Shielding linearised-gravity∗ 7 1 Robert Beig†, Piotr T. Chru´sciel‡ 0 University of Vienna 2 r March 14, 2017 a M 2 1 Abstract ] c We present an elementary argument that one can shield linearised q gravitational fields using linearised gravitational fields. This is done - r by using third-order potentials for the metric, which avoids the need g to solve singular equations in shielding or gluing constructions for the [ linearised metric. 2 v 6 Contents 8 4 0 1 Introduction 2 0 1. 2 Shielding linearised gravity 3 0 2.1 The Cauchy problem for linearised gravity . . . . . . . . . . . 4 7 2.2 Third order potentials . . . . . . . . . . . . . . . . . . . . . . 6 1 : 2.3 Shielding gravitational Cauchy data . . . . . . . . . . . . . . 7 v i X 3 Shielding Maxwell fields 9 r a 4 The Weyl tensor formulation 11 A Integrating 2-forms on R3 12 B Construction of the potential u 13 C A potential for the linearised Riemann tensor 15 ∗Preprint UWThPh-2016-31 †[email protected] ‡[email protected],URLhomepage.univie.ac.at/piotr.chrusciel 1 1 Introduction A fundamental property of Newtonian gravity is that the gravitational field cannot be localised in a bounded region. This is a simple consequence of the equation ∆φ = 4πGρ, where φ is the gravitational potential, G is Newton’s constant and ρ is the matter density: The requirement that ρ≥ 0, and the asymptotic behaviour −M/r of φ, where M is the total mass, implies that φ vanishes at large distances along a curve extending to infinity if and only if there is no matter whatsoever and φ ≡ 0. It is therefore extremely surprising that in general relativity, gravitational fields can be shielded away by gravitational fields, as proved recently in a remarkable paper by Carlotto and Schoen [5]. Since Newtonian gravity is part of the weak-field limit of general relativ- ity(indeed,thisisweak-fieldGRwithsmallvelocities), onewondersifasim- ilar screeningcan occur forlinearised relativity. As itturnsout, theanalysis of Carlotto and Schoen can bereadily generalised to linearised gravitational fields on cone-like sets as considered in [5] (compare [9]). This, however, requires sophistical mathematical machinery which imposes restrictions to the sets considered and, as an intermediate step, uses solutions blowing-up at the relevant boundaries, which leads to difficulties when trying to imple- ment the method numerically. The object of this note is to point out an alternative elementary method to perform gluings, or achieve screening of linearised gravity by linearised gravitational fields near a Minkowski back- ground. In particular, we give here a very simple proof that at any given time t, and givenanyopen setΩ ⊂ R3, everylinearised vacuumgravitational field h on {t}×R3 can be deformed to a new linearised vacuum field h˜ µν µν so that h˜ coincides with h on Ω and vanishes outside a slightly larger µν µν set. In other words, the gravitational field has been screened away outside of Ω, and this by using gravitational fields only: no matter fields, whether with positive or negative density, are needed. We emphasise that the construction of Carlotto-Schoen switches-off the gravitational field in sets which have a cone-like structure, whether in the linearised case or in the full treatment. In our approach no restrictions on the geometry of Ω occur, so that the screening can be done near any set. Our construction is likely to be useful for the numerical construction of initial data sets with interesting properties, by providing an efficient way of making gluings in the far-away zone, where nonlinear corrections become inessential. Here, as already pointed out, both the Corvino-Schoen and the 2 Carlotto-Schoen gluings require solving elliptic equations in spaces of func- tionswhicharesingularattheboundaryofthegluingregion(see[10,8]fora review), while our gluings are performed by explicit elementary integrations ((2.31)below),multiplicationbyacut-offfunction,andapplyingderivatives, once the metric has been put into transverse and traceless (TT)-gauge. The above leads one naturally to ask similar questions for electric and magnetic fields. Here we provide a simple proof that Maxwell fields can be shielded by Maxwell fields. Last but not least, we show how to perform the screeninginpractice,inthatweprovethatallsolutionsofsourcelessMaxwell equations in a bounded space-time region can be realised by manipulating charges and currents in an enclosing bounded region. 2 Shielding linearised gravity Consider R3+1 with a metric which, in the natural coordinates on R3+1, takes the form g = η +h , (2.1) µν µν µν where η denotes the Minkowski metric. Suppose that there exists a small constant ǫ such that we have |h |, |∂ h |, |∂ ∂ h |= O(ǫ). (2.2) µν σ µν σ ρ µν If we use the metric η to raise and lower indices one has 1 R = [∂ {∂ hα +∂ hα −∂αh }−∂ ∂ hα ]+O(ǫ2). (2.3) βδ α β δ δ β βδ δ β α 2 Coordinate transformations xµ 7→ xµ+ζµ, with |ζ |, |∂ ζ |, |∂ ∂ ζ |, |∂ ∂ ∂ ζ | = O(ǫ), (2.4) µ σ µ σ ρ µ σ ρ ν µ preserve (2.2), and lead to the gauge-freedom h 7→ h +∂ ζ +∂ ζ . (2.5) µν µν µ ν ν µ Imposing the wave-coordinates condition up to O(ǫ2) terms, ✷ xα = O(ǫ2), (2.6) g leads to 1 ∂ hβ = ∂ hβ +O(ǫ2), (2.7) β α α β 2 as well as 1 R = − ✷ h +O(ǫ2). (2.8) βδ η βδ 2 3 2.1 The Cauchy problem for linearised gravity InwhatfollowsweignoreallO(ǫ2)-termsintheequationsaboveandconsider the theory of a tensor field h with the gauge-freedom (2.5) and satisfying µν the equations 0= ∂ {∂ hα +∂ hα −∂αh }−∂ ∂ hα . (2.9) α β δ δ β βδ δ β α Solving the following wave equation 1 ✷ζ = −∂ hβ + ∂ hβ , α β α α β 2 where✷ ≡ ✷ isthewave-operator oftheMinkowskimetric,andperforming η (2.5) leads to a new tensor h , still denoted by the same symbol, such that µν 1 ∂ hβ = ∂ hβ , (2.10) β α α β 2 together with the usual wave equation for h: ✷h = 0. (2.11) βδ Solutions of this last equation are in one-to-one correspondence with their Cauchy data at t = 0. However, those data are not arbitrary, which can be seen as follows: Equations (2.10)-(2.11) imply 1 ✷(∂ hβ − ∂ hβ )= 0. (2.12) β α α β 2 It follows that (2.10) will hold if and only if 1 1 ∂ hβ − ∂ hβ = 0= ∂ ∂ hβ − ∂ hβ . (2.13) β α α β 0 β α α β (cid:18) 2 (cid:19)(cid:12)t=0 (cid:18) 2 (cid:19)(cid:12)t=0 (cid:12) (cid:12) (cid:12) (cid:12) Equivalently, taking (2.11) into account, ∂ (h +hi )| = 2∂ hi | , (2.14) 0 00 i t=0 i 0 t=0 ∂ h | = ∂ hj + 1∂ (h −hj ) , (2.15) 0 0i t=0 j i 2 i 00 j t=0 ∆h(cid:0)i | = ∂ ∂ hij| , (cid:1)(cid:12) (2.16) i t=0 i j t=0 (cid:12) ∂ ∂ hj −∂ hk δj = ∆h −∂ ∂ hj . (2.17) j 0 i 0 k i t=0 0i i j 0 t=0 (cid:0) (cid:1)(cid:12) (cid:0) (cid:1)(cid:12) (cid:12) (cid:12) The last two equations are of course the linearisations of the usual scalar and vector constraint equations. 4 There remains the freedom of choosing ζ | and ∂ ζ | . We choose α t=0 t α t=0 ∂ hk −2∂ hk −2∆ζ = 0, 0 k k 0 0 t=0 (cid:0) (cid:1)(cid:12) h00+2∂0ζ0 t=0 =(cid:12)0, (cid:0) (cid:1)(cid:12) h0i +∂iζ0+∂0ζ(cid:12)i t=0 = 0, Di hij − 13hk(cid:0)kδji +Diζj +Djζ(cid:1)i(cid:12)(cid:12)− 32Dkζkδji t=0 = 0, (2.18) (cid:0) (cid:1)(cid:12) where D ≡ Di ≡ ∂ in Cartesian coordinates. Indeed(cid:12), given any h and i i µν ∂ h | , the first equation can be solved for ζ | under suitable natural 0 µν t=0 0 t=0 conditionsonthedata; theseconddefines∂ ζ | ;thethirddefines∂ ζ | ; 0 0 t=0 0 i t=0 finally, the last equation is an elliptic equation for the vector field ζ | i t=0 which can be solved [7] if one assumes that the field 1 ∂ hi − hk δi (2.19) i j 3 k j t=0 (cid:0) (cid:1)(cid:12) (cid:12) belongs to a suitable weighted Sobolev or H¨older space, the precise require- ments being irrelevant for our purposes. We simply note that if some com- ponents of h behave as 1/r, then ζ will behave like lnr in general, which is ij likely to introduce lnr/r terms in the gauge-transformed metric. After per- forming this gauge-transformation, we end up with a tensor field h which µν satisfies 1 ∂ hk = h = h = ∂ hi − hk δi = 0. (2.20) 0 k t=0 00 t=0 0i t=0 i j 3 k j t=0 (cid:12) (cid:12) (cid:12) (cid:0) (cid:1)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) Inserting this into (2.14)-(2.17) we find ∂ h | = 0, (2.21) 0 00 t=0 ∂ h | = −1∂ hj | , (2.22) 0 0i t=0 6 i j t=0 ∆hi | = 0, (2.23) i t=0 ∂ ∂ hj −∂ hk δj = 0. (2.24) j 0 i 0 k i t=0 (cid:0) (cid:1)(cid:12) The further requirement that hi goes to zer(cid:12)o as r tends to infinity together i with the maximum principle gives hi | = 0. (2.25) i t=0 We conclude (compare [1]) that at any given time t = t every linearised 0 gravitational initial data set (h ,∂ h )| can be gauge-transformed to µν t µν t=t0 the TT-gauge: writing k = ∂ h , we have ij 0 ij hk | = ∂ hi | = kk = ∂ ki = 0. (2.26) k t=t0 i j t=t0 k i j 5 From what has been said and from uniqueness of solutions of the wave equation we also see that in this gauge we will have for all t h = h = hk = ∂ hi = 0, (2.27) 00 0i k i j which further implies that (2.26) is preserved by evolution. It should be pointed out that when the construction is carried out on the complement of a ball, e.g. because sources are present, or because we perform the construction at large distances only where the non-linearities become negligible, then (2.25) will not hold in general, and the trace of h ij will be non-trivial, with the usual expansion in terms of inverse powers of r, starting with 1/r-terms associated with the total mass of the configuration. In such cases our construction below still applies to the transverse-traceless part of the metric. 2.2 Third order potentials We will need the following result from [3], which can be summarised as follows: Let h be a symmetric, transverse and traceless tensor on R3, ij ∂ hi = 0 = hi . (2.28) i j i Thenthereexists asymmetrictraceless “thirdorderpotential” u suchthat ij h = P(u) , (2.29) mℓ mℓ where(hereg denotestheEuclideanmetricandDi theassociatedcovariant ij derivative) 1 1 P(u) := ǫ ij∂ ∆u −2∂ Dnu + g DnDku , (2.30) mℓ m i jℓ (ℓ j)n jℓ nk 2 2 (cid:0) (cid:1) and where u = u dxidxj can be constructed by the following procedure: ij Letting 1 σ (~x) := ǫ ℓh (λ~x)λ(1−λ)2dλ, (2.31) ijk ij ℓk Z 0 we set u = 2xmxnx σ +r2xmσ . (2.32) jℓ (j ℓ)mn m(jℓ) (This is clearly symmetric, and tracelessness is not very difficult to check. Other third-order-potentials u are possible, differing by an element of the kernel of P.) One way to see how (2.30) arises is to note that P(u) is, apart from a numerical factor, the linearisation at the flat metric of the 6 Cotton-York tensor in the direction of the trace-free tensor u. For (2.32), the formulae follow by successively integrating thrice the two-forms given in [3], at each step using the Poincar´e formula (3.7) below. We sketch the construction in Appendix A. The converse is also true: given any symmetric trace-free tensor u , the ij tensor field P(u) defined by (2.30) is symmetric, transverse and traceless (of which only the last property and the vanishing of the divergence on the first index are obvious). As an example, consider h describing a plane gravitational wave in ij TT-gauge propagating in direction ~k, h (~k) = ℜ H ei~k·~x , ∂ H = 0 = Hi = H kj, (2.33) ij ij ℓ ij i ij (cid:0) (cid:1) withpossiblycomplex coefficients H , whereℜdenotes therealpart. Then ij 1 σ = ℜ ǫ Hℓ eiλ~k·~x(λ−2λ2+λ3)dλ ijk ijℓ k Z (cid:0) 0 (cid:1) = ℜ Wǫ Hℓ , where (2.34) ijℓ k 2i(cid:0)ei~k·~x(~k·~x+(cid:1) 3i)−~k·~x(~k·~x−4i)+6 W(~x) = (2.35) (~k·~x)4 (which tends to 1/12 when ~k·~x tends to zero), u = ℜ W 2xmxix ǫ Hk −r2xiǫ Hk . (2.36) jℓ (j ℓ)mk i ik(j ℓ) (cid:16) (cid:17) (cid:0) (cid:1) As another example, consider the family of fields 2 u = ln(1+r2) D λ +D λ − Dkλ g . (2.37) ij i j j i k ij 3 (cid:0) (cid:1) Tensors of the form (2.37) with the ln(1+r2) term removed form the kernel of P for any λ (cf. Appendix A), which easily implies that if λ ∼ O(rσ) for i large r then h ∼ O(rσ−4), for all σ ∈ R. ij We also note that if h is compactly supported to start with, then u ij ij can also be chosen to be compactly supported; compare Appendix A. 2.3 Shielding gravitational Cauchy data We are ready to prove now a somewhat moregeneral version of our previous claim, that at any given time t, and given any region Ω ⊂ R3, every vacuum initial data set for the gravitational field (h ,k ) can be deformed to a new ij ij 7 vacuum initial data set (h˜ ,k˜ ) which coincides with (h ,k ) on Ω and ij ij ij ij vanishes outside of a slightly larger set. Indeed, consider any linearised gravitational field in the gauge (2.27). Denote by (h ,k ) theassociated Cauchy data at t, andlet (u ,v )denote ij ij ij ij the corresponding potentials discussed in Section 2.2, thus (h ,k ) = (P(u) ,P(v) ), (2.38) ij ij ij ij where P is the third-order differential operator of (2.30). Let Ω beany open subset of R3 and let Ω be any open set containing Ω. Let χ be any smooth Ω function which is identically equal to one on Ω and which vanishes outside e of Ω. Then the initial data set e (h˜ ,k˜ ) = (P(χ u) ,P(χ v) ) (2.39) ij ij Ω ij Ω ij satisfiesthevacuumconstraintequationseverywhere,coincideswith(h ,k ) ij ij in Ω and vanishes outside of Ω. When Ω is bounded, the new fields (h˜ ,k˜ ) can clearly be chosen to ij ij e vanish outside of a bounded set. For example, consider a plane wave so- lution as in (2.33). Multiplying the potentials (2.36) by a cut-off function χ which equals one on B(R ) and vanishes outside of B(R ) provides B(R1) 1 2 compactly supported gravitational data which coincide with the plane-wave ones in B(R ). (Alternatively one can replace~k·~x in the first line of (2.34), 1 or in (2.35)-(2.36), by ~k·~x χ .) In the limit ~k = 0, so that h is con- B(R1) ij stant and, e.g., k = 0, one obtains data which are Minkowskian in B(R ), ij 1 and outside of B(R ), and describe a burst of radiation localised in a spher- 2 ical shell. Note that the Minkowskian coordinates for the interior region are distinct from the ones for the outside region. The closest full-theory configuration to this would be Bartnik’s time symmetric initial data set [2] which are flat inside a ball of radius R , and which can be Corvino-Schoen 1 deformed to be Schwarzschildean outside of the ball of radius R ; here R 2 2 willbemuchlargerthanR ingeneral, butcanbemadeasclosetoR asde- 1 1 sired by making the free data available in Bartnik’s construction sufficiently small. For Ω’s which are not bounded it is interesting to enquire about fall-off properties of the shielded field. This will depend upon the geometry of Ω and the fall-off of the initial field: For cone-like geometries, as considered in [5, 9], and with h = O(1/r), µν the gravitational field in the screening region will fall-off again as O(1/r). This is rather surprising, as the gluing approach of [5] leads to a loss of decay even for the linear problem. One should, however, keep in mind that 8 the transition to the TT-gauge for a metric which falls-off as 1/r is likely to introduce lnr/r terms in the transformed metric, which will then propagate to the gluing region. As another example, consider the set Ω = (a,b) × R2, which is not covered by the methods of [5]. Our procedure in this case applies but if h = O(1/r), and if the cut-off function is taken to depend only upon the µν firstvariable of the productΩ = (a,b)×R2, one obtains a gravitational field h˜ vanishingoutside a slab Ω = (c,d)×R2, with [a,b] ⊂ (c,d), which might µν grow as r2lnr when receding to infinity within the slab. e So far we have been concentrating on “shielding”. But of course the above can be used to glue linearised field across a gluing region, by inter- polating the respective u’s to each other in the gluing zone. Equivalently, screen each of the fields which are glued to zero across the gluing region, and add the resulting new fields. 3 Shielding Maxwell fields Maxwell equations in Minkowski space-time share at least two features with linearised gravity: existence of constraint equations, and existence of gauge transformations. It might therefore be unsurprising that there exists a ver- sion oftheCarlotto-Schoen construction which appliesto theMaxwell equa- tions; compare [11, 10] for a discussion of the Maxwell equivalent of the Corvino-Schoen construction, which generalises without further ado to the Carlotto-Schoen setting. We wish to show here how to carry-out the shield- ing of Maxwell fields with Maxwell fields in an elementary way. Recall that solutions of the source-free Maxwell equations are in one-to- one correspondence with their initial data at time t; these are simply the electric field E~ and the magnetic field B~ at t. These fields are not arbitrary, but satisfy the constraints divE~ = divB~ = 0. (3.1) On R3, these imply the existence of vector potentials ~ω and A~ such that E~ = curl~ω, B~ = curlA~. (3.2) In fact there is an explicit formula for ~ω, 1 ω = ǫ xj Ek(λx)λdλ, (3.3) i jik Z 0 9 similarly for A~. Using (3.2) it is straightforward to show that at any given time t, and given any region Ω ⊂ R3, every sourceless Maxwell fields (E~,B~) can be deformed to new sourceless Maxwell fields which coincide with (E~,B~) on Ω and vanish outside a slightly larger set. Indeed, letting Ω and χ be Ω as in the paragraph following (2.38), the new Maxwell fields at t e E~ = curl(χ ~ω) and B~ = curl(χ A~) (3.4) Ω Ω aredivergence-free, coincidewiththeoriginalfieldsonΩ,andvanishoutside of Ω. One can solve the Cauchy problem for the Maxwell equations with the e new initial data (3.4) to obtain the associated space-time fields, if desired. The question then arises1 if every such configurations can be realised by an experimentalist in the lab. Here “an experimentalist” is defined as some- one whose laboratory equipment can produce any desired electric charges ρ and currents~j subject to the conservation law ∂ρ +div~j = 0. (3.5) ∂t These, inturn,willproduceMaxwell fieldsas dictated by theMaxwell equa- tions written in their tensorial special-relativistic form: ∂ Fµν = 4πjµ, where (jµ) = (ρ,~j). (3.6) ν More specifically, let us describe the lab as the following “world-volume”: U := [t ,t ]×Ω ⊂ R4. 0 3 The region within the labfwhere the desireed Maxwell fields need to be pro- duced will be the set U := [t ,t ]×Ω⊂ U , 1 2 with t0 < t1 ≤ t2 < t3 and Ω ⊂ Ω. Let Fµν befa source-free solution of the Maxwell equations in U, as needed to carry out the desired experiments. e The following prescription tells us what are the charges and currents outside of U which will produce a Maxwell field F˜ coinciding with F in µν µν U, out of a vacuum configuration at t ≤ t0: Let χU be a smooth function which is identically one on U and which vanishes outside of U. Let A be µ any four-vector potential associated with F , e.g. µν f 1 A (xα)= xν F (λxα)λdλ. (3.7) µ νµ Z 0 1Weare grateful to Peter Aichelburgfor pointing-out theissue to us. 10