Berwald Spacetime Overview

• Berwald spacetime is a Lorentz–Finsler geometry defined by a quadratic geodesic spray, making its nonlinear connection affine.
• Its structure simplifies Finsler gravitational field equations, enabling exact solutions in cosmological, Kundt, and VGR models.
• Rigidity results and synthetic geometric properties ensure that non-Riemannian features vanish under strict curvature and convexity conditions.

Berwald spacetime is a Lorentz–Finsler or pseudo-Finsler spacetime whose canonical nonlinear connection defines an affine connection on spacetime, so it is “closest” to pseudo-Riemannian geometry while still allowing a genuinely non-quadratic length measure (Fuster et al., 2018, Voicu et al., 3 Jun 2026). In local terms, it is a Finsler spacetime for which, at each base point, the Chern or Chern–Rund connection coefficients are constant as the tangent direction varies, or equivalently the geodesic spray is quadratic in the fibre variables (Caponio et al., 2024, Fuster et al., 2020). This affine reduction makes Berwald spacetimes a central class in Finsler gravity, in exact solution theory, in cosmological model building, and in rigidity results that sharply constrain non-Riemannian behavior under curvature assumptions (Fuster et al., 2018, Tayebi et al., 27 Jan 2026).

1. Defining structure

A Finsler spacetime is commonly formulated by a positively homogeneous Lagrangian $L$ on a conic domain of the tangent bundle, with vertical Hessian

$$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$

nondegenerate and of Lorentzian signature on timelike directions (Fuster et al., 2018, Fuster et al., 2020). In one formulation, $L$ is a real, smooth function on $TM\setminus\{0\}$, positively homogeneous of degree $r\ge 2$ in the fibre coordinates; in another, $L:A\to\mathbb R$ is defined on a conic open subbundle $A\subset TM\setminus\{0\}$, is positively homogeneous of degree two, and determines the causal structure through a proper conic timelike subset (Fuster et al., 2018, Fuster et al., 2020).

The geodesic spray coefficients are obtained from $L$. One explicit formula is

$$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$

while another equivalent convention writes

$$G^i(x,y)=\tfrac14\,g^{ij}(x,y)\bigl(y^k\partial_k\partial_{y^j}L-\partial_{x^j}L\bigr)$$

(Fuster et al., 2018, Fuster et al., 2020). A Berwald spacetime is then characterized by the quadraticity of the spray: $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$0 equivalently by the vanishing of third fibre derivatives of the spray, or by the fact that the induced nonlinear connection is linear in $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$1 (Fuster et al., 2018, Hohmann et al., 2020).

In the Chern–Rund formulation, $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$2 is called a Berwald spacetime if for each $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$3, the functions $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$4 are constant as $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$5 varies over $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$6; equivalently the Chern connection coefficients depend only on the foot-point $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$7, not on the direction $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$8 (Caponio et al., 2024). In particular, parallel transport is linear and preserves the Finsler structure $$g^L_{ab}(x,y)=\tfrac12\,\bar\partial_a\bar\partial_b\,L$$9 (Caponio et al., 2024). This is the decisive structural simplification: although the Finsler norm remains velocity dependent, the connection geometry reduces to an affine connection on the base manifold.

2. Equivalent criteria and constructive characterizations

A general necessary and sufficient criterion for Berwaldness is the first-order Berwald PDE. Given an auxiliary metric $L$0, define

$L$1

Then $L$2 is Berwald if and only if there exists a $L$3-tensor $L$4 such that

$L$5

and in that case

$L$6

(Pfeifer et al., 2019). This criterion is independent of the choice of auxiliary metric (Pfeifer et al., 2019).

For $L$7-type and $L$8-type Finsler spacetimes, the PDE reduces to explicit covariant-derivative conditions on the defining $L$9-form. If

$TM\setminus\{0\}$0

then Berwaldness is equivalent to a condition on $TM\setminus\{0\}$1; in particular, if $TM\setminus\{0\}$2, then $TM\setminus\{0\}$3, so any $TM\setminus\{0\}$4-Finsler Lagrangian is Berwald exactly when its defining $TM\setminus\{0\}$5-form is covariantly constant (Pfeifer et al., 2019). The Randers case re-obtains the criterion $TM\setminus\{0\}$6, while the m-Kropina or VSR case yields

$TM\setminus\{0\}$7

(Pfeifer et al., 2019).

A particularly important spacetime family is Very General Relativity (VGR), defined by

$TM\setminus\{0\}$8

A VGR spacetime is Berwald if and only if there exists a scalar function $TM\setminus\{0\}$9 such that

$r\ge 2$0

If $r\ge 2$1 is covariantly constant, then for any $r\ge 2$2-Finsler Lagrangian $r\ge 2$3 the spray reduces to the usual metric spray, hence $r\ge 2$4 is automatically Berwald (Fuster et al., 2018). These criteria make Berwald spacetimes unusually tractable within non-quadratic spacetime geometry.

3. Curvature, rigidity, and metrizability

In a Berwald spacetime, curvature tensors reduce to the curvature of the underlying affine connection. The affine curvature is

$r\ge 2$5

and the flag curvature is defined from the curvature endomorphism and the Finsler metric in the usual way (Hohmann et al., 2020, Heefer, 2024). Because the connection is $r\ge 2$6-only dependent, the hh-curvature, affine Ricci tensor, and associated flag curvature are much closer to pseudo-Riemannian curvature than in a general Finsler spacetime (Heefer, 2024).

This simplification leads to strong rigidity theorems. One result extends Numata’s and Szabó’s rigidity theorems to all dimensions: every Berwald manifold whose flag curvature $r\ge 2$7 is nowhere zero is necessarily Riemannian (Tayebi et al., 27 Jan 2026). The same source states that in Lorentzian-signature pseudo-Finsler geometry, if one assumes non-zero timelike or spacelike flag curvature everywhere, the same orthogonality argument forces the mean Cartan torsion to vanish, which implies that $r\ge 2$8 is derived from a pseudo-Riemannian metric; thus no genuinely non-Riemannian Berwald spacetime of everywhere non-zero flag curvature can exist (Tayebi et al., 27 Jan 2026). It follows that non-Riemannian Berwald spacetime models survive only in flat or sign-indefinite curvature regimes.

By contrast, the positive-definite metrizability theorem of Szabó does not extend in general to Finsler spacetimes. A large class of Berwald spacetimes has affine Ricci tensor that is not symmetric, and this asymmetry is possible because Finsler spacetime smoothness typically holds only on a proper conic subset of the slit tangent bundle (Fuster et al., 2020). When the Finsler Lagrangian is smooth on the entire slit tangent bundle, however, any Berwald spacetime has symmetric Ricci tensor (Fuster et al., 2020). The resulting picture is that Berwald spacetimes naturally interpolate between pseudo-Riemannian and torsion-free metric-affine geometry.

A different line of rigidity comes from averaging and indicatrix invariance. For Berwald spaces, the Levi-Civita connection of any Riemannian metric affine equivalent to the Berwald metric leaves invariant the indicatrix of the Finsler metric $r\ge 2$9, and conversely, if there is a Riemannian connection whose Levi-Civita leaves invariant by parallel transport the indicatrix of the Finsler structure, then the structure is Berwald (Torromé, 2012). In a related averaging approach, the invariance of the average metric along a homotopy in the space of Finsler structures over $L:A\to\mathbb R$0 yields the statement that any $L:A\to\mathbb R$1 regular Landsberg space is a Berwald space (Torrome, 2011). In Lorentzian signature, the same chain of arguments is stated to extend without essential change (Torromé, 2012).

4. Berwald spacetimes in Finsler gravity

Berwald spacetimes are especially prominent in Finsler gravity because the field equations simplify drastically when the nonlinear connection descends to an affine one. In one action-based approach, the gravitational action is an integral over the unit tangent bundle $L:A\to\mathbb R$2, built from the canonical Finsler curvature scalar $L:A\to\mathbb R$3; varying $L:A\to\mathbb R$4 gives a field equation which, in the Berwald case where the Landsberg tensor vanishes, reduces to

$L:A\to\mathbb R$5

with $L:A\to\mathbb R$6 independent of $L:A\to\mathbb R$7 and equal to the Ricci tensor of the underlying affine connection (Fuster et al., 2018). In particular, $L:A\to\mathbb R$8 solves the Berwald gravity equation (Fuster et al., 2018).

A second formulation, based on the action

$L:A\to\mathbb R$9

states that the Euler–Lagrange equation simplifies in a Berwald spacetime to the vanishing of the affine Ricci tensor,

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

(Heefer, 2024). This gives a very direct reduction of the vacuum sector to affine Ricci flatness.

A third action-based vacuum equation, also described as physically well motivated, takes the Berwald form

$A\subset TM\setminus\{0\}$1

and is explicitly said to be weaker than Ricci-flatness $A\subset TM\setminus\{0\}$2 (Voicu et al., 3 Jun 2026). This suggests that the precise Berwald vacuum sector depends on the adopted Finsler gravity action. What remains common to these formulations is that Berwald geometry suppresses the explicitly non-affine part of the field equations and makes exact solution theory feasible.

5. Symmetric models and exact solutions

The VGR construction provides explicit Berwald vacua. For Finslerian VSI spacetimes, one takes a Lorentzian Kundt–VSI metric in light-cone coordinates, sets $A\subset TM\setminus\{0\}$3 so that $A\subset TM\setminus\{0\}$4, and forms

$A\subset TM\setminus\{0\}$5

The Berwald condition holds, the spray acquires an explicit affine-plus-null correction, and the Berwald Ricci components satisfy the same vacuum-type equations as the Einstein Ricci tensor of the underlying VSI metric; the $A\subset TM\setminus\{0\}$6 subcase recovers gyratonic $A\subset TM\setminus\{0\}$7-waves and then ordinary $A\subset TM\setminus\{0\}$8-waves when $A\subset TM\setminus\{0\}$9 (Fuster et al., 2018). The same source gives homogeneous and isotropic VGR spacetimes with

$L$0

where the Berwald condition fixes

$L$1

and the vacuum equation admits only the spatially flat case $L$2 (Fuster et al., 2018).

Beyond the VGR ansatz, the most general spatially homogeneous and isotropic Berwald spacetime is described by a Lagrangian of the canonical form

$L$3

or, basis-independently,

$L$4

where $L$5 is an arbitrary smooth function of its argument, subject to nondegeneracy of the Hessian and Lorentzian signature on the timelike cone (Hohmann et al., 2020). These models contain FLRW as the special metric case $L$6, but encode additional velocity dependence through a zero-homogeneous function on the tangent bundle (Hohmann et al., 2020).

Spherical symmetry has recently produced nontrivial vacuum solutions. Among five non-Riemannian Finsler Lagrangian classes compatible with $L$7-invariant affine data, only one class is compatible with asymptotic flatness and a well-defined light-cone structure. In that class the Lagrangian can be written

$L$8

with $L$9 parallel with respect to the Levi-Civita connection of a pseudo-Riemannian metric $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$0, and the Berwald vacuum equation admits three families of non-Ricci-flat, asymptotically flat solutions (Voicu et al., 3 Jun 2026). These are stated to be the first non-trivial, exact spherically symmetric vacuum solutions in this setting (Voicu et al., 3 Jun 2026). Related exact Berwald vacuum solutions include Finsler–pp-waves in Brinkmann coordinates, for which the affine Ricci-flat condition becomes harmonicity of the profile $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$1 in the transverse plane (Heefer, 2024).

6. Global structure and synthetic geometry

Berwald spacetimes also admit global geometric results that are substantially stronger than in the generic Lorentz–Finsler case. For a connected weighted Finsler spacetime $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$2 with a complete timelike straight line and nonnegative weighted Ricci curvature in timelike directions, a splitting theorem gives

$$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$3

provided either the spacetime is timelike geodesically complete or it is Berwald and globally hyperbolic, together with the stated weighted completeness hypotheses (Caponio et al., 2024). In the Berwald case, the translations

$$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$4

lift to genuine Finsler isometries preserving $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$5, and geodesics split into the product of geodesics in $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$6 and $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$7 (Caponio et al., 2024). The Busemann function is correspondingly stronger in the Berwald setting: once the relevant maximum-principle argument is established, $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$8 and $$G^a(x,y)=\tfrac12\,g^{L\,ac}\bigl(y^d\partial_d\bar\partial_cL-\partial_cL\bigr),$$9, and its flow preserves the full Finsler structure rather than only a second-order approximation (Caponio et al., 2024).

A complementary synthetic characterization concerns local concavity of the time separation. For a Berwald spacetime, local concavity is equivalent to nonnegative flag curvature in timelike directions, and also equivalent to convexity of future capsules and of past capsules (Beran et al., 30 Sep 2025). In the formulation of that result, a Berwald spacetime is locally concave if every point has a convex normal neighborhood on which the time separation satisfies a Jensen-type inequality along pairs of geodesics (Beran et al., 30 Sep 2025). The paper explicitly describes this as the Lorentzian counterpart of Busemann convexity in metric geometry (Beran et al., 30 Sep 2025).

Taken together, these results locate Berwald spacetime at a distinctive intersection of affine geometry, non-quadratic causal structure, and Finsler gravitational dynamics. The class is broad enough to include VGR, cosmological, Kundt, $$G^i(x,y)=\tfrac14\,g^{ij}(x,y)\bigl(y^k\partial_k\partial_{y^j}L-\partial_{x^j}L\bigr)$$0-wave, and spherically symmetric vacuum models, yet rigid enough that curvature sign conditions, indicatrix invariance, and global splitting hypotheses often force reduction to pseudo-Riemannian or locally Minkowskian behavior (Fuster et al., 2018, Tayebi et al., 27 Jan 2026, Caponio et al., 2024).