Berwald-Finsler Spacetimes

• Berwald-Finsler spacetimes are Finsler geometries defined by 2-homogeneous Lagrangians with quadratic geodesic sprays that reduce to affine Berwald connections.
• They preserve pseudo-Riemannian features by enforcing connection coefficients that depend solely on position while allowing intrinsic direction dependence in the metric.
• Rigidity results, synthetic curvature characterizations, and symmetry classifications illustrate their key role in extending gravitational theories beyond standard metrics.

Searching arXiv for recent and foundational papers on Berwald-Finsler spacetimes and closely related Berwald/Finsler geometry. Berwald–Finsler spacetimes are Finsler spacetime structures whose canonical Finsler connection reduces to an affine connection on the base manifold, so that the geodesic spray is quadratic in the velocities and the corresponding connection coefficients depend only on position. In the pseudo-Finsler formulation used for spacetime geometry, the basic object is a 2-homogeneous Lagrangian $L(x,\dot x)$ on a conic subbundle of $TM\setminus\{0\}$, with non-degenerate Hessian $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$. The Berwald condition singles out those Finsler spacetimes that are closest to pseudo-Riemannian geometry while remaining genuinely direction-dependent in their metric sector (Voicu et al., 2023, Pfeifer et al., 2019). In positive-definite Finsler geometry Berwald spaces are tightly linked to affine and Riemannian structures, but in Lorentzian or pseudo-Finsler signature the corresponding spacetime theory exhibits both strong rigidity phenomena and genuine non-metrizable, metric-affine behavior (Fuster et al., 2020, Boonnam et al., 2018).

1. Definition and basic geometric structure

In the pseudo-Finsler spacetime framework, a pseudo-Finsler space $(M,L)$ is specified by a function

$L:\mathcal A\to\mathbb R$

on a conic subbundle $\mathcal A\subset TM\setminus\{0\}$, with positive 2-homogeneity,

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$

and non-degenerate fiber Hessian

$$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$

A Finsler spacetime is obtained when there exists a conic subbundle $\mathcal T\subset\mathcal A$ on which $g_{ab}$ has Lorentzian signature $TM\setminus\{0\}$0, $TM\setminus\{0\}$1, and the boundary of $TM\setminus\{0\}$2 is the light cone (Voicu et al., 2023). A closely related formulation starts from a continuous function $TM\setminus\{0\}$3, smooth on $TM\setminus\{0\}$4, positively homogeneous of degree $TM\setminus\{0\}$5, with a unit timelike shell $TM\setminus\{0\}$6 at each point ensuring a well-defined causal cone structure; the associated Finsler function is

$TM\setminus\{0\}$7

(Pfeifer et al., 2012).

A pseudo-Finsler space is of Berwald type if the geodesic spray coefficients are quadratic in the velocities,

$TM\setminus\{0\}$8

so that the nonlinear connection is linear,

$TM\setminus\{0\}$9

The coefficients $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$0 then define a torsion-free affine connection on $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$1, called the Berwald connection in this context (Voicu et al., 2023). Equivalent characterizations used across the literature include the direction-independence of the Chern connection coefficients, the linearity of parallel transport, and the vanishing of the hv-curvature of the Chern–Rund connection in the standard Finsler sense (Boonnam et al., 2018, Pfeifer et al., 2019).

This affine reduction is the precise sense in which Berwald–Finsler spacetimes are “closest” to pseudo-Riemannian spacetimes. The Finsler metric $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$2 may still depend on direction, and the Cartan tensor

$$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$3

may be nonzero, so the geometry need not be quadratic in $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$4 (Voicu et al., 2023). If $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$5, then $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$6 is pseudo-Riemannian and the Berwald connection is the Levi-Civita connection of $$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$7 (Voicu et al., 2023).

2. Canonical connection, curvature, and gravity equations

The canonical geodesic equation in Finsler spacetime geometry is

$$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$8

with spray coefficients

$$g_{ab}(x,\dot x)=\frac12\dot\partial_a\dot\partial_b L$$9

and nonlinear connection

$(M,L)$0

Its curvature is

$(M,L)$1

and the Finsler-Ricci scalar is

$(M,L)$2

(Voicu et al., 2023). In a Berwald spacetime this simplifies to

$(M,L)$3

where $(M,L)$4 is the Ricci tensor of the affine Berwald connection (Voicu et al., 2023).

The Landsberg trace also simplifies in the Berwald case. With

$(M,L)$5

Berwald spaces satisfy

$(M,L)$6

(Voicu et al., 2023). This reduction is decisive in the Finsler gravity equations derived from an action on the unit tangent bundle. In the formulation based on

$(M,L)$7

the field equation in vacuum reduces for Berwald spaces to

$(M,L)$8

The paper defines a pseudo-Finsler space to be Ricci-flat if

$(M,L)$9

and for proper Berwald spacetimes this is equivalent to the vacuum field equation (Voicu et al., 2023).

A parallel line of work formulates Finsler gravity directly from the scalar curvature $L:\mathcal A\to\mathbb R$0 of the Cartan nonlinear connection. In that framework the full field equation reduces, for Berwald spacetimes, to

$L:\mathcal A\to\mathbb R$1

with $L:\mathcal A\to\mathbb R$2 the homogeneity degree of $L:\mathcal A\to\mathbb R$3 (Fuster et al., 2018). In the metric case this is equivalent to Einstein vacuum equations; in general it is weaker than $L:\mathcal A\to\mathbb R$4, although for null-VGR Berwald spacetimes the two conditions coincide (Fuster et al., 2018).

This suggests a structural division. One branch emphasizes Ricci-flat Berwald spacetimes as the natural vacuum sector of Finsler gravity (Voicu et al., 2023). Another studies broader Berwald vacuum equations that still admit genuinely Finslerian solutions beyond the Ricci-flat case (Fuster et al., 2018).

3. Characterizations of Berwaldness

A central technical development is a necessary and sufficient first-order PDE for Berwaldness. If one decomposes a Finsler Lagrangian as

$L:\mathcal A\to\mathbb R$5

with respect to an auxiliary metric $L:\mathcal A\to\mathbb R$6, then $L:\mathcal A\to\mathbb R$7 is of Berwald type if and only if there exists a tensor $L:\mathcal A\to\mathbb R$8 such that

$L:\mathcal A\to\mathbb R$9

For such a Berwald Lagrangian, the spray is

$\mathcal A\subset TM\setminus\{0\}$0

(Pfeifer et al., 2019). The condition is intrinsic even though it is written using an auxiliary metric.

For $\mathcal A\subset TM\setminus\{0\}$1-Finsler geometries and $\mathcal A\subset TM\setminus\{0\}$2-Finsler spacetimes, where $\mathcal A\subset TM\setminus\{0\}$3 with

$\mathcal A\subset TM\setminus\{0\}$4

the Berwald condition reduces to a necessary and sufficient condition on the Levi-Civita covariant derivative of the defining 1-form $\mathcal A\subset TM\setminus\{0\}$5: $\mathcal A\subset TM\setminus\{0\}$6 A sufficient condition valid in all signatures is $\mathcal A\subset TM\setminus\{0\}$7 (Pfeifer et al., 2019). This recovers the classical Randers criterion and extends m-Kropina and VGR Berwald conditions.

For VGR spacetimes with

$\mathcal A\subset TM\setminus\{0\}$8

the Berwald condition becomes especially explicit: $\mathcal A\subset TM\setminus\{0\}$9 In the null case $$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$0, this simplifies to

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$1

(Fuster et al., 2018). This criterion yields both null and non-null Berwald examples, including VSI-type and cosmological VGR spacetimes.

A different characterization uses indicatrix invariance and averaging. In positive-definite Finsler geometry, a Finsler structure is Berwald if and only if its Chern connection coincides with the pull-back of its averaged connection, and equivalently if there exists a Riemannian metric whose Levi-Civita connection preserves the Finsler indicatrix under parallel transport (Torromé, 2012). This suggests a Lorentzian analogue for spacetime shells, although the noncompactness of Lorentzian indicatrices makes the averaging construction subtler. A plausible implication is that shell-preserving affine connections provide a natural diagnostic of Berwald behavior in Lorentz–Finsler settings, but the positive-definite proofs rely on compact indicatrices (Torromé, 2012).

4. Rigidity, metrizability, and generalized Berwald structures

Berwald geometry is simultaneously rigid and, in Lorentzian signature, less metrizable than its positive-definite counterpart. In positive-definite Finsler geometry, classical results assert strong rigidity. A complete Berwald manifold with nowhere vanishing flag curvature must be Riemannian, and in particular any Berwald space with flag curvature bounded below by a positive number is Riemannian (Boonnam et al., 2018). Under higher-rank, finite-volume, and nonpositive flag curvature assumptions, a complete connected Berwald space with irreducible universal cover is either locally symmetric or locally Minkowski (Wu, 2015). These theorems show that strong curvature or rank hypotheses sharply reduce the room for genuinely non-Riemannian Berwald structures.

In spacetime signature, the metrizability picture changes. The paper “On the Non-Metrizability of Berwald Finsler Spacetimes” proves that Szabó’s metrizability theorem does not extend to Finsler spacetimes in general (Fuster et al., 2020). The decisive obstruction is the possibility that the Ricci tensor of the affine Berwald connection is not symmetric. For a Berwald spacetime, the Ricci tensor is

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$2

and in large classes of Berwald spacetimes it satisfies

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$3

(Fuster et al., 2020). Since Levi-Civita Ricci tensors are symmetric, such affine structures cannot be Levi-Civita connections of any pseudo-Riemannian metric.

The same paper identifies a large $$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$4-type class,

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$5

for which Berwaldness is equivalent to

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$6

and whose Ricci antisymmetric part is

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$7

Whenever this does not vanish, the Berwald spacetime is non-metrizable (Fuster et al., 2020).

This leads to a structural reinterpretation: a Berwald Finsler spacetime naturally defines a torsion-free metric-affine geometry, with the affine connection decomposed as

$$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$8

where $$L(x,\lambda\dot x)=\lambda^2L(x,\dot x),\quad \lambda>0,$$9 is the non-metricity part relative to an arbitrary pseudo-Riemannian metric $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$0 (Fuster et al., 2020). Thus Berwald spacetimes are not generically pseudo-Riemannian in disguise; rather, they lie naturally in torsion-free metric-affine geometry.

A broader affine viewpoint is furnished by generalized Berwald metrics. In positive-definite Finsler geometry, a metric is generalized Berwald if there exists an affine connection, possibly with torsion, whose parallel transport preserves the Finsler function. Bartelmeß and Matveev prove that this is equivalent to monochromaticity, meaning all tangent normed spaces are linearly isometric (Bartelmeß et al., 2017). This suggests a useful distinction for spacetime applications: classical Berwald spacetimes are torsion-free affine Finsler structures, while generalized Berwald spacetimes admit possibly torsionful affine structures preserving the Finsler norm. The supplied data explicitly notes that this generalized Berwald perspective is informative for spacetime geometry, especially when torsion is regarded as physically meaningful (Bartelmeß et al., 2017).

5. Symmetric spacetime models: spherical and cosmological classes

The most detailed classifications currently available concern highly symmetric Berwald spacetimes. For spatial spherical symmetry, the pseudo-Finsler Lagrangian is constrained by the $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$1 Killing equations to depend on angular velocities only through

$$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$2

so

$$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$3

(Voicu et al., 2023). In the Berwald case, the most general $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$4-invariant affine connection involves a finite set of coefficient functions $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$5, and the corresponding curvature is encoded in functions $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$6 (Voicu et al., 2023).

The classification of $$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$7-invariant four-dimensional pseudo-Finsler Berwald structures yields six non-pseudo-Riemannian classes with power-law, exponential, one-variable, angular, or two-variable dependence on the velocities (Cheraghchi et al., 2022). Building on that classification, the Birkhoff theorem for Berwald Finsler spacetimes proves a sharp rigidity result: any four-dimensional spatially spherically symmetric Berwald pseudo-Finsler space that is Finsler-Ricci flat is either flat or pseudo-Riemannian (Voicu et al., 2023). In Lorentzian signature, the only non-flat vacuum solutions are therefore Schwarzschild. This extends the Jebsen–Birkhoff theorem to Berwald spacetimes and rules out genuinely Finslerian Ricci-flat spherically symmetric vacuum Berwald analogues of Schwarzschild (Voicu et al., 2023).

Cosmological symmetry admits a similarly strong classification. Imposing spatial homogeneity and isotropy yields the general cosmological Finsler form

$$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$8

If one further imposes Berwaldness, the cosmological Berwald condition reduces the possibilities drastically. Either the Lagrangian is purely quadratic and defines an FLRW metric, or, after a redefinition of time, it takes the form

$$g_{ab}(x,\dot x)=\frac{1}{2}\dot\partial_a\dot\partial_b L.$$9

with arbitrary smooth $\mathcal T\subset\mathcal A$0 (Hohmann et al., 2020). This is the most general nontrivial homogeneous and isotropic Berwald-Finsler spacetime according to that classification. The entire Finslerian deviation from FLRW is encoded in the zero-homogeneous function $\mathcal T\subset\mathcal A$1, while the affine structure remains that of a standard connection on spacetime (Hohmann et al., 2020).

A related cosmological example appears in VGR geometry. The most general homogeneous and isotropic VGR Berwald spacetime has

$\mathcal T\subset\mathcal A$2

with Berwaldness forcing

$\mathcal T\subset\mathcal A$3

In this class the geodesic spray becomes independent of $\mathcal T\subset\mathcal A$4, the curvature scalar is

$\mathcal T\subset\mathcal A$5

and the Berwald vacuum equation admits arbitrary scale factor $\mathcal T\subset\mathcal A$6 only in the spatially flat case $\mathcal T\subset\mathcal A$7 (Fuster et al., 2018). This contrasts sharply with standard FLRW vacuum dynamics.

6. Causality, curvature sign, and synthetic characterizations

Recent work establishes synthetic characterizations of curvature in Berwald spacetimes using time separation rather than smooth connection data. The paper “Concavity of spacetimes” proves that for a Berwald spacetime the following are equivalent: nonnegative timelike flag curvature, local concavity of the time separation, local timelike concavity, convex future capsules, and convex past capsules (Beran et al., 30 Sep 2025). Here the time separation is

$\mathcal T\subset\mathcal A$8

where the supremum is over future-directed causal curves (Beran et al., 30 Sep 2025).

This provides a Lorentzian analogue of Busemann convexity. In positive-definite geometry, local Busemann convexity characterizes nonpositive sectional or flag curvature in the Berwald setting. The Lorentzian counterpart replaces convexity of distance by concavity of time separation and nonpositive curvature by nonnegative timelike flag curvature (Beran et al., 30 Sep 2025). The characterization is new even for Lorentzian manifolds, where Berwaldness is automatic.

The same paper introduces future and past capsules,

$\mathcal T\subset\mathcal A$9

and proves that, in a Berwald spacetime, convexity of the unions of such sets along geodesics is equivalent to nonnegative timelike flag curvature (Beran et al., 30 Sep 2025). This gives a purely metric-causal signature of Berwald curvature bounds.

A plausible implication is that Berwald spacetimes form a natural bridge between smooth pseudo-Finsler geometry and synthetic Lorentzian geometry. The data explicitly states that these equivalences allow one to read off smooth curvature conditions from the metric-causal behavior of $g_{ab}$0, at least in the Berwald class (Beran et al., 30 Sep 2025).

7. Equivalence principle, singular models, and physical interpretation

A physically motivated subclass is provided by singular generalized Berwald spacetimes. In that framework, one starts with a Lorentzian metric $g_{ab}$1 and defines

$g_{ab}$2

with $g_{ab}$3 0-homogeneous in $g_{ab}$4, the positivity condition

$g_{ab}$5

and possibly a singular set in velocity space (Torromé, 2016). If $g_{ab}$6, the linear Berwald connection of $g_{ab}$7 coincides with the Levi-Civita connection of $g_{ab}$8, so $g_{ab}$9 is a generalized Berwald spacetime (Torromé, 2016).

These models preserve the null cones: $TM\setminus\{0\}$00 and share the same curvature and Einstein tensor as $TM\setminus\{0\}$01, while modifying proper time through the direction-dependent Lagrangian

$TM\setminus\{0\}$02

(Torromé, 2016). This makes them particularly suitable for discussing the equivalence principle. The paper argues that gravity, as a geometric interaction, should admit smooth free-fall coordinate systems for all freely falling observers, independently of their velocity. Generalized Berwald spacetimes satisfy this because their Berwald connection is affine, whereas generic Finsler spacetimes need not (Torromé, 2016).

In this setting, free-fall local coordinate systems are Fermi coordinates of the affine Berwald connection (Torromé, 2016). Since the affine structure is that of $TM\setminus\{0\}$03, one recovers the local inertial structure of GR, while direction dependence remains in chronometry rather than in geodesic trajectories or light cones. This suggests a conservative Finslerian extension of GR: local causal structure, geodesics, and curvature agree with an underlying Lorentzian metric, but proper time is direction dependent.

The same paper also observes that with vanishing cosmological constant the Einstein tensors of $TM\setminus\{0\}$04 and $TM\setminus\{0\}$05 coincide, whereas with a cosmological term one obtains a factor $TM\setminus\{0\}$06. The argument offered is that consistency then favors a very small cosmological constant when $TM\setminus\{0\}$07 (Torromé, 2016). This suggests a possible link between microscopic Finslerian anisotropy and the smallness of $TM\setminus\{0\}$08, although the data does not present it as a theorem.

Taken together, the supplied literature portrays Berwald–Finsler spacetimes as a sharply delimited sector of Finsler spacetime geometry: affine in their geodesic structure, often rigid under curvature assumptions, rich enough to admit non-metrizable metric-affine behavior, and tractable enough to support exact classifications under symmetry and synthetic curvature characterizations (Voicu et al., 2023, Fuster et al., 2020, Hohmann et al., 2020, Beran et al., 30 Sep 2025).