Spacetime symmetries and topology in bimetric relativity

Francesco Torsello,1, ∗ Mikica Kocic,1, † Marcus H¨og˚as,1, ‡ and Edvard M¨ortsell1, § 1Department of Physics & The Oskar Klein Centre, Stockholm University, AlbaNova University Centre, SE-106 91 Stockholm, Sweden We explore spacetime symmetries and topologies of the two metric sectors in Hassan–Rosen bimetric theory. We show that, in vacuum, the two sectors can either share or have separate spacetime symmetries. If stress–energy tensors are present, a third case can arise, with different spacetime symmetries within the same sector. This raises the question of the best definition of spacetime symmetry in Hassan–Rosen bimetric theory. We emphasize the possibility of imposing ansatzes and looking for solutions having different Killing vector fields or different isometries in the two sectors, which has gained little attention so far. We also point out that the topology of spacetime imposes a constraint on possible metric combinations. Keywords: Modified gravity, ghost-free bimetric theory, bimetric relativity, spacetime symmetries, isometries, Killing vector fields, topology

I. INTRODUCTION In this case, it is straightforward to talk about a space- time symmetry, since the whole system is invariant under Symmetries are fundamental in physics. The invari- the transformation. If this is not the case, however, the ance of physical quantities under a given transformation definition of a spacetime symmetry is ambiguous. Solu- allows physicists to simplify their models of Nature and tions having different KVFs and showing this ambiguity, to better understand them. in the context of the HR bimetric theory, are presented The first computed exact solution in general relativity in section II B. (GR) is the Schwarzschild solution [1], which was de- When finding particular solutions, the most common termined by assuming a static and spherically symmet- ansatzes presume the same isometries in both sectors ric spacetime. Schwarzschild’s approach has been widely (some exceptions can be found in [4,5]). This study and fruitfully used until our days. Indeed, the standard investigates when this is the most general ansatz, and if way for finding particular solutions of certain field equa- there are other possible ansatzes which can, in principle, tions is to impose some symmetries to the system, and lead to new solutions of the bimetric field equations. This deduce the appropriate ansatz for the dynamical fields. turns out to be the case. In (pseudo-)Riemannian geometry, spacetime symme- We also answer to the question if there are any con- tries are characterized by Killing vector fields (KVFs), straints on the solutions of HR bimetric theory, based on whose integral curves define paths along which the phys- the underlying topology of the spacetime. The answer ical quantities can be invariant. When this happens, the turns out to be yes, as we will discuss in section II A. standard terminology is that the physical quantity pos- The paper is organized in two main sections. In sec- sesses a symmetry along that path (or that the path de- tion II, we focus on HR bimetric theory, reviewed in sec- fines a “collineation” for the physical quantity). In the tion II A. We start by discussing explicit examples hav- case of a symmetry of the metric tensor, they coincide ing the described properties in section II B. These exam- with its isometries. We refer the reader to [2] for a basic ples should be thought of as a selection of cases showing treatment of spacetime symmetries and to [3] for a more different possible configurations of spacetime symmetries advanced approach. in HR theory. In principle, they may or may not pro- In this paper we explore the relation between the KVFs vide any other physical information. In section II C we of two metrics defined on the same differentiable man- arXiv:1710.06434v2 [gr-qc] 19 Apr 2018 study spacetime symmetries and we fit the examples into ifold. The motivation for this study comes from the a more general framework. We determine a conserved Hassan–Rosen (HR) bimetric theory (introduced in sec- vector current and state two propositions helpful for un- tion II), where this geometrical framework naturally ap- derstanding the structure of spacetime symmetries in the pears. When two metrics are concerned, the question of theory. We define the concepts of “bimetric spacetime whether an isometry of one metric is a spacetime sym- symmetry”, “sectoral spacetime symmetry” and “narrow metry for the whole system arises. We will see that, if spacetime symmetry” in bimetric relativity, as symme- the two metrics share their isometries, then the latter tries shared by all tensors, by the tensors in one metric will be spacetime symmetries for all other tensor fields. sector only and by some generic tensors, respectively. In AppendixA, we provide some strategies for constructing bimetric ansatzes having different KVFs. In section III, ∗ [email protected] we fit the results of section II into a wider perspective. † [email protected] In particular, we summarize the main geometric concepts ‡ [email protected] and results we have used in studying spacetime symme- § [email protected] tries. We explore the topic in a generic setup, i.e., we do 2

µ ρ µρ not assume any relation between the two metrics, other or, in index notation, S ρS ν := g fρν . The βn pa- than they are defined on the same differentiable manifold. rameters are generic real numbers and the en(S) are the µ After that, we state our conclusions. elementary symmetric polynomials of S ν , The reader specifically interested in HR bimetric the- ory can find all the related results in section II and Ap- e0 (S) = 1, pendixA. The paper being quite technical—especially e1 (S) = λ1 + λ2 + λ3 + λ4, section III—the reader interested mainly in the results e2 (S) = λ1λ2 + λ1λ3 + λ1λ4 + λ2λ3 + λ2λ4 + λ3λ4, and their discussion is invited to read the brief summaries at the end of section II and the beginning of section III, e3 (S) = λ1λ2λ3 + λ1λ2λ4 + λ1λ3λ4 + λ2λ3λ4, and the conclusions. e4 (S) = λ1λ2λ3λ4, (4)

µ where the λn are the eigenvalues of S ν . II. ISOMETRIES IN THE HASSAN-ROSEN Varying (1) with respect to gµν and fµν , we get the BIMETRIC THEORY bimetric field equations,

m4 1 Recently, there has been a noteworthy interest in con- µ µ µ Gg ν + 2 Vg ν = 2 Tg ν , (5a) structing a classical nonlinear theory of interacting spin- Mg Mg 2 fields. This interest culminated in the discovery of m4 1 G µ + V µ = T µ , (5b) two such theories which are free from the pathologi- f ν M 2 f ν M 2 f ν cal Boulware–Deser ghost [6], thanks to the particular f f µ form of the interaction potential between the spin-2 ten- with Gg,f ν being the Einstein tensors for gµν and fµν , µ sor fields. These theories are the de Rham–Gabadadze– Tg,f ν being the stress–energy tensors for the two inde- µ Tolley (dRGT) massive gravity [7,8], which was proven pendent matter sources and Vg,f ν , here referred to as to be ghost-free in [9], and the Hassan–Rosen (HR) bi- the “tensor potentials.” being equal to, metric theory [10, 11], whose unambiguous definition and spacetime interpretation are provided in [12]. 3 n µ X X n+k n−k We focus on the HR bimetric theory, describing two Vg ν = βn (−1) ek(S) S , (6a) interacting, dynamical and symmetric spin-2 fields, gµν n=0 k=0 3 n and fµν . It has seven degrees of freedom propagating µ X X n+k −1 −n+k five massive and two massless mode around proportional Vf ν = β4−n (−1) ek(S ) S . (6b) backgrounds [13–15]. One possible interpretation of this n=0 k=0 theory is to consider both the symmetric spin-2 fields Combining (5) with their own traces we obtain, as metrics defined on the same differentiable manifold. Here, we follow this interpretation and, therefore, aim to 1  1  1  understand the relations between the KVFs of the two R µ = m4 V δµ − V µ + T µ − T δµ , g ν M 2 2 g ν g ν g ν 2 g ν metrics and, consequently, between their isometries. g (7a)

A. Review of the theory, and topological constraint     µ 1 4 1 µ µ µ 1 µ Rf ν = 2 m Vf δ ν − Vf ν + Tf ν − Tf δ ν , Mf 2 2 The action of HR bimetric theory is [10], (7b) Z √ 1  4 2 4 αβ αβ αβ S = d x −g M Rg + Lg − m V (S) where T = g T , T = f T , V = g V and 2 g g αβ g f αβ f g αβ g V = f V αβ. 1  f αβ f 2 Noting that the stress–energy tensors T µ and + det (S) Mf Rf + Lf (1) g ν 2 µ Tf ν are divergence-less due to diffeomorphism in- R 4 √ where two Einstein–Hilbert actions constitute kinetic variance of the matter actions d x −g Lg and R 4 √ terms for both metrics, minimally coupled to two inde- d x −g det (S) Lf , the Bianchi constraints follows pendent matter sources described by the lagrangian den- from taking the divergence of (5), sities L , L (see, e.g., [16] regarding ghost-free matter g f ν ˜ ν couplings). The “bimetric potential” is defined as, ∇ν Vg µ = 0, ∇ν Vf µ = 0, (8) 4 where ∇ is the compatible covariant derivative of g X µ µν V (S) := βnen (S) . (2) and ∇˜ µ is the compatible covariant derivative of fµν . n=0 The following algebraic relation holds between the ten- Here, S is the square root matrix defined by, sor potentials and the bimetric potential [17],

1 µ µ µ S := g−1f
2 , (3) Vg ν + det (S) Vf ν = V (S) δ ν , (9) 3 which implies, fµν , with an anisotropic fluid coupled to gµν . The three cases concern metrics having different isometry groups. Vg + det (S) Vf = 4 V (S). (10) Note also that the solutions with the inhomogeneous and the anisotropic perfect fluids fit in the discussion in sec- 4 R 4 √ The interaction action −m d x −g V (S) is invariant tion II C about the definition of a spacetime symmetry under generic diffeomorphisms. This implies that the two in HR bimetric theory. Bianchi constraints are not independent, We remark that the topological constraints described in the previous section should be taken into account also ∇ V ν = − det (S) ∇˜ V ν . (11) ν g µ ν f µ for these solutions. This was proved in [18] for diffeomorphisms equal to the One way of finding solutions in GR is to put some as- identity at the boundary of the integration domain, and sumptions on a metric and then generate the correspond- µ 2 µ can be generalized to generic diffeomorphisms, thanks to ing stress–energy tensor defined through T ν := Mg G ν . the algebraic identity (9). This is called the “Synge’s method” [25, Chapter VIII]. Topological constraint. The importance of topology It is worth to stress that one has to be very lucky to when considering particular solutions in HR bimetric the- find an “acceptable” or non-pathological stress–energy ory was briefly pointed out in [19]. Here we extend the tensor with this method. A typical example is the gener- discussion, which is independent of any field equations alized Vaidya case where one assumes a specific form of and it is applicable to every theory involving more than the mass function, then trying to interpret the resulting one metric tensor on the same differentiable manifold. stress–energy tensor (an arbitrary mass function will not The action (1) is defined on a differentiable manifold, work in general, but some will be “sensible”; see section on which we have a set of smooth charts covering it (an 6 in [26]). In the bimetric case, we can similarly put µ atlas). The interaction term in the action constrains some assumptions on both gµν and S ν (or alternatively µ µρ the theory to have only one diffeomorphism invariance gµν and hµν so that S ν = g hρν ) and then generate µ µ (see, e.g., [18, 20]).1 Therefore, the chosen atlas must Tg ν and Tf ν using the bimetric field equations (5). This be shared by both metrics. Since an atlas uniquely de- is how the axially symmetric solution was obtained in termines the topology of the manifold [22, p. 20], the [27]. Moreover, one can also require some of the tensors two metrics must be compatible with this topology. We to vanish identically, restricting the ansatz; for instance, µ µ µ stress that topology is not determined by any field equa- Vg ν = Vf ν = 0 or one can require Tg ν = 0 but keeping µ tions, which, being differential, are always local [23, Fig- arbitrary Tf ν 6= 0. Of course, in the end the solution de- ure 31.5], [24]. pends on the judgment and taste of what is physical or We will clarify this issue in the next subsection when “sensible,” though some consistency checks must always discussing explicit examples. be satisfied, like conservation laws. Now, in general, gµν µ can have some collineations where S ν can spoil them by removing or introducing new collineations. Hence, gµν , µ µν µν B. Explicit examples with different KVFs fµν , S ν , Tg and Tf can have different isometries. We discuss three examples: We start out by discussing some explicit examples dis- (i) a vacuum solution of the bimetric field equations, playing different KVFs or different isometries, or both. In section II C, we see how they fit into the general re- (ii) a solution with two independent matter sources sults. coupled with gµν and fµν in a very peculiar way, In the literature, solutions of this kind were presented (iii) Einstein vacuum solutions with a freely specifiable in [4,5]. In particular, [4] considers a non-bidiagonal scalar function in one metric. ansatz with a Friedmann–Lemaˆıtre–Robertson–Walker (FLRW) gµν (homogeneous and isotropic) and an inho- We emphasize that these examples are meant to show mogeneous fµν , with a homogeneous perfect fluid cou- different possible configurations of spacetime symmetries pled to gµν . In their analysis, the authors of [5] discuss in HR theory, rather than being useful as descriptions of (i) bidiagonal cosmological solutions with an FLRW gµν physical systems. (homogeneous and isotropic) and a Lemaˆıtre fµν (only isotropic) with an inhomogeneous perfect fluid coupled to gµν , and (ii) an FLRW gµν and a Bianchi Type I Example I: Bi-Einstein vacuum solutions

The first example constitutes a family of non- bidiagonal solutions including black holes (BHs) which 1 Without the bimetric potential, the action (1) reduces to two are allowed to have different Killing horizons. In the decoupled copies of GR. In such case, the gauge group G of the theory would be the direct sum of two separate diffeomorphism bidiagonal case, corresponding to the Type I configura- groups, one acting on the g-sector only and the other on the f- tion in [12], this is forbidden by the proposition in [28] sector only [21]. The bimetric potential reduces G to its diagonal and its extension in [29, Appendix C]. The following so- subgroup [18]. lution being non-bidiagonal, is allowed to have different 4

Killing horizons. The same family of solutions was found The Ricci scalars are constants and the Kretschmann −6 g f in [13], but here we stress the fact that some of them, scalars diverge as r when r → 0, if rH and rH are although satisfying the bimetric field equations, must be non-zero. excluded due to the topological constraint presented in Both metrics can be diagonalized (not simulta- section II A. In addition, we discuss the solution in a new neously) to assume the usual Schwarzschild–anti–de perspective, analysing the isometries of the two metric Sitter/Schwarzschild–de Sitter (SAdS/SdS) form, by the sectors. general coordinate transformations, for gµν , We assume a spherically symmetric ansatz for the two −1 metrics and work in the outgoing Eddington–Finkelstein dv = dt˜+ G d˜r (16a) chart (v, r, θ, φ), dr = d˜r, (16b)   −G(r) 1 0 0 and for fµν ,  1 0 0 0  gµν =  2 , −1 ˆ −1
 0 0 r 0  dv = R0 dt + F dˆr (17a) 2 2 0 0 0 r sin (θ) −1 dr = R0 dˆr. (17b) −e2q(r)F (r) e2q(r) 0 0  e2q(r) 0 0 0 If we do not take into account the topological con- f =  , µν  0 0 R2(r)r2 0  straint of section II A, we can have many possible metric   combinations arising from (13) and (14). The horizon 0 0 0 R2(r)r2 sin2(θ) radii rg and rf are integration constants of the solu-  q(r)  H H e 0 0 0 g f tions, as in GR. If rH 6= 0 and rH = 0, we will have the 1 q(r) q(r)  µ  e (G(r) − F (r)) e 0 0  SAdS/SdS solution in gµν and the anti-de Sitter/de Sit- S ν = 2 , (12)  0 0 R(r) 0  ter (AdS/dS) in fµν , which do not share their isometry 0 0 0 R(r) groups. AdS and dS are both maximally symmetric [30], g f whereas SAdS and SdS are not. For rH = rH = 0, we can µ −1 where S ν is the principal square root of g f. still have AdS in one sector and dS in the other, depend- Assuming that G 6= F , from the field equations (5a), ing on the values of Λf and Λg, again implying different isometry groups. Also, one can fix two of the β parame-  q  ters, say β0 and β4, to set Λg = Λf = 0. In that case, we 1 2 R(r) ≡ R0 = −β2 ± β2 − β1β3 , have two Schwarzschild solutions having the same isom- β3 etry groups, but different KVFs. The KVFs for some q(r) = log(R0), of these solutions are explicitly computed in a Wolfram rg Λ (R ; {β }) Mathematica 11 [31] notebook, which is attached to the G(r) = 1 − H + g 0 i r2, r 3 paper and available at this link (click here).2 2 All these combinations satisfy the bimetric field equa- m 2

Λg(R0; {βi}) = β1β2 − β0β3 + 2R0 β2 − β1β3 , tions. However, we must also take into account the space- β3 (13) time topology, as explained in section II A. Since we use the same atlas for both sectors, they must have a common where i runs from 0 to 3. This solutions also satisfies the topology. To be precise, our Eddington–Finkelstein chart Bianchi constraints (8). Substituting the expressions in (v, r, θ, φ) does not cover the whole spacetime. However, (13) in the field equations (5b) results in, since the Penrose diagrams of all the possible solutions described by (13)-(14) are known [32–34], we know their f topology. Hence, we can build an atlas covering the whole r Λf (R0; {βi}) F (r) = 1 − H + r2, spacetime, uniquely determining the topology. As a re- r 3 sult, the remaining possible solutions are written in Ta- m2 Λf (R0; {βi}) = ± ble I. 3κ (2β2 + R0β3) We now discuss the relation between the KVFs and    β2β4 the isometry groups of the possible solutions. If gµν and × 2β1 β3 − β3 fµν are different, the isometry groups and the KVFs are  2  also different. Consider the case where gµν and fµν be- 4β2 β4 +R0 3β2β3 + β1β4 − . long to the same “class”— i.e., they have the same form, β3 but different numerical values of their parameters. First, (14) when 0 6= Λg 6= Λf 6= 0, the KVFs of gµν and fµν are dif- ferent, because they depend on the specific values of the The scalar invariants of the square root matrix are always finite, e.g.

4 2 Tr(S) = 4R0, det(S) = R0. (15) Explicitly, the link is https://tinyurl.com/y9r7nuuh. 5

f µν M S AdS dS SAdS SdS gµν

4 R M × • × × × 2 2 R × S S × × × • × 1 3 S × R AdS • × × × × 1 3 R × S dS × × × × × 1 1 2 S × R × S SAdS × • × × × 1 1 2 R × S × S SdS × × × × × Topology Solution 

Comments: (i) “M” stands for Minkowski and “S” stands for Schwarzschild. (ii) For AdS and SAdS, one can consider the universal covering in the timelike direction in order to 1 1 avoid violation of causality, “unwrapping” S into R . This would eliminate the closed timelike 4 curves present in AdS and SAdS, and would assign them the topologies, respectively, R and 2 2 R × S [32, 34]. 4 (iii) • stands for if we take the universal coverings of AdS and SAdS, having topologies R and 2 2 R × S (see next comment), otherwise it stands for ×. 2 (iv) S in the spatial topology of Schwarzschild, SAdS and SdS is necessary due to the presence of an event horizon, for stationary black holes (then, it holds also for static black holes) in 4 dimensions [2, p. 323],[35]. (v) Note that the topologies can be matched in some cases. For example, one can have a punctured 4 Minkowski with R \{r = 0} rather than M in the M and S case.

TABLE I. Metric combinations for the solution (13)–(14), which are allowed and forbidden by the topological constraint. cosmological constants (see e.g. [36] and [37] for the ex- agonal ansatz, plicit expressions of the KVFs of, respectively, AdS and −1 0 0 0  dS). However, their isometry group is the same. Second, 0 1 0 0 when Λ = Λ = 0, we have the bi-Schwarzschild or the g =  , g f µν  0 0 r2 0  bi-Minkowski solution. In this case, the isometry groups   0 0 0 r2 sin2(θ) are again the same, but the KVFs are different and re- lated by a diffeomorphism ϕ. If ξµ and ηµ are KVFs of −τ 0 2(t) 0 0 0  0 2 fµν and gµν , respectively, we have  0 ρ (r) 0 0  fµν =  2 ,  0 0 ρ (r) 0  µ β α µ ν α 2 2 ξ (ϕ (x )) = (ϕ∗) ν η (x ), (18) 0 0 0 ρ (r) sin (θ) τ 0(t) 0 0 0  with ϕ being the diffeomorphism. This means that the 0 ρ0(r) 0 0 Sµ =  , (19) components of the KVFs are different in the same chart ν  0 0 ρ(r)/r 0  xα = (v, r, θ, φ). We will return to (18) in section III. 0 0 0 ρ(r)/r µ −1 where S ν is the principal square root of g f and the prime means “derivative with respect to the argument.” µ Example II: Bi-Minkowski with “screened” matter sources The trace and the determinant of S ν are, 2ρ(r) Tr (S) = + ρ0(r) + τ 0(t), The relevance of this example resides in having met- r rics with the same isometry group, but different KVFs. ρ2(r) Also, it shows the possibility of having nongravitational det (S) = ρ0(r)τ 0(t), (20) r2 matter screened by the “effective stress–energy tensors” 0 0 µ µ so assuming τ (t), ρ (r) > 0, the square root is invertible Vg ν and Vf ν . and the pathologies described in Proposition 1 in [29, This example again exhibits a diffeomorphism between Section 3.1] will not appear. the g-sector and the f-sector, analogously to (18). We The change of coordinates which will put f in the consider two Minkowski metrics in a chart (t, r, θ, φ) µν standard Minkowski form is, which constitutes the usual spherical polar chart for gµν , 0 0 but not for fµν . In particular, we use the following bidi- dτ = τ (t)dt, dρ = ρ (r)dr. (21) 6

With the ansatz (19), the Bianchi constraints in (8) Analogous relations are valid for the other components are satisfied for every τ(t) and ρ(r). In vacuum, the field of the stress–energy tensors, even if they depend on time. equations (5) are reduced to, The metrics gµν and fµν share the SO(3) KVFs, since the coordinate transformation between the sectors does 4 4 m µ m µ not involve the angular coordinates [see (24)]. As we will 2 Vg ν = 0, 2 Vf ν = 0, (22) Mg Mf see in Proposition 2 in section II C, since the Lie deriva- µ tives of S ν with respect to the SO(3) KVFs are zero, the µ which is true only for zero tensor potentials Vg,f ν . How- SO(3) KVFs are collineations for all the tensors in this µ µ µ ever, we can couple two different generic stress–energy solution. However, the Lie derivatives of S ν ,Vg ν ,Vf ν µ µ tensors Tg ν and Tf ν to the metrics (see, e.g., [16]), with respect to all other KVFs of gµν and fµν are non- zero, so these KVFs are not collineations for the whole 4 4 m µ 1 µ m µ 1 µ system. 2 Vg ν = 2 Tg ν , 2 Vf ν = 2 Tf ν . (23) Mg Mg Mf Mf

µ µ and solve (23) for Tg ν and Tf ν . Note that the Example III: Einstein solutions with algebraically decoupled Bianchi constraints automatically imply that the stress– parameters energy tensors defined in this way are divergence- less. In addition, the stress–energy tensors always have With this example we want to show that requiring a 0 an energy density Tg 0(r) independent of time, and nonsingular geometry can constrain the KVFs of the met- a time-dependent isotropic non-homogeneous pressure, rics. As in the previous example, we consider the same 1 2 3 Tg 1(t, r) 6= Tg 2(t, r) = Tg 3(t, r). isometry group, but different KVFs. The field equations are thus formally satisfied, and the If we assume gµν to be an Einstein metric, then the µ two sectors share their isometry group, even if they have field equations in the g-sector are just Vg ν = 0. As ex- different KVFs. Indeed, (18) holds with the following plained in [19], this equation determines the (eigenvalues Jacobian matrix, of the) square root. There is a specific set of β-parameters µ for which Vg ν = 0 is satisfied with some of the square τ 0(t)−1 0 0 0 root eigenvalues λi left undetermined. We call this case 0 ρ0(r)−1 0 0 (ϕ ) µ = J −1
µ =  . (24) “algebraically decoupled”. Note that, even if we assume ∗ ν ν  0 0 1 0 gµν to be Einstein, fµν is not arbitrary. This is due to 0 0 0 1 the fact that, in HR theory, f = g S2 even if the metrics are dynamically decoupled. Hence, imposing V µ = 0 Two things should be remarked concerning this solu- g ν already determines f (or equivalently Sµ ), and makes tion. First, the two matter sources do not couple with µν ν the theory not equivalent to two separate copies of GR. the Einstein tensors of the metrics, but rather with the One algebraically decoupled case is when the two met- tensors potentials V µ and V µ . This does not mean g ν f ν rics are bidiagonal (Type I in [12]) and that we are decoupling gµν and fµν , retaining two copies of GR, because the Bianchi constraint (8) is an addi- 2 2 2 (β1β2 − β0β3) = 4(β1 − β0β3)(β2 − β1β3), tional requirement not present in GR. Therefore we can- 2 not choose any two metrics. Second, in this peculiar set β2 − β1β3 6= 0. (27) up we can have matter gravitationally decoupled from In this case, choosing the eigenvalues of the square root us, since its contribution is exactly canceled by tensor matrix to be potentials. We can require the stress–energy tensors not to diverge β1β2 − β0β3 at radial and time infinity or at finite values of t and r. λ1 = λ2 = λ3 = − 2 , (28) 2(β − β1β3) This can be achieved by setting, for example, 2 µ solves Vg ν = 0 with the last eigenvalue, λ4(x), left un- τ(t) = t + arcsinh(t), t ∈ µ R specified. The equation Vf ν = 0 becomes an equation ρ(r) = r + arcsinh(r), r > 0. (25) in the β-parameters and can be solved for β4. Let gµν be a specific Einstein metric; for definiteness This is a valid diffeomorphism from the Minkowski space- we let gµν be the Schwarzschild metric, time into itself, and the resulting stress–energy tensors 2 −1 2 2 2 are always finite and satisfy, gµν = −F dt + F dr + r dΩ , (29) g T 0 (r = 0) = ρ < ∞, 2 2 2 2 g 0 0 where dΩ = dθ + sin θdφ and F (r) := 1 − rH/r, T 0 (r = 0) = ρf < ∞, rendering the corresponding Einstein tensor to vanish. f 0 0 µ 0 g The only equation left to satisfy is Gf ν = 0, which lim Tg (r) = ρ∞ < ∞, r→∞ 0 becomes a set of differential equations for λ4(x). A set of 0 f solutions can be found through a separation of variables lim Tf (r) = ρ∞ < ∞. (26) r→∞ 0 ansatz. In particular, for such an ansatz, the arbitrary 7 function λ4(x) = λ(φ) solves the Einstein equation for obtain φ = φ(Φ), the scalars must be periodic in Φ with fµν . Hence the metric (29) together with period 2π. This is equivalent to φ(Φ = 2π) = φ(Φ = 0) + 2πn0 where n0 is a positive integer. 2 2 −1 2 2 2
2 2 2 2 fµν = λ1 −F dt + F dr + r dθ + λ (φ)r sin θdφ Note that we imposed Φ ∈ [0, 2π) since Φ is the az- (30) imuthal coordinate in the spherical polar chart of fµν . If Φ ∈ [0, Φ0) with Φ0 < 2π, we could identify the hypersur- solve the bimetric field equations, with λ(φ) being an faces Φ = 0 and Φ = Φ0 by demanding λ to have a period arbitrary function and with parameters specified above. of Φ0, but this would lead to a conical singularity in the Assuming that λ(φ) > 0, we can make the change of f-sector, as explained in [2, p. 214]. On the other hand, coordinates, if Φ ∈ [0, Φ0) with Φ0 > 2π, we would cover twice the region between 2π < Φ < Φ0. Then, we could restrict φ → Φ, dΦ = λ(φ)dφ (31a) Φ ∈ [0, 2π), but Φ and φ both being monotonic would t → T = λ1t (31b) lead to φ ∈ [0, φ0) with φ0 < 2π, i.e., a conical singu- larity in the g-sector. Introducing the conical singularity r → R = λ1r (31c) does not change the topology (since r = 0 is already not θ → Θ = θ (31d) part of the spacetime); however, the solution would not be Schwarzschild. where λ(φ) = Φ0(φ) and Φ(φ) is a monotonic function As an example, consider the functions Φ(φ) = φ + (and thus invertible). In the new chart, f is manifestly µν 1 sin φ and λ (φ) = Φ0(φ) = 1 + 1 cos φ. The function Schwarzschild, 2 4 2 λ4(φ) is strictly positive and periodic in φ with period   2 2π. Hence, Φ(φ) ∈ [0, 2π) is a monotonically increas- RH 2 dR fµν = − 1 − dT + ing function of φ, satisfying the above conditions. This R 1 − RH R function determines different KVFs for the metrics, ac- + R2 dΘ2 + sin2 ΘdΦ2
(32) cording to (34). Another example is when λ(φ) = c with c being a positive constant. In this case, λ(φ) is where we defined RH := λ1rh. Thus, fµν also being clearly periodic in φ and Φ = cφ. Concerning the prop- a Schwarzschild metric, gµν and fµν exhibit the same erty Φ(2π) = Φ(0) + 2πn, this demands that c = n is isometry group and determine the same topology (see a positive integer and φ(2π) = φ(0) + 2πn0 implies that section II A). The KVFs of gµν are 1/c = n0 is a positive integer. Hence c = 1, as discussed in [19]. η0 = ∂t, η1 = ∂φ, (33a) Note that this method of generating solutions did not η2 = cos φ ∂θ − cot θ sin φ ∂φ, (33b) rely on gµν being the Schwarzschild metric. The method η3 = − sin φ ∂θ − cot θ cos φ ∂φ (33c) generalizes straightforwardly to other Einstein metrics and other sets of β-parameters which have the property and the related KVFs of fµν are of leaving one or more of the square root eigenvalues λi undetermined. 1 λ1 ξ0 = ∂T = ∂t, ξ1 = ∂Φ = 0 ∂φ (34a) λ1 Φ (φ) C. Collineations in the Hassan-Rosen bimetric λ1 ξ = cos(Φ(φ))∂ − cot θ sin(Φ(φ))∂ , (34b) theory 2 θ Φ0(φ) φ λ 1 In this subsection we study general properties of space- ξ3 = − sin(Φ(φ))∂θ − 0 cot θ cos(Φ(φ))∂φ (34c) Φ (φ) time symmetries in HR bimetric theory. We start by reminding that the KVFs of a metric g If λ(φ) is non-constant, only the KVF generating time µν are defined to be the solutions of the Killing equation, translations is a KVF for both metrics. A final check reveals that the scalar invariants of the Lηgµν = ∇(µην) = 0, (35) square root S = diag(λ1, λ1, λ1, λ4(φ)) are all non-zero µ and finite, assuming that λ1, λ4(φ) > 0, i.e., the principal with Lη the Lie derivative along the vector field η and square root branch. ∇µ the covariant derivative compatible with gµν . The Contracting the KVFs (33) and (34) in different combi- parentheses are understood as the symmetrisation of the nations with the metrics yields a set of scalar fields, which indices they enclose. must be periodic in φ so that φ = 0 and φ = 2π give the The KVFs are completely specified by [2] same value. Demanding this to be the case is equivalent ρ ρ to λ(φ) being periodic in φ, i.e., λ(φ) = λ(φ + 2π), and v ∇ρηµ = v Lρµ, (36a) ρ ρ σ Φ(φ + 2π) = Φ(φ) + 2πn where n is an integer. How- v ∇ρLµν = −v Rµνρ ησ, (36b) ever, Φ(φ) is a monotonic (increasing) function of φ so n must be positive. Conversely, we must demand that if where Lµν := ∇µην is antisymmetric due to the Killing µ we invert the relation between the azimuth coordinates to equation Lµν = −Lνµ. The non-zero vector field v spec- 8

    ifies the integration path along which the differential sys- 1 4 1 ν ν ν 1 ν = 2 m Vg δµ − Vgµ + Tgµ − Tgδµ ην , tem is solved. As shown in [2], one can make use of (36b) Mg 2 2 to write (43a) 2 ν ρ ν ν ∇ ηµ = ∇ν L µ = −R µρ ην = −Rµ ην , (37) 2 f ν − ∇˜ ξµ = R µ ξν 2 ν     where ∇ := ∇ν ∇ . We can raise the index µ to obtain, 1 4 1 ν ν ν 1 ν = m Vf δµ − Vf µ + Tf µ − Tf δµ ξd. 2 µ µ ν M 2 2 2 ∇ η = −R ν η . (38) f (43b) Spacetime symmetries and field equations. A known result in GR is that a KVF is also a collineation for Therefore in vacuum, contrary to GR, the Laplacian of the Einstein tensor, the Riemann tensor and the stress- a KVF is not necessarily zero because of the contribu- energy tensor [3], tions from the tensor potentials [see (37)]. This reflects the more complicated structure of the theory, and makes β Lη gµν = 0 =⇒ Lη Rµνα = 0 the search for KVFs for vacuum solutions more difficult. µ =⇒ Lη Gµν = 0 However, for a generic vector field V and independently of any field equations, the “Komar identity” holds [38, 39] =⇒ Lη T µν = 0. (39) [ν µ] The first two results can be straightforwardly extended ∇µ∇ν ∇ V ≡ 0, (44) to the HR bimetric theory, whereas the third result can which, for a KVF, becomes not since it uses the Einstein equations. µ 2 µ
Consider the KVF η of the metric gµν (analogous ∇µ ∇ η ≡ 0, (45) computations can be performed for the f-sector). The Lie derivative of the field equations (5a) with respect to because ∇[ν ηµ] := (∇ν ηµ − ∇µην ) /2 = ∇ν ηµ due to the ηµ is, Killing equation (35). One could solve (45) for KVFs in 2 µ 4 µ µ the same way one would use (38) in GR. An analogous Mg Lη Gg ν + m Lη Vg ν = Lη Tg ν . (40) relation holds for the f-sector. An alternative proof for µ (44), simpler than the one in [39], is provided in appendix We know that Lη Gg ν = 0, hence, B. 4 µ µ m Lη Vg ν = Lη Tg ν . (41) So far, we have not assumed anything regarding the KVFs. Now we assume that the two metrics share a Therefore, in general, a KVF of one metric is neither a KVF Xµ. Then, substituting (7), (9) and (10) in (43a) collineation of the respective stress–energy tensor nor a and performing some algebra one has, collineation of its tensor potential (we showed this explic- 2 2 µ 2 ˜ 2 µ 4 µ itly in section II B). Also, this opens up the possibility Mg ∇ X + Mf det (S) ∇ X = −m V (S)X . (46) of finding solutions having a given KVF for one metric µ µ Fulfilling (46) is then a necessary condition for Xµ to be but different collineations for both Vg ν and Tg ν . As already mentioned, some solutions having this property a KVF for both metrics. It is an additional constraint to were presented in [4,5]. Therefore, we notice that the (45) when solving for shared KVFs. definition of a spacetime symmetry in HR bimetric the- Conserved currents determined by KVFs. Now con- µν µ ory is non-trivial. Suppose having a static and spherically sider the vector current Vg ην , where η is a KVF of µ µ g . Its divergence is zero, symmetric gµν and an axially symmetric Tg ν and Vg ν . µν Whether this should be considered a spherically symmet- ∇ V µν η
= η ∇ V µν + V µν ∇ η = 0, (47) ric system or an axially symmetric one remains an open µ g ν ν µ g g µ ν question. However, in “bimetric vacuum” (see [27] for and therefore it is a conserved current. The last equality µν a discussion about vacuum in HR bimetric theory), (41) follows from the Bianchi constraint for Vg , the symme- µν becomes, try of Vg and the antisymmetry of Lµν = ∇µην (the same result can be stated for the f-sector). This con- V µ = 0, (42) Lη g ν servation law is the same as that for the stress–energy and the KVFs of the metric are also collineations for the tensor corresponding tensor potential. ∇ T µν η
= 0, (48) In the following we analyze some properties of the µ g ν KVFs, useful for determining them in HR bimetric the- which is also true in GR.3 Considering the g-sector, we ory. First, (36) does not make use of the Einstein equa- can rewrite (47) in the following way, tions, so it holds for gµν and fµν separately. However, √ µν
the difference with GR is not only that we have sepa- ∂µ −gVg ην = 0. (49) rate differential systems for each metric. Inserting the bimetric field equations (7) in (37), we get, 2 g ν 3 Their validity is shown in the same way as for the current V µν η . − ∇ ηµ = R µ ην g ν 9

Separating the time and spatial derivatives and integrat- Proof. Since F is invertible, we can use the previous ing over a spacelike hypersurface V defined by x0 = Lemma in both directions. const., we obtain, Proposition 1 does not apply to our case, because the Z Z µ 3 √ 0ν
3 √ iν
tensor potentials, containing traces of S ν , are not in- ∂0 d x −gVg ην = − d x ∂i −gVg ην vertible functions of it. Therefore V V Z √ iν V µ = 0 =6 ⇒ Sµ = 0. (55) = − dσi −gVg ην , (50) Lξ g/f ν Lξ ν ∂V Therefore even in vacuum, as we saw in section II B, we µν where dσi is the two-surface element on ∂V. When Vg can have different KVFs in the two sectors. In such a tend to zero at spatial infinity, e.g., in asymptotically flat case, the isometry of gµν are spacetime symmetries for spacetimes with two Minkowski metrics, we can define tensors in the g-sector and the isometries of f will be 4 µν the conserved charge, spacetime symmetries for tensors in the f-sector. µ Z If gµν and S ν share their collineations, fµν shares 0 3 √ 0ν
0 V := d x −gVg ην , ∂0V = 0. (51) them too. This is a special case of a more general result V we are going to introduce after the following definitions. Regarding the meaning of V 0, we can consider an asymp- Definition 1. Consider metric fields g and f. Let g−1f totically flat static spacetime with a√ stress–energy tensor. be positive definite. We define the one-parameter set Then, both V 0 and P 0 := R d3x −g T 0ν η
would be −1 α V ν of Lorentzian metrics Γ = {hα = g(g f) , α ∈ R}, separately conserved. Therefore the total energy of the where the matrix power function is defined through Xα = 5 system would be the sum of the two, i.e. exp (α log X). Notice that g = h0 and f = h1.

0 0 µ E = P + V . (52) Definition 2. If the vector field ξ is a KVFs for gµν and fµν , we define it as a “bimetric isometry”. The defi- 4 00 Hence, in this case m Vg can be interpreted as the en- nition extends to any type of spacetime symmetries (e.g., ergy density due to the tensor potential. conformal vector fields).6 Structure of spacetime symmetries. We now state µ several claims which clarify the behavior of the isome- Definition 3. If ξ is a spacetime symmetry for all the tries and can help to define a symmetric spacetime in tensors in one metric sector, then we refer to it as a “sec- the HR bimetric theory. toral spacetime symmetry”. We start by stating the following (in matrix notation) Definition 4. If ξµ is a collineation for some tensors, we call it a “narrow collineation”. The definition extends to Lemma. Let ξ be a vector field, S a (1,1)-tensor field any spacetime symmetry, not only collineations. and F a matrix-valued function. If the Lie derivative of S with respect to ξ vanishes, then the Lie derivative of We now state F (S) vanishes too; that is, LξS = 0 implies LξF (S) = 0. Proposition 2. Consider a vector field ξ and two ar- Proof. Locally, one can always find a chart where a coor- bitrary hα and hβ such that Lξhα = 0 and Lξhβ = 0. 0 µ a a dinate x is aligned along ξ so that ξ = δ0 , Then Lξhγ = 0 for any γ ∈ R. µ ρ µ µ ρ ρ µ −1 LξS ν = ξ ∂ρS ν − (∂ρξ )S ν + (∂ν ξ )S ρ Proof. We define A = g f for readability. µ Our hypothesis is = ∂0S ν . (53)

0 Lξhα = Lξhβ = 0. (56) If LξS = 0, then S is not dependent on x , since ∂0S(x) = 0. Consequently F (S)(x) does not depend on Suppose α < β (otherwise switch them). We have 0 x , and ∂0F (S)(x) = 0. α α 0 = Lξhα = (Lξ g) A + g LξA , (57) Thanks to this Lemma, we know that which is equivalent to µ µ Lξ S ν = 0 =⇒ Lξ Vg,f ν = 0. (54) α α (Lξ g) A = −g LξA . (58) Proposition 1. If F is invertible then, LξS = 0 if and only if LξF (S) = 0.

5 The geometric mean of two symmetric matrices A and B is de- −1 1/2 fined by A # B = A(A B) . For any two hα and hβ we have 4 In HR bimetric theory, asymptotically flat spacetimes do not nec- hα # hβ = h(α+β)/2. See [12] for more details. essarily have two Minkowski metrics with the same components 6 According to this definition, the conformal vector fields found in [19]. [27] are bimetric. 10

From (56) and (58) it follows (b) The two metrics have different KVFs.

β β i. In vacuum, tensors in the g-sector share 0 = Lξhβ = (Lξ g) A + g LξA the symmetries with gµν , and those in the α β−α α β−α
= (Lξ g) A A + g Lξ A A f-sector share them with fµν . α β−α α β−α = −g (LξA ) A + g (LξA ) A ii. In the presence of stress–energy tensors, α β−α
tensors in the g-sector need not share the + g A Lξ A symmetries with gµν , and those in the f- β−α
= hα Lξ A . (59) sector need not share them with fµν . The matrix power function is invertible, so (59) and (iii) Case (a) can certainly be considered as an “au- Proposition1 imply thentic” spacetime symmetry in bimetric relativ- ity, i.e., a spacetime symmetry under which all the β−α
Lξ A = 0 =⇒ LξA = 0, (60) tensors in the theory are invariant. This is always the case in GR, since an isometry of the metric which, thanks to the Lemma, implies is a collineation for the Riemann, Ricci, Einstein γ and stress–energy tensor. However, we also high- LξA = 0 =⇒ Lξ (A ) = 0, ∀γ ∈ R. (61) light the other possibilities in bimetric relativity. Then, (58) tells us For example, consider two stationary metrics with different timelike KVFs in vacuum; in this case, it α (Lξ g) A = 0 =⇒ Lξ g = 0. (62) would be meaningful to consider this as a stationary spacetime, even if the KVFs are different. Hence, Therefore, in general, one may define an authentic spacetime γ γ symmetry in HR bimetric theory referring to the Lξhγ = (Lξ g) A + g Lξ (A ) = 0, (63) minimal symmetry group shared by the metrics, ∀γ ∈ R. and not to the associated symmetry vector fields. Corollary. An isometry is bimetric if, and only if, it is (iv) We proved that the two metrics share their KVFs an isometry for any two metrics in the set Γ. if, and only if, any two of the metrics in the set Γ =  −1
α hα = g g f : α ∈ R have the same KVF. Proposition2 tells us that, if any two of the hα share µ µ the KVFs, then every tensor field in the theory will be (v) If η and ξ are KVFs of gµν and fµν , respectively, invariant under the corresponding transformation. The then the following vector currents are covariantly corollary tells us that this is a bimetric spacetime sym- conserved, metry, and the whole spacetime is invariant under the transformation. In the opposite case, instead, the def- µν
˜  µν  ∇µ Vg ηµ ≡ 0, ∇µ V ξν ≡ 0, inition of spacetime symmetry needs more care. Note f that, when stress–energy tensors are present, tensor fields µν where V µν and V are the contributions from the within the same sector can have different collineations g f µ bimetric potential to the field equations [see (5)]. [compare with (41) in which the collineations of Vg ν and T µ are not necessarily isometries of the metric This is analogous to the current determined by a g ν stress–energy tensor, gµν ; they are narrow symmetries]. However in vacuum, thanks to (42), tensors within the same sector share the µν
˜  µν  collineations, which are then sectoral. ∇µ Tg ηµ ≡ 0, ∇µ Tf ξν ≡ 0.

In asymptotically flat spacetime, this leads to defin- D. Brief summary of section II ing an energy density contribution from the tensor µν µν potentials Vg and Vf , see (51)–(52). Here we summarize the main results of this section.

(i) We presented example solutions of the bimetric III. GEOMETRIC BACKGROUND AND field equations having different KVFs, different GENERAL ANALYSIS isometry groups or both; other methods for find- ing ansatzes with these properties are described in In this section, we study spacetime symmetries on a AppendixA. differentiable manifold endowed with two generic met- (ii) In HR bimetric theory, isometries arise in three con- rics, without imposing any field equations. We only make figurations: geometrical considerations and generalize the analysis in- troduced in the first section. A short non-technical sum- (a) The two metrics share the KVFs, and the lat- mary of the results is provided below, for the convenience ter define symmetries for all other tensors. of the reader not interested in the technicalities. 11

We want to understand if, starting with one metric deduce how the covariant derivatives of a tensor of any having some isometries, it is possible to constrain the rank relate, and how the Riemann and Ricci tensors de- isometries of the second metric. The results are: termined by the two covariant derivatives relate,

˜ σ σ σ α σ (i) Consider having two metrics on the same space- Rµνρ = Rµνρ − 2C [µ|αC |ν]ρ + 2∇[µ|C |ν]ρ, time, and suppose one metric has some symme- R˜ = R − 2Cα Cβ + 2∇ Cα , (66) tries. Then, the second metric need not to share µν µν [µ|β |α]ν [µ| |α]ν the symmetries with the first. Relations between where [µ|...|ν] denotes the antisymmetrization of µ and ν the KVFs of the two metrics can always be found, only. We notice that, in a coordinate basis, we can write but without additional requirements regarding the symmetries, these relations allow for any possible α α α C µν = Γ µν − Γ˜ µν , (67) symmetry configuration. where the Γ’s are the Christoffel symbols of the covariant (ii) One possible requirement is that the metrics have derivatives ∇ and ∇˜ , respectively. If the two covariant KVFs with the same character (timelike, spacelike µ µ derivative operators are compatible with two metrics g or null). For instance, a sufficient condition for the µν µ and fµν ,(67) can be rewritten as, metrics gµν and fµν to have timelike KVFs ξ and µ η (i.e., to be stationary) is that they satisfy (in 1   matrix notation) g = P T f P for some invertible P , Cα = gαβ ∇˜ g + ∇˜ g − ∇˜ g µν 2 µ βν ν µβ β µν and η = P ξ. The same is true if we require both 1 αβ metrics to be axially symmetric, and in general if = − f (∇µfβν + ∇ν fµβ − ∇βfµν ) . (68) we require KVFs with the same character. 2 (iii) Another requirement is that the metrics share some isometries. This allows us to relate the KVFs of the B. General properties of KVFs and the A-relation metrics. They need not to be the same, but they must satisfy the same commutation relations; e.g., Consider an open set of a differentiable manifold. Sup- if we require spherical symmetry for both metrics, pose ηµ and ξµ are two vector fields defined on the space- then the KVFs of both metrics must satisfy the time and non-vanishing in the open set. Then, suppose angular momentum commutation relations we can find a linear local map between them having the following form,7 ijk [Ji,Jj] =  Jk. (64) ξˆµ(xα) := ξµ(ϕβ (xα)) We also add here the topological constraint of sec- ∂ϕν = γµ ϕβ (xα)
(xα) ηρ (xα) tion II A, because it is as general as the previous state- ν ∂xρ ments. µ α ρ α = A ρ(x ) η (x ), (69) (iv) A spacetime can only have one topology, and the two metrics must be compatible with it. This re- where we have defined stricts the possible metric combinations, indepen- ∂ϕρ Aµ (xα) := γµ ϕβ (xα)
(xα), (70) dently on the symmetries. ν ρ ∂xν These outcomes can be applied to any modified theory to increase readability. In (69), we are applying two sep- of gravity relying on (pseudo-)Riemannian geometry. arate transformations to the spacetime and its tangent The rest of the section is quite technical; a discussion bundle. First, we apply the diffeomorphism ϕ to the of the results can be found in the conclusions. spacetime and we accordingly transform (push-forward) the vector field ηµ. Second, we apply a non-degenerate µ bundle map γ ν over the spacetime (see e.g. [40, p. 14] A. The relation between torsion-free covariant and [41, p. 116]).8 Following the terminology in [41, derivatives p. 87], we refer to ξµ and ηµ as “A-related” KVFs, where A refers to the composition of the two maps [see (70)]. On a differentiable manifold, given any two generic torsion-free covariant derivative operators ∇µ and ∇˜ µ, α we can define the tensor C µν , symmetric in µ and ν, such that: 7 The assumptions of non-vanishingness of the vector fields is not strictly necessary. We could allow for the vector fields to vanish ˜ α ∇µων = ∇µων − C µν ωα, (65) the same number of times in the open set. However, if they vanish a different number of times, the linear local map between with ωα any 1-form defined on the cotangent space of the them will not exist (a linear map must send 0 to 0). 8 µ manifold [2]. Knowing (65), one can straightforwardly A ν is a smooth automorphism of the tangent bundle onto itself. 12

µ We now assume η to be a KVF for gµν and write the as the standard vielbein does when one of the metric is Killing equation, Minkowski. Then, a similar treatment as the one per- formed in [42] is possible. There, it is shown that the Lηgµν = 2∇(µην) = 0, (71) standard vielbein must satisfy some specific constraints in order for the obtained metric to have a given set of with being the Lie derivative along ηµ, η := g ην , Lη µ µν isometries. In principle, as a standard vielbein connect and ∇ being the compatible covariant derivative oper- µ the Minkowski metric with a completely different and ator for g . We require ξµ to be a KVF for f , µν µν generic metric, the same holds in our set-up for the gen- ˜ eralized vielbein relating gµν and fµν . Therefore, there Lξfµν = 2∇ ξ = 0, (72) (µ ν) cannot be a general relation between the isometries of 11 ν the two metrics. with ξµ := fµν ξ and ∇˜ µ being the compatible covariant 9 derivative operator for fµν . Using (65) and (69) in (72), we can rewrite the Killing equation for f as follows, C. Constraints on the A-related KVFs and the µν isometry groups ˜ Lξfµν = 2∇(µξν) α
In this subsection we find the constraints on the map = 2 ∇(µξν) + C µν ξα µ A ν in (69) by requiring the KVFs of the two metrics  ρ σδ
α ρ σδ  = 2 ∇(µ fν)ρA σg ηˆδ + C µν fαρA σg ηˆδ to share certain properties. In particular, the properties will concern (i) the character of the KVFs and (ii) the = 2 ∇ f Aρ gσδ
+ Cα f Aρ gσδ ηˆ (µ ν)ρ σ µν αρ σ δ Lie algebra that they generate (we refer the reader to ρ σδ
+ 2 f(ν|ρA σg ∇|µ)ηˆδ . (73) [43, 44] for a rigorous treatment of the Lie groups and Lie algebras on a (pseudo-)Riemannian manifold). 10 Defining the tensor, Let’s consider case (i) first. We explore the relation δ ρ σδ between the character (timelike, spacelike, or null) of two Υν := fνρA σg , (74) µ A-related KVFs and how it depends on the map A ν in µ (73) becomes, (69). Specifically, we discuss one condition that A ν can satisfy in order for the A-related KVFs to have the same h δ α δi character with respect to their associated metric. Lξfµν = 2 ∇(µΥν) + C µν Υα ηδ This issue is of physical interest because it concerns δ
+ 2 Υ(ν| ∇|µ)ηδ = 0. (75) whether two metrics share or not some KVFs, both being timelike, spacelike or null with respect to the pertinent Note that we have not used the hypothesis ∇(µην) = 0, metric. For example, if the two metrics have different because it never appears explicitly in (75). Therefore, we timelike A-related KVFs, then they are both stationary can deduce that the isometries of gµν and fµν , in general, and could be both static (we will come back to this case are unrelated, as one might expect. later), but have different KVFs. Now, assuming (36) defines the KVFs of gµν , we can ξµ and ηµ having the same character with respect to find the general relation between them and their A- fµν and gµν means related KVFs of fµν by substituting (65) and (66) in the α β α β analog of (36) for fµν . However, we have not been able sign[gαβη η ] = sign[fαβξ ξ ]. (77) to deduce any information from the resulting equations other than that, in the generic case, the KVFs of two met- One case in which this relation is certainly satisfied is rics are unrelated, as already shown in (73). Nonetheless, when (in matrix notation) we have seen how to make use of (36b) in the case of HR bimetric theory. g = ATfA, (78) An interesting viewpoint is to consider the two metrics related by a “generalized” vielbein. In such an approach, i.e., A is a congruence between g and f. Stated dif- µ ferently, the map A between A-related KVFs is a local the generalized vielbein V ν would transform one metric into the other, e.g. isometry between g and f (i.e., it is an isometry locally at each point of the spacetime). Then, ρ σ fµν = V µ gρσV ν , (76) α β α β gαβη η = fαβξ ξ , (79)

9 The requirement of non-vanishingness for ηµ and ξµ is also mo- tivated by their utilization as KVFs of the metrics. If a KVF 11 The “generalized vielbein” is actually determined by the map vanishes at some point, a more detailed treatment should be in (69), which is not only a map between vector fields, but ac- carried out (see [3, Section 10.4]). tually between the two whole sectors. Indeed, we have already 10 δ µ δ The tensor Υν defined in (74) is the adjoint map of A σ. Specif- introduced the adjoint map Υν relating the 1-forms; then, every δ ically, ξν = Υν ηδ. tensor can be mapped from a sector into the other. 13

which guarantees that ξµ and ηµ always have the same D. Two particular cases character. However, we remark that this is a sufficient, but not necessary, condition to satisfy (77). In this subsection we discuss two relevant cases allow- Next, we consider case (ii). We address the question of ing us to understand more deeply our examples in sec- when two metrics share their isometry groups, i.e., they tion II B. determine the same spacetime symmetry. Even if the The first case, concerning an isomorphism between Lie µ µ µ two metrics have KVFs of the same character, we are algebras, is when γ ν = δ ν in (69) and the map A ν = µ µ not guaranteed that the metrics have the same isome- (ϕ∗) ν = (dϕ) ν is simply the differential map (push- try group. The simplest case concerns stationarity, be- forward) of a diffeomorphism ϕ from the spacetime to cause it only requires a timelike KVF [2]. Suppose that itself.14 Indeed, the push-forward is a linear map that µ one metric, for example gµν , possess a timelike KVF η . preserves the Lie bracket, (see Proposition 4.3.10 in [45]), µ Then, if A ν is a congruence between the metrics gµν i.e., for any two vector fields X and Y µ µ and fµν , the A-related KVF of η , i.e., ξ , will be time- like with respect to fµν . Therefore, both metrics will be ϕ∗ [X,Y ] = [ϕ∗X, ϕ∗Y ] , (81) stationary.12 However, this does not take into account staticity. Each of the metrics can be static (thus having i.e., ϕ-related KVFs. This case concerns metrics which a different isometry group), if, and only if, their time- have the same form in different coordinate systems (i.e., like KVFs are orthogonal to a congruence of spacelike related by a diffeomorphism). At first glance, this case hypersurfaces [2]. No general relation between the A- may seem trivial, but actually it is not, since the same related KVFs is manifest in this case. Therefore, one coordinate system can be, e.g. spherical for one metric metric could be only stationary and the other static; i.e., and not for the other. The second example and some different isometry groups, where the isometry group of cases within the the first example in section II B, as dis- a static metric is a subgroup of the stationary isometry cussed there, have ϕ-related KVFs, hence belonging to group, the latter including the time reversal. this category. In the most general case, i.e., when considering arbi- The second case we consider is when no diffeomorphism trary isometry groups, the relation between the KVFs in (69) is applied and

of two metrics can be analyzed in terms of the Lie al- µ µ µρ gebras of the isometry groups (which are Lie groups) of A ν = γ ν = f gρν . (82) the two metrics. Let {ηµ } , with δ dimension (n) n∈{1,...,δ} An example belonging to this category will be treated in of the isometry group, be a subset of the KVFs of gµν ν ν AppendixA. In this case, Υ µ = δµ and the two vector generating a Lie algebra defined by the Lie bracket, fields ξµ and ηµ have the same dual,

η , η  (F ) := η η (F )
− η η (F )
, ρ ρ (n) (m) (n) (m) (m) (n) ξµ = fµρξ = gµρη = ηµ. (83) (80) In addition, (75) reduces to, where m ∈ {1, ..., δ} and F is a C∞ scalar function de- h δ α δi δ
fined on the differentiable manifold. In order fµν to have Lξfµν = 2∇(µδν) + C µν δα ηδ + 2 δ(ν| ∇|µ)ηδ the same isometry group as gµν , a necessary condition is µ α α = C µν ηα + 2∇(µην) = C µν ηα that the set of A-related KVFs {A(ˆη(n))}n∈{1,...,δ} gen- erates a Lie algebra which is isomorphic to the one gen- ˜α α  = Γµν − Γ µν ηα = 0, (84) erated by the KVFs of gµν . This implies that the map µ A ν between A-related KVFs is a Lie algebra isomor- where, in the second line, we have used our original hy- phism. We remark that, even in the case of A-related pothesis in (71), i.e., ∇(µην) = 0. With (84) we show KVFs generating the same Lie algebra, they need not to µ µρ µ that, for A ν = f gρν , ξ is a KVF for f if, and only if, be the same. Indeed, a Lie algebra can have different gen- η is orthogonal to Cα .15 One particular solution of erators (i.e., different bases), and different choices deter- µ µν (84) is obtained when gµν and fµν share their compatible mine different KVFs, as we saw explicitly in the examples covariant derivatives, in section II B. Also, we recall that different Lie groups can have the same Lie algebra, and that is why Aµ be- α α ν ∇µ = ∇˜µ =⇒ Γ µν = Γ˜ . (85) ing a Lie algebra isomorphism is a necessary condition, µν but not sufficient to have the same isometry group.13

14 One can argue that this must be the case, because (active) dif- feomorphism of a manifold into itself can be thought as (passive) 12 This result relies on the fact that all 1-dimensional Lie algebras change of coordinates, which cannot modify the geometric prop- are isomorphic, because the Lie bracket is trivially zero. erties of the manifold. 13 15 α µ However, Lie groups sharing the Lie algebra have the same uni- Equivalently, (84) is an eigenequation for C µν and η must be α versal covering [44]. an eigenvector of C µν with eigenvalue zero. 14

It is well known that, given a metric on a differentiable In this case, the isometries of the metrics are spacetime manifold, its compatible covariant derivative is uniquely symmetries for the whole system and we call them “bi- determined (see e.g. [2, Theorem 3.1.1]). However, the metric symmetries.” It seems reasonable to consider bi- converse is not true, as was shown in [3, 46, 47]. Given a metric symmetries as “authentic” spacetime symmetries covariant derivative operator on a differentiable manifold, in bimetric relativity—i.e., symmetries for every tensors there can be more than one metric compatible with it. in the theory—but we stress that there are other possi- There are four possible cases, one of them being when bilities that can be considered as authentic. For instance, the two metrics are proportional (the most commonly we can have symmetries of one sector only, which we call encountered, according to [46, 47]). Arguably, this is “sectoral symmetries”, and symmetries of some tensors the least interesting case in physics, because, the metrics only, named “narrow symmetries”. Sectoral symmetries being proportional, their null cones (hence their causal can be regarded as authentic, in some cases. For exam- structures) are the same. ple, we can have two spherically symmetric metrics with different KVFs in vacuum (as we show in Example III in section II B). One can think about this as a spherically CONCLUSIONS symmetric spacetime. Therefore, it seems reasonable to define an authentic spacetime symmetries as the minimal We study spacetime symmetries and topology in the symmetry group shared by the sectors, without reference HR bimetric theory, where two metrics gµν and fµν are to the associated symmetry vector fields. defined on the same manifold. We determine the con- We clarify that a topological constraint limits the num- ditions for the metrics to share their isometries, but we ber of conceivable metric combinations. The unique dif- note they can also have different isometries. In detail, feomorphism invariance of the theory allows us to have we present a proposition stating that, if any two metrics only a single set of coordinate charts covering the space-  −1
α within the set hα := g g f : α ∈ R have the same time. Such a set is compatible with one, and only one KVF, then all other fields in the theory must have the topology, and both metrics must be compatible with it. same KVF. Also, we find a differential equation which We present examples showing the relations between determines a KVF shared by both metrics. the spacetime symmetries, the KVFs and the topologies We point out that many properties of spacetime sym- of the two sectors. We provide several methods to de- metries, valid in GR, are not valid in HR bimetric theory. termine possible ansatzes with metrics having different For instance, an isometry of one metric is not necessar- isometry groups, different KVFs or both in AppendixA. ily a collineation for the minimally coupled stress–energy Interesting open questions remain unanswered in our tensor, due to the tensor potentials V µ and V µ in the g ν f ν analysis. First, it would be interesting to find non- bimetric field equations. Also, in vacuum, a KVF ξµ does GR (analytical or numerical) solutions not sharing their not have to satisfy ∇2 ξµ = 0. However, we pointed out KVFs and/or their isometry groups. Second, our study another geometrical property of a KVF, following from does not involve the fixed point structure of the KVFs’ the Komar identity, i.e., ∇ ∇2 ξµ
= 0, for which we µ flow, and it would be desirable to understand the rela- provide an alternative proof in appendixB. tions between these structures in the two sectors. Third, In HR bimetric theory, concerning collineations of the we have not considered the relationship between KVFs tensors in the theory, three configurations are possible: of the two metrics and Lie point symmetries [48, p. 129] (i) The two metrics have the same KVFs, which define of the bimetric field equations. collineations for every tensor in both sectors. In principle, new ansatzes can be found and used in HR bimetric theory if we allow for different KVFs, or (ii) The two metrics do not have the same KVFs. different isometries of the two metrics, a fact that has gained little attention in the literature so far. Our results (a) In vacuum, tensors in the g-sector have the clarify that this is definitely a promising possibility to same collineations as g , and those in the f- µν enlarge the spectrum of (hopefully exact) solutions in sector the same as f . µν bimetric relativity. (b) When stress–energy tensors are present, ten- sors in the g-sector do not necessarily have the same collineations as gµν , and tensors in the f-sector do not necessarily have the same ACKNOWLEDGMENTS collineations as fµν . Therefore, contrary to GR, an isometry of the metric We are grateful to Ingemar Bengtsson for many re- is not necessarily a spacetime symmetry for all other ten- warding discussions, for reading the manuscript and pro- sors. This raises the intriguing question of what the best viding valuable comments. We also thank Fawad Hassan, definition of a spacetime symmetry in HR bimetric the- Bo Sundborg, Kjell Rosquist, Luis Apolo, Nico Winterg- ory is. If the two metrics share the isometry, then all ten- erst and Mikael von Strauss for many helpful and inspir- sors have the same spacetime symmetry, similar to GR. ing discussions. 15

µ Appendix A: How to determine ansatzes having DOFs in S ν , which can be determined by choosing its different KVFs diagonalizing matrix. Here we choose a constant Lorentz µ transformation Λ ν so that In the main text, we have studied some particular so- (Λ−1)µ Sρ Λσ = diag(λ , λ , λ , λ ). (A4) lutions of the bimetric field equations, focusing on their ρ σ µ 1 2 3 4 isometries and on the shared spacetime topology. We With this choice, fµν simply becomes a Minkowski met- concluded that there are both solutions where the met- ric, rics share their isometry group but not their KVFs, and −1,T ρ 2 2 2 2 −1 σ solutions where the metrics do not have the same isom- fµν = (Λ )µ diag(−λ1, λ2, λ3, λ4)ρσ(Λ ) ν . (A5) etry group. In this appendix we devise three different It is easy to show that fµν is a Minkowski metric, albeit methods of generating infinite sets of solutions belonging that the Cartesian coordinates of fµν are different from to these two categories. Hence the particular examples 0µ those of gµν . Defining a new set of coordinates x via studied in the main text are not unique. In the first example, the two metrics will be Einstein ∂x0µ = diag(λ , λ , λ , λ )Λ−1, (A6) metrics with vanishing cosmological constants and van- ∂xν 1 2 3 4 ishing stress–energy tensors in both sectors, i.e. this is a set of first-order linear partial differential equa- 0µ µ µ tions and always has a solution for x as an (invertible) Gg ν = Gf ν = 0, (A1a) µ µ µ function of x , so it is a valid coordinate transformation T = T = 0. (A1b) µ µ0 g ν f ν and fµ0ν0 = diag(−1, 1, 1, 1). Hence 0 = Gf ν = Gf ν0 . We conclude that The field equations to be solved are g = −dt2 + dx2 + dy2 + dz2, µ µ µν Vg ν = Vf ν = 0. (A2) h −1,T ρ 2 2 2 2 −1 σ i µ ν fµν = (Λ )µ diag(−λ1, λ2, λ3, λ4)ρσ(Λ ) ν dx dx We note that, if the field equations (5) hold, the Bianchi 02 02 02 02 constraints (8) are automatically satisfied.16 In the sec- = −dt + dx + dy + dz , (A7) ond example, we introduce stress–energy tensors in both µ solves the bimetric field equations. Note that Λ ν is an sectors to cancel the respective Einstein tensors and again arbitrary constant Lorentz matrix. Choosing different we are left to solve (A2). In the last example, we consider Lorentz matrices (i.e., choosing different Lorentz diffeo- a non-GR solution. morphisms between the sectors) yields different solutions. Since g and f are related via the diffeomorphism (A6) they must have the same symmetry group, as explained Method 1: Generating Minkowski solutions with in section III D. However, the Cartesian charts of g Lorentz transformations µν and fµν do not coincide generally and hence the KVFs of gµν and fµν may differ; they are related by the Lorentz We pick by hand the metric g to be the Minkowski µ µν transformation Λ ν . metric. Then there exists a coordinate chart xµ = As an example,√ let β0,2,3 = 1, √β1 = 0, β4 = 2, (t, x, y, z) such that the components of gµν are λ1,3 = (1 + 5)/2, and λ2,4 = (1 − 5)/2. This choice µ µ g = η = diag(−1, 1, 1, 1). (A3) of parameters satisfies Vg ν = Vf ν = 0. We choose µν µν the Lorentz matrix to be a boost of velocity 1/2 in the x- Note that there are ten degrees of freedom (DOFs) in direction combined with a rotation of angle −π/4 around each of the metrics and hence twenty in total. Choos- the x-axis. fµν becomes √ ing gµν as in (A3), its DOFs are completely determined.  √  − 1 9 + 5 5
− 2 5 0 0 Since f = gS2, the ten undetermined DOFs in f can 6 √ 3 √  2 5 1
 be redistributed to the square root matrix S. For Ein- − 3 6 9 − 5 5 0 0 fµν =  √ . stein metrics, the equations (A2) completely determine  0 0 3 − 5   2√ 2  the four eigenvalues λ1, ..., λ4 of the square root matrix 5 3 0 0 − 2 2 in terms of the β-parameters and one of β-parameters in (A8) terms of the others [19].17 This leaves six undetermined It is now straightforward to check whether or not the KVFs of gµν and fµν coincide. The KVF generating the rotational symmetry of gµν around the x-axis is η = 16 µ If we solve the bimetric field equations (5) without using the −z∂y + y∂z. The Lie derivative of fµν with respect to η Bianchi constraints (8), the solution will satisfy the Bianchi con- does not vanish. Rather, the KVF of fµν generating the straints automatically. The latter are enclosed and hidden in 0 0 rotational symmetry around the x -axis is ξ = −z ∂y0 + the field equations. However, the explicit use of them makes it 0 µ simpler to solve the field equations. y ∂z0 , or, expressed in the Cartesian chart x for gµν , 17 Unless for very a special set of the β-parameters, referred to as  √   √  algebraically decoupled in [19]. ξ = − 5y + 3z ∂y + −3y + 5z ∂z. (A9) 16

µ µ 2 µ Then, the metrics share their isometry group, but not all Tf ν to be Tf ν := Mf Gf ν , the bimetric field equa- of their KVFs. tions are satisfied. Concerning the Bianchi constraints, µ Note how the assumption of a constant Λ ν in (A5) they are satisfied by construction since both tensor po- insured gµν and fµν to be related via a diffeomorphism tentials vanishes and the stress–energy tensors in both µ and hence Gf ν to vanish. The method of generating sectors are defined to be the Einstein tensors of the two solutions could be generalized by retaining the spacetime metrics. dependence of the Lorentz matrix. In that case however, Note that the only equations that needs to be satis- µ T Gf ν = 0 does not hold automatically and becomes a fied when implementing the method are gS = (gS) to µ µ differential equation for Λ(x) ν . obtain a symmetric fµν metric and Vg,f ν (S) = 0. The When solving the bimetric field equations it is com- solution of the last two equations determines in the gen- µ mon to demand that a flat spacetime solution exists eral case the (constant) eigenvalues of S ν in terms of the in the theory. Usually this is achieved by demanding β-parameters and determines one of the β-parameters in 2 ηµν = gµν = c fµν to be a solution of the field equations, terms of the others. yielding the conditions, sometimes called asymptotic flat- To illustrate the method, let gµν be the Minkowski ness conditions, metric so that the ten DOFs in gµν are completely spec- ified. Then the Einstein tensor for gµν vanishes and ac- β0 = −3β1 − 3β2 − β3, µ cordingly Tg ν := 0. Furthermore, assume that β4 = −β1 − 3β2 − 3β3. (A10) −1 µ ρ σ (Λ (x)) ρS σΛ(x) ν = diag(λ1, λ2, λ3, λ4), (A12) If we are minimally coupled to gµν , the trajectory of a point-like test particle will be a solution of the geodesic where Λ(x) is a non-constant, local Lorentz matrix. Now, equation defined by gµν . Hence all measurable geomet- the ten remaining DOFs are distributed as four in the rical quantities will be contained in the metric g . Ac- µ µν eigenvalues of S ν and six in the Lorentz matrix. From cordingly, demanding gµν = ηµν is sufficient in order to (A12) we see that gS = (gS)T is satisfied by construction. obtain a universe that we would measure as flat. In fact, µ To satisfy Vg,f ν (S) = 0 we solve for the eigenvalues of even in the case where both metrics are Minkowski, the µ S ν and for β4, so that λi = λi(β0, ..., β4) and β4 = solutions (A3), (A8) shows that the β-parameters does β4(β0, ..., β3). For definiteness, let Λ(x) be a rotation not need to satisfy the conditions (A10). Therefore, de- around the z-axis with the rotation angle depending on manding (A10) is unnecessarily restrictive in order for z, a bi-flat solution to exist in the theory. On the other hand, demanding gµν = ηµν does constraint fµν to have 1 0 0 0 a specific form [19]. 0 cos(z) − sin(z) 0 Λ−1,T(z) =  . (A13) 0 sin(z) cos(z) 0 0 0 0 1 Method 2: Decoupled interaction terms

fµν becomes Let’s choose a Lorentzian metric gµν and use Synge’s method to define the stress–energy tensor minimally cou- fµν = µ pled to gµν . Then, Tg ν is proportional to the Einstein  2  −λ1 0 0 0 tensor of gµν , i.e. 2 2 2 2 1 2 2
 0 λ2 cos (z) + λ3 sin (z) 2 λ3 − λ2 sin(2z) 0   1 2 2
2 2 2 2 . µ 2 µ  0 2 λ3 − λ2 sin(2z) λ3 cos (z) + λ2 sin (z) 0  Tg ν := Mg Gg ν . (A11) 2 0 0 0 λ4 For this construction to yield a physical solution it is im- (A14) portant to pick the metric gµν yielding the stress–energy tensor in (A11) satisfying all the desired properties, e.g. The Einstein tensor for fµν can then be computed and we µ 2 µ the null energy condition (NEC). For example, if gµν define Tf ν := Mf Gf ν . The explicit expression is not 2 is the FLRW metric for a homogeneous and isotropic presented here but it is important to note that if λ2 6= 2 4/3 2 matter dominated universe, gµν = −dt + t d~x, then λ3, it is not vanishing. Thus, fµν is not the Minkowski the stress–energy as defined by (A11) is physical, since metric. Interestingly enough, the Ricci scalar of fµν is a we know that its Einstein tensor is proportional to constant depending on the square root eigenvalues λi diag(t−2, 0, 0, 0) and so it describes a pressure-less perfect µ 2 2 2 fluid. In such a case, the tensor potential Vg ν decouples (λ2 − λ3) µ Rf = − . (A15) from the field equations and we are left with Vg ν = 0. 2λ2λ2λ2 µ 2 3 4 Now let S ν be a rank (1,1) tensor such that gS = T µ (gS) and Vf ν (S) = 0. The first equation tells us that However, the Ricci tensor of fµν is not proportional to h := gS is a symmetric rank (0,2) tensor, implying that the identity matrix, the f metric, f := hS, is also symmetric. To be explicit, T T T T T T µ f := S h = S h = S gS = (gS) S = f. Now, defining Rf ν ∝ 17

0 0 0 0  rics, because the two metrics have to share their compat- 2 2 2 2 0 (λ2 + λ3) cos(2z) −(λ2 + λ3) sin(2z) 0 ible covariant derivative (see (84)). In HR bimetric the-  2 2 2 2 , 0 −(λ2 + λ3) sin(2z) −(λ2 + λ3) cos(2z) 0  ory, if the metrics are proportional, then the field equa- 2 2 0 0 0 −(λ2 − λ3) tions (5) reduce to two decoupled copies of the Einstein (A16) equations, i.e., gµν is a solution of the Einstein equa- 2 µ µ tions and fµν = c gµν , with c ∈ R; hence S ν = c δ ν . so f is not an (A)dS metric. Hence it is not maximally µ −2
µ µν We conclude that, assuming A ν = S ν , the ansatz symmetric and does not have the same isometry group having proportional metrics and proportional KVFs is as gµν . consistent. To summarize, due to screening from the stress–energy Note that S−2
µ is not a congruence between g tensors, the tensor potentials of g and f decouple ν µν µν µν and f in HR bimetric theory. Therefore, in this case, from the field equations. The interaction between the µν we are not guaranteed that (77) holds; however, Aµ be- two metrics vanishes, i.e., the metrics are effectively non- ν ing a congruence between g and f is a sufficient, but interacting. Hence it is not surprising that g and f µν µν µν µν not necessary condition for the KVFs of the two sectors could have completely different isometry groups. A note to have the same character. Therefore, there are solu- on the stress–energy tensor of f might also be in place. µν tions for which the KVFs are not related by a congruence Since T µ is defined to be proportional to the Einstein f ν between the metrics, but still have A-related KVFs with tensor for G µ , we are not guaranteed that such a stress- f ν the same character in the two sectors. energy tensor would satisfy, e.g., the energy conditions. In fact, as shown in [49], in HR bimetric theory in vac- The second potential ansatz we consider is, uum, there is a strong anti-correlation between the null µ µ µ µ −1
µ energy conditions in the g- and f-sector. However, it ϕ ν = δ ν ,A ν = γ ν = S ν . (A18) might be argued that from an observational viewpoint, the f-sector is insignificant as long as the gµν sector is for which we are assured that paired KVFs have the same µ well-behaved since only gµν and Tg ν are measurable. character with respect to the corresponding metric. In- deed, in HR bimetric theory, by definition we are guar- µ p −1 µ anteed that the square root S ν = g f ν exists [12], Method 3: Non-GR solutions and thus (in matrix notation),

Many works in the literature studied particular solu- f = STg S. (A19) tions of HR bimetric theory, both in vacuum and with matter sources (for a review, see e.g. [20] and references therein), and the usual approach is to choose some ansatz On the other hand, in this case (75) is not as simple as for the two metrics, motivated by an assumption about the previous case, since their isometries. The approach that we follow here is es- p sentially the same, but now we allow for different KVFs Υ = fS−1g−1 = f f −1gg−1 = ST, (A20) in the two sectors. In particular, we assume a specific µ form of the map A ν between the KVFs, and we solve which implies (coming back to the index notation), µ (75) for fµν (or, equivalently, S ν ). Afterwards, we check that the obtained ansatz satisfies the bimetric field equa- h T β α T βi tions (5). If so, then the determined ansatz describes a Lξfµν = 2∇(µS ν) − C µν S α ηβ solution of the theory, otherwise such a configuration is T β
not allowed in the theory. + 2S (ν| ∇|µ)ηβ = 0. (A21) We have discussed the possibility of having the same isometry group, but different KVFs. In this case, the We report here the solution of (A21) in the case of two metrics do not look explicitly symmetric in the same a static and spherically symmetric gµν in the spherical coordinate chart. Indeed, we can only choose one set of polar chart (t, r, θ, φ), generators of the Lie algebra of the isometry group, and if we choose the coordinate basis to be aligned with some −eq(r)F (r) 0 0 0  of the generators of the Lie algebra, then the components 0 F (r)−1 0 0 g =   (A22) of another set of generators will not look as simple in the  0 0 r2 0  same coordinate basis.   0 0 0 r2 sin2(θ) The first potential ansatz we consider is obtained by choosing, in (69), with the usual set of KVFs, assuming a handy expression µ µ µ µ µν −2
µ µ ϕ ν = δ ν ,A ν = γ ν = f gµν = S ν . for S ν . We first determine the most general expression µ ρ (A17) for S ν , obtained by the constraint that hµν = gµρS ν In section III D, we saw that in this case, the majority of (the symmetrizing quadratic form) is symmetric; the the cases (see [3, 46, 47]) provides two proportional met- symmetrisation of hµν automatically implies the same 18

µ for fµν . The resulting S ν is, The proof in [39] is based on the variation of the Einstein–Hilbert Lagrangian density. We now present   an alternative and simpler proof. S00 S01 S02 S03 If ∇[ν V µ] = 0, then the Komar identity is trivially  q 2   −e F S01 S11 S12 S13 true, hence we suppose it is not zero. By making use of  q  µ  e F S12  the commutation rule of covariant derivatives involving S ν =  − S S S ,  2 02 2 22 23  r F r  the Riemann curvature tensor [2, p. 39], we can rewrite  eqF S S  (B1) as − S 13 23 S  2 2 03 2 2 2 33 r sin (θ) F r sin (θ) sin (θ) [ν µ] (A23) ∇µ∇ν ∇ V ν µ µ ν = ∇µ∇ν ∇ V − ∇µ∇ν ∇ V where all the functions S depend on all the variables ν µ ν ρ µ µ ν σ ij = ∇ν ∇µ∇ V − Rµνρ ∇ V − Rµνσ ∇ V (t, r, θ, φ). Since solving (A21) with the general ansatz − ∇ ∇ ∇µV ν + R µ∇ρV ν + R ν ∇µV σ (A23) turns out to be quite involved, we assumed a sim- ν µ µνρ µνσ µ ν µ ρ µ ν σ pler form for S ν , setting to 0 some of its components. = ∇ν ∇µ∇ V − Rµρ∇ V + Rνσ∇ V µ ν σ ν µ ρ Here we report only one case which turns out, in the end, − ∇ν ∇µ∇ V − Rνσ∇ V + Rµρ∇ V to have proportional KVFs, with [ν µ] [µ ρ] [ν σ] µ = ∇ν ∇µ∇ V + Rµρ∇ V + Rνσ∇ V Let’s assume the following S ν , [ν µ] = ∇ν ∇µ∇ V , (B2)  S00 S01 0 0  q 2 µ −e F S01 S11 0 0 i.e., S ν =  , (A24)  0 0 S22 0  [ν µ] [ν µ] [ν µ] 0 0 0 S33 ∇µ∇ν ∇ V − ∇ν ∇µ∇ V = 2∇[µ∇ν]∇ V ≡ 0. (B3) where the functions Sij depend on all the variables. Solv- From (B3) it follows ing (A21) yields, [ν µ] [ν µ] ∇µ∇ν ∇ V = ∇(µ∇ν)∇ V ≡ 0, (B4) S00(t) 0 0 0 0 S (r) 0 0 since we are contracting a symmetric expression in µ, ν, Sµ =  11 , (A25) ν  0 0 λ 0 with an antisymmetric expression in µ, ν. This completes 0 0 0 λ the proof. For a KVF, ∇[ν ξµ] = ∇ν ξµ due to the Killing equation µ −1
µ ν with λ = const. Given that ξ = S ν η , we con- (35), therefore clude that the S−1-related KVFs are proportional in this [ν µ] 2 µ
framework. ∇µ∇ν ∇ ξ = ∇µ ∇ ξ ≡ 0. (B5) The field equations (5) and the Bianchi constraint (8) require S00(t) = const. This ansatz only contains three field variables, namely q(r), F (r) and S11(r), but the field equations and the Bianchi constraints constitute a system of seven coupled ordinary differential equations. Therefore, the system is over-determined and after solv- ing analytically for the three fields, there are algebraic equations and consistency relations to be satisfied. The solution of this differential–algebraic system (analytical or numerical) is left for future work.

Appendix B: Alternative proof for (44)

The Komar identity, presented in [38] and proved in [39], states that, for a generic vector field V µ,

[ν µ] ∇µ∇ν ∇ V ≡ 0. (B1)

This a purely geometrical identity, therefore it holds in any modified theory of gravity relying on (pseudo)- Riemannian geometry. 19

[1] K. Schwarzschild, Sitzungsber. Preuss. Akad. [27] M. Kocic, M. H¨og˚as,F. Torsello, and E. Mortsell, Wiss. Berlin (Math. Phys.) 1916, 189 (1916), (2017), arXiv:1708.07833 [hep-th]. arXiv:physics/9905030 [physics]. [28] C. Deffayet and T. Jacobson, Class. Quant. Grav. 29, [2] R. Wald, General Relativity (University of Chicago Press, 065009 (2012), arXiv:1107.4978 [gr-qc]. 2010). [29] F. Torsello, M. Kocic, and E. M¨ortsell, Phys. Rev. D 96, [3] G. Hall, Symmetries and Curvature Structure in Gen- 064003 (2017), arXiv:1703.07787 [gr-qc]. eral Relativity, World Scientific Lecture Notes in Physics [30] U. Moschella, “The de Sitter and anti-de Sitter Sight- (2004). seeing Tour,” in Einstein, 1905–2005: Poincar´eSeminar [4] M. S. Volkov, Phys. Rev. D 86, 061502 (2012). 2005 , edited by T. Damour, O. Darrigol, B. Duplantier, [5] H. Nersisyan, Y. Akrami, and L. Amendola, Phys. Rev. and V. Rivasseau (Birkh¨auserBasel, Basel, 2006) pp. D92, 104034 (2015), arXiv:1502.03988 [gr-qc]. 120–133. [6] D. G. Boulware and S. Deser, Phys. Rev. D6, 3368 [31] Wolfram Research, Inc., Mathematica, Version 11.0, (1972). Champaign, Illinois (2016). [7] C. de Rham and G. Gabadadze, Phys. Rev. D82, 044020 [32] S. W. Hawking and G. F. R. Ellis, The Large Scale Struc- (2010), arXiv:1007.0443 [hep-th]. ture of Space-Time, Cambridge Monographs on Mathe- [8] C. de Rham, G. Gabadadze, and A. J. Tolley, Phys. Rev. matical Physics (Cambridge University Press, 2011). Lett. 106, 231101 (2011), arXiv:1011.1232 [hep-th]. [33] G. W. Gibbons and S. W. Hawking, Phys. Rev. D 15, [9] S. F. Hassan and R. A. Rosen, Phys. Rev. Lett. 108, 2738 (1977). 041101 (2012), arXiv:1106.3344 [hep-th]. [34] C. Charmousis, “Introduction to anti de sitter black [10] S. F. Hassan and R. A. Rosen, JHEP 02, 126 (2012), holes,” in From Gravity to Thermal Gauge Theories: The arXiv:1109.3515 [hep-th]. AdS/CFT Correspondence, edited by E. Papantonopou- [11] S. F. Hassan and R. A. Rosen, JHEP 04, 123 (2012), los (Springer Berlin Heidelberg, Berlin, Heidelberg, 2011) arXiv:1111.2070 [hep-th]. pp. 3–26. [12] S. F. Hassan and M. Kocic, (2017), arXiv:1706.07806 [35] S. W. Hawking, Communications in Mathematical [hep-th]. Physics 25, 152 (1972). [13] D. Comelli, M. Crisostomi, F. Nesti, and L. Pilo, Phys. [36] M. Henneaux and C. Teitelboim, Communications in Rev. D85, 024044 (2012), arXiv:1110.4967 [hep-th]. Mathematical Physics 98, 391 (1985). [14] S. F. Hassan, A. Schmidt-May, and M. von Strauss, [37] S. Jamal, General Relativity and Gravitation 49, 88 JHEP 05, 086 (2013), arXiv:1208.1515 [hep-th]. (2017). [15] S. F. Hassan, A. Schmidt-May, and M. von Strauss, [38] A. Komar, Phys. Rev. 113, 934 (1959). Class. Quant. Grav. 30, 184010 (2013), arXiv:1212.4525 [39] W. R. Davis and M. K. Moss, Journal [hep-th]. of Mathematical Physics 7, 975 (1966), [16] C. de Rham, L. Heisenberg, and R. H. Ribeiro, Class. http://dx.doi.org/10.1063/1.1705011. Quant. Grav. 32, 035022 (2015), arXiv:1408.1678 [hep- [40] D. Husem¨oller, Fibre Bundles, Graduate Texts in Math- th]. ematics (Springer, 1994). [17] S. F. Hassan, A. Schmidt-May, and M. von Strauss, [41] J. Lee, Introduction to Smooth Manifolds, Graduate Int. J. Mod. Phys. D23, 1443002 (2014), arXiv:1407.2772 Texts in Mathematics (Springer, 2003). [hep-th]. [42] F. J. Chinea, Classical and Quantum Gravity 5, 135 [18] T. Damour and I. I. Kogan, Phys. Rev. D66, 104024 (1988). (2002), arXiv:hep-th/0206042 [hep-th]. [43] Y. Choquet-Bruhat, C. DeWitt-Morette, and [19] M. Kocic, M. H¨og˚as,F. Torsello, and E. Mortsell, M. Dillard-Bleick, Analysis, Manifolds, and Physics, pt. (2017), arXiv:1706.00787 [hep-th]. 1 (North-Holland Publishing Company, 1982). [20] A. Schmidt-May and M. von Strauss, J. Phys. A49, [44] B. Hall, Lie Groups, Lie Algebras, and Representations: 183001 (2016), arXiv:1512.00021 [hep-th]. An Elementary Introduction, Graduate Texts in Mathe- [21] N. Boulanger, T. Damour, L. Gualtieri, and M. Hen- matics (Springer, 2003). neaux, Nucl. Phys. B597, 127 (2001), arXiv:hep- [45] L. Conlon, Differentiable Manifolds, Modern Birkh¨auser th/0007220 [hep-th]. Classics (Birkh¨auserBoston, 2001). [22] S. Lang, Differential and Riemannian Manifolds, Grad- [46] G. S. Hall and C. B. G. McIntosh, International Journal uate Texts in Mathematics (Springer, 1995). of Theoretical Physics 22, 469 (1983). [23] C. Misner, K. Thorne, and J. Wheeler, Gravitation, [47] G. S. Hall, General Relativity and Gravitation 15, 581 Gravitation No. pt. 3 (W. H. Freeman, 1973). (1983). [24] R. W. Fuller and J. A. Wheeler, Phys. Rev. 128, 919 [48] H. Stephani, D. Kramer, M. MacCallum, C. Hoenselaers, (1962). and E. Herlt, Exact Solutions of Einstein’s Field Equa- [25] J. Synge, Relativity: the general theory, Series in physics tions, 2nd ed., Cambridge Monographs on Mathematical (North-Holland Pub. Co., 1960). Physics (Cambridge University Press, 2003). [26] N. Ibohal, Gen. Rel. Grav. 37, 19 (2005), arXiv:gr- [49] V. Baccetti, P. Martin-Moruno, and M. Visser, JHEP qc/0403098 [gr-qc]. 08, 148 (2012), arXiv:1206.3814 [gr-qc].