Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Proof of the cosmic no-hair conjecture in the T^3-Gowdy symmetric Einstein-Vlasov setting

Proof of the cosmic no-hair conjecture in the T^3-Gowdy symmetric Einstein-Vlasov setting The currently preferred models of the universe undergo accelerated expansion induced by dark energy. One model for dark energy is a positive cosmological constant. It is consequently of interest to study Einstein’s equations with a positive cosmological constant coupled to matter satisfying the ordinary energy conditions; the dominant energy condition etc. Due to the difficulty of analysing the behaviour of solutions to Einstein’s equations in general, it is common to either study situations with symmetry, or to prove stability results. In the present paper, we do both. In fact, we analyse, in detail, the future asymptotic behaviour of T -Gowdy symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant. In particular, we prove the cosmic no-hair conjecture in this setting. However, we also prove that the solutions are future stable (in the class of all solutions). Some of the results hold in a more general setting. In fact, we obtain conclusions concerning the causal structure of T -symmetric solutions, assuming only the presence of a positive cosmological constant, matter satisfying various energy conditions and future global existence. Adding the 3 0 assumption of T -Gowdy symmetry to this list of requirements, we obtain C -estimates for all but one of the metric components. There is consequently reason to expect that many of the results presented in this paper can be generalised to other types of matter. 1 Introduction At the end of 1998, two research teams studying supernovae of type Ia announced the unexpected conclusion that the universe is expanding at an accelerating rate; cf. [27, 18]. After the obser- vations had been corroborated by other sources, there was a corresponding shift in the class of solutions to Einstein’s equations used to model the universe. In particular, physicists attributed the acceleration to a form of matter they referred to as ’dark energy’. However, as the nature of the dark energy remains unclear, there are several models for it. The simplest one is that of a positive cosmological constant (which is the one we use in the present paper), but there are several other possibilities; cf., e.g., [24, 25, 26] and references cited therein for some examples. Combining the different observational data, the currently preferred model of the universe is spatially homoge- neous and isotropic (i.e., the cosmological principle is assumed to be valid), spatially flat, and has matter of the following forms: ordinary matter (usually modelled by a radiation fluid and dust), dark matter (often modelled by dust), and dark energy (often modelled by a positive cosmological constant). In the present paper, we are interested in the Einstein-Vlasov system. This corresponds to a different description of the matter than the one usually used. However, this system can also be used in order to obtain models consistent with observations; cf., e.g., [31, Chapter 28]. In fact, Vlasov matter has the property that it naturally behaves as radiation close to the singularity and as dust in the expanding direction, a desirable feature which is usually put in by hand when using perfect fluids to model the matter. The cosmic no-hair conjecture. The standard starting point in cosmology is the assumption of spatial homogeneity and isotropy. However, it is preferable to prove that solutions generally arXiv:1306.6223v2 [gr-qc] 16 May 2016 isotropise and that the spatial variation (as seen by observers) becomes negligible. This is expected to happen in the presence of a positive cosmological constant; in fact, solutions are in that case expected to appear de Sitter like to observers at late times. The latter expectation goes under the name of the cosmic no-hair conjecture; cf. Conjecture 11 for a precise formulation. The main objective when studying the expanding direction of solutions to Einstein’s equations with a positive cosmological constant is to verify this conjecture. Spatial homogeneity. Turning to the results that have been obtained in the past, it is natural to begin with the spatially homogeneous setting. In 1983, Robert Wald wrote a short, but remarkable, paper [40], in which he proves results concerning the future asymptotic behaviour of spatially homogeneous solutions to Einstein’s equations with a positive cosmological constant. In particular, he confirms that the cosmic no-hair conjecture holds. What is remarkable about the paper is the fact that he is able to obtain conclusions assuming only that certain energy conditions hold and that the solution does not break down in finite time. Concerning the symmetry type, the only issue that comes up in the argument is whether it is compatible with the spatial hypersurfaces of homogeneity having positive scalar curvature or not; positive scalar curvature of these hypersurfaces sometimes leads to recollapse. The results should be contrasted with the case of Einstein’s vacuum equation in the spatially homogeneous setting, where the behaviour is strongly dependent on the symmetry type. Since Wald does not prove future global existence, it is necessary to carry out a further analysis in order to confirm the picture obtained in [40] in specific cases. In the case of the Einstein-Vlasov system, this was done in [13]. It is also of interest to note that it is possible to prove results analogous to those of Wald for more general models for dark energy; cf., e.g., [24, 25, 26, 14]. Surface symmetry. Turning to the spatially inhomogeneous setting, there are results in the surface symmetric case with a positive cosmological constant; cf. [37, 38, 39, 15], and see [22] for a definition of surface symmetry. In this case, the isometry group (on a suitable covering space) is 3-dimensional. Nevertheless, the system of equations that result after symmetry reduction is 1 + 1-dimensional. However, the extra symmetries do eliminate some of the degrees of freedom. Again, the main results are future causal geodesic completeness and a verification of the cosmic no-hair conjecture. T -symmetry. A natural next step to take after surface symmetry is to consider Gowdy or T -symmetry. That is the purpose of the present paper. In particular, we prove future causal geodesic completeness of solutions to the T -Gowdy symmetric Einstein-Vlasov equations with a positive cosmological constant (note, however, the caveat concerning global existence stated in Subsection 1.1). Moreover, we verify that the cosmic no-hair conjecture holds. It is of interest to note that most of the arguments go through under the assumption of T -symmetry. However, in order to obtain the full picture in this setting, it is necessary to prove one crucial inequality, cf. Definition 1, which we have not yet been able to do in general. Stability. A fundamental question in the study of cosmological solutions is that of future stability: given initial data corresponding to an expanding solution, do small perturbations thereof yield maximal globally hyperbolic developments which are future causally geodesically complete and globally similar to the future? In the case of a positive cosmological constant, the first result was obtained by Helmut Friedrich; he proved stability of de Sitter space in 3 + 1 dimensions in [10]. Later, he and Michael Anderson generalised the result to higher (even) dimensions and to include various matter fields; cf. [11, 1]. Moreover, results concerning radiation fluids were obtained in [16]. However, conformal invariance plays an important role in the arguments presented in these papers. As a consequence, there seems to be a limitation of the types of matter models that can be dealt with using the corresponding methods. The paper [28] was written with the goal of developing methods that are more generally applicable. The papers [29, 36, 32, 34, 35, 12], in which the methods developed in [28] play a central role, indicate that this goal was achieved. In fact, a general future global non-linear stability result for spatially homogeneous solutions to the Einstein-Vlasov equations with a positive cosmological constant was obtained in [31], the ideas developed in [28] being at the core of the argument. In the present paper, we not only derive 2 3 detailed future asymptotics of T -Gowdy symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant. We also prove that all the resulting solutions are future stable in the class of all solutions (without symmetry assumptions). Outlook. As we describe in the next subsection, some of the results concerning T -Gowdy symmetric solutions hold irrespective of the matter model (as long as it satisfies the dominant energy condition and the non-negative pressure condition). As a consequence, we expect that it might be possible to derive detailed asymptotics in the case of the Einstein-Maxwell equations (with a positive cosmological constant), and in the case of the Einstein-Euler system (though the issue of shocks may be relevant in the latter case). Due to the stability results demonstrated in [36, 32, 34, 35], it might also be possible to prove stability of the corresponding solutions. 1.1 General results under the assumption of T -symmetry 2 2 T -symmetry. In the present paper, we are interested in T -symmetric solutions to Einstein’s equations. There are various geometric ways of imposing this type of symmetry (cf., e.g., [7, 33]), but for the purposes of the present paper, we simply assume the topology to be of the form I ×T , where I is an open interval contained in (0, ∞). If θ, x and y are ’coordinates’ on T and t is the coordinate on I, we also assume the metric to be of the form −1/2 λ/2 2 −1 2 P 2 −P 2 g = t e (−dt + α dθ ) + te [dx + Qdy + (G + QH)dθ] + te (dy + Hdθ) , (1) where the functions α > 0, λ, P , Q, G and H only depend on t and θ; cf., e.g., [33]. Note that translation in the x and y directions defines a smooth action of T on the spacetime (as well as on each constant t-hypersurface). Moreover, the metric is invariant under this action, and the corresponding orbits are referred to as the symmetry orbits, given by {t} × {θ} × T . Note that the area of the symmetry orbits is proportional to t. For this reason, the foliation of the spacetime corresponding to the metric form (1) is referred to as the constant areal time foliation. The case of T -Gowdy symmetry corresponds to the functions G and H being independent of time; again, there is a more geometric way of formulating this condition: the spacetime is said to be Gowdy symmetric if the so-called twist quantities, given by α β γ δ α β γ δ J = ǫ X Y ∇ X , K = ǫ X Y ∇ Y , (2) αβγδ αβγδ vanish, where X = ∂ and Y = ∂ are Killing fields of the above metric and ǫ is the volume form. x y A basic question to ask concerning T -symmetric solutions to Einstein’s equations is whether the maximal globally hyperbolic development of initial data admits a constant areal time foliation which is future global. There is a long history of proving such results. The first one was obtained by Vincent Moncrief, cf. [17], in the case of vacuum solutions with T -Gowdy symmetry. The case of T -symmetric vacuum solutions with and without a positive cosmological constant have also been considered in [8] and [5] respectively. Turning to Vlasov matter, [2] contains an analysis of the existence of foliations in the T -Gowdy symmetric Einstein-Vlasov setting. The corresponding results were later extended to the T -symmetric case in [3]. However, from our point of view, the most relevant result is that of [33]. Due to the results of this paper, there is, given T -symmetric initial data to the Einstein-Vlasov equations with a positive cosmological constant, a future global foliation of the spacetime of the form (1). In other words I = (t , ∞). Moreover, if the distribution function is not identically zero, then t = 0. Finally, if the initial data have Gowdy symmetry, then the same is true of the development. Strictly speaking, the future global existence result in [33] is based on the observation that the argument should not be significantly different from the proofs in [5, 8, 3]. It would be preferable to have a complete proof of future global existence in the case of interest here, but we shall not provide it in this paper. Results. Turning to the results, it is of interest to note that some of the conclusions can be obtained without making detailed assumptions concerning the matter content. For that reason, let us, for the remainder of this subsection, assume that we have a solution to Einstein’s equations 3 with a positive cosmological constant, where the metric is of the form (1), the existence interval I is of the form (t , ∞) and the matter satisfies the dominant energy condition and the non- negative pressure condition; recall that the matter is said to satisfy the dominant energy condition if T (u, v) ≥ 0 for all pairs u, v of future directed timelike vectors (where T is the stress energy tensor associated with the matter); and that it is said to satisfy the non-negative pressure condition if T (w, w) ≥ 0 for every spacelike vector w. To begin with, there is a constant C > 0 such that −3 1 α(t, θ) ≤ Ct for all (t, θ) ∈ [t +2, ∞)×S ; cf. Proposition 42. In fact, this conclusion also holds if we replace the cosmological constant with a non-linear scalar field with a positive lower bound; cf. Remark 43. One particular consequence of this estimate for α is that the θ-coordinate of a causal curve converges. Moreover, observers whose θ-coordinates converge to different θ-values are asymptotically unable to communicate. In this sense, there is asymptotic silence. In the case of Gowdy symmetry, more can be deduced. In fact, for every ǫ > 0, there is a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞)×S ; cf. Proposition 44. This estimate turns out to be of crucial importance also in the general T -symmetric case. For this reason, we introduce the following terminology. Definition 1. A metric of the form (1) which is defined for t > t for some t ≥ 0 is said to have 0 0 λ-asymptotics if there, for every ǫ > 0, is a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞) × S . Remark 2. All Gowdy solutions have λ-asymptotics under the above assumptions; cf. Proposi- tion 44. Proposition 3. Consider a T -symmetric solution to Einstein’s equations with a positive cos- mological constant. Assume that the matter satisfies the dominant energy condition and the non- negative pressure condition. Assume, moreover, that the corresponding metric admits a foliation of the form (1), on I × T , where I = (t , ∞) and t ≥ 0. Finally, assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0 such that 1 0 −1/2 λ(t, ·) + 3 ln t − 2 ln ≤ Ct , 4Λ −3/2 −1/2 0 0 t hα (t, ·)i + kQ(t, ·)k + kP (t, ·)k ≤ C, C C −3/2 1 1 kH (t, ·)k + kG (t, ·)k ≤ Ct t L t L for all (t, θ) ∈ [t , ∞) × S . Remark 4. The choice t = t + 2 may seem unnatural. However, we need to stay away from t 1 0 0 (since we do not control the solution close to t ). Moreover, in some situations we need to know that ln t is positive and bounded away from zero. Since t = 0 for most solutions, it is therefore natural to only consider the interval t ≥ t + 2 in the study of the future asymptotics. Remark 5. If h is a scalar function on S , we use the notation hhi = hdθ. (3) 2π Sometimes, we shall use the same notation for a scalar function h on I ×S . In that case, hhi is the function of t defined by hh(t, ·)i. Finally, if p¯ ∈ R , we shall also use the notation hp¯i. However, 2 1/2 in that case, hp¯i = (1 + |p¯| ) ; cf. Remark 19. 4 Proof. The statement is a consequence of Lemmas 46, 47 and 48. In particular, in the case of a T -Gowdy symmetric solution, there is asymptotic silence in the sense that the θxy-coordinates of a causal curve converge, and causal curves whose asymptotic θxy-coordinates differ are asymptotically unable to communicate; cf. Proposition 49. 1.2 Results in the Einstein-Vlasov setting In order to be able to draw detailed conclusions, we need to restrict our attention to a specific type of matter. In the present paper, we study the Einstein-Vlasov system. A general description of Vlasov matter. Intuitively, Vlasov matter gives a statistical descrip- tion of an ensemble of collections of particles. In practice, the matter is described by a distribution function defined on the space of states of particles. The possible states are given by the future directed causal vectors (here and below, we assume the Lorentz manifolds under consideration to be time oriented). Usually, one distinguishes between massive and massless particles. In the latter case, the distribution function is defined on the future light cone, and in the former case, it is defined on the interior. In the present paper, we are interested in the massive case, and we assume all the particles to have unit mass (for a description of how to reduce the case of varying masses to the case of all particles having unit mass, see [6]). As a consequence, the distribution function is a non-negative function on the mass shell P, defined to be the set of future directed unit timelike vectors. In order to connect the matter to Einstein’s equations, we need to associate a stress energy tensor with the distribution function. It is given by Vl T (r) = fp p µ . (4) α β P αβ r Pr In this expression, P denotes the set of future directed unit timelike vectors based at the spacetime point r. In other words, if T ∈ T M is a future directed timelike vector, then ref r P = {v ∈ T M | g(v, v) = −1, g(T , v) < 0}. r r ref Moreover, the Lorentz metric g induces a Riemannian metric on P , and µ denotes the corre- r P sponding volume form; cf. (18) below for a coordinate representation of µ . Finally, p denotes P α the components of the one form obtained by lowering the index of p ∈ P using the Lorentz metric g. Clearly, it is necessary to demand some degree of fall off of the distribution function f in order for the integral (4) to be well defined. In the present paper, we shall mainly be interested in the case that the distribution function has compact support in the momentum directions (for a fixed spacetime point). However, in Subsections 1.3–1.7 we shall consider a somewhat more general situation. Turning to the equation the distribution function has to satisfy, it is given by Lf = 0. (5) Here L denotes the vector field induced on the mass shell by the geodesic flow; cf. (19) below for a coordinate representation of it. An alternate way to formulate this equation is to demand that f be constant along γ˙ for every future directed unit timelike geodesic γ. The intuitive interpretation of the Vlasov equation (5) is that collisions between particles are neglected. It is of interest to note that if f satisfies the Vlasov equation, then the stress energy tensor is divergence free. To conclude, the Einstein-Vlasov equations with a positive cosmological constant consist of (5) and Vl Ein + Λg = T , (6) Vl where T is given by the right hand side of (4) and Λ is a positive constant. Moreover, Ein = Ric − Sg 5 is the Einstein tensor, where Ric is the Ricci tensor and S is the scalar curvature of a Lorentz manifold (M, g). The above description is somewhat brief, and the reader interested in more details is referred to, e.g., [9, 23, 4, 31]. 2 2 Vlasov matter under the assumption of T -symmetry. In the case of T -symmetry, it is convenient to use a symmetry reduced version of the distribution function. Introduce, to this end, the orthonormal frame 1/4 −λ/4 1/4 −λ/4 1/2 e =t e ∂ , e = t e α (∂ − G∂ − H∂ ), 0 t 1 θ x y (7) −1/2 −P/2 −1/2 P/2 e =t e ∂ , e = t e (∂ − Q∂ ). 2 x 3 y x Since the distribution function f is defined on the mass shell, it is convenient to parametrise this set; note that the manifolds we are interested in here are parallelisable (i.e., they have a global frame). An element in P can be written v e , where 0 1 2 2 2 3 2 1/2 v = [1 + (v ) + (v ) + (v ) ] . As a consequence, we can think of f as depending on v , i = 1, 2, 3, and the base point. However, due to the symmetry requirements, the distribution function only depends on the tθ-coordinates of the base point. As a consequence, the distribution function can be considered to be a function of 1 2 2 (t, θ, v), where v = (v , v , v ). In what follows, we shall abuse notation and denote the symmetry 1 3 reduced function, defined on I × S × R , by f. A symmetry reduced version of the equations is to be found in Section 2. Remark 6. In the T -symmetric setting, we always assume the distribution function f to have compact support when restricted to constant t-hypersurfaces. Under the assumptions made in the present paper, f has this property, assuming the initial datum for f to have compact support. The first question to ask concerning T -symmetric solutions is that of existence of constant areal time foliations for an interval of the form (t , ∞). However, due to previous results, cf. [33], we know that T -symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant are future global in this setting (keeping the caveat stated in Subsection 1.1 in mind). In other words, there is a t ≥ 0 such that the solution admits a foliation of the form (1) on I × T , where I = (t , ∞). Consequently, the issue of interest here is that of the asymptotics. Unfortu- nately, we are unable to derive detailed asymptotics for all T -symmetric solutions. However, we do obtain results for solutions with λ-asymptotics; recall that all T -Gowdy symmetric solutions fall into this class. Theorem 7. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics and let t = t + 2. Then 0 1 0 there are smooth functions α > 0, P , Q , G and H on S , and, for every 0 ≤ N ∈ Z, a ∞ ∞ ∞ ∞ ∞ constant C > 0 such that −3/2 tkH (t, ·)k N + tkG (t, ·)k N + kH(t, ·) − H k N + kG(t, ·) − G k N ≤ C t , (8) t t ∞ ∞ N C C C C −1 tkP (t, ·)k N + tkQ (t, ·)k N + kP (t, ·) − P k N + kQ(t, ·) − Q k N ≤ C t , (9) t t ∞ ∞ N C C C C α 3 3 −2 + + λ + ≤ C t , (10) t N α t t N N C C 3 −1 t α(t, ·) − α + λ(t, ·) + 3 ln t − 2 ln ≤ C t , (11) ∞ N N 4Λ −1/2 for all t ≥ t . Define f via f (t, θ, v) = f(t, θ, t v). Then there is an R > 0 such that 1 sc sc suppf (t, ·) ⊆ S × B (0) sc R 6 3 for all t ≥ t , where B (0) is the ball of radius R in R . Moreover, there is a smooth, non-negative 1 R 1 3 function with compact support, say f , on S × R , such that sc,∞ −1 tk∂ f (t, ·)k N 1 3 + kf (t, ·) − f k N 1 3 ≤ C t t sc sc sc,∞ N C (S ×R ) C (S ×R ) for all t ≥ t . Turning to the geometry, let g¯(t, ·) and k(t, ·) denote the metric and second funda- mental form induced by g on the hypersurface {t} × T , and let g¯ (t, ·) denote the components of ij g¯(t, ·) with respect to the vectorfields ∂ = ∂ , ∂ = ∂ and ∂ = ∂ etc. Then 1 θ 2 x 3 y −1 −1 −1 kt g¯ (t, ·) − g¯ k N + kt k − Hg¯ k N ≤ C t (12) ij ∞,ij ij ∞,ij N C C 1/2 for all t ≥ t , where H = (Λ/3) and 2 P∞ 2 −P∞ 2 g¯ = dθ + e [dx + Q dy + (G + Q H )dθ] + e (dy + H dθ) . (13) ∞ ∞ ∞ ∞ ∞ ∞ 4Λα Moreover, the solution is future causally geodesically complete. The proof of the above theorem is to be found in Section 10. It is of interest to record how the spacetime appears to an observer. In particular, we wish to prove the cosmic no-hair conjecture in the present setting. The rough statement of this conjecture is that the spacetime appears de Sitter like to late time observers. However, in order to be able to state a theorem, we need a formal definition. Before proceeding to the details, let us provide some intuition. Let 2 2Ht g = −dt + e g¯ , (14) dS E 1/2 4 where H = (Λ/3) and g¯ denotes the standard flat Euclidean metric. Then (R , g ) corre- E dS sponds to a part of de Sitter space. It may seem more reasonable to consider de Sitter space itself. However, as far as the asymptotic behaviour of de Sitter space is concerned, (14) is as good a model as de Sitter space itself. Consider a future directed and inextendible causal curve in 4 0 3 (R , g ), say γ = (γ , γ¯), defined on (s , s ). Then γ¯(s) converges to some x¯ ∈ R as s → s −. dS − + 0 + Moreover, γ(s) ∈ C for all s, where x ¯ ,Λ −1 −Ht C = {(t, x¯) : |x¯ − x¯ | ≤ H e }. x ¯ ,Λ 0 In practice, it is convenient to introduce a lower bound on the time coordinate and to introduce a margin in the spatial direction. Moreover, it is convenient to work with open sets. We shall therefore be interested in sets of the form −1 −Ht C = {(t, x¯) : t > T, |x¯| < KH e }; (15) Λ,K,T note that x¯ can be translated to zero by an isometry. Since we are interested in the late time behaviour of solutions, it is natural to restrict attention to sets of the form C for some K ≥ 1 Λ,K,T and T > 0. Definition 8. Let (M, g) be a time oriented, globally hyperbolic Lorentz manifold which is future causally geodesically complete. Assume, moreover, that (M, g) is a solution to Einstein’s equations with a positive cosmological constant Λ. Then (M, g) is said to be future asymptotically de Sitter like if there is a Cauchy hypersurface Σ in (M, g) such that for every future oriented and inextendible causal curve γ in (M, g), the following holds: − + • there is an open set D in (M, g), such that J (γ) ∩ J (Σ) ⊂ D, and D is diffeomorphic to C for a suitable choice of K ≥ 1 and T > 0, Λ,K,T −1 −Ht • using ψ : C → D to denote the diffeomorphism; letting R(t) = KH e ; using Λ,K,T g¯ (t, ·) and k (t, ·) to denote the metric and second fundamental form induced on S = dS dS t {t} × B (0) by g ; using g¯(t, ·) and k(t, ·) to denote the metric and second fundamental dS R(t) 7 ∗ ∗ form induced on S by ψ g (where ψ g denotes the pullback of g by ψ); and letting N ∈ N, the following holds: ¯ ¯ lim kg¯ (t, ·) − g¯(t, ·)k N + kk (t, ·) − k(t, ·)k N = 0. (16) dS dS C (S ) C (S ) t t dS dS t→∞ Remark 9. In the definition, we use the notation 1/2 i i j j 1 l 1 l im jn khk N = sup g¯ · · ·g¯ g¯ g¯ ∇ · · · ∇ h ∇ · · ·∇ h dS,i j dS,i j ij mn C (S ) 1 1 l l dS dS dS dS dS dS dS l=0 for a covariant 2-tensor field h on S , where ∇ denotes the Levi-Civita connection associated t dS with g¯ (t, ·). Note also that, even though R(t) shrinks to zero exponentially, the diameter of S , dS t as measured with respect to g¯ (t, ·), is constant. dS Remark 10. In some situations it might be more appropriate to adapt the Cauchy hypersurface Σ to the causal curve γ; i.e., to first fix γ and then Σ. The above definition leads to a formal statement of the cosmic no-hair conjecture. Conjecture 11 (Cosmic no-hair). Let A denote the class of initial data such that the corre- sponding maximal globally hyperbolic developments (MGHD’s) are future causally geodesically complete solutions to Einstein’s equations with a positive cosmological constant Λ (for some fixed matter model). Then generic elements of A yield MGHD’s that are future asymptotically de Sitter like. Remark 12. It is probably necessary to exclude certain matter models in order for the statement to be correct. Moreover, the statement, as it stands, is quite vague; there is no precise definition of the notion generic. However, what notion of genericity is most natural might depend on the situation. Remark 13. The Nariai spacetimes, discussed, e.g., in [28, pp. 126-127], are time oriented, globally hyperbolic, causally geodesically complete solutions to Einstein’s vacuum equations with a positive cosmological constant that do not exhibit future asymptotically de Sitter like behaviour. They are thus potential counterexamples to the cosmic no-hair conjecture. There is a similar example in the Einstein-Maxwell setting (with a positive cosmological constant) in [28, p. 127]. However, both of these examples are rather special, and it is natural to conjecture them to be unstable. Nevertheless, they constitute the motivation for demanding genericity. Finally, we are in a position to phrase a result concerning the cosmic no hair conjecture in the T -Gowdy symmetric setting. The proof of the theorem below is to be found in Section 10. Theorem 14. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics. Then the solution is future asymptotically de Sitter like; i.e., the cosmic no-hair conjecture holds. Remark 15. Recall that all T -Gowdy symmetric solutions have λ-asymptotics. Remark 16. In the particular case of interest here, the equality (16) can actually be improved to the estimate −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k N + kk (τ, ·) − k(τ, ·)k N ≤ C e dS dS N C (S ) C (S ) τ τ dS dS for all τ > T and a suitable constant C . Remark 17. The main estimate needed to prove the theorem is (12). In situations where such an estimate holds, it is thus to be expected that the solution is future asymptotically de Sitter like. 8 1.3 Stability, notation and function spaces Let us now turn to the subject of stability. Combining Theorem 7 with the results of [31], it turns out to be possible to prove that the solutions to which Theorem 7 applies are also future stable. In the present subsection, we begin by introducing the terminology necessary in order to make a formal statement of this result. Let (M, g) be a time oriented n+1-dimensional Lorentz manifold. We say that (x, U) are canonical local coordinates if ∂ 0 is future oriented timelike on U and g(∂ i| , ∂ j| ), i, j = 1, . . ., n, are the x x r x r components of a positive definite metric for every r ∈ U; cf. [31, p. 87]. If p ∈ P for some r ∈ U, we then define Ξ (p) = Ξ (p ∂ | ) = [x(r), p¯], (17) x x x r 1 n where p¯ = (p , . . . , p ). Note that Ξ are local coordinates on the mass shell. If f is defined on the −1 mass shell, we shall use the notation f = f ◦ Ξ . It is also convenient to introduce the notation ¯ ¯ p according to Ξ (p) = [x(r), p (p)], assuming p ∈ P . With this notation, the measure µ can x x x r P be written 1/2 |g (r)| µ = − ι dp¯ , (18) Pr p ◦ ι x,0 r where |g | is the determinant of the metric g, when expressed with respect to the x-coordinates; ι : P → P is the inclusion; p (p) are the components of p with respect to the coordinates x; and r r p (p) = g p (p). The reader interested in a derivation of (18) is referred to [31, Section 13.3]. x,α x,αβ Let us also note that the operator L is given by ∂ ∂ α i α β L = p − Γ p p (19) x αβ x x α i ∂x ∂¯ p with respect to the above coordinates. In order to proceed, we need to introduce function spaces for the distribution functions. Recall, to that end, [31, Definition 7.1, p. 87]: Definition 18. Let 1 ≤ n ∈ Z, µ ∈ R, (M, g) be a time oriented n + 1-dimensional Lorentz manifold and P be the set of future directed unit timelike vectors. The space D (P) is defined to consist of the smooth functions f : P → R such that, for every choice of canonical local coordinates (x, U), n + 1-multiindex α and n-multiindex β, the derivative ∂ ∂ f (where x symbolises the first x p¯ n + 1 and p¯ the last n variables), considered as a function from x(U) to the set of functions from R to R, belongs to 2 n C[x(U), L (R )]. (20) µ +|β| 2 n 2 Remark 19. The space L (R ) is the weighted L -space corresponding to the norm 1/2 2µ 2 khk = hp¯i |h(p¯)| dp¯ , (21) 2 1/2 where hp¯i = (1 + |p¯| ) ; recall the comments made in Remark 5. Remarks 20. If f ∈ D (P) for some µ > n/ 2 + 1, then the stress energy tensor is a well defined smooth function; cf. [31, Proposition 15.37, p. 246]. Moreover, the stress energy tensor is divergence free if f satisfies the Vlasov equation. It is worth pointing out that it is possible to introduce more general function spaces, corresponding to a finite degree of differentiability; cf. [31, Definition 15.1, p. 234]. However, the above definition is sufficient for our purposes. The above function spaces are suitable when discussing functions on the mass shell. However, we also need to introduce function spaces for the initial datum for the distribution function. If (¯x, U) are local coordinates on a manifold Σ, we introduce local ¯ ¯ ¯ coordinates on T Σ by Ξ (p¯ ∂ i| ) = (¯x(ξ), p¯) in analogy with (17). Moreover, if f is defined on ¯ x ¯ x −1 ¯ ¯ ¯ T Σ, we shall use the notation f = f ◦ Ξ . Let us recall [31, Definition 7.5, p. 89]: ¯ x ¯ x 9 ∞ Definition 21. Let 1 ≤ n ∈ Z, µ ∈ R and Σ be an n-dimensional manifold. The space D (T Σ) is defined to consist of the smooth functions f : T Σ → R such that, for every choice of local coordinates (¯x, U), n-multiindex α and n-multiindex β, the derivative ∂ ∂ f (where x¯ symbolises ¯ x x ¯ p¯ the first n and p¯ the last n variables), considered as a function from ¯x(U) to the set of functions from R to R, belongs to 2 n C[¯x(U), L (R )]. µ +|β| Remark 22. According to the criteria appearing in Definitions 18 and 21, we need to verify continuity conditions for every choice of local coordinates. However, it turns out to be sufficient to consider a collection of local coordinates covering the manifold of interest; cf. [31, Lemma 15.9, p. 235] and [31, Lemma 15.19, p. 237]. Finally, in order to be able to state a stability result, we need a norm. Recall, to this end, [31, Definition 7.7, pp. 89–90]: Definition 23. Let 1 ≤ n ∈ Z, 0 ≤ l ∈ Z, µ ∈ R and Σ be a compact n-dimensional manifold. Let, moreover, χ¯ , i = 1, ..., j, be a finite partition of unity subordinate to a cover consisting of coordinate neighbourhoods, say (¯x , U ). Then k · k l is defined by i i Vl,μ   1/2 j Z X X 2µ +2|β| α β 2 ¯  ¯ ¯ ¯ ¯  kfk l = h̺¯i χ¯ (ξ)(∂ ∂ f ) (ξ, ̺¯)dξd̺¯ (22) i ¯ ¯ x H ̺¯ i Vl,μ ¯ x (U )×R i i i=1 |α|+|β|≤l ¯ ¯ ∞ for each f ∈ D (T Σ). Remark 24. Clearly, the norm depends on the choice of partition of unity and on the choice of coordinates. However, different choices lead to equivalent norms. Here, we are mainly interested in the case Σ = T , in which case it is neither necessary to introduce local coordinates nor a partition of unity. 1.4 The Einstein-Vlasov-non-linear scalar field system In the present paper, we are mainly interested in the Einstein-Vlasov system with a positive cosmological constant. However, in the proof of future stability of T -Gowdy symmetric solutions, we use two results. First, we use the fact that solutions that start out close to de Sitter space are future stable. Second, we use Cauchy stability. There are results of this type in the literature. However, they are formulated in the Einstein-Vlasov-non-linear scalar field setting. In order to make it clear that the statements appearing in the literature can be applied in our setting, it is therefore necessary to briefly describe the Einstein-Vlasov-non-linear scalar field system. This is the purpose of the present subsection. In 3 + 1-dimensions, the Einstein-Vlasov-non-linear scalar field system can be written R − T + (trT )g = 0, (23) αβ αβ αβ α ′ ∇ ∇ φ − V ◦ φ = 0, (24) Lf = 0; (25) cf. [31, (7.13)–(7.15), p. 91]. In these equations, φ ∈ C (M) is referred to as the scalar field; V : R → R is a smooth function referred to as the potential; ∇ is the Levi-Civita connection associated with the metric g; and sf Vl T = T + T , αβ αβ αβ Vl where T is defined in (4) and sf γ T = ∇ φ∇ φ − ∇ φ∇ φ + V (φ) g . α β γ αβ αβ 10 ′ Assuming V to be such that V (0) = 0, it is consistent to demand that φ be zero in (24). Moreover, sf if φ = 0, then T = −V (0)g. Letting Λ = V (0), the equations (23)–(25) then reduce to the Einstein-Vlasov system with a positive cosmological constant Λ, assuming V (0) > 0. In order to prove future stability in the Einstein-Vlasov-non-linear scalar field setting, it is not sufficient ′ ′′ to demand that V (0) = 0 and V (0) > 0. It is also of interest to know that V (0) > 0. We shall therefore make this assumption from now on. Given V such that V (0) = 0, V (0) > 0 and ′′ V (0) > 0, it is convenient to introduce 1/2 H = (V (0)/3) (26) and ′′ 2 χ = V (0)/H ; (27) cf. [31, (7.9) and (7.10), p. 90]. Note that in the non-linear scalar field setting, we always assume V (0) to be positive and we equate it with Λ. In particular, (26) is thus consistent with previous definitions of H; cf., e.g, the statement of Theorem 7. In case we are interested in the Einstein- Vlasov system with a positive cosmological constant Λ, it is sufficient to choose V to be V (φ) = Λ + Λφ . (28) ′ ′′ 1/2 Then V (0) = Λ > 0, V (0) = 0 and V (0) = 2Λ > 0. Moreover, H = (Λ/3) and χ = 6. Clearly, (28) is an arbitrary choice; there are many other possibilities. Let us now recall the definition of initial data given in [31, Definition 7.11, pp. 93–94] (note that the dimension n is here assumed to equal 3): Definition 25. Let 5/2 < µ ∈ R. Initial data for (23)–(25) consist of an oriented 3-dimensional ¯ ¯ manifold Σ, a non-negative function f ∈ D (T Σ), a Riemannian metric g¯, a symmetric covariant ¯ ¯ ¯ 2-tensor field k and two functions φ and φ on Σ, all assumed to be smooth and to satisfy 0 1 ij 2 2 Vl ¯ ¯ ¯ ¯ ¯ ¯ ¯ ∇ φ ∇ φ + 2V (φ ) + 2ρ , (29) r¯− k k + (trk) = φ + ij 0 i 0 0 Vl ¯ ¯ ¯ ¯ ¯ ∇ k − ∇ (trk) = φ ∇ φ − J , (30) ji i 1 i 0 where ∇ is the Levi-Civita connection of g¯, r¯ is the associated scalar curvature, indices are raised Vl Vl and lowered by g¯ and ρ and J are given by (33) and (34) below respectively. Given initial data, the initial value problem is that of finding a solution (M, g, f, φ) to (23)–(25) (in other words, an 4- dimensional manifold M, a smooth time oriented Lorentz metric g on M, a non-negative function ∞ ∞ f ∈ D (P) and a φ ∈ C (M) such that (23)–(25) are satisfied), and an embedding i : Σ → M such that ∗ ∗ −1 ¯ ¯ i g = g¯, φ ◦ i = φ , f = i (f ◦ pr ) i(Σ) and if N is the future directed unit normal and κ is the second fundamental form of i(Σ), then ¯ ¯ i κ = k and (Nφ) ◦ i = φ . Such a quadruple (M, g, f, φ) is referred to as a development of the initial data, the existence of an embedding i being tacit. If, in addition to the above conditions, i(Σ) is a Cauchy hypersurface in (M, g), the quadruple is said to be a globally hyperbolic development. Remark 26. The map pr is the diffeomorphism from the mass shell above i(Σ) to the tangent i(Σ) space of i(Σ) defined by mapping a vector v to its component perpendicular to the normal of i(Σ). ¯ ¯ Remark 27. If φ = φ = 0, the equations (29) and (30) become 0 1 ij 2 Vl ¯ ¯ ¯ r¯− k k + (trk) = 2Λ + 2ρ , (31) ij Vl ¯ ¯ ¯ ∇ k − ∇ (trk) = −J . (32) ji i These are the constraint equations for the Einstein-Vlasov system with a positive cosmological constant Λ. 11 The energy density and current induced by the initial data are given by Vl 1/2 ¯ ¯ ρ (ξ) = f(p¯)[1 + g¯(p¯, p¯)] µ¯ , (33) g¯,ξ ¯Σ Vl ¯ ¯ ¯ ¯ J (X) = f(p¯)g¯(X, p¯)µ¯ ¯. (34) g¯,ξ T¯Σ ¯ ¯ In these expressions, ξ ∈ Σ, X ∈ T¯Σ, µ¯ ¯ is the volume form on T¯Σ induced by g¯ and p¯ ∈ T¯Σ. ξ g¯,ξ ξ ξ It is important to note that under the assumptions of the above definition, the energy density is a smooth function and the current is a smooth one-form field on Σ; cf. [31, Lemma 15.40, p. 246]. Given initial data, there is a unique maximal globally hyperbolic development thereof; cf. [31, Corollary 23.44, p. 418] and [31, Lemma 23.2, p. 398]. The definition of a maximal globally hyperbolic development is given by [31, Definition 7.14, p. 94]: Definition 28. Given initial data for (23)–(25), a maximal globally hyperbolic development of the data is a globally hyperbolic development (M, g, f, φ), with embedding i : Σ → M, such that ′ ′ ′ ′ if (M , g , f , φ ) is any other globally hyperbolic development of the same data, with embedding ′ ′ ′ i : Σ → M , then there is a map ψ : M → M which is a diffeomorphism onto its image such that ∗ ′ ∗ ′ ∗ ′ ′ ψ g = g , ψ f = f , ψ φ = φ and ψ ◦ i = i. It is worth noting that the maximal globally hyperbolic development is independent of the param- eter µ . The above discussion of the initial value problem for the Einstein-Vlasov-non-linear scalar field system is somewhat brief, and the reader interested in a more detailed discussion is referred to [31, Chapter 7]. 1.5 Future stability in the spatially homogeneous and isotropic setting In the proof of stability of the T -Gowdy symmetric solutions, we need to refer to [31, Theo- rem 7.16, pp. 104–106]. However, the statement of this theorem is based on terminology introduced in [31]. Moreover, in the statement of Theorem 35, we refer to the conclusions of [31, Theorem 7.16]. For this reason, we here provide not only the notational background, but also the statement of [31, Theorem 7.16]. However, the reader interested in a discussion giving a justification for why the particular formulation of the theorem is natural is referred to [31, Sections 7.6–7.7]. The rough idea of the statement is to only make local assumptions concerning the initial data and to derive future global conclusions concerning the solution. Given a 3-manifold Σ, we therefore focus on a local coordinate patch (¯x, U). Here U is the neighbourhood in which we make assumptions in the statement of the theorem. The conditions on the initial data are phrased in terms of Sobolev norms on U. Given a tensor field T on Σ, we therefore define   1/2 3 3 X X X α i ···i −1 2 1 s   (35) kTk l = |∂ T ◦ ¯x | dx¯ . H (U) j ···j 1 r ¯ x(U) i ,...,i =1 j ,...,j =1 1 s 1 r |α|≤l In this expression, the components of T are computed with respect to the coordinates ¯x and the derivatives are taken with respect to ¯x. In what follows, norms of the type kTk l are always H (U) computed using a particular choice of local coordinates. The choice we have in mind should be clear from the context. In the formulation of Theorem 29, we also use the notation   1/2 X X α −1 2   k∂ g¯k l = |∂ ∂ g¯ ◦ ¯x | dx¯ . (36) m m ij H (U) ¯ x(U) i,j=1 |α|≤l To measure the local size of the distribution function, we need a weighted Sobolev norm. However, it is also necessary to allow the freedom to rescale the momentum variable in the definition of the 12 norm. Since we have already motivated the need for this rescaling freedom in [31, Subsection 7.6.1, pp. 100–102], we shall not do so here. Given a constant w, we simply define the local norm for the distribution function by   1/2 Z Z w −w 2|β| w 2µ +2|β| α β 2 ¯  ¯ ¯ ¯  (37) kfk l = (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯ . ¯ ¯ x H (U) p¯ Vl,μ R ¯ x(U) |α|+|β|≤l −1 ¯ ¯ ¯ ¯ Here Ξ are the coordinates on TU associated with ¯x (cf. Subsection 1.3), and f = f ◦ Ξ . ¯ x ¯ x ¯ x Given the above notation, [31, Theorem 7.16, pp. 104–106] takes the following form for n = 3. Theorem 29. Let 5/2 < µ ∈ R and 7/2 < k ∈ Z. Let V be a smooth function on R such that ′ ′′ V (0) = V > 0, V (0) = 0 and V (0) > 0. Let H, χ > 0 be defined by (26) and (27) respectively and let K ≥ 0. There is an ε > 0, depending only on µ and V , such that if Vl ¯ ¯ ¯ ¯ • (Σ, g¯, k, f, φ , φ ) are initial data for (23)–(25) with dimΣ = 3, 0 1 ¯ ¯ • x : U → B (0) are local coordinates with x(U) = B (0), 1 1 • the inequality −2K |e g¯ − δ | ≤ ε (38) ij ij holds on U for all i, j = 1, ..., n, where K is defined by e = 4/H, • using the notation introduced in (35) and (36), the inequality 2 −1 −K ¯ ¯ ¯ Vl H k∂ g¯k k + Hkk − Hg¯k k + kφ k k +1 + H kφ k k ≤ εe j 0 0 0 0 1 0 (39) H (U) H (U) H (U) H (U) j=1 holds, • using the notation introduced in (37), the inequality w 2 5/2 −3K/2−K ¯ Vl kfk k ≤ H ε e (40) H (U) Vl,μ holds with w = K + K , Vl then the maximal globally hyperbolic development (M, g, f, φ) of the initial data has the property −1 that if i : Σ → M is the associated embedding, then all causal geodesics that start in i◦¯x [B (0)] 1/4 are future complete. Furthermore, there is a t < 0 and a smooth map ψ : (t , ∞) × B (0) → M, (41) − 5/8 −1 which is a diffeomorphism onto its image, such that all causal curves that start in i ◦ ¯x [B (0)] 1/4 remain in the image of ψ to the future, and g, f and φ have expansions of the form (42)–(55) −1 ¯ ¯ in the solid cylinder [0, ∞) × B (0) when pulled back by ψ. Finally, ψ(0, ξ) = i ◦ ¯x (ξ) for 5/8 ξ ∈ B (0). In the formulae below, Latin indices refer to the natural Euclidean coordinates on 5/8 B (0) and t is the natural time coordinate on the solid cylinder. Let ζ = 4χ/9, 5/8 3 1/2 [1 − (1 − ζ) ] ζ ∈ (0, 1) λ = pre ζ ≥ 1 and λ = min{1, λ }. There is a smooth Riemannian metric ̺¯ on B (0) and, for every l ≥ 0, m pre 5/8 a constant K such that 2Ht+2K ij ij −2Ht−2K −2λ Ht ke g (t, ·) − ̺¯ k l + ke g (t, ·) − ̺¯ k l ≤ K e , (42) ij ij l C C −2Ht−2K −2λ Ht ke ∂ g (t, ·) − 2H̺¯ k l ≤ K e , (43) t ij ij l 13 ij l for every l ≥ 0 and t ≥ 0. Here ̺¯ denotes the components of the inverse of ̺¯. Furthermore, C denotes the C -norm on B (0). Turning to g , there is a b > 0 and, for every l ≥ 0, a constant 5/8 0m K such that −bHt kg (t, ·) − v k + k∂ g (t, ·)k l ≤ K e , (44) 0m m 0 0m C l for all l ≥ 0 and t ≥ 0, where ij v = ̺¯ γ (45) m imj and γ denote the Christoffel symbols of the metric ̺¯, given by imj γ = (∂ ̺¯ + ∂ ̺¯ − ∂ ̺¯ ). imj i jm j im m ij Let k denote the components of the second fundamental form (induced on the constant-t hyper- ij surfaces) with respect to the standard coordinates on B (0). If λ < 1, there is, for every l ≥ 0, 5/8 m a constant K such that −2λ Ht kg (t, ·) + 1k l + k∂ g (t, ·)k l ≤ K e , 00 C 0 00 C l −2Ht−2K −2λ Ht ¯ m ke k (t, ·) − H̺¯ k l ≤ K e ij ij C l for every l ≥ 0 and t ≥ 0. If λ = 1, there is, for every l ≥ 0, a constant K such that m l −2Ht k[∂ g + 2H(g + 1)](t, ·)k l ≤ K e , 0 00 00 C l 2 1/2 −2Ht kg (t, ·) + 1k l ≤ K (1 + t ) e , 00 C l −2Ht−2K 2 1/2 −2Ht ke k (t, ·) − H̺¯ k l ≤ K (1 + t ) e ij ij C l λ Ht pre for every l ≥ 0 and t ≥ 0. In order to describe the asymptotics concerning φ, let ϕ = e φ. If ζ < 1, there is a smooth function ϕ , a constant b > 0 and, for every l ≥ 0, a constant K such 0 l that −bHt kϕ(t, ·) − ϕ k l + k∂ ϕk l ≤ K e (46) 0 C 0 C l for all l ≥ 0 and t ≥ 0. If ζ = 1, there are smooth functions ϕ and ϕ , a constant b > 0 and, for 0 1 every l ≥ 0, a constant K such that −bHt k∂ ϕ(t, ·) − ϕ k l + kϕ(t, ·) − ϕ t − ϕ k l ≤ K e (47) 0 1 1 0 l C C for all l ≥ 0 and t ≥ 0. Finally, if ζ > 1, there is an anti symmetric matrix A, given by 0 δH A = , −δH 0 1/2 where δ = 3(ζ − 1) /2, smooth functions ϕ and ϕ , a constant b > 0 and, for every l ≥ 0, a 0 1 constant K such that δHϕ ϕ −At 0 −bHt e (t, ·) − ≤ K e (48) ∂ ϕ ϕ 0 1 l for all l ≥ 0 and t ≥ 0. In order to describe the asymptotics for the distribution function, let −1 x = ψ . Then (x, U) are canonical local coordinates, where U = ψ[(t , ∞) × B (0)]. − 5/8 −1 Let f = f ◦ Ξ and −2Ht−K−K Vl h(t, x¯, q¯) = f (t, x¯, e q¯). (49) Introduce, moreover, the notation   1/2 Z Z 2µ +2|β| α β 2 ¯  ¯  kfk l = hp¯i |∂ ∂ f(x¯, p¯)| dp¯dx¯ H [B (0)×R ] x ¯ p¯ 5/8 Vl,μ B (0) R 5/8 |α|+|β|≤l 14 ∞ 3 for f ∈ C [B (0) × R ]. Then there is a constant b > 0 and, for every l, a constant K such 5/8 l that −bHt k∂ h(t, ·)k l ≤ K e (50) t l H [B (0)×R ] 5/8 Vl,μ ∞ 3 holds for all l ≥ 0 and t ≥ 0. There is also a function h ∈ C [B (0) × R ], a constant b > 0 5/8 and, for every l, a constant K such that khk l < ∞, H [B (0)×R ] 5/8 Vl,μ −bHt kh(t, ·) − hk l 3 ≤ K e (51) H [B (0)×R ] 5/8 Vl,μ hold for all l ≥ 0 and t ≥ 0. Furthermore, h ≥ 0. Concerning the stress energy tensor associated with the Vlasov matter, there is a b > 0 and, for every l ≥ 0, a constant K such that the estimates 3(Ht+K ) Vl 1/2 −bHt Vl e T − h|̺¯| dq¯ ≤ K e , (52) 3 l R C 3(Ht+K ) Vl 1/2 −bHt Vl ¯ e T + q h|̺¯| dq¯ ≤ K e , (53) i l 0i 2Ht+3K Vl Vl e T ≤ K (54) ij l hold for all l ≥ 0 and all t ≥ 0, where |̺¯| denotes the absolute value of the determinant of ̺¯, K−K j Vl ¯ ¯ q = v + e ̺¯ q¯ i i ij and v is defined in (45). Finally, if µ > 9/2, there is a constant b > 0 and, for every l ≥ 0, a constant K such that 3(Ht+K ) Vl 1/2 −bHt Vl ¯ ¯ ¯ e T − hq q |̺¯| dq¯ ≤ K e (55) i j l ij holds for all l ≥ 0 and t ≥ 0. Remark 30. In case one is only interested in the Einstein-Vlasov setting with a positive cosmolog- ical constant, more detailed information can be obtained; cf. [31, Proposition 32.8, pp. 609–611]. 1.6 Cauchy stability In what follows, we also need a Cauchy stability result in the Einstein-Vlasov-non-linear scalar field setting. There are such results in the literature; cf. [31]. However, for the convenience of the reader, we introduce the necessary terminology and quote the relevant result in the present subsection. To begin with, we need to introduce the notion of a background solution; cf. [31, Definition 24.2, p. 421]. In the 3-dimensional case, this definition takes the following form. Definition 31. Let 5/2 < µ ∈ R, Σ be a closed 3-dimensional manifold, and let g be a smooth time oriented Lorentz metric on M = I × Σ, where I is an open interval. Let ∂ denote differentiation with respect to the first coordinate and assume that g(∂ , ∂ ) = g < 0 and that the hypersurfaces t t 00 Σ = {t} × Σ are spacelike with respect to g for t ∈ I. Finally, assume that φ ∈ C (M) and f ∈ D (P), together with g, satisfy (23)–(25). Then (M, g, f, φ) is called a background solution. Remark 32. In the case of T -symmetric solutions, the metric is of the form (1). Moreover, the distribution functions of interest have compact support on constant time hypersurfaces. As a consequence, it is clear that the T -symmetric solutions we consider in the present paper are background solutions in the sense of the above definition. Next, we introduce the notion of induced initial data on constant t hypersurfaces; cf. [31, Defini- tion 24.3, p. 421]. In the 3-dimensional case, this definition takes the following form. 15 Definition 33. Let 5/2 < µ ∈ R, Σ be a closed 3-dimensional manifold, and let g be a smooth time oriented Lorentz metric on M = I×Σ, where I is an open interval. Let, furthermore, φ ∈ C (M), f ∈ D (P) and assume that (g, f, φ) solve (23)–(25). Let t ∈ I and assume Σ = {t} × Σ to be spacelike with respect to g. Let, furthermore, κ be the second fundamental form and N be the future directed unit normal of Σ . Finally, let ι : Σ → M be defined by ι (x¯) = (t, x¯) and t t t ∗ ∗ ∗ −1 ∗ ∗ ¯ ¯ ¯ ¯ g¯ = ι g, k = ι κ, f = ι (f ◦ pr ), φ = ι φ, φ = ι (Nφ). 0 1 t t t Σ t t ¯ ¯ ¯ ¯ Then (g¯, k, f, φ , φ ) are referred to as the initial data induced on Σ by (g, f, φ), or simply the 0 1 t initial data induced on Σ if the solution is understood from the context. Finally, we are in a position to formulate the Cauchy stability result we need here; cf. [31, Corollary 24.10, p. 432]. In the 3-dimensional case, this result takes the following form. Theorem 34. Let 5/2 < µ ∈ R and 5/2 < l ∈ Z. Let (M , g , f , φ ) be a background bg bg bg bg solution with M = I × Σ and recall the notation Σ, Σ etc. from Definition 31 (the interval bg bg t which was denoted by I in Definition 31 will here be denoted by I ). Assume that 0 ∈ I and bg bg ¯ ¯ ¯ ¯ let (g¯ , k , f , φ , φ ) be the initial data induced on Σ by (g , f , φ ). Make a choice bg bg bg bg,0 bg,1 0 bg bg bg of H (T Σ)-norms and a choice of Sobolev norms k · k l on tensor fields on Σ. Let J ⊂ I be bg Vl,µ H ¯ ¯ ¯ ¯ a compact interval and let ǫ > 0. Then there is a δ > 0 such that if (Σ, g¯, k, f, φ , φ ) are initial 0 1 data for the Einstein-Vlasov-non-linear scalar field system satisfying ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ kg¯ − g¯ k l+1 + kk − k k l + kφ − φ k l+1 + kφ − φ k l + kf − f k l ≤ δ, bg H bg H 0 bg,0 H 1 bg,1 H bg H (T Σ) Vl,μ then there is an open interval I containing 0 and a solution (g, f, φ) to (23)–(25) on M = I × Σ such that ¯ ¯ ¯ ¯ • the initial data induced on Σ by (g, f, φ) are given by (g¯, k, f, φ , φ ), 0 0 1 • ∂ is timelike with respect to g and Σ is a spacelike Cauchy hypersurface with respect to g t t for all t ∈ I, • J ⊂ I and if the initial data induced on Σ (for t ∈ I ∩ I) by (g, f, φ) and (g , f , φ ) t bg bg bg bg ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ are denoted by (g¯ , k , f , φ , φ ) and (g¯ , k , f , φ , φ ) respectively, then t t t t,0 t,1 bg,t bg,t bg,t bg,t,0 bg,t,1 ¯ ¯ ¯ ¯ kg¯ − g¯ k l+1 + kk − k k l + kφ − φ k l+1 t bg,t H t bg,t H t,0 bg,t,0 H (56) ¯ ¯ ¯ ¯ +kφ − φ k l + kf − f k l ≤ ǫ t,1 bg,t,1 H t bg,t H (T Σ) Vl,μ for all t ∈ J. 1.7 Stability of T -Gowdy symmetric solutions Combining Theorems 7, 29 and 34 yields a future stability result for the T -symmetric solutions considered in Theorem 7. Moreover, the solutions are stable in the Einstein-Vlasov-non-linear scalar field setting. Theorem 35. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant Λ. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics. Choose a t ∈ I and 3 3 ∗ let i : T → I × T be given by i(x¯) = (t, x¯). Let g¯ = i g and let k denote the pullback (under bg bg i) of the second fundamental form induced on i(T ) by g. Let, moreover, ∗ −1 f = i (f ◦ pr ). bg 3 i(T ) Make a choice of µ > 5/2, a choice of norms as in Definition 23 and a choice of Sobolev norms on tensorfields on T . Let, in addition, V : R → R be a smooth function such that V (0) = Λ, 16 ′ ′′ 3 ¯ ¯ ¯ ¯ V (0) = 0 and V (0) > 0. Then there is an ǫ > 0 such that if (T , g¯, k, f, φ , φ ) are initial data 0 1 ∞ 3 ¯ ¯ for (23)–(25), with f ∈ D (TT ), satisfying ¯ ¯ ¯ ¯ ¯ ¯ kg¯ − g¯ k 5 + kk − k k 4 + kf − f k 4 + kφ k 5 + kφ k 4 ≤ ǫ, bg H bg H bg H 0 H 1 H Vl,μ then the maximal globally hyperbolic development (M, g, f, φ) of the initial data is future causally geodesically complete. Moreover, there is a Cauchy hypersurface Σ in (M, g) such that for each point of Σ, there is a neighbourhood (¯x, U) such that Theorem 29 applies. In particular, the asymptotics stated in Theorem 29 thus hold. Remark 36. Up to the point where we appeal to Theorem 29, Cauchy stability applies. It should thus be possible to obtain detailed control over the perturbed solutions for the entire future. The interested reader is encouraged to write down the details. ¯ ¯ Remark 37. The function f has compact support, but f need not have compact support. bg The proof is to be found in Section 10. 1.8 Outline Finally, let us give an outline of the paper. In Section 2, we write down the equations in the case that the metric takes the form (1) (though the reader interested in a derivation is referred to Appendix A). In Section 3, we then collect the conclusions which are not dependent on the particular type of matter model (as long as it satisfies the dominant energy condition and the non- negative pressure condition). The section ends with conclusions concerning the causal structure of T -Gowdy symmetric spacetimes. Turning to the more detailed conclusions, we specialise to the case of Vlasov matter. The natural first step is to derive light cone estimates; i.e., to consider the behaviour along characteristics. This is the subject of Section 4. As opposed to the vacuum case, we need to control the characteristics associated with the Vlasov equation at the same time as the first derivatives of the metric components. Fortunately, the e - and e -components 2 3 of the momentum are controlled automatically due to the symmetry. However, an argument is required in the case of the e -component. In order to obtain control of higher order derivatives, we need to take derivatives of the characteristic system (associated with the Vlasov equation; i.e. with the geodesic flow). Naively, this should require control of second order derivatives of the metric functions, something we do not have. Nevertheless, by an appropriate choice of variables, controlling first order derivatives turns out to be sufficient. It is of interest to note that a similar choice was already suggested in [2, Lemma 3, p. 363]; cf. also [3, Lemma 3, p. 257]. However, in the present setting, it is not sufficient to derive a system involving only first order derivatives of the metric functions. We also need to be able to use the system to derive the desired type of asymptotics for the derivatives of the characteristic system. It turns out to be possible to do this, and we write down the required arguments in Section 6. After we have obtained this conclusion, it turns out to be possible to proceed inductively in order to derive higher order estimates for the characteristic system and the metric components. The required arguments are written down in Sections 7 and 8. In order to obtain the desired conclusions concerning the distribution function, it turns out to be convenient to consider L -based energies. This subject is treated in Section 9. Finally, in Section 10, we prove the main theorems of the paper. As an appendix to the paper, we include a derivation of Einstein’s equations as well as of the Vlasov equation; cf. Appendix A. We also provide a summary of the most important notation in Appendix B. 2 Symmetry assumptions and equations In this paper, we study T -symmetric solutions of Einstein’s equations. Since it will turn out to be convenient to express the equations using the orthonormal frame (7), let us introduce the notation ρ = T (e , e ), J = −T (e , e ), P = T (e , e ), S = T (e , e ), (57) 0 0 i 0 i i i i ij i j 17 where we do not sum over any indices; here, and below, we tacitly assume Latin indices to range from 1 to 3 and Greek indices to range from 0 to 3. It is also convenient to introduce the notation 5/2 1/2 P−λ/2 5/2 1/2 −P−λ/2 J = −t α e (G + QH ), K = QJ − t α e H . (58) t t t Note that these objects are the twist quantities introduced in (2); cf. Appendix A.3. In order to derive Einstein’s equations, it is useful to calculate the Einstein tensor for a metric of the form (1). The corresponding, somewhat lengthy, computations are to be found in Section A. Using the above notation, the calculations yield the conclusion that the 00 and 11-components of Einstein’s equations can be written λ/2−P 2 λ/2+P 2 α e J e (K − QJ) 2 2 2P 2 2 λ − 2 = t P + αP + e (Q + αQ ) + + (59) t θ t θ 5/2 5/2 α t t 1/2 λ/2 +4t e (ρ + Λ), λ/2−P 2 λ/2+P 2 e J e (K − QJ) 2 2 2P 2 2 λ = t P + αP + e (Q + αQ ) − − (60) t θ t θ 5/2 5/2 t t 1/2 λ/2 +4t e (P − Λ), respectively. The 22-component minus the 33-component can be written −1/2 λ/2−P 2 α e J −1/2 1/2 −1/2 2P 2 2 ∂ (tα P ) =∂ (tα P ) + tα e (Q − αQ ) + t t θ θ t θ 5/2 2t (61) −1/2 λ/2+P 2 α e (K − QJ) 1/2 λ/2 −1/2 − + t e α (P − P ). 2 3 5/2 2t The 22-component plus the 33-component can be written α 3 −1/2 1/2 −1/2 2 2P 2 2 2P 2 ∂ tα λ − 2 − =∂ tα λ − tα P + e Q − α(P + e Q ) t t θ θ t t θ θ α t λ/2−P 2 λ/2+P 2 e J e (K − QJ) (62) −1/2 − 2tα + 7/2 7/2 t t −1/2 1/2 λ/2 −1/2 + α λ + 2t e α (2Λ − P − P ). t 2 3 The 01, 02, 03, 12 and 13-components are equivalent to 2P 1/2 λ/2 −1/2 λ = 2t(P P + e Q Q ) − 4t e α J , (63) θ t θ t θ 1 5/4 −1/2 P/2+λ/4 J = 2t α e J , (64) θ 2 5/4 −1/2 −P/2+λ/4 5/4 −1/2 P/2+λ/4 K = 2t α e J + 2t α e QJ , (65) θ 3 2 5/4 λ/4+P/2 J = −2t e S , (66) t 12 5/4 λ/4+P/2 5/4 −P/2+λ/4 K = −2t e QS − 2t e S , (67) t 12 13 respectively. Finally, the 23-component reads −1/2 2P 1/2 2P −5/2 −1/2 λ/2+P 1/2 −1/2 λ/2+P ∂ (tα e Q )− ∂ (tα e Q ) = t α e J(K − QJ) + 2t α e S . (68) t t θ θ 23 For future reference, it is also of interest to note that −P +λ/2 2 P +λ/2 2 α e J e (K − QJ) 1/2 λ/2 1/2 λ/2 = − − − 4t e Λ − 2t e (ρ − P ), (69) 5/2 5/2 t t 2 2 2P 2 2 1/2 λ/2 λ − = t P + αP + e (Q + αQ ) + 2t e (ρ + P ). (70) t 1 t θ t θ 18 2.1 Preliminary calculations Since the metric components only depend on two variables, it is natural to derive estimates by integrating along characteristics. In the present subsection, we record a general calculation which is of interest in that context. To begin with, let us define 1/2 2 2P 2 ∂ = ∂ ± α ∂ , A = (∂ P ) + e (∂ Q) . (71) ± t θ ± ± ± One reason for introducing A is the equality (72) derived below; since the right hand side only contains first derivatives of the metric components, it is possible to integrate along the characteristics to control A . Lemma 38. Consider a T -symmetric solution to Einstein’s equations with a cosmological con- stant Λ such that the metric takes the form (1). Then 2 α 2 1/2 2P ∂ A = − − A ∓ α (P ∂ P + e Q ∂ Q) ± ∓ ∓ θ ∓ θ ∓ t α t −P +λ/2 2 P +λ/2 2 λ/2 e (K − QJ) e J(K − QJ) (72) e J + ∂ P − ∂ P + 2 e ∂ Q ∓ ∓ ∓ 7/2 7/2 7/2 t t t −1/2 λ/2 −1/2 λ/2 P + 2t e (P − P )∂ P + 4t e S e ∂ Q. 2 3 ∓ 23 ∓ Remark 39. In this calculation, the cosmological constant need not be positive. Proof. The statement follows from a lengthy computation. Let us, however, for the benefit of the reader, write down some of the intermediate steps. Using (61), we obtain 1 α 2P 2 2 ∂ ∂ P = − P + ∂ P + e (Q − αQ ) ± ∓ t ∓ t θ t 2α (73) −P +λ/2 2 P +λ/2 2 e J e (K − QJ) −1/2 λ/2 + − + t e (P − P ). 2 3 7/2 7/2 2t 2t Similarly, due to (68), we obtain λ/2−P 1 α e J(K − QJ) −1/2 λ/2−P ∂ ∂ Q = − Q + ∂ Q − 2(Q P − αQ P ) + + 2t e S . (74) ± ∓ t ∓ t t θ θ 23 7/2 t 2α Combining (73) and (74) with the fact that 2 2 2 −4(Q P − αP Q )∂ Q + 2∂ P (∂ Q) = −2∂ P (Q − αQ ), t t θ θ ∓ ± ∓ ∓ t θ a calculation yields the conclusion of the lemma. 2.2 Vlasov matter The equations (59)–(70) hold in general. However, we are here particularly interested in matter of Vlasov type. In order to derive the relevant form of the Vlasov equation, recall the conventions concerning f introduced in Subsection 1.2. Recall, moreover, the fact that the Vlasov equation is equivalent to f being constant along future directed unit timelike geodesics. As a consequence, it can be calculated (cf. Appendix A.7) that the Vlasov equation takes the form 1/2 1 2 3 ∂f α v ∂f 1 1 2α 1 v v 1/2 0 1 1/2 P + − α λ v + λ − − v − α e Q θ t θ 0 0 ∂t v ∂θ 4 4 α t v 3 2 2 2 1 (v ) − (v ) ∂f 1/2 −7/4 λ/4 −P/2 2 P/2 3 + α P − t e e Jv + e (K − QJ)v 0 1 2 v ∂v 1 2 1 1 1 v v ∂f 2 1/2 − P + v + α P t θ 0 2 2 t 2 v ∂v 1 3 1 1 1 1 v v v ∂f 3 1/2 P 2 1/2 − P v − α P + e v Q + α Q = 0. t θ t θ 0 0 3 2 t 2 v v ∂v (75) 19 Turning to the stress energy tensor, it satisfies T (e , e ) = v v f dv, (76) µ ν µ ν 3 −v where v = η v and η = diag{−1, 1, 1, 1}. In particular, in the case of Vlasov, we thus have α αβ Z Z Z Z k 2 j k (v ) v v 0 k ρ = v f dv, P = f dv, J = v f dv, S = f dv, (77) k k jk 0 0 3 3 v 3 3 v R R R R where j, k = 1, 2, 3. 3 Preliminary conclusions concerning the asymptotics In the present section, we are interested in T -symmetric solutions to Einstein’s equations such that the corresponding metric admits a foliation of the form (1) on I × T , where I = (t , ∞) and t ≥ 0. For the sake of brevity, we shall below refer to solutions of this form as future global, and we shall speak of t and t = t + 2 without further introduction. 0 1 0 It is useful to begin by recalling the following consequences of the non-negative pressure condition and the dominant energy condition. Lemma 40. Consider a solution to Einstein’s equations with a cosmological constant Λ and a metric of the form (1). Let ρ, P , J and S , i, j = 1, 2, 3, be defined by (57). If the stress energy i i ij tensor satisfies the non-negative pressure condition, then, for i = 1, 2, 3, 0 ≤ P . (78) If the stress energy tensor satisfies the dominant energy condition, then, for i, j = 1, 2, 3, 0 ≤ ρ, (79) |P | ≤ ρ, (80) |J | ≤ ρ, (81) |S | ≤ ρ. (82) ij Proof. By definition, P = T (e , e ). Since e is a spacelike vector field, the non-negative pressure i i i i condition implies that (78) holds. The dominant energy condition states that T (u, v) ≥ 0 for future directed timelike vectors u and v. By continuity, this inequality also holds for future directed causal vectors. Since e is future directed timelike, ρ = T (e , e ) ≥ 0, so that (79) follows. Note that 0 0 0 e ± e is a future directed causal vector field. In particular, 0 i 0 ≤ T (e − e , e + e ) = ρ + J − J − S . 0 i 0 j i j ij Since S is symmetric, adding this inequality with the one obtained by interchanging i and j ij yields the conclusion that S ≤ ρ. Similarly, ij 0 ≤ T (e ± e , e ± e ) = ρ ∓ J ∓ J + S . 0 i 0 j i j ij Adding the two inequalities yields −S ≤ ρ. Thus (82) holds. The proof of (80) is similar. Finally, ij 0 ≤ T (e , e ± e ) = ρ ∓ J , 0 0 i i so that (81) holds. Before deriving estimates describing the asymptotics of solutions, let us make the following remark. 20 Remark 41. In what follows, the constants appearing in the estimates we state are allowed to depend on the solution, unless otherwise indicated. Proposition 42. Given a future global solution to Einstein’s equations with a cosmological con- stant Λ > 0, T -symmetry and a stress energy tensor satisfying the dominant energy condition, there is a constant C > 0 such that −3 α(t, θ) ≤ Ct (83) for all (t, θ) ∈ [t , ∞) × S . Remark 43. The same conclusion holds if we replace the cosmological constant with a non- linear scalar field with a potential with a positive lower bound; in other words, if we set Λ = 0 o sf o and consider stress energy tensors of the form T = T + T , where T is the stress energy tensor sf associated with matter fields satisfying the dominant energy condition, and T is the stress energy tensor associated with a non-linear scalar field with a potential V having a positive lower bound. Proof. Due to (70) and the fact that the matter satisfies the dominant energy condition (so that (80) holds), we conclude that λ − α /α ≥ 0. There is thus a c > 0 such that t t 0 −1/2 λ/2 (α e )(t, θ) ≥ c for all (t, θ) ∈ [t , ∞) × S . Combining this observation with (69) and (80), we obtain −1/2 −1/2 1/2 −1/2 λ/2 1/2 ∂ α = − α ≥ 2t α e Λ ≥ c t t 1 2α for some constant c > 0 and all (t, θ) ∈ [t , ∞) × S . Integrating this inequality, we obtain the 1 1 conclusion of the proposition. In the Gowdy case, the second and third terms on the right hand side of (60) are zero, and as a consequence, we can extract more information. In fact, we have the following observation. Proposition 44. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -Gowdy symmetry and matter satisfying the non-negative pressure condition. Then there is, for every ǫ > 0, a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞) × S . Proof. Let λ = λ + 3 lnt − 2 ln . (84) 4Λ Then (60) with J = K = 0 yields 2 2 2P 2 2 1/2 λ/2 λ/2 ∂ λ = t P + αP + e (Q + αQ ) + 4t e P + (1 − e ). t t θ 1 t θ Since P ≥ 0 due to the non-negative pressure condition, cf. (78), we conclude that λ/2 ∂ λ ≥ (1 − e ). For every ǫ > 0, there is thus a T such that λ(t, θ) ≥ −ǫ for all (t, θ) ∈ [T, ∞)×S . The proposition follows. 21 In order to proceed, it is convenient to introduce an energy: −1/2 1/2 λ/2 E = tα λ − 2 − 4t e Λ dθ bas t 1 α −1/2 λ/2−P 2 α e J 2 −1/2 2 2 2P 2 2 = t α P + αP + e (Q + αQ ) + (85) t θ t θ 3/2 1 t −1/2 λ/2+P 2 α e (K − QJ) 3/2 −1/2 λ/2 + + 4t α e ρ dθ. 3/2 Let us motivate this particular choice. The energy E is quite similar to the energy defined bas in [3, (42), p. 251]. However, there is one fundamental difference. The integrand in the energy −1/2 2 defined in [3] contains a term of the form α U , where U = (P + ln t)/2. Using U instead of P as a variable is convenient in global existence arguments, since some of the formulae become less involved. However, the variable U is poorly adapted to the actual asymptotics of solutions. The −2 reason for this is that, in the end, it turns out that P converges to zero as t . The dominant term in U is thus ln t/2. Using U instead of P in the energy, the best estimate of E one could bas 3/2 1/2 hope for would be E (t) ≤ Ct . Below, we prove that E ≤ Ct ; cf. Lemma 46. In addition bas bas to this, it is possible to derive a good estimate for the time derivative of E ; cf. the proof of bas Lemma 45 below. On a more general level, it is natural to ask why it is necessary to use L -based energy estimates at all. Since the problem is 1+1-dimensional, should it not be sufficient to consider the behaviour along characteristics? The problem in our setting is that we wish to derive detailed quantitative information for arbitrary initial data. In particular, we are not in a situation where we can use bootstrap arguments. For this reason, we need to proceed step by step. First, it is necessary to derive not only rough, but quite detailed, control of some of the metric components, in particular λ. This leads, for example, to estimates of the form (96) and (97). Only once we have estimates of this form is it meaningful to turn to the characteristic system; cf., e.g., the last two terms on the right hand side of (109) and the proof of Lemma 52. In case the metric has λ-asymptotics (recall Definition 1), it turns out to be possible to estimate E . bas Lemma 45. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then for every a > 1/2, there is a constant C > 0 such that E (t) ≤ C t (86) bas a for all t ≥ t . Proof. Due to (62), we obtain h  i −1/2 1/2 λ/2 1/2 1/2 2 2P 2 ∂ tα λ − 2 − 4t e Λ =∂ (tα λ ) + 2tα (P + e Q ) t t θ θ θ θ −1/2 λ/2−P 2 −1/2 λ/2+P 2 3 α e J 3 α e (K − QJ) − − 5/2 5/2 2 2 t t (87) 5/2 −1/2 λ/2 2 2P 2 2 2P 2 − 2t α e Λ[P + e Q + α(P + e Q )] t t θ θ 1/2 −1/2 λ/2 + t α e (3ρ + P − 2P − 2P ) 1 2 3 2 −1/2 λ − 4t α e Λ(ρ + P ). Since the matter satisfies the non-negative pressure condition, we know that P ≥ 0, cf. (78), so that Z Z dE bas 1/2 2 2P 2 5/2 1/2 λ/2 2 2P 2 ≤ 2tα (P + e Q )dθ − 2t α e Λ(P + e Q )dθ θ θ θ θ dt 1 1 S S Z Z 1/2 −1/2 λ/2 2 −1/2 λ + t α e (3ρ + P )dθ − 4t α e Λ(ρ + P )dθ. 1 1 1 1 S S 22 Using the consequences of Lemma 40 and the fact that the solution has λ-asymptotics, we conclude that for every a > 1/2, there is a T ≥ t such that dE a bas ≤ E bas dt t for all t ≥ T . As a consequence, E (t) ≤ Ct for t ≥ t . bas 1 Using the estimate for E derived in Lemma 45, it is possible to extract more information bas concerning the asymptotics. Lemma 46. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that −1/2 λ(t, ·) + 3 ln t − 2 ln ≤ Ct , (88) 4Λ 1/2 E (t) ≤ Ct (89) bas for all t ≥ t . a 1/2 −3/2 Proof. Due to the estimate E (t) ≤ C t , the fact that α ≤ Ct , (60) and (80), we conclude bas a that 1/2 λ/2 a−5/2 hλ i = −4t he iΛ + O(t ); (90) recall that the notation hλ i was introduced in Remark 5. Due to (63) and (81), we also have −1/2 2 2 2P 2 2 1/2 −1/2 λ/2 |λ | ≤ tα [P + αP + e (Q + αQ )] + 4t α e ρ. θ t θ t θ Due to (86), we thus obtain a−1 |λ |dθ ≤ Ct . (91) Recall that λ is defined in (84) and note that, due to (90), λ/2 a−5/2 hλ i = (1 − he i) + O(t ). (92) Let us first prove that hλi converges to zero. Let, to this end, ǫ > 0. Since the solution has λ-asymptotics, we know that there is a T such that hλi(t) ≥ −ǫ for all t ≥ T . In order to prove ˆ ˆ that there is a T such that hλi(t) ≤ ǫ for all t ≥ T , let us assume that hλi(t) ≥ ǫ for some t. Due to (91), we conclude that λ(t, θ) ≥ ǫ/2 for all θ ∈ S (assuming t to be large enough). Inserting this information into (92), we conclude that ǫ/4 hλ i ≤ (1 − e ), assuming t to be large enough. Since the right hand side is negative and non-integrable, we conclude that hλi has to decay until it is smaller than ǫ (assuming the starting time t to be large enough). Moreover, hλi cannot exceed ǫ at a later time. In order to obtain a quantitative estimate, note that hλi/2 a−2 hλ i = (1 − e ) + O(t ), where we have used the fact that λ is bounded to the future as well as (91). As a consequence, 6 1 2 2 a−2 ˆ ˆ ˆ ˆ ˆ ˆ ˆ ∂ hλi =2hλihλ i = hλi 1 − 1 + hλi + O(hλi ) + O(t hλi) t t t 2 3 1 2 3 a−2 ˆ ˆ ˆ = − hλi + O(hλi ) + O(t hλi). t t 23 Let 0 < b < 1 − a and define 2b 2 E = t hλi . Then dE 2b 3 1 −1 b+a−1 1/2 = E − E + O(hλiE) + t O(t E ). dt t t t As a consequence, there is a constant C > 0 such that ∂ E ≤ 0 when E ≥ C and t is large enough. In particular, E is thus bounded to the future. For every 0 < b < 1/2, there is thus a constant C such that −b λ(t, ·) + 3 lnt − 2 ln ≤ C t 4Λ for all t ≥ t . Due to this estimate, we can return to the argument presented in the proof of 1/2 Lemma 45 and obtain the improvement E (t) ≤ Ct for t ≥ t . As a consequence, we can go bas 1 through the above arguments with a = 1/2 and b = 1/2. The lemma follows. Lemma 47. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that −3/2 −1/2 t hα (t, ·)i + kQ(t, ·)k 0 + kP (t, ·)k 0 ≤ C (93) C C for all t ≥ t . Proof. Estimate, using (69), (84), (88), (89) and Lemma 40, D E −1/2 −P +λ/2 2 −1/2 P +λ/2 2 1 α 1 α e J α e (K − QJ) −1/2 −1/2 ∂ hα i = − α = + dθ 5/2 5/2 2 α 2π 1 2t 2t 1/2 −1/2 λ/2 1/2 −1/2 λ/2 + [2t α e Λ + t α e (ρ − P )]dθ 2π 1 −1/2 1/2 −1/2 λ/2 ≤Ct + 2t α e Λdθ 2π 1 3 3 −1/2 λ/2 −1/2 −1/2 −3/2 −1/2 −1/2 ≤Ct + he α i ≤ hα i + Ct hα i + Ct . 2t 2t −1/2 Let A = hα i + t. Then dA 3 −1/2 −3/2 = ∂ hα i + 1 ≤ A + Ct A. dt 2t Consequently, A(t) 3 ln ≤ ln t + C , A(t ) 2 −1/2 3/2 so that hα i ≤ Ct for t ≥ t . Combining this estimate with (89) yields Z Z Z 1/2 1/2 1/2 2 −1/2 −3/4 3/4 |P |dθ ≤ α P dθ α dθ ≤ Ct t ≤ C (94) θ θ 1 1 1 S S S for t ≥ t . On the other hand, using (83) and (89) yields Z Z 1/2 1 1 2 −3/2 |∂ hPi| = |hP i| ≤ |P |dθ ≤ P dθ ≤ Ct . (95) t t t 2π 1 2π 1 S S Consequently, hPi is bounded to the future. Combining these two observations, we conclude that kP (t, ·)k 0 ≤ C for all t ≥ t . Combining this estimate for P with the bound (89), it is possible to derive L estimates for Q and Q analogous to (94) and (95). Consequently, Q is bounded to the future. θ t The lemma follows. 24 Lemma 48. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that λ/2−P 2 e J −2 ≤ Ct , (96) 5/2 P +λ/2 2 e (K − QJ) −2 ≤ Ct (97) 5/2 for all t ≥ t . Moreover, for t ≥ t , 1 1 −3/2 kH k 1 + kG k 1 ≤ Ct . t L t L Proof. Combining (64), (81), (88), (89) and (93), we conclude that Z Z Z 5/4 −1/2 λ/4 5/4 −3/2 3/4 3/2 −1/2 λ/2 |J |dθ ≤ Ct α e ρdθ ≤ Ct t t t α e ρdθ ≤ Ct. (98) 1 1 1 S S S The spatial variation of J is consequently not greater than Ct. Combining (66), (82), (83), (88), (89) and (93) yields Z Z Z 5/4 λ/4 5/4 −3/2 3/4 −3/2 3/2 −1/2 λ/2 −1/2 |J |dθ ≤ Ct e ρdθ ≤ Ct t t t t α e ρdθ ≤ Ct . (99) 1 1 1 S S S As a consequence, 1/2 |hJi| ≤ Ct . (100) Combining (98) and (100) yields kJ(t, ·)k 0 ≤ Ct. (101) Due to (88), (93) and (101), we conclude that (96) holds. Let us now turn to K − QJ. To begin with, note that the L -norm of Q is bounded to the future. The argument to prove this statement is similar to (94), keeping in mind that (93) holds. Moreover, Z Z Z Z |K − Q J − QJ |dθ ≤ |K − QJ |dθ + |Q J|dθ ≤ |K − QJ |dθ + Ct, θ θ θ θ θ θ θ θ 1 1 1 1 S S S S where we have used (101) and the fact that the L -norm of Q is bounded. On the other hand, (64) and (65) yield 5/4 −1/2 −P/2+λ/4 K − QJ = 2t α e J . θ θ 3 Keeping (81) in mind, we can thus argue as in the proof of (98) in order to conclude that |K − Q J − QJ |dθ ≤ Ct. (102) θ θ θ In particular, the spatial variation of K − QJ is bounded by Ct. On the other hand, the L -norm −3/2 of Q can be be bounded by Ct ; cf. (95) and (93). Combining this observation with (101) yields Z Z Z Z −1/2 |K − Q J − QJ |dθ ≤ |K − QJ |dθ + |Q J|dθ ≤ |K − QJ |dθ + Ct . (103) t t t t t t t t 1 1 1 1 S S S S Moreover, due to (66) and (67), 5/4 −P/2+λ/4 K − QJ = −2t e S . t t 13 25 Keeping (82) in mind, we can proceed as in (99) in order to obtain −1/2 |K − QJ |dθ ≤ Ct . t t 1/2 Due to (103) and this estimate, the mean value of K − QJ cannot grow faster than Ct . Com- bining this observation with (102) yields kK − QJk 0 ≤ Ct. (104) Keeping (88) and (93) in mind, we obtain (97). Using the fact that (58) holds, we conclude that Z Z −5/2 −1/2 P +λ/2 |H |dθ ≤ t α e |K − QJ|dθ. 1 1 S S Combining this inequality with (88), (93) and (104) yields the desired L -estimate for H . A similar argument for G yields the remaining conclusion of the lemma. 3.1 Causal structure of T -Gowdy symmetric solutions It is of interest to note that in the T -Gowdy symmetric case, it is sufficient to assume future global existence and energy conditions in order to conclude that there is asymptotic silence. In fact, we have the following result. Proposition 49. Consider a future global T -Gowdy symmetric solution to Einstein’s equations with a cosmological constant Λ > 0 and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C, depending only on the solution, such that if γ(s) = [s, θ(s), x(s), y(s)] = [s, γ¯(s)] is a causal curve, then 2 −3 |γ¯(s)| ≤ Cs (105) for s ≥ t . In particular, there is a point x¯ ∈ T such that 1 0 −1/2 d[γ¯(s), x¯ ] ≤ Cs (106) for all s ≥ t , where d is the standard metric on T . Proof. The causality of the curve is equivalent to the estimate −1 2 3/2 P−λ/2 2 3/2 −P−λ/2 2 ˙ ˙ ˙ α θ + s e [x˙ + Qy˙ + (G + QH)θ] + s e (y˙ + Hθ) ≤ 1. (107) Note that in the case of Gowdy symmetry, G and H are time-independent. In particular, they are thus bounded. Due to (93) we also know that Q is bounded for t ≥ t . On the other hand, combining (107) with (83), (88) and (93) yields −3/2 |θ| ≤ Cs , −3/2 |y˙ + Hθ| ≤ Cs , −3/2 |x˙ + Qy˙ + (G + QH)θ| ≤ Cs for s ≥ t . Thus (105) holds, an estimate which implies (106). 26 4 Light cone estimates In the presence of matter of Vlasov type, it is necessary to consider the characteristic system in parallel with the light cone estimates for the metric components. Let us therefore begin by writing down the characteristic system. It is given by dΘ V 1/2 = α , (108) ds V 1 2 3 dV 1 1 α 1 V V 1/2 0 1 1/2 P = − α λ V − λ − 2 − V + α e Q (109) θ t θ ds 4 4 α s V 3 2 2 2 −P/2+λ/4 P/2+λ/4 1 (V ) − (V ) e J e (K − QJ) 1/2 2 3 − α P + V + V , 0 7/4 7/4 2 V s s 2 1 2 dV 1 1 1 V V 2 1/2 = − P + V − α P , (110) t θ ds 2 s 2 V 3 1 3 1 2 dV 1 1 1 V V V V 3 1/2 P 2 1/2 P = − − P V + α P − e Q V − α e Q . (111) t θ t θ 0 0 ds 2 s 2 V V 1/2 Note that in this system of equations, functions such as α should be evaluated at [s, Θ(s)]. In view of the Vlasov equation (75), it is clear that the distribution function is constant along characteristics. It is important to note that only in the case of V is it necessary to carry out an 2 3 analysis; concerning V and V we automatically obtain the following estimate. Lemma 50. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution 0 0 has λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the 1 0 solution, such that if Θ, V is a solution to (108)–(111) with initial data Θ(t ), V (t ) such that 1 1 [t , Θ(t ), V (t )] is in the support of f, then 1 1 1 2 3 −1/2 |V (s)| + |V (s)| ≤ Cs for all s ≥ t . Remark 51. As mentioned in Remark 6, we tacitly assume f(t , ·) to have compact support. Proof. Due to (110) and (111), it can be verified that 1/2 P/2 2 1/2 P/2 2 1/2 −P/2 3 s e V , s Qe V + s e V (112) are conserved along characteristics. Since we know P and Q to be uniformly bounded, cf. Lemma 47, we obtain the conclusion of the lemma. Let us now turn to V . To begin with, we have the following estimate. Lemma 52. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution 0 0 has λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the 1 0 solution, such that if Θ, V is a solution to (108)–(111) with initial data Θ(t ), V (t ) such that 1 1 [t , Θ(t ), V (t )] is in the support of f, then 1 1 1 1 2 1 d(V ) 1 |V | 1 2 −1/2 λ/2 1 2 ≤ − (V ) + Cs e (Q ) ds s V (113) 1 1 |V | |V | −1 1/2 −3/2 1 2 −2 1 + CsF + Cs F + Cs (V ) + Cs |V | 0 0 V V for all s ≥ t , where 1 1 1 2 3 Q (t) := sup{|v | : (t, θ, v , v , v ) ∈ suppf} (114) 27 and F (t) = sup A (t, θ) + sup A (t, θ), (115) + − 1 1 θ∈S θ∈S where A is defined in (71). Proof. Due to (109), we have 1 2 1 2 3 d(V ) 1 1 α 1 V V V 1 2 1/2 P 1/2 0 1 = − α λ V V − λ − 2 − (V ) + 2α e Q θ t θ ds 2 2 α s V 3 2 2 2 −P/2+λ/4 P/2+λ/4 (V ) − (V ) e J e (K − QJ) 1/2 1 1 2 1 3 − α P V + 2 V V + 2 V V . 0 7/4 7/4 s s −2 1 However, due to Lemmas 48 and 50, we can estimate the last two terms by Cs |V |. We thus have 1 2 1 d(V ) 1 1 α 1 |V | 1/2 0 1 1 2 −1 1/2 −2 1 ≤ − α λ V V − λ − 2 − (V ) + Cs F + Cs |V |, θ t ds 2 2 α s V where we have used Lemma 50. Due to (59) and (63), the sum of the first and the second term on the right hand side can be written −P +λ/2 2 P +λ/2 2 e J e (K − QJ) 1/2 λ/2 0 1 1 1/2 λ/2 1 2 1 2 1 2 2s e (J V − ρV )V − 2s e Λ(V ) − (V ) − (V ) 5/2 5/2 2s 2s 1 1 1/2 2P 0 1 2 2 2P 2 2 1 2 1 2 − sα (P P + e Q Q )V V − s P + αP + e (Q + αQ ) (V ) + (V ) . t θ t θ t θ t θ 2 2s (116) Note that 1 1 1/2 0 1 2 2 1 2 1/2 0 1 2 2 1 2 −sα P P V V − s[P + αP ](V ) ≤sα |P P |V |V | − s[P + αP ](V ) t θ t θ t θ t θ 2 2 1/2 1 0 1 ≤sα |P P ||V |(V − |V |) t θ 1/2 2 1 2 − s(|P | − α |P |) (V ) . t θ Combining this estimate with a similar estimate for Q, we conclude that the second and third last terms in (116) can be estimated by 2 2 3 2 1 1 + (V ) + (V ) |V | 1/2 2P 1 sα (|P P | + e |Q Q |)|V | ≤ CsF . t θ t θ 0 1 0 V + |V | V Due to Lemma 46, we have 1 1 1/2 λ/2 −3/2 −2s e Λ + = − + O(s ). 2s s Combining the above observations with Lemma 48, we conclude that 1 2 d(V ) 1 2 1/2 λ/2 0 1 1 ≤ − (V ) + 2s e (J V − ρV )V ds s (117) 1 1 |V | |V | −1 1/2 −3/2 1 2 −2 1 + CsF + Cs F + Cs (V ) + Cs |V |. 0 0 V V Let us estimate the term 1/2 λ/2 0 1 1 1/2 λ/2 − + 1 T := 2s e (J V − ρV )V = 2s e (I + I )V , (118) 1 1 28 where Z Z − 1 0 0 1 1 2 3 I (s) = [v V (s) − v V (s)]f(s, Θ(s), v)dv dv dv , R −∞ Z Z + 1 0 0 1 1 2 3 I (s) = [v V (s) − v V (s)]f(s, Θ(s), v)dv dv dv . R 0 1 1 1 − There are two cases to be distinguished, V (s) ≥ 0 and V (s) < 0. When V (s) ≥ 0, I is non-positive and can be dropped. Furthermore, for v ≥ 0, 1 2 0 2 0 2 1 2 (v ) (V ) − (v ) (V ) 1 0 0 1 v V − v V = 1 0 0 1 v V + v V 1 2 2 2 3 2 1 2 2 2 3 2 (v ) (1 + (V ) + (V ) ) (V ) (1 + (v ) + (v ) ) = − 1 0 0 1 1 0 0 1 v V + v V v V + v V 1 2 2 3 2 v (1 + (V ) + (V ) ) ≤ . 2 3 2 −1/2 3 −1/2 Letting E = {(v , v ) : |v | ≤ Cs , |v | ≤ Cs }, where C is the constant appearing in vel Lemma 50, we obtain Z Z 1 2 2 3 2 v (1 + (V ) + (V ) ) 1/2 λ/2 1 1 2 3 T ≤2kf(t , ·)k s e V dv dv dv 1 1 ∞ E 0 vel (119) Z 1 Q 1 1 V V −1/2 λ/2 1 1 −1/2 λ/2 1 2 ≤Cs e v dv ≤ Cs e (Q ) , 0 0 V V 1 1 where Q (t) is defined as in the statement of the lemma. When V < 0, an analogous argument can be given and it follows that in both cases |V | −1/2 λ/2 1 2 T ≤ Cs e (Q ) . (120) Combining (117), (118) and (120) yields the conclusion of the lemma. Lemma 53. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 3 2 2 2P 2 2 1 2 t kP + αP + e (Q + αQ )k + t[Q (t)] ≤ C t θ t θ for all t ≥ t . Proof. Let us use (72) to derive an estimate for F . Note, to begin with, that 2 α 2 5 1/2 λ/2 −3/2 − − ≤ − + 4t e Λ ≤ − + O(t ), t α t t where we have used (69), (80) and (88). Note also that the second term on the right hand side of (72) can be written 1 2 1 1 2 2P 2 (A − A ) + α(P + e Q ) ≤ A + (A + A ). ∓ ± ∓ + − θ θ 2t t t 2t Combining these estimates with Lemma 48, we conclude that 4 1 1/2 −3/2 −3 ∂ A ≤ − A + Ct A + (A + A ) + Ct A ± ∓ ∓ ∓ + − t 2t −1/2 λ/2 −1/2 λ/2 P + 2t e (P − P )∂ P + 4t e S e ∂ Q. 2 3 ∓ 23 ∓ 29 Moreover, due to (77) and Lemma 50, −2 1 −2 1 |P | ≤ Ct ln(1 + Q ), |S | ≤ Ct ln(1 + Q ) k 23 for k = 2, 3. As a consequence, 4 1 1/2 1/2 −3/2 −3 −4 1 ∂ A ≤ − A + Ct A + (A + A ) + Ct A + Ct A ln(1 + Q ). ± ∓ ∓ ∓ + − ∓ ∓ t 2t Defining A = t A + t, (121) ± ± we obtain 1/2 −3/2 −2 1 ˆ ˆ ˆ ˆ ∂ A ≤ (A + A ) + Ct A + Ct A ln(1 + Q ). ± ∓ + − ∓ 2t Introducing ˆ ˆ ˆ F (t) = sup A (t, θ) + sup A (t, θ), (122) + − 1 1 θ∈S θ∈S we obtain −3/2 −2 1/2 1 ˆ ˆ ˆ ˆ ˆ F (t) ≤ F (t ) + F (s) + Cs F (s) + Cs F (s) ln(1 + Q ) ds. (123) Introducing 1 1 2 1/2 1 1 2 1/2 R (s) = [s(V (s)) + 1] , Q (s) = [s(Q (s)) + 1] , (124) (113) implies that 1 2 d(R ) −2 1 2 −2 −5/2 −3/2 1 2 ˆ ˆ ˆ ˆ ≤ Cs (Q ) + Cs F + Cs F + Cs (Q ) . ds Integrating this inequality from t to t and taking the supremum over initial data belonging to the support of f, we obtain 1 2 1 2 −2 −3/2 1 2 ˆ ˆ ˆ ˆ [Q (t)] ≤ [Q (t )] + Cs F + Cs (Q ) ds. (125) Adding (123) and (125) and introducing 1 2 ˆ ˆ G = F + (Q ) , (126) we obtain −3/2 G(t) ≤ G(t ) + G(s) + Cs G(s) ds. In particular, G(t) ≤ Ct, so that F (t) ≤ Ct and Q is bounded. Returning to (113) with this information in mind, we conclude that 1 2 d(R ) −3/2 1 2 ≤ Cs (Q ) . ds By arguments similar to ones given above, we conclude that Q is bounded. The lemma follows. 30 5 Intermediate estimates Before proceeding, it is useful to collect the estimates that follow from the above arguments. Lemma 54. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that −1 λ(t, ·) + 3 lnt − 2 ln ≤ Ct , (127) 4Λ α 3 3 −2 + + λ + ≤ Ct , (128) α t t 0 0 C C 1/2 −2 α λ ≤ Ct (129) for all t ≥ t . Moreover, −2 kJ k 0 + kK k 0 ≤ Ct , (130) t C t C P +λ/2 2 −P +λ/2 2 e (K − QJ) e J −4 + ≤ Ct , (131) 5/2 5/2 t t 0 0 C C kJ k 0 + kK k 0 ≤ C, (132) θ θ C C P +λ/2 2 −P +λ/2 2 e (K − QJ) e J −4 ∂ + ∂ ≤ Ct (133) θ θ 5/2 5/2 t t 0 0 C C for all t ≥ t . Finally, 1/2 −3/2 kρk 0 + t kJ k 0 + tkP k 0 + tkS k 0 ≤ Ct (134) C i C i C im C for all t ≥ t . Remark 55. Note that as a consequence of (128), −1/2 λ/2 −2 ∂ (α e ) ≤ Ct for all t ≥ t . Proof. Due to (77) and Lemmas 50 and 53, the estimate (134) holds. Combining Lemmas 48 and 53 with (60) and (134), we conclude that 3 3 λ/2 −2 ∂ λ = − e + O(t ), t t where we have used the notation (84). Combining this observation with Lemma 46, we conclude that there is a constant C such that 2 2 −2 ˆ ˆ ˆ ∂ λ ≤ − λ + Ct |λ|. 2 2 Introducing L = t λ , we obtain 1 C 1/2 ∂ L ≤ − L + L . t t In particular, it is clear that L decreases once it exceeds a certain value. As a consequence, L is bounded, and we obtain (127). Combining (60), (69), (127) and (134) and Lemmas 48 and 53, we then obtain (128). As a consequence of this estimate, t α converges to a strictly positive function. In particular, there are constants C > 0, i = 1, 2, such that C ≤ t α(t, θ) ≤ C (135) 1 2 31 for all t ≥ t . Due to (63), (88), (134) and Lemma 53, we obtain (129). Returning to (66) and (67), keeping (88), (93) and (134) in mind, we conclude that (130) holds. As a consequence, it is clear that J and K are bounded. Combining this observation with (88) and (93) yields (131). Due to (64), (65), (88), (93), (134) and (135), we also obtain (132). Due to (93), (135) and Lemma 53, we know that P and Q are bounded for t ≥ t . Moreover, λ is bounded for t ≥ t due to (129) θ θ 1 θ 1 and (135). Combining these observations with (127), (132) and the fact that J, K, Q and P are bounded, we obtain (133). 6 Derivatives of the characteristic system Solutions to the Vlasov equation can be expressed in terms of the initial datum for the distribution function and appropriate solutions to the characteristic system (108)–(111). In order to see this, let us begin by introducing the notation Θ, V for the solution to (108)–(111) corresponding to the initial data Θ(t; t, θ, v) = θ, V (t; t, θ, v) = v. (136) Here we write Θ(s; t, θ, v) and V (s; t, θ, v) in order to clarify the dependence on the initial data. In particular, Θ(s; t, θ, v) and V (s; t, θ, v), considered as functions of the parameter s, thus constitute a solution to (108)–(111). The purpose of the variables (t, θ, v) appearing after the semi-colon is simply to indicate that the relations (136) hold. We shall use the notation dΘ/ds and dV/ds to indicate differentiation with respect to the first variable. Moreover, we shall take ∂ Θ, ∂ Θ, ∂ iΘ t θ v etc. to denote differentiation with respect to the variables appearing after the semi-colon. Given a fixed τ ∈ (t , ∞), where (t , ∞) is the existence interval of the solution to the Einstein- 0 0 Vlasov system under consideration, we know that f(t, θ, v) = f[τ, Θ(τ; t, θ, v), V (τ; t, θ, v)]. (137) Since f(τ, ·) is a smooth function with compact support, it is sufficient to estimate the derivatives of solutions to the characteristic system in order to estimate the derivatives of f. Unfortunately, differentiating the characteristic system leads to second order derivatives of P , Q etc., quantities over which we have no control. However, using the ideas introduced in [2], this problem can be circumvented. In fact, let ∂ be a shorthand for ∂ , ∂ or ∂ i and let t θ −1/2 λ/2 Ψ = α e ∂Θ, (138) 2 2 3 2 α 1 (V ) − (V ) 1 1 −1/2 1/2 λ/2 0 −1/2 0 Z = ∂V + α λ − 2 − 4s e Λ V − α P V t t 0 2 1 2 4 α 2 (V ) − (V ) 2 2 3 2 0 2 3 1 (V ) − (V ) V V V 1 −1/2 P + P V − α e Q (139) θ t 0 2 1 2 0 2 1 2 2 (V ) − (V ) (V ) − (V ) 1 2 3 V V V +e Q ∂Θ, 0 2 1 2 (V ) − (V ) 2 2 2 Z = ∂V + P V ∂Θ, (140) 3 3 3 P 2 Z = ∂V − P V − e Q V ∂Θ. (141) θ θ 1 2 3 It is then possible to derive an ODE for (Ψ, Z , Z , Z ) such that the coefficients are controlled due to previous arguments. The definitions (138)–(141) differ slightly from those of [2]. The reason for this is that in the present context, it is not sufficient to know that no second order derivatives of P , Q etc. occur; we need to analyse, in detail, all the terms that appear, and to use the resulting system in order to derive specific estimates for ∂Θ and ∂V . The relevant result is the following. 32 2 Lemma 56. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that dZ 1 1 j = − Z + c Ψ + c Z , (142) 1,θ 1,j ds 2s dZ 1 2 2 = − Z + c Z , (143) 2,2 ds 2s dZ 1 3 2 3 = − Z + c Z + c Z , (144) 3,2 3,3 ds 2s dΨ = c Ψ + c Z , (145) θ,θ θ,i ds where Einstein’s summation convention applies to i and j, 1/2 −3/2 |c (s; t, θ, v)| + |c (s; t, θ, v)| + |c (s; t, θ, v)| + s |c (s; t, θ, v)| ≤ Cs , i,j θ,i i,θ θ,θ 1 3 and the estimate holds for all (t, θ, v) ∈ [t , ∞)× S × R in the support of f and for all s ∈ [t , t]. 1 1 Proof. Let us begin by noting that P/2 1/2 2 1/2 P/2 2 e s Z = ∂(s e V ), 1/2 −P/2 3 P/2 1/2 2 1/2 P/2 2 1/2 −P/2 3 s e Z + e Qs Z = ∂(s Qe V + s e V ). Since the quantities appearing in (112) are preserved along characteristics, we obtain P/2 1/2 2 e s Z = 0, (146) ds 1/2 −P/2 3 P/2 1/2 2 s e Z + e Qs Z = 0. (147) ds We also have 1 1 dΨ 1 α 1 V V 1/2 λ/2 = λ − Ψ + α λ Ψ + e ∂ . (148) t θ 0 0 ds 2 α 2 V V 1 0 i In the end, we shall express ∂(V /V ) in terms of Z and Ψ. However, there is no immediate gain in doing so here. The most cumbersome part of the argument is to compute the derivative of Z . This calculation can be divided into several parts. Let us begin by considering d dV (∂V ) = ∂ . ds ds When calculating the right hand side, it is convenient to divide the result into terms which include a ∂V factor, i = 1, 2, 3, and terms which do not. Combining Lemmas 50, 53 and 54, the terms which include such a factor can be written 1 i − ∂V + c ∂V , 2s −2 1 where c (s) = O(s ); note that the factor in front of V in (109) is given by −2 − + O(s ). 2s It is straightforward to calculate the remaining terms, and we conclude that 2 3 d 1 1 α V V 1 1/2 0 1 1/2 P (∂V ) = − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 4 4 α V 3 2 2 2 1 (V ) − (V ) 1/2 −7/4 λ/4 −P/2 2 − ∂ (α P ) ∂Θ + s ∂ (e e J)V ∂Θ θ θ θ 2 V −7/4 λ/4 P/2 3 1 i + s ∂ [e e (K − QJ)]V ∂Θ − ∂V + c ∂V , θ i 2s 33 −2 where c (s) = O(s ). Combining an argument which is identical to the proof of (133) with Lemma 50, we can estimate the factors multiplying ∂Θ in the third and fourth last terms on the right hand side. In fact, we obtain 2 3 d 1 1 1 α V V 1 1 1/2 0 1 1/2 P (∂V ) = − ∂V − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 2s 4 4 α V 3 2 2 2 1 (V ) − (V ) 1/2 i − ∂ (α P ) ∂Θ + c ∂Θ + c ∂V , θ θ θ i 2 V −2 −3 where c (s) = O(s ) and c (s) = O(s ). As a next step, it is of interest to consider the terms i θ α 1 that arise when d/ds hits a V in the second term in the definition of Z . Before writing down the result, let us note that, due to Lemmas 50, 53 and 54, i 0 dV 1 dV i −2 −2 = − V + O(s ), = O(s ). (149) ds 2s ds Due to these observations, Lemmas 50, 53 and 54, as well as (135) and the definition of Z , we α 1 conclude that when d/ds hits a V in the second term in Z , the resulting expression can be −2 written c ∂Θ, where c (s) = O(s ). θ θ Note that every term appearing in the second term in the definition of Z can be written in the form −1/2 h(·, Θ)ψ(V )α (·, Θ)∂Θ. (150) We have already estimated the terms that arise when d/ds hits ψ. Let us therefore consider the terms that arise when the derivative hits the remaining factors. Omitting the arguments, we need to consider V α 1/2 −1/2 h + α h ψα ∂Θ − hψ∂Θ t θ 0 3/2 V 2α 1 1 1 α V α V V θ θ 1/2 −1/2 + hψ − α ∂Θ + α ∂Θ + ∂ (151) 3/2 0 1/2 0 0 2α V 2α V V 1 1 V V −1/2 = ∂ (α h) + h ψ∂Θ + hψ∂ . t θ 0 0 V V −2 In all the terms of interest, hψ = O(s ), where we have used Lemmas 50, 53 and 54. As a consequence, this expression can be written −1/2 i ∂ (α h) + h ψ∂Θ + c ∂V , t θ i −2 where c (s) = O(s ). Adding up the above observations, we conclude that 1   2 3 dZ 1 1 1 α V V 1 1/2 0 1 1/2 P = − ∂V − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 2s 4 4 α V h  i 3 2 2 2 1 (V ) − (V ) 1 α 1/2 −1/2 1/2 λ/2 0 − ∂ (α P ) ∂Θ + ∂ α λ − 2 − 4s e Λ V ∂Θ θ θ t t 2 V 4 α 2 2 3 2 1 α 1 (V ) − (V ) 1/2 λ/2 1 −1/2 0 + ∂ λ − 2 − 4s e Λ V ∂Θ − ∂ (α P )V ∂Θ θ t t t 0 2 1 2 4 α 2 (V ) − (V ) 2 2 3 2 2 2 3 2 1 (V ) − (V ) 1 (V ) − (V ) 1 1 − P V ∂Θ + P V ∂Θ tθ tθ (152) 0 2 1 2 0 2 1 2 2 (V ) − (V ) 2 (V ) − (V ) 1 2 2 2 3 2 0 2 3 1 (V ) (V ) − (V ) V V V 1/2 −1/2 P + ∂ (α P ) ∂Θ − ∂ (α e Q ) ∂Θ θ θ t t 0 0 2 1 2 0 2 1 2 2 V (V ) − (V ) (V ) − (V ) 1 2 3 1 2 3 V V V V V V P P − ∂ (e Q ) ∂Θ + ∂ (e Q ) ∂Θ θ t t θ 0 2 1 2 0 2 1 2 (V ) − (V ) (V ) − (V ) 1 1 2 3 V V V V 1/2 P i + ∂ (α e Q ) ∂Θ + c ∂Θ + c ∂V , θ θ θ i 0 0 2 1 2 V (V ) − (V ) 34 −2 −2 where c (s) = O(s ) and c (s) = O(s ). How to interpret the terms in this equation should be θ i clear from (150) and (151). However, there is one term which is slightly ambiguous, namely the sixth one on the right hand side of (152). For the sake of clarity, let us point out that the factor in front of V ∂Θ in this term should be interpreted as the time derivative of h  i 1 α −1/2 1/2 λ/2 α λ − 2 − 4t e Λ , 4 α evaluated at [s, Θ(s)]. The expression (152) can be simplified somewhat. To begin with, the terms involving P cancel. Moreover, tθ P P P −∂ (e Q ) + ∂ (e Q ) = e (P Q − P Q ). θ t t θ t θ θ t Using Lemma 53 and (135), we can estimate this expression in order to conclude that the sum of −3 the fourth and fifth last terms is of the form c ∂Θ, where c = O(s ) (in order to obtain this θ θ conclusion, we have also used Lemma 50). Using Lemma 53, (127), (129) and (135), the sum of −2 the third and the seventh term can be written c ∂Θ, where c = O(s ). Since θ θ 3 2 2 2 1 2 2 2 3 2 2 2 3 2 (V ) − (V ) (V ) (V ) − (V ) (V ) − (V ) − + = V 0 0 0 2 1 2 0 2 1 2 V V (V ) − (V ) (V ) − (V ) 2 3 1 1 2 3 0 2 3 V V V V V V V V V + = , 0 0 0 2 1 2 0 2 1 2 V V (V ) − (V ) (V ) − (V ) 1/2 the terms involving a factor of ∂ (α P ) can be written θ θ 2 2 3 2 1 (V ) − (V ) 1/2 0 ∂ (α P )V ∂Θ, θ θ 0 2 1 2 2 (V ) − (V ) 1/2 P and the terms involving a factor of ∂ (α e Q ) can be written θ θ 0 2 3 V V V 1/2 P ∂ (α e Q ) ∂Θ. θ θ 0 2 1 2 (V ) − (V ) Combining these observations yields h  i dZ 1 1 α 1 1 −1/2 1/2 λ/2 0 1/2 0 = − ∂V + ∂ α λ − 2 − 4s e Λ V ∂Θ − ∂ (α λ )V ∂Θ t t θ θ ds 2s 4 α 4 2 2 3 2 1 (V ) − (V ) −1/2 1/2 0 − [∂ (α P ) − ∂ (α P )]V ∂Θ t t θ θ (153) 0 2 1 2 2 (V ) − (V ) 0 2 3 V V V −1/2 P 1/2 P i − [∂ (α e Q ) − ∂ (α e Q )] ∂Θ + c ∂Θ + c ∂V , t t θ θ θ i 0 2 1 2 (V ) − (V ) −2 −2 where c (s) = O(s ) and c (s) = O(s ). Combining (61), (135) and Lemmas 53 and 54 yields θ i −1/2 1/2 −1 ∂ (α P ) − ∂ (α P ) = O(t ). t t θ θ Due to this estimate and Lemmas 53 and 50, the fourth term on the right hand side of (153) can −2 be written c ∂Θ, where c (s) = O(s ). Keeping (68) in mind, a similar argument yields the θ θ same conclusion concerning the fifth term on the right hand side of (153). Finally, keeping (87) in mind, a similar argument yields the conclusion that the combination of the second and third −3/2 terms on the right hand side of (153) can be written c ∂Θ, where c (s) = O(s ). To conclude, θ θ dZ 1 1 i = − ∂V + c ∂Θ + c ∂V , (154) θ i ds 2s −3/2 −2 where c (s) = O(s ) and c (s) = O(s ). On the other hand, due to (127) and (135), the θ i −1/2 λ/2 function α e can be bounded from above and below by positive constants (for t ≥ t ). 35 In other words, ∂Θ and Ψ are interchangeable when deriving equations of the form (142)–(145). Moreover, due to (135) and Lemmas 50, 53 and 54, i i Z = ∂V + c Ψ, (155) i,θ −1/2 where c (s) = O(s ). We are now in a position to prove the lemma. To begin with, combining i,θ (146), (147) and Lemma 53 yields (143) and (144). Combining (148) and (155) with Lemmas 50, 53 and 54 yields (145). Finally, combining (154) and (155) yields (142). Lemma 57. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 2 1/2 i 2 (ln s) |∂ Θ(s; t, θ, v)| + s ∂ V (s; t, θ, v) ≤ C(ln t) (156) θ θ 1 3 for all s ∈ [t , t] and (t, θ, v) ∈ [t , ∞) × S × R in the support of f. 1 1 Remark 58. It is possible to use arguments similar to the ones given below in order to derive estimates for ∂ Θ, ∂ iΘ etc. t v Proof. Let i 1/2 i 2 ˆ ˆ Z (s; t, θ, v) = s Z (s; t, θ, v), Ψ(s; t, θ, v) = (ln s) Ψ(s; t, θ, v). (157) Then, due to Lemma 56, dZ 1/2 −2 j ˆ ˆ = c s (ln s) Ψ + c Z , 1,θ 1,j ds dZ = c Z , 2,2 ds dZ 2 3 ˆ ˆ = c Z + c Z , 3,2 3,3 ds dΨ 2 −1/2 2 i ˆ ˆ ˆ = Ψ + c Ψ + c s (ln s) Z θ,θ θ,i ds s lns for s ∈ [t , t], with coefficients as in Lemma 56. Introducing i 2 2 ˆ ˆ ˆ E = (Z ) + (Ψ) , (158) i=1 we conclude that there is a constant C > 0, depending only on the solution, such that dE 1 ≥ −C E ds s(ln s) for s ∈ [t , t]. As a consequence, ˆ ˆ E(s; t, θ, v) ≤ CE(t; t, θ, v) (159) for s ∈ [t , t]. Let us now assume ∂ = ∂ . Then 1 θ Ψ(t; t, θ, v) = O[(ln t) ]. Moreover, i 1/2 i Z (t; t, θ, v) = [t ∂ V + O(1)Ψ](t; t, θ, v) = O(1). As a consequence, E(t; t, θ, v) = O[(ln t) ]. Thus (159) implies (156); note that the estimate for ∂ Θ is immediate and that i i −1/2 |∂ V | ≤ |Z | + Cs |∂ Θ|. θ θ The lemma follows. 36 7 Higher order light cone estimates Before proceeding to the higher order light cone estimates, let us record some consequences of the estimates obtained in Lemma 57. Lemma 59. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 1/2 −3/2 2 k∂ ρk 0 + t k∂ J k 0 + tk∂ S k 0 + tk∂ P k 0 ≤ Ct (ln t) , (160) θ C θ i C θ ij C θ k C −3/2 ∂ ≤ Ct , (161) ≤ C (162) α C for all t ≥ t . Remark 60. Note that ∂ (α /α) = ∂ (α /α) = ∂ ∂ ln α. θ t t θ t θ Proof. The estimate (160) follows from the fact that (137) and (156) hold and the fact that i −1/2 |v | ≤ Ct in the support of f(t, ·). Consider (69). Since Lemmas 53 and 54 together with 1 −1/2 (135) imply that J, K, Q and P are bounded in C and that λ is O(t ), the first two terms −4 1 −1/2 on the right hand side of (69) are O(t ) in C . Since λ is O(t ), the θ-derivative of the third −3/2 term on the right hand side of (69) is O(t ). Due to (160), the θ-derivative of the last term is better. Thus (161) holds, so that α α θ t −3/2 ∂ ≤ ∂ ≤ Ct . t θ 0 0 α α C C Integrating this estimate yields (162). In what follows, we shall proceed inductively in order to derive estimates for higher order deriva- tives. Let us therefore assume that we have a T -symmetric solution to the Einstein-Vlasov equations with a cosmological constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. As- 0 0 sume, moreover, that the solution has λ-asymptotics and let t = t +2. Let us make the following 1 0 inductive assumption. Inductive assumption 61. For some 1 ≤ N ∈ Z, there are constants 0 ≤ m ∈ Z and C , j j j = 1, . . ., N, (depending only on N and the solution) such that j j ∂ V ∂ Θ 1/2 m s (s; t, θ, v) + (s; t, θ, v) ≤ C (ln t) , (163) j j ∂θ ∂θ 3/2 3/2 kP k N−1 + kQ k N−1 + t kP k N−1 + t kQ k N−1 ≤ C , (164) θ θ t t N−1 C C C C 1 3 for all j = 1, . . ., N, (t, θ, v) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]. 1 1 Remarks 62. The induction hypothesis holds for N = 1. In what follows, C and m will change j j from line to line. However, they are only allowed to depend on N and the solution. In this section we prove that, given that Inductive assumption 61 holds, then (164) holds with N replaced by N + 1. In the next section, we close the induction argument by proving that (163) holds with j replaced by N + 1. We shall need the following consequences of the inductive assumption. 37 2 Lemma 63. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there are constants C , j = 0, . . ., N, and m , depending only on N and the j N solution, such that 1/2 −3/2 m kρk N + t kJ k N + tkP k N + tkS k N ≤ C t (ln t) , (165) i i im N C C C C 1/2 l+1 t k∂ λk 0 ≤ C , (166) −1 j α ∂ α ≤ C , (167) kJ k N + kK k N ≤ C (ln t) , (168) θ θ N C C l+1 −3/2 ∂ ≤ C t (169) α C for t ≥ t , 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and i, m = 1, 2, 3. Proof. For N = 1, the conclusions follow from Lemmas 53, 54 and 59, (135) and the equations (64) and (65). We may thus, without loss of generality, assume that N ≥ 2. An immediate consequence of the inductive assumption is that, for 0 ≤ j ≤ N and t ≥ t , ∂ f ≤ C (ln t) ; ∂θ cf. (137) and (163). As a consequence of this estimate, we obtain (165). In order to obtain control of the θ-derivatives of α and λ, we need to proceed inductively. Let us make the inductive assumption that −1 α ∂ α ≤ C , (170) j −1/2 k∂ λk ≤ C t (171) C j for 1 ≤ j ≤ l < N. Note that we know the inductive hypothesis to be true for l = 1. Differentiating (63) l times with respect to θ and appealing to (164), (165), (170) and (171), we conclude that (171) holds with j replaced by l + 1. In order to improve our knowledge concerning α, let us begin by improving our estimates for the θ-derivatives for J and K. Differentiating (64) and (65) 0 ≤ j ≤ l times and using (164), (165), (170) and (171), we conclude that j+1 j+1 k∂ Jk 0 + k∂ Kk 0 ≤ C (ln t) (172) C C j θ θ for t ≥ t and 0 ≤ j ≤ l. Differentiating (69) l + 1 times with respect to θ, using (164), (165), (172) as well as the fact that (171) holds for 1 ≤ j ≤ l + 1, we obtain α α t θ l+1 l −3/2 ∂ = ∂ ∂ ≤ C t (173) t l 0 0 α C α C for t ≥ t . Thus ∂ ≤ C α C for t ≥ t . Combining this estimate with the inductive hypothesis, we conclude that (170) holds with j replaced by l + 1. Thus (171) and (170) hold for 1 ≤ j ≤ N. We thus conclude that (166), (167) and (168) hold. In addition, (173) implies that (169) holds. We are now in a position to derive higher order light cone estimates. Lemma 64. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 38 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there is a constant C > 0, depending only on N and the solution, such that 3/2 N N 3/2 P N P N t k∂ P k 0 + k∂ P k 0 + t ke ∂ Q k 0 + ke ∂ Q k 0 ≤ C (174) t C θ C t C θ C N θ θ θ θ for all t ≥ t . As a consequence, (164) holds with N replaced by N + 1. Proof. Let us begin by pointing out that if N = 1, then the lower bound is larger than the upper bound in some of the sums below. In that case, the corresponding sum should be equated with −3 −3 zero. Moreover, terms which are bounded by Ct for t ≥ t will sometimes be written O(t ). Let us compute N N 1/2 N N 1/2 1/2 1/2 N+1 ∂ [∂ P ∓ ∂ (α P )] =∂ P ∓ ∂ α P + α P ± α ∂ P ± t θ tt θ tθ t θ θ θ θ 2α N+1 1/2 1/2 − α ∂ (α P ) N−1 1 α 1 α t t N N−j j 1/2 N 1/2 =∂ P ∓ β ∂ ∂ (α P ) ∓ ∂ (α P ) tt j θ θ θ θ θ θ 2 α 2 α j=0 N−2 Nα N−j j+1 N 1/2 N 1/2 1/2 ∓ ∂ P ∓ β ∂ (α )∂ (P ) − ∂ [α ∂ (α P )] t j t θ θ θ θ θ θ 1/2 2α j=0 N−2 Nα θ N−j j+1 N 1/2 1/2 1/2 + ∂ (α P ) + β ∂ (α )∂ (α P ), θ j θ θ θ θ 1/2 2α j=0 −3 where the β are binomial coefficients. Note that all the sums are O(t ) due to Inductive as- sumption 61, Lemma 63 and (135). Let us use (61) in order to compute N 1/2 1/2 N ∂ [P − α ∂ (α P )] =∂ P − αP − P tt θ θ tt θθ θ θ θ N−1 1 α α t N−j t j N N = − ∂ P + ∂ P + β ∂ ∂ P t t j t θ θ θ θ t 2α 2α j=0 N−1 N−j 2P j 2 2 + β ∂ (e )∂ (Q − αQ ) θ θ t θ j=0 N−1 N−j j 2P 1/2 1/2 + e β (∂ (Q − α Q ))(∂ (Q + α Q )) j t θ t θ θ θ j=1 P +λ/2 2 e (K − QJ) 2P N 1/2 N 1/2 N + 2e [Q ∂ Q − α Q ∂ (α Q )] − ∂ t t θ θ θ θ θ 7/2 2t −P +λ/2 2 e J N −1/2 N λ/2 + t ∂ [e (P − P )]. + ∂ 2 3 θ θ 7/2 2t −3 Due to Inductive assumption 61, Lemmas 63 and 54, and (135), the sums are O(t ), as well as the last three terms on the right hand side. We thus obtain 1 α Nα t θ N N 1/2 N N N 1/2 N N 1/2 ∂ [∂ P ∓ ∂ (α P )] = − ∂ P + [∂ P ∓ ∂ (α P )] ∓ [∂ P ∓ ∂ (α P )] ± t θ t t θ t θ θ θ θ θ θ θ θ 1/2 t 2α 2α 2P N 1/2 N 1/2 −3 + 2e [Q ∂ Q − α Q ∂ (α Q )] + O(t ). t t θ θ θ θ Introducing N N 1/2 2 2P N N 1/2 2 A = [∂ P ± ∂ (α P )] + e [∂ Q ± ∂ (α Q )] , (175) N+1,± t θ t θ θ θ θ θ 39 we conclude that 5 2 N N 1/2 2 N N 1/2 2 N 1/2 N N 1/2 ∂ [∂ P ∓ ∂ (α P )] ≤ − [∂ P ∓ ∂ (α P )] ∓ ∂ (α P )[∂ P ∓ ∂ (α P )] ± t θ t θ θ t θ θ θ θ θ θ θ θ t t 1/2 −3 −3/2 + C t A + C t (A + A ), N N N+1,+ N+1,− N+1,∓ where we have used Inductive assumption 61, Lemma 63, (128) and (135). Let us now consider N−1 1 α 1 α t t N N 1/2 N N−j j 1/2 N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] =∂ Q ∓ β ∂ ∂ (α Q ) ∓ ∂ (α Q ) ± t θ tt j θ θ θ θ θ θ θ θ 2 α 2 α j=0 N−2 Nα θ N−j j N 1/2 N 1/2 1/2 ∓ ∂ Q ∓ β ∂ (α )∂ Q − ∂ [α ∂ (α Q )] t j tθ θ θ θ θ θ θ 1/2 2α j=0 N−2 Nα θ N−j j+1 N 1/2 1/2 1/2 + ∂ (α Q ) + β ∂ (α )∂ (α Q ). θ j θ θ θ θ 1/2 2α j=0 −3 As above, all the sums are O(t ) due to Inductive assumption 61, Lemma 63 and (135). Compute, using (68), N−1 1 α α t N−j t j N 1/2 1/2 N N N ∂ [Q − α ∂ (α Q )] = − ∂ Q + ∂ Q + β ∂ ∂ Q − 2(∂ P )Q tt θ θ t t j t t t θ θ θ θ θ θ t 2α 2α j=0 N N 1/2 1/2 1/2 N 1/2 − 2P (∂ Q ) + 2∂ (α P )α Q + 2α P ∂ (α Q ) t t θ θ θ θ θ θ θ N−1 N−1 X X N−j j N−j j 1/2 1/2 − 2 β (∂ P )(∂ Q ) + 2 β (∂ (α P ))(∂ (α Q )) j t t j θ θ θ θ θ θ j=1 j=1 λ/2−P e J(K − QJ) N −1/2 N λ/2−P + ∂ + 2t ∂ (e S ). θ θ 7/2 −3 Due to Inductive assumption 61, Lemmas 63 and 54, and (135), the sums are O(t ), as well as the last two terms on the right hand side. Thus 1 α Nα t θ N N 1/2 N N N 1/2 N N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] = − ∂ Q + [∂ Q ∓ ∂ (α Q )] ∓ [∂ Q ∓ ∂ (α Q )] ± t θ t t θ t θ θ θ θ θ θ θ θ 1/2 t 2α 2α N N N 1/2 1/2 − 2(∂ P )Q − 2P (∂ Q ) + 2∂ (α P )α Q θ t t t θ t θ θ θ 1/2 N 1/2 −3 + 2α P ∂ (α Q ) + O(t ). θ θ Consequently, 5 2 N N 1/2 2 N N 1/2 2 N 1/2 N N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] ≤ − [∂ Q ∓ ∂ (α Q )] ∓ ∂ (α Q )[∂ Q ∓ ∂ (α Q )] ± t θ t θ θ t θ θ θ θ θ θ θ θ t t 1/2 −3 −3/2 + C t A + C t (A + A ), N N N+1,+ N+1,− N+1,∓ where we have used Inductive assumption 61, Lemma 63, (128) and (135). Adding up the above estimates, we conclude that 5 1 1 ∂ A ≤ − A + (A − A ) + (A + A ) ± N+1,∓ N+1,∓ N+1,∓ N+1,± N+1,+ N+1,− t 2t t 1/2 −3/2 −3 + C t (A + A ) + C t A . N N+1,+ N+1,− N N+1,∓ Let us introduce 7/2 1/2 ˆ ˆ ˆ ˆ ˆ ˆ A = t A + t , F = sup A , F = F + F . (176) N+1,± N+1,± N+1,± N+1,± N+1 N+1,+ N+1,− θ∈S 40 Then −3/2 ˆ ˆ ˆ ˆ ∂ A ≤ A + C t (A + A ). ± N+1,∓ N+1,± N N+1,+ N+1,− 2t Integrating this differential inequality, taking the supremum etc., we obtain −3/2 ˆ ˆ ˆ ˆ F (t) ≤ F (t ) + F (s) + C s F (s) ds. N+1 N+1 1 N+1 N N+1 2s 1/2 Thus F (t) ≤ C t . Combining this estimate with Inductive assumption 61, Lemma 63 and N+1 N (135), we obtain (174). 8 Higher order derivatives of the characteristic system In the previous section we showed that (164) holds with N replaced by N + 1; i.e., that 3/2 3/2 kP k N + kQ k N + t kP k N + t kQ k N ≤ C (177) θ C θ C t C t C N holds for all t ≥ t . We also need to prove that (163) holds with j replaced with N + 1. Before stating the relevant result, let us make the following preliminary observation. Lemma 65. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there are constants C and m , j = 0, . . ., N, depending only on N and the j j solution, such that 1/2 −3/2 m kρk N + t kJ k N + tkP k N + tkS k N ≤ C t (ln t) , (178) i i im N C C C C j j+1 −1 1/2 α ∂ α + t k∂ λk 0 ≤ C , (179) θ θ kJ k N + kK k N ≤ C (ln t) , (180) θ C θ C N l+1 l+1 −3/2 ∂ + k∂ λ k 0 ≤ C t , (181) t C l θ θ α C 1/2 λ/2 −2 λ − 2 − 4t e Λ ≤ C t , (182) t N α C −2 λ − ≤ C t (183) t N for t ≥ t , 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and i, m = 1, 2, 3. Moreover, using the notation j i j i i j i j Ψ = ∂ Ψ, Z = ∂ Z , V = ∂ V , Θ = ∂ Θ (184) j j j j θ θ θ θ (where the ∂-operator used to define Z and Ψ is given by ∂ ), there are functions c , i = 1, 2, 3, θ i,θ such that the following estimates hold: 1/2 m |Ψ (s; t, θ, v)| + s |Z (s; t, θ, v)| ≤ C (ln t) , (185) l l l −1/2 λ/2 m |Ψ (s; t, θ, v) − (α e )[s, Θ(s; t, θ, v)]Θ (s; t, θ, v)| ≤ C (ln t) , (186) j j+1 j i i −1/2 m |Z (s; t, θ, v) − V (s; t, θ, v) − (c Ψ )(s; t, θ, v)| ≤ C s (ln t) , (187) i,θ j j j j+1 −1/2 |c (s; t, θ, v)| ≤ C s (188) i,θ 0 1 3 for all (t, θ, v) ∈ [t , ∞) × S × R in the support of f, 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and s ∈ [t , t]. 1 1 Remark 66. Due to (179), j p p |∂ α | ≤ C α p,j for all (t, θ) ∈ [t , ∞) × S , p ∈ R and 0 ≤ j ≤ N. In particular, spatial derivatives of powers of α can thus effectively be ignored. In the derivation of the estimates below, it is useful to keep this observation in mind. 41 Proof. Combining Lemma 63 with (177), (60) and (63), we obtain (178)–(181); recall that P , Q, J and K are bounded to the future. The estimate (182) is a consequence of Lemma 63, (177) and the fact that −P +λ/2 2 α e J 1/2 λ/2 2 2 2P 2 2 λ − 2 − 4t e Λ =t P + αP + e (Q + αQ ) + t θ t θ 5/2 α t P +λ/2 2 e (K − QJ) 1/2 λ/2 + + 4t e ρ; 5/2 cf. (69) and (70). For similar reasons, (183) holds; cf. (70). Turning to (185)–(188), let us note that ∂ Ψ(s; t, θ, v) can, up to numerical factors, be written as a sum of terms of the form k −1/2 λ/2 ∂ (α e )[s, Θ(s; t, θ, v)]Θ (s; t, θ, v)· · · Θ (s; t, θ, v), i i θ k+1 where i + · · · + i = j + 1. Since k ≤ j ≤ N, the first factor is bounded due to (179). Due to 1 k+1 Inductive assumption 61, the factors Θ can be estimated by C(ln t) if i ≤ N. The only way i j a factor Θ could occur is if k = 0 and all the derivatives hit Θ in the definition of Ψ. Due to N+1 1 these observations, (186) and the estimate |Ψ (s; t, θ, v)| ≤ C (ln t) , l l 0 ≤ l ≤ N − 1, follow. The proof of the remaining estimates is similar in nature, but somewhat more involved. We now finish the induction argument by proving that (163) holds with j replaced by N + 1. Lemma 67. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then (163) holds with j replaced by N + 1. Proof. The strategy of the proof is very similar to that of the proof of Lemma 56. The idea is to derive a system of ODE’s for Z and Ψ analogous to (142)–(145), and then to use arguments similar to those presented in the proof of Lemma 57. Deriving appropriate equations for Z , i = 2, 3, turns out to be relatively easy; this is due to (146) and (147). In fact, we obtain the desired conclusions concerning Z , i = 2, 3, without much effort; cf. (191) and (192) below. Deriving an equation for Ψ also turns out to be quite easy; cf. (190) below. Similarly to the proof of Lemma 56, the main difficulty consists of deriving an equation for Z . Once the desired equation has been obtained, we rescale Z and Ψ according to (199) and introduce an energy according to (200); note that these definitions are analogous to the ones given in the proof of Lemma 57. Finally, the equations imply a differential inequality for the energy E which can be integrated to yield the desired estimate. Before proceeding to a proof of the statement of the lemma, it is of interest to introduce some nota- 1 1 tion. Let b be a C function on M = (t , ∞)×S . Evaluating this function along a characteristic, we obtain B(s; t, θ, v) = b[s, Θ(s; t, θ, v)]. Differentiating B with respect to θ, we obtain ∂B ∂b ∂Θ (s; t, θ, v) = [s, Θ(s; t, θ, v)] (s; t, θ, v). (189) ∂θ ∂θ ∂θ On the other hand, distinguishing between B and b is quite cumbersome in the arguments that we are about to carry out. As a consequence, we shall write b when we mean B. Moreover, we shall use the notation ∂ b as a shorthand for ∂ B, whereas ∂ b should be interpreted as the first factor ϑ θ θ 42 on the right hand side of (189) and ∂ b should be interpreted as the function mapping (s; t, θ, v) to ∂b [s, Θ(s; t, θ, v)]. ∂t In particular, we thus have b = b Θ . Finally, let us point out that if in some expression a ϑ θ 1 ϑ-derivative hits a V or a Θ, it is to be interpreted as an ordinary θ-derivative. Note, to begin with, that 1 1 1 dΨ 1 α 1 V ∂V V 1/2 λ/2 λ/2 i i = λ − Ψ + α λ Ψ + e − e V ∂V t θ 0 0 0 3 ds 2 α 2 V V (V ) i=1 due to (148). Differentiating this equality N times with respect to ϑ, we obtain 1 1 dΨ 1 α 1 V V N t 1/2 λ/2 N+1 λ/2 i i −2 m = λ − Ψ + α λ Ψ +e −e V V +O[s (ln t) ], t N θ N N+1 0 0 0 3 ds 2 α 2 V V (V ) i=1 where we have used (135), Inductive assumption 61 and Lemma 65. Due to (187), this equation can be written dΨ N N i −2 m = c Ψ + c Z + O[s (ln t) ], (190) θ,θ θ,i N ds N −2 N −3/2 2 where c = O(s ), c = O(s ) and we sum over i but not N. Turning to Z , we have θ,θ θ,i 2 1 dZ 1 1 V 2 1/2 2 = − Z − P + α P Z ; t θ ds 2s 2 V cf. (146). Differentiating this equality N times with respect to ϑ, we obtain dZ 1 1 V N 2 1/2 2 −2 m = − Z − P + α P Z + O[s (ln t) ], t θ N N ds 2s 2 V where we have used (135), (177), Inductive assumption 61 and Lemma 65. As a consequence, 1/2 P/2 2 −3/2 m s e Z = O[s (ln t) ]. ds Integrating this equality from s to t, we obtain (assuming N ≥ 1) 1/2 P/2 2 m − s e Z (s; t, θ, v) = O[(ln t) ]; note that 1/2 P/2 2 1/2 P/2 N+1 t e Z (t; t, θ, v) = t e (∂ P )v = O(1) due to (177). In particular, we thus have 2 −1/2 m |Z (s; t, θ, v)| ≤ C s (ln t) (191) for s ∈ [t , t]. Turning to Z , we have 3 1 1 dZ 1 1 V V 3 1/2 3 P 1/2 2 = − Z + P + α P Z − e Q + α Q Z ; t θ t θ 0 0 ds 2s 2 V V cf. (147) and (146). Differentiating N times with respect to ϑ, we obtain 3 1 1 V V dZ 1 1 N 3 1/2 3 P 1/2 2 −2 m = − Z + P + α P Z − e Q + α Q Z + O[s (ln t) ]. t θ t θ N N N 0 0 ds 2s 2 V V 43 −2 m Due to (191), the third term on the right hand side is O[s (ln t) ]. We can thus proceed as in the proof of (191) in order to obtain 3 −1/2 m |Z (s; t, θ, v)| ≤ C s (ln t) (192) for s ∈ [t , t]. Finally, we need to derive an equation for Z . Just as in the derivation of the equation for Z , it is natural to divide the analysis into several steps. Consider, to begin with, dV j+1 ds for 0 ≤ j ≤ N. All the terms appearing in dV /ds can be written h ψ◦V . When differentiating an k l expression of this form, the terms that arise are (up to numerical factors) of the form ∂ h∂ ψ ◦V . ϑ θ −2 m If both k and l are ≥ 1, the resulting term is O[s (ln t) ]. If all the derivatives hit ψ, we obtain (after summing over all the terms appearing in dV /ds) 1 i −2 m − V + c V + O[s (ln t) ] j+1 j+1 2s −2 where c = O(s ) and we sum over i but not j. If all the derivatives hit h, we obtain (after summing over all the terms appearing in dV /ds) 2 3 1 1 α V V j+1 j+1 t j+1 1/2 0 1 1/2 P − ∂ (α λ )V − ∂ λ − 2 V + ∂ (α e Q ) θ t θ ϑ ϑ ϑ 4 4 α V 3 2 2 2 1 (V ) − (V ) j+1 j 1/2 −3 m − ∂ (α P ) + c Θ + O[s (ln t) ], θ j+1 ϑ θ 2 V −3 where c = O(s ) and we have used (132); note that, due to (177) and Lemma 65, we control N + 1 θ-derivatives of the first factor in each of the last two terms appearing on the right hand side of (109). Adding up, we conclude that dV 1 1 1 α j+1 j j j+1 j+1 t 1 i 1/2 0 1 ∂ = − V + c V + c Θ − ∂ (α λ )V − ∂ λ − 2 V j+1 θ t j+1 j+1 ϑ i θ ϑ ϑ ds 2s 4 4 α 2 3 3 2 2 2 V V 1 (V ) − (V ) j+1 j+1 1/2 P 1/2 −2 m + ∂ (α e Q ) − ∂ (α P ) + O[s (ln t) ], θ θ ϑ ϑ 0 0 V 2 V (193) j j −2 −3 where c = O(s ) and c = O(s ) and we sum over i but not j. i θ The second term in the definition of Z consists of a sum of terms of the form −1/2 h ψ ◦ V α Θ . (194) The relevant h’s are 1 α 1 1 1/2 λ/2 1/2 P 1/2 P h = λ − 2 − 4s e Λ , h = − P , h = α P , h = −e Q , h = α e Q , 1 t 2 t 3 θ 4 t 5 θ 4 α 2 2 and the relevant ψ’s are 2 2 3 2 2 2 3 2 0 2 3 (V ) − (V ) (V ) − (V ) V V V 0 0 1 ψ =V , ψ = V , ψ = V , ψ = , 1 2 3 4 0 2 1 2 0 2 1 2 0 2 1 2 (V ) − (V ) (V ) − (V ) (V ) − (V ) 1 2 3 V V V ψ = . 0 2 1 2 (V ) − (V ) We want to differentiate (194) with respect to s and then N times with respect to ϑ. Before turning to the details, let us record the following estimate: j j 1/2 −3/2 m s |(∂ h )(s; t, θ, v)| + |(∂ h )(s; t, θ, v)| ≤ C s (ln t) (195) 1 i N ϑ ϑ i=2 44 1 3 for 0 ≤ j ≤ N, (t, v, θ) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]. In the case of 1 1 h , i = 2, . . ., 5, (195) is an immediate consequence of the inductive hypothesis, (135), (177) and Lemma 65, and in the case of h , it is a consequence of (182). We also have 5 5 X X j+1 −1 m |ψ (s; t, θ, v)| + s |ψ (s; t, θ, v)| ≤ C, |(∂ ψ ◦ V )(s; t, θ, v)| ≤ C s (ln t) (196) 1 i i j i=2 i=1 1 3 for 0 ≤ j ≤ N − 1, (t, v, θ) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]; this is an 1 1 immediate consequence of the inductive hypothesis. Let us consider the term that arises when d/ds hits the ψ-factor in (194). Note, to this end, that dV −3/2 m ∂ ≤ C s (ln t) ds for all i = 1, 2, 3 and all 0 ≤ j ≤ N; for j = 0, the estimate is a consequence of (149); for j ≥ 1 and i = 1, it is a consequence of (193); and in the case of i = 2, 3, it follows immediately from (110), (111) and the induction hypothesis. Due to the above observations, dψ ◦ V N −1/2 N −2 m ∂ h α Θ = c Θ + O[s (ln t) ], 1 N+1 ϑ θ ds N −2 where c = O(s ) and h is one of h , . . . , h . When the s-derivative hits the remaining terms in 1 5 (194) (not ψ), we obtain " # 1 1 1 V ∂V V N −1/2 i i ∂ ∂ (α h) + h ψΘ + hψ − hψ V ∂V ; t θ 1 0 0 0 3 V V (V ) i=1 cf. (151). Due to (195) and (196), this expression can be written N −1/2 N i −2 mN ∂ ∂ (α h) + h ψΘ + c V + O[s (ln t) ], t θ 1 ϑ i N+1 i=1 N −2 1 where c = O(s ). Differentiating the second term appearing in the definition of Z once with respect to s and N times with respect to ϑ, we obtain (by adding up the above) 5 3 X X N −1/2 N i N −2 m ∂ ∂ (α h ) + (∂ h ) ψ Θ + c V + c Θ + O[s (ln t) ], (197) t i θ i i 1 N+1 ϑ i N+1 θ i=1 i=1 N −2 N −2 where c = O(s ) and c = O(s ). In order to obtain the desired equation we need to add θ i this expression to (193) with j = N. However, before doing so, note that 1 1 N+1 1/2 0 N 1/2 0 −2 m − ∂ (α λ )V = − ∂ [∂ (α λ )V Θ ] + O[s (ln t) ] θ θ θ 1 4 4 etc. Due to this observation, we can argue as in the proof of Lemma 56. In particular, we obtain 1 1 a formula analogous to (152): the difference is that ∂V should be replaced by V in the first N+1 term on the right hand side of (152); that ∂ should be applied to all but the first and last two terms on the right hand side of (152); and that the last two terms should be replaced by ones analogous to the last three terms on the right hand side of (197). Proceeding as in the proof of Lemma 56, the corresponding expression can be simplified; cf. the derivation of (153). Most of the steps involved in the derivation of (153) consist of algebraic manipulations. However, there are two exceptions. The combination of the fourth and fifth last terms on the right hand side of (152) can be written 1 2 3 V V V e (P Q − P Q ) Θ . t θ θ t 1 0 2 1 2 (V ) − (V ) 45 The analogous expression in the present setting is 1 2 3 V V V N P N −3 m ∂ e (P Q − P Q ) Θ = c Θ + O[s (ln t) ], t θ θ t 1 N+1 ϑ θ 0 2 1 2 (V ) − (V ) N −3 where c = O(s ) and we have used (177), Inductive assumption 61 and Lemma 65. The combination of the third and the seventh term on the right hand side of (152) can be written 1/2 λ/2 1 −∂ (s e Λ)V Θ . θ 1 In the present setting, the analogous term is N 1/2 λ/2 1 N −2 m −∂ [∂ (s e Λ)V Θ ] = c Θ + O[s (ln t) ], θ 1 N+1 ϑ θ N −2 where c = O(s ) and we have used Inductive assumption 61 and Lemma 65. Summing up, we obtain h  i dZ 1 1 α 1 N 1 N −1/2 1/2 λ/2 0 1/2 0 = − V + ∂ ∂ α λ − 2 − 4s e Λ V Θ − ∂ (α λ )V Θ t t 1 θ θ 1 N+1 ϑ ds 2s 4 α 4 2 2 3 2 1 (V ) − (V ) −1/2 1/2 0 − [∂ (α P ) − ∂ (α P )]V Θ t t θ θ 1 0 2 1 2 2 (V ) − (V ) 0 2 3 V V V −1/2 P 1/2 P −[∂ (α e Q ) − ∂ (α e Q )] Θ t t θ θ 1 0 2 1 2 (V ) − (V ) N N i −2 m + c Θ + c V + O[s (ln t) ], N+1 θ i N+1 (198) N −2 N −2 where c = O(s ), c = O(s ) and we sum over i but not over N. The term of importance is the second one on the right hand side. If all the ϑ-derivatives hit Θ , the resulting term can be N N −3/2 dealt with as in the proof of Lemma 56, and we obtain a c Θ -term, where c = O(s ). In N+1 θ θ the case of all the remaining terms, it is possible to use the equations (as in the proof of Lemma 56) −3/2 m in order to obtain terms of the form O[s (ln t) ]. Let us go through the argument in detail for 2 2 3 2 1 (V ) − (V ) N −1/2 1/2 0 ∂ − [∂ (α P ) − ∂ (α P )]V Θ . t t θ θ 1 0 2 1 2 2 (V ) − (V ) Using (61), this expression can be written −1/2 λ/2−P 2 −1/2 λ/2+P 2 1 1 α e J α e (K − QJ) N −1/2 −1/2 2P 2 2 − ∂ − α P + α e (Q − αQ ) + − ϑ t θ 7/2 7/2 2 s 2s 2s 2 2 3 2 (V ) − (V ) −1/2 λ/2 −1/2 0 N −2 m +s e α (P − P ) V Θ = c Θ + O[s (ln t) ], 2 3 1 N+1 0 2 1 2 (V ) − (V ) −2 where c = O(s ) and we have used (135), (177), Inductive assumption 61 and Lemma 65. Due to this argument, and similar ones for the remaining terms in (198), we obtain dZ 1 N 1 N N i −3/2 m = − V + c Θ + c V + O[s (ln t) ], N+1 N+1 θ i N+1 ds 2s N −3/2 N −3/2 where c = O(s ), c = O(s ) and we sum over i but not over N. Due to (186) and (187), θ i we conclude that dZ 1 N 1 N N i −3/2 m = − Z + c Ψ + c Z + O[s (ln t) ], N θ i N ds 2s N −3/2 N −3/2 where c = O(s ), c = O(s ) and we sum over i but not over N. Combining this equation with (190), (191) and (192) yields dΨ 2 N N −1/2 2 1 −2 2 m ˆ ˆ ˆ = Ψ + c Ψ + c s (ln s) Z + O[s (ln s) (ln t) ], N N θ,θ θ,1 N ds s ln s dZ N N 1/2 −2 N 1 −1 m ˆ ˆ = c s (ln s) Ψ + c Z + O[s (ln t) ], 1,θ N 1,1 N ds 46 N −2 N −3/2 N −3/2 N −3/2 where c = O(s ), c = O(s ), c = O(s ), c = O(s ), there is no summation θ,θ θ,1 1,θ 1,1 over N and we have used the notation 1 1/2 1 2 ˆ ˆ Z (s; t, θ, v) = s Z (s; t, θ, v), Ψ (s; t, θ, v) = (ln s) Ψ (s; t, θ, v). (199) N N N N Introducing the energy 2 1 2 ˆ ˆ ˆ E = (Ψ ) + (Z ) , (200) N N we conclude that dE C N N 1/2 −1 m ˆ ˆ ≥ − E − C s (ln t) E . N N ds s(ln s) Letting r be such that r (t ) = 0 and its derivative is the first factor in the first term on the N N 1 right hand side, we obtain dE N 1/2 −1 m ≥ −C s (ln t) E , ds 1/2 where E = exp(−r )E . Dividing by E and integrating from s to t, we obtain N N N 1/2 1/2 E (s; t, θ, v) ≤ E (t; t, θ, v) + C (ln t) . N N However, the first term on the right hand side can be estimated by C (ln t) . Combining the resulting estimate with (186), (187), (191) and (192), we conclude that (163) holds with j = N + 1. Corollary 68. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Let 0 ≤ k ∈ Z. Then there is a constant C , depending only on 1 0 k k and the solution, such that 1/2 1/2 −2 kP k k + kQ k k + t kα λ k k ≤ C t (201) t t θ k C C C for all t ≥ t . Moreover, α 3 3 −2 + + λ + ≤ C t t k α t t k k C C for all t ≥ t . Proof. By combining Lemmas 64 and 67, we know that Inductive assumption 61 holds for all N. In particular, we thus know that the conclusions of Lemma 65 hold for all N. Combining this information with (61) and (68) yields −1/2 −1/2 −1/2 k∂ (tα P )k k + k∂ (tα Q )k k ≤ C t . t t t t k C C As a consequence, −1/2 −1/2 1/2 ktα P k k + ktα Q k k ≤ C t . t t k C C Due to this estimate, as well as (135) and (179), we can proceed inductively in order to conclude that −2 kP k k + kQ k k ≤ C t . t t k C C Combining this estimate with (63), (135), (177) and Lemma 65, we obtain (201). Combining (201) with (60), (69) and (128), we obtain the final estimate stated in the corollary. 47 9 Energy estimates for the distribution function In the proof of the existence of f , cf. Theorem 7, a natural first step is to estimate L -based sc,∞ energies for f. In the process of deriving such estimates, it is useful to consider equations for the derivatives of the distribution function. Such equations take the following general form: 1/2 1 ∂h α v ∂h 1 ∂h + − v = R. (202) 0 i ∂t v ∂θ 2t ∂v In case h = f, R is given by ∂f R = L , (203) ∂v where 2 3 3 2 2 2 1 1 2α 3 v v 1 (v ) − (v ) 1 1/2 0 1 1/2 P 1/2 L = α λ v + λ − − v − α e Q + α P θ t θ θ 0 0 4 4 α t v 2 v −7/4 λ/4 −P/2 2 P/2 3 −t e e Jv + e (K − QJ)v , (204) 1 2 1 1 v v 2 2 1/2 L = P v + α P , (205) t θ 2 2 v 1 3 1 1 1 v v v 3 3 1/2 P 2 1/2 L = − P v − α P + e v Q + α Q . (206) t θ t θ 0 0 2 2 v v The energies we shall consider are Z Z −|β| l β 2 −1/2 −3/2 E [h](t) = t |∂ ∂ h(t, θ, v)| α t dvdθ. (207) θ v 1 3 S R l+|β|≤k We shall also use the notation E = E . −1/2 −3/2 Remarks 69. The purpose of the factor α t is to simplify some of the terms that result upon carrying out partial integrations. We could equally well consider energies of the form Z Z −|β| 1/2 2µ +2|β| l β 2 H [f](t) = t ht vi |∂ ∂ f(t, θ, v)| dvdθ θ v 1 3 S R l+|β|≤k for µ ≥ 0; cf. [31]. However, there is a constant C > 1, depending only on the solution, µ and β, such that −1 1/2 2µ +2|β| C ≤ ht vi ≤ C for t ≥ t (where t is defined as in the statement of previous lemmas) and (t, θ, v) in the support 1 1 of f. As a consequence, the corresponding weight is of no practical importance. Lemma 70. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Let h be a smooth solution to (202) (where the function α is 1 0 the object appearing in the Einstein-Vlasov equations and R is some function) which has compact support when restricted to compact time intervals. Then there is a constant C > 0, depending only on the solution to the Einstein-Vlasov equations, such that Z Z dE[h] 3 −1/2 −3/2 −2 ≤ − E[h] + 2 hRα t dvdθ + Ct E[h] dt 2t 1 3 S R for all t ≥ t . Remark 71. It is important to note that the constant C does not depend on h. Moreover, R should be thought of as being defined by (202). In particular, due to the assumptions concerning h, the function R is smooth and has compact support when restricted to compact time intervals. 48 Proof. Differentiating E with respect to time, we obtain Z Z Z Z dE 3 α −1/2 −3/2 2 −1/2 −3/2 = 2 h∂ hα t dvdθ + h − − α t dvdθ. (208) dt 1 3 1 3 2t 2α S R S R Due to (128), we can estimate the second term on the right hand side. Consider the first term on the right hand side of (208). Using (202), it can be written Z Z 1/2 1 α v ∂h 1 ∂h i −1/2 −3/2 2 h − + v + R α t dvdθ. 0 i 1 3 v ∂θ 2t ∂v S R The term involving ∂ h can be integrated to zero. The term involving R we leave as it is. What remains is to estimate the term Z Z Z Z 1 ∂h 3 i −1/2 −3/2 2 −1/2 −3/2 v α t dvdθ = − h α t dvdθ. 2t 1 3 ∂v 2t 1 3 S R S R The lemma follows. Let us turn to the higher order derivatives of the distribution function. Lemma 72. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Fix 0 ≤ k ∈ Z. Then there is a constant C > 0, depending 1 0 k only on k and the solution to the Einstein-Vlasov equations, such that dE [f] 3 −3/2 ≤ − E [f] + C t E [f] k k k dt 2t 3/2 for all t ≥ t . In particular, t E [f] is bounded to the future. 1 k Proof. Differentiating (202) with h = f, we obtain 1/2 1 1/2 1 ∂f α v ∂f 1 ∂f α v 1 β,l β,l β,l i β l β l i β l + − v = ∂ ∂ R + ∂ , ∂ ∂ f − v ∂ i, ∂ ∂ f, (209) v θ v θ v v θ 0 i 0 ∂t v ∂θ 2t ∂v v 2t β l where we use the notation f = ∂ ∂ f and assume that |β| +l ≤ k. Let us denote the right hand β,l v θ side of (209) by R . Due to Lemma 70, it is of interest to estimate β,l Z Z −|β| −1/2 −3/2 2 t f R α t dvdθ. (210) β,l β,l 1 3 S R By an inductive argument, it can be proven that the third term on the right hand side of (209) is given by |β|f /2t. The corresponding contribution to (210) is thus β,l |β| −|β| t E[f ]. β,l Turning to the second term on the right hand side of (209), it can (up to numerical factors) be written as a sum of terms of the form 1/2 1 α v β l β l +1 1 1 2 2 ∂ ∂ ∂ ∂ f, v v θ θ where β +β = β, l +l = l and |β | +l ≥ 1. Note that the first factor can always be estimated 1 2 1 2 1 1 −3/2 −2 by C t . In case β = 0, it can be estimated by C t (on the support of f). Due to these k 1 k observations, we have 1/2 1 X α v −|β|/2 β l −2 −|β |/2 t ∂ , ∂ ∂ f ≤ C t t |f |. θ k β ,l v θ 1 1 l +|β |≤k 1 1 49 The corresponding contribution to (210) can thus be estimated by −2 C t E [f]. k k Finally, let us consider the first term on the right hand side of (209). Since R is given by (203), β l the expression ∂ ∂ R is given by the sum of v θ i β l L ∂ i∂ ∂ f (211) v v θ and terms which (up to numerical factors) can be written l l β 1 i β 2 1 2 (∂ ∂ L )∂ i∂ ∂ f, (212) v v v θ θ where |β | + l ≥ 1. The contribution to (210) from (211) can be written 1 1 Z Z −|β| i 2 −1/2 −3/2 t L (∂ if )α t dvdθ. v β,l 1 3 S R i i Integrating partially with respect to v and keeping in mind that the L are given by (204)–(206), we conclude that this expression can be estimated by −3/2 C t E [f], (213) k k where we have used Lemma 65 and Corollary 68. Let us now consider the contribution arising from terms of the form (212). It is natural to divide these terms into two different categories. Either β = 0, or β 6= 0. In case β = 0, the expression (212) can be estimated by 1 1 1 −3/2 −1/2 β 2 C t t |∂ i∂ ∂ f|. v v In case β 6= 0, the expression (212) can be estimated by −3/2 β l 2 2 C t |∂ i∂ ∂ f|. k v In order to obtain these estimates, we have appealed to Lemma 65 and Corollary 68. As a consequence, the contribution to (210) from terms of the form (212) can be estimated by (213). Adding up the above observations, we conclude that d |β| dE[f ] |β| 3 β,l −|β| −|β| −|β| −|β| −|β| t E[f ] = − t E[f ] + t ≤ − t E[f ] − t E[f ] β,l β,l β,l β,l dt t dt t 2t Z Z −|β| −1/2 −3/2 −2 −|β| + 2 t R f α t dvdθ + Ct t E[f ] β,l β,l β,l 1 3 S R −|β| −3/2 ≤ − t E[f ] + C t E [f]. β,l k k 2t Summing over β and l, we obtain dE [f] 3 −3/2 ≤ − E [f] + C t E [f]. k k k dt 2t The lemma follows. In order to obtain a better understanding for the asymptotics, it is convenient to rescale the distribution function according to −1/2 f (t, θ, v) = f(t, θ, t v). sc We have the following conclusions concerning f . sc 50 2 Lemma 73. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Fix 0 ≤ k ∈ Z. Then there is a constant C, depending only on 1 0 1 3 the solution, such that in order for (t, θ, v) ∈ [t , ∞) × S × R to be in the support of f , v has 1 sc to satisfy |v| ≤ C. Moreover, there is a constant C > 0, depending only on k and the solution to the Einstein-Vlasov equations, such that −2 k∂ f (t, ·)k k 1 3 ≤ C t t sc k C (S ×R ) for all t ≥ t . In particular, there is thus a smooth, non-negative function with compact support, 1 3 say f , on S × R such that sc,∞ −1 kf (t, ·) − f k k 1 3 ≤ C t sc sc,∞ k C (S ×R ) for all t ≥ t . Proof. The statement concerning the support is an immediate consequence of Lemmas 50 and 53. In order to derive the desired estimates, let us compute −1/2 −1/2 i −1/2 (∂ f )(t, θ, v) = (∂ f)(t, θ, t v) − t v (∂ if)(t, θ, t v). t sc t v 2t Using the Vlasov equation, we conclude that 1/2 −1/2 1 α t v ∂ f = − ∂ f + R , t sc θ sc sc −1/2 ht vi where i −1/2 −1/2 1/2 i −1/2 R (t, θ, v) = L (t, θ, t v)(∂ if)(t, θ, t v) = t L (t, θ, t v)(∂ if )(t, θ, v) sc v v sc and the L are defined in (204)–(206). Introducing i 1/2 i −1/2 L (t, θ, v) = t L (t, θ, t v), sc we thus have 1/2 −1/2 1 α t v ∂ f = − ∂ f + L ∂ if . (214) t sc θ sc v sc sc −1/2 ht vi Due to the properties of the support of f , Lemma 65 and Corollary 68, there is a constant C sc k for each 0 ≤ k ∈ Z such that β l i −2 |(∂ ∂ L )(t, θ, v)| ≤ C t (215) v θ sc l+|β|≤k 1 3 for all i = 1, 2, 3 and (t, θ, v) ∈ [t , ∞) × S × R in the support of f . In order to estimate the 1 sc k −3/2 derivatives of f in C , it is convenient to translate the estimate E ≤ C t into an estimate sc k k for f . However, sc Z Z Z Z l β 2 −|β| l β −1/2 2 |(∂ ∂ f )(t, θ, v)| dθdv = t |(∂ ∂ f)(t, θ, t v)| dθdv sc θ v θ v 1 3 1 3 S R S R Z Z 3/2 −|β| l β 2 3/2 =t t |(∂ ∂ f)(t, θ, v)| dθdv ≤ Ct E (t) ≤ C , k k θ v 1 3 S R assuming l+|β| ≤ k. Due to this estimate and Sobolev embedding, we conclude that all derivatives of f are bounded for t ≥ t . Combining this observation with (214) and (215), we conclude that sc 1 β l −2 |∂ ∂ ∂ f | ≤ C t t sc k v θ l+|β|≤k for t ≥ t . The lemma follows. 51 10 Proof of the main theorems Finally, we are in a position to prove the main theorems. Let us begin with Theorem 7. Theorem 7. To begin with, the conclusions concerning the distribution function are direct conse- quences of Lemma 73. Turning to H and G, we have −5/2 −1/2 P +λ/2 −5/2 −1/2 −P +λ/2 H = −t α e (K − QJ), G = −QH − t α e J; (216) t t t cf. (58). In order to estimate H and G , it is useful to note that P and Q are bounded in every t t C -norm for t ≥ t ; this follows by integrating (201). Combining this observation with (66), (67), (88) and Lemma 65 yields −2 m kJ k N + kK k N ≤ C t (ln t) . t t N C C Thus J and K are uniformly bounded in C . Combining this observation with (216), (135), (88), the bound on P and Q in every C norm and Lemma 65, we conclude that (8) holds. Due to Corollary 68, we know that (9) and (10) hold. Combining (10) with (88) and (135) yields (11). Let us turn to the second fundamental form. By definition, 1/4 −λ/4 1/4 −λ/4 1/4 −λ/4 ¯ ¯ k = k(∂ , ∂ ) = h∇ e , ∂ i = h∇ (t e ∂ ), ∂ i = t e h∇ ∂ , ∂ i = t e ∂ g , ij i j ∂ 0 j ∂ t j ∂ t j t ij i i i where we have used the fact that ∂ and ∂ are perpendicular. In what follows, we would like to t i prove that kk − Hg¯ k N ≤ C , (217) ij ij N 1/2 where H = (Λ/3) . Consider the spatial components of the metric (1). If a time derivative hits one of P , Q, G or H in such a component, then the resulting expression is bounded in C after 1/4 −λ/4 it has been multiplied by t e /2; this is due to (8), (9) and (11). As a consequence, what we need to consider are the components of the tensor field 1 1 1 α 1/4 −λ/4 −1/2 λ/2 −1 2 P 2 t e − + λ − t e α dθ + e [dx + Qdy + (G + QH)dθ] 2 2t 2 α 1 1 1 3 α 3 −P 2 −3/4 −λ/4 −1/4 −1 λ/4 2 + e (dy + Hdθ) = t e g¯ + t α e λ + − + dθ , 2 2 2 t α t where g¯ is the spatial part of the metric. Note that the components of the second term on the right hand side are bounded in C ; this is a consequence of (10) and (11). Moreover, −3/4 −λ/4 −λ/4 t e = He , where λ is defined in (84). In order to prove (217), it is thus sufficient to demonstrate that −λ/4 k(e − 1)g¯ k N ≤ C . ij N −λ/4 N −1 N However, t(e − 1) is bounded in C due to (11) and t g¯ is bounded in C due to (8), (9) ij and (11). Thus (217) holds. Let us define g¯ by (13); note that this is a smooth Riemannian metric on T . Moreover, −1 −1 kt g¯ (t, ·) − g¯ k N ≤ C t (218) ij ∞,ij N due to (8), (9) and (11). Combining this estimate with (217), we obtain (12). The proof of future causal geodesic completeness is not very complicated, given the above estimates. It is, e.g., possible to proceed as in the proof of [28, Propositions 3 and 4, pp. 189–191]. However, we shall not write down the details, since the result follows from the proof of Theorem 35. Let us now turn to the proof of the cosmic no-hair conjecture. 52 Theorem 14. We need to verify that the conditions stated in Definition 8 are fulfilled. To begin with, note that Σ = {t} × T is a Cauchy hypersurface for each t ∈ (t , ∞). An argument is t 0 required in order to justify this statement, but since the details are quite standard (cf., e.g., the proof of [30, Proposition 20.3, p. 215], in particular [30, p. 217]), we have omitted the details. Let γ = (γ , γ¯) be a future directed and inextendible causal curve, defined on I = (s , s ). γ − + Reparametrising the curve, if necessary, we can assume that γ (s) = s and that I = (t , ∞). Due γ 0 to the causality of the curve, we know that i j −2 ˙ ˙ g¯ [γ(t)]γ¯ (t)γ¯ (t) ≤ −g [γ(t)] ≤ Ct ij 00 for t ≥ t , where t = t + 2 and we have used (11). Combining this estimate with (12), we 1 1 0 conclude that there is a constant K > 1 (independent of the curve γ, as long as γ (t) = t) such that i j 2 −2 −3 ˙ ˙ g¯ [γ¯(t)]γ¯ (t)γ¯ (t) ≤ K H t ∞,ij 3 −1 −1/2 for all t ≥ t . In particular, there is thus an x¯ ∈ T such that d [γ¯(t), x¯ ] ≤ K H t for all 1 0 ∞ 0 0 t ≥ t , where d is the topological metric on T induced by g¯ . Let ǫ > 0 denote the injectivity 1 ∞ ∞ inj radius of (T , g¯ ); the injectivity radius of a point p of a Riemannian manifold, denoted inj(p), is defined in [20, Definition 9.2, p. 142], and the injectivity radius of a Riemannian manifold is the infimum of the injectivity radii of the points of the manifold; that ǫ > 0 follows from the inj continuity of inj, cf. [20, p. 178]; readers interested in a more quantitative bound on the injectivity radius are referred to [20, Lemma 51, p. 319]. Then, given x¯ ∈ T , there are normal coordinates on B (x¯), where distances are computed using d ; cf. [19, pp. 72–73] for a definition of normal ǫ ∞ inj 2 −2 −2 coordinates. Fix t > K H ǫ +1 (note that t is independent of the curve). Due to the above − − inj arguments and definitions, − + 3 −1 −1/2 J (γ) ∩ J (Σ ) ⊆ {(t, x¯) ∈ I × T : t ≥ t , d (x¯, x¯ ) ≤ K H t } (219) t − ∞ 0 0 −1/2 −1 and the closed ball of radius K H t (with respect to d ) and centre x¯ is contained in the 0 ∞ 0 domain of definition of normal coordinates ¯x with centre at x¯ . Denote the set appearing on the right hand side of (219) by D . Let us define t ,K ,x ¯ − 0 0 2Hτ −1 ¯ ¯ ψ(τ, ξ) = [e , ¯x (ξ)]. Then −1 3 −1 −Hτ ¯ ¯ ψ (D ) = {(τ, ξ) ∈ J × R : τ ≥ T , |ξ| ≤ K H e }, t ,K ,x ¯ 0 0 − 0 0 −1 −1 where T = H ln t /2, and J = (τ , ∞), where τ = H ln t /2; if t = 0, then τ = −∞. 0 − 0 0 0 0 0 Letting T be slightly smaller than T and K be slightly larger than K , the map ψ is still defined 0 0 −1 −Hτ on C ; cf. (15). In analogy with Definition 8, let D = ψ(C ) and R(τ) = KH e . Λ,K,T Λ,K,T Due to the above arguments, we have already verified all of the requirements of Definition 8 (with Σ = Σ etc.) but the last one; i.e., that (16) hold. In order to proceed, let ¯ g denote the components of g¯ with respect to the coordinates ¯x. Let ∞,ij ∞ 2Hτ 2Hτ ¯ ¯ g (τ, ·) and k (τ, ·) denote the components of g¯(e , ·) and k(e , ·), respectively, with respect ij ij ¯ ¯ ¯ to the coordinates x. Moreover, consider g , g (τ, ·) and k (τ, ·) to be functions on the image ∞,ij ij ij of ¯x; i.e., on B (0), with the origin corresponding to x¯ . Note that the estimates (217) and (218) ǫ 0 inj hold with g¯ replaced by g¯ etc., assuming the domain on which the C norm is computed is ij ij suitably restricted. In particular, letting S be as in Definition 8, the following estimate holds: −2Hτ −2Hτ −2Hτ ke k (τ, ·) − H¯ g k N + ke g¯ (τ, ·) − g¯ k N ≤ C e ij ∞,ij ij ∞,ij N C (S ) C (S ) τ τ for all τ ≥ T . Note that g¯ (0) = δ , (∂ g¯ )(0) = 0 ∞,ij ij l ∞,ij by the definition of the coordinates x. As a consequence, if ξ ∈ S , then −Hτ ¯ ¯ |(∂ g¯ )(ξ)| = [(∂ g¯ )(sξ)]ds ≤ Ce . l ∞,ij l ∞,ij ds 53 Moreover, −2Hτ |g¯ (ξ) − δ | ≤ Ce ∞,ij ij for τ ≥ T and ξ ∈ S . In particular, we thus have −2Hτ −2Hτ −2Hτ ke k (t, ·) − Hδ k 0 + ke ¯ g (t, ·) − δ k 0 ≤ Ce ij ij C (S ) ij ij C (S ) τ τ for all τ ≥ T . Letting g¯ (τ, ·) and k (τ, ·) be defined as in Definition 8, we conclude, in particular, dS dS that −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k 0 + kk (τ, ·) − k(τ, ·)k 0 ≤ Ce dS dS C (S ) C (S ) τ τ dS dS for all τ ≥ T . In fact, due to the above estimates, we have −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k N + kk (τ, ·) − k(τ, ·)k N ≤ C e dS dS N C (S ) C (S ) τ τ dS dS for all τ ≥ T . The theorem follows. Finally, we are in a position to prove Theorem 35. Theorem 35. The idea of the proof is to demonstrate that for late enough t, there is a neighbour- hood of each point in {t} × T such that Theorem 29 applies in the neighbourhood; combining this observation with Cauchy stability, cf. Theorem 34, then yields the desired result. Fixing N, there is an ǫ > 0 and a constant C such that for every x¯ ∈ T , there are normal coordinates ¯x on W = B (x¯) with respect to g¯ , where distances on T are measured using the ǫ ∞ topological metric induced by g¯ . Moreover, if g¯ are the components of g¯ with respect to ¯x ∞ ∞,ij ∞ ij ij and g¯ are the components of the inverse, then the derivatives of g¯ and g¯ up to order N with ∞,ij ∞ ∞ respect to the coordinates ¯x on W are bounded by C . Moreover, the derivatives of ¯x, considered −1 as functions of (θ, x, y), up to order N + 1 are bounded by C . Similarly, the derivatives of ¯x up to order N + 1 are bounded by C . The arguments required in order to prove the above statements are similar to those presented in the proof of [31, Lemma 34.9, p. 650]. The important point is that we obtain uniform bounds which hold regardless of the base point. K −K 1/2 Define K by the condition e = 4/H and define the coordinates ¯y = e t ¯x on W . Note that the range of ¯y is B −K 1/2 (0). For t large enough (the bound being independent of the base point e t ǫ −K 1/2 x¯), we consequently have e t ǫ > 1. From now on, we assume t to be large enough that this is −1 the case. Moreover, we assume the coordinates ¯y to be defined on the image of B (0) under ¯y . Let g¯ denote the components of g¯(t, ·) with respect to the coordinates ¯y. Moreover, let g¯ ij ∞,ij denote the components of g¯ with respect to the coordinates ¯x. Due to (218), we have α −2K −1 −1−|α|/2 |∂ [(e g¯ − g¯ ) ◦ ¯y ](ξ)| ≤ C t ¯ ij ∞,ij N for ξ ∈ B (0) and |α| ≤ N; note that −1 −1 K −1/2 ¯ ¯ ¯y (ξ) = ¯x (e t ξ). −1 Since ¯ g ◦ ¯y (0) = δ , ∞,ij ij −2K −1 −1/2 |e ¯ g ◦ ¯y − δ | ≤ C t (220) ij ij N on B (0); in particular, (38) holds with a margin for t large enough. Similarly, −1 −1 l −1 |∂ (g¯ ◦ ¯y )(u)| = ∂ ∂ (g¯ ◦ ¯y )(su)u ds ≤ Ct m ∞,ij l m ∞,ij l=1 on B (0). To conclude, we thus have −2K −1 −1 ke (∂ g¯ ◦ ¯y )k N−1 ≤ C t . (221) k ij N C [B (0)] 54 Due to (217), we also have −1 −1 ¯ ¯ k(k − Hg ) ◦ y k N ≤ C t , (222) ij ij C [B (0)] N ¯ ¯ where k denotes the components of the second fundamental form k calculated using the coor- ij dinates ¯y. In the end, we shall choose K = ln t/2. As a consequence, (221) and (222) imply Vl that (39) holds with a margin (note that the T -symmetric background solution is such that φ , i = 0, 1, vanish), assuming N ≥ 5; note that in order to prove Theorem 35, it is sufficient to apply Theorem 29 with k = 4. Let us turn to the distribution function. First of all, recall that if Σ is a spacelike hypersurface in a Lorentz manifold, and f is a distribution function defined on the mass shell, then the initial datum for the distribution function (denoted f and defined on T Σ) induced by f on Σ is given by −1 f = f ◦ pr , where pr is the projection from the mass shell over Σ to T Σ; in other words, if p ∈ P for some r ∈ Σ and N is the future directed unit normal to Σ at r, then pr (p) is the element of T Σ r r corresponding to p + g(p, N )N . In our case, we are interested in the hypersurface Σ = {t} × T . r r If z = (t, θ, x, y), i α f p¯ e | = f (p e | ) = f(t, θ, p¯), i α z z 1 2 3 where p¯ = (p¯ , p¯ , p¯ ), 0 1 2 2 2 3 2 1/2 p = [1 + (p¯ ) + (p¯ ) + (p¯ ) ] i i and p = p¯ . However, in the application of Theorem 29, we need to express f with respect to the coordinates ¯y. Consequently, we are interested in i i j ¯ ¯ ¯ f(z¯, p¯) = f p¯ ∂ i = f p¯ A (z¯) e | = f[t, θ, v(z¯, p¯)], ¯ y j z z where z = (t, z¯), i 1 i 2 i 3 v(z¯, p¯) = (p¯ A (z¯), p¯ A (z¯), p¯ A (z¯)) i i i and A is defined by the requirement that ∂ i = A (z¯) e | . ¯ y j z z Thus ∂¯z K −1/2 K −1/2 A (z¯) = ∂ i , e | = e t ∂ i| , e | = e t (z) ∂ l| , e | , j j j ¯ y ¯ x ¯ z i z z z z z z i ∂¯x where ¯z correspond to the standard coordinates on the torus (which are locally well defined). In 1 2 3 particular, ∂ = ∂ , ∂ = ∂ and ∂ = ∂ . Due to the observations made at the beginning of ¯ z θ ¯ z x ¯ z y the proof, (7) and Theorem 7, it is clear that all derivatives of A up to order N are uniformly bounded on the domain of ¯y, the bound being independent of the base point x¯ and the time t (assuming t to be sufficiently large). What we need to estimate is Z Z −w 2|β| w 2µ +2|β| α 2 ¯ ¯ ¯ (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯, ¯ p¯ y R ¯ y(U) |α|+|β|≤k cf. (37) and (40), where −1 −1 K −1/2 1 −1 K −1/2 −1 K −1/2 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ f (ξ, p¯) = f[¯y (ξ), p¯] = f[¯x (e t ξ), p¯] = f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)] −1 1 −1 and the constant w remains to be specified. Note that all derivatives of ¯x and ¯z ◦ ¯x up to order N are uniformly bounded. As a consequence, γ 1 −1 K −1/2 −1 K −1/2 −1/2 ¯ ¯ ¯ ¯ ¯ ∂ f (ξ, p¯) = ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]ψ (t ξ) ¯ y γ p¯ v |γ|=|β| 55 for functions ψ with bounded derivatives; note that ∂v −1 K −1/2 i −1 K −1/2 ¯ ¯ (¯x (e t ξ), p¯) = A ◦ ¯x (e t ξ). ∂p¯ ¯ ¯ As a consequence, ∂ ∂ f (ξ, p¯) consists of sums of terms of the form ¯ p¯ y −|α|/2 l γ+δ 1 −1 K −1/2 −1 K −1/2 −1/2 λ ¯ ¯ ¯ t ∂ ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]φ (t ξ)p¯ , γ,δ,l θ v λ 1 λ 2 λ 3 λ 1 2 3 where |λ| = |δ|, |γ| = |β|, l + |δ| ≤ |α|, φ are bounded functions and p¯ = (p¯ ) (p¯ ) (p¯ ) . γ,δ,l On the other hand, we know that 1/2 f(t, θ, v) = f (t, θ, t v), sc where f converges to a smooth function with compact support with respect to every every C - sc norm. Moreover, we know that f has uniformly compact support. Note also that sc l γ+δ (|γ|+|δ|)/2 l γ+δ 1/2 ∂ ∂ f(t, θ, v) = t ∂ ∂ f (t, θ, t v). sc θ v θ v Since there is a uniform constant C > 1 (independent of t (large enough) and the base point x¯) such that −1 −1 K −1/2 C |p¯| ≤ |v(¯x (e t ξ), p¯)| ≤ C|p¯|, we conclude that −|α|/2 l γ+δ 1 −1 K −1/2 −1 K −1/2 −1/2 λ ¯ ¯ ¯ t |∂ ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]φ (t ξ)p¯ | γ,δ,l θ v (|γ|−|α|)/2 1/2 ≤C t χ(t p¯), α,β where χ is a smooth function with compact support. As a consequence, α (|β|−|α|)/2 1/2 ¯ ¯ |∂ ∂ f (ξ, p¯)| ≤ C t χ(t p¯) ¯ ¯ y α,β p¯ for ξ ∈ B (0). Let us now define w = K + K , where K = ln t/2. Then 1 Vl Vl Z Z −w 2|β| w 2µ +2|β| α 2 ¯ ¯ ¯ (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯ ¯ p¯ ¯ y R ¯ y(U) Z Z 2 −2|β|K −|β| K 1/2 2µ +2|β| |β|−|α| 2 1/2 −|α|−3/2 ≤C e t he t p¯i t χ (t p¯)dξdp¯ ≤ C t . ,µ α,β α,β R B (0) The square root of the right hand side of this expression should be compared with the right hand side of (40): 2 5/2 −3K/2−K 2 5/2 −3K/2 −1/2 Vl H ε e = H ε e t . Clearly, we have a margin. As a consequence, for t large enough there is, for every x¯ ∈ T , a neighbourhood of x¯ such that (40) holds with a margin. Due to (220), (221) and (222), we also know that (38) and (39) hold with a margin. We can thus apply Theorem 29 with the given µ > 5/2 3 −1 and k = 4. In addition, the covering of T obtained by taking the neighbourhoods ¯y [B (0)] 0 1/4 corresponding to varying base points x¯ has a finite subcovering. Appealing to Cauchy stability, Theorem 34, we conclude that there is an ǫ > 0 with the properties stated in the theorem. Acknowledgements The second author would like to acknowledge the support of the G¨oran Gustafsson Foundation, the G¨oran Gustafsson Foundation for Research in Natural Sciences and Medicine, and the Swedish Research Council. This article was in part written during a stay of both authors at the Erwin Schr¨odinger Institute in Vienna during the programme ’Dynamics of General Relativity: Black Holes and Asymptotics’. 56 A Derivation of the equations The purpose of this appendix is to compute the Einstein tensor associated with the metric (1), and to derive an expression for the Vlasov equation. Let us begin by expressing g with respect to suitable one-form fields. Let 0 −1/4 λ/4 ξ = t e dt, 1 −1/4 λ/4 −1/2 ξ = t e α dθ, 2 1/2 P/2 ξ = t e (dx + Qdy + (G + QH)dθ), 3 1/2 −P/2 ξ = t e (dy + Hdθ). With respect to these one-form fields, the metric can be written 0 0 i i g = −ξ ⊗ ξ + ξ ⊗ ξ . i=1 α α Using the orthonormal frame {e } introduced in (7), it can be verified that ξ (e ) = δ . α β A.1 Commutators Let us compute the commutators, in other words the functions γ such that βζ [e , e ] = γ e . β ζ α βζ α α α Clearly, γ = −γ . Consequently, it is sufficient to compute γ for β < ζ. By a straightforward βζ ζβ βζ computation, 0 1/4 −λ/4 1/2 γ = t e α λ , (223) 1 α 1 1 1/4 −λ/4 γ = − t e λ − 2 − , (224) 4 α t 2 −3/2 −P/2 γ = t e J, (225) 3 −3/2 P/2 γ = t e (K − QJ), (226) where we have used the notation (58). Turning to γ , the only α for which this object is non-zero is α = 2, and we then have 2 1/4 −λ/4 −1 γ = − t e t + P . Turning to γ , the only α’s for which this object is non-zero is α = 2 and α = 3, and we then have 2 1/4 −λ/4 P γ = −t e e Q , 3 1/4 −λ/4 −1 γ = − t e t − P . Turning to γ , the only α for which this object is non-zero is α = 2, and we then have 2 1/4 −λ/4 1/2 γ = − t e α P . Turning to γ , the only α’s for which this object is non-zero is α = 2 and α = 3, and we then have 2 1/4 −λ/4 1/2 P γ = −t e α e Q , 3 1/4 −λ/4 1/2 γ = t e α P . 57 α Finally, γ = 0. For future reference, let us make the following observations: A −3/4 −λ/4 γ = −t e , 0A i 1 −3/4 −λ/4 γ = γ − t e , 0i 01 α 1 −3/4 −λ/4 γ = γ − t e , 0α 01 γ = 0, 1A γ = 0, 1i α 0 γ = −γ , 1α 01 where Greek indices range from 0 to 3, lower case Latin indices range from 1 to 3 and capital Latin indices range from 2 to 3; moreover, Einstein’s summation convention is enforced. Non-zero components. Note that in order for γ to be non-zero, one of the following conditions βζ has to be satisfied: • {β, ζ} = {0, 1}, • one of β and ζ is in {0, 1}; the other is in {2, 3}; and α is in {2, 3}. A.2 Connection coefficients Define the connection coefficients Γ by the relation βζ ∇ e = Γ e , e ζ α β βζ where ∇ is the Levi-Civita connection associated with the metric g. Note that 0 i Γ = −h∇ e , e i, Γ = h∇ e , e i, e ζ 0 e ζ i βζ β βζ β since the frame is orthonormal. Let us begin by recording some symmetries of this object. Symmetries of the connection coefficients. Due to the metricity of the connection and the fact that the basis is orthonormal, we have Γ = 0 (no summation over α). βα For similar reasons, Γ is anti-symmetric in the indices i and j. Moreover, since [e , e ] is per- i j αj pendicular to e , we have Γ = h∇ e , e i = −he , ∇ e i = −he , ∇ e i = h∇ e , e i = Γ . e 0 i 0 e i 0 e j e 0 j j0 j j i i i0 i i 0 Thus, Γ is symmetric in i and j. Note that the computation also demonstrates that Γ = Γ . j0 j0 ij Similarly, since [e , e ] = 0, we have A B A B Γ = h∇ e , e i = −he , ∇ e i = −he , ∇ e i = h∇ e , e i = Γ . eB 1 A 1 eB A 1 eA B eA 1 B B1 A1 A A 1 In particular, Γ is symmetric in the indices A and B, and Γ = −Γ . B1 B1 AB Connection coefficients including two or more zero indices. That Γ = 0 follows from α0 the above. Moreover, using the Koszul formula, it can be computed that 0 1 0 Γ = Γ = γ 01 00 01 0 A and that the remaining components satisfy Γ = Γ = 0. 0A 00 Connection coefficients including exactly one zero index. As already mentioned, Γ is 0j i 0 i i antisymmetric and Γ = Γ is symmetric. It is thus sufficient to compute Γ for i < j and Γ j0 ij 0j j0 for i ≤ j. We have 1 A Γ = − γ , 0A 01 2 2 Γ = γ . 03 03 58 Moreover, i i Γ = −γ (no summation over i), i0 0i 1 A Γ = − γ , A0 01 2 2 Γ = − γ . 30 03 Connection coefficients including no zero index. Note that, due to the Koszul formula and the properties of the commutators, the only Γ ’s which are non-zero are the ones that have one jk B 1 A index equalling 1 and two indices in {2, 3}. Moreover, since Γ = −Γ is symmetric and Γ A1 AB 1B is antisymmetric, it is sufficient to calculate that 2 2 Γ = γ , 13 13 A A Γ = −γ (no summation over A), A1 1A 2 2 Γ = − γ . 31 13 For future reference, it is of interest to note that α 0 Γ = γ , (227) α1 01 Γ = 0, (228) i1 α 1 −3/4 −λ/4 Γ = −γ + t e , (229) α0 01 Γ = 0. (230) αA A.3 Twist quantities The quantities J and K have been defined in two different ways; one definition is given in (2) and another is given in (58). In the present subsection, we wish to verify that these two definitions yield the same result. In the proof of this statement, it is useful to introduce different notation for the different definitions. Let us therefore denote the J and the K defined in (2) by J and tw K respectively. The quantities defined in (58) we, however, still refer to as J and K. Due to tw the fact that we have calculated the connection coefficients using the orthonormal frame {e }, it is convenient to carry out the computations relative to this frame. Note, to this end, that 1/2 P/2 X = t e e , 1/2 −P/2 1/2 P/2 Y = t e e + t e Qe , 3 2 α α where X = ∂ and Y = ∂ . In particular, if X and Y denote the components of X and Y with x y respect to the frame {e }, then 2 1/2 P/2 α • X = t e and X = 0 for α 6= 2, 2 1/2 P/2 3 1/2 −P/2 0 1 • Y = t e Q, Y = t e and Y = Y = 0. Consequently, α β ζ δ 2 3 ζ δ 0 1 1 0 1 0 J = ǫ X Y ∇ X = ǫ X Y ∇ X = t(∇ X − ∇ X ) = −t(∇ X + ∇ X ), (231) tw αβζδ 23ζδ 0 1 where the indices are frame indices, and we assume the orientation of M to be such that ǫ = 1. Similarly, 1 0 K = −t(∇ Y + ∇ Y ). (232) tw 0 1 59 1 What remains is thus to calculate ∇ X etc. However, β β β ζ ζ β ζ ∇ X = ξ (∇ X) = ξ [e (X )e + X ∇ e ] = e (X ) + X Γ . α e α ζ e ζ α α α αζ The calculation for Y is the same. In particular, 1 0 ζ 1 0 2 1 0 2 2 ∇ X + ∇ X = X (Γ + Γ ) = X (Γ + Γ ) = −X γ . 0 1 0ζ 1ζ 02 12 01 Combining this calculation with (225) and (231) yields 2 2 3/2 P/2 −3/2 −P/2 J = tX γ = t e t e J = J. tw Thus J = J . Next, let us calculate tw 1 0 ζ 1 0 2 1 0 3 1 0 2 2 3 3 ∇ Y + ∇ Y = Y (Γ + Γ ) = Y (Γ + Γ ) + Y (Γ + Γ ) = −Y γ − Y γ . 0 1 0ζ 1ζ 02 12 03 13 01 01 Combining this observation with (225), (226) and (232) yields 2 2 3 3 3/2 P/2 −3/2 −P/2 3/2 −P/2 −3/2 P/2 K = tY γ + tY γ = t e Qt e J + t e t e (K − QJ) = K. tw 01 01 To conclude, we thus have J = J and K = K. tw tw A.4 Auxiliary computations In order to simplify future calculations, let us make some observations concerning the derivatives of the connection coefficients. Note, to begin with, that all γ (no summation) can be written 0i i 1/4 −λ/4 γ = −t e f 0i for suitably chosen functions f . Thus, it can be computed that 1 1 3 i 1/4 −λ/4 −1/2 −e (γ ) = − t e ∂ − exp − λ + ln α − ln t tα f 0 t i 0i 4 2 4 1 α 1 1/4 −λ/4 i 1/4 −λ/4 −1 i −1/2 −λ/2 1/2 −1/2 = t e λ − 2 − γ + t e t γ + t e α ∂ (tα f ), t t i 0i 0i 4 α t i 1 where we do not sum over i. However, the factor in front of γ in the first term is given by −γ 0i i 2 3 and the factor in front of γ in the second term is given by −γ − γ . Consequently, 02 03 0i 0 α 0 −1/2 −λ/2 1/2 −1/2 e (Γ ) + Γ Γ = t e α ∂ (tα f ), 0 t i ii α0 ii where we sum over α but not over i. In particular, we thus have 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (λ − 2α α − t )], (233) 0 t t t 11 α0 11 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (t + P )], (234) 0 t t 22 α0 22 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (t − P )]. (235) 0 t t 33 α0 33 0 2 3 1/4 −λ/4 Next, note that γ , γ and γ all can be written as h = t e f for suitably chosen functions i i 01 12 13 f . Moreover, 1/4 −λ/4 1/2 1/4 −λ/4 1/2 −λ/2 1/2 0 e (h ) = t e α ∂ (t e f ) = t e α ∂ f − γ h . 1 i θ i θ i i α 0 Since Γ = γ , we conclude that α1 01 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα λ ), (236) 1 θ θ 00 α1 00 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = − t e α ∂ (tα P ), (237) 1 θ θ 22 α1 22 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα P ). (238) 1 33 α1 33 θ θ 60 2 2 The expressions γ and γ require a somewhat different treatment. However, similar arguments 03 13 yield 1 1 1 1 α −1/2 −λ/2−P 1/2 1/2 2P 3 2 2 e (Γ ) + Γ Γ = − t e α ∂ (tα e Q ) − (γ − γ )γ , (239) 1 θ θ 23 23 α1 13 12 13 2 2 1 1 0 0 α −1/2 −λ/2−P 1/2 −1/2 2P 3 2 2 e (Γ ) + Γ Γ = t e α ∂ (tα e Q ) + (γ − γ )γ . (240) 0 t t 23 23 α0 03 02 03 2 2 A.5 Ricci curvature The Ricci curvature is given by X X Ric(e , e ) = ǫ hR e , e i = ǫ h∇ ∇ e − ∇ ∇ e − ∇ e , e i β ζ α e e ζ α α e e ζ e e ζ ζ α α β α β β α [e ,e ] α β α α δ δ δ λ = ǫ h∇ (Γ e ) − ∇ (Γ e ) − γ Γ e , e i α e δ e δ λ α α βζ β αζ αβ δζ δ δ λ δ δ λ δ λ = ǫ he (Γ )e + Γ Γ e − e (Γ )e − Γ Γ e − γ Γ e , e i α α δ λ β δ λ λ α βζ βζ αδ αζ αζ βδ αβ δζ α δ α α δ α δ α =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ , α β βζ βζ αδ αζ αζ βδ αβ δζ where ǫ = −1 and ǫ = 1. Let us begin by computing the 00 component: 0 i α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 0 α 0 00 00 αδ α0 α0 0δ α0 δ0 β β 1 1 α α α α =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ . 1 0 00 00 α1 α0 α0 0β α0 β0 In order to simplify the expression, let us note that 1 1 α −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα λ ), 1 θ θ 00 00 α1 where we have used (236). Moreover, due to (229), α 1 −3/4 −λ/4 1 −1/2 −λ/2 −1 −e (Γ ) = −e (−γ + t e ) = e (γ ) + t e (λ + 3t ). 0 0 0 t α0 01 01 Compute, using the symmetries of the connection coefficients, β α 0 i i 0 j i 1 0 0 2 −Γ Γ = −Γ Γ − Γ Γ − Γ Γ = −Γ Γ = −(γ ) . α0 0β i0 00 00 0i i0 0j 00 01 01 Finally, β β j α i 0 i i 0 2 1 1 A 1 B A −γ Γ = − γ Γ = γ Γ + γ Γ = (γ ) + γ Γ + γ Γ + γ Γ α0 β0 i0 β0 0i 00 0i j0 01 01 10 01 A0 0A B0 0 2 1 2 A 1 B A =(γ ) − (γ ) + γ Γ + γ Γ 01 01 01 A0 0A B0 1 1 1 1 0 2 1 2 2 2 3 2 2 2 3 2 2 2 2 3 =(γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − γ γ . 01 01 01 01 02 03 03 03 02 2 2 2 2 Since γ = 0, we obtain 1 1 −1/2 −λ/2 1/2 1/2 −1/2 −λ/2 −1 1 Ric(e , e ) = t e α ∂ (tα λ ) + t e (λ + 3t ) + e (γ ) 0 0 θ θ t 0 4 4 1 1 1 1 2 2 2 3 2 2 2 3 2 2 2 − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) . 01 01 02 03 03 2 2 2 In order to simplify this expression, it is useful combine (229) and (233) in order to conclude 1 1 1 1 2 −1/2 −λ/2 1/2 −1/2 −1 −1 −1/2 −λ/2 −1 −1 e (γ )−(γ ) = − t e α ∂ [tα (λ −2α α −t )]+ t e (λ −2α α −t ). 0 01 01 t t t t t 4 4 61 Let us now compute 1 1 1 2 2 3 2 2 2 3 2 2 2 − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) 01 01 02 03 03 2 2 2 λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 2 2P 2 −3/2 −λ/2 = − t e P + e Q + + − t e . t t 7/2 7/2 2 2 t t Thus −1/2 −λ/2 1/2 1/2 −1/2 −1 −1 Ric(e , e ) = t e α ∂ (tα λ ) − ∂ [tα (λ − 2α α − t )] 0 0 θ θ t t t λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2P 2 − t e P + e Q + + t t 7/2 7/2 2 t t 1 α −1/2 −λ/2 + t e λ − . 2 α Let us compute, using (227), (230) and the antisymmetry of Γ in i and j, that αj α β α α β α β α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 1 α 0 01 01 αβ α1 α1 0β α0 β1 β β 0 0 α 0 α α =e (Γ ) + Γ Γ − e (γ ) − Γ Γ − γ Γ 0 0 01 01 α0 01 0β β1 α1 α0 0 α β α β α =Γ Γ − Γ Γ − γ Γ . 01 α0 α1 0β α0 β1 Compute, using (229), 0 α 0 1 −3/4 −λ/4 0 1 0 −3/4 −λ/4 Γ Γ = γ (−γ + t e ) = −γ γ + γ t e . 01 α0 01 01 01 01 01 Moreover, using the symmetries of the connection coefficients, β α 0 1 i j 0 1 −Γ Γ = −Γ Γ − Γ Γ = γ γ . α1 0β 11 00 j1 0i 01 01 Thus 0 α α 0 −3/4 −λ/4 −1/2 −λ/2 1/2 Γ Γ − Γ Γ = γ t e = t e α λ . 01 α0 α1 0β 01 Finally, β β j α i 0 1 i 2 2 2 3 3 2 3 3 −γ Γ =γ Γ = γ Γ + γ Γ = γ Γ + γ Γ + γ Γ + γ Γ α0 β1 β1 01 01 j1 02 21 03 21 02 31 03 31 0i 0i 1 1 2 2 2 2 3 3 1/2 −λ/2 1/2 2P = − γ γ − γ γ − γ γ = − t e α (P P + e Q Q ). t θ t θ 02 12 03 13 03 13 2 2 Thus −1/2 −λ/2 1/2 2P Ric(e , e ) = t e α [λ − 2t(P P + e Q Q )]. 0 1 θ t θ t θ Let us compute, using (230), β β β α α α α α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 A α 0 0A αβ αA 0β βA 0A αA α0 j j A 1 α 0 1 1 0 i 0 1 i = − e (γ ) + Γ Γ − Γ Γ − Γ Γ − Γ Γ + γ Γ + γ Γ 01 0A α1 1A 00 0A 01 jA 01 0A 0j 0i iA A A B B B 1 = − e (γ ) + (Γ + Γ )Γ + γ Γ . 01 B1 1A 01 01 BA Let us begin by considering the case A = 2. Compute 1 1 1 2 −5/4 −P/2−λ/4 1/2 2 2 − e (γ ) = − t e α J − γ γ , 1 θ 01 12 01 2 2 2 1 1 2 B B 2 2 2 3 (Γ + Γ )Γ = − γ γ − γ γ , B1 12 01 12 01 13 01 2 2 B 1 2 2 3 2 γ Γ = γ γ + γ γ . 01 B2 01 12 01 13 62 Thus −5/4 −P/2−λ/4 1/2 Ric(e , e ) = − t e α J . 0 2 θ Compute 1 1 1 1 3 −5/4 P/2−λ/4 1/2 2 3 2 2 − e (γ ) = − t e α (K − QJ ) + γ γ − γ γ , 1 θ θ 01 12 01 13 01 2 2 2 2 3 B B 3 3 (Γ + Γ )Γ = − γ γ , B1 13 01 13 01 B 1 2 2 3 3 γ Γ = γ γ + γ γ . 01 B3 01 13 01 13 2 3 Keeping in mind that γ = −γ , we conclude that 12 13 −5/4 P/2−λ/4 1/2 Ric(e , e ) = − t e α (K − QJ ). 0 3 θ θ Compute, using the symmetries of the connection coefficients, α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 1 1 α 1 11 11 αδ α1 α1 1δ α1 δ1 0 0 α 0 A 0 0 i i 0 0 1 0 =e (Γ ) + Γ Γ − e (γ ) − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ 0 1 11 11 α0 01 01 1A i1 10 i1 1j 01 01 01 11 A 0 B A − γ Γ − γ Γ 01 A1 A1 B1 1 1 0 0 α 0 A A A A 1 0 0 2 =e (Γ ) + Γ Γ − e (γ ) + γ γ − γ γ + γ Γ − (γ ) 0 1 11 11 α0 01 01 01 01 01 01 11 01 4 4 1 0 A 0 B A − γ Γ − γ Γ − γ Γ 01 11 01 A1 A1 B1 0 0 α 0 0 2 A 0 B A =e (Γ ) + Γ Γ − e (γ ) − (γ ) − γ Γ − γ Γ . 0 1 11 11 α0 01 01 01 A1 A1 B1 Due to (233) and (236), we know that the sum of the first four terms is given by −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) . t t t θ θ It can also be computed that A 0 −3 −P 2 P 2 −γ Γ = t [e J + e (K − QJ) ], 01 A1 B A 1/2 −λ/2 2 2P 2 −γ Γ = − t e α(P + e Q ). A1 B1 θ θ Thus −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 Ric(e , e ) = t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) 1 1 t t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 1/2 −λ/2 2 2P 2 + t e + − t e α(P + e Q ). θ θ 7/2 7/2 2 t t 2 Let us turn to α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 1 A α 1 1A 1A αδ αA αA 1δ α1 δA 0 0 α 1 α 0 i i 0 i j 0 i =e (Γ ) + Γ Γ + Γ Γ − Γ Γ − Γ Γ − Γ Γ − γ Γ 1A 1A α0 1A α1 iA 10 0A 1i jA 1i i1 0A i 0 i j − γ Γ − γ Γ 01 iA j1 iA A 1 B B 1 1 1 B 1 1 0 B 0 = − e (γ ) + Γ Γ − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 01 A0 B0 A0 B0 0A 10 0A B0 01 1A 01 BA 63 Compute that 1 1 1 2 −5/4 −λ/4−P/2 3 2 2 2 − e (γ ) = − t e J − γ γ − γ γ , 0 t 01 03 01 02 01 2 2 2 1 1 1 0 −Γ Γ − γ Γ = 0, 02 10 01 12 1 1 B 1 B 1 2 2 2 3 −Γ Γ − Γ Γ = − γ γ − γ γ , 20 B0 02 B0 02 01 03 01 2 2 3 1 1 1 B B 0 2 2 2 3 3 2 Γ Γ − γ Γ = γ γ + γ γ + γ γ . 20 B0 01 B2 01 02 01 03 01 03 2 2 2 Adding up, we obtain −5/4 −λ/4−P/2 Ric(e , e ) = − t e J . 1 2 t Compute that 1 1 1 1 3 −5/4 P/2−λ/4 2 2 2 3 3 3 − e (γ ) = − t e (K − QJ ) − γ γ − γ γ − γ γ , 0 t t 01 03 01 02 01 03 01 2 2 2 2 1 1 1 0 −Γ Γ − γ Γ = 0, 03 10 01 13 B 1 B 1 3 3 −Γ Γ − Γ Γ = − γ γ , 30 B0 03 B0 03 01 3 1 1 1 B B 0 3 3 3 2 2 2 Γ Γ − γ Γ = γ γ + γ γ + γ γ . 30 B0 01 B3 01 03 01 02 01 03 2 2 2 Adding up, we obtain −5/4 P/2−λ/4 Ric(e , e ) = − t e (K − QJ ). 1 3 t t Compute α δ α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − Γ Γ − γ Γ A B α AB AB αδ αB Aδ αA δB 0 1 0 α 1 α δ α δ α =e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 0 1 AB AB AB α0 AB α1 αB Aδ αA δB Note that if A = B = 2, the first four terms can be written −1/2 −λ/2 1/2 −1/2 −1 1/2 t e α ∂ [tα (t + P )] − ∂ (tα P ) ; t t θ θ cf. (234) and (237). Let us therefore turn to δ α δ α 0 i i 0 i j 2 0 2 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α2 2δ α2 δ2 i2 20 02 2i j2 2i 02 22 12 22 i i i 1 A A 1 2 0 2 1 = − Γ (Γ + Γ ) − Γ Γ − Γ Γ − γ Γ − γ Γ . 20 20 02 A2 21 12 2A 02 22 12 22 However, it can be computed that 1 1 i i i −3 −P 2 1/2 −λ/2 2P 2 2 2 −Γ (Γ + Γ ) = − t e J − t e e Q − (γ ) , 20 20 02 t 02 2 2 1 A A 1 1/2 −λ/2 2P 2 2 2 −Γ Γ − Γ Γ = t e αe Q + (γ ) , A2 21 12 2A θ 12 2 0 2 1 2 2 2 2 −γ Γ − γ Γ = (γ ) − (γ ) . 02 22 12 22 02 12 Adding up, λ/2−P 2 1 1 e J −1/2 −λ/2 1/2 −1/2 −1 1/2 1/2 −λ/2 Ric(e , e ) = t e α ∂ [tα (t + P )] − ∂ (tα P ) − t e 2 2 t t θ θ 7/2 2 2 t 1/2 −λ/2 2P 2 2 − t e e (Q − αQ ). t θ 64 Next, consider 0 1 0 α 1 α δ α δ α Ric(e , e ) =e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 2 3 0 1 23 23 23 α0 23 α1 α3 2δ α2 δ3 Due to (239) and (240), the sum of the first four terms can be written h i 1 1 1 −1/2 −λ/2−P 1/2 −1/2 2P 1/2 2P 3 2 2 3 2 2 t e α ∂ (tα e Q ) − ∂ (tα e Q ) − (γ − γ )γ + (γ − γ )γ . t t θ θ 13 12 13 03 02 03 2 2 2 Compute δ α δ α 0 i i 0 i j 2 0 2 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α3 2δ α2 δ3 i3 20 03 2i j3 2i 02 23 12 23 i i i 1 A A 1 2 0 2 1 = − (Γ + Γ )Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 30 03 20 A3 21 13 2A 02 23 12 23 However, 1 1 i i i 3 2 3 2 −(Γ + Γ )Γ = − γ γ − γ γ , 30 03 20 01 01 03 03 2 2 1 A A 1 3 2 −Γ Γ − Γ Γ = γ γ , A3 21 13 2A 13 13 1 1 2 0 2 1 2 2 2 2 −γ Γ − γ Γ = γ γ − γ γ . 02 23 12 23 02 03 12 13 2 2 Adding up, we obtain h i −1/2 −λ/2−P 1/2 −1/2 2P 1/2 2P −3 Ric(e , e ) = t e α ∂ (tα e Q ) − ∂ (tα e Q ) − t J(K − QJ). 2 3 t t θ θ 2 2 Finally, let us consider 0 1 0 α 1 α δ α δ α Ric(e , e ) = e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 3 3 0 1 33 33 33 α0 33 α1 α3 3δ α3 δ3 Due to (235) and (238), the sum of the first four terms is −1/2 −λ/2 1/2 −1/2 −1 1/2 t e α ∂ [tα (t − P )] + ∂ (tα P ) . t t θ θ Let us therefore compute δ α δ α 0 i i 0 i A 0 A 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α3 3δ α3 δ3 i3 30 03 3i j3 03 A3 13 A3 3i i i i 1 A A 1 A 0 A 1 = − (Γ + Γ )Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 30 03 30 A3 31 13 3A 03 A3 13 A3 On the other hand i i i 3 2 3 2 −(Γ + Γ )Γ = − (γ ) − (γ ) , 30 03 30 01 03 1 A A 1 3 2 −Γ Γ − Γ Γ = (γ ) , A3 31 13 3A 13 1 1 A 0 A 1 2 2 3 2 2 2 3 2 −γ Γ − γ Γ = (γ ) + (γ ) − (γ ) − (γ ) . 03 A3 13 A3 03 03 13 13 2 2 Adding up, we obtain 1 1 δ α δ α −3 P 2 1/2 −λ/2 2P 2 2 −Γ Γ − γ Γ = − t e (K − QJ) + t e e (Q − αQ ). α3 3δ α3 δ3 t θ 2 2 Thus −1/2 −λ/2 1/2 −1/2 −1 1/2 Ric(e , e ) = t e α ∂ [tα (t − P )] + ∂ (tα P ) 3 3 t t θ θ λ/2+P 2 1 e (K − QJ) 1 1/2 −λ/2 1/2 −λ/2 2P 2 2 − t e + t e e (Q − αQ ). t θ 7/2 2 t 2 65 A.6 The Einstein tensor Adding up the above, we conclude that the scalar curvature S is given by −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 S = t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) t t t θ θ 1/2 −λ/2 2 2P 2 2 2P 2 + t e P + e Q − α(P + e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 + t e + − t e λ . 7/2 7/2 2 t t 2 As a consequence, if Ein = Ein(e , e ), then αβ α β λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2 2P 2 2 Ein = − t e P + αP + e (Q + αQ ) + + t θ t θ 7/2 7/2 4 t t 1 α −1/2 −λ/2 + t e λ − 2 , 4 α λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2 2P 2 2 Ein = − t e P + αP + e (Q + αQ ) − − t θ t θ 7/2 7/2 4 t t −1/2 −λ/2 + t e λ , 1 1 α 3 1 −1/2 −λ/2 1/2 −1/2 1/2 Ein = t e α ∂ tα P − λ + + − ∂ tα P − λ 22 t t t θ θ θ 2 2 α 2t 2 1/2 −λ/2 2 2P 2 2 2P 2 − t e P + 3e Q − α(P + 3e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e 3 + + t e λ . 7/2 7/2 4 t t 4 The last diagonal component is given by 1 1 α 3 1 −1/2 −λ/2 1/2 −1/2 1/2 Ein = − t e α ∂ tα P + λ − − − ∂ tα P + λ 33 t t t θ θ θ 2 2 α 2t 2 1/2 −λ/2 2 2P 2 2 2P 2 − t e P − e Q − α(P − e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e + 3 + t e λ . 7/2 7/2 4 t t 4 It is of interest to note that −1/2 −λ/2 1/2 −1/2 1/2 1/2 −λ/2 2P 2 2 Ein − Ein =t e α ∂ (tα P ) − ∂ (tα P ) − t e e (Q − αQ ) 22 33 t t θ θ t θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 − t e − . 7/2 7/2 2 t t We also have 1 α 3 −1/2 −λ/2 1/2 −1/2 1/2 Ein + Ein = − t e α ∂ tα λ − 2 − − ∂ tα λ 22 33 t t θ θ 2 α t 1/2 −λ/2 2 2P 2 2 2P 2 − t e P + e Q − α(P + e Q ) t t θ θ λ/2−P 2 λ/2+P 2 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e + + t e λ . 7/2 7/2 t t 2 The remaining components of the Einstein tensor equal the corresponding components of the Ricci tensor, and have consequently already been computed. Using the above calculations, the expressions (59)–(68) for Einstein’s equations, Ein + Λg = T , follow. 66 A.7 The Vlasov equation The distribution function f characterising the Vlasov matter is defined on the mass shell. The mass shell, in its turn, is given by the future directed unit timelike vectors. Since a tangent vector in this set can be written v e , where 0 1 2 2 2 3 2 1/2 v = [1 + (v ) + (v ) + (v ) ] , we can think of f as depending on v , i = 1, 2, 3, and the base point. However, due to the symmetry requirements, the distribution function only depends on the tθ-coordinates of the base point. As a consequence, the distribution function can be considered to be a function of (t, θ, v), 1 2 3 where v = (v , v , v ). In order to derive an equation for f, recall that the Vlasov equation is equivalent to f being constant along future directed unit timelike geodesics. Consider, therefore, a future directed unit timelike geodesic γ(s) = [t(s), θ(s), x(s), y(s)] 2 α in a T -symmetric spacetime. Define the functions v (s) by the equality γ˙ (s) = v (s) e | . γ(s) Note that dt 1/4 −λ/4 0 (s) = t (s)(e ) ◦ γ(s) v (s), (241) ds dθ 1/4 −λ/4 1/2 1 (s) = t (s)(e α ) ◦ γ(s) v (s). (242) ds Let ν(s) = [t(s), θ(s), v(s)], h = f ◦ ν, 1 2 3 where v(s) = [v (s), v (s), v (s)]. The requirement that f be constant along geodesics is equivalent to the requirement that dh/ds = 0 regardless of the choice of future directed unit timelike geodesic γ. On the other hand, dh ∂f dt ∂f dθ ∂f dv = ◦ ν + ◦ ν + ◦ ν . ds ∂t ds ∂θ ds ∂v ds i=1 Keeping (241) and (242) in mind, the requirement that dh/ds = 0 is equivalent to the requirement that 1 i ∂f v ∂f v˙ ∂f 1/2 ◦ ν + α ◦ γ ◦ ν + ◦ ν = 0. 0 1/4 −λ◦γ/4 0 i ∂t v ∂θ t e v ∂v i=1 In order to derive an expression for v˙, note that α α β α β µ α β µ α 0 = γ¨ = (v e ) = v˙ e + v ∇ e = v˙ e + v v ∇ e = (v˙ + v v Γ )e . α α γ˙ β α e β α β µ ds The geodesic equation can thus be written α β µ α v˙ = −v v Γ . βµ Using this formula, it can be calculated that 1 0 0 0 1 0 1 2 0 2 3 0 3 2 2 2 3 3 3 2 2 3 v˙ = −γ v v + γ v v + γ v v + γ v v − γ v v − γ v v − γ v v . 01 01 01 01 12 13 13 Using the formulae for γ , we conclude that λµ 1 2 3 v˙ 1 1 α 1 v v 1/2 0 1 1/2 P = − α λ v − λ − 2 − v + α e Q θ t θ 1/4 −λ/4 0 0 t e v 4 4 α t v 3 2 2 2 1 (v ) − (v ) 1/2 −7/4 λ/4 −P/2 2 P/2 3 − α P + t e e Jv + e (K − QJ)v , 2 v 67 where we have omitted composition with γ for the sake of brevity. We also have 2 2 0 2 2 1 2 v˙ = γ v v + γ v v , 02 12 so that 2 1 2 v˙ 1 1 1 v v 2 1/2 = − P + v − α P . t θ 1/4 −λ/4 0 0 t e v 2 t 2 v Finally, 3 2 0 2 3 0 3 2 1 2 3 1 3 v˙ = γ v v + γ v v + γ v v + γ v v , 03 03 13 13 so that 3 1 3 1 v˙ 1 1 1 v v v 3 1/2 P 2 1/2 = − − P v + α P − e v Q + α Q . t θ t θ 0 0 1/4 −λ/4 0 2 t 2 v v t e v Adding up the above computations, we conclude that the Vlasov equation is equivalent to the requirement that (75) holds. B Notation Metric variables, twist quantities, cosmological constant, frame, manifold. 1. α > 0, λ, P , Q, G and H. These are the functions characterising the metric; cf. (1). 2. λ is defined in (84). 3. J and K. These are the twist quantities defined in (2). They also satisfy (58). 1/2 4. Λ and H. Λ is the positive cosmological constant and H = (Λ/3) . 5. {e } is the orthonormal frame defined in (7). 6. t , t . In the situations of interest in this paper, the metric (1) is defined on (t , ∞) × T , 0 1 0 where t ≥ 0. When speaking of a T -symmetric solution, we take it for granted that t is 0 0 defined in this way. Moreover, t = t + 2. 1 0 Variables for the characteristic system. 1 2 3 1. Θ, V , V , V are the basic variables of the characteristic system (108)–(111). The symbols Θ(s; t, θ, v), V (s; t, θ, v) denote a solution to the characteristic system corresponding to initial data (t, θ, v). In other words, Θ(s; t, θ, v), V (s; t, θ, v), considered as functions of s, are solutions to (108)–(111). Moreover, Θ(t; t, θ, v) = θ, V (t; t, θ, v) = v. 2. Ψ and Z , i = 1, 2, 3. Given a choice of derivative (∂ , ∂ or ∂ i), say ∂, the variables Ψ and t θ 1 2 3 Z = (Z , Z , Z ) are defined by (138)–(141). ˆ ˆ 3. Ψ and Z are the rescaled versions of Ψ and Z, and they are defined in (157). i i 4. Ψ , Z , V and Θ are the higher order derivatives of Ψ, Z, V and Θ. They are defined in j j j j (184). 1 1 ˆ ˆ 5. Z and Ψ are the rescaled versions of Z and Ψ , and they are defined in (199). N N N N Notation, matter quantities. 1. ρ, J , P and S . The quantities ρ, J , P and S are defined in general in (57). In the case i i ij i i ij of Vlasov matter, they are defined in (77). 68 Notation, Vlasov matter. 1. P denotes the mass shell (the set of future directed unit timelike vectors). 2. f denotes the distribution function. For T -symmetric solutions, f, however, denotes the symmetry reduced version of the distribution function. In other words, f is considered to 1 2 3 be a function of t, θ, v , v and v , where, if p is an element of the mass shell, (t, θ, x, y) is the base point of p, and v are the components of p relative to the orthonormal frame {e }. −1/2 3. f is the rescaled distribution function. It is given by f (t, θ, v) = f(t, θ, t v). sc sc Vl 4. T is the stress energy tensor associated with the Vlasov matter. It is given by (4). The αβ components of the stress energy tensor with respect to the frame {e } are given by (57) and (77); see also (76). i i 5. L . The functions L are given by (204)–(206). The initial value formulation. 1. pr is the projection defined in Remark 26. Vl Vl 2. ρ and J are defined in (33) and (34). Auxiliary notation. 1. hhi. If h is a scalar function, its mean value is denoted hhi; cf. (3). 2 1/2 2. hp¯i. If p¯ is a vector, hp¯i = (1 + |p¯| ) . Energies controlling the metric variables, the distribution function and solutions to the characteristic system. 1. The H -norm is defined in (22). Vl,µ 2. ∂ and A are defined in (71). They are given by ± ± 1/2 2 2P 2 ∂ = ∂ ± α ∂ , A = (∂ P ) + e (∂ Q) . ± t θ ± ± ± 3. E is the L -based energy introduced in (85). bas 1 1 4. Q controls the size of the support of f in the v -direction; cf. (114). It is given by 1 1 1 2 3 Q (t) := sup{|v | : (t, θ, v , v , v ) ∈ suppf}. 5. F . The sup-norm energy F is introduced in (115). It is given by F (t) = sup A (t, θ) + sup A (t, θ). + − 1 1 θ∈S θ∈S ˆ ˆ 6. A and F are introduced in (121) and (122) respectively. They are given by ˆ ˆ ˆ ˆ A = t A + t, F (t) = sup A (t, θ) + sup A (t, θ). ± ± + − 1 1 θ∈S θ∈S 1 1 7. R and Q are defined in (124). They are given by 1 1 2 1/2 1 1 2 1/2 R (s) = [s(V (s)) + 1] , Q (s) = [s(Q (s)) + 1] . 69 8. G is defined in (126). 9. E is introduced in (158). It is given by i 2 2 ˆ ˆ ˆ E = (Z ) + (Ψ) . i=1 10. A is introduced in (175). It is given by N+1,± N N 1/2 2 2P N N 1/2 2 A = [∂ P ± ∂ (α P )] + e [∂ Q ± ∂ (α Q )] . N+1,± t θ t θ θ θ θ θ ˆ ˆ 11. A and F are introduced in (176). They are given by N+1,± N+1 7/2 1/2 ˆ ˆ ˆ ˆ A = t A + t , F = sup A + sup A . N+1,± N+1,± N+1 N+1,+ N+1,− 1 1 θ∈S θ∈S 12. E is defined in (200). It is given by 2 1 2 ˆ ˆ ˆ E = (Ψ ) + (Z ) . N N 13. E and E. These energies control suitably weighted Sobolev norms of the distribution function. E is defined in (207). Moreover, E = E . k 0 References [1] Anderson, M. T.: Existence and Stability of even-dimensional asymptotically de Sitter spaces. Ann. Henri Poincar´e 6, 801–820 (2005) [2] Andr´easson, H.: Global Foliations of Matter Spacetimes with Gowdy Symmetry. Commun. Math. Phys. 206, 337-366 (1999) [3] Andr´easson, H.; Rendall, A.D.; Weaver, M.: Existence of CMC and constant areal time folia- tions in T symmetric spacetimes with Vlasov matter. Comm. Partial Differential Equations 29, 237-262 (2004) [4] Andr´easson, H.: The Einstein-Vlasov System/Kinetic Theory, Living Rev. Relativity 14, 4. URL (cited on 2013-04-23): http://www.livingreviews.org/lrr-2011-4 (2011) [5] Berger, B. K.; Chru´sciel, P. T.; Isenberg, J.; Moncrief, V.: Global foliations of vacuum space- times with T isometry. Ann. Physics 260, no. 1, 117148 (1997) [6] Calogero, S., Heinzle, J. M.: Oscillations toward the singularity of LRS Bianchi type IX cosmological models with Vlasov matter. SIAM J. Appl. Dyn. Syst. 9, 1244–1262 (2010) [7] Chru´sciel, P. T. On space-times with U(1) × U(1) symmetric compact Cauchy surfaces. Ann. Physics 202, no. 1, 100150 (1990) [8] Clausen, A.; Isenberg, J.: Areal foliation and asymptotically velocity-term dominated behavior in T symmetric space-times with positive cosmological constant. J. Math. Phys. 48, no. 8, 082501 (2007) [9] Ehlers, J.: Survey of general relativity theory, in Israel, W., ed., Relativity, Astrophysics, and Cosmology, Proceedings of the summer school held 14–26 August 1972 at the Banff Centre, Banff, Alberta, Atrophysics and Space Science Library, vol. 38, pp. 1–125, (Reidel, Dordrecht, Netherlands; Boston, U.S.A., 1973). 70 [10] Friedrich, H.: On the existence of n-geodesically complete or future complete solutions of Einstein’s field equations with smooth asymptotic structure. Commun. Math. Phys. 107, 587– 609 (1986) [11] Friedrich, H.: On the global existence and the asymptotic behavior of solutions to the Einstein–Maxwell–Yang–Mills equations. J. Differential Geom. 34, no. 2, 275–345 (1991) [12] Isenberg, J.; Luo, X.: Power Law Inflation with Electromagnetism. Ann. Physics 334, 420-454 (2013) [13] Lee, H.: Asymptotic behaviour of the Einstein–Vlasov system with a positive cosmological constant. Math. Proc. Camb. Phil. Soc. 137, 495–509 (2004) [14] Lee, H.: The Einstein–Vlasov system with a scalar field. Ann. Henri Poincar´e 6, 697–723 (2005) [15] LeFloch, P. G.; Tchapnda, S. B: Plane-symmetric spacetimes with positive cosmological con- stant. The case of stiff fluids. Adv. Theor. Math. Phys. 15, no. 4, 11151140 (2011) [16] Lu¨bbe, C.; Valiente Kroon, J. A.: A conformal approach for the analysis of the non-linear stability of radiation cosmologies. Ann. Physics 328, 1–25 (2013) [17] Moncrief, V.: Global properties of Gowdy spacetimes with T × R topology. Ann. Physics 132, no. 1, 87-107 (1981) [18] Perlmutter, S. et. al.: Measurements of Ω and Λ from 42 high-redshift supernovae. Astrophys. J. 517, 565–586 (1999) [19] O’Neill, B.: Semi Riemannian Geometry. Academic Press, Orlando (1983) [20] Petersen, P.: Riemannian geometry, Springer Verlag, New York (2006) [21] Rein, G.: On future geodesic completeness for the Einstein-Vlasov system with hyperbolic symmetry. Math. Proc. Cambridge Philos. Soc. 137, 237–244 (2004) [22] Rendall, A. D.: Crushing singularities in spacetimes with spherical, plane and hyperbolic symmetry. Class. Quantum Grav. 12 1517 (1995) [23] Rendall, A.D.: An introduction to the Einstein–Vlasov system, in Chru´sciel, P.T., ed. Math- ematics of Gravitation, Part I: Lorentzian Geometry and Einstein Equations, Proceedings of the Workshop on Mathematical Aspects of Theories of Gravitation, held in Warsaw, February 29 - March 30, 1996, Banach Center Publications, vol. 41, pp. 35–68, (Polish Academy of Sciences, Institute of Mathematics, Warsaw, Poland, 1997). [24] Rendall, A. D.: Accelerated cosmological expansion due to a scalar field whose potential has a positive lower bound. Class. Quant. Grav. 21, 2445–2454 (2004) [25] Rendall, A. D.: Intermediate inflation and the slow-roll approximation. Class. Quant. Grav. 22, 1655–1666 (2005) [26] Rendall, A. D.: Dynamics of k-essence. Class. Quant. Grav. 23, 1557–1570 (2006) [27] Riess, A. G. et al.: Observational Evidence from Supernovae for an Accelerating Universe and a Cosmological Constant. Astron. J. 116, 1009–1038 (1998) [28] Ringstr¨om, H.: Future stability of the Einstein non-linear scalar field system. Invent. math. 173, 123–208 (2008) [29] Ringstr¨om, H.: Power law inflation. Commun. Math. Phys. 290, 155–218 (2009) 71 [30] Ringstr¨om, H.: The Cauchy problem in General Relativity, European Mathematical Society, Zu¨rich (2009) [31] Ringstr¨om, H.: On the Topology and Future Stability of the Universe, Oxford University Press, Oxford (2013) [32] Rodnianski, I., Speck, J.: The nonlinear future stability of the FLRW family of solutions to the irrotational Euler–Einstein system with a positive cosmological constant. J. Eur. Math. Soc. (JEMS) 15, no. 6, 2369–2462 (2013) [33] Smulevici, J.: On the area of the symmetry orbits of cosmological spacetimes with toroidal or hyperbolic symmetry. Anal. PDE 4, no. 2, 191–245 (2011) [34] Speck, J.: The nonlinear future stability of the FLRW family of solutions to the Euler-Einstein system with a positive cosmological constant. Selecta Math. (N.S.) 18, no. 3, 633–715 (2012) [35] Speck, J.: The stabilizing effect of spacetime expansion on relativistic fluids with sharp results for the radiation equation of state. Arch. Ration. Mech. Anal. 210, no. 2, 535–579 (2013) [36] Svedberg, C.: Future Stability of the Einstein–Maxwell–Scalar Field System. Ann. Henri Poincar´e 12, No. 5, 849–917 (2011) [37] Tchapnda, S.B.; Rendall, A. D.: Global existence and asymptotic behaviour in the future for the Einstein-Vlasov system with positive cosmological constant. Classical Quantum Gravity 20, 3037–3049 (2003) [38] Tchapnda, S. T., Noutchegueme, N.: The surface–symmetric Einstein–Vlasov system with cosmological constant. Math. Proc. Cambridge Phil. Soc. 138, 541–553 (2005) [39] Tchapnda, S. B.: The plane symmetric Einstein-dust system with positive cosmological con- stant. J. Hyperbolic Differ. Equ. 5, no. 3, 681692 (2008) [40] Wald, R.: Asymptotic behaviour of homogeneous cosmological models in the presence of a positive cosmological constant. Phys. Rev. D 28, 2118–2120 (1983) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Mathematics arXiv (Cornell University)

Proof of the cosmic no-hair conjecture in the T^3-Gowdy symmetric Einstein-Vlasov setting

Mathematics , Volume 2021 (1306) – Jun 26, 2013

Loading next page...
 
/lp/arxiv-cornell-university/proof-of-the-cosmic-no-hair-conjecture-in-the-t-3-gowdy-symmetric-RrcFEc0O7B
ISSN
1435-9855
eISSN
ARCH-3343
DOI
10.4171/JEMS/623
Publisher site
See Article on Publisher Site

Abstract

The currently preferred models of the universe undergo accelerated expansion induced by dark energy. One model for dark energy is a positive cosmological constant. It is consequently of interest to study Einstein’s equations with a positive cosmological constant coupled to matter satisfying the ordinary energy conditions; the dominant energy condition etc. Due to the difficulty of analysing the behaviour of solutions to Einstein’s equations in general, it is common to either study situations with symmetry, or to prove stability results. In the present paper, we do both. In fact, we analyse, in detail, the future asymptotic behaviour of T -Gowdy symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant. In particular, we prove the cosmic no-hair conjecture in this setting. However, we also prove that the solutions are future stable (in the class of all solutions). Some of the results hold in a more general setting. In fact, we obtain conclusions concerning the causal structure of T -symmetric solutions, assuming only the presence of a positive cosmological constant, matter satisfying various energy conditions and future global existence. Adding the 3 0 assumption of T -Gowdy symmetry to this list of requirements, we obtain C -estimates for all but one of the metric components. There is consequently reason to expect that many of the results presented in this paper can be generalised to other types of matter. 1 Introduction At the end of 1998, two research teams studying supernovae of type Ia announced the unexpected conclusion that the universe is expanding at an accelerating rate; cf. [27, 18]. After the obser- vations had been corroborated by other sources, there was a corresponding shift in the class of solutions to Einstein’s equations used to model the universe. In particular, physicists attributed the acceleration to a form of matter they referred to as ’dark energy’. However, as the nature of the dark energy remains unclear, there are several models for it. The simplest one is that of a positive cosmological constant (which is the one we use in the present paper), but there are several other possibilities; cf., e.g., [24, 25, 26] and references cited therein for some examples. Combining the different observational data, the currently preferred model of the universe is spatially homoge- neous and isotropic (i.e., the cosmological principle is assumed to be valid), spatially flat, and has matter of the following forms: ordinary matter (usually modelled by a radiation fluid and dust), dark matter (often modelled by dust), and dark energy (often modelled by a positive cosmological constant). In the present paper, we are interested in the Einstein-Vlasov system. This corresponds to a different description of the matter than the one usually used. However, this system can also be used in order to obtain models consistent with observations; cf., e.g., [31, Chapter 28]. In fact, Vlasov matter has the property that it naturally behaves as radiation close to the singularity and as dust in the expanding direction, a desirable feature which is usually put in by hand when using perfect fluids to model the matter. The cosmic no-hair conjecture. The standard starting point in cosmology is the assumption of spatial homogeneity and isotropy. However, it is preferable to prove that solutions generally arXiv:1306.6223v2 [gr-qc] 16 May 2016 isotropise and that the spatial variation (as seen by observers) becomes negligible. This is expected to happen in the presence of a positive cosmological constant; in fact, solutions are in that case expected to appear de Sitter like to observers at late times. The latter expectation goes under the name of the cosmic no-hair conjecture; cf. Conjecture 11 for a precise formulation. The main objective when studying the expanding direction of solutions to Einstein’s equations with a positive cosmological constant is to verify this conjecture. Spatial homogeneity. Turning to the results that have been obtained in the past, it is natural to begin with the spatially homogeneous setting. In 1983, Robert Wald wrote a short, but remarkable, paper [40], in which he proves results concerning the future asymptotic behaviour of spatially homogeneous solutions to Einstein’s equations with a positive cosmological constant. In particular, he confirms that the cosmic no-hair conjecture holds. What is remarkable about the paper is the fact that he is able to obtain conclusions assuming only that certain energy conditions hold and that the solution does not break down in finite time. Concerning the symmetry type, the only issue that comes up in the argument is whether it is compatible with the spatial hypersurfaces of homogeneity having positive scalar curvature or not; positive scalar curvature of these hypersurfaces sometimes leads to recollapse. The results should be contrasted with the case of Einstein’s vacuum equation in the spatially homogeneous setting, where the behaviour is strongly dependent on the symmetry type. Since Wald does not prove future global existence, it is necessary to carry out a further analysis in order to confirm the picture obtained in [40] in specific cases. In the case of the Einstein-Vlasov system, this was done in [13]. It is also of interest to note that it is possible to prove results analogous to those of Wald for more general models for dark energy; cf., e.g., [24, 25, 26, 14]. Surface symmetry. Turning to the spatially inhomogeneous setting, there are results in the surface symmetric case with a positive cosmological constant; cf. [37, 38, 39, 15], and see [22] for a definition of surface symmetry. In this case, the isometry group (on a suitable covering space) is 3-dimensional. Nevertheless, the system of equations that result after symmetry reduction is 1 + 1-dimensional. However, the extra symmetries do eliminate some of the degrees of freedom. Again, the main results are future causal geodesic completeness and a verification of the cosmic no-hair conjecture. T -symmetry. A natural next step to take after surface symmetry is to consider Gowdy or T -symmetry. That is the purpose of the present paper. In particular, we prove future causal geodesic completeness of solutions to the T -Gowdy symmetric Einstein-Vlasov equations with a positive cosmological constant (note, however, the caveat concerning global existence stated in Subsection 1.1). Moreover, we verify that the cosmic no-hair conjecture holds. It is of interest to note that most of the arguments go through under the assumption of T -symmetry. However, in order to obtain the full picture in this setting, it is necessary to prove one crucial inequality, cf. Definition 1, which we have not yet been able to do in general. Stability. A fundamental question in the study of cosmological solutions is that of future stability: given initial data corresponding to an expanding solution, do small perturbations thereof yield maximal globally hyperbolic developments which are future causally geodesically complete and globally similar to the future? In the case of a positive cosmological constant, the first result was obtained by Helmut Friedrich; he proved stability of de Sitter space in 3 + 1 dimensions in [10]. Later, he and Michael Anderson generalised the result to higher (even) dimensions and to include various matter fields; cf. [11, 1]. Moreover, results concerning radiation fluids were obtained in [16]. However, conformal invariance plays an important role in the arguments presented in these papers. As a consequence, there seems to be a limitation of the types of matter models that can be dealt with using the corresponding methods. The paper [28] was written with the goal of developing methods that are more generally applicable. The papers [29, 36, 32, 34, 35, 12], in which the methods developed in [28] play a central role, indicate that this goal was achieved. In fact, a general future global non-linear stability result for spatially homogeneous solutions to the Einstein-Vlasov equations with a positive cosmological constant was obtained in [31], the ideas developed in [28] being at the core of the argument. In the present paper, we not only derive 2 3 detailed future asymptotics of T -Gowdy symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant. We also prove that all the resulting solutions are future stable in the class of all solutions (without symmetry assumptions). Outlook. As we describe in the next subsection, some of the results concerning T -Gowdy symmetric solutions hold irrespective of the matter model (as long as it satisfies the dominant energy condition and the non-negative pressure condition). As a consequence, we expect that it might be possible to derive detailed asymptotics in the case of the Einstein-Maxwell equations (with a positive cosmological constant), and in the case of the Einstein-Euler system (though the issue of shocks may be relevant in the latter case). Due to the stability results demonstrated in [36, 32, 34, 35], it might also be possible to prove stability of the corresponding solutions. 1.1 General results under the assumption of T -symmetry 2 2 T -symmetry. In the present paper, we are interested in T -symmetric solutions to Einstein’s equations. There are various geometric ways of imposing this type of symmetry (cf., e.g., [7, 33]), but for the purposes of the present paper, we simply assume the topology to be of the form I ×T , where I is an open interval contained in (0, ∞). If θ, x and y are ’coordinates’ on T and t is the coordinate on I, we also assume the metric to be of the form −1/2 λ/2 2 −1 2 P 2 −P 2 g = t e (−dt + α dθ ) + te [dx + Qdy + (G + QH)dθ] + te (dy + Hdθ) , (1) where the functions α > 0, λ, P , Q, G and H only depend on t and θ; cf., e.g., [33]. Note that translation in the x and y directions defines a smooth action of T on the spacetime (as well as on each constant t-hypersurface). Moreover, the metric is invariant under this action, and the corresponding orbits are referred to as the symmetry orbits, given by {t} × {θ} × T . Note that the area of the symmetry orbits is proportional to t. For this reason, the foliation of the spacetime corresponding to the metric form (1) is referred to as the constant areal time foliation. The case of T -Gowdy symmetry corresponds to the functions G and H being independent of time; again, there is a more geometric way of formulating this condition: the spacetime is said to be Gowdy symmetric if the so-called twist quantities, given by α β γ δ α β γ δ J = ǫ X Y ∇ X , K = ǫ X Y ∇ Y , (2) αβγδ αβγδ vanish, where X = ∂ and Y = ∂ are Killing fields of the above metric and ǫ is the volume form. x y A basic question to ask concerning T -symmetric solutions to Einstein’s equations is whether the maximal globally hyperbolic development of initial data admits a constant areal time foliation which is future global. There is a long history of proving such results. The first one was obtained by Vincent Moncrief, cf. [17], in the case of vacuum solutions with T -Gowdy symmetry. The case of T -symmetric vacuum solutions with and without a positive cosmological constant have also been considered in [8] and [5] respectively. Turning to Vlasov matter, [2] contains an analysis of the existence of foliations in the T -Gowdy symmetric Einstein-Vlasov setting. The corresponding results were later extended to the T -symmetric case in [3]. However, from our point of view, the most relevant result is that of [33]. Due to the results of this paper, there is, given T -symmetric initial data to the Einstein-Vlasov equations with a positive cosmological constant, a future global foliation of the spacetime of the form (1). In other words I = (t , ∞). Moreover, if the distribution function is not identically zero, then t = 0. Finally, if the initial data have Gowdy symmetry, then the same is true of the development. Strictly speaking, the future global existence result in [33] is based on the observation that the argument should not be significantly different from the proofs in [5, 8, 3]. It would be preferable to have a complete proof of future global existence in the case of interest here, but we shall not provide it in this paper. Results. Turning to the results, it is of interest to note that some of the conclusions can be obtained without making detailed assumptions concerning the matter content. For that reason, let us, for the remainder of this subsection, assume that we have a solution to Einstein’s equations 3 with a positive cosmological constant, where the metric is of the form (1), the existence interval I is of the form (t , ∞) and the matter satisfies the dominant energy condition and the non- negative pressure condition; recall that the matter is said to satisfy the dominant energy condition if T (u, v) ≥ 0 for all pairs u, v of future directed timelike vectors (where T is the stress energy tensor associated with the matter); and that it is said to satisfy the non-negative pressure condition if T (w, w) ≥ 0 for every spacelike vector w. To begin with, there is a constant C > 0 such that −3 1 α(t, θ) ≤ Ct for all (t, θ) ∈ [t +2, ∞)×S ; cf. Proposition 42. In fact, this conclusion also holds if we replace the cosmological constant with a non-linear scalar field with a positive lower bound; cf. Remark 43. One particular consequence of this estimate for α is that the θ-coordinate of a causal curve converges. Moreover, observers whose θ-coordinates converge to different θ-values are asymptotically unable to communicate. In this sense, there is asymptotic silence. In the case of Gowdy symmetry, more can be deduced. In fact, for every ǫ > 0, there is a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞)×S ; cf. Proposition 44. This estimate turns out to be of crucial importance also in the general T -symmetric case. For this reason, we introduce the following terminology. Definition 1. A metric of the form (1) which is defined for t > t for some t ≥ 0 is said to have 0 0 λ-asymptotics if there, for every ǫ > 0, is a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞) × S . Remark 2. All Gowdy solutions have λ-asymptotics under the above assumptions; cf. Proposi- tion 44. Proposition 3. Consider a T -symmetric solution to Einstein’s equations with a positive cos- mological constant. Assume that the matter satisfies the dominant energy condition and the non- negative pressure condition. Assume, moreover, that the corresponding metric admits a foliation of the form (1), on I × T , where I = (t , ∞) and t ≥ 0. Finally, assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0 such that 1 0 −1/2 λ(t, ·) + 3 ln t − 2 ln ≤ Ct , 4Λ −3/2 −1/2 0 0 t hα (t, ·)i + kQ(t, ·)k + kP (t, ·)k ≤ C, C C −3/2 1 1 kH (t, ·)k + kG (t, ·)k ≤ Ct t L t L for all (t, θ) ∈ [t , ∞) × S . Remark 4. The choice t = t + 2 may seem unnatural. However, we need to stay away from t 1 0 0 (since we do not control the solution close to t ). Moreover, in some situations we need to know that ln t is positive and bounded away from zero. Since t = 0 for most solutions, it is therefore natural to only consider the interval t ≥ t + 2 in the study of the future asymptotics. Remark 5. If h is a scalar function on S , we use the notation hhi = hdθ. (3) 2π Sometimes, we shall use the same notation for a scalar function h on I ×S . In that case, hhi is the function of t defined by hh(t, ·)i. Finally, if p¯ ∈ R , we shall also use the notation hp¯i. However, 2 1/2 in that case, hp¯i = (1 + |p¯| ) ; cf. Remark 19. 4 Proof. The statement is a consequence of Lemmas 46, 47 and 48. In particular, in the case of a T -Gowdy symmetric solution, there is asymptotic silence in the sense that the θxy-coordinates of a causal curve converge, and causal curves whose asymptotic θxy-coordinates differ are asymptotically unable to communicate; cf. Proposition 49. 1.2 Results in the Einstein-Vlasov setting In order to be able to draw detailed conclusions, we need to restrict our attention to a specific type of matter. In the present paper, we study the Einstein-Vlasov system. A general description of Vlasov matter. Intuitively, Vlasov matter gives a statistical descrip- tion of an ensemble of collections of particles. In practice, the matter is described by a distribution function defined on the space of states of particles. The possible states are given by the future directed causal vectors (here and below, we assume the Lorentz manifolds under consideration to be time oriented). Usually, one distinguishes between massive and massless particles. In the latter case, the distribution function is defined on the future light cone, and in the former case, it is defined on the interior. In the present paper, we are interested in the massive case, and we assume all the particles to have unit mass (for a description of how to reduce the case of varying masses to the case of all particles having unit mass, see [6]). As a consequence, the distribution function is a non-negative function on the mass shell P, defined to be the set of future directed unit timelike vectors. In order to connect the matter to Einstein’s equations, we need to associate a stress energy tensor with the distribution function. It is given by Vl T (r) = fp p µ . (4) α β P αβ r Pr In this expression, P denotes the set of future directed unit timelike vectors based at the spacetime point r. In other words, if T ∈ T M is a future directed timelike vector, then ref r P = {v ∈ T M | g(v, v) = −1, g(T , v) < 0}. r r ref Moreover, the Lorentz metric g induces a Riemannian metric on P , and µ denotes the corre- r P sponding volume form; cf. (18) below for a coordinate representation of µ . Finally, p denotes P α the components of the one form obtained by lowering the index of p ∈ P using the Lorentz metric g. Clearly, it is necessary to demand some degree of fall off of the distribution function f in order for the integral (4) to be well defined. In the present paper, we shall mainly be interested in the case that the distribution function has compact support in the momentum directions (for a fixed spacetime point). However, in Subsections 1.3–1.7 we shall consider a somewhat more general situation. Turning to the equation the distribution function has to satisfy, it is given by Lf = 0. (5) Here L denotes the vector field induced on the mass shell by the geodesic flow; cf. (19) below for a coordinate representation of it. An alternate way to formulate this equation is to demand that f be constant along γ˙ for every future directed unit timelike geodesic γ. The intuitive interpretation of the Vlasov equation (5) is that collisions between particles are neglected. It is of interest to note that if f satisfies the Vlasov equation, then the stress energy tensor is divergence free. To conclude, the Einstein-Vlasov equations with a positive cosmological constant consist of (5) and Vl Ein + Λg = T , (6) Vl where T is given by the right hand side of (4) and Λ is a positive constant. Moreover, Ein = Ric − Sg 5 is the Einstein tensor, where Ric is the Ricci tensor and S is the scalar curvature of a Lorentz manifold (M, g). The above description is somewhat brief, and the reader interested in more details is referred to, e.g., [9, 23, 4, 31]. 2 2 Vlasov matter under the assumption of T -symmetry. In the case of T -symmetry, it is convenient to use a symmetry reduced version of the distribution function. Introduce, to this end, the orthonormal frame 1/4 −λ/4 1/4 −λ/4 1/2 e =t e ∂ , e = t e α (∂ − G∂ − H∂ ), 0 t 1 θ x y (7) −1/2 −P/2 −1/2 P/2 e =t e ∂ , e = t e (∂ − Q∂ ). 2 x 3 y x Since the distribution function f is defined on the mass shell, it is convenient to parametrise this set; note that the manifolds we are interested in here are parallelisable (i.e., they have a global frame). An element in P can be written v e , where 0 1 2 2 2 3 2 1/2 v = [1 + (v ) + (v ) + (v ) ] . As a consequence, we can think of f as depending on v , i = 1, 2, 3, and the base point. However, due to the symmetry requirements, the distribution function only depends on the tθ-coordinates of the base point. As a consequence, the distribution function can be considered to be a function of 1 2 2 (t, θ, v), where v = (v , v , v ). In what follows, we shall abuse notation and denote the symmetry 1 3 reduced function, defined on I × S × R , by f. A symmetry reduced version of the equations is to be found in Section 2. Remark 6. In the T -symmetric setting, we always assume the distribution function f to have compact support when restricted to constant t-hypersurfaces. Under the assumptions made in the present paper, f has this property, assuming the initial datum for f to have compact support. The first question to ask concerning T -symmetric solutions is that of existence of constant areal time foliations for an interval of the form (t , ∞). However, due to previous results, cf. [33], we know that T -symmetric solutions to the Einstein-Vlasov equations with a positive cosmological constant are future global in this setting (keeping the caveat stated in Subsection 1.1 in mind). In other words, there is a t ≥ 0 such that the solution admits a foliation of the form (1) on I × T , where I = (t , ∞). Consequently, the issue of interest here is that of the asymptotics. Unfortu- nately, we are unable to derive detailed asymptotics for all T -symmetric solutions. However, we do obtain results for solutions with λ-asymptotics; recall that all T -Gowdy symmetric solutions fall into this class. Theorem 7. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics and let t = t + 2. Then 0 1 0 there are smooth functions α > 0, P , Q , G and H on S , and, for every 0 ≤ N ∈ Z, a ∞ ∞ ∞ ∞ ∞ constant C > 0 such that −3/2 tkH (t, ·)k N + tkG (t, ·)k N + kH(t, ·) − H k N + kG(t, ·) − G k N ≤ C t , (8) t t ∞ ∞ N C C C C −1 tkP (t, ·)k N + tkQ (t, ·)k N + kP (t, ·) − P k N + kQ(t, ·) − Q k N ≤ C t , (9) t t ∞ ∞ N C C C C α 3 3 −2 + + λ + ≤ C t , (10) t N α t t N N C C 3 −1 t α(t, ·) − α + λ(t, ·) + 3 ln t − 2 ln ≤ C t , (11) ∞ N N 4Λ −1/2 for all t ≥ t . Define f via f (t, θ, v) = f(t, θ, t v). Then there is an R > 0 such that 1 sc sc suppf (t, ·) ⊆ S × B (0) sc R 6 3 for all t ≥ t , where B (0) is the ball of radius R in R . Moreover, there is a smooth, non-negative 1 R 1 3 function with compact support, say f , on S × R , such that sc,∞ −1 tk∂ f (t, ·)k N 1 3 + kf (t, ·) − f k N 1 3 ≤ C t t sc sc sc,∞ N C (S ×R ) C (S ×R ) for all t ≥ t . Turning to the geometry, let g¯(t, ·) and k(t, ·) denote the metric and second funda- mental form induced by g on the hypersurface {t} × T , and let g¯ (t, ·) denote the components of ij g¯(t, ·) with respect to the vectorfields ∂ = ∂ , ∂ = ∂ and ∂ = ∂ etc. Then 1 θ 2 x 3 y −1 −1 −1 kt g¯ (t, ·) − g¯ k N + kt k − Hg¯ k N ≤ C t (12) ij ∞,ij ij ∞,ij N C C 1/2 for all t ≥ t , where H = (Λ/3) and 2 P∞ 2 −P∞ 2 g¯ = dθ + e [dx + Q dy + (G + Q H )dθ] + e (dy + H dθ) . (13) ∞ ∞ ∞ ∞ ∞ ∞ 4Λα Moreover, the solution is future causally geodesically complete. The proof of the above theorem is to be found in Section 10. It is of interest to record how the spacetime appears to an observer. In particular, we wish to prove the cosmic no-hair conjecture in the present setting. The rough statement of this conjecture is that the spacetime appears de Sitter like to late time observers. However, in order to be able to state a theorem, we need a formal definition. Before proceeding to the details, let us provide some intuition. Let 2 2Ht g = −dt + e g¯ , (14) dS E 1/2 4 where H = (Λ/3) and g¯ denotes the standard flat Euclidean metric. Then (R , g ) corre- E dS sponds to a part of de Sitter space. It may seem more reasonable to consider de Sitter space itself. However, as far as the asymptotic behaviour of de Sitter space is concerned, (14) is as good a model as de Sitter space itself. Consider a future directed and inextendible causal curve in 4 0 3 (R , g ), say γ = (γ , γ¯), defined on (s , s ). Then γ¯(s) converges to some x¯ ∈ R as s → s −. dS − + 0 + Moreover, γ(s) ∈ C for all s, where x ¯ ,Λ −1 −Ht C = {(t, x¯) : |x¯ − x¯ | ≤ H e }. x ¯ ,Λ 0 In practice, it is convenient to introduce a lower bound on the time coordinate and to introduce a margin in the spatial direction. Moreover, it is convenient to work with open sets. We shall therefore be interested in sets of the form −1 −Ht C = {(t, x¯) : t > T, |x¯| < KH e }; (15) Λ,K,T note that x¯ can be translated to zero by an isometry. Since we are interested in the late time behaviour of solutions, it is natural to restrict attention to sets of the form C for some K ≥ 1 Λ,K,T and T > 0. Definition 8. Let (M, g) be a time oriented, globally hyperbolic Lorentz manifold which is future causally geodesically complete. Assume, moreover, that (M, g) is a solution to Einstein’s equations with a positive cosmological constant Λ. Then (M, g) is said to be future asymptotically de Sitter like if there is a Cauchy hypersurface Σ in (M, g) such that for every future oriented and inextendible causal curve γ in (M, g), the following holds: − + • there is an open set D in (M, g), such that J (γ) ∩ J (Σ) ⊂ D, and D is diffeomorphic to C for a suitable choice of K ≥ 1 and T > 0, Λ,K,T −1 −Ht • using ψ : C → D to denote the diffeomorphism; letting R(t) = KH e ; using Λ,K,T g¯ (t, ·) and k (t, ·) to denote the metric and second fundamental form induced on S = dS dS t {t} × B (0) by g ; using g¯(t, ·) and k(t, ·) to denote the metric and second fundamental dS R(t) 7 ∗ ∗ form induced on S by ψ g (where ψ g denotes the pullback of g by ψ); and letting N ∈ N, the following holds: ¯ ¯ lim kg¯ (t, ·) − g¯(t, ·)k N + kk (t, ·) − k(t, ·)k N = 0. (16) dS dS C (S ) C (S ) t t dS dS t→∞ Remark 9. In the definition, we use the notation 1/2 i i j j 1 l 1 l im jn khk N = sup g¯ · · ·g¯ g¯ g¯ ∇ · · · ∇ h ∇ · · ·∇ h dS,i j dS,i j ij mn C (S ) 1 1 l l dS dS dS dS dS dS dS l=0 for a covariant 2-tensor field h on S , where ∇ denotes the Levi-Civita connection associated t dS with g¯ (t, ·). Note also that, even though R(t) shrinks to zero exponentially, the diameter of S , dS t as measured with respect to g¯ (t, ·), is constant. dS Remark 10. In some situations it might be more appropriate to adapt the Cauchy hypersurface Σ to the causal curve γ; i.e., to first fix γ and then Σ. The above definition leads to a formal statement of the cosmic no-hair conjecture. Conjecture 11 (Cosmic no-hair). Let A denote the class of initial data such that the corre- sponding maximal globally hyperbolic developments (MGHD’s) are future causally geodesically complete solutions to Einstein’s equations with a positive cosmological constant Λ (for some fixed matter model). Then generic elements of A yield MGHD’s that are future asymptotically de Sitter like. Remark 12. It is probably necessary to exclude certain matter models in order for the statement to be correct. Moreover, the statement, as it stands, is quite vague; there is no precise definition of the notion generic. However, what notion of genericity is most natural might depend on the situation. Remark 13. The Nariai spacetimes, discussed, e.g., in [28, pp. 126-127], are time oriented, globally hyperbolic, causally geodesically complete solutions to Einstein’s vacuum equations with a positive cosmological constant that do not exhibit future asymptotically de Sitter like behaviour. They are thus potential counterexamples to the cosmic no-hair conjecture. There is a similar example in the Einstein-Maxwell setting (with a positive cosmological constant) in [28, p. 127]. However, both of these examples are rather special, and it is natural to conjecture them to be unstable. Nevertheless, they constitute the motivation for demanding genericity. Finally, we are in a position to phrase a result concerning the cosmic no hair conjecture in the T -Gowdy symmetric setting. The proof of the theorem below is to be found in Section 10. Theorem 14. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics. Then the solution is future asymptotically de Sitter like; i.e., the cosmic no-hair conjecture holds. Remark 15. Recall that all T -Gowdy symmetric solutions have λ-asymptotics. Remark 16. In the particular case of interest here, the equality (16) can actually be improved to the estimate −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k N + kk (τ, ·) − k(τ, ·)k N ≤ C e dS dS N C (S ) C (S ) τ τ dS dS for all τ > T and a suitable constant C . Remark 17. The main estimate needed to prove the theorem is (12). In situations where such an estimate holds, it is thus to be expected that the solution is future asymptotically de Sitter like. 8 1.3 Stability, notation and function spaces Let us now turn to the subject of stability. Combining Theorem 7 with the results of [31], it turns out to be possible to prove that the solutions to which Theorem 7 applies are also future stable. In the present subsection, we begin by introducing the terminology necessary in order to make a formal statement of this result. Let (M, g) be a time oriented n+1-dimensional Lorentz manifold. We say that (x, U) are canonical local coordinates if ∂ 0 is future oriented timelike on U and g(∂ i| , ∂ j| ), i, j = 1, . . ., n, are the x x r x r components of a positive definite metric for every r ∈ U; cf. [31, p. 87]. If p ∈ P for some r ∈ U, we then define Ξ (p) = Ξ (p ∂ | ) = [x(r), p¯], (17) x x x r 1 n where p¯ = (p , . . . , p ). Note that Ξ are local coordinates on the mass shell. If f is defined on the −1 mass shell, we shall use the notation f = f ◦ Ξ . It is also convenient to introduce the notation ¯ ¯ p according to Ξ (p) = [x(r), p (p)], assuming p ∈ P . With this notation, the measure µ can x x x r P be written 1/2 |g (r)| µ = − ι dp¯ , (18) Pr p ◦ ι x,0 r where |g | is the determinant of the metric g, when expressed with respect to the x-coordinates; ι : P → P is the inclusion; p (p) are the components of p with respect to the coordinates x; and r r p (p) = g p (p). The reader interested in a derivation of (18) is referred to [31, Section 13.3]. x,α x,αβ Let us also note that the operator L is given by ∂ ∂ α i α β L = p − Γ p p (19) x αβ x x α i ∂x ∂¯ p with respect to the above coordinates. In order to proceed, we need to introduce function spaces for the distribution functions. Recall, to that end, [31, Definition 7.1, p. 87]: Definition 18. Let 1 ≤ n ∈ Z, µ ∈ R, (M, g) be a time oriented n + 1-dimensional Lorentz manifold and P be the set of future directed unit timelike vectors. The space D (P) is defined to consist of the smooth functions f : P → R such that, for every choice of canonical local coordinates (x, U), n + 1-multiindex α and n-multiindex β, the derivative ∂ ∂ f (where x symbolises the first x p¯ n + 1 and p¯ the last n variables), considered as a function from x(U) to the set of functions from R to R, belongs to 2 n C[x(U), L (R )]. (20) µ +|β| 2 n 2 Remark 19. The space L (R ) is the weighted L -space corresponding to the norm 1/2 2µ 2 khk = hp¯i |h(p¯)| dp¯ , (21) 2 1/2 where hp¯i = (1 + |p¯| ) ; recall the comments made in Remark 5. Remarks 20. If f ∈ D (P) for some µ > n/ 2 + 1, then the stress energy tensor is a well defined smooth function; cf. [31, Proposition 15.37, p. 246]. Moreover, the stress energy tensor is divergence free if f satisfies the Vlasov equation. It is worth pointing out that it is possible to introduce more general function spaces, corresponding to a finite degree of differentiability; cf. [31, Definition 15.1, p. 234]. However, the above definition is sufficient for our purposes. The above function spaces are suitable when discussing functions on the mass shell. However, we also need to introduce function spaces for the initial datum for the distribution function. If (¯x, U) are local coordinates on a manifold Σ, we introduce local ¯ ¯ ¯ coordinates on T Σ by Ξ (p¯ ∂ i| ) = (¯x(ξ), p¯) in analogy with (17). Moreover, if f is defined on ¯ x ¯ x −1 ¯ ¯ ¯ T Σ, we shall use the notation f = f ◦ Ξ . Let us recall [31, Definition 7.5, p. 89]: ¯ x ¯ x 9 ∞ Definition 21. Let 1 ≤ n ∈ Z, µ ∈ R and Σ be an n-dimensional manifold. The space D (T Σ) is defined to consist of the smooth functions f : T Σ → R such that, for every choice of local coordinates (¯x, U), n-multiindex α and n-multiindex β, the derivative ∂ ∂ f (where x¯ symbolises ¯ x x ¯ p¯ the first n and p¯ the last n variables), considered as a function from ¯x(U) to the set of functions from R to R, belongs to 2 n C[¯x(U), L (R )]. µ +|β| Remark 22. According to the criteria appearing in Definitions 18 and 21, we need to verify continuity conditions for every choice of local coordinates. However, it turns out to be sufficient to consider a collection of local coordinates covering the manifold of interest; cf. [31, Lemma 15.9, p. 235] and [31, Lemma 15.19, p. 237]. Finally, in order to be able to state a stability result, we need a norm. Recall, to this end, [31, Definition 7.7, pp. 89–90]: Definition 23. Let 1 ≤ n ∈ Z, 0 ≤ l ∈ Z, µ ∈ R and Σ be a compact n-dimensional manifold. Let, moreover, χ¯ , i = 1, ..., j, be a finite partition of unity subordinate to a cover consisting of coordinate neighbourhoods, say (¯x , U ). Then k · k l is defined by i i Vl,μ   1/2 j Z X X 2µ +2|β| α β 2 ¯  ¯ ¯ ¯ ¯  kfk l = h̺¯i χ¯ (ξ)(∂ ∂ f ) (ξ, ̺¯)dξd̺¯ (22) i ¯ ¯ x H ̺¯ i Vl,μ ¯ x (U )×R i i i=1 |α|+|β|≤l ¯ ¯ ∞ for each f ∈ D (T Σ). Remark 24. Clearly, the norm depends on the choice of partition of unity and on the choice of coordinates. However, different choices lead to equivalent norms. Here, we are mainly interested in the case Σ = T , in which case it is neither necessary to introduce local coordinates nor a partition of unity. 1.4 The Einstein-Vlasov-non-linear scalar field system In the present paper, we are mainly interested in the Einstein-Vlasov system with a positive cosmological constant. However, in the proof of future stability of T -Gowdy symmetric solutions, we use two results. First, we use the fact that solutions that start out close to de Sitter space are future stable. Second, we use Cauchy stability. There are results of this type in the literature. However, they are formulated in the Einstein-Vlasov-non-linear scalar field setting. In order to make it clear that the statements appearing in the literature can be applied in our setting, it is therefore necessary to briefly describe the Einstein-Vlasov-non-linear scalar field system. This is the purpose of the present subsection. In 3 + 1-dimensions, the Einstein-Vlasov-non-linear scalar field system can be written R − T + (trT )g = 0, (23) αβ αβ αβ α ′ ∇ ∇ φ − V ◦ φ = 0, (24) Lf = 0; (25) cf. [31, (7.13)–(7.15), p. 91]. In these equations, φ ∈ C (M) is referred to as the scalar field; V : R → R is a smooth function referred to as the potential; ∇ is the Levi-Civita connection associated with the metric g; and sf Vl T = T + T , αβ αβ αβ Vl where T is defined in (4) and sf γ T = ∇ φ∇ φ − ∇ φ∇ φ + V (φ) g . α β γ αβ αβ 10 ′ Assuming V to be such that V (0) = 0, it is consistent to demand that φ be zero in (24). Moreover, sf if φ = 0, then T = −V (0)g. Letting Λ = V (0), the equations (23)–(25) then reduce to the Einstein-Vlasov system with a positive cosmological constant Λ, assuming V (0) > 0. In order to prove future stability in the Einstein-Vlasov-non-linear scalar field setting, it is not sufficient ′ ′′ to demand that V (0) = 0 and V (0) > 0. It is also of interest to know that V (0) > 0. We shall therefore make this assumption from now on. Given V such that V (0) = 0, V (0) > 0 and ′′ V (0) > 0, it is convenient to introduce 1/2 H = (V (0)/3) (26) and ′′ 2 χ = V (0)/H ; (27) cf. [31, (7.9) and (7.10), p. 90]. Note that in the non-linear scalar field setting, we always assume V (0) to be positive and we equate it with Λ. In particular, (26) is thus consistent with previous definitions of H; cf., e.g, the statement of Theorem 7. In case we are interested in the Einstein- Vlasov system with a positive cosmological constant Λ, it is sufficient to choose V to be V (φ) = Λ + Λφ . (28) ′ ′′ 1/2 Then V (0) = Λ > 0, V (0) = 0 and V (0) = 2Λ > 0. Moreover, H = (Λ/3) and χ = 6. Clearly, (28) is an arbitrary choice; there are many other possibilities. Let us now recall the definition of initial data given in [31, Definition 7.11, pp. 93–94] (note that the dimension n is here assumed to equal 3): Definition 25. Let 5/2 < µ ∈ R. Initial data for (23)–(25) consist of an oriented 3-dimensional ¯ ¯ manifold Σ, a non-negative function f ∈ D (T Σ), a Riemannian metric g¯, a symmetric covariant ¯ ¯ ¯ 2-tensor field k and two functions φ and φ on Σ, all assumed to be smooth and to satisfy 0 1 ij 2 2 Vl ¯ ¯ ¯ ¯ ¯ ¯ ¯ ∇ φ ∇ φ + 2V (φ ) + 2ρ , (29) r¯− k k + (trk) = φ + ij 0 i 0 0 Vl ¯ ¯ ¯ ¯ ¯ ∇ k − ∇ (trk) = φ ∇ φ − J , (30) ji i 1 i 0 where ∇ is the Levi-Civita connection of g¯, r¯ is the associated scalar curvature, indices are raised Vl Vl and lowered by g¯ and ρ and J are given by (33) and (34) below respectively. Given initial data, the initial value problem is that of finding a solution (M, g, f, φ) to (23)–(25) (in other words, an 4- dimensional manifold M, a smooth time oriented Lorentz metric g on M, a non-negative function ∞ ∞ f ∈ D (P) and a φ ∈ C (M) such that (23)–(25) are satisfied), and an embedding i : Σ → M such that ∗ ∗ −1 ¯ ¯ i g = g¯, φ ◦ i = φ , f = i (f ◦ pr ) i(Σ) and if N is the future directed unit normal and κ is the second fundamental form of i(Σ), then ¯ ¯ i κ = k and (Nφ) ◦ i = φ . Such a quadruple (M, g, f, φ) is referred to as a development of the initial data, the existence of an embedding i being tacit. If, in addition to the above conditions, i(Σ) is a Cauchy hypersurface in (M, g), the quadruple is said to be a globally hyperbolic development. Remark 26. The map pr is the diffeomorphism from the mass shell above i(Σ) to the tangent i(Σ) space of i(Σ) defined by mapping a vector v to its component perpendicular to the normal of i(Σ). ¯ ¯ Remark 27. If φ = φ = 0, the equations (29) and (30) become 0 1 ij 2 Vl ¯ ¯ ¯ r¯− k k + (trk) = 2Λ + 2ρ , (31) ij Vl ¯ ¯ ¯ ∇ k − ∇ (trk) = −J . (32) ji i These are the constraint equations for the Einstein-Vlasov system with a positive cosmological constant Λ. 11 The energy density and current induced by the initial data are given by Vl 1/2 ¯ ¯ ρ (ξ) = f(p¯)[1 + g¯(p¯, p¯)] µ¯ , (33) g¯,ξ ¯Σ Vl ¯ ¯ ¯ ¯ J (X) = f(p¯)g¯(X, p¯)µ¯ ¯. (34) g¯,ξ T¯Σ ¯ ¯ In these expressions, ξ ∈ Σ, X ∈ T¯Σ, µ¯ ¯ is the volume form on T¯Σ induced by g¯ and p¯ ∈ T¯Σ. ξ g¯,ξ ξ ξ It is important to note that under the assumptions of the above definition, the energy density is a smooth function and the current is a smooth one-form field on Σ; cf. [31, Lemma 15.40, p. 246]. Given initial data, there is a unique maximal globally hyperbolic development thereof; cf. [31, Corollary 23.44, p. 418] and [31, Lemma 23.2, p. 398]. The definition of a maximal globally hyperbolic development is given by [31, Definition 7.14, p. 94]: Definition 28. Given initial data for (23)–(25), a maximal globally hyperbolic development of the data is a globally hyperbolic development (M, g, f, φ), with embedding i : Σ → M, such that ′ ′ ′ ′ if (M , g , f , φ ) is any other globally hyperbolic development of the same data, with embedding ′ ′ ′ i : Σ → M , then there is a map ψ : M → M which is a diffeomorphism onto its image such that ∗ ′ ∗ ′ ∗ ′ ′ ψ g = g , ψ f = f , ψ φ = φ and ψ ◦ i = i. It is worth noting that the maximal globally hyperbolic development is independent of the param- eter µ . The above discussion of the initial value problem for the Einstein-Vlasov-non-linear scalar field system is somewhat brief, and the reader interested in a more detailed discussion is referred to [31, Chapter 7]. 1.5 Future stability in the spatially homogeneous and isotropic setting In the proof of stability of the T -Gowdy symmetric solutions, we need to refer to [31, Theo- rem 7.16, pp. 104–106]. However, the statement of this theorem is based on terminology introduced in [31]. Moreover, in the statement of Theorem 35, we refer to the conclusions of [31, Theorem 7.16]. For this reason, we here provide not only the notational background, but also the statement of [31, Theorem 7.16]. However, the reader interested in a discussion giving a justification for why the particular formulation of the theorem is natural is referred to [31, Sections 7.6–7.7]. The rough idea of the statement is to only make local assumptions concerning the initial data and to derive future global conclusions concerning the solution. Given a 3-manifold Σ, we therefore focus on a local coordinate patch (¯x, U). Here U is the neighbourhood in which we make assumptions in the statement of the theorem. The conditions on the initial data are phrased in terms of Sobolev norms on U. Given a tensor field T on Σ, we therefore define   1/2 3 3 X X X α i ···i −1 2 1 s   (35) kTk l = |∂ T ◦ ¯x | dx¯ . H (U) j ···j 1 r ¯ x(U) i ,...,i =1 j ,...,j =1 1 s 1 r |α|≤l In this expression, the components of T are computed with respect to the coordinates ¯x and the derivatives are taken with respect to ¯x. In what follows, norms of the type kTk l are always H (U) computed using a particular choice of local coordinates. The choice we have in mind should be clear from the context. In the formulation of Theorem 29, we also use the notation   1/2 X X α −1 2   k∂ g¯k l = |∂ ∂ g¯ ◦ ¯x | dx¯ . (36) m m ij H (U) ¯ x(U) i,j=1 |α|≤l To measure the local size of the distribution function, we need a weighted Sobolev norm. However, it is also necessary to allow the freedom to rescale the momentum variable in the definition of the 12 norm. Since we have already motivated the need for this rescaling freedom in [31, Subsection 7.6.1, pp. 100–102], we shall not do so here. Given a constant w, we simply define the local norm for the distribution function by   1/2 Z Z w −w 2|β| w 2µ +2|β| α β 2 ¯  ¯ ¯ ¯  (37) kfk l = (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯ . ¯ ¯ x H (U) p¯ Vl,μ R ¯ x(U) |α|+|β|≤l −1 ¯ ¯ ¯ ¯ Here Ξ are the coordinates on TU associated with ¯x (cf. Subsection 1.3), and f = f ◦ Ξ . ¯ x ¯ x ¯ x Given the above notation, [31, Theorem 7.16, pp. 104–106] takes the following form for n = 3. Theorem 29. Let 5/2 < µ ∈ R and 7/2 < k ∈ Z. Let V be a smooth function on R such that ′ ′′ V (0) = V > 0, V (0) = 0 and V (0) > 0. Let H, χ > 0 be defined by (26) and (27) respectively and let K ≥ 0. There is an ε > 0, depending only on µ and V , such that if Vl ¯ ¯ ¯ ¯ • (Σ, g¯, k, f, φ , φ ) are initial data for (23)–(25) with dimΣ = 3, 0 1 ¯ ¯ • x : U → B (0) are local coordinates with x(U) = B (0), 1 1 • the inequality −2K |e g¯ − δ | ≤ ε (38) ij ij holds on U for all i, j = 1, ..., n, where K is defined by e = 4/H, • using the notation introduced in (35) and (36), the inequality 2 −1 −K ¯ ¯ ¯ Vl H k∂ g¯k k + Hkk − Hg¯k k + kφ k k +1 + H kφ k k ≤ εe j 0 0 0 0 1 0 (39) H (U) H (U) H (U) H (U) j=1 holds, • using the notation introduced in (37), the inequality w 2 5/2 −3K/2−K ¯ Vl kfk k ≤ H ε e (40) H (U) Vl,μ holds with w = K + K , Vl then the maximal globally hyperbolic development (M, g, f, φ) of the initial data has the property −1 that if i : Σ → M is the associated embedding, then all causal geodesics that start in i◦¯x [B (0)] 1/4 are future complete. Furthermore, there is a t < 0 and a smooth map ψ : (t , ∞) × B (0) → M, (41) − 5/8 −1 which is a diffeomorphism onto its image, such that all causal curves that start in i ◦ ¯x [B (0)] 1/4 remain in the image of ψ to the future, and g, f and φ have expansions of the form (42)–(55) −1 ¯ ¯ in the solid cylinder [0, ∞) × B (0) when pulled back by ψ. Finally, ψ(0, ξ) = i ◦ ¯x (ξ) for 5/8 ξ ∈ B (0). In the formulae below, Latin indices refer to the natural Euclidean coordinates on 5/8 B (0) and t is the natural time coordinate on the solid cylinder. Let ζ = 4χ/9, 5/8 3 1/2 [1 − (1 − ζ) ] ζ ∈ (0, 1) λ = pre ζ ≥ 1 and λ = min{1, λ }. There is a smooth Riemannian metric ̺¯ on B (0) and, for every l ≥ 0, m pre 5/8 a constant K such that 2Ht+2K ij ij −2Ht−2K −2λ Ht ke g (t, ·) − ̺¯ k l + ke g (t, ·) − ̺¯ k l ≤ K e , (42) ij ij l C C −2Ht−2K −2λ Ht ke ∂ g (t, ·) − 2H̺¯ k l ≤ K e , (43) t ij ij l 13 ij l for every l ≥ 0 and t ≥ 0. Here ̺¯ denotes the components of the inverse of ̺¯. Furthermore, C denotes the C -norm on B (0). Turning to g , there is a b > 0 and, for every l ≥ 0, a constant 5/8 0m K such that −bHt kg (t, ·) − v k + k∂ g (t, ·)k l ≤ K e , (44) 0m m 0 0m C l for all l ≥ 0 and t ≥ 0, where ij v = ̺¯ γ (45) m imj and γ denote the Christoffel symbols of the metric ̺¯, given by imj γ = (∂ ̺¯ + ∂ ̺¯ − ∂ ̺¯ ). imj i jm j im m ij Let k denote the components of the second fundamental form (induced on the constant-t hyper- ij surfaces) with respect to the standard coordinates on B (0). If λ < 1, there is, for every l ≥ 0, 5/8 m a constant K such that −2λ Ht kg (t, ·) + 1k l + k∂ g (t, ·)k l ≤ K e , 00 C 0 00 C l −2Ht−2K −2λ Ht ¯ m ke k (t, ·) − H̺¯ k l ≤ K e ij ij C l for every l ≥ 0 and t ≥ 0. If λ = 1, there is, for every l ≥ 0, a constant K such that m l −2Ht k[∂ g + 2H(g + 1)](t, ·)k l ≤ K e , 0 00 00 C l 2 1/2 −2Ht kg (t, ·) + 1k l ≤ K (1 + t ) e , 00 C l −2Ht−2K 2 1/2 −2Ht ke k (t, ·) − H̺¯ k l ≤ K (1 + t ) e ij ij C l λ Ht pre for every l ≥ 0 and t ≥ 0. In order to describe the asymptotics concerning φ, let ϕ = e φ. If ζ < 1, there is a smooth function ϕ , a constant b > 0 and, for every l ≥ 0, a constant K such 0 l that −bHt kϕ(t, ·) − ϕ k l + k∂ ϕk l ≤ K e (46) 0 C 0 C l for all l ≥ 0 and t ≥ 0. If ζ = 1, there are smooth functions ϕ and ϕ , a constant b > 0 and, for 0 1 every l ≥ 0, a constant K such that −bHt k∂ ϕ(t, ·) − ϕ k l + kϕ(t, ·) − ϕ t − ϕ k l ≤ K e (47) 0 1 1 0 l C C for all l ≥ 0 and t ≥ 0. Finally, if ζ > 1, there is an anti symmetric matrix A, given by 0 δH A = , −δH 0 1/2 where δ = 3(ζ − 1) /2, smooth functions ϕ and ϕ , a constant b > 0 and, for every l ≥ 0, a 0 1 constant K such that δHϕ ϕ −At 0 −bHt e (t, ·) − ≤ K e (48) ∂ ϕ ϕ 0 1 l for all l ≥ 0 and t ≥ 0. In order to describe the asymptotics for the distribution function, let −1 x = ψ . Then (x, U) are canonical local coordinates, where U = ψ[(t , ∞) × B (0)]. − 5/8 −1 Let f = f ◦ Ξ and −2Ht−K−K Vl h(t, x¯, q¯) = f (t, x¯, e q¯). (49) Introduce, moreover, the notation   1/2 Z Z 2µ +2|β| α β 2 ¯  ¯  kfk l = hp¯i |∂ ∂ f(x¯, p¯)| dp¯dx¯ H [B (0)×R ] x ¯ p¯ 5/8 Vl,μ B (0) R 5/8 |α|+|β|≤l 14 ∞ 3 for f ∈ C [B (0) × R ]. Then there is a constant b > 0 and, for every l, a constant K such 5/8 l that −bHt k∂ h(t, ·)k l ≤ K e (50) t l H [B (0)×R ] 5/8 Vl,μ ∞ 3 holds for all l ≥ 0 and t ≥ 0. There is also a function h ∈ C [B (0) × R ], a constant b > 0 5/8 and, for every l, a constant K such that khk l < ∞, H [B (0)×R ] 5/8 Vl,μ −bHt kh(t, ·) − hk l 3 ≤ K e (51) H [B (0)×R ] 5/8 Vl,μ hold for all l ≥ 0 and t ≥ 0. Furthermore, h ≥ 0. Concerning the stress energy tensor associated with the Vlasov matter, there is a b > 0 and, for every l ≥ 0, a constant K such that the estimates 3(Ht+K ) Vl 1/2 −bHt Vl e T − h|̺¯| dq¯ ≤ K e , (52) 3 l R C 3(Ht+K ) Vl 1/2 −bHt Vl ¯ e T + q h|̺¯| dq¯ ≤ K e , (53) i l 0i 2Ht+3K Vl Vl e T ≤ K (54) ij l hold for all l ≥ 0 and all t ≥ 0, where |̺¯| denotes the absolute value of the determinant of ̺¯, K−K j Vl ¯ ¯ q = v + e ̺¯ q¯ i i ij and v is defined in (45). Finally, if µ > 9/2, there is a constant b > 0 and, for every l ≥ 0, a constant K such that 3(Ht+K ) Vl 1/2 −bHt Vl ¯ ¯ ¯ e T − hq q |̺¯| dq¯ ≤ K e (55) i j l ij holds for all l ≥ 0 and t ≥ 0. Remark 30. In case one is only interested in the Einstein-Vlasov setting with a positive cosmolog- ical constant, more detailed information can be obtained; cf. [31, Proposition 32.8, pp. 609–611]. 1.6 Cauchy stability In what follows, we also need a Cauchy stability result in the Einstein-Vlasov-non-linear scalar field setting. There are such results in the literature; cf. [31]. However, for the convenience of the reader, we introduce the necessary terminology and quote the relevant result in the present subsection. To begin with, we need to introduce the notion of a background solution; cf. [31, Definition 24.2, p. 421]. In the 3-dimensional case, this definition takes the following form. Definition 31. Let 5/2 < µ ∈ R, Σ be a closed 3-dimensional manifold, and let g be a smooth time oriented Lorentz metric on M = I × Σ, where I is an open interval. Let ∂ denote differentiation with respect to the first coordinate and assume that g(∂ , ∂ ) = g < 0 and that the hypersurfaces t t 00 Σ = {t} × Σ are spacelike with respect to g for t ∈ I. Finally, assume that φ ∈ C (M) and f ∈ D (P), together with g, satisfy (23)–(25). Then (M, g, f, φ) is called a background solution. Remark 32. In the case of T -symmetric solutions, the metric is of the form (1). Moreover, the distribution functions of interest have compact support on constant time hypersurfaces. As a consequence, it is clear that the T -symmetric solutions we consider in the present paper are background solutions in the sense of the above definition. Next, we introduce the notion of induced initial data on constant t hypersurfaces; cf. [31, Defini- tion 24.3, p. 421]. In the 3-dimensional case, this definition takes the following form. 15 Definition 33. Let 5/2 < µ ∈ R, Σ be a closed 3-dimensional manifold, and let g be a smooth time oriented Lorentz metric on M = I×Σ, where I is an open interval. Let, furthermore, φ ∈ C (M), f ∈ D (P) and assume that (g, f, φ) solve (23)–(25). Let t ∈ I and assume Σ = {t} × Σ to be spacelike with respect to g. Let, furthermore, κ be the second fundamental form and N be the future directed unit normal of Σ . Finally, let ι : Σ → M be defined by ι (x¯) = (t, x¯) and t t t ∗ ∗ ∗ −1 ∗ ∗ ¯ ¯ ¯ ¯ g¯ = ι g, k = ι κ, f = ι (f ◦ pr ), φ = ι φ, φ = ι (Nφ). 0 1 t t t Σ t t ¯ ¯ ¯ ¯ Then (g¯, k, f, φ , φ ) are referred to as the initial data induced on Σ by (g, f, φ), or simply the 0 1 t initial data induced on Σ if the solution is understood from the context. Finally, we are in a position to formulate the Cauchy stability result we need here; cf. [31, Corollary 24.10, p. 432]. In the 3-dimensional case, this result takes the following form. Theorem 34. Let 5/2 < µ ∈ R and 5/2 < l ∈ Z. Let (M , g , f , φ ) be a background bg bg bg bg solution with M = I × Σ and recall the notation Σ, Σ etc. from Definition 31 (the interval bg bg t which was denoted by I in Definition 31 will here be denoted by I ). Assume that 0 ∈ I and bg bg ¯ ¯ ¯ ¯ let (g¯ , k , f , φ , φ ) be the initial data induced on Σ by (g , f , φ ). Make a choice bg bg bg bg,0 bg,1 0 bg bg bg of H (T Σ)-norms and a choice of Sobolev norms k · k l on tensor fields on Σ. Let J ⊂ I be bg Vl,µ H ¯ ¯ ¯ ¯ a compact interval and let ǫ > 0. Then there is a δ > 0 such that if (Σ, g¯, k, f, φ , φ ) are initial 0 1 data for the Einstein-Vlasov-non-linear scalar field system satisfying ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ kg¯ − g¯ k l+1 + kk − k k l + kφ − φ k l+1 + kφ − φ k l + kf − f k l ≤ δ, bg H bg H 0 bg,0 H 1 bg,1 H bg H (T Σ) Vl,μ then there is an open interval I containing 0 and a solution (g, f, φ) to (23)–(25) on M = I × Σ such that ¯ ¯ ¯ ¯ • the initial data induced on Σ by (g, f, φ) are given by (g¯, k, f, φ , φ ), 0 0 1 • ∂ is timelike with respect to g and Σ is a spacelike Cauchy hypersurface with respect to g t t for all t ∈ I, • J ⊂ I and if the initial data induced on Σ (for t ∈ I ∩ I) by (g, f, φ) and (g , f , φ ) t bg bg bg bg ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ are denoted by (g¯ , k , f , φ , φ ) and (g¯ , k , f , φ , φ ) respectively, then t t t t,0 t,1 bg,t bg,t bg,t bg,t,0 bg,t,1 ¯ ¯ ¯ ¯ kg¯ − g¯ k l+1 + kk − k k l + kφ − φ k l+1 t bg,t H t bg,t H t,0 bg,t,0 H (56) ¯ ¯ ¯ ¯ +kφ − φ k l + kf − f k l ≤ ǫ t,1 bg,t,1 H t bg,t H (T Σ) Vl,μ for all t ∈ J. 1.7 Stability of T -Gowdy symmetric solutions Combining Theorems 7, 29 and 34 yields a future stability result for the T -symmetric solutions considered in Theorem 7. Moreover, the solutions are stable in the Einstein-Vlasov-non-linear scalar field setting. Theorem 35. Consider a T -symmetric solution to the Einstein-Vlasov system with a positive cosmological constant Λ. Choose coordinates so that the corresponding metric takes the form (1) on I × T , where I = (t , ∞). Assume that the solution has λ-asymptotics. Choose a t ∈ I and 3 3 ∗ let i : T → I × T be given by i(x¯) = (t, x¯). Let g¯ = i g and let k denote the pullback (under bg bg i) of the second fundamental form induced on i(T ) by g. Let, moreover, ∗ −1 f = i (f ◦ pr ). bg 3 i(T ) Make a choice of µ > 5/2, a choice of norms as in Definition 23 and a choice of Sobolev norms on tensorfields on T . Let, in addition, V : R → R be a smooth function such that V (0) = Λ, 16 ′ ′′ 3 ¯ ¯ ¯ ¯ V (0) = 0 and V (0) > 0. Then there is an ǫ > 0 such that if (T , g¯, k, f, φ , φ ) are initial data 0 1 ∞ 3 ¯ ¯ for (23)–(25), with f ∈ D (TT ), satisfying ¯ ¯ ¯ ¯ ¯ ¯ kg¯ − g¯ k 5 + kk − k k 4 + kf − f k 4 + kφ k 5 + kφ k 4 ≤ ǫ, bg H bg H bg H 0 H 1 H Vl,μ then the maximal globally hyperbolic development (M, g, f, φ) of the initial data is future causally geodesically complete. Moreover, there is a Cauchy hypersurface Σ in (M, g) such that for each point of Σ, there is a neighbourhood (¯x, U) such that Theorem 29 applies. In particular, the asymptotics stated in Theorem 29 thus hold. Remark 36. Up to the point where we appeal to Theorem 29, Cauchy stability applies. It should thus be possible to obtain detailed control over the perturbed solutions for the entire future. The interested reader is encouraged to write down the details. ¯ ¯ Remark 37. The function f has compact support, but f need not have compact support. bg The proof is to be found in Section 10. 1.8 Outline Finally, let us give an outline of the paper. In Section 2, we write down the equations in the case that the metric takes the form (1) (though the reader interested in a derivation is referred to Appendix A). In Section 3, we then collect the conclusions which are not dependent on the particular type of matter model (as long as it satisfies the dominant energy condition and the non- negative pressure condition). The section ends with conclusions concerning the causal structure of T -Gowdy symmetric spacetimes. Turning to the more detailed conclusions, we specialise to the case of Vlasov matter. The natural first step is to derive light cone estimates; i.e., to consider the behaviour along characteristics. This is the subject of Section 4. As opposed to the vacuum case, we need to control the characteristics associated with the Vlasov equation at the same time as the first derivatives of the metric components. Fortunately, the e - and e -components 2 3 of the momentum are controlled automatically due to the symmetry. However, an argument is required in the case of the e -component. In order to obtain control of higher order derivatives, we need to take derivatives of the characteristic system (associated with the Vlasov equation; i.e. with the geodesic flow). Naively, this should require control of second order derivatives of the metric functions, something we do not have. Nevertheless, by an appropriate choice of variables, controlling first order derivatives turns out to be sufficient. It is of interest to note that a similar choice was already suggested in [2, Lemma 3, p. 363]; cf. also [3, Lemma 3, p. 257]. However, in the present setting, it is not sufficient to derive a system involving only first order derivatives of the metric functions. We also need to be able to use the system to derive the desired type of asymptotics for the derivatives of the characteristic system. It turns out to be possible to do this, and we write down the required arguments in Section 6. After we have obtained this conclusion, it turns out to be possible to proceed inductively in order to derive higher order estimates for the characteristic system and the metric components. The required arguments are written down in Sections 7 and 8. In order to obtain the desired conclusions concerning the distribution function, it turns out to be convenient to consider L -based energies. This subject is treated in Section 9. Finally, in Section 10, we prove the main theorems of the paper. As an appendix to the paper, we include a derivation of Einstein’s equations as well as of the Vlasov equation; cf. Appendix A. We also provide a summary of the most important notation in Appendix B. 2 Symmetry assumptions and equations In this paper, we study T -symmetric solutions of Einstein’s equations. Since it will turn out to be convenient to express the equations using the orthonormal frame (7), let us introduce the notation ρ = T (e , e ), J = −T (e , e ), P = T (e , e ), S = T (e , e ), (57) 0 0 i 0 i i i i ij i j 17 where we do not sum over any indices; here, and below, we tacitly assume Latin indices to range from 1 to 3 and Greek indices to range from 0 to 3. It is also convenient to introduce the notation 5/2 1/2 P−λ/2 5/2 1/2 −P−λ/2 J = −t α e (G + QH ), K = QJ − t α e H . (58) t t t Note that these objects are the twist quantities introduced in (2); cf. Appendix A.3. In order to derive Einstein’s equations, it is useful to calculate the Einstein tensor for a metric of the form (1). The corresponding, somewhat lengthy, computations are to be found in Section A. Using the above notation, the calculations yield the conclusion that the 00 and 11-components of Einstein’s equations can be written λ/2−P 2 λ/2+P 2 α e J e (K − QJ) 2 2 2P 2 2 λ − 2 = t P + αP + e (Q + αQ ) + + (59) t θ t θ 5/2 5/2 α t t 1/2 λ/2 +4t e (ρ + Λ), λ/2−P 2 λ/2+P 2 e J e (K − QJ) 2 2 2P 2 2 λ = t P + αP + e (Q + αQ ) − − (60) t θ t θ 5/2 5/2 t t 1/2 λ/2 +4t e (P − Λ), respectively. The 22-component minus the 33-component can be written −1/2 λ/2−P 2 α e J −1/2 1/2 −1/2 2P 2 2 ∂ (tα P ) =∂ (tα P ) + tα e (Q − αQ ) + t t θ θ t θ 5/2 2t (61) −1/2 λ/2+P 2 α e (K − QJ) 1/2 λ/2 −1/2 − + t e α (P − P ). 2 3 5/2 2t The 22-component plus the 33-component can be written α 3 −1/2 1/2 −1/2 2 2P 2 2 2P 2 ∂ tα λ − 2 − =∂ tα λ − tα P + e Q − α(P + e Q ) t t θ θ t t θ θ α t λ/2−P 2 λ/2+P 2 e J e (K − QJ) (62) −1/2 − 2tα + 7/2 7/2 t t −1/2 1/2 λ/2 −1/2 + α λ + 2t e α (2Λ − P − P ). t 2 3 The 01, 02, 03, 12 and 13-components are equivalent to 2P 1/2 λ/2 −1/2 λ = 2t(P P + e Q Q ) − 4t e α J , (63) θ t θ t θ 1 5/4 −1/2 P/2+λ/4 J = 2t α e J , (64) θ 2 5/4 −1/2 −P/2+λ/4 5/4 −1/2 P/2+λ/4 K = 2t α e J + 2t α e QJ , (65) θ 3 2 5/4 λ/4+P/2 J = −2t e S , (66) t 12 5/4 λ/4+P/2 5/4 −P/2+λ/4 K = −2t e QS − 2t e S , (67) t 12 13 respectively. Finally, the 23-component reads −1/2 2P 1/2 2P −5/2 −1/2 λ/2+P 1/2 −1/2 λ/2+P ∂ (tα e Q )− ∂ (tα e Q ) = t α e J(K − QJ) + 2t α e S . (68) t t θ θ 23 For future reference, it is also of interest to note that −P +λ/2 2 P +λ/2 2 α e J e (K − QJ) 1/2 λ/2 1/2 λ/2 = − − − 4t e Λ − 2t e (ρ − P ), (69) 5/2 5/2 t t 2 2 2P 2 2 1/2 λ/2 λ − = t P + αP + e (Q + αQ ) + 2t e (ρ + P ). (70) t 1 t θ t θ 18 2.1 Preliminary calculations Since the metric components only depend on two variables, it is natural to derive estimates by integrating along characteristics. In the present subsection, we record a general calculation which is of interest in that context. To begin with, let us define 1/2 2 2P 2 ∂ = ∂ ± α ∂ , A = (∂ P ) + e (∂ Q) . (71) ± t θ ± ± ± One reason for introducing A is the equality (72) derived below; since the right hand side only contains first derivatives of the metric components, it is possible to integrate along the characteristics to control A . Lemma 38. Consider a T -symmetric solution to Einstein’s equations with a cosmological con- stant Λ such that the metric takes the form (1). Then 2 α 2 1/2 2P ∂ A = − − A ∓ α (P ∂ P + e Q ∂ Q) ± ∓ ∓ θ ∓ θ ∓ t α t −P +λ/2 2 P +λ/2 2 λ/2 e (K − QJ) e J(K − QJ) (72) e J + ∂ P − ∂ P + 2 e ∂ Q ∓ ∓ ∓ 7/2 7/2 7/2 t t t −1/2 λ/2 −1/2 λ/2 P + 2t e (P − P )∂ P + 4t e S e ∂ Q. 2 3 ∓ 23 ∓ Remark 39. In this calculation, the cosmological constant need not be positive. Proof. The statement follows from a lengthy computation. Let us, however, for the benefit of the reader, write down some of the intermediate steps. Using (61), we obtain 1 α 2P 2 2 ∂ ∂ P = − P + ∂ P + e (Q − αQ ) ± ∓ t ∓ t θ t 2α (73) −P +λ/2 2 P +λ/2 2 e J e (K − QJ) −1/2 λ/2 + − + t e (P − P ). 2 3 7/2 7/2 2t 2t Similarly, due to (68), we obtain λ/2−P 1 α e J(K − QJ) −1/2 λ/2−P ∂ ∂ Q = − Q + ∂ Q − 2(Q P − αQ P ) + + 2t e S . (74) ± ∓ t ∓ t t θ θ 23 7/2 t 2α Combining (73) and (74) with the fact that 2 2 2 −4(Q P − αP Q )∂ Q + 2∂ P (∂ Q) = −2∂ P (Q − αQ ), t t θ θ ∓ ± ∓ ∓ t θ a calculation yields the conclusion of the lemma. 2.2 Vlasov matter The equations (59)–(70) hold in general. However, we are here particularly interested in matter of Vlasov type. In order to derive the relevant form of the Vlasov equation, recall the conventions concerning f introduced in Subsection 1.2. Recall, moreover, the fact that the Vlasov equation is equivalent to f being constant along future directed unit timelike geodesics. As a consequence, it can be calculated (cf. Appendix A.7) that the Vlasov equation takes the form 1/2 1 2 3 ∂f α v ∂f 1 1 2α 1 v v 1/2 0 1 1/2 P + − α λ v + λ − − v − α e Q θ t θ 0 0 ∂t v ∂θ 4 4 α t v 3 2 2 2 1 (v ) − (v ) ∂f 1/2 −7/4 λ/4 −P/2 2 P/2 3 + α P − t e e Jv + e (K − QJ)v 0 1 2 v ∂v 1 2 1 1 1 v v ∂f 2 1/2 − P + v + α P t θ 0 2 2 t 2 v ∂v 1 3 1 1 1 1 v v v ∂f 3 1/2 P 2 1/2 − P v − α P + e v Q + α Q = 0. t θ t θ 0 0 3 2 t 2 v v ∂v (75) 19 Turning to the stress energy tensor, it satisfies T (e , e ) = v v f dv, (76) µ ν µ ν 3 −v where v = η v and η = diag{−1, 1, 1, 1}. In particular, in the case of Vlasov, we thus have α αβ Z Z Z Z k 2 j k (v ) v v 0 k ρ = v f dv, P = f dv, J = v f dv, S = f dv, (77) k k jk 0 0 3 3 v 3 3 v R R R R where j, k = 1, 2, 3. 3 Preliminary conclusions concerning the asymptotics In the present section, we are interested in T -symmetric solutions to Einstein’s equations such that the corresponding metric admits a foliation of the form (1) on I × T , where I = (t , ∞) and t ≥ 0. For the sake of brevity, we shall below refer to solutions of this form as future global, and we shall speak of t and t = t + 2 without further introduction. 0 1 0 It is useful to begin by recalling the following consequences of the non-negative pressure condition and the dominant energy condition. Lemma 40. Consider a solution to Einstein’s equations with a cosmological constant Λ and a metric of the form (1). Let ρ, P , J and S , i, j = 1, 2, 3, be defined by (57). If the stress energy i i ij tensor satisfies the non-negative pressure condition, then, for i = 1, 2, 3, 0 ≤ P . (78) If the stress energy tensor satisfies the dominant energy condition, then, for i, j = 1, 2, 3, 0 ≤ ρ, (79) |P | ≤ ρ, (80) |J | ≤ ρ, (81) |S | ≤ ρ. (82) ij Proof. By definition, P = T (e , e ). Since e is a spacelike vector field, the non-negative pressure i i i i condition implies that (78) holds. The dominant energy condition states that T (u, v) ≥ 0 for future directed timelike vectors u and v. By continuity, this inequality also holds for future directed causal vectors. Since e is future directed timelike, ρ = T (e , e ) ≥ 0, so that (79) follows. Note that 0 0 0 e ± e is a future directed causal vector field. In particular, 0 i 0 ≤ T (e − e , e + e ) = ρ + J − J − S . 0 i 0 j i j ij Since S is symmetric, adding this inequality with the one obtained by interchanging i and j ij yields the conclusion that S ≤ ρ. Similarly, ij 0 ≤ T (e ± e , e ± e ) = ρ ∓ J ∓ J + S . 0 i 0 j i j ij Adding the two inequalities yields −S ≤ ρ. Thus (82) holds. The proof of (80) is similar. Finally, ij 0 ≤ T (e , e ± e ) = ρ ∓ J , 0 0 i i so that (81) holds. Before deriving estimates describing the asymptotics of solutions, let us make the following remark. 20 Remark 41. In what follows, the constants appearing in the estimates we state are allowed to depend on the solution, unless otherwise indicated. Proposition 42. Given a future global solution to Einstein’s equations with a cosmological con- stant Λ > 0, T -symmetry and a stress energy tensor satisfying the dominant energy condition, there is a constant C > 0 such that −3 α(t, θ) ≤ Ct (83) for all (t, θ) ∈ [t , ∞) × S . Remark 43. The same conclusion holds if we replace the cosmological constant with a non- linear scalar field with a potential with a positive lower bound; in other words, if we set Λ = 0 o sf o and consider stress energy tensors of the form T = T + T , where T is the stress energy tensor sf associated with matter fields satisfying the dominant energy condition, and T is the stress energy tensor associated with a non-linear scalar field with a potential V having a positive lower bound. Proof. Due to (70) and the fact that the matter satisfies the dominant energy condition (so that (80) holds), we conclude that λ − α /α ≥ 0. There is thus a c > 0 such that t t 0 −1/2 λ/2 (α e )(t, θ) ≥ c for all (t, θ) ∈ [t , ∞) × S . Combining this observation with (69) and (80), we obtain −1/2 −1/2 1/2 −1/2 λ/2 1/2 ∂ α = − α ≥ 2t α e Λ ≥ c t t 1 2α for some constant c > 0 and all (t, θ) ∈ [t , ∞) × S . Integrating this inequality, we obtain the 1 1 conclusion of the proposition. In the Gowdy case, the second and third terms on the right hand side of (60) are zero, and as a consequence, we can extract more information. In fact, we have the following observation. Proposition 44. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -Gowdy symmetry and matter satisfying the non-negative pressure condition. Then there is, for every ǫ > 0, a T > t such that λ(t, θ) ≥ −3 lnt + 2 ln − ǫ 4Λ for all (t, θ) ∈ [T, ∞) × S . Proof. Let λ = λ + 3 lnt − 2 ln . (84) 4Λ Then (60) with J = K = 0 yields 2 2 2P 2 2 1/2 λ/2 λ/2 ∂ λ = t P + αP + e (Q + αQ ) + 4t e P + (1 − e ). t t θ 1 t θ Since P ≥ 0 due to the non-negative pressure condition, cf. (78), we conclude that λ/2 ∂ λ ≥ (1 − e ). For every ǫ > 0, there is thus a T such that λ(t, θ) ≥ −ǫ for all (t, θ) ∈ [T, ∞)×S . The proposition follows. 21 In order to proceed, it is convenient to introduce an energy: −1/2 1/2 λ/2 E = tα λ − 2 − 4t e Λ dθ bas t 1 α −1/2 λ/2−P 2 α e J 2 −1/2 2 2 2P 2 2 = t α P + αP + e (Q + αQ ) + (85) t θ t θ 3/2 1 t −1/2 λ/2+P 2 α e (K − QJ) 3/2 −1/2 λ/2 + + 4t α e ρ dθ. 3/2 Let us motivate this particular choice. The energy E is quite similar to the energy defined bas in [3, (42), p. 251]. However, there is one fundamental difference. The integrand in the energy −1/2 2 defined in [3] contains a term of the form α U , where U = (P + ln t)/2. Using U instead of P as a variable is convenient in global existence arguments, since some of the formulae become less involved. However, the variable U is poorly adapted to the actual asymptotics of solutions. The −2 reason for this is that, in the end, it turns out that P converges to zero as t . The dominant term in U is thus ln t/2. Using U instead of P in the energy, the best estimate of E one could bas 3/2 1/2 hope for would be E (t) ≤ Ct . Below, we prove that E ≤ Ct ; cf. Lemma 46. In addition bas bas to this, it is possible to derive a good estimate for the time derivative of E ; cf. the proof of bas Lemma 45 below. On a more general level, it is natural to ask why it is necessary to use L -based energy estimates at all. Since the problem is 1+1-dimensional, should it not be sufficient to consider the behaviour along characteristics? The problem in our setting is that we wish to derive detailed quantitative information for arbitrary initial data. In particular, we are not in a situation where we can use bootstrap arguments. For this reason, we need to proceed step by step. First, it is necessary to derive not only rough, but quite detailed, control of some of the metric components, in particular λ. This leads, for example, to estimates of the form (96) and (97). Only once we have estimates of this form is it meaningful to turn to the characteristic system; cf., e.g., the last two terms on the right hand side of (109) and the proof of Lemma 52. In case the metric has λ-asymptotics (recall Definition 1), it turns out to be possible to estimate E . bas Lemma 45. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then for every a > 1/2, there is a constant C > 0 such that E (t) ≤ C t (86) bas a for all t ≥ t . Proof. Due to (62), we obtain h  i −1/2 1/2 λ/2 1/2 1/2 2 2P 2 ∂ tα λ − 2 − 4t e Λ =∂ (tα λ ) + 2tα (P + e Q ) t t θ θ θ θ −1/2 λ/2−P 2 −1/2 λ/2+P 2 3 α e J 3 α e (K − QJ) − − 5/2 5/2 2 2 t t (87) 5/2 −1/2 λ/2 2 2P 2 2 2P 2 − 2t α e Λ[P + e Q + α(P + e Q )] t t θ θ 1/2 −1/2 λ/2 + t α e (3ρ + P − 2P − 2P ) 1 2 3 2 −1/2 λ − 4t α e Λ(ρ + P ). Since the matter satisfies the non-negative pressure condition, we know that P ≥ 0, cf. (78), so that Z Z dE bas 1/2 2 2P 2 5/2 1/2 λ/2 2 2P 2 ≤ 2tα (P + e Q )dθ − 2t α e Λ(P + e Q )dθ θ θ θ θ dt 1 1 S S Z Z 1/2 −1/2 λ/2 2 −1/2 λ + t α e (3ρ + P )dθ − 4t α e Λ(ρ + P )dθ. 1 1 1 1 S S 22 Using the consequences of Lemma 40 and the fact that the solution has λ-asymptotics, we conclude that for every a > 1/2, there is a T ≥ t such that dE a bas ≤ E bas dt t for all t ≥ T . As a consequence, E (t) ≤ Ct for t ≥ t . bas 1 Using the estimate for E derived in Lemma 45, it is possible to extract more information bas concerning the asymptotics. Lemma 46. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that −1/2 λ(t, ·) + 3 ln t − 2 ln ≤ Ct , (88) 4Λ 1/2 E (t) ≤ Ct (89) bas for all t ≥ t . a 1/2 −3/2 Proof. Due to the estimate E (t) ≤ C t , the fact that α ≤ Ct , (60) and (80), we conclude bas a that 1/2 λ/2 a−5/2 hλ i = −4t he iΛ + O(t ); (90) recall that the notation hλ i was introduced in Remark 5. Due to (63) and (81), we also have −1/2 2 2 2P 2 2 1/2 −1/2 λ/2 |λ | ≤ tα [P + αP + e (Q + αQ )] + 4t α e ρ. θ t θ t θ Due to (86), we thus obtain a−1 |λ |dθ ≤ Ct . (91) Recall that λ is defined in (84) and note that, due to (90), λ/2 a−5/2 hλ i = (1 − he i) + O(t ). (92) Let us first prove that hλi converges to zero. Let, to this end, ǫ > 0. Since the solution has λ-asymptotics, we know that there is a T such that hλi(t) ≥ −ǫ for all t ≥ T . In order to prove ˆ ˆ that there is a T such that hλi(t) ≤ ǫ for all t ≥ T , let us assume that hλi(t) ≥ ǫ for some t. Due to (91), we conclude that λ(t, θ) ≥ ǫ/2 for all θ ∈ S (assuming t to be large enough). Inserting this information into (92), we conclude that ǫ/4 hλ i ≤ (1 − e ), assuming t to be large enough. Since the right hand side is negative and non-integrable, we conclude that hλi has to decay until it is smaller than ǫ (assuming the starting time t to be large enough). Moreover, hλi cannot exceed ǫ at a later time. In order to obtain a quantitative estimate, note that hλi/2 a−2 hλ i = (1 − e ) + O(t ), where we have used the fact that λ is bounded to the future as well as (91). As a consequence, 6 1 2 2 a−2 ˆ ˆ ˆ ˆ ˆ ˆ ˆ ∂ hλi =2hλihλ i = hλi 1 − 1 + hλi + O(hλi ) + O(t hλi) t t t 2 3 1 2 3 a−2 ˆ ˆ ˆ = − hλi + O(hλi ) + O(t hλi). t t 23 Let 0 < b < 1 − a and define 2b 2 E = t hλi . Then dE 2b 3 1 −1 b+a−1 1/2 = E − E + O(hλiE) + t O(t E ). dt t t t As a consequence, there is a constant C > 0 such that ∂ E ≤ 0 when E ≥ C and t is large enough. In particular, E is thus bounded to the future. For every 0 < b < 1/2, there is thus a constant C such that −b λ(t, ·) + 3 lnt − 2 ln ≤ C t 4Λ for all t ≥ t . Due to this estimate, we can return to the argument presented in the proof of 1/2 Lemma 45 and obtain the improvement E (t) ≤ Ct for t ≥ t . As a consequence, we can go bas 1 through the above arguments with a = 1/2 and b = 1/2. The lemma follows. Lemma 47. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that −3/2 −1/2 t hα (t, ·)i + kQ(t, ·)k 0 + kP (t, ·)k 0 ≤ C (93) C C for all t ≥ t . Proof. Estimate, using (69), (84), (88), (89) and Lemma 40, D E −1/2 −P +λ/2 2 −1/2 P +λ/2 2 1 α 1 α e J α e (K − QJ) −1/2 −1/2 ∂ hα i = − α = + dθ 5/2 5/2 2 α 2π 1 2t 2t 1/2 −1/2 λ/2 1/2 −1/2 λ/2 + [2t α e Λ + t α e (ρ − P )]dθ 2π 1 −1/2 1/2 −1/2 λ/2 ≤Ct + 2t α e Λdθ 2π 1 3 3 −1/2 λ/2 −1/2 −1/2 −3/2 −1/2 −1/2 ≤Ct + he α i ≤ hα i + Ct hα i + Ct . 2t 2t −1/2 Let A = hα i + t. Then dA 3 −1/2 −3/2 = ∂ hα i + 1 ≤ A + Ct A. dt 2t Consequently, A(t) 3 ln ≤ ln t + C , A(t ) 2 −1/2 3/2 so that hα i ≤ Ct for t ≥ t . Combining this estimate with (89) yields Z Z Z 1/2 1/2 1/2 2 −1/2 −3/4 3/4 |P |dθ ≤ α P dθ α dθ ≤ Ct t ≤ C (94) θ θ 1 1 1 S S S for t ≥ t . On the other hand, using (83) and (89) yields Z Z 1/2 1 1 2 −3/2 |∂ hPi| = |hP i| ≤ |P |dθ ≤ P dθ ≤ Ct . (95) t t t 2π 1 2π 1 S S Consequently, hPi is bounded to the future. Combining these two observations, we conclude that kP (t, ·)k 0 ≤ C for all t ≥ t . Combining this estimate for P with the bound (89), it is possible to derive L estimates for Q and Q analogous to (94) and (95). Consequently, Q is bounded to the future. θ t The lemma follows. 24 Lemma 48. Consider a future global solution to Einstein’s equations with a cosmological constant Λ > 0, T -symmetry, λ-asymptotics, and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C > 0 such that λ/2−P 2 e J −2 ≤ Ct , (96) 5/2 P +λ/2 2 e (K − QJ) −2 ≤ Ct (97) 5/2 for all t ≥ t . Moreover, for t ≥ t , 1 1 −3/2 kH k 1 + kG k 1 ≤ Ct . t L t L Proof. Combining (64), (81), (88), (89) and (93), we conclude that Z Z Z 5/4 −1/2 λ/4 5/4 −3/2 3/4 3/2 −1/2 λ/2 |J |dθ ≤ Ct α e ρdθ ≤ Ct t t t α e ρdθ ≤ Ct. (98) 1 1 1 S S S The spatial variation of J is consequently not greater than Ct. Combining (66), (82), (83), (88), (89) and (93) yields Z Z Z 5/4 λ/4 5/4 −3/2 3/4 −3/2 3/2 −1/2 λ/2 −1/2 |J |dθ ≤ Ct e ρdθ ≤ Ct t t t t α e ρdθ ≤ Ct . (99) 1 1 1 S S S As a consequence, 1/2 |hJi| ≤ Ct . (100) Combining (98) and (100) yields kJ(t, ·)k 0 ≤ Ct. (101) Due to (88), (93) and (101), we conclude that (96) holds. Let us now turn to K − QJ. To begin with, note that the L -norm of Q is bounded to the future. The argument to prove this statement is similar to (94), keeping in mind that (93) holds. Moreover, Z Z Z Z |K − Q J − QJ |dθ ≤ |K − QJ |dθ + |Q J|dθ ≤ |K − QJ |dθ + Ct, θ θ θ θ θ θ θ θ 1 1 1 1 S S S S where we have used (101) and the fact that the L -norm of Q is bounded. On the other hand, (64) and (65) yield 5/4 −1/2 −P/2+λ/4 K − QJ = 2t α e J . θ θ 3 Keeping (81) in mind, we can thus argue as in the proof of (98) in order to conclude that |K − Q J − QJ |dθ ≤ Ct. (102) θ θ θ In particular, the spatial variation of K − QJ is bounded by Ct. On the other hand, the L -norm −3/2 of Q can be be bounded by Ct ; cf. (95) and (93). Combining this observation with (101) yields Z Z Z Z −1/2 |K − Q J − QJ |dθ ≤ |K − QJ |dθ + |Q J|dθ ≤ |K − QJ |dθ + Ct . (103) t t t t t t t t 1 1 1 1 S S S S Moreover, due to (66) and (67), 5/4 −P/2+λ/4 K − QJ = −2t e S . t t 13 25 Keeping (82) in mind, we can proceed as in (99) in order to obtain −1/2 |K − QJ |dθ ≤ Ct . t t 1/2 Due to (103) and this estimate, the mean value of K − QJ cannot grow faster than Ct . Com- bining this observation with (102) yields kK − QJk 0 ≤ Ct. (104) Keeping (88) and (93) in mind, we obtain (97). Using the fact that (58) holds, we conclude that Z Z −5/2 −1/2 P +λ/2 |H |dθ ≤ t α e |K − QJ|dθ. 1 1 S S Combining this inequality with (88), (93) and (104) yields the desired L -estimate for H . A similar argument for G yields the remaining conclusion of the lemma. 3.1 Causal structure of T -Gowdy symmetric solutions It is of interest to note that in the T -Gowdy symmetric case, it is sufficient to assume future global existence and energy conditions in order to conclude that there is asymptotic silence. In fact, we have the following result. Proposition 49. Consider a future global T -Gowdy symmetric solution to Einstein’s equations with a cosmological constant Λ > 0 and a stress energy tensor satisfying the dominant energy condition and the non-negative pressure condition. Then there is a constant C, depending only on the solution, such that if γ(s) = [s, θ(s), x(s), y(s)] = [s, γ¯(s)] is a causal curve, then 2 −3 |γ¯(s)| ≤ Cs (105) for s ≥ t . In particular, there is a point x¯ ∈ T such that 1 0 −1/2 d[γ¯(s), x¯ ] ≤ Cs (106) for all s ≥ t , where d is the standard metric on T . Proof. The causality of the curve is equivalent to the estimate −1 2 3/2 P−λ/2 2 3/2 −P−λ/2 2 ˙ ˙ ˙ α θ + s e [x˙ + Qy˙ + (G + QH)θ] + s e (y˙ + Hθ) ≤ 1. (107) Note that in the case of Gowdy symmetry, G and H are time-independent. In particular, they are thus bounded. Due to (93) we also know that Q is bounded for t ≥ t . On the other hand, combining (107) with (83), (88) and (93) yields −3/2 |θ| ≤ Cs , −3/2 |y˙ + Hθ| ≤ Cs , −3/2 |x˙ + Qy˙ + (G + QH)θ| ≤ Cs for s ≥ t . Thus (105) holds, an estimate which implies (106). 26 4 Light cone estimates In the presence of matter of Vlasov type, it is necessary to consider the characteristic system in parallel with the light cone estimates for the metric components. Let us therefore begin by writing down the characteristic system. It is given by dΘ V 1/2 = α , (108) ds V 1 2 3 dV 1 1 α 1 V V 1/2 0 1 1/2 P = − α λ V − λ − 2 − V + α e Q (109) θ t θ ds 4 4 α s V 3 2 2 2 −P/2+λ/4 P/2+λ/4 1 (V ) − (V ) e J e (K − QJ) 1/2 2 3 − α P + V + V , 0 7/4 7/4 2 V s s 2 1 2 dV 1 1 1 V V 2 1/2 = − P + V − α P , (110) t θ ds 2 s 2 V 3 1 3 1 2 dV 1 1 1 V V V V 3 1/2 P 2 1/2 P = − − P V + α P − e Q V − α e Q . (111) t θ t θ 0 0 ds 2 s 2 V V 1/2 Note that in this system of equations, functions such as α should be evaluated at [s, Θ(s)]. In view of the Vlasov equation (75), it is clear that the distribution function is constant along characteristics. It is important to note that only in the case of V is it necessary to carry out an 2 3 analysis; concerning V and V we automatically obtain the following estimate. Lemma 50. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution 0 0 has λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the 1 0 solution, such that if Θ, V is a solution to (108)–(111) with initial data Θ(t ), V (t ) such that 1 1 [t , Θ(t ), V (t )] is in the support of f, then 1 1 1 2 3 −1/2 |V (s)| + |V (s)| ≤ Cs for all s ≥ t . Remark 51. As mentioned in Remark 6, we tacitly assume f(t , ·) to have compact support. Proof. Due to (110) and (111), it can be verified that 1/2 P/2 2 1/2 P/2 2 1/2 −P/2 3 s e V , s Qe V + s e V (112) are conserved along characteristics. Since we know P and Q to be uniformly bounded, cf. Lemma 47, we obtain the conclusion of the lemma. Let us now turn to V . To begin with, we have the following estimate. Lemma 52. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution 0 0 has λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the 1 0 solution, such that if Θ, V is a solution to (108)–(111) with initial data Θ(t ), V (t ) such that 1 1 [t , Θ(t ), V (t )] is in the support of f, then 1 1 1 1 2 1 d(V ) 1 |V | 1 2 −1/2 λ/2 1 2 ≤ − (V ) + Cs e (Q ) ds s V (113) 1 1 |V | |V | −1 1/2 −3/2 1 2 −2 1 + CsF + Cs F + Cs (V ) + Cs |V | 0 0 V V for all s ≥ t , where 1 1 1 2 3 Q (t) := sup{|v | : (t, θ, v , v , v ) ∈ suppf} (114) 27 and F (t) = sup A (t, θ) + sup A (t, θ), (115) + − 1 1 θ∈S θ∈S where A is defined in (71). Proof. Due to (109), we have 1 2 1 2 3 d(V ) 1 1 α 1 V V V 1 2 1/2 P 1/2 0 1 = − α λ V V − λ − 2 − (V ) + 2α e Q θ t θ ds 2 2 α s V 3 2 2 2 −P/2+λ/4 P/2+λ/4 (V ) − (V ) e J e (K − QJ) 1/2 1 1 2 1 3 − α P V + 2 V V + 2 V V . 0 7/4 7/4 s s −2 1 However, due to Lemmas 48 and 50, we can estimate the last two terms by Cs |V |. We thus have 1 2 1 d(V ) 1 1 α 1 |V | 1/2 0 1 1 2 −1 1/2 −2 1 ≤ − α λ V V − λ − 2 − (V ) + Cs F + Cs |V |, θ t ds 2 2 α s V where we have used Lemma 50. Due to (59) and (63), the sum of the first and the second term on the right hand side can be written −P +λ/2 2 P +λ/2 2 e J e (K − QJ) 1/2 λ/2 0 1 1 1/2 λ/2 1 2 1 2 1 2 2s e (J V − ρV )V − 2s e Λ(V ) − (V ) − (V ) 5/2 5/2 2s 2s 1 1 1/2 2P 0 1 2 2 2P 2 2 1 2 1 2 − sα (P P + e Q Q )V V − s P + αP + e (Q + αQ ) (V ) + (V ) . t θ t θ t θ t θ 2 2s (116) Note that 1 1 1/2 0 1 2 2 1 2 1/2 0 1 2 2 1 2 −sα P P V V − s[P + αP ](V ) ≤sα |P P |V |V | − s[P + αP ](V ) t θ t θ t θ t θ 2 2 1/2 1 0 1 ≤sα |P P ||V |(V − |V |) t θ 1/2 2 1 2 − s(|P | − α |P |) (V ) . t θ Combining this estimate with a similar estimate for Q, we conclude that the second and third last terms in (116) can be estimated by 2 2 3 2 1 1 + (V ) + (V ) |V | 1/2 2P 1 sα (|P P | + e |Q Q |)|V | ≤ CsF . t θ t θ 0 1 0 V + |V | V Due to Lemma 46, we have 1 1 1/2 λ/2 −3/2 −2s e Λ + = − + O(s ). 2s s Combining the above observations with Lemma 48, we conclude that 1 2 d(V ) 1 2 1/2 λ/2 0 1 1 ≤ − (V ) + 2s e (J V − ρV )V ds s (117) 1 1 |V | |V | −1 1/2 −3/2 1 2 −2 1 + CsF + Cs F + Cs (V ) + Cs |V |. 0 0 V V Let us estimate the term 1/2 λ/2 0 1 1 1/2 λ/2 − + 1 T := 2s e (J V − ρV )V = 2s e (I + I )V , (118) 1 1 28 where Z Z − 1 0 0 1 1 2 3 I (s) = [v V (s) − v V (s)]f(s, Θ(s), v)dv dv dv , R −∞ Z Z + 1 0 0 1 1 2 3 I (s) = [v V (s) − v V (s)]f(s, Θ(s), v)dv dv dv . R 0 1 1 1 − There are two cases to be distinguished, V (s) ≥ 0 and V (s) < 0. When V (s) ≥ 0, I is non-positive and can be dropped. Furthermore, for v ≥ 0, 1 2 0 2 0 2 1 2 (v ) (V ) − (v ) (V ) 1 0 0 1 v V − v V = 1 0 0 1 v V + v V 1 2 2 2 3 2 1 2 2 2 3 2 (v ) (1 + (V ) + (V ) ) (V ) (1 + (v ) + (v ) ) = − 1 0 0 1 1 0 0 1 v V + v V v V + v V 1 2 2 3 2 v (1 + (V ) + (V ) ) ≤ . 2 3 2 −1/2 3 −1/2 Letting E = {(v , v ) : |v | ≤ Cs , |v | ≤ Cs }, where C is the constant appearing in vel Lemma 50, we obtain Z Z 1 2 2 3 2 v (1 + (V ) + (V ) ) 1/2 λ/2 1 1 2 3 T ≤2kf(t , ·)k s e V dv dv dv 1 1 ∞ E 0 vel (119) Z 1 Q 1 1 V V −1/2 λ/2 1 1 −1/2 λ/2 1 2 ≤Cs e v dv ≤ Cs e (Q ) , 0 0 V V 1 1 where Q (t) is defined as in the statement of the lemma. When V < 0, an analogous argument can be given and it follows that in both cases |V | −1/2 λ/2 1 2 T ≤ Cs e (Q ) . (120) Combining (117), (118) and (120) yields the conclusion of the lemma. Lemma 53. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 3 2 2 2P 2 2 1 2 t kP + αP + e (Q + αQ )k + t[Q (t)] ≤ C t θ t θ for all t ≥ t . Proof. Let us use (72) to derive an estimate for F . Note, to begin with, that 2 α 2 5 1/2 λ/2 −3/2 − − ≤ − + 4t e Λ ≤ − + O(t ), t α t t where we have used (69), (80) and (88). Note also that the second term on the right hand side of (72) can be written 1 2 1 1 2 2P 2 (A − A ) + α(P + e Q ) ≤ A + (A + A ). ∓ ± ∓ + − θ θ 2t t t 2t Combining these estimates with Lemma 48, we conclude that 4 1 1/2 −3/2 −3 ∂ A ≤ − A + Ct A + (A + A ) + Ct A ± ∓ ∓ ∓ + − t 2t −1/2 λ/2 −1/2 λ/2 P + 2t e (P − P )∂ P + 4t e S e ∂ Q. 2 3 ∓ 23 ∓ 29 Moreover, due to (77) and Lemma 50, −2 1 −2 1 |P | ≤ Ct ln(1 + Q ), |S | ≤ Ct ln(1 + Q ) k 23 for k = 2, 3. As a consequence, 4 1 1/2 1/2 −3/2 −3 −4 1 ∂ A ≤ − A + Ct A + (A + A ) + Ct A + Ct A ln(1 + Q ). ± ∓ ∓ ∓ + − ∓ ∓ t 2t Defining A = t A + t, (121) ± ± we obtain 1/2 −3/2 −2 1 ˆ ˆ ˆ ˆ ∂ A ≤ (A + A ) + Ct A + Ct A ln(1 + Q ). ± ∓ + − ∓ 2t Introducing ˆ ˆ ˆ F (t) = sup A (t, θ) + sup A (t, θ), (122) + − 1 1 θ∈S θ∈S we obtain −3/2 −2 1/2 1 ˆ ˆ ˆ ˆ ˆ F (t) ≤ F (t ) + F (s) + Cs F (s) + Cs F (s) ln(1 + Q ) ds. (123) Introducing 1 1 2 1/2 1 1 2 1/2 R (s) = [s(V (s)) + 1] , Q (s) = [s(Q (s)) + 1] , (124) (113) implies that 1 2 d(R ) −2 1 2 −2 −5/2 −3/2 1 2 ˆ ˆ ˆ ˆ ≤ Cs (Q ) + Cs F + Cs F + Cs (Q ) . ds Integrating this inequality from t to t and taking the supremum over initial data belonging to the support of f, we obtain 1 2 1 2 −2 −3/2 1 2 ˆ ˆ ˆ ˆ [Q (t)] ≤ [Q (t )] + Cs F + Cs (Q ) ds. (125) Adding (123) and (125) and introducing 1 2 ˆ ˆ G = F + (Q ) , (126) we obtain −3/2 G(t) ≤ G(t ) + G(s) + Cs G(s) ds. In particular, G(t) ≤ Ct, so that F (t) ≤ Ct and Q is bounded. Returning to (113) with this information in mind, we conclude that 1 2 d(R ) −3/2 1 2 ≤ Cs (Q ) . ds By arguments similar to ones given above, we conclude that Q is bounded. The lemma follows. 30 5 Intermediate estimates Before proceeding, it is useful to collect the estimates that follow from the above arguments. Lemma 54. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that −1 λ(t, ·) + 3 lnt − 2 ln ≤ Ct , (127) 4Λ α 3 3 −2 + + λ + ≤ Ct , (128) α t t 0 0 C C 1/2 −2 α λ ≤ Ct (129) for all t ≥ t . Moreover, −2 kJ k 0 + kK k 0 ≤ Ct , (130) t C t C P +λ/2 2 −P +λ/2 2 e (K − QJ) e J −4 + ≤ Ct , (131) 5/2 5/2 t t 0 0 C C kJ k 0 + kK k 0 ≤ C, (132) θ θ C C P +λ/2 2 −P +λ/2 2 e (K − QJ) e J −4 ∂ + ∂ ≤ Ct (133) θ θ 5/2 5/2 t t 0 0 C C for all t ≥ t . Finally, 1/2 −3/2 kρk 0 + t kJ k 0 + tkP k 0 + tkS k 0 ≤ Ct (134) C i C i C im C for all t ≥ t . Remark 55. Note that as a consequence of (128), −1/2 λ/2 −2 ∂ (α e ) ≤ Ct for all t ≥ t . Proof. Due to (77) and Lemmas 50 and 53, the estimate (134) holds. Combining Lemmas 48 and 53 with (60) and (134), we conclude that 3 3 λ/2 −2 ∂ λ = − e + O(t ), t t where we have used the notation (84). Combining this observation with Lemma 46, we conclude that there is a constant C such that 2 2 −2 ˆ ˆ ˆ ∂ λ ≤ − λ + Ct |λ|. 2 2 Introducing L = t λ , we obtain 1 C 1/2 ∂ L ≤ − L + L . t t In particular, it is clear that L decreases once it exceeds a certain value. As a consequence, L is bounded, and we obtain (127). Combining (60), (69), (127) and (134) and Lemmas 48 and 53, we then obtain (128). As a consequence of this estimate, t α converges to a strictly positive function. In particular, there are constants C > 0, i = 1, 2, such that C ≤ t α(t, θ) ≤ C (135) 1 2 31 for all t ≥ t . Due to (63), (88), (134) and Lemma 53, we obtain (129). Returning to (66) and (67), keeping (88), (93) and (134) in mind, we conclude that (130) holds. As a consequence, it is clear that J and K are bounded. Combining this observation with (88) and (93) yields (131). Due to (64), (65), (88), (93), (134) and (135), we also obtain (132). Due to (93), (135) and Lemma 53, we know that P and Q are bounded for t ≥ t . Moreover, λ is bounded for t ≥ t due to (129) θ θ 1 θ 1 and (135). Combining these observations with (127), (132) and the fact that J, K, Q and P are bounded, we obtain (133). 6 Derivatives of the characteristic system Solutions to the Vlasov equation can be expressed in terms of the initial datum for the distribution function and appropriate solutions to the characteristic system (108)–(111). In order to see this, let us begin by introducing the notation Θ, V for the solution to (108)–(111) corresponding to the initial data Θ(t; t, θ, v) = θ, V (t; t, θ, v) = v. (136) Here we write Θ(s; t, θ, v) and V (s; t, θ, v) in order to clarify the dependence on the initial data. In particular, Θ(s; t, θ, v) and V (s; t, θ, v), considered as functions of the parameter s, thus constitute a solution to (108)–(111). The purpose of the variables (t, θ, v) appearing after the semi-colon is simply to indicate that the relations (136) hold. We shall use the notation dΘ/ds and dV/ds to indicate differentiation with respect to the first variable. Moreover, we shall take ∂ Θ, ∂ Θ, ∂ iΘ t θ v etc. to denote differentiation with respect to the variables appearing after the semi-colon. Given a fixed τ ∈ (t , ∞), where (t , ∞) is the existence interval of the solution to the Einstein- 0 0 Vlasov system under consideration, we know that f(t, θ, v) = f[τ, Θ(τ; t, θ, v), V (τ; t, θ, v)]. (137) Since f(τ, ·) is a smooth function with compact support, it is sufficient to estimate the derivatives of solutions to the characteristic system in order to estimate the derivatives of f. Unfortunately, differentiating the characteristic system leads to second order derivatives of P , Q etc., quantities over which we have no control. However, using the ideas introduced in [2], this problem can be circumvented. In fact, let ∂ be a shorthand for ∂ , ∂ or ∂ i and let t θ −1/2 λ/2 Ψ = α e ∂Θ, (138) 2 2 3 2 α 1 (V ) − (V ) 1 1 −1/2 1/2 λ/2 0 −1/2 0 Z = ∂V + α λ − 2 − 4s e Λ V − α P V t t 0 2 1 2 4 α 2 (V ) − (V ) 2 2 3 2 0 2 3 1 (V ) − (V ) V V V 1 −1/2 P + P V − α e Q (139) θ t 0 2 1 2 0 2 1 2 2 (V ) − (V ) (V ) − (V ) 1 2 3 V V V +e Q ∂Θ, 0 2 1 2 (V ) − (V ) 2 2 2 Z = ∂V + P V ∂Θ, (140) 3 3 3 P 2 Z = ∂V − P V − e Q V ∂Θ. (141) θ θ 1 2 3 It is then possible to derive an ODE for (Ψ, Z , Z , Z ) such that the coefficients are controlled due to previous arguments. The definitions (138)–(141) differ slightly from those of [2]. The reason for this is that in the present context, it is not sufficient to know that no second order derivatives of P , Q etc. occur; we need to analyse, in detail, all the terms that appear, and to use the resulting system in order to derive specific estimates for ∂Θ and ∂V . The relevant result is the following. 32 2 Lemma 56. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that dZ 1 1 j = − Z + c Ψ + c Z , (142) 1,θ 1,j ds 2s dZ 1 2 2 = − Z + c Z , (143) 2,2 ds 2s dZ 1 3 2 3 = − Z + c Z + c Z , (144) 3,2 3,3 ds 2s dΨ = c Ψ + c Z , (145) θ,θ θ,i ds where Einstein’s summation convention applies to i and j, 1/2 −3/2 |c (s; t, θ, v)| + |c (s; t, θ, v)| + |c (s; t, θ, v)| + s |c (s; t, θ, v)| ≤ Cs , i,j θ,i i,θ θ,θ 1 3 and the estimate holds for all (t, θ, v) ∈ [t , ∞)× S × R in the support of f and for all s ∈ [t , t]. 1 1 Proof. Let us begin by noting that P/2 1/2 2 1/2 P/2 2 e s Z = ∂(s e V ), 1/2 −P/2 3 P/2 1/2 2 1/2 P/2 2 1/2 −P/2 3 s e Z + e Qs Z = ∂(s Qe V + s e V ). Since the quantities appearing in (112) are preserved along characteristics, we obtain P/2 1/2 2 e s Z = 0, (146) ds 1/2 −P/2 3 P/2 1/2 2 s e Z + e Qs Z = 0. (147) ds We also have 1 1 dΨ 1 α 1 V V 1/2 λ/2 = λ − Ψ + α λ Ψ + e ∂ . (148) t θ 0 0 ds 2 α 2 V V 1 0 i In the end, we shall express ∂(V /V ) in terms of Z and Ψ. However, there is no immediate gain in doing so here. The most cumbersome part of the argument is to compute the derivative of Z . This calculation can be divided into several parts. Let us begin by considering d dV (∂V ) = ∂ . ds ds When calculating the right hand side, it is convenient to divide the result into terms which include a ∂V factor, i = 1, 2, 3, and terms which do not. Combining Lemmas 50, 53 and 54, the terms which include such a factor can be written 1 i − ∂V + c ∂V , 2s −2 1 where c (s) = O(s ); note that the factor in front of V in (109) is given by −2 − + O(s ). 2s It is straightforward to calculate the remaining terms, and we conclude that 2 3 d 1 1 α V V 1 1/2 0 1 1/2 P (∂V ) = − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 4 4 α V 3 2 2 2 1 (V ) − (V ) 1/2 −7/4 λ/4 −P/2 2 − ∂ (α P ) ∂Θ + s ∂ (e e J)V ∂Θ θ θ θ 2 V −7/4 λ/4 P/2 3 1 i + s ∂ [e e (K − QJ)]V ∂Θ − ∂V + c ∂V , θ i 2s 33 −2 where c (s) = O(s ). Combining an argument which is identical to the proof of (133) with Lemma 50, we can estimate the factors multiplying ∂Θ in the third and fourth last terms on the right hand side. In fact, we obtain 2 3 d 1 1 1 α V V 1 1 1/2 0 1 1/2 P (∂V ) = − ∂V − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 2s 4 4 α V 3 2 2 2 1 (V ) − (V ) 1/2 i − ∂ (α P ) ∂Θ + c ∂Θ + c ∂V , θ θ θ i 2 V −2 −3 where c (s) = O(s ) and c (s) = O(s ). As a next step, it is of interest to consider the terms i θ α 1 that arise when d/ds hits a V in the second term in the definition of Z . Before writing down the result, let us note that, due to Lemmas 50, 53 and 54, i 0 dV 1 dV i −2 −2 = − V + O(s ), = O(s ). (149) ds 2s ds Due to these observations, Lemmas 50, 53 and 54, as well as (135) and the definition of Z , we α 1 conclude that when d/ds hits a V in the second term in Z , the resulting expression can be −2 written c ∂Θ, where c (s) = O(s ). θ θ Note that every term appearing in the second term in the definition of Z can be written in the form −1/2 h(·, Θ)ψ(V )α (·, Θ)∂Θ. (150) We have already estimated the terms that arise when d/ds hits ψ. Let us therefore consider the terms that arise when the derivative hits the remaining factors. Omitting the arguments, we need to consider V α 1/2 −1/2 h + α h ψα ∂Θ − hψ∂Θ t θ 0 3/2 V 2α 1 1 1 α V α V V θ θ 1/2 −1/2 + hψ − α ∂Θ + α ∂Θ + ∂ (151) 3/2 0 1/2 0 0 2α V 2α V V 1 1 V V −1/2 = ∂ (α h) + h ψ∂Θ + hψ∂ . t θ 0 0 V V −2 In all the terms of interest, hψ = O(s ), where we have used Lemmas 50, 53 and 54. As a consequence, this expression can be written −1/2 i ∂ (α h) + h ψ∂Θ + c ∂V , t θ i −2 where c (s) = O(s ). Adding up the above observations, we conclude that 1   2 3 dZ 1 1 1 α V V 1 1/2 0 1 1/2 P = − ∂V − ∂ (α λ )V ∂Θ − ∂ λ − 2 V ∂Θ + ∂ (α e Q ) ∂Θ θ θ θ t θ θ ds 2s 4 4 α V h  i 3 2 2 2 1 (V ) − (V ) 1 α 1/2 −1/2 1/2 λ/2 0 − ∂ (α P ) ∂Θ + ∂ α λ − 2 − 4s e Λ V ∂Θ θ θ t t 2 V 4 α 2 2 3 2 1 α 1 (V ) − (V ) 1/2 λ/2 1 −1/2 0 + ∂ λ − 2 − 4s e Λ V ∂Θ − ∂ (α P )V ∂Θ θ t t t 0 2 1 2 4 α 2 (V ) − (V ) 2 2 3 2 2 2 3 2 1 (V ) − (V ) 1 (V ) − (V ) 1 1 − P V ∂Θ + P V ∂Θ tθ tθ (152) 0 2 1 2 0 2 1 2 2 (V ) − (V ) 2 (V ) − (V ) 1 2 2 2 3 2 0 2 3 1 (V ) (V ) − (V ) V V V 1/2 −1/2 P + ∂ (α P ) ∂Θ − ∂ (α e Q ) ∂Θ θ θ t t 0 0 2 1 2 0 2 1 2 2 V (V ) − (V ) (V ) − (V ) 1 2 3 1 2 3 V V V V V V P P − ∂ (e Q ) ∂Θ + ∂ (e Q ) ∂Θ θ t t θ 0 2 1 2 0 2 1 2 (V ) − (V ) (V ) − (V ) 1 1 2 3 V V V V 1/2 P i + ∂ (α e Q ) ∂Θ + c ∂Θ + c ∂V , θ θ θ i 0 0 2 1 2 V (V ) − (V ) 34 −2 −2 where c (s) = O(s ) and c (s) = O(s ). How to interpret the terms in this equation should be θ i clear from (150) and (151). However, there is one term which is slightly ambiguous, namely the sixth one on the right hand side of (152). For the sake of clarity, let us point out that the factor in front of V ∂Θ in this term should be interpreted as the time derivative of h  i 1 α −1/2 1/2 λ/2 α λ − 2 − 4t e Λ , 4 α evaluated at [s, Θ(s)]. The expression (152) can be simplified somewhat. To begin with, the terms involving P cancel. Moreover, tθ P P P −∂ (e Q ) + ∂ (e Q ) = e (P Q − P Q ). θ t t θ t θ θ t Using Lemma 53 and (135), we can estimate this expression in order to conclude that the sum of −3 the fourth and fifth last terms is of the form c ∂Θ, where c = O(s ) (in order to obtain this θ θ conclusion, we have also used Lemma 50). Using Lemma 53, (127), (129) and (135), the sum of −2 the third and the seventh term can be written c ∂Θ, where c = O(s ). Since θ θ 3 2 2 2 1 2 2 2 3 2 2 2 3 2 (V ) − (V ) (V ) (V ) − (V ) (V ) − (V ) − + = V 0 0 0 2 1 2 0 2 1 2 V V (V ) − (V ) (V ) − (V ) 2 3 1 1 2 3 0 2 3 V V V V V V V V V + = , 0 0 0 2 1 2 0 2 1 2 V V (V ) − (V ) (V ) − (V ) 1/2 the terms involving a factor of ∂ (α P ) can be written θ θ 2 2 3 2 1 (V ) − (V ) 1/2 0 ∂ (α P )V ∂Θ, θ θ 0 2 1 2 2 (V ) − (V ) 1/2 P and the terms involving a factor of ∂ (α e Q ) can be written θ θ 0 2 3 V V V 1/2 P ∂ (α e Q ) ∂Θ. θ θ 0 2 1 2 (V ) − (V ) Combining these observations yields h  i dZ 1 1 α 1 1 −1/2 1/2 λ/2 0 1/2 0 = − ∂V + ∂ α λ − 2 − 4s e Λ V ∂Θ − ∂ (α λ )V ∂Θ t t θ θ ds 2s 4 α 4 2 2 3 2 1 (V ) − (V ) −1/2 1/2 0 − [∂ (α P ) − ∂ (α P )]V ∂Θ t t θ θ (153) 0 2 1 2 2 (V ) − (V ) 0 2 3 V V V −1/2 P 1/2 P i − [∂ (α e Q ) − ∂ (α e Q )] ∂Θ + c ∂Θ + c ∂V , t t θ θ θ i 0 2 1 2 (V ) − (V ) −2 −2 where c (s) = O(s ) and c (s) = O(s ). Combining (61), (135) and Lemmas 53 and 54 yields θ i −1/2 1/2 −1 ∂ (α P ) − ∂ (α P ) = O(t ). t t θ θ Due to this estimate and Lemmas 53 and 50, the fourth term on the right hand side of (153) can −2 be written c ∂Θ, where c (s) = O(s ). Keeping (68) in mind, a similar argument yields the θ θ same conclusion concerning the fifth term on the right hand side of (153). Finally, keeping (87) in mind, a similar argument yields the conclusion that the combination of the second and third −3/2 terms on the right hand side of (153) can be written c ∂Θ, where c (s) = O(s ). To conclude, θ θ dZ 1 1 i = − ∂V + c ∂Θ + c ∂V , (154) θ i ds 2s −3/2 −2 where c (s) = O(s ) and c (s) = O(s ). On the other hand, due to (127) and (135), the θ i −1/2 λ/2 function α e can be bounded from above and below by positive constants (for t ≥ t ). 35 In other words, ∂Θ and Ψ are interchangeable when deriving equations of the form (142)–(145). Moreover, due to (135) and Lemmas 50, 53 and 54, i i Z = ∂V + c Ψ, (155) i,θ −1/2 where c (s) = O(s ). We are now in a position to prove the lemma. To begin with, combining i,θ (146), (147) and Lemma 53 yields (143) and (144). Combining (148) and (155) with Lemmas 50, 53 and 54 yields (145). Finally, combining (154) and (155) yields (142). Lemma 57. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 2 1/2 i 2 (ln s) |∂ Θ(s; t, θ, v)| + s ∂ V (s; t, θ, v) ≤ C(ln t) (156) θ θ 1 3 for all s ∈ [t , t] and (t, θ, v) ∈ [t , ∞) × S × R in the support of f. 1 1 Remark 58. It is possible to use arguments similar to the ones given below in order to derive estimates for ∂ Θ, ∂ iΘ etc. t v Proof. Let i 1/2 i 2 ˆ ˆ Z (s; t, θ, v) = s Z (s; t, θ, v), Ψ(s; t, θ, v) = (ln s) Ψ(s; t, θ, v). (157) Then, due to Lemma 56, dZ 1/2 −2 j ˆ ˆ = c s (ln s) Ψ + c Z , 1,θ 1,j ds dZ = c Z , 2,2 ds dZ 2 3 ˆ ˆ = c Z + c Z , 3,2 3,3 ds dΨ 2 −1/2 2 i ˆ ˆ ˆ = Ψ + c Ψ + c s (ln s) Z θ,θ θ,i ds s lns for s ∈ [t , t], with coefficients as in Lemma 56. Introducing i 2 2 ˆ ˆ ˆ E = (Z ) + (Ψ) , (158) i=1 we conclude that there is a constant C > 0, depending only on the solution, such that dE 1 ≥ −C E ds s(ln s) for s ∈ [t , t]. As a consequence, ˆ ˆ E(s; t, θ, v) ≤ CE(t; t, θ, v) (159) for s ∈ [t , t]. Let us now assume ∂ = ∂ . Then 1 θ Ψ(t; t, θ, v) = O[(ln t) ]. Moreover, i 1/2 i Z (t; t, θ, v) = [t ∂ V + O(1)Ψ](t; t, θ, v) = O(1). As a consequence, E(t; t, θ, v) = O[(ln t) ]. Thus (159) implies (156); note that the estimate for ∂ Θ is immediate and that i i −1/2 |∂ V | ≤ |Z | + Cs |∂ Θ|. θ θ The lemma follows. 36 7 Higher order light cone estimates Before proceeding to the higher order light cone estimates, let us record some consequences of the estimates obtained in Lemma 57. Lemma 59. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Then there is a constant C > 0, depending only on the solution, 1 0 such that 1/2 −3/2 2 k∂ ρk 0 + t k∂ J k 0 + tk∂ S k 0 + tk∂ P k 0 ≤ Ct (ln t) , (160) θ C θ i C θ ij C θ k C −3/2 ∂ ≤ Ct , (161) ≤ C (162) α C for all t ≥ t . Remark 60. Note that ∂ (α /α) = ∂ (α /α) = ∂ ∂ ln α. θ t t θ t θ Proof. The estimate (160) follows from the fact that (137) and (156) hold and the fact that i −1/2 |v | ≤ Ct in the support of f(t, ·). Consider (69). Since Lemmas 53 and 54 together with 1 −1/2 (135) imply that J, K, Q and P are bounded in C and that λ is O(t ), the first two terms −4 1 −1/2 on the right hand side of (69) are O(t ) in C . Since λ is O(t ), the θ-derivative of the third −3/2 term on the right hand side of (69) is O(t ). Due to (160), the θ-derivative of the last term is better. Thus (161) holds, so that α α θ t −3/2 ∂ ≤ ∂ ≤ Ct . t θ 0 0 α α C C Integrating this estimate yields (162). In what follows, we shall proceed inductively in order to derive estimates for higher order deriva- tives. Let us therefore assume that we have a T -symmetric solution to the Einstein-Vlasov equations with a cosmological constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. As- 0 0 sume, moreover, that the solution has λ-asymptotics and let t = t +2. Let us make the following 1 0 inductive assumption. Inductive assumption 61. For some 1 ≤ N ∈ Z, there are constants 0 ≤ m ∈ Z and C , j j j = 1, . . ., N, (depending only on N and the solution) such that j j ∂ V ∂ Θ 1/2 m s (s; t, θ, v) + (s; t, θ, v) ≤ C (ln t) , (163) j j ∂θ ∂θ 3/2 3/2 kP k N−1 + kQ k N−1 + t kP k N−1 + t kQ k N−1 ≤ C , (164) θ θ t t N−1 C C C C 1 3 for all j = 1, . . ., N, (t, θ, v) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]. 1 1 Remarks 62. The induction hypothesis holds for N = 1. In what follows, C and m will change j j from line to line. However, they are only allowed to depend on N and the solution. In this section we prove that, given that Inductive assumption 61 holds, then (164) holds with N replaced by N + 1. In the next section, we close the induction argument by proving that (163) holds with j replaced by N + 1. We shall need the following consequences of the inductive assumption. 37 2 Lemma 63. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there are constants C , j = 0, . . ., N, and m , depending only on N and the j N solution, such that 1/2 −3/2 m kρk N + t kJ k N + tkP k N + tkS k N ≤ C t (ln t) , (165) i i im N C C C C 1/2 l+1 t k∂ λk 0 ≤ C , (166) −1 j α ∂ α ≤ C , (167) kJ k N + kK k N ≤ C (ln t) , (168) θ θ N C C l+1 −3/2 ∂ ≤ C t (169) α C for t ≥ t , 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and i, m = 1, 2, 3. Proof. For N = 1, the conclusions follow from Lemmas 53, 54 and 59, (135) and the equations (64) and (65). We may thus, without loss of generality, assume that N ≥ 2. An immediate consequence of the inductive assumption is that, for 0 ≤ j ≤ N and t ≥ t , ∂ f ≤ C (ln t) ; ∂θ cf. (137) and (163). As a consequence of this estimate, we obtain (165). In order to obtain control of the θ-derivatives of α and λ, we need to proceed inductively. Let us make the inductive assumption that −1 α ∂ α ≤ C , (170) j −1/2 k∂ λk ≤ C t (171) C j for 1 ≤ j ≤ l < N. Note that we know the inductive hypothesis to be true for l = 1. Differentiating (63) l times with respect to θ and appealing to (164), (165), (170) and (171), we conclude that (171) holds with j replaced by l + 1. In order to improve our knowledge concerning α, let us begin by improving our estimates for the θ-derivatives for J and K. Differentiating (64) and (65) 0 ≤ j ≤ l times and using (164), (165), (170) and (171), we conclude that j+1 j+1 k∂ Jk 0 + k∂ Kk 0 ≤ C (ln t) (172) C C j θ θ for t ≥ t and 0 ≤ j ≤ l. Differentiating (69) l + 1 times with respect to θ, using (164), (165), (172) as well as the fact that (171) holds for 1 ≤ j ≤ l + 1, we obtain α α t θ l+1 l −3/2 ∂ = ∂ ∂ ≤ C t (173) t l 0 0 α C α C for t ≥ t . Thus ∂ ≤ C α C for t ≥ t . Combining this estimate with the inductive hypothesis, we conclude that (170) holds with j replaced by l + 1. Thus (171) and (170) hold for 1 ≤ j ≤ N. We thus conclude that (166), (167) and (168) hold. In addition, (173) implies that (169) holds. We are now in a position to derive higher order light cone estimates. Lemma 64. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 38 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there is a constant C > 0, depending only on N and the solution, such that 3/2 N N 3/2 P N P N t k∂ P k 0 + k∂ P k 0 + t ke ∂ Q k 0 + ke ∂ Q k 0 ≤ C (174) t C θ C t C θ C N θ θ θ θ for all t ≥ t . As a consequence, (164) holds with N replaced by N + 1. Proof. Let us begin by pointing out that if N = 1, then the lower bound is larger than the upper bound in some of the sums below. In that case, the corresponding sum should be equated with −3 −3 zero. Moreover, terms which are bounded by Ct for t ≥ t will sometimes be written O(t ). Let us compute N N 1/2 N N 1/2 1/2 1/2 N+1 ∂ [∂ P ∓ ∂ (α P )] =∂ P ∓ ∂ α P + α P ± α ∂ P ± t θ tt θ tθ t θ θ θ θ 2α N+1 1/2 1/2 − α ∂ (α P ) N−1 1 α 1 α t t N N−j j 1/2 N 1/2 =∂ P ∓ β ∂ ∂ (α P ) ∓ ∂ (α P ) tt j θ θ θ θ θ θ 2 α 2 α j=0 N−2 Nα N−j j+1 N 1/2 N 1/2 1/2 ∓ ∂ P ∓ β ∂ (α )∂ (P ) − ∂ [α ∂ (α P )] t j t θ θ θ θ θ θ 1/2 2α j=0 N−2 Nα θ N−j j+1 N 1/2 1/2 1/2 + ∂ (α P ) + β ∂ (α )∂ (α P ), θ j θ θ θ θ 1/2 2α j=0 −3 where the β are binomial coefficients. Note that all the sums are O(t ) due to Inductive as- sumption 61, Lemma 63 and (135). Let us use (61) in order to compute N 1/2 1/2 N ∂ [P − α ∂ (α P )] =∂ P − αP − P tt θ θ tt θθ θ θ θ N−1 1 α α t N−j t j N N = − ∂ P + ∂ P + β ∂ ∂ P t t j t θ θ θ θ t 2α 2α j=0 N−1 N−j 2P j 2 2 + β ∂ (e )∂ (Q − αQ ) θ θ t θ j=0 N−1 N−j j 2P 1/2 1/2 + e β (∂ (Q − α Q ))(∂ (Q + α Q )) j t θ t θ θ θ j=1 P +λ/2 2 e (K − QJ) 2P N 1/2 N 1/2 N + 2e [Q ∂ Q − α Q ∂ (α Q )] − ∂ t t θ θ θ θ θ 7/2 2t −P +λ/2 2 e J N −1/2 N λ/2 + t ∂ [e (P − P )]. + ∂ 2 3 θ θ 7/2 2t −3 Due to Inductive assumption 61, Lemmas 63 and 54, and (135), the sums are O(t ), as well as the last three terms on the right hand side. We thus obtain 1 α Nα t θ N N 1/2 N N N 1/2 N N 1/2 ∂ [∂ P ∓ ∂ (α P )] = − ∂ P + [∂ P ∓ ∂ (α P )] ∓ [∂ P ∓ ∂ (α P )] ± t θ t t θ t θ θ θ θ θ θ θ θ 1/2 t 2α 2α 2P N 1/2 N 1/2 −3 + 2e [Q ∂ Q − α Q ∂ (α Q )] + O(t ). t t θ θ θ θ Introducing N N 1/2 2 2P N N 1/2 2 A = [∂ P ± ∂ (α P )] + e [∂ Q ± ∂ (α Q )] , (175) N+1,± t θ t θ θ θ θ θ 39 we conclude that 5 2 N N 1/2 2 N N 1/2 2 N 1/2 N N 1/2 ∂ [∂ P ∓ ∂ (α P )] ≤ − [∂ P ∓ ∂ (α P )] ∓ ∂ (α P )[∂ P ∓ ∂ (α P )] ± t θ t θ θ t θ θ θ θ θ θ θ θ t t 1/2 −3 −3/2 + C t A + C t (A + A ), N N N+1,+ N+1,− N+1,∓ where we have used Inductive assumption 61, Lemma 63, (128) and (135). Let us now consider N−1 1 α 1 α t t N N 1/2 N N−j j 1/2 N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] =∂ Q ∓ β ∂ ∂ (α Q ) ∓ ∂ (α Q ) ± t θ tt j θ θ θ θ θ θ θ θ 2 α 2 α j=0 N−2 Nα θ N−j j N 1/2 N 1/2 1/2 ∓ ∂ Q ∓ β ∂ (α )∂ Q − ∂ [α ∂ (α Q )] t j tθ θ θ θ θ θ θ 1/2 2α j=0 N−2 Nα θ N−j j+1 N 1/2 1/2 1/2 + ∂ (α Q ) + β ∂ (α )∂ (α Q ). θ j θ θ θ θ 1/2 2α j=0 −3 As above, all the sums are O(t ) due to Inductive assumption 61, Lemma 63 and (135). Compute, using (68), N−1 1 α α t N−j t j N 1/2 1/2 N N N ∂ [Q − α ∂ (α Q )] = − ∂ Q + ∂ Q + β ∂ ∂ Q − 2(∂ P )Q tt θ θ t t j t t t θ θ θ θ θ θ t 2α 2α j=0 N N 1/2 1/2 1/2 N 1/2 − 2P (∂ Q ) + 2∂ (α P )α Q + 2α P ∂ (α Q ) t t θ θ θ θ θ θ θ N−1 N−1 X X N−j j N−j j 1/2 1/2 − 2 β (∂ P )(∂ Q ) + 2 β (∂ (α P ))(∂ (α Q )) j t t j θ θ θ θ θ θ j=1 j=1 λ/2−P e J(K − QJ) N −1/2 N λ/2−P + ∂ + 2t ∂ (e S ). θ θ 7/2 −3 Due to Inductive assumption 61, Lemmas 63 and 54, and (135), the sums are O(t ), as well as the last two terms on the right hand side. Thus 1 α Nα t θ N N 1/2 N N N 1/2 N N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] = − ∂ Q + [∂ Q ∓ ∂ (α Q )] ∓ [∂ Q ∓ ∂ (α Q )] ± t θ t t θ t θ θ θ θ θ θ θ θ 1/2 t 2α 2α N N N 1/2 1/2 − 2(∂ P )Q − 2P (∂ Q ) + 2∂ (α P )α Q θ t t t θ t θ θ θ 1/2 N 1/2 −3 + 2α P ∂ (α Q ) + O(t ). θ θ Consequently, 5 2 N N 1/2 2 N N 1/2 2 N 1/2 N N 1/2 ∂ [∂ Q ∓ ∂ (α Q )] ≤ − [∂ Q ∓ ∂ (α Q )] ∓ ∂ (α Q )[∂ Q ∓ ∂ (α Q )] ± t θ t θ θ t θ θ θ θ θ θ θ θ t t 1/2 −3 −3/2 + C t A + C t (A + A ), N N N+1,+ N+1,− N+1,∓ where we have used Inductive assumption 61, Lemma 63, (128) and (135). Adding up the above estimates, we conclude that 5 1 1 ∂ A ≤ − A + (A − A ) + (A + A ) ± N+1,∓ N+1,∓ N+1,∓ N+1,± N+1,+ N+1,− t 2t t 1/2 −3/2 −3 + C t (A + A ) + C t A . N N+1,+ N+1,− N N+1,∓ Let us introduce 7/2 1/2 ˆ ˆ ˆ ˆ ˆ ˆ A = t A + t , F = sup A , F = F + F . (176) N+1,± N+1,± N+1,± N+1,± N+1 N+1,+ N+1,− θ∈S 40 Then −3/2 ˆ ˆ ˆ ˆ ∂ A ≤ A + C t (A + A ). ± N+1,∓ N+1,± N N+1,+ N+1,− 2t Integrating this differential inequality, taking the supremum etc., we obtain −3/2 ˆ ˆ ˆ ˆ F (t) ≤ F (t ) + F (s) + C s F (s) ds. N+1 N+1 1 N+1 N N+1 2s 1/2 Thus F (t) ≤ C t . Combining this estimate with Inductive assumption 61, Lemma 63 and N+1 N (135), we obtain (174). 8 Higher order derivatives of the characteristic system In the previous section we showed that (164) holds with N replaced by N + 1; i.e., that 3/2 3/2 kP k N + kQ k N + t kP k N + t kQ k N ≤ C (177) θ C θ C t C t C N holds for all t ≥ t . We also need to prove that (163) holds with j replaced with N + 1. Before stating the relevant result, let us make the following preliminary observation. Lemma 65. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then there are constants C and m , j = 0, . . ., N, depending only on N and the j j solution, such that 1/2 −3/2 m kρk N + t kJ k N + tkP k N + tkS k N ≤ C t (ln t) , (178) i i im N C C C C j j+1 −1 1/2 α ∂ α + t k∂ λk 0 ≤ C , (179) θ θ kJ k N + kK k N ≤ C (ln t) , (180) θ C θ C N l+1 l+1 −3/2 ∂ + k∂ λ k 0 ≤ C t , (181) t C l θ θ α C 1/2 λ/2 −2 λ − 2 − 4t e Λ ≤ C t , (182) t N α C −2 λ − ≤ C t (183) t N for t ≥ t , 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and i, m = 1, 2, 3. Moreover, using the notation j i j i i j i j Ψ = ∂ Ψ, Z = ∂ Z , V = ∂ V , Θ = ∂ Θ (184) j j j j θ θ θ θ (where the ∂-operator used to define Z and Ψ is given by ∂ ), there are functions c , i = 1, 2, 3, θ i,θ such that the following estimates hold: 1/2 m |Ψ (s; t, θ, v)| + s |Z (s; t, θ, v)| ≤ C (ln t) , (185) l l l −1/2 λ/2 m |Ψ (s; t, θ, v) − (α e )[s, Θ(s; t, θ, v)]Θ (s; t, θ, v)| ≤ C (ln t) , (186) j j+1 j i i −1/2 m |Z (s; t, θ, v) − V (s; t, θ, v) − (c Ψ )(s; t, θ, v)| ≤ C s (ln t) , (187) i,θ j j j j+1 −1/2 |c (s; t, θ, v)| ≤ C s (188) i,θ 0 1 3 for all (t, θ, v) ∈ [t , ∞) × S × R in the support of f, 0 ≤ j ≤ N, 0 ≤ l ≤ N − 1 and s ∈ [t , t]. 1 1 Remark 66. Due to (179), j p p |∂ α | ≤ C α p,j for all (t, θ) ∈ [t , ∞) × S , p ∈ R and 0 ≤ j ≤ N. In particular, spatial derivatives of powers of α can thus effectively be ignored. In the derivation of the estimates below, it is useful to keep this observation in mind. 41 Proof. Combining Lemma 63 with (177), (60) and (63), we obtain (178)–(181); recall that P , Q, J and K are bounded to the future. The estimate (182) is a consequence of Lemma 63, (177) and the fact that −P +λ/2 2 α e J 1/2 λ/2 2 2 2P 2 2 λ − 2 − 4t e Λ =t P + αP + e (Q + αQ ) + t θ t θ 5/2 α t P +λ/2 2 e (K − QJ) 1/2 λ/2 + + 4t e ρ; 5/2 cf. (69) and (70). For similar reasons, (183) holds; cf. (70). Turning to (185)–(188), let us note that ∂ Ψ(s; t, θ, v) can, up to numerical factors, be written as a sum of terms of the form k −1/2 λ/2 ∂ (α e )[s, Θ(s; t, θ, v)]Θ (s; t, θ, v)· · · Θ (s; t, θ, v), i i θ k+1 where i + · · · + i = j + 1. Since k ≤ j ≤ N, the first factor is bounded due to (179). Due to 1 k+1 Inductive assumption 61, the factors Θ can be estimated by C(ln t) if i ≤ N. The only way i j a factor Θ could occur is if k = 0 and all the derivatives hit Θ in the definition of Ψ. Due to N+1 1 these observations, (186) and the estimate |Ψ (s; t, θ, v)| ≤ C (ln t) , l l 0 ≤ l ≤ N − 1, follow. The proof of the remaining estimates is similar in nature, but somewhat more involved. We now finish the induction argument by proving that (163) holds with j replaced by N + 1. Lemma 67. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t +2. Assume, moreover, that Inductive assumption 61 holds for some 1 0 1 ≤ N ∈ Z. Then (163) holds with j replaced by N + 1. Proof. The strategy of the proof is very similar to that of the proof of Lemma 56. The idea is to derive a system of ODE’s for Z and Ψ analogous to (142)–(145), and then to use arguments similar to those presented in the proof of Lemma 57. Deriving appropriate equations for Z , i = 2, 3, turns out to be relatively easy; this is due to (146) and (147). In fact, we obtain the desired conclusions concerning Z , i = 2, 3, without much effort; cf. (191) and (192) below. Deriving an equation for Ψ also turns out to be quite easy; cf. (190) below. Similarly to the proof of Lemma 56, the main difficulty consists of deriving an equation for Z . Once the desired equation has been obtained, we rescale Z and Ψ according to (199) and introduce an energy according to (200); note that these definitions are analogous to the ones given in the proof of Lemma 57. Finally, the equations imply a differential inequality for the energy E which can be integrated to yield the desired estimate. Before proceeding to a proof of the statement of the lemma, it is of interest to introduce some nota- 1 1 tion. Let b be a C function on M = (t , ∞)×S . Evaluating this function along a characteristic, we obtain B(s; t, θ, v) = b[s, Θ(s; t, θ, v)]. Differentiating B with respect to θ, we obtain ∂B ∂b ∂Θ (s; t, θ, v) = [s, Θ(s; t, θ, v)] (s; t, θ, v). (189) ∂θ ∂θ ∂θ On the other hand, distinguishing between B and b is quite cumbersome in the arguments that we are about to carry out. As a consequence, we shall write b when we mean B. Moreover, we shall use the notation ∂ b as a shorthand for ∂ B, whereas ∂ b should be interpreted as the first factor ϑ θ θ 42 on the right hand side of (189) and ∂ b should be interpreted as the function mapping (s; t, θ, v) to ∂b [s, Θ(s; t, θ, v)]. ∂t In particular, we thus have b = b Θ . Finally, let us point out that if in some expression a ϑ θ 1 ϑ-derivative hits a V or a Θ, it is to be interpreted as an ordinary θ-derivative. Note, to begin with, that 1 1 1 dΨ 1 α 1 V ∂V V 1/2 λ/2 λ/2 i i = λ − Ψ + α λ Ψ + e − e V ∂V t θ 0 0 0 3 ds 2 α 2 V V (V ) i=1 due to (148). Differentiating this equality N times with respect to ϑ, we obtain 1 1 dΨ 1 α 1 V V N t 1/2 λ/2 N+1 λ/2 i i −2 m = λ − Ψ + α λ Ψ +e −e V V +O[s (ln t) ], t N θ N N+1 0 0 0 3 ds 2 α 2 V V (V ) i=1 where we have used (135), Inductive assumption 61 and Lemma 65. Due to (187), this equation can be written dΨ N N i −2 m = c Ψ + c Z + O[s (ln t) ], (190) θ,θ θ,i N ds N −2 N −3/2 2 where c = O(s ), c = O(s ) and we sum over i but not N. Turning to Z , we have θ,θ θ,i 2 1 dZ 1 1 V 2 1/2 2 = − Z − P + α P Z ; t θ ds 2s 2 V cf. (146). Differentiating this equality N times with respect to ϑ, we obtain dZ 1 1 V N 2 1/2 2 −2 m = − Z − P + α P Z + O[s (ln t) ], t θ N N ds 2s 2 V where we have used (135), (177), Inductive assumption 61 and Lemma 65. As a consequence, 1/2 P/2 2 −3/2 m s e Z = O[s (ln t) ]. ds Integrating this equality from s to t, we obtain (assuming N ≥ 1) 1/2 P/2 2 m − s e Z (s; t, θ, v) = O[(ln t) ]; note that 1/2 P/2 2 1/2 P/2 N+1 t e Z (t; t, θ, v) = t e (∂ P )v = O(1) due to (177). In particular, we thus have 2 −1/2 m |Z (s; t, θ, v)| ≤ C s (ln t) (191) for s ∈ [t , t]. Turning to Z , we have 3 1 1 dZ 1 1 V V 3 1/2 3 P 1/2 2 = − Z + P + α P Z − e Q + α Q Z ; t θ t θ 0 0 ds 2s 2 V V cf. (147) and (146). Differentiating N times with respect to ϑ, we obtain 3 1 1 V V dZ 1 1 N 3 1/2 3 P 1/2 2 −2 m = − Z + P + α P Z − e Q + α Q Z + O[s (ln t) ]. t θ t θ N N N 0 0 ds 2s 2 V V 43 −2 m Due to (191), the third term on the right hand side is O[s (ln t) ]. We can thus proceed as in the proof of (191) in order to obtain 3 −1/2 m |Z (s; t, θ, v)| ≤ C s (ln t) (192) for s ∈ [t , t]. Finally, we need to derive an equation for Z . Just as in the derivation of the equation for Z , it is natural to divide the analysis into several steps. Consider, to begin with, dV j+1 ds for 0 ≤ j ≤ N. All the terms appearing in dV /ds can be written h ψ◦V . When differentiating an k l expression of this form, the terms that arise are (up to numerical factors) of the form ∂ h∂ ψ ◦V . ϑ θ −2 m If both k and l are ≥ 1, the resulting term is O[s (ln t) ]. If all the derivatives hit ψ, we obtain (after summing over all the terms appearing in dV /ds) 1 i −2 m − V + c V + O[s (ln t) ] j+1 j+1 2s −2 where c = O(s ) and we sum over i but not j. If all the derivatives hit h, we obtain (after summing over all the terms appearing in dV /ds) 2 3 1 1 α V V j+1 j+1 t j+1 1/2 0 1 1/2 P − ∂ (α λ )V − ∂ λ − 2 V + ∂ (α e Q ) θ t θ ϑ ϑ ϑ 4 4 α V 3 2 2 2 1 (V ) − (V ) j+1 j 1/2 −3 m − ∂ (α P ) + c Θ + O[s (ln t) ], θ j+1 ϑ θ 2 V −3 where c = O(s ) and we have used (132); note that, due to (177) and Lemma 65, we control N + 1 θ-derivatives of the first factor in each of the last two terms appearing on the right hand side of (109). Adding up, we conclude that dV 1 1 1 α j+1 j j j+1 j+1 t 1 i 1/2 0 1 ∂ = − V + c V + c Θ − ∂ (α λ )V − ∂ λ − 2 V j+1 θ t j+1 j+1 ϑ i θ ϑ ϑ ds 2s 4 4 α 2 3 3 2 2 2 V V 1 (V ) − (V ) j+1 j+1 1/2 P 1/2 −2 m + ∂ (α e Q ) − ∂ (α P ) + O[s (ln t) ], θ θ ϑ ϑ 0 0 V 2 V (193) j j −2 −3 where c = O(s ) and c = O(s ) and we sum over i but not j. i θ The second term in the definition of Z consists of a sum of terms of the form −1/2 h ψ ◦ V α Θ . (194) The relevant h’s are 1 α 1 1 1/2 λ/2 1/2 P 1/2 P h = λ − 2 − 4s e Λ , h = − P , h = α P , h = −e Q , h = α e Q , 1 t 2 t 3 θ 4 t 5 θ 4 α 2 2 and the relevant ψ’s are 2 2 3 2 2 2 3 2 0 2 3 (V ) − (V ) (V ) − (V ) V V V 0 0 1 ψ =V , ψ = V , ψ = V , ψ = , 1 2 3 4 0 2 1 2 0 2 1 2 0 2 1 2 (V ) − (V ) (V ) − (V ) (V ) − (V ) 1 2 3 V V V ψ = . 0 2 1 2 (V ) − (V ) We want to differentiate (194) with respect to s and then N times with respect to ϑ. Before turning to the details, let us record the following estimate: j j 1/2 −3/2 m s |(∂ h )(s; t, θ, v)| + |(∂ h )(s; t, θ, v)| ≤ C s (ln t) (195) 1 i N ϑ ϑ i=2 44 1 3 for 0 ≤ j ≤ N, (t, v, θ) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]. In the case of 1 1 h , i = 2, . . ., 5, (195) is an immediate consequence of the inductive hypothesis, (135), (177) and Lemma 65, and in the case of h , it is a consequence of (182). We also have 5 5 X X j+1 −1 m |ψ (s; t, θ, v)| + s |ψ (s; t, θ, v)| ≤ C, |(∂ ψ ◦ V )(s; t, θ, v)| ≤ C s (ln t) (196) 1 i i j i=2 i=1 1 3 for 0 ≤ j ≤ N − 1, (t, v, θ) ∈ [t , ∞) × S × R in the support of f and s ∈ [t , t]; this is an 1 1 immediate consequence of the inductive hypothesis. Let us consider the term that arises when d/ds hits the ψ-factor in (194). Note, to this end, that dV −3/2 m ∂ ≤ C s (ln t) ds for all i = 1, 2, 3 and all 0 ≤ j ≤ N; for j = 0, the estimate is a consequence of (149); for j ≥ 1 and i = 1, it is a consequence of (193); and in the case of i = 2, 3, it follows immediately from (110), (111) and the induction hypothesis. Due to the above observations, dψ ◦ V N −1/2 N −2 m ∂ h α Θ = c Θ + O[s (ln t) ], 1 N+1 ϑ θ ds N −2 where c = O(s ) and h is one of h , . . . , h . When the s-derivative hits the remaining terms in 1 5 (194) (not ψ), we obtain " # 1 1 1 V ∂V V N −1/2 i i ∂ ∂ (α h) + h ψΘ + hψ − hψ V ∂V ; t θ 1 0 0 0 3 V V (V ) i=1 cf. (151). Due to (195) and (196), this expression can be written N −1/2 N i −2 mN ∂ ∂ (α h) + h ψΘ + c V + O[s (ln t) ], t θ 1 ϑ i N+1 i=1 N −2 1 where c = O(s ). Differentiating the second term appearing in the definition of Z once with respect to s and N times with respect to ϑ, we obtain (by adding up the above) 5 3 X X N −1/2 N i N −2 m ∂ ∂ (α h ) + (∂ h ) ψ Θ + c V + c Θ + O[s (ln t) ], (197) t i θ i i 1 N+1 ϑ i N+1 θ i=1 i=1 N −2 N −2 where c = O(s ) and c = O(s ). In order to obtain the desired equation we need to add θ i this expression to (193) with j = N. However, before doing so, note that 1 1 N+1 1/2 0 N 1/2 0 −2 m − ∂ (α λ )V = − ∂ [∂ (α λ )V Θ ] + O[s (ln t) ] θ θ θ 1 4 4 etc. Due to this observation, we can argue as in the proof of Lemma 56. In particular, we obtain 1 1 a formula analogous to (152): the difference is that ∂V should be replaced by V in the first N+1 term on the right hand side of (152); that ∂ should be applied to all but the first and last two terms on the right hand side of (152); and that the last two terms should be replaced by ones analogous to the last three terms on the right hand side of (197). Proceeding as in the proof of Lemma 56, the corresponding expression can be simplified; cf. the derivation of (153). Most of the steps involved in the derivation of (153) consist of algebraic manipulations. However, there are two exceptions. The combination of the fourth and fifth last terms on the right hand side of (152) can be written 1 2 3 V V V e (P Q − P Q ) Θ . t θ θ t 1 0 2 1 2 (V ) − (V ) 45 The analogous expression in the present setting is 1 2 3 V V V N P N −3 m ∂ e (P Q − P Q ) Θ = c Θ + O[s (ln t) ], t θ θ t 1 N+1 ϑ θ 0 2 1 2 (V ) − (V ) N −3 where c = O(s ) and we have used (177), Inductive assumption 61 and Lemma 65. The combination of the third and the seventh term on the right hand side of (152) can be written 1/2 λ/2 1 −∂ (s e Λ)V Θ . θ 1 In the present setting, the analogous term is N 1/2 λ/2 1 N −2 m −∂ [∂ (s e Λ)V Θ ] = c Θ + O[s (ln t) ], θ 1 N+1 ϑ θ N −2 where c = O(s ) and we have used Inductive assumption 61 and Lemma 65. Summing up, we obtain h  i dZ 1 1 α 1 N 1 N −1/2 1/2 λ/2 0 1/2 0 = − V + ∂ ∂ α λ − 2 − 4s e Λ V Θ − ∂ (α λ )V Θ t t 1 θ θ 1 N+1 ϑ ds 2s 4 α 4 2 2 3 2 1 (V ) − (V ) −1/2 1/2 0 − [∂ (α P ) − ∂ (α P )]V Θ t t θ θ 1 0 2 1 2 2 (V ) − (V ) 0 2 3 V V V −1/2 P 1/2 P −[∂ (α e Q ) − ∂ (α e Q )] Θ t t θ θ 1 0 2 1 2 (V ) − (V ) N N i −2 m + c Θ + c V + O[s (ln t) ], N+1 θ i N+1 (198) N −2 N −2 where c = O(s ), c = O(s ) and we sum over i but not over N. The term of importance is the second one on the right hand side. If all the ϑ-derivatives hit Θ , the resulting term can be N N −3/2 dealt with as in the proof of Lemma 56, and we obtain a c Θ -term, where c = O(s ). In N+1 θ θ the case of all the remaining terms, it is possible to use the equations (as in the proof of Lemma 56) −3/2 m in order to obtain terms of the form O[s (ln t) ]. Let us go through the argument in detail for 2 2 3 2 1 (V ) − (V ) N −1/2 1/2 0 ∂ − [∂ (α P ) − ∂ (α P )]V Θ . t t θ θ 1 0 2 1 2 2 (V ) − (V ) Using (61), this expression can be written −1/2 λ/2−P 2 −1/2 λ/2+P 2 1 1 α e J α e (K − QJ) N −1/2 −1/2 2P 2 2 − ∂ − α P + α e (Q − αQ ) + − ϑ t θ 7/2 7/2 2 s 2s 2s 2 2 3 2 (V ) − (V ) −1/2 λ/2 −1/2 0 N −2 m +s e α (P − P ) V Θ = c Θ + O[s (ln t) ], 2 3 1 N+1 0 2 1 2 (V ) − (V ) −2 where c = O(s ) and we have used (135), (177), Inductive assumption 61 and Lemma 65. Due to this argument, and similar ones for the remaining terms in (198), we obtain dZ 1 N 1 N N i −3/2 m = − V + c Θ + c V + O[s (ln t) ], N+1 N+1 θ i N+1 ds 2s N −3/2 N −3/2 where c = O(s ), c = O(s ) and we sum over i but not over N. Due to (186) and (187), θ i we conclude that dZ 1 N 1 N N i −3/2 m = − Z + c Ψ + c Z + O[s (ln t) ], N θ i N ds 2s N −3/2 N −3/2 where c = O(s ), c = O(s ) and we sum over i but not over N. Combining this equation with (190), (191) and (192) yields dΨ 2 N N −1/2 2 1 −2 2 m ˆ ˆ ˆ = Ψ + c Ψ + c s (ln s) Z + O[s (ln s) (ln t) ], N N θ,θ θ,1 N ds s ln s dZ N N 1/2 −2 N 1 −1 m ˆ ˆ = c s (ln s) Ψ + c Z + O[s (ln t) ], 1,θ N 1,1 N ds 46 N −2 N −3/2 N −3/2 N −3/2 where c = O(s ), c = O(s ), c = O(s ), c = O(s ), there is no summation θ,θ θ,1 1,θ 1,1 over N and we have used the notation 1 1/2 1 2 ˆ ˆ Z (s; t, θ, v) = s Z (s; t, θ, v), Ψ (s; t, θ, v) = (ln s) Ψ (s; t, θ, v). (199) N N N N Introducing the energy 2 1 2 ˆ ˆ ˆ E = (Ψ ) + (Z ) , (200) N N we conclude that dE C N N 1/2 −1 m ˆ ˆ ≥ − E − C s (ln t) E . N N ds s(ln s) Letting r be such that r (t ) = 0 and its derivative is the first factor in the first term on the N N 1 right hand side, we obtain dE N 1/2 −1 m ≥ −C s (ln t) E , ds 1/2 where E = exp(−r )E . Dividing by E and integrating from s to t, we obtain N N N 1/2 1/2 E (s; t, θ, v) ≤ E (t; t, θ, v) + C (ln t) . N N However, the first term on the right hand side can be estimated by C (ln t) . Combining the resulting estimate with (186), (187), (191) and (192), we conclude that (163) holds with j = N + 1. Corollary 68. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Let 0 ≤ k ∈ Z. Then there is a constant C , depending only on 1 0 k k and the solution, such that 1/2 1/2 −2 kP k k + kQ k k + t kα λ k k ≤ C t (201) t t θ k C C C for all t ≥ t . Moreover, α 3 3 −2 + + λ + ≤ C t t k α t t k k C C for all t ≥ t . Proof. By combining Lemmas 64 and 67, we know that Inductive assumption 61 holds for all N. In particular, we thus know that the conclusions of Lemma 65 hold for all N. Combining this information with (61) and (68) yields −1/2 −1/2 −1/2 k∂ (tα P )k k + k∂ (tα Q )k k ≤ C t . t t t t k C C As a consequence, −1/2 −1/2 1/2 ktα P k k + ktα Q k k ≤ C t . t t k C C Due to this estimate, as well as (135) and (179), we can proceed inductively in order to conclude that −2 kP k k + kQ k k ≤ C t . t t k C C Combining this estimate with (63), (135), (177) and Lemma 65, we obtain (201). Combining (201) with (60), (69) and (128), we obtain the final estimate stated in the corollary. 47 9 Energy estimates for the distribution function In the proof of the existence of f , cf. Theorem 7, a natural first step is to estimate L -based sc,∞ energies for f. In the process of deriving such estimates, it is useful to consider equations for the derivatives of the distribution function. Such equations take the following general form: 1/2 1 ∂h α v ∂h 1 ∂h + − v = R. (202) 0 i ∂t v ∂θ 2t ∂v In case h = f, R is given by ∂f R = L , (203) ∂v where 2 3 3 2 2 2 1 1 2α 3 v v 1 (v ) − (v ) 1 1/2 0 1 1/2 P 1/2 L = α λ v + λ − − v − α e Q + α P θ t θ θ 0 0 4 4 α t v 2 v −7/4 λ/4 −P/2 2 P/2 3 −t e e Jv + e (K − QJ)v , (204) 1 2 1 1 v v 2 2 1/2 L = P v + α P , (205) t θ 2 2 v 1 3 1 1 1 v v v 3 3 1/2 P 2 1/2 L = − P v − α P + e v Q + α Q . (206) t θ t θ 0 0 2 2 v v The energies we shall consider are Z Z −|β| l β 2 −1/2 −3/2 E [h](t) = t |∂ ∂ h(t, θ, v)| α t dvdθ. (207) θ v 1 3 S R l+|β|≤k We shall also use the notation E = E . −1/2 −3/2 Remarks 69. The purpose of the factor α t is to simplify some of the terms that result upon carrying out partial integrations. We could equally well consider energies of the form Z Z −|β| 1/2 2µ +2|β| l β 2 H [f](t) = t ht vi |∂ ∂ f(t, θ, v)| dvdθ θ v 1 3 S R l+|β|≤k for µ ≥ 0; cf. [31]. However, there is a constant C > 1, depending only on the solution, µ and β, such that −1 1/2 2µ +2|β| C ≤ ht vi ≤ C for t ≥ t (where t is defined as in the statement of previous lemmas) and (t, θ, v) in the support 1 1 of f. As a consequence, the corresponding weight is of no practical importance. Lemma 70. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Let h be a smooth solution to (202) (where the function α is 1 0 the object appearing in the Einstein-Vlasov equations and R is some function) which has compact support when restricted to compact time intervals. Then there is a constant C > 0, depending only on the solution to the Einstein-Vlasov equations, such that Z Z dE[h] 3 −1/2 −3/2 −2 ≤ − E[h] + 2 hRα t dvdθ + Ct E[h] dt 2t 1 3 S R for all t ≥ t . Remark 71. It is important to note that the constant C does not depend on h. Moreover, R should be thought of as being defined by (202). In particular, due to the assumptions concerning h, the function R is smooth and has compact support when restricted to compact time intervals. 48 Proof. Differentiating E with respect to time, we obtain Z Z Z Z dE 3 α −1/2 −3/2 2 −1/2 −3/2 = 2 h∂ hα t dvdθ + h − − α t dvdθ. (208) dt 1 3 1 3 2t 2α S R S R Due to (128), we can estimate the second term on the right hand side. Consider the first term on the right hand side of (208). Using (202), it can be written Z Z 1/2 1 α v ∂h 1 ∂h i −1/2 −3/2 2 h − + v + R α t dvdθ. 0 i 1 3 v ∂θ 2t ∂v S R The term involving ∂ h can be integrated to zero. The term involving R we leave as it is. What remains is to estimate the term Z Z Z Z 1 ∂h 3 i −1/2 −3/2 2 −1/2 −3/2 v α t dvdθ = − h α t dvdθ. 2t 1 3 ∂v 2t 1 3 S R S R The lemma follows. Let us turn to the higher order derivatives of the distribution function. Lemma 72. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Fix 0 ≤ k ∈ Z. Then there is a constant C > 0, depending 1 0 k only on k and the solution to the Einstein-Vlasov equations, such that dE [f] 3 −3/2 ≤ − E [f] + C t E [f] k k k dt 2t 3/2 for all t ≥ t . In particular, t E [f] is bounded to the future. 1 k Proof. Differentiating (202) with h = f, we obtain 1/2 1 1/2 1 ∂f α v ∂f 1 ∂f α v 1 β,l β,l β,l i β l β l i β l + − v = ∂ ∂ R + ∂ , ∂ ∂ f − v ∂ i, ∂ ∂ f, (209) v θ v θ v v θ 0 i 0 ∂t v ∂θ 2t ∂v v 2t β l where we use the notation f = ∂ ∂ f and assume that |β| +l ≤ k. Let us denote the right hand β,l v θ side of (209) by R . Due to Lemma 70, it is of interest to estimate β,l Z Z −|β| −1/2 −3/2 2 t f R α t dvdθ. (210) β,l β,l 1 3 S R By an inductive argument, it can be proven that the third term on the right hand side of (209) is given by |β|f /2t. The corresponding contribution to (210) is thus β,l |β| −|β| t E[f ]. β,l Turning to the second term on the right hand side of (209), it can (up to numerical factors) be written as a sum of terms of the form 1/2 1 α v β l β l +1 1 1 2 2 ∂ ∂ ∂ ∂ f, v v θ θ where β +β = β, l +l = l and |β | +l ≥ 1. Note that the first factor can always be estimated 1 2 1 2 1 1 −3/2 −2 by C t . In case β = 0, it can be estimated by C t (on the support of f). Due to these k 1 k observations, we have 1/2 1 X α v −|β|/2 β l −2 −|β |/2 t ∂ , ∂ ∂ f ≤ C t t |f |. θ k β ,l v θ 1 1 l +|β |≤k 1 1 49 The corresponding contribution to (210) can thus be estimated by −2 C t E [f]. k k Finally, let us consider the first term on the right hand side of (209). Since R is given by (203), β l the expression ∂ ∂ R is given by the sum of v θ i β l L ∂ i∂ ∂ f (211) v v θ and terms which (up to numerical factors) can be written l l β 1 i β 2 1 2 (∂ ∂ L )∂ i∂ ∂ f, (212) v v v θ θ where |β | + l ≥ 1. The contribution to (210) from (211) can be written 1 1 Z Z −|β| i 2 −1/2 −3/2 t L (∂ if )α t dvdθ. v β,l 1 3 S R i i Integrating partially with respect to v and keeping in mind that the L are given by (204)–(206), we conclude that this expression can be estimated by −3/2 C t E [f], (213) k k where we have used Lemma 65 and Corollary 68. Let us now consider the contribution arising from terms of the form (212). It is natural to divide these terms into two different categories. Either β = 0, or β 6= 0. In case β = 0, the expression (212) can be estimated by 1 1 1 −3/2 −1/2 β 2 C t t |∂ i∂ ∂ f|. v v In case β 6= 0, the expression (212) can be estimated by −3/2 β l 2 2 C t |∂ i∂ ∂ f|. k v In order to obtain these estimates, we have appealed to Lemma 65 and Corollary 68. As a consequence, the contribution to (210) from terms of the form (212) can be estimated by (213). Adding up the above observations, we conclude that d |β| dE[f ] |β| 3 β,l −|β| −|β| −|β| −|β| −|β| t E[f ] = − t E[f ] + t ≤ − t E[f ] − t E[f ] β,l β,l β,l β,l dt t dt t 2t Z Z −|β| −1/2 −3/2 −2 −|β| + 2 t R f α t dvdθ + Ct t E[f ] β,l β,l β,l 1 3 S R −|β| −3/2 ≤ − t E[f ] + C t E [f]. β,l k k 2t Summing over β and l, we obtain dE [f] 3 −3/2 ≤ − E [f] + C t E [f]. k k k dt 2t The lemma follows. In order to obtain a better understanding for the asymptotics, it is convenient to rescale the distribution function according to −1/2 f (t, θ, v) = f(t, θ, t v). sc We have the following conclusions concerning f . sc 50 2 Lemma 73. Consider a T -symmetric solution to the Einstein-Vlasov equations with a cosmo- logical constant Λ > 0 and existence interval (t , ∞), where t ≥ 0. Assume that the solution has 0 0 λ-asymptotics and let t = t + 2. Fix 0 ≤ k ∈ Z. Then there is a constant C, depending only on 1 0 1 3 the solution, such that in order for (t, θ, v) ∈ [t , ∞) × S × R to be in the support of f , v has 1 sc to satisfy |v| ≤ C. Moreover, there is a constant C > 0, depending only on k and the solution to the Einstein-Vlasov equations, such that −2 k∂ f (t, ·)k k 1 3 ≤ C t t sc k C (S ×R ) for all t ≥ t . In particular, there is thus a smooth, non-negative function with compact support, 1 3 say f , on S × R such that sc,∞ −1 kf (t, ·) − f k k 1 3 ≤ C t sc sc,∞ k C (S ×R ) for all t ≥ t . Proof. The statement concerning the support is an immediate consequence of Lemmas 50 and 53. In order to derive the desired estimates, let us compute −1/2 −1/2 i −1/2 (∂ f )(t, θ, v) = (∂ f)(t, θ, t v) − t v (∂ if)(t, θ, t v). t sc t v 2t Using the Vlasov equation, we conclude that 1/2 −1/2 1 α t v ∂ f = − ∂ f + R , t sc θ sc sc −1/2 ht vi where i −1/2 −1/2 1/2 i −1/2 R (t, θ, v) = L (t, θ, t v)(∂ if)(t, θ, t v) = t L (t, θ, t v)(∂ if )(t, θ, v) sc v v sc and the L are defined in (204)–(206). Introducing i 1/2 i −1/2 L (t, θ, v) = t L (t, θ, t v), sc we thus have 1/2 −1/2 1 α t v ∂ f = − ∂ f + L ∂ if . (214) t sc θ sc v sc sc −1/2 ht vi Due to the properties of the support of f , Lemma 65 and Corollary 68, there is a constant C sc k for each 0 ≤ k ∈ Z such that β l i −2 |(∂ ∂ L )(t, θ, v)| ≤ C t (215) v θ sc l+|β|≤k 1 3 for all i = 1, 2, 3 and (t, θ, v) ∈ [t , ∞) × S × R in the support of f . In order to estimate the 1 sc k −3/2 derivatives of f in C , it is convenient to translate the estimate E ≤ C t into an estimate sc k k for f . However, sc Z Z Z Z l β 2 −|β| l β −1/2 2 |(∂ ∂ f )(t, θ, v)| dθdv = t |(∂ ∂ f)(t, θ, t v)| dθdv sc θ v θ v 1 3 1 3 S R S R Z Z 3/2 −|β| l β 2 3/2 =t t |(∂ ∂ f)(t, θ, v)| dθdv ≤ Ct E (t) ≤ C , k k θ v 1 3 S R assuming l+|β| ≤ k. Due to this estimate and Sobolev embedding, we conclude that all derivatives of f are bounded for t ≥ t . Combining this observation with (214) and (215), we conclude that sc 1 β l −2 |∂ ∂ ∂ f | ≤ C t t sc k v θ l+|β|≤k for t ≥ t . The lemma follows. 51 10 Proof of the main theorems Finally, we are in a position to prove the main theorems. Let us begin with Theorem 7. Theorem 7. To begin with, the conclusions concerning the distribution function are direct conse- quences of Lemma 73. Turning to H and G, we have −5/2 −1/2 P +λ/2 −5/2 −1/2 −P +λ/2 H = −t α e (K − QJ), G = −QH − t α e J; (216) t t t cf. (58). In order to estimate H and G , it is useful to note that P and Q are bounded in every t t C -norm for t ≥ t ; this follows by integrating (201). Combining this observation with (66), (67), (88) and Lemma 65 yields −2 m kJ k N + kK k N ≤ C t (ln t) . t t N C C Thus J and K are uniformly bounded in C . Combining this observation with (216), (135), (88), the bound on P and Q in every C norm and Lemma 65, we conclude that (8) holds. Due to Corollary 68, we know that (9) and (10) hold. Combining (10) with (88) and (135) yields (11). Let us turn to the second fundamental form. By definition, 1/4 −λ/4 1/4 −λ/4 1/4 −λ/4 ¯ ¯ k = k(∂ , ∂ ) = h∇ e , ∂ i = h∇ (t e ∂ ), ∂ i = t e h∇ ∂ , ∂ i = t e ∂ g , ij i j ∂ 0 j ∂ t j ∂ t j t ij i i i where we have used the fact that ∂ and ∂ are perpendicular. In what follows, we would like to t i prove that kk − Hg¯ k N ≤ C , (217) ij ij N 1/2 where H = (Λ/3) . Consider the spatial components of the metric (1). If a time derivative hits one of P , Q, G or H in such a component, then the resulting expression is bounded in C after 1/4 −λ/4 it has been multiplied by t e /2; this is due to (8), (9) and (11). As a consequence, what we need to consider are the components of the tensor field 1 1 1 α 1/4 −λ/4 −1/2 λ/2 −1 2 P 2 t e − + λ − t e α dθ + e [dx + Qdy + (G + QH)dθ] 2 2t 2 α 1 1 1 3 α 3 −P 2 −3/4 −λ/4 −1/4 −1 λ/4 2 + e (dy + Hdθ) = t e g¯ + t α e λ + − + dθ , 2 2 2 t α t where g¯ is the spatial part of the metric. Note that the components of the second term on the right hand side are bounded in C ; this is a consequence of (10) and (11). Moreover, −3/4 −λ/4 −λ/4 t e = He , where λ is defined in (84). In order to prove (217), it is thus sufficient to demonstrate that −λ/4 k(e − 1)g¯ k N ≤ C . ij N −λ/4 N −1 N However, t(e − 1) is bounded in C due to (11) and t g¯ is bounded in C due to (8), (9) ij and (11). Thus (217) holds. Let us define g¯ by (13); note that this is a smooth Riemannian metric on T . Moreover, −1 −1 kt g¯ (t, ·) − g¯ k N ≤ C t (218) ij ∞,ij N due to (8), (9) and (11). Combining this estimate with (217), we obtain (12). The proof of future causal geodesic completeness is not very complicated, given the above estimates. It is, e.g., possible to proceed as in the proof of [28, Propositions 3 and 4, pp. 189–191]. However, we shall not write down the details, since the result follows from the proof of Theorem 35. Let us now turn to the proof of the cosmic no-hair conjecture. 52 Theorem 14. We need to verify that the conditions stated in Definition 8 are fulfilled. To begin with, note that Σ = {t} × T is a Cauchy hypersurface for each t ∈ (t , ∞). An argument is t 0 required in order to justify this statement, but since the details are quite standard (cf., e.g., the proof of [30, Proposition 20.3, p. 215], in particular [30, p. 217]), we have omitted the details. Let γ = (γ , γ¯) be a future directed and inextendible causal curve, defined on I = (s , s ). γ − + Reparametrising the curve, if necessary, we can assume that γ (s) = s and that I = (t , ∞). Due γ 0 to the causality of the curve, we know that i j −2 ˙ ˙ g¯ [γ(t)]γ¯ (t)γ¯ (t) ≤ −g [γ(t)] ≤ Ct ij 00 for t ≥ t , where t = t + 2 and we have used (11). Combining this estimate with (12), we 1 1 0 conclude that there is a constant K > 1 (independent of the curve γ, as long as γ (t) = t) such that i j 2 −2 −3 ˙ ˙ g¯ [γ¯(t)]γ¯ (t)γ¯ (t) ≤ K H t ∞,ij 3 −1 −1/2 for all t ≥ t . In particular, there is thus an x¯ ∈ T such that d [γ¯(t), x¯ ] ≤ K H t for all 1 0 ∞ 0 0 t ≥ t , where d is the topological metric on T induced by g¯ . Let ǫ > 0 denote the injectivity 1 ∞ ∞ inj radius of (T , g¯ ); the injectivity radius of a point p of a Riemannian manifold, denoted inj(p), is defined in [20, Definition 9.2, p. 142], and the injectivity radius of a Riemannian manifold is the infimum of the injectivity radii of the points of the manifold; that ǫ > 0 follows from the inj continuity of inj, cf. [20, p. 178]; readers interested in a more quantitative bound on the injectivity radius are referred to [20, Lemma 51, p. 319]. Then, given x¯ ∈ T , there are normal coordinates on B (x¯), where distances are computed using d ; cf. [19, pp. 72–73] for a definition of normal ǫ ∞ inj 2 −2 −2 coordinates. Fix t > K H ǫ +1 (note that t is independent of the curve). Due to the above − − inj arguments and definitions, − + 3 −1 −1/2 J (γ) ∩ J (Σ ) ⊆ {(t, x¯) ∈ I × T : t ≥ t , d (x¯, x¯ ) ≤ K H t } (219) t − ∞ 0 0 −1/2 −1 and the closed ball of radius K H t (with respect to d ) and centre x¯ is contained in the 0 ∞ 0 domain of definition of normal coordinates ¯x with centre at x¯ . Denote the set appearing on the right hand side of (219) by D . Let us define t ,K ,x ¯ − 0 0 2Hτ −1 ¯ ¯ ψ(τ, ξ) = [e , ¯x (ξ)]. Then −1 3 −1 −Hτ ¯ ¯ ψ (D ) = {(τ, ξ) ∈ J × R : τ ≥ T , |ξ| ≤ K H e }, t ,K ,x ¯ 0 0 − 0 0 −1 −1 where T = H ln t /2, and J = (τ , ∞), where τ = H ln t /2; if t = 0, then τ = −∞. 0 − 0 0 0 0 0 Letting T be slightly smaller than T and K be slightly larger than K , the map ψ is still defined 0 0 −1 −Hτ on C ; cf. (15). In analogy with Definition 8, let D = ψ(C ) and R(τ) = KH e . Λ,K,T Λ,K,T Due to the above arguments, we have already verified all of the requirements of Definition 8 (with Σ = Σ etc.) but the last one; i.e., that (16) hold. In order to proceed, let ¯ g denote the components of g¯ with respect to the coordinates ¯x. Let ∞,ij ∞ 2Hτ 2Hτ ¯ ¯ g (τ, ·) and k (τ, ·) denote the components of g¯(e , ·) and k(e , ·), respectively, with respect ij ij ¯ ¯ ¯ to the coordinates x. Moreover, consider g , g (τ, ·) and k (τ, ·) to be functions on the image ∞,ij ij ij of ¯x; i.e., on B (0), with the origin corresponding to x¯ . Note that the estimates (217) and (218) ǫ 0 inj hold with g¯ replaced by g¯ etc., assuming the domain on which the C norm is computed is ij ij suitably restricted. In particular, letting S be as in Definition 8, the following estimate holds: −2Hτ −2Hτ −2Hτ ke k (τ, ·) − H¯ g k N + ke g¯ (τ, ·) − g¯ k N ≤ C e ij ∞,ij ij ∞,ij N C (S ) C (S ) τ τ for all τ ≥ T . Note that g¯ (0) = δ , (∂ g¯ )(0) = 0 ∞,ij ij l ∞,ij by the definition of the coordinates x. As a consequence, if ξ ∈ S , then −Hτ ¯ ¯ |(∂ g¯ )(ξ)| = [(∂ g¯ )(sξ)]ds ≤ Ce . l ∞,ij l ∞,ij ds 53 Moreover, −2Hτ |g¯ (ξ) − δ | ≤ Ce ∞,ij ij for τ ≥ T and ξ ∈ S . In particular, we thus have −2Hτ −2Hτ −2Hτ ke k (t, ·) − Hδ k 0 + ke ¯ g (t, ·) − δ k 0 ≤ Ce ij ij C (S ) ij ij C (S ) τ τ for all τ ≥ T . Letting g¯ (τ, ·) and k (τ, ·) be defined as in Definition 8, we conclude, in particular, dS dS that −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k 0 + kk (τ, ·) − k(τ, ·)k 0 ≤ Ce dS dS C (S ) C (S ) τ τ dS dS for all τ ≥ T . In fact, due to the above estimates, we have −2Hτ ¯ ¯ kg¯ (τ, ·) − g¯(τ, ·)k N + kk (τ, ·) − k(τ, ·)k N ≤ C e dS dS N C (S ) C (S ) τ τ dS dS for all τ ≥ T . The theorem follows. Finally, we are in a position to prove Theorem 35. Theorem 35. The idea of the proof is to demonstrate that for late enough t, there is a neighbour- hood of each point in {t} × T such that Theorem 29 applies in the neighbourhood; combining this observation with Cauchy stability, cf. Theorem 34, then yields the desired result. Fixing N, there is an ǫ > 0 and a constant C such that for every x¯ ∈ T , there are normal coordinates ¯x on W = B (x¯) with respect to g¯ , where distances on T are measured using the ǫ ∞ topological metric induced by g¯ . Moreover, if g¯ are the components of g¯ with respect to ¯x ∞ ∞,ij ∞ ij ij and g¯ are the components of the inverse, then the derivatives of g¯ and g¯ up to order N with ∞,ij ∞ ∞ respect to the coordinates ¯x on W are bounded by C . Moreover, the derivatives of ¯x, considered −1 as functions of (θ, x, y), up to order N + 1 are bounded by C . Similarly, the derivatives of ¯x up to order N + 1 are bounded by C . The arguments required in order to prove the above statements are similar to those presented in the proof of [31, Lemma 34.9, p. 650]. The important point is that we obtain uniform bounds which hold regardless of the base point. K −K 1/2 Define K by the condition e = 4/H and define the coordinates ¯y = e t ¯x on W . Note that the range of ¯y is B −K 1/2 (0). For t large enough (the bound being independent of the base point e t ǫ −K 1/2 x¯), we consequently have e t ǫ > 1. From now on, we assume t to be large enough that this is −1 the case. Moreover, we assume the coordinates ¯y to be defined on the image of B (0) under ¯y . Let g¯ denote the components of g¯(t, ·) with respect to the coordinates ¯y. Moreover, let g¯ ij ∞,ij denote the components of g¯ with respect to the coordinates ¯x. Due to (218), we have α −2K −1 −1−|α|/2 |∂ [(e g¯ − g¯ ) ◦ ¯y ](ξ)| ≤ C t ¯ ij ∞,ij N for ξ ∈ B (0) and |α| ≤ N; note that −1 −1 K −1/2 ¯ ¯ ¯y (ξ) = ¯x (e t ξ). −1 Since ¯ g ◦ ¯y (0) = δ , ∞,ij ij −2K −1 −1/2 |e ¯ g ◦ ¯y − δ | ≤ C t (220) ij ij N on B (0); in particular, (38) holds with a margin for t large enough. Similarly, −1 −1 l −1 |∂ (g¯ ◦ ¯y )(u)| = ∂ ∂ (g¯ ◦ ¯y )(su)u ds ≤ Ct m ∞,ij l m ∞,ij l=1 on B (0). To conclude, we thus have −2K −1 −1 ke (∂ g¯ ◦ ¯y )k N−1 ≤ C t . (221) k ij N C [B (0)] 54 Due to (217), we also have −1 −1 ¯ ¯ k(k − Hg ) ◦ y k N ≤ C t , (222) ij ij C [B (0)] N ¯ ¯ where k denotes the components of the second fundamental form k calculated using the coor- ij dinates ¯y. In the end, we shall choose K = ln t/2. As a consequence, (221) and (222) imply Vl that (39) holds with a margin (note that the T -symmetric background solution is such that φ , i = 0, 1, vanish), assuming N ≥ 5; note that in order to prove Theorem 35, it is sufficient to apply Theorem 29 with k = 4. Let us turn to the distribution function. First of all, recall that if Σ is a spacelike hypersurface in a Lorentz manifold, and f is a distribution function defined on the mass shell, then the initial datum for the distribution function (denoted f and defined on T Σ) induced by f on Σ is given by −1 f = f ◦ pr , where pr is the projection from the mass shell over Σ to T Σ; in other words, if p ∈ P for some r ∈ Σ and N is the future directed unit normal to Σ at r, then pr (p) is the element of T Σ r r corresponding to p + g(p, N )N . In our case, we are interested in the hypersurface Σ = {t} × T . r r If z = (t, θ, x, y), i α f p¯ e | = f (p e | ) = f(t, θ, p¯), i α z z 1 2 3 where p¯ = (p¯ , p¯ , p¯ ), 0 1 2 2 2 3 2 1/2 p = [1 + (p¯ ) + (p¯ ) + (p¯ ) ] i i and p = p¯ . However, in the application of Theorem 29, we need to express f with respect to the coordinates ¯y. Consequently, we are interested in i i j ¯ ¯ ¯ f(z¯, p¯) = f p¯ ∂ i = f p¯ A (z¯) e | = f[t, θ, v(z¯, p¯)], ¯ y j z z where z = (t, z¯), i 1 i 2 i 3 v(z¯, p¯) = (p¯ A (z¯), p¯ A (z¯), p¯ A (z¯)) i i i and A is defined by the requirement that ∂ i = A (z¯) e | . ¯ y j z z Thus ∂¯z K −1/2 K −1/2 A (z¯) = ∂ i , e | = e t ∂ i| , e | = e t (z) ∂ l| , e | , j j j ¯ y ¯ x ¯ z i z z z z z z i ∂¯x where ¯z correspond to the standard coordinates on the torus (which are locally well defined). In 1 2 3 particular, ∂ = ∂ , ∂ = ∂ and ∂ = ∂ . Due to the observations made at the beginning of ¯ z θ ¯ z x ¯ z y the proof, (7) and Theorem 7, it is clear that all derivatives of A up to order N are uniformly bounded on the domain of ¯y, the bound being independent of the base point x¯ and the time t (assuming t to be sufficiently large). What we need to estimate is Z Z −w 2|β| w 2µ +2|β| α 2 ¯ ¯ ¯ (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯, ¯ p¯ y R ¯ y(U) |α|+|β|≤k cf. (37) and (40), where −1 −1 K −1/2 1 −1 K −1/2 −1 K −1/2 ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ f (ξ, p¯) = f[¯y (ξ), p¯] = f[¯x (e t ξ), p¯] = f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)] −1 1 −1 and the constant w remains to be specified. Note that all derivatives of ¯x and ¯z ◦ ¯x up to order N are uniformly bounded. As a consequence, γ 1 −1 K −1/2 −1 K −1/2 −1/2 ¯ ¯ ¯ ¯ ¯ ∂ f (ξ, p¯) = ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]ψ (t ξ) ¯ y γ p¯ v |γ|=|β| 55 for functions ψ with bounded derivatives; note that ∂v −1 K −1/2 i −1 K −1/2 ¯ ¯ (¯x (e t ξ), p¯) = A ◦ ¯x (e t ξ). ∂p¯ ¯ ¯ As a consequence, ∂ ∂ f (ξ, p¯) consists of sums of terms of the form ¯ p¯ y −|α|/2 l γ+δ 1 −1 K −1/2 −1 K −1/2 −1/2 λ ¯ ¯ ¯ t ∂ ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]φ (t ξ)p¯ , γ,δ,l θ v λ 1 λ 2 λ 3 λ 1 2 3 where |λ| = |δ|, |γ| = |β|, l + |δ| ≤ |α|, φ are bounded functions and p¯ = (p¯ ) (p¯ ) (p¯ ) . γ,δ,l On the other hand, we know that 1/2 f(t, θ, v) = f (t, θ, t v), sc where f converges to a smooth function with compact support with respect to every every C - sc norm. Moreover, we know that f has uniformly compact support. Note also that sc l γ+δ (|γ|+|δ|)/2 l γ+δ 1/2 ∂ ∂ f(t, θ, v) = t ∂ ∂ f (t, θ, t v). sc θ v θ v Since there is a uniform constant C > 1 (independent of t (large enough) and the base point x¯) such that −1 −1 K −1/2 C |p¯| ≤ |v(¯x (e t ξ), p¯)| ≤ C|p¯|, we conclude that −|α|/2 l γ+δ 1 −1 K −1/2 −1 K −1/2 −1/2 λ ¯ ¯ ¯ t |∂ ∂ f[t,¯z ◦ ¯x (e t ξ), v(¯x (e t ξ), p¯)]φ (t ξ)p¯ | γ,δ,l θ v (|γ|−|α|)/2 1/2 ≤C t χ(t p¯), α,β where χ is a smooth function with compact support. As a consequence, α (|β|−|α|)/2 1/2 ¯ ¯ |∂ ∂ f (ξ, p¯)| ≤ C t χ(t p¯) ¯ ¯ y α,β p¯ for ξ ∈ B (0). Let us now define w = K + K , where K = ln t/2. Then 1 Vl Vl Z Z −w 2|β| w 2µ +2|β| α 2 ¯ ¯ ¯ (e ) he p¯i |∂ ∂ f | (ξ, p¯)dξdp¯ ¯ p¯ ¯ y R ¯ y(U) Z Z 2 −2|β|K −|β| K 1/2 2µ +2|β| |β|−|α| 2 1/2 −|α|−3/2 ≤C e t he t p¯i t χ (t p¯)dξdp¯ ≤ C t . ,µ α,β α,β R B (0) The square root of the right hand side of this expression should be compared with the right hand side of (40): 2 5/2 −3K/2−K 2 5/2 −3K/2 −1/2 Vl H ε e = H ε e t . Clearly, we have a margin. As a consequence, for t large enough there is, for every x¯ ∈ T , a neighbourhood of x¯ such that (40) holds with a margin. Due to (220), (221) and (222), we also know that (38) and (39) hold with a margin. We can thus apply Theorem 29 with the given µ > 5/2 3 −1 and k = 4. In addition, the covering of T obtained by taking the neighbourhoods ¯y [B (0)] 0 1/4 corresponding to varying base points x¯ has a finite subcovering. Appealing to Cauchy stability, Theorem 34, we conclude that there is an ǫ > 0 with the properties stated in the theorem. Acknowledgements The second author would like to acknowledge the support of the G¨oran Gustafsson Foundation, the G¨oran Gustafsson Foundation for Research in Natural Sciences and Medicine, and the Swedish Research Council. This article was in part written during a stay of both authors at the Erwin Schr¨odinger Institute in Vienna during the programme ’Dynamics of General Relativity: Black Holes and Asymptotics’. 56 A Derivation of the equations The purpose of this appendix is to compute the Einstein tensor associated with the metric (1), and to derive an expression for the Vlasov equation. Let us begin by expressing g with respect to suitable one-form fields. Let 0 −1/4 λ/4 ξ = t e dt, 1 −1/4 λ/4 −1/2 ξ = t e α dθ, 2 1/2 P/2 ξ = t e (dx + Qdy + (G + QH)dθ), 3 1/2 −P/2 ξ = t e (dy + Hdθ). With respect to these one-form fields, the metric can be written 0 0 i i g = −ξ ⊗ ξ + ξ ⊗ ξ . i=1 α α Using the orthonormal frame {e } introduced in (7), it can be verified that ξ (e ) = δ . α β A.1 Commutators Let us compute the commutators, in other words the functions γ such that βζ [e , e ] = γ e . β ζ α βζ α α α Clearly, γ = −γ . Consequently, it is sufficient to compute γ for β < ζ. By a straightforward βζ ζβ βζ computation, 0 1/4 −λ/4 1/2 γ = t e α λ , (223) 1 α 1 1 1/4 −λ/4 γ = − t e λ − 2 − , (224) 4 α t 2 −3/2 −P/2 γ = t e J, (225) 3 −3/2 P/2 γ = t e (K − QJ), (226) where we have used the notation (58). Turning to γ , the only α for which this object is non-zero is α = 2, and we then have 2 1/4 −λ/4 −1 γ = − t e t + P . Turning to γ , the only α’s for which this object is non-zero is α = 2 and α = 3, and we then have 2 1/4 −λ/4 P γ = −t e e Q , 3 1/4 −λ/4 −1 γ = − t e t − P . Turning to γ , the only α for which this object is non-zero is α = 2, and we then have 2 1/4 −λ/4 1/2 γ = − t e α P . Turning to γ , the only α’s for which this object is non-zero is α = 2 and α = 3, and we then have 2 1/4 −λ/4 1/2 P γ = −t e α e Q , 3 1/4 −λ/4 1/2 γ = t e α P . 57 α Finally, γ = 0. For future reference, let us make the following observations: A −3/4 −λ/4 γ = −t e , 0A i 1 −3/4 −λ/4 γ = γ − t e , 0i 01 α 1 −3/4 −λ/4 γ = γ − t e , 0α 01 γ = 0, 1A γ = 0, 1i α 0 γ = −γ , 1α 01 where Greek indices range from 0 to 3, lower case Latin indices range from 1 to 3 and capital Latin indices range from 2 to 3; moreover, Einstein’s summation convention is enforced. Non-zero components. Note that in order for γ to be non-zero, one of the following conditions βζ has to be satisfied: • {β, ζ} = {0, 1}, • one of β and ζ is in {0, 1}; the other is in {2, 3}; and α is in {2, 3}. A.2 Connection coefficients Define the connection coefficients Γ by the relation βζ ∇ e = Γ e , e ζ α β βζ where ∇ is the Levi-Civita connection associated with the metric g. Note that 0 i Γ = −h∇ e , e i, Γ = h∇ e , e i, e ζ 0 e ζ i βζ β βζ β since the frame is orthonormal. Let us begin by recording some symmetries of this object. Symmetries of the connection coefficients. Due to the metricity of the connection and the fact that the basis is orthonormal, we have Γ = 0 (no summation over α). βα For similar reasons, Γ is anti-symmetric in the indices i and j. Moreover, since [e , e ] is per- i j αj pendicular to e , we have Γ = h∇ e , e i = −he , ∇ e i = −he , ∇ e i = h∇ e , e i = Γ . e 0 i 0 e i 0 e j e 0 j j0 j j i i i0 i i 0 Thus, Γ is symmetric in i and j. Note that the computation also demonstrates that Γ = Γ . j0 j0 ij Similarly, since [e , e ] = 0, we have A B A B Γ = h∇ e , e i = −he , ∇ e i = −he , ∇ e i = h∇ e , e i = Γ . eB 1 A 1 eB A 1 eA B eA 1 B B1 A1 A A 1 In particular, Γ is symmetric in the indices A and B, and Γ = −Γ . B1 B1 AB Connection coefficients including two or more zero indices. That Γ = 0 follows from α0 the above. Moreover, using the Koszul formula, it can be computed that 0 1 0 Γ = Γ = γ 01 00 01 0 A and that the remaining components satisfy Γ = Γ = 0. 0A 00 Connection coefficients including exactly one zero index. As already mentioned, Γ is 0j i 0 i i antisymmetric and Γ = Γ is symmetric. It is thus sufficient to compute Γ for i < j and Γ j0 ij 0j j0 for i ≤ j. We have 1 A Γ = − γ , 0A 01 2 2 Γ = γ . 03 03 58 Moreover, i i Γ = −γ (no summation over i), i0 0i 1 A Γ = − γ , A0 01 2 2 Γ = − γ . 30 03 Connection coefficients including no zero index. Note that, due to the Koszul formula and the properties of the commutators, the only Γ ’s which are non-zero are the ones that have one jk B 1 A index equalling 1 and two indices in {2, 3}. Moreover, since Γ = −Γ is symmetric and Γ A1 AB 1B is antisymmetric, it is sufficient to calculate that 2 2 Γ = γ , 13 13 A A Γ = −γ (no summation over A), A1 1A 2 2 Γ = − γ . 31 13 For future reference, it is of interest to note that α 0 Γ = γ , (227) α1 01 Γ = 0, (228) i1 α 1 −3/4 −λ/4 Γ = −γ + t e , (229) α0 01 Γ = 0. (230) αA A.3 Twist quantities The quantities J and K have been defined in two different ways; one definition is given in (2) and another is given in (58). In the present subsection, we wish to verify that these two definitions yield the same result. In the proof of this statement, it is useful to introduce different notation for the different definitions. Let us therefore denote the J and the K defined in (2) by J and tw K respectively. The quantities defined in (58) we, however, still refer to as J and K. Due to tw the fact that we have calculated the connection coefficients using the orthonormal frame {e }, it is convenient to carry out the computations relative to this frame. Note, to this end, that 1/2 P/2 X = t e e , 1/2 −P/2 1/2 P/2 Y = t e e + t e Qe , 3 2 α α where X = ∂ and Y = ∂ . In particular, if X and Y denote the components of X and Y with x y respect to the frame {e }, then 2 1/2 P/2 α • X = t e and X = 0 for α 6= 2, 2 1/2 P/2 3 1/2 −P/2 0 1 • Y = t e Q, Y = t e and Y = Y = 0. Consequently, α β ζ δ 2 3 ζ δ 0 1 1 0 1 0 J = ǫ X Y ∇ X = ǫ X Y ∇ X = t(∇ X − ∇ X ) = −t(∇ X + ∇ X ), (231) tw αβζδ 23ζδ 0 1 where the indices are frame indices, and we assume the orientation of M to be such that ǫ = 1. Similarly, 1 0 K = −t(∇ Y + ∇ Y ). (232) tw 0 1 59 1 What remains is thus to calculate ∇ X etc. However, β β β ζ ζ β ζ ∇ X = ξ (∇ X) = ξ [e (X )e + X ∇ e ] = e (X ) + X Γ . α e α ζ e ζ α α α αζ The calculation for Y is the same. In particular, 1 0 ζ 1 0 2 1 0 2 2 ∇ X + ∇ X = X (Γ + Γ ) = X (Γ + Γ ) = −X γ . 0 1 0ζ 1ζ 02 12 01 Combining this calculation with (225) and (231) yields 2 2 3/2 P/2 −3/2 −P/2 J = tX γ = t e t e J = J. tw Thus J = J . Next, let us calculate tw 1 0 ζ 1 0 2 1 0 3 1 0 2 2 3 3 ∇ Y + ∇ Y = Y (Γ + Γ ) = Y (Γ + Γ ) + Y (Γ + Γ ) = −Y γ − Y γ . 0 1 0ζ 1ζ 02 12 03 13 01 01 Combining this observation with (225), (226) and (232) yields 2 2 3 3 3/2 P/2 −3/2 −P/2 3/2 −P/2 −3/2 P/2 K = tY γ + tY γ = t e Qt e J + t e t e (K − QJ) = K. tw 01 01 To conclude, we thus have J = J and K = K. tw tw A.4 Auxiliary computations In order to simplify future calculations, let us make some observations concerning the derivatives of the connection coefficients. Note, to begin with, that all γ (no summation) can be written 0i i 1/4 −λ/4 γ = −t e f 0i for suitably chosen functions f . Thus, it can be computed that 1 1 3 i 1/4 −λ/4 −1/2 −e (γ ) = − t e ∂ − exp − λ + ln α − ln t tα f 0 t i 0i 4 2 4 1 α 1 1/4 −λ/4 i 1/4 −λ/4 −1 i −1/2 −λ/2 1/2 −1/2 = t e λ − 2 − γ + t e t γ + t e α ∂ (tα f ), t t i 0i 0i 4 α t i 1 where we do not sum over i. However, the factor in front of γ in the first term is given by −γ 0i i 2 3 and the factor in front of γ in the second term is given by −γ − γ . Consequently, 02 03 0i 0 α 0 −1/2 −λ/2 1/2 −1/2 e (Γ ) + Γ Γ = t e α ∂ (tα f ), 0 t i ii α0 ii where we sum over α but not over i. In particular, we thus have 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (λ − 2α α − t )], (233) 0 t t t 11 α0 11 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (t + P )], (234) 0 t t 22 α0 22 0 α 0 −1/2 −λ/2 1/2 −1/2 −1 e (Γ ) + Γ Γ = t e α ∂ [tα (t − P )]. (235) 0 t t 33 α0 33 0 2 3 1/4 −λ/4 Next, note that γ , γ and γ all can be written as h = t e f for suitably chosen functions i i 01 12 13 f . Moreover, 1/4 −λ/4 1/2 1/4 −λ/4 1/2 −λ/2 1/2 0 e (h ) = t e α ∂ (t e f ) = t e α ∂ f − γ h . 1 i θ i θ i i α 0 Since Γ = γ , we conclude that α1 01 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα λ ), (236) 1 θ θ 00 α1 00 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = − t e α ∂ (tα P ), (237) 1 θ θ 22 α1 22 1 α 1 −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα P ). (238) 1 33 α1 33 θ θ 60 2 2 The expressions γ and γ require a somewhat different treatment. However, similar arguments 03 13 yield 1 1 1 1 α −1/2 −λ/2−P 1/2 1/2 2P 3 2 2 e (Γ ) + Γ Γ = − t e α ∂ (tα e Q ) − (γ − γ )γ , (239) 1 θ θ 23 23 α1 13 12 13 2 2 1 1 0 0 α −1/2 −λ/2−P 1/2 −1/2 2P 3 2 2 e (Γ ) + Γ Γ = t e α ∂ (tα e Q ) + (γ − γ )γ . (240) 0 t t 23 23 α0 03 02 03 2 2 A.5 Ricci curvature The Ricci curvature is given by X X Ric(e , e ) = ǫ hR e , e i = ǫ h∇ ∇ e − ∇ ∇ e − ∇ e , e i β ζ α e e ζ α α e e ζ e e ζ ζ α α β α β β α [e ,e ] α β α α δ δ δ λ = ǫ h∇ (Γ e ) − ∇ (Γ e ) − γ Γ e , e i α e δ e δ λ α α βζ β αζ αβ δζ δ δ λ δ δ λ δ λ = ǫ he (Γ )e + Γ Γ e − e (Γ )e − Γ Γ e − γ Γ e , e i α α δ λ β δ λ λ α βζ βζ αδ αζ αζ βδ αβ δζ α δ α α δ α δ α =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ , α β βζ βζ αδ αζ αζ βδ αβ δζ where ǫ = −1 and ǫ = 1. Let us begin by computing the 00 component: 0 i α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 0 α 0 00 00 αδ α0 α0 0δ α0 δ0 β β 1 1 α α α α =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ . 1 0 00 00 α1 α0 α0 0β α0 β0 In order to simplify the expression, let us note that 1 1 α −1/2 −λ/2 1/2 1/2 e (Γ ) + Γ Γ = t e α ∂ (tα λ ), 1 θ θ 00 00 α1 where we have used (236). Moreover, due to (229), α 1 −3/4 −λ/4 1 −1/2 −λ/2 −1 −e (Γ ) = −e (−γ + t e ) = e (γ ) + t e (λ + 3t ). 0 0 0 t α0 01 01 Compute, using the symmetries of the connection coefficients, β α 0 i i 0 j i 1 0 0 2 −Γ Γ = −Γ Γ − Γ Γ − Γ Γ = −Γ Γ = −(γ ) . α0 0β i0 00 00 0i i0 0j 00 01 01 Finally, β β j α i 0 i i 0 2 1 1 A 1 B A −γ Γ = − γ Γ = γ Γ + γ Γ = (γ ) + γ Γ + γ Γ + γ Γ α0 β0 i0 β0 0i 00 0i j0 01 01 10 01 A0 0A B0 0 2 1 2 A 1 B A =(γ ) − (γ ) + γ Γ + γ Γ 01 01 01 A0 0A B0 1 1 1 1 0 2 1 2 2 2 3 2 2 2 3 2 2 2 2 3 =(γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − γ γ . 01 01 01 01 02 03 03 03 02 2 2 2 2 Since γ = 0, we obtain 1 1 −1/2 −λ/2 1/2 1/2 −1/2 −λ/2 −1 1 Ric(e , e ) = t e α ∂ (tα λ ) + t e (λ + 3t ) + e (γ ) 0 0 θ θ t 0 4 4 1 1 1 1 2 2 2 3 2 2 2 3 2 2 2 − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) . 01 01 02 03 03 2 2 2 In order to simplify this expression, it is useful combine (229) and (233) in order to conclude 1 1 1 1 2 −1/2 −λ/2 1/2 −1/2 −1 −1 −1/2 −λ/2 −1 −1 e (γ )−(γ ) = − t e α ∂ [tα (λ −2α α −t )]+ t e (λ −2α α −t ). 0 01 01 t t t t t 4 4 61 Let us now compute 1 1 1 2 2 3 2 2 2 3 2 2 2 − (γ ) − (γ ) − (γ ) − (γ ) − (γ ) 01 01 02 03 03 2 2 2 λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 2 2P 2 −3/2 −λ/2 = − t e P + e Q + + − t e . t t 7/2 7/2 2 2 t t Thus −1/2 −λ/2 1/2 1/2 −1/2 −1 −1 Ric(e , e ) = t e α ∂ (tα λ ) − ∂ [tα (λ − 2α α − t )] 0 0 θ θ t t t λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2P 2 − t e P + e Q + + t t 7/2 7/2 2 t t 1 α −1/2 −λ/2 + t e λ − . 2 α Let us compute, using (227), (230) and the antisymmetry of Γ in i and j, that αj α β α α β α β α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 1 α 0 01 01 αβ α1 α1 0β α0 β1 β β 0 0 α 0 α α =e (Γ ) + Γ Γ − e (γ ) − Γ Γ − γ Γ 0 0 01 01 α0 01 0β β1 α1 α0 0 α β α β α =Γ Γ − Γ Γ − γ Γ . 01 α0 α1 0β α0 β1 Compute, using (229), 0 α 0 1 −3/4 −λ/4 0 1 0 −3/4 −λ/4 Γ Γ = γ (−γ + t e ) = −γ γ + γ t e . 01 α0 01 01 01 01 01 Moreover, using the symmetries of the connection coefficients, β α 0 1 i j 0 1 −Γ Γ = −Γ Γ − Γ Γ = γ γ . α1 0β 11 00 j1 0i 01 01 Thus 0 α α 0 −3/4 −λ/4 −1/2 −λ/2 1/2 Γ Γ − Γ Γ = γ t e = t e α λ . 01 α0 α1 0β 01 Finally, β β j α i 0 1 i 2 2 2 3 3 2 3 3 −γ Γ =γ Γ = γ Γ + γ Γ = γ Γ + γ Γ + γ Γ + γ Γ α0 β1 β1 01 01 j1 02 21 03 21 02 31 03 31 0i 0i 1 1 2 2 2 2 3 3 1/2 −λ/2 1/2 2P = − γ γ − γ γ − γ γ = − t e α (P P + e Q Q ). t θ t θ 02 12 03 13 03 13 2 2 Thus −1/2 −λ/2 1/2 2P Ric(e , e ) = t e α [λ − 2t(P P + e Q Q )]. 0 1 θ t θ t θ Let us compute, using (230), β β β α α α α α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 0 A α 0 0A αβ αA 0β βA 0A αA α0 j j A 1 α 0 1 1 0 i 0 1 i = − e (γ ) + Γ Γ − Γ Γ − Γ Γ − Γ Γ + γ Γ + γ Γ 01 0A α1 1A 00 0A 01 jA 01 0A 0j 0i iA A A B B B 1 = − e (γ ) + (Γ + Γ )Γ + γ Γ . 01 B1 1A 01 01 BA Let us begin by considering the case A = 2. Compute 1 1 1 2 −5/4 −P/2−λ/4 1/2 2 2 − e (γ ) = − t e α J − γ γ , 1 θ 01 12 01 2 2 2 1 1 2 B B 2 2 2 3 (Γ + Γ )Γ = − γ γ − γ γ , B1 12 01 12 01 13 01 2 2 B 1 2 2 3 2 γ Γ = γ γ + γ γ . 01 B2 01 12 01 13 62 Thus −5/4 −P/2−λ/4 1/2 Ric(e , e ) = − t e α J . 0 2 θ Compute 1 1 1 1 3 −5/4 P/2−λ/4 1/2 2 3 2 2 − e (γ ) = − t e α (K − QJ ) + γ γ − γ γ , 1 θ θ 01 12 01 13 01 2 2 2 2 3 B B 3 3 (Γ + Γ )Γ = − γ γ , B1 13 01 13 01 B 1 2 2 3 3 γ Γ = γ γ + γ γ . 01 B3 01 13 01 13 2 3 Keeping in mind that γ = −γ , we conclude that 12 13 −5/4 P/2−λ/4 1/2 Ric(e , e ) = − t e α (K − QJ ). 0 3 θ θ Compute, using the symmetries of the connection coefficients, α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 1 1 α 1 11 11 αδ α1 α1 1δ α1 δ1 0 0 α 0 A 0 0 i i 0 0 1 0 =e (Γ ) + Γ Γ − e (γ ) − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ 0 1 11 11 α0 01 01 1A i1 10 i1 1j 01 01 01 11 A 0 B A − γ Γ − γ Γ 01 A1 A1 B1 1 1 0 0 α 0 A A A A 1 0 0 2 =e (Γ ) + Γ Γ − e (γ ) + γ γ − γ γ + γ Γ − (γ ) 0 1 11 11 α0 01 01 01 01 01 01 11 01 4 4 1 0 A 0 B A − γ Γ − γ Γ − γ Γ 01 11 01 A1 A1 B1 0 0 α 0 0 2 A 0 B A =e (Γ ) + Γ Γ − e (γ ) − (γ ) − γ Γ − γ Γ . 0 1 11 11 α0 01 01 01 A1 A1 B1 Due to (233) and (236), we know that the sum of the first four terms is given by −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) . t t t θ θ It can also be computed that A 0 −3 −P 2 P 2 −γ Γ = t [e J + e (K − QJ) ], 01 A1 B A 1/2 −λ/2 2 2P 2 −γ Γ = − t e α(P + e Q ). A1 B1 θ θ Thus −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 Ric(e , e ) = t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) 1 1 t t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 1/2 −λ/2 2 2P 2 + t e + − t e α(P + e Q ). θ θ 7/2 7/2 2 t t 2 Let us turn to α δ α α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − e (Γ ) − Γ Γ − γ Γ 1 A α 1 1A 1A αδ αA αA 1δ α1 δA 0 0 α 1 α 0 i i 0 i j 0 i =e (Γ ) + Γ Γ + Γ Γ − Γ Γ − Γ Γ − Γ Γ − γ Γ 1A 1A α0 1A α1 iA 10 0A 1i jA 1i i1 0A i 0 i j − γ Γ − γ Γ 01 iA j1 iA A 1 B B 1 1 1 B 1 1 0 B 0 = − e (γ ) + Γ Γ − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 01 A0 B0 A0 B0 0A 10 0A B0 01 1A 01 BA 63 Compute that 1 1 1 2 −5/4 −λ/4−P/2 3 2 2 2 − e (γ ) = − t e J − γ γ − γ γ , 0 t 01 03 01 02 01 2 2 2 1 1 1 0 −Γ Γ − γ Γ = 0, 02 10 01 12 1 1 B 1 B 1 2 2 2 3 −Γ Γ − Γ Γ = − γ γ − γ γ , 20 B0 02 B0 02 01 03 01 2 2 3 1 1 1 B B 0 2 2 2 3 3 2 Γ Γ − γ Γ = γ γ + γ γ + γ γ . 20 B0 01 B2 01 02 01 03 01 03 2 2 2 Adding up, we obtain −5/4 −λ/4−P/2 Ric(e , e ) = − t e J . 1 2 t Compute that 1 1 1 1 3 −5/4 P/2−λ/4 2 2 2 3 3 3 − e (γ ) = − t e (K − QJ ) − γ γ − γ γ − γ γ , 0 t t 01 03 01 02 01 03 01 2 2 2 2 1 1 1 0 −Γ Γ − γ Γ = 0, 03 10 01 13 B 1 B 1 3 3 −Γ Γ − Γ Γ = − γ γ , 30 B0 03 B0 03 01 3 1 1 1 B B 0 3 3 3 2 2 2 Γ Γ − γ Γ = γ γ + γ γ + γ γ . 30 B0 01 B3 01 03 01 02 01 03 2 2 2 Adding up, we obtain −5/4 P/2−λ/4 Ric(e , e ) = − t e (K − QJ ). 1 3 t t Compute α δ α δ α δ α Ric(e , e ) =e (Γ ) + Γ Γ − Γ Γ − γ Γ A B α AB AB αδ αB Aδ αA δB 0 1 0 α 1 α δ α δ α =e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 0 1 AB AB AB α0 AB α1 αB Aδ αA δB Note that if A = B = 2, the first four terms can be written −1/2 −λ/2 1/2 −1/2 −1 1/2 t e α ∂ [tα (t + P )] − ∂ (tα P ) ; t t θ θ cf. (234) and (237). Let us therefore turn to δ α δ α 0 i i 0 i j 2 0 2 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α2 2δ α2 δ2 i2 20 02 2i j2 2i 02 22 12 22 i i i 1 A A 1 2 0 2 1 = − Γ (Γ + Γ ) − Γ Γ − Γ Γ − γ Γ − γ Γ . 20 20 02 A2 21 12 2A 02 22 12 22 However, it can be computed that 1 1 i i i −3 −P 2 1/2 −λ/2 2P 2 2 2 −Γ (Γ + Γ ) = − t e J − t e e Q − (γ ) , 20 20 02 t 02 2 2 1 A A 1 1/2 −λ/2 2P 2 2 2 −Γ Γ − Γ Γ = t e αe Q + (γ ) , A2 21 12 2A θ 12 2 0 2 1 2 2 2 2 −γ Γ − γ Γ = (γ ) − (γ ) . 02 22 12 22 02 12 Adding up, λ/2−P 2 1 1 e J −1/2 −λ/2 1/2 −1/2 −1 1/2 1/2 −λ/2 Ric(e , e ) = t e α ∂ [tα (t + P )] − ∂ (tα P ) − t e 2 2 t t θ θ 7/2 2 2 t 1/2 −λ/2 2P 2 2 − t e e (Q − αQ ). t θ 64 Next, consider 0 1 0 α 1 α δ α δ α Ric(e , e ) =e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 2 3 0 1 23 23 23 α0 23 α1 α3 2δ α2 δ3 Due to (239) and (240), the sum of the first four terms can be written h i 1 1 1 −1/2 −λ/2−P 1/2 −1/2 2P 1/2 2P 3 2 2 3 2 2 t e α ∂ (tα e Q ) − ∂ (tα e Q ) − (γ − γ )γ + (γ − γ )γ . t t θ θ 13 12 13 03 02 03 2 2 2 Compute δ α δ α 0 i i 0 i j 2 0 2 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α3 2δ α2 δ3 i3 20 03 2i j3 2i 02 23 12 23 i i i 1 A A 1 2 0 2 1 = − (Γ + Γ )Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 30 03 20 A3 21 13 2A 02 23 12 23 However, 1 1 i i i 3 2 3 2 −(Γ + Γ )Γ = − γ γ − γ γ , 30 03 20 01 01 03 03 2 2 1 A A 1 3 2 −Γ Γ − Γ Γ = γ γ , A3 21 13 2A 13 13 1 1 2 0 2 1 2 2 2 2 −γ Γ − γ Γ = γ γ − γ γ . 02 23 12 23 02 03 12 13 2 2 Adding up, we obtain h i −1/2 −λ/2−P 1/2 −1/2 2P 1/2 2P −3 Ric(e , e ) = t e α ∂ (tα e Q ) − ∂ (tα e Q ) − t J(K − QJ). 2 3 t t θ θ 2 2 Finally, let us consider 0 1 0 α 1 α δ α δ α Ric(e , e ) = e (Γ ) + e (Γ ) + Γ Γ + Γ Γ − Γ Γ − γ Γ . 3 3 0 1 33 33 33 α0 33 α1 α3 3δ α3 δ3 Due to (235) and (238), the sum of the first four terms is −1/2 −λ/2 1/2 −1/2 −1 1/2 t e α ∂ [tα (t − P )] + ∂ (tα P ) . t t θ θ Let us therefore compute δ α δ α 0 i i 0 i A 0 A 1 −Γ Γ − γ Γ = − Γ Γ − Γ Γ − Γ Γ − γ Γ − γ Γ α3 3δ α3 δ3 i3 30 03 3i j3 03 A3 13 A3 3i i i i 1 A A 1 A 0 A 1 = − (Γ + Γ )Γ − Γ Γ − Γ Γ − γ Γ − γ Γ . 30 03 30 A3 31 13 3A 03 A3 13 A3 On the other hand i i i 3 2 3 2 −(Γ + Γ )Γ = − (γ ) − (γ ) , 30 03 30 01 03 1 A A 1 3 2 −Γ Γ − Γ Γ = (γ ) , A3 31 13 3A 13 1 1 A 0 A 1 2 2 3 2 2 2 3 2 −γ Γ − γ Γ = (γ ) + (γ ) − (γ ) − (γ ) . 03 A3 13 A3 03 03 13 13 2 2 Adding up, we obtain 1 1 δ α δ α −3 P 2 1/2 −λ/2 2P 2 2 −Γ Γ − γ Γ = − t e (K − QJ) + t e e (Q − αQ ). α3 3δ α3 δ3 t θ 2 2 Thus −1/2 −λ/2 1/2 −1/2 −1 1/2 Ric(e , e ) = t e α ∂ [tα (t − P )] + ∂ (tα P ) 3 3 t t θ θ λ/2+P 2 1 e (K − QJ) 1 1/2 −λ/2 1/2 −λ/2 2P 2 2 − t e + t e e (Q − αQ ). t θ 7/2 2 t 2 65 A.6 The Einstein tensor Adding up the above, we conclude that the scalar curvature S is given by −1/2 −λ/2 1/2 −1/2 −1 −1 1/2 S = t e α ∂ [tα (λ − 2α α − t )] − ∂ (tα λ ) t t t θ θ 1/2 −λ/2 2 2P 2 2 2P 2 + t e P + e Q − α(P + e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 + t e + − t e λ . 7/2 7/2 2 t t 2 As a consequence, if Ein = Ein(e , e ), then αβ α β λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2 2P 2 2 Ein = − t e P + αP + e (Q + αQ ) + + t θ t θ 7/2 7/2 4 t t 1 α −1/2 −λ/2 + t e λ − 2 , 4 α λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 2 2 2P 2 2 Ein = − t e P + αP + e (Q + αQ ) − − t θ t θ 7/2 7/2 4 t t −1/2 −λ/2 + t e λ , 1 1 α 3 1 −1/2 −λ/2 1/2 −1/2 1/2 Ein = t e α ∂ tα P − λ + + − ∂ tα P − λ 22 t t t θ θ θ 2 2 α 2t 2 1/2 −λ/2 2 2P 2 2 2P 2 − t e P + 3e Q − α(P + 3e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e 3 + + t e λ . 7/2 7/2 4 t t 4 The last diagonal component is given by 1 1 α 3 1 −1/2 −λ/2 1/2 −1/2 1/2 Ein = − t e α ∂ tα P + λ − − − ∂ tα P + λ 33 t t t θ θ θ 2 2 α 2t 2 1/2 −λ/2 2 2P 2 2 2P 2 − t e P − e Q − α(P − e Q ) t t θ θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e + 3 + t e λ . 7/2 7/2 4 t t 4 It is of interest to note that −1/2 −λ/2 1/2 −1/2 1/2 1/2 −λ/2 2P 2 2 Ein − Ein =t e α ∂ (tα P ) − ∂ (tα P ) − t e e (Q − αQ ) 22 33 t t θ θ t θ λ/2−P 2 λ/2+P 2 1 e J e (K − QJ) 1/2 −λ/2 − t e − . 7/2 7/2 2 t t We also have 1 α 3 −1/2 −λ/2 1/2 −1/2 1/2 Ein + Ein = − t e α ∂ tα λ − 2 − − ∂ tα λ 22 33 t t θ θ 2 α t 1/2 −λ/2 2 2P 2 2 2P 2 − t e P + e Q − α(P + e Q ) t t θ θ λ/2−P 2 λ/2+P 2 e J e (K − QJ) 1 1/2 −λ/2 −1/2 −λ/2 − t e + + t e λ . 7/2 7/2 t t 2 The remaining components of the Einstein tensor equal the corresponding components of the Ricci tensor, and have consequently already been computed. Using the above calculations, the expressions (59)–(68) for Einstein’s equations, Ein + Λg = T , follow. 66 A.7 The Vlasov equation The distribution function f characterising the Vlasov matter is defined on the mass shell. The mass shell, in its turn, is given by the future directed unit timelike vectors. Since a tangent vector in this set can be written v e , where 0 1 2 2 2 3 2 1/2 v = [1 + (v ) + (v ) + (v ) ] , we can think of f as depending on v , i = 1, 2, 3, and the base point. However, due to the symmetry requirements, the distribution function only depends on the tθ-coordinates of the base point. As a consequence, the distribution function can be considered to be a function of (t, θ, v), 1 2 3 where v = (v , v , v ). In order to derive an equation for f, recall that the Vlasov equation is equivalent to f being constant along future directed unit timelike geodesics. Consider, therefore, a future directed unit timelike geodesic γ(s) = [t(s), θ(s), x(s), y(s)] 2 α in a T -symmetric spacetime. Define the functions v (s) by the equality γ˙ (s) = v (s) e | . γ(s) Note that dt 1/4 −λ/4 0 (s) = t (s)(e ) ◦ γ(s) v (s), (241) ds dθ 1/4 −λ/4 1/2 1 (s) = t (s)(e α ) ◦ γ(s) v (s). (242) ds Let ν(s) = [t(s), θ(s), v(s)], h = f ◦ ν, 1 2 3 where v(s) = [v (s), v (s), v (s)]. The requirement that f be constant along geodesics is equivalent to the requirement that dh/ds = 0 regardless of the choice of future directed unit timelike geodesic γ. On the other hand, dh ∂f dt ∂f dθ ∂f dv = ◦ ν + ◦ ν + ◦ ν . ds ∂t ds ∂θ ds ∂v ds i=1 Keeping (241) and (242) in mind, the requirement that dh/ds = 0 is equivalent to the requirement that 1 i ∂f v ∂f v˙ ∂f 1/2 ◦ ν + α ◦ γ ◦ ν + ◦ ν = 0. 0 1/4 −λ◦γ/4 0 i ∂t v ∂θ t e v ∂v i=1 In order to derive an expression for v˙, note that α α β α β µ α β µ α 0 = γ¨ = (v e ) = v˙ e + v ∇ e = v˙ e + v v ∇ e = (v˙ + v v Γ )e . α α γ˙ β α e β α β µ ds The geodesic equation can thus be written α β µ α v˙ = −v v Γ . βµ Using this formula, it can be calculated that 1 0 0 0 1 0 1 2 0 2 3 0 3 2 2 2 3 3 3 2 2 3 v˙ = −γ v v + γ v v + γ v v + γ v v − γ v v − γ v v − γ v v . 01 01 01 01 12 13 13 Using the formulae for γ , we conclude that λµ 1 2 3 v˙ 1 1 α 1 v v 1/2 0 1 1/2 P = − α λ v − λ − 2 − v + α e Q θ t θ 1/4 −λ/4 0 0 t e v 4 4 α t v 3 2 2 2 1 (v ) − (v ) 1/2 −7/4 λ/4 −P/2 2 P/2 3 − α P + t e e Jv + e (K − QJ)v , 2 v 67 where we have omitted composition with γ for the sake of brevity. We also have 2 2 0 2 2 1 2 v˙ = γ v v + γ v v , 02 12 so that 2 1 2 v˙ 1 1 1 v v 2 1/2 = − P + v − α P . t θ 1/4 −λ/4 0 0 t e v 2 t 2 v Finally, 3 2 0 2 3 0 3 2 1 2 3 1 3 v˙ = γ v v + γ v v + γ v v + γ v v , 03 03 13 13 so that 3 1 3 1 v˙ 1 1 1 v v v 3 1/2 P 2 1/2 = − − P v + α P − e v Q + α Q . t θ t θ 0 0 1/4 −λ/4 0 2 t 2 v v t e v Adding up the above computations, we conclude that the Vlasov equation is equivalent to the requirement that (75) holds. B Notation Metric variables, twist quantities, cosmological constant, frame, manifold. 1. α > 0, λ, P , Q, G and H. These are the functions characterising the metric; cf. (1). 2. λ is defined in (84). 3. J and K. These are the twist quantities defined in (2). They also satisfy (58). 1/2 4. Λ and H. Λ is the positive cosmological constant and H = (Λ/3) . 5. {e } is the orthonormal frame defined in (7). 6. t , t . In the situations of interest in this paper, the metric (1) is defined on (t , ∞) × T , 0 1 0 where t ≥ 0. When speaking of a T -symmetric solution, we take it for granted that t is 0 0 defined in this way. Moreover, t = t + 2. 1 0 Variables for the characteristic system. 1 2 3 1. Θ, V , V , V are the basic variables of the characteristic system (108)–(111). The symbols Θ(s; t, θ, v), V (s; t, θ, v) denote a solution to the characteristic system corresponding to initial data (t, θ, v). In other words, Θ(s; t, θ, v), V (s; t, θ, v), considered as functions of s, are solutions to (108)–(111). Moreover, Θ(t; t, θ, v) = θ, V (t; t, θ, v) = v. 2. Ψ and Z , i = 1, 2, 3. Given a choice of derivative (∂ , ∂ or ∂ i), say ∂, the variables Ψ and t θ 1 2 3 Z = (Z , Z , Z ) are defined by (138)–(141). ˆ ˆ 3. Ψ and Z are the rescaled versions of Ψ and Z, and they are defined in (157). i i 4. Ψ , Z , V and Θ are the higher order derivatives of Ψ, Z, V and Θ. They are defined in j j j j (184). 1 1 ˆ ˆ 5. Z and Ψ are the rescaled versions of Z and Ψ , and they are defined in (199). N N N N Notation, matter quantities. 1. ρ, J , P and S . The quantities ρ, J , P and S are defined in general in (57). In the case i i ij i i ij of Vlasov matter, they are defined in (77). 68 Notation, Vlasov matter. 1. P denotes the mass shell (the set of future directed unit timelike vectors). 2. f denotes the distribution function. For T -symmetric solutions, f, however, denotes the symmetry reduced version of the distribution function. In other words, f is considered to 1 2 3 be a function of t, θ, v , v and v , where, if p is an element of the mass shell, (t, θ, x, y) is the base point of p, and v are the components of p relative to the orthonormal frame {e }. −1/2 3. f is the rescaled distribution function. It is given by f (t, θ, v) = f(t, θ, t v). sc sc Vl 4. T is the stress energy tensor associated with the Vlasov matter. It is given by (4). The αβ components of the stress energy tensor with respect to the frame {e } are given by (57) and (77); see also (76). i i 5. L . The functions L are given by (204)–(206). The initial value formulation. 1. pr is the projection defined in Remark 26. Vl Vl 2. ρ and J are defined in (33) and (34). Auxiliary notation. 1. hhi. If h is a scalar function, its mean value is denoted hhi; cf. (3). 2 1/2 2. hp¯i. If p¯ is a vector, hp¯i = (1 + |p¯| ) . Energies controlling the metric variables, the distribution function and solutions to the characteristic system. 1. The H -norm is defined in (22). Vl,µ 2. ∂ and A are defined in (71). They are given by ± ± 1/2 2 2P 2 ∂ = ∂ ± α ∂ , A = (∂ P ) + e (∂ Q) . ± t θ ± ± ± 3. E is the L -based energy introduced in (85). bas 1 1 4. Q controls the size of the support of f in the v -direction; cf. (114). It is given by 1 1 1 2 3 Q (t) := sup{|v | : (t, θ, v , v , v ) ∈ suppf}. 5. F . The sup-norm energy F is introduced in (115). It is given by F (t) = sup A (t, θ) + sup A (t, θ). + − 1 1 θ∈S θ∈S ˆ ˆ 6. A and F are introduced in (121) and (122) respectively. They are given by ˆ ˆ ˆ ˆ A = t A + t, F (t) = sup A (t, θ) + sup A (t, θ). ± ± + − 1 1 θ∈S θ∈S 1 1 7. R and Q are defined in (124). They are given by 1 1 2 1/2 1 1 2 1/2 R (s) = [s(V (s)) + 1] , Q (s) = [s(Q (s)) + 1] . 69 8. G is defined in (126). 9. E is introduced in (158). It is given by i 2 2 ˆ ˆ ˆ E = (Z ) + (Ψ) . i=1 10. A is introduced in (175). It is given by N+1,± N N 1/2 2 2P N N 1/2 2 A = [∂ P ± ∂ (α P )] + e [∂ Q ± ∂ (α Q )] . N+1,± t θ t θ θ θ θ θ ˆ ˆ 11. A and F are introduced in (176). They are given by N+1,± N+1 7/2 1/2 ˆ ˆ ˆ ˆ A = t A + t , F = sup A + sup A . N+1,± N+1,± N+1 N+1,+ N+1,− 1 1 θ∈S θ∈S 12. E is defined in (200). It is given by 2 1 2 ˆ ˆ ˆ E = (Ψ ) + (Z ) . N N 13. E and E. These energies control suitably weighted Sobolev norms of the distribution function. E is defined in (207). Moreover, E = E . k 0 References [1] Anderson, M. T.: Existence and Stability of even-dimensional asymptotically de Sitter spaces. Ann. Henri Poincar´e 6, 801–820 (2005) [2] Andr´easson, H.: Global Foliations of Matter Spacetimes with Gowdy Symmetry. Commun. Math. Phys. 206, 337-366 (1999) [3] Andr´easson, H.; Rendall, A.D.; Weaver, M.: Existence of CMC and constant areal time folia- tions in T symmetric spacetimes with Vlasov matter. Comm. Partial Differential Equations 29, 237-262 (2004) [4] Andr´easson, H.: The Einstein-Vlasov System/Kinetic Theory, Living Rev. Relativity 14, 4. URL (cited on 2013-04-23): http://www.livingreviews.org/lrr-2011-4 (2011) [5] Berger, B. K.; Chru´sciel, P. T.; Isenberg, J.; Moncrief, V.: Global foliations of vacuum space- times with T isometry. Ann. Physics 260, no. 1, 117148 (1997) [6] Calogero, S., Heinzle, J. M.: Oscillations toward the singularity of LRS Bianchi type IX cosmological models with Vlasov matter. SIAM J. Appl. Dyn. Syst. 9, 1244–1262 (2010) [7] Chru´sciel, P. T. On space-times with U(1) × U(1) symmetric compact Cauchy surfaces. Ann. Physics 202, no. 1, 100150 (1990) [8] Clausen, A.; Isenberg, J.: Areal foliation and asymptotically velocity-term dominated behavior in T symmetric space-times with positive cosmological constant. J. Math. Phys. 48, no. 8, 082501 (2007) [9] Ehlers, J.: Survey of general relativity theory, in Israel, W., ed., Relativity, Astrophysics, and Cosmology, Proceedings of the summer school held 14–26 August 1972 at the Banff Centre, Banff, Alberta, Atrophysics and Space Science Library, vol. 38, pp. 1–125, (Reidel, Dordrecht, Netherlands; Boston, U.S.A., 1973). 70 [10] Friedrich, H.: On the existence of n-geodesically complete or future complete solutions of Einstein’s field equations with smooth asymptotic structure. Commun. Math. Phys. 107, 587– 609 (1986) [11] Friedrich, H.: On the global existence and the asymptotic behavior of solutions to the Einstein–Maxwell–Yang–Mills equations. J. Differential Geom. 34, no. 2, 275–345 (1991) [12] Isenberg, J.; Luo, X.: Power Law Inflation with Electromagnetism. Ann. Physics 334, 420-454 (2013) [13] Lee, H.: Asymptotic behaviour of the Einstein–Vlasov system with a positive cosmological constant. Math. Proc. Camb. Phil. Soc. 137, 495–509 (2004) [14] Lee, H.: The Einstein–Vlasov system with a scalar field. Ann. Henri Poincar´e 6, 697–723 (2005) [15] LeFloch, P. G.; Tchapnda, S. B: Plane-symmetric spacetimes with positive cosmological con- stant. The case of stiff fluids. Adv. Theor. Math. Phys. 15, no. 4, 11151140 (2011) [16] Lu¨bbe, C.; Valiente Kroon, J. A.: A conformal approach for the analysis of the non-linear stability of radiation cosmologies. Ann. Physics 328, 1–25 (2013) [17] Moncrief, V.: Global properties of Gowdy spacetimes with T × R topology. Ann. Physics 132, no. 1, 87-107 (1981) [18] Perlmutter, S. et. al.: Measurements of Ω and Λ from 42 high-redshift supernovae. Astrophys. J. 517, 565–586 (1999) [19] O’Neill, B.: Semi Riemannian Geometry. Academic Press, Orlando (1983) [20] Petersen, P.: Riemannian geometry, Springer Verlag, New York (2006) [21] Rein, G.: On future geodesic completeness for the Einstein-Vlasov system with hyperbolic symmetry. Math. Proc. Cambridge Philos. Soc. 137, 237–244 (2004) [22] Rendall, A. D.: Crushing singularities in spacetimes with spherical, plane and hyperbolic symmetry. Class. Quantum Grav. 12 1517 (1995) [23] Rendall, A.D.: An introduction to the Einstein–Vlasov system, in Chru´sciel, P.T., ed. Math- ematics of Gravitation, Part I: Lorentzian Geometry and Einstein Equations, Proceedings of the Workshop on Mathematical Aspects of Theories of Gravitation, held in Warsaw, February 29 - March 30, 1996, Banach Center Publications, vol. 41, pp. 35–68, (Polish Academy of Sciences, Institute of Mathematics, Warsaw, Poland, 1997). [24] Rendall, A. D.: Accelerated cosmological expansion due to a scalar field whose potential has a positive lower bound. Class. Quant. Grav. 21, 2445–2454 (2004) [25] Rendall, A. D.: Intermediate inflation and the slow-roll approximation. Class. Quant. Grav. 22, 1655–1666 (2005) [26] Rendall, A. D.: Dynamics of k-essence. Class. Quant. Grav. 23, 1557–1570 (2006) [27] Riess, A. G. et al.: Observational Evidence from Supernovae for an Accelerating Universe and a Cosmological Constant. Astron. J. 116, 1009–1038 (1998) [28] Ringstr¨om, H.: Future stability of the Einstein non-linear scalar field system. Invent. math. 173, 123–208 (2008) [29] Ringstr¨om, H.: Power law inflation. Commun. Math. Phys. 290, 155–218 (2009) 71 [30] Ringstr¨om, H.: The Cauchy problem in General Relativity, European Mathematical Society, Zu¨rich (2009) [31] Ringstr¨om, H.: On the Topology and Future Stability of the Universe, Oxford University Press, Oxford (2013) [32] Rodnianski, I., Speck, J.: The nonlinear future stability of the FLRW family of solutions to the irrotational Euler–Einstein system with a positive cosmological constant. J. Eur. Math. Soc. (JEMS) 15, no. 6, 2369–2462 (2013) [33] Smulevici, J.: On the area of the symmetry orbits of cosmological spacetimes with toroidal or hyperbolic symmetry. Anal. PDE 4, no. 2, 191–245 (2011) [34] Speck, J.: The nonlinear future stability of the FLRW family of solutions to the Euler-Einstein system with a positive cosmological constant. Selecta Math. (N.S.) 18, no. 3, 633–715 (2012) [35] Speck, J.: The stabilizing effect of spacetime expansion on relativistic fluids with sharp results for the radiation equation of state. Arch. Ration. Mech. Anal. 210, no. 2, 535–579 (2013) [36] Svedberg, C.: Future Stability of the Einstein–Maxwell–Scalar Field System. Ann. Henri Poincar´e 12, No. 5, 849–917 (2011) [37] Tchapnda, S.B.; Rendall, A. D.: Global existence and asymptotic behaviour in the future for the Einstein-Vlasov system with positive cosmological constant. Classical Quantum Gravity 20, 3037–3049 (2003) [38] Tchapnda, S. T., Noutchegueme, N.: The surface–symmetric Einstein–Vlasov system with cosmological constant. Math. Proc. Cambridge Phil. Soc. 138, 541–553 (2005) [39] Tchapnda, S. B.: The plane symmetric Einstein-dust system with positive cosmological con- stant. J. Hyperbolic Differ. Equ. 5, no. 3, 681692 (2008) [40] Wald, R.: Asymptotic behaviour of homogeneous cosmological models in the presence of a positive cosmological constant. Phys. Rev. D 28, 2118–2120 (1983)

Journal

MathematicsarXiv (Cornell University)

Published: Jun 26, 2013

There are no references for this article.