Linear-quadratic optimization problem and the frequency theorem for periodic systems. II (Q1814379): Difference between revisions

[For part I see the author, Sib. Math. J. 27, No. 4 (1990); translation from Sib. Math. Zh. 27, No. 4(158), 181-200 (1986; Zbl 0648.93037).] We prove propositions stated in part I concerning the `` frequency theorem''. Basically, we consider a topological problem associated with the initial optimization problem. The optimization problem was reduced in part I to the investigation of some ``associated'' linear \(T\)-periodic Hamiltonian system. The original problem has a solution if and only if the associated Hamiltonian system meets two conditions: total instability and nonoscillation. Total instability follows from Pontryagin's maximum principle, while being nonoscillatory is analogous to Jacobi's condition in the calculus of variations. We study the structure of the set of Hamiltonians having these properties. We show that in the Banach space of all Hamiltonians \(H(t)=H(t)^*=H(t+T)\), the set of Hamiltonians corresponding totally unstable systems factors into a countable number of domains \({\mathfrak N}_ k\); we determine the properties that uniquely distinguish the Hamiltonians in each such region. In Section 2 we also prove that the regions \({\mathfrak N}_ k\) \((k=0,\pm1,\pm2,\dots)\) of total instability are affinely homeomorphic and that equations with \(H(\cdot)\in{\mathfrak N}_ 0\) are nonoscillatory, while equations with \(H(\cdot)\in {\mathfrak N}_ k\), \(k\neq 0\), are oscillatory. Thus, solvability of the original problem is equivalent to the condition \(H(\cdot)\in{\mathfrak N}_ 0\), where \(H(\cdot)\) is the Hamiltonian of the associated system. At the same time, it turns out that solvability of the equations of the maximum principle (which are satisfied by an optimal process) is equivalent to the condition \(H(\cdot)\in\bigcup_ k{\mathfrak N}_ k\backslash{\mathcal P}\), where \({\mathcal P}\) is some ``surface'' in the space of Hamiltonians. In this sense, there is a large ``gap'' between the necessary conditions implied by the maximum principle and the necessary and sufficient conditions of Theorem 2 of part I. The presence of nonempty sets \({\mathfrak N}_ k\), \(k\neq 0\), implies that the condition of being nonoscillatory, which is absent in stationary problems, is essential in the periodic case --- in contrast to the stationary case, it does not follow from total instability. In Section 3 we show that for \(H(\cdot)\in{\mathfrak N}_ k\) the index \(k\) of a region of total instability is the same as the Maslow index of some appropriate closed curve in the Lagrangian Grassmannian.