--- title: Almost Finsler Manifolds Overview url: https://www.emergentmind.com/topics/almost-finsler-manifolds type: topic --- # Almost Finsler Manifolds Overview Almost Finsler manifolds arise in several closely related but non-identical extensions of classical Finsler geometry. In one direct recent formulation, a partial or almost Finsler manifold is a triple \((M,S,F)\) with \(M\) smooth, \(S\subset TM\) a closed conelike subset containing the zero section, and \(F:TM-S\to(0,\infty)\) smooth and positively \(1\)-homogeneous, such that each fiber \((T_xM,S\cap T_xM,F|_x)\) is a partial or almost Minkowski space; the almost case requires positive definiteness of the fiberwise fundamental tensor away from the slit [2508.21744]. Other works use the expression for even-dimensional pseudo-Finsler manifolds carrying local almost hypercomplex structures [1305.5996], or for Finsler manifolds considered up to almost isometry through their forward quasi-metric and smooth semi-Lipschitz cone [1911.07991]. This suggests that the term is context-dependent, but in each usage it marks a controlled weakening, enrichment, or equivalence-based reformulation of the standard Finsler framework. ## 1. Foundational setting and competing definitions A standard Finsler metric on a smooth manifold \(M\) is a function \(F:TM\setminus\{0\}\to(0,\infty)\) that is positively \(1\)-homogeneous in the fiber variables and whose fundamental tensor \[ g_{ij}(x,y)=\frac{1}{2}\frac{\partial^2(F^2)(x,y)}{\partial y^i\partial y^j} \] is positive definite for every nonzero \(y\). In the formulation of Javaloyes–Sánchez, two basic relaxations are separated: a conic pseudo-Finsler metric allows the domain to be a conic open subset \(A\subset TM\) and permits the fundamental tensor to be indefinite, while a conic Finsler metric is a conic pseudo-Finsler metric with positive-definite fundamental tensor on \(A\setminus\{0\}\); a pseudo-Finsler metric is the full-domain case \(A=TM\) with no positivity requirement, and a standard Finsler metric is recovered when \(A=TM\) and \(g_y\succ 0\) [1111.5066]. The more recent partial/almost formalism changes the domain language from a conic open set to a slit \(S\subset TM\). A partial Finsler manifold \((M,S,F)\) is defined on \(TM-S\), where \(S\) is closed, conelike, nonempty, and contains the zero section; an almost Finsler manifold is the corresponding positive-definite case. The fiberwise model is a partial or almost Minkowski space, so the fundamental tensor is still obtained from the second fiber derivatives of \(F^2/2\), but positivity is only required away from the slit [2508.21744]. These definitions are not interchangeable. In the even-dimensional pseudo-Finsler literature, an “almost Finsler manifold” may instead mean an even-dimensional pseudo-Finsler manifold \(F^{2n}=(M,M',F^*)\) with \(M'\subset TM\) open, \(\pi(M')=M\), \(0(M)\cap M'=\varnothing\), and \(F^*\) positively \(2\)-homogeneous, together with local almost hypercomplex structures built from a non-linear connection [1305.5996]. In the Banach–Stone-type theory of almost isometries, “almost Finsler manifolds” are effectively Finsler manifolds studied modulo almost isometry, with the forward quasi-metric and the cone \(SC_{1^{-}^1}\) playing the central role [1911.07991]. ## 2. Slit geometry, positivity, and generalized local models For a partial Minkowski space \((V,S,F)\), the slit \(S\subset V\) is conelike, meaning \(S=\lambda S\) for all \(\lambda>0\). The associated bilinear form at \(y\in V-S\), \[ (u,v)_y:=\frac12\lim_{s,t\to 0}\frac{F^2(y+su+tv)}{st}, \] is the fiberwise Hessian of \(F^2/2\); positive definiteness of this form is exactly the additional condition defining an almost Minkowski space. The indicatrix is the hypersurface \(F^{-1}(\{1\})\), and the solid indicatrix is \(F^{-1}([0,1])\). Because \(F\) is \(1\)-homogeneous, the indicatrix determines the norm [2508.21744]. A partial Finsler manifold need not remain positive-definite on all of \(TM-S\). To isolate the genuinely positive region, the extended slit is defined by \[ ES=S\cup \{y\in TM-S: g_y \text{ is not positive definite}\}. \] Then \((M,ES,F)\) is an almost Finsler manifold, called the truncation of \((M,S,F)\). This construction makes explicit that nonpositive eigenvalues of \(g_y\) are treated as excluded directions rather than admissible anisotropies in the almost category [2508.21744]. The conic pseudo-Finsler framework clarifies why this truncation is geometrically significant. For Minkowski conic pseudo-norms, the paper of Javaloyes–Sánchez develops a unit-ball/indicatrix characterization in terms of the closed unit ball \(B\subset A\) and indicatrix \(S=\partial B\): strong convexity of the indicatrix characterizes conic Minkowski norms, convexity yields the triangle inequality, and strict convexity yields the strict triangle inequality. At the same time, conic and pseudo-Finsler generalizations exhibit genuine pathologies. For conic Finsler metrics, forward and backward balls are open, but they may fail to form a basis unless the metric is lower bounded; the induced separation \(d_F\) may be discontinuous or even vanish between distinct points. In pseudo-Finsler geometry, if the structure is not Finsler, then degeneracy of \(g\) must occur somewhere on every tangent space where positivity fails [1111.5066]. This combination of slit-based exclusion and conic-domain restriction shows that “almost” in Finsler geometry is not merely terminological. It identifies a regime in which one keeps homogeneity and differential tensor calculus but relaxes full-domain availability or strong convexity, often at the price of subtler topology of balls, weaker distance behavior, or singular directions. ## 3. Bipartite spaces, Randers-type models, and characteristic tensors A particularly explicit class of almost/partial Finsler structures is furnished by bipartite spaces. Let \((M,\rho)\) be Riemannian, with \[ \rho(y)=\sqrt{r_{ij}(x)y^iy^j}, \] and let \(s_{ij}(x)\) be a symmetric nonnegative \((0,2)\)-tensor whose eigenvalues lie in \([0,1)\). Writing \[ \sigma(x,y)=\sqrt{s_{ij}(x)y^iy^j}, \qquad S_x=\ker(s_x), \qquad S=\bigcup_x S_x, \] the bipartite norms are \[ F^\pm=\rho\pm \sigma. \] The pairs \((M,S,F^+)\) and \((M,S,F^-)\) are partial Finsler manifolds; their indicatrices are called the lemon and the apple, respectively. The slit coincides with the space of fixed points under scaling in the set-theoretic sense used in the paper, and \(S_x^f\) is a sphere of dimension \(\dim S_x-1\) [2508.21744]. Two especially important special cases are the \(\mathbf{a}\)- and \(\mathbf{b}\)-spaces. Given a nonzero vector \(b\), decompose \(y=y_{\parallel}+y_{\perp}\) into the components parallel and perpendicular to \(b\). The \(\mathbf{a}\)-spaces are defined by \[ F_b^\pm(y)=\|y\|\pm \|b\|\,\|y_{\parallel}\|, \] with slit \(S=b^\perp\); they are reversible, and on manifolds the corresponding slit is \(\bigcup_{x\in M}(a(x))^\perp\). The \(\mathbf{b}\)-spaces are defined by \[ F_b^\pm(y)=\|y\|\pm \|b\|\,\|y_{\perp}\|, \] with slit \(S=\mathbb{R}b\); \(F_b^+\) is always an almost Minkowski norm, while \(F_b^-\) is almost Minkowski only for \(\dim V=2\) [2508.21744]. The relation with Randers geometry is exact at the level of indicatrix unions. If \(\alpha(x,y)=\sqrt{a_{ij}(x)y^iy^j}\) and \(\beta(x,y)=b_i(x)y^i\), then \(F=\alpha+\beta\) is a Randers metric under the usual norm bound. For the \(\mathbf{a}\)-spaces, when \(\langle b,y\rangle>0\) one has \(F_b^\pm(y)=F_{\pm b}(y)\), and when \(\langle b,y\rangle<0\) one has \(F_b^\pm(y)=F_{\mp b}(y)\). Hence the indicatrix union of the almost Finsler \(\mathbf{a}\)-manifolds equals the union of the two Randers indicatrices. By contrast, the \(\mathbf{b}\)-space indicatrix union is a spindle toroid; in dimension \(3\) it is a \(2\)-dimensional spindle toroid formed by the apple and lemon glued along the fixed-point circle \(S^f\) [2508.21744]. The central tensorial contribution of this framework is the construction of characteristic \(3\)-tensors generalizing the Matsumoto tensor. With \[ p_j=F_j,\qquad h_{jk}=F F_{jk},\qquad C_{jkl}=\frac14(F^2)_{jkl},\qquad I_j=g^{kl}C_{jkl}, \] and \(\Delta=F-\rho\), the bipartite characteristic tensor is \[ S_{jkl}=C_{jkl} -\frac{1}{\kappa}\sum_{(jkl)} \left[ I_j+\frac{F^2}{(F-\Delta)\Delta}g^{mn}\Delta_m\Delta_{nj} \right] \left[ h_{kl}-\frac{F^2}{\Delta}\Delta_{kl} \right], \] with \[ \kappa=n+1-\frac{F^2}{\Delta}g^{kl}\Delta_{kl}. \] For bipartite almost Finsler manifolds, \(S_{jkl}=0\). In the \(\mathbf{b}\)-space case there is a simpler vanishing tensor, \[ B_{jkl}=C_{jkl} -\frac{1}{\kappa_b}\sum_{(jkl)} I_j \left[ h_{kl}-\frac{F^2}{\Delta}\Delta_{kl} \right], \] and \(B_{jkl}=0\) on all \(\mathbf{b}\)-spaces. In the Randers and \(\mathbf{a}\)-space limit, \(\Delta_{jk}=0\), so \(S_{jkl}\) reduces to the Matsumoto tensor \[ M_{jkl}=C_{jkl}-\frac{1}{n+1}\sum_{(jkl)}I_j h_{kl}, \] recovering the classical vanishing criterion for Randers geometry [2508.21744]. ## 4. Almost isometries and the functional reconstruction viewpoint A different use of “almost” keeps the underlying object a connected Finsler manifold \((\mathcal{X},F)\) but weakens the morphisms. In the nonreversible case the fundamental metric object is the forward quasi-metric \[ d^{+}(x,y)=\inf_{\gamma}\int_0^1 F(\gamma(t),\dot\gamma(t))\,dt, \] with backward distance \(d^{-}(x,y)=d^{+}(y,x)\) and symmetrized metric \(d^s(x,y)=d^{+}(x,y)\vee d^{+}(y,x)\). The dual norm on cotangent spaces is \[ F^*(\xi)=\sup_{v\neq 0}\frac{\xi(v)}{F(x,v)}. \] A smooth function is forward semi-Lipschitz with constant \(L\) if \[ f(y)-f(x)\le L\,d^+(x,y)\quad\forall x,y, \] and for \(C^1\) functions this is equivalent to the pointwise differential bound \(F^*(df_x)\le L\) [1911.07991]. The relevant function space is the convex partially ordered cone \[ SC_{\mathrm{SLip}^1}(\mathcal{X})=\{f\in C^\infty(\mathcal{X}): F^*(df_x)\le 1\ \text{for all }x\}, \] together with its open subcone \[ SC_{1^{-}^1}(\mathcal{X})=\{f\in C^\infty(\mathcal{X}): F^*(df_x)<1\ \text{for all }x\}. \] This cone simultaneously records differentiable information through \(df\) and quasi-metric information through \(d^+\). The fundamental global-local identity is \[ \|f\|_s:=\sup_{x\neq y}\frac{f(y)-f(x)}{d^+(x,y)} =\sup_{x\in\mathcal{X}}F^*(df_x). \] Almost isometries are defined by preservation of the triangular function \[ \operatorname{Tr}_d(x_1,x_2,x_3)=d(x_1,x_2)+d(x_2,x_3)-d(x_1,x_3). \] A bijection \(T:\mathcal{X}\to\mathcal{Y}\) is an almost isometry if it preserves \(\operatorname{Tr}\), equivalently if there exists a potential \(\phi:\mathcal{Y}\to\mathbb{R}\) such that \[ d^+_{\mathcal{Y}}(T(x),T(y)) = d^+_{\mathcal{X}}(x,y)+\phi(T(x))-\phi(T(y)). \] It is strict if it also satisfies a two-sided multiplicative control \[ c^{-1}d^+_{\mathcal{X}}(x,y)\le d^+_{\mathcal{Y}}(T(x),T(y))\le c\,d^+_{\mathcal{X}}(x,y) \] for some \(c\ge 1\). In the smooth Finsler setting, every almost isometry arises from a diffeomorphism and a smooth potential by \[ G=T^*(F)-d(\phi\circ T^{-1}), \] and strictness is equivalent to the associated potentials having semi-Lipschitz norm strictly less than \(1\) [1911.07991]. The main Banach–Stone-type theorem states that if \((\mathcal{X},d^+_{\mathcal{X}})\) and \((\mathcal{Y},d^+_{\mathcal{Y}})\) are connected, second countable, bicomplete Finsler manifolds and \[ \mathcal{T}:SC_{1^{-}^1}(\mathcal{Y})\to SC_{1^{-}^1}(\mathcal{X}) \] is an isomorphism of convex partially ordered sets, then there exist a constant \(a>0\), a smooth \(\phi=\mathcal{T}0\) with \(\|\phi\|_s<1\), and a diffeomorphism \(T:\mathcal{X}\to\mathcal{Y}\) such that \((\mathcal{X},d^+_{\mathcal{X}})\) is almost isometric to \((\mathcal{X},d'_{\mathcal{X}})\), \(T\) is an isometry between \((\mathcal{X},a\cdot d'_{\mathcal{X}})\) and \((\mathcal{Y},d^+_{\mathcal{Y}})\), and \[ \mathcal{T}f(x)=c\,f(T(x))+\phi(x),\qquad c=a^{-1}. \] In compact connected Finsler manifolds every almost isometry is strict, while in the reversible case the potential must be constant, so the classification collapses to composition operators with global scaling. In this sense, almost Finsler manifolds understood up to almost isometry are classified by the order-convex structure of \(SC_{1^{-}^1}\) [1911.07991]. ## 5. Even-dimensional pseudo-Finsler manifolds and local almost hypercomplex structures In another precise tradition, an almost Finsler manifold is an even-dimensional pseudo-Finsler manifold \(F^{2n}=(M,M',F^*)\). Here \(M\) is a smooth \(2n\)-manifold, \(M'\) is an open submanifold of \(TM\) satisfying \(\pi(M')=M\) and \(0(M)\cap M'=\varnothing\), and \(F^*:M'\to\mathbb{R}\) is smooth, positively \(2\)-homogeneous, and has associated quadratic form of signature \((q,2n-q)\) with \(0