Learning and search are analyzed in hybrid automata that support continuous and differentiable change as in Lagrangian mechanics. Discrete approximations of that change are investigated in simple domains such as gridworlds where Legendre transforms and Lie algebras are considered.