12 views

Uploaded by forizsl

Gravitational waves

- Gravitational Waves: An Introduction
- Gravitational Waves in Matter
- Detecting Fractal Dimensions via Primordial Gravitational Wave Astronomy. Fractal Cosmology.
- Chris Van Den Broeck- A ‘warp drive’ with more reasonable total energy requirements
- 0401117.pdf
- Artificial generation of accustic and gravitational waves in the atmosphere
- gravitational wave
- On the Vacuum Propagation of Gravitational Waves (WWW.OLOSCIENCE.COM)
- Gravitational Waves in Relativistic Theory of Gravitation
- Communication System Using Gravitational Waves - US6300614B1
- Gravitational Wave Detection
- The Schwarzschild solution and its implications for gravitational waves
- Gravitational Waves
- PhysRevD.27.2885
- definision physic stpm sem1
- Quantum State of the Universe
- Sung-Won Kim- Scattering problem of scalar wave in wormhole geometry
- Choral Speaking Script 2016
- Surfer Grids
- Gravity Power generation

You are on page 1of 90

JONATHAN LUK AND IGOR RODNIANSKI Abstract. In this paper, we initiate the rigorous mathematical study of the problem of impulsive gravitational spacetime waves. We construct such spacetimes as solutions to the characteristic initial value problem of the Einstein vacuum equations with a data curvature delta singularity. We show that in the resulting spacetime, the delta singularity propagates along a characteristic hypersurface, while away from that hypersurface the spacetime remains smooth. Unlike the known explicit examples of impulsive gravitational spacetimes, this work in particular provides the rst construction of an impulsive gravitational wave of compact extent and does not require any symmetry assumptions. The arguments in the present paper also extend to the problem of existence and uniqueness of solutions to a larger class of non-regular characteristic data.

Contents 1. Introduction 1.1. First Version of the Theorem 1.2. Strategy of the Proof 1.3. Outline of the Paper 2. Setting and Equations 2.1. Double Null Foliation 2.2. The Coordinate System 2.3. The Equations 2.4. Integration and Norms 2.5. Precise Statement of the Main Theorem 3. The Initial Data 3.1. Approximation Procedure 4. A Priori Estimates 4.1. Estimates for Metric Components 4.2. Estimates for Transport Equations 4.3. Sobolev Embedding 4.4. Commutation Formulae 4.5. General Elliptic Estimates for Hodge Systems 4.6. Lp (S ) Estimates for the Ricci Coecients 4.7. L2 (S ) Estimates for Curvature 4.8. Elliptic Estimates for Ricci Coecients 4.9. Energy Estimates for Curvature 4.10. Propagation of Regularity 4.11. End of Proof of Theorem 4 4.12. Additional Estimates for an Impulsive Gravitational Wave 5. Convergence 5.1. Norms

1

2 5 7 11 11 12 12 13 15 16 17 19 25 27 29 30 30 31 32 35 36 43 49 49 49 54 57

5.2. Estimates for the Dierence of the Metrics 5.3. Transport Equations for the Dierence Quantities 5.4. Estimates for the Ricci Coecient Dierence 5.5. Estimates for the Curvature Dierence 5.6. Additional Estimates on the Dierence of the Metrics 5.7. The Limiting Spacetime Metric 5.8. Uniqueness 5.9. Propagation of Regularity 5.10. The Limiting Spacetime for an Impulsive Gravitational Wave References

58 64 68 74 77 80 84 85 86 89

1. Introduction In this paper, we initiate the study of the characteristic initial value problem for impulsive gravitational spacetimes in general relativity. These were considered to be solutions of the vacuum Einstein equations R = 0 with a delta singularity in the Riemann curvature tensor supported on a null hypersurface. Their historical origin can be traced back to the cylindrical waves of Einstein-Rosen [15], the plane waves of Brinkmann [7] and the explicit impulsive gravitational spacetimes of Penrose [35]. Impulsive gravitational waves have often been studied within the class of plane fronted gravitational waves. This class of explicit solutions to the vacuum Einstein equations has been rst studied by Brinkmann [7] and interest in them has been revived in later decades by the work of Bondi-Pirani-Robinson [6]. They were later classied geometrically by JordonEhlers-Kundt [20]. Among this class are the pp -waves (plane fronted waves with parallel rays) that were discovered by Brinkmann [7], for which the metric takes the form g = 2dudr + H (u, X, Y )du2 + dX 2 + dY 2 , and the Einstein vacuum equations imply that 2H 2H + = 0. (1) X 2 Y 2 These include the special case of sandwich waves, where H is compactly supported in u. Since (1) is linear, it follows that pp -waves enjoy a principle of linear superposition. pp -waves have a plane symmetry and originally impulsive gravitational waves have been thought of as a limiting case of the pp -wave with the function H admitting a delta singularity in the variable u. Precisely, explicit impulsive gravitational spacetimes were discovered and studied by Penrose [35], who gave the metric in the following double null coordinate form: g = 2dudu + (1 u(u))dx2 + (1 + u(u))dy 2 , where is the Heaviside step function. In the Brinkmann coordinate system, the metric has the pp -wave form and an obvious delta singularity: g = 2dudr (u)(X 2 Y 2 )du2 + dX 2 + dY 2 ,

where (u) is the Dirac delta. Despite the presence of the delta singularity for the metric in Brinkmann coordinates, the corresponding spacetime is Lipschitz and it is only the Riemann curvature tensor (specically, the only non-trivial component of it) that has a delta function supported on the plane null hypersurface u = 0. This spacetime turns out to possess remarkable global geometric properties [34]. In particular, it exhibits a strong focusing property such that the whole null cone emanating from a point in the past of u = 0 refocuses to a single line in the future of u = 0 (See Figure 1). In [34], this property forms the basis of Penroses argument that global hyperbolicity fails in this spacetime. Penroses impulsive spacetime is plane symmetric, non-asymptotically at and the delta curvature singularity is supported on the 3-dimensional innite plane {(u, u, x, y ) : u = 0}. It has been long debated whether the strong focussing property and the resulting lack of global hyperbolicity is directly tied to the innite extent of the impulsive gravitational wave (See Yurtsever [43]).

Figure 1. Focusing in Penroses impulsive gravitational spacetime While the interest in impulsive gravitational spacetimes had been high due to the availability of explicit solutions, their global geometric properties, superposition properties and their limiting relation to general spacetimes (see Penrose [35] , Aichelburg-Sexl [1]), the rst study of general spacetimes satisfying the Einstein equations and admitting possible three dimensional delta singularities was undertaken by Taub [42], who derived a system of consistency relations linking the metric, curvature tensor and the geometry of the singular hypersurface. Impulsive gravitational spacetimes also arise as high-frequency limiting spacetimes considered by Choquet-Bruhat [8]. We refer the readers to [16], [17], [3], [5] and the references therein for a more detailed exposition on the physics literature. In this paper, we begin the study of impulsive gravitational spacetimes viewed in the context of an (characteristic) initial value problem. We consider the data, prescribed on an outgoing null hypersurface u = 0 and assume that the component of the curvature has a delta singularity supported on a two dimensional surface S0,us , which can be thought of as the intersection of the hypersurfaces u = 0 and u = us . Observe that in the Penroses explicit impulsive solution, the curvature component has precisely this type of behavior

when restricted to u = 0. Unlike the explicit impulsive spacetimes, which have only been constructed in plane symmetry and thus have innite spatial extent, we will consider the case of S0,us with compact topology, more precisely a sphere. The data on the incoming null hypersurface is prescribed to be smooth but otherwise without any smallness assumptions. With this data, we show that a unique spacetime satisfying the vacuum Einstein equations in an appropriate sense can be constructed locally. Moreover, the delta singularity propagates along a null hypersurface emanating from the initial singularity on S0,us and the spacetime is smooth away from this null hypersurface (See Figure 2).

H us S0,us H0 H0

S0,0

Figure 2. Propagation of Singularity Our main result on the propagation of an impulsive gravitational wave is described by Theorem 1. Suppose the following hold for the initial data set: The data on H 0 is smooth. The data on H0 is smooth except across a two sphere S0,us , where the traceless part of the second fundamental form has a jump discontinuity. Then (a) Given such initial data and suciently small, there exists a unique spacetime (M, g ) endowed with a double null foliation u, u and solves the characteristic initial value problem for the vacuum Einstein equations in the region 0 u u , 0 u u whenever u , u . (b) Dene H us to be the incoming null hypersurface emanating from S0,us . Then the curvature components AB = R(eA , e4 , eB , e4 ) are measures with a singular atom supported on the hypersurface H us . All other components of the curvature tensor can be dened in L . Moreover, the solution is smooth away from H us . For an impulsive gravitational wave, standard local existence shows that the spacetime is smooth in 0 u < us . The problem of constructing an impulsive gravitational wave lies in making sense of the solution in the whole region with us u u and showing that the singularity propagates along the characteristic hypersurface H us and that the spacetime is smooth after the impulse. Remark 1 (Larger class of initial data). The proof introduces a new type of energy estimates for the Einstein vacuum equations which allows some components of the Riemann curvature tensor not to be in L2 . This discovery allows us to consider the problem of local existence and uniqueness for a larger class of non-regular data. This includes data with Riemann curvature tensor that can only be understood as a conormal distribution. Remark 2 (Uniqueness). Penroses construction of explicit impulsive gravitational spacetimes is based on the gluing approach, which has later been used to generate other explicit

solutions of the Einstein equations. In the general case of the characteristic initial value problem for an impulsive gravitational wave, an appropriate adaptation of the gluing philosophy would allow us to construct weak solutions with undetermined uniqueness. In this paper, however, we construct strong solutions in the sense that that uniqueness can also be established. To prove uniqueness, we establish a priori estimates for a larger class of admissible initial data. This allows us to show that the solution is unique among all C 0 limits of smooth solutions to the vacuum Einstein equations. Remark 3 (Propagation of singularity). Our theorem gives a precise description of the propagation of the initial singularity. Such problems have been extensively studied for semilinear equations (see for example [4], [31]). Our result can be formally compared to the works of Majda on the propagation of shocks [29], [30] for systems of conservation laws and the subsequent [2], [32], [33], [14]. In these works, a short time existence, uniqueness and regularity result was established for initial data with a jump discontinuity across a surface and a precise description of the propagation of the singularity was also given. We establish an analogous result of propagation of singularity for a nonlinear system of quasilinear hyperbolic equations. However, contrary to [29], [30] the singularity that is considered in the present paper is not a shock, as it propagates along the characteristics. Moreover, unlike in [29], [30], [2],[32], [33], [14], where the problem is reformulated as an initial-boundary value problem and uniqueness is known only within the class of piecewise smooth solutions, our solution is also unique among limits of smooth solutions. In order to achieve this, the special structure of the Einstein equations in the double null foliation gauge has been heavily exploited. The construction of spacetimes from non-regular characteristic initial data consistent with that of an impulsive gravitational wave had been known only under symmetry assumptions [9], [26], [27]. The work of Christodoulou, who solved the characteristic initial value problem for data with bounded variation for the spherically symmetric Einstein-scalar eld system [9], can be thought of as a rst result in that direction. In particular, in this work, the second derivatives of the scalar eld, which formally is analogous to the curvature, is allowed to have a delta singularity. The study of data with bounded variation turned out to have important consequences in the global structure of spacetimes and the resolution of the cosmic censorship conjecture for the spherically symmetric Einstein-scalar eld system [10], [11]. The construction of distributional solutions for the vacuum Einstein equations that include plane impulsive gravitational wave was carried out in [26], [27]. The present paper is the rst work that provides a consistent study of the initial value problem of impulsive gravitational spacetimes, including their existence, uniqueness and propagation of singularity/regularity. Going beyond spacetimes which represent a single impulsive gravitational wave, colliding impulsive gravitational waves had been studied by Khan-Penrose [21] and Szekeres [41]. In these explicit solutions, the spacetimes possess two null hypersurfaces with curvature delta singularity with a transverse intersection, representing the nonlinear interaction of two impulsive gravitational waves. The study of the characteristic initial value problem for the colliding impulsive gravitational waves will be carried out by the authors in a subsequent paper. 1.1. First Version of the Theorem. Our general approach is based on energy estimates and transport equations in the double null foliation gauge. This general approach in the double null foliation gauge has been carried out in [22], [12] and [24].

The spacetime in question will be foliated by families of outgoing and incoming null hypersurfaces Hu and H u respectively. Their intersection is assumed to be a 2-sphere denoted by Su,u . Dene a null frame {e1 , e2 , e3 , e4 }, where e3 and e4 are null, as indicated in Figure 3, and e1 , e2 are tangent to the two spheres Su,u . e4 is tangent to Hu and e3 is tangent to H u.

e3 e4

H0

H0

S0,0

Figure 3. The Basic Setup and the Null Frame Decompose the Riemann curvature tensor with respect to this frame: AB = R(eA , e4 , eB , e4 ), AB = R(eA , e3 , eB , e3 ), 1 1 A = R(eA , e4 , e3 , e4 ), A = R(eA , e3 , e3 , e4 ), 2 2 1 1 = R(e4 , e3 , e4 , e3 ), = R(e4 , e3 , e4 , e3 ) 4 4 In the context of impulsive gravitational spacetimes, the component can only be understood as a measure. Dene also the following Ricci coecients with respect to the null frame: AB = g (DA e4 , eB ), 1 A = g (D3 eA , e4 ), 2 1 = g (D4 e3 , e4 ), 4 1 A = g (DA e4 , e3 ) 2 AB = g (DA e3 , eB ), 1 A = g (D4 eA , e3 ) 2 1 = g (D3 e4 , e3 ), 4

Dene (resp. ) to be the traceless part of (resp. ). For the problem of the propagation of impulsive gravitational waves, we prescribe initial data on H0 such that has a jump discontinuity across S0,us but smooth otherwise. On H 0 , we prescribe the initial data to be smooth but without any smallness assumptions. As mentioned before, we will prove local existence and uniqueness for a class of data more general than that for the impulsive gravitational wave. More precisely, we require that and its angular derivatives are merely bounded. However, we do not require the derivative of in the e4 direction even to be dened. Theorem 2. Suppose the characteristic initial data are smooth on H 0 . On H0 , the characteristic initial data are determined by and A . Let A be the coordinates on the two sphere

foliating H0 . Suppose, in every coordinate patch |( )i AB |, |( )k ( )i (tr, A )| C. u i 3 k1 i3 Then there exists suciently small such that the unique solution to the vacuum Einstein equations (M, g ) endowed with a double null foliation u, u exists in 0 u , 0 u . Associated to the spacetime a coordinate system (u, u, 1 , 2 ) exists, relative to which the spacetime is in particular Lipschitz and retains higher regularity in the angular directions. Remark 4. We will also explicitly construct a class of initial data that satisfy the assumptions of Theorem 1 in Section 3. This construction of initial data will indicate an easy modication leading to an even more general class satisfying the assumptions of Theorem 2. This Theorem in particular implies Theorem 1(a). An additional argument, based on the estimates derived in the proof of Theorem 2, will be carried out in Section 5.10 to prove part (b) of Theorem 1. The statement of Theorem 2 will be made precise in Section 2.5. In particular, the smoothness assumptions of Theorem 2 can be weakened and we will dene in what sense the vacuum Einstein equations are satised. Remark 5 (Comparison to local existence results). Without symmetry assumptions, all known proofs of existence of spacetimes satisfying the Einstein equations are based on L2 type estimates for the curvature tensor and its derivatives or the metric components and their derivatives. Even with the recent resolution of the L2 curvature conjecture by KlainermanRodnianski-Szeftel [25], [37], [38], [39], [40], the Riemann curvature tensor has to be at least in L2 . (The classical local existence result of Hughes-Kato-Marsden [19] requires the metric 5 to be in H s , with s 2 .) In this paper, we prove local existence and uniqueness under the assumption that the spacetime is merely Lipschitz, which in terms of dierentiability is even one derivative weaker than the L2 curvature conjecture. Of course the Lipschitz assumption refers to the worst possible behavior observed in our data and our result heavily relies on the structure of the Einstein equations which allows us to eciently exploit the better behavior of the other components. 1.2. Strategy of the Proof. For an impulsive gravitational wave, the curvature tensor can only be dened as a measure and is not in L2 . This is one of the main challenges of this work. Let denote the curvature components and denote the Ricci coecients. The key new observation of this paper is that the L2 -type energy estimates for the components of the Riemann curvature tensor

u u

+

Hu Hu

H0

+

H0

+

0 0 Su

,u

du du

coupled together with the null transport equations for the Ricci coecients 3 = + , 4 = +

2

can be renormalized and closed avoiding the L -non-integrable components of curvature. 1.2.1. Renormalized Energy Estimates. The diculty in carrying out the above argument is that in our setting, not all curvature components are dened in L2 . Even at the level of initial data, is only dened as a measure. Therefore, we need to prove L2 energy estimates without involving . To this end, we need to introduce and estimate the renormalized curvature components. The idea of renormalizing the curvature components has been introduced in

[23]. Unlike this work, there the renormalization was used to obtain estimates for the Ricci coecients. The classical way to derive energy estimates for the Einstein equations is via the BelRobinson tensor. In view of our renormalization, we avoid the use of the Bel-Robinson tensor and instead prove the energy estimates directly from the Bianchi identities (8). We note that the derivation of the energy estimates directly from the Bianchi equations without using the Bel Robinson tensor has also appeared in the work of Holzegel [18] in a dierent setting. The challenge and motivation of this method is the derivation of estimates not involving the singular component of curvature . To illustrate this, we rst prove the energy estimates for on Hu and for (, ) on H u by considering the following set of Bianchi equations: 1 4 = div + ... 2 1 + ... 4 = div + 2 3 = + + ... However, the curvature component still appears in the nonlinear terms in these equations. In order to deal with this problem, we renormalize and . Dene 1 , = 2 Using the equation 4 = + ..., we notice that the rst two equations become 4 = div + ..., 4 = div + ... 1 =+ . 2

At the same time, the equation for can be re-written in terms of of and : 3 = + + + ... Now we have a set of renormalized Bianchi equations that does not contain . Using these equations, we derive the renormalized energy estimate

u u

2 +

Hu Hu

( , )2

H0

2 +

H0

( , )2 +

0 0 Su

,u

+ du du ,

in which does not appear in the error term. The same philosophy can be applied for the remaining curvature components (, , , ). As a consequence, we obtain a set of L2 curvature estimates which do not explicitly couple to the singular curvature component . We say explicitly that there is still a remaining possibility that the Ricci coecients appearing in the nonlinear error terms for the energy estimates may depend on .

1.2.2. Estimates for the Ricci Coecients. In order to close the estimates, it is necessary to show that all the Ricci coecients can be estimated without any knowledge of . The most dangerous component is , which naively would have to be estimated using the transport equation 4 + tr = 2 . We take an alternate route and estimate from the equation 1 1 3 + tr = + 2 tr + , 2 2 or 1 1 1 div = tr ( ) ( tr) . 2 2 2 As we shall see, the loss of information of is accompanied by the loss of information of on H u . This presents yet an additional challenge in estimating the Ricci coecients. 1.2.3. Higher Order Energy Estimates. Another diculty arises considering the fact that in order to close the energy estimates, we need to prove higher derivative estimates for the curvature components. However, the derivatives of some curvature components along e4 are not dened in L2 initially. We will therefore only use angular covariant derivatives as commutators and will prove estimates only for the L2 norms of the angular covariant derivatives of the renormalized curvature components. We will show rstly that this procedure does not introduce terms that cannot be estimated in L2 (in particular will not appear) and secondly that all the energy estimates can be closed only using the estimates of the angular derivatives of the renormalized curvature components. 1.2.4. Existence and Uniqueness. Since we work initial data with very low regularity, the a priori estimates that we have derived along the lines of Sections 1.2.1, 1.2.2 and 1.2.3 do not immediately imply the existence and uniqueness of the solutions. Instead, we need to approximate the data by a sequence of smooth data and show rst that they have a common domain of existence and second that they converge in this domain. The fact that the solutions to the sequence of smooth data have a common domain of existence follows from the a priori estimates that do not involve and its derivatives as outlined above. Since the approximating data are smooth, we can conclude that the approximating solutions are also smooth and exist in a common domain. Once we have proved the uniform a priori estimates, we proceed to prove that the sequence of solutions converges. To this end, we consider the equations for the dierence of the Ricci coecients and curvature components. We identify the spacetimes in this sequence by the value of their coordinate functions and derive equations for the dierence of the Ricci coecients and curvature components. Our a priori estimates heavily relies on the structure of the Einstein equations which allows us to eliminate any dependence of the component. A priori, there is no reason to think that this structure is preserved when considering dierence of these solutions, which is necessary to show that our end result is a strong solution to the Einstein equations. It is a remarkable fact that it turns out that in the equations for the dierence of the Ricci coecients and curvature components, indeed does not appear. Thus the a priori estimates we have proved are sucient to control the dierence of the Ricci coecients and curvature components from the dierence of the initial value. This proves that the sequence of solutions converge. Even though we face the standard challenge

10

of loss of derivative in the quasilinear equations, our estimates are still sucient to show that the sequence converges. The constructed limiting spacetime is not smooth. In particular, second derivative of the metric in the e4 direction is not even dened. We will, however, show that the Ricci curvature tensor is better behaved than a general second derivative of the metric and that the limiting spacetime satises the vacuum Einstein equations R = 0 in the L2 sense. The estimates for the dierence of the Ricci coecients and curvature components imply that the constructed spacetime is the unique solution to the Einstein equations among the class of spacetimes admitting a double null foliation and satisfying strong enough a priori bounds. Moreover, we can prove uniqueness of the constructed solution among all limits of smooth spacetimes, i.e, any spacetime that arises as a C 0 limit of smooth solutions to the vacuum Einstein equations with initial data converging to the given initial data must coincide with the constructed spacetime. It can be observed that the above argument for the a priori estimates, as well as that for showing existence and uniqueness, does not use the fact that initially, is a measure whose singular support is on S0,us . In eect, the argument avoids completely, and can be used for data such that is much rougher. Since the argument used only estimates on the Ricci coecients and the curvature components other than , it can be used to handle initial data satisfying only the assumptions of Theorem 2. 1.2.5. Regularity and Propagation of Singularity. In the setting of Theorem 1, i.e., that of an impulsive gravitational wave, the theorem gives a precise description of the propagation of singularity. The a priori estimates imply that all the curvature components other than are bounded. Here, we are interested in proving two additional statements: rstly, is a measure that indeed has a delta singularity on H us ; secondly, the spacetime is smooth away from H us . To show that is a measure, we approximate the data for by a sequence of smooth data n such that n is of size 2n in a region |u us | 2n . The spacetimes constructed for such data are smooth and therefore allow us to use the previously avoided L2 estimate for the component of curvature. This estimate imply that in the constructed spacetime, in the region |u us | 2n , the L2 u norm of n are uniformly bounded; while in the region n n 2 |u us | 2 , the Lu norm of n is bounded by 2 2 . By Cauchy-Schwarz, we have that the L1 u norms of n are uniformly bounded. This allows us to show that in the limiting spacetime, the curvature component is a measure. To show that the singular part of is a delta function supported on the null hypersurface H us , we notice that = 4 tr 2 . Therefore, it suces to show that has a jump discontinuity across H us and smooth everywhere else. This can be proved using the equation 1 1 3 + tr = + 2 tr + 2 2 and the fact that on the initial hypersurface H0 , has a jump discontinuity across S0,us and smooth everywhere else.

11

In order to show that the spacetime is smooth away from H us , we will estimate the higher regularity of all the curvature components in that region. We rst use the Bianchi equations + 3( + ). 3 + tr = + 2 + + 2 Integrating this equation and using Gronwalls inequality, we obtain for any u = us that sup

u iI

iI

iI

The regularity of is thus inherited from the initial data, which is smooth away from u = us . Once we have estimates for , we consider the equation 1 + ) + ( + 4 ). 3 + tr = + 4 3( 2 Integrating as before, we see that also inherits the regularity from the initial data. Higher derivatives estimates for can be derived analogously by dierentiating this equation. The other components of curvature can be controlled in a similar fashion. Notice that this procedure results in a loss derivatives. In particular, in order to control the N -th derivative of the curvature, one needs 2N derivatives initially. 1.3. Outline of the Paper. Finally, we give the outline of the remainder of the paper. In the next Section, we give a careful introduction of the setting, describing the double null foliation, the coordinate system, the equations and relevant notations. We will state a more precise version of Theorem 2, which we will call Theorem 3. In Section 3, we provide a construction of the initial data set satisfying the conditions in Theorem 1 and exhibit a sequence of smooth data approximating the data with a curvature delta singularity. In Sections 4 and 5, we prove Theorem 3. In Section 4, we prove that for smooth initial data satisfying the assumptions of Theorem 3, a unique spacetime exists in a region depending only on the constants in the assumptions. This in particular implies that the approximating sequence of initial data constructed in Section 3 gives rise to spacetimes with a common and uniform region of existence (identied by a choice of a double null coordinate system). In Section 5, we study the equations for the dierence of two spacetimes. This allows us to show convergence of solutions with converging initial data and conclude the existence part of Theorem 3. In Section 5.7, we examine the regularity of the limiting spacetime and show that it is a solution to the vacuum Einstein equations. In Section 5.8, we conclude the uniqueness part of Theorem 3. Finally, in Section 5.10, we return to the proof of part (b) of Theorem 1, giving a precise description of the propagation of singularity. Acknowledgments: The authors would like to thank Mihalis Dafermos for valuable discussions. 2. Setting and Equations Our setting is the characteristic initial value problem with data given on the two characteristic hypersurfaces H0 and H 0 intersecting at the sphere S0,0 . The spacetime will be a solution to the Einstein equations constructed in a neighborhood of S0,0 bounded by the two hypersurfaces.

12

e3

e4

H0

H0

S0,0

Figure 4. The Basic Setup and the Null Frame 2.1. Double Null Foliation. For a spacetime in a neighborhood of S0,0 , we dene a double null foliation as follows: Let u and u be solutions to the eikonal equation g u u = 0, L = 2g u, g u u = 0, L = 2g u. satisfying the initial conditions u = 0 on H0 and u = 0 on H 0 . Let These are null and geodesic vector elds. Dene 22 = g (L , L ). Dene e3 = L , e4 = L to be the normalized null pair such that g (e3 , e4 ) = 2 and L = 2 L , L = 2 L to be the so-called equivariant vector elds. We will denote the level sets of u as Hu and the level sets of u and H u . By virtue of the eikonal equations, Hu and H u are null hypersurface. Notice that the sets dened by xed values of (u, u) are 2-spheres. We denote such spheres by Su,u . They are intersections of the hypersurfaces Hu and H u . The integral ows of L and L respect the foliation Su,u . 2.2. The Coordinate System. On a spacetime in a neighborhood of S0,0 , we dene a coordinate system (u, u, 1 , 2 ) as follows: On the sphere S0,0 , dene a coordinate system (1 , 2 ) for the sphere such that on each coordinate patch the metric is smooth, bounded and positive denite. Then we dene the coordinates on the initial hypersurfaces by requiring A A = 0 on H 0 , and = 0 on H0 . u u We now dene the coordinate system in the spacetime in a neighborhood of S0,0 by letting u and u to be solutions to the eikonal equations: g u u = 0, and dene 1 , 2 by L / L A = 0, g u u = 0,

13

where L / L denote the restriction of the Lie derivative to T Su,u (See [12]). Relative to the coordinate system, the null pair e3 and e4 can be expressed as e3 = 1 + bA A u , e4 = 1 , u

for some bA such that bA = 0 on H 0 , while the metric g takes the form g = 22 (du du + du du) + AB (dA bA du) (dB bB du). 2.3. The Equations. We will recast the Einstein equations as a system for Ricci coecients and curvature components associated to a null frame e3 , e4 dened above and an orthonormal frame e1 , e2 tangent to the 2-spheres Su,u . Using the indices A, B to denote 1, 2, we dene the Ricci coecients relative to the null fame: AB = g (DA e4 , eB ), AB = g (DA e3 , eB ), 1 1 A = g (D3 eA , e4 ), A = g (D4 eA , e3 ) 2 2 (2) 1 1 = g (D3 e4 , e3 ), = g (D4 e3 , e4 ), 4 4 1 A = g (DA e4 , e3 ) 2 where DA = De(A) . We also introduce the null curvature components, AB = R(eA , e4 , eB , e4 ), AB = R(eA , e3 , eB , e3 ), 1 1 A = R(eA , e4 , e3 , e4 ), A = R(eA , e3 , e3 , e4 ), (3) 2 2 1 1 = R(e4 , e3 , e4 , e3 ), = R(e4 , e3 , e4 , e3 ) 4 4 Here R denotes the Hodge dual of R. We denote by the induced covariant derivative operator on Su,u and by 3 , 4 the projections to Su,u of the covariant derivatives D3 , D4 , see precise denitions in [22]. Observe that, 1 1 = 4 (log ), = 3 (log ), 2 2 (4) A = A + A (log ), A = A + A (log ) Let (1) (2) denote an arbitrary contraction of the tensor product of (1) and (2) with respect to the metric . We also dene ((1) (2) )AB := A B + B A AB ((1) (2) ) for one forms A , A , ((1) (2) )AB := / AB ( 1 )CD AC BD for symmetric two tensors AB , AB . For totally symmetric tensors, the div and curl operators are dened by the formulas (div )A1 ...Ar := B BA1 ...Ar , (curl )A1 ...Ar := / BC B CA1 ...Ar , where / is the volume form associated to the metric . Dene also the trace to be (tr)A1 ...Ar1 := ( 1 )BC BCA1 ...Ar1 .

(1) (2) (1) (2) (1) (2) (1) (2) (1) (2)

14

We separate the trace and traceless part of and . Let and be the traceless parts of and respectively. Then and satisfy the following null structure equations:

(5)

The other Ricci coecients satisfy the following null structure equations:

4 = ( ) 3 = ( ) + 3 4 = 2 + | |2 4 3 3 = 2 + | |2 + 4 1 ( ) ( + ) 4 1 ( ) ( + ) 4 1 | + |2 + 8 1 | + |2 + 8 1 2 1 2 (6)

1 div = tr 2 1 div = tr + 2

1 tr) , 2 1 tr) + 2

(7)

15

with K the Gauss curvature of the surfaces S . The null curvature components satisfy the following null Bianchi equations: 1 + ) + ( + 4 ), 3 + tr = + 4 3( 2 4 + 2tr = div 2 + , 3 + tr = + 2 + + 2 + 3( + ), 1 3 2 , 4 + tr = div + 2 2 1 3 2 , 3 + tr = div + 2 2 (8) 3 1 4 + tr = div + + 2 , 2 2 3 1 3 + tr = div + 2 , 2 2 3( ), 4 + tr = + + 2 + 2 3 + 2tr = div 2 + , 1 4 + tr = + 4 3( ) + ( 4 ) 2 where denotes the Hodge dual on Su,u . In the sequel, we will use capital Latin letters A {1, 2} for indices on the spheres Su,u and Greek letters {1, 2, 3, 4} for indices in the whole spacetime. In the following it will be useful to apply a schematic notation. We will let denote an arbitrary tensoreld, a Ricci coecient and a null curvature component dierent from . We will simply write (or , etc.) to denote an arbitrary contraction. The use of this notation is reserved for the cases when the precise nature of the contraction is not important to the argument. Moreover, when using this schematic notation, we will neglect all constant factors. 2.4. Integration and Norms. Let U be a coordinate patch on S0,0 and pU be a partition of unity in DU such that pU is supported in DU . Given a function , the integration on Su,u is given by the formula:

:=

Su,u U

pU

det d1 d2 .

u u 0 u u 0

:=

Du,u U 0

pU

=2

U 0

pU 2

:=

Hu U 0

16

and :=

Hu U 0

With these denitions of integration, we can dene the norms that we will use. Let be a tensoreld. For 1 p < , dene ||||p Lp (Su,u ) := ||||p Lp (Hu ) := ||||p Lp (H Dene also the L

u)

:=

norm by

1/2 ||||L (Su,u ) := sup < , > (). Su,u

2.5. Precise Statement of the Main Theorem. With the notations introduced in this Section, we give a precise version of the statement of Theorem 2: Theorem 3. Suppose the initial data set for the characteristic initial value problem is given on H0 for 0 u u and on H 0 for 0 u u such that c | det |(

i 3 Su,0

|, | det

S0,u

| C, | C,

i )

Su,0

| + |(

i )

u

S0,u

O0 :=

i 3

u i3

R0 :=

i 2

{,,, }

u {,,,}

u

Then for suciently small depending only on C and c, there exists a spacetime (M, g ) endowed with a double null foliation u, u that solves the characteristic initial value problem to the vacuum Einstein equations in 0 u u , 0 u u for u , u . The metric is continuous and takes the form g = 22 (du du + du du) + AB (dA bA du) (dB bB du). (M, g ) is a C 0 limit of smooth solutions to the vacuum Einstein equations and is the unique spacetime solving the characteristic initial value problem among all C 0 limits of smooth solutions. Moreover, 0 0 4 g, g Cu Cu L (S ), u 2 2 2 0 0 2 Cu L (S ), g, g, g Cu 2 2 u u g, (( 1 )AB ( )AB ) L u Lu L (S ), u u u

17

2 2 4 g, g L u Lu L (S ). u uu 2 In the (u, u, 1 , 2 ) coordinates, the Einstein equations are satised in L u Lu L (S ). Furthermore, the higher angular dierentiability in the data results in higher angular dierentiability. We prove this Theorem in two steps. First, we show that if the initial data are smooth and the bounds for the initial data in the assumption of Theorem 3 hold, then an can be chosen depending only on the bounds in the assumption of Theorem 3 such that a smooth spacetime solving the Einstein equations exists in 0 u and 0 u . This is formulated as Theorem 4, and is proved in Section 4. Then, we show that a sequence of solutions to the Einstein equations with a converging sequence of smooth initial data, satisfying the assumptions of Theorem 3 with uniform constants C and c, converges. The limit spacetime satises the Einstein equations and has the properties stated in Theorem 3. This is formulated as Theorem 6 and is proved in Section 5. We furthermore show that the solution is unique and that the regularity in the angular directions persists in Section 5. 3. The Initial Data In this Section, we construct data satisfying the constraint equations such that has a jump discontinuity across a two sphere. We also construct a sequence of smooth data satisfying the constraint equations approaching the data with discontinuous . We derive precise bounds for the Ricci coecients and the curvature components for the initial data in this approximating sequence. On the initial characteristic hypersurfaces, we x =1 identically on H0 and H 0 . This implies = 0 on H0 , and = 0 on H 0 , and = = , and have to satisfy the equations on H0 and H 0 . (9)

L / L = 2,

L / L = 2, (10) 1 |2 (11) L / L tr = (tr)2 | 2 1 L / L tr = (tr)2 | |2 (12) 2 Here L / denotes the restriction of the Lie derivative to T Su,u . This is a notion intrinsic to the null hypersurfaces. Prescribe AB , A , tr and tr on the two sphere S0,0 . Relative to the coordinate system 1 2 ( , ), we require that on S0,0 , |( )i (AB , A , tr, tr)| C. iI +3 On the initial incoming hypersurface H 0 , we have the coordinate system (u, 1 , 2 ) and on the initial outgoing hypersurface, we have the coordinate system (u, 1 , 2 ) as constructed in

18

Section 2.2. On H 0 , we prescribe smoothly the conformal class of the metric AB satisfying det AB = 1. We require the estimates |( )j ( )i AB | C. u j J +1 iI +3 On S0,0 , since AB and AB are in the same conformal class, AB = 2 AB . (10) implies that, AB = 2AB = 2 AB + trAB . u On the other hand Since ( 1 )AB we can identify 1 AB , AB = 2 2 u and tr = Notice also that 2 . u AB = 2 AB + 2 AB . u u u AB = log det = 0, u u

1 1 AC 1 BD | |2 ) ( ) AB CD = ( 4 u u depends only on and is in L . Thus, (12) can be re-written as 2 1 1 AC 1 BD + ( ) ( ) CD = 0. AB u2 8 u u This ordinary dierential equation can be solved up to u u as long as u for suciently small . The ODE also implies that u , = 0, and thus the solution is regular. On the initial outgoing hypersurface H0 we now prescribe the conformal class of the metric AB satisfying det AB = 1 such that its u derivative has a jump discontinuity. To this end we dene smooth matrices (( )1 )AB and (( )2 )AB , such that (( )1 )AB is positive denite with determinant equals to 1 and both (( )1 )AB and (( )2 )AB satisfy |(

kK +1 iI +3

k i ) ( ) (( )j )AB | C. u

Fix us 2 , where > 0 is a small parameter depending on the constants C above and will be determine later. We prescribe the initial data for such that its u derivative has a jump discontinuity across the sphere S0,us . Let AB = (( )1 )AB + (u us )(( )2 )AB 1{uus } . For suciently small depending on C , AB is positive denite for 0 u . We then dene 1 AB = . det AB

19

On S0,0 , since AB and AB are in the same conformal class, AB = 2 AB . (9) implies that AB = 2AB = 2 AB + trAB . u On the other hand AB = 2 AB + 2 AB . u u u ( 1 )AB we can identify 1 AB = 2 AB , 2 u and tr = (13) AB = log det = 0, u u

Since

2 . (14) u With this identication, has a jump discontinuity at u = us while tr is continuous. Notice also that 1 1 AC 1 BD | |2 ) ( ) AB CD = ( 4 u u depends only on and is in L . Thus, (11) can be re-written as 2 1 1 AC 1 BD + ( ) ( ) CD = 0. (15) AB u2 8 u u This ordinary dierential equation can be solved up to u u as long as u for suciently small . (15) also implies that u , = 0, and thus the metric on the spheres remain regular. 3.1. Approximation Procedure. We now introduce the approximation procedure to construct a sequence of smooth initial data approaching the data described above. Consider a 0 that is supported in [1, 1] and is identically 1 in [ 1 , 1 ]. Let C0 (R) function h 2 2 (x) = h is smooth. Now let Note that h

n

0 ( x) h 0 (2x) x 0 h 0 x<0

hn (x) =

j =

(2n x). h

We note that hn is supported in {x 2(n+1) } and hn := hn hn1 is supported in {2(n+1) x 2n }. Moreover, hn 1{x0} in Lp for every p < as n and xhn x1{x0} in Lp for every p as n . At the n-th step, we dene (n , n , trn ) = (, , tr) on S0,0 and , tr) on H 0 ( n , trn ) = (

20

as before. Dene )2 )AB on H0 , ( n )AB = (( )1 )AB + (u us )hn (u us )(( and ( n )AB = 1 det( )n ( n )AB on H0 .

Notice that for suciently small, we have uniform upper bounds for each component of and their angular derivatives and a uniform lower bound for det . Moreover, it is easy to see that for n := n n1 , we have n | C 2(1+k)n , for us u us + 2n , k K + 1, |( )k ( )i u iI +3 ( Thus k i = 0, ) ( ) u n ( |( for u us + 2n , for i I + 3, k K + 1,

1 1 k i p n | du) p C 2(1+k p )n for k K + 1. ) ( ) u 0 Moreover, we know that n in the sense that 1 1 n )|p du) p C 2(1+k p )n for k K + 1. ( |( )k ( )i ( u 0

With the denition of ( n )AB , we can solve (15) for n and use (13) and (14) to solve for n and (tr)n . This then allows us to use the null structure equations and the Bianchi identities to solve for all Ricci coecients and curvature components, as well as their derivatives. In the following we will use the notation that a subscript n denotes that metric component, Ricci coecient or curvature component associated to n . Moreover, we will use the notation n = n n1 . We now derive the bounds for each of the n s and n s in the initial data. By uniqueness of the constraint ODEs, all n = 0 for u us . We therefore only derive bounds for us u . From (15), we have c n C,

kK +2 iI +3

|(

k i ) ( ) n | C u

thus |(

kK iI +3

i ) (, tr)n | C.

kK +2 iI +3

k i ) ( ) | C 2(2+k)n , u n

for us u us + 2n ,

21

kK iI +3

for us u us + 2n , for u us + 2n ;

kK iI +3

kK iI +3

for us u us + 2n for u us + 2n .

kK iI +3

From the equation for on H0 together with the Codazzi equation, we have 1 3 4 = div tr tr. 2 2 Notice that since = 1 on H0 , we can write 4 = + . u Thus, we can integrate to get the estimate |( )k ( )i n | C. u kK +1 iI +2 For the dierence n = n n1 , |(

kK +1 iI +2

k i ) ( ) n | C 2n . u

Notice that we have used the control on the metric to relate div to the coordinates derivatives. Since on H0 , = = , we have |( )k ( )i (n , n )| C. u kK +1 iI +2 and |(

kK +1 iI +2

k i ) ( ) (n , n )| C 2n . u

22

and the Gauss equation 1 1 K = + trtr, 2 4 we have 4 (tr) + trtr = 2K 2div + | |2 . Since the estimates for the Gauss curvature K can be derived from the estimates for and its derivatives alone, by the estimates derived above and the initial estimate for tr on S0,0 , |(

kK +1 iI +1

k i ) ( ) trn | C. u

and |(

kK +1 iI +1

k i ) ( ) trn | C 2n . u

kK +1 iI +1

k i ) ( ) | C. u n

and |(

kK +1 iI +1

k i ) ( ) | C 2n . u n

kK 1 iI +3

k i ) ( ) n | C 2(1+k)n , u |( k i ) ( ) n | C 2n , u

for us u us + 2n , for u us + 2n ,

kK 1 iI +3

kK iI +2

k i ) ( ) n | C for us u us + 2n , u k i ) ( ) n | C 2n for u us + 2n . u

|(

kK iI +2

23

we have |(

kK iI +1

|(

kK iI +1

kK iI +1

k i ) ( ) n | C for us u us + 2n , u |( i ) | C 2n for u us + 2n . n

kK iI +1

iI +2

i ) | C on S0,0 . n

kK +1 iI

k i ) ( ) | C, u n

|(

kK +1 iI

k i ) ( ) n | C 2n . u

I +1

|(

kK +1 i=0 I +1

k i ) ( ) | C. u n

|(

kK +1 i=0

k i ) ( ) | C 2n . u n

1 Together with = 0 which follows from = 2 4 (log ), we have the bounds in L for all the Ricci coecients and null curvature components on H0 . A similar procedure allows us to solve for the Ricci coecients and the null curvature components on H 0 . Recall that for an impulsive gravitational wave, we take the data on H 0 to be xed, i.e., n = and n = on H 0 .

24

j Finally, we see that on H0 , we can dene j 3 n and 3 n for j J by taking the values j of j 3 n and 3 n restricted to S0,0 and solve the ODEs as above. Notice that they can be solved as long as I is suciently large depending on J . This allows us to conclude

Proposition 1. For every I 2, J, K 0, there exists characteristic initial data for an impulsive gravitational wave such that on H 0 :

k i sup ||j 3 4 ||L2 (S0,u ) CI,J,K , j J kK iI +1 u

(16) (17)

i sup ||j 3 ||L2 (S0,u ) CI,J , j J iI +1 u

(18) (19)

k i sup ||j 3 4 ||L2 (S0,u ) CI,J,K , j J kK iI +1 u

and for suciently small, is positive denite on H0 and H 0 with bounded angular derivatives up to the (I + 2)-nd derivative. There exists a sequence of of smooth initial data approaching the characteristic initial data described above such that (16)-(19) hold for n and n , and for H0 \ {us u us + 2n },

k i sup ||j 3 4 n ||L2 (S0,u ) CI,J,K , j J kK iI +1 u

(20) (21)

i n sup ||j 3 n ||L2 (S0,u ) CI,J,K 2 . j J i I u

(22)

This sequence of data converges in the sense that on H 0 , the data is xed; and on H0 , for n = n n1 and n = n n1 ,

i n sup ||j , 3 n ||L2 (S0,u ) CI,J 2 j J iI +1 u

(23) (24)

25

on H0 \ {us u us + 2n },

k i n sup ||j , 3 4 n ||L2 (S0,u ) CI,J,K 2 j J kK iI +1 u n k i . sup ||j 3 4 n ||L2 (S0,u ) CI,J,K 2 j J kK iI {,,, } u

(25) (26)

Remark 6. In the context of Theorem 3, a corresponding sequence of n and n can be constructed. As opposed to the case of an impulsive gravitational wave, such a sequence will satisfy (16)-(19) and (23)-(24), but it will not obey the improved estimates of Proposition 1 which holds on H0 \ {us u us + 2n }. This additional regularity is only relevant for the argument that the impulsive gravitational wave is smooth away from H us . Proposition 1 in particular constructs the set of initial data satisfying the assumptions of Theorem 1. In the following, we will focus on the case I = 2, J = K = 0. This level of regularity is sucient to prove all the estimates and to construct a spacetime metric satisfying the Einstein equations. 4. A Priori Estimates In this section, we begin the proof of Theorem 3. We show that for any smooth characteristic initial data satisfying the estimates in the assumption of Theorem 3, there is an depending only on the constants in the estimates in the assumptions such that the unique resulting spacetime admits a double null foliation (u, u) and remains smooth with precise estimates on its Ricci coecients and curvature components in 0 u u and 0 u u for u, u . In particular, no assumptions have to be made on the size of . In the context of Theorem 3, our Theorem in this Section asserts that a region of existence can be found independent of n for any smooth approximating sequence of data satisfying the bounds in the assumptions of Theorem 3. We rst dene the norms that we will work with. Let R=

i2

{,,, }

u {,,,}

u

u,u

and i,2 = O

H {tr,,,,, tr}

u H {tr,,,,tr, } 2 1

u

We write O=

Oi,2 +

i=0 i=0

Oi,4 + O0, .

The following is the main theorem in this section: Theorem 4. Suppose the initial data set for the characteristic initial value problem is smooth and satises c | det Su,0 | C, |( )i Su,0 | C, i 3

26

O0 :=

i 2

u i2

u

R0 :=

i2

{,,, } {,,,}

||i ||L2 (H 0 ) C.

Then for suciently small depending only on C and c, there exists a unique spacetime (M, g ) endowed with a double null foliation (u, u) that solves the Einstein equations in the region 0 u u , 0 u u , whenever u , u . Moreover, in this region, the following norms are bounded above by a constant C depending only on C and c: 3,2 , R < C . O, O It follows by [36] that for every smooth initial data set, a unique smooth spacetime satisfying the Einstein equations exists in a small neighborhood of S0,0 . In order to show that the size of this region only depends on the constants in the assumption of Theorem 4, it suces to establish a priori control in the region 0 u , 0 u . The Theorem then follows from a standard last slice argument (See [28]). In this section, all estimates will be proved under the following bootstrap assumption: O0, 0 . (A)

We now outline the steps in proving a priori estimates. In Section 4.1, we will rst estimate the metric components under the bootstrap assumption (A). In Section 4.2-4.5, we will derive some preliminary estimates and formulae. In Section 4.2, we provide Propositions which gives Lp estimates for general quantities satisfying transport equations. They will be used to control the Ricci coecients and curvature components. In Section 4.3, we establish Sobolev Embedding Theorems in our setting. In Section 4.4, we state some formulae for the commutators [4 , ] and [3 , ]. They will then be used to obtain higher order estimates for general quantities satisfying transport equations. In Section 4.5, we prove elliptic estimates for general quantities obeying Hodge systems. After these preliminary estimates, we estimate the Ricci coecients and the curvature components. This proceeds in three steps: 3,2 < together imply that STEP 1: In Section 4.6, we prove that R < and O O C (O0 ), where C (O0 ) is a constant depending only on the initial norm O0 . These estimates are proved via the null structure equations. 3,2 STEP 2: In Section 4.8, we use elliptic estimates to prove that R < implies O C (O0 , R0 , R), where C (O0 , R0 , R) is a constant depending both on the initial norm and R. STEP 3: In Section 4.9, we derive the energy estimates and prove the boundedness of R by the initial data. These steps provide bounds for the Ricci coecients up to three angular derivatives and the curvature components up to two angular derivatives depending only on the constants in the assumptions of Theorem 4. In Section 4.10, we show that this implies that higher regularity propagates and that the spacetime remains smooth in a region depending only

27

on the constants in the assumptions of Theorem 4. This allows us to conclude the proof of Theorem 4 in Section 4.11. Finally, in Section 4.12, we return to the sequence of smooth spacetimes arising from the sequence of data approximating the data of an impulsive gravitational wave as described in Section 3. We show that these spacetimes obey estimates in addition to those given in Theorem 4. These bounds will be applied in Section 5.10 to establish the regularity and singularity properties of an impulsive gravitational spacetime. 4.1. Estimates for Metric Components. We rst show that we can control under the bootstrap assumption (A): Proposition 2. There exists

0

0 (0 )

0,

1 2. 2 Proof. Consider the equation 1 1 1 1 = 4 log = 4 1 = . (27) 2 2 2 u Fix u. Notice that both and are scalars and therefore the L norm is independent of the metric. We can integrate equation (27) using the fact that 1 = 1 on H0 to obtain

u

0

|| ||L (Su ,u ) du C 0 .

This implies both the upper and lower bounds for for suciently small . We then show that we can control under the bootstrap assumption (A): Proposition 3. Consider a coordinate patch U on S0,0 and dene Uu,0 to be a coordinate patch on Su,0 given by the one-parameter dieomorphism generated by L. Dene Uu,u to be the image of Uu,0 under the one-parameter dieomorphism generated by L. Dene also DU = 0u ,0u Uu,u . For small enough depending on initial data and 0 , there exists C and c depending only on initial data such that the following pointwise bounds hold for in DU : c det C. Moreover, in DU , |AB |, |( 1 )AB | C. Proof. The rst variation formula states that L / L = 2. In coordinates, this means AB = 2AB . u From this we derive that log(det ) = tr. u Dene 0 (u, u, 1 , 2 ) = (0, u, 1 , 2 ).

u

| det det(0 )| C

0

|tr|du C 0 .

28

This implies that the det is bounded above and below. Let be the larger eigenvalue of . Clearly, C sup ,

A,B =1,2

(28)

A,B =1,2

Then

u

|AB (0 )AB | C

0

|AB |du C 0 .

Using the upper bound (28), we thus obtain the upper bound for |AB |. The upper bound for |( 1 )AB | follows from the upper bound for |AB | and the lower bound for det . A consequence of the previous Proposition is an estimate on the surface area of each two sphere Su,u . Proposition 4. sup |Area(Su,u ) Area(Su,0 )| C 0 .

u,u

Proof. This follows from the fact that chosen to be appropriately small.

is

With the estimate on the volume form, we can now show that the Lp norms dened with respect to the metric and the Lp norms dened with respect to the coordinate system are equivalent. Proposition 5. Given a covariant tensor A1 ...Ar on Su,u , we have

r p/2 < , > Su,u i=1 Ai =1,2

|A1 ...Ar |p

det d1 d2 .

We can also control b under the bootstrap assumption, thus controlling the full spacetime metric: Proposition 6. In the coordinate system (u, u, 1 , 2 ), |bA | C 0 . Proof. bA satises the equation bA = 42 A . u This can be derived from [L, L] = bA . u A

29

4.2. Estimates for Transport Equations. The estimates for the Ricci coecients and the null curvature components are derived from the null structure equations and the null Bianchi equations respectively. In order to use the equations, we need a way to obtain estimates from the null transport type equations. Such estimates require the boundedness of tr and tr, which is consistent with our bootstrap assumption (A). Below, we state two Propositions which provide Lp estimates for general quantities satisfying transport equations either in the e3 or e4 direction. Proposition 7. There exists we have

0

0 (0 )

u

u u

Lp (Su,u )

Lp (Su ,u )

,u )

Proof. The following identity holds for any scalar f : d du Similarly, we have d du Hence, taking f = ||p , we have

u

f=

Su,u Su,u

df + trf du

=

Su,u

(e4 (f ) + trf ) .

f=

Su,u Su,u

e3 (f ) + trf .

,u )

+

u u Su,u

p2

+

u Su

,u

(29)

The Proposition is proved using Cauchy-Schwarz on the sphere and the L bounds for and tr (tr) which are provided by Proposition 2 and the bootstrap assumption (A) respectively. The above estimates also hold for p = : Proposition 8. There exists

0

0 (0 )

u

0,

we have

u u

,u )

du

Proof. This follows simply from integrating along the integral curves of L and L, and the estimate on in Proposition 2.

30

4.3. Sobolev Embedding. Under the bootstrap assumption (A), the Sobolev Embedding hold on a 2-sphere Su,u . Proposition 9. There exists

0

0 (0 )

1

0,

we have

||||L4 (Su,u ) C

i=0

Proof. We rst prove this for scalars. Given our coordinate system, the desired Sobolev Embedding follows from standard Sobolev Embedding Theorems and the lower and upper bound of the volume form. Thus, the proposition holds for scalars. Now, for being a 2 tensor, let f = ||2 + . Then < , > ||f ||L4 (Su,u ) C ||f ||L2 (Su,u ) + || ||L2 (Su,u ) C ||f ||L2 (Su,u ) + ||||L2 (Su,u ) . 2 2 || + The Proposition can be achieved by sending 0. We can also prove the Sobolev Embedding Theorem for the L norm: Proposition 10. There exists

0

0 (0 )

0,

we have

2

||||L (Su,u ) C

i=0

Proof. The rst statement follows from coordinate considerations as in Proposition 9. The second statement follows from applying the rst and Proposition 9. 4.4. Commutation Formulae. We have the following formula from [22]: Proposition 11. The commutator [4 , ] acting on an (0, r) S-tensor is given by [4 , B ]A1 ...Ar =[D4 , DB ]A1 ...Ar + (B log )4 A1 ...Ar ( 1 )CD BD C A1 ...Ar

r r

i=1

1 CD

BD A A1 ...A i C...Ar +

i

i=1

By induction, we get the following schematic formula for repeated commutations: Proposition 12. Suppose 4 = F0 . Let 4 i = Fi . Then Fi

i1 +i2 +i3 =i

i1 ( + )i2 i3 F0 +

i1 +i2 +i3 +i4 =i

i1 ( + )i2 i3 i4

+

i1 +i2 +i3 +i4 =i1 i1

i1 ( + )i2 i3 i4 .

where by ( + )i2 we mean the sum of all terms which is a product of i2 factors, each factor being j ( + ) for some j and that the sum of all j s is i1 , i.e., i1 ( + )i2 =

31

j1 +...+ji2 =i1

Gi

i1 +i2 +i3 =i

i1 ( + )i2 i3 G0 +

i1 +i2 +i3 +i4 =i

i1 ( + )i2 i3 i4

+

i1 +i2 +i3 +i4 =i1

i1 ( + )i2 i3 i4 .

Proof. The proof is by induction. We will prove it for the rst statement. The proof of second one is analogous. The formula obviously holds for i = 0. Assume that the statement is true for i < i0 . Fi0 =[4 , ]i0 1 + Fi0 1 i0 + ( + )4 i0 1 + i0 1 + ( + )i0 1 +

i1 +i2 +i3 =i0

i1 ( + )i2 i3 F0 i1 ( + )i2 i3 i4

i1 +i2 +i3 +i4 =i0

+ +

i1 ( + )i2 i3 i4 .

i1 +2i2 +i3 +i4 =i0 1

First notice that the rst, third and fourth terms are acceptable. Then plug in the formula for Fi0 1 = 4 i0 1 and get the result. The following further simplied version is useful for our estimates in the next section: Proposition 13. Suppose 4 = F0 . Let 4 i = Fi . Then Fi

i1 +i2 +i3 =i i

i1 i2 i3 F 0 +

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 .

i1 +i2 +i3 =i

i1 i2 i3 G0 +

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 .

Proof. We replace and using the Codazzi equations, which schematically looks like = + , = + . 4.5. General Elliptic Estimates for Hodge Systems. We recall the denition of the divergence and curl of a symmetric covariant tensor of an arbitrary rank: (div )A1 ...Ar = B BA1 ...Ar , (curl )A1 ...Ar = / BC B CA1 ...Ar , where / is the volume form associated to the metric . Recall also that the trace is dened to be (tr)A1 ...Ar1 = ( 1 )BC BCA1 ...Ar1 . The following elliptic estimate is standard (See for example [13] or [12]):

32

Proposition 14. Let be a totally symmetric r + 1 covariant tensoreld on a 2-sphere (S2 , ) satisfying div = f, curl = g, tr = h. Suppose also that ||i K ||L2 (S ) < .

i1

Then for i 3,

i1

|| ||

L2 (S )

C(

k 1

|| K ||

L2 (S )

)(

j =0

For the special case that a symmetric traceless 2-tensor, we only need to know its divergence: Proposition 15. Suppose is a symmetric traceless 2-tensor satisfying div = f. Suppose moreover that ||i K ||L2 (S ) < .

i 1

Then, for i 3,

i1

||i ||L2 (S ) C (

k 1

||k K ||L2 (S ) )(

j =0

Proof. In view of Proposition 14, this Proposition follows from curl = f. This is a direct computation using that fact that is both symmetric and traceless. 4.6. Lp (S ) Estimates for the Ricci Coecients. We continue to work under the bootstrap assumptions (A). In this Subsection, we pursue step 1 of the proof as outlined in the beginning of the Section. We show if that the curvature norm R and the third order Ricci 3,2 are bounded, then so is the Ricci coecient norm O. In particular, our coecient norm O bootstrap assumption (A) and all the estimates in the last section are veried as long as R 3,2 are controlled. and O Proposition 16. Suppose Then there exists

0

where O0 is the initial norm for the Ricci coecients dened in Theorem 4. Proof. To prove the Proposition, we need to estimate

i1

Oi,4 and

i2

the Oi,4 norms. Using the null structure equations for all Ricci coecients except , we have 3 =

{tr, ,, } {,,,}

+ ,

33

and 4 =

{,,, }

+ ,

{tr,, } where the on the right hand sides denote any Ricci coecients. We will also use the null structure equations commuted with angular derivatives:

i 3 =

,, } {tr, {,,,} i1 +i2 +i3 =i

i1 i2 i3 +

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 , ,

i1 +i2 +i3 +i4 =i i1 i2 i3 i4

i 4

{tr,, }

=

{,,, } i1 +i2 +i3 =i

i1

i2

i3

(30)

where we have applied the notation dened in Proposition 12. By Sobolev Embedding in Proposition 9 and the bootstrap assumption (A), we have

2

|| ||L4 (Su,u ) 0

i1 +i2 +i3 1 i=0

i1

i2

i3

1

i1 +i2 +i3 +i4 1 i=0

Oi,4 .

(31)

1

i=0 1

(32) Oi,4 .

O0 + C

1 2

0 R + C (0 )

i=0

In order to close this estimate, we need to estimate . We like to avoid the use of the equation 4 = + since we do not have an estimate for singular curvature component . Instead, we use the equation 3 = + . Commuting with angular derivatives, we get 3 i =

i1 +i2 +i3 =i

i1 i2 i3 +1 +

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 .

(33)

1 u 1

i=0 0

i=0

Oi,4 .

u 0

1 2

3,2 . O

1

Therefore,

1

i=0

1 2

3,2 + C (0 ) O

i=0

Oi,4 .

(34)

34

1 1

Oi,4 O0 + C

i=0

1 2

0 R + C O2,2 + C

1 2

3,2 + C (0 ) O

i=0

Oi,4 .

(35)

By choosing

1

i=0

where C (O0 ) does not depend on 0 , once is chosen to be small enough. We now move to estimate the Oi,2 norms. Applying Proposition 7 to (30), we obtain

2

i=0 2 1

(36) Oi,4 )2 .

O0 + C

1 2

0 R + 0

i=0

Oi,2 + (

i=0

2 2 1

|| ||

i=0

L2 (S

u,u )

C (O0 ) + C

1 2

3,2 + C 0 O

i=0

Oi,2 + (

i=0

Oi,4 )2 .

(37)

2 2 1

Oi,2 O0 + C

i=0

1 2

0 R + 0

i=0

Oi,2 + (

i=0

Oi,4 )2 .

Choosing

2

0

i=0

2 1

Oi,2 C (O0 ) + (

i=0 i=0

Oi,4 )2 .

(38)

For

1 2

Oi,4 +

i=0 i=0

Oi,2 C (O0 ).

Finally, we use Sobolev embedding in Proposition 10 to get the necessary bounds for O0,

1

i=0

3,2 are controlled. Furthermore, the bootstrap assumption (A) is veried as long as R and O since the choice of 0 depend only on O0 , the can be set to depend only on O0 , R and 3,2 . O

35

4.7. L2 (S ) Estimates for Curvature. In this Subsection, we prove that , , , and their rst derivatives can be controlled in L2 (S ). The argument relies on Proposition 7 and the Bianchi equations (8). The results of this Section will give the bounds for the Gauss curvature and its rst derivatives appearing in Proposition 14. Proposition 17. Suppose Then there exists

0

3 ,2 < . R < , O 3,2 ) such that for 0 , = 0 (O0 , R, O ||i ||L2 (Su,u ) C (R0 ).

i1 {,,, }

{,,,}

+

{,,,,}

{,,,}

2 +

( + ).

{,,,,}

i 1

i2 {,,,}

u i1 1

i2 2 {,,,}

||i2 ||L2 (H u )

+ C sup

u i1 1

i2 2 {,,,}

1

i=0 i 1

i1

1 2

C (O0 )R + C (O0 )

3 2

i 1

We now derive L2 (S ) estimate for K and its derivative, which will in particular allow us to apply Proposition 14. Proposition 18. Suppose Then there exists

0

3 ,2 < . R < , O 3,2 ) such that for 0 , = 0 (O0 , R, O ||i K ||L2 (S ) C (O0 , R0 )

i1

36

Proof. We use the equation 1 1 K = + tr tr, 2 4 and bound term by term using Propositions 16 and 17. 4.8. Elliptic Estimates for Ricci Coecients. This Subsection contains step 2 of the proof of the a priori estimates outlined in the beginning of the Section. We bound the third derivatives of the Ricci coecients. They cannot be estimated by transport equations alone as before because that would require the control for the three derivatives of curvature, which we do not have. Instead, we need to use elliptic estimates, following ideas in [13], [22], [12], [24]. We hope to nd carefully chosen combination of := + such that satises transport equations 3 = H + + H , (39) or 4 = H + + H , (40) }, H {, , , } and H where H {tr, , , , , tr}, H {tr, , , , tr, {, , , }. These transport equations will then be coupled to the estimates from Hodge systems of the type D = + + to recover the control for the third derivatives of the Ricci coecients. In order to apply this strategy, it is desirable for (39) and (40) to have two special structures. Firstly, should not appear on the right hand side of the equation so that can be estimated by these transport equations without a loss of derivative. Secondly, we hope that the terms H and H do not enter the 3 equation (respectively, H and H do not enter the 4 equation). If this is satised for equations (39) and (40), then the integrated error terms can be controlled by the ux term that are bounded on the corresponding hypersurface. Indeed, the third derivatives of most the Ricci coecients can be obtained in this way by constructing approximate variables which satisfy either (39) or (40). An extra diculty arises when we estimate 3 . More precisely, the variable that allows us to retrieve estimates for derivatives of satises an equation of the form 3 = H + + H + , (41) which has an undesirable term . Since we prove estimates without any information on the singular curvature component , we automatically lose control of 2 in L2 (H ). Therefore, when obtaining the estimates from this transport equation using Proposition 7, the error term arising from in (41) cannot be bounded. In order to overcome this problem, we prove a weaker bound for 3 . More precisely, we will only control it in L2 (H ) but will not obtain any bounds of its L2 (H ) norm. We will moreover show that this is sucient to close the energy estimates. We rst bound the variables that satisfy equations of the form (39) and (40). Proposition 19. Let satisfy the equation 3 = H + + H , or 4 = H + + H ,

37

where H {tr, , , , , tr}, H {tr, , , , tr, }, H {, , , } and H {, , , }. Suppose 3 ,2 < . R < , O Then there exists

0

0 (O0 , R, O3,2 ) 2

0.

i=0 u,u

Proof. Notice that the assumption allows us to choose sition 16 holds. For satisfying

i1 +i2 +i3 +i4 =i

(i1 i2 i3 i4 +1 H + i1 i2 +1 i3 i4 + i1 i2 i3 i4 H ),

Take i 2. We estimate term by term. For the rst two terms, except for the factor with 3 H , can be estimated using the O norm, which is bounded in view of Proposition 16. The 3,2 norm. We have term 3 H can be controlled by the O

2 ||i1 i2 i3 i4 +1 H + i1 i2 +1 i3 i4 ||L1 C (O0 )(1 + u L (Su,u ) 1 2

3,2 ). O

The curvature term can be estimated using Proposition 16 and the denition of the curvature norm, since the component does not appear:

2 C (O0 )(1 + ||i1 i2 i3 i4 H ||L1 u L (Su,u ) 1 2

R).

2

1 3 ,2 ) . sup ||i ||L2 (Su,u ) C (O0 ) + C (O0 ) 2 (R + O

i=0

i1 +i2 +i3 +i4 =i

(i1 i2 i3 i4 +1 H + i1 i2 +1 i3 i4 + i1 i2 i3 i4 H ),

2

1 3 ,2 ) . sup ||i ||L2 (Su,u ) C (O0 ) + C (O0 ) 2 (R + O

i=0

We now write down the explicit quantities following [24] and show that the estimates for some of the Ricci coecients can be recovered from via elliptic equations.

38

Proposition 20. Assume R < , Let denote any of the following quantities: tr, tr, := div , := div . 3 ,2 < . O

3,2 ) such that for Then satises (39) or (40). Moreover, there exists 0 = 0 (O0 , R, O 0 , we have 1 sup ||3 (tr, tr)||L2 (Su,u ) C (O0 ) 2 ,

u,u

u

1 2

+ R), + R),

u

1 2

}. where, as before, H {tr, , , , , tr}, H {tr, , , , tr, Proof. Consider the following equations: 3 tr = H H , 4 tr = H H . These equations do not involve a curvature term. After commuting with an angular derivative, we have 3 tr = H + , 4 tr = H + . Thus tr and tr satisfy the requirement to be one of the quantities. Thus, by denition, ||3 tr||L2 (Su,u ) C ||2 ||L2 (Su,u ) . Moreover, ||3 tr||L2 (Hu ) C and ||3 tr||L2 (H u ) C

1 2 1 2

(42) (43)

From the estimates for 3 tr and 3 tr, we can obtain the estimates for 3 and 3 . This is because of the Codazzi equations: 1 div = tr + , 2 1 div = tr + + , 2 Since and are symmetric traceless 2-tensors and we have the estimates for the Gauss curvature in Proposition 18, we can apply the elliptic estimates for symmetric traceless 2-tensors in Proposition 15 ||3 ||L2 (Hu ) C (||3 tr||L2 (Hu ) + ||2 ||L2 (Hu ) +

i1 +i2 2

39

i1 +i2 2

C we have

1 2

u i2

1

1

(44)

||3 ||L2 (Hu ) C 2 ||2 ||L2 (Su,u ) + C R + C (O0 ) 2 . Similarly, using the Codazzi equation for , we get ||3 ||L2 (H u ) C Dene = div , , = div

1 2

(45) (46)

1 where as before = 2 . Notice that div and div satisfy the following equations:

4 div = div + + + H , 3 div = div + + + H , Recall the Bianchi equation: 1 3 + + 2 , 4 + tr = div 2 2 3 1 + 2 , 3 + tr = div 2 2 In the 4 equation , the term appears. This is undesirable because we do not have good estimates for . Instead we renormalize the equations using 1 . := 2 Now, we write down the equations for instead of and . Since 4 = + , the 4 equations now do not contain . Schematically, we have 4 = div + + H + 3 , 3 = div + + H + 3 , where all the curvature components are now acceptable. Thus and satisfy the following equations: 4 = H + H + , 3 = H + H + , Thus and also satisfy the schematic equation for . We now show that we can get the estimates 3 and 3 from that of 2 and 3 . To do this, we consider the following div-curl systems div = + , curl = + , div = + , curl = + .

40

Notice that and can be estimated by R on both Hu and H u . Then, using (44) and Proposition 14, we have ||3 (, )||L2 (Hu ) C and ||3 (, )||L2 (H u ) C

1 2

1

(47) (48)

1 2

The Proposition thus follows from (42), (43), (45), (46), (47), (48) and Proposition 19. The above construction of provides the desired estimates for the third derivatives of all 3,2 norm, the Ricci coecients with the exception of and . Therefore, to estimate the O 3 3 it remains to prove estimates for and . To this end we follow the strategy in [24]. Dene two auxiliary quantities and to be solutions to 1 3 = , 2 1 4 = , 2 with zero boundary conditions along H0 and H 0 respectively. For the remainder of the Section, will also be used to denote and ; H will also be used to denote ; and H will also be used to denote . It is easy to see that Proposition 21. Suppose Then there exists

0

i 2

0,

||i ( , )||L2 (Su,u ) C. Proof. Since 1 3 = , 2 1 4 = , 2 with zero data. It follows from Proposition 7 that ||i ( , )||L2 (Su,u ) C (O0 ) 2 R.

i 2

1

appropriately.

Now are now ready to use elliptic estimates to control 3 and 3 . As before, our strategy is to use a combination = + that can be estimated using a transport equation. However, in this case does not satisfy an equation of the type in Proposition 19. Proposition 22. Suppose Then there exists

0

3 ,2 < . R < , O 3,2 ) such that whenever = 0 (O0 , R, O ||3 (, )||L2 (Hu ) C (O0 )( ||3 (, )||L2 (H u ) C (O0 )(

1 2 1 2

0,

we have

+ R), + R),

41

Proof. Let 1 := + , 2 1 := + . 2 We have 1 3 = + + + , 2 1 3 = + + , 2 + = 3 + . Therefore 3 = + + . Similar, using 1 4 = + + + , 2 1 4 = + + , 2 and the Bianchi equation + = 4 + , we have 3 = + + . (50) The equations for and do not have the favorable structure of the equations of the other variables. In particular, for the 3 equation, and appear as source terms. The loss of the L2 energy estimate for is automatically accompanied by the loss of estimate for in L2 (H ). Therefore, 2 cannot be controlled in L2 (H u ). This means that we cannot estimate 2 in L2 (H u ). Nevertheless, as we now show, we can estimate 2 in L2 (Hu ). This estimate will be sucient in the sequel. Commuting (49) with angular derivatives to get 3 2 =

i1 +i2 +i3 2

(49)

i1 i2 i3 +

i1 +i2 +i3 3

i1 i2 i3 .

Notice that we now allow to represent the terms and . Since by Proposition 21 i and i satisfy the same estimates as any other for i 2, those terms can be estimated in the same way as in Proposition 19. The new terms that have not been estimated in Proposition 19 are 3 , i1 i2 i3 and 3 (, , ). We estimate them by

i1 +i2 +i3 2

2 2 || 3 ||L1 C (O0 ) 2 ||3 ||L2 , u L (S ) u L (S ) 1

and ||

i1 +i2 +i3 2

2 i1 i2 i3 ||L1 C (O0 ) u L (Su,u ) 1 2 2 ||i ||L2 , u L (Su,u )

i2

1

and

2 2 || 3 (, )||L1 C (O0 ) 2 ||3 (, )||L2 . u L (S ) u L (S )

42

Then, using Proposition 7 and the estimates in the proof of Proposition 19, we have ||2 ||L2 (Su,u ) C (O0 )(1 + +

1 2 1 2

3,2 + O

1 2

R+

1 2

i2

This holds for any xed u, u. Integrating this in L2 u , we get ||2 ||L2 (Hu ) C (O0 ) 2 (1 + +

i 2

1 1 2

2 2 ||i ||L2 u Lu L (Su,u ) 1 2

R,

i2

and

2 2 ||3 (, , )||L2 ) u Lu L (Su,u ) 1 2

3,2 + O

1 2

1 1 2

3 ,2 + O

1 2

R+

1 2

1 2

2 2 ||3 ||L ). u Lu L (Su,u ) (52) We show that we can recover 3 , 3 , 3 and 3 by considering the div-curl system for and . We have 1 div = div + div , 2 curl = 0, 1 div = curl + curl , 2 curl = 0, 1 div = div div , 2 curl = 0, 1 div = curl curl , 2 curl = 0. Thus, by Proposition 14 and 18, we have

1 2

3,2 + O

1 2

R+

1 2

1 2

i 2

43

i 2

||i ||L2 (H u ) ).

Using (51) and (52), we can thus choose suciently small to conclude the Proposition. Putting these all together gives Proposition 23. Assume R < . Then there exists

0

1 2

+ R).

Proof. Impose the bootstrap assumption 3,2 1 . O (53) Under this assumption, we can choose depending on O0 , R and 1 such that the conclusions in Propositions 19, 20 and 22 hold. Thus, 3,2 C (O0 )( O

1 2

+ R).

This improves the bootstrap assumption (53) for 1 suciently large. Moreover, 1 can be chosen to depend only on O0 and R. Thus, 0 depends only on O0 and R. 4.9. Energy Estimates for Curvature. In this Subsection, we carry out step 3 in the proof of the a priori estimates as outlined in the beginning of the Section. We prove energy estimates for the curvature components and their rst two derivatives, thus providing bounds for R and closing all the estimates. These energy estimates will be derived after renormalization directly from the Bianchi equations. This is analogous to [18], where a similar procedure without curvature renormalization has been carried out to derive energy estimates for all components of curvature. We remind the reader of our challenge to avoid any information on the non-L2 -integrable curvature component in our context. We recall the following two Propositions from [28], which can be proved by a direct computation. Proposition 24. Suppose 1 and 2 are r tensorelds, then 1 4 2 +

Du,u Du,u

2 4 1 =

H u (0,u)

1 2

H 0 (0,u)

1 2 + 1 2 +

H0 (0,u) (1)

(2)

1 3 2 +

Du,u Du,u

2 3 1 =

Hu (0,u)

1 2

(1) A1 A2 ...Ar

and an r 1 tensoreld

Du,u

Ar (2) A1 ...Ar1 +

Du,u

( + )(1) (2) .

Du,u

By using the Bianchi equations and the integration by parts formula in Proposition 24 and 25, we can prove the following energy estimates. The main new ingredient is the observation that can be renormalized away in the energy estimates. We can therefore control , , , , without any knowledge of

44

Proposition 26. The following L2 estimates for the curvature components hold for i 2: |i |2 +

{,,, } Hu {,,,} Hu

|i |2 |i |2 +

{,,,} Hu

1 2

{,,, } Hu

|i |2 + +

Du,u

C (O0 ).

i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 i1 i2 i3 K i4

i1 +i2 +i3 +i4 =i1

+

Du,u

i i

Du,u i1 +i2 +i3 +i4 =i+1

i1 i2 i3 i4 .

Proof. We rst consider the following sets of Bianchi equations: 3 + tr = + 2 + + 2 + 3( + ), 3 1 4 + tr = div + 2 , 2 2 1 3 + + 2 , 4 + tr = div 2 2 In the 4 equations for and , the term appears. This is undesirable because we do not have good estimates for . Instead we renormalize both and as in the proof of Proposition 20. Dene 1 := + , 2 1 . := 2 Now, we write down the equations for and instead of and . Since 4 = + , the equations now do not contain . Schematically, we have 4 = div + + + 3 , 4 = div + + + 3 , (54)

where {, , , , } is now one of the acceptable curvature components. Rewrite also the rst equation as 3 = + + + . (55) Applying Proposition 25 yields the following: < , 3 > =

Du,u

< , + > + < , > < div , > + < div , > + < , + >

Du,u

= =

Du,u

45

Applying Proposition 24 then yields the energy estimates for the 0-th derivative of the curvature: | |2 +

Hu Hu

2 + 2 2 + 2 +

Hu Du,u

Hu

| |2 +

( + + 3 ).

i1 +i2 +i3 +i4 =i

i4 i1 i2 i3 +

i1 +i2 +i3 +i4 +i5 =i1

, ) + i1 i2 i3 K i4 i5 (

i1 +i2 +i3 =i

i1 i2 i3 ( ),

and 4 i div i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 +

i1 +i2 +i3 +i4 +i5 =i1

i1 i2 i3 K i4 i5 +

i1 +i2 +i3 =i

i1 i2 i3 ( ),

and 4 i div i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 +

i1 +i2 +i3 +i4 +i5 =i1

i1 i2 i3 K i4 i5 +

i1 +i2 +i3 =i

i1 i2 i3 ( ),

where (div i )A1 ...Ai = (i+1 )B A1 ...Ai B . We integrate by parts using Proposition 24 and 25 as in the 0-th derivative case to get |i |2 +

Hu Hu

|i (, )|2 |i (, )|2 +

Hu Du,u

Hu

|i |2 + +

Du,u

i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4

i

i1 +i2 +i3 +i4 =i1

i1 i2 i3 K i4 i1 i2 i3 i4 .

i1 +i2 +i3 +i4 =i+1

+

Du,u

46

We then consider the following set of Bianchi equations. Notice that does not appear in any equations in this set. 3 1 2 , 3 + tr = div + 2 2 3 1 3 + tr = div + 2 , 2 2 4 + tr = + + 2 + 2 3( ), From these we can derive the following estimates: |i (, )|2 +

Hu Hu

|i |2 |i |2 +

Hu Du,u

Hu

|i (, )|2 + +

Du,u

i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4

i

i1 +i2 +i3 +i4 =i1

i1 i2 i3 K i4 .

Finally, we look at the following set of Bianchi equations: 3 + 2tr = div 2 + , 1 4 + tr = + 4 3( ) + ( 4 ) 2 Again, we notice the absence of the terms. Hence we have the estimates: |i |2 +

Hu Hu

|i |2 |i |2 +

Hu Du,u

Hu

|i |2 + +

Du,u

i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4

i

i1 +i2 +i3 +i4 =i1

i1 i2 i3 K i4 .

{,,, } Hu { , ,,} Hu

|2 |i

can be bounded by the right hand side in the statement of the Proposition. To conclude, we use the fact that = + ,

47

and thus |i |2

{,,,} Hu

{ , ,,} Hu

|2 + |i

i1 +i2 2

||i1 i2 ||L2 (H u )

1 2

{ , ,,} Hu

|2 |i +

C (O0 ).

Finally, we prove the boundedness of the norm R: Proposition 27. There exists

0

0 (O0 , R0 )

0,

R C (O0 , R0 ). Proof. Assume as a bootstrap assumption: R 2 . Thus we can apply Propositions 16, 18 and 23. By Proposition 26, we have

2

(56)

(

i=0 {,,, } Hu

(i )2 +

{,,,} Hu

1 2

(i )2 ) i i1 i2 i3 i4

i1 +i2 +i3 +i4 2

C R2 0+ +

C (O0 ) +

Du,u i2

i

Du,u i2 i1 +i2 +i3 +i4 1

i1 i2 i3 K i4 (57)

i1 +i2 +i3 +i4 3 i1 i2 i3 i4

+

Du,u i2

(i )2 +

i2 Du,u

C R2 0+C +

i1 +i2 +i3 +i4 1

i1 i2 i3 i4

i1 +i2 +i3 +i4 2 Du,u 2

i1 i2 i3 K i4

Du,u

+

i1 +i2 +i3 +i4 3 Du,u

i1 i2 i3 i4

i 2

1 2

R,

since i can be estimated either in L2 (Hu ) or L2 (H u ). By Sobolev Embedding in Propositions 9 and 10, we thus have

2 4 2 ||i ||L2 + ||||L2 C u Lu L (S ) u Lu L (S ) 1 2

R,

i 1

48

i 3 i2

2 4 ||i ||L2 + u Lu L (S ) 2 ||i ||L2 C u Lu L (S ) 1 2

3,2 , O

i 1

i3 i 2

2 4 ||i ||L2 + u Lu L (S ) 2 ||i ||L2 C (O0 , R0 ) 2 R. u Lu L (S ) 1

i 1

i1 +i2 +i3 +i4 2 Du,u 2

i1 ,i2 1,i3 1

(

i1 2

1 2

u,u

u,u i3 2

i2 2

i1 +i2 +i3 1 Du,u 2

i1 +i2 +i3 1

(

i1 2

1 2

u,u

u,u i3 2

i2 1

i1 +i2 +i3 +i4 3 Du,u 2

i1 +i2 +i3 +i4 3

(

i1 3

1 2

u,u

u,u i3 3

i2 2

2

(

i=0 {,,, } Hu

(i )2 +

{,,,} Hu

(i )2 ) C R2 0+

1 2

C (O0 ) +

1 2

C (O0 , R0 )R.

49

By choosing

This improves (56) for 2 suciently large. Moreover, 2 can be chosen to depend only on O0 and R0 . Thus the choice of depends only on O0 and R0 . 4.10. Propagation of Regularity. Up to this point, we have been considering spacetimes with smooth characteristic initial data. In the context of an impulsive gravitational wave, our estimates in this Section apply to spacetimes arising from the smooth data approximating the given singular data. For these spacetimes, we can also prove that estimates for the higher order derivatives of the curvature. In view of the bounds that we have already obtained, this follows from standard arguments. In particular, energy estimates for the higher order derivatives can be derived as in Section 4.9. In this case, since the equations are linear in the highest order derivatives, one can use the bounds from the previous Sections together with Gronwalls inequality to obtain the desired estimates. We omit the details and present nal result. Proposition 28. Suppose, in addition to the assumptions of Theorem 4, the bound (

j J kK iI {,,,} k i ||j 3 4 ||L2 (H 0 ) + {,,, } k j ||i 3 4 ||L2 (H0 ) ) DI,J,K .

holds initially for some I, J, K . Then, the following bounds hold for 0 u u , 0 u u , (

j J kK iI {,,,} k i ||j 3 4 ||L2 (H u ) + {,,, } k i ||j 3 4 ||L2 (Hu ) ) DI,J,K .

for u , u , where can be chosen as in Theorem 4 and the constant DI,J,K depends only on the size of the initial data norm in the assumption of Theorem 4 and DI,J,K . Remark 7. Notice that for approximating solutions for an impulsive gravitational wave, the constants in the initial norms DI,J,K depend on n. However, the initial data satisfy uniform estimates for each I 2, J 0 with K = 0. Therefore, the corresponding uniform estimates will hold in the region 0 u u and 0 u u . 4.11. End of Proof of Theorem 4. Once we have the estimates for all the higher derivatives, the proof of Theorem 4 is standard. We refer, for example, to Section 6 of [28] for details. 4.12. Additional Estimates for an Impulsive Gravitational Wave. In this Subsection, we focus our attention on the sequence of spacetimes (Mn , gn ) with characteristic initial data converging to that of an impulsive gravitational wave as constructed in Section 3. The conclusion of Theorem 4 can be applied to (Mn , gn ). The initial data of the sequence of spacetimes, however, possess an additional property that n and its angular derivatives are bounded uniformly in L1 u . In this Subsection, we prove that in the constructed spacetimes (Mn , gn ), the same estimates hold for all u u with u . Moreover, we show that n concentrates around us and if n is initially more regular uniformly in n away from us , it is also uniformly more regular away from us for 0 u u . We begin by proving energy estimates for n in L2 (Hu ) and n in L2 (H u ) using the pair of Bianchi equations for (3 n , 4 )n ) in (8).

50

Hu Hu

|i n |2 |i |2 +

Hu Du,u

Hu

|i n |2 + +

Du,u

i (n , n )

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4

i (n , n )

i1 +i2 +i3 +i4 =i

i1 i2 i3 K i4 ,

where here, unlike in other places, we use to denote all possible curvature components, including n . Proof. Consider the following Bianchi equations: 1 + ) + ( + 4 ), 3 + tr = + 4 3( 2 4 + 2tr = div 2 + , Using Proposition 25, < n , 3 n > = = = < n , n > + < n , > < div n , n > + < n , > < 4 n , n > + < n , > + < n , >

Hu Hu

|n |2

Hu

|n |2 +

Hu

|n |2 +

Du,u

< (n , n ), >

We use the commutation formula, and note that the special structure is preserved in the highest order: 3 i i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 +

i1 +i2 +i3 +i4 +i5 =i1

i1 i2 i3 K i4 i5 +

i1 +i2 +i3 =i

i1 i2 i3 ( ),

4 i div i

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4 +

i1 +i2 +i3 +i4 +i5 =i1

i1 i2 i3 K i4 i5 +

i1 +i2 +i3 =i

i1 i2 i3 ( ),

51

Performing the integration by parts as before using Propositions 24 and 25, we have |i n |2 +

Hu Hu

|i n |2 |i n |2 +

Hu Du,u

Hu

|i n |2 + +

Du,u

i (n , n )

i1 +i2 +i3 +i4 =i

i1 i2 i3 i4

i (n , n )

i1 +i2 +i3 +i4 =i

i1 i2 i3 K i4 .

4.12.1. Before the Impulse. We rst apply Proposition 29 for the region 0 u us . Notice that the initial norm u

s

i 2

i 2 (sup ||i n ||2 L2 (Hu (0,us )) + sup || n ||L2 (H u ) ) C (O0 , R0 ), i2 u uus

independent of n. Proof. By Proposition 29 and all the estimates in the previous section, we have

i 2 (||i n ||2 L2 (Hu (0,us )) + sup || n ||L2 (H u ) ) i 2 uus u

0

i 2 (sup ||i n ||2 L2 (Hu (0,us )) + sup || n ||L2 (H u ) ) C (O0 , R0 ), i2 u uus

as claimed. 4.12.2. The Impulse Region. For an impulsive gravitational wave, the initial data represented by the component of curvature and its derivative are not in L2 . By the construction of Section 3, the approximating sequence n satises initially

us +2n i 2 us

2 ||i n ||2 L2 (S0,u ) du C 2 . n

In the following Proposition, we show that this bound is propagated. Proposition 31. (sup ||i n ||2 L2 (Hu (us ,us +2n )) +

i 2 u

n

sup

us uus +2n

52

Proof. The proof is the same as Proposition 30, except that the initial data for i can only n be bounded by 2 2 in L2 (H0 (us , us + 2n )). 4.12.3. After the Impulse. For us + 2n u u , since the initial data satisfy

u i 2 us +2n

independent of n, we again have a better estimate. An extra challenge arises from the fact that the estimates derived in Proposition 31 depends on n. This can be overcome if we allow a loss in derivative: Proposition 32. ( sup ||i n ||2 L2 (Hu (us +2n ,u )) +

i1 uu

sup

us +2n uu

independent of n. Proof. To bound the left hand side, we can apply the argument as in Proposition 30 to get ( sup ||i n ||2 L2 (Hu (us +2n ,u )) +

i 1 uu

n

sup

us +2n uu

The problem in proving Proposition 32 with a bound independent of n is that according to Proposition 31, ||i n ||L2 (H u ) may grow in n. However, we can still obtain the desired s bounds for n and n (note that the argument below will not apply to 2 n unless extra assumptions on the initial data are used) by using Proposition 17, which implies that sup ||i n ||L2 (Su,u ) C (O0 , R0 ).

i 1 u,u

i 1

n ) s +2

C (O0 , R0 ).

We can also prove estimates for higher derivatives away from us . The inuence of a propagating curvature impulse results in a loss of derivatives in such estimates compared to the assumed regularity of the initial data. By the remark after Proposition 28, we have estimates, uniform in n, for the 3 and derivatives for and . We now show that we can also have uniform in n estimates for the 4 derivatives for and for u us + 2n . By the Bianchi equations (8), this in turn can be reduced to showing estimates for 4 derivatives of . This is proved in the following Proposition: Proposition 33. For us + 2n u , if the initial data set satises

i ||k 4 n ||L2 (H 0 ) CI,K . kK iI {,,,}

sup

kK iI u {,,, }

53

then

i (sup ||k 4 n ||L2 (Hu (us +2n , )) + iI 1 kmin{K,

I i1 2

sup

us +2n u

where CI,K is independent of n. Proof. The main step is to prove that on H us +2n , we have the estimate

i ||k 4 n ||L2 (H u

n ) s +2

independent of n. For k = 0, we can commute with angular derivatives and integrate along the u direction using the Bianchi equation 3 + tr = + 2 + + 2 + 3( + ). Since the right hand side has one more angular derivative, this integration loses a derivative, allowing us only to prove ||i n ||L2 (H u +2n ) C

s

iI 1

Now we consider also the 4 derivatives. Dierentiate the equation by k 4 i and commute ] on the left hand side. Moreover, except for the component , whenever we see [3 , k 4 i k1 4 , we substitute an appropriate Bianchi equation from (8). Then we get an equation

i k1 i+2 3 k + ... 4 = 4

where ... denote terms that are lower order in terms of derivatives. We then integrate this along the u direction. Thus

i ||k 4 n ||L2 (H u

n ) s +2

n ) s +2

+ ...

i ||k 4 n ||L2 (H u

n ) s +2

C ||i+2k n ||L2 (H u

n ) s +2

+ ...

Now have reduced to the can with only angular derivatives. By the above, we need i+2k +1 I . Thus we can prove

i ||k 4 n ||L2 (H u iI 1 kmin{K,

I i1 2 n ) s +2

C.

(58)

The conclusion thus follows from a standard energy estimate type argument as in Proposition 29 and using (58) as the initial data on H us +2n Combining the estimates in Propositions 30, 31, 33 and Cauchy-Schwarz, we have the following uniform L1 u estimate. Proposition 34.

u

sup

uu iI 0

where CI is independent of n. Proposition 34 is crucial in showing that the limiting spacetime will have dened as a nite measure with a singular atom at us .

54

5. Convergence In this Section, we show that a sequence of initial data satisfying uniform estimates of Theorem 3 with converging initial data gives rise to a sequence of converging spacetimes. The convergence will be understood as follows: the spacetime will be identied in the system of double null coordinates (u, u, 1 , 2 ) and convergence will be established for the sequence of the corresponding spacetime metrics. In view of the quasilinear nature of the Einstein equations, we can only hope to prove convergence of our approximating spacetimes in a norm with one derivative fewer than the a priori estimates that we established for them. To get estimates of metric, we use the fact that the metric components satisfy inhomogeneous transport equations with right hand side expressed as Ricci coecients. The estimates for the dierence of Ricci coecients and curvature components are derived by considering the system of dierence equations obtained from the original system of transport, elliptic and Bianchi equations. As was the case for the a priori estimates, the challenge for the dierence system is a lack of any a priori control of the dierence of the singular curvature components . For the approximating sequence of initial data, n is not a Cauchy sequence in L2 (H0 ). In the proof of the a priori estimates, we handled the lack of information of the component of curvature via a renormalization procedure and the observation that the system satised by the Ricci coecients and the renormalized curvature components can be closed without any reference to the component of curvature. Potentially, this property may fail when we consider the dierence system. It is however a remarkable fact as we will show below that the dierence equations still possess the same structure. The following is the main Theorem in which we estimate the dierence of the metrics, Ricci coecients and curvature components of two spacetimes: Theorem 5. Suppose we have two sets of initial data (1) and (2) satisfying the conditions in Theorem 4 with the same constants C and c. By Theorem 4, we can solve for vacuum spacetimes (M(1) , g (1) ) and (M(2) , g (2) ) corresponding to the initial data sets (1) and (2) in the region 0 u u and 0 u u for u , u . Let (u, u, 1 , 2 ) be the coordinate system introduced in Section 2.2 such that the metrics take the form g (i) = 2((i) )2 (du du + du du) + ( (i) )AB (dA (b(i) )A du) (dB (b(i) )B du), where = 1 and bA = 0 on H0 and H 0 . We can now identify the two spacetimes by identifying points with the same value of coordinate functions. Dene g = g (1) g (2) , = (1) (2) and = (1) (2) to be the dierence of the metric, the dierence of the Ricci coecients and the dierence of the curvature components respectively. If the data satisfy sup |( )i AB (u = 0)| + sup |( )i AB (u = 0)| a, u u (

i 1 =,

u i1

u

i 1

u

sup || (tr, tr) ||L2 (Su,0 ) + sup ||2 (tr, tr) ||L2 (S0,u ) a,

u u

55

i 1 {,,, }

2 + ||i ||L2 u L (S0,u )

{,,,}

where the angular covariant derivative and all the norms are dened with respect to the spacetime (1). Then the following estimates hold in {0 u u } {0 u u }: sup |(

u,u

i ) | C a, AB

=, i1 u,u

i1 u

sup ||2 (, , , ) ||L2 (Hu ) + sup ||2 ( , , , ) ||L2 (H u ) + sup ||2 (tr, tr) ||L2 (Su,u ) C a,

u u u,u

i 1 {,,, }

2 + sup ||i ||L2 u L (Su,u )

{,,,} u

for some constant C depending only on C and c and independent of a. By iterating Theorem 5, we can reduce its proof to the region 0 u , 0 u for some suciently small independent of a. Most of this Section will be devoted to a proof of Theorem 5 for 0 u and 0 u . The proof of Theorem 5 will be carried out in Sections 5.1, 5.2, 5.3, 5.4 and 5.5. Theorem 5 implies the following convergence result: Theorem 6. Suppose we have a sequence of initial data that coincide on H 0 and satisfy the assumptions of Theorem 4 with uniform constants C and c on H0 and H 0 . By Theorem 4, for every initial data in the sequence, a unique smooth solution to the vacuum Einstein equations (Mn , gn ) exists in 0 u u , 0 u u for u , u and the metric takes the following form in the coordinate system (u, u, 1 , 2 ): gn = 2(n )2 (du du + du du) + (n )AB (dA (bn )A du) (dB (bn )B du), where = 1 and bA = 0 on H0 and H 0 . Denote the Ricci coecients and curvature components by n and n respectively. Identify the spacetimes in the sequence the value of the coordinate functions (u, u, 1 , 2 ). Dene also gn = gn gn1 , n = n n1 and n = n n1 . If sup |( )i (AB )n (u = 0)| an , u sup ||i n ||L2 (S0,u ) an ,

=, i1 u

i 1

u

56

i1 {,,, }

2 + ||i n ||L2 u L (S0,u )

{,,,}

for some an such that an < , then the spacetime metrics converge uniformly to a continuous limiting spacetime metric g g = 2( )2 (du du + du du) + ( )AB (dA (b )A du) (dB (b )B du) in the region 0 u u , 0 u u . Moreover, 4 ( gn , gn ) converge to ( g , g ) in L u Lu L (S ), u u 2 2 2 2 2 2 2 ( 2 gn , gn , 2 gn ) converge to ( 2 g , g , 2 g ) in L u Lu L (S ), u u u u 1 AB 1 AB p0 ) (n )AB )) converge to ( g , (( ) ( )AB )) in L ( gn , ((n u Lu L (S ), u u u u u u 2 2 2 2 p0 4 ( gn , gn ) converge to ( g , g ) in L u Lu L (S ). u uu u uu As a consequence, in the limiting spacetime, 0 0 4 g , g Cu Cu L (S ), u 2 2 2 0 0 2 g , g , g C u Cu L (S ), 2 u u2 1 AB g , (( ) ( )AB ) L u Lu L (S ), u u u 2 2 4 g , g L u Lu L (S ). u uu Remark 8. Notice that in the case of initial data of the impulsive gravitational wave, we 2 have i n converging in Lp u L (S ) for any 2 p < , but not for p = . We thus take a sequence of initial data that converges in a topology which is consistent with i n converging 2 in Lp L ( S ) . u The convergence of the Ricci coecients and the acceptable curvature components follows from Theorem 5. In Section 5.7, we will show that the convergence of the Ricci coecients and the acceptable curvature components imply the asserted convergence of the metric and the regularity property of the limiting spacetime. The above convergence theorem is strong enough to show that the limiting spacetime is a solution to the Einstein equations: Theorem 7. Suppose all the assumptions of Theorem 6 hold. Then the limiting spacetime 2 metric satises the Einstein equations in L u Lu L (S ). This will also be proved in Section 5.7. Moreover, this limiting spacetime solution is the unique solution to the vacuum Einstein equations. Theorem 8. The solution to the Einstein equations given by Theorems 6 and 7 is unique among spacetimes that arise as C 0 limit of smooth solutions to the vacuum Einstein equations.

57

A more precise version of the uniqueness theorem is formulated as Proposition 73 and will be proved in Section 5.8. Moreover, if the initial data is assumed to be more regular, the limit spacetime metric is more regular: Theorem 9. Suppose, in addition to the assumptions of Theorem 6, the bounds (

j J iI {,,,} i ||j 3 n ||L2 (H 0 ) + {,,, } j ||i 3 n ||L2 (H0 ) ) C

j J +2 imin{I,j 2}

j i 2 ) ( ) g L u Lu L (S ), u

(

j J +2 imin{I,j 2}

j i p0 2 )( ) g L u Lu L (S ). u u

This will be proved in Section 5.9. For general initial data satisfying the assumptions of Theorem 3, using the construction in Section 3, there exists an approximating sequence of smooth initial data satisfying the assumption of Theorem 6. Thus, the combination of Theorems 4, 6, 7, 9 and 8 together imply Theorem 3. The remainder of this Section will be organized as follows: Theorem 5 is proved in Sections 5.1-5.5. After the denition of the norms in Section 5.1, the proof is carried out in three steps: STEP 1 (Section 5.2): The dierence of the metric components are estimated assuming the bounds for the dierence of the Ricci coecients. STEP 2 (Section 5.4): The dierence of the Ricci coecients are controlled assuming the estimates of the dierence of the curvature components. This relies on the transport equations for the dierence quantities derived in Section 5.3. STEP 3 (Section 5.5): Finally, the bounds for the dierence of curvature components are obtained, closing all the estimates for Theorem 5. In Section 5.6, additional estimates are derived for the metric components. These additional estimates will be used together with Theorem 5 to construct a limiting spacetime and to obtain Theorem 6 in Section 5.7. In this Section, Theorem 7 is also proved, showing that the limiting spacetime satises the Einstein equations. In Section 5.8, we formulate and prove a precise version of Theorem 8, establishing uniqueness of the limiting spacetime. In Section 5.9, Theorem 9 is proved, showing additional regularity in the spacetime with more regular initial data. Finally, in Section 5.10, we return to the case of an impulsive gravitational wave and obtain extra regularity properties for these spacetimes. 5.1. Norms. We begin the proof of Theorem 5. Dene the following L2 norms for the dierence of the null curvature components or their renormalized versions: R =

i1

u i 1

u

58

Dene the following norm for dierence of the the Ricci coecients: O =

i1

u i 1

u,u

These norms will be used to estimate the dierence of the null Ricci coecients except for those involving the highest derivatives, for which the estimates are weaker. We therefore also introduce the norms: = sup ||2 (, O , , ) ||L2 (H ) + sup ||2 ( , , , ) ||L2 (H ) + sup ||2 (tr, tr) ||L2 (S ) ,

u

u

u,u

u,u

are dierence counterparts of the norms R, O and O Notice that the norms R , O and O for the dierence quantities. Note, however, that the former provides control of one fewer derivatives than the latter. This is due to the fact that convergence will be proved in a norm which is one derivative weaker than the corresponding norms for the a priori estimates. Dene also the following norms:

4 O = sup(||(, , , , tr, tr) ||L2 + ||(, , , , tr, tr) ||L2 ). u L (Su,u ) u L (Su,u )

It follows by Sobolev Embedding in Proposition 9 and 10 that Proposition 35. + O ). O C (O 5.2. Estimates for the Dierence of the Metrics. In this section, we show that the dierence of the metrics and their coordinate angular derivatives can be controlled by the norms. Recall that the metrics take the form O and O g (i) = 2((i) )2 (du du + du du) + ( (i) )AB (dA (b(i) )A du) (dB (b(i) )B du), for i = 1, 2. Proposition 36. sup |( , (1 ) )(u, u)| C 2 O .

u,u

1

Proof. Recall that 1 1 1 1 = 4 log = 4 1 = . 2 2 2 u 1 (1 ) = . 2 u By integrating along the u direction, noticing that = 0 on H 0 , and using Cauchy-Schwarz we get 1 |(1 ) | C 2 O . In order to get the estimate for , we note that ( + (2) )1 = ((1) )1 = (1 )(2) + (1 ) . Therefore, = (((2) )1 + ()1 ) )1 (2) = (1 ) 1+

(1 ) (1 )(2)

Hence

59

1

Using the estimates for , we also have estimates for . Proposition 37. satises the following pointwise bounds: sup |(AB , (( 1 )AB ) )(u, u)| a + C 2 Ou .

u,u

1

Proof. The components of solve the following ODE: AB = 2AB . u This implies log(det ) = tr. u From this we can derive an equation for (det ) : 1 (det ) = u (det )(2) (det ) (det )(1) + (det )(1) (det )(2) (tr) u .

By Proposition 3, we have uniform upper and lower bounds on det and uniform estimates 1 for and u (det ). The previous Proposition gives | | C 2 O . Moreover, |tr |du O by denition. Thus, 1 (59) |(det ) | C 2 O . We can also derive an equation for (AB ) : (AB ) = 2(AB ) . u Thus |(AB ) (u, u)| |(AB ) (u, 0)| + C

0 u

By assumption, |(AB ) (u, 0)| a. By Proposition 3, matrices (1) and (2) are uniformly non-degenerate matrices. Hence, regardless of whether we dene the L norm with respect to (1) or (2) , we have |( AB ) | C sup || ||L (Su,u ) .

A,B =1,2 u,u

Therefore, |(AB ) | a + C 2 O . Now, (59) and (60) together also imply the pointwise bound in coordinates for |(( 1 )AB ) | a + C 2 O . This estimate allows us to conclude that the Lp norms dened with respect to either metric 1 (1) or (2) dier only by a + C 2 O .

1 1

(60)

60

(1) (2)

Proposition 38. Given any tensor, we can dene its Lp (S ) norms ||||Lp (Su,u ) and ||||Lp (Su,u ) with respect to the rst and second metric. Suppose ||||Lp (Su,u ) < . Then

(1) (2) (1)

|||||Lp (Su,u ) ||||Lp (Su,u ) | C (a + 2 O )||||Lp (Su,u ) . In particular, is also in Lp with respect to the second metric. Proof. This follows from the pointwise control of (AB ) , (( 1 )AB ) and (det ) in coordinates.

In a similar manner, we can control the ( A ) derivatives of .

(1)

Proposition 39.

1 )AB ||L4 (Su,u ) Ca + C 2 O , C u,u 4 where L (S ) is understood as the L4 norm for a scalar function and by Proposition 38 can be dened with respect either to the metric (1) or (2) .

sup ||(

Proof. By the equation AB = 2AB u and the assumption, we have ||( C )i AB (u, u)||L4 (Su,u ) Ca + C Since we have ) | C |(, ) | + C | ||| + C ||| | C |(, ) | + C | D AB | + C | |, C where is the connection coecients on the spheres with respect to . Moreover, by CauchySchwarz, 1 ||( C ) ||L4 (Su,u ) du C 2 O . Thus by (61), we have ||( C )AB (u, u)||L4 (Su,u ) u 1 Ca + C 2 O + C || , , , D AB , ||L4 (Su,u ) du . 0 Using already established estimates, Gronwalls inequality and Cauchy-Schwarz, we thus have 1 ||( C )AB (u, u)||L4 (Su,u ) Ca + C 2 O . |(

u

(

0

1 (log ) = ( + ), 2

61

A consequence of the above Proposition is the estimates on the dierence of the connection coecients: Proposition 40. sup || ||L4 (Su,u ) Ca + C 2 O .

u,u

1

Proof. This follows from the fact that can be expressed as a linear combination and its rst angular derivative. This implies that (1) and (2) are comparable in the following sense: Proposition 41. Let p 4. Suppose is a tensor. Then ||((1) ) ((2) )||Lp (Su,u ) C (a + 2 O )||||L (Su,u ) . Proof. Since we have estimates for AB and ( 1 )AB . It suce to estimate ((1) ) Using Proposition 40, we have ||((1) )

A A 1

((2) )

((2) )

||Lp (Su,u )

||(1) (2) ||Lp (Su,u ) C || ||L4 (Su,u ) ||||L (Su,u ) C (a + 2 O )||||L (Su,u ) . We now show that the dierence between the angular covariant derivatives of two tensors is comparable to the angular covariant derivative of the dierence: Proposition 42. Let p 4. Suppose (1) and (2) are tensors dened on spacetimes (1) and (2) respectively. Then ||(1) ( ) () ||Lp (Su,u ) C (a + 2 O )||||L (Su,u ) . Proof. It is easy to see that (1) ( ) () = (1) (2) + (2) (2) . The Proposition thus follows from Proposition 41. We now estimate the coordinate angular derivative of in L4 (S ): Proposition 43. We have sup ||

u,u

1 2O . 4 (S || Ca + C L ) u,u A 1 1

62

1 4 Thus, in order to estimate A , we need to estimate A in Lu L (S ). Since by the a priori estimates in the previous section we have || ||L (S ) C , it suces by Proposition 42 to estimate u i1 0

u i1 0

The conclusion thus follows. The above Proposition allows us to estimate the second coordinate angular derivatives of in L2 (S ). Proposition 44. sup ||(

u,u

1 2 ), )AB ||L2 (Su,u ) Ca + C 2 (O + O C D

where L2 (S ) is understood as the L2 norms for scalar functions. Proof. By the equation AB = 2AB , u and the a priori estimates in the previous section, we have 2 ||( C D )AB (u, u)||L2 (Su,u ) u 2 2 Ca + C ( C D ) ||L2 (Su,u ) + ( C D ) ||L2 (Su,u ) du 0 u +C ( C ) ||L2 (Su,u ) + ( C ) ||L2 (Su,u ) du . 0 The last term has already been estimated in the proof of Proposition 39:

u

(62)

(

0

1

Ca + C 2 O . Since 1 (log ) = ( + ), 2

63

we have 2 ) ||L2 (Su,u ) C D C || (, ) ||L2 (Su,u ) + C || ||L2 (Su,u ) ||||L (Su,u ) + C || ||L2 (Su,u ) || |||L (Su,u ) + C || ||L4 (Su,u ) || ||L4 (Su,u ) + C ||||L (Su,u ) || ||L2 (Su,u ) C ||i (, ) ||L2 (Su,u ) + C ||( )i ||L2 (Su,u ) + C ||( )i ||L2 (Su,u ) . i 1 i1 i1 ||( , we have Using Propositions 43 and the denition of the norms O and O

u i 1 0

u

1 )+C C 2 (O + O u

||(

0

1

2 ). ) ||L2 (Su,u ) du C (O + O C D

)+C Ca + C 2 (O + O

0

||(

1 2 ). )AB ||L2 (Su,u ) Ca + C 2 (O + O C D

The proof of the previous Proposition also shows that Proposition 45. sup ||(

u,u

1 2 ), ) ||L2 (Su,u ) Ca + C 2 (O + O C D

Proposition 44 implies the following estimates of K , the dierence of the Gauss curvatures: Proposition 46. K satises the following bounds:

1 ). sup ||K ||L2 (Su,u ) Ca + C 2 (O + O

u,u

Proof. This follows from the estimates of the derivatives of in Propositions 39 and 44. Moreover, Proposition 44 allows us to conclude that the covariant second derivative with respect to connection (1) and connection (2) are comparable:

64

Proposition 47. Let p 2. Suppose is a tensor. Then ||((1) )2 ((2) )2 ||Lp (Su,u ) C (a + 2 O )

i1

1

Proof. By Propositions 40, 44 and 46. This implies that taking the covariant second derivatives of the dierence is comparable to taking the dierence of the covariant second derivatives: Proposition 48. Let p 2. Suppose (1) and (2) are tensors dened on spacetimes (1) and (2) respectively. Then ||((1) )2 ( ) (2 ) ||Lp (Su,u ) C (a + 2 O )

i1

1

Proof. The proof follows as that of Proposition 42, using Proposition 47 instead of Proposition 41. We now prove the estimates for b : Proposition 49. sup ||(bA ) ||L (Su,u ) , sup ||

u,u u,u

The Proposition thus follows from dierentiating the equation by , integrating in u and using the the bounds in Propositions 36, 43 and 45 and the denitions of the norms.

5.3. Transport Equations for the Dierence Quantities. In this Subsection, we show how to derive estimates for the dierence of the quantities (1) and (2) satisfying transport (1) (2) (1) (2) equations in the 4 , 4 or 3 , 3 directions. This should be compared with the previous Subsection where we considered dierence quantities satisfying transport equations, but unlike in this Subsection, the corresponding transport equations held with respect to the , derivatives rather than 4 , 3 , which themselves depend on the spacetimes under u u consideration. In order to estimate the quantities associated to the dierence between that of dierent spacetimes, we need to derive a transport equation for from the transport equation for . Proposition 50. Consider spacetimes satisfy the hypothesis of Theorem 4 and 5. Let be a (0, r) S-tensor. Suppose 4 = F . Then 4 F + Similarly, suppose 3 = G. Then 3 G +

(1) (1)

(1 ) 4 + 1 ( 1 ) . 1

(1 ) 3 + 1 (bA ) A + 1 ( 1 ) + 1 b . 1

65

i denotes that the original Ai in the i-th slot of the tensor is removed. This equation where A holds in both spacetimes (1) and (2). Then, we derive the equations for . We will write schematically without the exact constant depending only on r: (4 )(1) ( 1 )(1) (1 )(1) (1) (1) (2) u ( 1 )(1) (1 )(1) (1) (1) (1 )(2) ( 1 )(2) (1 )(2) (2) (2) = (1 )(1) u u (2) (1 ) + ( 1 1 ) (2) u (2) + ( 1 1 ) (2) =F (1 ) u (1 ) =F 1 (1) (4 )(1) (2) + (1 )(2) ( 1 ) (2) . ( ) = (1 )(1) For the 3 equations, we write in coordinates 3 A1 ...Ar =1 A1 ...Ar + 1 bB B A1 ...Ar u 1 B C 1 CD 1 ( ) CD A1 ...A i C...Ar . i C...Ar + b BAi A1 ...A

We derive the dierence equations as before. However, e3 in the respective spacetimes are not parallel to each other and we therefore have extra terms. (3 )(1) + (1 )(1) b(1) ( 1 )(1) (1 )(1) (1) + (1 )(1) b(1) (1) (1) (2) u = (1 )(1) + (1 )(1) b(1) ( 1 )(1) (1 )(1) (1) + (1 )(1) b(1) (1) (1) u (1 )(2) + (1 )(2) b(2) ( 1 )(2) (1 )(2) (2) + (1 )(2) b(2) (2) (2) u (2) (1 ) + ( 1 1 ) (2) (1 b) (2) (1 b) (2) u 1 ( ) (1) =G 1 (1) 3 (1) (1 )(2) b (2) + (1 )(2) ( 1 ) (2) (1 )(2) (b) (2) . ( ) = (1 )(1)

This, together with the commutation estimates in Proposition 12, gives the following estimates for (i )n .

66

i1 +i2 +i3 =i

(i1 (, )i2 i3 F ) +

i1 +i2 +i3 +i4 =i

i1 +i2 +i3 +i4 =i

+ (1 )

i1 +i2 +i3 =i

i1 (, )i2 i3 F +

i

+ ( 3 (i )

i1 +i2 +i3 +i4 =i

i1 +i2 +i3 =i

+ (1 )

i1 +i2 +i3 =i

i1 (, )i2 i3 G +

i1 +i2 +i3 +i4 =i i

i1 (, )i2 i3 i4

+b

i+1

+ ((

) + b ) .

1 2

Proof. According to the already established estimates, we have follows directly from Propositions 12 and 50.

We now use Propositions 7 and 51 to obtain estimates from a transport equation. Proposition 52. Suppose 4 = F . Then ||i ||L2 (Su,u )

i 1 u

i 1

0

1 2

i 1

+ O )( + C (O + O

i1

u u

i 1 u

i 1

0

1 2

i 1

+ O )( + C (O + O

i1

u u

Proof. We use the pointwise estimates for , and b proved in the last section. By Propositions 42, it suces to estimate ||(i ) ||L2 (Su,u )

i 1

since the dierence can be estimated by )(||||L4 (S ) + ||||L (Su,u ) ). ||(i ) ( )||L2 (Su,u ) C (O + O u,u

i 1

67

We can estimate ||(i ) ||L2 (Su,u ) using Proposition 7 and the equation in Proposition 51. Recall the formula in Proposition 51 4 (i )

i1 +i2 +i3 =i

(i1 (, )i2 i3 F ) +

i1 +i2 +i3 +i4 =i

i1 +i2 +i3 +i4 =i

+ (1 )

i1 +i2 +i3 =i

i1 (, )i2 i3 F +

i

+ (

) .

u

0

i1 +i2 +i3 1 u

C

0

(

i1 1

1 ||(, )||i L (S

u,u

)) i2 1

C

i1

u 0

i1 +i2 +i3 1 u

C

0

(

i1 1 i2 2 u

u,u

) i3 1

+C

0

(

i1 1

i2 1 i3 1

3 ||i1 (, , tr, )||i L (S

u,u

|| ) )( i4 1

i4

C

i1

The third term is easier to estimate since by Proposition 36, (1 ) can be estimated in L :

u 0

1 2

i1 +i2 +i3 1

)( C (O + O

i 1

u u

i 1

u u

The fourth term can be controlled in the same way as the third term, since ( 1 ) can be estimated in L2 uL :

u 0

1 2

i1

)( C (O + O

i 1

u u

68

Putting all these estimates together using Proposition 7, we have ||i ||L2 (Su,u )

i1

i1

i1

1 2

+ O )( + C (O + O

i 2

u u

Take to be suciently small depending only on C but not a, we can absorb the third term to the left hand side to get ||i ||L2 (Su,u )

i 1

i 1

i 1

1 2

+ O )( + C (O + O

i2

u u

5.4. Estimates for the Ricci Coecient Dierence. The results of the previous Subsection can now be used to estimate the norm O in terms of R . In this Subsection and the subsequent Section 5.5, in order to simplify notation, we will omit the superindices (1) and (2) in all the background quantities, for example, 3 , , , ... Proposition 53. There exists and C depending only on the a priori estimates in Theorem 4 (but independent of a) such that

1 1 . O Ca + C 2 R + C 2 O

Proof. We control O using the null structure equations. First, we consider the Ricci coecients tr, , . They all satisfy 3 equations of the form 3 = + such that , do not appear in the terms on the right hand side; and that does not appear as . This is important because and cannot be controlled by C O in the L2 (S ) norms and cannot be controlled by C R in L2 (H ). In order to estimate using this equation, we need to estimate the curvature term: ) + C 2 R . 2 ||(i ) ||L1 C 2 (O + O u L (S )

i1

1 1

69

2 ||(i ( )) ||L1 u L (Su,u )

i 1

1 ) + C 1 2 C 2 (O + O

i1 +i2 1

u u

+ C 2

i1 + i2 1

1 2

u u

i 1 u

i 1

(63)

1 + O ) + C 1 2R . a + C 2 (O + O

We now consider the equation for , which schematically looks like 3 = + + + , where = , are the good components as above. The term or can be estimated since can be put in an appropriate Lp (S ) norm and be controlled by C O . However, the p most dicult terms are and since and cannot be bounded in L1 u L (S ) by C O . Thus, by Proposition 52, we have ||i ||L2 (Su,u )

i1

C

i 1

i1 +i2 1

2 ||i ||L2 u L (Su,u )

i 1 u

C

i1

0

1 2

+ O ), a + C R + C (O + O (64) since we can control i by C O after integrating along the u direction. We can estimate in a similar manner as . satises 3 = + + ,

70

norm and using already obtained estimates, we have where = , . Putting in the O ||i ||L2 (Su,u )

i 1

C

i1

i1 +i2 1

2 ||i ||L2 u L (Su,u )

i1 u

i1

0

1 2

(65)

+ O ). a + C (O + O It remains to consider tr, , . They satisfy 4 equations of the form: 4 = + , where can be any Ricci coecients and can be any curvature components = , , i.e., all the terms can be controlled in L2 (H ) by R . Moreover, since we are now integrating in u, all terms can be estimated by C O . By Proposition 52, we have

u

i 1 i1

0

O du .

Therefore,

1 1 + O ). ||i (tr, , ) ||L2 (Su,u ) a + C 2 R + C 2 (sup O + O

i1

1 1 + O ). O Ca + C 2 R + C 2 (O 1

1 1 . O Ca + C 2 R + C 2 O

Notice that in particular and C depend only on the a priori estimates from Theorem 4 and are independent of a. by R . Recall from Propositions 19 and 20 that O was We now move on to estimate O controlled using a combination of transport equations for and Hodge systems. The norm can be dealt with in a similar fashion. In order to perform this scheme, we need show O where we controlled from , the dierence that as was with the case of the norm O satises elliptic equations with as a source. This is given by the following Proposition: Proposition 54. Let be a (0, r)-tensoreld. Suppose div = f, on each of the spacetimes. Then div f + ( 1 ) + , curl = g, tr = h

71

curl g + (/ ) , tr h + ( 1 ) . Proof. This is a straightforward coordinate computation. Based on the estimates we have on , ( 1 ) and , and using the estimates in Proposition 14, we have Proposition 55. Let (1) and (2) be totally symmetric r + 1 covariant tensorelds on 2spheres (S2 , (1) ), (S2 , (2) ) respectively satisfying div (i) = f (i) , for i = 1, 2. Then ||2 ||L2 (S ) Ca

i1

tr(i) = h(i) ,

(||i f ||L2 (S ) + ||i g ||L2 (S ) + ||i h||L2 (S ) + ||||L2 (S ) ) (||i f ||L2 (S ) + ||i g ||L2 (S ) + ||i h ||L2 (S ) + || ||L2 (S ) )

i1

+C

Proof. This is a direct consequence of Propositions 54 and 14. . First, we dene and derive estimates for using We can now begin to estimate O transport equations. Let denote (tr) , (tr) , , , , . We have the following estimates: Proposition 56.

1 + R ). ||i (tr, tr, , ) ||L2 (Su,u ) Ca + C 2 (O

i1

Proof. From the proof of Proposition 20, we know that each of tr, tr, , satises either 3 = H + 3 + H , or 4 = H + 3 + H , where as in Section 4.8, we use the notation H {tr, , , , , tr}, H {tr, , , , tr, }, H {, , , } and H {, , , }. We focus on those s satisfying the 4 equation. The other case can be treated analogously. By Proposition 52, we need to control

2 ||(i ( H + 3 + H )) ||L1 , u L (Su,u )

i1

i1

i1

We rst estimate the terms. Among those we rst look at the term with two derivatives on the Ricci coecient. We have, using Proposition 53, 1 1 + C || ||L1 L (S ) + C 2 +2 ||( 2 H ) ||L1 L2 (S ) C O ||2 ||L2 L2 (S ) C (a + O R ).

u u,u u u,u

u,u

72

The other term with Ricci coecients can be estimated by O and thus by Proposition 53:

1 1 ). 2 ||( i1 i2 ) ||L1 C (a + 2 R + 2 O u L (Su,u )

i1 +i2 1

2 ||( i1 i2 i3 H ) ||L1 u L (Su,u )

i1 +i2 +i3 1

1 1 ) + C 1 2 C (a + 2 R + 2 O 2 ||i H ||L2 u L (Su,u )

i1

). C (a + R + O The terms without can be estimated using the a priori estimates in Theorem 4: (||i ( H + 3 + H )||L2 (Su,u ) + ||i ||L2 (Su,u ) ) C.

i1

1 2

1 2

1 + R ). ||i (tr, tr, , ) ||L2 (Su,u ) Ca + C 2 (O

i 1

We now consider the dierence quantities and . As in Proposition 22, they will only satisfy L2 (Hu ) and L2 (H u ) estimates. As in the proof of Proposition 22, in what follows, we allow also to be and . Proposition 57. + R ), ||i ||L2 (Hu ) C 2 (a + O

i 1

1 + R ). ||i ||L2 (H u ) C 2 (a + O 1

i 1

Proof. Recall equations (49) and (50): 3 = + 3 + , or 4 = + 3 + . We focus on . As is easier. The only terms that are new compared to the proof of Proposition 56

2 ||i ( ) ||L1 and u L (Su,u ) 2 ||i (, ) ||L1 . u L (Su,u )

i 1

i 2

Thus using the estimates in the proof of Proposition 56, we have +R)+C 2 ( ||i ||L2 (Su,u ) Ca+C 2 (O

i1 i1

1 1 2 ||i ||L2 + u L (Su,u ) 2 ||i (, ) ||L2 ). u L (Su,u )

i2

73

i1

1 + R ) + C 1 2( C 2 a + C (O 2 2 ||i ||L2 + u Lu L (Su,u ) 2 2 ||i (, ) ||L2 ). u Lu L (Su,u )

i 1

i2

1 + R ). ||i ||L2 (Hu ) C 2 a + C (O

i1

Similarly

1 + R ). ||i ||L2 (Hu ) C 2 a + C (O

i1

We use Propositions 56 and 57 and elliptic estimates to derive all the estimates for 2 . Proposition 58. Ca + C R . O Proof. We will rst prove

1 + CR . sup ||2 (, , , ) ||L2 (Hu ) + sup ||2 ( , , , ) ||L2 (H u ) Ca + C 2 O

We have the following div-curl systems: div (, , , ) = (tr, , ) + + (, , ), curl (, , , ) = (tr, , ) + + (, , ), and div ( , , , ) = (tr, , ) +

=,

+ ( , , ), + ( , , ).

=,

curl ( , , , ) = (tr, , ) +

u i 1

u i 1

u i1

u

and the lower order terms involving the Ricci coecients ||i ( )||L2 (Hu ) +

i1 i1 =,

||i ( )||L2 (H u ) .

1 + R ). C 2 (a + O

74

The curvature terms can be estimated using the denition of R by CR . The remaining terms can be bounded by ||i ( )||L2 (Hu )

i1

i 1

1 2

i 1

1 2

u u

1

Therefore, together with the estimates for 2 tr and 2 tr in Proposition 56, we have

1 Ca + C 2 + CR . O O

By choosing suciently small depending on C so that C 2 , we have Ca + C R . O Notice that C and depend only on the a priori estimates in Theorem 4 and are both independent of a. 5.5. Estimates for the Curvature Dierence. In order to nish the proof of Theorem 5, we need to estimate R by Ca. This will be proved using an energy-type estimate. Proposition 59.

i1 {,,, } Hu

(i )2 +

{ , ,,} Hu

(i )2

i1 +i2 +i3 1

||( i1 i2 i3 ) ||2 + 2 2 L2 u Lu L (S )

i1 +i2 +i3 2

||( i1 i2 i3 ) ||2 . 2 2 L2 u Lu L (S )

Proof. The energy estimates for the curvature dierence follows from the dierence Bianchi equations. Once these are derived, the energy estimates will be obtained as in Section 4 by integration by parts in appropriate subsets of the dierence Bianchi system. In the proof below, we concentrate on the most dicult case which involves renormalized curvature components. The estimates for the other components can be derived in a similar fashion. Recall that 3 + tr = + 2 + + 2 + 3( + ), 3 1 4 + tr = div + 2 , 2 2 3 1 4 + tr = div + + 2 , 2 2

75

As before, we renormalize the equations. Dene 1 := + , 2 1 . := 2 Schematically, we have 3 = + + + + 3 , 4 = div + + + 3 , 4 = div + + + 3 . From these we can derive equations for , and . By Proposition 50, (1 ) 3 + 1 (bA ) A + 1 ( 1 ) + 1 b , 1 (1 ) 4 = (4 ) + 4 + 1 ( 1 ) , 1 (1 ) 4 = (4 ) + 4 + 1 ( 1 ) . 1 3 = (3 ) + Moreover, we can write schematically (( , )) = ( , ) + ( , ) (div ) = div + + . Therefore, 3 = + + ( + + 3 ) + F , 4 = div + ( + + 3 ) + F , 4 = div + ( + + 3 ) + F , where F , F and F satisfy the bound

2 2 ||i F ||L2 C 2 (a + R ). u Lu L (S ) 1

i 1

Therefore, multiplying the rst equation by , integrating by parts and using the other equations, we have | |2 +

Hu 2 Hu

(| |2 + | |2 )

1 2 1 2 Ca + ||( , , )( + + 3 ) ||L1 + ||( , , )F, , ||L1 u Lu L (S ) u Lu L (S ) 2 2 Ca2 + ||( , , )||L2 + ||( + + 3 ) ||2 + +||F ||2 2 2 2 2 L2 L2 u Lu L (S ) u Lu L (S ) u Lu L (S )

76

We now look at the equation the rst covariant angular derivative for the curvature components: 3 = 2 + 2 + K ( , ) + 2 + + + 2 + 2 + 4 , 4 = div + K + 2 + + + 2 + 2 + 4 , 4 = div + K + 2 + + + 2 + 2 + 4 . A similar argument as before involving writing the equations for ( , , ) and integrating by parts yield | |2 +

Hu Hu

(| |2 + | |2 )

From this one can conclude Proposition 60. R Ca. Proof. By the previous Proposition, we need to estimate ||(K ) ||2 , 2 2 L2 u Lu L (S ) , ||( i1 i2 i3 ) ||2 2 2 L2 u Lu L (S )

i1 +i2 +i3 1

||( i1 i2 i3 ) ||2 2 2 L2 u Lu L (S )

i1 +i2 +i3 2

By Proposition 46 and the denition of the norm R and the a priori estimates from Theorem 4, we have ||K ||2 + ||K ||2 C 2 (R )2 . ||(K ) ||2 2 2 2 2 2 2 L2 L2 L2 u Lu L (S ) u Lu L (S ) u Lu L (S ) To estimate the remaining terms, we will repeatedly invoke Proposition 42 to exchange () and ( ) (or ( ) and ( ) etc.), at the expense of an error term 2 (R )2 . ||( i1 i2 i3 ) ||2 2 2 L2 u Lu L (S )

i1 +i2 +i3 1

C 2 (R )2 + C (

i1 1 i2 2

2 ||i1 ||i )2 ( L u Lu L (S )

i3 1

+ C(

i1 1 2

||

i1

C (a + (R )2 ),

77

where the second term is estimated by Theorem 4 and the denition of R ; and the last term is estimated by Theorem 4 and Proposition 53. Finally, we estimate using Theorem 4 and Proposition 53 ||( i1 i2 i3 ) ||2 2 2 L2 u Lu L (S )

i1 +i2 +i3 1

C 2 (R )2 + C (

i1 1 i2 2

2 )2 ( ||i1 ||i L u Lu L (S )

||i3 ||2 ) 2 2 L2 u Lu L (S )

i3 2

C (a + (R ) ), Putting these together, we have (R )2 Ca2 + C (R )2 . The conclusion follows by choosing suciently small depending only on C . This implies all the quantities that we have estimated in this section can be estimated by Ca: Proposition 61. 2 , R Ca. 2 4 g ||L g ||L , || ,O ,O u Lu L (S ) u Lu L (S ) 2 Proof. This is a direct consequence of Propositions 36, 37, 39, 43, 44, 45, 49, 53, 56, 58 and 60.

||g ||L , || u Lu L (S )

This concludes the proof of Theorem 5. 5.6. Additional Estimates on the Dierence of the Metrics. In this Subsection, we derive some additional estimates on the dierence of the metrics. They follow directly from the bounds already derived. This will be used in Section 5.7 to show that the limiting spacetime metric is in the space asserted in Theorem 6 and satises the Einstein equations. Proposition 62. ( , , b )||L4 (Su,u ) Ca. u Proof. satises a transport equation in the L direction: || L / L = 2. From this we can derive an equation for : = bA A (bA ) A + 2() . u Estimating the right hand side using Proposition 61, we have ||L4 (Su,u ) Ca. u also satises a transport equation in the L direction: || 1 1 1 = 4 log = 4 1 = ( + bA A )1 . 2 2 2 u

78

(1 ) . By direct estimates using Proposition 61, As for , we can derive an equation for u we have || ||L4 (Su,u ) Ca. u Finally, we move to u b . b does not satisfy a transport equation. We therefore resort to the equation bA = 42 A . u Applying u , we get 2 bA = 4 (2 A ). uu u = 0 on H 0 . Thus Since (b )A = 0 on H 0 , b u u b ||L4 (Su,u ) C || (2 A ) ||L4 (Su,u ) du . u u 0 By Proposition 61, the right hand side Ca.

||

Proof. We prove this Proposition by taking an extra u derivative of the equations in the proof of Proposition 62. By L / L = 2, we have 2 A = 2 () (b ). 2 u u u A Notice that = 3 . u Thus using the null structure equation for 3 , and estimating directly, we get

|| To estimate

2 , u2

we use

1 = ( + bA A )1 . 2 u A Dierentiating with respect to u , using the estimates for u b and the null structure equation for 3 , we get 2 || 2 ||L2 (Su,u ) Ca. u Finally, dierentiating bA = 42 A u twice in the u direction, we get 3 bA 2 = 4 (2 A ). uu2 u2

79

Using the null structure equations as well as the Bianchi equation for 3 , we see that all the terms on the right hand side can be bounded in L2 (Su,u ). Thus we can integrate to get || 2 b ||L2 (Su,u ) Ca. u2

It is also possible to prove estimates for the mixed second derivatives Proposition 64. || 2 ( , , b )||L2 (Su,u ) Ca. uA

A

Proof. This can be proved in a similar way as Proposition 62, taking derivative.

instead of

On the other hand, u derivatives can only be taken once and can only be estimated after 0 taking the Lp u norm: Proposition 65. || ( , , b )||Lp Ca. 0 u L (Su,u ) u

Proof. We can directly estimate the right hand side of the equations = 2() , u 1 (1 ) = , 2 u A (b ) = 4(2 A ) . u The mixed u and derivatives can also be estimates: Proposition 66. || Proof. Take

A

A (b ) = 4(2 A ) , u and apply the estimates in Proposition 61. Similarly for the mixed u and u derivatives:

80

u

A (b ) = 4(2 A ) , u and use the null structure equations. We do not have estimates for the second u derivatives of the metric, since they can only be understood as measures even for the initial data. However, the following substitute is useful in showing that the limit spacetime that we construct satises the Einstein equations, in particular, Ruu = 0. Proposition 68. || Ca. (( 1 )AB AB ) ||Lp 0 u L (Su,u ) u u AB . u

tr u

1 |2 2 tr, 4 tr + (tr)2 = | 2 we have tr = ( (tr)2 | |2 2 tr) . u 2 The conclusion follows from estimating the right hand side directly using Proposition 61. 5.7. The Limiting Spacetime Metric. In this Subsection, we prove Theorems 6 and 7. We therefore assume that the conditions of Theorem 6 hold. Consider the sequence of metric gn = 2(n )2 (du du + du du) + (n )AB (dA (bn )A du) (dB (bn )B du). By Proposition 61, n , bn and n converge uniformly to their limiting values , b and and are therefore continuous functions of (u, u, 1 , 2 ) and dene a continuous limiting spacetime metric g = 2( )2 (du du + du du) + ( )AB (dA (b )A du) (dB (b )B du). (66)

To complete the proof of Theorem 6, we need to demonstrate the desired regularity of the limiting spacetime. This follows easily from the estimates in the previous Subsections.

81

Proposition 69. 4 gn , gn ) converge to ( g , g ) in L u Lu L (S ), u u 2 2 2 2 2 2 2 gn , 2 gn ) converge to ( 2 g , g , 2 g ) in L ( 2 gn , u Lu L (S ), u u u u p0 1 AB 1 AB ) ) (n )AB )) converge to ( g , (( ( )AB )) in L ( gn , ((n u Lu L (S ), u u u u u u 2 2 2 2 p0 4 ( gn , gn ) converge to ( g , g ) in L u Lu L (S ). u uu u uu ( Proof. That 4 gn , gn converge in L u Lu L (S ) u follows from Propositions 61 and 62 respectively. That 2 2 2 2 gn , 2 gn , gn converge in L u Lu L (S ) 2 u u follows from Propositions 61, 63 and 64 respectively. That 1 AB p0 gn , ((n ) (n )AB ) converge in L u Lu L (S ) u u u follows from Propositions 65 and 68 respectively. That 2 2 p0 4 gn , gn converge in L u Lu L (S ) u uu follows from Propositions 64 and 5.6 respectively. Proposition 70. 0 0 4 g , g Cu Cu L (S ), u 2 2 2 0 0 2 g , g , g C u Cu L (S ), 2 u u2 1 AB g , (( ) ( )AB ) L u Lu L (S ), u u u 2 2 4 g , g L u Lu L (S ). u uu Proof. To prove the rst two statements, notice that since gn are smooth, the convergence p 0 0 p in L u Lu L (S ) implies that the limit is in Cu Cu L (S ). For the latter two statements, we use the fact that for p0 < , if fn f in Lp0 and fn is uniformly bounded in L , then f L . Proposition 70 allows us to conclude that all rst derivatives of g can be dened and belong to an appropriate space. These norms also allow us to conclude that the product of 2 any two rst derivatives of the metric belongs to L u Lu L (S ). For the second derivatives, 2 Proposition 70 shows that all second derivatives of the metric except u 2 (, b, ) are dened

0 2 as functions in L Lp u L (S ). The second derivative u2 (, b, ) is merely a distribution. As we will see below, its denition is not necessary to make sense of the Einstein equations.

2

82

We now show that all components of the curvature for the limiting spacetime, except for 2 RuAuB , RuuuA and Ruuuu , can be dened at worst as functions in L u Lu L (S ). For this we use the coordinate denition of the Riemann curvature tensor: R = g ( + (67) ). x x Proposition 71. In the limiting spacetime, all components of the Riemann curvature tensor, dened by the coordinate expression (67) expect for RuAuB , RuuuA and Ruuuu , can be dened 2 as functions in L u Lu L (S ). Moreover, R(L, L, L, eA ) and R(L, L, L, L) can be dened as 2 functions in L u Lu L (S ). Proof. Using (67), it is easy to see that if , , , = u, the expression has at most one u derivative. Thus, by the regularity properties of g , these components of the Riemann 2 curvature tensor are well-dened as L u Lu L (S ) functions. In the case that exactly one of , , , is equal to u, we can assume, by the symmetry properties of the Riemann curvature tensor, that = u. Then, the formula for the component of the Riemann curvature tensor has at most one u derivative and by the regularity properties, 2 it is well-dened as L u Lu L (S ) functions. Notice, however, that strictly speaking, at the level of regularity that we have, the symmetry properties of the Riemann curvature tensor may not hold. Consider, for example, the component R12u1 . Using the formula above, we would have a term g1A (g Au g12,u ),u , which is not dened. However, the sum of all terms containing two u derivatives is equal to the expression g1 (g u g12,u ). u which can be rearranged, up to terms with at most one u derivative, in the form (g1 g u g12,u ). u u This expression vanishes since g1 g = 0. The above calculation is a reection of the symmetry properties of the Riemann curvature tensor and provides an appropriate distributional denition of R for which the symmetry properties formally hold. We have therefore shown that all components of curvature with the exception of RuAuB , 2 RuuuA and Ruuuu are dened as L u Lu L (S ) functions. However, replacing the double null coordinate system (u, u, 1 , 2 ) by the null frame (L, L, e1 , e2 ), we can show that additional 2 components of curvature are dened as functions in L u Lu L (S ). To see this, we can write R(L, L, L, eA ) = RuuuA + bB RuBuA , and R(L, L, L, L) = Ruuuu + bB RuuuB + bA RuAuu + bA bB RuAuB = Ruuuu + 2bA RuAuu + bA bB RuAuB . Note that 1 1 Ruuuu = guu,uu + good terms = bA bB gAB,uu + good terms, 2 2 1 1 RuuuA = gAu,uu + good terms = bB gAB,uu + good terms, 2 2 1 RuBuA = gAB,uu + good terms, 2

83

where good terms denote terms with at most one u derivative. Therefore, the terms with two u derivatives cancel in R(L, L, L, A) and R(L, L, L, L) and they can be dened as 2 L u Lu L (S ) functions. We now verify that the limiting spacetime that we have constructed satises the vacuum Einstein equations in the sense that relative to the system of double null coordinates (u, u, 1 , 2 ), 2 R = 0 in L u Lu L (S ). For this, we use the coordinate denition of the Ricci curvature: R = + . x x Proposition 72. The limiting spacetime satises the vacuum Einstein equations in the sense that relative to the system of double null coordinates (u, u, 1 , 2 ),

2 R = 0 in L u Lu L (S ).

. Proof. Now, notice that with the metric given by (66), we have

1 uu 1 uu 1 u1 1 u2 (g ) = (g ) = (g ) = (g ) = 0.

Therefore, u does not contain a term with g,u . In other words, the expressions for Ruu , RuA , RAB for A, B = 1, 2 do not contain two u derivatives. Therefore, we have Ruu = RuA = RAB = 0

2 as functions in L u Lu L (S ). For the component Ruu , there are terms involving two u derivatives of the metric. In particular, we have u u Ruu = uu + terms involving at most one u derivative of the metric. u u uu From this expression, we see that the terms involving two u derivatives of the metric cancel. Thus we also have Ruu = 0 2 as functions in Lu Lu L (S ). Similarly, for RAu , we have

u u Au + terms involving at most one u derivative of the metric. u u uA As for Ruu , because of the cancellation, we have RAu = RuA = 0

2 as functions in L u Lu L (S ). It now remains to study Ruu . Indeed this term involves two u derivatives of the metric and there are no cancellations to remove this term. We have u A A Ruu = uu u + +terms involving at most one u derivative of the metric. uu + u u A uu u Au The rst two terms cancel each other. Note that

A uu = 0

84

since guu = guB = 0, We are thus left with g Au = 0. A AB Au = (g gAB ). u u u 2 It might seem that this term behaves like u However, this 2 which we cannot control. 4 particular combination of derivatives, by Proposition 68, is in fact in Lu Lu L (S ). Thus 2 Ruu = 0 in L u Lu L (S ).

5.8. Uniqueness. In the Subsection, we show that the spacetime constructed in Theorem 6 is the unique spacetime among the class of continuous spacetimes which arise as C 0 limits of smooth spacetime solutions to the vacuum Einstein equations. In view of Theorem 5, uniqueness follows easily if the spacetime solution is assumed to be in the class of spacetimes satisfying the conclusions of 5. Nevertheless, we prove uniqueness within a larger class of spacetimes for which it is not a priori assumed that the spacetime is Lipschitz, or that it is more regular in the u and the angular directions. The following is a precise formulation of the uniqueness theorem (Theorem 8) and is the main result in this Subsection: Proposition 73. Let (M(1) , g (1) ) and (M(2) , g (2) ) be two C 0 Lorentzian spacetimes in double null coordinates, dened in 0 u u , 0 u u and taking the form g (i) = 2((i) )2 (du du + du du) + ( (i) )AB (dA (b(i) )A du) (dB (b(i) )B du) such that = 1 and bA = 0 on H0 and H 0 and (1) |H0 = (2) |H0 and (1) |H 0 = (2) |H 0 (i) (i) for i = 1, 2, there exists a sequence of smooth spacetimes (Mn , gn ) such that for (i) every n, gn is dened in 0 u u and 0 u u and takes the following form in double null coordinates:

(i) i) 2 (i) A (i) A B B gn = 2(( n ) (du du + du du) + (n )AB (d (bn ) du) (d (bN ) du) ( i)

(i)

in C 0 in 0 u u , 0 u u ,

(2) The initial data for gn satisfy the assumptions of Theorem 4 uniformly, (i) (3) The initial data for gn converges to the initial data for g (i) , i.e.,

(i) gn |H0 g (i) |H0 (i) and gn |H 0 g (i) |H 0

in the norms in the assumptions of Theorem 5. Then, if u , u , where is as given in Theorem 4, g (1) = g (2) in 0 u u , 0 u u Proof. By Property (3) in the assumptions, for every i, there exists ni such that sup |( )i (AB )ni (u = 0)| + sup ||( )i (AB )ni (u = 0)| 2i , u u (

i 1 =,

u i1

u

85

i 1

u

sup || (tr, tr)ni ||L2 (Su,0 ) + sup ||2 (tr, tr)ni ||L2 (S0,u ) 2i ,

2 u u

2 ||i ni ||L2 2i . u L (S0,u )

{,,, } i1

By Theorem 5, there exists C depending only on the uniform bounds in the assumption (2) such that (1) (2) |gn gn | C 2i . i i Therefore, in C 0 norm, (1) (2) |g (1) g (2) | C |gn gn | C 2i . i i Since this holds for every i, we have g (1) = g (2) .

5.9. Propagation of Regularity. Using Proposition 28, we show that if the initial data is more regular in the 3 and directions, then so is the limiting spacetime. Proposition 74. Suppose, in addition to the assumptions of Theorem 5, we have the following estimates for the initial data: (

j J iI {,,,} i ||j 3 ||L2 (H 0 ) + {,,, } j ||i 3 ||L2 (H0 ) ) C.

Then (

j J iI {,,,} i sup ||j 3 ||L2 (H u ) + u {,,, } j sup ||i 3 ||L2 (Hu ) ) C . u

Proof. By Proposition 28, for an approximating sequence of spacetimes, the following bound holds independent of n: (

j J iI {,,,} i sup ||j 3 n ||L2 (H u ) + u {,,, } j sup ||i 3 n ||L2 (Hu ) ) C . u

The conclusion thus follows. Using the equations = 2, u 1 1 = , 2 u A b = 42 A , u together with Proposition 74, we can show that g is in the desired space as indicated in Theorem 9. The details are straightforward and will be omitted.

86

5.10. The Limiting Spacetime for an Impulsive Gravitational Wave. In the Subsection, we show that for the case of an impulsive gravitational wave, the spacetime is more regular than a general spacetime with merely bounded , thus concluding the proof of Theorem 1. In particular, we show that is well dened as a measure which is the weak limit of n . All other curvature components can be dened in L . Moreover, the singularity of is supported on a null hypersurface and the constructed spacetime is smooth away from this null hypersurface. 5.10.1. Limit of the Curvature Component . We rst show that n has a well dened limit as a measure. Proposition 75. In the case of an impulsive gravitational wave, for every u and every = (1 , 2 ) S2 , (u, ) is a measure on [0, u ], dened as the limit of n (u, ) Proof. By Proposition 34, for every u and , n has uniformly bounded L1 u norms. Thus to show convergence of n , we only need to show that for every u and every = (1 , 2 ) S2 ,

u

n (u, u , )du

0

converges to a limit (u, )([0, u)) as n for all u such that (u, )([0, u)) is continuous. By the equation 4 + tr = 2 , we have u u 1 n (u, u , )du = ( n + trn n + 2n n )(u, u , )du . n u 0 0 Integrating by parts and using 1 = 2n , u n we derive

u 1 1 n (u, )du = ( n )(u, u, ) ( n )(u, u = 0, ) + n n 0 0 u u

0

converges. Thus, in order to show weak convergence of n , we need to show pointwise 1 convergence for n for all u such that (u, )([0, u)) is continuous. By the construction n of the data for an impulsive gravitational wave in Section 3, we know that for the initial data ( n )AB (u = 0, u, ) C 2n for u us or u us + 2n . Using the equation 1 1 3 + tr = + 2 tr + , 2 2 one sees that ( n )AB (u, u, ) C 2n for u us or u us + 2n . 1 Therefore, n converges for all u = us . By Proposition 6, n also converges for all u = us . n

87

It remains to show that the limit (u, )([0, u)) is discontinuous at u = us . In view of (68), it suces to show that has a jump discontinuity across H us . This follows from the equation 1 1 3 + tr = + 2 tr + , 2 2 Rewriting in coordinates, we have 1 + bA A ) 1 ( 1 1 + tr u 2 1 + . = + 2 tr 2 Consider the coordinate system ( u, u, 1 , 2 ) such that u = u bA A . Since |bA | C , this is well-dened in a local coordinate patch. Thus, 1 1 + tr ) = + 2 tr 1 + . u 2 2 Notice that is continuous and the right hand side of the equation is also continuous. Since also has a jump for the initial data, ( u0 , u, ) has a jump discontinuity for u = us , discontinuity across u = us . 1 ( 5.10.2. Control of the Curvature Components , , , , . Except for the curvature component which can only be dened as a measure, all other curvature components can be dened as L u Lu L (S ) functions:

Proposition 76. , , , , can be dened as functions in L u Lu L (S ). Proof. That , , , can be dened as functions as L u Lu L (S ) follows from Proposition 74 and the Sobolev Embedding Theorem. To show that can be dened as a L u Lu L (S ) function, we consider the following Bianchi equation:

3 =

{, }

+

{,,, }

Since , , , , can be bounded in L u Lu L (S ), it follows from Proposition 7 and Gron walls inequality that can be estimated in L u Lu L (S ), as desired.

5.10.3. Smoothness of Spacetime away from H us . We now show that the spacetime is smooth away from the null hypersurface u = us . In the region u < us before the impulse, this follows from standard theory of local-wellposedness. We focus on the region after the impulse, where u > us . In the following Proposition, we prove uniform estimates for the 4 and derivatives of in the region u > us . The bounds for all derivatives of the curvature components follow from a combination of the estimates on , Theorem 9 and the Bianchi equations. Proposition 77. Suppose the data for the impulsive gravitational wave satisfy the estimates as given in Proposition 1 for some I 2, K 0. Then

i lim sup ||k u,u )) CI,K . 4 ||L2 (Hu ( iI 1 kmin{K,

I i1 2

u us +

88

i sup ||k 4 n ||L2 (Hu (us +2n , )) CI,K . iI 1 kmin{K,

I i1 2

(69)

Fix > 0. Take n large enough such that 2n < . Then (69) implies that

i sup ||k 4 n ||L2 (Hu (us +, )) CI,K . iI 1 kmin{K,

I i1 2

The conclusion follows from the fact that this bound is uniform in . As a consequence, we show Proposition 78. The limiting spacetime is smooth away from the the null hypersurface H us . Remark 9. In Proposition 77, we have only used the fact that for u > us , the initial data set is smooth. Therefore, suppose we have a initial data set satisfying the assumptions of Theorem 3, with the additional assumption that it is smooth for u > u for some u , we can also prove that the spacetime is smooth in 0 u , u <u . 5.10.4. Decomposition of into Singular and Regular Parts. Finally, we show that has a delta singularity on the incoming null hypersurface H us . Proposition 79. can be decomposed as = (us )s + r , where (us ) is the scalar delta function supported on the null hypersurface H us , s = s (u, ) = 0 belongs to L u L (S ) and r belongs to Lu Lu L (S ). Proof. Dene s (u, ) := lim+ 1 (u, u, ) lim 1 (u, u, ),

uus uus

and r := (us )s . We now show that s and r have the desired property. By Theorem 6, s belongs to L has a jump discontinuity across u = us , u L (S ). That s = 0 follows from the fact that which is proved in Proposition 75. It remains to show that r belongs to L u Lu L (S ). To show this, we consider the measure of the half open interval [0, u) using the measure r (u, ): (r (u, ))([0, u)) =(1 )(u, u, ) lim+ (1 )(u, u , ) + lim (1 )(u, u , ) (1 )(u, u = 0, )

u us u us u

+

0

u

= lim

u us u

u u

(1 )(u, u , )du u

+

0

89

(1 )(u, u, ) is a bounded function away from the the hypersurface By Proposition 77, u H us . Thus (r (u, ))([0, u)) can be expressed as an integral over [0, u) of an L u Lu L (S ) function, as desired.

References

[1] P. C. Aichelburg and R. U. Sexl. On the gravitational eld of a massless particle. Gen. Rel. Grav., 2:279284, 1971. [2] S. Alinhac. Ondes de rar efaction pour des syst` emes quasi-lin eaires hyperboliques multidimensionnels. In Journ ees Equations aux D eriv ees Partielles (Saint Jean de Monts, 1988), pages Exp. No. VIII, 7. Ecole Polytech., Palaiseau, 1988. [3] C. Barrabes and P. A. Hogen. Singular null hypersurfaces in general relativity. World Scientic, London, 2003. [4] M. Beals. Propagation and interaction of singularities in nonlinear hyperbolic problems. Progress in Nonlinear Dierential Equations and their Applications, 3. Birkh auser Boston Inc., Boston, MA, 1989. [5] J. Bicak. Selected solutions of einsteins eld equations: their role in general relativity and astrophysics. Lect. Notes Phys., 540:1126, 2000. [6] H. Bondi, F. A. E. Pirani, and I. Robinson. Gravitational waves in general relativity. III. Exact plane waves. Proc. Roy. Soc. Lond. A, 251(1267):519533, 1959. [7] M. W. Brinkmann. On Riemann spaces conformal to Euclidean space. Proc. Natl. Acad. Sci. U.S.A., 9:13, 1923. [8] Y. Choquet-Bruhat. Construction de solutions radiatives approch ees des equations dEinstein. Comm. Math. Phys., 12:1635, 1969. [9] D. Christodoulou. Bounded variation solutions of the spherically symmetric Einstein-scalar eld equations. Comm. Pure Appl. Math., 46(8):11311220, 1993. [10] D. Christodoulou. Examples of naked singularity formation in the gravitational collapse of a scalar eld. Ann. of Math. (2), 140(3):607653, 1994. [11] D. Christodoulou. The instability of naked singularities in the gravitational collapse of a scalar eld. Ann. of Math. (2), 149(1):183217, 1999. [12] D. Christodoulou. The formation of black holes in general relativity. EMS Monographs in Mathematics. European Mathematical Society (EMS), Z urich, 2009, arXiv:0805.3880. [13] D. Christodoulou and S. Klainerman. The global nonlinear stability of the Minkowski space, volume 41 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ, 1993. [14] J.-F. Coulombel and P. Secchi. The stability of compressible vortex sheets in two space dimensions. Indiana Univ. Math. J., 53(4):9411012, 2004. [15] A. Einstein and N. Rosen. On gravitational waves. J. Franklin Inst., 223:4354, 1937. [16] J. B. Griths. Colliding plane waves in general relativity. Oxford UP, Oxford, 1991. [17] J. B. Griths and J. Podolsky. Exact space-times in Einsteins general relativity. Cambridge UP, Cambridge, 2009. [18] G. Holzegel. Ultimately Schwarzschildean spacetimes and the black hole stability problem. 2010, arXiv:1010.3216. [19] T. J. R. Hughes, T. Kato, and J. E. Marsden. Well-posed quasi-linear second-order hyperbolic systems with applications to nonlinear elastodynamics and general relativity. Arch. Rational Mech. Anal., 63(3):273294 (1977), 1976. [20] P. Jordan, J. Ehlers, and W. Kundt. Strenge l osungen der feldgleichungen der allgemeinen relativit atstheorie. Akad. Wiss. Lit. Mainz, Abh. Math. Naturwiss., 2, 1960. [21] K. A. Khan and R. Penrose. Scattering of two impulsive gravitational plane waves. Nature, 229:185186, 1971. [22] S. Klainerman and F. Nicol` o. The evolution problem in general relativity, volume 25 of Progress in Mathematical Physics. Birkh auser Boston Inc., Boston, MA, 2003. [23] S. Klainerman and I. Rodnianski. Causal geometry of Einstein-vacuum spacetimes with nite curvature ux. Invent. Math., 159(3):437529, 2005. [24] S. Klainerman and I. Rodnianski. On the formation of trapped surfaces. 2009, arXiv:0912.5097.

90

[25] S. Klainerman, I. Rodnianski, and J. Szeftel. The Bounded L2 Curvature Conjecture. 2012, arXiv:1204.1767. [26] P. G. LeFloch and J. Smulevici. Global geometry of T 2 -symmetric spacetimes with weak regularity. C. R. Math. Acad. Sci. Paris, 348(21-22):12311233, 2010. [27] P. G. LeFloch and J. M. Stewart. The characteristic initial value problem for plane symmetric spacetimes with weak regularity. Class. Quant. Grav., 28:145019, 2011, 1004.2343. [28] J. Luk. On the local existence for the characteristic initial value problem in general relativity. 2011, arXiv:1107.0898. [29] A. Majda. The existence of multidimensional shock fronts. Mem. Amer. Math. Soc., 43(281):v+93, 1983. [30] A. Majda. The stability of multidimensional shock fronts. Mem. Amer. Math. Soc., 41(275):iv+95, 1983. [31] G. M etivier. The Cauchy problem for semilinear hyperbolic systems with discontinuous data. Duke Math. J., 53(4):9831011, 1986. [32] G M etivier. Stability of multi-dimensional weak shocks. Comm. Partial Dierential Equations, 15(7):9831028, 1990. [33] G. M etivier. Ondes soniques. J. Math. Pures Appl. (9), 70(2):197268, 1991. [34] R. Penrose. A remarkable preperty of plane waves in general relativity. Rev. Mod. Phys., 37:215220, 1965. [35] R. Penrose. The geometry of impulsive gravitational waves. In General relativity (papers in honour of J. L. Synge), pages 101115. Clarendon Press, Oxford, 1972. [36] A. D. Rendall. Reduction of the characteristic initial value problem to the Cauchy problem and its applications to the Einstein equations. Proc. Roy. Soc. London Ser. A, 427(1872):221239, 1990. [37] J. Szeftel. Parametrix for wave equations on a rough background I: regularity of the phase at initial time . 2012, arXiv:1204.1768. [38] J. Szeftel. Parametrix for wave equations on a rough background II: construction and control at initial time . 2012, arXiv:1204.1769. [39] J. Szeftel. Parametrix for wave equations on a rough background III: space-time regularity of the phase . 2012, arXiv:1204.1770. [40] J. Szeftel. Parametrix for wave equations on a rough background IV: control of the error term . 2012, arXiv:1204.1771. [41] P. Szekeres. Colliding gravitational waves. Nature, 228:11831184, 1970. [42] A. H. Taub. Space-times with distribution-valued curvature tensors. J. Math. Phys., 21(6):14231431, 1980. [43] U. Yurtsever. Colliding almost-plane gravitational waves: Colliding exact plane waves and general properties of almost-plane-wave spacetimes. Phys. Rev. D, 33:28032817, 1988. Department of Mathematics, Princeton University, Princeton NJ 08544 E-mail address : jluk@math.princeton.edu Department of Mathematics, Princeton University, Princeton NJ 08544 and Department of Mathematics, MIT, Cambridge, MA 02139 E-mail address : irod@math.princeton.edu, irod@math.mit.edu

- Gravitational Waves: An IntroductionUploaded bysflima2000
- Gravitational Waves in MatterUploaded byocean66
- Detecting Fractal Dimensions via Primordial Gravitational Wave Astronomy. Fractal Cosmology.Uploaded byLuke Lee
- Chris Van Den Broeck- A ‘warp drive’ with more reasonable total energy requirementsUploaded byUtnyu
- 0401117.pdfUploaded byBhanuka Samarakoon
- Artificial generation of accustic and gravitational waves in the atmosphereUploaded byNOLINKNEWS
- gravitational waveUploaded byManodeep Mondal
- On the Vacuum Propagation of Gravitational Waves (WWW.OLOSCIENCE.COM)Uploaded byFausto Intilla
- Gravitational Waves in Relativistic Theory of GravitationUploaded byFerry Yudha Pratama
- Communication System Using Gravitational Waves - US6300614B1Uploaded byJohn Carter
- Gravitational Wave DetectionUploaded byChris Hollebon
- The Schwarzschild solution and its implications for gravitational wavesUploaded bythenarmis9384
- Gravitational WavesUploaded byDevendra Pratap Singh
- PhysRevD.27.2885Uploaded byestebanmtzv
- definision physic stpm sem1Uploaded byJason Choo
- Quantum State of the UniverseUploaded byJJ
- Sung-Won Kim- Scattering problem of scalar wave in wormhole geometryUploaded byRtpom
- Choral Speaking Script 2016Uploaded bypain_rinnegan
- Surfer GridsUploaded byMuhammadFatirAz-Zamakhsyahri
- Gravity Power generationUploaded byArslan Naveed
- Origin of Gravity and Mass - A New ThinkingUploaded byJohn Paily
- zidu guo - science fair research paperUploaded byapi-392691300
- Beekle HenryUploaded byisquare77
- Tugas Besar CaeUploaded byJati Pamungkas
- S. Krasnikov- A traversable wormholeUploaded byHerftez
- Some Keys for Gold Making AlchemyUploaded byJorge Daniel Ferreira Granados
- Cos Grove Space TimeUploaded bysrfairchild

- Be Here Now - NatureUploaded byforizsl
- Delirium Quantum - Or, Where I Will Take Quantum Mechanics if It Will Let MeUploaded byforizsl
- On Participatory RealismUploaded byforizsl
- Conformal Carroll Groups 1403.4213Uploaded byforizsl
- Some Negative Remarks on Operational Approaches to Quatum TheoryUploaded byforizsl
- The Physical Interpretation of DaseinisationUploaded byforizsl
- Time Crystals From Minimum Time UncertaintyUploaded byforizsl
- What is a Thing - Topos Theory in the Foundations of PhysicsUploaded byforizsl
- Cauchy Problem as a Two-surface Based Geometrodynamics 1409.4914Uploaded byforizsl
- A Renormalizable Theory of Quantum Gravity - Renormalization Proof of the Gauge Theory of Volume Preserving Diffeomorphisms 1308.2384Uploaded byforizsl
- How Well Do We Understand the Thermal History of the Universe - Implications of the Recent BICEP2 DataUploaded byforizsl
- Quantum Theory From Rules on Information AcquisitionUploaded byforizsl
- Conf Schedule SmallUploaded byforizsl
- 't Hooft the Firewall Transformation for Black Holes and Some of Its Implications 1612.08640Uploaded byforizsl
- The Problem of Space in the Light of Relativity - The Views of H. Weyl and E. CartanUploaded byforizsl
- Irreducibility of the Ashtekar-Isham-Lewandowski RepresentationUploaded byforizsl
- !Mermin_Physics-QBism puts the scientist back into science_Nature_507421a.pdfUploaded byforizsl
- A New Approach to the Hard Problem - 07.14 - Final - KI - TSC 2014Uploaded byforizsl
- Antimatter and the Second Law of Thermodynamics 1704.01750Uploaded byforizsl
- Quantum Theory From Rules on Information AcquisitionUploaded byforizsl
- Testing Loop Quantum CosmologyUploaded byforizsl
- The Atoms of Spacetime and the Cosmological ConstantUploaded byforizsl
- -A Critique of the Determination of the Energy-momentum of a System From the Equations of Motion of Matter in the General Theory of RelativityUploaded byforizsl
- !Spectral Properties of the Quantum Mixmaster Universe 1703.08462Uploaded byforizsl
- The 1965 Penrose Singularity TheoremUploaded byforizsl
- ! Konishi Time and a temporally statistical geometrodynamics (quantum brain dynamics) 2368.pdfUploaded byforizsl
- Back to Parmenides 2106Uploaded byforizsl
- Quantum Bayesianism at the PerimeterUploaded byforizsl
- 't Hooft the Firewall Transformation for Black Holes and Some of Its Implications 1612.08640Uploaded byforizsl
- Quantum Measurement and the Paulian IdeaUploaded byforizsl

- Ma English ProjectUploaded bySubodh Sonawane
- Chandogya Upanisad Selections From Chapter 8Uploaded bytimsmith1081574
- Neurophysiology of Anxiety DisordersUploaded byShivan A.C.
- Conducting Marketing Research and Market DemandsUploaded byJune Angeli Muyco
- 2433_75bcUploaded byJoe Dalton
- FPTree-09Uploaded bySuganyaEaswaran
- GlaucusResearch-National Beverage-Short-Nasdaq FIZZ Sept 28 2016Uploaded bychasingthevig
- Enterprise DBA Part 1A - Architecture and Administration - Volume 1 • Instructor GuideUploaded byMarcos Perez
- Philosophy of Science Handout - The Problem of InductionUploaded byKinez72
- IELTS Problem - Solution Essay Writing.pdfUploaded bynavdeep kaur
- what rabbit owners need to know about myxomatosis and VHDUploaded bybuddies_bunnies
- United States v. Donald Marsh, 4th Cir. (2012)Uploaded byScribd Government Docs
- HOUSE HEARING, 109TH CONGRESS - FREEDOM IN THE WORKPLACEUploaded byScribd Government Docs
- IT Enabled HRUploaded bytubai25
- BatteriesUploaded bydeanweipa
- ITM Guidelines for Answering a Case Study Assignment 2016Uploaded bySandy
- Barton FreemasonryUploaded byIon Cristian
- Child Literacy Case StudyUploaded byJessica Fiero
- Staff AccountantUploaded byapi-78139028
- Manager/Director/AVPUploaded byapi-121377554
- Bible ComparisonsUploaded byOzancan Demirışık
- B07MGJ2LJRUploaded byjkruise
- Defining the New Spirituality by George D. ChryssidesUploaded byAna Dajić
- Cisco IOS Quick ReferenceUploaded bySel Espinosa
- 21 January 2015 ConstiUploaded byreylaxa
- 26_summary of MafaUploaded byhina1234
- the millionaire next door e-portfolioUploaded byapi-240741436
- Sample Proposal BudgetsUploaded byanamu_online
- sanipro.pptUploaded byYus Ani
- All Adventures GamesUploaded byМилош Ђорђевић