Arxiv:1910.00148V3 [Gr-Qc] 7 Dec 2019 Sound Waves in Earth’S Atmosphere [13, 14]

Luminal Propagation of Gravitational Waves in Scalar-tensor Theories: The Case for Torsion

JoséBarrientos,2, 3, 4, ∗ Fabrizio Cordonier-Tello,5, † CristóbalCorral,6, ‡ Fernando Izaurieta,2, § Perla Medina,2, 7, ¶ Eduardo Rodr´ıguez,8, ∗∗ and Omar Valdivia1, †† 1Facultad de Ciencias, Universidad Arturo Prat, 1110939 Iquique, Chile 2Departamento de F´ısica, Universidad de Concepción,Casilla 160-C, 4070105 Concepción,Chile 3Departamento de Enseñanzade las Ciencias Básicas, Universidad Católica del Norte, Larrondo 1281, 1781421 Coquimbo, Chile 4Institute of Mathematics of the Czech Academy of Sciences, Zitná25,ˇ 11567 Praha 1, Czechia 5Arnold Sommerfeld Center for Theoretical Physics, Ludwig-Maximilians-Universität München,Theresienstraße 37, 80333 Munich, Germany 6Departamento de F´ısica, Universidad de Santiago de Chile, Avenida Ecuador 3493, EstaciónCentral, 9170124 Santiago, Chile 7Centro de Estudios Cient´ıficos (CECs), Avenida Arturo Prat 514, 5110466 Valdivia, Chile 8Departamento de F´ısica, Universidad Nacional de Colombia, 111321 Bogotá,Colombia (Dated: December 10, 2019) Scalar-tensor gravity theories with a nonminimal Gauss–Bonnet coupling typically lead to an anomalous propagation speed for gravitational waves, and have therefore been tightly constrained by multimessenger observations such as GW170817/GRB170817A. In this paper we show that this is not a general feature of scalar-tensor theories, but rather a consequence of assuming that spacetime torsion vanishes identically. At least for the case of a nonminimal Gauss–Bonnet coupling, removing the torsionless condition restores the canonical dispersion relation and therefore the correct propagation speed for gravitational waves. To achieve this result we develop a new approach, based on the first-order formulation of gravity, to deal with perturbations on these Riemann–Cartan geometries.

PACS numbers: 04.50.+h Keywords: Nonvanishing Torsion, Gravitational Waves, Riemann–Cartan Geometry, Gauss–Bonnet Coupling

I. INTRODUCTION topological invariants, e.g., the Pontryagin or Gauss– Bonnet (GB) terms, motivated by effective field theo- The multimessenger measurements of the GW170817 ries, string theory, and particle physics [15]. From a event by the LIGO/Virgo Collaboration [1] and the phenomenological viewpoint, the scalar-Pontryagin mod- gamma-ray burst GRB 170817A by Fermi and other ification to GR—also known as Chern–Simons modified observatories [2] have provided a strong limit of about gravity—is an interesting extension that might explain one part in 1015 on the difference between the propaga- the flat galaxy rotation curves dispensing with dark mat- tion speed of gravitational waves (GW) and the speed of ter [16], while leaving the propagation speed of GWs un- light [3,4]. This observation imposes severe constraints affected [17]. This interaction generates nontrivial effects on different viable alternatives to general relativity (GR) when rotation is included [18–23], providing a smoking aimed at explaining the dark sector of the Universe by gun in future GW detectors [24–27]. The couplings be- means of degrees of freedom beyond the metric ones. tween scalar fields and the GB term, on the other hand, In particular, many scalar-tensor theories of the Horn- have been studied in different setups and several solu- deski/Galileon type predicted, at least in some regimes, tions that exhibit spontaneous scalarization have been an anomalous propagation speed for GWs [5–12], and reported [28–41]. Their stability, however, depends on even in some cases an anomalous propagation speed for the choice of the coupling between the scalar field and arXiv:1910.00148v3 [gr-qc] 7 Dec 2019 sound waves in Earth’s atmosphere [13, 14]. This obser- the GB term [42–44]. In spite of this, the theory is ex- vation implies that, depending on the type of coupling perimentally disadvantaged from an astrophysical view- that the scalar fields develop with the geometry, some of point, since it develops an anomalous propagation speed these theories have been disfavored by the observational for GWs [45]. data. Scalar-tensor theories have been formulated in geome- A particular interaction that has been widely stud- tries that depart from the pseudo-Riemannian framework ied in the literature is the coupling of scalar fields to several times in the past. In particular, the gravitational role of Riemann–Cartan (RC) geometries, characterized by curvature and torsion, was first discussed by Cartan ∗ [email protected] and Einstein themselves [46], and later on in the frame- † [email protected] work of gauge theories of gravitation [47–50]. Within ‡ [email protected] § fi[email protected] its simplest formulation—the Einstein–Cartan–Sciama– ¶ [email protected] Kibble (ECSK) theory—torsion is a nonpropagating field ∗∗ [email protected] sourced only by the spin density of matter. The non- †† [email protected] minimal coupling of scalar fields to geometry dramati- 2 cally changes this conclusion. As shown in Ref. [51], the standard Lorenz gauge fixing for the trace-reversed per- typical Horndeski/Galileon couplings and second-order turbation to this setting. SectionV describes how to derivatives are generic sources of torsion, even in the ab- separate low- and high-frequency terms. We follow the sence of any spin density. When scalar fields are coupled approach of Ref. [75], with appropriate modifications for to the Nieh–Yan topological invariant [52], a regulariza- the case of RC geometry. SectionVI focuses on the lead- tion procedure of the axial anomaly in RC spacetimes ing high-frequency term to prove that torsion restores can be prescribed [53–57], and a torsion-descendent ax- the canonical dispersion relation, including speed, for ion that might solve the strong CP problem in a gravita- the metric mode of GWs. The eikonal approximation tional fashion is predicted [58–60]. The nonminimal cou- is used to show that the new torsional mode (variously pling to the Gauss–Bonnet invariant, on the other hand, called “torsionon” [76], “roton” [77], or “gravity W and can be motivated from dimensional reduction of string- Z bosons” [78]) propagates interacting with the polariza- generated gravity models [61], and it could drive the late- tion of the standard metric mode, generalizing the results time acceleration of the Universe in the absence of the of Ref. [69]. Finally, conclusions and further comments cosmological constant [62–64]. The first-order formula- are given in Sec.VII, while many details on the calcula- tion of Chern–Simons modified gravity produces inter- tions are provided in AppendixA. esting phenomenology when coupled with fermions [65], and it has been shown that the different nonminimal couplings support four-dimensional black string config- II. SCALAR-TENSOR MODEL WITH GAUSS–BONNET COUPLING urations in vacuum, possessing locally AdS3 × R geometries with nontrivial torsion [66]. Remarkably, some of these models can be regarded as a zero-parameter ex- Let M be a four-dimensional spacetime manifold with tension of GR [67], whose cosmological implications have metric signature (−, +, +, +). We shall consider a scalar- been recently studied in Ref. [68]. In general, assuming a tensor theory whose independent dynamical fields are a a µ 2 torsion-free condition reduces the number of independent the vierbein one-form e = e µdx , the spin connec- ab ab µ fields, making the torsionless theory an entirely different tion one-form ω = ω µdx , and a complex zero-form dynamical system from the torsionful one. scalar field φ, with φ¯ being its complex conjugate. The In this work, we show that it is only the torsionless ver- Lagrangian four-form describing the scalar-tensor theory sion of the scalar-tensor theory based on the scalar-GB with Gauss–Bonnet coupling is given by coupling that predicts an anomalous propagation speed 1 ab c d for GWs. When torsion is taken into account as a right- L = abcdR ∧ e ∧ e ful component of geometry, GWs generically propagate 4κ4 at the speed of light, and hence those torsional theories 1 a b c d − (Λ + κ4V ) abcde ∧ e ∧ e ∧ e survive unfalsified by multimessenger astronomy. Since 4!κ4 the dispersion relation for electromagnetic waves (EMW) ¯ 3 1 ab cd − dφ ∧ ∗dφ − abcdR ∧ R , (1) also remain unmodified by torsion, both EMWs and GWs 8κ4 Λ + κ4V move along null geodesics, even on a background with ab ab a cb nonvanishing torsion. This does not mean, however, that where R = dω + ω c ∧ ω is the Lorentz curvature torsion is wholly undetectable; as shown in Ref. [69], tor- two-form, and V stands for the scalar field’s potential, sion affects the propagation of polarization for GWs.1 which is assumed to depend only on the magnitude of Thus, at least for this case, the recent observational data φ, i.e., V = V (|φ|). Throughout this article, we work only disfavor the torsionless version of the theory, but in the context of RC geometry, meaning that the vier- not its more general torsional relative. bein and spin connection are considered as independent Our article is organized as follows. SectionII presents degrees of freedom, and therefore the torsion two-form a a a a b the main line of reasoning, where we introduce the La- T = De = de + ω b ∧ e does not need to vanish. The grangian that defines the theory and give general argu- Lagrangian depends only on first-order derivatives of the ments on why the torsionless version of the GB cou- spin connection and it does not contain derivatives of pling changes the speed of GWs, while the most gen- the vierbein: we do not include any explicit torsional eral dynamical torsion case does not. SectionsIII–VI terms [79]. The coupling constant κ4 is related to New- prove this statement in detail. In Sec.III, we define ton’s gravitational constant GN through κ4 = 8πGN , some mathematical operators that greatly simplify the and the cosmological constant is denoted by Λ. analysis of GWs on an RC geometry and study their The Lagrangian (1) allows for propagating torsion in properties and algebra. In Sec.IV, we use a Lorentz- vacuum and can be regarded as both a particular case covariant version of the Lie derivative to generalize the of Horndeski’s theory [80] and as a generalization of dynamical Chern–Simons modified gravity [15, 81], both of

1 Similar results have been found in teleparallel gravity theo- 2 ries [70–72] and in f(R) theories with a nonminimal coupling The vierbein is related to the spacetime metric gµν through a b to the Nieh–Yan term [73, 74]. gµν = ηabe µe ν , with ηab = diag (−, +, +, +). 3 which set the torsion equal to zero at the outset (but The field equations set Ea, Eab, E, and E¯ to zero on M, see Ref. [51] for the torsional version of Horndeski’s the- while the boundary conditions are given by the vanishing ory). The theory defined by Eq. (1) actually differs of Bab, B, and B¯ on ∂M. Furthermore, our 1/ (Λ + κ4V ) from the standard torsional ECSK theory only in the GB coupling allows the field equations (5)–(7) to be fully nonminimal coupling 1/ (Λ + κ4V ) with the GB density. compatible with the boundary conditions (8)–(9). In par- When V = const., this last term becomes a topologi- ticular, Eab = DBab, and the system admits the maxi- cal invariant proportional to the Euler characteristic and mally symmetric solution in vacuum does not contribute to the field dynamics in the bulk, although it becomes relevant in the regularization of φ = φ0, (11) Noether charges for asymptotically locally anti-de Sit- 1 Rab = e2σ0 ea ∧ eb, (12) ter spacetimes [82, 83]. In the general case, namely l2 V 6= const., this term contributes to the field equations T a = 0, (13) acting as a source of torsion [51, 58–64, 66]. The particular nonminimal coupling with the GB term we use is but where φ0 = const. and σ0 = σ (φ0). This solution de- one choice; the results regarding the speed of GWs are scribes a spacetime of constant curvature and zero tor- still valid even if the 1/ (Λ + κ4V ) coupling is replaced sion, where the (constant) scalar field plays no role. by an arbitrary function of (the magnitude of) the scalar When scalar-tensor gravity theories are treated within field, f (|φ|). Our choice, however, has several important the first-order formalism, torsion propagates in vacuum algebraic and physical properties which lead to a much sourced by the derivatives of the scalar fields (for fur- more transparent treatment. ther details, see Ref. [51]). As a matter of fact, the field To start with, this choice for the nonminimal coupling equation E = 0 [cf. Eq. (6)] can be rewritten as with the GB term allows the Lagrangian to be written in ab a much more compact way, ∂σ 1 T p = −l2e−2σ ep ∧ ed∗ d |φ| ∧ Rab ∧ ec 2 ∂ |φ| 2 abcd l ab cd ¯ L = − abcdF ∧ F − dφ ∧ ∗dφ, (2) 8κ4 pq − ∗ [2eq ∧ ∗ (d |φ| ∧ R )] . (14) where 3 κ 1 Λ = , e2σ = 1 + 4 V,F ab = e−σRab − eσea ∧ eb. The propagating nature of torsion becomes manifest in l2 Λ l2 this equation, since its right-hand side possesses deriva- (3) tives of the torsion through Rab. This can be seen from The independent stationary variations of L with respect the decomposition of the two-form curvature into their 4 to ea, ωab, φ and φ¯ yield Riemannian and torsional pieces δL = δea ∧ E + δωab ∧ E + δφE¯ + δφ¯E ab ˚ab ˚ ab a cb a ab R = R + Dκ + κ c ∧ κ , (15) ab ¯ + d δω ∧ Bab + δφB¯ + δφB , (4) where R˚ab is the canonical Riemann curvature two-form where and κab = ωab − ˚ωab is the contorsion tensor one-form, a a b 1 σ b cd related to the two-form torsion as T = κ b ∧ e . Ea = e abcde ∧ F 2κ4 To understand why torsion restores the speed of light 1 of GWs for the scalar-GB coupling, let us go back to the + ZbZ¯ + Z¯bZ − |Z|2 δb ∗e , (5) 2 a a a b Lagrangian (1). Since we are working in the context of 2 RC geometry, the vierbein and the spin connection are l −σ cd Eab = −D e abcdF , (6) independent fields. This means that the GB coupling 4κ4 term does not depend on the vierbein, namely 2 1 φ ∂ l ab cd E = d∗dφ − abcdF ∧ F , (7) 1 2 |φ| ∂ |φ| 4κ4 ab cd δe abcdR ∧ R = 0. (16) 2 Λ + κ4V l −σ cd Bab = − e abcdF , (8) 4κ4 Therefore, the field equation for the vierbein, Ea = 0 B = −∗dφ, (9) [cf. Eq. (4)], is insensitive to its presence. In fact, this equation can be cast into the Einstein–Hilbert form as and3 κ Za = −∗ (ea ∧ ∗dφ) , Z¯a = −∗ ea ∧ ∗dφ¯ . (10) Rbc ∧ ed = 4 eb ∧ ec ∧ edT e, (17) abcd 3 bcde a

3 See Deﬁnition2 in Sec.III for the mathematical properties of the operator −∗ (ea ∧ ∗ . 4 We use the notation X˚ to denote the “torsionless version” of X. 4

a where Tb is an effective stress-energy tensor given by goal, we must prove that the torsional terms hidden in Eq. (17) do not change the GW dispersion relation and a a ¯ ¯a 2 3 2σ a speed. One also has to deal with the fact that torsion is Tb = Z Zb + Z Zb − |Z| + 2 e δb . (18) κ4l a propagating field in the nonminimally coupled theory. The torsional mode interacts with the standard metric ab c Since GWs arise from perturbations to abcdR ∧ e , as mode, and it is not a priori obvious whether it changes in the usual torsionless case, the GB coupling cannot their speed. possibly contribute to them. In order to prove this point, in the following sections How does the torsionless condition so dramatically al- we provide a complete treatment of GWs on a spacetime ter the behavior of GWs? To see why, note that naively with torsion. The necessary mathematical scaffolding is a imposing T = 0 in the field equations [cf. Eqs. (5)–(9)] developed in Sec.III. Then, in Sec.VI we come back does not lead to the standard torsionless case; instead, to Eq. (17) to show that torsion restores the canonical we get a constant scalar field. The torsionless condition dispersion relation and speed for GWs. is a constraint on the geometry, and as such it must be imposed through the addition of a Lagrangian multiplier two-form Ma to the Lagrangian (1), III. MATHEMATICAL INTERMEZZO a L 7→ LM = L − T ∧ Ma. (19) In this section, we introduce the mathematical tools It is this modified Lagrangian, LM , which reproduces that allow us to describe perturbations and waves in the the standard torsionless dynamics. The field equations context of RC geometry. derived from δLM = 0 read

(M) Ea = Ea − DMa = 0, (20) 1 A. A superalgebra of differential operators E(M) = E − (M ∧ e − M ∧ e ) = 0, (21) ab ab 2 a b b a E¯(M) = E¯ = 0, (22) The differential operators we define appeared origi- nally in Refs. [51, 69, 76]; here we briefly review them (M) E = E = 0, (23) for the benefit of the reader who may be unfamiliar with T a = 0. (24) them. We also show that these operators form a superalgebra and identify its associated super-Jacobi identity, Equation (21) can be solved for the Lagrangian multiplier which, beyond the merely aesthetic, eases the study of to find the eikonal limit of GWs on an RC geometry. 1 For the sake of generality, in this section we work on M a = ∗ 2e ∧ ∗Eba + ea∗ e ∧ e ∧ ∗Ebc . (25) b 2 b c a d-dimensional manifold M endowed with an RC geometry and a metric tensor with η− negative and d − η− ab Since Eab includes the Lorentz curvature two-form R , positive eigenvalues. In the rest of the paper we restrict the term DMa in Eq. (20) turns out to be proportional to ourselves to d = 4 and a spacetime signature η− = 1. The derivatives of Rab. It is straightforward to see that such space of differential p-forms on M is denoted by Ωp (M). (M) terms in Ea make a nonzero leading-order contribu- tion in the eikonal limit for perturbations, and therefore Definition 1 (Hodge star operator). The Hodge star p d−p modify their dispersion relation and the GW speed. operator [84] is a linear map, ∗ :Ω (M) → Ω (M), p The lesson to be learned from this analysis is that im- that takes a differential p-form α ∈ Ω (M), posing the torsionless condition a priori is very differ- 1 ent from imposing it a posteriori: the torsionless theory, µ1 µp α = αµ ···µ dx ∧ · · · ∧ dx , (26) where T a = 0 from the outset, has fewer degrees of free- p! 1 p dom and it constitutes therefore a different dynamical system from the full torsional theory. Imposing the tor- and maps it into its Hodge dual, ∗α ∈ Ωd−p (M), defined sionless condition on the field equations of the torsional by theory implies, in the nonminimally coupled case, reduc- p ing the scalar field to triviality. Even if the Lagrangians |g| ∗α = αµ1···µp dxµp+1 ∧ · · · ∧ dxµd , (27) for both theories may look superficially identical, they are p!(d − p)! µ1···µd inherently different theories; as shown above, the torsionless condition amounts to a constraint on the dynamics. where g is the determinant of the metric tensor and

In the case of the GB coupling, the price of such a con- µ1···µd is the totally antisymmetric Levi-Civita pseu- straint is the anomalous speed for GWs. dotensor. While certainly plausible, we still have to rigorously show that Eq. (17) leads to the canonical dispersion re- Deﬁnition 2. The operators Ia1···aq :Ωp (M) → lation for GWs, including their speed. To achieve this Ωp−q (M) act on p-forms to produce (p − q)-forms ac- 5 cording to the rule5 Deﬁnition 8 (Generalized Beltrami–Laplace wave oper-

(d−p)(p−q)+η ator). We define the generalized Beltrami–Laplace wave Ia1···aq = (−1) − ∗ (ea1 ∧ · · · ∧ eaq ∧ ∗ , (28) p p operator B :Ω (M) → Ω (M) by [51] where ∗ is the Hodge star operator introduced in Defini- a B = −DaD , (35) tion1. The most important case is q = 1, where Da is the derivative operator introduced in Defini- a d(p−1)+η− a I = (−1) ∗ (e ∧ ∗ , (29) tion3. which acts as a coderivative, satisfying the same sign- Lemma 9. The operators introduced in Definitions7 corrected Leibniz rule as the exterior derivative. and8 satisfy the following generalized Weitzenböck iden- p p tity for an RC geometry: Definition 3. We define Da :Ω (M) → Ω (M) to be the derivative operator given by [51] 2 a dR = B + IaD I , (36) Da = {Ia, D} = IaD + DIa, (30) where Ia is the coderivative operator introduced in Defini- where Ia is the coderivative operator introduced in Def- tion2, with q = 1, and D stands for the Lorentz-covariant 6 inition2, with q = 1, and D stands for the Lorentz- exterior derivative, D = d+ω. The proof to this Lemma covariant exterior derivative, D = d + ω. for the case of RC geometry was given in Refs. [51, 69].

The Da derivative plays a major role in the study of Lemma 10. The Da derivative introduced in Defini- GWs on RC geometries. It satisfies Leibniz’s rule (with- tion3 satisfies the following useful commutation relation out sign correction) and has many useful properties (see, with the Hodge star operator: e.g., Lemmas4 and5 below). b [Da, ∗] = IaT ∧ Ib∗. (37) Lemma 4. Let ∇µ = ∂µ + Γµ be the usual spacetime covariant derivative for the general (not necessarily tor- Most importantly, the operators Da,Ia, and D give rise sionless) affine connection Γρ . We have to a superalgebra of differential operators, where the cur- µσ vature and torsion play the role of structure “constants.” b Da = ∇a + IaT ∧ Ib, (31) This makes sense because Ia and D are odd (“fermionic”) operators, where Ia and Da are the operators introduced in Defini- µ p p+1 tions2 and3, and ∇a = ea ∇µ. Note that Eq. (31) D:Ω (M) → Ω (M) , (38) implies that Da and ∇a coincide in the torsionless case, p p−1 Ia :Ω (M) → Ω (M) , (39) D˚a = ∇˚a. while D is an even (“bosonic”) operator, Lemma 5. Equation (30) can be inverted to yield a p p a a Da :Ω (M) → Ω (M) . (40) D = e ∧ Da − T ∧ Ia, (32)

Theorem 11. The operators Da, Ia, and D (defined where Ia and Da are the operators introduced in Defini- tions2 and3. above) close on themselves, satisfying the following superalgebra: Definition 6. We define the generalized covariant coderivative D‡ :Ωp (M) → Ωp−1 (M) by [51] {Ia, D} = Da, (41) ‡ a {Ia, Ib} = 0, (42) D = −IaDI , (33) {D, D} = 2D2, (43) where I is the coderivative operator introduced in Def- a [I , D ] = −T c I , (44) inition2, with q = 1, and D stands for the Lorentz- a b ab c 2 2 covariant exterior derivative, D = d + ω. [D, Db] = D Ib − IbD , (45) 2 2 2 2 Definition 7 (Generalized de Rham–Laplace wave oper- [Da, Db] = IabD + D Iab + IaD Ib − IbD Ia c c ator). We define the generalized de Rham–Laplace wave − (DT ab ∧ Ic + T abDc) , (46) p p operator dR :Ω (M) → Ω (M) by [51] ‡ ‡ dR = D D + DD , (34) 6 where D‡ is the generalized covariant coderivative in- The torsionless (pseudo-) Riemannian case of this Lemma has been known for a long time (see, e.g., Ref. [85, Ch. V, Sec. B.4], troduced in Definition6 and D stands for the Lorentz- Ref. [86, Ch. 6.3] and Ref. [87]), but the original source of this covariant exterior derivative, D = d + ω. result for torsionless geometries is unknown to the authors. In fact, we have not been able to find any evidence of it ever appearing in the work of the Austrian mathematician Roland Weitzenböck (1885–1955). If the reader knows the real origin of 5 These operators were first defined in Ref. [51], where they were the Weitzenböck identity, we would be glad to be contacted and denoted as Σa1···aq . to learn about its actual attribution. 6 where D2 acts not as a differential operator but as one For instance, the Lorentz-covariant Lie derivatives of that, by virtue of the Bianchi identities, gives rise to the vielbein ea and the scalar field φ respectively read terms proportional to the Lorentz curvature two-form; 2 a a b 2 a 2 a a b a a a a e.g., D e = R b ∧ e , D R b = 0, and D T = R b ∧ T . Lξe = (IξD + DIξ) e = IξT + Dξ , (51) Proof. The (anti)commutation relations (41)–(45) are all Lξφ = (IξD + DIξ) φ = Iξdφ. (52) straightforward to prove. To show that Eq. (46) holds, it It is clear that, when acting on p-forms that behave as suffices to notice that [Da, Db] = [Da, {D, Ib}] and to use the super-Jacobi identity a scalar under LLTs, the Lorentz-covariant Lie derivative reduces to the standard one given by Cartan’s for-

{D, [Ib, Da]} + [Da, {D, Ib}] − {Ib, [Da, D]} = 0. (47) mula (48). One can check directly that

a a a b £ξe = Lξe + λ e , (53) In as few words as possible, and at the risk of gloss- b ing over some important subtleties, one may say that the £ξφ = Lξφ, (54) ab ab ab study of GWs in the context of RC geometry is very sim- £ξω = Lξω − Dλ , (55) ilar to the standard Riemannian case, but using the new derivative Da instead of the standard torsionless space- where λab = −I ωab plays the role of an infinitesimal ˚ ξ time covariant derivative ∇µ. local Lorentz parameter. This means that the difference between the usual Lie derivative and its Lorentz- covariant version amounts to an infinitesimal LLT. B. Lorentz-covariant Lie derivative The Lorentz-covariant Lie derivative is the suitable operator to define black hole entropy as the Noether charge The usual Lie derivative is not Lorentz covariant. For at the horizon in the first-order formalism, since the stan- instance, while the vielbein transforms as a vector under dard Lie derivative does not produce the correct transfor- local Lorentz transformations (LLT), its Lie derivative mation law for the vierbein at the bifurcation surface [94]. does not. Since LLTs are an essential part of our construction [e.g., the Lagrangian (1) is invariant under this gauge symmetry], we define a modified Lorentz-covariant IV. PERTURBATIONS ON A version of the Lie derivative that fixes this problem. RIEMANN–CARTAN GEOMETRY AND GAUGE FIXING Definition 12 (Cartan’s formula). When acting on a differential p-form, the Lie derivative operator along a vector field ξ~ is given by Cartan’s formula [88], A. Lie draggings and Lorentz transformations

£ξ = Iξd + dIξ, (48) In the usual torsionless GW treatment, it proves useful to define the trace-reversed version of the metric pertur- 7 where Iξ is the contraction operator. bation, Definition 13 (Lorentz-covariant Lie derivative). When ¯ 1 acting on a differential p-form that behaves as a tensor hµν = hµν − hgµν , (56) under LLTs, the Lorentz-covariant Lie derivative opera- 2 ~ tor along a vector field ξ is given by the formula [89–93] and then to perform a wisely chosen infinitesimal diffeomorphism on the metric, Lξ = IξD + DIξ, (49) ˚ ˚ where D is the Lorentz-covariant exterior derivative. On gµν 7→ gµν + ∇µξν + ∇ν ξµ, (57) the other hand, the Lorentz-covariant Lie derivative of the spin connection one-form ωab is defined as in order to arrive at the Lorenz gauge-fixing condition,

ab ab ∇˚ h¯µν = 0. (58) Lξω = IξR , (50) µ where Rab is the Lorentz curvature two-form. It is not trivial to generalize this procedure for the case of RC geometry. A first generalization was put forward in Ref. [51], but, while correct, it proved to be far from the best choice: the final result was a cumbersome inho- 7 Also called the interior product and denoted by ıξ or ξy. For our mogeneous GW equation with many torsional couplings. purposes, it proves most convenient to write I as [cf. Eq. (29)] ξ In this section, we use the Da derivative (see Definition3 d(p−1)+η− µ Iξ = (−1) ∗ (ξ ∧ ∗ , where ξ = ξµdx is the one-form in Sec.III) to provide a generalized Lorenz gauge fixing ~ µ dual to the vector field ξ = ξ ∂µ. in an optimal way. 7

An infinitesimal Lie dragging (LD) on the fields of the with theory corresponds to + 1 b c L ea = e [ξ (IbDcea + IaDceb) + Dbξa + Daξb]  a a ξ 2 δLD (ξ) e = Lξe ,  1 ab ab ˚ ˚ b LD : (59) = Dbξa + Daξb e δLD (ξ) ω = Lξω , 2   δ (ξ) φ = L φ, 1 LD ξ = D˚ξ + D˚ ξ , 2 a a where Lξ denotes the Lorentz-covariant Lie derivative operator along a vector field ξ~ (see Definition 13 in − 1 b c Sec.IIIB). Since the Lagrangian (1) is Lorentz invari- Lξ ea = e [ξ (IbDcea − IaDceb) + Dbξa − Daξb] ant, we have that 2 1 = − [D ξ − D ξ + (T − T ) ξc] eb, 2 a b b a abc bac LξL = dIξL a ab ¯ µ a = Lξe ∧ Ea + Lξω ∧ Eab + LξφE¯ + LξφE and where ξ = ξµdx = ξae is the one-form dual to the vector ξ~ = ξµ∂ = ξa~e . + d L ωab ∧ B + L φB¯ + L φ¯B . (60) µ a ξ ab ξ ξ Defining the antisymmetric parameter From this result we conclude that in a Lorentz-invariant 1 λ˜ = [D ξ − D ξ + (T − T ) ξc] , (68) Lagrangian, the only important piece of the LD is the ab 2 a b b a abc bac one described by the Lξ operator. Additionally, under a an infinitesimal LLT, the fields transform according to it is clear that Lξe contains a residual Lorentz transformation  a a b δLLT (λ) e = λ be ,  a + a ã b ab ab Lξe = Lξ e − λ be . (69) LLT : δLLT (λ) ω = −Dλ , (61)   δLLT (λ) φ = 0. A similar residual Lorentz freedom is found when Lξ acts on ωab and φ. The commutator of the infinitesimal LDs and LLTs, Therefore, we consider the final set of modified in- once applied to any gravitational field, form the Lie al- finitesimal Lie draggings (MLD) given by Lξ and a gebra counter-Lorentz transformation,  a + a a ã b [δLLT (λ1) , δLLT (λ2)] = δLLT (λ3) , (62)  δMLD(ξ)e = Lξ e = Lξe + λ be ,  ˜ MLD : δ (ξ)ωab = L+ωab = L ωab − Dλãb, (70) [δLLT (λ) , δLD (ξ)] = δLD ξ , (63) MLD ξ ξ  + ¯ ¯  δMLD(ξ)φ = Lξ φ = Lξφ, [δLD (ξ1) , δLD (ξ2)] = δLLT λ + δLD ξ , (64) with ab a cb a cb ã a b where we have defined λ3 = λ1cλ2 −λ2cλ1 , ξ = λ bξ , ¯ab ab ¯a a 1 λ = Iξ Iξ R , and ξ = Iξ Iξ T . The commutator be- + 2 1 1 2 L ea = D˚ξa + D˚aξ , (71) tween two LDs shows that curvature and torsion appear ξ 2 1 as “structure functions” of the algebra. Remarkably, this L+ωab = I Rab − D Daξb − Dbξa + T abc − T bac ξ , algebra closes off shell regardless of the dimensionality of ξ ξ 2 c the spacetime, its internal group, the field content of the (72) theory, and even in cases with restricted symmetries [93]. + Lξ φ = Iξdφ. (73) Furthermore, the invariance of the Lagrangian (1) under arbitrary LDs and LLTs implies the Noether identities This is the set of transformations we will use to generalize the standard gauge fixing of GWs. b bc ¯ DEa = IaT ∧ Eb + IaR ∧ Ebc + Daφ E + Daφ E¯, (65)

DEab = e[a ∧ Eb], (66) B. Lie draggings vs. gauge transformations respectively, which are also referred to as the contracted Bianchi identities. Before moving on, we would like to point out a com- The Lξ operator is a well-defined Lorentz-covariant mon misunderstanding regarding the interpretation of an version of the Lie derivative operator, but it still includes infinitesimal LD as a harmless “gauge transformation” on some residual Lorentz freedom. To see this, note that it the fields. First, the Lagrangian, the vierbein, the spin a is possible to write the action of Lξ on e in terms of the connection, and the scalar field, although invariant un- Dc derivative as der diffeomorphisms by virtue of being differential forms, transform nontrivially under infinitesimal LDs. Second, a + a − a Lξe = Lξ e + Lξ e , (67) diffeomorphism invariance is not a gauge symmetry in 8 the sense that there is no principal bundle involved (in The one-form Vab corresponds to a purely torsional per- sharp contrast to the local Lorentz symmetry). Third, turbation mode, independent of Ha. as shown in Sec.IIIB, the Lie derivative of a Lorentz- In terms of the perturbations, torsion and curvature tensor p-form does not transform covariantly under LLTs. behave as Therefore, it is certainly more suitable to take the LDs ¯ (1) (2) and LLTs as the fundamental symmetries of the theory Ta 7→ Ta = Ta + Ta + Ta , (81) in the first-order formalism. Rab 7→ R¯ab = Rab + Rab + Rab , (82) Given a well-behaved theory for a field ψ, with field (1) (2) equations written symbolically as E (ψ) = 0, we have with that (1) b 1 b Ta = Vab ∧ e − Ia H ∧ Tb , (83) E (ψ + Lξψ) = E (ψ) + LξE (ψ) = 0. (74) 2 1 1 T (2) = V ∧ Hb + I Hc ∧ I Hb ∧ T , (84) This means that it is possible to map in an invertible way a 2 ab 4 a c b an on-shell configuration into a different one satisfying Rab = DU ab + DV ab, (85) some practical condition we are interested in: the gauge (1) (1) ab ab a a cb cb fixing. Thus, solving the field equations for the latter is R(2) = DU(2) + U(1)c + V c ∧ U(1) + V . (86) equivalent to solving them for the former, and it is only in this restricted sense that an LD can be identified with We can always perform an MLD on the background a gauge transformation. and a GW perturbation simultaneously, 1 ea 7→ e¯a = ea + L+ea + Ha, (87) C. Perturbations on a Riemann–Cartan geometry ξ 2 defining a new GW as Infinitesimal LDs act on perturbations on the RC geometry in a way similar to the standard Riemannian case. 1 1 H0a = Ha + L+ea, (88) In Ref. [76], it was shown that they can be described up 2 2 ξ to second order through the perturbations in the vierbein and the spin connection given by and therefore we have what we might call the “gauge transformation,” 1 ea 7→ e¯a = ea + Ha, (75) 0 ˚ ˚ 2 Ha 7→ Ha = Ha + Dξa + Daξ. (89) ab ab ab ab ab ω 7→ ω¯ = ω + U (H, ∂H) + V . (76) It is possible to use this relation to prove that a a µ Here, H = H µdx is a one-form describing the vierbein 1 1 DaH0 − dH0 = DaD˚ ξ+IaDDI˚ ξ+DaH − dH, (90) perturbation, which is related to the canonical metric a 2 a a a 2 perturbation gµν 7→ gµν + hµν through µ a where ξ = ξµdx and H = H a. Since it is always µ a a ρ 1 ρ µ 1 ρ λ µ possible to find a one-form field ξ = ξµdx satisfying H µ = e ρ h ν − h µh ν + h λh µh ν + ··· , a ˚ a ˚ a 1 4 8 D Daξ + I DDIaξ + D Ha − 2 dH = 0, we can always (77) choose Ha such that a 1 a 1 hµν = e µ + H µ Haν . (78) D Ha − dH = 0. (91) 4 a 2 Without loss of generality, its orthonormal-frame com- This is the RC-geometry generalization of the standard µ¯ ponents can be taken as symmetric, Hba = Hab for any Lorenz gauge fixing ∇˚ hµν = 0 on the trace-reversed ¯ 1 Lorentz-invariant theory [76]. variable hµν = hµν − 2 gµν h of standard Riemannian ge- The perturbation on the spin connection comes in two ometry. pieces, Uab (H, ∂H) and Vab. The one-form Uab can be In the following sections, we will use the mathematical a (1) (2) 3 tools we have developed in Sec.III and the condition (91) written in terms of H as Uab = Uab + Uab + O H , where to study the propagation of GWs in the nonminimal GB- coupling case. 1 U (1) = − (I DH − I DH ) , (79) ab 2 a b b a 1 V. SIZES AND FREQUENCIES U (2) = I (DH ∧ Hc) ab 8 ab c 1 h i − I U (1) ∧ Hc − I U (1) ∧ Hc . (80) Even in the standard torsionless case, it is a nontrivial 2 a bc b ac task to separate GWs from the background geometry. In 9 general, one must consider an expansion in two kind of meaning that the torsional modes are much weaker than variables: amplitudes and frequencies. A GW is well de- the metric ones. fined only in the regime when small and rapidly changing The curvature perturbations (85)–(86) include 1/2 perturbations move over a slowly varying background. and 1/ terms on each order, We follow the approach of Ref. [75, Ch. 1.5] as closely as 1 H possible, but considering a nonvanishing torsion. DU ab ∼ , (100) (1) 2 2 Let us normalize the analysis by choosing a vierbein of L H components ab 1 H DV ∼ 2 , (101) a L V e ∼ 1, (92) µ 1 H2 DU ab + U a ∧ U cb ∼ , (102) (2) (1)c (1) 2 2 describing a slowly changing geometry over a character- L H istic length scale L. Since the torsion two-form is given 1 H2 by T a = dea + ωa ∧ eb, we infer that both torsion and U a ∧ V cb + V a ∧ U cb ∼ , (103) b (1)c c (1) L2 the spin connection must be of magnitude H 2 a cb H V c ∧ V ∼ . (104) a 1 ab 1 L2 |T | ∼ , ω ∼ , (93) L L At this point, since Eq. (17) has the same form as the while the Lorentz curvature turns out to be of magnitude canonical Einstein–Hilbert equations, and since the leading terms in the expansions are the metric modes, much 1 ab of the analysis goes exactly as in the standard case. The R ∼ 2 . (94) L field equations must be split into low- and high-frequency Let us label the amplitude scales for perturbations as8 pieces. From the low-frequency piece it is straightforward to prove that a ab |H | ∼ H, V ∼ V, (95) H H 1, (105) with H 1 and V 1. These perturbations change and taking this into consideration, the high-frequency rapidly in the wavelength scales λH and λV , piece of Eq. (17) to leading and subleading orders cor- a H ab V responds just to |∂H | ∼ , ∂V ∼ . (96) λ λ H V ab c abcdR(1) ∧ e = 0. (106) These wavelengths are small compared with the scale L of the background geometry, In Sec.VI, we analyze the behavior of GWs and torsional modes predicted by Eq. (106). λ λ = H 1, = V 1. (97) H L V L VI. THE DISPERSION RELATION, SPEED In order to relate the perturbation scales H and V , AND POLARIZATION OF GRAVITATIONAL let us observe that the torsion components are 1/L times WAVES smaller than those of the vierbein. Since the perturbed ¯ (1) torsion is given by [cf. Eqs. (83)–(84)] Ta = Ta + Ta + A. The wave equation and torsional obstruction to (2) 3 (1) (2) Ta +O H , it is natural to expect Ta and Ta to be the transverse-traceless gauge also 1/L times smaller than the vierbein perturbations, The left-hand side of Eq. (106) can be written as (see H H2 T (1) ∼ , T (2) ∼ . (98) AppendixA for the algebraic details) a L a L ab c 1 p m This is the same as requiring the perturbation scales of abcnR(1) ∧ e = InWm − ηmnIpW ∗e , (107) V ab and Ha to be related by 2 H where V ∼ , (99) L a a a Wm = −DaD Hm + [Da, Dm] H + 2IaDV m. (108)

It is clear that Eq. (106) and Eq. (107) imply Wm = 0 8 a as the equation for GWs, and therefore from now on the Later on we use H to denote the trace H a, which is of course unrelated to H as the scale for metric perturbations. We can equation only hope that the reader will be able to tell one from the other according to context. a a a − DaD Hm + [Da, Dm] H + 2IaDV m = 0, (109) 10 will be the protagonist of our analysis. a wisely chosen LD we may get a trace that is H times a As in the torsionless case, the commutator [Da, Dm] H smaller than the typical magnitude of the metric pertur- gives rise to some inhomogeneous terms with curvature bation. and torsion, In the next section, we will use the eikonal limit of GWs to obtain valuable information. For instance, from a a b [Da, Dm] H = Iam R b ∧ H simple inspection of Eq. (111) we can see that the dis- ab ab + Ia R Hbm + RmbH persion relation for Hm will not be modified at leading ab a b order by torsion. This means that Hm propagates at − DTabmH + TabmD H . (110) the speed of light on null geodesics, as in the standard Replacing these inhomogeneous terms and using the torsionless case. Even further, in Eq. (110) the terms a b ab ab ab generalized Weitzenböck identity of Lemma9 (see also Iam R b ∧ H ,Ia R Hbm + RmbH , and DTabmH 2 Ref. [69]), Wm can be written in terms of the generalized are all of order H/L and irrelevant to the eikonal limit at de Rham–Laplace wave operator (cf. Definition7) as both leading and subleading order. However, the terms a b a TabmD H and 2IaDV m in Eq. (111) modify the prop- a b Wm = dRHm + Iam R b ∧ H agation of GW polarization and the conservation of the − DT Hab + T DaHb + 2I DV a . (111) “number of rays” at subleading order, generalizing the abm abm a m results of Ref. [69] for fields obeying the homogeneous ‡ ‡ This implies that the analysis of Ref. [69] has to be equation D D + DD Hm = 0. In the next section we generalized to include the extra inhomogeneous terms in analyze these affirmations in detail. Eq. (111). A second important observation regarding p Wm is that its “trace,” IpW , consists only of torsional inhomogeneous terms besides the wave operator acting B. The eikonal limit of gravitational waves a on H = H a, p a abc d IpW = −DaD H + T DcHab + 2TbcdH a Let us write the vierbein and torsional perturbations a ab 10 ab H and V as + 2IabDV . (112) Ha = eiθHa,V ab = eiθVab, (115) In general lines, the metric mode of GWs behaves sim- ilarly to the standard torsionless case, albeit with an im- where θ is a rapidly changing (on a characteristic scale portant and subtle difference. Let us observe that, in the λ) real phase, which, for simplicity, we take to be the a a b standard case, besides the Lorenz gauge fixing (58), it is same for both geometrical modes. Here, H = H be ab ab c possible to perform (in a vacuum region only) an addi- and V = V ce correspond to slowly changing (on tional gauge transformation to render h traceless: the a characteristic scale L) one-forms with complex-valued µν a ab transverse-traceless gauge. In our case, things are a bit components H b and V c. more complicated. Under an infinitesimal LD generated In terms of the characteristic scales λ and L we define ˜ a the eikonal parameter = λ/L. From Eqs. (99) and (105) by ξ, the trace of the metric perturbation, H = H a, changes as one can easily show that it satisfies ã V H 1. (116) H 7→ H˜ = H + 2D˚aξ . (113) In terms of these, the transverse condition (91) becomes Naively it may seem possible to choose a ξ˜ such that ˜ ˜ a ˚ ˜ a ˚ ˜ 1 1 H = 0, as long as ξ also satisfies D Daξ + I DDIaξ = 0 D Ha − dH = ieiθ k Ha − kH in order not to spoil the Lorenz condition (91). The a 2 a 2 problem with such a construction is Eq. (112). A trace- 1 a + eiθ D Ha − dH , (117) less H would create an unphysical constraint between a 2 the perturbations and torsion, while Wm corresponds to T abc D H + 2T Hd + 2I DV ab = 0, (114) c ab bcd a ab 1 W = eiθ k kaH − 2i ka D H + T Hb and therefore, in general we must have H 6= 0.9 Torsion m a m a m 2 abm thus creates an obstruction to the popular transverse- 1 −I (k ∧ Va ) + H D ka traceless gauge. However, comparing orders of magni- a m 2 m a tude in the terms of Eq. (112), we conclude that using a a a − DaD Hm + [Da, Dm] H + 2IaDV m , (118)

9 In the standard GR torsionless case, the constraint in Eq. (114) vanishes identically and it is possible to impose h = 0 in a vac- 10 uum region. Some further conditions, such as h0µ = 0, are only The physical perturbations correspond to the real parts of these possible on a ﬂat background even in the torsionless case. complex quantities. 11

a b where the wave one-form k is given by This result is equivalent to k D˚ak = 0, and since D˚a = ∇˚a, we get a µ k = dθ = kae = kµdx . (119) a b k ∇˚ak = 0. (126) Equation (118) generalizes Eq. (61) of Ref. [69] to take into account the inhomogeneous terms in Eq. (111). This means that the metric mode of GWs travels along We can expand Ha and Vab as null geodesics, not null auto-parallels. At subleading order, torsion gives rise to an anomalous ∞ ∞ X X propagation of polarization. In order to analyze it, let us Ha = Ha , Vab = Vab , (120) a (n) (n) parametrize the wave polarization and amplitude as H(0) n=0 n=0 ab and V(0) through where Ha and Vab are of order n. The leading orders, (n) (n) Ha = HP a, (127) Ha and Vab , correspond to dominant, λ-independent (0) (0) (0) ab ab pieces. The subleading order, n = 1, describes the prop- V(0) = VQ . (128) agation of polarization, while terms with n ≥ 2 corre- Here, wave polarization is described by the one-forms spond to higher-order deviations from the geometric op- P a = P a ec and Qab = Qab ec, while wave amplitude tics limit. c c is described by H and V. The polarization components To leading order, the equation W = 0 implies the m are complex valued, P a ,Qab ∈ , while the amplitude canonical dispersion relation for the metric mode, c c C scalars are real, H, V ∈ R. The polarization forms are a µ normalized as kak = kµk = 0, (121) P¯ ∧ ∗P a = v , (129) and the standard transverse condition, a (4) ¯ ab Qab ∧ ∗Q = v(4), (130) a 1 kaH(0) − kH(0) = 0. (122) 1 a b c d 2 where v(4) = 4! abcde ∧ e ∧ e ∧ e is the volume four- form and a bar above a quantity denotes its complex This means that, to leading order in the dispersion rela- conjugate. This implies the following normalization on tion, there is no diﬀerence with the standard GR torsion- the wave polarization and amplitude: less case. However, at subleading order, Wm = 0 gives rise to new interactions with torsion, ¯(0) a 2 ∗ Ha ∧ ∗H(0) = −H , (131) 1 (0) 0 = ka D H(0) + T Hb ∗ V¯ ∧ ∗Vab = −V2. (132) a m 2 abm (0) ab (0)

a 1 (0) a Let J = H2k be the “number-of-rays” current den- − Ia k ∧ V m + H Dak , (123) (0) 2 m sity one-form. When considering the eikonal limit and standard Riemannian geometry, this form is conserved, and the transverse condition assumes the form d†J = 0. However, this is no longer true for a geometry 1 with nonvanishing torsion, as was first shown in Ref. [69] D Ha − dH = 0. (124) a (0) 2 (0) for the homogeneous wave equation case. The current situation is similar, but new terms have to be added. † ab c a It may seem strange to have the usual dispersion rela- Since d J = Tabcη J −DaJ , let us start by computing tion (121) even in this case. On a geometry with nonvan- a 2 a 2 a ishing torsion, one may expect GWs to propagate on null DaJ = Da H k + H Dak . (133) auto-parallels.11 However, this is not the case: Eq. (121) implies that GWs propagate on null geodesics, regardless Using Eq. (131), Lemma 10 and Eq. (127), we have that of the background torsion. 2 c c Da H = −∗ DaH¯c ∧ ∗H + DaHc ∧ ∗H¯ Differentiating Eq. (121) and using ka = Daθ, we find 2 bc − H TbcaΠ , (134) a c k (Dakb + T abkc) = 0. (125) where 1 Πab = ηab − P¯caP b + P caP¯ b . (135) 2 c c 11 It is worth noticing that, when torsion is present, geodesics and auto-parallels do not necessarily coincide: while the former are This allows us to write curves of extremal length with respect to the metric, the latter a a c a c are curves over which a vector is parallel transported with re- DaJ = −∗ k DaH¯c ∧ ∗H + k DaHc ∧ ∗H¯ spect to itself according to the connection (see Ref. [49] for a ab c 2 a discussion). − TabcΠ J + H Dak . (136) 12

Using Eq. (123), we find have been dramatically constrained, with the scalar-GB coupling a particular example of this class. In the present a V ab c article we showed that dispensing with the torsionless DaJ = Θc − TabcΠ J , (137) H condition—usually assumed in gravitational theories— allows the latter case to be reconciled with observations, with in stark contrast with the usual torsionless case. pq ba pq ba Our starting point is a fairly standard scalar-tensor Θc = Q¯c Ppq + Q¯ aPbc + Qc P¯pq + Q aP¯bc . (138) gravity theory, defined by the Lagrangian (1). Crucially, From here, we can see that J is no longer conserved in removing the torsionless condition makes the vierbein the eikonal limit, that is and the spin connection independent degrees of freedom, and the same Lagrangian can give rise to two radically V different dynamical theories, depending on whether or d†J = T ηab + Πab − Θ J c. (139) abc H c not this hypothesis is assumed a priori. One common misconception is to consider the torsional case as an ex- Observe that, even on backgrounds with vanishing tor- otic and small departure from the torsionless case. That sion, a nonzero torsional perturbation will lead to a may be true for some of the ECSK phenomenology, but breakdown in the number-of-rays conservation, d†J 6= 0. it is definitely not the case for large sectors of the Horn- Another consequence of torsion is that the polarization deski Lagrangian [51]. Another misconception is the be- one-form P a is no longer parallel transported along the lief that in order to recover the standard Riemannian GW trajectory. This can be seen by using Eqs. (123) geometry, it suffices to impose T a = 0 in all equations. and (137), which allow us to write Both misconceptions arise from the failure to recognize that the torsionless condition is a strong constraint on a b 1 ab c 0 = k DaPm + TcbmP − TabcΠ Pm k the geometry, in the sense that we must add a Lagrange 2 multiplier to impose it, as shown in Eqs. (19)–(25) and V 1 Ref. [51]. In this sense, the torsionless condition amounts + Θ kcP − I (k ∧ Qa ) , (140) H 2 c m a m to an unproven hypothesis. Imposing it means adding a hypothesis to the theory and a constraint to the geome- which in component language reads try. The dynamics derived from the field equations makes c 1 c p p ab k ∇˚cPmn = k TmpcP n + TnpcP m + TabcΠ Pmn it plausible to expect that GWs propagate at the speed 2 of light.12 To prove it, we have developed some new V a + (Qcmn + Qcnm + ηmcQna mathematical tools to study the wave operator on an RC H geometry and a new approach to study perturbations on a + ηncQma − ΘcPmn)] . (141) a background with dynamical torsion. Besides the standard metric mode, there is an additional one associated From Eq. (141) we observe that the polarization compo- to the torsional degrees of freedom. We showed how to nents P are parallel transported along the trajectory mn use the generalized Lie derivative to find a “generalized of the GW (i.e., the relation kc∇˚ P = 0 is fulfilled) c mn Lorenz gauge fixing” for the case of nonvanishing torsion. only in the Riemannian geometry case, when both back- These provide the necessary mathematical tools to tackle ground torsion and its perturbations vanish, T = 0, abc the problem of GWs on a dynamical torsion background, V ab = 0. In the general case, the propagation of po- which allowed us to determine their speed in the current larization will be disturbed along the GW trajectory by theory. interactions with background torsion and “rotons.” After a general analysis of sizes and frequencies, we To summarize, torsion does not change the dispersion found the inhomogeneous gravitational wave equation for relation (121), and therefore it changes neither the speed this theory [cf. Eq. (109)], and we discovered that torsion nor the direction of propagation of GWs. At leading obstructs the popular transverse-traceless gauge. Then, order in the eikonal approximation, its propagation re- we proceeded with the geometric optics (eikonal) ap- mains the same as in standard GR. The only difference proximation. To leading order, we recovered the canon- lies in how torsion changes the propagation of polariza- ical dispersion relation k kµ = 0, implying that GWs tion along the GW trajectory [cf. Eq. (141)]. µ travel on null geodesics. The point is subtle but essential. EMWs always travel on null geodesics, regardless of any background torsion [69]. If GWs would have trav- VII. CONCLUSIONS AND FUTURE POSSIBILITIES

Observational data coming from multimessenger as- 12 We notice that a similar behavior occurs in the ﬁrst-order for- tronomy indicate, to a very high precision, that GWs mulation of Chern–Simons modiﬁed gravity [65], even though its travel at the speed of light. Due to this, several scalar- metric formulation is compatible with the luminal propagation tensor theories that predict an anomalous GW speed of GWs, as shown in Ref. [17]. 13

(out) Outgoing GW, Pmn

GW detector

(in) Incoming GW, Pmn

FIG. 1. Conceptual diagram of a possible way to map torsion “clouds” inside a region of space by measuring changes in the polarization of incoming/outgoing GWs. Each small triangle represents a GW detector. eled on null auto-parallels, this might lead to an unob- Let us consider space-based GW detectors distributed served delay between GW/EMW, even if both of them on the surface of a vast spherical region, as suggested in were traveling at the same speed. Our results show that Fig.1 (e.g., GW detectors orbiting the Sun at a distance this is not the case: EMWs and GWs travel at the same of several AU at different angles with the ecliptic plane). speed and on the same kind of trajectory. If each detector is capable of measuring GW polariza- Torsion does affect GWs at subleading order, though. tion, the whole set of detectors could make a “torsion In the torsionless case, both EMWs and GWs satisfy tomography” of sorts of the enclosed region, compar- two eikonal limit conditions: (i) the number-of-rays cur- ing the polarization components Pmn of incoming and rent density is conserved, and (ii) polarization is parallel outgoing GWs. For instance, the degree of rotation of transported along the null geodesic trajectory. In the GW polarization would encode information about tor- current torsional context, these conditions are satisfied sion within the sphere. This technique may seem (and by EMWs but they are no longer valid for GWs: tor- probably is) far fetched, given current technological pos- sion breaks down the conservation of the number of rays sibilities. However, from a certain point of view, it is and GW polarization is no longer parallel transported also a very conservative idea. One may regard it as the along the null geodesic. As discussed in Ref. [69], this scaled-up gravitational version of Faraday’s 1845 setup behavior is generic of waves propagating on a torsional for measuring what is now known as Faraday rotation background, and not just a feature of the GB coupling. (see Fig.2). In this effect, certain transparent dielectric materials present circular birefringence when a magnetic This fact opens up many questions regarding how tor- field goes through them. The different speeds of the two sion modifies the propagation phenomenology of metric circular polarization modes have the effect of rotating the and torsional polarization, which lie beyond the scope of plane of polarization of light going through the dielec- this paper. The fact that GW polarization is affected by tric parallel to the magnetic field, providing information torsion means, however, that one may envision ways to on the properties of the material. In the gravitational use it to study the distribution of torsion. 14

ICYT) grant 72160340. He wishes to thank Dieter Lüstfor his kind hospitality at the Arnold Sommerfeld Center for Theoretical Physics in Munich. The work of C. C. is supported by Proyecto POSTDOC DICYT, Código041931CM POSTDOC, Universidad de Santiago de Chile. F. I. is funded by Fondo Nacional de Desarrollo Cient´ıficoy Tecnológico(FONDECYT) grant 1180681. P. M. is supported by ComisiónNacional de Investigación Cient´ıfica y Tecnológica(CONICYT) grant 21161574. O. V. is supported by Project VRIIP0062-19, Univer- sidad Arturo Prat, Chile.

Appendix A: Detailed derivation of the wave equation

Replacing Eqs. (85), (79) and (32) in Eq. (106), it is straightforward to show that FIG. 2. Figure from Michael Faraday’s diary [95], Septem- ber 13, 1845. A piece of transparent glass (“silico borate 1 1 of lead”) rotated the plane of polarization of the light go- ab c p m abcnR(1) ∧ e = Wmn − ηmnW p ∗e , (A1) ing through the glass in the same direction as the applied 2 2 magnetic field. Faraday did not understand at the time the mechanism behind this phenomenon, but he knew that the where change in polarization was a probe codifying important infor- m p h am ami mation on the ray of light, the nature of the magnetic field, W n = (InDa − IaDn + T naIp) U(1) + V . (A2) and the material itself. After many other tests, he famously wrote in entry 7718, September 30, 1845: “Still, I have at last Using the commutator from Eq. (44), succeeded in illuminating a magnetic curve or line of force and in magnetizing a ray of light.” This experiment was the first T p ∧ I = I D − D I , (A3) indication of a deeper relationship between optics and elec- na p a n n a tromagnetism, two decades before Maxwell’s prediction [96] W m is given by of light being an EMW in 1865. n

m h am ami W n = (InDa − DnIa) U(1) + V . (A4) version, the change in the propagation of polarization is caused not by birefringence, since all GW polarization (1) It is possible to prove that the Uab term in Eq. (79) modes travel on null geodesics, but by the interaction can be rewritten as with torsion. There exists, of course, the possibility of such an experiment indicating that torsion vanishes. In (1) 1 U = − (DaHb − DbHa) , (A5) such a case, the torsionless condition would become a ab 2 valuable clue from nature on the kind of solutions of our and therefore theories that match reality rather than an untested hypothesis on their structure. 1h W = − I D DaH + I D D Ha mn 2 n a m n a m i + D I (DaH − D Ha) ACKNOWLEDGMENTS n a m m a + (InDa − DnIa) V m. (A6) The authors wish to thank Yuri Bonder, Jens Boos, Fabrizio Canfora, Oscar Castillo-Felisola, Norman Cruz, The “Lorenz gauge fixing” in Eq. (91) can be rewritten NicolásGonzález,Julio Oliva, Miguel Pino, Francisca as

Ram´ırez, Patricio Salgado, Sebastián Salgado, Jorge a a a Zanelli, and Alfonso Zerwekh for enlightening discus- ImDaH + Ia (D Hm − DmH ) = 0, (A7) sions and comments. J. B. acknowledges financial support from ComisiónNacional de InvestigaciónCient´ıfica and therefore y Tecnológica (CONICYT) grant 21160784 and also 1h W = − I D DaH + I [D , D ] Ha thanks the Institute of Mathematics of the Czech mn 2 n a m n a m Academy of Sciences, where part of this work was car- ai a ried out. F. C.-T. was supported by Comisión Na- + (InDm − DnIm) DaH + (InDa − DnIa) V m. cional de InvestigaciónCient´ıficay Tecnológica(CON- (A8) 15

However, using Eq. (30) it is straightforward to show in this gauge we have that that 1 (I D − D I ) D Ha = (I D − DI ) dH = 0. n m n m a 2 mn mn (A10) Inserting this result into Eq. (A8), we get the inhomogeneous wave equation

I D − D I = I D − DI , (A9) 1 n m n m mn mn W = (−I D DaH + I [D , D ] Ha) mn 2 n a m n a m a + (IanD − DIan) V m 1 = I (−D DaH + [D , D ] Ha + 2I DV a ) . 2 n a m a m a m (A11) a 1 and since from the “Lorenz gauge ﬁxing” DaH = 2 dH,

[1] B. P. Abbott et al. (LIGO Scientific, Virgo), Phys. Rev. [14] E. Babichev and A. Lehébel, JCAP 1812, 027 (2018), Lett. 119, 161101 (2017), arXiv:1710.05832 [gr-qc]. arXiv:1810.09997 [gr-qc]. [2] B. P. Abbott et al. (LIGO Scientific, Virgo, Fermi [15] S. Alexander and N. Yunes, Phys. Rep. 480, 1 (2009), GBM, INTEGRAL, IceCube, AstroSat Cadmium Zinc arXiv:0907.2562 [hep-th]. Telluride Imager Team, IPN, Insight-Hxmt, ANTARES, [16] K. Konno, T. Matsuyama, Y. Asano, and S. Tanda, Swift, AGILE Team, 1M2H Team, Dark Energy Camera Phys. Rev. D 78, 024037 (2008), arXiv:0807.0679 [gr-qc]. GW-EM, DES, DLT40, GRAWITA, Fermi-LAT, ATCA, [17] A. Nishizawa and T. Kobayashi, Phys. Rev. D 98, 124018 ASKAP, Las Cumbres Observatory Group, OzGrav, (2018), arXiv:1809.00815 [gr-qc]. DWF (Deeper Wider Faster Program), AST3, CAAS- [18] K. Konno, T. Matsuyama, and S. Tanda, Phys. Rev. D TRO, VINROUGE, MASTER, J-GEM, GROWTH, 76, 024009 (2007), arXiv:0706.3080 [gr-qc]. JAGWAR, CaltechNRAO, TTU-NRAO, NuSTAR, Pan- [19] D. Grumiller and N. Yunes, Phys. Rev. D 77, 044015 STARRS, MAXI Team, TZAC Consortium, KU, Nordic (2008), arXiv:0711.1868 [gr-qc]. Optical Telescope, ePESSTO, GROND, Texas Tech [20] K. Konno, T. Matsuyama, and S. Tanda, Prog. Theor. University, SALT Group, TOROS, BOOTES, MWA, Phys. 122, 561 (2009), arXiv:0902.4767 [gr-qc]. CALET, IKI-GW Follow-up, H.E.S.S., LOFAR, LWA, [21] N. Yunes and F. Pretorius, Phys. Rev. D 79, 084043 HAWC, Pierre Auger, ALMA, Euro VLBI Team, Pi of (2009), arXiv:0902.4669 [gr-qc]. Sky, Chandra Team at McGill University, DFN, AT- [22] H. Ahmedov and A. N. Aliev, Phys. Rev. D 82, 024043 LAS Telescopes, High Time Resolution Universe Sur- (2010), arXiv:1003.6017 [hep-th]. vey, RIMAS, RATIR, SKA South Africa/MeerKAT), As- [23] Y. Brihaye and E. Radu, Phys. Lett. B 764, 300 (2017), trophys. J. 848, L12 (2017), arXiv:1710.05833 [astro- arXiv:1610.09952 [gr-qc]. ph.HE]. [24] S. Alexander and N. Yunes, Phys. Rev. Lett. 99, 241101 [3] B. P. Abbott et al. (LIGO Scientific, Virgo, Fermi- (2007), arXiv:hep-th/0703265 [hep-th]. GBM, INTEGRAL), Astrophys. J. 848, L13 (2017), [25] S. Alexander, L. S. Finn, and N. Yunes, Phys. Rev. D arXiv:1710.05834 [astro-ph.HE]. 78, 066005 (2008). [4] E. A. Huerta et al., Nat. Rev. Phys. 1, 600 (2019). [26] N. Yunes, K. Yagi, and F. Pretorius, Phys. Rev. D 94, [5] D. Bettoni, J. M. Ezquiaga, K. Hinterbichler, and 084002 (2016), arXiv:1603.08955 [gr-qc]. M. Zumalacárregui, Phys. Rev. D 95, 084029 (2017), [27] S. H. Alexander and N. Yunes, Phys. Rev. D 97, 064033 arXiv:1608.01982 [gr-qc]. (2018), arXiv:1712.01853 [gr-qc]. [6] J. M. Ezquiaga and M. Zumalacárregui, Phys. Rev. Lett. [28] M. Gurses, Gen. Rel. Grav. 40, 1825 (2008), 119, 251304 (2017), arXiv:1710.05901 [astro-ph.CO]. arXiv:0707.0347 [gr-qc]. [7] J. M. Ezquiaga and M. Zumalacárregui, Front. Astron. [29] L. N. Granda, Mod. Phys. Lett. A 27, 1250018 (2012), Space Sci. 5, 44 (2018), arXiv:1807.09241 [astro-ph.CO]. arXiv:1108.6236 [hep-th]. [8] T. Baker, E. Bellini, P. G. Ferreira, M. Lagos, J. Noller, [30] L. N. Granda, Int. J. Theor. Phys. 51, 2813 (2012), and I. Sawicki, Phys. Rev. Lett. 119, 251301 (2017), arXiv:1109.1371 [gr-qc]. arXiv:1710.06394 [astro-ph.CO]. [31] L. N. Granda and E. Loaiza, Int. J. Mod. Phys. D 21, [9] J. Sakstein and B. Jain, Phys. Rev. Lett. 119, 251303 1250002 (2012), arXiv:1111.2454 [hep-th]. (2017), arXiv:1710.05893 [astro-ph.CO]. [32] P. Kanti, R. Gannouji, and N. Dadhich, Phys. Rev. D [10] L. Heisenberg and S. Tsujikawa, JCAP 1801, 044 (2018), 92, 083524 (2015), arXiv:1506.04667 [hep-th]. arXiv:1711.09430 [gr-qc]. [33] D. D. Doneva and S. S. Yazadjiev, Phys. Rev. Lett. 120, [11] C. D. Kreisch and E. Komatsu, JCAP 1812, 030 (2018), 131103 (2018), arXiv:1711.01187 [gr-qc]. arXiv:1712.02710 [astro-ph.CO]. [34] H. O. Silva, J. Sakstein, L. Gualtieri, T. P. Sotiriou, [12] S. Nojiri and S. D. Odintsov, Phys. Lett. B 779, 425 and E. Berti, Phys. Rev. Lett. 120, 131104 (2018), (2018), arXiv:1711.00492 [astro-ph.CO]. arXiv:1711.02080 [gr-qc]. [13] S. Mukherjee and S. Chakraborty, Phys. Rev. D 97, [35] G. Antoniou, A. Bakopoulos, and P. Kanti, Phys. Rev. 124007 (2018), arXiv:1712.00562 [gr-qc]. D 97, 084037 (2018), arXiv:1711.07431 [hep-th]. 16

[36] G. Antoniou, A. Bakopoulos, and P. Kanti, Phys. Rev. [65] S. Alexander and N. Yunes, Phys. Rev. D 77, 124040 Lett. 120, 131102 (2018), arXiv:1711.03390 [hep-th]. (2008), arXiv:0804.1797 [gr-qc]. [37] D. D. Doneva and S. S. Yazadjiev, JCAP 1804, 011 [66] A. Cisterna, C. Corral, and S. del Pino, Eur. Phys. J. C (2018), arXiv:1712.03715 [gr-qc]. 79, 400 (2019), arXiv:1809.02903 [gr-qc]. [38] D. D. Doneva, S. Kiorpelidi, P. G. Nedkova, E. Pa- [67] S. Alexander, M. Cortês, A. R. Liddle, J. Magueijo, pantonopoulos, and S. S. Yazadjiev, Phys. Rev. D 98, R. Sims, and L. Smolin, Phys. Rev. D 100, 083506 104056 (2018), arXiv:1809.00844 [gr-qc]. (2019), arXiv:1905.10380 [gr-qc]. [39] A. Bakopoulos, G. Antoniou, and P. Kanti, Phys. Rev. [68] S. Alexander, M. Cortês, A. R. Liddle, J. Magueijo, D 99, 064003 (2019), arXiv:1812.06941 [hep-th]. R. Sims, and L. Smolin, Phys. Rev. D 100, 083507 [40] Y. S. Myung and D.-C. Zou, Int. J. Mod. Phys. D 28, (2019), arXiv:1905.10382 [gr-qc]. 1950114 (2019), arXiv:1903.08312 [gr-qc]. [69] J. Barrientos, F. Izaurieta, E. Rodr´ıguez, and O. Val- [41] G. Antoniou, A. Bakopoulos, P. Kanti, B. Kleihaus, and divia, arXiv:1903.04712 [gr-qc]. J. Kunz, arXiv:1904.13091 [hep-th]. [70] M. Hohmann, C. Pfeifer, J. L. Said, and U. Ualikhanova, [42] Y. S. Myung and D.-C. Zou, Phys. Rev. D 98, 024030 Phys. Rev. D 99, 024009 (2019), arXiv:1808.02894 [gr- (2018), arXiv:1805.05023 [gr-qc]. qc]. [43] J. L. Blázquez-Salcedo,D. D. Doneva, J. Kunz, and [71] I. Soudi, G. Farrugia, V. Gakis, J. Levi Said, and S. S. Yazadjiev, Phys. Rev. D 98, 084011 (2018), E. N. Saridakis, Phys. Rev. D 100, 044008 (2019), arXiv:1805.05755 [gr-qc]. arXiv:1810.08220 [gr-qc]. [44] H. O. Silva, C. F. B. Macedo, T. P. Sotiriou, L. Gualtieri, [72] S. Bahamonde, K. F. Dialektopoulos, V. Gakis, and J. Sakstein, and E. Berti, Phys. Rev. D 99, 064011 J. Levi Said, arXiv:1907.10057 [gr-qc]. (2019), arXiv:1812.05590 [gr-qc]. [73] F. Bombacigno and G. Montani, Phys. Rev. D 97, 124066 [45] Y. Gong, E. Papantonopoulos, and Z. Yi, Eur. Phys. J. (2018), arXiv:1804.03897 [gr-qc]. C 78, 738 (2018), arXiv:1711.04102 [gr-qc]. [74] F. Bombacigno and G. Montani, Phys. Rev. D 99, 064016 [46] R. Debever, ed., Elie Cartan and Albert Einstein: Letters (2019), arXiv:1810.07607 [gr-qc]. on Absolute Parallelism, 1929–1932 (Princeton Univer- [75] M. Maggiore, Gravitational Waves. Vol. 1: Theory and sity Press, Princeton, NJ, 1979). Experiments, Oxford Master Series in Physics (Oxford [47] T. W. B. Kibble, J. Math. Phys. 2, 212 (1961). University Press, Oxford, 2007). [48] D. W. Sciama, Rev. Mod. Phys. 36, 463 (1964), [Erra- [76] F. Izaurieta, E. Rodr´ıguez, and O. Valdivia, Eur. Phys. tum: Rev. Mod. Phys. 36, 1103 (1964)]. J. C 79, 337 (2019), arXiv:1901.06400 [gr-qc]. [49] F. W. Hehl, P. von der Heyde, G. D. Kerlick, and J. M. [77] F. W. Hehl, in Cosmology and Gravitation: Spin, Nester, Rev. Mod. Phys. 48, 393 (1976). Torsion, Rotation, and Supergravity, NATO Advanced [50] M. Blagojevićand F. W. Hehl, eds., Gauge Theories of Study Institutes Series, Vol. 58, edited by P. G. Gravitation (World Scientific, Singapore, 2013). Bergmann and V. D. Sabbata (Springer, Boston, MA, [51] J. Barrientos, F. Cordonier-Tello, F. Izaurieta, P. Med- 1980) pp. 5–61. ina, D. Narbona, E. Rodr´ıguez, and O. Valdivia, Phys. [78] J. Boos and F. W. Hehl, Int. J. Theor. Phys. 56, 751 Rev. D 96, 084023 (2017), arXiv:1703.09686 [gr-qc]. (2017), arXiv:1606.09273 [gr-qc]. [52] H. T. Nieh and M. L. Yan, J. Math. Phys. 23, 373 (1982). [79] A. Mardones and J. Zanelli, Class. Quant. Grav. 8, 1545 [53] S. Mercuri, Phys. Rev. Lett. 103, 081302 (2009), (1991). arXiv:0902.2764 [gr-qc]. [80] G. W. Horndeski, Int. J. Theor. Phys. 10, 363 (1974). [54] O. Chandia and J. Zanelli, Phys. Rev. D 55, 7580 (1997), [81] R. Jackiw and S. Y. Pi, Phys. Rev. D 68, 104012 (2003), arXiv:hep-th/9702025 [hep-th]. arXiv:gr-qc/0308071 [gr-qc]. [55] Y. N. Obukhov, E. W. Mielke, J. Budczies, and F. W. [82] R. Aros, M. Contreras, R. Olea, R. Troncoso, and Hehl, Found. Phys. 27, 1221 (1997), arXiv:gr-qc/9702011 J. Zanelli, Phys. Rev. Lett. 84, 1647 (2000), arXiv:gr- [gr-qc]. qc/9909015 [gr-qc]. [56] D. Kreimer and E. W. Mielke, Phys. Rev. D 63, 048501 [83] R. Aros, M. Contreras, R. Olea, R. Troncoso, and (2001), arXiv:gr-qc/9904071 [gr-qc]. J. Zanelli, Phys. Rev. D 62, 044002 (2000), arXiv:hep- [57] O. Chandia and J. Zanelli, Phys. Rev. D 63, 048502 th/9912045 [hep-th]. (2001), arXiv:hep-th/9906165 [hep-th]. [84] H. Flanders, Differential Forms with Applications to the [58] M. Lattanzi and S. Mercuri, Phys. Rev. D 81, 125015 Physical Sciences (Dover Publications, New York, 1989). (2010), arXiv:0911.2698 [gr-qc]. [85] Y. Choquet-Bruhat, C. DeWitt-Morette, and [59] O. Castillo-Felisola, C. Corral, S. Kovalenko, I. Schmidt, M. Dillard-Bleick, Analysis, Manifolds and Physics, and V. E. Lyubovitskij, Phys. Rev. D 91, 085017 (2015), 2nd ed., Vol. I: Basics (North Holland Publishing arXiv:1502.03694 [hep-ph]. Company, Amsterdam, 1982). [60] G. K. Karananas, Eur. Phys. J. C 78, 480 (2018), [86] P. A. Griffiths and J. Harris, Principles of Algebraic Ge- arXiv:1805.08781 [hep-th]. ometry (Wiley, New York, 1978). [61] O. Castillo-Felisola, C. Corral, S. del Pino, and [87] D. Bini, C. Cherubini, R. T. Jantzen, and R. Ruffini, F. Ram´ırez, Phys. Rev. D 94, 124020 (2016), Int. J. Mod. Phys. D 12, 1363 (2003). arXiv:1609.09045 [gr-qc]. [88] M. Nakahara, Geometry, Topology and Physics, 3rd ed. [62] A. Toloza and J. Zanelli, Class. Quant. Grav. 30, 135003 (Taylor & Francis, London, 2016). (2013), arXiv:1301.0821 [gr-qc]. [89] F. W. Hehl, J. D. McCrea, E. W. Mielke, and [63] J. L. Espiro and Y. Vásquez, Gen. Rel. Grav. 48, 117 Y. Ne’eman, Phys. Rep. 258, 1 (1995), arXiv:gr- (2016), arXiv:1410.3152 [gr-qc]. qc/9402012 [gr-qc]. [64] A. Cid, F. Izaurieta, G. Leon, P. Medina, and D. Nar- [90] Y. N. Obukhov and G. F. Rubilar, Phys. Rev. D 74, bona, JCAP 1804, 041 (2018), arXiv:1704.04563 [gr-qc]. 064002 (2006), arXiv:gr-qc/0608064 [gr-qc]. 17

[91] Y. N. Obukhov and G. F. Rubilar, Phys. Rev. D 76, [94] T. Jacobson and A. Mohd, Phys. Rev. D 92, 124010 124030 (2007), arXiv:0712.3547 [hep-th]. (2015), arXiv:1507.01054 [gr-qc]. [92] Y. N. Obukhov and G. F. Rubilar, Phys. Lett. B 660, [95] M. Faraday, Faraday’s Diary, edited by T. Martin, Vol. 4 240 (2008), arXiv:0712.3551 [hep-th]. (HR Direct, Riverton, UT, 2009). [93] C. Corral and Y. Bonder, Class. Quant. Grav. 36, 045002 [96] J. Clerk Maxwell, Phil. Trans. R. Soc. 155, 459 (1865). (2019), arXiv:1808.01497 [gr-qc].