---
title: Directionally Dependent Finsler Metrics
url: https://www.emergentmind.com/topics/directionally-dependent-finsler-metrics
type: topic
---

# Directionally Dependent Finsler Metrics

Directionally dependent Finsler metrics are Finsler structures whose length function depends not only on the base point but also on the tangent direction. In local coordinates this dependence is encoded by the fundamental tensor
\[
g_{ij}(x,y)=\frac{1}{2}\frac{\partial^2 F^2}{\partial y^i\partial y^j},
\]
which varies with the fiber variable $y$ rather than only with $x$; this is the defining anisotropy absent in Riemannian geometry [2107.09227]. On surfaces, one especially rigid realization occurs on $1/n$-translation structures, where tangent planes are canonically identified with $\mathbb{R}^2$ only up to translations and rotations by $2\pi/n$, so compatible norms must be $\mathbb{Z}/n\mathbb{Z}$-invariant and produce singular flat geometries with cone points, CAT(0) geodesics, multi-foliations, and Liouville currents recovering directional length data [2604.02243].

## 1. General structure of directional dependence

A Finsler structure on a smooth manifold is a continuous map $F:TM\to[0,\infty)$ that is smooth away from the zero section, positively $1$-homogeneous in the fiber, positive definite, and strongly convex in the sense that the fiberwise Hessian of $F^2$ is positive definite for nonzero tangent vectors [2509.12660]. The associated length of a piecewise $C^1$ curve $\gamma$ is
\[
L(\gamma)=\int F(\gamma(t),\dot\gamma(t))\,dt.
\]
This framework permits anisotropy because the norm on each $T_xM$ need not arise from an inner product and may vary with direction inside a single fiber [2204.01034].

Directional dependence should not be conflated with non-reversibility. A Finsler metric is reversible when $F(x,-y)=F(x,y)$ and non-reversible otherwise; the latter typically yields a quasi-metric with $d_F(x,y)\neq d_F(y,x)$ [2509.12660]. However, several important anisotropic metrics remain reversible. On $1/n$-translation surfaces, the $\theta$-web norms are symmetric because they are built from sums of absolute values, so they are directionally dependent without being asymmetric [2604.02243]. A common misconception is therefore that anisotropy necessarily means orientation dependence in the sense of forward/backward distance asymmetry; the literature distinguishes these notions sharply.

From the connection-theoretic viewpoint, anisotropy also appears in the dependence of Christoffel-type data on $y$. Javaloyes, Sánchez, and Villaseñor formulate this through anisotropic connections $\nabla$ on $M$, with coefficients $\Gamma^a{}_{ij}(x,y)$ and induced nonlinear connection $N^a{}_i=\Gamma^a{}_{ij}y^j$, and identify vertically trivial Finsler connections canonically with anisotropic connections [2107.05986]. Minguzzi’s metrical approach shows that the vertical Hessian nature of $g$, the Barthel nonlinear connection, and the Chern and Cartan connections can be derived from compatibility axioms rather than postulated a priori [2107.09227].

## 2. Compatible metrics on $1/n$-translation surfaces

A $1/n$-translation structure on a surface $\Sigma$ is a $(\mathbb{R}^2\rtimes\mathbb{Z}/n\mathbb{Z},\mathbb{R}^2)$-structure on $\Sigma\setminus P$, where $P$ is a discrete set of cone points and the transition maps are Euclidean translations composed with rotations by multiples of $2\pi/n$ [2604.02243]. Equivalently, one may use complex charts with changes of coordinates
\[
z\mapsto e^{2\pi i k/n}z+c.
\]
The resulting flat cone metric has cone angles
\[
\theta(p)=\frac{2(k+n)\pi}{n}\ge 2\pi,
\]
and the universal cover carries a CAT(0) metric. For closed $\Sigma$ there is a combinatorial Gauss–Bonnet identity,
\[
\sum_{p\in P}(\theta(p)-2\pi)=4\pi(g-1).
\]

The same structure is equivalent to a complex structure $X$ together with a holomorphic $n$-differential $q$. In natural coordinates away from zeros, $q=dw^n$, while near a zero of order $k$ one has
\[
q=u^k\,du^n,
\]
which induces the cone angle $2(k+n)\pi/n$ [2604.02243]. This identification places directionally dependent Finsler metrics on $1/n$-translation surfaces in direct contact with the geometry of holomorphic differentials.

A compatible Finsler metric is induced by a $\mathbb{Z}/n\mathbb{Z}$-invariant norm on $\mathbb{R}^2$. Formally, if $\|\cdot\|_Q$ is invariant under rotations of order $n$, then under the canonical identification $T_x\Sigma\simeq\mathbb{R}^2$ one sets
\[
F|_{T_x\Sigma}=\|\cdot\|_Q.
\]
The model family is given by $\theta$-web norms. For
\[
v_{[\theta]}=\left\{\exp\!\left(i\left(\theta+\frac{2\pi k}{n}\right)\right)\;\middle|\;k=0,\ldots,n-1\right\},
\]
define
\[
\|u\|_\theta=\sum_{w\in v_{[\theta+\frac{\pi}{2}]}}|\langle u,w\rangle|,
\qquad
F_\theta(u)=\sum_{w\in v_{[\theta+\frac{\pi}{2}],p}}|\langle u,w\rangle|.
\]
Their unit balls are convex, generally polygonal, and invariant under rotation by $2\pi/n$ [2604.02243]. In these models anisotropy is exactly the dependence of length on the direction of $u$ relative to the web directions.

## 3. Geodesics, singularities, and homotopy minimization

For compatible metrics on $1/n$-translation surfaces, geodesics are defined globally as minimizers of the length functional in a homotopy class, rather than through a local Euler–Lagrange ODE. This choice is essential because non-strictly convex norms, especially polyhedral ones, can produce non-uniqueness of local minimizing directions [2604.02243]. The length functional is
\[
L(\gamma)=\int_0^1 F(\gamma'(t))\,dt.
\]

The central geodesic statement is Theorem A: for any $\mathbb{Z}/n\mathbb{Z}$-invariant norm on $\mathbb{R}^2$ and every $1/n$-translation surface $\Sigma$, the piecewise-straight CAT(0) geodesics—concatenations of saddle connections meeting cone points with angles at least $\pi$ on both sides—are geodesics for the induced Finsler distance and realize the smallest length in relative or free homotopy classes [2604.02243]. Thus the anisotropy changes the length assigned to a path but does not change the support of the minimizing representative in a homotopy class. This is a distinctive feature of the $1/n$-translation setting.

At cone points, admissibility is governed by the $\pi$-angle rule: adjacent saddle connections along a geodesic must meet with angle at least $\pi$ on both sides. Closed regular geodesics occur in one-parameter families forming immersed cylinders, and the extremal boundary curves of such cylinders are singular [2604.02243]. The presence of cylinders later reappears in the current-theoretic picture as atomic contributions.

A related but different piecewise flat theory, due to Xu and Deng, makes the direction dependence at singularities even more explicit. On piecewise flat Finsler surfaces, the curvature at a vertex can depend on an incoming or outgoing tangent direction, and extendability of a one-sided geodesic through a vertex is characterized by the sign of a direction-dependent curvature $K(x,v)$ [1608.05896]. This comparison shows that directional dependence may act either by reweighting fixed CAT(0) supports, as on $1/n$-translation surfaces, or by altering extension behavior at singular vertices.

## 4. Multi-foliations and Liouville currents

The current-theoretic component of the theory begins with multi-foliations. In $\mathbb{R}^2$, the $\theta$-multi-foliation is the union of all oriented straight lines in the directions $\theta+2\pi k/n$, and its transverse measure on a tangent vector $v$ is
\[
\sum_{k=0}^{n-1}\left|\left\langle v,\,v_{\theta+\frac{2k\pi}{n}+\frac{\pi}{2}}\right\rangle\right|
=
\sum_{w\in v_{[\theta+\frac{\pi}{2}]}}|\langle v,w\rangle|.
\]
Pulling this construction back through the $1/n$-translation atlas yields a multi-foliation on $\Sigma\setminus P$, and the associated length functional is
\[
\ell_\theta(\gamma)=\int \sum_{w\in v_{[\theta+\frac{\pi}{2}],\,\gamma(t)}}|\langle \gamma'(t),w\rangle|\,dt.
\]
For closed curves, $\ell_\theta([\gamma])$ is the infimum over the free homotopy class [2604.02243].

An oriented geodesic current on $\Sigma$ is a $\pi_1(\Sigma)$-invariant locally finite Borel measure on
\[
G^+(\widetilde\Sigma)=\big((\partial\widetilde\Sigma)\times(\partial\widetilde\Sigma)\big)\setminus\Delta.
\]
For each $\theta$, the theory constructs an oriented current $\mu_\theta$ supported on the endpoints of straight leaves of the $\theta$-multi-foliation. On a small box $C\times D$ in endpoint space, using an orthogonal segment $I$, the measure is defined by
\[
\mu_\theta(C\times D)=|I\cap L(C\times D)|.
\]
Carathéodory’s extension theorem then produces a unique measure, and cylinder leaves appear as atoms [2604.02243].

The key Crofton-type statement is
\[
i(\mu_\theta^s,\delta_\gamma)=\ell_\theta([\gamma]),
\]
where $\mu_\theta^s$ is the symmetrization of $\mu_\theta$ and $i(\cdot,\cdot)$ is Bonahon’s intersection pairing. For a general compatible norm, one has an integral representation
\[
\|v\|=\int_{S^1}\sum_{w\in v_{[\theta+\frac{\pi}{2}]}}|\langle v,w\rangle|\,d\nu(\theta),
\]
and therefore a Liouville current
\[
L=\int_{S^1}\mu_\theta^s\,d\nu(\theta)
\]
satisfying
\[
i(L,\delta_\gamma)=\ell_F([\gamma])
\]
for every closed curve $\gamma$ [2604.02243]. Existence and uniqueness follow from the current-theoretic rigidity of marked length spectra.

Multi-foliations generalize measured foliations to the $n$-fold rotational setting. When the $1/n$-structure is induced by a $1/m$-structure with $n=mk$, the current decomposes accordingly, and for $n=1,2$ the symmetrized current is a measured lamination with self-intersection zero [2604.02243]. For higher $n$, the support obeys a combinatorial crossing theorem: if $k=\lfloor n/2\rfloor$, then $\mu_\theta$ has many positive $k$-crossings but no positive $(k+1)$-crossing.

## 5. Model examples and comparative frameworks

The explicit web metrics already show how the symmetry order $n$ controls the shape of the infinitesimal unit ball [2604.02243].

| Case | Formula | Unit ball |
|---|---|---|
| $n=1$ | $F_\theta(u)=|\langle u,v_{\theta+\frac{\pi}{2}}\rangle|$ | strip |
| $n=2$ | $F_\theta(u)=|\langle u,v_{\theta+\frac{\pi}{2}}\rangle|+|\langle u,-v_{\theta+\frac{\pi}{2}}\rangle|$ | square |
| $n=3$ | $F_\theta(u)=\sum_{k=0}^{2}\left|\left\langle u,v_{\theta+\frac{\pi}{2}+\frac{2k\pi}{3}}\right\rangle\right|$ | regular hexagon |

Compared with Riemannian geometry, where the unit ball is an ellipse and geodesics are typically studied through local Euler–Lagrange equations, these compatible Finsler metrics are governed by non-elliptic unit balls with $\mathbb{Z}/n\mathbb{Z}$ symmetry, and direction dependence is encoded by projections onto web normals [2604.02243]. This difference is structural rather than cosmetic: polyhedral norms make global minimization more natural than local differential-equation formulations.

Other settings realize directional dependence differently. On convex projective surfaces, Danciger and Stecker introduced an asymmetric Finsler metric whose loop lengths are logarithms of top eigenvalues, and for large cubic differentials the metric converges to a Finsler metric with triangular unit balls
\[
F^\Delta_\mu(x,v)=\max\{2\Re(\alpha(v)):\alpha^3=\mu_x\},
\]
so logarithms of trace functions are approximated by Finsler lengths directly defined from the cubic differential [2309.10290]. This provides a higher-rank analogue of web-controlled anisotropy.

In complex Finsler geometry, Zhong proved that the unit ball admits no $\mathrm{Aut}(B_n)$-invariant strongly pseudoconvex complex Finsler metric other than a constant multiple of the Poincaré–Bergman metric, while the polydisk admits infinitely many $\mathrm{Aut}(P_n)$-invariant complete strongly convex Kähler-Berwald metrics
\[
F_{t,k}(z;v)=\sqrt{\frac{1}{1+t}\left[\sum_{l=1}^n |v_l|^2(1-|z_l|^2)^{-2}+t\sum_{l=1}^n |v_l|^{2k}(1-|z_l|^2)^{-2k}\right]},
\]
whose fundamental tensor depends explicitly on $v$ through $|v_i|^{2k-2}$ [2110.12436]. In yet another direction, position-dependent logarithmic and square-root metrics on $\mathbb{R}_{>0}^n$ are reversible and Euclidean-isometric, whereas the paper on Finsler TDA introduces genuinely non-reversible $\alpha+\beta$ metrics with
\[
F(x,y)=\alpha_f(x,y)+\sum_i b_i(x)f_i(x)^{1/\alpha}y^i
\]
and forward/backward balls that differ [2509.12660].

## 6. Misconceptions, rigidity, and open directions

Two recurrent misconceptions are explicitly contradicted by the literature. First, directional dependence does not force asymmetry: the $\theta$-web norms on $1/n$-translation surfaces are reversible, and several position-dependent information-geometric metrics are reversible as well [2604.02243]. Second, anisotropic geodesics need not be determined by a local spray formulation. For compatible norms on $1/n$-translation surfaces, especially polyhedral ones, the theory adopts homotopy-minimizing geodesics precisely because non-strict convexity can destroy local uniqueness; the global CAT(0) representative is the canonical object [2604.02243].

The principal open questions in the $1/n$-translation setting are also stated explicitly. For non-strictly convex norms, the global CAT(0) selection principle resolves existence of minimizers, but finer variational and geodesic-spray descriptions remain open. The support of multi-foliation currents is maximal among $(\lfloor n/2\rfloor+1)$-crossing-free currents, and whether this crossing property characterizes multi-foliation currents in general is posed as an intriguing question. The support combinatorics of these currents are also expected to interact with rigidity and flexibility questions in higher-rank settings and with boundaries of Hitchin components [2604.02243].

Within the broader Finsler framework, anisotropic connection theory clarifies a complementary issue: how directional dependence should be differentiated and parallel transported. Javaloyes, Sánchez, and Villaseñor show that vertically trivial Finsler connections are canonically equivalent to anisotropic connections, and that the Levi-Civita–Chern anisotropic connection is characterized by preservation of the fundamental tensor under observer-dependent parallel transport [2107.05986]. This suggests that global current-theoretic constructions and local anisotropic-connection constructions address different but compatible layers of the same geometry.

Directionally dependent Finsler metrics therefore form a broad class rather than a single model. In the surface theory of $1/n$-translation structures, they are controlled by $\mathbb{Z}/n\mathbb{Z}$-invariant norms, CAT(0) geodesics, multi-foliations, and Liouville currents. In convex projective, complex, and information-theoretic settings, they appear through triangular, polynomial, or asymmetric unit balls and through $y$-dependent fundamental tensors. Across these settings, the common principle is that the infinitesimal unit ball replaces the ellipsoid of Riemannian geometry, and the geometry records not only where a vector is based, but also in which direction it points.

Source: https://www.emergentmind.com/topics/directionally-dependent-finsler-metrics