Classification of non-Riemannian doubled-yet-gauged spacetime

Assuming $\mathbf{O}(D,D)$ covariant fields as the `fundamental' variables, Double Field Theory can accommodate novel geometries where a Riemannian metric cannot be defined, even locally. Here we present a complete classification of such non-Riemannian spacetimes in terms of two non-negative integers, $(n,\bar{n})$, $0\leq n+\bar{n}\leq D$. Upon these backgrounds, strings become chiral and anti-chiral over $n$ and $\bar{n}$ directions respectively, while particles and strings are frozen over the $n+\bar{n}$ directions. In particular, we identify $(0,0)$ as Riemannian manifolds, $(1,0)$ as non-relativistic spacetime, $(1,1)$ as Gomis-Ooguri non-relativistic string, $(D{-1},0)$ as ultra-relativistic Carroll geometry, and $(D,0)$ as Siegel's chiral string. Combined with a covariant Kaluza-Klein ansatz which we further spell, $(0,1)$ leads to Newton-Cartan gravity. Alternative to the conventional string compactifications on small manifolds, non-Riemannian spacetime such as $D=10$, $(3,3)$ may open a new scheme of the dimensional reduction from ten to four.

1 Introduction Ever since Einstein formulated his theory of gravity, i.e. General Relativity (GR), by employing differential geometryà la Riemann, the Riemannian metric, g µν , has been privileged to be the only geometric and thus gravitational field. All other fields are meant to be 'extra matters'. On the other hand, string theory suggests us to put a two-form gauge potential, B µν , and a scalar dilaton, φ, on an equal footing along with the metric. Forming the massless sector of closed strings, this triplet of objects is ubiquitous in all string theories. Further, a genuine stringy symmetry, T-duality, can mix the three of them [1,2], thus hinting at the existence of Stringy Gravity which should take the entire closed string massless sector as geometric and gravitational. After series of pioneering works on 'doubled sigma models' [3][4][5][6][7][8] and 'double field theory' (DFT) [9][10][11][12][13] (cf. [14][15][16] for reviews), such an idea of Stringy Gravity has materialized. 1 1 Strictly speaking, string theory does not predict General Relativity but its own gravity, i.e. Stringy Gravity. 1 The word 'double' above refers to the fact that doubled (D +D)-dimensional coordinates are used for the description of D-dimensional physical spacetime. While such a usage was historically first made in the case of a torus background -by introducing a dual coordinate conjugate to the string winding momentum -the doubled coordinates are far more general and can be applied to any compact or non-compact spacetime, and to not only string but also particle theories.
Stringy Gravity of our interest adopts the doubled-yet-gauged coordinate system [17] which meets two properties. Firstly, an O(D, D) group is a priori postulated, having the invariant constant "metric", (1.1) Along with its inverse, J AB , the invariant metric can be used to freely raise and lower the O(D, D) vector indices (capital letters, A, B, . . .). Secondly, the doubled coordinates are gauged by an equivalence relation, where ∆ A is an arbitrary 'derivative-index-valued' vector. This means that its superscript index must be identifiable as that of a derivative, ∂ A = J AB ∂ B . For example, with arbitrary functions, Φ 1 , Φ 2 belonging to the theory, ∆ A = Φ 1 ∂ A Φ 2 . The equivalence relation can be realized by requiring that all the fields or functions in Stringy Gravity -such as Φ 1 , Φ 2 , physical fields, local symmetry parameters, and their arbitrary derivatives -should be invariant under the coordinate gauge symmetry shift, In this way, a single physical point is not represented by a point, as in ordinary Riemannian geometry, but as a gauge orbit in the doubled coordinate system.
The above coordinate gauge symmetry invariance is equivalent to the so-called 'section condition' in DFT, With respect to the off block-diagonal form of the O(D, D) metric (1.1), the doubled coordinates split into two parts: x A = (x µ , x ν ) and ∂ A = (∂ µ , ∂ ν ), such that ∂ A ∂ A = 2∂ µ∂ µ . The general solution to the section condition is then given by∂ µ ≡ 0, up to O(D, D) rotations [10,11].
Diffeomorphism covariance in doubled-yet-gauged spacetime reads and for a covariant tensor (or tensor density with weight ω), The latter corresponds to the passive counterpart of the "generalized Lie derivative",L ξ ,à la Siegel [10].
The whole massless sector of closed strings, or stringy gravitons, can be represented by a unit-weighted scalar density, e −2d , and a symmetric projector, The complementary, orthogonal projector,P AB = J AB − P AB , satisfies, from (1.7), PP = 0,P 2 =P . The difference of the two projectors sets a symmetric O(D, D) element, known as the DFT-metric (or "generalized metric"), These O(D, D) covariant defining properties of the stringy gravitational fields can be conveniently solved by the conventional variables,  [20][21][22], and have been shown to provide a natural geometry for the 2 However, a fully covariant four-indexed "Riemann-like" curvature has been argued not to exist [18,19]. This absence is, in a way, consistent with the fact that there exists no locally inertial frame for an extended object, i.e. a string, where the stringy Christoffel connection (2.17) might vanish completely: Equivalence Principle holds for particles not strings.

Scope of the paper
It is the purpose of the present paper to classify completely DFT backgrounds, by deriving the most general solution to the defining property of the stringy gravitational field, or (1.8). Our classification is given in terms of two non-negative integers, (n,n), 0 ≤ n +n ≤ D. Except for the (0, 0) case, these are generically non-Riemannian.
Since various DFTs and the relevant doubled sigma models have been constructed, strictly in terms of the O(D, D) covariant fields without referring to the Riemannian ones {g µν , B µν , φ}, 4 our result can be readily and unambiguously applicable to these models which include, e.g. coupling to the Standard Model [33], higher spin [34], fluctuation or Noether analyses [21,[35][36][37], the doubled-yet-gauged Green-Schwarz superstring action [22] and the maximally supersymmetric D = 10 DFT [38]. In particular, this last example, with the Killing spinor equations therein, may lead to a new scheme of the dimensional reduction from ten to four, by assuming the six-dimensional internal space to be non-Riemannian, alternative to the traditional string compactification on 'small' Riemannian manifolds [39]. Further applications can be found in the holographic correspondences between Newton-Cartan gravity and condensed matter physics, e.g. [40,41].

Organization of the paper
The rest of the paper is structured as follows. In section 2, we classify the DFT-metric and spell the corresponding DFT-vielbeins. We discuss the dynamics of point particle and string upon these backgrounds.
We also spell a path integral definition of the proper length in doubled-yet-gauged spacetime as well as a covariant Kaluza-Klein ansatz for DFT. In section 3, we discuss various applications, such as Gomis-Ooguri non-relativistic string, non-relativistic and ultra-relativistic geometries, Siegel's chiral string and Newton-Cartan gravity. Appendix contains the technical derivation of the main result. 3 cf. [24,25] for U-duality analogs. 4 Yet, there are quite a few works in the literature which do not meet this criterion, relying explicitly on the Riemannian variables.
Our results are thus not applicable therein.
2 General results

Classification of the DFT-metric
As recalled in the introduction, the DFT-metric is by definition a symmetric O(D, D) element, satisfying the following relation: Our main result consists in providing a full classification for DFT-metrics by solving the above defining properties: the most general solution is characterized by two non-negative integers, (n,n), 0 ≤ n +n ≤ D, and assumes the following form: where i, j = 1, 2, · · · , n andī, = 1, 2, · · · ,n. The variables, H µν , K µν , B µν , X i µ , Y ν j ,Xī µ ,Ȳ ν  , must meet the following properties: -H µν and K µν are symmetric tensors 3) whose kernels are spanned by X i µ ,Xī ν and Y µ j ,Ȳ ν  , respectively -a completeness relation, -the skew-symmetry of B-field, While the derivation is carried out in Appendix, some comments are in order. From (2.4), (2.5) and the linear independency of X i µ ,Xī ν , orthonormal as well as algebraic relations follow With the choice of the section,∂ µ ≡ 0, the doubled-yet-gauged diffeomorphisms (1.5), (1.6), or the generalized Lie derivative of the DFT-metric, cf. (A.4), imply that the variables transform covariantly as where L ξ denotes the ordinary, i.e. undoubled, Lie derivative with the local parameter, ξ ν , being part of the doubled vector field, ξ A = (ξ µ , ξ ν ). Our (n,n)-classification of the DFT-metric having the explicit parametrization (2.2) is particularly useful for the choice of the section,∂ µ ≡ 0. For example, the action for a massless scalar field reads (c.f. [43]) For couplings to generic tensors or Yang-Mills fields, we refer to [18,21,36,44,45]. However, if the (n,n) DFT-metric (2.2) admits an isometry direction, there appears arbitrariness in choosing the section. In this case, our parametrization (2.2) may be modified, see e.g. [32,46].
Clearly, constant (n,n) DFT-metric (2.2) and DFT-dilaton, d, solve the equations of motion of DFT. Thus, our (n,n) classification also accounts for non-Riemannian 'flat' backgrounds. It is worthwhile to note that the characteristic value, (n,n), may change point-wise in a given doubled-yet-gauged curved spacetime, typically at a "Riemannian singular point". Further, O(D, D) rotations (along isometry directions) can also change the value of (n,n), for example, the (0, 0) fundamental string backgroundà la Dabholkar et al. [47] can be mapped to (1, 1) by certain O(D, D) rotations [20] (cf. [24]). However, the trace of a DFT-metric, It is instructive to note that the B-field contributes to the DFT-metric by an O(D, D) conjugation, such that the contribution is 'Abelian', in the following sense: (2.12) Further, the precise expression of the (n,n) DFT-metric (2.2) as well as the fundamental algebraic relations, Firstly under obvious GL(n) × GL(n) rotations, secondly under the transformation of only the B-field having two arbitrary skew-symmetric local parameters, m ij = −m ji ,mī = −mī, (2.14) and lastly under the following somewhat less trivial transformations of {Y µ i ,Ȳ μ ı , K µν , B µν }, where V µi andV µī are arbitrary local parameters. In fact, the latter two transformations, (2.14) and (2.15), can be unified into The dynamics of the DFT-metric and the DFT-dilaton is dictated by the Euler-Lagrange equations of DFT. The expression of the (n,n) DFT-metric (2.2) may then be inserted into the known stringy extension of the Christoffel symbol to lead to covariant derivatives and curvatures [18]. Yet, the trace (2.10) of the (n,n) DFT-metric can be nontrivial, and this calls for some revision of the previous result:

Particle and string on (n,n) doubled-yet-gauged spacetime
While the notion of doubled-yet-gauged spacetime might sound somewhat mysterious, it is possible to define proper length and hence to show that it is a 'metric space'. To do so, we first note that the usual infinitesimal Thus, the naive contraction with the DFT-metric, dx A dx B H AB , cannot give any sensible definition of proper length in doubled-yet-gauged spacetime. To cure the problem, we need to gauge dx A explicitly, introducing a connection, A A , which should satisfy the same property as the coordinate gauge symmetry Provided the connection transforms appropriately, Dx A becomes a well-behaved i.e. covariant vector [20], We propose then to define the proper distance in doubled-yet-gauged spacetime by path integral [48], By letting∂ µ ≡ 0 and therefore A A ≡ (A µ , 0), we may solve the constraints and write That is to say, only the half of the doubled coordinates, i.e.x µ directions, are gauged. Furthermore, with the Riemannian DFT-metric (1.9), we get [20] Thus, after integrating out the auxiliary connection, our proposal (2.23) reduces -at least classically -to indeed the formula (2.23) measures the distance between two 'gauge orbits'.
The exponent in (2.23) immediately sets the action for a point particle in doubled-yet-gauged spacetime, or its square-root free einbein formulation [42], It also easily extends to (Nambu-Goto type) area and volume, which in turn provides the doubled-yet-gauged string action [20] (c.f. [8] and also [22] for the extension to the Green-Schwarz superstring), Besides the constraint imposed by the auxiliary potential, A A , the equation of motion of the former particle action can be spelled in terms of the stringy Christoffel connection (2.17), On the other hand, for a string propagating on the (0, 0) Riemannian background, the auxiliary potential, A A , implies the self-duality (i.e. chirality) over the entire doubled spacetime [20], 29) and the Euler-Lagrangian equation of x A gets simplified to give the stringy geodesic equation, which extends (2.28), yet with a different numerical factor in front of the connection, 2 versus 1.
For a generic non-Riemannian background, the analysis is more subtle which we investigate hereafter.
We substitute the generic (n,n) DFT-metric (2.2) into the covariant actions, and move from doubled formalism to undoubled one. One useful identity which generalizes (2.25) from Riemannian (0, 0) to a generic which reads more explicitly for particles, Note that, in accordance with the completeness relation (2.5), the auxiliary vector potential decomposes as Integrating out H µν A τ ν gives the on-shell relation, which implies that the 'dual' conjugate momenta are trivial along D − n −n number ofx µ directions.
On the other hand, integrating out the remaining components, Y ρ i A τ ρ andȲ ρ ı A τ ρ , we acquire constraints on the x µ coordinates, Namely, the particle freezes over n +n directions on the physical section formed by x µ coordinates.
For string, combining the useful identity (2.31) with the topological term in the action (2.27), we can reduce the world-sheet Lagrangian, where for shorthand notation we set Now, integrating out H µν A αν we obtain the on-shell relation, and integrating out Y ν i A αν ,Ȳ ν ı A αν , we obtain chiral constraints, Namely, string becomes chiral over n directions and anti-chiral overn directions on the section coordinatized by x µ . The chirality further implies that strings which meet boundary conditions (periodic, Neumann or Dirichlet) are also frozen, or localized, over the n +n directions, similarly to the particle case (2.35).

DFT-vielbeins for (n,n) doubled-yet-gauged spacetime
In order to couple to fermions [33,49] or R-R sector [50], as well as for supersymmetrizations [38,51,52], it is necessary to introduce a pair of DFT-vielbeins, V Ap andV Ap , from which one can construct the projectors, where η pq andηpq are the two constant metrics of twofold local Lorentz symmetries for two distinct locally inertial frames, one for the left and the other for the right closed string modes [53]. 5 To ensure the symmetric, orthogonal and completeness properties of the projectors (1.7), the DFT-vielbeins must satisfy their own defining properties: with the eigenvalues, (η, +η) and (η, −η).
The main result of this subsection is the construction of the DFT-vielbeins, V Ap andV Ap , for the general (n,n) DFT-metric (2.2). They are given by 2D × (D + n −n) and 2D × (D − n +n) matrices respectively, (D − n +n) × D matrices, such that 1 ≤ p ≤ D + n −n and 1 ≤p ≤ D − n +n. Explicitly, with the smaller range of indices, 1 ≤ a,ā ≤ D−n −n and 1 ≤ i ≤ n, 1 ≤ī ≤n as before, the matrices read where {h a µ , k ν b } and {hā µ ,k νb } are two sets of the "square-roots" of H µν and K µν , The 'total' twofold local Lorentz symmetry group is clearly Spin(t + n, s + n) × Spin(s +n, t +n), with t + s + n +n = D, where (t, s) is the signature of H µν and K µν . The corresponding constant metrics are η pq andηpq respectively, while η ab andηāb are (t + s) × (t + s) sub-blocks of them, of which the signatures are numerically opposite to each other [18], (2.45) There are defining properties of {h a µ , k ν b } and {hā µ ,k νb }, in accordance with (2.4), (2.5): (2.46) It follows that and where the superscript T converts column vectors to row ones. As expected, P AB andP AB are respectively free of the barred and unbarred variables, {Xī,Ȳ} and {X i , Y j }.
In a parallel manner to (2.11), the B-field contributes to the DFT-vielbeins through O(D, D) multiplications, (2.49) For consistency, the trace of the DFT-metric reads The symmetry of the DFT-metric (2.16) extends to DFT-vielbeins: (2.51) As seen from the doubled-yet-gauged actions for particle and string (2.32), (2.36), as well as the coupling to a scalar field (2.9), it is not the full signatures of the spin group,

Kaluza-Klein ansatz for DFT
The ordinary Kaluza-Klein ansatz for a Riemannian metric can be 'block-diagonalized', (2.53) for which we decomposeD = D + D, such that and set an off-block-diagonal so(D, D) element, Explicitly, we haveŴ 3 = 0 and

Applications
The case of (0, 0) admits a well-defined Riemannian metric and hence corresponds to Riemannian geometry, or to "Generalized Geometry" [63][64][65][66][67][68] when equipped with the pair of DFT-vielbeins. In this section, we discuss various applications of other (n,n) backgrounds and identify the corresponding geometries.

Maximally non-Riemannian (D, 0) : Siegel's chiral string
In the maximally non-Riemannian case of (D, 0), with i = 1, 2, · · · , D, we can view X i µ as a nondegenerate D × D square matrix. Then from (2.7) and we conclude that X j λ Y µ j is actually an identity, Thus, in the case of (D, 0), we have The corresponding DFT-vielbein, V Ap (2.42) and the Spin(D, D) metric are also 2D ×2D square matrices, On the other hand,V Ap is trivial.
The resulting string action is completely chiral on the D-dimensional section (2.36) [20], From the conventional (0, 0) set-up, noting the sign difference,
An interesting choice then appears to be Spin(4, 6) × Spin (6, 4). Such a choice can encompass sixdimensional (3, 3) non-Riemannian 'internal' spacetime, while maintaining the ordinary four-dimensional Minkowskian 'external' spacetime. As analyzed in subsection 2.2, point particles and strings freeze on the (3, 3) internal spacetime and this may imply a natural dimensional reduction of string theory from ten to four, alternative to the conventional compactification on 'small' Riemannian manifolds, e.g. CY 3 . The latter will be of interest to analyze the Killing spinor equations in [38] for the D = 10 (3, 3) DFT-vielbeins (2.42).

(1, 1) : Non-relativistic limità la Gomis-Ooguri
In this subsection, we identify (1, 1) as the non-relativistic limità la Gomis-Ooguri [23]. We start by considering a generic Riemannian metric which depends explicitly on the speed of light, c, where T µ and S ν are orthogonal time-like and space-like vectors, Essentially, (3.9) is the 'covariantized' form of the ordinary Kaluza-Klein ansatz for the Riemannian metric (2.53) as The inverse of the metric is then given by where the variables, N ν and Υ µν , meet by construction, 7 Now, we introduce an ansatz for the B-field in a similar manner, 14) and require that the Riemannian DFT-metric (1.9) should be non-singular in the non-relativistic, large c limit. In (3.14), without loss of generality we may set B µ to be orthogonal to N ν , i.e. N µ B µ = 0. Further, B 0 µν denotes the zeroth order in c which is arbitrary and should survive once the limit is taken, as expected from the 'Abelian' nature of the B-field from (2.11) and (2.12).
Clearly in the limit of c → ∞, the inverse, g µν , is regular. We only need to ensure both g −1 B and g − Bg −1 B to be non-singular. The former implies In turn, Bg −1 B cannot be quadratically singular, and hence for the regularity of g − Bg −1 B, the leading power of g must be first order in c, i.e. the apparent second order term in (3.9) must be trivial, Therefore, the nontrivial cancellation of diverging terms inside g − Bg −1 B takes place at the first order, which reads Contraction of the quantity inside the square bracket with N ν gives It follows that g − Bg −1 B is non-singular as

(3.20)
After all, the DFT-metric becomes completely regular, which can easily and precisely be identified as the (1, 1) type of the classification (2.2) as As demonstrated in [20], constant flat background belonging to this type generates the Gomis-Ooguri nonrelativistic string [23] (see also [22] for its Green-Schwarz superstring extension). Thus, a generic (1, 1) DFT-metric provides a curved spacetime generalization of the non-relativistic string.

(D − 1, 0) : Ultra-relativistic or Carroll
The Riemannian metric (3.9) in the previous section defines the proper length. Rescaling the metric by an overall factor of c −2 , it becomes the Riemannian metric for the proper time: where the variables should satisfy (3.10) and (3.13), which we recall here Clearly, the expression of g µν in (3.23) indicates the possibility of taking a small c i.e. ultra-relativistic limit, as the inverse remains non-singular, yet degenerate having rank one, In this subsection, we propose a (D − 1, 0) DFT-metric as the ultra-relativistic 'completion' of the above degenerate inverse (3.25), where all the variables are from (3.24). It is easy to check that this ansatz satisfies the defining properties of the DFT-metric (2.1) and H A A = 2Υ µν Φ µν = 2(D − 1). Note the identification, From (2.35), particles freeze over almost all the directions except one, This is in agreement with the ultra-relativistic limit of Riemannian geodesicsà la Bergshoeff et al. [72].
Namely, particles cannot move faster than light, and thus must freeze in the ultra-relativistic limit, c → 0.
In fact, (D −1, 0) forms a Carroll structure [73,74]: Φ µν is known as a Carrollian metric, i.e. a rank (D −1) covariant metric whose kernel is spanned by the Carroll vector, N ν , and T µ is a principal connection. The Carrollian boost symmetry [74] is given, with an arbitrary local parameter, V µ , by The ordinary Kaluza-Klein ansatz (2.53) treats the two block-diagonal Riemannian metrics, g and g , asymmetrically. Exchanging the two will lead to an alternative Kaluza-Klein ansatz. In this subsection, we consider such an alternative ansatz for the Riemannian metric (3.9), which reads with the inverse, (3.31) Clearly the inverse of the Riemannian metric allows non-singular large c limit, of which the rank is D − 1.
The DFT-metric which completes this degenerate inverse is then with Here the upper and lower signs correspond to (1, 0) and (0, 1) respectively.
The transformation (2.15) reduces to with U µ = Υ µν V ν ∈ Ker(T ). This transformation is sometimes referred to as a Milne transformation or a Galilean boost in the literature [77].
From (2.35), particles freeze over the time direction only, so that the observerẋ µ is said to be space-like. This is naturally dual to the ultra-relativistic Carroll dynamics (3.28) where time flows but all spatial directions freeze.
In order to account for the dynamics of time-like observers (for which time flows), one needs to introduce external forces, as done in the following subsection within the ambient framework of a null Kaluza-Klein reduction.
Then the 'external' DFT-metric, H AB , must be of the (n,n) = (0, 1) type 8 , i.e. the lower sign in (3.33), which ensuresĤÂÂ = 2(n −n) = 2(n + n −n −n ) = 0. We let (ỹ, y) denote the primed coordinates, (x 1 , x 1 ), and write for the ambient doubled coordinates, The ambient DFT-metric (2.60) then takes the following form, where, using the notations of section 2.3, we set a (D + 1)-dimensional Spin(s + 1, t + 1) vector, 10 8 The alternative choice of (n ,n ) = (0, 1) obtained by setting H A B ≡ −J A B will involve replacingP by −P in (2.60), and accordingly the external DFT-metric, H, will need to be of (1, 0) type. 9 In terms of ordinary Lie derivative, L ξĝμν = 0, L ξBμν = −2∂ [μξν] , and (3.39) means ξμξνĝμν = 0. such that, from (2.42), It is also convenient to define from (2.43), (2.45), Note the identification of the conventions, Now, with the lower sign choice of (3.33), plugging (3.40) into the master doubled-yet-gauged action for a point particle (2.26), we obtain in a similar fashion to (2.32), where we set for shorthand notation as well as for a convenient field redefinition to replace A τ µ , Note that the very last term in (3.45) is a perfect square which vanishes afterhā µ Λ µ being integrated out as Since y is the coordinate for the isometry direction, it serves as a Lagrange multiplier: it forces the conjugate momentum of y, or p, to be constant, Integrating out Λ gives a constraint, such that the time is generically not frozen, c.f. (3.36). Further, integrating out the auxiliary field, A τỹ inside D τỹ determines the velocity, with (3.47), (3.48), (3.49), (3.50) The einbein imposes the Hamiltonian constraint, From (3.50) and (3.51), it follows that: That is to say, whenever p = 0,ẏ is completely fixed by the dynamics of the x µ coordinates. The auxiliary variable, Λ, is also fixed in the same manner.
Making use of the world-line diffeomorphisms, we hereafter normalize the einbein: such that τ coincides with the proper length.
The equation of motion for x µ reads now where we defined for simplicity, the field strengths Computing the contractions, N µ E µ ,ẋ µ E µ , respectively, we obtain the time derivative of the auxiliary variable, and a consistency relation among the constraints (3.49), (3.51), All together, the combination, Υ λµ E µ + N λĖ Λ , fully determines the acceleration, where γ λ µν denotes the following coefficients, We emphasize that the dynamics of the D-dimensional coordinates x µ as prescribed by (3.59) is independent of the Kaluza-Klein direction, y. Geometrically, this means that one can interpret x µ as coordinates on the quotient manifold of the ambient spacetime by the light-like direction along the vector field, ξμ∂μ = ∂ y .
In the special case where T µν vanishes (i.e. the one-form, T µ , is closed) and W + is a (non-vanishing) constant, the expression (3.59) simplifies tö of which the right-hand side can be interpreted as the Lorentz plus Coulomb forces. In this particular case, the coefficients (3.60) are the ones associated to the so-called 'special Galilean connection' for the field of observers, N µ , (cf. e.g. [76]). In accordance with the usual Riemannian ambient approach of [79][80][81] (cf. also [74,82,84,85]), the resulting dynamical trajectories (3.61) can be interpreted as Newton-Cartan geodesics. These are of two different types, depending on the value of p : • p = 0 (Space-like observer).
In this case, the constraint 11 , T µẋ µ = 0, holds as a consequence of (3.49) so that we recover the case 11 From the ambient perspective, the constraint, Tµẋ µ = 0, implies that the dynamics becomes restricted to a D-dimensional hypersurface of the ambient manifold, transverse to the null isometry vector field, ξμ∂μ = ∂y. Such a light-like hypersurface is naturally endowed with a Carrollian metric structure [73], and the equations of motion (3.61) together with (3.50) and (3.56) can be naturally interpreted as geodesics associated to a suitable Carrollian connection induced by the ambient metric structure (cf. [73,83]  • p = 0 (Time-like observer).
In this case, τ is parametrized to ensure e = 1 pW + T µẋ µ ≡ 1 such that the observerẋ µ is time-like.
Equation (3.61) can thus be reformulated as where the coefficientsγ λ µν are defined aŝ The connection associated to the coefficients (3.63) is naturally interpreted as a Newtonian connection [78], i.e. a torsion-free connection compatible with the Leibnizian structure (Υ µν , T µ ) such that the associated field strength, F µν , is closed.
In summary, assuming the triviality of T µν and W + , the doubled-yet-gauged particle action (2.26) with the ambient (D+1)-dimensional Kaluza-Klein ansatz (2.60) reproduces the full content of Newtonian dynamics (unifying the space-like and time-like cases) on the D-dimensional manifold quotient along the light-like direction, y.
In principle, the assumption regarding the triviality of the variables, T µν and W + , should be examined by considering the on-shell dynamics of the DFT-metric, i.e. the Euler-Lagrangian equations of DFT. In the present work, we have focused on the kinematical side of the DFT-metric and the subsequent particle and string dynamics on the background. We leave the investigation of the dynamical aspect of the (n,n) DFTmetric for future work. From our perspective, the DFT action and its full equations of motion determine universally and unambiguously all the dynamics of the (n,n) backgrounds, including (0, 0) Riemannian General Relativity and (0, 1) Newton-Cartan gravity, in a unifying manner.
The defining condition (A.1) reads explicitly, The generalized Lie derivative of the DFT-metric, c.f. (1.6), leads to Without loss of generality then, introducing a skew-symmetric B-field, 12 we may put It follows that, with some vector field, Y µ , H µ ν takes the form, We proceed with a new symmetric variable, K µν = K νµ , The last relation in (A. 3) gives Contracting this with X µ shows Lastly we impose the skew-symmetric condition of H µρ H ρ ν , which gives with (A.11), K µρ Y ρ X ν + K νρ Y ρ X µ = 0 , (A.13) and hence in particular, contracting with Y µ , we have (A.14) Substituting this back into (A.13), we conclude that Y µ K µρ Y ρ must be trivial, and hence in fact from (A.14), We may perform a field redefinition, Y µ → ±Y µ , in order to remove the sign factor in the normalization of (A.12). After all, the most general form of the DFT-metric in the 'least' degenerate case takes the form, (A.16) 12 The ambiguity in introducing the B-field through (A.8) amounts to the symmetry of the final result (2.16).
of which the variables must meet The above analysis can be straightforwardly extended to the most general degenerate cases, where there are On the other hand, contraction of (A.23) with (Y j ·X k )Y µ k leads to where we set Y k ·M ·Y i ≡ Y ρ k M ρσ Y σ i for shorthand notation. Substituting into (A.23), we get which, after contraction with Y µ l Y ν m and from (A.25), can be seen to be equivalent to Now, we may recast (A.26) into It is crucial to note, from the symmetric property, Y k ·M ·Y j = Y j ·M ·Y k , that the free indices, µ and ν, below are symmetric, and further from the skew-symmetric property (A.28), that the free indices, ν and ρ, below are skewsymmetric, Therefore, if we perform a field redefinition, We then let which nicely satisfies Finally, we perform a similarity transformation, (X i µ , Y ν j ) → (S i k X k µ , Y ν k S −1k j ), which leaves Y µ i X i ν invariant but diagonalizes Y ρ i X j ρ with the eigenvalues of either +1 or −1. We then let N = n +n in order to denote the numbers of the +1 and −1 eigenvalues of Y ρ i X j ρ . If the corresponding eigenvalue is −1, we further perform a field redefinition, (Xī µ ,Ȳ ν ı ) := (X i µ , −Y ν i ), which involves the change of the index from i toī. In this way, we arrive at the most general form of the DFT-metric, (2.2), classified by two non-negative integers, n,n.
It is also worthwhile to decompose the B-field utilizing the completeness relation (2.5), for which we set