gives

P 1 = K = , (1.4.84) 2 x=y=0 ρ1ρ2 where K is the Gauß curvature. In three dimensions there are 3 independent pairs, 12, 23, and 31, thereby the Riemann tensor 3·2 has 6 independent components: 3 with identical pairs and 2 = 3 with diﬀerent pairs (the cyclic

28 identity does not reduce the number of independent components). The Ricci tensor has also 6 components, which are related to the components of the Riemann tensor by P P = P γ − P γ + P γ − P γ + (γ γ − γ γ ). (1.4.85) αβγδ αγ βδ αδ βγ βδ αγ βγ αδ 2 αδ βγ αγ βδ Choosing the Cartesian coordinates at a given point, deﬁned by the condition

gαβ = diag(1, 1, 1), (1.4.86)

and diagonalizing Pαβ, which is equivalent to 3 rotations, brings Pαβ to the canonical form with 6 − 3 = 3 independent components. Consequently, the Riemann tensor in three dimensions has 3 physically independent components. The Gauß curvature of a surface perpendicular to the x3 axis is given by P1212 K = 2 . (1.4.87) γ11γ22 − γ12 In four dimensions there are 6 independent pairs, 01, 02, 03, 12, 23, and 31, thereby there are 6·5 6 components with identical pairs and 2 = 15 with diﬀerent pairs. The cyclic identity reduces the number of independent components by 1, thereby the Riemann tensor in four dimensions has generally 20 independent components. Choosing the Cartesian coordinates at a given point and applying 6 rotations brings Pijkl to the canonical form with 20 − 6 = 14 physically independent components. The Weyl tensor is deﬁned as 1 1 W = P − (P g + P g − P g − P g ) + P (g g − g g ). (1.4.88) iklm iklm 2 il km km il im kl kl im 6 il km im kl This tensor has all the symmetry and antisymmetry properties of the Riemann tensor, and is also traceless (any contraction of the Weyl tensor vanishes).

1.4.5 Metric geodesics Let us consider two points in spacetime, P and Q. Among curves that connect these points, one curve has the minimal value of the interval s = R ds, and is referred to as a metric geodesic. The equation of a metric geodesic is given by the condition that R ds be an extremum with the endpoints of the curve ﬁxed: Z Z Z δdxig dxj 1 Z δg dxidxj Z δ ds = δ (g dxidxk)1/2 = ij + ij = g ujδdxi ik ds 2 ds ij 1 Z Z Z 1 Z + g δxkuiujds = d(u δxi) − du δxi + g δxkuiujds 2 ij,k i i 2 ij,k Z du 1 Z = − i δxids + g δxiujukds = 0, (1.4.89) ds 2 jk,i

R i i i where we omit the total diﬀerential term d(uiδx ) because δx = 0 at the endpoints. Since δx is arbitrary, we obtain

d 1 Z duj 1 Z (g uj) − g ujukds = g + ukg uj − g ujukds ds ij 2 jk,i ij ds ij,k 2 jk,i duj = g + { m}g ujuk = 0 (1.4.90) ij ds j k im or, after multiplying (1.4.90) by gil:

D{}ul dul = + { l }ujuk = uiul = 0. (1.4.91) ds ds j k :i

29 The metric geodesic equation (1.4.91) can be written as

d2xi dxk dxl + { i } = 0. (1.4.92) ds2 k l ds ds Using (1.4.34) and (1.4.35), the aﬃne geodesic equation (1.2.56) can be written as

d2xi dxk dxl dxk dxl + { i } + 2S i = 0. (1.4.93) ds2 k l ds ds kl ds ds If the torsion tensor is completely antisymmetric then the last term in (1.4.93) vanishes and the aﬃne geodesic equation coincides with the metric geodesic equation. The equation of geodesic deviation with respect to the Levi-Civita connection is, analogously to (1.3.49),

D{}2vi = P i ujukvl. (1.4.94) ds2 jkl

If ζi is a Killing vector of the Levi-Civita connection then along a metric geodesic,

D{} (uiζ ) = uk(uiζ ) = uiukζ + ζ ukui = 0. (1.4.95) ds i i :k i:k i :k The ﬁrst term in the sum in (1.4.95) vanishes because of (1.4.52) and the second term vanishes because of the metric geodesic equation. Therefore, to each Killing vector of the Levi-Civita connec- i tion there corresponds a quantity u ζi which does not change along the metric geodesic, analogously to (1.2.85): i i k u ζi = giku ζ = const. (1.4.96)

1.4.6 Galilean frame of reference and Minkowski tensor At a given point, the nondegenerate (g 6= 0) metric tensor can be brought to a diagonal (canonical) form gik = diag(±1, ±1, ±1, ±1). Physical systems are described by the metric tensor with g < 0. Without loss of generality, we assume that the canonical form of the metric tensor is

ik ik gik = ηik = diag(1, −1, −1, −1), g = η = diag(1, −1, −1, −1). (1.4.97)

A frame of reference in which gik has the canonical form is referred to as Galilean. The transformation (1.2.46) with (1.2.49) brings a symmetric affine connection, thus the Christoffel symbols, to zero at a given point without changing the components of the metric tensor because of (1.2.47). Therefore, a frame of reference can be locally both geodesic and Galilean. Such a frame is called inertial. In this frame, the first derivatives of the metric tensor vanish because of (1.4.39). The corresponding metric tensor (1.4.97) is referred to as the Minkowski tensor. The square of the line element for this metric is ds2 = c2dt2 − dx2 − dy2 − dz2. (1.4.98) In a locally inertial frame the coordinates xi, not only the differentials dxi, are components of a contravariant vector. i In the absence of torsion, spacetime with a vanishing Riemann tensor P klm = 0 is flat. In the new coordinates ya (1.3.19), (1.3.20) gives

i k ab ∂x ∂x ia kb ab g (y) = gik(x) = gik(x)h h = η . (1.4.99) ∂ya ∂yb Therefore, in a ﬂat spacetime without torsion one can always ﬁnd a system of coordinates which is Galilean everywhere.

30 1.4.7 Riemann normal coordinates If the frame of reference is locally geodesic and Galilean at a given point, taken as the origin of the coordinates, then the metric tensor at a point near the origin depends on the derivatives of the metric at the origin. In this frame, the Christoﬀel symbols at the origin vanish. We expand the metric tensor up to quadratic terms: 1 1 g (xk) = g (0) + g (0)xk + g (0)xkxl = η + g (0)xkxl, (1.4.100) ij ij ij,k 2 ij,kl ij 2 ij,kl where the metric tensor at the origin is equal to the Minkowski tensor and the ﬁrst derivatives of the metric tensor at the origin vanish because of (1.4.39). We choose the coordinates such that

xi = ais (1.4.101)

for every metric geodesic curve passing through the origin and parametrized with the interval s, where ai is a constant four-vector and s = 0 at the origin. Such coordinates are referred to as the Riemann normal coordinates. Accordingly, the derivatives of xi with respect to s are

dxi d2xi d3xi (0) = ai, (0) = (0) = 0. (1.4.102) ds ds2 ds3 Consequently, the metric geodesic equation (1.4.92) gives

i j k {j k}(0)a a = 0, (1.4.103)

therefore the condition for the geodesic frame of reference (1.2.50) is satisﬁed:

i {j k}(0) = 0. (1.4.104)

Diﬀerentiating (1.4.92) with respect to s gives

d3xi d{ i } dxj dxk d2xj dxk + j k + 2{ i } = 0. (1.4.105) ds3 ds ds ds j k ds2 ds At the origin, the relations (1.4.102) reduce this equation to

d{ i } dxj dxk dxl dxj dxk j k = { i } = { i } (0)alajak = 0. (1.4.106) ds ds ds j k ,l ds ds ds j k ,l Therefore, the Christoﬀel symbols satisfy

i {(j k},l)(0) = 0. (1.4.107) In the geodesic frame of reference, the Riemann tensor (1.4.57) reduces to

i i i P jkl = {j l},k − {j k},l. (1.4.108)

Consequently, using (1.4.107) gives

i i i i i i i P jkl + P kjl = {j l},k − {j k},l + {k l},j − {k j},l = −3{j k},l, (1.4.109) which gives 1 { i } (0) = − (P i + P i )(0). (1.4.110) j k ,l 3 jkl kjl Diﬀerentiating (1.4.39) with respect to the coordinates and using vanishing of the ﬁrst derivatives of the metric tensor at the origin gives

m m gij,kl = {k j },lgmi + {k i },lgmj. (1.4.111)

31 Substituting (1.4.110) into this equation gives 1 1 1 g (0) = − (P + P + P + P ) = − (P − P + P ) = − (P − P ) ij,kl 3 ikjl ijkl jkil jikl 3 ikjl kijl jikl 3 ikjl kjil 1 = − (P + P ). (1.4.112) 3 ikjl iljk Consequently, the covariant metric tensor (1.4.100) in the Riemann normal coordinates at a point near the origin, in quadratic approximation, is given by 1 1 g (xk) = η − (P + P )(0)xkxl = η − P (0)xkxl. (1.4.113) ij ij 6 ikjl iljk ij 3 ikjl The deviation of the metric tensor from the Minkowski tensor is proportional to the curvature. The corresponding contravariant metric tensor is given by 1 gij(xk) = ηij + P i j (0)xkxl. (1.4.114) 3 k l Similar calculations lead to the expansion of the covariant metric tensor in quartic approximation: 1 1 1 2 g (xk) = η − P (0)xkxl − P (0)xkxlxm − P − P pP (0)xkxlxmxn. ij ij 3 ikjl 6 ikjl:m 20 ikjl:mn 45 ikl jmnp (1.4.115)

1.4.8 Intervals, proper time, and distances The form of the Minkowski tensor distinguishes the coordinate x0 from the rest of the coordinates xα, where the index α can be 1,2,3. The temporal coordinate x0 can be written as x0 = ct, where t is referred to as time and c is called the speed of propagation of interaction. The coordinates xα are spatial and span space. The set of 4 coordinates xi describe an event and span spacetime. The curve xi(λ), where λ is a parameter, is referred to as a world line of a given point. The quantitites dxα vα = (1.4.116) dt are the components of a three-dimensional vector, the velocity of this point. An inﬁnitesimal interval ds is timelike if ds2 > 0, spacelike if ds2 < 0, and null if ds2 = 0. In a Galilean frame of reference, the spatial coordinates are Cartesian (1.4.86). In this frame, the square of the line element (interval) between two inﬁnitesimally separated points (events) is

2 i k 2 2 X α α ds = ηikdx dx = c dt − dx dx , (1.4.117) α where dxi are infinitesimal coordinate differences between the two points. The square of the interval between two finitely separated points is

2 i k 2 2 X α α ∆s = ηik∆x ∆x = c ∆t − ∆x ∆x , (1.4.118) α where ∆xi are finite coordinate differences between the two points. If ∆s is timelike, one can always find a frame of reference in which the two events occur at the same place, ∆xα = 0. A frame of reference in which dxα = 0, thereby vα = 0, describes a point at rest and is referred to as the rest frame or the comoving frame. In this frame t = τ,

ds2 = c2dτ 2, (1.4.119) where τ is the proper time. If dxα 6= 0, thereby vα 6= 0, along a world line then the point moves or is in motion. The proper time for a moving point is equal to the time measured by a clock moving with this point. If ∆s is spacelike, one can always ﬁnd a frame of reference in which the two events

32 occur at the same time (are synchronous), ∆x0 = 0. If ds = 0 along a world line, this world line P α α 1/2 describes the propagation of a signal (interaction), with ( α v v ) = c. Equations (1.4.117) and (1.4.119) give 1 X dτ 2 = dt2 − dxαdxα, (1.4.120) c2 α so the proper time τ goes more slowly than the coordinate time t. If ∆s is timelike, the two events occur at diﬀerent times: t1 6= t2. If t2 > t1 then t2 is in the future with respect to t1 and t1 is in the past with respect to t2. The time of a measurement t0 is called the present time. All events for which t < t0 form the absolute past relative to the event O at the present (events in this region occur before O in all systems of reference). All events for which t > t0 form the absolute future relative to the event O at the present (events in this region occur after O in all systems of reference). Such a division into the absolute past and the absolute future with respect to O is possible only for events for which their intervals with respect to O are timelike, as shown in Figure 2. For O = (0, 0, 0, 0), these events (ct, x, y, z) lie within a cone (ct)2 − x2 − y2 − z2 = 0 which is called the null cone or light cone. All events for which their intervals with respect to O are spacelike are absolutely remote relative to O. The principle of causality states that any event O can be aﬀected only by events in the absolute past relative to O.

Figure 2: Light cone.

In the rest frame dxα = 0 gives uα = 0. At each point in space, the condition dxα = 0 gives the relation between the proper time and the coordinate time: 1√ dτ = g dx0, (1.4.121) c 00 which requires g00 ≥ 0. (1.4.122) The relation (1.4.14) gives 0 −1/2 u = (g00) . (1.4.123) The distance between two inﬁnitesimally separated points cannot be obtained by imposing dx0 because x0 transforms diﬀerently at these points. Instead, we consider a signal that leaves point α α 0 0 α 0 0 0 B(x + dx ) at x + dx−, reaching point A(x ) at x and coming back to point B at x + dx+, as shown in Figure 3. Accordingly, we have

2 0 2 0 α α β ds = g00(dx ) + 2g0αdx dx + gαβdx dx = 0 (1.4.124)

gives q 0 1 α α β dx± = (−g0αdx ± (g0αg0β − g00gαβ)dx dx ). (1.4.125) g00

33 The difference in the time coordinate between emitting and receiving the signal at point B is equal 0 0 √ to the difference between dx+ and dx− times g00/c, and the distance dl between points A and B is equal to this difference times c/2: 2 α β dl = γαβdx dx , (1.4.126) where g0αg0β γαβ = −gαβ + (1.4.127) g00 is the symmetric spatial metric tensor of spacetime, that is, the metric tensor of space. The event at point A at x0 is synchronized with the event at point B at the arithmetic mean of the time coordinates of emitting and receiving the signal: 1 x0 + (dx0 + dx0 ) = x0 + g dxα, (1.4.128) 2 − + α where g0α gα = − . (1.4.129) g00 Therefore, we have 0 α δx = gαδx , (1.4.130) 0 which is equivalent to δx0 = 0, is the difference in x between two synchronized infinitesimally separated points.

Figure 3: Distance.

In terms of (1.4.126) and (1.4.129), the square of the line element is equal to 2 0 α 2 2 ds = g00(dx − gαdx ) − dl , (1.4.131) The three-dimensional velocity (1.4.116), dxα vα = , (1.4.132) dτ is defined in terms of the synchronized proper time (corresponding to the difference in x0 between two synchronized infinitesimally separated points (1.4.130)): 1√ 1√ dτ = g (dx0 − δx0) = g (dx0 − g dxα). (1.4.133) c 00 c 00 α Therefore, the metric (1.4.131) becomes v2 ds2 = g (dx0 − g dxα)2 1 − , (1.4.134) 00 α c2 where v is the speed, α β 1/2 v = (γαβv v ) . (1.4.135) Using (1.4.131) in the definition of the four-velocity (1.2.57) gives α α v 0 1 α u = , u = + gαu , (1.4.136) p 2 2 √ p 2 2 c 1 − v /c g00 1 − v /c from which we also find √ 0 α g00 u0 = g00u + g0αu = . (1.4.137) p1 − v2/c2

34 1.4.9 Spatial vectors The spatial components of a contravariant four-vector Ai form a three-dimensional, spatial vector A: Ai = (A0,Aα) = (A0, A). (1.4.138) The contravariant four-vector index α is also the contravariant spatial-vector index. The covariant components of a spatial vector are related to the contravariant components by the spatial metric tensor (1.4.127) which raises and lowers indices of spatial vectors analogously to the metric tensor acting on four-vectors:

β Aα = γαβA , (1.4.139) α αβ B = γ Bβ, (1.4.140)

αβ where γ is the inverse of γαβ: αδ α γ γβδ = δβ . (1.4.141) A linear combination aA + bB of two spatial vectors A and B, where a and b are scalars, is a spatial vector C whose components are

α α α C = aA + bB ,Cα = aAα + bBα. (1.4.142)

The following formulae are satisﬁed:

γαβ = −gαβ, (1.4.143)

g = −g00s, (1.4.144) gα = −g0α, (1.4.145)

00 1 α g = − gαg , (1.4.146) g00 where s = detγαβ. (1.4.147) For example, contracting (1.4.127) with (1.4.143) gives

αδ αδ αδ g0β αi α0 αi α0 g0β γ γβδ = g gβδ − g g0δ = g gβi − g gβ0 − (g g0i − g g00) g00 g00 α α g0β α = δβ − δ0 = δβ , (1.4.148) g00 in accordance with (1.4.141). The components gα form a spatial vector g. The dot product or scalar product of two spatial vectors is

α β A · B = γαβA B . (1.4.149)

The square of a spatial vector A is A2 = A · A (1.4.150) and its norm, length, or magnitude is √ A = |A| = A2. (1.4.151)

The angle between two spatial vectors θ is deﬁned through

A · B = AB cosθ. (1.4.152)

In three-dimensional space, the completely antisymmetric permutation symbols are deﬁned as

123 0123 αβγ [αβγ] = = 1, = , ε123 = −ε0123 = 1, εαβγ = ε[αβγ]. (1.4.153)

35 The spatial analogues of (1.4.16) and (1.4.17) are √ 1 e = sε , eαβγ = √ αβγ . (1.4.154) αβγ αβγ s The cross product or vector product of two spatial vectors A and B, C = A × B is deﬁned as the spatial vector density, dual to the antisymmetric tensor

Cαβ = AαBβ − AβBα, (1.4.155) thereby giving 1 1 Cα = eαβγ C = eαβγ A B ,C = e Cβγ = e AβBγ , (1.4.156) 2 βγ β γ α 2 αβγ αβγ γ αβ αβγ Cαβ = eαβγ C ,C = e Cγ . (1.4.157) The permutation symbols satisfy

εαβγ εαδζ = δβδδγζ − δβζ δγδ, (1.4.158)

εαβγ εαβδ = 2δγδ, (1.4.159)

εαβγ εαβγ = 6, (1.4.160)

where δαβ is the Cartesian metric tensor,

αβ δαβ = δ = diag(1, 1, 1). (1.4.161)

The spatial covariant derivative ∇α acts on spatial vectors analogously to the metric covariant derivative acting on four-vectors:

β β β γ ∇αA = ∂αA + {γ α}γ A , (1.4.162) γ ∇αAβ = ∂αAβ − {β α}γ Aγ , (1.4.163)

δ where {α β}γ are the three-dimensional, spatial Christoﬀel symbols: 1 { δ } = γδγ (γ + γ − γ ). (1.4.164) α β γ 2 γα,β γβ,α αβ,γ The gradient operator is given by

α α αβ (grad) = (∇) = γ ∇β. (1.4.165)

The spatial components of a covariant-vector operator ∂i acting on a scalar φ form the gradient of φ: ∂φ ∂φ ∂φ ∂ φ = , = , ∇φ . (1.4.166) i c∂t ∂xα c∂t The divergence of a spatial vector A is, analogously to (1.4.45), 1 √ divA = ∇ · A = √ ∂ ( sAα). (1.4.167) s α The curl of a spatial vector A is deﬁned as the spatial vector density, dual to the antisymmetric tensor ∂αAβ − ∂βAα: 1 (curlA)α = (∇ × A)α = eαβγ (∂ A − ∂ A ) = eαβγ ∂ A . (1.4.168) 2 β γ γ β β γ The Laplace-Beltrami operator or Laplacian is the divergence of the gradient, 1 √ 4 = ∇2 = ∇ · ∇ = √ ∂ ( sγαβ∂ ). (1.4.169) s α β

36 The d’Alembert operator or d’Alembertian is deﬁned as 1 ∂2 = − 4. (1.4.170) c2 ∂t2 The time component of the dual hypersurface element (1.1.35) is equal to the spatial volume element dV : dS0 = dV. (1.4.171) The spatial analogue of the Gauß-Stokes theorem (1.1.39) is Gauß’ theorem: ∂ df ↔ dV , (1.4.172) α ∂xα where ? dfα = df0α (1.4.173) is the spatial component of the dual surface element (1.1.33), perpendicular to the xα axis. In a locally Galilean frame of reference, the covariant and contravariant components of a spatial vector are identical because γαβ = δαβ. (1.4.174) In this frame, we refer to the Cartesian coordinates x1, x2, x3 as x, y, z. These coordinates form the three-dimensional radius vector x. The following formulae are satisﬁed:

A × B = −B × A, (1.4.175) A · (B × C) = B · (C × A) = C · (A × B), (1.4.176) A × (B × C) = B(A · C) − C(A · B), (1.4.177) (A · B)2 + (A × B)2 = A2B2, (1.4.178) curl gradφ = 0, (1.4.179) div curl A = 0, (1.4.180) grad(φψ) = gradφ ψ + φ gradψ, (1.4.181) grad(A · B) = (A · ∇)B + (B · ∇)A + A × curl B +B × curl A, (1.4.182) div(φA) = gradφ · A + φ div A, (1.4.183) curl(φA) = gradφ × A + φ curl A, (1.4.184) div(A × B) = B · curl A − A · curl B, (1.4.185) curl(A × B) = (B · ∇)A − (A · ∇)B + A div B − B div A, (1.4.186) curl curl A = grad div A − 4A, (1.4.187) where α (A · ∇)B = A ∂αB. (1.4.188) The vector product of two spatial vectors A and B satisﬁes

A × B = AB sinθ n, (1.4.189) where n is a unit vector perpendicular to both A and B, in the direction given by the right-handed corkscrew rule.

1.4.10 Embedded hypersurfaces A surface embedded in a three-dimensional space consists of points whose radius vectors are vector functions of two parameters ξα, where the index α can be 1 or 2: x = x(ξ1, ξ2). A vector ∂x ∂ x = (1.4.190) α ∂ξα

37 is tangent to the surface. We deﬁne the induced or intrinsic metric tensor on the surface as

γαβ = ∂αx · ∂βx. (1.4.191)

The length element dl on the surface is given by the ﬁrst fundamental form: ∂x ∂x dl2 = dx · dx = · dξαdξβ = γ dξαdξβ, (1.4.192) ∂ξα ∂ξβ αβ and the area element is given by p 1 2 dS = detγαβdξ dξ . (1.4.193) The inverse intrinsic metric tensor γαβ is deﬁned according to

αδ α γ γβδ = δβ . (1.4.194) We deﬁne the unit normal vector to a surface as ∂ x × ∂ x n = 1 2 , n · n = 1. (1.4.195) |∂1x × ∂2x| This vector is perpendicular to a tangent vector:

∂αx · n = 0. (1.4.196)

If the surface is curved, then the normal vectors at two close points on the surface are not parallel. The change of the normal vector is given by the extrinsic curvature tensor:

Kαβ = ∂α∂βx · n. (1.4.197)

The extrinsic curvature is symmetric, Kαβ = Kβα. (1.4.198) Diﬀerentiating the relation (1.4.196) with respect to ξβ and using (1.4.197) gives

Kαβ = −∂αx · ∂βn. (1.4.199)

α β α The quantity Kαβdξ dξ is the second fundamental form. The intrinsic Christoﬀel symbols {β γ }, symmetric in the lower indices, are constructed from the intrinsic metric tensor analogously to the spatial Christoﬀel symbols (1.4.164) constructed from the spatial metric tensor. They are used to construct the covariant derivative ∇i acting on the vectors tangent to the surface, analogously to (1.4.162) and (1.4.163). The covariant derivatives acting on x and n are equal to the partial derivatives:

∇αx = ∂αx, ∇αn = ∂αn. (1.4.200)

The second derivatives of x, which are the ﬁrst derivatives of the tangent vectors, satisfy the Gauß equation: γ ∂α∂βx = {α β}∂γ x + Kαβn, (1.4.201) which can be written in a covariant form:

∇α∇βx = Kαβn. (1.4.202)

Multiplying this equation by n gives (1.4.197). The ﬁrst derivatives of the normal vector satisfy the Weingarten equation: β βγ ∂αn = −Kα ∂βx = −Kαγ γ ∂βx, (1.4.203) which can be written in a covariant form:

β ∇αn = −Kα ∇βx. (1.4.204)

38 Multiplying this equation by ∂γ x and using (1.4.191) gives (1.4.199). The intrinsic metric tensor and the extrinsic curvature can also be written in a covariant form:

γαβ = ∇αx · ∇βx, (1.4.205)

Kαβ = ∇α∇βx · n. (1.4.206) Using the Gauß equation, the relation

∂α∂β∂γ x = ∂β∂α∂γ x (1.4.207) can be written as δ δ ∂α({β γ }∂δx + Kβγ n) = ∂β({α γ }∂δx + Kαγ n). (1.4.208) Eﬀecting the diﬀerentiation and using again the Gauß equation gives

δ δ δ ∂δx∂α{β γ } + {β γ }{α δ}∂x + {β γ }Kαδn + ∂αKβγ n + Kβγ ∂αn δ δ δ = ∂δx∂β{α γ } + {α γ }{β δ}∂x + {α γ }Kβδn + ∂βKαγ n + Kαγ ∂jn. (1.4.209)

Multiplying this equation by ∂ζ x and using (1.4.191), (1.4.196), and (1.4.199) gives

δ δ δ δ γδζ ∂α{β γ } + γζ {β γ }{α δ} − Kαζ Kβγ − γδζ ∂β{α γ } − γζ {α γ }{β δ} + Kβζ Kαγ = 0, (1.4.210) which is equivalent to the Gauß equation:

r γαβ = K αKγβ − K βKγα, (1.4.211)

where r γαβ is the intrinsic curvature tensor constructed from the intrinsic Christoﬀel symbols analogously to the Riemann tensor (1.4.57) constructed from the Levi-Civita connection. Multiplying (1.4.209) by n and using (1.4.196) and ∂αn · n = 0 gives the Codazzi-Mainardi-Peterson equation:

δ δ {β γ }Kαδ + ∂αKβγ = {α γ }Kβδ + ∂βKαγ , (1.4.212) which can be written in a covariant form:

∇αKβγ = ∇βKαγ . (1.4.213) The Gauß curvature is defined as detK K K − K K K = αβ = 11 22 12 21 . (1.4.214) detγαβ γ11γ22 − γ12γ21 Using the Gauß equation, it leads to the Gauß theorem: r K = 1212 , (1.4.215) detγαβ which is consistent with (1.4.80) and (1.4.84). A curve on a surface consists of points whose radius vectors depend on a parameter t: x = x(ξ1(t), ξ2(t)). Such a curve is geodesic if it satisfies the metric geodesic equation analogous to (1.4.92): d2ξα dξβ dξγ + { α } = 0. (1.4.216) dt2 β γ dt dt A geodesic curve also satisfies d2x d dx dx ∼ n, · = 0. (1.4.217) dt2 dt dt dt A hypersurface embedded in a four-dimensional spacetime consists of points whose coordinates are functions of three parameters ξα, where the index α can be 1, 2, or 3: xi = xi(ξ1, ξ2, ξ3). Equivalently, these coordinates satisfy an equation of constraint:

f(xi) = 0, (1.4.218)

39 where f is a function of the coordinates. The normal vector to this hypersurface is given by ∂f n = . (1.4.219) i ∂xi All inﬁnitesimal displacements dxi along such a hypersurface satisfy, according to (1.4.219),

i df = nidx = 0. (1.4.220)

The normal vector (1.4.219) is orthogonal to the hypersurface:

ijkl ninj:k = 0, (1.4.221) where ijkl is the completely antisymmetric permutation symbol. This condition is equivalent to 1 n n = (n n + n n + n n − n n − n n − n n ) = 0. (1.4.222) [i j:k] 6 i j:k j k:i k i:j k j:i i k:j j i:k If the normal vector to a hypersurface is timelike, then the hypersurface is spacelike. Such a normal vector can be normalized: i n ni = 1, (1.4.223) which gives i n ni:k = 0. (1.4.224) In this case, the four-velocity of a point in spacetime can be taken as the normal vector:

ni = ui. (1.4.225)

If the four-velocity has only the time component, then the hypersurface is a hypersurface of constant time and represents a volume in space, in which the point exists at this time. A division of spacetime into such hypersurfaces is referred to as a foliation of spacetime. We consider a spacelike hypersurface. We deﬁne the projection tensor onto the hypersurface:

i i i h j = δj − n nj, (1.4.226) which is orthogonal to ni: i j h jn = 0. (1.4.227) The projection tensor satisﬁes i j i h jh k = h k. (1.4.228) The indices in the projection tensor can be raised or lowered by the metric tensor:

j ik i jk ik i k ki hik = h kgij = gik − nink = hki, h = h jg = g − n n = h . (1.4.229)

ik The tensors hik and h are symmetric and not inverse to one another. The projection ⊥ of a tensor T onto a hypersurface is deﬁned as the contraction of the tensor T with the projection tensor through all indices. For example, the projections of vectors are

i i k k ⊥ V = h kV , ⊥ Vi = h iVk. (1.4.230) These projections are tangent vectors to the hypersurface. The projection of the metric tensor gives

k l ij i j kl ij ⊥ gij = h ih jgkl = hij, ⊥ g = h kh lg = h . (1.4.231) Consequently, the relation i j i j hij ⊥ V ⊥ V = gij ⊥ V ⊥ V (1.4.232) shows that the tensor hij is the intrinsic metric tensor γij on the hypersurface, analogously to (1.4.191): γij = hij. (1.4.233)

40 The inverse intrinsic metric tensor γij is defined as in (1.4.194). The projection of the normal vector vanishes: ⊥ ni = 0. (1.4.234) If the normal vector to the hypersurface is timelike, then the projections of tensors have only spatial i i components. Using the tensor n nj instead of h j projects a tensor onto the direction of the normal vector. The projection of the covariant derivative (with respect to a torsionless affine connection) of a vector defines the intrinsic covariant derivative of a vector on the hypersurface:

l l i l j DkV =⊥ ∇kV = h kh j∇iV . (1.4.235)

The intrinsic covariant derivative of the intrinsic metric tensor vanishes:

Dkγij =⊥ ∇kγij =⊥ ∇k(gij − ninj) = − ⊥ (ni∇knj + nj∇kni) = 0, (1.4.236)

which is a consequence of the metric compatibility of the affine connection (1.4.5). Accordingly, the intrinsic covariant derivative is constructed from the intrinsic Christoffel symbols, which are constructed from the intrinsic metric tensor. If ∇k is related to the Levi-Civita connection of the metric gij, then Dk is related to the Levi-Civita connection of the intrinsic metric γij. If the parallel transport of the normal vector to a hypersurface along a vector W i =⊥ V i on the hypersurface does not vanish, i j W ∇in 6= 0, (1.4.237) then the hypersurface is curved. Such a hypersurface has a nonzero extrinsic curvature tensor, defined as k l Kij = − ⊥ ∇inj = −h ih j∇knl. (1.4.238) Using (1.4.224) and (1.4.226), the extrinsic curvature is equal to

k k l l k Kij = −(δi − n ni)(δj − n nj)nl:k = −nj:i + nin nj:k. (1.4.239)

The extrinsic curvature is a tensor with only spatial components:

j Kijn = 0. (1.4.240)

Antisymmetrizing the indices in the extrinsic curvature and using (1.4.239) gives

k k Kij − Kji = ni:j − nj:i + n ninj:k − n njni:k. (1.4.241)

The term on the right-hand side is equal to the term in (1.4.222) contracted with nk, which vanishes. Consequently, the extrinsic curvature is symmetric, as in (1.4.198). This symmetry also results from (1.4.219): Kij = − ⊥ ∇i∂jf = − ⊥ ∇j∂if = Kji. (1.4.242) For the Levi-Civita connection, (1.4.51) gives 1 K = − ⊥ n = − ⊥ n = − ⊥ L g , (1.4.243) ij j:i (i:j) 2 n ij

i where Ln is the Lie derivative of the metric tensor along the vector n . If the normal vector is the four-velocity of a point in spacetime, then (1.4.239) gives

Du K = −u + u j . (1.4.244) ij j;i i ds The contraction of the extrinsic curvature tensor gives the extrinsic curvature scalar:

ij K = Kijγ . (1.4.245)

41 For a spacelike hypersurface, the spatial coordinates on this hypersurface can be taken as the parameters ξα. Diﬀerentiating the equation of constraint for a hypersurface f(xi(ξα)) = 0 with respect to ξα gives ∂f ∂f ∂xi ∂xi = = n = 0. (1.4.246) ∂ξα ∂xi ∂ξα i ∂ξα Diﬀerentiating covariantly this equation with respect to ξβ gives

∇n ∂xi ∇2xi ∇ ∂f ∂xi ∇2f ∇n i +n = +n ∇ ∇ xi = +n ∇ ∇ xi = β +n ∇ ∇ xi = 0, ∂ξβ ∂ξα i ∂ξβ∂ξα ∂xi ∂ξβ ∂ξα i β α ∂ξα∂ξβ i β α ∂ξα i β α (1.4.247) where the covariant derivatives ∇α are constructed from the metric tensor γαβ and the corresponding Levi-Civita connection. Accordingly, using (1.4.238), we obtain

i Kαβ = −∇αnβ = ni∇β∇αx , (1.4.248) which is consistent with the extrinsic curvature tensor for a surface (1.4.206). The intrinsic covariant derivative of a vector W i =⊥ V i on a hypersurface is

m m n n DjWk =⊥ ∇jWk = (δj − n nj)(δk − n nk)∇mWn m n m n = ∇jWk − n nj∇mWk − n nk∇jWn + n njn nk∇mWn m n m n = ∇jWk − n nj∇mWk + nkW ∇jnn + n njn nk∇mWn, (1.4.249)

i i i where we used niW = 0, which gives n ∇jWi = −W ∇jni. Consequently, the second derivative is

DiDjWk =⊥ (∇iDjWk) =⊥ (∇i ⊥ ∇jWk) m n m n =⊥ ∇i∇jWk+ ⊥ ∇i(−n nj∇mWk + nkW ∇jnn + n njn nk∇mWn) n n =⊥ ∇i∇jWk+ ⊥ ∇inkW ∇jnn =⊥ ∇i∇jWk + KikKjnW , (1.4.250) where we used (1.4.234) and (1.4.238). The commutator of intrinsic covariant derivatives gives the intrinsic curvature tensor: l [Di,Dj]Wk = −r kijWl, (1.4.251) whereas the commutator of covariant derivatives gives the Riemann tensor, according to (1.4.56). Therefore, antisymmetrizing the indices i, j in (1.4.250) gives

l l l l r kijWl =⊥ P kijWl − KikKjlW + KjkKilW , (1.4.252) which leads to ⊥ Plkij = rlkij + KikKjl − KjkKil. (1.4.253)

This equation is consistent with (1.4.211) for Plkij = 0, satisﬁed for a curved surface in a ﬂat space. i The projection of n Pijkl is given by

i i i ⊥ (n Pijkl) =⊥ (∇l∇knj − ∇k∇lnj) =⊥ (∇l(Kkj − nkn nj:i) − ∇k(Klj − nln nj:i)) i =⊥ (∇lKkj − ∇kKlj + (∇knl − ∇lnk)n nj:i) = DlKkj − DkKlj, (1.4.254) where we used (1.4.219) and (1.4.234). This equation is consistent with (1.4.213) for Plkij = 0, satisfied for a curved surface in a flat space. Equations (1.4.253) and (1.4.254) are referred to as the Gauß-Codazzi equations. If the normal vector to a hypersurface is spacelike, then the hypersurface is timelike. An example of such a hypersurface is a hypersurface on which a given spatial coordinate is constant. The normal vector can be normalized: i n ni = −1. (1.4.255) The projection tensor onto a timelike hypersurface differs from (1.4.226) by a sign:

i i i h j = δj + n nj, (1.4.256)

42 whereas all other deﬁnitions are the same as for spacelike hypersurfaces. If a hypersurface is spacelike or timelike, and forms a boundary between two submanifolds in spacetime, then the intrinsic metric tensor γij and the extrinsic curvature tensor Kij are continuous across the hypersurface. Consequently, the ﬁrst and second fundamental forms are continuous across the hypersurface. These two covariant conditions are referred to as the Darmois junction conditions. Equivalently, the metric tensor gij and its derivatives gij,k are continuous across the hypersurface. These two conditions are referred to as the Lichnerowicz junction conditions.

1.4.11 Event horizon If the normal vector (1.4.219) to a hypersurface is a null vector,

i nin = 0, (1.4.257) then this hypersurface is a null hypersurface. Equations (1.4.220) and (1.4.257) indicate that ni lies itself on the null hypersurface to which it is normal,

dxi ∝ ni, (1.4.258) which also gives 2 i i ds = dxidx ∝ nin = 0. (1.4.259) Therefore, all world lines on a null hypersurface are null. The light cones at the points of such a hypersurface are tangent to this hypersurface. Since all physical world lines must lie within the local light cones, the forward-time motion through a null hypersurface can occur in only one direction. To avoid any discontinuities, this direction is the same for all points on such a hypersurface. A null hypersurface is therefore an event horizon: a boundary in spacetime beyond which events cannot aﬀect events on the other side. All laws of classical physics are known to be time-symmetric, that is, symmetric under the transformation t → −t. However, the existence of event horizons, which are solutions to these laws and provide boundary conditions for spacetime, violates this symmetry. The unidirectional character of the motion through an event horizon can be used to deﬁne the past and future: the arrow of time. References: [1, 2, 3, 4, 5].

1.5 Tetrad and spin connection 1.5.1 Tetrad In addition to the coordinate systems, at each point in spacetime we can set up four linearly inde- i pendent vectors ea such that i eaeib = ηab, (1.5.1) where a, b = 0, 1, 2, 3 are Lorentz indices and ηab = diag(1, −1, −1, −1) is the coordinate-invariant Minkowski metric tensor in a locally geodesic frame of reference at this point. This set of four vectors is referred to as a tetrad. The inverse tetrad eai satisﬁes

i b b eaei = δa, (1.5.2) i a i eaek = δk. (1.5.3)

ik The coordinate metric tensors gik and g are related to the Minkowski metric tensor through the tetrad:

a b gik = ei ekηab, (1.5.4) ik i k ab g = eaeb η , (1.5.5) where ηab satisﬁes bc b ηacη = δa. (1.5.6)

43 Any vector V can be speciﬁed by its components V i with respect to the coordinate system or by the coordinate-invariant projections V a of the vector onto the tetrad ﬁeld:

a a i i V = ei V ,Va = eaVi, (1.5.7) i i a a V = eaV ,Vi = ei Va, (1.5.8)

ab and similarly for tensors and densities with more indices. We can use ηab and its inverse η to lower ik and raise Lorentz indices, as we use gik and its inverse g to lower and raise coordinate indices. a Let us consider the determinant of the matrix composed from the components of the tetrad ei ,

a e = |ei |. (1.5.9)

This determinant is related to the determinant g of the metric tensor gik, using (1.5.4), by

e = p|g|. (1.5.10)

The diﬀerential and derivatives of the determinant (1.5.9) are given, analogously to (1.4.18) and (1.4.19), by

i a a i de = eeadei = −eei dea, (1.5.11) i a a i e,k = eeaei,k = −eei ea,k. (1.5.12) The variation of (1.5.9) is thus, analogously to (1.4.20), equal to

i a a i δe = eeaδei = −eei δea. (1.5.13) Similarly to (1.4.21), the covariant derivative of (1.5.9) vanishes:

e;j = 0. (1.5.14)

1.5.2 Lorentz transformation The relation (1.5.4) imposes 10 constraints on the 16 components of the tetrad, leaving 6 components i i arbitrary. If we change from one tetrad ea to another,e ˜b, then the vectors of the new tetrad are linear combinations of the vectors of the old tetrad:

i b i e˜a = Λ aeb. (1.5.15)

i The relation (1.5.4) applied to the tetrad ﬁelde ˜b, a b gik =e ˜i e˜kηab, (1.5.16)

b imposes on the matrix Λ a the orthogonality condition: c d Λ aΛ bηcd = ηab. (1.5.17)

b We refer to Λ a as a Lorentz matrix, and to a transformation of form (1.5.15) as the Lorentz transformation.

1.5.3 Tetrad transport A natural choice for the zeroth component of a tetrad at a given point is

i i e0 = u . (1.5.18) Along a world line this tetrad should be transported such that the zeroth component always coincides with the four-velocity. The Fermi-Walker transport of a tetrad is deﬁned as

∇ei Du Dui a = −uiej j + ej u . (1.5.19) ds a ds ds a j

44 Putting a = 0 in (1.5.19) gives ∇ui Dui = , (1.5.20) ds ds so the Fermi-Walker transport of the four-velocity is equivalent to its covariant change and thus (1.5.18) is valid at all points. This transport does not change the orthogonality relation for tetrads (1.5.1) because (1.5.19) gives ∇ (ei e ) = 0. (1.5.21) ds a ib

1.5.4 Spin connection We deﬁne i i i i j ω ak = ea;k = ea,k + Γj kea. (1.5.22) The quantities a a k a k k j ω bi = ekω bi = ek(eb,i + Γj ieb) (1.5.23) transform like vectors under coordinate transformations. We can extend the notion of covariant dif- ab ferentiation to quantities with Lorentz coordinate-invariant indices by regarding ω i as a connection, referred to as Lorentz or spin connection. For a contravariant Lorentz vector

a a a b V |i = V ,i + ω biV , (1.5.24)

i where |i is the covariant derivative of such a quantity with respect to x . The covariant derivative a of a scalar V Wa coincides with its ordinary derivative:

a a (V Wa)|i = (V Wa),i, (1.5.25) which gives the covariant derivative of a covariant Lorentz vector:

b Wa|i = Wa,i − ω aiWb. (1.5.26)

The chain rule infers that the covariant derivative of a Lorentz tensor is equal to the sum of the corresponding ordinary derivative of this tensor and terms with spin connection corresponding to each Lorentz index:

ab... ab... a eb... b ae... e ab... e ab... T cd...|i = T cd...,i + ω eiT cd... + ω eiT cd... + · · · − ω ciT ed... − ω diT ce... − .... (1.5.27)

We assume that the covariant derivative |i is total, that is, also recognizes coordinate indices, acting on them like ;i. For a tensor with both coordinate and Lorentz indices

aj... aj... a ej... j al... e aj... l aj... T bk...|i = T bk...,i + ω eiT bk... + Γl iT bk... + · · · − ω biT ek... − Γk iT bl... − .... (1.5.28)

A total covariant derivative of a tetrad is

i i i j b i ea|k = ea,k + Γj kea − ω akeb = 0, (1.5.29) because of (1.5.22). Therefore, total covariant differentiation commutes with converting between a coordinate and Lorentz indices. Equation (1.5.29) determines the spin connection ω bi in terms of the affine connection, tetrad and its ordinary derivatives, in accordance with (1.5.23). Conversely, the affine connection is determined by the spin connection, tetrad and its derivatives:

j j a j Γi k = ω ik + ei,kea. (1.5.30) The torsion tensor is then related to these quantities by

j j a j S ik = ω [ik] + e[i,k]ea, (1.5.31)

45 and the torsion vector is k a k Si = ω [ik] + e[i,k]ea. (1.5.32) Metric compatibility of the aﬃne connection leads to

a b a b c c gik;j = gik|j = ei ekηab|j = −ei ek(ω ajηcb + ω bjηac) = −(ωkij + ωikj) = 0, (1.5.33) so the spin connection is antisymmetric in its ﬁrst two indices:

a a ω bi = −ωb i. (1.5.34) Accordingly, the spin connection has 24 independent components. The contortion tensor is related to the spin connection by Cijk = ωijk + ∆ijk, (1.5.35) where a a a ∆ijk = eiae[j,k] − ejae[i,k] − ekae[i,j] (1.5.36) are the Ricci rotation coefficients. The first term on the right-hand side in (1.5.35) is expected because both the contortion tensor and spin connection are antisymmetric in their first two indices. The quantities i i i i j $ ak = ea:k = ea,k + {j k}ea (1.5.37) form the Levi-Civita spin connection and are related to the Ricci rotation coefficients by (1.5.35) with Cijk = 0, $ijk = −∆ijk, (1.5.38) so Cijk = ωijk − $ijk. (1.5.39)

1.5.5 Tetrad representation of curvature tensor The commutator of the covariant derivatives of a tetrad with respect to the aﬃne connection is

k k σ l k 2ea;[ji] = R lijea + 2S ijea;l. (1.5.40) This commutator can also be expressed in terms of the spin connection:

k k k b kb b k ea;[ji] = ω a[j;i] = (eb ω a[j);i] = ωba[jω i] + ω a[j;i]eb kb b k l k = ωba[jω i] + ω a[j,i]eb + S ijω al. (1.5.41) Consequently, the curvature tensor with two Lorentz and two coordinate indices depends only on the spin connection and its ordinary derivatives:

a a a a c a c R bij = ω bj,i − ω bi,j + ω ciω bj − ω cjω bi. (1.5.42) Because the spin connection is antisymmetric in its ﬁrst two indices, the tensor (1.5.42) is antisymmetric in its ﬁrst two (Lorentz) indices, like the Riemann tensor. The contraction of the curvature tensor (1.5.42) with a tetrad gives the Ricci tensor with one Lorentz and one coordinate index:

a i Rbj = R bijea. (1.5.43)

a The contraction of the tensor R i with a tetrad gives the Ricci scalar,

a i ab i j R = R iea = R ijeaeb. (1.5.44) The Riemann tensor with two Lorentz and two coordinate indices depends on the Levi-Civita connection (1.5.37) the same way the curvature tensor depends on the aﬃne connection:

a a a a c a c P bij = $ bj,i − $ bi,j + $ ci$ bj − $ cj$ bi. (1.5.45)

46 The contraction of (1.5.45) with a tetrad gives the Riemannian Ricci tensor with one Lorentz and one coordinate index: a i Pbj = P bijea. (1.5.46) a The contraction of the tensor P i with a tetrad gives the Riemann scalar,

a i ab i j P = P iea = P ijeaeb. (1.5.47)

References: [3, 4, 6, 7, 8].

1.6 Lorentz group Lorentz transformations relate diﬀerent tetrads at a given point in spacetime, where the metric tensor ijkl ijkl can be brought to the Galilean form: gik = ηik. Accordingly, we have = e , εijkl = eijkl, αβγ αβγ = e , and εαβγ = eαβγ .

1.6.1 Subgroups of Lorentz group and Einstein principle of relativity

A composition of two Lorentz transformations Λ1 and Λ2,

a a c Λ b = Λ(1)cΛ(2)b, (1.6.1)

a satisﬁes (1.5.17), thereby it is a Lorentz transformation. The Kronecker symbol δb also satisﬁes (1.5.17), thereby it can be regarded as the identity Lorentz transformation. Therefore, Lorentz transformations form a group, referred to as the Lorentz group. Taking the determinant of the relation (1.5.17) gives a |Λ b| = ±1. (1.6.2) a a A Lorentz transformation with |Λ b| = 1 is proper and with |Λ b| = −1 is improper. Proper Lorentz transformations form a group because the determinant of the product of two proper Lorentz transformations is 1. Improper Lorentz transformations include the parity transformation P

a Λ b(P ) = diag(1, −1, −1, −1), t → t, x → −x, (1.6.3) and the time reversal T a Λ b(T ) = diag(−1, 1, 1, 1), t → −t, x → x. (1.6.4) 0 0 0 0 The relation (1.5.17) gives Λ 0Λ 0 − Λ αΛ α = 1, thereby

0 |Λ 0| ≥ 1. (1.6.5)

0 i Lorentz transformations with Λ 0 ≥ 1 are orthochronous and form a group. If x is a timelike vector, i 00 0 0 0 α x xi > 0, then for an orthochronous transformation x = Λ 0x + Λ αx , q q 0 α 0 0 β β 0 2 0 2 0 0 |Λ αx | ≤ Λ αΛ αx x < (Λ 0) (x ) = |Λ 0x |. (1.6.6)

Therefore, the time component of a timelike vector does not change the sign under orthochronous transformations. Einstein’s special principle of relativity states that physical laws do not change their form under transformations within the orthochronous proper subgroup of the Lorentz group. Equivalently, physical laws have the same form in all admissible inertial frames of reference. The special principle of relativity is a special case of the general principle of relativity, in which arbitrary diﬀerentiable coordinate transformations are restricted to linear transformations (orthochronous proper Lorentz transformations) between inertial frames of reference. Under the parity transformation, the spatial components of contravariant and covariant vectors, which form spatial vectors, change the sign. The permutation symbols do not change under this transformation. Accordingly, the spatial components of dual vector densities, such as the components of a vector product (1.4.157) or a curl (1.4.168), do not change the sign. Such quantities, that

47 transform under proper Lorentz transformations like vectors and do not change the sign in their spatial components under the parity transformation, are referred to as axial vectors or pseudovectors. Similarly, the scalar contraction of the Levi-Civita symbol and a tensor changes the sign, while a scalar does not. Quantities that transform under proper Lorentz transformations like scalars and change the sign under the parity transformation are referred to as pseudoscalars.

1.6.2 Inﬁnitesimal Lorentz transformations Let us consider an inﬁnitesimal Lorentz transformation

µ µ µ Λ ν = δν + ν , (1.6.7)

µ where ν are inﬁnitesimal quantities. The relation (1.5.17) gives

µν = −νµ, (1.6.8) where the indices are raised and lowered using the Minkowski metric tensor. Therefore, Lorentz transformations are given by 6 independent antisymmetric parameters µν . The corresponding transformation of a contravariant vector Aµ is 1 1 A0µ = Aµ + µ Aν = Aµ + ρσ(δµη − δµη )Aν = Aµ + ρσJ µ Aν , (1.6.9) ν 2 ρ σν σ ρν 2 νρσ where µ µ µ Jνρσ = δρ ησν − δσ ηρν . (1.6.10)

We deﬁne matrices Jρσ such that µ µ (Jρσ)ν = Jνρσ. (1.6.11) Therefore, in the matrix notation (with Aµ treated as a column), 1 A0 = 1 + ρσJ A. (1.6.12) 2 ρσ

The 6 matrices Jρσ are the inﬁnitesimal generators of the vector representation of the Lorentz group. The explicit form of the generators of the Lorentz group in the vector representation is

 0 −1 0 0   0 0 −1 0   −1 0 0 0   0 0 0 0  J01 =   ,J02 =   ,  0 0 0 0   −1 0 0 0  0 0 0 0 0 0 0 0  0 0 0 −1   0 0 0 0   0 0 0 0   0 0 −1 0  J03 =   ,J12 =   ,  0 0 0 0   0 1 0 0  −1 0 0 0 0 0 0 0  0 0 0 0   0 0 0 0   0 0 0 0   0 0 0 1  J23 =   ,J31 =   . (1.6.13)  0 0 0 −1   0 0 0 0  0 0 1 0 0 −1 0 0

1.6.3 Generators and Lie algebra of Lorentz group The commutator of the generators of the Lorentz group in the vector representation is given, using (1.6.10) and (1.6.11), by

µ µ λ µ λ µ [Jκτ ,Jρσ]ν = (Jκτ )λ(Jρσ)ν − (Jρσ)λ(Jκτ )ν = (−Jκρητσ − Jτσηκρ + Jκσητρ + Jτρηκσ)ν , (1.6.14) so [Jκτ ,Jρσ] = −Jκρητσ − Jτσηκρ + Jκσητρ + Jτρηκσ. (1.6.15)

48 The relation (1.6.15) constitutes the Lie algebra of the Lorentz group. If a set of qantitites φ transforms under a Lorentz transformation Λ with a matrix D(Λ)

φ → D(λ)φ, (1.6.16) then D is a representation of the Lorentz group if

D(I) = I,D(Λ1Λ2) = D(Λ1)D(Λ2), (1.6.17)

where I denotes the identity transformation, and Λ1 and Λ2 are two Lorentz transformations. There- fore, we have D(Λ−1) = D−1(Λ), (1.6.18) where Λ−1 is the Lorentz transformation to Λ: ΛΛ−1 = I. For an inﬁnitesimal Lorentz transformation in any representation, 1 D(Λ) = I + ρσJ , (1.6.19) 2 ρσ according to (1.6.12). The relation

−1 −1 D(Λ1Λ2Λ1 ) = D(Λ1)D(Λ2)D (Λ1) (1.6.20) gives (1.6.15), valid for any representation of the Lorentz group. −1 If Λ1 and Λ2 are two group transformations then Λ3 = Λ1Λ2Λ1 is a group transformation. If −1 Λ2 = I + 2G2 is an infinitesimal group transformation with generator G2 then Λ3 = I + 2Λ1G2Λ1 −1 is an infinitesimal group transformation with generator G3 = Λ1G2Λ1 . If Λ1 = I + 1G1 is an infinitesimal group transformation with generator G1 then, neglecting terms in 1 of higher order, G3 = G2 +1[G1,G2], thereby [G1,G2] is a generator. For a finite number N of linearly independent N generators, a general infinitesimal group transformation is Λ = I + Σa=1aGa. Because [Ga,Gb] is a N generator, it is a linear combination of the N generators: [Ga,Gb] = Σc=1fabcGc, where fabc are the structure constants of the Lie algebra of the given group. For the Lorentz group, aGa = D(Λ) − I, where D(Λ) is given by (1.6.19).

1.6.4 Rotations and boosts Rotations are proper orthochronous Lorentz transformations with

0 α 0 Λ α = Λ 0 = 0, Λ 0 = 1. (1.6.21) Rotations act only on the spatial coordinates xα and form a group, referred to as the rotation group. Boosts are proper orthochronous Lorentz transformations with

α Λ β = 0. (1.6.22) We deﬁne 1 J = e J βγ , (1.6.23) α 2 αβγ Kα = J0α, (1.6.24) and 1 ϑ = e βγ , (1.6.25) α 2 αβγ ηα = 0α. (1.6.26)

The explicit form of the generators of the rotation group Jα in the vector representation is  0 0 0   0 0 1   0 −1 0  J1 =  0 0 −1  ,J2 =  0 0 0  ,J3 =  1 0 0  . (1.6.27) 0 1 0 −1 0 0 0 0 0

49 For an inﬁnitesimal Lorentz transformation (1.6.19)

D = I + ϑ · J + η · K. (1.6.28)

A ﬁnite Lorentz transformation can be regarded as a composition of successive identical inﬁnites- imal Lorentz transformations:

n θ·J+η·K D = limn→∞(I + θ · J/n + η · K/n) = e . (1.6.29)

The ﬁnite parameters θ, η are the canonical parameters for a given Lorentz transformations. For a ﬁnite Lorentz transformation, (1.6.19) gives

1 ρσ J D(Λ) = e 2 ρσ , (1.6.30) so ∂D(Λ) Jµν = . (1.6.31) ∂µν Λ=I The explicit form of a ﬁnite Lorentz transformation in the vector representation is

 1 0 0 0   1 0 0 0 

θJ1  0 1 0 0  θJ2  0 cosθ 0 sinθ  R1 = e =   ,R2 = e =   ,  0 0 cosθ −sinθ   0 0 1 0  0 0 sinθ cosθ 0 −sinθ 0 cosθ  1 0 0 0   coshη sinhη 0 0 

θJ3  0 cosθ −sinθ 0  ηK1  sinhη coshη 0 0  R3 = e =   ,B1 = e =   ,  0 sinθ cosθ 0   0 0 1 0  0 0 0 1 0 0 0 1  coshη 0 sinhη 0   coshη 0 0 sinhη 

ηK2  0 1 0 0  ηK3  0 1 0 0  B2 = e =   ,B3 = e =   ,  sinhη 0 coshη 0   0 0 1 0  0 0 0 1 sinhη 0 0 coshη (1.6.32)

α where Rα denotes a rotation about the x axis and Bα denotes a boost along this axis. The canonical parameters θ and η are respectively referred to as the angle of rotation and rapidity. The parameters ϑ and η in (1.6.25) and (1.6.26) are thus respectively inﬁnitesimal values of the angle of rotation and rapidity. A rotation about any axis, say z, by an angle θ turns the two other axes, x and y, into new axes, x0 and y0, such that the angle between x and x0 (or y and y0) (1.4.152) is θ. The rotation group is compact: θα ∈ [0, 2π] and θ = 2π ⇔ θ = 0. The explicit form of a ﬁnite rotation in the three-dimensional vector representation is

 1 0 0   cosθ 0 sinθ  R1(θ) =  0 cosθ −sinθ  ,R2(θ) =  0 1 0  , 0 sinθ cosθ −sinθ 0 cosθ  cosθ −sinθ 0  R3(θ) =  sinθ cosθ 0  . (1.6.33) 0 0 1

For instance,    0      Vx Vx Vx Vxcosθ − Vysinθ 0  Vy  →  Vy  = R3  Vy  =  Vxsinθ + Vycosθ  . (1.6.34) 0 Vz Vz Vz Vz The relation (1.6.31) gives ∂Rα(θ) Jα = . (1.6.35) ∂θ θ=0

50 The orthogonality relation (1.5.17) applied to any of the rotation matrices (1.6.33) shows that a rotation matrix R is orthogonal, that is, its transpose RT is equal to its inverse R−1:

T −1 T T Rα = Rα ,RαRα = Rα Rα = I, (1.6.36) where I is the identity matrix. The commutation relation (1.6.15) gives

[Jα,Jβ] = eαβγ Jγ , (1.6.37)

[Jα,Kβ] = eαβγ Kγ , (1.6.38)

[Kα,Kβ] = −eαβγ Jγ . (1.6.39)

Therefore, rotations do not commute and form a nonabelian group, rotations and boosts do not commute, and boosts do not commute. Changing the order of two nonparallel boosts is equivalent to applying a rotation, referred to as the Thomas-Wigner rotation. The structure constants of the Lie algebra of the rotation group are fabc = eabc. Moreover, the square of the generators of rotation,

2 J = JαJα, (1.6.40) commutes with Jα:

2 [J ,Jβ] = [Jα,Jβ]Jα + Jα[Jα,Jβ] = eαβγ (Jγ Jα + JαJγ ) = 0. (1.6.41)

Deﬁnining 1 L = (J + iK), (1.6.42) 2 1 Q = (J − iK), (1.6.43) 2 gives

[Lα,Lβ] = eαβγ Lγ , (1.6.44)

[Qα,Qβ] = eαβγ Qγ , (1.6.45)

[Lα,Qβ] = 0, (1.6.46)

so the Lorentz group is isomorphic with the product of two complex rotation groups. Accordingly, the Lorentz group can be regarded as the group of four-dimensional rotations in the Minkowski space, or the group of tetrad rotations.

1.6.5 Poincarégroup Under an infinitesimal coordinate transformation (1.2.66) in a locally flat spacetime, (1.4.51) gives

ηik → ηik − ξi,k − ξk,i. (1.6.47)

i Therefore, the tensor ηik is invariant under (1.2.66) (isometric) if ξ is a Killing vector,

ξ(i,k) = 0, (1.6.48) which has the solution i ik i ξ = xk + , (1.6.49) where ik and i are constant. The ﬁrst term on the right-hand side of (1.6.49) corresponds to a Lorentz rotation described by 6 parameters ik satisfying (1.6.8). The second term on the right-hand side of (1.6.49) corresponds to a translation. A combination of two translations does not change if their order is reversed, thereby translations commute:

[Tµ,Tν ] = 0, (1.6.50)

51 α α where Tµ is the generator of translation. The relations (1.6.37) and (1.6.38) mean that J and K are spatial vectors under rotations. Spatial translations are spatial vectors under rotations, while a time translation is a scalar:

[Jα,Tβ] = eαβγ Tγ , (1.6.51)

[Jα,T0] = 0. (1.6.52) The last relation indicates that the generators of rotations, like the generators of spatial translations, correspond to conserved quantities, which are quantities that do not change in time. The covariant generalization of (1.6.51) and (1.6.52) is

[Jµν ,Tρ] = Tµηνρ − Tν ηµρ. (1.6.53) The relations (1.6.15), (1.6.50) and (1.6.53) constitute the Lie algebra of the inhomogeneous Lorentz or Poincar´egroup. In particular,

[Kα,Tβ] = −T0δαβ, (1.6.54)

[Kα,T0] = −Tα. (1.6.55) The last relation indicates that the generators of boosts do not correspond to conserved quantities. For an inﬁnitesimal rotation about the z axis,

(I + ϑJz)f(ct, x) = D(Rz(ϑ))f(ct, x) = f(ct, Rz(ϑ)x) ≈ f(ct, x − ϑy, ϑx + y, z) ∂f ∂f = f(ct, x) − ϑy + ϑx , (1.6.56) ∂x ∂y or ∂ ∂ J = x − y , (1.6.57) z ∂y ∂x which gives the diﬀerential representation of rotations:

Jα = eαβγ xβ∂γ . (1.6.58) For an inﬁnitesimal boost along the z axis,

(I + ηKz)f(ct, x) = D(Bz(η))f(ct, x) = f(Bz(η)(ct, x)) ≈ f(ct + ηz, y, z + ηct) ∂f ∂f = f(ct, x) + ηz + ηct , (1.6.59) c∂t ∂z or ∂ ∂ K = z + ct , (1.6.60) z c∂t ∂z which gives the differential representation of boosts: ∂ ∂ K = x + ct . (1.6.61) α α c∂t ∂xα The relation for an infinitesimal translation, analogous to (1.6.19), is µ D(t) = I + Tµ, (1.6.62) so a finite translation is given by µ D(t) = e Tµ . (1.6.63) Translation in (1.6.49) can also be written as ν ν ν tµ()x = x + δµ. (1.6.64) The relation analogous to (1.6.35) is ∂tµ() Tµ = . (1.6.65) ∂ =0 The differential representation of a translation is thus ∂ T = . (1.6.66) µ ∂xµ

52 1.6.6 Invariants of Lorentz and Poincar´egroup Analogously to (1.6.41),

2 [L ,Lβ] = 0, (1.6.67) 2 [Q ,Qβ] = 0, (1.6.68)

so L2 and Q2 commute with all 6 generators of the Lorentz group. Consequently, J 2 + K2 and J · K commute with all generators of the Lorentz group, that is, are the invariants or Casimir operators of the Lorentz group. The Casimir operators of Lorentz group do not commute with the generators of translation Tµ, thereby they are not the invariants of the Poincar´egroup. Instead, the mass operator

2 µ m = −T Tµ (1.6.69) and 2 µ W = W Wµ, (1.6.70) where W µ is the Pauli-Lubańskipseudovector 1 W µ = eµνρσJ T , (1.6.71) 2 ρσ ν commute with all generators of the Poincarégroup, thereby they are the Casimir operators of the Poincarégroup. The Pauli-Lubańskipseudovector obeys the commutation relations

[Tµ,Wν ] = 0, (1.6.72)

[Jµν ,Wρ] = Wµηνρ − Wν ηµρ, (1.6.73) µ ν µνρσ [W ,W ] = e WρTσ. (1.6.74)

The relation (1.6.73) is analogous to (1.6.53) because W µ behaves like a vector under proper Lorentz transformations. We deﬁne the four-momentum operator

Pµ = iTµ, (1.6.75)

whose time component is the energy operator P0 = iT0 and spatial components form the momentum operator Pα = iTα. We deﬁne the angular four-momentum operator

Mµν = iJµν , (1.6.76)

whose spatial components form the angular momentum operator

Mα = iJα. (1.6.77)

Therefore, the following relations are satisﬁed:

[Mµν ,Mρσ] = −i(Mµρηνσ + Mνσηµρ − Mµσηνρ − Mνρηµσ), (1.6.78)

[Pµ,Pν ] = 0, (1.6.79)

[Mµν ,Pρ] = i(Pµηνρ − Pν ηµρ), (1.6.80) 2 µ m = P Pµ, (1.6.81) 1 W µ = − eµνρσM P , (1.6.82) 2 ρσ ν [Pµ,Wν ] = 0, (1.6.83)

[Mµν ,Wρ] = i(Wµηνρ − Wν ηµρ), (1.6.84) µ ν µνρσ [W ,W ] = −ie WρPσ, (1.6.85)

[Mα,Mβ] = ieαβγ Mγ . (1.6.86)

53 1.6.7 Relativistic kinematics The quantitites vα (1.4.116) form the three-dimensional vector of velocity:

dx v = , (1.6.87) dt where x is the radius vector. The magnitude of the velocity is equal to the speed (1.4.135):

|v| = v. (1.6.88)

Let us consider a boost in the direction of the z axis

x0i = e−ηK3 xi, (1.6.89)

where xi and x0i have a form of a column (4×1 matrix), and eηK3 is given by (1.6.32). Therefore, the coordinates in an inertial K-system (unprimed) are related to the coordinates in an inertial K0-system (primed) by

ct = ct0coshη + z0sinhη, x = x0, y = y0, z = z0coshη + ct0sinhη. (1.6.90)

Let us consider the origin of the K0-system, x0 = y0 = z0 = 0, in the K-system. Therefore, we have

ct = ct0coshη, z = ct0sinhη, (1.6.91)

dz 0 which gives the relation between the rapidity η and speed V = dt of K relative to K: tanhη = β, (1.6.92) where V V β = , β = . (1.6.93) c c Accordingly, coshη = γ and sinhη = βγ, where

V 2 −1/2 γ = 1 − . (1.6.94) c2 The relations (1.6.90) become

V t = γ t0 + z0 , c2 x = x0, y = y0, z = γ(z0 + V t0), (1.6.95)

and are referred to as a special Lorentz transformation in the z-direction. The reverse transformation is V t0 = γ t − z , c2 x0 = x, y0 = y, z0 = γ(z − V t). (1.6.96)

For a boost along an arbitrary direction, the spatial vector x = (x, y, z) transforms such that its 0 2 component parallel to the velocity V = cβ of K relative to K, xk = (x · V)V/V (similarly for

54 primed), behaves like z in (1.6.95) and its component perpendicular to V, x⊥ = x − xk, behaves like x in (1.6.95):

V · x0 t = γ t0 + , c2 0 x⊥ = x⊥, 0 0 xk = γ(xk + Vt ), (1.6.97) so (γ − 1)(V · x0)V x = γ(x0 + Vt0) + x0 = γVt0 + x0 + . (1.6.98) k ⊥ V 2 Therefore, the transformation law for the coordinates in two inertial frames of reference is ! ct γ γβ ct0 = (γ−1)β 0 , (1.6.99) x γβ 1 + β2 β x

or equivalently ! ct0 γ −γβ ct 0 = (γ−1)β . (1.6.100) x −γβ 1 + β2 β x The matrix in (1.6.100) is called a boost matrix. In the local Minkowski spacetime, contravariant vectors transform like xi, according to (1.6.97) and (1.6.99), ! W 0 γ γβ W 00 = (γ−1)β 0 , (1.6.101) W γβ 1 + β2 β W

covariant vectors transform such that they remain related to contravariant vectors by the Minkowski metric tensor, and tensors transform like products of vectors. For example, if V = cβzˆ is parallel to the z axis, a tensor of rank (0,2) transforms according to

2 2 T00 = γ(T000 + βT030 ) = γ (T0000 + βT3000 + βT0030 + β T3030 ),

T0⊥ = γ(T00⊥0 + βT30⊥0 ), 2 2 T03 = γ(T030 + βT000 ) = γ (T0030 + βT3030 + βT0000 + β T3000 ),

T⊥⊥ = T⊥0⊥0 ,

T3⊥ = γ(T30⊥0 + βT00⊥0 ), 2 2 T33 = γ(T330 + βT300 ) = γ (T3030 + βT0030 + βT3000 + β T0000 ), (1.6.102)

T where the index ⊥ denotes either 1 or 2, and the transposed components Tik = Tki transform like the transpositions of the right-hand sides in (1.6.102). If Tik is antisymmetric then T03 = T0030 . The relations (1.6.95) can be written as V dt = γ dt0 + dz0 , c2 dx = dx0, dy = dy0, dz = γ(dz0 + V dt0), (1.6.103)

which gives

0 vx vx = 0 2 , γ(1 + V vz/c ) 0 vy vy = 0 2 , γ(1 + V vz/c ) 0 vz + V vz = 0 2 , (1.6.104) 1 + V vz/c

55 where dx dx0 v = , v0 = . (1.6.105) dt dt0 Two special Lorentz transformations in the same direction commute because of (1.6.39). If a Lorentz 0 00 transformation from K to K has parameters β1 and γ1, and a Lorentz transformation from K to 0 00 K has parameters β2 and γ2, then a Lorentz transformation from K to K has parameters β3 and γ3 such that β1 + β2 β3 = , γ3 = γ1γ2(1 + β1β2). (1.6.106) 1 + β1β2 For a boost along an arbitrary direction, (1.6.99) gives the Lorentz transformation of velocities:

v0 + γV + (γ − 1)(v0 · V)V/V 2 v = . (1.6.107) γ(1 + v0 · V/c2)

If v0 = |V0| = c then v = |V| = c, in agreement with the constancy of the speed of propagation of interaction. Let us consider two points at rest in an inertial frame of reference K with positions z1 and z2, 0 thereby the distance between them is ∆z = z2 − z1. In the inertial frame K , moving relative to K 0 0 0 0 0 0 in the z-direction with speed V , z1 = γ(z1 +V t1) and z2 = γ(z2 +V t2), thereby if t1 = t2 is the time 0 0 0 at which we measure (simultaneously) the positions of the two points then ∆z = γ(z2 − z1) = γ∆z . Therefore, the length of an object in K0, whose length in the rest frame K is l (proper length), is

l l0 = < l, (1.6.108) γ

which is referred to as the Lorentz-FitzGerald contraction. The volume of an object in K0, whose volume in the rest frame K is V (proper volume), is

V V 0 = . (1.6.109) γ Let us suppose that there are two rods of equal lengths, moving parallel relative to each other. From the point of view of an observer moving with the first rod, the second one is shorter, and from the point of view of an observer moving with the second rod, the first one is shorter. There is no contradiction in this statement because the positions of both ends of a rod must be measured simultaneously and the simultaneity is not invariant: from the transformation law (1.6.95) it follows that if δt = 0 then δt0 6= 0 and if δt0 = 0 then δt 6= 0. Let us consider a clock (any mechanism with a periodic or evolutionary behavior) at rest in K0 0 0 0 with position z ; the time difference between two events with t1 and t2, as measured by this clock, 0 0 0 0 0 2 0 0 2 is ∆t = t2 − t1. In the frame K, t1 = γ(t1 + V z /c ) and t2 = γ(t2 + V z /c ), thereby

0 0 ∆t = t2 − t1 = γ∆t > ∆t . (1.6.110)

Therefore, the rate of time is slower for moving clocks than those at rest (time dilation), in agreement with (1.4.120) and (1.4.126), from which c2dτ 2 = c2dt2 − dl2 and 1 dτ = dt. (1.6.111) γ

Let us suppose that there are two clocks linked to the inertial frames K and K0, and that when the clock in K passes by the clock in K0 the readings of the two clocks coincide. From the point of view of an observer in K clocks in K0 go more slowly, and from the point of view of an observer in K0 clocks in K go more slowly. There is no contradiction in this statement because to compare the rates of the two clocks in K and K0 we must compare the readings of the same moving clock in K0 with diﬀerent clocks in K; we require several clocks in one frame and one in the other, thus the measurement process is not symmetric with respect to the two frames of reference. The clock that

56 goes more slowly is the one which is being compared with diﬀerent clocks in the other frame. The time interval measured by a clock is equal to the integral 1 Z ∆t = ds (1.6.112) c along its world line. Since the world line is a straight line for a clock at rest and a curved line for a clock moving such that it returns to the starting point, the integral R ds taken between two world points has its maximum value if it is taken along the straight line connecting these two points. For a Lorentz transformation with speed V = |V|, (1.6.107) gives

v0sinθ0 tanθ = , (1.6.113) γ(v0cosθ0 + V )

where θ is the angle between v and V, and θ0 is the angle between v0 and V. If v = v0 = c then

cosθ0 + V cosθ = c , (1.6.114) V 0 1 + c cosθ which is referred to as the aberration of a signal. Let us suppose that an observer in frame K 1 c measures a periodic signal with period T , frequency ν = T and wavelength λ = ν , propagating in the −z direction; the number of pulses in time dt is n = νdt. A second observer in frame K0, moving in the z direction with speed V relative to the ﬁrst one, travels a distance V dt and measures V dt 0 V 0 0 dt λ more pulses: n = ν(1 + c )dt. Because the time interval dt with respect to K is dt = γ , the 0 0 V frequency of the signal in K is ν = γν(1 + c ) or ν0 = eην. (1.6.115)

This dependence of the frequency of a signal on a frame of reference is referred to as the Doppler eﬀect. When c → ∞ (at which γ → 1) the above formulae, referring to relativistic kinematics, reduce to their nonrelativistic limit. The Lorentz transformation (1.6.99) reduces to the Galilei transformation,

t = t0, x = x0 + Vt0, (1.6.116)

so the time is an absolute (invariant) quantity in nonrelativistic (Newtonian) physics. Any two Galilei transformations commute. The transformation law for velocities (1.6.107) reduces to the simple addition of vectors, v = v0 + V. (1.6.117)

1.6.8 Four-acceleration In the Galilean frame, the line element in (1.4.117) is equal to

P vαvα 1/2 v2 1/2 cdt ds = cdt 1 − α = cdt 1 − = . (1.6.118) c2 c2 γ

This line element is a special case of that in (1.4.134) for g00 = 1 and g0α = 0. The corresponding diﬀerential of the proper time is equal to (1.6.111). In a locally inertial frame of reference, the components of the four-velocity in the Cartesian coordinates are

dx0 cdt dxα dxα γ u0 = = = γ, uα = = = vα, (1.6.119) ds ds ds cdt/γ c

which can be written as γ γ ui = γ, v , u = γ, − v , (1.6.120) c i c

57 where v is the velocity (1.6.87) and γ = √ 1 (confer (1.6.94)). The spatial components uα in 1−v2/c2 (1.6.119) coincide with those in (1.4.136), whereas u0 in (1.6.119) is a special case of that in (1.4.136) for g00 = 1 and g0α = 0. We deﬁne the four-acceleration:

Dui D2xi wi = = = ukui . (1.6.121) ds ds2 ;k This vector is orthogonal to ui because of (1.4.14): 1 D wiu = (uiu ) = 0, (1.6.122) i 2 ds i thus having 3 independent components. In a locally inertial frame of reference, the four-acceleration is given by dui d2xi wi = = = ukui . (1.6.123) ds ds2 ,k Its components in the Cartesian coordinates are

du0 dγ dx0 γ dγ 1 d 1 d 1 v2 −2 v dv w0 = = = = (γ2) = = 1 − · ds cdt ds c dt 2c dt 2c dt 1 − v2/c2 c2 c3 dt γ4 = v · a, c3 duα duα dx0 γ d γ2 vα dγ γ2 γ4 wα = = = (γvα) = aα + γ = aα + (v · a)vα, (1.6.124) ds cdt ds c2 dt c2 c2 dt c2 c4 which can be written as γ4 γ2 γ4 γ4 γ2 γ4 wi = v · a, a + (v · a)v , w = v · a, − a − (v · a)v , (1.6.125) c3 c2 c4 i c3 c2 c4 where a is the three-dimensional acceleration vector: dvα d2xα dv d2x aα = = , a = = . (1.6.126) dt dt2 dt dt2 The invariant square of the four-acceleration is thus

γ8 γ2 γ4 2 γ4 γ2 wiw = (v · a)2 − a + (v · a)v = − a2 + (v · a)2 . (1.6.127) i c6 c2 c4 c4 c2 If v = 0 at a given instant of time, the corresponding frame of reference is referred to as the instantaneous rest frame. In this frame

a2 wiw = − , (1.6.128) i c4 so 2p i a0 = c −w wi (1.6.129) is the magnitude of the acceleration in the instantaneous rest frame, called the proper acceleration. Along an affine geodesic, the four-acceleration with respect to the affine connection (1.6.121) vanishes because of (1.2.59). Along a metric geodesic, the four-acceleration with respect to the Levi-Civita connection (defined by (1.6.121) with colon instead of semicolon) vanishes because of (1.4.91). The equation of geodesic deviation (1.3.52) determines the relative four-acceleration of two bodies moving along two infinitely close affine geodesics. Let us suppose that a noninertial frame K0 moves with velocity v relative to an inertial frame of reference K. If the velocity of K0 changes by dv0 relative to the initial frame K0, then it changes by dv relative to K. In the nonrelativistic limit, the two changes are equal, dv = dv0, and K0 does

58 not rotate with respect to K. In relativistic kinematics, these changes are diﬀerent because of the Thomas-Wigner rotation. The velocity of K0 relative to K after the change, v + dv, is equal to dv0 boosted by v. Using the Lorentz transformation (1.6.107), in which v0 is replaced with dv0 and V is replaced with v, we obtain

dv0 + γv + (γ − 1)(dv0 · v)v/v2 v + dv = , (1.6.130) γ(1 + dv0 · v/c2)

where γ = (1 − v2/c2)−1/2. Keeping only terms linear in dv0, we have

dv0 γ − 1 v v + dv = + v(1 − dv0 · v/c2) + (dv0 · v) , (1.6.131) γ γ v2 which gives v × dv0 v × dv = . (1.6.132) γ The angle of inﬁnitesimal rotation from the nonrelativistic sum v + dv0 to the relativistic v + dv determines the relativistic rotation of K0 with respect to K. Using (1.4.189) with sin(dθ) ≈ dθ and dθ = n dθ, where n is a unit vector parallel to the axis of rotation, gives

(v + dv0) × (v + dv) v × (dv − dv0) 1 − γ dθ = ≈ = (v × dv). (1.6.133) v2 v2 v2 Consequently, we obtain the angular velocity of the Thomas precession:

dθ γ − 1 γ2 a × v Ω = = (a × v) = , (1.6.134) dt v2 γ + 1 c2 where a = dv/dt is the acceleration of K0 relative to K. When v c, a × v Ω ≈ . (1.6.135) 2c2

References: [2, 3].

1.7 Spinors 1.7.1 Spinor representation of Lorentz group Let γa be the coordinate-invariant 4×4 Dirac matrices deﬁned as

a b b a ab γ γ + γ γ = 2η I4, (1.7.1) where I4 is the unit 4×4 matrix (4 is the lowest dimension for which (1.7.1) has solutions). Accord- i i a ingly, the spacetime-dependent Dirac matrices, γ = eaγ , satisfy

i j j i ij γ γ + γ γ = 2g I4. (1.7.2)

Under a tetrad rotation, (1.5.15) gives a a b γ˜ = Λ bγ . (1.7.3) Let L be a 4×4 matrix such that

a a b −1 a −1 γ = Λ bLγ L = Lγ˜ L , (1.7.4)

−1 −1 −1 where L is the matrix inverse to L: LL = L L = I4. The condition (1.7.4) represents the constancy of the Dirac matrices γa under the combined tetrad rotation and transformation γ → LγL−1. We refer to L as the spinor representation of the Lorentz group. The relation (1.7.4) gives

59 a the matrix L as a function of the Lorentz matrix Λ b. For an inﬁnitesimal Lorentz transformation (1.6.7), the solution for L is 1 1 L = I + Gab,L−1 = I − Gab, (1.7.5) 4 2 ab 4 2 ab where Gab are the generators of the spinor representation of the Lorentz group: 1 Gab = (γaγb − γbγa). (1.7.6) 4 A spinor ψ is deﬁned as a quantity that, under tetrad rotations, transforms according to

ψ˜ = Lψ. (1.7.7)

An adjoint spinor ψ¯ is deﬁned as a quantity that transforms according to

ψ¯˜ = ψL¯ −1, (1.7.8)

so the product ψψ¯ is a scalar: ψ¯˜ψ˜ = ψψ.¯ (1.7.9) The indices of the γa and L that are implicit in the 4×4 matrix multiplication in (1.7.1), (1.7.2) and (1.7.4) are spinor indices. The relation (1.7.4) shows that the Dirac matrices γa can be regarded as quantities that have, in addition to the invariant index a, one spinor index and one adjoint-spinor index. The product ψψ¯ transforms like the Dirac matrices:

ψ˜ψ¯˜ = LψψL¯ −1. (1.7.10)

The spinors ψ and ψ¯ can be used to construct tensors. For example, ψγ¯ aψ transforms like a contravariant Lorentz vector:

¯ a ¯ −1 a b −1 a ¯ b ψγ ψ → ψL Λ bLγ L Lψ = Λ bψγ ψ. (1.7.11)

1.7.2 Spinor connection The derivative of a spinor does not transform like a spinor: ˜ ψ,i = Lψ,i + L,iψ. (1.7.12)

If we introduce the spinor connection Γi that transforms according to

−1 −1 Γ˜i = LΓiL + L,iL , (1.7.13)

then a covariant derivative of a spinor,

ψ;i = ψ,i − Γiψ, (1.7.14)

is a spinor: ˜ ˜ ˜ −1 −1 ψ;i = ψ,i − Γ˜iψ = Lψ,i + L,iψ − (LΓiL + L,iL )Lψ = Lψ;i. (1.7.15) Because ψψ¯ is a scalar, ¯ ¯ (ψψ);i = (ψψ),i, (1.7.16) the chain rule for covariant diﬀerentiation gives the covariant derivative of an adjoint spinor, ¯ ¯ ¯ ψ;i = ψ,i + ψΓi. (1.7.17)

We also have ¯ ¯ ψ|i = ψ;i, ψ|i = ψ;i. (1.7.18)

60 The Dirac matrices γa transform like ψψ¯, whose covariant derivative is ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ (ψψ);i = ψ;iψ + ψψ;i = (ψψ),i − Γiψψ + ψψΓi = (ψψ),i − [Γi, ψψ]. (1.7.19) Therefore, the covariant derivative of a Dirac matrix is a a a a γ ;i = γ ,i − [Γi, γ ] = −[Γi, γ ], (1.7.20) which gives j j j j k j γ ;i = γ |i = γ ,i + Γk iγ − [Γi, γ ]. (1.7.21) Accordingly, we obtain a a b a γ |i = ω biγ − [Γi, γ ]. (1.7.22) ¯ i The quantity ψγ ψ|i transforms under Lorentz rotations like a scalar: ¯ i ¯ −1 i −1 ¯ i ψγ ψ|i → ψL Lγ L Lψ|i = ψγ ψ|i. (1.7.23)

The relation ηab|i = 0 infers that a γ |i = 0, (1.7.24) a because the Dirac matrices γ only depend on ηab. Multiplying both sides of (1.7.22) by γa from the left gives a b a ωabiγ γ − γaΓiγ + 4Γi = 0. (1.7.25) We seek the solution of (1.7.25) in the form 1 Γ = − ω γaγb − A , (1.7.26) i 4 abi i

where Ai is a spinor-tensor quantity with one vector index. Substituting (1.7.26) to (1.7.25), together a b c ab with the identity γcγ γ γ = 4η , gives a − γaAiγ + 4Ai = 0, (1.7.27)

so Ai is an arbitrary vector multiple of I4. Therefore, the spinor connection Γi is given, up to the addition of an arbitrary vector multiple of I4, by the Fock-Ivanenko coeﬃcients: 1 1 Γ = − ω γaγb = − ω Gab. (1.7.28) i 4 abi 2 abi Using the deﬁnition (1.5.22), we can also write (1.7.28) as 1 1 Γ = − ej [γ , γc] = [γj , γ ]. (1.7.29) i 8 c;i j 8 ;i j For the Levi-Civita connection, the covariant derivative of a spinor (1.7.14) becomes

{} ψ:i = ψ,i − Γi ψ, (1.7.30) and the covariant derivative of an adjoint spinor (1.7.17) becomes ¯ ¯ ¯ {} ψ:i = ψ,i + ψΓi , (1.7.31) {} where the Levi-Civita spinor connection Γi is given, similarly to (1.7.28), by 1 1 Γ{} = − $ γaγb = − $ Gab. (1.7.32) i 4 abi 2 abi Substituting (1.5.39) into (1.7.32) gives 1 1 Γ = Γ{} − C γjγk = Γ{} − C γ[jγk]. (1.7.33) i i 4 jki i 4 jki Accordingly, (1.7.30) and (1.7.31) yield 1 ψ = ψ + C γiγjψ, (1.7.34) ;k :k 4 ijk 1 ψ¯ = ψ¯ − C ψγ¯ iγj. (1.7.35) ;k :k 4 ijk

61 1.7.3 Curvature spinor The commutator of total covariant derivatives of a spinor is

k k ψ|ji − ψ|ij = (ψ|j),i − Γiψ|j − Γj iψ|k − (ψ|i),j + Γjψ|i + Γi jψ|k k k = −Γj,iψ + ΓiΓjψ + Γi,jψ − ΓjΓiψ + 2S ijψ|k = Kijψ + 2S ijψ|k, (1.7.36)

where Kij = −Kji is deﬁned as Kij = Γi,j − Γj,i + [Γi, Γj]. (1.7.37) Substituting (1.7.13) to (1.7.37) gives

−1 −1 K˜ij = Γ˜i,j − Γ˜j,i + [Γ˜i, Γ˜j] = L(Γi,j − Γj,i + [Γi, Γj])L = LKijL , (1.7.38)

a so Kij transforms under tetrad rotations like the Dirac matrices γ , that is, Kij is a spinor with one spinor index and one adjoint-spinor index. We refer to Kij as the curvature spinor. The relation (1.7.24) leads to k γ |i = 0. (1.7.39) Therefore, the commutator of covariant derivatives of the spacetime-dependent Dirac matrices vanishes: k k l l k k k l k 2γ |[ji] = R lijγ + 2S ijγ |l + [Kij, γ ] = R lijγ + [Kij, γ ] = 0. (1.7.40)

Multiplying both sides of (1.7.40) by γk from the left gives

k l k Rklijγ γ + γkKijγ − 4Kij = 0. (1.7.41)

We seek the solution of (1.7.41) in the form 1 K = R γkγl + B , (1.7.42) ij 4 klij ij

where Bij is a spinor-tensor quantity with two vector indices. Substituting (1.7.42) to (1.7.41) gives

k γkBijγ − 4Bij = 0, (1.7.43)

so Bij is an antisymmetric-tensor multiple of I4. The tensor Bij is related to the vector Ai in (1.7.26) by Bij = Aj,i − Ai,j + [Ai,Aj]. (1.7.44)

Because ψ has no indices other than spinor indices, Ai is a vector and [Ai,Aj] = 0. The invariance of (1.7.41) under the addition of an antisymmetric-tensor multiple Bij of the unit matrix to the curvature spinor is related to the invariance of (1.7.25) under the addition of a vector multiple Ai of the unit matrix to the spinor connection. Setting Ai = 0, which corresponds to the Fock-Ivanenko spinor connection, gives Bij = 0. Therefore, the curvature spinor Kij is given, up to the addition of an arbitrary antisymmetric-tensor multiple of I4, by 1 1 K = R γkγl = R Gkl. (1.7.45) ij 4 klij 2 klij

References: [3, 4].

Spacetime is a fabric in which various fields representing matter exist. These fields can be described by vectors, tensors, and spinors. They satisfy the equations derived from two fundamental principles: the principle of relativity and the principle of least action. The physics of fields is referred to as field theory and constitutes Chapter 2 (Fields).

62 2 Fields 2.1 Principle of least action The most general formulation of the law that governs the dynamics of classical systems is Hamilton’s principle of least action, according to which every classical system is characterized by a definite scalar-density function L, and the dynamics of the system is such that a certain condition is satisfied. i Let φA(x ) be a set of physical fields (indexed by A), being differentiable functions of the coordinates, and let L be a Lorentz covariant quantity constructed from the fields φA and their derivatives. Let us consider a scalar quantity 1 Z S = LdΩ, (2.1.1) c

where the integration is over some region in locally Minkowski spacetime. Let δφA be arbitrary and independent, small changes of φA (regarded as a dynamical variable) over the region of integration, which vanish on the boundary. Then the change in S can be written as X δS = δAS, (2.1.2) A where 1 Z δ S = F δφ dΩ. (2.1.3) A c A A The principle of least action states that the dynamics of a physical system is given by the condition the scalar S be a local minimum. Therefore, any infinitesimal change in the dynamics of the system does not alter the value of S: δS = 0 (2.1.4) (S is a local extremum). The condition (2.1.4) is referred to as the principle of stationary action, which is the necessary part of the principle of least action. If L is covariant and φA transform covariantly under the Lorentz group, the variational condition (2.1.4) gives the Lorentz covariant equations FA = 0. (2.1.5) These equations are also invariant for any other transformations (internal symmetries) for which L is invariant. L is referred to as the Lagrangian density, S is the action functional, δS = 0 is the principle of least action, and (2.1.5) are the field equations. The field equations of a physical system are the result of the action being a local extremum. The condition that the action be a local minimum imposes additional restrictions on possible choices for S. The number of independent field equations for a given system is referred to as the number of the degrees of freedom representing this system. In most physical cases L contains only φA and their first derivatives. A Lagrangian density containing higher derivatives can always be written in terms of first derivatives by increasing the number of the components φA. Let us consider a physical system in the Galilean frame of reference. If L depends only on φ and ∂iφ, L = L(φ, φ,i), then 1 Z ∂L ∂L 1 Z ∂L ∂L δS = δφ + δ(φ,i) dΩ = δφ + (δφ),i dΩ c ∂φ ∂(φ,i) c ∂φ ∂(φ,i) 1 Z ∂L ∂L ∂L = δφ − ∂i δφ + ∂i δφ dΩ. (2.1.6) c ∂φ ∂(φ,i) ∂(φ,i) The last term in the integrand in the second line of (2.1.6) is a divergence. Its four-volume integral can be transformed, using the Gauß-Stokes theorem (1.1.39), into a hypersurface integral over the boundary of the integration region. Since δφ = 0 on the boundary, this term does not contribute to the variation of the action: 1 Z ∂L ∂L Z ∂L 1 Z ∂L ∂L δS = − ∂i δφdΩ + δφdSi = − ∂i δφdΩ. (2.1.7) c ∂φ ∂(φ,i) ∂(φ,i) c ∂φ ∂(φ,i)

63 If δS = 0 for arbitrary variations δφ that vanish on the boundary, then

∂L ∂L − ∂i = 0. (2.1.8) ∂φ ∂(φ,i) Deﬁning the variational derivative of L with respect to φ,

δL ∂L ∂L = − ∂i , (2.1.9) δφ ∂φ ∂(φ,i)

we can write (2.1.8) as δL = 0. (2.1.10) δφ

The set of equations (2.1.8), for each field φA, is referred to as the Lagrange equations. There is some arbitrariness in the choice of L; adding to it the divergence of an arbitrary vector density or multiplying it by a constant produces the same field equations. If a system consists of two noninteracting parts A and B, with corresponding Lagrangian densitites LA(φA, ∂φA) and LB(φB, ∂φB), then the Lagrangian density for this system is the sum LA + LB. This additivity of the Lagrangian density means that the field equations for either of the two parts do not involve quantities pertaining to the other part. If LA also depends on φB and/or ∂φB, and/or LB depends on φA and/or ∂φA, then the subsystems A and B interact. References: [1, 2, 3].

2.2 Action for gravitational field Let us consider a Lagrangian density L that depends on the affine (or spin) connection and its first derivatives. Such Lagrangian density can be decomposed into the covariant part Lg that contains derivatives of the affine/spin connection, which is referred to as the Lagrangian density for the gravitational field, and the covariant part Lm that does not contain these derivatives, which is referred to as the Lagrangian density for matter:

L = Lg + Lm. (2.2.1)

The simplest covariant scalar that can be constructed from the affine/spin connection and its first derivatives is the Ricci scalar R. The corresponding Lagrangian√ density for the gravitational field is proportional to the product of R and the scalar density −g:

1 √ 1 √ L = − −gR = − −g P − gik(2Cl + Cj Cl − Cl Cm ) , (2.2.2) g 2κ 2κ il:k ij kl im kl where κ is Einstein’s gravitational constant and we used (1.4.67). The action for the gravitational field is thus 1 Z 1 Z √ S = L dΩ = − R −gdΩ. (2.2.3) g c g 2κc The metric tensor and the affine connection are two fundamental quantities describing a gravitational field. Since the affine connection is metric-compatible, given by (1.4.34), it is a function of the metric tensor, its derivatives and the torsion tensor. Accordingly, the metric and torsion tensors are dynamical variables in varying the action. Equivalently, the tetrad and spin connection can be taken as dynamical variables. Let us consider the Riemannian part of the Lagrangian density for the gravitational field (2.2.2), which is proportional to the Riemann scalar P : 1 √ L{} = − −gP. (2.2.4) g 2κ

64 √ i The scalar density −gP is linear in first derivatives of the Christoffel symbols {k l}: √ √ ik l l m l m l −gP = −gg ({i k},l − {i l},k + {i k }{m l} − {i l }{m k}) √ ik l l √ ik √ ik l l √ ik = ( −gg {i k}),l − {i k}( −gg ),l − ( −gg {i l}),k + {i l}( −gg ),k √ ik m l m l + −gg ({i k }{m l} − {i l }{m k}). (2.2.5) √ We can therefore subtract from −gP total derivatives without altering the field equations, replacing it by a noncovariant quantity G that does not contain first derivatives of the Christoffel symbols:

√ √ ik l √ ik l G = −gP − ( −gg {i k}),l + ( −gg {i l}),k l √ ik l √ ik √ ik m l m l = {i l}( −gg ),k − {i k}( −gg ),l + −gg ({i k }{m l} − {i l }{m k}) l √ ik j √ ik √ i jk √ k ij = {i l} ( −gg ):k + {j k} −gg − −g{j k}g − −g{j k}g l √ ik j √ ik √ i jk √ k ij −{i k} ( −gg ):l + {j l} −gg − −g{j l}g − −g{j l}g √ ik m l m l l j √ ik √ i jk + −gg ({i k }{m l} − {i l }{m k}) = {i l} {j k} −gg − −g{j k}g √ k ij l j √ ik √ i jk √ k ij − −g{j k}g − {i k} {j l} −gg − −g{j l}g − −g{j l}g √ ik m l m l √ ik m l m l + −gg ({i k }{m l} − {i l }{m k}) = −gg ({i l }{m k} − {i k }{m l}). (2.2.6) We also define G G = √ = gik({ m}{ l } − { m}{ l }). (2.2.7) −g i l m k i k m l The Riemannian part (2.2.4) of the Lagrangian density for the gravitational field reduces accordingly to 1 1 √ L{} = − G = − −gG. (2.2.8) g 2κ 2κ Any coordinate transformation results in variations of gik, thereby 1 Z 1 Z √ S{} = L{}dΩ = − P −gdΩ (2.2.9) g c g 2κc is not necessarily a minimum with respect to these variations (only an extremum) because not all δgik correspond to actual variations of the gravitational field. In order to exclude the variations δgik resulting from changing the coordinates, we must impose on the metric tensor 4 arbitrary constraints. If we choose g0α = 0, |gαβ| = const, (2.2.10) then G becomes 1 G = − g00gαβgγδg g . (2.2.11) 4 αγ,0 βδ,0 In the locally Galilean frame of reference, gαβ = −δαβ, thereby 1 G = − g00(g )2. (2.2.12) 4 αβ,0

00 {} For physical systems, g > 0. Therefore, in order for Sg to have a minimum, κ must be positive, {} otherwise an arbitrarily rapid change of gαβ in time would result in an arbitrarily low value of Sg and there would be no minimum of S. References: [2, 3].

2.3 Matter 2.3.1 Metric energy-momentum tensor The variation of the action for matter, 1 Z S = L dΩ, (2.3.1) m c m

65 with respect to the metric tensor: 1 Z 1 Z δS = T δgijdΩ = − T ijδg dΩ, (2.3.2) m 2c ij 2c ij

deﬁnes the metric energy-momentum density Tij. This tensor density is symmetric:

Tij = Tji. (2.3.3) Equivalently, we have δL ∂L ∂L T = 2 m = 2 m − 2∂ m . (2.3.4) ij δgij ∂gij k ij ∂(g ,k) The metric energy-momentum tensor is deﬁned as T T = √ ij . (2.3.5) ij −g

2.3.2 Tetrad energy-momentum tensor The variation of the matter action (2.3.1) with respect to the tetrad: 1 Z δS = T aδei dΩ, (2.3.6) m c i a a deﬁnes the tetrad energy-momentum density Ti . Equivalently a i δLm = Ti δea (2.3.7) or a δLm Ti = i . (2.3.8) δea The corresponding tensor density with two coordinate indices is a Tij = eajTi . (2.3.9) The tetrad energy-momentum tensor is deﬁned as T t = ij . (2.3.10) ij e This tensor is generally not symmetric.

2.3.3 Canonical energy-momentum density

A matter Lagrangian density Lm can be written as Lm = eL, where L is a scalar. If L depends on matter fields φ and their covariant derivatives φ|i, then such fields are said to be minimally coupled to the affine connection. If these fields are written in terms of Lorentz indices instead of vector i indices, then the tetrad appears in L only through a covariant combination eaφ|i. Varying L with respect to the tetrad gives, using (1.5.13), a i ∂L i a i ∂Lm a i δLm = eδL − eei Lδea = e φ|iδea − Lmei δea = φ|i − ei Lm δea. (2.3.11) ∂φ|a ∂φ|a The last term in (2.3.11), a ∂Lm a Θi = φ|i − ei Lm, (2.3.12) ∂φ|a is referred to as the canonical energy-momentum density. The corresponding tensor density with two coordinate indices is

i ∂Lm i ∂Lm i Θj = φ|j − δjLm = φ|j − δjLm. (2.3.13) ∂φ|i ∂φ,i Comparing (2.3.11) with (2.3.7) shows that the canonical energy-momentum density is identical with the tetrad energy-momentum density: a a Θi = Ti . (2.3.14)

66 2.3.4 Spin tensor The variation of the matter action (2.3.1) with respect to the spin connection, 1 Z δS = S iδωab dΩ, (2.3.15) m 2c ab i

i deﬁnes the spin density Sab : i δLm ∂Lm Sab = 2 ab = 2 ab , (2.3.16) δω i ∂ω i which is antisymmetric in the Lorentz indices:

i i Sab = −Sba . (2.3.17)

The second equality in (2.3.16) is satisfied because a matter Lagrangian density Lm may depend on ab the spin connection but not on its derivatives; a scalar density depending on derivatives of ω i is a ab i Lagrangian density for the gravitational field. The variations δω i are independent of δea, thereby the spin density is independent of the energy-momentum density. The relation (1.5.35) indicates that the spin density with three coordinate indices, which is antisymmetric in the first two indices, is generated by the contortion tensor:

k k δLm Sij = −Sji = 2 ij . (2.3.18) δC k

Accordingly, the variation of Lm with respect to the torsion tensor,

jk δLm τi = 2 i , (2.3.19) δS jk is a homogeneous linear function of the spin connection because of (1.4.35):

δL δL ∂Clmn τ = 2 m = 2 m = S (δlδmδn + δmδn δl + δnδmδl ) ijk δSijk δClmn ∂Sijk lmn i [j k] i [j k] i [j k] = Sijk − Sjki + Skij, (2.3.20)

Sijk = τ[ij]k, (2.3.21) antisymmetric in the last two indices: τijk = −τikj. (2.3.22)

The variation of Lm with respect to the metric-compatible aﬃne connection in the metric-aﬃne variational formulation of gravity is equivalent to the variation with respect to the torsion (or contortion) tensor. ab The spin connection ω i appears in Lm only through covariant derivatives of φ, in a combination ∂L − Γiφ, where ∂φ,i 1 Γ = − ω Gab (2.3.23) i 2 abi is the connection in the covariant derivative of φ:

φ|i = φ,i − Γiφ. (2.3.24)

i Consequently, the spin density Sab is identical with

i i ∂Lm Σab = −Σba = Gabφ, (2.3.25) ∂φ,i referred to as the canonical spin density. The spin tensor is deﬁned as S s = ijk . (2.3.26) ijk e

67 2.3.5 Belinfante-Rosenfeld relation The total variation of the matter action with respect to geometrical variables is either 1 Z 1 Z δS = dΩT aδei + dΩS iδωab (2.3.27) m c i a 2c ab i or 1 Z 1 Z δS = dΩT δgik + dΩτ ikδSj . (2.3.28) m 2c ik 2c j ik The relation (1.5.5) gives 1 Z 1 Z Z dΩT δgik = dΩT (δei ek + ei δek)ηab) = dΩT ekaδei , (2.3.29) 2 ik 2 ik a b a b ik a and (1.5.31) gives 1 Z 1 Z dΩτ ikδSj = dΩτ ik δ(ej e ωab ) + δea ej + ea δej 2 j ik 2 j a ib k i,k a i,k a 1 Z = dΩ τ liδ(ej e )ωab + τ iδωab + (τ ikej δea) − (τ ikej ) δea + τ ikea δej 2 j a lb i ab i j a i ,k j a ,k i j i,k a 1 Z = dΩ τ lkωcb e δej + τ liωab ej δe + τ iδωab − (τ ikej ) δea + τ lmeb δej 2 j k lb c j i a lb ab i j a ,k i j l,m b 1 Z 1 Z + dS τ ikej δea = dΩ −τ lkωcb e ei ej δea + τ ilωb ejδea + τ iδωab 2 k j a i 2 j k lb c a i j al b i ab i ik i jk ij b ik a lm b i j a −(τa |k − S jkτa − 2Sjτa + ω akτb )δei − τj el,mebeaδei 1 Z = dΩ τ iδωab − τ ik ej δea + 2S τ ijδea . (2.3.30) 2 ab i j ;k a i j a i Comparing (2.3.27) with (2.3.28) leads to Z 1 Z 1 Z 1 Z dΩT aδei + dΩS iδωab = dΩT δgik + dΩ τ iδωab − τ ik ej δea i a 2 ab i 2 ik 2 ab i j ;k a i Z 1 Z 1 Z +2S τ ijδea = dΩT ekaδei + dΩτ iδωab + dΩτ jk eaδei j a i ik a 2 ab i 2 i ;k j a Z kj a b i − dΩSjτb ekei δea. (2.3.31)

ab i The terms in (2.3.31) with δω i give (2.3.21), while the terms with δea give 1 T a = T eka + τ jk ea − S τ aj (2.3.32) i ik 2 i ;k j j i or 1 T = T − ∇ (S j − S j + Sj ) + S (S j − S j + Sj ). (2.3.33) ik ik 2 j ik k i ik j ik k i ik Equation (2.3.33) is referred to as the Belinfante-Rosenfeld relation between the metric and tetrad energy-momentum densites. The Belinfante-Rosenfeld relation can be written, after dividing by e, as a tensor equation: 1 T = t − ∇∗(s j − s j + sj ), (2.3.34) ik ik 2 j ik k i ik ∗ where ∇j is given by (1.2.45). In the absence of spin, (2.3.33) and (2.3.34) reduce to

Tik = Tik, tik = Tik. (2.3.35)

References: [2, 3, 4, 7, 9]

68 2.4 Symmetries and conservation laws 2.4.1 Noether theorem Let us consider a physical system, described by a Lagrangian density L that depends on matter i fields φ, their first derivatives φ,i, and the coordinates x . The change of the Lagrangian density δL under an infinitesimal coordinate transformation (1.2.66) is thus ¯ ∂L ∂L ∂L i δL = δφ + δ(φ,i) + i ξ , (2.4.1) ∂φ ∂φ,i ∂x ¯ where the changes δφ and δ(φ,i) are brought about by the transformation (1.2.66) and ∂ denotes i partial differentiation with respect to x at constant φ and φ,i. The variation δL brought about by this transformation is also given by (1.2.75):

i δL = −ξ ,iL. (2.4.2) Using the Lagrange equations (2.1.8) and the identities

∂¯L ∂L ∂L L,i = i + φ,i + φ,ji, (2.4.3) ∂x ∂φ ∂φ,j j δ(φ,i) = (δφ),i − ξ ,iφ,j, (2.4.4) we bring (2.4.1) to i ∂L j δL = ξ L,i + (δφ − ξ φ,j) . (2.4.5) ∂φ,i ,i Combining (2.4.2) and (2.4.5) gives the conservation law,

i J ,i = 0, (2.4.6) for the Noether current:

i i ∂L j i ∂L ¯ J = ξ L + (δφ − ξ φ,j) = ξ L + δφ. (2.4.7) ∂φ,i ∂φ,i Equation (2.4.6) represents the Noether theorem, which states that to each continuous symmetry of a Lagrangian density there corresponds a conservation law.

2.4.2 Conservation of spin The Lorentz group is the group of tetrad rotations. Since a physical matter Lagrangian density Lm(φ, φ,i) is invariant under local, proper Lorentz transformations, it is invariant under tetrad rotations: ∂Lm ∂Lm a i 1 i ab δLm = δφ + δ(φ,i) + Ti δea + Sab δω i = 0, (2.4.8) ∂φ ∂φ,i 2 where the changes δ are brought about by a tetrad rotation. Under integration of (2.4.8) over spacetime, the ﬁrst two terms vanish because of the Lagrange equations for φ (2.1.8): Z 1 T aδei + S iδωab dΩ = 0. (2.4.9) i a 2 ab i

a For an inﬁnitesimal Lorentz transformation (1.6.7), the tetrad ei changes by

a a a a b a a δei =e ˜i − ei = Λ bei − ei = i, (2.4.10)

i a j and the tetrad ea, because of the identity δ(ei ea) = 0, according to

i i δea = − a. (2.4.11)

69 The spin connection changes by

ab a jb a jb a jb a cb a jb a bc ab δω i = δ(ej ω i) = jω i − ej ;i = cω i − ej |i + cω i = − |i. (2.4.12)

Substituting (2.4.11) and (2.4.12) to (2.4.9), together with partial integration (1.2.43), gives

Z 1 Z 1 − T ai + S iab dΩ = − T ij + S kij dΩ i a 2 ab |i ij 2 ij |k Z 1 = −T − S S k + S k ijdΩ = 0. (2.4.13) [ij] k ij 2 ij ;k

Since the inﬁnitesimal Lorentz rotation ij is arbitrary, we obtain the covariant conservation law for the spin density (6 equations):

k k Sij ;k = Tij − Tji + 2SkSij . (2.4.14) Dividing this law by e gives the conservation law for the spin tensor:

∗ k ∇ksij = tij − tji. (2.4.15) The conservation law (2.4.15) also results from antisymmetrizing the Belinfante-Rosenfeld relation k (2.3.34) with respect to the indices i, k. If we use the metric-compatible aﬃne connection Γi j, which ab is invariant under tetrad rotations, instead of the spin connection ω i as a variable in Lm, then we ab i must replace the term with δω i in (2.4.8) by a term with δ(ea,j).

2.4.3 Conservation of metric energy-momentum The metric and torsion tensors can be taken, instead of the tetrad and spin connection, as the dynamical variables describing spacetime. Under an inﬁnitesimal coordinate transformation (1.2.66), the matter Lagrangian density Lm(φ, φ,i) changes according to

∂Lm ∂Lm ∂Lm ik ∂Lm ik δLm = δφ + δ(φ,i) + ik δg + ik δ(g ,l) ∂φ ∂φ,i ∂g ∂g ,l

∂Lm j ∂Lm j + j δS ik + j δ(S ik,l). (2.4.16) ∂S ik ∂S ik,l

1 R The matter action Sm = c Lm(φ, φ,i)dΩ is a scalar, thereby it does not change under this transformation: Z 1 ∂Lm ∂Lm ∂Lm ik ∂Lm ik δSm = δφ + δ(φ,i) + ik δg + ik δ(g ,l) c ∂φ ∂φ,i ∂g ∂g ,l ∂Lm j ∂Lm j + j δS ik + j δ(S ik,l) dΩ = 0. (2.4.17) ∂S ik ∂S ik,l

The ﬁrst two terms in (2.4.17) vanish because of the Lagrange equations for φ (2.1.8). If the ik j variations δg and S ik vanish on the boundary of the region of integration, then

1 Z ∂L ∂L 1 Z ∂L ∂L δS = m − ∂ m δgikdΩ + m − ∂ m δSj dΩ m c ∂gik l ∂gik c j l j ik ,l ∂S ik ∂S ik,l 1 Z δL 1 Z δL 1 Z 1 Z = m δgikdΩ + m δSj dΩ = T δgikdΩ + τ ikδSj dΩ c δgik c j ik 2c ik 2c j ik δS ik 1 Z 1 Z = − T ikδg dΩ + τ ikδSj dΩ = 0. (2.4.18) 2c ik 2c j ik

70 The components of the metric tensor change because of an inﬁnitesimal coordinate transformation (1.2.66), thereby the corresponding variation of the metric tensor is given by (1.4.51): ¯ δgik = δgik = −Lξgik = −2ξ(i:k), (2.4.19)

and the variation of the torsion tensor is given by a Lie derivative,

j ¯ j j j l l j l j l j δS ik = δS ik = −LξS ik = ξ ,lS ik − ξ ,iS lk − ξ ,kS il − ξ S ik,l. (2.4.20) The variation of the matter action under (1.2.66) is therefore equal to 1 Z 1 Z δS = δS¯ = − T ikδg¯ dΩ + τ ikδS¯ j dΩ = 0. (2.4.21) m m 2c ik 2c j ik The ﬁrst term on the right of (2.4.21) is 1 Z 1 Z 1 Z 1 Z − T ikδg¯ dΩ = T ikξ dΩ = (T ikξ ) dΩ − T ik ξ dΩ 2c ik c i:k c i :k c :k i 1 Z 1 Z 1 Z 1 Z = (T ikξ ) dΩ − T ik ξ dΩ = T ikξ dS − T k ξldΩ. (2.4.22) c i ,k c :k i c i k c l :k The second term on the right of (2.4.21) is 1 Z 1 Z τ ikδS¯ j dΩ = (τ ikξjSl ) − (τ ikξlSj ) − (τ ikξlSj ) dΩ 2c j ik 2c j ik ,l j lk ,i j il ,k 1 Z + −(τ ikSl ) ξj + (τ ikSj ) ξl + (τ ikSj ) ξl − τ ikSj ξldΩ 2c j ik ,l j lk ,i j il ,k j ik,l 1 Z 1 Z 1 Z = τ ikξjSl dS − τ ikξlSj dS − τ ikξlSj dS 2c j ik l 2c j lk i 2c j il k 1 Z + −(2τ ikSj ) − (τ ijSk ) − τ ikSj ξldΩ. (2.4.23) 2c j li ,k l ij ,k j ik,l

If the variations ξi of the coordinates vanish on the boundary of the region of integration, then (2.4.21) becomes 1 Z δS = − 2T k + (2τ ikSj + τ ijSk ) + τ ikSj ξldΩ = 0. (2.4.24) m 2c l :k j li l ij ,k j ik,l

Since the variations ξi are arbitrary, (2.4.24) gives the covariant conservation law for the metric energy-momentum density (4 equations):

k ik j 1 ij k 1 ik j Tl :k + τj S li + τl S ij + τj S ik,l = 0. (2.4.25) 2 ,k 2 √ Dividing the conservation law (2.4.25) by −g gives √ −g k ik j 1 ij k ik j 1 ij k ,k 1 ik j Tl :k + tj S li + tl S ij + tj S li + tl S ij √ + tj S ik,l 2 ,k 2 −g 2

k ik j 1 ij k im j k ik j m 1 ij k m = Tl :k + tj S li + tl S ij − tj S li{m k} + tj S mi{l k } + tm S ij{l k } 2 :k 2 1 1 1 1 − t ijSm { k } + t ikSj + t ijSk { m } + t ikSj − t ikSm { j } 2 l i j m k j li 2 l ij m k 2 j ik:l 2 j ik m l 1 1 + t ikSj { m} + t ikSj { m} = 0, (2.4.26) 2 j mk i l 2 j im k l where τ t = √ijk . (2.4.27) ijk −g

71 We therefore obtain the conservation law for the metric energy-momentum tensor:

k ik j 1 ij k 1 ik j Tl + tj S li + tl S ij + tj S ik:l = 0. (2.4.28) 2 :k 2

If the matter Lagrangian density does not depend on the torsion tensor, then tijk = 0 and (2.4.25) reduces to ik T :k = 0. (2.4.29) Equivalently, (2.4.28) reduces to ik T :k = 0. (2.4.30) R ij ¯ ij ¯ In this case, vanishing of T δgijdΩ in (2.4.21) does not imply T = 0, because 10 variations δgij are not all independent; they are functions of 4 independent variations ξi.

2.4.4 Conservation of tetrad energy-momentum

The matter action Sm is invariant under inﬁnitesimal translations of the coordinate system (1.2.66). The corresponding changes of the tetrad and spin connection are given by Lie derivatives, ¯ i i i j j i δea = −Lξea = ξ ,jea − ξ ea,j, (2.4.31) ¯ ab ab j ab j ab δω i = −Lξω i = −ξ ,iω j − ξ ω i,j. (2.4.32) Equation (2.4.9) becomes Z 1 T aδe¯ i + S iδω¯ ab dΩ = 0. (2.4.33) i a 2 ab i If the variations ξi of the coordinates vanish on the boundary of the region of integration, then substituting (2.4.31) and (2.4.32) into (2.4.33) gives Z 1 1 T aξi ej − T aξjei − S iξj ωab − S iξjωab dΩ i ,j a i a,j 2 ab ,i j 2 ab i,j Z 1 1 = −T j − T aej + (S jωab ) − S jωab ξidΩ = 0. (2.4.34) i ,j j a,i 2 ab i ,j 2 ab j,i This equation is satisﬁed for an arbitrary vector ξi, thereby we obtain j ab j ab ab j a j Sab ,jω i + Sab (ω i,j − ω j,i) − 2Ti ,j − 2Tj ea,i j k j c j c ab j a j = (Sab |j − 2SkSab + Scb ω aj + Sac ω bj)ω i − 2Ti ,j − 2Tj ea,i j ab a cb a cb +Sab (−R ij + ω ciω j − ω cjω i) = 0, (2.4.35) which reduces to j k ab ab j j j jk jk (Sab |j − 2SkSab )ω i − R ijSab − 2Ti ;j + 4SjTi − 2Tjkω i + 4S iTjk k k jl kl j j j jk jk = (Sjl ;k − 2SkSjl )ω i − R ijSkl − 2Ti ;j + 4SjTi − 2Tjkω i + 4S iTjk = 0. (2.4.36) The conservation law for the spin density (2.4.14) brings (2.4.36) to the covariant conservation law for the tetrad energy-momentum density: 1 T j = 2S T j + 2Sj T k + S jRkl , (2.4.37) i ;j j i ki j 2 kl ji which is equivalent to 1 Tij = C iTjk + S Rklji. (2.4.38) :j jk 2 klj This law can be written as the conservation law for the tetrad energy-momentum tensor: 1 tij = C itjk + s Rklji. (2.4.39) :j jk 2 klj Equations (2.4.37) and (2.4.39) are equivalent to (2.4.25) and (2.4.28).

72 2.4.5 Conservation laws for Lorentz group Let us consider a physical system in which the gravitational field (torsion and curvature) can be neglected, described by a matter Lagrangian density Lm. The Lagrangian density Lm therefore depends on the coordinates only through matter fields φ and their first derivatives φ,i. Differentiating Lm gives, using the Lagrange equations (2.1.8), ∂Lm ∂Lm ∂Lm ∂Lm ∂Lm ∂iLm = φ,i + φ,ji = ∂j φ,i + φ,ji = ∂j φ,i . (2.4.40) ∂φ ∂φ,j ∂φ,j ∂φ,j ∂φ,j This equation can be written as a conservation law:

j θi ,j = 0, (2.4.41) for a quantity j ∂Lm j θi = φ,i − δi Lm. (2.4.42) ∂φ,j The conservation law (2.4.41) is a special case of (2.4.37) in the absence of torsion and curvature, expressed in the Galilean and geodesic frame. The quantity (2.4.42) is a special case of the canonical energy-momentum density (2.3.13) in the absence of torsion, expressed in the Galilean and geodesic frame. The Noether current (2.4.7) can be written as

i ∂Lm i j J = δφ − θj ξ . (2.4.43) ∂φ,i

If xi are Cartesian coordinates then for translations, ξi = i = const and δφ = 0, the current (2.4.7) is i i ∂Lm j J = Lm − φ,j. (2.4.44) ∂φ,i The conservation law (2.4.6) for this current is

j i θj ,i = 0, (2.4.45)

i i i j 1 ij which gives (2.4.41) because are arbitrary. For Lorentz rotations, ξ = jx and δφ = 2 ijG φ, where Gij are the generators of the Lorentz group, the Noether current (2.4.7) is i ij ∂Lm 1 kl jk kl ∂Lm i 1 ∂Lm J = xjLm + Gklφ − xkφ,j = xk φ,l − xkδl Lm + Gklφ . (2.4.46) ∂φ,i 2 ∂φ,i 2 ∂φ,i

The conservation law (2.4.6) for this current is kl ∂Lm i 1 ∂Lm φ,[lxk] − δ[lxk]Lm + Gklφ = 0. (2.4.47) ∂φ,i 2 ∂φ,i ,i

Because kl are arbitrary, this equation gives the conservation law,

i Mkl ,i = 0, (2.4.48)

for the angular momentum density:

i i i ∂Lm i i i Mkl = xkθl − xlθk + Gklφ = xkθl − xlθk + Σkl . (2.4.49) ∂φ,i The angular momentum density is antisymmetric in the ﬁrst two indices:

Mijk = −Mjik. (2.4.50)

73 This quantity is the sum, k k k Mij = Λij + Σij , (2.4.51) of two tensor densities: the orbital angular momentum density,

i i i Λkl = xkθl − xlθk , (2.4.52) and the intrinsic angular momentum density (canonical spin density) (2.3.25). The conservation law (2.4.48) gives

kli k li k li l ki l ki kli M ,i = δi θ + x θ ,i − δiθ − x θ ,i + Σ ,i = 0, (2.4.53) which reduces, by means of (2.4.41), to

i θkl − θlk − Σkl ,i = 0. (2.4.54) This equation is a special case of the conservation law for the spin density (2.4.14) in the absence of torsion, expressed in the Galilean and geodesic frame. The canonical energy-momentum density θik is not symmetric. However, the quantity j τik = θik + ∂jψik , (2.4.55) where 1 ψ j = − (Σ j − Σ j + Σj ), (2.4.56) ik 2 ik k i ik is symmetric:

j j j j j τik − τki = θik − θki + ∂j(ψik − ψki ) = Σik ,j + ∂j(ψik − ψki ) = 0. (2.4.57) Since (2.4.56) is antisymmetric in the last two indices,

ψikj = −ψijk, (2.4.58) the quantity (2.4.55) is also conserved:

ik ik ikj ik τ ,k = θ ,k + ψ ,jk = θ ,k = 0. (2.4.59)

The symmetric energy-momentum density τik is equal to the metric energy-momentum density (2.3.4), expressed in the Galilean and geodesic frame. Equation (2.4.55) is a special case of the Belinfante-Rosenfeld relation (2.3.34) in the absence of torsion, expressed in the Galilean and geodesic frame.

2.4.6 Momentum four-vector Integrating the conservation law (2.4.41) for the canonical energy-momentum density (2.4.42), which is satisﬁed if we neglect torsion and curvature, over a four-volume and using the Gauß-Stokes theorem (1.1.39) gives Z I ik ik θ ,kdΩ = θ dSk = 0, (2.4.60) where the integral on the right is taken over the closed hypersurface surrounding the four-volume. 0 If the hypersurface represented by the element dSk is taken as a hyperplane perpendicular to the x 0 axis (volume hypersurface), dSk = δkdV , then the closed hypersurface surrounds the four-volume between two hyperplanes at times t1 and t2:

I Z t Z t ik ik 2 i0 2 θ dSk = θ dSk = θ dV = 0. (2.4.61) t1 t1 We deﬁne the momentum four-vector or four-momentum of matter in the four-volume as 1 Z 1 Z 1 Z 1 Z P i = TikdS = Ti0dV = ΘikdS = Θi0dV, (2.4.62) c k c c k c

74 where we used (2.3.14). In the locally Galilean and geodesic frame of reference, the four-momentum (2.4.62) reduces to 1 Z 1 Z P i = θikdS = θi0dV, (2.4.63) c k c which is locally conserved because of (2.4.61):

i i i P |t1 = P |t2 ,P = const. (2.4.64)

1 i0 00 The components c θ form the four-momentum density. The component θ is referred to as the energy density, 00 ˙ ∂Lm W = θ = φ − Lm. (2.4.65) ∂φ˙ Integrating it over the volume gives the time component P 0 of the four-momentum, Z ∂L cP 0 = θ00dV = φ˙ − L, (2.4.66) ∂φ˙ where Z L = LmdV (2.4.67)

is the Lagrange function or Lagrangian. The covariant time component of cP i is referred to as the energy, E = cP0. (2.4.68) ˙ dφ Hereinafter, a dot above any quantity φ denotes the derivative of φ with respect to time, φ = dt , ¨ d2φ and two dots above φ denote second derivative of φ with respect to time, φ = dt2 . Consequently, the action of a physical system is equal to the time integral of the Lagrangian, Z S = Ldt. (2.4.69)

The components 1 Πα = θα0 (2.4.70) c form the momentum density Π. Integrating them over the volume gives the spatial components P α of the four-momentum, 1 Z P α = θα0dV, (2.4.71) c which form the momentum vector P:

P i = (P 0,P α) = (P 0, P). (2.4.72)

The conservation law (2.4.41) can be written as

1 ∂θ00 ∂θ0α + = 0, (2.4.73) c ∂t ∂xα 1 ∂θα0 ∂θαβ + = 0. (2.4.74) c ∂t ∂xβ Integrating these equations over a volume and using the Gauß’ theorem (1.4.172) gives

∂ Z I θ00dV = −c θ0αdf , (2.4.75) ∂t α ∂ Z 1 I θα0dV = − θαβdf . (2.4.76) ∂t c β

75 The integral of a three-dimensional vector V α over a two-dimensional surface represented by the H α element dfα, V dfα, is referred to as the ﬂux of this vector. Integrating the components of the energy current S, Sα = cθ0α, (2.4.77)

over dfα gives the energy ﬂux or power: I dE P = Sαdf = − . (2.4.78) α dt

The last equality results from (2.4.66), (2.4.68) and (2.4.75). Integrating the components θαβ, which represent the momentum current, over dfα gives the momentum ﬂux. The stress tensor is deﬁned as

σαβ = −θαβ. (2.4.79)

Its integral taken over dfα gives the vector opposite to the momentum ﬂux, called the surface force F : s I α αβ Fs = σ dfβ. (2.4.80)

The relations (2.4.71), (2.4.76), (2.4.79) and (2.4.80) equal the time derivative of the momentum P α α to the surface force Fs : ˙ α α P = Fs . (2.4.81) The components of the energy-momentum tensor form the following matrix:

W S θik = c . (2.4.82) cΠ −σαβ

ikj ik Adding ψ ,j, where (2.4.58) is satisﬁed, to θ preserves the conservation law (2.4.41) and brings θik to a symmetric form, τ ik. Using the Gauß-Stokes theorem (1.1.38) gives Z Z Z 1 Z 1 I τ ikdS = θikdS + ψikj dS = cP i + (ψikj dS − ψikj dS ) = cP i + ψikjdf ∗ , k k ,j k 2 ,j k ,k j 2 kj (2.4.83) where the last integral is taken over the surface which bounds the hypersurface. If this surface is located in a region, where matter is absent, then the surface integral vanishes. Consequently, replacing θik by τ ik does not change the four-momentum (2.4.63): 1 Z 1 Z P i = τ ikdS = τ i0dV. (2.4.84) c k c

2.4.7 Mass We define the mass of matter in a four-volume as (P iP )1/2 m = i , (2.4.85) c where P i is the four-momentum of the matter (2.4.62). This definition is analogous to the mass operator (1.6.81). If the matter satisfies the dominant energy condition (confer (2.5.10)), then i i i P Pi ≥ 0 and the mass is a real quantity. If P Pi > 0, then m > 0. If P Pi = 0, then m = 0.

2.4.8 Angular momentum four-tensor Integrating the conservation law (2.4.48) for the angular momentum density (2.4.49), which is sat- isﬁed if we neglect torsion and curvature, over a four-volume and using the Gauß-Stokes theorem (1.1.39) gives Z I ikj ikj M ,jdΩ = M dSj = 0, (2.4.86)

76 where the integral on the right is taken over the closed hypersurface surrounding the four-volume. If the hypersurface represented by the element dSj is taken as a volume hyperplane, then the closed hypersurface surrounds the four-volume between two hyperplanes at times t1 and t2:

I Z t Z t ikj ikj 2 ik0 2 M dSj = M dSj = M dV = 0. (2.4.87) t1 t1 The angular momentum four-tensor of matter in the four-volume, 1 Z 1 Z M ik = MikjdS = Mik0dV, (2.4.88) c j c is therefore locally conserved:

ik ik ik M |t1 = M |t2 ,M = const. (2.4.89) The angular momentum four-tensor, following (2.4.50), is antisymmetric:

M ik = −M ki. (2.4.90)

1 αβ0 The components c M form the spatial angular momentum density. Integrating them over the volume gives the components of the spatial angular momentum tensor, 1 Z M αβ = Mαβ0dV. (2.4.91) c Since the angular momentum tensor is antisymmetric,

M αβ = −M βα, (2.4.92) we can deﬁne the angular momentum pseudovector M: 1 M α = eαβγ M ,M = e M γ . (2.4.93) 2 βγ αβ αβγ Integrating (2.4.51) over a hypersurface gives

M ik = Lik + Sik, (2.4.94)

where 1 Z 1 Z 1 Z 1 Z Lik = ΛikjdS = (xiθkj − xkθij)dS = Λik0dV = (xiθk0 − xkθi0)dV (2.4.95) c j c j c c is the orbital angular momentum four-tensor and 1 Z 1 Z Sik = ΣikjdS = Σik0dV (2.4.96) c j c

is the intrinsic angular momentum four-tensor. Unlike M ik, these tensors are not separately conserved. These tensors are also antisymmetric:

Lik = −Lki,Sik = −Ski. (2.4.97)

Similarly to (2.4.93), we can deﬁne the orbital angular momentum pseudovector L and the intrinsic angular momentum pseudovector S: 1 Lα = eαβγ L ,L = e Lγ , (2.4.98) 2 βγ αβ αβγ 1 Sα = eαβγ S ,S = e Sγ , (2.4.99) 2 βγ αβ αβγ M α = Lα + Sα, M = L + S. (2.4.100)

77 The symmetry of τ ik can be written, using (2.4.59), as

ki ik i kl k il τ − τ = ∂l(x τ − x τ ) = 0. (2.4.101)

Integrating this equation over a four-volume and using the Gauß-Stokes theorem (1.1.39) gives

I Z t Z t i kl k il i kl k il 2 i k0 k i0 2 (x τ − x τ )dSl = (x τ − x τ )dSl = (x τ − x τ )dV = 0, (2.4.102) t1 t1 which shows the conservation of the quantity 1 Z 1 Z L˜ik = (xiτ kl − xkτ il)dS = (xiτ k0 − xkτ i0)dV = const. (2.4.103) c l c The symmetry of a second-rank tensor whose ordinary divergence is zero is therefore related to the local conservation of the orbital angular momentum four-tensor constructed from that tensor. Using (2.3.25), (2.4.55), and the Gauß-Stokes theorem (1.1.38) leads to Z Z Z ˜ik ikl i klj k ilj ik i klj k ilj cL = Λ dSl + (x ψ ,j − x τ ,j)dSl = cL + (x ψ ),j − (x ψ ),j dSl Z 1 − (ψkli − ψilk)dS = cLik + (xiψklj) dS − (xiψklj) dS − (xkψilj) dS + (xkψilj) dS l 2 ,j l ,l j ,j l ,l j Z 1 I + ΣikjdS = cM ik + (xiψklj − xkψilj)df ∗ . (2.4.104) j 2 lj If the integration surface is located in a region, where matter is absent, then the surface integral vanishes. Consequently, the quantity (2.4.103) is equal to the angular momentum four-tensor (2.4.88) if we neglect torsion and curvature. This equality shows that replacing θik in (2.4.52) by τ ik changes the values of (2.4.95) and thereby (2.4.88). The angular momentum four-tensor can also be written, using (2.4.63), in terms of P i: Z M ik = (xidP k − xkdP i) + Sik. (2.4.105)

In the absence of the intrinsic angular momentum, (2.4.105) reduces to Z M ik = (xidP k − xkdP i). (2.4.106)

The conservation of M 0α,

1Z Z 1 Z M α0 = xαθ00dV − x0 θα0dV + Sα0 = xαθ00dV − ctP α + Sα0 = const, (2.4.107) c c divided by the conserved P 0 gives

Sα0 Xα = V αt + + const, (2.4.108) P 0 where cP α V α = (2.4.109) P 0 and R xαθ00dV Xα = . (2.4.110) R θ00dV If the intrinsic angular momentum is constant, then the relation (2.4.108) describes a uniform motion of the center of inertia, whose coordinates are Xα, with velocity V α. The coordinates of the center of inertia (2.4.110) are not the spatial components of a four-dimensional vector.

78 2.4.9 Energy-momentum tensor for particles Let us consider matter which is distributed over a small region in space and consists of points with the coordinates xi, forming an extended body whose motion is represented by a world tube in spacetime. The motion of the body as a whole is represented by an arbitrary timelike world line γ inside the world tube, which consists of points with the coordinates Xi(τ), where τ is the proper time on γ. We deﬁne

δxi = xi − Xi, (2.4.111) dXi ui = , (2.4.112) ds

2 i j where ds = gijdX dX . The conservation law for the tetrad energy-momentum density (2.4.39) is 1 Tji + { j }Tik − C jTik − R jSikl = 0. (2.4.113) ,i i k ik 2 ikl Integrating (2.4.113) over the volume of the body at a constant time X0 and using Gauß’ theorem to eliminate surface integrals gives Z Z Z 1 Z Tj0 dV + { j }TikdV − C jTikdV − R jSikldV = 0, (2.4.114) ,0 i k ik 2 ikl where dV is a volume element. If a body is not spatially extended then it is referred to as a particle. In this case, the quantity (2.4.111) satisﬁes δxi = 0. (2.4.115) i i i The aﬃne connection Γj k, and consequently {j k}, C jk, and the curvature tensor in the integrands in (2.4.114), are therefore equal to their respective values at the point Xi. Consequently, we obtain

Z Z Z 1 Z Tj0 dV + { j } TikdV − C j TikdV − R j SikldV = 0. (2.4.116) ,0 i k ik 2 ikl We deﬁne the following integrals: Z M ik = u0 TikdV, (2.4.117) Z N ijk = u0 SijkdV. (2.4.118)

Since the integration domain is not spatially extended, these quantities are tensors, and can be represented as covariant hypersurface integrals: Z ik l ik M = u T dSl, (2.4.119) Z ijk l ijk N = u S dSl. (2.4.120)

Using these integrals and Z Z Z j0 j0 1 d j0 T ,0dV = T dV = 0 T dV (2.4.121) ,0 u ds

turns (2.4.116) into

d M j0 1 + { j }M (ik) − C jM [ik] − R jN ikl = 0. (2.4.122) ds u0 i k ik 2 ikl

79 The conservation law (2.4.113) gives 1 (xlTji) = Tjl − xl{ j }Tik + xlC jTik + xlR jSikm, (2.4.123) ,i i k ik 2 ikm l m ji m jl l jm l m j ik l m j ik (x x T ),i = x T + x T − x x {i k}T + x x Cik T 1 + xlxmR jSikn. (2.4.124) 2 ikn Integrating (2.4.123) over the volume of the body and using Gauß’ theorem to eliminate surface integrals gives Z Z Z Z 1 Z (xlTj0) dV = TjldV − xl{ j }TikdV + xlC jTikdV + xlR jSikmdV. (2.4.125) ,0 i k ik 2 ikm

In this relation, we use xi = Xi, which follows from (2.4.111) and (2.4.115). Substituting (2.4.112) l l 0 into (2.4.125) and using X ,0 = u /u gives

ul Z Z Z Z Z Tj0dV + Xl Tj0 dV = TjldV − Xl { j }TikdV + Xl C jTikdV u0 ,0 i k ik 1 Z + Xl R jSikmdV. (2.4.126) 2 ikm

This equation reduces, by means of (2.4.114), to

ul Z Z Tj0dV = TjldV. (2.4.127) u0

Using the deﬁnition (2.4.117) brings (2.4.127) to

ul M j0 = M jl. (2.4.128) u0 Putting l = 0 in (2.4.128) gives the identity. Integrating (2.4.124) over the volume of the body does not introduce new relations. The expressions analogous to (2.4.123) and (2.4.124) with higher multiples of xi do not introduce new relations as well. If the spin density vanishes, Sijk = 0, the particle is spinless. The conservation law for the spin density (2.4.14) gives in this case the symmetry of the energy-momentum density, Tik = Tki. The tensor density Tij is also equal to the metric energy-momentum density T ij according to (2.3.35). The quantity (2.4.117) is then symmetric:

M ik = M ki. (2.4.129)

Putting j = 0 in (2.4.128) gives ul M 0l = M 00. (2.4.130) u0 The relation (2.4.128) leads then to

ul ujul M jl = M 0j = M 00. (2.4.131) u0 (u0)2

The quantity (2.4.117) for a spinless particle is proportional to the product of the components of the four-velocity. Equations (2.4.62) and (2.4.117) give the four-momentum of a spinless particle:

1 Z M i0 M 0i ui P i = Ti0dV = = = M 00. (2.4.132) c cu0 cu0 c(u0)2

80 The mass (2.4.85) of the particle is therefore given by

P iP uiu M 00 2 m2 = i = i (M 00)2 = , (2.4.133) c2 (cu0)4 (cu0)2 leading to M 00 m = . (2.4.134) (cu0)2 Consequently, the four-momentum is given by

P i = mcui, (2.4.135)

and the mass satisﬁes P iu m = i . (2.4.136) c The four-momentum of a spinless particle is proportional to its four-velocity. The quantity (2.4.131) simpliﬁes to M ik = mc2uiuk. (2.4.137) This relation gives Z uiuk TikdV = mc2 (2.4.138) u0 or uiuk Tik(x) = mc2δ(x − x ) , (2.4.139) 0 u0

where δ(x − x0) is the spatial Dirac delta representing a point mass located at x0. Contracting (2.4.137) with ui gives the relations for the four-momentum and mass:

M iku P i = k , (2.4.140) c M iku u m = i k . (2.4.141) c2 Since M ik for a particle is a tensor, P i is a four-vector and m is a scalar. In a locally inertial frame of reference, (1.6.120) and (2.4.135) give E P i = (mcγ, mγv) = , p , (2.4.142) c so the energy and momentum of the particle are

E = mc2γ, (2.4.143) P = mγv. (2.4.144)

Accordingly, (2.4.133) gives E2 = (Pc)2 + (mc2)2. (2.4.145) In the rest frame of the particle, P = 0, (2.4.145) reduces to Einstein’s formula for the rest energy,

E = mc2. (2.4.146)

The formulae (2.4.143) and (2.4.144) give

Pc2 v = . (2.4.147) E Taking the diﬀerential of (2.4.145) gives EdE = c2P · dP, from which we obtain, using (2.4.147),

dE = v · dP. (2.4.148)

81 If a particle is massless, m = 0, then (2.4.145) and (2.4.147) give

E = P c, v = c. (2.4.149)

We deﬁne the mass density µ such that √ µ sdV = dm, (2.4.150)

where s is given by (1.4.147). The mass density for a particle located at xa is m µ(x) = √ δ(x − x ), (2.4.151) s a so (2.4.139) turns into √ uiuk Tik = µc2 s . (2.4.152) u0 Therefore, the energy-momentum tensor for a spinless particle is given by

uiuk µ(x)c dxi dxk uiuk ik 2 2 √ T (x) = µ(x)c √ 0 = √ = mc δ(x − xa) 0 g00u g00 ds dt −gu Z uiuk = mc2 √ δ(x − x (τ))dτ, (2.4.153) −g a

where xa(τ) is the particle’s wordline as a function of its proper time τ. For a system of particles, we have X uiuk T ik(x) = m c2δ(x − x )√ . (2.4.154) a a −gu0 a In the absence of torsion and in the locally Galilean frame of reference, the conservation law for the energy-momentum tensor is given by (2.4.41), thereby

i Tα ,i = 0. (2.4.155) Let us consider a closed system of particles which carry out a finite motion, in which all quantities vary over finite ranges. We define the average over a certain time interval τ of a function f of these ¯ 1 R τ ¯˙ 1 quantities as f = τ 0 fdt. The average of the derivative of a bounded quantity f = τ f(τ) − f(0) → 0 as τ → ∞. Therefore, averaging (2.4.155) over the time gives

¯ β Tα ,β = 0. (2.4.156) Multiplying (2.4.156) by xα and integrating over the volume gives, omitting surface integrals, Z Z α ¯ β ¯ α x Tα ,βdV = − Tα dV = 0. (2.4.157)

The average energy of the system (2.4.66) is thus Z Z ¯ ¯ 0 ¯ i E = T0 dV = Ti dV. (2.4.158)

Substituting (1.6.120) into (2.4.154) gives

X v2 1/2 T i(x) = m c2δ(x − x ) 1 − , (2.4.159) i a a c2 a

i so Ti ≥ 0. Putting (2.4.159) into (2.4.158) gives

X v2 1/2 E¯ = m c2 1 − , (2.4.160) a c2 a

82 which is referred to as the virial theorem. The equation of motion for a spinless particle follows from (2.4.122), which reduces to

d M j0 + { j }M ik = 0. (2.4.161) ds u0 i k Substituting (2.4.137) into (2.4.161) gives d (muj) + { j }muiuk = 0. (2.4.162) ds i k

Contracting (2.4.162) with uj yields dm duj + m u + { j }uiuku = 0. (2.4.163) ds ds j i k j l i Diﬀerentiating u u gli = 1 with respect to s gives dul dg dul 2 uig + ului li = 2 u + uluig uk = 0. (2.4.164) ds li ds ds l li,k Using 1 1 { j }uiuku = (g + g − g )uiukul = g uiukul (2.4.165) i k j 2 li,k lk,i ik,l 2 li,k turns (2.4.164) into duj u + { j }uiuku = 0. (2.4.166) ds j i k j Consequently, (2.4.163) reduces to dm = 0, (2.4.167) ds showing that the mass of a particle is constant. Taking this constancy into account, (2.4.162) reduces to the metric geodesic equation (1.4.91). A spinless particle moves in a gravitational ﬁeld along a metric geodesic, regardless of its mass. This phenomenon is referred to as the universality of free fall or weak equivalence principle.

2.4.10 Spin tensor for particles If the spin density does not vanish, we can use the conservation law for this quantity to determine the spin tensor and the energy-momentum tensor for a system of particles. The conservation law for the spin density (2.4.14) is

ijk i jlk j ilk [ij] S ,k − Γl kS + Γl kS − 2T = 0. (2.4.168) Integrating (2.4.168) over the volume of the body at a constant time X0 (analogously to the calculations in the preceding section) and using Gauß’ theorem to eliminate surface integrals gives Z Z Z Z ij0 i jlk j ilk [ij] S ,0dV − Γl kS dV + Γl kS dV − 2 T dV = 0. (2.4.169)

i For a particle, the aﬃne connection Γj k in the integrands in (2.4.169) is equal to its value at the point Xi. Consequently, we obtain Z Z Z Z ij0 i jlk j ilk [ij] S ,0dV − Γl k S dV + Γl k S dV − 2 T dV = 0. (2.4.170)

Using the integrals (2.4.117) and (2.4.118) turns (2.4.170) into

d N ij0 − Γ i N jlk + Γ j N ilk − 2M [ij] = 0. (2.4.171) ds u0 l k l k

83 The conservation law (2.4.168) gives

l ijk ijl l i jlk l j ilk l [ij] (x S ),k = S + x Γl kS − x Γl kS + 2x T . (2.4.172) Integrating (2.4.172) over the volume of the body and using Gauß’ theorem to eliminate surface integrals gives Z Z Z Z Z l ij0 ijl l i jmk l j imk l [ij] (x S ),0dV = S dV + x Γm kS dV − x Γm kS dV + 2 x T dV. (2.4.173)

Substituting (2.4.112) into (2.4.173) and using xi = Xi gives

ul Z Z Z Z Z Sij0dV + Xl Sij0 dV = SijldV + Xl Γ i SjmkdV − Γ j SimkdV u0 ,0 m k m k Z +2 T[ij]dV , (2.4.174) which reduces, by means of (2.4.169), to

ul Z Z Sij0dV = SijldV. (2.4.175) u0 Using the deﬁnition (2.4.118) brings (2.4.175) to

ul N ij0 = N ijl. (2.4.176) u0 Putting l = 0 in (2.4.176) gives the identity. The expressions analogous to (2.4.172) with higher multiples of xi does not introduce new relations. The relation (2.4.176) infers that the spin tensor for a system of particles satisﬁes

sijl = sijul, (2.4.177)

where ij ijl s = s ul. (2.4.178) If this tensor is orthogonal to ui, ij s uj = 0, (2.4.179) then it has 3 independent components. A system satisfying (2.4.177) and (2.4.179) is referred to as a spin ﬂuid. The spin tensor (2.4.177) is traceless, because of (2.4.179). In a locally Galilean, rest frame of reference, (2.4.179) becomes s0α = 0. (2.4.180)

In this frame, the 3 components of sij are spatial, sαβ, and are equivalent to 3 components of a spatial pseudovector: 1 sα = eαβγ s . (2.4.181) 2 βγ The relation (2.4.128) gives ul ul M jl = M 0j + 2 M [j0]. (2.4.182) u0 u0 Putting j = 0 in (2.4.182) gives (2.4.130). The relations (2.4.134) and (2.4.182) lead then to

ujul ul ul M jl = M 00 + 2 M [j0] = mc2ujul + 2 M [j0]. (2.4.183) (u0)2 u0 u0 The relation (2.4.171) gives

d N i00 2M [i0] = − Γ i N 0lk + Γ 0 N ilk. (2.4.184) ds u0 l k l k

84 Substituting this equation into (2.4.183) yields

ul d N j00 M jl = mc2ujul + − Γ j N 0ik + Γ 0 N jik . (2.4.185) u0 ds u0 i k i k

Using (2.4.176) brings (2.4.185) to

ul d N j00 uk uk M jl = mc2ujul + + Γ j N i00 + Γ 0 N ji0 . (2.4.186) u0 ds u0 u0 i k u0 i k

Consequently, the four-momentum (2.4.132) is

M i0 1 d N i00 uk uk P i = = mcui + + Γ i N l00 + Γ 0 N il0 . (2.4.187) cu0 cu0 ds u0 u0 l k u0 l k

The four-momentum of a particle with spin is not proportional to its four-velocity. il0 ul i00 If the spin density is completely antisymmetric, then (2.4.176) gives N = − u0 N and thus

N ijk = 0. (2.4.188)

Therefore, such a ﬁeld cannot be represented as a point or a system of points.

2.4.11 Relativistic ideal ﬂuids

In an arbitrary frame of reference, the metric energy-momentum tensor Tik describing isotropic spinless matter (without a preferred direction in its rest frame) can be decomposed into several parts. One part is proportional to uiuk, analogously to (2.4.152). Another part is proportional to the projection tensor (1.4.229): hik = gik − uiuk, (2.4.189) which is orthogonal to ui: k hiku = 0. (2.4.190) i The tensor Tik can also contain terms with covariant derivatives of u . Let us assume that Tik does not depend on derivatives of ui. Therefore, we have

Tik = uiuk − phik = ( + p)uiuk − pgik, (2.4.191) where a scalar is equal to the energy density W in the locally Galilean rest frame and a scalar p is ik the pressure. In this frame T = diag(, p, p, p) and the stress tensor σαβ = −pδαβ, thereby (2.4.80) gives I I F α = − p df α = − p nαdf. (2.4.192)

This equation, referred to as Pascal’s law, states that the force per unit surface df acting on a surface is parallel, with the opposite sign, to the outward normal vector of this surface nα, dF α/df = −pnα. The scalars and p can also be written as

i k = Tiku u , (2.4.193) 1 p = − T hik. (2.4.194) 3 ik Matter described by the tensor (2.4.191) represents an ideal ﬂuid. The relation (2.4.191) can be written as Tik = (cπi + pui)uk − pgik, (2.4.195) where 1 π = T uk = u (2.4.196) i c ik c i

85 is equal to the four-momentum density in the locally Galilean rest frame: 1 π = T . (2.4.197) i c i0 We also have π ui = . (2.4.198) i c The relation between and p is referred to as the equation of state. In the Galilean frame of reference, combining (1.6.120), (2.4.82), and (2.4.191) gives

+ pv2/c2 W = , (2.4.199) 1 − v2/c2 ( + p)v S = , (2.4.200) 1 − v2/c2 ( + p)v v σ = − α β − pδ . (2.4.201) αβ c2 − v2 αβ The relation (2.4.191) gives i T = T i = − 3p. (2.4.202) 2 2 0 α The component T00 = u0 + p(u0 − g00) is, by means of u0 = (g00dx + g0αdx )/ds, (1.4.126) and (1.4.127), equal to dl 2 T = u2 + pg , (2.4.203) 00 0 00 ds

so it is positive under physical conditions > 0, p > 0, and g00 > 0. If Tik depends also on derivatives of ui then matter described by the tensor (2.4.191) with the corresponding additional terms represents a viscous ﬂuid. Comparing (2.4.202) with (2.4.159) gives

X v2 1/2 − 3p = m c2 1 − , (2.4.204) a c2 a where the summation extends over all particles in unit volume, thereby p ≤ /3. In the nonrelativistic limit p ≈ 0, while in the ultrarelativistic limit (v → c) p → /3. Let us consider a system of noninteracting identical particles of mass m, which we call an ideal gas, with the number of particles in unit volume (number density or concentration) n, thereby

µ = nm. (2.4.205)

Comparing (2.4.191) in the locally Galilean rest frame with (2.4.153) gives the kinetic formulae for ideal gases:

= nmc2γ,¯ (2.4.206) nm p = γv2. (2.4.207) 3 The covariant conservation (2.4.30) of the metric energy-momentum tensor (2.4.191) gives

k i k i ik ( + p)u :ku + ( + p)u u :k = p,kg . (2.4.208)

Multiplying (2.4.208) by ui gives the equation of continuity:

k k ( + p)u :k = p,ku . (2.4.209) Substituting this equation into (2.4.208) gives the Euler equation:

D{}ui ( + p) = p hik. (2.4.210) ds ,k

86 If p,i ∝ ui (which includes the case p = const) then (2.4.210) reduces to the metric geodesic equation (1.4.91). Deﬁning a quantity n such that dn d = (2.4.211) n + p brings (2.4.209) to the conservation law

i (nu ):i = 0. (2.4.212) The quantity n thus represents the proper (in the rest frame) number density of particles composing the ﬂuid. The number of particles dN in a volume element dV in the rest frame of reference is equal to dN = n dV, (2.4.213) where n is the proper number density. In a frame of reference moving relative to the rest frame with velocity v, the same volume element is given by dV 0 = dV p1 − v2/c2 (1.6.109), and the number density is n0. Since dN 0 = n0 dV 0 and dN 0 = dN is an invariant, we have n n0 = . (2.4.214) p1 − v2/c2

In an arbitrary frame of reference, the tetrad energy-momentum density Tik describing isotropic matter with spin cannot be decomposed, because of its asymmetry, as in (2.4.191). However, it can be decomposed as in (2.4.195): √ Tik = −g (cpi + pui)uk − pgik , (2.4.215) where 1 p = √ T uk (2.4.216) i c −g ik is the corresponding four-momentum density in the locally Galilean rest frame. The conservation law for the spin density (2.4.15) gives

∗ k c(piuj − pjui) = ∇ksij . (2.4.217) Deﬁning the energy density analogously to (2.4.198) as

i = cpiu (2.4.218) and multiplying (2.4.217) by uj gives 1 p = u + ∇∗s kuj. (2.4.219) i c i c k ij Therefore, we obtain √ ∗ k l Tij = −g(uiuj − phij + ∇ksil u uj), (2.4.220) which, with the Belinfante-Rosenfeld relation (2.3.34), gives the metric energy-momentum tensor for isotropic matter with spin: 1 T = u u − ph + ∇∗s kulu − ∇∗(s k + 2sk ). (2.4.221) ij i j ij k il j 2 k ij (ij) Substituting (2.4.219) into (2.4.217) gives the dynamical equation for the spin tensor:

∗ k ∗ k l ∗ k l ∇ksij − ∇ksil u uj + ∇ksjl u ui = 0. (2.4.222)

j If sijk = 0 then (2.4.219) gives pi ∝ ui. Multiplying (2.4.222) by u gives the identity, thereby any 3 components of (2.4.222) are linear combinations of the other components. Therefore, we can impose k 3 constraints on sij .

87 2.4.12 Multipole expansion of spin tensor Let us consider matter which is distributed over a small but spatially extended region in space. In this case, the quantity (2.4.111) is small but does not vanish. Since the dimensions of the body are small, integrals with two or more factors δxi multiplying Tjk can be neglected. For the consistency of this approximation, we also neglect integrals with one or more factors δxi multiplying Sjkl because the corrections to the energy-momentum tensor from intrinsic spin are quadratic in the spin density (confer (2.5.19)). Since the integration is over the volume of the body at a constant time, we also have δx0 = 0. (2.4.223) The conservation law for the spin density is given by (2.4.168). Integrating it over the volume of the body gives (2.4.169). Since the aﬃne connection appears only in terms with the spin density, these terms in this approximation satisfy (2.4.115). Accordingly, the aﬃne connection in the integrands in (2.4.169) is equal to its value at the point Xi, yielding (2.4.170) and (2.4.171). We also have (2.4.172) and (2.4.173). However, we use xi = Xi + δxi in the term with Tjk. Instead of (2.4.174), we obtain

ul Z Z Z Z Z Sij0dV + Xl Sij0 dV = SijldV + Xl Γ i SjmkdV − Γ j SimkdV u0 ,0 m k m k Z Z +2 T[ij]dV + 2 δxlT[ij]dV. (2.4.224)

This equation reduces, by means of (2.4.169), to

ul Z Z Z Sij0dV = SijldV + 2 δxlT[ij]dV. (2.4.225) u0 We deﬁne the following integral: Z M ijk = −u0 δxiTjkdV. (2.4.226)

The relation (2.4.223) gives M 0jk = 0. (2.4.227) Using the deﬁnitions (2.4.118) and (2.4.226) brings (2.4.225) to

1 ul M l[ij] = − N ij0 − N ijl . (2.4.228) 2 u0

Putting l = 0 in (2.4.228) gives the identity because of (2.4.227). The expressions analogous to (2.4.172) with higher multiples of xi do not introduce new relations. If the spin density vanishes, the relation (2.4.228) gives M ijk = M ikj. (2.4.229)

2.4.13 Multipole expansion of energy-momentum tensor In the integrated conservation law for the tetrad energy-momentum density (2.4.114), we expand the aﬃne connection, which multiplies Tik, but not the curvature tensor, which multiplies Sjkl. This expansion in the linear approximation is

i i(0) i(0) l Γj k = Γj k + Γj k,l δx , (2.4.230) where the superscript (0) denotes a value at Xi. Accordingly, we have

i i (0) i (0) l {j k} = {j k} + {j k},l δx , (2.4.231) i i(0) i(0) l C jk = C jk + C jk,lδx . (2.4.232)

88 Substituting (2.4.231) and (2.4.232) into (2.4.114) and omitting the superscripts (0) gives Z Z Z Z j0 j ik j l ik j ik T ,0dV + {i k} T dV + {i k},l δx T dV − Cik T dV Z 1 Z −C j δxlTikdV − R j SikldV = 0. (2.4.233) ik ,l 2 ikl Using the deﬁnitions (2.4.117), (2.4.118), and (2.4.226) turns (2.4.233) into d M j0 1 + { j }M (ik) − { j } M l(ik) − C jM [ik] + C j M l[ik] − R jN ikl = 0, (2.4.234) ds u0 i k i k ,l ik ik ,l 2 ikl which generalizes (2.4.122). Integrating (2.4.123) over the volume of the body and using Gauß’ theorem to eliminate surface integrals gives (2.4.125). Substituting there (2.4.111) and (2.4.112), together with (2.4.231) and (2.4.232), gives ul Z Z Z Z Z Tj0dV + Xl Tj0 dV + (δxlTj0) dV = TjldV − Xl { j }TikdV u0 ,0 ,0 i k Z Z Z l j ik l j ik l j ik − δx {i k}T dV + X Cik T dV + δx Cik T dV 1 Z + Xl R jSikmdV. (2.4.235) 2 ikm This equation reduces, by means of (2.4.114), to l Z Z Z Z Z u j0 l j0 jl l j ik l j ik T dV + (δx T dV = T dV − δx {i k}T dV + δx Cik T dV, (2.4.236) u0 ,0

j j i where {i k} and Cik are evaluated at X and we omitted the superscripts (0). Using the deﬁnitions (2.4.117) and (2.4.226) brings (2.4.236) to ul d M lj0 M j0 − = M jl + { j }M lik − C jM lik, (2.4.237) u0 ds u0 i k ik which generalizes (2.4.128). Putting l = 0 in (2.4.237) gives the identity because of (2.4.227). Integrating (2.4.124) over the volume of the body and using Gauß’ theorem to eliminate surface integrals gives Z Z Z Z l m j0 m jl l jm l m j ik (x x T ),0dV = x T dV + x T dV − x x {i k}T dV Z 1 Z + xlxmC jTikdV + xlxmR jSikndV. (2.4.238) ik 2 ikn Substituting (2.4.111) and (2.4.112) into (2.4.238) gives ul Z ul Z um Z um Z Xm Tj0dV + δxmTj0dV + Xl Tj0dV + δxlTj0dV u0 u0 u0 u0 Z Z Z l m j0 l m j0 m l j0 +X X T ,0dV + X δx T ,0dV + X δx T ,0dV Z Z 1 Z = −XlXm { j }TikdV − C jTikdV − R jSikldV i k ik 2 ikl Z Z Z l jm m j ik m j ik +X T dV − δx {i k}T dV + δx Cik T dV Z Z Z m jl l j ik l j ik +X T dV − δx {i k}T dV + δx Cik T dV Z Z + δxmTjldV + δxlTjmdV. (2.4.239)

89 This equation reduces, by means of (2.4.114) and (2.4.236), to

ul Z um Z Z Z δxmTj0dV + δxlTj0dV = δxmTjldV + δxlTjmdV (2.4.240) u0 u0 or ul um M mi0 + M li0 = M mil + M lim. (2.4.241) u0 u0 The expressions analogous to (2.4.123) and (2.4.124) with higher multiples of xi do not introduce new relations.

2.4.14 Mathisson-Papapetrou-Dixon equations We deﬁne the following integral: 1 Z Lik = (δxiTk0 − δxkTi0)dV, (2.4.242) c which is analogous to the angular momentum four-tensor (2.4.106). This quantity is related to (2.4.226) by 1 Lik = (M ki0 − M ik0). (2.4.243) cu0 We also deﬁne 1 Z J ik = Lik + Sik0dV, (2.4.244) c which is analogous to the angular momentum four-tensor (2.4.105). This quantity is related to (2.4.118) and (2.4.226) by 1 J ik = (M ki0 − M ik0 + N ik0). (2.4.245) cu0 The relations (2.4.227) and (2.4.243) give

M j00 = −cu0Lj0. (2.4.246)

Let us consider spinless matter in spacetime without torsion. In this case, the symmetry relations (2.4.129) and (2.4.229) bring (2.4.234) to

d M j0 + { j }M ik − { j } M lik = 0. (2.4.247) ds u0 i k i k ,l The relation (2.4.237) reduces to

ul d M lj0 M j0 − = M jl + { j }M lik. (2.4.248) u0 ds u0 i k Antisymmetrizing (2.4.248) in the indices j, l and using (2.4.129) and (2.4.243) gives

uj ul dLjl 1 1 M l0 − M j0 + + { j }M lik − { l }M jik = 0. (2.4.249) cu0 cu0 ds c i k c i k In the absence of the external gravitational ﬁeld and neglecting the gravitational ﬁeld of the j body, we have {i k} = 0. The relation (2.4.247) reduces to dP i = 0, (2.4.250) ds where P i is given by (2.4.132). The integration of this equation gives the conservation of the momentum four-vector along a world line:

P i = const. (2.4.251)

90 The relation (2.4.249) reduces to

dLik + uiP k − ukP i = 0, (2.4.252) ds whose integration gives the conservation of the angular momentum four-tensor along a world line:

Lik + XiP k − XkP i = const. (2.4.253)

The tensor Lik is the intrinsic angular momentum of the body, while the tensor (in the absence of the gravitational ﬁeld) XiP k − XkP i is the orbital angular momentum associated with the motion of the body as a whole. If Lik = 0 then (2.4.252) gives P i ∝ ui, thereby (2.4.251) is equivalent to ui = const and thus Xi is a linear function of the proper time τ. If Lik 6= 0 then Xi can be given i α by 3 arbitrary functions of τ because u ui = 1. In the momentum rest frame, in which P = 0, uα 6= 0, thereby the body has an arbitrary internal motion. This arbitrariness is a consequence of 10 equations (2.4.250) and (2.4.252) for 13 independent components of ui, P i, and Lik. Putting j = 0 in (2.4.248) gives

ul d M l00 M 0l = M l0 = M 00 − − { 0 }M lik. (2.4.254) u0 ds u0 i k Substituting (2.4.254) into (2.4.248) gives

ul uj d M j00 d M lj0 M jl = M 00 − − { 0 }M jik − − { j }M lik. (2.4.255) u0 u0 ds u0 i k ds u0 i k Antisymmetrizing (2.4.255) in the indices j, l and using (2.4.129), (2.4.243), and (2.4.246) gives

dLjl uj dLl0 ul dLj0 1 uj 1 ul + − + { j } − { 0 } M lik − { l } − { 0 } M jik = 0. (2.4.256) ds u0 ds u0 ds c i k u0 i k c i k u0 i k

This relation contains the time derivative of the quantity Lik only, thereby it is an equation of motion for this quantity. The tensor M ijk can be determined from the relation (2.4.241). Taking the cyclic permutations of the indices i, l, m in (2.4.241), adding the ﬁrst and second of these relations, and subtracting the third gives ul ui um M [mi]0 + M [ml]0 + M (li)0 = M l[im] + M i[lm] + M m(il). (2.4.257) u0 u0 u0 This equation, using (2.4.229) and (2.4.243), reduces to

M (il)0 culLim + cuiLlm + 2um = 2M mil. (2.4.258) u0 Putting in this relation l = 0, and using (2.4.227) and (2.4.243), gives

2M mio = cu0Lim − cuiLm0 − cumLi0, (2.4.259)

which upon substitution into (2.4.258) gives cum 2M mil = −culLmi − cuiLml − (uiLl0 + ulLi0). (2.4.260) u0 Since Lik is a tensor, we also have D{}Lik dLik = Lik uj = Lik uj + { i }Llkuj + { k }Liluj = + { i }Llkuj + { k }Liluj. (2.4.261) ds :j ,j l j l j ds l j l j Substituting the last two equations into (2.4.256) gives

D{}Ljl uj D{}Ll0 ul D{}Lj0 + − = 0. (2.4.262) ds u0 ds u0 ds

91 Contracting this equation with ul gives

1 D{}Lj0 D{}Ljk uju D{}Lk0 = u + k , (2.4.263) u0 ds k ds u0 ds which upon substitution into (2.4.262) gives

D{}Ljl D{}Lkl D{}Ljk − uju − ulu = 0. (2.4.264) ds k ds k ds This equation is a covariant equation of motion for the intrinsic angular momentum four-tensor Lik, confirming that Lik is a tensor. It is analogous to (2.4.222). For a particle, the quantities M ik and P i are tensors. For matter distributed over a spatially extended region in space, these quantities are not tensors. In this case, we define the modified four-momentum as M j0 1 Πj = + { j }uiLk0. (2.4.265) cu0 u0 i k Accordingly, we define the modified mass analogously to (2.4.136):

Πiu m = i . (2.4.266) c Equation (2.4.254) gives, using (2.4.246) and (2.4.260),

ui dLi0 uiuj M i0 = M 00 + c + c{ 0 } ujLik + Lk0 u0 ds j k u0 ui D{}Li0 uiuj = M 00 + c − cuk({ i }Ll0 + { 0 }Lil) + c{ 0 } ujLik + Lk0 . (2.4.267) u0 ds l k l k j k u0 Therefore, we obtain

ui D{}Li0 M i0 + c{ i }Lj0uk = (M 00 + c{ 0 }Lj0uk) + c . (2.4.268) j k u0 j k ds

0 Contracting this equation with ui/u , and using (2.4.265) and (2.4.266), gives

1 u D{}Li0 m = (M 00 + c{ 0 }Lj0uk) + i . (2.4.269) (cu0)2 j k cu0 ds

Substituting this equation into (2.4.268) and using (2.4.263) yields

M i0 c D{}Lik = mc2ui − { i }Lj0uk + c u . (2.4.270) u0 u0 j k ds k Therefore, the quantity (2.4.265) is equal to

D{}Ljk Πj = mcuj + u . (2.4.271) ds k This equation shows that Πi is a four-vector, thereby m deﬁned in (2.4.266) is a scalar. Substituting (2.4.270) into (2.4.247), and using (2.4.248) and (2.4.260), gives

d D{}Lik D{}Ljl mcui +u +{ i }uk mcuj +u +Lmkuj({ i } +{ l }{ i }) = 0. (2.4.272) ds k ds j k l ds j k ,m j k l m

Using the Riemann tensor (1.4.57), this equation takes a covariant form:

D{} D{}Lik 1 mcui + u = − P i ujLmk (2.4.273) ds k ds 2 jmk

92 or D{}Πi 1 = − P i ujLmk. (2.4.274) ds 2 jmk The relation (2.4.271) brings (2.4.264) to D{}Llj = ujΠl − ulΠj. (2.4.275) ds Equations (2.4.274) and (2.4.275) are the Mathisson-Papapetrou-Dixon equations of motion for a spatially extended body in a gravitational field. For a particle, Lik = 0. In this case, (2.4.275) gives Πi proportional to ui. Consequently, (2.4.274) reduces to D{}ui/ds, which is the metric geodesic equation (1.4.91). The change of the mass (2.4.266) along a world line is, using (2.4.271) and (2.4.274), dm D{}m 1 D{}Πj 1 D{}u 1 D{}u 1 D{}Lji D{}u = = u + Πj j = Πj j = u j ds ds c j ds c ds c ds c ds i ds 1 D{}u D{}u = − Lji i j = 0, (2.4.276) c ds ds showing that the modified mass is constant along the world line. For a spatially extended body with the angular momentum four-tensor (2.4.242), the equation of motion (2.4.273) shows that the body does not move along a metric geodesic. Analogously to the Pauli-Lubańskipseudovector (1.6.71), we define a pseudovector: 1 Li = eijklu L , (2.4.277) 2 j kl which is orthogonal to ui, i L ui = 0. (2.4.278) i Differentiating (2.4.277) covariantly with respect to {j k} and using (2.4.264) gives D{}Li D{}uk D{}ui = −ui L + ukL , (2.4.279) ds ds k ds k thereby the covariant (with respect to the Levi-Civita connection) change of this pseudovector along the world line is equal to the corresponding Fermi-Walker transport. Multiplying (2.4.279) by Li and using (2.4.278) gives d(LiL ) i = 0, (2.4.280) ds thereby the change of the pseudovector Li along the world line is a four-rotation with a constant value of the magnitude (precession). For matter with spin in spacetime without torsion, the calculations are similar to those for spinless matter. The modified four-momentum is generalized to 1 1 Πj = P j + { j }(uiJ k0 + N 0ik) − C jN ik0, (2.4.281) cu0 i k 2cu0 ik and the modified mass is generalized to u u u m = j P j + j { j }(uiJ k0 + N 0ik) − j C jN ik0. (2.4.282) c c2u0 i k 2c2u0 ik The equation of motion (2.4.274) becomes D{}Πj 1 1 = − P j uiJ mk − N Cikl:j, (2.4.283) ds 2c imk 2c ikl and the equation of motion (2.4.275) becomes D{} 1 1 J lj = cujΠl − culΠj + Cl N jik + C lN ikj − Cj N lik − C jN ikl. (2.4.284) ds ik 2 ik ik 2 ik For a body with the angular momentum four-tensor (2.4.244), the last equation shows that the body does not move along a metric geodesic. References: [2, 3, 4, 6, 7, 9, 10].

93 2.5 Gravitational field equations 2.5.1 Einstein-Cartan action and equations The metric and torsion tensors are two independent, fundamental variables describing a gravitational field. The action for the gravitational field and matter is equal, following (2.2.3), to

1 Z √ S = S + S = − R −gdΩ + S . (2.5.1) g m 2κc m

The action (2.5.1) subjected to varying the metric and torsion tensors is called the Einstein-Cartan action for the gravitational field and matter. Using (1.4.67) and applying partial integration (1.4.48) gives 1 Z √ S = − P − gik(2Cl + Cj Cl − Cl Cm ) −gdΩ g 2κc il:k ij kl im kl 1 Z √ 1 I √ = − P − gik(Cj Cl − Cl Cm ) −gdΩ + Clk −gdS , (2.5.2) 2κc ij kl im kl κc l k where dSi is the element of the closed hypersurface surrounding the integration four-volume. The stationarity of action (2.1.4), which is a part of the principle of least action, is applied with a condition that the variations of the variables at the boundary of integration four-volume vanish. Accordingly, the variation of the hypersurface integral taken over this boundary in (2.5.2) vanishes. This integral therefore does not contribute to the field equations and can be omitted, which reduces (2.5.1) to 1 Z √ S = − P − gik(Cj Cl − Cl Cm ) −gdΩ + S . (2.5.3) 2κc ij kl im kl m √ Firstly, we vary (2.5.3) with respect to the metric tensor. Using (2.3.2) and the identity δ −g = 1 √ ik − 2 −ggikδg , which results from (1.4.20), gives 1 Z √ √ 1 √ δ S = − δP gik −g + P δgik −g − P −gg δgik dΩ g 2κc ik ik 2 ik 1 Z 1 √ − −Cj Cl + Cl Cm + g (Cjm Cl − Clj Cm ) −gδgikdΩ 2κc ij kl im kl 2 ik j ml m jl 1 Z √ + T −gδgikdΩ. (2.5.4) 2c ik We define the contravariant metric density, √ gik = −ggik, (2.5.5)

ik whose covariant derivative with respect to the Christoﬀel symbols vanishes: g :l = 0. In the ﬁrst term on the right-hand side of (2.5.4), using (1.4.70) and applying partial integration (1.4.48) brings this term to zero: Z Z I ik l l ik ik l ik l δPikg dΩ = (δ{i k}):l − (δ{i l}):k g dΩ = (g δ{i k}dSl − g δ{i l}dSk) = 0. (2.5.6)

We therefore obtain 1 Z √ δ S = − G −gδgikdΩ g 2κc ik 1 Z 1 √ − −Cj Cl + Cl Cm + g (Cjm Cl − Clj Cm ) −gδgikdΩ 2κc ij kl im kl 2 ik j ml m jl 1 Z √ + T −gδgikdΩ, (2.5.7) 2c ik

94 where Gik is the Einstein tensor (1.4.79). Secondly, we vary (2.5.3) with respect to the contortion tensor, which is equivalent to varying with respect to the torsion tensor. Using (2.3.18) and (2.3.26) gives 1 Z √ 1 Z √ δ S = − (Ckj − Clj δk) −gδCi dΩ + T −gδgikdΩ C κc i l i jk 2c ik 1 Z √ + s ik −gδCj dΩ. (2.5.8) 2c j ik The total variation of S is then δS = δgS + δCS. (2.5.9) ik j Since the variations δg and δC ik are independent, the stationarity of action (2.1.4) yields ik δgS = δCS = 0. The condition δgS = 0 for an arbitrary δg gives the ﬁrst Einstein-Cartan equation: Gik = κ(Tik + Uik), (2.5.10) where 1 1 U = Cj Cl − Cl Cj − g (Cjm Cl − CmjlC ) (2.5.11) ik κ ij kl ij kl 2 ik j ml ljm or 1 U = 4S S − 2Sj S l − 2Sj Sl + S S jl ik κ i k il jk il kj ijl k 1 − g (4SjS − 2Sl Snm − Sl S mn) . (2.5.12) 2 ik j mn l mn l

The ﬁrst Einstein-Cartan equation (2.5.10) can be written as

1 P = κ T + U − (T + U)g , (2.5.13) ik ik ik 2 ik where i i T = T i,U = U i. (2.5.14) j The condition δCS = 0 for an arbitrary δC ik gives κ Ck − δk Cl = s k. (2.5.15) [ji] [i j]l 2 ij This equation can be written as the Cartan equations or the second Einstein-Cartan equation: κ T j = Sj − S δj + S δj = − s j, (2.5.16) ik ik i k k i 2 ik

j where T ik is the modiﬁed torsion tensor. The relation (2.5.16) is equivalent to κ Sk = − (s k + δk s l), (2.5.17) ij 2 ij [i j]l κ Ck = (sk − s k − s k − g skl + δks l). (2.5.18) ij 2 ij ij j i ij l j il Combining (2.5.11) and (2.5.18) gives

1 1 1 U ik = κ −sij skl − sijlsk + sjlis k + gik(−4sl sjm + sjlms ) . (2.5.19) [l j] 2 jl 4 jl 8 j[m l] jlm

The tensor (2.5.19) represents a correction to the metric energy-momentum tensor from the spin contributions to the geometry of spacetime. It is quadratic in the spin tensor, thereby representing a spin-spin contact interaction. Accordingly, changing the signs of all the components of the spin tensor

95 does not affect this correction. The spin tensor also appears in Tik because Lm depends on torsion. The first Einstein-Cartan equation (2.5.10), in which Uik is given by (2.5.19), is a field equation with the combined energy-momentum tensor Tik + Uik as a source of the curvature. The conservation law (2.4.28) for the metric energy-momentum tensor is, upon substituting (2.5.10) and (2.5.15), equivalent to the the contracted Bianchi identity (1.4.78) for the Einstein tensor. This identity, applied to (2.5.10), gives the Riemannian conservation law for the combined energy-momentum tensor: ik ik (T + U ):k = 0. (2.5.20) This law is equivalent to (2.4.28) with (2.5.18). The relation (2.5.17) between the torsion and spin tensors is algebraic. Torsion at a given point in spacetime does not vanish only if matter is present at this point, represented in the Lagrangian density by a function which depends on torsion. If the matter Lagrangian density does not depend on torsion, then the spin tensor vanishes, and so does the torsion tensor. In vacuum, which is defined as the absence of matter, Tik = 0 and sijk = 0, the Riemannian Ricci tensor in (2.5.13) also vanishes:

Pik = 0. (2.5.21) Unlike the metric, which is related to matter through a differential field equation, torsion does not propagate in vacuum. The vanishing of Pik and Sijk at a given point in spacetime is a covariant i criterion for the absence of matter at this point. The Riemann tensor P jkl at such a point, however, can be different from zero.

2.5.2 Sciama-Kibble action The tetrad and spin connection, instead of the metric tensor and aﬃne connection, can be regarded as dynamical variables. The action (2.5.1) subjected to varying the tetrad and spin connection is called the Sciama-Kibble action. Using (2.3.27) gives 1 Z 1 Z 1 Z δS = − δ(eR)dΩ + T aδei dΩ + S iδωab dΩ. (2.5.22) 2κc c i a 2c ab i The Lagrangian density for the gravitational ﬁeld is given by (2.2.2), with the curvature scalar R given by (1.5.42) and (1.5.44):

i jb a a a c a c ij ab a cb eR = eeae (ω bj,i − ω bi,j + ω ciω bj − ω cjω bi) = 2eab(ω j,i + ω ciω j), (2.5.23) where ij [i j] eab = eea eb . (2.5.24) ij ij c ij c ij i kj j ik k ij This quantity satisﬁes eab|j = eab,j − ω ajecb − ω bjeac + Γk jeab + Γk jeab − Γk jeab = 0, which results from (1.5.29). Varying eR and omitting total derivatives which lead to hypersurface integrals gives, using (1.5.13),

a a i ij ab a cb δ(eR) = (2R i − Rei )eδea + 2eabδ(ω j,i + ω ciω j) a a i ij c ij c ij ab = (2R i − Rei )eδea + 2(eab,j − ω ajecb − ω bjeac)δω i a a i i kj ij ab = (2R i − Rei )eδea − 2(S kjeab + 2Sjeab)δω i. (2.5.25) The variation (2.5.22) is therefore equal to 1 Z 1 1 Z δS = − Ra − Rea eδei dΩ + (Si ekj + 2S eij )δωab dΩ κc i 2 i a κc kj ab j ab i 1 Z 1 Z + T aδei dΩ + S iδωab dΩ. (2.5.26) c i a 2c ab i

ab The condition δS = 0 for an arbitrary δω i gives κ Si − S ei + S ei = − S i, (2.5.27) ab a b b a 2e ab

96 which is equivalent to the second Einstein-Cartan equation (2.5.16). The condition δS = 0 for an i arbitrary δea gives 1 κ Ra − Rea = T a, (2.5.28) i 2 i e i which is equivalent to 1 R − Rg = κt . (2.5.29) ki 2 ik ik Substituting (2.5.16) and (2.5.29) into the conservation law for the spin tensor (2.4.15) gives

k k k k − 2(S ij;k − Si;j + Sj;i) = Rji − Rij − 4Sk(S ij − Siδj + Sjδi ), (2.5.30) which is equivalent to the contracted cyclic identity (1.4.72). Therefore, the contracted cyclic identity imposes the conservation law for the spin density. Substituting (2.5.16) and (2.5.29) into the conservation law for the tetrad energy-momentum tensor (2.4.39) gives 1 1 1 Rj − R = 2S Rj − Rδj + 2Sj Rk − Rδk − (Sj − S δj + S δj )Rkl , (2.5.31) i;j 2 ;i j i 2 i ki j 2 j kl k l l k ji which is equivalent to the contracted Bianchi identity (1.4.73). Therefore, the contracted Bianchi identity imposes the conservation law for the energy-momentum density. The gravitational field equations therefore contain the equations of motion of matter. Substituting (2.5.16) and (2.5.29) into the Belinfante-Rosenfeld relation (2.3.34) gives 1 1 κT = R − Rg + ∇∗(Sj + 2S δj − 2S j − 2Sjg ) = R − Rg ik ki 2 ik j ik k i (ik) ik ki 2 ik ∗ j l j lj +∇j (−C ki + C klδi − C lgik). (2.5.32) Combining (1.4.65), (1.4.67) and (2.5.32) gives 1 1 κT = P − P g + Cl − Cl + Cj Cl − Cj Cl − g (−2Clj ik ik 2 ik ki:l kl:i ki jl kl ji 2 ik l:j lj m mjl j j l l j l j j −C lC jm + C Cljm) − C ki:j − C ljC ki + C kjC li + C ijC kl + C kj:i l j lj j ml j l j lj −C kiC lj − gik(C l:j + C ljC m) − Cj(−C ki + C klδi − C lgik), (2.5.33) which is equivalent to the first Einstein-Cartan equation (2.5.10). Therefore, the relation between the Ricci tensor and the Riemannian Ricci tensor is equivalent to the Belinfante-Rosenfeld relation, whereas (2.5.29) is another form of the first Einstein-Cartan equation. Varying the action for the gravitational field and matter with respect to the metric tensor and the tensorial, antisymmetric part of the affine connection (torsion tensor) constitutes the metric-affine variational principle of stationary action.

2.5.3 Einstein-Hilbert action and Einstein equations In almost all physical situations, the second Einstein-Cartan equation gives a torsion tensor whose squares of the leading components are negligibly small in magnitude relative to the leading components of the Riemann tensor (confer (2.5.11)). In those situations, we can approximate the torsion tensor as zero. In this approximation, the affine connection is equal to the Levi-Civita connection. Varying the action for the gravitational field and matter with respect to the metric tensor, with the affine connection constrained to be equal to the Levi-Civita connection, constitutes the metric variational principle of stationary action. If the Lagrangian density for matter does not depend on the affine connection, then the spin density vanishes, and so does the torsion tensor. In this case, the metric-affine field equations reduce to the metric field equations and we can use the metric principle of stationary action. If the torsion tensor vanishes, the Einstein-Cartan action (2.5.1) reduces to 1 Z √ S = − P −gdΩ + S , (2.5.34) 2κc m

97 which corresponds to the Lagrangian density (2.2.4). The action (2.5.34) subjected to varying the metric tensor is called the Einstein-Hilbert action for the gravitational ﬁeld and matter. The Einstein-Hilbert action is a special case of the Einstein-Cartan action, where the aﬃne connection is constrained to be symmetric and thus equal to the Levi-Civita connection. Varying (2.5.34) with respect to the metric tensor gives, similarly to (2.5.7),

1 Z 1 √ 1 Z √ δ S = − P − P g −gδgikdΩ + T −gδgikdΩ. (2.5.35) g 2κc ik 2 ik 2c ik

ik Applying the stationarity of action δgS = 0 to (2.5.35) for an arbitrary δg gives the Einstein equations of the general theory of relativity: 1 G = P − P g = κT (2.5.36) ik ik 2 ik ik or 1 P = κ T − T g . (2.5.37) ik ik 2 ik √ √ Because δ R P −gdΩ = δ R G −gdΩ, where G is the noncovariant quantity (2.2.7), the left-hand side of the Einstein equations is √ 1 ∂( −gG) G = √ . (2.5.38) ik −g ∂gik The covariant conservation of the Einstein tensor (1.4.78) imposes the conservation of the metric energy-momentum tensor (2.4.30). The gravitational field equations therefore contain the equations of motion of matter, like for the Einstein-Cartan action. The Einstein equations (2.5.36) are 10 second-order partial differential equations for: 10 − 4 = 6 independent components of the metric tensor gik (the factor 4 is the number of the coordinates which can be chosen arbitrarily), 3 independent components of the four-velocity ui, and either or p (which are related to each other by the equation of state). The contracted Bianchi identity (1.4.78) gives the equations of motion of matter. In vacuum, the Einstein equations are 10 − 4 = 6 independent equations (the factor 4 is the number of constraints from the contracted Bianchi identity) for 6 independent components of the metric tensor gik. In the Einstein equations (2.5.36), the only second time-derivatives of gik are the derivatives of the spatial components of the metric tensor,g ¨αβ, and they appear only in the αβ components of the equations. Therefore, the initial values (at t = 0) for gαβ andg ˙αβ can be chosen arbitrarily. The first time-derivativesg ˙0α andg ˙00 appear only in the αβ components of the field equations (2.5.36). The 0α and 00 components of the field equations (2.5.36) give the initial values for g0α and g00. The undetermined initial values forg ˙0α andg ˙00 correspond to 4 degrees of freedom for a free gravitational field. A general gravitational field has 8 degrees of freedom: 4 degrees of freedom for a free gravitational field, 3 related to the four-velocity, and 1 related to the matter ( or p). The above analysis regarding gik also applies to the first Einstein-Cartan equation (2.5.10) because the torsion tensor is algebraically related to the matter. If gik = ηik, then the general theory of relativity is said to reduce to the special theory of relativity. The Einstein equations (2.5.36) are a special case of the first Einstein-Cartan equation (2.5.10). They are valid when the matter fields do not depend on the affine connection, for which the spin density vanishes and Uik = 0. They are also an accurate approximation of (2.5.10) when the matter fields depend on the connection but the tensor Uik can be neglected relative to Tik. In the metric- ab i affine variational principle of stationary action, in which the variations δω i are independent of δea, the spin density is independent of the energy-momentum density. The Einstein-Cartan equations contain the covariant conservation laws for the energy-momentum and spin tensors, which generalize the special-relativistic conservation laws (2.4.41) and (2.4.48). The angular momentum density in (2.4.48) contains both the orbital and intrinsic parts. In the metric variational principle, in which ab ab i the variations δω i = δ$ i are functions of the variations δea and their derivatives according to (1.5.37), the spin density is a function of the energy-momentum energy-momentum density. The Ein- stein equations contain only the covariant conservation law for the energy-momentum tensor, which

98 generalizes the special-relativistic conservation law (2.4.41) with a symmetric energy-momentum tensor. The resulting conservation law (2.4.48) contains the angular momentum density only with the orbital part that depends on the energy-momentum density. Accordingly, the metric variational principle does not account for the intrinsic angular momentum (spin) of matter. Consequently, the existence of spin (which does not depend on energy and momentum) requires the metric-aﬃne variational principle.

2.5.4 Utiyama action The action (2.5.34) subjected to varying the tetrad is called the Utiyama action. The Utiyama action is a special case of the Sciama-Kibble action, where the torsion tensor is approximated as zero and the spin connection is constrained to be equal to the Levi-Civita spin connection (1.5.37) which depends on the tetrad. Using (2.3.6) gives 1 Z 1 Z δS = − δ(eP )dΩ + T aδei dΩ. (2.5.39) 2κc c i a The Lagrangian density for the gravitational ﬁeld is given by (2.2.4), with the Riemann scalar P given by (1.5.45) and (1.5.47):

i jb a a a c a c ij ab a cb eP = eeae ($ bj,i − $ bi,j + $ ci$ bj − $ cj$ bi) = 2eab($ j,i + $ ci$ j). (2.5.40) i a Varying eP and omitting total derivatives gives in the absence of torsion, using δe = eeaδei and ij ij c ij c ij eab|j = eab,j − $ ajecb − $ bjeac = 0 (which results from (1.5.29)),

a a i ij ab a cb a a i δ(eP ) = (2P i − P ei )eδea + 2eabδ($ j,i + $ ci$ j) = (2P i − P ei )eδea ij c ij c ij ab a a i +2(eab,j − $ ajecb − $ bjeac)δ$ i = (2P i − P ei )eδea. (2.5.41) Equaling δS = 0 gives 1 P a − P ea = κt a, (2.5.42) i 2 i i which is equivalent to the Einstein equations (2.5.36) because of (2.3.5) and (2.3.34) (in the absence of torsion).

2.5.5 Einstein pseudotensor and principle of equivalence √ Since the noncovariant quantity G (2.2.6) differs from −gP by total divergences, we can use Gauß- Stokes theorem (1.1.39) to write the action for the gravitational field as 1 Z 1 I 1 I S = − G − gnp(Ci Cm − Ci Cm ) dΩ − gik{ l }dS + gik{ l }dS . g 2κc ni pm nm pi 2κc i k l 2κc i l k (2.5.43) The hypersurface integrals in (2.5.43) do not contribute to the field equations and can be omitted. The action for the gravitational field and matter (2.5.3) thus reduces to 1 Z 1 S = − G − gnp(Ci Cm − Ci Cm ) + L dΩ. (2.5.44) c 2κ ni pm nm pi m

The condition δgS = 0, which is equivalent to δ 1 − G − gnp(Ci Cm − Ci Cm ) + L = 0, (2.5.45) δgjl 2κ ni pm nm pi m gives the first Einstein-Cartan equation (2.5.10). Because G depends on gij and its first deriva- ij tives g ,k, we can construct a canonical energy-momentum density (2.4.42) corresponding to the 1 ik gravitational field, treating − 2κ G like Lm and g like a matter field φ: 1 ∂G t i = − gjl − δi G . (2.5.46) k 2κ jl ,k k ∂g ,i

99 √ This quantity is not a tensor density because G is not a scalar density. Its division by −g defines the Einstein energy-momentum pseudotensor for the gravitational field: t i 1 ∂G √k = − gjl − δi G , (2.5.47) −g 2κ jl ,k k ∂g ,i where G is the quantity (2.2.7). Differentiating (2.5.46) gives ∂G ∂G ∂G ∂G ∂G 2κt i = −∂ gjl − gjl + G = −∂ gjl − gjl + gjl k ,i i jl ,k jl ,ki ,k i jl ,k jl ,ki ∂gjl ,k ∂g ,i ∂g ,i ∂g ,i ∂g ,i ∂G ∂G ∂G δG + gjl = − ∂ gjl = gjl , (2.5.48) jl ,ik ∂gjl i jl ,k δgjl ,k ∂g ,i ∂g ,i which, using (2.5.11) and (2.5.45), leads to

1 √ np i m i m δ Lm + −gg (C C − C C ) 1 √ t i = 2κ ni pm nm pi gjl = (T + −gU )gjl . (2.5.49) k ,i δgjl ,k 2 jl jl ,k The Riemannian conservation law (2.5.20) gives √ √ 1 √ (T i + −gU i) = { l }(T i + −gU i) = glmg (T i + −gU i) k k ,i k i l l 2 im,k l l 1 √ = − glm (T + −gU ). (2.5.50) 2 ,k lm lm The total energy-momentum density for the gravitational ﬁeld and matter is given by √ √ −g t k + T k + −gU k = t k + G k. (2.5.51) i i i i κ i As a result of adding (2.5.49) and (2.5.50), the ordinary divergence of this quantity vanishes, as that in (2.4.59): √ i i √ i i −g i (tk + Tk + −gUk ),i = tk + Gk = 0. (2.5.52) κ ,i Integrating (2.5.52) over the four-volume and using the Gauß’ theorem (1.1.39) gives I i i √ i (tk + Tk + −gUk )dSi = 0. (2.5.53)

The corresponding four-momentum (2.4.63) of the gravitational ﬁeld and matter, which is not a vector (it transforms like a vector only for Lorentz transformations), is therefore conserved: 1 Z √ P = (t k + T k + −gU k)dS = const. (2.5.54) i c i i i k Using (1.4.32) and (1.4.43) gives

m ∂{i l } 1 m n m n mn rs = − (gl(rδs) δi + gi(rδs) δl − g gi(rgs)l), (2.5.55) ∂g ,n 2 l ∂{m l} 1 n rs = − grsδm. (2.5.56) ∂g ,n 2 Consequently, we obtain

l m l ∂G ik m ∂{m k} ik l ∂{i k } ik m ∂{m l} rs = 2g {i l } rs − g {m l} rs − g {i k } rs ∂g ,n ∂g ,n ∂g ,n ∂g ,n 1 1 = −{ n } + { l }δn + { l }δn − { l }gmng + { n}gjlg , (2.5.57) r s 2 s l r r l s m l rs 2 j l rs

100 which leads to

∂G rs √ i rs √ l ri l mi √ i jl √ rs g ,k = − −g{r s}g ,k + −g{r l}g ,k + {m l}g ( −g),k − {j l}g ( −g),k. (2.5.58) ∂g ,i The Einstein pseudotensor (2.5.47) can thus be written as

t i 1 √k = √ ({ i }glm − { l }gmi + δi G). (2.5.59) −g 2κ −g l m ,k m l ,k k

Accordingly, tik is not symmetric in the indices i, k. ik Since the derivatives g ,j are homogeneous linear functions of the Christoffel symbols, the Ein- stein pseudotensor (2.5.59) is a homogeneous quadratic function of the Christoffel symbols, so it vanishes in the locally Galilean and geodesic frame of reference. It can also differ from zero in the Minkowski spacetime (in the absence of the gravitational field) if we choose the coordinates such that the Christoffel symbols do not vanish. Therefore, the energy of the gravitational field is not absolutely localized in spacetime; it depends on the choice of the coordinates. A physically mean- ingful energy-momentum pseudotensor can be constructed if the coordinates are asymptotically (far from the sources of the field) Cartesian, so the Christoffel symbols tend asymptotically to zero. If we neglect torsion, then the gravitational field can be always eliminated locally by transforming the coordinate system to the locally Galilean and geodesic frame of reference in which the Einstein pseudotensor vanishes. This property of the gravitational field is referred to as the principle of equivalence. We can construct from P i (2.5.54) the angular momentum four-tensor (2.4.106): 1 Z √ √ M ik = xi(tkl + T kl + −gU kl) − xk(til + T il + −gU il)dS , (2.5.60) c l which is not a tensor (it transforms like a tensor only for Lorentz transformations). Because the quantity tik is not symmetric, the four-tensor (2.5.60) is not conserved. The conservation law (2.5.52) infers that the quantity (2.5.51) can be written as √ −g t l + G l = η li , (2.5.61) k κ k k ,i

li where ηk satisﬁes li il ηk = −ηk , (2.5.62) kl √ analogously to (2.4.58). The quantity ηi is proportional to −g and linear in ﬁrst derivatives of gik. Substituting (2.5.61) into (2.5.54) gives, using the Gauß-Stokes theorem (1.1.38) and (2.5.51), 1 Z 1 Z 1 I P = η kl dS = (η kl dS − η kl dS ) = η kldf ∗ , (2.5.63) i c i ,l k 2c i ,l k i ,k l 2c i kl

? where dfik is the element of the closed surface which bounds the hypersurface. If the hypersurface is 0 a volume hypersurface, dSk = δkdV , then the four-momentum of the gravitational ﬁeld and matter (2.5.63) in a given volume can be written as a surface integral: 1 I 1 I P = η 0αdf ∗ = η 0αdf , (2.5.64) i 2c i 0α c i α where dfα is the element of the closed surface which bounds the volume. k k If we neglect torsion, then Uik = 0 and the left-hand side of (2.5.51) reduces to ti + Ti . The k k conservation law (2.5.52) reduces to (ti + Ti ),k = 0. This quantity can also be written as (2.5.61), thereby the corresponding four-momentum of the gravitational ﬁeld and matter is also given by (2.5.64). kl The quantity ηi is not unique. The relation (2.5.61) is invariant under a transformation

l l klm ηik → ηik + ζi ,m, (2.5.65)

101 klm where ζi is a quantity satisfying klm kml ζi = −ζi . (2.5.66) We deﬁne 1 λiklm = (−g)(gikglm − gilgkm), (2.5.67) 2κ ikl iklm ilk h = λ ,m = −h . (2.5.68) The quantity (2.5.68) is equal, by means of (1.4.40) and (1.4.43), to 1 hikl = (−g)({ l }gkmgij + { k }gmngil − { m }gkngil 2κ j m m n m n k lm ij l mn ik m ln ik −{j m}g g − {m n}g g + {m n}g g ). (2.5.69) Equations (1.4.40), (1.4.43), (2.2.6) and (2.5.59) infer that 1 η kl = √ h kl (2.5.70) i −g i

kl satisfies (2.5.61). Taking ηi in (2.5.61), which is not related to (2.5.70) by (2.5.65), leads to a different energy-momentum pseudotensor. If we take 1 η kl = √ (2h kl − δkh jl + δlh jk), (2.5.71) i −g i i j i j then the right-hand side of (2.5.61) is given by √ −g η kl = (g − g )gkmgln. (2.5.72) i ,l κ in,m im,n The construction of a conserved four-momentum for the gravitational field and matter is possible because the Lagrangian density for the gravitational field Lg (2.2.2) is linear in the curvature tensor. Accordingly, it is linear in second derivatives of the metric tensor, so we can use the noncovariant ijkl quantity G (2.2.6). Another scalar density which is linear in curvature is Rijkl. Using (1.4.49), (1.4.64) and (1.4.75), and omitting a total derivative, this parity-violating expression reduces to iklm j −2 C kmCjil, which does not depend on the derivatives of the metric tensor and thus does not describe the gravitational field.

2.5.6 Møller pseudotensor

a The Riemann scalar P (1.5.47) is linear in ﬁrst derivatives of the Levi-Civita spin connection $ bi:

i j ab i j ab i j ab i j ab i j ac b eP = (eeaeb$ j),i − (eeaeb),i$ j − (eeaeb$ i),j + (eeaeb),j$ i + eeaeb$ i$c j i j ac b i j ab i j ab i j ac b −eeaeb$ j$c i = 2(eeaeb$ j),i − 2(eeaeb),i$ j + eeaeb$ i$c j i j ac b −eeaeb$ j$c i, (2.5.73) where we used (1.5.45). We can therefore subtract from eP total derivatives without altering the ﬁeld equations, replacing it by a noncovariant quantity M:

i j ab i j ac b i j ac b M = −2(eeaeb),i$ j + eeaeb$ i$c j − eeaeb$ j$c i k ij i aj i kj j ib j ik = −2e({k i}$ j + $ ai$ j − {k i}$ j + $ bi$ j − {k i}$ j) ic j ic j ia j ia j +e($ i$c j − $ j$c i) = e($ i$ aj − $ j$ ai), (2.5.74) using (1.4.43) and (1.5.37). We also deﬁne M M = = $ia $j − $ia $j . (2.5.75) e i aj j ai

102 The Riemannian part (2.2.4) of the Lagrangian density for the gravitational field reduces accordingly to 1 1 L{} = − M = − eM. (2.5.76) g 2κ 2κ Since the noncovariant quantity M (2.5.74) differs from eP by a total divergence, the action for the gravitational field and matter (2.5.3) is equivalent to 1 Z 1 S = − M − e(Cia Cj − Cia Cj ) + L dΩ. (2.5.77) c 2κ i aj j ai m Using (1.5.39), this action can be written as 1 Z e S = (ωia ωj − ωia ωj − 2ωia $j + 2ωia $j ) + L dΩ. (2.5.78) c 2κ i aj j ai i aj j ai m

The condition δeS = 0, which is equivalent to δ 1 np i m i m j − M − eg (C niC pm − C nmC pi) + Lm = 0, (2.5.79) δea 2κ

i gives the Einstein equations (2.5.28). Because of (1.5.22), M depends on the tetrad ea and its first i derivatives ea,j. Therefore, analogously to (2.5.46), we can construct a canonical energy-momentum 1 i density corresponding to the gravitational field, treating − 2κ M like Lm and ea like a matter field φ: 1 ∂M m i = − ej − δi M . (2.5.80) k 2κ j a,k k ∂ea,i This quantity is not a tensor density because M is not a scalar density. Its division by e defines the Møller energy-momentum pseudotensor for the gravitational field: m i 1 ∂M k = − ej − δi M . (2.5.81) e 2κ j a,k k ∂ea,i Differentiating (2.5.80) gives

i ∂M j ∂M j ∂M j ∂M j ∂M j 2κmk ,i = −∂i j ea,k − j ea,ki + M,k = −∂i j ea,k − j ea,ki + j ea,k ∂ea,i ∂ea,i ∂ea,i ∂ea,i ∂ea ∂M j ∂M ∂M j δM j + j ea,ki = j − ∂i j ea,k = j ea,k, (2.5.82) ∂ea,i ∂ea ∂ea,i δea which, using (2.5.11) and (2.5.79), leads to

1 np i m i m i δ Lm + 2κ eg (C niC pm − C nmC pi) j a a j mk ,i = j ea,k = (Tj + eUj )ea,k. (2.5.83) δea The conservation law (2.5.50) gives, using (1.5.5), 1 (T i + eU i) = − glm (T + eU ) = −ej (T a + eU a). (2.5.84) k k ,i 2 ,k lm lm a,k j j Adding (2.5.83) and (2.5.84) leads, using (2.5.28) and (2.5.32), to e (m i + T i + eU i) = ej (T a − T a) = ej ∇∗(Cia − 2Saδi + 2Siea). (2.5.85) k k k ,i a,k j j a,k κ j j j j

If we neglect torsion, then Uik = 0 and the total energy-momentum density for the gravitational ﬁeld and matter is given by The quantity e m k + T k = m k + G k. (2.5.86) i i i κ i

103 The ordinary divergence of this quantity vanishes, as that in (2.5.52):

i i (mk + Tk ),i = 0. (2.5.87) Integrating (2.5.87) over the four-volume and using the Gauß-Stokes theorem gives I i i (mk + Tk )dSi = 0. (2.5.88)

The corresponding four-momentum (2.4.63) of the gravitational ﬁeld and matter in this approximation, which is not a vector (it transforms like a vector only for Lorentz transformations), is therefore conserved: 1 Z P = (m k + T k)dS = const. (2.5.89) i c i i k Using (1.4.32), (1.4.43), (1.5.5), (1.5.10) and (1.5.37) gives

i ∂$ aj 1 i b l i l b ib l ib l il b il b m = (δmδaδj − δmeaej − e eagjm − e eamδj + g eamej + g δagjm), (2.5.90) ∂eb,l 2 i ∂$ ai b l l b m = δaδm − eaem. (2.5.91) ∂eb,l Consequently, we obtain

∂M l jb b jl lb bl lb m = 2(δm$ j − em$ j − $ m + $m + $m ). (2.5.92) ∂eb,l The Møller energy-momentum pseudotensor (2.5.81) can thus be written, using (1.5.37), as

m i 1 1 k = −$i $ja − $j $ ai + { i }$jl − { l }$ji + $ljig − $ijlg + δi M . (2.5.93) e κ ak j ak j l k j l k j kl,j jl,k 2 k

i Accordingly, mik is not symmetric in the indices i, k. We can construct from P (2.5.89) the angular momentum four-tensor (2.4.106): 1 Z M ik = xi(mkl + T kl) − xk(mil + T il)dS , (2.5.94) c l which is not a tensor (it transforms like a tensor only for Lorentz transformations). Because the quantity mik is not symmetric, the four-tensor (2.5.94) is not conserved. Since the Levi-Civita spin connection and the Christoffel symbols are homogeneous linear func- i tions of the derivatives ea,k, these derivatives are homogeneous linear functions of the Christof- fel symbols. Consequently, the Møller energy-momentum pseudotensor (2.5.93) is a homogeneous quadratic function of the Christoffel symbols, so it vanishes in the locally Galilean and geodesic frame of reference. This pseudotensor depends on the choice of both the coordinates and tetrad. To fix the tetrad, one can impose on it 6 constraints which are covariant under constant Lorentz transformations but not under general Lorentz transformations (otherwise such constraints would not fix the tetrad since Lorentz transformations are tetrad rotations).

2.5.7 Landau-Lifshitz energy-momentum pseudotensor In the locally Galilean and geodesic frame of reference, the relations (2.5.61) and (2.5.70) give

ik ikl G = κh ,l, (2.5.95)

where hikl is given by (2.5.67) and (2.5.68). The ﬁrst Einstein-Cartan equation (2.5.10) reduces in this frame to ikl ik ik h ,l = T + U . (2.5.96)

104 ik ik The Riemannian conservation law (2.5.20) reduces to (T + U ),k = 0, which is consistent with (2.5.96). In an arbitrary frame of reference, (2.5.95) is not valid. We deﬁne a quantity tik such that

ik ik ikl (−g)(κt + G ) = h ,l. (2.5.97)

Therefore, we have ik ik ik (−g)(t + T + U ) ,k = 0. (2.5.98) The corresponding four-momentum of the gravitational ﬁeld and matter is conserved: 1 Z P i = (−g)(tik + T ik + U ik)dS = const. (2.5.99) c k

The quantity tik is not a tensor density, thereby the conserved four-momentum P i (2.5.99) is not i a vector. The four-momentum P is not a√ vector even for Lorentz√ transformations, because of the factor −g instead of the weight-1 density −g. Dividing P i by −g at some ﬁxed point (a natural choice is inﬁnity) turns it into a vector with respect to Lorentz transformations. Using (2.5.97) turns (2.5.99), for the hypersurface dS0 = dV , into 1 Z 1 I 1 I P i = hikl dS = hikldf ∗ = hi0αdf . (2.5.100) c ,l k 2c kl c α

This formula differs from the four-momentum of the gravitational field and matter (2.5.64),√ constructed from the Einstein energy-momentum pseudotensor, by an additional factor −g in the integrand (confer (2.5.70)). The quantity tik is referred to as the Landau-Lifshitz energy-momentum pseudotensor for the gravitational field. The explicit expression for the Landau-Lifshitz pseudotensor is 1 tik = (gilgkm − gikglm)(2{ n }{ p } − { n}{ p } − { n }{ p }) 2κ l m n p l p m n l n m p il mn k p k p k p k p +g g ({l p}{m n} + {m n}{l p} − {n p}{l m} − {l m}{n p}) kl mn i p i p i p i p +g g ({l p}{m n} + {m n}{l p} − {n p}{l m} − {l m}{n p}) lm np i k i k +g g ({l n}{m p} − {l m}{n p}) (2.5.101)

or 1 1 (−g)tik = gik glm − gil gkm + gikg gln gpm 2κ ,l ,m ,l ,m 2 lm ,p ,n il kn mp kl in mp np il km −(g gmng ,pg ,l + g gmng ,pg ,l) + glmg g ,ng ,p 1 + (2gilgkm − gikglm)(2g g − g g )gnr gpq . (2.5.102) 8 np qr pq nr ,l ,m

This pseudotensor is symmetric in the indices i, k. The corresponding angular momentum of the gravitational ﬁeld and matter, constructed from P i as that in (2.4.106), is therefore conserved:

1 Z M ik = xi(tkl + T kl + U kl) − xk(til + T il + U il)(−g)dS = const. (2.5.103) c l √ Dividing M ik by −g at inﬁnity turns it into an antisymmetric tensor under Lorentz transformations. Using (2.5.68) and (2.5.99) turns (2.5.103), for the hypersurface dS0 = dV , into 1 Z 1 I M ik = (xiλklmn − xkλilmn )dS = (xiλklmn − xkλilmn )df ∗ c ,nm ,nm l 2c ,n ,n lm 1 I 1 I − (λklin − λilkn) dS = (xihk0α − xkhi0α + λi0αk)df . (2.5.104) c ,n l c α

105 Choosing the volume hypersurface dV = dS0 gives 1 Z M ik = xi(tk0 + T k0 + U k0) − xk(ti0 + T i0 + U i0)(−g)dV = const. (2.5.105) c

The conservation of M 0α in (2.5.105) divided by the conserved P 0 in (2.5.99) gives a uniform motion (2.4.108) (without the intrinsic angular momentum) of the center of inertia for the gravitational field and matter. The velocity of this motion is given by (2.4.109) and (2.5.99), and the coordinates of the center of inertia are R xα(t00 + T 00 + U 00)(−g)dV Xα = . (2.5.106) R (t00 + T 00 + U 00)(−g)dV These coordinates, like (2.4.110), are not the spatial components of a four-dimensional vector. The Einstein and Landau-Lifshitz pseudotensors are examples of quantities which in the absence ik ik of the gravitational field reduce to T +U , and which upon integration over dSk give a conservation of some quantity. There exists an infinite number of such pseudotensors, but the Landau-Lifshitz pseudotensor is the only one which contains only the first derivatives of gik and is also symmetric.

2.5.8 Palatini variation Instead of varying the action for the gravitational ﬁeld and matter with respect to the torsion j tensor, we can vary it with respect to the aﬃne connection (δΓi k is a tensor) and use the metric k compatibility of the connection (1.4.5). Varying Sg in (2.5.1) with respect to Γi j gives, by means of (1.3.44), 1 Z 1 Z √ δ S = − δR gikdΩ = − (δΓ l ) − (δΓ l ) − 2Sj δΓ l gik −gdΩ. (2.5.107) Γ g 2κc ik 2κc i k ;l i l ;k lk i j Partial integration and omitting total derivatives in (2.5.107) gives, using (1.2.43), 1 Z δ S = (δΓ l gik − 2S δΓ l gik − δΓ l gik + 2S δΓ l gik + 2Sj δΓ l gik)dΩ. (2.5.108) Γ g 2κc i k ;l l i k i l ;k k i l lk i j The variation of the action is thus 1 Z δ S = (δΓ l gik − 2S δΓ l gik − δΓ l gik + 2S δΓ l gik + 2Sj δΓ l gik)dΩ Γ 2κc i k ;l l i k i l ;k k i l lk i j 1 Z + Πi kδΓ j dΩ, (2.5.109) 2c j i k where i k δLm Π j = 2 j . (2.5.110) δΓi k Since the connection is metric-compatible, the condition δS = 0 gives κ gikS − δkSi + Ski = √ Πi k. (2.5.111) j j j 2 −g j

Comparing (2.5.111) with the second Einstein-Cartan equation (2.5.27) shows that

i k i k ik Π j = −S j = −Πj . (2.5.112) Contracting the indices i, j gives i k Π i = 0, (2.5.113) which also results from the invariance of the Lagrangian density under a projective transformation (1.2.61) (the symmetric part of the Ricci tensor is invariant under this transformation): 1 1 δL = δL = Πi kδΓ j = Πi kδjδA = 0. (2.5.114) m 2 j i k 2 j i k

106 The antisymmetry relation in (2.5.112) algebraically constrains possible forms of matter Lagrangians. Thereby, it is not a conservation law. If the matter Lagrangian density Lm does not depend on the affine connection, then the variation of the action with respect to the connection is referred to as the Palatini variation. In this case, (2.5.111) turns the torsion tensor into zero, so the connection is formed by the Christoffel symbols and the field equations are the Einstein equations (2.5.36).

2.5.9 Gravitational potential

If the metric tensor gij is approximately equal to the Minkowski metric tensor ηij, then the corresponding gravitational ﬁeld is weak. We can write 2φ g ≈ 1 + , (2.5.115) 00 c2 where φ is referred to as the gravitational potential. Therefore, nonrelativistic gravitational ﬁelds, corresponding to the limit c → ∞, are weak. Also u0 ≈ 1 and uα ≈ 0. In this limit, the leading component of the Levi-Civita connection is 1 ∂g 1 ∂φ { α } ≈ − gαβ 00 = , (2.5.116) 0 0 2 ∂xβ c2 ∂xα so the metric geodesic equation (1.4.91) reduces to

dv = g = −∇φ, (2.5.117) dt where g is the acceleration due to gravity. The quantity G in (2.2.7) reduces to 2 G = (∇φ)2. (2.5.118) c4 The leading component of the Riemannian Ricci tensor is

∂{ α } 1 ∂2φ 1 P ≈ 0 0 = = 4φ. (2.5.119) 00 ∂xα c2 ∂xα2 c2 The leading component of the energy-momentum tensor (2.4.153) is

2 T00 = µc . (2.5.120)

Therefore, the Einstein equations in the nonrelativistic limit reduce to the Poisson equation:

4φ = 4πGµ, (2.5.121)

where c4κ G = (2.5.122) 8π is Newton’s gravitational constant. In vacuum, where µ = 0, the Poisson equation reduces to the Laplace equation: 4φ = 0. (2.5.123) 0 α vα 2 In the nonrelativistic limit of an ideal ﬂuid, we have c → ∞, γ ≈ 1, u ∼ 1, u ≈ c , ≈ µc , and p . Consequently, the equation of continuity (2.4.209) reduces to

∂µ + div s = 0, (2.5.124) ∂t where s = µv (2.5.125)

107 is referred to as the mass current. Integrating (2.5.124) over the volume gives ∂ Z I µdV + s · df = 0. (2.5.126) ∂t This equation means that the change in time of the total mass inside a volume, m = R µdV , is balanced by the mass flux through the surface bounding this volume, representing the conservation of the total mass of a fluid. The Euler equation (2.4.210) reduces in this limit to ∂vα µ + vα vβ = µφ ηαβ + p ηαβ (2.5.127) ∂t ,β ,β ,β or dv ∂v µ = µ + (v · ∇)v = −µ∇φ − ∇p. (2.5.128) dt ∂t In the nonrelativistic limit, the total momentum (2.4.144) of a fluid is Z P = mv = µvdV. (2.5.129)

Integrating (2.5.128) over the volume gives its change in time: dP Z I = − ∇φdm − pdf. (2.5.130) dt Without pressure gradients, (2.5.128) reduces to (2.5.117).

2.5.10 Raychaudhuri equation Let us consider a congruence of particles with four-velocity ui. We deﬁne the expansion scalar θ, the traceless shear tensor σik, and the antisymmetric vorticity tensor ωik according to

i θ = u :i, (2.5.131) 1 σ = u − θh − w u , (2.5.132) ik (i:k) 3 ik (i k) ωik = u[i:k] − w[iuk], (2.5.133) where wi is the four-acceleration (1.6.121). Expansion has θ > 0 and contraction has θ < 0. These definitions give 1 u hj = σ + ω + θh . (2.5.134) i:j k ik ik 3 ik i i i l k j k k l Contracting u :jk − u :kj = P lkju with respect to the indices i, j gives θ:ku − u :kju = −Pklu u or k l k i k:i − Pklu u = θ:ku − w :i + ui:ku . (2.5.135) Defining 1 1 σ2 = σ σik, ω2 = ω ωik (2.5.136) 2 ik 2 ik gives 1 u uk:i = 2(σ2 − ω2) + θ2, (2.5.137) i:k 3 which brings (2.5.135) to the Raychaudhuri equation: dθ 1 = −2(σ2 − ω2) − θ2 + wi − P uiuk. (2.5.138) ds 3 :i ik For a perfect fluid, the Einstein equations give κ P uiuk = ( + 3p). (2.5.139) ik 2

108 We deﬁne four energy conditions. The null energy condition is satisﬁed if

i j Tijk k ≥ 0 (2.5.140) for any null, future-pointing vector ki. For a perfect ﬂuid, this condition gives

+ p ≥ 0. (2.5.141)

The weak energy condition is satisﬁed if

i j Tiju u ≥ 0 (2.5.142)

for any causal (null or timelike), future-pointing vector ui. For a perfect ﬂuid, this condition gives

+ p ≥ 0, ≥ 0. (2.5.143)

The strong energy condition is satisﬁed if 1 T − T g uiuj ≥ 0 (2.5.144) ij 2 ij

for any causal, future-pointing vector ui. For a perfect ﬂuid, this condition gives

+ p ≥ 0, + 3p ≥ 0. (2.5.145)

j The dominant energy condition is satisfied if the weak condition is satisfied and Tiju is a causal, future-pointing vector. For a perfect fluid, this condition gives

+ p ≥ 0, ≥ |p|. (2.5.146)

This condition guarantees that particles in a congruence do not move faster than light. The dominant condition infers the weak condition. The weak condition infers the null condition. The strong condition infers the null condition. If the strong condition is satisﬁed and particles in a congruence move without rotation and i acceleration (ωik = w = 0), then (2.5.138) and the Einstein equations give dθ c ≤ − θ2. (2.5.147) dτ 3 If θ is θ at τ = 0 then 0 cτ θ−1 ≥ θ−1 + . (2.5.148) 0 3

Therefore, if θ0 < 0 (initial contraction) then θ diverges to a curvature singularity, which is a point in spacetime where the density of matter and curvature are inﬁnite, as τ increases: θ → −∞. Singularities are unphysical and their appearance in a system indicates that a theory describing such a system is incomplete.

2.5.11 Relativistic spin ﬂuids For a spin ﬂuid, substituting (2.4.177) into the second Einstein-Cartan equation (2.5.17) and using (2.4.179) gives ∗ Si = 0, ∇i = ∇i. (2.5.149) Therefore, the corresponding metric energy-momentum tensor (2.4.221) reduces to 1 T = u u − ph − ∇ sk + ∇ s kulu − ∇ s k. (2.5.150) ij i j ij k (ij) k il j 2 k ij Accordingly, the torsion tensor is 1 Sj = − κs uj. (2.5.151) ik 2 ik

109 The last two terms on the right of the second line of (2.5.150) can be written, using (2.4.217), (2.4.219) and (2.4.179), as 1 1 ∇ s kulu − ∇ s k = c(p u − p u )ulu − c(p u − p u ) k il j 2 k ij i l l i j 2 i j j i 1 1 = c(p u + p u ) − u u = (∇ s kulu + ∇ s kulu ) = −∇ s luku 2 i j j i i j 2 k il j k jl i l k(i j) l k l k l k = −∇l(sk(iu )u uj) = −∇lsk(iu u uj) = −∇l(sk(iuj))u u k l = −∇l(s (iuj))uku . (2.5.152) The metric energy-momentum tensor (2.5.150) is then

ij i j ij l l k(i j) T = u u − ph − (δk + uku )∇l(s u ). (2.5.153)

The last term on the right of (2.5.153) can be decomposed according to (1.4.34) into

l l k(i j) l l {} k(i j) l l k m(i j) −(δk + uku )∇l(s u ) = −(δk + uku )∇l (s u ) − (δk + uku )(C mls u i k(m j) j k(i m) +C mls u + C mls u ). (2.5.154) This term reduces, by means of (2.4.179) and (2.5.15), to

l l {} k(i j) l i k(m j) j k(i m) l k m(i j) −(δk + uku )∇l (s u ) − δk(C mls u + C mls u ) − uku C mls u l l {} k(i j) i k(m j) j k(i m) = −(δk + uku )∇l (s u ) − C mks u − C mks u 1 = −(δl + u ul)∇{}(sk(iuj)) + κ(s ui + s iu + s iu )sk(muj) k k l 2 mk k m m k 1 + κ(s uj + s ju + s ju )sk(mui) 2 mk k m m k 1 = −(δl + u ul)∇{}(sk(iuj)) − κ(s skluiuj − siksj ). (2.5.155) k k l 2 kl k Therefore, (2.5.153) becomes 1 T ij = uiuj − phij − (δl + u ul)∇{}(sk(iuj)) − κs2uiuj + κsiksj , (2.5.156) k k l 2 k where 1 s2 = sijs > 0. (2.5.157) 2 ij Substituting (2.4.177) into (2.5.19) and using (2.4.179) gives 1 1 1 U ij = κs2uiuj + κs2gij − κsiksj . (2.5.158) 2 4 2 k Adding (2.5.156) and (2.5.158) brings the combined energy-momentum tensor T ij + U ij in the first Einstein-Cartan equation (2.5.10) to 1 1 T ij + U ij = − κs2 uiuj − p − κs2 hij − (δl + u ul)∇{}(sk(iuj)). (2.5.159) 4 4 k k l If the spin orientation of particles in a spin fluid is random then the macroscopic spacetime average of sij and of its gradients, such as of the last term on the right of (2.5.159), vanish. On the contrary, the terms that are quadratic in the spin tensor do not vanish after averaging. Therefore, the combined energy-momentum tensor of a macroscopic spin fluid describes a perfect fluid with the effective energy density 1 ˜ = (T + U )uiuj = − κs2 (2.5.160) ij ij 4

110 and the effective pressure 1 1 p˜ = − (T + U )hij = p − κs2. (2.5.161) 3 ij ij 4 If the spin orientation of particles (confer (2.6.46)) in a spin fluid is not random then the combined energy density of a macroscopic spin fluid is 1 1 ˜ = − κs2 − (δl + u ul)ui∇{}sk = − κs2 − ∇{}sk ui 4 k k l i 4 k i 1 1 = − κs2 + ski∇{}u = − κs2 + ski∂ u . (2.5.162) 4 [k i] 4 [k i] In a locally Galilean frame of reference which is also a rest frame, (2.5.162) becomes 1 1 ˜ = − κs2 + s · curl v. (2.5.163) 4 2 where s is the spatial spin-density pseudovector (2.4.181). The effective energy density (2.5.160) and pressure (2.5.161) can be negative if the quantity s2 is sufficiently large. Consequently, a spin fluid could violate the strong energy condition (2.5.145) and thus prevent a singularity. References: [1, 2, 3, 4, 6, 7, 9, 11].

2.6 Spinor ﬁelds 2.6.1 Dirac matrices The Dirac matrices γa deﬁned by (1.7.1) are complex. A particular solution of (1.7.1) is given by the Dirac representation:

α 0 I2 0 α 0 σ γ = , γ = α , (2.6.1) 0 −I2 −σ 0

where I2 is the unit 2×2 matrix and 0 1 0 −i 1 0 σ1 = , σ2 = , σ3 = (2.6.2) 1 0 i 0 0 −1

are the Pauli matrices (all indices are coordinate invariant). The Pauli matrices are traceless, tr(σα) = 0, and Hermitian, σα† = σα (the Hermitian conjugation of a matrix A is the combination of the complex conjugation and transposition, A† = A∗T ). They also satisfy

σασβ = δαβ + iεαβγ σγ , (2.6.3)

2 and their squares are (σα) = I2. The identity (2.6.3) gives the anticommutation relation hσ σ i σ α , β = iε γ , (2.6.4) 2 2 αβγ 2

σα so 2 form the lowest, two-dimensional representation of the angular momentum operator Mα (1.6.77). The properties of σα infer that the Dirac matrices are traceless, tr(γa) = 0, and satisfy

γa† = γ0γaγ0, γa∗ = γ2γaγ2, (2.6.5)

which gives γ0† = γ0 and γα† = −γα. Hereinafter, γ0, γ1, γ2 and γ3 refer to the Dirac matrices with a Lorentz index, γa. The relation (1.7.1) yields the total antisymmetry of γ0γ1γ2γ3:

γ0γ1γ2γ3 = γ[0γ1γ2γ3]. (2.6.6)

We deﬁne i γ5 = − e γaγbγcγd = iγ0γ1γ2γ3, (2.6.7) 24 abcd

111 which is traceless, tr(γ5) = 0, and Hermitian, γ5† = γ5. It also satisﬁes

a 5 5 2 5 {γ , γ } = 0, (γ ) = I4, γ |i = 0, (2.6.8) where the last relation results from (1.4.22) and (1.7.24). In the Dirac representation,

0 I γ5 = γ5∗ = 2 . (2.6.9) I2 0

The anticommutation relation (1.7.1) gives

a γ γa = 4I4, (2.6.10) a b b γ γ γa = −2γ , (2.6.11) a b c bc γ γ γ γa = 4η I4, (2.6.12) a b c d d c b γ γ γ γ γa = −2γ γ γ , (2.6.13) a b c ab c bc a ac b abcd 5 γ γ γ = η γ + η γ − η γ + ie γdγ , (2.6.14) {γa, γ[bγc]} = 2γ[aγbγc]. (2.6.15)

The Dirac representation is not unique; the relation (1.7.1) is invariant under a similarity transformation γa → SγaS−1, where S is a nondegenerate (det S 6= 0) matrix. Accordingly, ψ → Sψ and I −I ψ¯ → ψS¯ −1. Taking S = √1 2 2 turns the Dirac representation into the chiral or Weyl 2 I2 I2 representation, in which

α 0 0 I2 α 0 σ 5 −I2 0 γ = , γ = α , γ = . (2.6.16) I2 0 −σ 0 0 I2

For an inﬁnitesimal Lorentz transformation (1.6.7), the relations (1.7.5) and (1.7.6) give L = 1 a b b a † † † I4 + 8 ab(γ γ − γ γ ). Using (AB) = B A gives 1 L† = I + (γb†γa† − γa†γb†), (2.6.17) 4 8 ab which is equal to L−1 (so L is unitary) for rotations and equal to L for boosts. The relation (2.6.5) leads thus to 1 1 L†γ0 = γ0 + (γb†γa† − γa†γb†)γ0 = γ0 − γ0(γaγb − γbγa) = γ0L−1. (2.6.18) 8 ab 8 ab Therefore, the quantity ψ†γ0 transforms under (1.7.7) like an adjoint spinor:

ψ†γ0 → ψ†L†γ0 = ψ†γ0L−1. (2.6.19)

Accordingly, we can associate these two quantities:

ψ¯ = ψ†γ0. (2.6.20)

The spinors ψ and ψ†γ0 can be used to construct tensors, as in (1.7.11): ψ†γ0ψ transforms like a scalar, ψ†γ0γiψ is a vector, ψ†γ0γ[iγj]ψ is an antisymmetric tensor, ψ†γ0γ5ψ is a pseudoscalar, and ψ†γ0γiγ5ψ is a pseudovector. Higher-rank tensors constructed from ψ and ψ†γ0 reduce to the above 5 kinds of tensors because of (2.6.14). To show that ψγ¯ 5ψ transforms like a pseudoscalar, we substitute (1.7.4) into (2.6.7) and use (2.6.6), which gives

5 0 1 2 3 a b c d −1 0 1 2 3 [a b c d] −1 γ = iΛ aΛ bΛ cΛ dLγ γ γ γ L = iΛ aΛ bΛ cΛ dLγ γ γ γ L . (2.6.21)

Using (1.1.25) and γ[aγbγcγd] = −ieabcdγ5, (2.6.22)

112 which results from (2.6.7), gives

5 abcd 0 1 2 3 5 −1 a 5 −1 γ = e Λ aΛ bΛ cΛ dLγ L = det(Λ b)Lγ L . (2.6.23)

Therefore, we have ¯ 5 ¯ −1 a 5 −1 a ¯ 5 ψγ ψ → ψL det(Λ b)Lγ L Lψ = det(Λ b)ψγ ψ, (2.6.24) which is the transformation law for a Lorentz scalar density. Similarly,

¯ c 5 ¯ −1 c a d 5 −1 a c ¯ d 5 ψγ γ ψ → ψL Λ ddet(Λ b)Lγ γ L Lψ = det(Λ b)Λ dψγ γ ψ, (2.6.25) which is the transformation law for a Lorentz vector density. We deﬁne the chirality projection operators

I ± γ5 P = 4 ,P + P = I ,P 2 = I ,P P = P P = 0. (2.6.26) ± 2 + − 4 ± 4 + − − +

They project a spinor ψ into the right-handed spinor ψR and left-handed spinor ψL,

ψR = P+ψ, ψL = P−ψ, ψ = ψR + ψL. (2.6.27)

If we split a spinor ψ into two two-component parts,

u ψ = , (2.6.28) v then, in the chiral representation:

u 0 ψ = , ψ = . (2.6.29) L 0 R v

An inﬁnitesimal rotation is described by a Lorentz matrix (1.6.7) with 0α = 0. The corresponding spinor transformation matrix (1.7.5) in the Dirac representation is, using (1.6.25),

1 1 i σα 0 L = I + γαγβ = I + e γαγβϑ = I − ϑ . (2.6.30) 4 4 αβ 4 4 αβγ γ 4 2 α 0 σα

Therefore, the (unitary) spinor transformation matrix for a ﬁnite rotation by an angle ϑ about an axis parallel to a unit vector n, ϑ = ϑn, is " # i σ 0 ϑ ϑ σ 0 L = exp − ϑ · = cos I − i sin n · , (2.6.31) 2 0 σ 2 4 2 0 σ where σ is a spatial vector composed from the Pauli matrices σα. If we split a spinor ψ into two parts u and v (2.6.28), then both u and v transform under rotations according to

u → Su, v → Sv, (2.6.32) where ϑ ϑ S = cos I − i sin n · σ. (2.6.33) 2 2 2 A rotation by a full angle 2π changes the sign of a spinor. A rotation by 4π brings a spinor to its original position. An inﬁnitesimal boost is described by a Lorentz matrix (1.6.7) with αβ = 0. The corresponding spinor transformation matrix (1.7.5) in the Dirac representation is, using (1.6.26),

1 1 1 0 σα L = I + γ0γα = I + η γ0γα = I + η . (2.6.34) 4 2 0α 4 2 α 4 2 α σα 0

113 Therefore, the spinor transformation matrix for a ﬁnite boost with a rapidity η along an axis parallel to a unit vector n, η = ηn, is " # 1 0 σ η η 0 σ L = exp η · = cosh I + sinh n · . (2.6.35) 2 σ 0 2 4 2 σ 0

Using (1.6.92) gives 1 (1 + γ)I2 γβ · σ L = p . (2.6.36) 2(1 + γ) γβ · σ (1 + γ)I2 If this boost transforms a particle of mass m from rest to a motion with momentum p and energy E then (2.6.36) is equivalent, because of (2.4.143) and (2.4.144), to

2 1 (E + mc )I2 cσ · p L = p 2 . (2.6.37) 2mc2(E + mc2) cσ · p (E + mc )I2

2.6.2 Lagrangian density and spin tensor for spinor field A Lagrangian density for dynamical spinor fields must contain first derivatives of spinors and be ¯ i real. The simplest scalar containing derivatives of spinors is quadratic in ψ, ψγ ψ;i, where ψ;i is the covariant derivative of ψ (1.7.14). Its kinetic part, containing derivatives of spinors, is equal to ¯ i ψγ ψ,i and is complex. Since the complex conjugate of this quantity is

¯ i ∗ ¯ i † † i† ¯† ¯ 0 i† 0 ¯ i (ψγ ψ,i) = (ψγ ψ,i) = ψ,iγ ψ = ψ,iγ γ γ ψ = ψ,iγ ψ, (2.6.38)

¯ i ¯ i ¯ i ¯ i both scalars ψγ ψ,i + ψ,iγ ψ and i(ψγ ψ,i − ψ,iγ ψ) are real. The former scalar is, however, equal ¯ i to a total divergence (ψγ ψ),i, thereby a Lagrangian density proportional to such term does not contribute to ﬁeld equations. Therefore, the simplest kinetic part of a spinor Lagrangian density is ¯ i ¯ i proportional to i(ψγ ψ,i − ψ,iγ ψ). Another scalar that can be used in a spinor Lagrangian density is proportional to ψψ¯ , which is real:

(ψψ¯ )∗ = (ψψ¯ )† = (ψ†γ0ψ)† = ψ†γ0ψ†† = ψψ.¯ (2.6.39)

The simplest Lagrangian density for a spinor ﬁeld is thus given by ie ie L = (ψγ¯ iψ − ψ¯ γiψ) − meψψ¯ = ei (ψγ¯ aψ − ψ¯ γaψ) − meψψ,¯ (2.6.40) ψ 2 ;i ;i 2 a ;i ;i where m is a real scalar constant, called the spinor mass. It is referred to as the Dirac Lagrangian density. Putting the deﬁnition of the covariant derivative of a spinor (1.7.14) into (2.6.40) gives ie ie L = (ψγ¯ iψ − ψ¯ γiψ) − ψ¯{γi, Γ }ψ − meψψ.¯ (2.6.41) ψ 2 ,i ,i 2 i

Using the Fock-Ivanenko coeﬃcients (1.7.28) as the spinor connection Γi turns (2.6.41) into ie ie L = (ψγ¯ iψ − ψ¯ γiψ) + ω ψ¯{γi, γaγb}ψ − meψψ¯ ψ 2 ,i ,i 8 abi ie ie = (ψγ¯ iψ − ψ¯ γiψ) + ω ψ¯{γi, γ[aγb]}ψ − meψψ.¯ (2.6.42) 2 ,i ,i 8 abi The spin density (2.3.16) corresponding to the Dirac Lagrangian density (2.6.42) is, by means of (2.6.15), δL ie Sijk = 2 ψ = ψγ¯ [iγjγk]ψ. (2.6.43) δωijk 2 The spin density (2.6.43) is completely antisymmetric,

Sijk = S[ijk], (2.6.44)

114 and independent of the spinor mass m. The corresponding spin tensor is also completely antisymmetric, i sijk = ψγ¯ [iγjγk]ψ = s[ijk] = −eijkls , (2.6.45) 2 l where 1 si = ψγ¯ iγ5ψ (2.6.46) 2 is the spin-density pseudovector. The second Einstein-Cartan equation (2.5.17) with the spin tensor (2.6.45) gives a completely antisymmetric torsion tensor,

iκ S = − ψγ¯ γ γ ψ, (2.6.47) ijk 4 [i j k] so Si = 0. Therefore, the contortion tensor is, using (2.6.14), κ κ C = S = e ψγ¯ lγ5ψ = e sl. (2.6.48) ijk ijk 4 ijkl 2 ijkl The torsion tensor is dual to a pseudovector that is proportional to the Dirac spin-density pseudovector (2.6.46). The pseudovector density

i ¯ i 5 i jA = eψγ γ ψ = 2es (2.6.49)

is called the axial Dirac current. The axial Dirac current and the pseudovector (2.6.46) are real because

(ψγ¯ iγ5ψ)∗ = (ψ†γ0γiγ5ψ)† = ψ†γ5γi†γ0ψ = ψ†γ5γ0γiψ = ψ†γ0γiγ5ψ = ψγ¯ iγ5ψ. (2.6.50)

Therefore, the spin tensor (2.6.45) is also real, in accordance with the reality of (2.6.40). For a system of spinor fields, the Lagrangian density is equal to the sum of the Lagrangian densities (2.6.40) for each spinor and a Lagrangian density representing fields that carry the interaction between the spinors. The Lagrangian density for the interaction fields does not depend on derivatives of spinors. Since Cijk appears only in the additive, kinetic term in the Lagrangian density, ie i ¯ a ¯ a 2 ea(ψγ ψ;i − ψ;iγ ψ), the spin tensor for a system of spinor fields is additive. Accordingly, the spin tensor for such a system is also completely antisymmetric. The complete antisymmetry of the spin density (2.6.44) leads to (2.4.188), which is consistent with (2.6.45) only if ψ = 0. Therefore, a spinor field cannot be a point particle or a system of point particles.

2.6.3 Dirac equation Varying (2.6.41) with respect to ψ¯ and omitting total divergences gives

i δL = δψ¯eγkψ + (eγkψ) − e{Γ , γk}ψ − emδψψ.¯ (2.6.51) ψ 2 ,k ,k k The stationarity of the action δS = 0 under δψ¯ therefore gives i eγkψ + (eγkψ) − e{Γ , γk}ψ − emψ = 0. (2.6.52) 2 ,k ,k k Substituting k k k k k k (eγ ψ),k = eγ ψ,k + eγ ;kψ − 2eSkγ ψ = eγ ψ,k + e[Γk, γ ]ψ (2.6.53) into (2.6.52) gives the Dirac equation:

k k k iγ ψ,k − iγ Γkψ − mψ = iγ ψ;k − mψ = 0. (2.6.54)

115 Putting (2.6.48) into (1.7.34) yields

κ iκ γkψ = γkψ + e (ψγ¯ lγ5ψ)γkγiγjψ = γkψ + e (ψγ¯ lγ5ψ)eijkmγ γ5ψ ;k :k 16 ijkl :k 16 ijkl m 3iκ = γkψ − (ψγ¯ lγ5ψ)γ γ5ψ. (2.6.55) :k 8 l Therefore, the Dirac equation (2.6.54) becomes 3κ iγkψ + (ψγ¯ γ5ψ)γkγ5ψ = mψ. (2.6.56) :k 8 k Varying (2.6.41) with respect to ψ and using the stationarity of the action under δψ gives the adjoint Dirac equation: ¯ k ¯ k ¯ ¯ k ¯ − iψ,kγ − iψΓkγ − mψ = −iψ;kγ − mψ = 0, (2.6.57) which is equivalent to the adjoint conjugate of (2.6.56), 3κ − iψ¯ γk + (ψγ¯ γ5ψ)ψγ¯ kγ5 = mψ.¯ (2.6.58) :k 8 k Equations (2.6.56) and (2.6.58) are cubic in spinor fields. They are also the field equations corresponding to an effective metric Lagrangian density with a quartic, axial-axial spin-spin interaction: ie 3κe L{} = (ψγ¯ iψ − ψ¯ γiψ) − meψψ¯ + (ψγ¯ γ5ψ)(ψγ¯ kγ5ψ). (2.6.59) ψ 2 :i :i 16 k Subtracting (2.6.57) multiplied by ψ from (2.6.54) multiplied by ψ¯ gives, using (1.7.18) and (1.7.39), 1 (ψγ¯ iψ) = (ψγ¯ iψ) = (eψγ¯ iψ) = 0. (2.6.60) |i :i e ,i The vector density i ¯ i jV = eψγ ψ, (2.6.61) called the vector Dirac current, is thus conserved:

∂j0 ji = V + ∂ jα = 0. (2.6.62) V,i c∂t α V The vector Dirac current is real because

(ψγ¯ iψ)∗ = (ψ†γ0γiψ)† = ψ†γi†γ0ψ = ψ†γ0γiψ = ψγ¯ iψ. (2.6.63)

Its time component, 0 ¯ 0 † jV = eψγ ψ = eψ ψ, (2.6.64) is thus real and positive. The Dirac equation (2.6.54) gives

j k j − γ (γ ψ;k)|j = imγ ψ;j (2.6.65) or, because of (1.7.39),

j k (j k) j k 2 − γ γ ψ;kj = −γ γ ψ;kj − γ γ ψ|[kj] = m ψ. (2.6.66)

The relations (1.7.36) and (1.7.45) turn (2.6.66) into 1 1 ψ i + m2ψ = γjγk(K ψ + 2Sl ψ ) = γjγkγiγlR ψ + γjγkSl ψ , (2.6.67) ;i 2 kj kj ;l 8 ilkj kj ;l

116 where Kij is the curvature spinor. If a spinor is equal either to its left-handed projection, ψ = ψL, or right-handed projection, ψ = ψR, then it is called a Weyl spinor. Multiplying (2.6.54) by P± and using (2.6.8) gives i i iP±γ ψ;i = iγ P∓ψ;i = mP±ψ (2.6.68) or i L(R) R(L) iγ ψ;i = mψ . (2.6.69) Therefore, if ψ is a Weyl spinor then m = 0. The nonlinear, cubic terms in (2.6.56) and (2.6.58) represent a spinor self-interaction, corresponding to a spin-spin interaction in the tensor (2.5.19). At densities satisfying κψψ¯ m, these equations can be approximated by linear equations:

k iγ ψ:k = mψ, (2.6.70) ¯ k ¯ −iψ:kγ = mψ. (2.6.71) Equation (2.6.67) reduces to the Klein-Gordon-Fock equation: 1 ψ i + m2ψ = γjγkγiγlP ψ, (2.6.72) :i 8 ilkj

where Pijkl is the Riemann tensor.

2.6.4 Energy-momentum tensor for spinor ﬁeld

i Varying the Lagrangian density (2.6.40) with respect to ea gives the tetrad energy-momentum density for a Dirac spinor field, ie T a = (ψγ¯ aψ − ψ¯ γaψ − eaψγ¯ jψ + eaψ¯ γjψ) + meeaψψ.¯ (2.6.73) i 2 ;i ;i i ;j i ;j i The conservation law (2.4.39) applied to the energy-momentum density (2.6.73) gives the Dirac equations (2.6.54) and (2.6.57). We can obtain the metric energy-momentum tensor for such a field using the Belinfante-Rosenfeld relation (2.3.34), the spin tensor (2.6.45), and the torsion tensor in (2.6.48). We can also derive the metric energy-momentum tensor by varying (2.6.40) with respect to gik: 2 δL T = √ ψ . (2.6.74) ik −g δgik Using an identity δγj 1 = δj γ , (2.6.75) δgik 2 (i k) which results from the definition of the Dirac matrices (1.7.2), leads to i T = (ψγ¯ ψ − ψ¯ γ ψ − g ψγ¯ jψ + g ψ¯ γjψ) + mg ψψ.¯ (2.6.76) ik 2 (i ;k) ;(i k) ik ;j ik ;j ik The conservation law (2.4.30) applied to the energy-momentum tensor (2.6.76) gives the Dirac equations (2.6.54) and (2.6.57). Substituting the Dirac equation (2.6.54) into the (2.6.76) gives i T = (ψδ¯ j γ ψ − ψ¯ δj γ ψ). (2.6.77) ik 2 (i k) ;j ;j (i k) Putting (1.7.34) and (1.7.35) with (2.6.48) into (2.6.77) yields i κ T = (ψδ¯ j γ ψ − ψ¯ δj γ ψ) + (−s s + sls g ). (2.6.78) ik 2 (i k) :j :j (i k) 2 i k l ik Substituting the completely antisymmetric spin tensor (2.6.45) into (2.5.19) gives κ 1 κ U ik = sijlsk − giksjlms = (2sisk + sls gik). (2.6.79) 4 jl 2 jlm 4 l

117 The combined energy-momentum tensor for a Dirac field is thus i 3κ T + U = (ψδ¯ j γ ψ − ψ¯ δj γ ψ) + sls g . (2.6.80) ik ik 2 (i k) :j :j (i k) 4 l ik The tensor (2.6.80) is equal to the energy-momentum tensor for the effective metric Lagrangian density (2.6.59): {} 2 δL T = √ ψ . (2.6.81) ik −g δgik The first term on the right of (2.6.80) is the Riemannian part of the energy-momentum tensor for a Dirac field and can be macroscopically averaged as an ideal fluid with the energy density and 0 i pressure p. In the comoving frame of reference, in which s = 0 because of siu = 0, the second term 3κ 2 on the right of (2.6.80) is equal to − 4 s gik, where s is the spatial spin pseudovector. The average value of s2 is proportional to n2, where n is the concentration of spinor particles. The averaged second term on the right of (2.6.80) has therefore a negative contribution to the energy density, 0 > ˜ ∝ n2, and a positive contribution to the pressure,p ˜ = −˜.

2.6.5 Discrete symmetries of spinors The spinor representation of the parity transformation (1.6.3) is given by

0 LP = CP γ , (2.6.82)

where CP = const. Substituting (2.6.82) into (1.7.4) with L = LP gives, using (2.6.5),

a a 0 b 0 a b† γ = Λ bγ γ γ = Λ bγ , (2.6.83)

a a which is satisﬁed if Λ b = Λ b(P ). Since the double parity transformation is equivalent to the 2 identity transformation, LP = I4, we have CP = ±1. The spinor representation of the time-reversal transformation (1.6.4) is given by 0 5 LT = CT γ γ , (2.6.84) where CT = const. Substituting (2.6.84) into (1.7.4) with L = LT gives, using (2.6.5) and (2.6.8),

a a 0 5 b 5 0 a b† γ = Λ bγ γ γ γ γ = −Λ bγ , (2.6.85)

a a which is satisﬁed if Λ b = Λ b(T ). Since the double time-reversal transformation is equivalent to 2 the identity transformation, LT = I4, we have CT = ±i. The charge conjugation of a spinor ψ is deﬁned as

ψc = −iγ2ψ∗, ψ∗ = −iγ2ψc. (2.6.86)

The double charge-conjugation transformation is equivalent to the identity transformation:

(ψc)c = −iγ2(ψc)∗ = −iγ2(−iγ2ψ∗)∗ = γ2γ2∗ψ = ψ. (2.6.87)

The charge conjugation of the left-handed projection of a spinor is the right-handed projection of the charge conjugation of the spinor and vice versa (confer (2.6.27)):

c ∗ 5 2 5 2 5 ∗ 5 2 ∗ (I4 ∓ γ )ψ = −iγ (I4 ∓ γ )ψ = −iγ (I4 ∓ γ )ψ = −i(I4 ± γ )γ ψ 5 c = (I4 ± γ )γ . (2.6.88)

References: [3, 4, 7, 12].

118 2.7 Electromagnetic field 2.7.1 Gauge invariance and electromagnetic potential The Lagrangian density (2.6.40) is a real combination of the complex Dirac matrices γi and spinors ψ, ψ¯. We define a gauge transformation of the first type of the spinor fields,

ψ → ψ0 = eieαψ, ψ¯ → ψ¯0 = e−ieαψ,¯ (2.7.1) where e is a real scalar constant, called the spinor electric charge. If α is a real scalar constant, then (2.6.40) is invariant under (2.7.1). Such a transformation is called global. If α = α(xi) is a scalar function of the coordinates, then (2.6.40) is not invariant under (2.7.1) because

0 ieα ψ;µ = e (ψ;µ + ieα,µψ). (2.7.2) Such a transformation is called local. For a local transformation, we must introduce a compensating vector ﬁeld Aµ, called the electromagnetic potential, such that the Weyl electromagnetic covariant derivative Dµ = ∇µ − ieAµ (2.7.3) of a spinor ψ, Dµψ = ψ;µ − ieAµψ, (2.7.4) transforms under (2.7.1) like ψ: 0 ieα Dµψ = e Dµψ. (2.7.5) This requirement gives 0 0 0 ieα ψ;µ − ieAµψ = e (ψ;µ − ieAµψ), (2.7.6) which, with (2.7.1) and (2.7.2), yields the transformation law for the electromagnetic potential,

0 Aµ → Aµ = Aµ + α,µ. (2.7.7) This law is called a gauge transformation of the second type. The adjoint conjugation of (2.7.4) is ¯ ¯ ∗ ¯ Dµψ = ψ;µ + ieAµψ. (2.7.8) The scalar ψψ¯ is invariant under (2.7.1), thereby ¯ ¯ ¯ ¯ Dµψψ + ψDµψ = Dµ(ψψ) = ∂µ(ψψ), (2.7.9) which constrains the electromagnetic potential to be real:

∗ Aµ = Aµ. (2.7.10) The time component of Aµ, φ = A0, is called the electric potential and the spatial components Aα form the magnetic potential A: Aµ = (φ, A). (2.7.11) The gauge transformation (2.7.7) reads ∂α φ0 = φ + , A0 = A − ∇α. (2.7.12) c∂t In the local Minkowski spacetime, Aµ transforms according to (1.6.101), ! φ γ γβ φ0 = (γ−1)β 0 . (2.7.13) A γβ 1 + β2 β A

The gauge-invariant modiﬁcation of the Dirac Lagrangian density (2.6.40) is ie ie L = ei (ψγ¯ aD ψ − D ψγ¯ aψ) − meψψ¯ = ei (ψγ¯ aψ − ψ¯ γaψ) − meψψ¯ − eA eψγ¯ iψ. (2.7.14) ψ 2 a i i 2 a ;i ;i i

119 The spin density corresponding to the Lagrangian density (2.7.14) remains equal to (2.6.43); it is independent of the spinor electric charge e. The electromagnetic potential corresponds, up to the multiplication by an arbitrary constant, to the vector multiple of I4 in the formula for the spinor connection (1.7.26). The electromagnetic potential is analogous to the affine connection: it modifies a derivative of a spinor so such derivative transforms like a spinor under unitary gauge transformations of the first type, while the connection modifies a derivative of a tensor so such derivative transforms like a tensor under coordinate transformations. The gauge-invariant modification of the Dirac equation (2.6.54) is

k k iγ ψ;k + eAkγ ψ = mψ, (2.7.15) whose adjoint conjugate generalizes (2.6.57):

¯ k ¯ k ¯ − iψ;kγ + eAkψγ = mψ. (2.7.16)

The gauge-invariant modiﬁcation of the Dirac equation (2.6.56) is 3κ iγkψ + eA γkψ = mψ − (ψγ¯ 5γ ψ)γ5γkψ, (2.7.17) :k k 8 k whose adjoint conjugate generalizes (2.6.58): 3κ − iψ¯ γk + eA ψγ¯ k = mψ¯ − (ψγ¯ 5γ ψ)ψγ¯ 5γk. (2.7.18) :k k 8 k Taking the complex conjugate of (2.7.17) gives, using (2.6.50), 3κ − iγk∗ψ∗ + eA γk∗ψ∗ = mψ∗ − (ψγ¯ 5γ ψ)γ5∗γk∗ψ∗. (2.7.19) :k k 8 k The relations (1.7.18) and (1.7.24) give

c c 2 ∗ 2 ∗ 2 ∗ ψ;k = ψ|k = (−iγ ψ )|k = −iγ ψ|k = −iγ ψ;k. (2.7.20) Substituting (2.6.86) and (2.7.20) into (2.7.19) gives, using (2.6.5) and (2.6.9), 3κ − iγ2γkγ2(−iγ2)ψc + eA γ2γkγ2(−iγ2)ψc = m(−iγ2)ψc − (ψγ¯ 5γ ψ)γ5γ2γkγ2(−iγ2)ψc. :k k 8 k (2.7.21) Using (2.6.8), the relation (2.7.21) becomes 3iκ γ2γkψc + ieA γ2γkψc = −imγ2ψc + (ψγ¯ 5γ ψ)γ2γ5γkψc. (2.7.22) :k k 8 k Multiplying (2.7.22) by −iγ2 from the left brings this equation to 3κ iγkψc − eA γkψc = mψc − (ψγ¯ 5γ ψ)γ5γkψc. (2.7.23) :k k 8 k The Hermitian conjugate of (2.6.86) gives

ψT = ψ∗† = iψc†γ2† = −iψc†γ2. (2.7.24)

Thus we obtain, using (2.6.50),

¯ 5 † 0 5 ∗ T 2 0 2 5 2 2 ∗ c† 2 2 0 2 5 2 2 2 c ψγ γkψ = (ψ γ γ γkψ) = ψ γ γ γ γ γ γkγ ψ = (−iψ γ )γ γ γ γ γ γkγ (−iγ ψ ) c† 0 2 5 2 c c 5 c = −ψ γ γ γ γ γkψ = −ψ γ γkψ . (2.7.25)

Substituting (2.7.25) into (2.7.23) gives the Dirac equation for the charge-conjugate spinor ﬁeld ψc: 3κ iγkψc − eA γkψc = mψc + (ψcγ5γ ψc)γ5γkψc. (2.7.26) :k k 8 k

120 Comparing (2.7.26) with (2.7.17) shows that ψ and ψc correspond to the same value of the mass m and to the opposite values of the electric charge, e and −e. Accordingly, the charge-conjugation transformation does not change the mass of a spinor, but changes the sign of its electric charge. The ﬁeld equations for ψ and ψc are asymmetric because of the opposite signs of the corresponding cubic terms relative to the mass terms. This asymmetry is related to the fact that the scalar ψψ¯ changes sign under the charge-conjugation transformation:

ψcψc = −ψψ,¯ (2.7.27) whereas the Lorentz square of ψγ¯ 5γkψ does not change sign:

c 5 k c c 5 c ¯ 5 k ¯ 5 (ψ γ γ ψ )(ψ γ γkψ ) = (ψγ γ ψ)(ψγ γkψ). (2.7.28)

The first two terms in the effective metric Lagrangian density (2.6.59) are thus antisymmetric under charge conjugation, while the last, four-fermion term is symmetric. Therefore, (ψ, m, e) and (ψc, m, −e) are asymmetric under the charge-conjugation transformation and do not satisfy the same field equation. Torsion generates an asymmetry between a spinor and its charge conjugate. At densities satisfying κψψ¯ m, the cubic terms in (2.7.17) and (2.7.26) can be neglected. In this approximation, (ψ, m, e) and (ψc, m, −e) are symmetric under the charge-conjugation transformation and satisfy the same field equation.

2.7.2 Electromagnetic ﬁeld tensor The commutator of total covariant derivatives of a spinor is given by (1.7.36) with the curvature spinor Kij given by (1.7.42), where the tensor Bij is related to the vector Ai in (1.7.26) by (1.7.44). Therefore, the commutator of the electromagnetic covariant derivatives of a spinor, [Di,Dj]ψ, is given by (1.7.36) with the curvature spinor 1 K = R γkγl + ieF I , (2.7.29) ij 4 klij ij 4 where the antisymmetric tensor

Fij = Aj,i − Ai,j = Aj:i − Ai:j (2.7.30) is referred to as the electromagnetic ﬁeld tensor. The electromagnetic ﬁeld tensor is analogous to the curvature tensor: it appears in the expression for the commutator of electromagnetic covariant derivatives of a spinor, while the curvature tensor appears in the expression for the commutator of coordinate-covariant derivatives of a tensor. Substituting (2.7.7) into (2.7.30) gives

0 Fij = Fij, (2.7.31) so the electromagnetic field tensor is gauge invariant. The definition (2.7.30) is equivalent to the first Maxwell-Minkowski equation

Fij,k + Fjk,i + Fki,j = Fij:k + Fjk:i + Fki:j = 0 (2.7.32) or ijkl ijkl Fjk,l = Fjk:l = 0. (2.7.33) We deﬁne a spatial vector E whose covariant components are related to the 0α components of the electromagnetic ﬁeld tensor (2.7.30):

Eα = F0α, (2.7.34)

and a spatial tensor Bαβ equal to the spatial part of (2.7.30): 1 √ B = F ,Bα = − √ αβγ B ,B = − sε Bγ , (2.7.35) αβ αβ 2 s βγ αβ αβγ

121 where s is given by (1.4.147). The component of (2.7.32) with all spatial indices, Bαβ,γ + Bβγ,α + Bγα,β = 0, gives, using (1.4.167), div B = 0. (2.7.36)

The components of (2.7.32) with one temporal index, Bαβ,0 +Eα,β −Eβ,α = 0, gives, using (1.4.168), √ 1 ∂( sB) curl E = − √ . (2.7.37) c s ∂t The spatial vector E is called the electric field and the spatial pseudovector B is the magnetic field. In the locally geodesic and Galilean frame of reference, these fields depend on the components of the electromagnetic potential (2.7.11) according to (2.7.30): ∂A E = − − ∇φ, (2.7.38) c∂t B = ∇ × A, (2.7.39)

and they are invariant under (2.7.12). The tensor Fij is given by   0 Ex Ey Ez  −Ex 0 −Bz By  Fij =   , E = (F01,F02,F03), B = (F32,F13,F21), (2.7.40)  −Ey Bz 0 −Bx  −Ez −By Bx 0 and transforms under Lorentz transformations according to (1.6.102). Therefore, the electric and magnetic ﬁelds transform according to 1 − γ E = γ(E0 − β × B0) + (β · E0)β, (2.7.41) β2 1 − γ B = γ(B0 + β × E0) + (β · B0)β. (2.7.42) β2 In this frame, (2.7.36) and (2.7.37) become the ﬁrst pair of the Maxwell equations:

div B = 0, (2.7.43) ∂B curl E = − . (2.7.44) c∂t Applying the div operator to (2.7.39) gives (2.7.43) and applying the curl operator to (2.7.38) gives (2.7.44). Applying the div operator to (2.7.44) gives (2.7.43). Integrating the ﬁrst pair of the Maxwell equations over the volume and surface area, respectively, gives I B · df = 0, (2.7.45)

I ∂ Z E · dl = − B · df . (2.7.46) c∂t

The integral H A · df is the flux of a vector A through the surface f and the integral H A · dl is called the circulation of A along the contour l. Therefore, the flux of the magnetic field through a closed surface vanishes and the circulation of the electric field along a contour, which is called the electromotive force, is equal to the minus time derivative of the flux of the magnetic field through the surface enclosed by this contour (Faraday’s law). In a locally Galilean frame of reference, the simplest invariants (under proper Lorentz transformations) of the electromagnetic field are quadratic in Fij:

ij 2 2 ijkl FijF = 2(B − E ) = const, e FijFkl = 8E · B = const. (2.7.47) If the vectors E and B are mutually perpendicular in frame K, E · B = 0, then they are mutually perpendicular in other inertial frames. If E and B are equal in magnitude in K, B2 − E2 = 0, then

122 they are equal in magnitude in other inertial frames. The transformation laws (2.7.41) and (2.7.42) imply that if E0 = 0 in the frame K0 then in the frame K

E = −γβ × B0 = −β × B, (2.7.48) and if B0 = 0 in the frame K0 then in the frame K

B = γβ × E0 = β × E. (2.7.49)

If the vectors E and B are mutually perpendicular in K, but not equal in magnitude, then there exists a frame K0 in which the ﬁeld is either electric, B0 = 0 (if E > B), or magnetic, E0 = 0 (if E < B). The velocity of K0 relative to K is perpendicular to E and B, and it is equal in magnitude B E to respectively either c E or c B . Equivalently, if one of the vectors E, B vanishes in one frame of reference then these vectors are mutually perpendicular in other inertial frames. Except for the case where the vectors E and B are mutually perpendicular and equal in magnitude, there exist frames in which these vectors are parallel to each other at a given point. These frames move relative to one another with velocities parallel to both vectors. One of such frames, K0 (in which E0 k B0), has a velocity V relative to K which is perpendicular to both vectors E and B. Substituting the formulae 1 − γ E0 = γ(E + β × B) + (β · E)β, (2.7.50) β2 1 − γ B0 = γ(B − β × E) + (β · B)β, (2.7.51) β2

which are inverse to (2.7.41) and (2.7.42), into the condition E0 × B0 = 0 and using β = kE × B, where k is a constant of proportionality, gives E + k(E × B) × B × B − k(E × B) × E = 0 or 1+β2 k = E2+B2 , thereby V E × B c = . (2.7.52) V 2 E2 + B2 1 + c2

2.7.3 Lagrangian density for electromagnetic field The simplest gauge-invariant Lagrangian density representing the electromagnetic field is a linear √ ij ijkl combination of terms quadratic in Fij: −gFijF and FijFkl, which a locally Galilean frame of reference reduce to (2.7.47). The second, parity-violating term is a total divergence because of (2.7.33): ijkl ijkl FijFkl = 2( FijAl),k, (2.7.53) so it does not contribute to the field equations. Therefore, the Lagrangian density for the electromagnetic field is given by 1 √ L = − −gF F ij, (2.7.54) EM 16π ij 1 where the Gaußian factor 16π sets the units of Ai. In the locally geodesic and Galilean frame of reference, (2.7.54) becomes 1 L = (E2 − B2). (2.7.55) EM 8π Therefore, in order for the action S to have a minimum, there must be the minus sign in front of the right-hand side of (2.7.54). Otherwise an arbitrarily rapid change of A in time would result in an arbitrarily large value of E, according to (2.7.38), and thus an arbitrarily low value of S, thereby the action would have no minimum. A generalization of the tensor (2.7.30) to the covariant k ˜ k derivatives with respect to the affine connection Γi j, Fij = Aj;i − Ai;j = Fij + 2S ijAk, is not gauge invariant, thereby the torsion tensor cannot appear in a gauge-invariant Lagrangian density which is quadratic in Fij. Therefore, the electromagnetic field, unlike spinor fields, does not couple to torsion. Accordingly, the electromagnetic field is not minimally coupled to the affine connection.

123 2.7.4 Electromagnetic current and electric charge We deﬁne the electromagnetic current density cδL ji = − m , (2.7.56) δAi and the electromagnetic current vector ji ji = √ . (2.7.57) −g 0 The invariance of the action under an arbitrary inﬁnitesimal gauge transformation δAi = Ai − Ai = φ,i gives, upon partial integration and omitting a total divergence, 1 Z 1 Z 1 Z δS = − jiδA dΩ = − jiφ dΩ = ji φdΩ = 0, (2.7.58) c2 j c2 ,i c2 ,i so the electromagnetic current is conserved,

i i j ,i = 0, j :i = 0. (2.7.59)

For spinor matter, we use the gauge-invariant Lagrangian density (2.7.14): Lm = Lψ. The electromagnetic current (2.7.56) for the spinor ﬁeld is thus proportional to the conserved vector Dirac current (2.6.61): i ¯ i i j = eceψγ ψ = ecjV . (2.7.60) Let us consider matter which is distributed over a small region in space, as in section 2.4.9. Integrating (2.7.59) over the volume hypersurface and using Gauß’ theorem to eliminate surface integrals gives Z 0 j ,0dV = 0. (2.7.61)

The conservation law (2.7.59) also gives

k i k i k i k i k (x j ),i = x ,ij + x j ,i = δi j = j , (2.7.62) which, upon integrating over the volume hypersurface and using Gauß’ theorem to eliminate surface integrals, gives Z Z xkj0dV = jkdV. (2.7.63) ,0 Using (2.4.111) and (2.4.112) turns (2.7.63) into

k Z Z Z u 0 k 0 k 0 j dV + δx j dV = j dV. (2.7.64) u ,0

R k 0 i A particle located at xa satisfies δx j dV = 0. Consequently, j (x) is proportional to δ(x − xa), giving uk jk = j0. (2.7.65) u0 The electromagnetic current density of a particle is proportional to its four-velocity, analogously to (2.4.130). We define the electric charge density ρ such that cρ j0 = √ . (2.7.66) g00 The electric charge density is not a tensor density. We define the electric charge e such that √ ρ sdV = de. (2.7.67)

124 The electric charge density for particles with charges ea located at xa is

X ea ρ(x) = √ δ(x − x ). (2.7.68) s a a

R 0 R i A quantity j dV , which is equal to j dSi for a volume hypersurface and thereby is a scalar, is thus Z Z X √ cea X j0dV = −g√ √ δ(x − x )dV = c e . (2.7.69) g s a a a 00 a The electric charge is therefore a scalar. Therefore, the electromagnetic current vector for a system of charged particles is k X cu ea jk(x) = √ δ(x − x ), (2.7.70) u0 −g a a analogously to (2.4.154). The relation (2.7.61) represents the conservation of the total electric charge of a physical system. For a particle moving along a wordline xa(τ), we have cuk e Z uk jk(x) = √ δ(x − x ) = ec √ δ(x − x (τ))dτ. (2.7.71) u0 −g a −g a

ui In the locally geodesic and Galilean frame of reference, u0 = (1, v/c), which gives ji = (cρ, j), (2.7.72)

where j is the spatial electric current density vector,

j = ρv. (2.7.73)

The conservation law (2.7.59) in this frame has the form of the equation of continuity, as (2.6.62): ∂ρ ji = + ∇ · j = 0. (2.7.74) ,i ∂t

For one particle located at x0(t), ρ(x) = eδ(x − x0), (2.7.74) is explicitly satisﬁed since ∂ρ ∂ ∂ ∂ = e δ(x − x0) = ev · δ(x − x0) = −ev · δ(x − x0) ∂t ∂t ∂x0 ∂x ∂ = − · evδ(x − x ) = −∇ · j, (2.7.75) ∂x 0

dx0 where v = dt . For a system of charged particles, we also have Z X jdV = eava, (2.7.76) a

dxa where va = dt is the velocity of the particle of charge ea. The equation of continuity (2.7.74) represents, upon integrating over the volume, the conservation of the total electric charge: ! ∂ Z I ρdV + j · df = 0. (2.7.77) ∂t

The relation (2.7.60) gives ρ = eψ†ψ, j = ecψ¯γψ, (2.7.78) where γ is a spatial vector composed from the Dirac matrices γα. If we identify the electric charge e in (2.7.67) with the spinor electric charge e in (2.7.78), then we ﬁnd an integral constraint on a Dirac spinor ﬁeld: Z √ ψ†ψ sdV = 1. (2.7.79)

125 2.7.5 Maxwell equations The Lagrangian density for the electromagnetic field and charged matter is the sum of (2.7.54) and √ i the term − −gAij arising from (2.7.56): 1 √ 1√ L = − −gF F ik − −gA jk, (2.7.80) EM+q 16π ik c k where we omitted terms corresponding to the gravitational field and matter fields which do not depend on Ak. Varying (2.7.80) with respect to the electromagnetic potential Ak and omitting a total divergence gives

1 √ jk 1 √ jk δ L = − −gF ikδF − δA = − −gF ik(δA − δA ) − δA A EM+q 8π ik c k 8π k,i i,k c k 1 √ jk 1 √ 1√ = −gF ikδA − δA = ( −gF ik) δA − −gjkδA , (2.7.81) 4π k,i c k 4π ,i k c k

so the principle of least action δS = 0 for arbitrary variations δAk yields the second Maxwell- Minkowski equation √ 4π ( −gF ik) = jk (2.7.82) ,i c or 4π F ik = jk. (2.7.83) :i c The variation (2.7.81) can also be written as

1 jk 1 jk δ L = − eF ik(δA − δA ) − δA = − eF ikδA − δA . (2.7.84) A EM+q 8π k:i i:k c k 4π k:i c k Accordingly, we have ∂L ∂L 1 EM+q = EM+q = − eF ik. (2.7.85) ∂Ak:i ∂Ak,i 4π

The Lagrange equations (2.1.8) with L = LEM+q for the ﬁeld Ak

∂L ∂L − ∂i = 0, (2.7.86) ∂Ak ∂(Ak,i)

are equivalent to (2.7.82) with (2.7.56). The electromagnetic field equation (2.7.82) infers that ji i is conserved, j ,i = 0, which corresponds to the conservation of the total electric charge, but does not constrain the motion of particles. Therefore, a configuration of charged particles producing the electromagnetic field can be arbitrary, subject only to the condition that the total charge be conserved, unlike a configuration of particles producing the gravitational field which is not arbitrary but constrained by the gravitational field equations. The Lagrangian density for the electromagnetic field and a charged spinor is e ie L = − F F ik + ei (ψγ¯ aD ψ − D ψγ¯ aψ) − meψψ.¯ (2.7.87) 16π ik 2 a i i For an infinitesimal gauge transformation, α 1, (2.7.1)) and (2.7.7) give

¯ ¯ i δψ = ieαψ, δψ = −ieαψ, δAk = α,k, ξ = 0. (2.7.88)

If this transformation is global, α = const, then the corresponding Noether current (2.4.7) is

i ∂L ∂L ¯ ∂L e ik ie ¯ i ie ¯ i J = δAk + δψ + δψ ¯ = − F δAk + ψγ δψ − δψγ ψ ∂Ak,i ∂ψ,i ∂ψ,i 4π 2 2 α = −eeψγ¯ iψα = − ji, (2.7.89) c

126 where we used (2.7.60). Consequently, the conservation law (2.4.6) gives the conservation (2.7.59) of the electromagnetic current associated with the spinor. We deﬁne

α √ 0α D = − g00F , (2.7.90) √ 1√ 1 Hαβ = g F αβ,H = − sε Hβγ ,Hαβ = −√ αβγ H . (2.7.91) 00 α 2 αβγ s γ

ij αβ αi βj The relations F0α = g0igαjF and F = g g Fij give then

Eα β Dα = √ + g Hαβ, (2.7.92) g00 Hαβ Bαβ = √ − gαEβ + gβEα, (2.7.93) g00 or, in the spatial-vector notation, E D = √ − g × H, (2.7.94) g00 H B = √ + g × E. (2.7.95) g00 Using (1.4.144) brings the temporal component of (2.7.82) to 1 √ √ ( sDα) = 4πρ (2.7.96) s ,α or div D = 4πρ. (2.7.97) The spatial components of (2.7.82) read 1 √ 1 √ dxα √ ( sHαβ) + √ ( sDα) = −4πρ (2.7.98) s ,β s ,0 dx0 or √ 1 ∂( sD) 4π curl H = √ + j. (2.7.99) c s ∂t c The conservation law (2.7.59) reads √ 1 ∂( sρ) √ + div j = 0. (2.7.100) s ∂t In the locally geodesic and Galilean frame of reference, (2.7.94) and (2.7.95) reduce to D = E, (2.7.101) B = H. (2.7.102) In this frame, (2.7.97) and (2.7.99) become the second pair of the Maxwell equations: div E = 4πρ, (2.7.103) ∂E 4π curl B = + j. (2.7.104) c∂t c Applying the div operator to (2.7.104) and using (2.7.103) gives (2.7.74). Integrating the second pair of the Maxwell equations over the volume and surface area, respectively, gives I E · df = 4πq, (2.7.105)

I ∂ Z 4π Z B · dl = E · df + j · df. (2.7.106) c∂t c

127 Therefore, the flux of the electric field through a closed surface is proportional to the total charge inside the volume enclosed by the surface f (Gauß’ law) and the circulation of the magnetic field along a contour is equal to the time derivative of the flux of the electric field through the surface enclosed by this contour, called the displacement current, plus the surface integral of the current vector (the Ampère-Ørsted law). The two pairs of the Maxwell equations are linear in the fields E and B. The sum of any two solutions of the Maxwell equations is also a solution of these equations. Therefore, the electromagnetic field of a system of sources (particles) is the sum of the fields from each source. The additivity of the electromagnetic field is referred to as the principle of superposition.

2.7.6 Energy-momentum tensor for electromagnetic ﬁeld The variation with respect to the metric tensor (2.3.4) of the Lagrangian density (2.7.54), 1 √ 1 √ 1 √ δ L = − F F δ( −ggilgkm) = −gg F F ikδglm − −gF F gilδgkm g EM 16π ik lm 32π lm ik 8π ik lm 1 √ 1 = −g g F F lm − F jF δgik, (2.7.107) 8π 4 ik lm i kj

gives the metric energy-momentum tensor (2.3.5) for the electromagnetic ﬁeld:

1 1 T = g F F lm − F jF . (2.7.108) ik 4π 4 ik lm i kj

The corresponding energy density W , energy current S called the Poynting vector, and stress tensor σαβ called the Maxwell stress tensor, are given in the locally geodesic and Galilean frame of reference, using (2.4.82), by 1 W = (E2 + B2), (2.7.109) 8π c S = E × B, (2.7.110) 4π 1 1 σ = E E + B B − δ (E2 + B2) . (2.7.111) αβ 4π α β α β 2 αβ

Multiplying (2.7.44) by B and (2.7.104) by E and adding these scalar products gives

1 ∂E 1 ∂B 4π E · + B · = − j · E − (B · curl E − E · curl B), (2.7.112) c ∂t c ∂t c from which we obtain 1 ∂ 4π (E2 + B2) = − j · E − div(E × B) (2.7.113) 2c ∂t c or ∂W + j · E + div S = 0. (2.7.114) ∂t

Under Lorentz transformations, W , S and σαβ transform like the corresponding components of a tensor of rank (0,2) (2.4.82), according to (1.6.102). The energy-momentum tensor for the electromagnetic ﬁeld is traceless,

ik Tikg = 0, (2.7.115)

so (2.4.159) and the virial theorem (2.4.160) remain unchanged if the particles interact electromag- netically. The condition (2.7.115) also gives, using (2.4.202),

= 3p, (2.7.116)

128 so (2.4.204) shows that the free electromagnetic ﬁeld is ultrarelativistic. In a frame of reference, in which the vectors E and B are parallel to one another or one of them vanishes (and the x axis is along the direction of these vectors), the nonzero components of the tensor Tik are

T00 = −T11 = T22 = T33 = W. (2.7.117) If the vectors E (along the x axis) and B (along the y axis) are mutually perpendicular and equal in magnitude then T00 = T03 = T33 = W. (2.7.118) The variation with respect to the tetrad (2.3.8) of the Lagrangian density (2.7.54), 1 1 1 δ L = − F F ηabηcdδ(eei ejekel ) = eeaF F jkδei − eF F ajδei , (2.7.119) e EM 16π ij kl a b c d 16π i jk a 4π ij a gives the tetrad energy-momentum tensor (2.3.10) for the electromagnetic ﬁeld: 1 1 t a = eaF F lm − F F aj . (2.7.120) i 4π 4 i lm ij

The corresponding tensor tik is equal to (2.7.108). This equality is a consequence of the reduced Belinfante-Rosenfeld relation (2.3.35), which is valid if the spin tensor (2.3.26) is equal to zero. The spin tensor for the electromagnetic field vanishes because the Lagrangian density (2.7.54) does not depend on the torsion tensor. The canonical energy-momentum density (2.3.13) for the electromagnetic field is given by j ∂LEM j Θi = Ak:i − δi LEM. (2.7.121) ∂Ak:j The relation (2.7.85) gives 1 1 Θ j = − eF jkA + eδjF F kl. (2.7.122) i 4π k:i 16π i kl The canonical energy-momentum density is not identical with the tetrad energy-momentum density j eti corresponding to (2.7.120) because the relation (2.3.14) is valid only for fields that are minimally coupled to the affine connection. If the Lagrangian density for the electromagnetic field were minimally coupled to the affine ˜ 1 ˜ ˜ij ˜ 1 ˜ik jk 1 ˜ik connection, LEM = − 16π eFijF , then δALEM+q = − 4π eF δAk;i − c δAk = − 4π e(2SiF − ˜ik jk F ;i)δAk − c δAk, where a total divergence was omitted. The resulting field equation would be ∗ ˜ik 4π k ˜ 1 ˜ik j 1 ˜ik ∇i F = c j . The variation δCLEM+q = − 4π eF δS ikAj = − 4π eF AjδCjik gives the spin 1 ˜ tensor, sijk = − 2π A[iFj]k. The second Einstein-Cartan equation gives the corresponding tor- κ ˜ κ l ˜ κ k ˜ sion tensor, Skij = 4π A[iFj]k + 8π A gk[jFi]l, leading to Fij = Aj,i − Ai,j + 4π A A[iFj]k and ˜ik 4π k κ l ik F ;i = c j + 8π A FilF . In the presence of spinors, Skij would have another part (2.6.48), κ ˜ κ l κ l ˜ κ k ˜ k l Skij = 4π A[iFj]k + 8π A gk[jFi]l + 2 eijkls , leading to Fij = Aj,i − Ai,j + 4π A A[iFj]k + κeijklA s . ˜ 1 1 ˜ ˜lm ˜ j ˜ The metric energy-momentum tensor for LEM is Tik = 4π ( 4 gikFlmF −Fi Fkj). The Belinfante- 1 ∗ ˜ j 1 ˜ j Rosenfeld relation (2.3.34) gives tik = Tik − 4π ∇j (AiFk ) = Tik − 4π Ai;jFk , where the sourceless ∗ ˜ik field equation ∇i F = 0 was used. The corresponding canonical energy-momentum density is ˜ j ∂LEM j ˜ 1 jk 1 j kl j Θ = Ak;i − δ LEM = − eF˜ Ak;i + eδ F˜klF˜ = et , showing that the relation (2.3.14) i ∂Ak;j i 4π 16π i i is valid.

2.7.7 Lorentz force Let us consider a charge particle interacting with the electromagnetic ﬁeld. The total energy- momentum tensor for the particle and electromagnetic ﬁeld is covariantly conserved, which gives the motion of the particle. The electromagnetic part yields, by means of (2.7.32) and (2.7.83), 1 1 1 1 1 T k = F F lm − F F kl − F F kl = − F F lm − F F lm i :k 4π 2 lm:i il:k il :k 4π 2 mi:l 2 il:m 1 1 −F F kl − F F kl = F F kl = − F jl. (2.7.123) il:k il :k 4π il :k c il

129 The particle part gives, using (2.4.153),

k ! k 2 uiu Ti :k = µc √ 0 , (2.7.124) g00u :k so we obtain k ! 2 uiu 1 l µc √ 0 − Filj = 0. (2.7.125) g00u c :k Multiplying (2.7.125) by ui and using (2.7.65) gives

k ! 2 u µc √ 0 , (2.7.126) g00u :k which turns (2.7.125) into k l 2 u 1 u µc √ 0 ui:k = Filρ√ 0 (2.7.127) g00u c g00u or D{}ui e mc = F iju , (2.7.128) ds c j which is the equation of motion of a particle of mass m and charge e in the electromagnetic ﬁeld Fij. Multiplying (2.7.128) by ui gives the identity, thereby (2.7.128) has 3 independent components. The right-hand side of (2.7.128) is referred to as the Lorentz force. D{} d u0 d In the locally geodesic and Galilean frame of reference, we have ds = ds = c dt and the four-velocity is given by (1.6.120). The 3 independent spatial components of (2.7.128) are, using (2.4.135), dP α duα e = mc = eF α0 + F αβv . (2.7.129) dt dt c β In the spatial-vector notation, this equation of motion is given by dP e = F = eE + v × B, (2.7.130) dt L c

where P is the momentum of the particle (2.4.144) and FL is the spatial vector of the Lorentz force. The temporal component of (2.7.128) is

dP 0 du e = mc 0 = F vα. (2.7.131) dt dt c 0α In the spatial-vector notation, this equation of motion is given by dE = ev · E, (2.7.132) dt where E is the energy of the particle (2.4.143), which also results from multiplying (2.7.130) by v and using (2.4.148). Integrating (2.7.114) over the volume gives ∂ Z Z I W dV + j · EdV + S · df = 0, (2.7.133) ∂t which, with (2.7.76) and (2.7.132), yields the conservation of the total energy (2.4.75) of the electromagnetic ﬁeld and particles: ! ∂ Z X I W dV + E + S · df = 0. (2.7.134) ∂t a a

References: [2, 3].

130 2.8 Action for particles

The energy-momentum tensor for a point particle of mass m located at the radius vector r0 is given by (2.4.154): uiuk T ik(r) = mc2δ(r − r )√ , (2.8.1) 0 −gu0 so the variation of the action with respect to the metric tensor (2.3.2) gives

1 Z √ mc Z uiuk mc Z δS = − T ikδg −gdΩ = − δg dx0 = − uiukδg ds 2c ik 2 u0 ik 2 ik Z i k Z Z mc δgikdx dx p = − = −mc δ g dxidxk = −mcδ ds. (2.8.2) 2 ds ik

Therefore, the action for a free (interacting only with the gravitational ﬁeld) particle is

Z 2 S = −mc ds, (2.8.3) 1 where 1 and 2 denote the world points corresponding to the arrival of the particle at the initial and final position. The equation of motion of a free particle is thus the metric geodesic equation (1.4.91). The action for a system of noninteracting particles is the sum of the actions corresponding to each particle: X Z S = − mac dsa. (2.8.4) a A particle interacts with other particles through fields that carry the interaction. Consequently, we must add to (2.8.4) the action describing fields that carry the interaction. For the electromagnetic interaction, such a field is described by the electromagnetic potential and couples to the electromagnetic current. The electromagnetic current vector for a point particle of charge e located at the radius vector r0 is given by (2.7.70):

cuk e jk(r) = √ δ(r − r ). (2.8.5) u0 −g 0

Substituting (2.8.5) into the second term of (2.7.80) gives the action for this coupling:

1 Z √ e Z uk e Z S = − −gA jkdV dx0 = − A dx0 = − A dxk, (2.8.6) e c2 k c k u0 c k so the total action for a particle of mass m and charge e interacting with the electromagnetic potential Ai is Z e Z S = −mc ds − A dxi. (2.8.7) c i For a system of particles, the total action is the sum of the actions (2.8.7) for each particle. Under the gauge transformation (2.7.7), the action (2.8.7) changes by the integral of a total diﬀerential:

Z e Z e Z S0 = −mc ds − A dxi − dα, (2.8.8) c i c

so the conditions δS = 0 and δS0 = 0 are equivalent, and the corresponding equations of motion are gauge invariant. The variation of (2.8.3) with respect to the coordinates xi gives, using (1.4.89), (1.4.90) and (1.4.91), Z D{}u Z δS = mc i δxids − d(u δxi) . (2.8.9) ds i

131 R i The variation of Aidx is Z Z Z Z Z i i i j i i δ Aidx = δAidx + Aiδdx = Ai,jδx dx + Aidδx Z Z Z Z Z j i i i j i i = Ai,jδx dx + d(Aiδx ) − dAiδx = Ai,jδx dx + d(Aiδx ) Z Z Z j i j i i − Ai,jdx δx = Fijdx δx + d(Aiδx ) Z Z j i i = Fiju δx ds + d(Aiδx ). (2.8.10)

Therefore, the variation of (2.8.7) is, using the four-momentum pl = mcul (2.4.135):

Z D{}u e Z Z e Z δS = mc i δxids − F ujδxids − mc d(u δxi) − d(A δxi) ds c ij i c i Z D{}p e Z Z e = i − F uj δxids − d p δxi + A δxi ds c ij i c i Z 2 {} Z 2 D pi e j i e i = − Fiju δx ds − pi + Ai δx , (2.8.11) 1 ds c c 1 where the limits 1 and 2 denote the endpoints of the particle’s world line. The principle of least action δS = 0 for arbitrary δxi vanishing at the endpoints gives the Lorentz equation of motion of a particle of mass m and charge e in the electromagnetic ﬁeld Fij (2.7.128). The action for a particle (2.8.4) and its generalization (2.8.7) determine the Lagrangian for the particle. This Lagrangian satisﬁes the Lagrange equations that are analogous to (2.1.8). References: [2, 3].

A particle is a special case of a ﬁeld existing in spacetime. The physics of particles and their systems, such as rigid bodies and ideal ﬂuids, is referred to as mechanics and will constitute future Chapters 3 (Particles) and 4 (Special Cases). This material is logically presented in The Course of Theoretical Physics by L. D. Landau and E. M. Lifshitz [2, 13, 14].

132 References

[1] E. Schr¨odinger, Space-Time Structure (Cambridge University Press, 1950).

[2] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Pergamon, 1975). [3] E. A. Lord, Tensors, Relativity and Cosmology (McGraw-Hill, 1976). [4] F. W. Hehl, P. von der Heyde, and G. D. Kerlick, Phys. Rev. D 10, 1066 (1974); F. W. Hehl, P. von der Heyde, G. D. Kerlick, and J. M. Nester, Rev. Mod. Phys. 48, 393 (1976); V. de Sabbata and M. Gasperini, Introduction to Gravitation (World Scientiﬁc, 1986); V. de Sabbata and C. Sivaram, Spin and Torsion in Gravitation (World Scientiﬁc, 1994). [5] R. M. Wald, General Relativity (University of Chicago Press, 1984). [6] R. Utiyama, Phys. Rev. 101, 1597 (1956).

[7] T. W. B. Kibble, J. Math. Phys. 2, 212 (1961); D. W. Sciama, in Recent Developments in General Relativity (Pergamon, 1962), p. 415; Rev. Mod. Phys. 36, 463 (1964); 36, 1103 (1964). [8] C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation (Freeman, 1973). [9] F. W. Hehl, Phys. Lett. A 36, 225 (1971); Gen. Relativ. Gravit. 4, 333 (1973); 5, 491 (1974).

[10] A. Papapetrou, Proc. Roy. Soc. London A 209, 248 (1951); K. Nomura, T. Shirafuji, and K. Hayashi, Prog. Theor. Phys. 86, 1239 (1991); N. J. Poplawski, Phys. Lett. B 690, 73 (2010); 727, 575 (2013). [11] C. Møller, Ann. Phys. 4, 347 (1958). [12] F. W. Hehl and B. K. Datta, J. Math. Phys. 12, 1334 (1971).

[13] L. D. Landau and E. M. Lifshitz, Mechanics (Pergamon, 1976). [14] L. D. Landau and E. M. Lifshitz, Fluid Mechanics (Pergamon, 1982); L. D. Landau and E. M. Lifshitz, Theory of Elasticity (Pergamon, 1986); L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media (Pergamon, 1984).

133