betaFeedback

WorksMathematics

Jacobi fields and Jacobi sections for semisprays on Lie algebroids: transverse variations and equivalences

Abstract

We study the relationship between Jacobi sections and Jacobi fields associated with semisprays on Lie algebroids. First, we characterize compatible transverse variations in terms of Lie algebroid morphisms. This characterization leads to a variational description of Jacobi sections: a section along an integral curve of a semispray is a Jacobi section if and only if it arises as the infinitesimal variation of a compatible transverse variation of a dynamical variation satisfying a suitable condition on its image in the base. We also investigate the relationship between Jacobi sections and Jacobi fields through the anchor map. We prove that every Jacobi section projects to a Jacobi field, and establish a converse when the corresponding base variation is contained in a single leaf of the characteristic foliation.

Introduction

Jacobi fields arise naturally from the study of variations through geodesics on Riemannian manifolds or on manifolds with affine connections. They are the variational vector fields associated with such variations and describe the infinitesimal behavior of geodesics under perturbations of their initial conditions. Equivalently, Jacobi fields can be characterized as vector fields JJ along any geodesic curve γ\gamma satisfying the Jacobi equation:

Dt2J+R(J,γ˙)γ˙=0,D_t^2J+R(J,\dot{\gamma})\dot{\gamma}=0,

where DtD_t is the covariant derivative along γ\gamma and RR denotes the curvature tensor. Jacobi fields play an important role in several fundamental results in Riemannian geometry, including the Rauch comparison theorem, the Myers theorem, and the Cartan–Hadamard theorem.

The Levi-Civita connection on a Riemannian manifold (M,g)(M,g) induces a spray G∈X(TM)\mathcal{G}\in\mathfrak{X}(TM) which is called the geodesic spray. Then, any geodesic curve on MM is the projection of an integral curve of G\mathcal{G} by the canonical tangent bundle projection. Indeed, there exists a correspondence between variations of geodesics on MM and variations by integral curves on G\mathcal{G} and their corresponding infinitesimal variations. Hence, one can extend the study of variations of integral curves and infinitesimal variations of any spray or semispray on the tangent bundle. The semisprays on the tangent bundle are very important objects in mechanics since their integral curves describe the dynamics on the manifold governed by second-order differential equations. In this setting, Bučataru [4] derived the concept of Jacobi fields for a spray on the tangent bundle by proving that the infinitesimal variation of a variation of integral curves satisfies a kind of Jacobi equation which involves the curvature of the nonlinear connection induced by the spray. Later, Cariñena et al. [6] define the notion of Jacobi field for semisprays on the tangent bundle. Their definition of Jacobi fields is based on the infinitesimal variations of variations of integral curves on the base (i.e., projection on the base manifold of integral curves of the semisprays). They also derive a Jacobi-type equation for Jacobi fields of semisprays in terms of the covariant derivative along the semispray and the generalized Jacobi endomorphism defined by the nonlinear connections induced by the vertical endomorphism on the tangent bundle. Anahory Simões et al. introduce the notion of Jacobi fields and the corresponding Jacobi equation for nonholonomic mechanical systems using nonholonomic connections in [1].

The notion of Jacobi field has also been extended to Lie algebroids in different contexts, but mostly using a variational approach. For example, Boucetta introduced the notion of Jacobi sections along geodesic curves on a Riemannian Lie algebroid, as the sections along a given geodesic curve satisfying a Jacobi-type equation defined by the covariant derivative and the curvature associated with the Levi-Civita connection of the metric ([3]). Furthermore, Boucetta also proved that his definition of Jacobi sections admits a variational approach (see [3]). In [13], Jóźwikowski introduces Jacobi vector fields for Lagrangian systems on skew-symmetric algebroids using the Legendre mapping and the Tulczyjew differential. Despite the name, they are sections of the algebroid along a base curve. For semisprays on Lie algebroids, Cariñena, Gheorghiu and Martínez [6] define Jacobi sections along an integral curve of the semispray by means of a special kind of Lie algebroid morphisms and derive for them a Jacobi equation of the form ∇∇W+Φ(W)=0\nabla\nabla W + \Phi(W) = 0 which is expressed in terms of the dynamical covariant derivative ∇\nabla and the Jacobi endomorphism Φ\Phi of the semispray. Their constructions, including a variational characterization, are carried out on the prolongation. Whether this variational characterization can be reformulated in terms of variations of integral curves on the Lie algebroid itself, as in Boucetta’s setting, is the question we address in this paper.

In [2], we introduced the notion of Jacobi fields for semisprays on Lie algebroids: we consider a variation by integral curves of a given integral curve of the semispray, then we project the variation to the base manifold and define a Jacobi field as the variational vector field of the projected variation. Our definition recovers the classical definition of Jacobi fields in the case of the Riemannian spray of a Riemannian manifold. Actually, when we restrict to the case of sprays on a general Lie algebroid, we proved that they are in a one-to-one correspondence with the linear connections on Lie algebroids: given a spray, there exists a unique torsionless linear connection such that the integral curves of the spray are precisely the geodesic curves of the linear connection. Then, we can extend Boucetta’s definition of a Jacobi section along a geodesic (integral) curve to the case of a linear connection not necessarily given by a Riemannian metric on the Lie algebroid. At this point, we can associate to a given spray on a Lie algebroid its Jacobi sections and its Jacobi fields. We also show that every Jacobi section projects via the anchor map to a Jacobi field and, conversely, every Jacobi field tangent to a leaf of the characteristic foliation is the image of a Jacobi section.

The goal of this paper is to relate the Jacobi sections defined by Cariñena, Gheorghiu and Martínez [6] with the Jacobi fields introduced in [2]. Since the constructions of [6] are carried out on the prolongation of the Lie algebroid, while the notion of Jacobi fields is defined in terms of variations of integral curves of the semispray on the Lie algebroid itself, our first task is to translate between the two languages.

First, in Theorem 5.4 we show that a vector bundle morphism Φ=ζ dt+κ ds\Phi= \zeta\,dt + \kappa\,ds from TI~×TIT\widetilde{I} \times TI into AA is a Lie algebroid morphism if and only if ζ\zeta is a variation by admissible curves, with base contained in a leaf of the characteristic foliation, and κ\kappa is a compatible transverse variation of ζ\zeta. Then, we prove our first main theorem (Theorem 6.5): a section is a Jacobi section along an integral curve of a semispray if and only if it is the infinitesimal variation of a compatible transverse variation of a variation by integral curves of the semispray. This variational characterization is analogous to the one given by Boucetta in [3] for geodesics of Riemannian Lie algebroids. A technical, but relevant, ingredient of the proof is that the lift of a vector field to the prolongation of the tangent Lie algebroid is a Lie algebroid morphism (Proposition 3.5).

In our formulation, the compatibility condition is the only ingredient whose definition still relies on the prolongation; we leave open the question of describing it intrinsically on the Lie algebroid.

The second step is to compare Jacobi sections with Jacobi fields. Our second main result states that the anchor map relates both notions: every Jacobi section projects to a Jacobi field (Theorem 7.1), and, conversely, every Jacobi field arising from a variation whose projection to the base lies in a single leaf of the characteristic foliation is the image of a Jacobi section (Theorem 7.2). In [2] this relation was proved for sprays, where Jacobi sections can be defined through the torsionless linear connections associated with the spray, and it was stated in [2 Remark 6] that the same relation holds for semisprays. Here we establish it for arbitrary semisprays, using the Jacobi sections of [6], which do not require a linear connection.

The paper is organized as follows. In § 2, we recall some basic facts on Lie algebroids. § 3 is devoted to the geometry of Lie algebroid prolongations and the constructions needed throughout the paper. In § 4, we introduce semisprays, dynamical variations and the Jacobi fields of [2]. In § 5 we characterize variations of admissible curves with compatible transverse variations in terms of Lie algebroid morphisms. In § 6 we use this characterization to obtain a variational description of the Jacobi sections of [6] along integral curves of a semispray. In § 7 we relate Jacobi sections and Jacobi fields through the anchor map, and establish the conditions under which a Jacobi field is the projection of a Jacobi section. Finally, in § 8 we present an example illustrating this correspondence.

Preliminaries on Lie algebroids

Definition 2.1. A Lie algebroid is a triple (p:A→M,[⋅,⋅]A,ρ)(p : A \to M,[\cdot,\cdot]_A,\rho) where p:A→Mp : A \to M is a vector bundle with a real Lie algebra bracket [⋅,⋅]A:Γ(A)×Γ(A)→Γ(A)[\cdot,\cdot]_A : \Gamma(A) \times\Gamma(A) \to\Gamma(A) defined on the C∞(M)C^\infty(M)-module of sections Γ(A)\Gamma(A) and with a vector bundle homomorphism ρ:A→TM\rho: A \to TM, called anchor map, satisfying the Leibniz rule

[σ,fη]A=f[σ,η]+(Lρ(σ)f)η,[\sigma,f\eta]_A=f[\sigma,\eta]+(L_{\rho(\sigma)}f)\eta,

for all σ,η∈Γ(A)\sigma,\eta\in\Gamma(A), f∈C∞(M)f \in C^\infty(M).

Since the anchor map ρ:A→TM\rho: A \to TM is a bundle morphism, it induces a morphism of C∞(M)C^\infty(M)-modules ρ:Γ(A)→X(M)\rho: \Gamma(A) \to\mathfrak{X}(M). Moreover, an immediate consequence of this definition is that the anchor map ρ\rho is a Lie algebra homomorphism:

ρ([σ,η]A)=[ρ(σ),ρ(η)]TM,σ,η∈Γ(A).\rho([\sigma,\eta]_A)=[\rho(\sigma),\rho(\eta)]_{TM},\qquad\sigma,\eta\in\Gamma(A).

The image of the anchor map D:=ρ(A)\mathcal{D}:=\rho(A) is a smooth generalized distribution on MM which is integrable in the sense of Stefan-Sussmann [19, 20]. The integral submanifolds of D\mathcal{D} define a singular foliation of MM called the characteristic foliation.

If we take local coordinates (U,x1,…,xn)(U,x^{1},\ldots,x^{n}) on MM and a local basis {ei}i=1r\{e_i\}_{i=1}^{r} of sections Γ(U)\Gamma(U) we obtain local coordinates of AA, called adapted coordinates:

A∋a↦(x1,…,xn,y1,…,yr),(1)A \ni a \mapsto(x^{1},\ldots,x^{n},y^{1},\ldots,y^{r}), \tag*{(1)}

where a=yj⋅ej(x)a=y^{j}\cdot e_j(x). The smooth functions ρji,Cijk\rho^i_j,C^k_{ij} on C∞(U)C^\infty(U) defined by

ρ(ej)=ρji∂∂xi,[ei,ej]A=Cijkek(2)\rho(e_j)=\rho^i_j\frac{\partial}{\partial x^i},\qquad[e_i,e_j]_A=C^k_{ij}e_k \tag*{(2)}

are called the structure functions of the Lie algebroid. These functions satisfy the so-called structure equations on the Lie algebroid:

ρji∂ρlk∂xi−ρli∂ρjk∂xi=ρikCjli,∑cyclic⁡(j,l,s)(ρji∂Clsk∂xi−CjikClsi)=0.(3)\rho^i_j\frac{\partial\rho^k_l}{\partial x^i}-\rho^i_l\frac{\partial\rho^k_j}{\partial x^i}=\rho^k_iC^i_{jl},\qquad\sum_{\operatorname{cyclic}(j,l,s)}\left(\rho^i_j\frac{\partial C^k_{ls}}{\partial x^i}-C^k_{ji}C^i_{ls}\right)=0. \tag*{(3)}

Now, we present some basic examples of Lie algebroids.

Example 2.2. For any manifold MM, the tangent bundle τM:TM→M\tau_M:TM\to M is a Lie algebroid with the standard Lie bracket of vector fields and anchor map given by the identity.

Example 2.3. Let (M,Π)(M,\Pi) be a Poisson manifold, i.e., a manifold MM together with bivector Π∈X2(M)\Pi\in\mathfrak{X}^{2}(M) satisfying the Jacobi identity

[Π,Π]SCH=0.[\Pi,\Pi]_{\mathrm{SCH}}=0.

The Poisson bivector Π\Pi induces the following vector bundle morphism

Π♯:T∗M→TM,Π♯(ω):=iωΠ.\Pi^\sharp:T^*M\to TM,\qquad\Pi^\sharp(\omega):=\mathord{\mathrm{i}}_{\omega}\Pi.

Moreover, it is well known that the following bracket

[⋅,⋅]Π:Ω(M)×Ω(M)→Ω(M),[σ,η]Π:=LΠ♯(σ)η−LΠ♯(η)σ−d(Π(σ,η)),[\cdot,\cdot]_{\Pi}:\Omega(M)\times\Omega(M)\to\Omega(M),\qquad[\sigma,\eta]_{\Pi}:=\mathcal{L}_{\Pi^\sharp(\sigma)}\eta-\mathcal{L}_{\Pi^\sharp(\eta)}\sigma-\mathrm{d}(\Pi(\sigma,\eta)),

is a Lie bracket on the R\mathbb{R}-linear space of 11-forms Ω(M)\Omega(M). Furthermore, the canonical cotangent bundle projection πM:T∗M→M\pi_M:T^*M\to M is a Lie algebroid with Lie bracket [⋅,⋅]Π[\cdot,\cdot]_{\Pi} and anchor map Π♯:T∗M→TM\Pi^\sharp:T^*M\to TM. This example is discussed in full detail in [10].

Example 2.4. An involutive distribution D⊂TM\mathcal{D}\subset TM is a Lie algebroid with the inclusion D↪TM\mathcal{D}\hookrightarrow TM as anchor map and the usual bracket of vector fields as Lie bracket.

Example 2.5. Let g∋ξ↦ξM∈X(M)\mathfrak{g}\ni\xi\mapsto\xi_M\in\mathfrak{X}(M) be a (right) action of the real Lie algebra g\mathfrak{g} on the manifold MM. Its associated Lie algebroid consists of the trivial vector bundle pr⁡1:M×g→M\operatorname{pr}_1:M\times\mathfrak{g}\to M, with anchor map ρ:M×g→TM\rho:M\times\mathfrak{g}\to TM defined by

ρ(m,ξ):=ξM(m).\rho(m,\xi):=\xi_M(m).

Identifying the sections of pr⁡1\operatorname{pr}_1 with the set of g\mathfrak{g}-valued functions v:M→gv:M\to\mathfrak{g}, one can define a Lie bracket by

[v,w](m):=[v(m),w(m)]+⟨dw,(v(m))M⟩(m)−⟨dv,(w(m))M⟩(m).[v,w](m):=[v(m),w(m)]+\langle\mathrm{d}w,(v(m))_M\rangle(m)-\langle\mathrm{d}v,(w(m))_M\rangle(m).

A section of the dual bundle π:A∗→M\pi:A^* \to M will be referred to as an AA-form. More generally, elements of Γ(∧kA∗)\Gamma(\wedge^k A^*) will be called AA-forms of degree kk.

The set of 1-forms Γ(A∗)\Gamma(A^*) can be identified with the space of linear functions Clin∞(A)C^\infty_{\mathrm{lin}}(A) as follows: for every θ∈Γ(A∗)\theta\in\Gamma(A^*),

lθ(a):=⟨θp(a),a⟩.(4)l_\theta(a) := \langle\theta_{p(a)},a\rangle. \tag*{(4)}

The algebraic structures on a Lie algebroid allow us to define an exterior differential operator dA:Γ(∧∙A∗)→Γ(∧∙A∗)d_A:\Gamma(\wedge^\bullet A^*) \to\Gamma(\wedge^\bullet A^*), defined over the C∞(M)C^\infty(M)-module of AA-forms Γ(∧∙A∗)\Gamma(\wedge^\bullet A^*) as follows: if ω∈Γ(∧kA∗)\omega\in\Gamma(\wedge^k A^*)

dAω(σ0,…,σk):=∑i=0k(−1)iLρ(σi)(ω(σ0,…,σi^,…,σk))+∑0≤i<j≤k(−1)i+jω([σi,σj]A,σ0,…,σi^,…,σj^,…,σk),(5)\begin{aligned} d_A\omega(\sigma_0,\ldots,\sigma_k) := {}& \sum_{i=0}^{k}(-1)^i\mathcal{L}_{\rho(\sigma_i)}(\omega(\sigma_0,\ldots,\widehat{\sigma_i},\ldots,\sigma_k))\\ &+ \sum_{0\leq i<j\leq k}(-1)^{i+j}\omega([\sigma_i,\sigma_j]_A,\sigma_0,\ldots,\widehat{\sigma_i},\ldots,\widehat{\sigma_j},\ldots,\sigma_k), \tag*{(5)} \end{aligned}

where σ0,…,σk∈Γ(A)\sigma_0,\ldots,\sigma_k \in\Gamma(A). The operator dAd_A is a graded derivation of degree one and a coboundary operator, i.e., dA2=0d_A^2=0, of the cochain complex (Γ(∧∙A∗),dA)(\Gamma(\wedge^\bullet A^*),d_A).

Definition 2.6. Given two Lie algebroids (p1:A1→M1,[⋅,⋅]A1,ρ1)(p_1:A_1\to M_1,[\cdot,\cdot]_{A_1},\rho_1) and (p2:A2→M2,[⋅,⋅]A2,ρ2)(p_2:A_2\to M_2,[\cdot,\cdot]_{A_2},\rho_2), a Lie algebroid morphism Φ:A1→A2\Phi:A_1\to A_2 over φ:M1→M2\varphi:M_1\to M_2 is a vector bundle homomorphism Φ\Phi covering φ\varphi such that the induce map on forms:

Φ∗:Γ(∧kA2∗)→Γ(∧kA1∗),\Phi^*:\Gamma(\wedge^k A_2^*)\to\Gamma(\wedge^k A_1^*),

defined by ⟨Φ∗(σ),a⟩=⟨σφ(m),Φ(a)⟩\langle\Phi^*(\sigma),a\rangle=\langle\sigma_{\varphi(m)},\Phi(a)\rangle, commutes with the exterior differential

dA1∘Φ∗=Φ∗∘dA2.d_{A_1}\circ\Phi^*=\Phi^*\circ d_{A_2}.

The properties of Lie algebroid morphisms are summarized in the next proposition.

Proposition 2.7. The set of Lie algebroids together with Lie algebroid morphisms form a category.

This is a standard fact which can be easily proven; see, for example, [14].

Example 2.8. On any Lie algebroid (p:A→M,[⋅,⋅]A,ρ)(p:A\to M,[\cdot,\cdot]_A,\rho), the anchor map ρ:A→TM\rho:A\to TM is a Lie algebroid morphism between the Lie algebroid itself and the tangent bundle. From the formula (5) for the differential exterior operator, one can deduce

ρ∗∘d=dA∘ρ∗.\rho^*\circ d=d_A\circ\rho^*.

The geometry of Prolongation of a Lie algebroid

The prolongation of a Lie algebroid over its bundle projection turns out to be the natural analogue of the tangent bundle of a tangent bundle. It provides the natural geometric framework for the study of semisprays, second-order dynamics and variational objects on Lie algebroids. Here, we develop only the tools necessary for the paper. Further details on the Lie algebroid prolongation can be found in [7, 12, 17, 18].

For a given Lie algebroid (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho), its prolongation over AA is the vector bundle pA:TA→Ap_A:\mathcal{T}A\to A where

TA:={(a,b,v)∈A×A×TA∣p(a)=p(b); ρ(b)=Tap(v)}(6)\mathcal{T}A:=\{(a,b,v)\in A\times A\times TA\mid p(a)=p(b);\ \rho(b)=T_ap(v)\} \tag*{(6)}

and projection bundle given by

pA(a,b,v):=a.(7)p_A(a,b,v):=a. \tag*{(7)}

The fiber over a∈Aa\in A is isomorphic to the vector space

(pA)−1(a)≃{(b,v)∈Ap(a)×TaA∣ρ(b)=Tap(v)}.(p_A)^{-1}(a)\simeq\{(b,v)\in A_{p(a)}\times T_aA\mid\rho(b)=T_ap(v)\}.

We also consider the map p^:TA→A\hat p:\mathcal{T}A\to A given by projection onto the second factor:

p^(a,b,v):=b.(8)\hat p(a,b,v):=b. \tag*{(8)}

With the aid of the map p^:TA→A\hat p:\mathcal{T}A\to A, we can define a special kind of sections of TA\mathcal{T}A called projectable sections.

Definition 3.1. A section Z∈Γ(TA)Z\in\Gamma(\mathcal{T}A) is said to be projectable if there exists σ∈Γ(A)\sigma\in\Gamma(A) such that

p^∘Z=σ∘p.(9)\hat p\circ Z=\sigma\circ p. \tag*{(9)}

One can deduce that every projectable section ZZ has the form

Z=(id⁡A,σ∘p,X),Z=(\operatorname{id}_A,\sigma\circ p,X),

where σ∈Γ(A)\sigma\in\Gamma(A) and X∈X(A)X\in\mathfrak{X}(A) are such that XX and ρ(σ)\rho(\sigma) are pp-related vector fields.

Lie algebroid structure of TA\mathcal{T}A. The key property that allows us to define a Lie algebroid structure on TA\mathcal{T}A is the fact that every section of TA\mathcal{T}A is locally written as a linear combination of projectable sections. The Lie bracket can be easily defined in terms of projectable sections and extended by R\mathbb{R}-linearity and the Leibniz rule.

To obtain a local basis of sections on TA\mathcal{T}A, we have to consider the local coordinate system on the prolongation TA\mathcal{T}A constructed as follows: fix an adapted coordinate system (xi,yk)(x^i,y^k) on AA associated with a local chart (U,x1,x2,…,xn)(U,x^1,x^2,\ldots,x^n) on MM and a basis {el}\{e_l\} of local sections on AA. Now, we choose a point (a,b,v)∈TA(a,b,v)\in\mathcal{T}A with a,b∈p−1(U)a,b\in p^{-1}(U). Since p(a)=p(b)p(a)=p(b), the points a,b∈Aa,b\in A have coordinates (xi,uk)(x^i,u^k) and (xi,yj)(x^i,y^j), respectively. Now, the condition ρ(b)=Tap(v)\rho(b)=T_ap(v) implies that vv has the form

v=yjρji(x)∂∂xi∣a+vl∂∂yl∣a.v=y^j\rho^i_j(x)\left.\frac{\partial}{\partial x^i}\right|_a+v^l\left.\frac{\partial}{\partial y^l}\right|_a.

Thus, one can assign

TA∋(a,b,v)↦(xi,uk,yj,vl),(10)\mathcal{T}A\ni(a,b,v)\mapsto(x^i,u^k,y^j,v^l), \tag*{(10)}

as coordinates. It follows from here that rank⁡TA=2rank⁡A\operatorname{rank}\mathcal{T}A=2\operatorname{rank}A.

Then, we define a local basis {Ei,Vj}\{\mathcal{E}_i,\mathcal{V}_j\} of sections of TA\mathcal{T}A by formulas:

Ej(a):=(a,ej(p(a)),ρji(x)∂∂xi∣a),andVk(a):=(a,0,∂∂yk∣a).(11)\mathcal{E}_j(a):=\left(a,e_j(p(a)),\rho^i_j(x)\left.\frac{\partial}{\partial x^i}\right|_a\right),\qquad\text{and}\qquad \mathcal{V}_k(a):=\left(a,0,\left.\frac{\partial}{\partial y^k}\right|_a\right). \tag*{(11)}

These local sections are projectable: p^∘Ej=ej∘p\hat p\circ\mathcal{E}_j=e_j\circ p and p^∘Vk=0\hat p\circ\mathcal{V}_k=0.

Finally, the bracket of the set of projectable sections is presented in the next definition.

Definition 3.2. Let Zi=(id⁡A,σi∘p,Xi)Z_i=(\operatorname{id}_A,\sigma_i\circ p,X_i) be projectable sections of TATA with σi∈Γ(A)\sigma_i\in\Gamma(A) for i=1,2i=1,2. The Lie bracket of Z1Z_1 and Z2Z_2 is

[Z1,Z2](a):=(a,[σ1,σ2]∘p(a),[X1,X2](a)).(12)[Z_1,Z_2](a):=(a,[\sigma_1,\sigma_2]\circ p(a),[X_1,X_2](a)). \tag*{(12)}

In particular, the Lie brackets of the elements of the basis (11) are given by

[Ei,Ej]=CijkEk,[Ei,Yk]=0,[Yj,Yk]=0.(13)[\mathcal{E}_i,\mathcal{E}_j]=C_{ij}^{k}\mathcal{E}_k,\qquad[\mathcal{E}_i,\mathcal{Y}_k]=0,\qquad[\mathcal{Y}_j,\mathcal{Y}_k]=0. \tag*{(13)}

The anchor map ρ^:TA→TA\widehat{\rho}:TA\to TA is just the projection onto the third factor,

ρ^(a,b,v):=v,(14)\widehat{\rho}(a,b,v):=v, \tag*{(14)}

and can also be defined locally in terms of the basis (11) by

ρ^(Ej)=ρji∂∂xi,ρ^(Yk)=∂∂yk.(15)\widehat{\rho}(\mathcal{E}_j)=\rho_j^i\frac{\partial}{\partial x^i},\qquad\widehat{\rho}(\mathcal{Y}_k)=\frac{\partial}{\partial y^k}. \tag*{(15)}

Therefore, the triple (pA:TA→A,[⋅,⋅]TA,ρ^)(p_A:TA\to A,[\cdot,\cdot]_{TA},\widehat{\rho}), where the bracket [⋅,⋅]TA[\cdot,\cdot]_{TA} is defined over projectable sections by (12) and the anchor map ρ^\widehat{\rho} given by (14), is a Lie algebroid called the prolongation Lie algebroid of AA.

In the following result, we state an important property.

Proposition 3.3. The map p^:TA→A\widehat{p}:TA\to A given by 3.3 is a Lie algebroid morphism over the vector bundle projection p:A→Mp:A\to M, i.e., p^\widehat{p} is a vector bundle morphism satisfying

p^∗∘dTA=dA∘p^∗.\widehat{p}^{*}\circ d_{TA}=d_A\circ\widehat{p}^{*}.

The proof of this proposition can be found in [18].

Lie algebroid morphisms can be lifted to the prolongations of Lie algebroids, as we briefly discuss. Assume that (pi:Ai→Mi,[⋅,⋅]Ai,ρi)(p_i:A_i\to M_i,[\cdot,\cdot]_{A_i},\rho_i), with i=1,2i=1,2, are two Lie algebroids. Assume that Φ:A1→A2\Phi:A_1\to A_2 is a vector bundle morphism over φ:M1→M2\varphi:M_1\to M_2. Then, the map TΦ:TA1→TA2T\Phi:TA_1\to TA_2 given by

TΦ(a,b,v)=(Φ(a),Φ(b),TaΦ(v))(16)T\Phi(a,b,v)=(\Phi(a),\Phi(b),T_a\Phi(v)) \tag*{(16)}

is also a vector bundle morphism over Φ\Phi. Moreover, TΦT\Phi inherits from Φ\Phi the property of being a Lie morphism, as we establish in the following proposition. The proof of this fact and more properties of the map TΦT\Phi can be found in [16].

Proposition 3.4. If Φ:A1→A2\Phi:A_1\to A_2 is a morphism of Lie algebroid then the map TΦ:TA1→TA2T\Phi:TA_1\to TA_2 (16) is a morphism of Lie algebroids,

In the prolongation of a Lie algebroid (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho), there exists map χA:TA→TA\chi_A:TA\to TA with the property of χA∘χA=Id⁡\chi_A\circ\chi_A=\operatorname{Id}, called the canonical involution [9, 18, 12, 16]. This map is given by

χA(a,b,v)=(b,a,v‾),(17)\chi_A(a,b,v)=(b,a,\overline{v}), \tag*{(17)}

where v‾∈TbA\overline{v}\in T_bA such that ρ(a)=Tbp(v‾)\rho(a)=T_b p(\overline{v}) and

v‾(lθ)=v(lθ)+dAθ(a,b),(18)\overline{v}(l_\theta)=v(l_\theta)+d_A\theta(a,b), \tag*{(18)}

with lθl_\theta the linear function (4) defined for every section θ∈Γ(A∗)\theta\in\Gamma(A^*).

Given a section η∈Γ(A)\eta\in\Gamma(A), one can define the complete lift of η\eta as the section ηc∈Γ(TA)\eta^c\in\Gamma(TA) defined by

ηc(a)=χA(η∘p(a),a,Tp(a)η(ρ(a))).(19)\eta^c(a)=\chi_A(\eta\circ p(a),a,T_{p(a)}\eta(\rho(a))). \tag*{(19)}

The prolongation of the tangent Lie algebroid. In this part, we discuss in some detail the prolongation of the tangent Lie algebroid of a smooth manifold. In particular, we are interested in showing how a vector field on the manifold induces a Lie algebroid morphism from the tangent Lie algebroid into its prolongation. This fact plays a crucial role in the proof of the main theorem in § 6.

Let MM be a smooth manifold, and consider its tangent Lie algebroid (τM:TM→M,[⋅,⋅]Lie,id⁡TM)(\tau_M : TM \to M, [\cdot,\cdot]_{\mathrm{Lie}}, \operatorname{id}_{TM}). The prolongation of the tangent Lie algebroid consists of the vector bundle (τM)TM:TTM→TM(\tau_M)^{TM} : TTM \to TM, where

TTM={(v,w,V)∈TM×TM×TTM∣τM(v)=τM(w), TτM(V)=w},TTM = \left\{(v,w,V) \in TM \times TM \times TTM \mid\tau_M(v) = \tau_M(w),\ T\tau_M(V) = w\right\},

and (τM)TM(v,w,V)=v(\tau_M)^{TM}(v,w,V)=v. A projectable section on Γ(TTM)\Gamma(TTM) is of the form Z(v)=(v,X(τM(v),W(v)),Z(v)=(v,X(\tau_M(v),W(v)), with X∈X(M)X \in\mathfrak{X}(M), W∈X(M)W \in\mathfrak{X}(M) satisfying TτMW=X∘τMT_{\tau_M}W=X\circ\tau_M. So, the bracket of two projectable sections Z1,Z2∈Γ(TTM)Z_1,Z_2 \in\Gamma(TTM) is given by

[Z1,Z2](v)=(v,[X1,X2]∘τM(v),[W1,W2]∘(v)).[Z_1,Z_2](v)=\left(v,[X_1,X_2]\circ\tau_M(v),[W_1,W_2]\circ(v)\right).

Finally, the anchor map ρ^TTM:TTM→TTM\widehat{\rho}_{TTM}:TTM\to TTM is the map ρ^TTM(v,w,V)=V\widehat{\rho}_{TTM}(v,w,V)=V. Observe that ρ^TTM\widehat{\rho}_{TTM} is an isomorphism of Lie algebroids with inverse mapping (ρ^TTM)−1:TTM→TTM(\widehat{\rho}_{TTM})^{-1}:TTM\to TTM given by

(ρ^TTM)−1(V)=(τTM(V),TτM(V),V).(\widehat{\rho}_{TTM})^{-1}(V)=\left(\tau_{TM}(V),T\tau_M(V),V\right).

From the previous observations, we have the following consequences:

  • For every section Z∈Γ((TTM)∗)Z \in\Gamma((TTM)^*) there exist a section ς∈Ω(TM)\varsigma\in\Omega(T M) such that (ρ^TTM)∗ς=Z(\widehat{\rho}_{TTM})^*\varsigma=Z.

  • (ρ^TTM)∗∘dTTM=dTTM∘(ρ^TTM)∗(\widehat{\rho}_{TTM})^* \circ d_{TTM}=d_{TTM}\circ(\widehat{\rho}_{TTM})^*.

Let X∈X(M)X \in\mathfrak{X}(M) be a vector field on MM. For every v∈TMv \in TM, the point (X∘τM(v),v,TvX(v))(X\circ\tau_M(v),v,T_vX(v)) belongs to TTMTTM. Thus, every vector field X∈X(M)X \in\mathfrak{X}(M) induces a map CX:TM→TTMC_X:TM\to TTM, given by

CX(v):=(X(τM(v)),v,TτM(v)X(v)).(20)C_X(v):=\left(X(\tau_M(v)),v,T_{\tau_M(v)}X(v)\right). \tag*{(20)}

By straightforward computations, one can show that CXC_X makes the following diagram

TM→CXTTMτM↓↓(τM)TMM→XTM\begin{CD} TM @>{C_X}>> TTM \\ @V{\tau_M}VV @VV{(\tau_M)^{TM}}V \\ M @>{X}>> TM \end{CD}

commutative and CXC_X is linear on each fiber. Furthermore, this mapping has the following property.

Proposition 3.5. For every X∈X(M)X \in\mathfrak{X}(M), the map CXC_X given by (20) is a Lie algebroid morphism over XX, i.e., the pair (CX,X)(C_X,X) is a vector bundle morphism such that

dTM∘CX∗=CX∗∘dTTM.(21)d_{TM}\circ C_X^*=C_X^*\circ d_{TTM}. \tag*{(21)}

Proof. First, we show the compatibility of CXC_X with the anchor maps. This means the commutativity of the diagram

TM→CXTTMρTM=id⁡↓↓ρ^TTMTM→TXTTM\begin{CD} TM @>{C_X}>> TTM \\ @V{\rho_{TM}=\operatorname{id}}VV @VV{\widehat{\rho}_{TTM}}V \\ TM @>{TX}>> TTM \end{CD}

Indeed, for every v∈TMv \in TM we have

ρ^TTM(CX(v))=ρ^TTM(X(τM(v)),v,TτM(v)X(v))=TτM(v)X(v)=TX(ρTM(v)).\widehat{\rho}_{TTM}(C_X(v))=\widehat{\rho}_{TTM}\left(X(\tau_M(v)),v,T_{\tau_M(v)}X(v)\right)=T_{\tau_M(v)}X(v)=TX(\rho_{TM}(v)).

Thus, we have ρ^TTM∘CX=TX\widehat{\rho}_{TTM}\circ C_X=TX. From here, we only need to show that relation (21) holds for 1-sections. We must recall that the vector field X∈X(M)X\in\mathfrak{X}(M) is a smooth map X:M→TMX:M\to TM and then its tangent map TX:TM→TTMTX:TM\to TTM is a morphism on the respective tangent Lie algebroids, which implies that (TX)∗∘dTTM=dTM∘(TX)∗(TX)^*\circ d_{TTM}=d_{TM}\circ(TX)^*. Now, let Z∈Γ((TTM)∗)Z\in\Gamma((TTM)^*) and ς∈Ω(TM)\varsigma\in\Omega(TM) such that (ρ^TTM)∗ς=Z(\widehat{\rho}_{TTM})^*\varsigma=Z. By direct computation, we have

dTM∘CX∗(Z)=dTM∘CX∗((ρ^TTM)∗ς)=dTM∘(ρ^TTM∘CX)∗(ς)=dTM∘(TX)∗(ς),=(TX)∗∘dTTM(ς)=(ρ^TTM∘CX)∗∘dTTM(ς)=CX∗∘(ρ^TTM)∗∘dTTM(ς),=CX∗∘dTM((ρ^TTM)∗ς)=CX∗∘dTM(Z).\begin{aligned} d_{TM}\circ C_X^*(Z) &=d_{TM}\circ C_X^*((\widehat{\rho}_{TTM})^*\varsigma) =d_{TM}\circ(\widehat{\rho}_{TTM}\circ C_X)^*(\varsigma) =d_{TM}\circ(TX)^*(\varsigma),\\ &=(TX)^*\circ d_{TTM}(\varsigma) =(\widehat{\rho}_{TTM}\circ C_X)^*\circ d_{TTM}(\varsigma) =C_X^*\circ(\widehat{\rho}_{TTM})^*\circ d_{TTM}(\varsigma),\\ &=C_X^*\circ d_{TM}((\widehat{\rho}_{TTM})^*\varsigma) =C_X^*\circ d_{TM}(Z). \end{aligned}

This completes the proof.

Semisprays, dynamical variations and Jacobi fields

First, we introduce the fundamental notion of admissible curve on a Lie algebroid (p:A→M,[⋅,⋅]A,ρ)(p:A\to M,[\cdot,\cdot]_A,\rho).

Definition 4.1. Let (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho) be a Lie algebroid. A smooth curve ϑ:I⊂R→A\vartheta:I\subset\mathbb{R}\to A is said to be admissible if

dγdt(t)=ρ(ϑ(t)),(22)\frac{d\gamma}{dt}(t)=\rho(\vartheta(t)), \tag*{(22)}

where γ=p∘ϑ\gamma=p\circ\vartheta. The curve γ\gamma is called the base curve of ϑ\vartheta.

In adapted coordinates (2.1), a curve ϑ(t)=(xi(t),yj(t))\vartheta(t)=(x^i(t),y^j(t)) on AA is admissible if and only if x˙i=ρji(x)yj\dot{x}^i=\rho^i_j(x)y^j. In the literature these curves are also called AA-paths, see [11].

It follows from the definition that the base curve of an admissible curve is completely contained in a leaf of the characteristic foliation. Moreover, every curve on MM contained in a leaf of the characteristic foliation is the base curve of some admissible curve, as we prove in the following lemma.

Lemma 4.2. Let L\mathcal{L} be a leaf of the characteristic foliation of a Lie algebroid (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho) and let γ:I→M\gamma:I\to M be a smooth curve entirely contained in L\mathcal{L}. Then, there exists an admissible curve ϑ:I→A\vartheta:I\to A such that

ρ(ϑ(t))=dγdt.\rho(\vartheta(t))=\frac{d\gamma}{dt}.

Proof. Let AL=p−1(L)A_{\mathcal{L}}=p^{-1}(\mathcal{L}). Denote by pL:AL→Lp_{\mathcal{L}}:A_{\mathcal{L}}\to\mathcal{L} the pull-back of AA via the immersion to ιL:L↪M\iota_{\mathcal{L}}:\mathcal{L}\hookrightarrow M. Since the Lie bracket of section Γ(A)\Gamma(A) induces a Lie bracket on the space of sections of pL:AL→Lp_{\mathcal{L}}:A_{\mathcal{L}}\to\mathcal{L}, it has a transitive Lie algebroid structure with anchor map ρL\rho_{\mathcal{L}} induced by ρ\rho, ρL:=ρ∘ιL\rho_{\mathcal{L}}:=\rho\circ\iota_{\mathcal{L}}. We also consider the Lie algebroid induced on K=ker⁡(ρL)\mathcal{K}=\ker(\rho_{\mathcal{L}}). Hence, we get an exact sequence of vector bundles over L\mathcal{L}

0→K→AL≅TL⊕ker⁡ρL→TL→0.(23)0\to\mathcal{K}\to A_{\mathcal{L}}\cong T\mathcal{L}\oplus\ker\rho_{\mathcal{L}}\to T\mathcal{L}\to0. \tag*{(23)}

Because this exact sequence always splits, there exists a right splitting of ρL\rho_{\mathcal{L}}, that is, a vector bundle map h:TL→ALh:T\mathcal{L}\to A_{\mathcal{L}} such that ρL∘h=id⁡TL\rho_{\mathcal{L}}\circ h=\operatorname{id}_{T\mathcal{L}}. Now, we choose ϑ(t):=h(dγdt(t))\vartheta(t):=h\left(\frac{d\gamma}{dt}(t)\right). By construction, p(ϑ(t))=γ(t)p(\vartheta(t))=\gamma(t) and ϑ\vartheta is an admissible curve,

ρ(ϑ(t))=ρL(h(dγdt(t)))=dγdt(t).\rho(\vartheta(t))=\rho_{\mathcal{L}}\left(h\left(\frac{d\gamma}{dt}(t)\right)\right)=\frac{d\gamma}{dt}(t).

Example 4.3. Consider the tangent bundle τM:TM→M\tau_M:TM\to M with its structure of Lie algebroid described in Example 2.2. Every smooth curve γ:I→M\gamma:I\to M defines an admissible curve on TMTM given by ϑ(t)=γ˙(t)\vartheta(t)=\dot{\gamma}(t).

Let ϑ:I→A\vartheta:I\to A be an admissible curve on the Lie algebroid (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho). A variation of ϑ\vartheta by admissible curves is a smooth map ζ:(−ε,ε)×I→A\zeta:(-\varepsilon,\varepsilon)\times I\to A, such that:

  • (i) ζ(0,t)=ϑ(t)\zeta(0,t)=\vartheta(t),

  • (ii) for any s∈(−ε,ε)s\in(-\varepsilon,\varepsilon), the curve t↦ζ(s,t)t\mapsto\zeta(s,t) is admissible.

The admissible curves are relevant objects on a Lie algebroid because any admissible curve is related to the velocity of its projection onto the base manifold by the anchor map. However, admissibility is a kinematic condition and, by itself, does not prescribe an evolution law. To describe dynamics on a Lie algebroid in terms of admissible curves, one may consider vector fields whose integral curves are admissible. This leads to the notion of semisprays on Lie algebroids, which play a role analogous to that of second-order differential equations on a tangent bundle.

Definition 4.4. A vector field V∈X(A)\mathcal{V}\in\mathcal{X}(A) on the Lie algebroid (p:A→M,[⋅,⋅],ρ)(p:A\to M,[\cdot,\cdot],\rho) is called a semispray if

Tp∘V=ρ,(24)T p\circ\mathcal{V}=\rho, \tag*{(24)}

where Tp:TA→TMT p:TA\to TM denotes the tangent map of the projection. A semispray V∈X(A)\mathcal{V}\in\mathcal{X}(A) is said to be a spray if the vector field V\mathcal{V} satisfies the homogeneous condition:

mk∗V=kV,for k>0,(25)m_k^*\mathcal{V}=k\mathcal{V},\qquad\text{for }k>0, \tag*{(25)}

where mk:A→Am_k:A\to A is the dilation map given by mk(a):=kam_k(a):=ka.

Remark 4.5. Sprays on Lie algebroids, understood as vector fields on AA that lift the anchor, also play a central role in [5], where a spray is used to construct a local Lie groupoid integrating AA. In that setting, the flow of the spray gives rise to a path-exponential map with values in the groupoid, while the exponential map considered in [2] takes values in the leaf of the base manifold.

We denote by ϕt\phi^t the local flow of a semispray V∈X(A)\mathcal{V} \in\mathfrak{X}(A). If ϑ:I→A\vartheta: I \to A is an integral curve of V\mathcal{V}, its projection γ(t)=p∘ϑ(t)\gamma(t) = p \circ\vartheta(t) onto MM will be called an integral curve in the base of V\mathcal{V}. A direct computation gives

dγdt(t)=ddt(p∘ϑ(t))=Tp(dϑdt)=Tp(V(ϑ(t)))=ρ(ϑ(t)).\frac{d\gamma}{dt}(t) = \frac{d}{dt}(p \circ\vartheta(t)) = Tp\left(\frac{d\vartheta}{dt}\right) = Tp(\mathcal{V}(\vartheta(t))) = \rho(\vartheta(t)).

Thus, every integral curve of a semispray is admissible.

In adapted coordinates for AA, a semispray V\mathcal{V} has the form

V(x,y)=yjρji(x)∂∂xi+νk(x,y)∂∂yk.\mathcal{V}(x,y) = y^j\rho_j^i(x)\frac{\partial}{\partial x^i} + \nu^k(x,y)\frac{\partial}{\partial y^k}.

for some local functions νk(x,y)\nu^k(x,y) on AA. If V(x,y)\mathcal{V}(x,y) is a spray, then νk(x,y)=νlnk(x)ylyn\nu^k(x,y) = \nu^k_{ln}(x)y^l y^n. Note that every spray V\mathcal{V} vanishes at the zero section of the vector bundle p:A→Mp : A \to M, that is, V(x,0)=0\mathcal{V}(x,0) = 0. If ϑ(t)=(xi(t),yk(t))\vartheta(t) = (x^i(t),y^k(t)) is an integral curve of V\mathcal{V} in local coordinates, then

x˙i=yjρji(x),\dot{x}^i = y^j\rho_j^i(x),
y˙k=νk(x,y).\dot{y}^k = \nu^k(x,y).

Remark 4.6. The homogeneity condition (4.4) for a spray V\mathcal{V} can be expressed in terms of the Liouville vector field E∈X(A)\mathbf{E} \in\mathfrak{X}(A), which locally has the form

E(x,y)=yk∂∂yk.\mathbf{E}(x,y) = y^k\frac{\partial}{\partial y^k}.

Indeed, a semispray V∈X(A)\mathcal{V} \in\mathfrak{X}(A) is a spray if and only if

[E,V]=V.[\mathbf{E},\mathcal{V}] = \mathcal{V}.

Now we fix an integral curve ϑ:I→A\vartheta: I \to A of a semispray V∈X(A)\mathcal{V} \in\mathfrak{X}(A) and the integral curve in the base γ(t)=p∘ϑ(t)\gamma(t) = p \circ\vartheta(t). A dynamical variation of ϑ\vartheta is a variation by integral curves of V\mathcal{V}, i.e. a smooth map ζ:(−ε,ε)×I→A\zeta: (-\varepsilon,\varepsilon) \times I \to A such that: (i) ζ(0,t)=ϑ(t)\zeta(0,t) = \vartheta(t), and (ii) for any s∈(−ε,ε)s \in(-\varepsilon,\varepsilon), the curve ϑs(t):=ζ(s,t)\vartheta_s(t) := \zeta(s,t) is an integral curve of V\mathcal{V}. In particular, ζ\zeta is a variation of ϑ(t)\vartheta(t) by admissible curves.

Then, we consider the smooth map α:(−ε,ε)×I→M\alpha: (-\varepsilon,\varepsilon) \times I \to M, given by

α:=p∘ζ,\alpha:= p \circ\zeta,

which is a variation by smooth curves of γ(t)\gamma(t). We will call α(s,t)\alpha(s,t) a dynamical variation in the base. Since ζ\zeta is a variation by integral curves of V\mathcal{V}, we have the following property:

ρ(ζ(s,t))=∂α∂t(s,t).(26)\rho(\zeta(s,t)) = \frac{\partial\alpha}{\partial t}(s,t). \tag*{(26)}

Let

Z(t)=∂ζ∂s∣s=0,andJ(t)=∂α∂s∣s=0.(27)Z(t) = \left.\frac{\partial\zeta}{\partial s}\right|_{s=0}, \qquad\text{and} \qquad J(t) = \left.\frac{\partial\alpha}{\partial s}\right|_{s=0}. \tag*{(27)}

be the variational vector fields of variations ζ\zeta and α\alpha respectively. In analogy with what happens in a Riemannian manifold (M,g)(M,g), one can say that the vector field

J(t)=∂α∂s∣s=0J(t)=\left.\frac{\partial\alpha}{\partial s}\right|_{s=0}

is a Jacobi field along the curve in the base γ\gamma, because it is the variational vector field of the dynamical variation in the base α(s,t)\alpha(s,t). However, such a characterization requires the existence of an explicit dynamical variation. It is therefore more convenient to formulate an equivalent definition directly in terms of vector fields along the corresponding integral curves. To state an equivalent notion but independent of dynamical variations, we need to look for a relation between the variational vector fields Z(t)Z(t) and J(t)J(t). In [2], it is proved that the variational vector fields of the dynamical variations ζ\zeta and α\alpha, respectively, are related by the equation

Tρ(Z(t))=Jc(t),(28)T\rho(Z(t))=J^{c}(t), \tag*{(28)}

where Tρ:TA→TTMT\rho:TA\to TTM denotes the tangent map of the anchor ρ\rho and Jc(t)∈Γ(TTM∣γ)J^{c}(t)\in\Gamma(TTM|_{\gamma}) denotes the complete lift of the vector field J(t)∈Γ(TM∣γ)J(t)\in\Gamma(TM|_{\gamma}).

Here, we give the definition of Jacobi fields for semisprays on Lie algebroids presented in [2].

Definition 4.7. Let V∈X(A)\mathcal V\in\mathcal X(A) be a semispray on the Lie algebroid (p:A→M,[⋅,⋅]A,ρ)(p:A\to M,[\cdot,\cdot]_{A},\rho). Fix an integral curve in the base γ:[0,T]→M\gamma:[0,T]\to M corresponding to an integral curve ϑ:[0,T]⊂R→A\vartheta:[0,T]\subset\mathbb R\to A of V\mathcal V, i.e., γ(t)=p(ϑ(t))\gamma(t)=p(\vartheta(t)). A vector field J(t)J(t) along γ\gamma is a Jacobi field if there exists a vector field Z(t)Z(t) along ϑ(t)\vartheta(t) satisfying

(LVZ)(t)=0,(29)(\mathcal L_{\mathcal V}Z)(t)=0, \tag*{(29)}
Tρ(Z(t))=Jc(t).(30)T\rho(Z(t))=J^{c}(t). \tag*{(30)}

Remark 4.8. Condition (4.8) for a vector field Z(t)Z(t) along the integral curve ϑ(t)\vartheta(t) implies that Z(t)Z(t) is the variational vector field of some variation of ϑ(t)\vartheta(t) by integral curves of V\mathcal V which is equivalent to

TφVt(ξ)=Z(t),T\varphi_{\mathcal V}^{t}(\xi)=Z(t),

for some ξ∈Tϑ(0)A\xi\in T_{\vartheta(0)}A and TφVtT\varphi_{\mathcal V}^{t} is the tangent map of φVt(⋅)\varphi_{\mathcal V}^{t}(\cdot) (see [6]). Thus, we define Jacobi fields as variational vector fields of variations by integral curves in the base.

Remark 4.9. Condition (4.9) implies Tp(Z(t))=J(t)Tp(Z(t))=J(t), where Tp:TA→TMTp:TA\to TM denotes the tangent map of p:A→Mp:A\to M.

Example 4.10. Let (M,g)(M,g) be a Riemannian manifold and ∇\nabla the Levi-Civita connection of the metric gg. On the tangent bundle τM:TM→M\tau_M:TM\to M with its canonical structure of Lie algebroid, the Riemannian metric induces the so-called geodesic spray Vg∈X(TM)\mathcal V_g\in\mathcal X(TM) [8], which is given locally by

Vg(x,y)=yi∂∂xi−Γijk(x)yiyj∂∂yk,\mathcal V_g(x,y)=y^i\frac{\partial}{\partial x^i}-\Gamma^k_{ij}(x)y^iy^j\frac{\partial}{\partial y^k},

where Γijk\Gamma^k_{ij} are the Christoffel symbols of ∇\nabla. The integral curves of Vg\mathcal V_g are the tangent lifts of geodesic curves on MM. Hence, if ζ(s,t)\zeta(s,t) is a dynamical variation of Vg\mathcal V_g, then α(s,t)=p∘ζ(s,t)\alpha(s,t)=p\circ\zeta(s,t) is a variation by geodesic curves. Therefore, the vector field

J(t)=∂α∂s(0,t)J(t)=\left.\frac{\partial\alpha}{\partial s}(0,t)\right.

along γ(t)=α(0,t)\gamma(t)=\alpha(0,t) is the classical variational characterization of Jacobi fields in Riemannian geometry. As is well known, these variational vector fields are equivalently characterized as solutions of the classical Jacobi equation

∇γ˙2J+R(J,γ˙)γ˙=0,\nabla_{\dot{\gamma}}^{2}J+R(J,\dot{\gamma})\dot{\gamma}=0,

where RR denotes the curvature of the Levi-Civita connection, [15].

On the other hand, one can verify that

(LVgJc)(t)=0.(\mathcal{L}_{\mathcal{V}_{g}}J^{c})(t)=0.

Since id⁡TM\operatorname{id}_{TM} is the anchor map in the tangent Lie algebroid, the complete lift JcJ^{c} satisfies condition (30) automatically. Hence, the vector field J(t)J(t) is also a Jacobi field along γ(t)\gamma(t) in the sense of Definition 4.7.

The previous example extends naturally from geodesic sprays to arbitrary semisprays on the tangent bundle.

Example 4.11. Consider the tangent Lie algebroid (τM:TM→M,[⋅,⋅]Lie,id⁡TM)(\tau_M:TM\to M,[\cdot,\cdot]_{\mathrm{Lie}},\operatorname{id}_{TM}), where [⋅,⋅]Lie[\cdot,\cdot]_{\mathrm{Lie}} stands for the canonical Lie bracket of vector fields on MM. Let V∈X(TM)\mathcal{V}\in\mathfrak{X}(TM) be a semispray on TMTM. In adapted coordinates, V\mathcal{V} takes the following form

V(x,y)=yi∂∂xi+Vk(x,y)∂∂yk.\mathcal{V}(x,y)=y^{i}\frac{\partial}{\partial x^{i}}+\mathcal{V}^{k}(x,y)\frac{\partial}{\partial y^{k}}.

This kind of vector field is also known as a second-order differential equation on TMTM. Let ϑ:I→TM\vartheta:I\to TM be an integral curve of the semispray V\mathcal{V} and let γ=p∘ϑ\gamma=p\circ\vartheta. By Definition 4.7, a vector field J(t)J(t) along γ\gamma is a Jacobi field if and only if

(LVJc)(t)=0.(\mathcal{L}_{\mathcal{V}}J^{c})(t)=0.

Therefore, in this case, we have recovered the definition of Jacobi fields for second-order differential equations on the tangent bundle given by Cariñena, Gheorghiu and Martínez in [6], (Definition 2.6).

Compatible transverse variations and Lie algebroid morphisms

In the previous section, Jacobi fields for semisprays on Lie algebroids were introduced from a purely dynamical viewpoint as variational vector fields associated with dynamical variations of integral curves of a semispray. To connect this construction to the notion of Jacobi sections introduced by Cariñena, Gheorghiu and Martínez, we require additional geometric ingredients. This section introduces compatible transverse variations, which naturally complement dynamical variations and provide the appropriate framework for establishing the correspondence developed in the following sections.

Here, we consider variations ζ:(−ε,ε)×I→A\zeta:(-\varepsilon,\varepsilon)\times I\to A of admissible curves on AA with the property

(A) the variation α(s,t)=p(ζ(s,t))\alpha(s,t)=p(\zeta(s,t)) of smooth curves on MM is contained in a fixed leaf L\mathcal{L} of the characteristic foliation.

Definition 5.1. Given a variation ζ\zeta by admissible curves satisfying the property (A)(A), a transverse variation to ζ\zeta is a smooth map ϰ:(−ε,ε)×I→A\varkappa: (-\varepsilon,\varepsilon) \times I \to A such that:

(i) p(ζ(s,t))=p(ϰ(s,t))p(\zeta(s,t)) = p(\varkappa(s,t)),

(ii) ρ(ϰ)=∂α∂s\rho(\varkappa) = \frac{\partial\alpha}{\partial s}.

The existence of a transverse variation to a variation ζ\zeta by admissible curves is ensured by any right splitting h:TL→ALh : TL \to A_L of the short exact sequence of Lie algebroids (23). In fact, the smooth mapping ϰ(s,t):=h(∂α∂s)\varkappa(s,t) := h\left(\frac{\partial\alpha}{\partial s}\right) is a transverse variation of ζ(s,t)\zeta(s,t). If ρ\rho is injective, there is a unique transverse variation to a given variation by admissible curves. On the contrary, a variation ζ\zeta admits many transverse variations to it.

Let ϑ:I→A\vartheta: I \to A be an admissible curve and γ=p∘ϑ\gamma= p \circ\vartheta. A smooth curve ς:I→A\varsigma: I \to A such that p∘ς=γp \circ\varsigma= \gamma is called a section over γ\gamma. The set of sections over γ\gamma will be denoted by Γγ(A)\Gamma_\gamma(A). Now, we consider the map Ξϑ:Γγ(A)→Γ(TA∣ϑ)\Xi_\vartheta: \Gamma_\gamma(A) \to\Gamma(TA|_\vartheta) given by

Ξϑ(ς)(t)=ρ^(χA(ς(t),ϑ(t),dςdt(t))).(31)\Xi_\vartheta(\varsigma)(t) = \widehat{\rho}\left(\chi_A\left(\varsigma(t),\vartheta(t),\frac{d\varsigma}{dt}(t)\right)\right). \tag*{(31)}

where ρ^:TA→TA\widehat{\rho} : \mathcal{T}A \to TA denotes the anchor map of the prolongation (14) and χA:TA→TA\chi_A : \mathcal{T}A \to\mathcal{T}A is the canonical involution (17).

The complete lift of a section ς:I→A\varsigma: I \to A over γ\gamma is defined by

ςc(t)=χA(ς(t),ϑ(t),dςdt(t)).(32)\varsigma^c(t) = \chi_A\left(\varsigma(t),\vartheta(t),\frac{d\varsigma}{dt}(t)\right). \tag*{(32)}

The map (31) can be written in terms of the complete lift by Ξϑ(ς)(t)=ρ^(ςc(t))\Xi_\vartheta(\varsigma)(t) = \widehat{\rho}(\varsigma^c(t)). The map (31) is extended to the case of a variation ζ:(−ε,ε)×I→A\zeta: (-\varepsilon,\varepsilon) \times I \to A of admissible curves on AA and α=p∘ζ\alpha= p \circ\zeta. A smooth map c:(−ε,ε)×I→Ac : (-\varepsilon,\varepsilon) \times I \to A such that p∘c=αp \circ c = \alpha is called a section over α\alpha and the set of sections over α\alpha will be denoted by Γα(A)\Gamma_\alpha(A). Then, we define the map Ξζ:Γα(A)→Γ(TA∣ζ)\Xi_\zeta: \Gamma_\alpha(A) \to\Gamma(TA|_\zeta) by

Ξζ(c):=ρ^(χA(c,ζ,∂c∂t)).(33)\Xi_\zeta(c) := \widehat{\rho}\left(\chi_A\left(c,\zeta,\frac{\partial c}{\partial t}\right)\right). \tag*{(33)}

Then, we introduce the tensor of variation of a variation ζ:(−ε,ε)×I→A\zeta: (-\varepsilon,\varepsilon) \times I \to A of admissible curves as the map Δζ:Γα(A)→Γ(TA∣ζ)\Delta_\zeta: \Gamma_\alpha(A) \to\Gamma(TA|_\zeta) given by

Δζ(ϰ):=∂ζ∂s−Ξζ(ϰ).(34)\Delta_\zeta(\varkappa) := \frac{\partial\zeta}{\partial s} - \Xi_\zeta(\varkappa). \tag*{(34)}

Definition 5.2. Given a variation ζ\zeta by admissible curves satisfying the property (A)(A), a transverse variation to ζ\zeta is said to be compatible if

Δζ(ϰ)=0.(35)\Delta_\zeta(\varkappa) = 0. \tag*{(35)}

The following result asserts the existence of a compatible transverse variation.

Proposition 5.3. Given a variation ζ(s,t)\zeta(s,t) by admissible curves with property (A) and ϰ0:(−ε,ε)→A\varkappa_0:(-\varepsilon,\varepsilon)\to A such that

ρ(ϰ0)(s)=∂∂s(p∘ζ)(s,0),\rho(\varkappa_0)(s)=\frac{\partial}{\partial s}(p\circ\zeta)(s,0),

there exists a unique transverse variation ϰ\varkappa of ζ\zeta such that

Δζ(ϰ)=0,ϰ(s,0)=ϰ0(s).(36)\Delta_\zeta(\varkappa)=0,\qquad\varkappa(s,0)=\varkappa_0(s). \tag*{(36)}

Proof. Suppose that the variation ζ\zeta, the mapping α=p∘ζ\alpha=p\circ\zeta and the curve ϰ0\varkappa_0 have the following local representations:

ζ(s,t)=∑j=iryj(s,t)ej(α(s,t)),ϰ0(s)=∑j=irz0j(s)ejα(s,0),\zeta(s,t)=\sum_{j=i}^{r}y^j(s,t)e_j(\alpha(s,t)),\qquad\varkappa_0(s)=\sum_{j=i}^{r}z_0^j(s)e_j\alpha(s,0),
α(s,t)=(x1(s,t),…,xn(s,t)).(37)\alpha(s,t)=(x^1(s,t),\ldots,x^n(s,t)). \tag*{(37)}

By equation (34), the variation ϰ(s,t)=∑j=irzj(s,t)ejα(s,t)\varkappa(s,t)=\sum_{j=i}^{r}z^j(s,t)e_j\alpha(s,t) is compatible if and only if for each s∈(−ε,ε)s\in(-\varepsilon,\varepsilon) the coefficient functions zj(s,t)z^j(s,t) satisfy the following time-dependent linear system of differential equations

∂zk∂t+yiCijkzj=∂yk∂s,for all k=1,…,r,(38)\frac{\partial z^k}{\partial t}+y^iC_{ij}^{k}z^j=\frac{\partial y^k}{\partial s},\qquad\text{for all } k=1,\ldots,r, \tag*{(38)}
zk(s,0)=z0k(s).(39)z^k(s,0)=z_0^k(s). \tag*{(39)}

Thus, the existence and uniqueness of ϰ\varkappa follow from the existence and uniqueness theorem of solutions of linear differential systems. In order to prove that ϰ\varkappa is a transverse variation of ζ\zeta, we need to prove that

∂xk∂s(s,t)=zj(s,t)ρjk(α(s,t)).(40)\frac{\partial x^k}{\partial s}(s,t)=z^j(s,t)\rho_j^k(\alpha(s,t)). \tag*{(40)}

By assumption, ∂xk∂t=yj(s,t)ρjk(α(s,t))\frac{\partial x^k}{\partial t}=y^j(s,t)\rho_j^k(\alpha(s,t)). Then,

∂∂t(∂xk∂s)=∂∂s(∂xk∂t)=∂yj∂sρjk+yj∂ρjk∂xl∂xl∂s,\frac{\partial}{\partial t}\left(\frac{\partial x^k}{\partial s}\right)=\frac{\partial}{\partial s}\left(\frac{\partial x^k}{\partial t}\right)=\frac{\partial y^j}{\partial s}\rho_j^k+y^j\frac{\partial\rho_j^k}{\partial x^l}\frac{\partial x^l}{\partial s},

by equation (38),

=∂zj∂tρjk+ylClrjzrρjk+yj∂ρjk∂xl∂xl∂s,=\frac{\partial z^j}{\partial t}\rho_j^k+y^lC_{lr}^{j}z^r\rho_j^k+y^j\frac{\partial\rho_j^k}{\partial x^l}\frac{\partial x^l}{\partial s},

using the structure equations (3),

=∂∂t(zjρjk)+yj∂ρjk∂xl(∂xl∂s−zjρlk).=\frac{\partial}{\partial t}(z^j\rho_j^k)+y^j\frac{\partial\rho_j^k}{\partial x^l}\left(\frac{\partial x^l}{\partial s}-z^j\rho_l^k\right).

Hence, for all s∈(−ε,ε)s\in(-\varepsilon,\varepsilon) we have the following time-dependent linear system of differential equations

∂∂t(∂xk∂s−zjρjk)=yj∂ρjk∂xl(∂xl∂s−zjρlk),for all k=1,…,n.\frac{\partial}{\partial t}\left(\frac{\partial x^k}{\partial s}-z^j\rho_j^k\right)=y^j\frac{\partial\rho_j^k}{\partial x^l}\left(\frac{\partial x^l}{\partial s}-z^j\rho_l^k\right),\qquad\text{for all } k=1,\ldots,n.

Since ∂xk∂s(s,0)−zj(s,0)ρjk(α(s,0))=0\frac{\partial x^k}{\partial s}(s,0)-z^j(s,0)\rho_j^k(\alpha(s,0))=0, by uniqueness of solution of linear system of differential equations the relation (40) holds. □\square

Let I,I~⊂RI,\widetilde{I} \subset\mathbb{R} be two open intervals. Consider the tangent bundle τI~×I:TI~×TI→I~×I\tau_{\widetilde{I}\times I}:T\widetilde{I}\times TI\to\widetilde{I}\times I with its canonical Lie algebroid structure. Assume that Φ:TI~×TI→A\Phi:T\widetilde{I}\times TI\to A is a vector bundle morphism over the smooth map α:I~×I→M\alpha:\widetilde{I}\times I\to M. Denote the coordinates (s,t)(s,t) in I~×I\widetilde{I}\times I and let ζ(s,t)=Φ(∂∂t∣(s,t))\zeta(s,t)=\Phi\left(\frac{\partial}{\partial t}\big|_{(s,t)}\right) and ϰ(s,t)=Φ(∂∂s∣(s,t))\varkappa(s,t)=\Phi\left(\frac{\partial}{\partial s}\big|_{(s,t)}\right). Then, the vector bundle morphism can be written

Φ(s,t)=ζ(s,t)dt+ϰ(s,t)ds.(41)\Phi(s,t)=\zeta(s,t)dt+\varkappa(s,t)ds. \tag*{(41)}

Since Φ\Phi is vector bundle morphism over α\alpha, we have that

p∘ζ=p∘ϰ=α.p\circ\zeta=p\circ\varkappa=\alpha.

The following question concerns the conditions under which the morphism Φ\Phi (41) is a morphism of Lie algebroids. In the next result, we present sufficient and necessary conditions under which Φ\Phi is a morphism of Lie algebroids. Thus, these conditions are related to variations of admissible curves with compatible transversal conditions.

Theorem 5.4. The morphism Φ\Phi (41) is a morphism of Lie algebroids if and only if ζ(s,t)\zeta(s,t) is a variation of admissible curves satisfying the property (A) and ϰ(s,t)\varkappa(s,t) is a compatible transverse variation of ζ\zeta.

Proof. First, we assume that Φ\Phi is a Lie algebroid morphism. Then, d(α∗f)=Φ∗dAfd(\alpha^{*}f)=\Phi^{*}d_{A}f, for every f∈C∞(M)f\in C^{\infty}(M). A direct computation gives

⟨df,ρ(ζ)⟩=⟨dAf,ζ⟩=⟨Φ∗dAf,∂∂t⟩=⟨d(α∗f),∂∂t⟩=⟨α∗df,∂∂t⟩=⟨df,∂α∂t⟩.\langle df,\rho(\zeta)\rangle=\langle d_{A}f,\zeta\rangle=\left\langle\Phi^{*}d_{A}f,\frac{\partial}{\partial t}\right\rangle=\left\langle d(\alpha^{*}f),\frac{\partial}{\partial t}\right\rangle=\left\langle\alpha^{*}df,\frac{\partial}{\partial t}\right\rangle=\left\langle df,\frac{\partial\alpha}{\partial t}\right\rangle.

Thus, ∂α∂t=ρ(ζ)\frac{\partial\alpha}{\partial t}=\rho(\zeta) and then, ζ(s,t)\zeta(s,t) is a variation of admissible curves. By similar computations, we get that ∂α∂s=ρ(ϰ)\frac{\partial\alpha}{\partial s}=\rho(\varkappa), which implies that ζ(s,t)\zeta(s,t) satisfies the property (A) and ϰ(s,t)\varkappa(s,t) is a transverse variation of ζ\zeta. Now, for every θ∈Γ(A∗)\theta\in\Gamma(A^{*}), we have d(Φ∗θ)=Φ∗(dAθ)d(\Phi^{*}\theta)=\Phi^{*}(d_{A}\theta). By one hand, we have

Φ∗(dAθ)(∂∂s,∂∂t)=dAθ(ϰ,ζ)(42)\Phi^{*}(d_{A}\theta)\left(\frac{\partial}{\partial s},\frac{\partial}{\partial t}\right)=d_{A}\theta(\varkappa,\zeta) \tag*{(42)}

On the other hand,

d(Φ∗θ)(∂∂s,∂∂t)=∂∂s(lθ∘ζ)−∂∂t(lθ∘ϰ)=⟨d(ζ∗lθ),∂∂s⟩−⟨d(ϰ∗lθ),∂∂t⟩=⟨ζ∗d(lθ),∂∂s⟩−⟨ϰ∗d(lθ),∂∂t⟩=∂∂s(lθ)−∂∂t(lθ).(43)\begin{aligned} d(\Phi^{*}\theta)\left(\frac{\partial}{\partial s},\frac{\partial}{\partial t}\right)=\frac{\partial}{\partial s}(l_{\theta}\circ\zeta)-\frac{\partial}{\partial t}(l_{\theta}\circ\varkappa)=\left\langle d(\zeta^{*}l_{\theta}),\frac{\partial}{\partial s}\right\rangle-\left\langle d(\varkappa^{*}l_{\theta}),\frac{\partial}{\partial t}\right\rangle\\ &=\left\langle\zeta^{*}d(l_{\theta}),\frac{\partial}{\partial s}\right\rangle-\left\langle\varkappa^{*}d(l_{\theta}),\frac{\partial}{\partial t}\right\rangle=\frac{\partial}{\partial s}(l_{\theta})-\frac{\partial}{\partial t}(l_{\theta}). \tag*{(43)} \end{aligned}

Combining (42) and 5.13, we get

∂ζ∂s(lθ)=∂ϰ∂t(lθ)+dAθ(ϰ,ζ)\frac{\partial\zeta}{\partial s}(l_{\theta})=\frac{\partial\varkappa}{\partial t}(l_{\theta})+d_{A}\theta(\varkappa,\zeta)

Taking into account (3.13), we obtain

χA(ϰ,ζ,∂ϰ∂t)=(ζ,ϰ,∂ζ∂s).\chi_{A}\left(\varkappa,\zeta,\frac{\partial\varkappa}{\partial t}\right)=\left(\zeta,\varkappa,\frac{\partial\zeta}{\partial s}\right).

By (3.9) and (5.3), Ξζ(ϰ)=∂ζ∂s\Xi_{\zeta}(\varkappa)=\frac{\partial\zeta}{\partial s}. Hence, Δζ(ϰ)=0\Delta_{\zeta}(\varkappa)=0 and then ϰ\varkappa is a compatible transverse variation of ζ\zeta. The converse of this result also holds because all the previous arguments are reversible.

Under the assumptions of Theorem 5.4, we assume that ζ\zeta is a variation of the admissible curve ϑ:I→A\vartheta:I\to A, i.e., ζ(0,t)=ϑ(t)\zeta(0,t)=\vartheta(t). Let κ∈Γγ(A)\kappa\in\Gamma_\gamma(A), with γ=p∘ϑ\gamma=p\circ\vartheta, given by κ(t)=κ(0,t)\kappa(t)=\kappa(0,t). Then, the infinitesimal variation Z(t)Z(t) of the variations ζ(s,t)\zeta(s,t) is given by

Z(t)=∂ζ∂s(0,t)=Ξϑ(κ)(t)=ρ^(κϑc(t)).(44)Z(t)=\frac{\partial\zeta}{\partial s}(0,t)=\Xi_{\vartheta}(\kappa)(t)=\widehat{\rho}(\kappa^{c}_{\vartheta}(t)). \tag*{(44)}

The previous theorem shows that compatible transverse variations admit an intrinsic characterization in terms of Lie algebroid morphisms. This characterization will allow us to reinterpret the variational definition of Jacobi sections introduced by Cariñena, Gheorghiu and Martínez entirely in terms of compatible transverse variations.

Remark 5.5. The idea that a pair (ζ,κ)(\zeta,\kappa) defining a homotopy of AA-paths should be viewed as a morphism TI×TI→ATI\times TI\to A already appears in [11 §1.3], and a characterization of Lie algebroid morphisms in terms of the canonical involution is given in [16 Prop. 2]. Theorem 5.4 makes this precise in the setting needed here: it gives a coordinate-free proof that a vector bundle morphism Φ=ζ dt+κ ds\Phi=\zeta\,dt+\kappa\,ds is a Lie algebroid morphism if and only if ζ\zeta is a variation by admissible curves and κ\kappa is a compatible transverse variation of ζ\zeta, with compatibility expressed through the variation tensor Δζ\Delta_\zeta.

One can generate variations of admissible curves with compatible transverse variation by choosing a section σ∈Γ(A)\sigma\in\Gamma(A) and a smooth curve ν:I~→L⊂M\nu:\widetilde{I}\to L\subset M, where LL is a leaf of the characteristic foliation of ρ\rho. Following Cariñena, Gheorghiu and Martínez in [6], we say that the flow of a section σ\sigma is the pair (ϕt,φt)(\phi_t,\varphi_t) where ϕt\phi_t denotes the local flow of the vector field ρ(σ)\rho(\sigma) and φt\varphi_t the flow of ρ^(σc)\widehat{\rho}(\sigma^c). Here, σc\sigma^c is the complete lift of σ\sigma defined in (3.14). The flow (ϕt,φt)(\phi_t,\varphi_t) of a section σ∈Γ(A)\sigma\in\Gamma(A) has the following properties:

  • The pair (ϕt,φt)(\phi_t,\varphi_t) is a morphism of Lie algebroids.

  • σ∘ϕt=φt∘σ\sigma\circ\phi_t=\varphi_t\circ\sigma.

  • The vector fields ρ(σ)\rho(\sigma) and ρ^(σc)\widehat{\rho}(\sigma^c) are pp-related.

We define ζ:I~×I→A\zeta:\widetilde{I}\times I\to A by

ζ(s,t):=σ(ϕt(ν(s))).(45)\zeta(s,t):=\sigma(\phi_t(\nu(s))). \tag*{(45)}

where II is the common interval of definition of the flow ϕt\phi_t throughout the points ν(s)\nu(s) with s∈I~s\in\widetilde{I}. Observe that α(s,t)=p∘ζ(s,t)=ϕt(ν(s))\alpha(s,t)=p\circ\zeta(s,t)=\phi_t(\nu(s)). Then

∂α∂t(s,t)=ρ(σ(ϕt(ν(s))))=ρ(ζ(s,t)).\frac{\partial\alpha}{\partial t}(s,t)=\rho(\sigma(\phi_t(\nu(s))))=\rho(\zeta(s,t)).

Therefore, ζ(s,t)=σ(ϕt(ν(s)))\zeta(s,t)=\sigma(\phi_t(\nu(s))) is a variation of admissible curves satisfying the property (A). By Lemma 4.2, there exists a smooth curve κ0:I~→A\kappa_0:\widetilde{I}\to A such that p∘κ0=νp\circ\kappa_0=\nu and dνds=ρ(κ0)\frac{d\nu}{ds}=\rho(\kappa_0). By Proposition 5.3 there exists a compatible transverse variation κ:I~×I→A\kappa:\widetilde{I}\times I\to A such that κ(s,0)=κ0(s)\kappa(s,0)=\kappa_0(s).

The following result states an explicit formula for a compatible transverse variation of ζ\zeta.

Lemma 5.6. The map κ:I~×I→A\kappa:\widetilde{I}\times I\to A given by

κ(s,t)=ϕt(κ0(s)),\kappa(s,t)=\phi_t(\kappa_0(s)),

is a compatible transverse variation of ζ\zeta (5.15).

Proof. First, we shall prove that κ\kappa is a transverse variation of ζ\zeta. Using that ϕt\phi_t is a morphism of vector bundles over φt\varphi_t, we get

p(κ(s,t))=p(ϕt(κ0(s)))=φt(p(κ0(s)))=φt(ν(s))=p(ζ(s,t)).(46)p(\kappa(s,t))=p(\phi_t(\kappa_0(s)))=\varphi_t(p(\kappa_0(s)))=\varphi_t(\nu(s))=p(\zeta(s,t)). \tag*{(46)}

Since (ϕt,φt)(\phi_t,\varphi_t) is a morphism of Lie algebroids, ϕt∗dAf=dA∘φt∗f\phi_t^*d_Af=d_A\circ\varphi_t^*f, for all f∈C∞(M)f\in C^\infty(M). Then,

⟨df,ρ(κ(s,t))⟩=⟨dAf,ϕt(κ0(s))⟩=⟨ϕt∗dAf,κ0(s)⟩=⟨dA∘φt∗f,κ0(s)⟩,=⟨d∘φt∗f,ρ(κ0(s))⟩=⟨φt∗∘df,dνds(s)⟩=⟨df,∂∂s(φt(ν(s)))⟩.\begin{aligned} \langle df,\rho(\kappa(s,t))\rangle &=\langle d_Af,\phi_t(\kappa_0(s))\rangle =\langle\phi_t^*d_Af,\kappa_0(s)\rangle =\langle d_A\circ\varphi_t^*f,\kappa_0(s)\rangle,\\ &=\langle d\circ\varphi_t^*f,\rho(\kappa_0(s))\rangle =\langle\varphi_t^*\circ df,\frac{d\nu}{ds}(s)\rangle =\left\langle df,\frac{\partial}{\partial s}(\varphi_t(\nu(s)))\right\rangle. \end{aligned}

Due to ff is arbitrary, ∂∂s(φt(ν(s)))=ρ(κ(s,t))\frac{\partial}{\partial s}(\varphi_t(\nu(s)))=\rho(\kappa(s,t)) which implies that κ\kappa is a transverse variation of ζ\zeta.

Now, we shall prove that κ\kappa is compatible. By direct computation, we get

∂κ∂t=ρ^(σc(κ(s,t))),\frac{\partial\kappa}{\partial t}=\widehat{\rho}(\sigma^c(\kappa(s,t))),

by (3.14) we have,

∂κ∂t=ρ^(χA(ζ(s,t),κ(s,t),∂ζ∂s(s,t))).\frac{\partial\kappa}{\partial t} =\widehat{\rho}\left(\chi_A\left(\zeta(s,t),\kappa(s,t),\frac{\partial\zeta}{\partial s}(s,t)\right)\right).

This equality implies

χA(ζ(s,t),κ(s,t),∂ζ∂s(s,t))=(κ(s,t),ζ(s,t),∂κ∂t(s,t)).\chi_A\left(\zeta(s,t),\kappa(s,t),\frac{\partial\zeta}{\partial s}(s,t)\right) =\left(\kappa(s,t),\zeta(s,t),\frac{\partial\kappa}{\partial t}(s,t)\right).

Since χA∘χA=id\chi_A\circ\chi_A=\mathrm{id}, we obtain

χA(κ(s,t),ζ(s,t),∂κ∂t(s,t))=(ζ(s,t),κ(s,t),∂ζ∂s(s,t)).\chi_A\left(\kappa(s,t),\zeta(s,t),\frac{\partial\kappa}{\partial t}(s,t)\right) =\left(\zeta(s,t),\kappa(s,t),\frac{\partial\zeta}{\partial s}(s,t)\right).

It follows from here that ∂ζ∂s(s,t)=Ξζ(κ)\frac{\partial\zeta}{\partial s}(s,t)=\Xi_\zeta(\kappa) which is equivalent to Δζ(κ)=0\Delta_\zeta(\kappa)=0. This proves that κ\kappa is compatible.

Furthermore, if Φ:TI~×TI→A\Phi:T\widetilde{I}\times TI\to A is a morphism of Lie algebroids over the map α:I~×I→M\alpha:\widetilde{I}\times I\to M if given by

Φ(s,t)=σ(α(s,t))dt+κ(s,t)ds.(47)\Phi(s,t)=\sigma(\alpha(s,t))dt+\kappa(s,t)ds. \tag*{(47)}

with σ∈Γ(A)\sigma\in\Gamma(A), then the map α(s,t)\alpha(s,t) represents a variation of integral curves of the vector field ρ(σ)\rho(\sigma),

∂α∂t(s,t)=ρ(σ)(α(s,t)).\frac{\partial\alpha}{\partial t}(s,t)=\rho(\sigma)(\alpha(s,t)).

That is, we can write α(s,t)=φt(ν(s))\alpha(s,t)=\varphi_t(\nu(s)) for some smooth curve ν:I~→L⊂M\nu:\widetilde{I}\to L\subset M.

The previous discussions motivate the following definition, which is also presented in [6].

Definition 5.7. A 1-parameter family of integral curves of a section σ∈Γ(A)\sigma\in\Gamma(A) is a Lie algebroid morphism Φ:TI~×TI→A\Phi: T\widetilde{I} \times TI \to A over α:I~×I→M\alpha: \widetilde{I} \times I \to M of the form

Φ(s,t)=σ(α(s,t)) dt+κ(s,t) ds.(48)\Phi(s,t)=\sigma(\alpha(s,t))\,dt+\kappa(s,t)\,ds. \tag*{(48)}

The section κ:I→A\kappa: I \to A given by κ(t)=κ(0,t)\kappa(t)=\kappa(0,t) is called the infinitesimal variation of the 1-parameter family Φ\Phi.

Jacobi sections for SODEs on Lie algebroids

Once we have established the geometric framework of compatible transverse variations, we are ready to revisit the notion of a Jacobi section introduced by Cariñena, Gheorghiu and Martínez [6].

Let (p:A→M,[⋅,⋅]A,ρ)(p : A \to M,[\cdot,\cdot]_{A},\rho) and consider the prolongation of the Lie algebroid pA:TA→Ap_{A} : \mathcal{T}A \to A given by (6) and (7).

Definition 6.1. A second-order differential equation (SODE) of the Lie algebroid AA is a section S∈Γ(TA)S \in\Gamma(\mathcal{T}A) in the prolongation given by

S(a)=(a,a,V(a)).(49)S(a)=(a,a,\mathcal{V}(a)). \tag*{(49)}

By an integral curve of a second-order differential equation SS is an integral curve of the vector field ρ^(S)=V∈X(A)\widehat{\rho}(S)=\mathcal{V}\in\mathfrak{X}(A).

There is a close relation between semisprays and second-order differential equations on a Lie algebroid. It follows from the definition of the prolongation of a Lie algebroid that if S∈Γ(TA)S \in\Gamma(\mathcal{T}A) is a SODE, then, the vector field ρ^(S)=V\widehat{\rho}(S)=\mathcal{V} is a semispray on AA, i.e., Tp∘V=ρTp\circ\mathcal{V}=\rho. Conversely, every semispray V∈X(A)\mathcal{V}\in\mathfrak{X}(A) induces a SODE by formula (49). Hence, an integral curve of a SODE on AA is precisely an integral curve of a semispray on AA.

Remark 6.2. An equivalent definition to Definition 6.1 is given in [18] in terms of the vertical endomorphism on the prolongation and the Euler section. This definition is analogous to the one given for the notion of second-order differential equation on the tangent bundle of a smooth manifold. Throughout this paper we adopt Definition 6.1 because it establishes the correspondence between semisprays on Lie algebroids and second-order differential equations on the prolongation that will be used below.

Let S∈Γ(TA)S \in\Gamma(\mathcal{T}A) be a SODE on AA and suppose that Θ:I~×I→TA\Theta: \widetilde{I}\times I\to\mathcal{T}A is a 1-parameter family of integral curves of SS. Now, we recall the notion of Jacobi section for second-order differential equations due to Cariñena, Gheorghiu and Martínez, which was formulated in terms of infinitesimal variations of 1-parameter families of integral curves.

Definition 6.3. Given an integral curve ϑ:I→A\vartheta: I\to A of a SODE SS on the Lie algebroid AA, a section κ\kappa of AA along γ=p∘ϑ\gamma=p\circ\vartheta is a Jacobi section along ϑ\vartheta if its complete lift κϑc(t)\kappa_{\vartheta}^{c}(t) of κ\kappa is the infinitesimal variation of Θ:I~×I→TA\Theta:\widetilde{I}\times I\to\mathcal{T}A, a 1-parameter family of integral curves of SS.

Remark 6.4. Cariñena, Gheorghiu and Martínez also characterized Jacobi sections as solutions of a second-order equation, ∇∇W+Φ(W)=0\nabla\nabla W+\Phi(W)=0, where the dynamical covariant derivative ∇\nabla and the Jacobi endomorphism Φ\Phi of the semispray are defined by means of the prolongation [6 Thm. 3.7]. In this paper we focus on their variational characterization, which we reformulate in terms of variations of integral curves on AA.

One of the purposes of this paper is to define Jacobi sections in an equivalent form to Definition 6.3 but in terms of dynamical variations of integral curves and compatible transverse variations. To achieve this goal, we provide a geometric interpretation of 1-parameter families of integral curves of SODEs.

If Θ:T(I~×I)→TA\Theta:T(\widetilde{I}\times I)\to TA is a 1-parameter family of integral curves of SS, then, Θ\Theta has the following form

Θ=S(ζ(s,t))dt+B(s,t)ds.(50)\Theta=S(\zeta(s,t))dt+B(s,t)ds. \tag*{(50)}

Then, Θ:T(I~×I)→TA\Theta:T(\widetilde{I}\times I)\to TA is a morphism of Lie algebroids over the map ζ:I~×I→A\zeta:\widetilde{I}\times I\to A. Besides, ζ\zeta is a variation by integral curves of the semisprays V=ρ^(S)V=\hat{\rho}(S) or a dynamical variation of the integral curve ϑ(t):=ζ(0,t)\vartheta(t):=\zeta(0,t). The coefficient B(s,t)B(s,t) in (6.2) has the following form

B(s,t)=(ζ(s,t),X(s,t),∂ζ∂s(s,t)).(51)B(s,t)=\left(\zeta(s,t),\mathcal{X}(s,t),\frac{\partial\zeta}{\partial s}(s,t)\right). \tag*{(51)}

This easily follows from the following identities: pA(B(s,t))=pA(S(ζ(s,t)))=ζ(s,t)p_A(B(s,t))=p_A(S(\zeta(s,t)))=\zeta(s,t) and ρ^(B(s,t))=Θ(∂∂s)=∂ζ∂s\hat{\rho}(B(s,t))=\Theta\left(\frac{\partial}{\partial s}\right)=\frac{\partial\zeta}{\partial s}. Moreover, since, B(s,t)∈TAB(s,t)\in TA, we also have p(X(s,t))=p(ζ(s,t))=α(s,t)p(\mathcal{X}(s,t))=p(\zeta(s,t))=\alpha(s,t) and ρ(X)=Tp(∂ζ∂s(s,t))=∂α∂s(s,t)\rho(\mathcal{X})=T p\left(\frac{\partial\zeta}{\partial s}(s,t)\right)=\frac{\partial\alpha}{\partial s}(s,t). This means that X\mathcal{X} is a transverse variation of the dynamical variation ζ\zeta of ϑ\vartheta. Then, according to the Definition 6.3, the section κ(t)=X(0,t)\kappa(t)=\mathcal{X}(0,t) along γ=p∘ϑ\gamma=p\circ\vartheta is a Jacobi section because of

B(0,t)=(ϑ(t),κ(t),Z(t))=χA(X,ϑ,dκdt(t))=κϑc(t).B(0,t)=(\vartheta(t),\kappa(t),Z(t))=\chi_A\left(\mathcal{X},\vartheta,\frac{d\kappa}{dt}(t)\right)=\kappa_{\vartheta}^{c}(t).

The previous discussions give rise to the relations between 1-parameter families of integral curves of a SODE and Jacobi sections with dynamical variations and their compatible transverse variations. In fact, we have the following result.

Theorem 6.5. Let S∈Γ(TA)S\in\Gamma(TA) be a SODE on AA and ϑ:I→A\vartheta:I\to A an integral curve of SS. A section κ\kappa of AA along γ=p∘ϑ\gamma=p\circ\vartheta is a Jacobi section along ϑ\vartheta if and only if there exists a dynamical variation ζ:I~×I→A\zeta:\widetilde{I}\times I\to A of ϑ\vartheta satisfying property (A) and a compatible transverse variation X:I~×I→A\mathcal{X}:\widetilde{I}\times I\to A of ζ\zeta such that X(0,t)=κ(t)\mathcal{X}(0,t)=\kappa(t).

Proof. First, we assume that κ\kappa is a Jacobi section along ϑ\vartheta. Then, the complete lift of κ(t)\kappa(t) is the infinitesimal variation of Θ:T(I~×I)→TA\Theta:T(\widetilde{I}\times I)\to TA, a 1-parameter family of integral curves of SS. Assume that Θ\Theta is given by (6.2) with coefficient BB of the form (6.3). Recall the map p^:TA→A\hat{p}:TA\to A given by (3.3) and define Φ:I~×I→A\Phi:\widetilde{I}\times I\to A by Φ:=p^∘Θ\Phi:=\hat{p}\circ\Theta. By Proposition 2.7 and Proposition 3.3, Φ\Phi is a Lie algebroid morphism over α=p∘ζ\alpha=p\circ\zeta. A direct computation gives, Φ(∂∂t)=p^(S(ζ(s,t)))=ζ(s,t)\Phi\left(\frac{\partial}{\partial t}\right)=\hat{p}(S(\zeta(s,t)))=\zeta(s,t) and Φ(∂∂s)=p^(B(s,t))=X(s,t)\Phi\left(\frac{\partial}{\partial s}\right)=\hat{p}(B(s,t))=\mathcal{X}(s,t). Hence,

Φ=ζ(s,t)dt+X(s,t)ds.(52)\Phi=\zeta(s,t)dt+\mathcal{X}(s,t)ds. \tag*{(52)}

By Theorem 5.4, ζ:I~×I→A\zeta:\widetilde{I}\times I\to A is a variation by admissible curves satisfying property (A) and X:I~×I→A\mathcal{X}:\widetilde{I}\times I\to A is a compatible transverse variation such that X(0,t)=κ(t)\mathcal{X}(0,t)=\kappa(t). Moreover, Θ(∂∂t)=S(ζ(s,t))\Theta\left(\frac{\partial}{\partial t}\right)=S(\zeta(s,t)) and p^\hat{p} maps it to ∂ζ∂t\frac{\partial\zeta}{\partial t}, so ∂ζ∂t=V(ζ)\frac{\partial\zeta}{\partial t}=V(\zeta). Hence, for each s∈I~s\in\widetilde{I}, the curve ζ(s,⋅)\zeta(s,\cdot) is an integral curve of V=ρ^(S)V=\hat{\rho}(S), which implies that ζ:I~×I→A\zeta:\widetilde{I}\times I\to A is a dynamical variation of ϑ\vartheta.

Now, we shall prove the converse. Assume that ζ:I~×I→A\zeta:\widetilde{I}\times I\to A is a dynamical variation of ϑ\vartheta satisfying property (A) and ϰ:I~×I→A\varkappa:\widetilde{I}\times I\to A is a compatible transverse variation such that ϰ(0,t)=κ(t)\varkappa(0,t)=\kappa(t), and construct the maps Φ\Phi by (6.4). It follows from Theorem 5.4 that Φ\Phi is a Lie algebroid morphism. Consider the mapping TΦ:TT(I~×I)→TAT\Phi:T T(\widetilde{I}\times I)\to T A defined over the corresponding prolongations of Lie algebroids by formula (3.11). By Proposition 3.4, TΦT\Phi is also a Lie algebroid morphism. Then, we choose the vector field X=∂∂t∈X(I~×I)X=\frac{\partial}{\partial t}\in\mathfrak{X}(\widetilde{I}\times I) and consider the map C∂∂t:T(I~×I)→TT(I~×I)C_{\frac{\partial}{\partial t}}:T(\widetilde{I}\times I)\to T T(\widetilde{I}\times I) by means of formula (3.15). C∂∂tC_{\frac{\partial}{\partial t}} is a Lie algebroid morphism by Proposition 3.5. Thus, the map Θ:T(I~×I)→TA\Theta:T(\widetilde{I}\times I)\to T A given by

Θ:=TΦ∘C∂∂t\Theta:=T\Phi\circ C_{\frac{\partial}{\partial t}}

is also a Lie algebroid morphism. Evaluating the morphism Θ\Theta at the section ∂∂t\frac{\partial}{\partial t}, we get

Θ(∂∂t)=TΦ∘C∂/∂t(∂∂t)=(ζ(s,t),ϰ(s,t),T(Φ(∂/∂t))(∂∂t))=(ζ(s,t),ϰ(s,t),∂ζ∂t)=S(ζ(s,t)).\begin{aligned} \Theta\left(\frac{\partial}{\partial t}\right)=T\Phi\circ C_{\partial/\partial t}\left(\frac{\partial}{\partial t}\right)=\left(\zeta(s,t),\varkappa(s,t),T\left(\Phi(\partial/\partial t)\right)\left(\frac{\partial}{\partial t}\right)\right) \\ &=\left(\zeta(s,t),\varkappa(s,t),\frac{\partial\zeta}{\partial t}\right)=S(\zeta(s,t)). \end{aligned}

Analogously, evaluating the morphism Θ\Theta at the section ∂∂s\frac{\partial}{\partial s}, we obtain

Θ(∂∂s)=(Φ(∂∂t),Φ(∂∂s),Tζ(∂∂s))=(ζ(s,t),ϰ(s,t),∂ζ∂s)=B(s,t).\Theta\left(\frac{\partial}{\partial s}\right)=\left(\Phi\left(\frac{\partial}{\partial t}\right),\Phi\left(\frac{\partial}{\partial s}\right),T\zeta\left(\frac{\partial}{\partial s}\right)\right)=\left(\zeta(s,t),\varkappa(s,t),\frac{\partial\zeta}{\partial s}\right)=B(s,t).

This implies that Θ\Theta is a 1-parameter family of integral curves of SS. Then, κ(t)\kappa(t) is a Jacobi section along ϑ\vartheta because

B(0,t)=(ϑ(t),κ(t),Z(t))=χA(ϰ,ϑ,dκdt(t))=κϑg(t).B(0,t)=(\vartheta(t),\kappa(t),Z(t))=\chi_A\left(\varkappa,\vartheta,\frac{d\kappa}{dt}(t)\right)=\kappa^g_{\vartheta}(t).

From Theorem 6.5, one can conclude that dynamical variations with compatible transverse variations provide the natural variational mechanism underlying Jacobi sections associated with semisprays on Lie algebroids. With this result, dynamical variations and compatible transverse variations become the fundamental geometric objects, while the prolongation appears naturally as the geometric framework supporting the Definition 6.3, given originally in [6] by Cariñena et al.

In consequence, the original definition of Cariñena, Gheorghiu and Martínez admits an equivalent formulation, expressed entirely in terms of dynamical variations and compatible transverse variations. Thus, in view of Theorem 6.5, Jacobi sections for semisprays (or SODEs) on Lie algebroids can equivalently be defined as follows.

Definition 6.6. Given an integral curve ϑ:I→A\vartheta:I\to A of a SODE SS on the Lie algebroid AA, a section κ\kappa of AA along γ=p∘ϑ\gamma=p\circ\vartheta is a Jacobi section along ϑ\vartheta if there exist a dynamical variation ζ:I~×I→A\zeta:\widetilde{I}\times I\to A of ϑ\vartheta satisfying property (A) and a compatible transverse variation ϰ:I~×I→A\varkappa:\widetilde{I}\times I\to A of ζ\zeta such that ϰ(0,t)=κ(t)\varkappa(0,t)=\kappa(t).

Remark 6.7. It follows from Theorem 6.5 and [6 Thm. 3.7] that the Jacobi sections of Definition 6.6 are exactly the solutions of the Jacobi equation ∇∇κ+Φ(κ)=0\nabla\nabla\kappa+\Phi(\kappa)=0 of the semispray. In contrast with the Jacobi equation of [2, 3], which is formulated through the curvature of a linear connection and therefore requires a spray, this characterization holds for arbitrary semisprays.

The anchor correspondence between Jacobi sections and Jacobi fields

In this section, we clarify the relationship between Jacobi fields presented in Section 4, and the notion of Jacobi sections revised in Section 6. The variational interpretation of Jacobi sections established in Theorem 6.5 will provide us with the appropriate framework to relate both notions. Indeed, while Jacobi fields arise from arbitrary dynamical variations of integral curves of a semispray, Jacobi sections correspond to those dynamical variations satisfying Property (A) and admitting compatible transverse variations. This additional geometric structure allows one to compare both notions through the anchor map. In this section, we establish this correspondence and characterize the circumstances under which a Jacobi field can be realized by a Jacobi section.

Our first result states the relation between Jacobi sections and Jacobi fields.

Theorem 7.1. Let V∈X(A)\mathcal{V} \in\mathfrak{X}(A) be a semispray on a Lie algebroid AA, ϑ:I→A\vartheta: I \to A an integral curve of V\mathcal{V} and γ=p∘ϑ\gamma= p \circ\vartheta. If κ(t)\kappa(t) is a Jacobi section along ϑ\vartheta, then

J(t)=ρ(κ(t))(53)J(t) = \rho(\kappa(t)) \tag*{(53)}

is a Jacobi field of V\mathcal{V} along γ\gamma.

Proof. Suppose that κ(t)\kappa(t) is a Jacobi section along ϑ\vartheta. By Theorem 6.5, there exists a dynamical variation ζ:I~×I→A\zeta: \widetilde{I} \times I \to A of ϑ\vartheta satisfying property (A) and a compatible transverse variation ϰ:I~×I→A\varkappa: \widetilde{I} \times I \to A of ζ\zeta such that ϰ(0,t)=κ(t)\varkappa(0,t) = \kappa(t). Then, α(s,t)=p∘ζ(s,t)\alpha(s,t) = p \circ\zeta(s,t) is a variation of γ\gamma by integral curves in the base. Thus, J(t)=∂α∂s(0,t)J(t) = \frac{\partial\alpha}{\partial s}(0,t) is a Jacobi field along of γ\gamma. Since ϰ\varkappa is a transverse variation, we have ∂α∂s(s,t)=ρ(ϰ(s,t))\frac{\partial\alpha}{\partial s}(s,t) = \rho(\varkappa(s,t)). Therefore,

ρ(κ(t))=ρ(ϰ(0,t))=∂α∂s(0,t)=J(t).\rho(\kappa(t)) = \rho(\varkappa(0,t)) = \frac{\partial\alpha}{\partial s}(0,t) = J(t).

In summary, the previous theorem shows that a Jacobi section of a Lie algebroid is projected under the anchor map into a Jacobi field. However, the converse of the theorem holds under an additional hypothesis about the dynamical variations.

Theorem 7.2. Let V∈X(A)\mathcal{V} \in\mathfrak{X}(A) be a semispray on a Lie algebroid AA, ϑ:I→A\vartheta: I \to A an integral curve of V\mathcal{V} and γ=p∘ϑ\gamma= p \circ\vartheta. Suppose that J(t)J(t) is a Jacobi field arising from a variation α:I~×I→M\alpha: \widetilde{I} \times I \to M by integral curves in the base whose image is completely contained in a leaf L⊂M\mathcal{L} \subset M of the characteristic foliation, then there exists a Jacobi section κ(t)\kappa(t) along ϑ\vartheta such that

J(t)=ρ(κ(t)).(54)J(t) = \rho(\kappa(t)). \tag*{(54)}

Proof. Recall (Definition 4.7 and Remark 4.8) that Jacobi fields are variational vector fields of dynamical variations in the base. By assumption, J(t)=∂α∂s(0,t)J(t) = \frac{\partial\alpha}{\partial s}(0,t) for a variation α(s,t)=p∘ζ(s,t)\alpha(s,t) = p \circ\zeta(s,t), where ζ:I~×I→A\zeta: \widetilde{I} \times I \to A is a dynamical variation of ϑ\vartheta. Since the image of α\alpha is completely contained in a leaf L\mathcal{L} of the characteristic foliation, ζ\zeta satisfies the property (A). By Lemma 4.2, there is a curve ϰ0:I~→A\varkappa_0:\widetilde{I}\to A such that ρ(ϰ0(s))=∂α∂s(s,0)\rho(\varkappa_0(s))=\frac{\partial\alpha}{\partial s}(s,0). By Proposition 5.3, there exists a compatible transverse variation ϰ:I~×I→A\varkappa:\widetilde{I}\times I\to A such that ϰ(s,0)=ϰ0(s)\varkappa(s,0)=\varkappa_0(s). By Theorem 6.5, ϰ(t)=ϰ(0,t)\varkappa(t)=\varkappa(0,t) is a Jacobi section along ϑ\vartheta. Finally, we have

ρ(ϰ(t))=ρ(ϰ(0,t))=∂ϑ∂s(0,t)=J(t).\rho(\varkappa(t))=\rho(\varkappa(0,t))=\frac{\partial\vartheta}{\partial s}(0,t)=J(t).

The theorems above completely describe the correspondence between Jacobi fields and Jacobi sections for arbitrary semisprays. We must emphasize that Jacobi sections always project through the anchor map to Jacobi fields. But not every Jacobi field comes, via the anchor, from some Jacobi sections. This phenomenon comes from the fact that the Jacobi fields arise from arbitrary dynamical variations of integral curves, whereas Jacobi sections arise from the additional geometric structure of dynamical variations satisfying Property (A) together with compatible transverse variations. Hence, the Jacobi field for semisprays on Lie algebroids presented in Section 4 is a geometric structure that considers a broader class of dynamical variations than the Jacobi sections.

We conclude this section by comparing with the spray case treated in [2]. In this paper, inspired by [3], Jacobi sections were defined as the sections along an integral curve satisfying a Jacobi equation written through the curvature of a linear connection. This is possible because a spray determines a unique torsionless linear connection whose geodesics are the integral curves of the spray [2 Thm. 13]. Without homogeneity, a semispray induces no linear connection, and this approach is no longer available.

For sprays, Jacobi sections can also be obtained from both dynamical variations and compatible transverse variations [2 Prop. 17], [3 Prop. 3.17]. In Theorem 6.5, we show that this variational description does not depend on homogeneity: it holds for the Jacobi sections of [6] of an arbitrary semispray. Homogeneity is only needed to write the Jacobi equation through a linear connection (compare Remark 6.7).

An illustration of the correspondence between Jacobi sections and Jacobi fields

In this section we present examples illustrating the variational framework developed in the previous sections. Our main goal is to show how compatible transverse variations arise naturally in concrete situations and how they provide the geometric mechanism relating dynamical variations, Jacobi sections and Jacobi fields. In particular, the examples illustrate the correspondence established in Section 7 and complement the abstract theory with explicit constructions.

Let M=RM=\mathbb{R} and the abelian Lie algebra g=R2=span⁡{e1,e2}\mathfrak{g}=\mathbb{R}^{2}=\operatorname{span}\{e_{1},e_{2}\}. Consider the infinitesimal action of g\mathfrak{g} on R\mathbb{R} defined by

Ψ(e1)=x∂∂x,Ψ(e2)=0.\Psi(e_{1})=x\frac{\partial}{\partial x},\qquad\Psi(e_{2})=0.

In this example, we study the action Lie algebroid given by the vector bundle

p:M×R2→Mp:M\times\mathbb{R}^{2}\to M

with coordinates (x,y,z)(x,y,z) where the elements on the fiber through xx is given by a=ye1+ze2a=ye_1+ze_2. The anchor map ρ:M×R2→TM\rho:M\times\mathbb{R}^2\to TM is

ρ(x,y,z)=xy∂∂x.\rho(x,y,z)=xy\frac{\partial}{\partial x}.

The bracket of sections can be defined in terms of the global basis sections σ1(x)=(x,1,0)\sigma_1(x)=(x,1,0), σ2(x)=(x,0,1)\sigma_2(x)=(x,0,1) by [σ1,σ2]=0[\sigma_1,\sigma_2]=0, and extended to arbitrary sections by bilinearity, skew-symmetry and the Leibniz rule. The characteristic foliation has three components: L−=(−∞,0)L_- = (-\infty,0), L0={0}L_0=\{0\}, and L+=(0,∞)L_+=(0,\infty). Since ρ(σ1)=x∂∂x\rho(\sigma_1)=x\frac{\partial}{\partial x} and ρ(σ2)=0\rho(\sigma_2)=0, the semisprays have the following form

V(x,y,z)=xy∂∂x+ν1(x,y,z)∂∂y+ν2(x,y,z)∂∂z.\mathcal{V}(x,y,z)=xy\frac{\partial}{\partial x}+\nu_1(x,y,z)\frac{\partial}{\partial y}+\nu_2(x,y,z)\frac{\partial}{\partial z}.

For simplicity, we consider the semispray corresponding to ν1=ν2≡0\nu_1=\nu_2\equiv0. The corresponding integral curves satisfy the following system

x˙=xy,y˙=0,z˙=0.\dot{x}=xy,\qquad\dot{y}=0,\qquad\dot{z}=0.

Hence, the solution through the point (x0,y0,z0)(x_0,y_0,z_0) at t=0t=0 is the curve ϑ:R→M×R2\vartheta:\mathbb{R}\to M\times\mathbb{R}^2

ϑ(t)=(x0ey0t,y0,z0)\vartheta(t)=(x_0e^{y_0t},y_0,z_0)

with γ(t)=x0ey0t\gamma(t)=x_0e^{y_0t}. If x0>0x_0>0, then the dynamical variation ζ:R×R→M×R2\zeta:\mathbb{R}\times\mathbb{R}\to M\times\mathbb{R}^2 of ϑ\vartheta

ζ(s,t)=(x0e(y0+rs)t+us,y0+rs,z0+vs),\zeta(s,t)=\left(x_0e^{(y_0+rs)t+us},y_0+rs,z_0+vs\right),

where r,u,vr,u,v are real constants, satisfies the property (A), because the image α(s,t)=x0e(y0+rs)t+us\alpha(s,t)=x_0e^{(y_0+rs)t+us} is contained in L+L_+. By Theorem 7.2, the Jacobi field along γ\gamma

J(t)=∂α∂s(0,t)=(x0(rt+u)ey0t)∂∂x,J(t)=\frac{\partial\alpha}{\partial s}(0,t)=\left(x_0(rt+u)e^{y_0t}\right)\frac{\partial}{\partial x},

is in the image of the anchor map. On the other hand, the map ϰ:R×R→M×R2\varkappa:\mathbb{R}\times\mathbb{R}\to M\times\mathbb{R}^2 defined by

ϰ(s,t)=(x0e(y0+rs)t+us,rt+u,vt)\varkappa(s,t)=\left(x_0e^{(y_0+rs)t+us},rt+u,vt\right)

is a transverse variation of ζ\zeta. Using equation (38), we can see that ϰ\varkappa is also compatible. So, κ(t)=(x0ey0t,u+rt,vt)\kappa(t)=(x_0e^{y_0t},u+rt,vt) is a Jacobi section along ϑ\vartheta and

ρ(κ(t))=J(t).\rho(\kappa(t))=J(t).

Therefore, this example provides an explicit illustration of the correspondence established in Theorem 7.2: the Jacobi field J(t)J(t) arises as the anchor image of the Jacobi section κ(t)\kappa(t), obtained from a compatible transverse variation.

To emphasize that Property (A) is an essential hypothesis in Theorem 7.2, we now consider a dynamical variation that does not satisfy this property. The map

ζ~(s,t)=(se(y0+rs)t,y0+rs,z0+us)\widetilde{\zeta}(s,t)=\left(se^{(y_0+rs)t},y_0+rs,z_0+us\right)

is a dynamical variation of the integral curve ϑ~(t)=(0,y0,z0)\tilde{\vartheta}(t) = (0, y_0, z_0). Observe that the image of α~(s,t)=se(y0+rs)t\tilde{\alpha}(s,t) = s e^{(y_0+rs)t} is not contained in any single characteristic leaf, since it intersects L−L_{-}, L0L_0, and L+L_{+}. The Jacobi field associated with this variation is

J~(t)=ey0t∂∂x\tilde{J}(t) = e^{y_0t}\frac{\partial}{\partial x}

Since γ~(t)=p∘ϑ~(t)≡0\tilde{\gamma}(t) = p \circ\tilde{\vartheta}(t) \equiv0, any section along γ~\tilde{\gamma} has the following form ς(t)=(0,ς1(t),ς2(t))\varsigma(t) = (0, \varsigma_1(t), \varsigma_2(t)). Hence, there is no Jacobi section κ(t)\kappa(t) along ϑ~(t)\tilde{\vartheta}(t) such that ρ(κ(t))=J~(t)\rho(\kappa(t)) = \tilde{J}(t). Although α~\tilde{\alpha} generates a Jacobi field, its image is not contained in a single characteristic leaf, and therefore no compatible transverse variation can exist, and hence no Jacobi section can be associated with it.

Acknowledgements. SECIHTI supported this research under the Grant CF-2023-G-279.

References

  1. [1]Anahory Simões, A., Marrero, J. C., Martín de Diego, D. (2022). Jacobi fields in nonholonomic mechanics. Journal of Physics A: Mathematical and Theoretical, 55(4), 045202.arxiv.org/abs/2004.10457
  2. [2]Avendaño-Camacho, M., Kama-Mamani, J. (2026). Jacobi fields, sprays and connections on Lie algebroids . Bol. Soc. Mat. Mex. 32, 18. https://doi.org/10.1007/s40590-025-00845-2
  3. [3]Boucetta, M.(2011). Riemannian geometry of Lie algebroids. Journal of the Egyptian Mathematical Society. 19(1-2), 57–70.arxiv.org/abs/0806.3522
  4. [4]Bucătaru, I. (1999). The Jacobi fields for a spray on the tangent bundle. Novi Sad J. Math, 29(3), 69–78.
  5. [5]Cabrera, A., Mărcuț, I., Salazar, M. A. (2020). On local integration of Lie brackets. Journal für die reine und angewandte Mathematik (CrelIes Journal), 2020 (760), 267–293.
  6. [6]Cariñena, J.F., Gheorghiu, I., Martínez, E. (2015). Jacobi fields for second-order differential equations on Lie algebroids. In: Conference Publications, vol. 2015, pp. 213–222. doi: 10.3934/proc.2015.0213DOI
  7. [7]Cariñena, J. F., da Costa, J. M. N., Santos, P. (2007). Reduction of Lagrangian mechanics on Lie algebroids. Journal of Geometry and Physics, 57(3), 977–990.DOI
  8. [8]Chavel, I.: Riemannian geometry: a modern introduction. Cambridge University Press (1995)
  9. [9]Cortés, J., de León, M., Marrero, J. C., Martín de Diego, D., Martínez, E. (2006). A survey of Lagrangian mechanics and control on Lie algebroids and groupoids. International Journal of Geometric Methods in Modern Physics, 3(3), 509–558.DOI
  10. [10]Crainic, M., Loja Fernandes, R., Mărcuț, I. (2021). Lectures on Poisson Geometry, vol. 217. American Mathematical Society.DOI
  11. [11]Crainic, M., Fernandes, R. L. (2003). Integrability of Lie brackets. Annals of Mathematics, 157(2), 575-620. https://doi.org/10.4007/annals.2003.157.575
  12. [12]de León, M., Marrero, J. C., Martínez, E. (2005). Lagrangian submanifolds and dynamics on Lie algebroids. Journal of Physics A: Mathematical and General, 38(24), R241–R308.DOI
  13. [13]Jóźwikowski, M. (2013). Jacobi vector fields for Lagrangian systems on algebroids. Int. J. Geom. Methods Mod. Phys., 10(05), 1350011
  14. [14]Kosmann-Schwarzbach, Y., Laurent-Gengoux, C., Weinstein, A. (2008). Modular Classes of Lie Algebroid Morphisms. Transformation Groups, 13, 727–755 . https://doi.org/10.1007/s00031-008-9032-y
  15. [15]Lee, J.M.: Introduction to Riemannian manifolds. Springer 2 (2018).DOI
  16. [16]Martínez, E. (2008). Variational calculus on Lie algebroids, ESAIM: Control, Optimisation and Calculus of Variations, 14 no. 2, pp. 356-380. doi: 10.1051/cocv:2007056.DOI
  17. [17]Martínez, E., Mestdag, T., Sarlet, W. (2002). Lie algebroid structures and Lagrangian systems on affine bundles. Journal of Geometry and Physics, 44(1), 70-95.DOI
  18. [18]Martínez, E. (2001). Lagrangian Mechanics on Lie Algebroids, Acta Appl. Math. 67 , 295–320.DOI
  19. [19]P. Stefan, Accessible sets, orbits, and foliations with singularities, Proc. London Math. Soc. 29 (1974), 699–713.DOI
  20. [20]H. J. Sussmann, Orbits of families of vector fields and integrability of distributions, Trans. Amer. Math. Soc. 180 (1973), 171–188.DOI

Paper details

vlnc:W_562vvs8j
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents