You are on page 1of 37

ROTATING BLACK HOLES

The Geometry of Kerr Spacetime


Anusar Farooqui
Copyright All rights reserved by the author. Commercial distribution and sale of the document is in violation of international copyright law. However, feel free to copy and distribute this document free of charge. Caveat emptor This document contains material pertaining to the geometry of Kerr black holes. The goal of the author in creating this document was to learn the material himself. They are neither complete nor is there any guarantee of correctness of the exposition. Important sections will be skimmed or skipped. For a more substantial treatment of the material the reader is referred to Barrett ONeills excellent text, The Geometry of Kerr black holes. Notational remark Throughout this document we will follow the Einstein summation convention without exception. Which is, if the same index appears both as a subscript and a superscript on one side of an equation then we suppress the summation sign, e.g., n i f i f i1 x fxi will appear simply as x fxi . This will also apply, mutatis mutandis to equations with multiple indices: no summation signs will ever appear for such indices without an explanation.

1
1.1

Preliminaries
Manifolds

Dierentiable Structures A topological manifold is a topological space locally homeomorphic to Rn . An atlas is a collection of smooth coordinate charts that cover the manifold and are glued together consistently with smooth transition maps on overlapping neighborhoods. An atlas endows a unique dierentiable structure to a topological manifold which is then called a smooth manifold. A map : M N between manifolds is smooth if, with coordinate charts and , the composition maps 1 are smooth. The set of all smooth, real-valued functions f : M R is a commutative ring under addition and multiplication of functions denoted by FpM q or more explicity by C V pM ; Rq. A dieomorphism is a smooth map : M N with a smooth inverse 1 , in which case M and N are said to be dieomorphic. Tangent Vectors A tangent vector X to M at x the Leibnizian property

M is an R-linear map X : FpM q R with


; df, g

X rf g s  Xf g pxq f pxqXg

FpM q

(1.1.1)

where we supress x in the subscript and the brackets for clarity. The set of all tangent vectors to a manifold M at a point x is denoted by Tx pM q. This is a vector space under scaler multiplication by real numbers and addition of functions. Let x U an open subset of M and : U Rn be a chart with local coordinates tx1 , . . . , xn u. Then the partials of the coordinate expression f 1 f f of f are tangent vectors t fx1 , . . . , fxn u and form a basis for the vector space Tx pM q. The disjoint union of the tangent spaces at all points of the manifold is itself a manifold called the tangent bundle T M  xM Tx pM q. The projection f , . . . , f q : 1 pU q R2n : T M M sends Xx x and px1 , . . . , xn , fx1 f xn is a chart on T M . Curves A curve in a manifold M is a smooth map : I M where I tangent vector psq to M at psq for s I is dened by 9
9 psqrf s 

R.

The

dpf q psq ds

; df

FpM q

(1.1.2)

Dierential maps Given a mapping : M N , for each point x M there is a linear transformation d|x : Tx M Tpxq N called the dierential map or the push-forward 2

dened by |x rf s  dpxqrf s  X rf

; dX

Tx M and f FpM q

(1.1.3)

An equivalent characterization is that d preserves tangents, i.e., d sends psq p qI psq. One can regard it as a single map d : T M T N of 9 tangent spaces. Vector Fields A vector eld X is a cross section of T M , i.e., a smooth map X : M T M such that X  idM . X is smooth means that Xf is smooth whenever f is smooth. The set XpM q of vector elds on M is a module over FpM q, the ring of real-valued functions on M . X saties the Leibnizian property (1) and is therefore a derivation X . In terms of local coordinates, we can write X  X i ff i with the Einstein summation convention. This is the fundamental x link between the invariant description of X and its tensor description as an ntuple of functions X i  X pxi q. The Lie bracket [X, Y ] of vector elds X and Y is characterized by

rX, Y sf  X pY f q Y pXf q ; df FpM q rX, rY, Z ss rY, rZ, X ss rZ, rX, Y ss  0


Integral Curves

(1.1.4)

This operation is bilinear over R, skew-symmetric and satises the Jacobi identity: (1.1.5)

A vector eld on a manifold has another interpretation as a dierential equation whose solutions are curves, called integral curves, that at each point have the tangent specied by the vector eld. That is psq  X psq 9 In local coordinates,
9 psq

; ds I

(1.1.6)

f f xi d  ds pxi q ff i x
psqrxi s 9

(1.1.7) (1.1.8)

Thus the vector equation becomes a linear system of ordinary dierential equations:
: implicit

denition

d i px q  X i px1 , . . . , xn q ; di  1, 2, . . . , n (1.1.9) ds Since X is smooth, local existence and uniqueness follows from standard results, so that for each point x M, h! integral curve x , with the largest possible domain starting at x : x p0q  x. And integral curves , of X can meet only if one is a reparametrization of the other, i.e., psq  ps constq. A vector eld is complete if each of its maximal integral curves is dened on the entire real line. Suppose X is complete. For each s R, let s : M N be the smooth mapping that sends each point x x psq. Clearly, 0  idM and st  s t ; ds, t R. Since it has a smooth inverse s , each s is a dieomorphism. The collection ts : s Ru is called the ow of X. If X is not complete its ow is still dened locally:

dx M, h a neighborhood U of x and 0, such that dy U, y is dened at least on p, q so that s : U M is well dened with the above properties. In this case, it is called a local ow.
Remark 1.0. We will learn in the next chapter that a vector eld on a semiRiemannian manifold is Killing if its ow has the property that for each s, s is not just a dieomorphism but an isometry. Killing vector elds are extremely important for the geometry of Kerr black holes. One-forms The cotangent space to M at x, is the dual of the tangent space whose elements are called covectors:

The disjoint union of cotangent spaces is called the cotangent bundle, T M  M . A section of the cotangent bundle is called a one-form. That is, a x Tx 1-form is a covector eld, i.e., it assigns to each x M a covector x Tx M . pM q is a module over FpM q The space of all 1-forms or covector elds on M , X
under pointwise addition and multiplication by real valued functions on M . A vector eld X XpM q and a one form X pM q combine naturally to give a real valued function X FpM q. Each function f FpM q gives rise to a 1-form, its dierential df . The cotangent space has a basis of coordinate f f one-forms tdx1 , . . . , dxn u, dual to t fx1 , . . . , fxn u and each one-form can be expressed as  p ff i qdxi . In particular, computing in local coordinates, we x recover the multilinear calculus formula for the total dierential of a function: ff df  fxi dxi . Remark 1.0. We the following transformation rules under change of local coordinates: 4

Tx M

 pTx M q  tx : Tx M R,

x linearu

(1.1.10)

f  fyj f fxi f xi f y j fxi dyj dxi  fyj


Bundles A (left) action of a Lie group G on a manifold F : G G F F, pg, xq gx such that g phpxqq  ghpxq and with the identity e G ex  x ; dx F

(1.1.11) (1.1.12)

1 F , is a smooth map

; dg, h G and x F

For a xed g G, the map g : F F, x gx is a dieromorphism with the smooth inverse g1 . The action of a Lie group is said to be eective if g  idM g  e (1.1.13) S1 In Kerr applications, the circle group acts eectively on the unit 2-sphere, 1 S 2 by rotation around the z-axis as follows. For R, we have R px, y, z q  px cos y sin , x sin y cos , z q (1.1.14)

A smooth ber bundle consists of a smooth map : E B, a Lie group acting on a manifold: G 1 F , and a collection of smooth mappings such that 1. Local product condition: Each : 1 pU q U F is a dieomorphism for some open neighborhood U B such that p1 pb, xqq  b ; dx F and b U , i.e., the following diagram commutes 1 p U q /U ttt t tt ttt Proj.  ytt U

2. Overlap condition: If , have intersecting domains U there is a smooth map : U V G such that pb, xq  pb, pbq pxqq ; db U

in B,

and x F

Terminology: is the projection, E is the total manifold, B is the base, F is the ber, G is the structure group and is the set of bundle coordinate systems . For brevity, one often writes G , F

/E

/B

/ Total manifold / Base. 1 Fiber That is, Structure group A vector bundle of rank k over M , is a bundle F E M with ber F  Rk with the general linear group as the structure group, in which case we have the local trivializationX 1 pU q /U ss s s ss ss Proj.  yss U

Rk

and the restriction of to 1 pxq is the linear isomorphism 1 pxq ! txuRk ! Rk . In the case of vector bundles, we usually denote the bundle by E M or simply by E. For instance, the tangent bundle T M M is a vector bundle as is the cotangent bundle T M M . If there exists a map such that M ww w w ww w  {ww Proj. M E
/

Rk

then is called a global trivialization and E is said to be a trivial bundle.

1.2

Tensors

The Invariant approach Let s and r be nonnegative integers, not both zero. A tensor eld of type (r, s) on M is an F-multilinear function A : X X X X F looooooomooooooon looooomooooon
r s

with the convention that a tensor eld of type (0, 0) is just a function f F. Thus, for every choice of r covector elds (1-forms) t1 , . . . , r u and s vector elds tX1 , . . . , Xs u, Ap1 , . . . , r , X1 , . . . , Xs q is a real valued function and A is linear over F in each argument separately. In particular, a function f can be factored out of each slot.
: Denition

Let Tr denote the set of all type (r, s) tensor elds on M . Clearly, T0  1 X : X F such that X p q  pX q , d. So we also have T1 X. Only tensors of the same type 0 can be added but arbitrary tensors can be multiplied using the tensor product. For example, if A is of type (1, 1) and B is of type (2, 3) the A B is given by
s X . Now X is the dual of X , indeed, we have X

pA B qp, , X, Y, Z q  Ap, X qB p, Y, Z q

(1.2.1)

for all covector elds , and vector elds X, Y, Z on M . The tensor product is F-bilinear and associative but may not be commutative. However, if A is covariant, that is, of type (0, s) and B is contravariant, that is, of type (r, 0), then A and B commute. The Classical approach Let tx1 , . . . , xn u be local coordinates for a chart on a neighborhood U M . Then, on U , the components of A Tr are the real valued functions computed s as follows: f f (1.2.2) Ai1 ,...,ir  Apdxi1 , . . . , dxir , j , . . . , js q j1 ,...,js

fx 1

fx

The superscripts are contravariant indices and the subscripts are covariant indices. The operations given in the previous section can be easily computed in local coordinates. For example, if A is a tensor eld of type (1, 1) and we are given  i dxi and X  X i ff i then x Ap, X q  Ai i X j j (1.2.3)

where we have applied the Einstein summation convention to multi-indices for the rst time and supressed i,j . The components of a sum of (r, s) tensors are the sum of components. The components of a tensor product are the product of the components in tensorial sense: if A is of type (1, 1) and B is type (1, 2) then A B has components:

pA B qij m  Aik B jm k

(1.2.4)

Remark 1.0. Its useful to dene tensors invariantly, since then we dont have to check that they are well dened. But they are more explicit and easier to compute in local coordinates. Also, there is the built in error detection feature called index balance: that the same unsummed subscripts and superscripts must appear on each side of a tensor equation. Frame Fields A frame eld tE1 , . . . , En u, is an alternative basis for the tangent space Tx M for each x U M , where each basis element can be expressed in terms of the f f coordinate frame t fx1 , . . . , fxn u as follows 7

Ei by

The coframe eld consists of the dual 1-forms t 1 , . . . , n u dened implicitly i pEj q  i j For any vector eld X, we have the following duality formula X (1.2.6)

 Eij ff j x

(1.2.5)

 i pX qEi

(1.2.7)

The components of a tensor eld can be expressed relative to a frame eld, replacing the coordinate frames in (16) by the new basis vectors. One advantage of using generalized frame elds is that we can often dene them over larger domains. For instance, these is a globally dened frame eld on a torus while a minimum of 4 charts are required to construct an atlas. Change of coordinates is replaced by change of frame elds. Given a frame eld tEi u with dual t i u and a new frame eld tFj u with dual tj u on the same domain U . There are unique n n-matrix valued functions a, b on U such that Fj By duality bi ak k j

 ak Ek j

and

 bim m

(1.2.8)

 ji

(1.2.9)

Note that in the special case of coordinate frames, a and b are just the Jacobian fyi f xi matrices fxj and fyj . Akin to the coordinate frames case, we have a tensor transformation rule for change of frame elds that determines a unique tensor in the new frame eld. For example, if A is a (1, 1) tensor with components Ai j relative to tEi u, then its components in the new frame eld tFi u are given by Api , Fj q  Apbi , ak Ek q  bi ak Ak j j Clearly, the tensor A is independent of the choice of frame eld. Contraction This is a generalization of the concept of the trace of a linear operator to tensors. For a choice of a contravariant index and a covariant index of the components of A, take the trace. For example, is A is of type (2, 3) then choosing the rst contravariant superscript index and the second covariantsubscript index, the 1 resulting tensor C2 A has components Aji k and is called a contraction of A. In general, contraction reduces type from (r, s) to (r 1, s 1). (1.2.10)

Tensors at a point Let x M . Then an (r, s) tensor at the point x function

M is an R-multilinear

A : Tx pM q Tx pM q Tx pM q Tx pM q R looooooooooooomooooooooooooon looooooooooooomooooooooooooon
r s

(1.2.11)

A tensor eld on M determines at each point x M , a tensor Ax called its value at x. One can further replace, Tx pM q by an nite dimensional real vector space V and Tx pM q by V and dene (r, s) tensors as before. In these terms, an inner product g on V is a symmetric, positive denite (0, 2) tensor on V . g is scalar product if, instead of positive deniteness we just require it to be nondegenerate: g pv, wq  0 , dw V v  0.

1.3

Dierential Geometry

Metric Tensors A metric tensor on a manifold M is a smooth covariant (0, 2) tensor that assigns to each point on the manifold, a scalar product on the tangent space, M x gx Tx M . Then pM, g q is called a semi-Riemannian manifold. For vector elds V, W on M , g pV, W q is a real valued function, sometimes denoted by xV, W y. If g is positive denite (the scaler products gx are inner products), then g is a Riemannian metric and pM, g q is a Riemannian manifold. If each gx is a Lorentz scaler product (see section 1.5), then g is a Lorentz metric and pM, g q is called a Lorentz manifold. On a chart U with local coordinates tx1 , . . . , xn u, the components of the metric tensor are

f f  gp ff , fx q  x ff , fx y (1.3.1) x x Hence, we can write g  gij dxi dxj . The metric tensor is often replaced by the line element ds2  q, which assigns to each point x the associated quadratic form qx of gx . So the value of ds2 on a vector eld V is g pV, V q  xV, V y. The
gij
i j i j

metric tensor can be obtained from the line element by polarization. Locally, we can express the line element as ds2

 gij dxi dxj

(1.3.2)

Indeed, given a vector eld V , we have

ds2 pV q

 xV, V y f  xV i ff , V j fx y x f i j f  V V x fx , fx y  gij V i V j  gij dxi pV qdxj pV q  pgij dxi dxj qpV q


i j i j

Key Examples For 0 n, let Rn be the manifold Rn furnished with the line element ds2  i pdxi q2 , where 4 1 , di  1 , di

Rn is called a semi-Euclidean n-space of index . When  0 this is just or dinary Euclidean space, which is obviously a Riemannian manifold. And when  1 and n 2, it is Minkowski n-space, a Lorentz manifold. The standard metric on Rn is denoted by gstd . The pullback of gstd under the local coordinate maps endows every smooth manifold with a cannonical Riemannian metric also denoted by gstd . Locally, this is clear. One only needs to check for the consistency on the overlap to demonstrate this very important existence result. Isometries A dieomorphism : M N of semi-Riemannian manifolds is an isometry if it is preserves scarlar products:

xdpvq, dpwqy  xv, wy

for all tangent vectors v, w

(1.3.3)

Then M and N are isometric and are geometrically indistinguishable. Suppose ds2  gij dy i dy j on V N and ds2  gij dxi dxj on U M . Then, we have N M the following useful criterion in terms of local coordinates. The mapping : U

is an isometry on U if ds2 pulls back to ds2 : N M pds2 q  ds2 N M (1.3.4)

10

The Levi-Civita Connection 1.3.1 The Fundamental Theorem of semi-Riemannian Geometry. On a semi-Riemannian manifold there exists a unique function : X X X such that 1. 2. 3.
V

pW q is F-linear in V pW q is R-linear in W pf W q  V rf sW f
V V

pW q

;df

4. rV, W s 

pW q
X V, W

pV q
XW

5. X xV, W y  x

y xV,

pW q is the covariant derivative of W


f  k f x ij

k

is called the Levi-Civita connection of M . For vector elds V and W , with respect to V . is described by Christoel symbols, real-valued funck for all i, j. Then, 1 g 2

In local coordinates, tions k such that f ij f xi

k ij

9 For a curve in M with $ 0 and a vector eld Y , the covariant derivative of Y along is dened by Y I  Y which can be written locally as 9

fgi fgij fgj & f xj f xl f xi k dY ij ds


i

(1.3.5)

DY dt

YI 

dY k ds

dxj ds

&

where in a slight abuse of notation we have written xi in place of xi . If Y I  0, the vector eld is said to be parallel for parameter values s and t, Y ptq is said to be obtained by parallel transport along . The tangent 9 9 of is a vector eld on , and the covariant derivative of is the acceleration 9  D  of the curve. : 9 9 dt Geodesics A curve is a geodesic if its acceleration is zero:  0. Straight lines 9 9 in Euclidean space generalize to geodesics. In local coordinates, a geodesic is charaterized by the geodesic equations:
: 9 9 xk k xi xj ij

f f xk

(1.3.6)

0

for all k 11

 1, . . . , n

(1.3.7)

Where we write xi for xi and supress the argument for pq. One uses standard results from ODE theory to obtain the following result. 1.3.2 Local Existence and Uniqueness for Geodesics. Given a point x M and a tangent vector X Tx M , there exists a unique geodesic : I M passing through x with tangent vector X. If every geodesic can be extended over the entire real line R then M is said to be geodesically complete. A reprametrization psq  phpsqq of a geodesic is a also a geodesic only if it is ane, i.e., hpsq  as b. A pregeodesic is a curve that admits a reparametrization as a geodesic. Here is a useful reformulation of the geodesic equations. Let L : T M R be a smooth function on the tangent bundle. A curve on M is extremal for L if there is an atlas for T M such that the Euler-Lagrange equations hold: d ds

1 For a semi-Riemannian manifold M , let L : T M R, 2 x, y. 1 9 9 In local coordinates, L  2 gij px1 , . . . , xn qxi xj . The resulting Euler-Lagrange equations are

f L p q  f L f xi f xi
9 9

'

, di  1, . . . , n

(1.3.8)

$ 1 fgij 9i 9j d 9 gkl xl  xx , dk  1, . . . , n (1.3.9) ds 2 f xk These equations are equivalent to the geodesic equations. This formulation will be extremely useful in computing geodesics in Kerr spacetimes.
Type-changing On a semi-Riemannian manifold, the metric tensor can be used to change any tensor of type (r, s) to that of type (rI , sI ) with r s  rI sI . They are regarded as being equivalent. It suces to show how to change a tensor of type (r, s) to that of type (r 1, s 1) and (r 1, s 1). Let A be a (r, s) tensor eld with components i Aj1 ,...,ir 1 ,...,js

relative to a frame eld tEi u. We change A to type (r 1, s 1) by lowering an index to obtain the components for the new tensor as follows:
i Aj1 ,...,,...,ir 1 ,...,,...,js
1 r 1 s

 g Aij ,...,,...,i (1.3.10) ,.........,j for a choice of contravariant index ti1 , . . . , ir u, where g  xE , E y are
the components of the metric tensor relative to the frame eld. Similarly, we change A to type (r 1, s 1) by raising an index
i Aj1 ,...,,...,irs 1 ,...,,...,j

 g Aij ,.........,i ,...,,...,j


1 1

r s

(1.3.11)

where g is the inverse of the matrix g . 12

Lemma 1.0. Let tEi u be an othonormal frame eld, so gij  i j j , with j  1. The eect of raising or lowering an index i on the components of a tensor is just multiplication by i . On a semi-Riemannian manifold, the contraction operation can be extended to two indices of the same variance. One needs only to raise or lower one of them and use the natural contraction. Curvature The Riemannian curvature tensor of a semi-Riemannian manifold M is the function R : X X X X given by RXY Z

Zq

XZ

rX,Y s Z

(1.3.12)

If the Lie bracket of vector elds X and Y is zero then the Riemannian curvature tensor measures the failure of X and Y to commute. 1.3.3 Symmetries of the curvature tensor. For all vector elds X, Y, Z, W 1. RXY Z

 RY X Z

2. xRXY Z, W y  xRXY W, Z y 3. RXY Z RY Z X

RZX Y  0

4. xRXY Z, W y  xRZW X, Y y In terms of the frame eld tEi u, the components of the curvature tensor are

 i pRE E Ej q Using the duality formula X  i pX qEi one obtains


i Rjk
k

(1.3.13)

i  Rjk fxi (1.3.14) R f f f xk f x f xj Due to the symmetries of the curvature tensor, it has only one nonzero contraction modulo the sign called the Ricci curvature tensor denoted by Ric. By convention, the sign is specied by the following lemma:

Lemma 1.0. In terms of an arbitrary frame eld, the components of Ric are
k  RicpEi , Ej q  Rikj In the orthonormal case, RicpX, Y q  k xRXE Ek , Y y.

Rij

(1.3.15)

13

Further contracting Ric one obtains the scaler curvature S  g ij Rij . The sectional curvature K, a real valued function on the set of all 2-planes tangent to M is dened as follows. For a basis tX, Y u of , we have K pq 

If S is a surface and x S, the sectional curvature K is just the classical Gaussian curvature of S at x. The Riemannian curvature tensor R has the following symmetry properties with the covariant derivative:

xRXY X, Y y xX, X yxX, X y xX, Y y2

(1.3.16)

R XY

 pRXY Z q R p
XR Y Z

pX qY Z RX pY q Z RXY
RqZX

pZ q

(1.3.17)

and the Bianchi curvature identities

q p

Z R XY

0

(1.3.18)

The various forms of curvature are geometric invariants, i.e., they are preserved by isometries. Killing Vector Fields Let X be a vector eld on a semi-Riemannian manifold M with ow ts u. If each s is not just a dieomorphism but an isometry then X is called a Killing vector eld or sometimes an innitisimal isometry. 1.3.4 Proposition. The following are equivalent: 1. X is a Killing vector eld 2. X xY, Z y  xrX, Y s, Z y xY, rX, Z sy 3.

X, Z y  x

Z X, Y

We have the following important conservation lemma: Lemma 1.0. If X is a Killing vector eld on M and is a geodesic, then the 9 scalar product xX, y is constant along .
: Proof. Since tion.

d  0, ds xX, y  x
9

9 X, 9

y  0 by part 3 of the previous proposi-

If M is connected, then any isometry of M is completely determined by its dierential map dx at a single point x M . So a Killing vector eld is completely determined by its values on a arbitrarily small neighborhood of a single point. 14

1.4

Extending Manifolds

Coordinate Extension Let U be a semi-Riemannian manifold that we want to extend. Find a dieomorphism : U V M a smooth or analytic manifold. Choose a metric g0 on V that makes an isometry. Extend the tensor eld g0 to a larger set W V such that g |V  g0 and g is still a metric tensor on W . Let pU, q be a coordinate chart with local coordinates tx1 , . . . , xn u and let gij be the components of a metric tensor g on U . So is a dieomorphism of U onto its image pU q Rn . Assign pU q the metric that makes an isometry. The components gij have the same expressions as before but with tx1 , . . . , xn u now considered as the natural Cartesian coordinates of Rn . Now it we may be able to extend the functions gij to an open set containing pU q and in the analytic case such extensions are unique. Gluing Topological Spaces Given two semi-Riemannian manifolds M and N , together with an isometry from U M to V N . There is a natural way to glue them together along U  V producing a new manifold Q. We call tM, N, u the gluing data and the matching map. Consider M and N as topological spaces. Dene on M N , the equivalence relation  dened by

The natural injections and are homeomorphisms onto their respective images. If M : M P and N : N P agree on M N then they dene well a map : Q P . And if M and N are continuous, and suppose |M  M and |M  M , then is continous as well. Note that even if M and N are Hausdor, Q need not be. As a counterexample, consider M  N  R and U  V  tt : t 0u with the identity map. We say that the gluing data tM, N, : U V u sases the Hausdor condition if there does not exist any sequence txn u in U such that both lim xn

rx  y or x  pyq or y  pxqs Then Q  M N  M N { . The natural injections : M Q and : N Q combine to give the natural projection : M N Q for a quotient space. In particular, a subset S Q is open both 1 pS q M and 1 pS q N are open. Equivalently, a map : Q X is continuous both and are continuous.

xy

M U

and

lim pxn q N

If the gluing data satises the Hausdor condition and if M and N are Hausdor, then so is Q.

15

Gluing Semi-Riemannian Manifolds For smooth manifolds it suces to assign Q the atlas consisting of all charts of M  pM q and N  pN q. Its straightforward to check that they agree on the overlaps and the transition maps are smooth. Moreover, the same holds for analytic manifolds as well. Furthermore, on semi-Riemannian manifolds with the matching map an isometry the metric tensors agree on M N and hence we obtain a metric tensor on Q. We will use this such extensions often in the construction of maximal extensions of Kerr spacetimes.

1.5

Lorentz Vector Spaces

Let V be a vector space over the reals of dimension n. A scalar product g on V is a symmetric, nondegenerate bilinear form on V . Vectors v and w are orthogonal if g pv, wq  xv, wy  0 and we write v u w. g is nondegenerate asserts that zero is the only vector orthogonal to all the vectors, i.e., v u V v  0. A useful criterion for nondegeneracy is that one, and hence every, basis tv1 , . . . , vn u for V the matrix gij  g pvi , vj q is invertible.

pV, gq is called a scalar product space. If g is positive denite, it is an inner product and V is then called an inner product space. The norm is dened as 2 |v|  |xv, vy| 1 . If |u|  1 then u is called a unit vector and a set of orthogonal unit vectors is called orthonormal. Every scalar product space has an orthonormal basis. Indeed, any orthonormal set can be enlarged to an orthonormal basis.
Associated to an orthonormal basis te1 , . . . , en u are n numbers i  such that xei , ej y  i j j . Modulo the order, every orthonormal set has the same signs. p1 , . . . , n q is called the signature of V , usually listing the negative signs rst. The number of negative signs is called the index of V . A Lorentz vector space is a scalar product space of dimension n index  1. A vector v in a Lorentz vector space V is spacelike timelike null if xv, v y 0 or v  0 if xv, v y 0 and v $ 0 if xv, v y  0

2 with

The type that a vector falls into is called its causal character. A vector orthogonal to a timelike vector z is spacelike. Thus, z u is an inner product space and V  Rz z u . For v V , writing v  au x for a timelike vector u and a spacelike vector x we obtain

xv, vy  a2 |x|2

(1.5.1)

Since a cone is a subset closed under multiplication by positive scalars, the set of all null vectors of V is called the nullcone. 16

The nullcone has two components and with = . Each component is a cone dieomorphic to R S n2 . and are also called nullcones and there is no invariant way to distinguish between them. For n 3, S n2 is connected, hence the disjoint sets and are actually the connected components of the nullcone. Since S 0  t1u when n  2, and are not connected. The set I of all timelike vectors in V is called the timecone. It has two components I and I . Each is an open convex cone with I  I . The boundary of the timecone is the nullcone along with the zero vector: I  t0u. Two timelike vectors z and w are in the same timecone xz, wy 0. A vector that is not spacelike is called causal. A subspace W of V is spacelike timelike null if g |W is positive denite if g |W is nondegenerate and has index 1 if g |W is degenerate

and its type is called the causal character of W . Note that t0u is spacelike. W is spacelike it is an inner product space. For n 2, W is timelike W is a Lorentz vector space. In a Lorentz vector space, orthogonal null vectors are collinear. Let W be a null subspace of V . Then, there is a nonzero vector d W , unique upto multiplication by nonzero scalars, such that d u W . In particular, d is null. Moreover, every vector in W Rd is spacelike. For a spacelike subspace W , extending an orthonormal basis to one for V shows that W u is timelike and vice-versa so that V  W W u . It follows that W is null W u is null as well.

1.6

General Relativity

Let S be a 3-dimensional Riemannian manifold called Space. A particle in space is a curve : I S where we think of t I as the time parameter. The classic Newtonian case is a mass m located at the origin in S  R3 , and particles of various masses m 3 m move in S according to Newtons laws of motion and the gravitational law. Now consider the Riemannian manifold S R with R as the time axis. Call this manifold spacetime. Each point px, tq is called an event. A particle is a curve t p pxq, tq in spacetime. Now replace the Riemannian line element d 2 dt2 by the Lorentz line element d 2 dt2 . With S  R3 and setting m  0 gives us Minkowski spacetime. A particle moving solely under the inuence of the central mass is said to be free falling. The key idea of relativity is that freely falling particles are geodesics in spacetime. Spacetimes Let M be a Lorentz manifold, whose points are called events. We time-orient M by selecting continuously, for the tangent space Tx M at each point x M 17

one of two components of the timecone I calling it the future timecone or simply the futurecone. The others are called past timecones. So, a spacetime X is a connected time-oriented 4-dimensional Lorentz manifold. A timelike tangent vector is z is future-pointing if it is in a future timecone, otherwise its past-pointing. A nullcone in the boundary of each future timecone is a future nullcone, and its vectors are future-pointing. A non-spacelike curve or vector eld is future-pointing if all its tangent vectors are future-pointing. Einsteins insight was that physical invariants of gravitational elds are precisely the geometric invariants of spacetime. Therefore, no concepts expressed in terms of coordinates are of physical signicance unless they are independent of the coordinate descriptions. Particles A material particle in a spacetime M is a future-pointing, timelike curve. The proper time of is its arc-length function, and its mass is m  || 0. 9 Light or electromagnetic radiation has a special place is relativity. A photon or lightlike particle in a spacetime M is a future-pointing null geodesic. A 9 lightlike particle has no proper time since | |  0. A material particle is freely falling if it is a geodesic. Moreover, it has a constant mass m since, for a geodesic x, y  m2 is constant. 9 9 d The tangent vector  ds of a material particle is called energy-momentum 9 4-vector and denoted by p. Reparametrizing by its proper time and denoting it by we have p d ds  m d ds since m

d d 
ds

ds

The unit vector d is called the 4-velocity of the particle. In an abuse of notation ds we often write just for when the parametrization is clear from the context. For a lightlike particle , the energy-momentum 4-vector is just p  . 9

Curvature In special relativity, the curvature vanishes and spacetime is at. A vaccum spacetime has no sources of gravity, which is equivalent, via Einsteins equation, to the vanishing of Ricci curvature. Such a spacetime is called Ricci at. Thus, a model of gravitational eld of a single star must be Ricci at away from the source of gravity. Of course, the Riemannian curvature need not be zero. When Ricci curvature does not vanish, it describes the motion of the source of gravity. Such models are useful in cosmology.
: Henceforth,

this will serve as the denition for spacetime.

18

Stationary Observers An observer in spacetime is simply a material particle parametrized by proper time. Observers can send and receive messages and keep track of their proper time. In special relativity, a freely falling observer can impose his proper time and space on the entire Minkowski spacetime but in general relativity, we need an entire family of observers to get analogous results. A observer eld U on a spacetime M is a future-pointing unit vector eld. Each integral curve of U is an observer called a U-observer. An observer eld U is stationary if there exists a smooth function f 0 on M such that f U is a Killing vector eld. If U is also hypersurface orthogonal, i.e. U u is integrable, then U is static. A spacetime is absolutely stationary if it has a unique stationary observer eld, in which case the U -observers are said to be at rest. If U is static, the integral manifolds of U u are 3-dimensional spacelike, submanifolds that are isometric under the ow, and hence constitute a common space for the U -observers. The gravitational eld outside a single star is stationary if the physical properties of the star are not changing. It it is also not rotating, the spacetime is static. Suppose we have local coordinates  px0 , x1 , x2 , x3 q, where the coordinate f f f f vector eld fx0 is timelike future-pointing and spant fx1 , fx2 , fx3 u is spacelike. 0 Then, the coordinate slices tx  constantu are 3-dimensional Riemannian manifolds. Moreover, we have the following

f Lemma 1.0. If px0 , x1 , x2 , x3 q are local coordinates such that fx0 is timelike fg future-pointing and suppose fxij  0 for all i, j. Then, the observer eld U  0 f 0 {| f 0 | is stationary. Conversely, for a stationary U such a coordinate system fx fx can be found at every point. If in addition g0i  0 for i 0, then the observer eld U is static and conversely. f f Proof. fxij  0 for all i, j fx0 is a Killing vector eld. Then, f U  fx0 0 f 0 |. Conversely, let f 0 be a function such that f U is Killing. Since with f  | fx f U is nonvanishing, at each point there is a coordinate system px0 , x1 , x2 , x3 q fg f f such that fx0  f U . But fx0 Killing fxij  0 for all i, j. 0
Instantaneous Observers An instantaneous observer at an event p M , is a future pointing timelike unit vector u Tp M . u has the instantaneous information common to every ordinary observer whose 4-velocity is u as it passes through p. In particular, u knows Tp M ! Ru uu ! Ru R3 . There is a natural way for u to measure the speed of any particle as it passes through p at proper time 0 . If is a 9 material particle, then p0 q  au x, where x uu . Note that a is the instantaneous rate at which us time, t, is increasing relative to s time , so we write a  dt{d . Similarly, |x| is the rate at which the arc length in uu is increasing relative to , so |v |  d {d . Thus, u measures the speed of at event p as 19

fg

d  d{d  |x| 1 (1.6.1) s d {dt a Where the last inquality holds since is timelike 0 x, y|p  a2 |x|2 . 9 9 For a lightlike particle , since is null we have 0  x, y  a2 |x|2 , and 9 9 9 therefore d {ds  1. So instantaneous observers always measure the speed of light to be c  1. Now let u and v be two instantaneous observers at an event p M . u writes u  1 u 0 and d {ds  0 so he is at rest. Then writing v  a u x he concludes that v is moving since u $ v x $ 0. But v writes v  1 v 0 and u  b v y, concluding that he is at rest and it is u that is moving. Now a  dtu {dtv , but 1  xu, v y  a2 |x|2 so a  p1 |x|2 q1{2 1. So according to u, his time runs faster than vs. Since u sees himself at rest and v moving, he claims: Moving clocks run slower. v measures dtv {dtu 1, so his time runs faster than us, but he agrees with the slogan. Their disagreements derive from their dierent decompositions of Tp M into space and time. Special Relativity Special relativity is general relativity of a spacetime isometric to Minkowski spacetime R4 , where the subscript denotes the index. For an arbitrary spacetime 1 M , special relativity obtains in each tangent space Tp M  R4 . 1 Let be a material particle with energy momentum vector p  d{ds  m d{d , and u be an innitisimal observer at some event p  p0 q. Let be the projection of onto the space uu . Tp M

 Ru uu p q  tp qu p q

where t is the observers time and is the proper time of the particle. Hence, dt d  d u d d d and since this is a timelike unit vector we get

d Now v  |d {dt|  d {dt is the Newtonian speed of as measured by u and by chain rule |d {d |  v dt{d . Thus, dt{d  p1 v 2 q1{2 . Consequently, pm In the binomial expansion E d d

1  c

dt 2

d d

m u 1 v2

c
1 2

m d 2 dt 1v

 mp1 v2 q1{2  m

mv 2 Opv 4 q

1 the second term 2 mv 2 is kinetic energy, and Einstein declared this scalar E to be the energy of the particle measured by u, with rest mass m  E |v0 as merely one form of energy. This is the famous formula E  mc2 .

20

1.7

Submanifolds

Smooth Submanifolds A manifold P is a submanifold of a manifold M if it is a topological subspaceX with a smooth inclusion map : P M whose push-forward, is one-toone. If P has dimension one less than M it is called a hypersurface. We say that submanifold is closed if it is a closed set of M . Every smooth manifold is dieomorphic to a submanifold of some Euclidean space. The following is a useful way to get submanifolds: Lemma 1.0. Let f be a smooth real valued function on M . If the push-forward f  df is nonvanishing on a level set tf  constu, then the level set is a closed hypersurface of M . As an example, consider the function f pxq  i pxi q2 on Rn . For r 0, the level set tf  ru is a closed hypersurface, the pn 1q-sphere of radius r. Remark 1.0. Let S 2 be the unit sphere in R3 . Let denote colatitude and longitude. We treat as circular so that the coordinate system , covers all of the sphere except the poles (0, 0, 1). is undened at the poles, but its coordinate vector eld ff is well dened and smooth on the entire sphere and is zero at the poles. Dening p0, 0, 1q  0 and p0, 0, 1q  extends to the entire sphere, with 0 . At the poles, is only continuous but sin and cos are analytic on all of S 2 . Foliations A distribution of dimension k on a smooth manifold M is a smooth eld of k-planes on M . That is, assigns to each point p M a k-dimensional subspace p of Tp M , and locally has a basis of k smooth vector elds. Thus, a distribution is a subbundle of the tangent bundle T M . A submanifold P of M such that Tp P  p for all p P is called an integral submanifold of . If there is an integral manifold of through every point x M , then is said to be integrable. A vector eld V is said to be in a distribution if Vp p for all p M , i.e., if it is a section of the subbundle . 1.7.1 Frobenius Theorem. A distribution on M is integrable

tX, Y rX, Y s u

(1.7.1)

Through each point of M there is a unique largest connected integral manifold of . The collection of all such maximal integral manifolds on is called a foliation of dimension k  dim(). For example, on Rn 0 the spheres tf  rur0 constitute a pn 1q-dimensional foilation, with consisting of their tangent spaces.
: So

it has the relative topology.

21

Every one-dimensional distribution is integrable since the bracket condition is trivial. Such foliations are called congruences. Every nonvanishing vector eld determines a unique congruence on the manifold. For a given one dimensional distribution , such a vector eld need not exist; if it does, is said to be orientable. Then supplies a direction at each point p M , i.e., an oriented tangent line p Tp M . Hypersurfaces If a curve meets a hypersurface S at ps0 q S the meeting is transverse if ps0 q is not tangent to S. For a closed hypersurface S in M , a neighborhood N 9 of S is tubular if there is a line bundle : E S over S and a dieomorphism : E N such that p0p q  p, dp S. Then, N is foliated by curves that cut transversally across S. A hypersurface is said to be two-sided if there exists a continuous vector eld Z of M on S such that Z is never tangent to S, and is therefore nowhere vanishing. Let N be a tubular neighborhood of S. Suppose S is two sided, then, N S has two connected components. Otherwise, S is called one-sided and N S is connected. A closed hypersurface S in M separates M if M S is not connected. Lemma 1.0. Let S be a closed connected hypersurface of M. If S is two-sided, then, M S has at most two components. On the hand, if S is one-sided then M S is connected. Submanifolds of Lorentz Manifolds A submanifold P of M is spacelike, timelike or null provided each tangent space Tx P of P is spacelike, timelike or null respectively, as a vector subspace of Tx M . A spacelike submanifold is intrinsically a Riemannian manifold and a timelike one is a Lorentz manifold. On the other hand, for a null submanifold P , the restriction of g to P is degenerate, so there is not much to say about this case even though they show up all the time in GR. A submanifold N is nontimelike if it has no timelike tangent vectors. If a timelike curve meets N then it must cut transversally. Now, every nontimelike hypersurface S is two-sided, for being time orientable, M admits a globally dened timelike vector eld Z that cannot be tangent to S. Proposition 1.7.2. Let S be a closed connected nontimelike hypersurface in a spacetime M. If N is a tubular neighborhood of S, then the two components of N S can be denoted by N and N in such a way that any future-pointing timelike curve with ps0 q S crosses S transversally from N to N . Corollary 1.7.3. Let S be a closed connected nontimelike hypersurface in a spacetime M. If S separates M, then M S has exactly two components and 22

these can be denoted C and C so that any future pointing timelike curve that meets S tranversally crosses S from C to C . A nonvanishing vector eld X is hypersurface-orthogonal if X u is integrable. Then its integral manifolds are hypersurface normal to X. For example, the radial vector eld X  xi ff i in Rn 0 has this property since X is everywhere x orthogonal to the spheres tx : |x|  constu. Totally Geodesic Submanifolds A submanifold P of a semi-Riemannian manifold M is totally geodesic if vector elds X, Y are tangent to P , then the covariant derivative X Y is also tangent to P . Examples are hyperplanes in Rn and great spheres in S n . Proposition 1.7.4. Let P be a totally geodesic submanifold of M. If a geodesic 9 of M meets P at a point ps0 q where ps0 q is tangent to P, then psq remains in P for s near s0 . Of course, need not be contained in P globally. However, if the submanifold is closed there is no escape: Corollary 1.7.5. If a geodesic : I M is tangent to a closed, totally geodesic submanifold P at a single point of I, then lies entirely in P. The following is a prime source for closed, totally geodesic submanifolds. Theorem 1.7.6. The following two are always closed, totally geodesic submanifolds of M : 1. The set of xed points of an isometry : M 2. The set of zeros of a Killing vector eld. Remark 1.0. A connected, totally geodesic submanifold of a Lorentz manifold has causal character since parallel transport preserves character. Remark 1.0. The failure of a submanifold to be totally geodesic can be measured by its shape tensor or the second fundamental form S. For X, Y XpM q tangent to P , dene S pX, Y q  nor X Y , the component normal to P X . Consequently, P is totally geodesic

M .

S  0

1.8

Cartan Computations

Let tE1 , . . . , En u be an orthonormal frame eld on an open neighborhood of a semi-Riemannian manifold M , with orthonormal coframe eld t 1 , . . . , n u. Such coframe elds can be directly recognized from the line element ds2 as follows.
: It

is easy to check that S is a tensor, i.e., it is well dened and F P -bilinear.

p q

23

Lemma 1.0. One forms t 1 , . . . , n u are a coframe eld on M and t1 , . . . , n u is the signature of M if and only if

such that for all tangent vectors Xx

 i p i q 2 (1.8.1) The connection forms for a frame eld tE1 , . . . , En u are the one-forms t i j ut1i,j nu
ds2 i j pX q  i p Hence, by the duality formula
X Ej X Ej

, di, j

(1.8.2)

 jE

(1.8.3)

In the Cartan approach, connection forms replace Christoel symbols. The matrix p i j pX qq tells us how the frame is changing in the direction X. Lemma: The First Structural Equation 1.0. d i Lemma 1.0. i j In particular, i i

 i  i j j i

(1.8.4)

(1.8.5)

Consequently, all connection forms are determined by t i j uij . Furthermore, for a coframe eld t 1 , . . . , n u, the connection forms t i j ut1i,j nu are the unique one-forms that satisfy equations 1.8.4 and 1.8.5. For tangent vectors X, Y let i j pX, Y q be the matrix of curvature operator RXY relative to the frame tEi u. Explicitly, So ti j ut1i,j nu are 2-forms called the curvature forms of the frame elds. Theorem: The Second Structural Equation 1.0. i j RXY pEj q  i j pX, Y qEi , dj (1.8.6)

0

 di j i j

, di, j

(1.8.7)

As with the connection forms, lowering an index of gives ij  i i j , which by symmetry of curvature is skew-symmetric in i and j. Hence, Corollary 1.0. i j In particular,
i i

 0.

 i j j i

(1.8.8)

24

The above results imply that considered as matrix valued forms i j and i j take their values in the appropriate Lie algebraX . The components of the curvature tensor R with respect to the frame eld are Ri jk Hence, Rijk are

 i Ri jk . The sectional curvatures of the frame eld 2-planes


K pij q  K pEi , Ej q  i j Rijij

 i j pEk , E q

(1.8.9)

(1.8.10)  j Ri jij 2 and the curvature invariant can be written as  i j k pRijk q . Remark 1.0. If M is a Riemannian manifold (  0) with frame eld tEi u,

raising and lowering indices does not require a sign change. Hence, the matrices pi j q1i,jn and pi j q1i,jn are skew-symmetric. Remark 1.0. If M is a Lorentz manifold, then a sign change is required for the timelike index. The skew-symmetry of i j leads to i 0 but i j and similarly for i j . Corollary 1.0. Let u, v be orthogonal coordinates in a semi-Riemannian surface S. Write x ff , ff y  E  1 e2 , x ffv , ffv y  G  e g 2 where 1 and 2 are 1, u u and e, g 0. The Gaussian curvature of S is given by

 0 i
for i, j

(1.8.11)

 j i

(1.8.12)

ev 1 1 2 geu u KS  eg g v

'

(1.8.13)

: For Minkowski spacetime M Rn this is possibly the Lie algebra o1 n of the Lorentz 1 group O1 n which is the group of isometries of M.

pq

pq

25

The Kerr Spacetime

Kerr spacetime is the standard relativistic model of the gravitational eld of a rotating star or indeed any spherically symmetric massive body. The qualitative character of the spacetime depends on the mass and rate of rotation of the massive body, the most interesting case being when the rotation is really slow. To describe the situation it is natural to use spherical coordinates r, , on R3 and a time coordinate t R. We interpret r as the distance to the center of the star, as colattitude and as longitute, with the star rotating in the positive direction. These coordinates are called the Boyer-Lindquist coordinates.

2.1

The Kerr Metric

Kerr spacetime depends on two parameters, the mass m 0 and the angular momentum per unit mass a $ 0. If we set a  0, we obtain the Schwartzchild spacetime. Further setting m  0, we obtain Minkowski spacetime M. The following two functions are crucial: 2

 r2 a2 cos2  r2 2mr a2

(2.1.1)

Setting a  0 immediately makes it clear that is the analog of r in the Schwartzchild case and is a generalization of the Schwartzchild horizon function hprq  1 m{r.
2  1 2m a2 r2 r r The Boyer-Lindquist form of the line element is

(2.1.2)

ds2

 gtt dt2 grr dr2 g d2 g d2 2gt ddt

(2.1.3)

The tensor components of the Kerr metric are presented in table 2.1 along with the Schwartzchild and Minkowski ones for comparison. Table 2.1: The Kerr metric Schwartzchild Boyer-Lindquist (1966) (1916) 1 2m{r 1 2mr{2 r1 2m{rs1 2 { r2 2  % r2 sin2 0
sin r2 a2 2mra2
2 2

gtt grr g g gij pi $ j q

Minkowski (1908) 1 1 r2 r2 sin2 0

gt  gt  2mra sin2 {2 zero otherwise

sin2

26

Remark 2.0. We make the following observations from Table 2.1 1. When a  0 the Kerr metric reduces to the Schwartzchild metric. 2. Since the coodinates t and do not appear in the formulas the coordinate f vector elds ft and ff are Killing vector elds. The isometries of the time ow express the time-invariance of the model and ff expresses axial symmetry. 3. The double change t t, backwards reverses the rotation.

gives an isometry:

running time

4. Kerr spacetime is asymtotically at. Letting r spacetime M.

V we obtain Minkowski

Lemma 2.0. The Boyer-Lindquist metric components satify the following metric identities: g a sin2 gt
2

ag pr

gt a sin gtt
2

agt pr2 a2 qgtt

a qgt
2

 pr2 a2 q sin2  a sin2  a sin2 

(2.1.4) (2.1.5) (2.1.6) (2.1.7)

The Canonical Kerr vector elds are

V W

 pr2 a2 q fft a ff  ff a sin2 fft 

(2.1.8) (2.1.9)

And in terms of the canonical Kerr vector elds, the metric identities of Lemma 2.0 can be written more simply as

xV, ff y xV, fft y xW, ff y xW, fft y

a sin2

(2.1.10) (2.1.11) (2.1.12) (2.1.13)

  pr2 a2 q sin2  a sin2


27

Lemma 2.0.

xV, V y  xW, W y  xV, W y 

a sin2 sin 0
2 2

(2.1.14) (2.1.15) (2.1.16)

These components actually contitute a metric tensor as the following lemma demostrates. Lemma 2.0. For the Boyer-Linquist coordinates,
2 1. gtt g gt

 sin2

2. detpgij q  4 sin2 This shows that the Boyer-Lindquist coordinates give a nondegenerate metric tensor with Lorentz signature. We examine the causal character of the cannonical Kerr vector elds.

2.2

Boyer-Lindquist Blocks

We have seen that with a  0 the Kerr metric reduces to the Schwartzchild one. For a 0 we have the following terminology: Table 2.2: The Rotation Types Rotation Type Roots of the horizon function  r2 2mr a2 Slowly rotating Kerr spacetime r  m pm2 a2 q1{2 Extreme Kerr spacetime double root r  m Rapidly rotating Kerr spacetime No real roots

Rate of Rotation a2 m2 a2  m2 a2 m2

The zeros of the horizon function will determine the horizons. But since this form of the metric fails when  0, we will need to enlarge the spacetime before we gure out the horizons. In contrast to the Schwartzchild case, the Kerr metric does not fail when r  0, so the usual spherical coordinate condition r 0 can be dropped, letting r R. Thus, r and t are cartesian coordinates for an entire plane R2 , with spherical coordinates , on the 2-sphere S 2 . Now, consider the product manifold R2 the present coordinates or metric fails:

S2.

There are 3 subsets on which

1. On the horizons H :  0. For the fast Kerr there are no horizons, for extreme Kerr there is a single horizon r  m, and for slow Kerr there are two horizons r  r and r  r .

28

2. On the ring singularity  R S 1 : 2

 0 r  0 and

cos  0.

3. On the axis A: sin  0. For spherical coordinates on R3 , sin  0 gives the z-axis but since now r R, there are two such axes. In the sphere, S 2 , sin  0 picks out the poles, and then in the R2 S 2 each pole gives a complete axis enduring through time. Question: Can the Kerr metric be extended over any of these? Kinda, the Kerr metric can be analytically extended over the axes A but not over the horizons or the ring singularity. A look at the line element given by 2.1.3, shows that the metric components of Table 2.1 are nonsingular on the axes. The dierentials dt and dr are well dened on the axis in R2 S 2 but d and d are singular at the poles of the 2-sphere and hence on the axis. However, the combination sin2 d is extendible since it can be written in cartesian coordinates as xdy ydx. And d can be dealt with by using the line element d 2 of the 2-sphere. 4 On R2 S 2 H

, the Kerr line element can be written as: ds2  p2 {qdr2 2 d 2 a2 sin4 d2 dt2 2mr2 pdt a sin2 dq2

(2.2.1)

Spacetimes are by denition connected, so we must look into the connected components of R2 S 2 H . This domain is cut into components by horizons. The Boyer-Lindquist blocks I, II and III are the following open subsets of R2 S 2 : 1. For slow Kerr, where there are two horizons r I II III : : : r r r r r r r r

 r m

2. For extreme Kerr, where there is a single horizon r I III : :

m m

3. For fast Kerr, since there is no horizon, R2 S 2 itself can be regarded as a single block I = III. As we will show later, maximally extended Kerr spacetimes are constructed by gluing Boyer-Lindquist blocks together along horizons. In the fast case, it turns out that R2 S 2 is already a maximal spacetime. Remark 2.0. 1.

f f and W are always spacelike.

29

2.

f fr is spacelike on I III and timelike on II.

3. V is timelike on I

III and spacelike on II.

30

3
3.1

Petrov Classication
The Hodge Star Operator

For a basis te0 , e1 , e2 , e3 u of V , the choice of sign of the 4-vector e0 e1 e2 e3  denes an orientation on of V . Then there exists a unique linear operator : 2 2 , called the Hodge star operator, such that the Hodge identity holds:

2 pV q (3.1.1) In the case of the spacetime bivector case n  4, p  2,  1, the Hodge star is a complex structure on 2 . Explicitly, pa ibq  a bpq ; da, b R (3.1.2) that is, i  . This makes 2 a complex vector space of dimension 3 with basis say te0 e1 , e0 e2 , e0 e3 u. With this structure we usually denote it by 2 . Moreover, g is a complex scalar product on 2 . Furthermore, the Weyl cx cx curvature operator C : 2 2 commutes with the Hodge star operator *

 x, y

; d,

and is thus a complex linear operator.

Lemma 3.0. Each tracefree curvature C $ 0 on a Lorentz vector space V is in exactly one of the following 5 Petrov types, expressed in terms of the induced linear operator C on 2 : cx Type I Type D Type II Type N Type III C has 3 distinct eigenvalues. C us diagonalizable with exactly 2 distinct eigenvalues. C has exactly 2 distinct eigenvalues, each with geometric multiplicity 1. C has a unique eigenvalue, necessarily 0, with geometric multiplicity 2. C has a unique eigenvalue, necessarily 0, with geometric multiplicity 1.

31

4
4.1

Kerr Geodesics
First-integrals
9 9 xk k ij xi xj 9

Explicit solutions to the geodesic equations

0

are hard to come by. The best information is provided by rst-integrals, expressions in rst derivatives that are constant on each geodesic . There is always one: 9 9 q  x, y A Killing vector eld provide another, so in Kerr we get two more energy E and angular momentum L. There is 4th one called Carters constant Q. Since we have 4 independent rst-integrals for our 4-dimensional manifold we have complete integrability.

A rst-integral for the geodesics of a semi-Riemannian manifold M is a smooth real-valued function on the tangent bundle T M such that for each geodesic in M the function s p psqq is a constant, p q. Equivalently, 9 the smooth assignment of a constant p q to each geodesic in M determines a function : T M R, p q where is a geodesic with initial velocity . In local coordinates, a rst-integral becomes a rst-order ODE with additive constant p q. The line element ds2 is a rst-integral since evaluated on it gives the 9 9 9 constant q  x, y. We can always reparametrize to set q  1 or 0. For a free falling particle, setting q  1 means parametrization by proper time 9 and is the 4-velocity. For a massive particle with mass m, we can sometimes parametrize so that q  x, y  m2 , so that is the energy-momentum 9 9 9 vector. Recall the following conservation lemma: Lemma 4.1. If X is a Killing vector eld on M and is a geodesic, then the 9 scalar product xX, y is constant along . Hence, the function x, Xx y for all Tx M is a rst-integral. There are exactly two, independent Killing vector elds in Kerr spacetime, given in Boyer-Lindquist coordinates by ft and f . The constant E is called the energy of and

 x, ft y
9

9 L  x, f y

is called the angular momentum of . Recall that if a material particle has mass m 0 and proper time , then  m d{d . Thus, for the r 1, i.e. 9 the Kerr exterior we have 32

9 L  x, f y  mg d{d

 mr2 sin2 d{d  mR2 d{d

where R is the orthogonal distance to the axis. This is essentially the Newtonian formula for angular momentum. Next,

 x, ft y  mgtt dt{d  mdt{d 2mM {r where the rst term mdt{d comprises of both the mass energy and the kinetic energy. The second term 2mM {r represents the Newtonian potential energy
E
9

of a particle with mass m at a distance r from a star of mass M . For material particles, the rst-integrals E and L provide the relativistic versions of the conservation of energy and angular momentum, and in this context are called the constants of motion.

Remark 4.2. 1. For a non-spacelike geodesic outside the ergosphere, i.e. where ft is timelike, E $ 0. Furthermore, if both and ft are future-pointing timelike, has positive energy: E 0. 2. Since f  0 on the axis of rotation, any geodesic that meets the axis has L  0. A geodesic are said to have a direct orbit if L 0 and a retrograde orbit if L 0. 3. For c, d R, the eect of a reparametrization psq pcs dq is just E cE and L cL. Setting q  1 for timelike geodesics and q  1 for spacelike geodesics, normalizes E and L. For null geodesics only the ratio E : L is important. Our goal is to use the above rst-integrals to nd rst-order ODEs for the Boyer-Lindquist coordinates psq and tpsq for an arbitrary geodesic. Recall that we have the canonical vector elds V  pr2 a2 qft af and W  f a sin2 ft . These are the rotational versions of ft and f . Denition 4.3. If is a geodesic with energy E and angular momentum L, let
9 Pprq  x, V y  pr2 a2 qE La

and

9 Dpq  x, W y  L Ea sin2

9 Lemma 4.4. In Boyer-Lindquist coordinates, with a geodesic 9 f tft , we have 9 D 9 at9 pr2 a2 q  sin2 and P 9 t a sin2  9

 rfr f
9 9

Proof. It suces to plug in the denitions and use the metric identities. 33

9 9 We can solve the equations in the above lemma for and t, since the determinant of coecients is 2 0. So that we have

Lemma 4.5. The Boyer-Lindquist rst-order geodesic equations for and t are  9 t
9

1 pr2 a2 qP aD 2

1 D 2 sin2

aP

& &

9 9 Remark 4.6. The above ODEs express and t as functions of r and only, giving us a way to compute explicit integral formulas for psq and tpsq when we are given rpsq and psq.

4.2

The Carter constant


 &  &

We begin by expressing q in terms of the coordinate functions rpsq and psq. Lemma 4.7. q  2 1 2 9 9 r 2

12

P2

D sin2

The following important result is due to Brandon Carter Theorem 4.8 (Carter). Associated with each Kerr geodesic is a constant K such that the coordinates r and satisfy
9 Rprq : 4 r2 9 pq : 4 2

 pqr2 Kq P2
2

D  K qa2 cos2 sin2

Let V  pr2 vector elds.

Remark 4.9. The above rst-order geodesic equations are called the radial equation and the colatitude equation.

a2 qft af and W  a sin2 ft f be the canonical Kerr

Example 4.10. We compute the rst integrals q, L, E, K for principal null V 9 geodesics . For the Boyer-Lindquist parametrization  fr . First note that, since is a null geodesic q  x, y  0 9 9

34

Furthermore, we have

xV, f y  gt pr2 a2 q g a  2 & 2  2aM r 2sin pr2 a2 q pr2 a2 q 2aM r 2sin sin2 a  2aM r 2sin  
and  
2

&    r2 a2 a2 sin2 a sin2 pr2 a2 q   &


2

a sin r a sin2

a 2M r

xV, ft y  gtt pr2 a2 q gt a  2 2M r  1 2 pr2 a2 q 2aM r 2sin a


2  pr2 a2 q 2M r r a2 a2  2  sin 2     

&  

Therefore, and

9 L : x, f y  a sin2 9 E : x, ft y  1

To compute K, rst note that D  L aE sin2  0 So that both D and q vanish. So the radial equation pq  K qa2 cos2 implies that K  0 as well. A geodesic is completely determined by its initial position s0 and the tangent vector s0 . In Kerr geometry, the four rst-integrals can replace s0 . 9 9
9 Lemma 4.11. A geodesic is uniquely determined by the initial data r0 , 0 , 0 , t0 ; sgn r0 , sgn 0 9 and the rst integrals q, E, L, K.

D sin2

We have the corresponding existence result. The rst integrals q, E, L, K are well-dened functions on the tangent bundle T M , hence so are P and D, and then R and . For example, if Xx Tx M , then E pXx q  xX, ft y, and DpXx q  LpXx q aE pXx q sin2 ppxqq. 35

Theorem 4.12. Let p0 be a point o-axis in a Boyer-Lindquist block. Given any numbers q , E , L , K such that the corresponding values of R and are nonnegative at p0 , there exists a geodesic starting at p0 for which q , E , L , K are the rst integrals. Furthermore, when rp0q2 $ 0 both signs can be realized 9 9 9 for rp0q; similarly for p0q. Remark 4.13. The above lemma shows that away from the axes, equator and horizon the four rst integrals are independent.

Recall that the principal plane : spantfr , V u is timelike and u is spacelike. Let and u be the components of in and u respectively. If then is called a principal geodesic. Lemma 4.14. If is a geodesic with rst integrals K and q, then
9 9 K  qr2 2 x , y

Corollary 4.15. Let be a Kerr geodesic. 1. If is timelike, then K the equator.

0, and K  0

is a principal geodesic in

2. If is null, then K 0, and K  0 3. If is spacelike, then K the axis.

is principal. lies entirely in

qa2 , and K  qa2

Remark 4.16. We see the K 0 for null and timelike geodesics. There are surprisingly many lazy geodesics, i.e., without energy or angular momentum. These appear only in the totally geodesic submanifolds: polar plane, horizon or axis. In fact, if is not contained in the horizon or axis, then the Boyer-lindquist coordinates are applicable and we have LE

 0 P  D  0 t   0
9 9

that is, is tangent to a polar plane and hence remains in it. In the null case Lemma 4.17. For a Kerr null geodesic , q  0, and we have 1. K  0

is principal. is in a timelike polar plane. is in axis but not on the horizon.

2. L  E

 0 but K 0 $ 0

3. K  L  0 but E 4. L  E

 K  0

is a restphoton. 36

Remark 4.18. The restphotons are uniquely characterized as the geodesics whose rst-integrals all vanish. Any function of rst-integrals is a rst-integral. A useful one, Q, also called the Carter constant, is dened by K  Q pL aE q2 Lemma 4.19. The radial equation can be written as pq  Q cos2 a2 pE 2 qq

L2 sin2

&

37

You might also like