Finsler–Laplacians: A Family of Operators

• Finsler–Laplacians are Laplace-type operators defined on Finsler manifolds, where the metric varies with both position and direction.
• They include various formulations such as nonlinear divergence-form, linear dynamical, and horizontal Hodge-type operators tailored for different analytic and geometric needs.
• Their study reveals insights into anisotropic spectral theory, heat flow, and comparison geometry, with applications ranging from eigenvalue estimates to shape analysis.

Finsler–Laplacians are Laplace-type operators attached to Finsler structures, where the metric depends on both position and direction rather than only on position as in Riemannian geometry. In the literature, the term does not denote a single canonical operator. It includes nonlinear divergence-form operators on functions, linear operators defined by averaging second derivatives along the geodesic flow, and horizontal Hodge-type operators on sphere bundles or prolongations. The defining features are direction-dependent fundamental tensors, Legendre transforms, geodesic dynamics, and an explicit dependence on the chosen measure or volume form (Ohta et al., 2011, Barthelmé, 2011, Mirshafeazadeh et al., 2019).

1. Conceptual scope and nonuniqueness

A recurrent point in the literature is that, unlike the Riemannian case, there is no unique Finslerian extension of the Laplace operator. One strand works with the nonlinear operator

$\Delta u=\operatorname{div}_m(\nabla u),$

where the gradient is defined by Legendre duality and the divergence depends on a chosen measure (Ohta et al., 2011). Another strand, due to Barthelmé, defines a linear, symmetric, elliptic second-order Laplace operator by averaging second derivatives over directions using an angle form produced from the contact-geometric structure of the geodesic flow (Barthelmé, 2011). Further constructions act not on functions on the base manifold, but on horizontal forms on the unit sphere bundle $SM$ or on prolongations of holomorphic Lie algebroids (Mirshafeazadeh et al., 2019, Ionescu, 2017). The paper on the “Finslerian sphere” explicitly remarks that several possibilities are available, mentioning Bao–Lackey, Shen, and Barthelmé, and then fixes Barthelmé’s definition for its analysis (Li, 2018).

This diversity is reinforced by measure dependence. The divergence-form theory is built relative to a smooth reference measure $m$, or in specific applications the canonical Hausdorff volume density, while Barthelmé’s dynamical operator is tied to the Holmes–Thompson volume (Ohta et al., 2011, Mester et al., 2023, Barthelmé, 2011). The “Finslerian sphere” example also stresses the coexistence of Busemann–Hausdorff and Holmes–Thompson volume forms and notes that its adopted Laplacian is related to the Holmes–Thompson volume form (Li, 2018).

This suggests that “Finsler–Laplacian” is best treated as a family name rather than a single operator.

Family	Defining mechanism	Typical setting
Nonlinear divergence-form	$\Delta u=\operatorname{div}_m(\nabla u)$	Weighted Finsler manifolds
Linear dynamical operator	Averaging $L_X^2(\pi^*f)$ over directions	Homogenized tangent bundle $HM$
Horizontal Hodge-type operator	$d_H\delta_H+\delta_H d_H$	Horizontal forms on $SM$
Applied Randers-derived FLBO	$$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$	Shape analysis on meshes

2. Nonlinear divergence-form Finsler–Laplacians

In the metric-measure framework, the Finsler gradient is defined by Legendre duality. For a differentiable function $u$,

$SM$0

and on the set where $SM$1,

$SM$2

Given a positive smooth measure $SM$3, locally $SM$4, the divergence is

$SM$5

and the Laplacian is

$SM$6

This is the basic operator in the work of Ohta and Sturm and in the nonlinear spectral theory of compact reversible Finsler metric measure manifolds (Ohta et al., 2011, Kristály et al., 2019).

Its nonlinearity is structural. The inverse tensor $SM$7 is evaluated at $SM$8 itself, so the map $SM$9 is nonlinear unless $m$0 is induced by a Riemannian metric. In local weighted-divergence form one has

$m$1

which makes the operator quasilinear/nonlinear (Kristály et al., 2019). The same basic definition is used in the eigenvalue estimates of Yin–He–Shen, where

$m$2

with divergence taken with respect to an arbitrary volume form $m$3 (Yin et al., 2012).

Several papers develop linearization along a reference direction. If $m$4, one defines

$m$5

so that $m$6 (Ohta et al., 2011). This linearized operator is the correct substitute for a fixed diffusion generator in Bochner formulas, gradient estimates, and local comparison arguments.

The divergence-form viewpoint also appears in more specialized settings. On bounded domains of complete Finsler manifolds, the nonsmooth Dirichlet inclusion paper fixes the canonical Hausdorff volume density

$m$7

defines

$m$8

and uses the weak identity

$m$9

as the basis of a nonsmooth variational theory (Mester et al., 2023).

A Euclidean but genuinely anisotropic model is the operator

$\Delta u=\operatorname{div}_m(\nabla u)$0

associated with a norm $\Delta u=\operatorname{div}_m(\nabla u)$1 on $\Delta u=\operatorname{div}_m(\nabla u)$2. This paper proves that $\Delta u=\operatorname{div}_m(\nabla u)$3 acts as a linear operator on $\Delta u=\operatorname{div}_m(\nabla u)$4-radially symmetric smooth functions, where $\Delta u=\operatorname{div}_m(\nabla u)$5 is the dual norm, and that

$\Delta u=\operatorname{div}_m(\nabla u)$6

for $\Delta u=\operatorname{div}_m(\nabla u)$7 (Akagi et al., 2017). That result does not turn the general operator into a linear one; it isolates a special symmetry class on which the anisotropic nonlinearity collapses to the classical radial Laplacian.

3. Linear and dynamical constructions

Barthelmé’s “A natural Finsler--Laplace operator” defines a linear operator directly from geodesic dynamics. The construction begins on the homogenized tangent bundle

$\Delta u=\operatorname{div}_m(\nabla u)$8

with Hilbert form $\Delta u=\operatorname{div}_m(\nabla u)$9 and Reeb field $L_X^2(\pi^*f)$0, where $L_X^2(\pi^*f)$1 is the generator of the geodesic flow. From the canonical volume

$L_X^2(\pi^*f)$2

one obtains a unique base volume form $L_X^2(\pi^*f)$3 and an $L_X^2(\pi^*f)$4-form $L_X^2(\pi^*f)$5 on $L_X^2(\pi^*f)$6 such that

$L_X^2(\pi^*f)$7

The operator is then

$L_X^2(\pi^*f)$8

It is a second-order differential operator, elliptic, symmetric with respect to $L_X^2(\pi^*f)$9, unitarily equivalent to a Schrödinger operator, and equal to the Laplace–Beltrami operator in the Riemannian case. The paper also identifies $HM$0 with the Holmes–Thompson volume and defines the associated energy

$HM$1

On compact manifolds this yields a standard discrete spectral theory and min–max characterization (Barthelmé, 2011).

The four-dimensional Finslerian Reissner–Nordström paper gives a concrete worked example of this linear operator on a non-Riemannian curved Finsler manifold, but only on a distinguished two-dimensional Randers-type angular subspace, the “Finslerian sphere.” The metric is

$HM$2

and the exact specialized operator $HM$3 is given explicitly, together with its first-order expansion in $HM$4 (Li, 2018). The same paper emphasizes that its construction is not a general intrinsic theory of Finsler Laplacians: the operator is imported from Barthelmé’s definition and specialized to a particular constant-flag-curvature Randers metric.

An application-oriented linearization appears in the shape-analysis paper on “Finsler-Laplace-Beltrami Operators.” Starting from the nonlinear Finsler heat equation

$HM$5

for a Randers metric

$HM$6

the paper derives, under short-time, small-$HM$7, and eikonal-type assumptions, the simplified linear diffusion

$HM$8

Its homogeneous part yields the FLBO

$HM$9

The paper is explicit that this FLBO is not the most general Finsler Laplacian in the mathematical literature; it is a tractable linear operator extracted from a simplified Randers heat analysis (Weber et al., 2024).

4. Horizontal Hodge Laplacians and bundle-level operators

A different branch of the subject replaces scalar diffusion on the base manifold by Hodge-type operators on forms over auxiliary bundles. On the unit sphere bundle

$d_H\delta_H+\delta_H d_H$0

the paper on harmonic vector fields defines horizontal $d_H\delta_H+\delta_H d_H$1-forms

$d_H\delta_H+\delta_H d_H$2

introduces the horizontal differential $d_H\delta_H+\delta_H d_H$3, the horizontal co-differential $d_H\delta_H+\delta_H d_H$4, and the horizontal Laplacian

$d_H\delta_H+\delta_H d_H$5

With respect to the canonical volume form $d_H\delta_H+\delta_H d_H$6 on $d_H\delta_H+\delta_H d_H$7, one has the adjointness relation

$d_H\delta_H+\delta_H d_H$8

and the Hodge-type theorem

$d_H\delta_H+\delta_H d_H$9

This framework yields a definition of harmonic vector fields through the associated horizontal 1-form and leads to a Bochner–Yano type classification theorem based on the harmonic Ricci scalar (Mirshafeazadeh et al., 2019).

On holomorphic Lie algebroids, the prolongation $SM$0 carries horizontal and vertical splittings induced by a Chern–Finsler nonlinear connection. The paper on “Laplace operators on holomorphic Lie algebroids” defines, for functions on the prolongation,

$SM$1

with explicit local formulas

$SM$2

$SM$3

For horizontal $SM$4-forms it defines

$SM$5

again with local expressions in terms of the Chern–Finsler connection (Ionescu, 2017).

These constructions show that “Finsler–Laplacian” can also mean a horizontal Hodge or Kodaira-type operator rather than a scalar Laplacian on functions. A common misconception is to identify all Finsler Laplacians with nonlinear divergence-form operators on the base manifold. The bundle-level literature is explicitly different in domain, connection, and analytic role (Mirshafeazadeh et al., 2019, Ionescu, 2017).

5. Spectral theory and model geometries

For Barthelmé’s linear operator, the compact theory closely parallels the Riemannian case. The spectrum of $SM$6 is discrete, with an infinite unbounded sequence of real eigenvalues of finite multiplicity, orthogonal eigenspaces, and complete eigenfunctions in $SM$7. The corresponding Rayleigh quotient is

$SM$8

and in dimension $SM$9 the operator obeys the conformal covariance law

$$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$0

(Barthelmé, 2011).

The spectral comparison paper shows that this linear Finsler–Laplacian is stable under pointwise bi-Lipschitz comparison of Finsler norms, but with constants depending on quasireversibility. If $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$1 and $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$2 satisfy

$$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$3

then for every $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$4 there exists $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$5, depending on $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$6, the quasireversibility constants, and the dimension, such that

$$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$7

On a surface of genus $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$8, if $$-\operatorname{div}_X(D_{\mathcal F_x^*}\nabla_X u)$$9 is $u$0-quasireversible, then

$u$1

The same paper shows that the dependence on quasireversibility is essential by constructing Randers metrics on any surface with arbitrarily large

$u$2

This is presented as a genuinely non-Riemannian phenomenon (Barthelmé et al., 2012).

For nonlinear divergence-form operators, higher spectral theory is more delicate because the operator is not linear. The paper “Nonlinear spectrums of Finsler manifolds” addresses this by defining min–max eigenvalues using faithful dimension pairs. The energy functional is

$u$3

and eigenfunctions are critical points of $u$4 on the normalized $u$5-sphere. The paper proves that Lusternik–Schnirelmann and Krasnosel'skii constructions are faithful and produce actual weak eigenvalues, whereas the modified Lebesgue covering dimension pair is not faithful and gives

$u$6

This is an explicit controversy inside nonlinear spectral theory: not every topological min–max scheme yields a meaningful higher spectrum (Kristály et al., 2019).

Explicit model geometries clarify how anisotropy alters spectra. On the “Finslerian sphere,” the eigenfunctions of the specialized Barthelmé operator take the perturbative form

$u$7

and the eigenvalue correction depends on both $u$8 and $u$9. The paper emphasizes that $SM$00 retains eigenvalue $SM$01, while $SM$02 are shifted, reflecting the surviving $SM$03-axis symmetry and the breaking of the full spherical degeneracy (Li, 2018).

Lower bounds for the first eigenvalue have also been extended to the nonlinear operator $SM$04. Under weighted Ricci and $SM$05-curvature assumptions, one has the Lichnerowicz-type estimate

$SM$06

and in the case $SM$07 and $SM$08,

$SM$09

Under $SM$10, the paper proves the Zhong–Yang-type bound

$SM$11

These are obtained by combining the Finsler Bochner formula with gradient estimates adapted to the nonlinear setting (Yin et al., 2012).

6. Heat flow, comparison geometry, and applications

For the nonlinear divergence-form Laplacian, the decisive analytic tool is the Finslerian Bochner–Weitzenböck formula. On a weighted Finsler manifold, Ohta and Sturm prove

$SM$12

together with the dimensional inequality

$SM$13

From these formulas they derive Li–Yau type gradient estimates, parabolic Harnack inequalities, and Bakry–Émery gradient estimates for the nonlinear heat equation

$SM$14

(Ohta et al., 2011).

The semigroup paper extends this program to a Bakry–Ledoux framework in the nonlinear setting. Its central observation is that the true heat flow is nonlinear, while the relevant semigroup estimates are formulated through the linearized semigroup $SM$15 along a heat trajectory. Under $SM$16, finite $SM$17 and $SM$18, and an additional noncompact regularity assumption, it proves the $SM$19-gradient estimate

$SM$20

and shows that, modulo the stated assumptions, the following are equivalent: $SM$21, the Bochner inequality, the improved Bochner inequality, the $SM$22-gradient estimate, and the $SM$23-gradient estimate. The same machinery yields Bakry–Ledoux’s Gaussian isoperimetric inequality on Finsler manifolds, including the non-reversible case (Ohta, 2016).

A newer development studies comparison geometry for the reference-vector Laplacian

$SM$24

rather than only $SM$25. Under forward completeness, finite misalignment, lower bounds on the mixed weighted Ricci curvature, and bounds on non-Riemannian tensors, the paper proves the pointwise distance estimate

$SM$26

wherever the distance function $SM$27 is smooth. This new Laplacian comparison theorem is then used to derive global and local Li–Yau type estimates for the Finslerian Schrödinger equation

$SM$28

on compact and noncompact forward complete Finsler manifolds (Shen, 2023).

Normed-space models and variational problems supply further applications. For

$SM$29

the Cauchy problem paper proves an optimal sufficient condition for measure initial data in the Finsler heat equation and exhibits the anisotropic Gaussian

$SM$30

as an explicit solution (Akagi et al., 2017). The nonsmooth variational paper treats the Dirichlet inclusion

$SM$31

on bounded domains of complete Finsler manifolds and proves a threshold-type result: only the trivial solution for small $SM$32, and two different nontrivial nonnegative weak solutions for large $SM$33 (Mester et al., 2023). In geometric data analysis, the FLBO of the Randers-based shape-analysis paper is used as an anisotropic spectral operator for heat kernels, filtering, and correspondence estimation, but that paper is explicit that its construction is a specialized, application-driven linearization rather than a general theory of Finsler Laplacians (Weber et al., 2024).

Across these directions, the central fault line remains clear. Some Finsler–Laplacians are nonlinear metric-measure operators on functions, some are linear dynamical averages over the geodesic flow, and some are horizontal Hodge operators on auxiliary bundles. Their coexistence is not a defect of the literature; it reflects the fact that Finsler geometry admits several natural analytic structures, each preserving a different part of the Riemannian Laplacian paradigm.