General techniques for constructing variational integrators (Q693191)

In this survey paper two techniques for constructing discrete variational integrators suited to dynamical systems that can be described by a Lagrangian function are studied. In the first technique, that leads to a Galerkin approach, the action associated to the Lagrangian function \( L( q(t), \dot{q}(t))\) in the time step \([0,h]\) is approximated by a numerical quadrature \( L_S (g ) := h \sum_{i=1}^s b_i L (g(c_i h), \dot{g}(c_i h))\) with \(s\) nodes \(c_i\) in \([0,1]\) and weights \(b_i\) and then the extremization is carried out in a suitable space of interpolatory functions. A complete theoretical study of this technique that includes the order and preservation properties is presented and to illustrate it a discretization of the Lagrangian of a free rigid body is given. The second technique relies on the discretization of the exact discrete Lagrangian in terms of Jacobi's solution of the Hamilton-Jacobi equation. Now the discretization requires not only a numerical quadrature but also a one-step numerical integrator for differential equations in a similar way to shooting methods for boundary-value problems. A study of some properties of the integrator that are inherited by the variational integrator is given as well some implementation issues. Finally, the technique is illustrated with numerical experiments of Kepler's and pendulum problems.