INSTITUTE OF PHYSICS PUBLISHING Class. Quantum Grav.

20 (2003) 38153826

CLASSICAL AND QUANTUM GRAVITY PII: S0264-9381(03)63062-5

Dynamics in non-globally-hyperbolic static spacetimes: II. General analysis of prescriptions for dynamics
Akihiro Ishibashi1 and Robert M Wald2
1 Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK 2 Enrico Fermi Institute and Department of Physics, University of Chicago, 5640 S. Ellis Avenue, Chicago, IL 60637-1433, USA

Received 6 May 2003 Published 31 July 2003 Online at stacks.iop.org/CQG/20/3815 Abstract It was previously shown by one of us that in any static, non-globally-hyperbolic, spacetime, it is always possible to dene a sensible dynamics for a Klein Gordon scalar eld. The prescription proposed for doing so involved viewing the spatial derivative part, A, of the wave operator as an operator on a certain L2 Hilbert space H and then dening a positive, self-adjoint operator on H by taking the Friedrichs extension (or other positive extension) of A. However, this analysis left open the possibility that there could be other inequivalent prescriptions of a completely different nature that might also yield satisfactory denitions of the dynamics of a scalar eld. We show here that this is not the case. Specically, we show that if the dynamics agrees locally with the dynamics dened by the wave equation, if it admits a suitable conserved energy and if it satises certain other specied conditions, then it must correspond to the dynamics dened by choosing some positive, self-adjoint extension of A on H. Thus, subject to our requirements, the previously given prescription is the only possible way of dening the dynamics of a scalar eld in a static, non-globally-hyperbolic, spacetime. In a subsequent paper, this result will be applied to the analysis of scalar, electromagnetic and gravitational perturbations of anti-de Sitter spacetime. By doing so, we will determine all possible choices of boundary conditions at innity in anti-de Sitter spacetime that give rise to sensible dynamics. PACS numbers: 04.20.Cv, 04.20.Dw, 04.20.Ex

0264-9381/03/163815+12$30.00 2003 IOP Publishing Ltd Printed in the UK

3815

3816

A Ishibashi and R M Wald

1. Introduction Let (M, gab ) be a spacetime and let KleinGordon equation a a m2 = 0. be a real3 scalar eld in this spacetime that satises the (1)

It is well known (see, e.g., [1]) that if (M, gab ) is globally hyperbolic and if is a Cauchy surface for (M, gab ), then for each (0 , 0 ) C ( ) C ( ), there exists a unique C (M) that satises equation (1) and is such that | = 0 and t a a | = 0 , where t a is a given (not necessarily unit) normal to . However, no similar result holds, in general, when (M, gab ) fails to be globally hyperbolic. Given suitable smooth data for at some time (i.e., on a spacelike slice, ), the partial differential equation (1) need not, in general, admit any solutions corresponding to this data and, even when a solution exists, it need not be unique. Given the innite variety of singularities and causal pathologies that can occur in an arbitrary non-globally-hyperbolic spacetime, it is far from clear how to even attempt to give a unique prescription for dening dynamics. Nevertheless, it is of interest to analyse dynamics in non-globally-hyperbolic spacetimes and determine whether there exist prescriptions that give rise to acceptable, unique dynamics, since this might give hints as to whether and how the spacetime singularities themselves might be resolved in quantum gravity. It is well known that classical general relativity predicts the occurrence of singularities. If these singularities are resolved by quantum gravity, then one might expect to have well-dened, deterministic predictions in situations corresponding to the presence of classical singularities. As a rst step in the direction of analysing dynamics in non-globally-hyperbolic spacetimes, attention was restricted in [2] to static, non-globally-hyperbolic spacetimes that possess a hypersurface orthogonal to the static Killing eld, t a , such that the orbits of t a are complete and each orbit intersects once and only once. No causal pathologies occur in such spacetimes (since they are automatically stably causal), but there is still an innite variety of possible singular behaviour. It was shown in [2] that in such spacetimes, a sensible dynamics can always be dened (for sufciently nice initial data) in the following manner: In a static spacetime, the KleinGordon equation (1) can be written in the form 2 /t 2 = A , where t denotes the Killing parameter and A = V D a (V Da ) + m2 V 2 , (3) associated with the where V = (t a ta )1/2 and Da is the derivative operator on induced metric on . View A as an operator (with domain C0 ( )) on the Hilbert is chosen to be V 1 times the natural space H = L2 ( , ), where the measure on volume element on associated with the induced metric. Then A is a positive, symmetric operator on H. Consequently, it admits at least one positive self-adjoint extension namely, the Friedrichs extension. Choose some positive, self-adjoint extension, AE . Given (0 , 0 ) C0 ( ) C0 ( ), for each t R we dene t = cos AE t 0 + AE
1/2 1/2 1/2 1/2 1/2 1/2 sin AE t 0 .

(2)

(4)

Here cos AE t and AE sin AE t are bounded operators dened via the spectral theorem. Since, clearly, we have (0 , 0 ) H H, it follows that for all t, t is a well-dened
3 The analysis of the dynamics of a complex scalar eld follows immediately from that of the real scalar eld by treating the real and imaginary parts separately.

Dynamics in non-globally-hyperbolic static spacetimes: II

3817

element of H. It was shown in [2] that there exists a unique C (M) such that for all t, | t = t and t a a | t = t , where t denotes the time translate of by t. Furthermore, satises equation (1) throughout M, and also the additional properties enumerated in the next section. Therefore, equation (4) provides a satisfactory prescription for dening the dynamics associated with equation (1), at least for initial data in C0 ( ) C0 ( ). Note that different choices of self-adjoint extensions, AE , give rise to inequivalent prescriptions for dening dynamical evolution. (In essence, they correspond to inequivalent choices of boundary conditions at the singularity.) Thus, if A has more than one positive selfadjoint extension, there will be more than one choice of dynamical evolution law as dened by the prescription equation (4). On the other hand, if A has a unique self-adjoint extension as occurs in some cases of interest [3, 4]the general prescription equation (4) yields unique evolution. However, the analysis of [2] leaves open the possibility that, in addition to the freedom of choosing different self-adjoint extensions of A, there could exist other acceptable prescriptions for dening dynamics that are not even of the form of equation (4). If that were the case, one could not conclude that dynamical evolution is uniquely dened even in situations where A has a unique self-adjoint extension. The main purpose of this paper is to give a general analysis of the possibilities for dening dynamics in static, non-globally-hyperbolic spacetimes, and show thatsubject to certain requirementsthe prescription equation (4) is the only possible one. In the following section, we shall specify the requirements that we impose upon the dynamics. In section 3, we shall then prove that these requirements imply that the dynamics is of the form (4). Applications to dening dynamics in anti-de Sitter spacetime will be given in a subsequent paper. 2. Assumptions concerning dynamics Let (M, gab ) be a static (but non-globally-hyperbolic) spacetime, i.e., (M, gab ) possesses an everywhere timelike hypersurface orthogonal Killing eld, t a , whose orbits are complete. We further assume that there exists a slice orthogonal to t a such that every orbit of t a intersects once and only once. Let t denote image of under the static isometry by parameter t; it follows that { t } provides a foliation of M. We shall label each point p M by (t, x), where x denotes the intersection with of the orbit of t a through p, and t denotes the parameter of p along this orbit starting from x. It follows that t is a global time function on M, and hence (M, gab ) is stably causal. We consider the KleinGordon equation (1), which, as noted above, can be written in the form (2), with A being the differential operator given by equation (3). We seek a rule for providing us with a solution to equation (1) associated with any allowed initial data on . We do not know in advance the widest possible class of initial data that should be allowed, but we demand that it at least include all data on that is smooth and of compact support. Thus, our goal is the following: we seek a suitable prescription such that given any (0 , 0 ) C0 ( ) C0 ( ), we obtain a unique C (M) such that depends linearly 0 ), satises equation (1) throughout M, and is such that | = 0 and t a a | = 0 . on (0 , We now state the additional requirements that we impose upon the prescription so that it is suitable. First, we require that the solution must be compatible with causality in the following sense. Assumption 1 (causality). Let K0 denote the support of the initial data on K0 = supp(0 ) supp( 0 ). , i.e., let (5)

3818

A Ishibashi and R M Wald

Then we require that supp( ) J + (K0 ) J (K0 ), where J denotes the causal future/past. Let Dene and t ( /t)| t . The following lemma will be useful in many places in our analysis. Lemma 2.1. Suppose that the dynamics satises assumption 1. Then there exists a > 0 such that t and t are each in C0 ( t ) for all |t| < . Proof. Since is smooth, clearly t and t are smooth for all t, so we need only show that they have compact support for sufciently small t. We show rst that there exists a > 0 such that K0 int D( t ) for all |t| < , where D denotes the domain of dependence. Namely, if not, / we could nd a sequence {xn } K0 such that xn int D( tn ) for all n, where the sequence {tn } converges to zero. Since K0 is compact, there exists an accumulation point x K0 , and it follows that x int D( tn ) for sufciently large n. However, this is impossible, since clearly / x int D( ) and the time translation isometries are a continuous map from R M into M, so we must have x int D( t ) for sufciently small t. Now choose |t| < ; for deniteness we assume that t 0. Then, by the above result, we have J + (K0 ) J ( t ) int D( t ), so we may restrict attention to the globally-hyperbolic sub-spacetime int D( t ). But, since K0 is compact, it follows from a slight generalization of a standard theorem (see theorem 8.3.12 of [1]) that J + (K0 ) t is compact. It then follows immediately from assumption 1 that the supports of t and t are compact. Our second set of requirements ensures that our prescription is compatible with the spacetime symmetries. Let Tt : C (M) C (M) denote the natural action of the time translation isometries on smooth functions on M, i.e., for any smooth F : M M we dene (Tt F )(s, x) = F (s t, x).
be the solution corresponding to the initial data (0 , 0 ) C0 ( ) C0 ( ).

(6)

| t,

(7) (8)

(9)

Similarly, let P : C (M) C (M) denote the natural action of the time reection isometry on smooth functions on M, i.e., for any smooth F : M M we dene (P F )(t, x) = F (t, x). We require our prescription to satisfy the following properties: Assumption 2(i) (time translation invariance). Let be the solution associated with the data (0 , 0 ) on . Suppose that Kt supp(t ) supp( t ) is compactwhich, as shown above, must hold for all |t| < for some > 0. We require that if the time translated data (t , t ) are specied on , then the corresponding solution must be the time translate, Tt , of the original solution . Furthermore, if the data ( 0 , A0 ) is specied on which formally correspond to initial data for /tthen we demand that the corresponding solution is /t. Assumption 2(ii) (time reection invariance). If the data (0 , 0 ) are specied on which formally correspond to initial data for the time reverse of then we require that the corresponding solution is P . (10)

Dynamics in non-globally-hyperbolic static spacetimes: II

3819

Our nal set of requirements concerns the existence of a suitable conserved energy for solutions. Consider rst the case of a globally-hyperbolic, static spacetime. In this case, if (0 , 0 ) is of compact support, then (t , t ) will be of compact support for all t. For the unique solution, , corresponding to this initial data, dene E( , ) = 0 2 V 1 d + 0 A0 V 1 d (11)

= ( 0 , 0 )L2 + (0 , A0 )L2 ,
2

where the L inner product is dened using the volume element specied below equation (3), and the integrals clearly converge since all functions appearing in the integrals are smooth and of compact support. It is easily seen that E is positive denite, and it follows immediately from the KleinGordon equation (2) that E is conserved in the sense that E(Tt , Tt ) = E( , ). (12) We may therefore view E as a conserved inner product on the vector space, W , of solutions with initial data in C0 ( ) C0 ( ). More generally, if is a solution with initial data in C0 ( ) C0 ( ) and is a solution with initial data that is merely in C ( ) C ( ), then (13) E( , ) = (0 , 0 )L2 + (0 , A0 )L2 is also well dened and conserved. We shall require our prescription for dening dynamics in the non-globally-hyperbolic case to admit a conserved energy which reduces to equation (13) in appropriate cases. In the non-globally-hyperbolic case, a solution with data (0 , 0 ) in C0 ( ) C0 ( ) will not, in t will be of compact support for all t (although, as shown above, general, be such that t and t and t will be of compact support for all |t| < for some > 0). Consequently, if we dene W to be the vector space of solutions (as given by our prescription) with initial data (0 , 0 ) in C0 ( ) C0 ( ), then Tt will not map W into itself. It is therefore useful to enlarge the solution space W as follows. Let V denote the vector space of solutions to the KleinGordon equation that can be expressed as nite linear combinations of solutions of the form Tt for W and t R, i.e.,
V{ |

= Tt1

+ + Ttn

n,

W }.

(14)

(Note that in the globally-hyperbolic case, we have V = W , but in the non-globally-hyperbolic case, in general, W will be a proper subset of V .) Dene the initial data spaces, X and Y, for V by X {0 =
C0 (

| |

V }, V }.
C0 (

(15) (16)

Y { 0 = ( /t)| |

) X and ) Y . Several additional properties of the spaces Clearly, we have X and Y follow immediately from our previous assumptions. First, since the last condition of assumption 2(i) implies that if V then /t V , it follows immediately that Y X. Furthermore, assumption 2(ii) implies that if V then P V , and consequently, ( P ) V . It follows that if (0 , 0 ) are initial data for a solution in V , then (0 , 0) 0 ) are also initial data for a solution V . Consequently, the vector space of initial data and (0, for solutions in V is isomorphic to X Y . We require that our prescription for the dynamics be such that there exists a symmetric, positive denite bilinear map (i.e., an inner product) E : V V R such that the following properties hold:

3820

A Ishibashi and R M Wald

Assumption 3(i) (time translation invariance of E). E is time translation invariant in the sense that for all 1 , 2 V and all t we have E(Tt
1 , Tt 2)

= E(

1,

2 ).

(17)

Assumption 3(ii) (time reection invariance of E). E is time reection invariant in the sense that for all 1 , 2 V we have E(P
1, P 2)

= E(

1,

2 ).

(18)

Assumption 3(iii) (agreement with formula in the globally-hyperbolic case). If , V and has initial data in C0 ( ) C0 ( ), then E is given by the same formula as in the globally-hyperbolic case (see equation (13)), i.e., (19) E( , ) = ( 0 , 0 )L2 + (0 , A0 )L2 . Assumption 3(iv) (compatibility of convergence with respect to E with more elementary notions of convergence of initial data). Suppose that { n } is a sequence in V that is a Cauchy sequence with respect to the norm dened by E. Suppose, further, that there exists a V such that (n )0 and all of its spatial derivatives converge uniformly on compact subsets of to 0 and its corresponding spatial derivatives and, similarly, that ( n )0 and all of its spatial derivatives converge uniformly on compact subsets of to 0 and its corresponding spatial derivatives. Then we require that { n } converge to in the norm dened by E, i.e., we require that
n

lim E(

) = 0.

(20)

3. General analysis of dynamics In this section, we shall prove that any prescription for dening dynamics that satises the assumptions of the previous section must, in fact, be equivalent to the prescription of [2], obtained by choosing a self-adjoint extension of the operator A on L2 ( , ). It follows immediately from either assumption 3(iii) or assumption 3(iv) that E( , ) = 0 whenever the initial data for vanish. Since E is positive denite, it follows that solutions in V can be put in 11 correspondence with initial data in X Y , so we can equivalently view E as an inner product E : (X Y ) (X Y ) R on initial data space. We may break E up into a sum of four maps of the form Q : X X R, S : Y Y R, R : X Y R and T : Y X R, i.e., we may write (21) E([0 , 0 ], [0 , 0 ]) = Q(0 , 0 ) + S( 0 , 0 ) + R( 0 , 0 ) + T (0 , 0 ). By assumption 3(ii), E is invariant under the time reection map P. Since the action of P on initial data is to take (0 , 0 ) into (0 , 0 ), it follows immediately that the time reection invariance of E implies that R = T = 0. The positive deniteness of E on V implies that Q denes an inner product on X and that S denes an inner product on Y. In fact, we now show that the inner product, S, on Y is just the L2 inner product considered in the analysis of [2]. Lemma 3.1. Let Y and S : Y Y R be dened as above (see equations (16) and (21)). Then, under the assumptions stated in the previous section, we have Y L2 ( , ), where the measure is that arising from V 1 times the natural volume element on . Furthermore, for all , Y , we have S(, ) = (, )L2 . (22)

Dynamics in non-globally-hyperbolic static spacetimes: II

3821

Proof. Let Y and consider the solution V with initial data (0 = 0, 0 = ). Let {On } be a nested family of open subsets of with compact closure such that n On = . (Such a family can be constructed by putting a complete Riemannian metric on , choosing a point p and taking On to be the open ball of radius n in this metric about point p.) Let fn 1 everywhere, and fn (x) = 1 if {fn } be a sequence of functions in C0 ( ) with 0 / x On1 but fn (x) = 0 if x On+1 . Let n = fn . Let n W be the solution associated with the initial data (0, n ). Then we have S(, ) = E( , ) = E( n , = E( +
n,

n) n)

+ E( +2

n,

n)

+ E(
2 1

n,

n)

+ E(

n,

n)

fn (1 fn ) V (fn )2 V 1 d ,

d (23)

(fn )2 V 1 d

where we have used assumption 3(iii) to obtain an explicit form for E in the case where one of the arguments, n , lies in W . However, the last inequality in equation (23) implies that L2 ( , ), since, if not, given any C > 0 we could nd an n such that fn 2 2 > C. L This shows that Y L2 . But, since L2 , it follows that {n = fn } is a Cauchy sequence in L2 , i.e., { n } is a Cauchy sequence in the energy norm. Since {n } and all of its spatial derivatives clearly converge uniformly on all compact subsets of to and its corresponding spatial derivatives, it follows from assumption 3(iv) that { n } converges to in the energy norm. But this means that for all Y we have S(, ) = lim
n

(fn )2 V 1 d

2 L2 ,

(24)

which implies that the inner product on Y dened by S is just the L2 inner product, as we desired to show. Now complete V in the inner product E to obtain the real Hilbert space HE . Similarly, we complete X and Y in the inner products Q and S, respectively, to obtain real Hilbert spaces HX and HY = L2 ( , ), respectively. Clearly, we have HE = HX L2 ( , ). (25) Now, by assumption 3(i), for each t, Tt : V V is a norm preservingand, hence, boundedlinear map in the norm E, so it can be uniquely extended to a bounded linear map Tt : HE HE . By continuity, the extended Tt also is norm preserving. Furthermore, the range of Tt clearly includes V , so the range of Tt is dense in HE . Consequently, Tt : HE HE is unitary for each t. Furthermore, on V we have Tt Ts = Tt+s and, by continuity, this relation must hold on HE . Thus, Tt is a one-parameter unitary group on HE . Furthermore, we have the following proposition: Proposition 3.1. For all /t. V , Tt is strongly differentiable in HE , and its derivative is

Proof. Our task is to show that for all V , Tt lim = 0. (26) t0 t t E Since V consists of nite sums of time translates of elements of W (see equation (14)), it sufces to prove equation (26) for solutions of the form Ts for W . However, since Tt is a unitary group, it follows immediately that it sufces to prove equation (26) for W .

3822

A Ishibashi and R M Wald

In that case, we also have /t W and, by lemma 2.1, there exists a > 0 such that Tt W for all |t| < . Consequently, by assumption 3(iii), the energy norm appearing in equation (26) is given explicitly by equation (19). The result then follows immediately from standard results for solutions with smooth data of compact support to the partial differential equation (1) on the globally-hyperbolic spacetime int D( ). Corollary 1. Tt denes a strongly continuous one-parameter group on HE . Proof. We have already established that Tt is a one-parameter group on HE , so we need only show that it is strongly continuous. However, we have just shown that Tt is strongly differentiableand, hence, strongly continuousfor all V . Since V is dense in HE , given HE and given > 0 we can nd a V such that E < /3. Writing Tt Tt = Tt ( )( ) + (Tt
E,

),

(27) (28)

we immediately see that

E

2 /3 + Tt

so strong continuity of Tt on V implies strong continuity of Tt on HE . Since Tt is a strongly continuous one-parameter unitary group on HE , by Stones theorem (adapted here to the real Hilbert space case), there exists a skew adjoint map B : HE HE such that Tt = exp(t B). Furthermore, since Tt is strongly differentiable for all V we have V Dom(B). Since we have shown in proposition 3.1 that for all V the derivative of Tt is just /t, we have for all V (29) = B . t Using the isomorphism (25), we can view B as a skew-adjoint operator on HX L2 ( , ) and rewrite equation (29) as the pair of equations 0 = B11 0 B12 0 , (30) 0 = B21 0 B22 0 , where 0 ( 2 / 2 t)| ,
2

(31) (32)

and B11 : HX HX , B12 : L ( , ) HX , etc. Equations (30) and (31) hold for all 0 X and all 0 Y . It follows immediately that the restriction of B11 to X vanishes and that the restriction of B12 to Y L2 ( , ) must equal I , where I denotes the identity map on functions. (Note, however, that B12 is a map between different Hilbert spaces.) Similarly, the invariance of the dynamics under time reection (assumption 2(ii) above) implies that the restriction of B22 to Y must vanish. Thus, writing B = B21 , we have learned that there exists an operator B : HX L2 ( , ) with X Dom(B) such that for all 0 X we have 0 = B0 . (33)
Comparing with equation (2), we see that if 0 C0 ( ), then B = A, where A was dened by equation (3). Thus, when viewed as an operator from HX into L2 ( , ), B is an extension of A from the domain C0 ( ) to a domain that includes all of X. Finally, we note that the skew-adjointness of B on HE directly implies that for all 0 , 0 X and all 0 , 0 Y , we have (0 , 0 )HX + (0 , B0 )L2 = (0 , 0 )HX ( 0 , B0 )L2 . (34)

Dynamics in non-globally-hyperbolic static spacetimes: II

3823

But this equation can hold if and only if for all 0 X and all 0 Y we have ( 0 , B0 )L2 = (0 , 0 )HX . To make further progress, we need the following lemma: Lemma 3.2. Let W and, as before, write t = | t and t = ( /t)| t . Then t L2 ( , ) (and, hence, X L2 ( , )). Furthermore, t is strongly differentiable with respect to t as a vector in L2 ( , ) and dt /dt = t . More generally, dn t /dt n exists and equals ( n /t n )| t . Remark. We previously showed that Y L2 ( , ) and also that Y X. In addition, we previously showed that Tt is strongly differentiable with respect to t as a vector in HE . None of the results claimed in lemma 3.2 follow immediately from these previous results. Proof. We label points by (t, x) as explained at the beginning of section 2. Then at each xed x we have t 0 =
0 t

(35)

t dt =
0 t 0

[ 0 + ( t 0 )] dt ,

(36)

and hence t 0 t 0 = It follows that |t 0 t 0 |2 =

0 t 0 t 2

( t 0 )dt .

(37)

( t 0 )dt |1|2 dt
0 t t

| t 0 |2 dt (38)

=t
0

| t 0 |2 dt ,

where the Schwartz inequality was used in the second line. 2 Now let {fn } be as in the proof of lemma 3.1. We multiply equation (38) by fn and integrate over with respect to the volume element V 1 d . We obtain fn (t 0 t 0 )
2 L2

t =t
0 t t

V 1 d
0

dt |fn ( t 0 )|2

dt dt
0 t

V 1 d |fn ( t 0 )|2 V 1 d | t 0 |2 , Tt )

t t
0

dt E(Tt
t

4t
0

dtE( , ) (39)

= 4t 2 E( , ).

By the same argument as used in lemma 3.1, this inequality shows that (t 0 t 0 ) 2 0 are in C0 ( ), this proves that t L2 ( , ), as we L ( , ). However, since 0 and desired to show.

3824

A Ishibashi and R M Wald

To prove the differentiability of t , we need to generalize and sharpen our above estimates. First, we repeat the same steps as those leading to equation (39) but now use arbitrary times t and t rather than t and 0 to nd that for all t and all t we have t t (t t) t t t (t t) t
2 L2

4(t t)2 E( , ). 4(t t)2 E( /t, /t),

(40)

Applying this result to the solution /t, we obtain

2 L2

(41)

which implies that at any xed t, there exists a C > 0 and a > 0 such that for all |t t| < , we have t t
L2

C|t t|.
2

(42)

On the other hand, now that we know that t L ( , ) for all t, the same steps as those led to the third line of equation (39) now yield t t (t t) t
2 L2

(t t)
t

dt

t t

2 L2 .

(43)

Substituting the estimate (42), we nd that for all |t t| < , we have t t (t t) t

2 L2

C (t t)4 .

(44)

Dividing this equation by (t t)2 and taking the limit t t, we immediately see that the strong derivative of t in L2 ( , ) exists and is equal to t . The results for higher time derivatives of t follow immediately by applying the same arguments to n /t n . We now prove our main theorem: Theorem 3.1. Consider any prescription for assigning a solution, , to any initial data (0 , 0 ) C0 ( ) C0 ( ) such that all the assumptions stated in section 2 are satised. 2 Let A : L ( , ) L2 ( , ) be the operator (3) dened on the domain C0 ( ). Then 2 2 there exists a positive, self-adjoint extension, AE : L ( , ) L ( , ), of A, such that the dynamics dened by equation (4) agrees with that given by the prescription. Proof. We have already learned that for all V , there exists an operator B : HX L2 ( , ) such that equation (33) holds. Furthermore, we know that B is an extension of A to a domain that contains X and is such that equation (35) holds for all X and 0 Y . We 2 also know that Y X L ( , ). Now let C denote the restriction of B to the domain Y and view C as a map C : L2 ( , ) L2 ( , ) rather than a map of HX into L2 ( , ). Since C0 ( ) Y , we 2 see that C is densely dened on L ( , ) and equation (35) shows that, when viewed as an operator on L2 ( , ), C is positive and symmetric on the domain Y. Therefore, there exists a positive, self-adjoint extension of C, which we shall denote as D : L2 ( , ) L2 ( , ). Consider rst a solution W which is of the form = /t for some W . Then for all t we have t = t , so t Y (and, of course, by denition of Y, we have t Y for all t). For each t, dene t = cos(D 1/2 t)t D 1/2 sin(D 1/2 t) t . (45)

Note that this formula corresponds to using the given prescription to evolve forward by t to obtain (t , t ), and then using the prescription (4) with AE = D to evolve backwards by t. Since by lemma 3.2 t and t are strongly differentiable in t and since

Dynamics in non-globally-hyperbolic static spacetimes: II

3825

t Y Dom(D) Dom(D 1/2 ), it follows that t is strongly differentiable in t and its derivative as a vector in L2 ( , ) is given by dt dt d t = D 1/2 sin(D 1/2 t)t + cos(D 1/2 t) cos(D 1/2 t) t D 1/2 sin(D 1/2 t) . dt dt dt (46) However, by lemma 3.2 we have dt /dt = t . Furthermore, we have d t (47) = t = Bt = Dt , dt where the rst equality follows from lemma 3.2, the second from equation (33) and the third from the fact that D agrees with B for all vectors in Y. We therefore obtain dt /dt = 0 and thus t = 0 = 0 , i.e., we have found that 0 = cos(D 1/2 t)t D 1/2 sin(D 1/2 t) t . In a similar manner, it follows that 0 = cos(D 1/2 t) t D 1/2 sin(D 1/2 t)(Dt ). These equations can be inverted to yield t = cos(D 1/2 t)0 + D 1/2 sin(D 1/2 t) 0 . (50) (49) (48)

This proves that all solutions of the form = /t are as claimed in the theorem, with AE = D. Now consider a solution W arising from arbitrary initial data (0 , 0 ) C0 ( ) C0 ( ). Let denote the solution dened by equation (4) with AE = D. It was proved in [2] that is a smooth solution to the KleinGordon equation (1) with initial data (0 , 0 ). It also follows from the analysis of [2] that d t /dt exists (in the strong Hilbert space sense in L2 ( , )) and equals ( /t)| t , where t | t . However, differentiation of equation (4) yields d t = cos(D 1/2 t) 0 + D 1/2 sin(D 1/2 t)(D0 ). (51) dt Equation (51) shows that /t is the solution given by equation (4) with AE = D for the initial data ( 0 , A0 ). However, by assumption 2(i), the solution with initial data ( 0 , A0 ) as given by our prescription is /t. Furthermore, we have just shown above that for solutions of the form /t, our prescription must agree with the prescription given by equation (4) with AE = D. It therefore follows that /t = /t. However, since | = | , it follows immediately that = , i.e., our prescription for obtaining a solution associated with arbitrary data (0 , 0 ) C0 ( )C0 ( ) agrees with the prescription dened by equation (4) with AE = D. Acknowledgments We wish to thank Stefan Hollands and Itai Seggev for helpful discussions. This research was initiated at the Yukawa Institute, and RMW wishes to thank the Yukawa Institute for its hospitality. AI thanks the Enrico Fermi Institute for its hospitality. This research was supported in part by the Japan Society for the Promotion of Science and by NSF grant PHY 00-90138 to the University of Chicago.

3826

A Ishibashi and R M Wald

References
[1] [2] [3] [4] Wald R M 1984 General Relativity (Chicago: University of Chicago Press) Wald R M 1980 J. Math. Phys. 21 2802 Horowitz G T and Marolf D 1995 Phys. Rev. D 52 5670 Ishibashi A and Hosoya A 1999 Phys. Rev. D 60 104028