GEOMETRIC ANALYSIS OF ORI-TYPE SPACETIMES J.DIETZANDA.DIRMEIER2 ANDM.SCHERFNER2 Abstract. In 1993 A. Ori [1] presented spacetimes violating the chronology conditioninordertoanswerthequestionwhetheratimemachineconstruction has to violate the weak energy condition or not. Later, in 2005 [2], he con- structed a class of time machine solutions with compact vacuum core. Both classesincludeaninterestingglobalstructureanditispossibletoobtainclosed timelikecurves. Besideswefocusonthegeometricstructure,inparticularsym- metriesandgeodesics,iffeasible,andvisualizeseveralaspects. 2 1 0 2 1. Introduction n a The spacetimes analyzed here, known as Ori-type spacetimes, were foremost de- J scribedin[1]and[2]. Thefirstpaperpresentsatimemachinemodelinwhichclosed 9 timelike curves (CTCs) evolve in a bounded region of space from a well-behaved ] c spacelikeinitialslice. ThissliceS,justastheentirespacetime,isasymptoticallyflat q and topologically trivial. In addition, this model fulfills the weak energy condition - r on S and up until and beyond the time slice, displaying the causality violation. g [ Thesecondpaperusesparticularvacuumsolutionstoconstructtimemachinemod- 1 elswherethecausalityviolationoccursinsideanemptytorus,constitutingthecore v of the time machine. In that case the matter field surrounding the (empty) torus 9 2 satisfies the weak, dominant, and strong energy conditions. 9 1 Here we will comprehensively discuss the geometric and CTC structure of the class . 1 ofRicciflatchronologyviolatingspacetimesgivenin[2]. Theanalysisfortheother 0 class discussed in [1] will only be given as a supplement, since—although it is the 2 1 predecessor in some sense—the details are much harder to reveal because of the : more complicated metric tensor, so here the research is still going on. v i X The structure of other particular Ori-type spacetimes—pseudo Schwarzschild and r pseudo Kerr—are discussed in [3]. a Before starting with our main investigations we describe the important underlying time machine structure. 2. Preliminaries Particles move through spacetime along causal curves, null curves are the trajecto- riesofmasslessparticleswhereasparticleswithnon-vanishingmassaredescribedby timelike curves. Suppose now that we are given a spacetime M containing CTCs. The physical interpretation would be that a particle moving once around a CTC would encounter itself in its own past. A similar argument can be adduced in the 1 case of null curves. This is the reason why spacetimes with closed causal curves (CCCs) have been qualified as time machine spacetimes. While it is easy to construct spacetimes containing CTCs or CCCs, it is more diffi- culttofindexamplesofspacetimesthatare—tosomeextent—physicallyreasonable. In [5], A. Ori presents a list, setting constraints on what should be classified as a realistic time machine model. One essential notion in General Relativity is the Cauchy development or domain of dependence of a subset A of a spacetime M. It is connected with the problem of formulating the Einstein equations as a well posed initial value problem and the concept of global hyperbolicity. FromarealisticpointofviewitisdesirabletoobtainCCCsastheunevitableconse- quenceofaCauchydevelopment. However, theinterioroftheCauchydevelopment of an achronal subset A is globally hyperbolic and, therefore, excludes any CCCs. The most one can hope for is that there are points in the boundary of the Cauchy developmentwhichlieonCCCs. Thisisthecentralideaofatimemachinestructure (TM-structure) for a spacetime. We will consider M to be a four-dimensional manifold and a metric tensor g on M withsignature(−+++), suchthattheLorentzianmanifold(M,g)togetherwitha fixed time-orientation is a spacetime. Let V(M) denote the set of all points p∈M that lie on some closed causal curve in M. We shall refer to V(M) as the causality violatingregionorthetimemachine. Ifp∈V(M), wesaythatcausalityisviolated at p. The causality condition is said to hold on a subset U ⊂M if V(M)∩U =∅. We say that (M,g) has TM-structure if the following three properties hold for M: (TM1) ThereexistsanopensubsetU ofM onwhichthecausalityconditionholds, (TM2) there is a spacelike hypersurface S contained in U, such that (TM3) for an achronal, compact subset C ⊂ S the future domain of dependence of C contains points in its boundary at which causality is violated, i.e., D+(C)∩V(M)(cid:54)=∅. InthiscasethecausalityviolatingregionV(M)issaid to be compactly constructed. This seems to be a reasonable condition for any spacetime that may be viewed as a time machine model. Although there definitely are many more additional conditionsonecouldimpose,wewillfocusonTM-structurehere. Itshouldbenoted that models with TM-structure can satisfy the weak, strong and dominant energy condition (see [6]) for the matter content of the spacetime if properly constructed. Example: Consider the smooth manifold M :=S1×R3 with the global coordinate system (φ,x,y,z) on M, φ being a circular coordinate. Define the metric tensor on M by (1) g :=−dφ⊗dφ+dx⊗dx+dy⊗dy+dz⊗dz. Then(M,g)becomesavacuumspacetime,theso-called(four-dimensional)Lorentz cylinder if we require the timelike vector field ∂ to be future pointing. We see φ 2 immediately that all φ-coordinate lines are CTCs. Hence, every point of M lies on a closed timelike curve. It can be shown that V(M) = ∅ for any globally hyperbolic spacetime M (cf. [4], Chapter 14). On the other hand, in the case of the Lorentz cylinder, we have V(M)=M. Mainlybecauseofphysicalreasons(oreverydayexperience)wedonot want to consider the Lorentz cylinder as a spacetime with TM-structure. The idea behind a TM-structure is the following: One can think of D+(C) as the set of points p ∈ M that are predictable from C in the sense that no inextendible causal curve through q ∈D+(C) can avoid C. Because the set intD+(C) does not contain points at which causality is violated (for it is a subset of D+(C) on which the so-called strong causality condition holds), the most one can hope for is that there are points in the boundary of D+(C) contained in V(M). As it can be seen from the definition of the TM-structure, spacelike hypersurfaces are an important entity in the context of TM-structures. Therefore, we include the following criterion. Proposition 1. Let f: M →R be a smooth function on a Lorentzian manifold M and a∈R be in the image of f. Then the sets (2) H :={p∈f−1(a): (cid:104)(gradf) ,(gradf) (cid:105)>0} t p p and (3) H :={p∈f−1(a): (cid:104)(gradf) ,(gradf) (cid:105)<0} s p p are hypersurfaces in M and H is timelike, H spacelike. t s Proof: The set U := {p ∈ M: (cid:104)(gradf) ,(gradf) (cid:105) > 0} ⊂ M is open, hence a p p Lorentzianmanifoldintheusualmanner. BecauseofthedefiningpropertyofU the one form df| is nowhere vanishing. Therefore, H is a hypersurface in U, hence in U t M. If p∈H and u∈T H , we have t p t (4) g((gradf) ,u)=u(f)=u(f| )=0 p Ht since f is constant on H . This proves the decomposition t (5) T M =T H ⊕R(gradf) p p t p and because (gradf) is spacelike this implies that g| is of index 1, i.e., H p TpHt t is timelike. If we consider H , almost the same arguments apply, except that now s (gradf) istimelike. HenceT H isaspacelikesubspaceofT M,i.e,H isspacelike. p p s p s (cid:3) In coordinates x1,...,xn the differential df of f is df = ∂f dxi and ∂xi ∂f ∂f (6) (cid:104)gradf,gradf(cid:105)=gij . ∂xi∂xj Using this formula and applying the preceding proposition to the coordinate neigh- borhood, one obtains: 3 Proposition 2. If x1,...,xn are local coordinates of a Lorentzian manifold M, the coordinate slices xk =const for 1≤k ≤n are (i) spacelike hypersurfaces if gkk <0, (ii) timelike hypersurfaces if gkk >0 for all points p in the coordinate neighborhood. 3. A Class of Ricci Flat Chronology Violating Spacetimes Here we investigate spacetimes described in [2] and we start with the smooth man- ifold M := R3×S1 and introduce coordinates (T,x,y,φ) on M, where T, x and y arenaturalcoordinatesonR3 andφisacircularcoordinateonS1. Foranysmooth function f: R2×S1 →R we define the metric tensor g on M by (7) g :=−dT ⊗ dφ+dx⊗dx+dy⊗dy+(f(x,y,φ)−T)dφ⊗dφ, s with ⊗ being the symmetrized tensor product. Emphasizing the role of f, we s write M for the semi-Riemannian manifold (M,g). The periodicity of f in its f third argument guarantees that g is well defined on all of M . Since the matrix of f thecomponentfunctionsofg hasdet[g ]=−1,themetrictensorisnon-degenerate ij and has Lorentzian signature. The Ricci tensor of M is given by f 1(cid:18)∂2f ∂2f(cid:19) (8) Ric=− + dφ⊗dφ, 2 ∂x2 ∂y2 so we assume that f satisfies the relation (9) f +f =0 xx yy to get Ricci flat spacetimes M . As the timelike unit vector field f 1 (10) ∂ + (f −T +1)∂ φ 2 T shows, M is time orientable. f 3.1. General Properties. Beforewemakeanexplicitchoiceforf andembarkon a more detailed study of a special example of the described class, we shall consider the general case. 3.1.1. Closed Timelike Curves in M . Of course, it is the topological factor S1 in f M and the metric component g = f −T that is responsible for the appearance f φφ of CTCs. Given T, x and y the curves (11) [−π,π]→M, s(cid:55)→γ (s):=(T,x,y,s) T,x,y are closed and timelike provided (12) f(x,y,s)−T <0 for all s∈[−π,π]. 4 Depending on the explicit form of f, this is a condition on the coordinates (T,x,y) determining a non-empty subset of R3 that consists of points sitting on CTCs of the form (11). By Prop. 2, the hypersurfaces H defined by T =const are T  (cid:41) timelike >0, (13) if gTT =T −f is spacelike <0, respectively. Therefore, any timelike curve in H , in particular CTCs, must be T contained in that part of H where T − f > 0. The same result may also be T obtained by calculating g(γ(cid:48),γ(cid:48)) directly with the condition dT(γ(cid:48)) = 0. We will seelaterforonechoiceoff howthesetwoingredients—theregionofCTCsandthe causal character of H —allow for a TM-structure of M . T f 3.1.2. KillingVectorFieldsofM . LetLbetheLiederivative,suchthattheKilling f equation for a vector field X is given by L g = 0. Furthermore we denote the X algebra of Killing vector fields on some open subset U ⊂ M of a manifold M by i(U). We will make repeated use of the following proposition, the proof of which can be found in [10], p. 61. Here R denotes the curvature tensor of M. Proposition 3. If X ∈i(M) is a Killing field, then L D(k)R=0 for all k ∈N, X i.e., the Lie derivative in the direction of X of any covariant derivative of the Riemann curvature tensor vanishes. Let L denote the linear system (L D(k)R) k X p for some p∈M and k ∈N. Then dimi(M)≤dimkerL . k For arbitrary f, depending on all three coordinates x, y and φ, there is no reason to expect an abundance of non-trivial, linearly independent Killing vector fields on M . There is only one obvious and non-tivial Killing vector field, which is actually f not even globally defined. Proposition 4. For arbitrary f there is only one local Killing vector field K on M . In coordinates, f (14) K =e−φ2∂T. Proof: First of all, note that K cannot be extended to all of M . Formally not f entirelycorrect,wemaysaythatthisisduetothefactthatthecomponentfunction e−φ2 of K is not properly periodic in φ. If φ takes values in (−π,π) the problem arises at points p∈M which are not covered by the coordinate system (T,x,y,φ). f In order to have an atlas of M at our disposal, we use a second coordinate system f (T,x,y,ψ)withψmappinginto(0,2π). Weshallrefertothefirstcoordinatesystem as A and to the second as B. The transformation formula (cid:40) φ if 0<φ<π, (15) ψ(φ)= φ+2π if −π <φ<0, 5 implies the coordinate expression  e−ψ2∂T if 0<φ<π, (16) K = e−ψ2−π∂T if −π <φ<0, for K in system B. If K were extendible to all of M , using B, the two limits f (17) lim Kp =e−π2∂T ψ(cid:37)π and (18) lim Kp = lim e−ψ2−π∂T =e−32π∂T ψ(cid:38)π ψ(cid:38)π would necessarily be the same. Next we verify that K is indeed a Killing vector field. Generally,foravectorfieldX :=G∂ +H∂ +P∂ +Q∂ theKillingequation T x y φ L g =0 is equivalent to the following system of PDE’s: X Q =0 T H −Q =0 T x P −Q =0 T y P =0 y H =0 x P +H =0 x y G −fQ +TQ +Q =0 T T T φ G −fQ +TQ −H =0 x x x φ G −fQ +TQ −P =0 y y y y −2G +2Q f +Qf −2TQ +Hf +Pf −G=0 φ φ φ φ x y The assumption H = P = Q = 0 and the supposition that G depends only on φ reduces this system of PDEs to the simple ordinary differential equation (19) G+2G =0. φ Hence G(φ) = e−φ2 and K is a Killing vector field. It remains to show that there is a choice for f, such that there is no other solution to the Killing equation which is not a constant multiple of K. This is the point where Prop. 3 comes in. If we choose (20) f(x,y,φ):=exycosφ, the equation L DR = 0 evaluated at (T,x,y,φ) = (1,1,1,0) becomes a linear X system with kernel of dimension 1. (cid:3) The fact that K cannot be extended to a global Killing field is due to the topology of M . If we consider the universal covering k: R4 → M of M , assign to R4 the f f f pull-back metric k∗g and use global coordinates (T,x,y,z) on R4, the vector field (21) K˜ :=e−z2∂z is a global Killing field on R4, irrespective of the explicit form of f. Changing the z-coordinate to the periodic coordinate φ implies loosing the global Killing field K˜. 6 3.2. One Special Choice for f. We are now going to consider a special example of the class of spacetimes described so far by choosing f to be a (22) f(x,y,φ):= (x2−y2), 2 where a is a positive constant. For notational convenience we drop the subscript f in M from now on. In order to obtain a coordinate system which is better suited f to the description of the region of CTCs in M we use the transformation: (23) R3×S1 →R3×S1, (T,x,y,φ)(cid:55)→(t,x,y,φ) with a (24) t:=T − (x2−y2)+eρ2. 2 Here, e is another positive constant and ρ2 :=x2+y2. It is clear that (t,x,y,φ) is a coordinate system. A short calculation yields (25) g =−dt⊗ dφ+(2e−a)xdx⊗ dφ+(2e+a)ydy⊗ dφ s s s +dx⊗dx+dy⊗dy+(eρ2−t)dφ⊗dφ in the new coordinates. 3.2.1. CTCs in M. For fixed values of t, x and y the curves γ (s) = (t,x,y,s) t,x,y are closed and  spacelike for t<eρ2, (26) null for t=eρ2, timelike for t>eρ2. For t < 0 all these curves are spacelike. At t = 0 the curve γ is a closed null 0,0,0 geodesic, as will follow from later results. For a given value of t all curves γ t,x,y lie in the hypersurface H given by t = const. The causal character of H can be t t inferred from (27) gtt =t+(2e−a)2x2+(2e+a)2y2−eρ2. If we suppose (28) (2e+a)2 <e, then H is spacelike throughout for t < 0. At t = 0, H is spacelike except at the t t closed null geodesic x=y =0. In the region t>0,  spacelike if (x,y) lies outside E , t (29) H is t timelike if (x,y) lies inside Et, where E is the ellipse in the x-y plane described by gtt =0 or t (30) t=[e−(2e−a)2]x2+[e−(2e+a)2]y2. Thus any CTC in H must be entirely contained in E (see Fig. 1). t t In [2], Ori describes in some detail how to construct a spacelike hypersurface S in the region {p ∈ M: t(p) ≤ 0} of M that contains a compact subset C, such that 7 y filled with CTC's t'>t t x Figure 1. A section φ = const in H and the circle of radius t t in the x-y plane filled with CTCs of the form γ (we have set t,x,y e=1). The growth of this circle with t is indicated by the arrows. The ellipse E confines the region of CTCs in H . t t the closed null curve γ is contained in D+(C), making M a candidate for a 0,0,0 spacetime with TM-structure. 3.2.2. Killing Vector Fields of M. We have already seen that due to the topology of M the Killing field K, which exists on M independently of the explicit form of f, isnotaglobalKillingfield. Wewillbeaccompaniedbythisproblemthroughout this section and, therefore, we concentrate our efforts on the part U of M covered by the coordinates (t,x,y,φ), where φ∈(−π,π). However, since the expression for f does not contain the coordinate φ we immediately conclude that the vector field ∂ is a global Killing field on M; it will be the only one. φ Proposition 5. The dimension of i(U) satisfies (31) dimi(U)≤6 and K6 =∂φ and K5 =e−φ2∂t are Killing fields on U. 8 Proof: We use Prop. 3 to get the upper bound for dimi(U). At the point p = (1,1,1,0) the linear system L R = 0 consists of the following four linearly X independent equations: a (L R) = X X 4241 2 1;2 a (L R) = − X X 4341 2 1;3 a (L R) =− X4 −(2e−a)aX +aX X 4242 2 1;2 1;4 a a (L R) = X − (2e−a)X −aX X 4342 2 1;2 2 1;3 2;3 Hence,theresultfollows. Since∂ =∂ ,thevectorfieldK isaKillingfieldbecause t T 5 of Prop. 4. (cid:3) The indices of K and K in the foregoing proposition indicate that indeed 5 6 dimi(U) = 6. In order to find the other four Killing fields we have to write down explicitly the equations given by the Killing equation L g =0. If we set X (32) X :=G∂ +H∂ +P∂ +Q∂ , t x y φ these are: (33) Q =0 t (34) H +(2e−a)xQ −Q =0 t t x (35) P +(2e+a)yQ −Q =0 t t y (36) −G +(2e−a)xH +(2e+a)yP +(eρ2−t)Q −Q =0 t t t t φ (37) H +(2e−a)xQ =0 x x (38) P +(2e+a)yQ +H +(2e−a)xQ =0 x x y x −G +(2e−a)H +(2e−a)xH +(2e+a)yP x x x (39) +(eρ2−t)Q +H +(2e−a)xQ =0 x φ φ (40) P +(2e+a)yQ =0 y y −G +(2e−a)xH +(2e+a)P +(2e+a)yP y y y (41) +(eρ2−t)Q +P +(2e+a)yQ =0 y φ φ −G−2tQ +2e(xH +yP)+2eρ2Q φ φ (42) +2(2e+a)yP +2(2e−a)xH −2G =0 φ φ φ The following two assumptions will simplify this system: • Q=0; • H and P are functions of φ only. 9 Then eqs. (33)-(35), (37), (38) and eq. (40) are trivially satisfied and we are left with: (43) G =0 t (44) −G +(2e−a)H +H =0 x φ (45) −G +(2e+a)P +P =0 y φ (46) −G+2e(xH +yP)+2(2e+a)yP +2(2e−a)xH −2G =0 φ φ φ At this point there are two similar cases. 1. Suppose P =0. The last three equations simplify to: (47) G =0 y (48) −G +(2e−a)H +H =0 x φ (49) −G+2exH +2(2e−a)xH −2G =0 φ φ Because of (43) and (47), G does not depend on t or y. Since H is a function of φ only, we deduce from (48) that (50) G(x,φ)=x(H (φ)+(2e−a)H(φ)), φ where we have set a possible integration constant equal to zero. Substituting this into (49) results in the ODE (51) 2H +H −aH =0. φφ φ The ansatz H(φ)=e−λφ leads to the polynomial equation (52) 2λ2−λ−a=0 with zeros 1(cid:0) √ (cid:1) 1(cid:0) √ (cid:1) (53) α:= 1+ 1+8a and β := 1− 1+8a , 4 4 and the solutions to (51) are (54) H (φ):=e−αφ and H (φ):=e−βφ. 1 2 For G we calculate (55) G (x,φ):=(2e−a−α)xe−αφ and G (x,φ):=(2e−a−β)xe−βφ. 1 2 With these choices for H and G all 10 components of the Killing equation are sat- isfied. 2. Suppose H =0. From (44), (45) and (46) we now get: (56) G =0 x (57) −G +(2e+a)P +P =0 y φ (58) −G+2eyP +2(2e+a)yP −2G =0 φ φ Reasoning as in the first case leads to (59) G(y,φ)=y(P (φ)+(2e+a)P(φ)) φ 10