The saddle point property in Hamiltonian systems (Q914145)

We consider optimal control problems which are autonomous except for a time dependent discount factor. Pontrjagin's maximum principle states that the optimal state and costate trajectories are solutions to an autonomous system of differential equations in (modified) Hamiltonian form. Under the assumption that a stationary solution of this system exists and that it has a stable manifold of maximal dimension, we derive local curvature properties of the Hamiltonian function at the stationary state. In particular, we show that a generalized curvature matrix is negative definite provided that the discount rate is zero or that the stationary state satisfies a symmetry condition. A geometrical interpretation of this result is given in terms of the convex-concave structure of the Hamiltonian function. Two examples show that the symmetry condition cannot be omitted from the theorem but that it is satisfied in some problems arising in optimal economic growth theory.