Subriemannian geometry, a variational approach (Q946809)

In Riemannian geometry, geodesics appear as critical points of the Riemann length functional. There are two important methods with which such critical points can be found: Hamilton-Jacobi and Hamiltonian formalisms. In Riemannian spaces both methods are equivalent. Recall that a sub-Riemannian space is defined as a triple \((M,D,g) \) where \(M\) is a \(C^{\infty }\)-differentiable manifold, \(D\) is a bracket generating (horizontal) distribution of constant rank \(k\) on \(M\), and \(g\) is a \(C^{\infty }\)-metric tensor on \(D\). A vector field \(X\) on \(M\) is called horizontal if \(X\in D\); a smooth curve \(\gamma \) in \(M\) is called horizontal if its velocity vector field \(\gamma ^{\prime }\) is horizontal. The length of a smooth horizontal curve \(\gamma \) is defined by \(g\) in the usual way. The Carnot-Carathéodory distance between a pair of points \(x,y\) of a sub-Riemannian space is defined as the greatest lower bound of lengths of smooth horizontal curves joining \(x\) to \(y\). The authors study Hamilton-Jacobi and Hamiltonian methods in sub-Riemannian spaces. A smooth real-valued solution \(S(\tau ,x) ,\tau \in \mathbb{R}\), \(x\in M\), of the Hamilton-Jacobi equation \(\frac{\partial S}{ \partial \tau }+H(\nabla S) =0\) with the initial condition \( S|_{\tau =0}=S_{0}\) is called an action function. One main result (Theorem 3.4) characterizes minimizers of the energy functional \( I(\varphi) =\int_{0}^{\tau }| \varphi ^{\prime }| _{g}dt\) in terms of the action functions. Namely, let \(S:[0,\tau] \times M\rightarrow \mathbb{R}\) be an action function. Then \( \varphi \) is a minimizer of the energy functional if and only if \(\varphi ^{\prime }=\nabla _{h}S|_{\varphi(t)}\), where \(\nabla _{h}\) is the horizontal gradient. In addition, the action function allows to calculate the minimum value of the energy functional. As a corollary, the authors prove that \ each horizontal curve \(\varphi \) satisfying \(\varphi ^{\prime }=\nabla _{h}S|_{\varphi(t)}\) is locally length minimizing. Moreover, if \(\varphi :[ 0,\tau] \rightarrow M\) is a horizontal length-minimizing curve joining \(p\) to \(q\) (and satisfying \( \varphi ^{\prime }=\nabla _{h}S|_{\varphi (t)}\)), then \(\ell (\varphi)=\) \(\sqrt{2\tau S(\tau ,q)-S(0,p)}\). A normal geodesic from \(x\in M\) to \(y\in M\) is defined as a solution of the Hamiltonian system: \(x^{\prime i}(s)=\frac{\partial H}{\partial p_{i}}\), \(p^{\prime i}(s)=-\frac{\partial H}{\partial x^{i}}\), where \(H\) is the Hamiltonian. The authors' main result connected with Hamiltonian formalism (Theorem 4.8) states that each normal geodesic is in fact a horizontal locally length-minimizing curve. The converse statement is not true, as shown earlier by \textit{R. Montgomery} [SIAM J. Control Optimization 32, No.~6, 1605--1620 (1994; Zbl 0816.49019)].