Premetric teleparallel theory of gravity and its local and linear constitutive law

We continue to investigate the premetric teleparallel theory of gravity (TG) with the coframe (tetrad) as gravitational potential. We start from the field equations and a local and linear constitutive law. We create a Tonti diagram of TG in order to disclose the structure of TG. Subsequently we irreducibly decompose the 6th order constitutive tensor under the linear group. Moreover, we construct the most general constitutive tensors from the metric and the totally antisymmetric Levi-Civita symbol, and we demonstrate that they encompass nontrivial axion and skewon type pieces. Using these tools, we derive for TG in the geometric-optics approximation propagating massless spin 0, 1, and 2 waves, including the special case of Einstein’s general relativity.


Introduction
Lately we followed in [18] the program of Kottler of 1922 to remove the metric tensor of spacetime, the gravitational potential within general relativity theory (GR), from the fun-damental laws of classical electromagnetism and gravity as far as possible. In particular, we applied this to the theory of gravity [31] in that we started from a translational gauge theory of gravity, also known as teleparallel theory of gravity (TG). 1 We assume that our readers are familiar with [31]. The TG approach is reviewed in Blagojević et al. [8], see also Maluf [36], and Aldrovandi and Pereira [1]. For a somewhat related tetrad approach to gravity, in which affine symmetry is exploited, one should compare Sławianowski et al. [54].
Because of the vanishing curvature, we can pick suitable frames such that Γ α β vanishes globally: In this 'teleparallel gauge,' the covariant exterior derivative taken with respect to the connection Γ α β is reduced to the ordinary exterior derivative. This will simplify our formalism. However, we will drop the star * over the equality sign in future since (1) is assumed to be valid throughout our paper.
The essence of the premetric approach can be formulated as follows. This universal field-theoretic scheme is based on conservation laws which hold true for the two types of variables: extensive fields ("how much?") and intensive ("how strong?") ones. These variables satisfy the fundamental equaions which are metric-free, whereas the metric comes in only via the linking equations which establish consitutive relations between the extensive and intensive variables.
Premetric electrodynamics [22] is based on the conservation laws of electric charge and magnetic flux which give rise to the fundamental equations d H = J and d F = 0.
Here H is the electromagnetic excitation 2-form (extensive variable) and F the electromagnetic field strength 2-form (intensive variable). By introducing the constitutive relation H = κ[F], one obtains a predictive physical theory.
The premetric gravity framework [31] can be constructed along the same lines by replacing the electric charge with a "gravitational charge" → mass → energy-momentum. In this introductory section, we provide a short overview of the premetric teleparallel approach.
As was already pointed out in [17, p. 52], "...for consistency, we cannot allow spinning matter (other than as test particles) in such a T 4 ..." that is, in a 4d Weitzenböck spacetime. In other words, the field equation (2) is valid only for hydrodynamic and for electromagnetic matter. A careful proof of this stipulation was given by Obukhov and Pereira 2 [47]. For matter with spin, the Dirac field, for example, the Lorentz group should also be gauged, which removes the teleparallelism constraint R α β = 0. Then one arrives at a Poincaré gauge theory operating in a Riemann-Cartan spacetime with torsion and with Cartan curvature R α β = 0.
The homogeneous field equation of TG reads with the untwisted vector-valued 2-form F α := dϑ α , the torsion of spacetime, which has the expansion Note that (4)  The analogy to the Maxwell equations of electrodynamics should be apparent, see [22]. Equation (2) represents four inhomogeneous Maxwell type equations and Eq. (4) four homogeneous Maxwell type equations. Since TG is a translational gauge theory, it has the four one-form potentials ϑ α = e i α dx i , where four is the number of generators of the translation group.

Local and linear constitutive law
In order to complete the two field equations of TG to a predictive system of equations, one has to adopt a constitutive law between excitation H and field strength F. The simplest assumption is that the functional H = κ[F] is local and linear, To deduce the corresponding component representation, we remember that the functional κ αβ acts on 2-forms and creates as response other 2-forms. Since any 2-form can be decomposed with respect to the 2-form basis ϑ νρ , it is sufficient to study the behavior of ϑ νρ under the application of κ αβ . Because of the assumed locality and linearity, we have Let us come back to (6). We decompose H α and F β and, by using (7), we find: By renaming some indices, we eventually derive the final formula see [31,Eq. (47)].
Accordingly, κ βγ α νρ μ has (6 × 4) 2 = 576 independent components. In the corresponding electrodynamics case, the constitutive law reads H βγ = 1 2 κ βγ νρ F νρ . Thus, by contrast, we have only 6 2 = 36 independent components. As we will see further down, if we study only reversible processes, then this number is appreciably downsized in both cases. A concise Hamiltonian formulation of teleparallel gravity was given by Ferraro and Guzmán [14]. Recently Hohmann et al. [26] studied teleparallel gravity, but instead of taking local and linear constitutive equations, they turned to local and nonlinear ones in order to incorporate f (T )theories into the general TG formalism. These investigations [26] are very helpful since they bring order into the plethora of f (T )-theories and make them more transparent. Such nonlinear models provide an interesting development of gravitational theory based on an analogy with Born-Infeld-Plebański electrodynamics. We believe, though, that there is, at the present time, no real need to push nonlinear constitutive laws, since gravity is nonlinear anyway, due to its self-interaction -and this in spite of a linear constitutive law, which guarantees the quasi-linearity of the emerging field equation.
Kostelecký and Mewes [33] investigated Lorentz and diffeomorphism violations in linearized gravity. In this context, they introduced tensors which are of a similar type as our constitutive tensor χ . Our group-theoretical treatment of χ in Sect. 3 is reminiscent of their method. However, in our article the full nonlinearity of gravity is treated in a premetric framework.
We would like to stress the following fact about teleparallelism theories: We have always a frame e α = e i α ∂ i and a coframe ϑ β = e j β dx j with us, that is, we can always change from holonomic to anholonomic indices by using e i α and e j β , respectively -and vice versa. This implies that all the indices in premetric TG are fundamentally equal, in particular, all those occurring in κ βγ α νρ μ in (9). The 'group' indices α and μ are of the same quality as the 'form' indices β, γ , ν, ρ. Contractions with all indices are always allowed. Of course, at the premetric stage, raising and lowering of indices is only possible with the totally antisymmetric Levi-Civita symbols αβγ δ = ±1, 0 and μνρσ = ∓1, 0, since no metric is available so far in our premetric framework.
The constitutive tensor density 3 χ βγ α νρ μ is equivalent to κ βγ α νρ μ , in particular, But there is still a third useful version of the constitutive tensor available. For the purpose of the irreducible decomposition under the GL(4, R), it is optimal to have the indices of the constitutive tensor all exclusively either in lower or in upper position. Since so far we have no metric at our disposal, we can only move the antisymmetric sets of indices. If we have a look at (9), we can rewrite it witȟ that is, We have the symmetrieš The merit of formulating a field theory only in terms of its field equations -here Eqs. (2) and (4) -and an associated constitutive law -here Eq. (6) -is that it covers both processes, irreversible and reversible ones. At first, like in the conventional treatment of general relativity [13], we turn our attention to reversible processes. From their definition it is clear that, for instance, periodic processes and those the equations of motion of which are formulated in a time symmetric way, are reversible. Time symmetry means that we can substitute in the field equations t by −t without changing them.
Thus dissipation is not allowed in reversible processes and we can define for each such process an energy function and, 4 Similarly, we have: Suppose that when a system under consideration changes from a state, α, to another state, α , the environment changes from β to β . If in some way it is possible to return the system from α to α and at the same time to return the environment from β to β, the process (α, β) → (α , β ) is said to be reversible. Ryogo Kubo [34, p. 61]. by a Legendre transformation, a Lagrangian. Accordingly, reversible processes can always be formulated by means of an action principle. For TG, the twisted Lagrange 4-form reads We substitute (5) and (8) into (17) and find: With the definition (12) and the volume 4-form vol, Eq. (18) can be rewritten as (19) or, by renaming the indices of the F's, equivalently as Consequently, only those components of the constitutive tensor enter the Lagrangian that satisfy the relations If we assume that the model is completely specified by the Lagrangian (17), we restrict our considerations to reversible processes. Then the relations (21) are necessary conditions, which correspond to the symmetry of a 24 × 24 matrix. Thus, the set of the 576 independent components of χ βγ α νρ μ reduces to only 300 ones. It is possible to rewrite the constitutive law in a more compact way. We introduce the 6-dimensional co-basis ϑ αβ → ϑ I in the space of 2-forms, with the collective indices I, J, · · · = (01, 02, 03; 23, 31, 12) = (1, 2, . . . , 6), see [22, p. 40]. Then excitation and field strength decompose as H α = H I α ϑ I and F β = F J β ϑ J , respectively, and the constitutive law reads, with the 300 independent components Thus, for the Lagrangian we find This closes our short introduction to TG. Like all classical theories, whose form is established, we can put TG into a Tonti type diagram [56] in order to clearly display its structure. This will be done for the first time in Sect. 2. Then we turn to a closer examination of the constitutive tensor of TG. In Sect. 3, we decompose it into smaller pieces, in particular into the irreducible pieces with respect to the linear group GL(4, R). In Sect. 4, metric dependent constitutive tensors will be addressed, in particular those which relate TG to general relativity. In Sect. 5, we will study the propagation of gravitational waves in TG within the geometric optics approximation. We will follow the procedure that we developed for electrodynamics [2,22,29]. Since in TG we have four generators of the gauge group, things become a bit more complicated than in electrodynamics. In Sect. 6, we will specialize these considerations on gravitational waves to metric dependent models.

Tonti diagram of the premetric teleparallel theory of gravity
Over the past decades, Tonti [56] developed a general classification program for classical and relativistic theories in physics, such as, e.g., for particle dynamics, electromagnetism, the mechanics of deformable media, fluid mechanics, thermodynamics, and gravitation. Here we will display in Fig. 1 for the first time an appropriate and consistent diagram of the teleparallel theory of gravity (TG). If a theory is well-understood, its configuration and its source variables can be clearly identified and their interrelationships displayed in the form of a Tonti diagram. Such a diagram defines what one may call the skeleton of a theory. In Tonti's book [56], for all classical theories, including the relativistic ones, a corresponding framework was established -and this step by step, based on an operational definition of the quantities involved.
Tonti [56, p. 402] has also displayed a diagram for relativistic gravitation. In Tonti's own words, it was an "attempt" of a diagram based on an ansatz for a tetrad theory of gravity by Kreisel and Treder, see [57, pp. Fig. 1 is based on our recent paper [31] and on the present one.
Let us have a look at our new diagram. The configuration variables of TG are the coordinates x i of the 4-dimensional spacetime (four 0-forms) and the coframes ϑ α (four 1-forms). By differentiation, we find the torsion F β (four 2-forms) and by further differentiation the homogeneous field equation of gravity d F β = 0 (four 3-forms), the right hand side of which vanishes.
The round boxes on the left column depict geometrical objects and the square boxes interrelate these geometrical objects. The formula d F β = 0, for example, corresponds to the first Bianchi identity of a teleparallel spacetime. In a Riemann-Cartan space, we have D F β = ϑ α ∧ R α β . In teleparallelism, the curvature vanishes, R α β = 0, and, in the For global variables, Tonti distinguishes spatial domains, such as volumes V, surfaces S, lines L, and points P, with respect to time he introduces instants I and intervals T. Furthermore, for the respective domains, he has inner (interior) and outer (exterior) orientation. Hence [ I × S], for example, refers to a time domain I with exterior orientation and a space domain S with interior orientation. And this domain [ I × S] supports the torsion two-form. The holonomic coordinates x j , to take another example, depend on an instant of time I and a spatial point P. i.e. [I × P]. Then we have to add the orientation. All exterior forms on the left columns are forms without twist, see [22], those on the right columns all carry twist. This can be read off from the corresponding orientations. This comes about as follows.
For configuration variables the associated space elements are endowed always with an interior orientation, whereas in the case of source variables it is the exterior orientation which plays a role. This behavior is found by phenomenologically examining the different theories. According to Tonti, the underlying theoretical reason for this correspondence is not clear, but phenomenology does not allow any other attributions. The configuration variables are related to the theory of chains of algebraic topology, whereas the source variables are associated to co-chains. For more details, we refer to the exhaustive monograph of Tonti [56].
Since the gravitational field itself carries energy-momentum, it is also the source of a new gravitational field, which likewise carries energy-momentum, etc. Thus, like general relativity, TG is an intrinsically nonlinear theory, even it carries a linear constitutive law. Within the field equation of TG, shows up explicitly and is a manifestation of this nonlinearity. By differentiation of the field equation, we find Thus, d (m) Σ α = −d (ϑ) Σ α is nonvanishing in general, which clearly shows up in our Tonti diagram. The Tonti diagram displayed in Fig. 1 was constructed for the premetric version of teleparallelism, no metric is involved at all. We developed the corresponding formalism in our previous paper in [31,Sec.II]. Above, in Sect. 1.1, we discussed already that teleparallel gravity is only equivalent to GR as long as the matter which is involved does not carry intrinsic spin. If this is the case, the energy-momentum of matter is described by the 3-form (m) σ α , with ϑ [α ∧ (m) σ β] = 0, i.e., the corresponding energy-momentum tensor is symmetric. Since ϑ α := g αγ ϑ γ , we need a metric for the specification of such an energy-momentum (m) σ α . And the metric g αβ induces an associated Levi-Civita connection 1-form Γ α β . The energy-momentum law, in the teleparallel gauge, reads d (m) Σ α = (e α F β ) ∧ (m) Σ β . If we dispense with the teleparallel gauge for the moment, the derivative d (m) Σ α can be substituted by D (m) Σ α , with D as the covariant exterior derivative operator with respect to the teleparallel connection Γ α β . Having now a metric available, we can assume that the teleparallel connection is metric compatible, that is, Dg αβ = 0. For the symmetric energy-momentum, we have then the energy-momentum law Due to a lemma of Meyer [38], the right side of (27) can be absorbed by the left side. This can be demonstrated by expanding the covariant derivative: Let us now introduce a tilde to denote the Riemannian objects and operators: Γ α β is the Christoffel connection, for example, and D the Riemannian covariant derivative. Since Γ α β = Γ α β − K α β , with the contortion 1-form K α β , we can rewrite (28) as The contortion, is related to the torsion via F β = K β γ ∧ ϑ γ . If we substitute this into (29), it can be recast into Because of the antisymmetry of the contortion, K (αβ) = 0, Eq. (30) simplifies to Now we recall that (m) σ α is symmetric, Consequently, The gravitational constant is denoted by . Also known as translation gauge theory of gravity This is the energy-momentum law of GR. Accordingly, our entry d (m) Σ α = (e α F β ) ∧ (m) Σ β in the Tonti diagram becomes, provided a symmetric energy-momentum tensor is prescribed, D (m) σ α = 0, well in accordance with the vanishing divergences in analogous Tonti diagrams.

Decompositions of the constitutive tensor
In this section we characterize the constitutive tensor due to its symmetry transformations relative to the general linear group GL(4, R) and the permutation (symmetry) group. We refer to Hamermesh [15] and to Barut and Raczka [5] as background information for group theory.

Two forms of the constitutive tensor
According to (11), the constitutive tensor density χ αβ μ γ δ ν is a order- 4 2 tensor density that is skew-symmetric in the two pairs of upper indices, see (13): In n dimensions, we have n n 2 n n 2 = 1 4 n 4 (n − 1) 2 independent components. For n = 4, it gives 576 relevant components. Due to this large number, it is instructive to consider decomposition of the constitutive tensor into smaller pieces. We apply Young's decomposition technique that yields irreducible decomposition under the group GL(4, R). For our conventions and the details of the Young decomposition, see "Appendix". For the tensor χ αβ μ γ δ ν , separate S 4 -permutations of four upper indices and S 2permutations of two lower indices are available, where S n denotes the n-dimensional permutation group. Correspondingly, we are dealing with the Cartesian product group Although the metric tensor is not available in our construction, we can, as we have shown in (15), lower two pairs of skew-symmetric indices by the Levi-Civita symbol. Thus, we have a order-0 6 tensoř with the identitieš This tensor is naturally decomposed under the permutation group S 6 . Both types of decompositions are invariant under the action of the basis transformation group GL(4, R).

S 4 × S 2 decomposition of χ αβ μ γ δ ν
Treating the covariant and contravariant indices as belonging to two separate tensor spaces, the irreducible decomposition of χ αβ μ γ δ ν is defined as a product of the irreducible decompositions. The corresponding Young diagrams are expanded as Here, the three first diagrams relate to the upper indices. This decomposition repeats the known irreducible decomposition of the electromagnetic constitutive tensor. The remaining two diagrams represent the symmetric and antisymmetric parts of the second order tensor. Collecting all possible combinations, we obtain the decomposition of the tensor χ αβ μ γ δ ν into 6 independent pieces: In tensor notation, the above decomposition reads Here we used the set of six projection operators [I ] P. We chose the labeling of I = 1, . . . , 6 such that it corresponds to the same sequence of Young diagrams as depicted in the second equality of Eq. (38). Explicitly, the S 4 ×S 2 -irreducible pieces of the constitutive tensor are expressed as In terms of independent components in n = 4 dimensions, this decomposition corresponds to It is straightforward to show that the above operators [I ] P = [I ] P αβκγ δ ωλμρσ ν are orthogonal projectors: In (39)-(44), we first decompose the gravitational constitutive tensor into three independent pieces relative to its four upper indices. This S 4 decomposition is done completely similar to electrodynamics. Then we extract relative to the pair of the lower indices the symmetric and antisymmetric parts. Since the principal and axion part of the electromagnetic constitutive tensor are reversible, the corresponding parts of the gravitational constitutive tensor are reversible when symmetrized in the lower indices. The electrodynamics the skewon part is irreversible. Thus its gravitational analog is reversible when antisymmetrized in the lower indices. The remaining three parts are irreversible (change their sign under permutation of triads of indices).
For completeness, we present here the results about the contributions of the irreducible pieces into the dispersion relation for the gravitational waves. These facts will be derived in Sect. 6.

S 6 decomposition ofχ αβμγ δν
Now we consider the decomposition of the constitutive tensor under the permutation of its six lower indices. We follow Hamermesh's [15] Eqs. (7-159) to . Then, for n = 4 dimensions, the symmetries (35) imply the Young decomposition ofχ αβμγ δν : Here we omitted vanishing diagrams in n = 4 (i.e. diagrams that contain antisymmetrization with respect to more than four indices). In terms of independent components, we have In tensorial language, this corresponds tǒ where again the expressions (I ) P ωλκρσ αβμγ δν denote orthogonal projectors, The explicit expressions of the seven terms (I )χ αβμγ δν are quite involved and we do not display them here; they can be found in "Appendix".

Lagrangian
Let us first consider how these independent pieces contribute to the gravitational Lagrangian (ϑ) Λ = − 1 2 F α ∧ H α . In components, we have Substitution of the constitutive law (11) yields Consequently only those parts of the constitutive tensor contribute to the Lagrangian that satisfy the "pair commutation" symmetries Accordingly, only the following terms are left over in the Lagrangian:

Energy-momentum current
The energy-momentum current of the coframe field is defined as [31] (ϑ) The constitutive law for the pure axion piece (101) is expressed as In this case, the energy-momentum current takes the form Consequently, the symmetric combination of the axionic part [5] χ βγ α νρ μ does not contribute to the energy-momentum current.
The tensors k αβ = −k βα and l αβ = −l βα are antisymmetric, that is, k (αβ) = 0, l (αβ) = 0 , whereas m αβ and n αβ are asymmetric. However, in the reversible case, m αβ and n αβ turn out to be symmetric, whereas the two other antisymmetric tensors turn out to be transposed to each other, k αβ = l βα . This can be recognized more clearly by collecting all the traces into the specific constitutive tensor This tensor is determined by four tensors of 2nd order: two of them are antisymmetric, L αβ = − L βα and K αβ = − K βα , that is, L (αβ) = 0, K (αβ) = 0, and the two other ones are general 2nd order tensors M αβ and N αβ . It is straightforward to establish one-to-one correspondences between these objects and the traces (105)-(108): The inverse relations read Decomposing the specific constitutive tensor (109) into reversible and irreversible (skewon) parts yields Here we introduced This confirms that the irreversible, the skewon part vanishes, In the general case, the specific constitutive trace tensor (109) is characterized by 5 nontrivial irreducible parts: [4] χ αβ γ μν ρ = [5] χ αβ γ μν ρ = 0, (126) [6] Here we denoted

Metric-dependent constitutive tensor
In general relativity as well as in teleparallelism, the existence of a metric tensor g is conventionally assumed. We choose the metric signature (+, −, −, −). The question is then how the constitutive tensor density χ αβ μ γ δ ν can be expressed in terms of the metric, namely in terms of its covariant and/or contravariant components g αβ and g γ δ , respectively. Since χ is a 6th order tensor of type 4 2 , it seems reasonable to start with an ansatz of a purely contravariant dimensionless 6th order tensor K αβμγ δν . Then the contravariant metric components g γ δ are the only metric components that have to be taken into account. In general, one can come up with a polynomial ansatz of arbitrary order in the metric tensor. However, since the resulting constitutive tensor is of 6th order, all indices of the metric factors except six should be necessarily contracted. Recalling that a contraction of the covariant and contravariant metric yields a Kronecker delta, we then find that a polynomial of an arbitrary order is automatically reduced to a general polynomial expression in the metric which is just cubic in g αβ .
Since χ αβ μ γ δ ν is antisymmetric in the index pairs αβ and γ δ, we have to antisymmetrize K correspondingly by subsequently putting brackets around the indices αβ and γ δ, respectively. Also recall that the metric is symmetric, hence its index ordering is optional: Note that the α 4 and the α 7 terms are already antisymmetric in γ δ. Thus, there is no need to put brackets [ ] around γ δ or δγ , respectively. We can now collect all terms by recalling the antisymmetry in both index pairs αβ and γ δ. We find, without any further intermediary step, the compact equation We know that χ is a density, like √ −g, with g := det g μν . Accordingly, we can identify the parity even part of the constitutive tensor density as follows, with as the gravitational constant. In our formalism up to now, including the Tonti diagram, was put to 1 for simplicity. Here we do display it for clarity. Thus, we find, with the three constants, the expression This proves our ansatz with three independent constants in [31,Eq. (80)]. 5 It is, indeed, the most general expression which, up to a factor of √ −g, turns out to be a cubic polynomial in the metric. 5 In [31], there was a typo in Eq. (80): the last plus sign + in this formula should have been a minus sign −. Moreover, for simplification, we changed now our conventions with respect to the β's slightly: 4β old 1 = β 1 , 8β old 2 = −β 2 , 8β old 3 = −β 3 .

Parity violating terms
Already for quite some time, see Pellegrini and Plebanski [48] and Müller-Hoissen and Nitsch [40], also parity violating (odd) constitutive tensors for torsion square Lagrangians have been considered in the literature; for recent reviews, see [3,7,32,45]. There it has been shown that two parity odd terms occur, which are both linear in the totally antisymmetric Levi-Civita symbol αβγ δ , which is, as we may recall, a tensor density: One can also take into account such additional parity violating terms in the framework of the Poincaré gauge theory of gravity. In that more general theory, it leads to interesting new cosmological models [4,25] and to new gravitational wave solutions [6,43]. We would like to understand the generality of the result in (140) in a similar way as we did it for the parity even case. For parity reasons, the Levi-Civita symbol has to be linearly or in odd powers in the corresponding ansatz. Since again, as in the even part of χ αβ μ γ δ ν , we start with an tensor expression of type 6 0 , we need additionally a tensor of type 2 0 . Clearly the contravariant components of the metric g αβ qualify for such a purpose. Thus, the simplest possible ansatz, see also (140), is to start with a 6th order tensor density as follows: M αβμγ δν = μ 1 g αμ ε βγ δν + μ 2 g αγ ε βμδν + μ 3 g αδ ε βμγ ν + μ 4 g αν ε βμγ δ + μ 5 g βμ ε αγ δν + μ 6 g βγ ε αμδν + μ 7 g βδ ε αμγ ν + μ 8 g βν ε αμγ δ + μ 9 g μγ ε αβδν + μ 10 g μδ ε αβγ ν + μ 11 g μν ε αβγ δ + μ 12 g γ ν ε αβμδ + μ 13 g δν ε αβμγ . (141) The μ 1 , μ 2 , . . . , μ 13 are arbitrary constants. We now lower again the indices μ and ν and antisymmetrize at the same time with respect to the index pairs αβ and γ δ, respectively: Having established the required antisymmetries, we can now identify the odd piece of the constitutive tensor as Here we have to recall an algebraic trick that we had already applied in discussing the electromagnetic 4th order constitutive tensor, see [22,Eq.
Looking now at (142), we have typically δ [α μ ε β]γ δν . With T α μ = δ α μ , we find By such methods, we can derive several identities for the Levi-Civita symbol in 4 dimensions: Let us come back to (143) with (142). We decompose the δε-terms in symmetric and antisymmetric pieces with respect to the lower indices: As a result, the δε-terms in the second and third lines of (142) can be rearranged as follows: Next, we write Consequently, the expression (151) transforms into where we used the identity (147). We recall that (154) belongs to the second and third line of (142). In the last step, it remains to use (146) in the last line of (154) and to substitute (148) into the last line of (142). Then, with the constants (μ 1 −μ 5 +μ 4 −μ 8 +μ 9 −μ 10 −μ 12 +μ 13 +4μ 11 ) , the result eventually reads Here α is dimensionless pseudoscalar. It is necessary in order to take into account the transformation law of this part under improper reflections. Therefore, the most general parity-odd part of the constitutive tensor in the end boils down to just the three independent terms with β 4 , β 5 , and β 6 . The β 4 -term, being totally antisymmetric in αβγ δ, represents an β 4 -axion, whereas the β 5 -term describes a β 5 -axion. They both represent reversible processes. In contrast, the β 6 -term corresponds to irreversibility. When contracted by F αβ μ and F γ δ ν , this term drops out from the general teleparallel Lagrangian. Moreover, one can reduce the number of independent terms in the Lagrangian by making use of the Nieh-Yan topological invariant [41,42,46].

GR || : the teleparallel equivalent of GR
We found in the framework of the teleparallel equivalent GR || of general relativity (GR) for the β's in [31, Eq. (88)] 6 , Accordingly, the three β's are related to GR in a quite definite way, and we get a feeling for their interpretation. The GR || Lagrangian is distinguished from the other teleparallelism Lagrangians as being locally Lorentz invariant, see, e.g., Cho [11] and Müller-Hoissen and Nitsch [40]. Any other additional term of even parity in the gravitational Lagrangian removes this local invariance.

Von der Heyde Lagrangian
Earlier, different parity even pieces of torsion square Lagrangians were compared by Muench et al. [39], see also the literature cited therein. A particular role played in these considerations the torsion square piece of a Lagrangian of von der Heyde [24]. Its constitutive tensor [17] reads For the corresponding constitutive law, we havě This law, in the teleparallel case, leads already to the correct Newtonian approximation. The constitutive tensor (165) carries the following β-values: As we saw, the teleparallel equivalent GR || has a slightly different constitutive law (164).
According to Mashhoon,8 "...the mere fact of postulating a nonlocal constitutive relation violates local Lorentz invariance." Still, the version of NLG of 2009 is based on the reversible and locally Lorentz invariant parameter set (164) and Mashhoon's version corresponds to an irreversible and possibly also locally Lorentz invariant parameter set, see the 2nd line of (168).

Propagation of gravitational waves
In this section, we discuss the wave propagation in our premetric teleparallel theory (TG) in linear approximation.

Geometric optics approximation
We start with the source-free field equations Notice that the source term Σ in (2) contains two independent expressions: the matter energy-momentum current (m) Σ and the gravitational energy-momentum current (ϑ) Σ. In matterfree regions, we have (m) Σ = 0. Similar to standard GR, we will assume, in addition, (ϑ) Σ = 0. This requirement is applicable for small waves and means linearization of the field equations.
Here the components q ν of the wave covector are determined by the differential of the phase ϕ function of the wave: dϕ = q ν ϑ ν . The second equation of (170) has the solution with an arbitrary tensor A β α . It is a gravitational analog of the electromagnetic potential. We observe the gauge invariance of this gravitational potential. An expression of the form A β α = q β C α does not contribute to the field strength F ρσ α . In other words, the model is invariant under the transformations with an arbitrary vector C α .
We substitute (171) into (170) and use the constitutive relatioň This yields the characteristic equation with fourth order characteristic tensor Inserting here the decomposition (38) of the constitutive tensor, we observe that the axion-type parts [5] χ μρ α νσ β and [6] χ μρ α νσ β do not enter the tensor M μ α ν β . Consequently, these parts do not contribute to the wave propagation as it was already outlined in Table 1.
In the reversible case, the characteristic tensor M μ α ν β satisfies the symmetry relations A generalized Fresnel equation can be derived as the condition for the solvability of the Eq. (174) along the lines of the algebraic computations of Itin [30].

Dispersion relation: general facts
Observe that the characteristic equation represents 16 equations for the 16 variables A ν β .
For a compact representation of this system, we denote a pair of upper and lower indices by a multi-index In this notation, the system (177) reads The gauge transformation (172) can be rewritten as The 4 linearly independent vectors C α imply that there are likewise 4 linearly independent vectors Q J . The identities In electromagnetism with only one gauge invariance constraint, there appears the 1st order adjoint matrix Adj(M) αβ . This matrix is expressed as a scalar function multiplied by a tensor product of q α , namely Adj(M) αβ = λ(q)q α q β . Thus, the electromagnetic dispersion relation takes the form λ(q) = 0. Similarly, in the case of the gravitational equation (184), we have (4) Adj(M) I 1 ···I 4 J 1 ···J 4 = Λ(Q)Q I 1 · · · Q I 4 Q J 1 · · · Q J 4 = 0.
(185) Thus, the gravitational dispersion relation takes the form Here Λ(q) is a homogeneous polynomial of the order 16 = 24−8 in the variable q α . We recall that in the electromagnetic case, the corresponding form is of 4th order. This fact yields birefringence in the wave propagation. In 3-dimensional elasticity theory, the dispersion relation is of 6th order, with 3 different waves in general. In our generalized gravitational model, there are 8 different waves in general.

Dispersion relation decomposed
In order to clarify the nature of the Eq. (177), we decompose the solution under the GL(4, R) irreducibly into the scalar Aδ β α and the traceless part A α β , Consequently, Eq. (177) decomposes into A similar decomposition can be performed with E μ α . Thus, we find the 1 + 15 equations or, explicitly, one equation and 15 equations Substituting (190) into (191), we derive the algebraic system for the traceless variable: where we introduced The latter tensor is apparently traceless for both pairs of indices: Accordingly, we find a system of 15 algebraic equations for 15 variables A ν μ . After one solves the homogeneous equation (192), one immediately finds the scalar A from (190).

Scalar waves as a special case
We consider now a special case of pure scalar waves. Let the field A ν β be identically zero: A ν β = 0. Hence, we are left with a system of two scalar equations The second equation has a non-trivial solution only if M α α β β = 0. Consequently we have a dispersion relation Using the double-trace tensor (105), we can write it as In the metric-dependent case (162), the tensor m αβ , up to a factor, coincides with the Lorentz metric tensor g αβ . Thus we recover the standard light cone of general relativity,

A separable case
We assume now that two terms in the left-hand side of Eqs. (190) and (191) vanish independently. In other words, we assume that the system separates into two independent subsystems. In [28], a similar type of consideration allowed to extract the teleparallel equivalent of GR from a set of metric based models. Thus we have 15 equations for 15 variables and 1 equation for 1 variable

Gauge conditions
In electromagnetism, gauge invariance can be restricted by applying a gauge fixing condition. The Lorenz gauge condition is the unique diffeomorphism invariant expression. For wave solutions, it takes the form A α q α = 0. Note that this condition can be formulated only on a metric manifold.
In teleparallel theory, we are able to formulate a similar condition in a premetric form: A more general gauge condition can be proposed in the form of Here K 1 and K 2 are arbitrary constants. The exceptional case for K 1 = 1 and K 2 = −1, namely is invariant under the gauge transformation A α β → A α β + q α C β and, accordingly, does not represent a gauge condition.

Gravitational waves in a metric model
Let us now specialize to the metric-dependent constitutive tensor (156). The analysis of the wave propagation in the geometric optics approximation is straightforward. Here we will confine our attention to the parity-even case (139). Recently a related paper was published by Hohmann et al. [27].

Characteristic system
We substitute the constitutive tensor (139) with its parity even pieces into (175). As a result, M β α ν μ (g) = χ βγ α νρ μ (g) × q γ q ρ takes the form We can immediately derive the contractions where q 2 := g αβ q α q β . Consequently, Furthermore, we verify that the tensor (204) has the following properties: Interestingly, for the GR || case of (164), the right-hand sides of (208) and (209) vanish, i.e., the tensor M β α ν μ turns out to be transversal to the wave covector q μ in all four indices.
Taking into account these properties, the wave propagation system (177) yields In the generic case, when 2β 1 − β 2 − β 3 = 0, we find the relation which implies We will use this in subsequent derivations.
Let us now turn to the decomposed equations (191) and (192), Using (205), (206), and A ν ν = 0, we can recast (191) into This is derived for the generic case by assuming 2β 1 − 3β 2 − β 3 = 0, otherwise (191) is trivial. Finally, we turn to the traceless part (192) of the wave propagation equation. Substituting (204)-(206) into (193), and then making use of (212), we can rewrite (192) in the equivalent form Quite remarkably, the coupling constant β 2 does not contribute. Decomposing (214) into symmetric and antisymmetric parts, we obtain the two equations Two important observations are in order. Firstly, we see that the symmetric A (μν) and skew-symmetric A [μν] variables decouple from each other and are governed by two separate dynamical equations. Secondly, the symmetric A (μν) mode remains the only one when coupling constants satisfy 2β 1 = − β 3 and, similarly, the antisymmetric A [μν] mode remains the only one when the coupling constants satisfy 2β 1 = β 3 . In the generic case (when 2β 1 = − β 3 and 2β 1 = β 3 ), both modes are dynamical. In principle, one can proceed by deriving the Fresnel equations for each mode in a way similar to classical electrodynamics. However, one can choose an alternative way and analyze the propagation equations by using the method of spin-projection operators. The corresponding tools were earlier developed in metric gravity [52] and in its teleparallel formulation [40]. Their application is straightforward. Ultimately we conclude that the wave equations describe the massless spin-0, spin-2 and spin-1 modes propagating along the light cones q 2 = 0.
6.2 Waves in GR || , the teleparallel equivalent of GR In order to understand more clearly what happens in GR || , it is instructive to analyze the wave propagation equation from scratch. Plugging (204) into (175), we find for (174), after some straightforward algebra and lowering the index μ, Contracting with g αβ , we find whereas contraction with q α yields In the generic case, we recover the earlier findings (211) and (210). Making use of this, we see that the first line in (217) vanishes, whereas the remaining equation reduces to the system (215) and (216). However, GR || represents a very special case when the coupling constants (164) are such that both 2β 1 −β 2 −β 3 = 0 and 2β 1 + β 3 = 0. As a result, the first two lines of (217) disappear, and the propagation equation has only a symmetric traceless part. The bottom line is that in GR || Eq. (217) reduces to after we introduce the familiar variable Accordingly, we indeed recover the result of GR with the massless spin-2 graviton propagating along the light cone. In other words, GR || is completely consistent with Einstein's theory with respect to the propagation of gravitational waves.

Conclusions and outlook
Recently, in Ref. [29], we developed a novel framework for a premetric teleparallel theory of gravity (TG). Here we continue this study on TG by paying attention specifically to TG models with a general local and linear constitutive law. In Sect. 2, we constructed the Tonti diagram of TG which explicitly displays the generic structure of the theory.
Our main new results are presented in Sect. 3 where the irreducible decomposition of the premetric constitutive tensor is established in full detail. This issue was not analysed in the earlier literature. Here we considered two types of such decompositions with respect to the permutation group, namely one related to S 4 × S 2 and the other one with respect to S 6 . The relations between these two decompositions are explicitly derived and the physical meaning of the different irreducible pieces clarified.
After establishing for the constitutive tensor these premetric results, we turned in Sect. 4 to a special case: The spacetime continuum is supposed to carry a metric tensor. We constructed the most general metric dependent constitutive tensor that is cubic in the metric tensor. It includes both, parity even and parity odd parts. Thereby extending earlier results, we for the first time obtained the most general family of teleparallel consitutive tensors. In particular, a parity odd Lagrangian was newly found, and its physical interpretation fixed.
In Sect. 5, the propagation of gravitational waves are derived for the premetric case in the geometrical optics approximation. Additional propagating modes of spin 1 and 0 are extracted. Our results are generally consistent with the recent findings reported in [27].
Subsequently, in Sect. 6, the metric case is addressed. The GR limit is naturally embedded in our formalism.
The results presented here can serve as a basis for the study of expanded gravitational models including axion and skewon effects. In particular, violation of Lorentz invariance can be meaningfully addressed in our premetric set-ups.
where S 136 denotes symmetrization between the 1st, 3rd, 6th index; similarly, A 34 denotes antisymmetrization between Fig. 2 The resolution of the identity on the 576-dimensional vector space Tχ has a block-diagonal structure in terms of the canonical Young projection operators {I } P : Tχ → T Ǐ χ with I = 1, . . . , 7. We depict these operators by 1×1 blocks, with the relative width of f λ I ×d I /576. Provided f λ I > 1, these operators can be further decomposed into noncanonical operators (I, j) P : T Ǐ χ → T Ǐ χ with j = 1, . . . , f λ I . We depict this internal ambiguity by inserting f λ I × f λ I smaller blocks the 3rd and 4th index, and so on. In the last lines we switched to the more convenient cycle notation. Note that in our conventions the Young symmetrizers are not normalized.
The exact expressions are lengthy, so we make some comments: (i) Using the Young decomposition technique, the 576dimensional tensor space ofχ αβγ μνρ , call it Tχ , can be decomposed into the direct sum of lower-dimensional vector spaces T Iχ with I = 1, . . . , 7. Denoting the projection operators onto these lower-dimensional subspaces via (I ) P : Tχ → T Ǐ χ , the identity element on Tχ can be written as We emphasize that the operators (I, j) P are not canonical, that is, they are only determined up to arbitrary linear transformations G : ρ(T Ǐ χ ) → ρ(T Ǐ χ ). For a graphical representation of statements (i)-(iii), see Fig. 2.
(247) It has been verified by computer algebra that the above expressions indeed belong to orthogonal subspaces.