Continuous and discrete Clebsch variational principles (Q1022432)

The authors deal with variational principles defined by the action (or cost for the optimal control problem) \[ S=\int l[\mathbf{\xi}(t)]dt, \eqno{(1)} \] whose Lagrangian (or cost functional) \(l: V\to \mathbb{R}\) is defined on vectors \(\mathbf{\xi}\) in a vector space \(V\), subject to a condition imposed by a velocity map from the vector space \(V\) to the tangent space \(T_{Q}M\) of a manifold \(M\) at the point \(\mathbf{Q}\;\) \(\;\mathcal{L}_{\xi}:V\times M \to T_{{Q}}M\). The velocity map \(\mathcal{L}_{\xi}\) introduces the dynamics, \[ \dot{\mathbf{Q}}(t)=\mathcal{L}_{\xi}\mathbf{Q}(t), \eqno{(2)} \] where \(\mathbf{\xi}\in V\) and \( \dot{\mathbf{Q}} \in T_{{Q}}M \) is tangent to the curve \(\mathbf{Q}(t)\) in the manifold \(M\). Such variational principles arise in two different contexts: \begin{itemize}\item[1.] The optimal control context, in which one seeks solutions for \(\mathbf{Q}(t)\) governed by the dynamics (2) that control the motion along a curve in an interval \(0\leq t \leq T,\) so as to minimize the cost in (1) for a given cost functional \(l[\mathbf{\xi}]\). \item[2.] The Hamilton's principle context, in which stationarity \(\delta S=0\) of the action in (1) implies dynamical equations for \(\mathbf{\xi}\) subject to the constraint imposed by the velocity map (2). The Clebsch approach enforces (2) through a Lagrange multiplier term in the action or cost. Doing so produces dynamical equations for \(\mathbf{Q}\) and for the Lagrange multiplier \(\mathbf{P}\) in terms of \(\mathbf{\xi}\), together with a formula for \(\mathbf{\xi}\) given in terms of \(\mathbf{Q}\) and \(\mathbf{P}\). The Lagrangian multiplier \(\mathbf{P}\) is also the canonically conjugate momentum for \(\mathbf{Q}\) in the corresponding Hamiltonian formulation, and the formula for \(\mathbf{\xi}\) in terms of \(\mathbf{Q}\) and \(\mathbf{P}\) has special significance in the Hamiltonian frame-work. It is shown that \(\mathbf{Q}\) and \(\mathbf{P}\) may be eliminated from the resulting Clebsch equations, if and only if the velocity map \(\mathcal{L}_{\xi}\) is a Lie algebra action on \(\mathbf{Q} \in M.\) The Clebsch framework for velocity maps thus has a clear connection with the theory of Euler-Poincaré reduction; namely, the equations obtained are the Euler-Poincaré equations on the dual of the Lie algebra. For the continuous time case where the velocity map is assumed to be a Lie derivative, this connection is not unexpected. Examples included the finite-dimensional rigid body equation, and two infinite-dimensional examples: the singular solutions of the EPDiff equation and the incompressible Euler equation. In the EPDiff example, the Clebsch framework derives the singular solutions as a family of momentum maps. Also it is shown how discrete Clebsch variational principles can be used to produce numerical methods for Clebsch equations.\end{itemize}