Discrete Optimal Control: Second Order Optimality Conditions (Q3147458)

The authors deal with the discrete optimal control problem \[ \begin{cases} \text{minimize} \;&{\mathcal F}(x,u)=K(x_0,x_{n+1}) + \sum_{k=0}^{N}g(k,x_{k+1},u_k),\\ \text{subject to}& \Delta x_k=f(k,x_{k+1},u_k),\;k\in\{0,\dots,N\},\quad \varphi(x_0,x_{N+1})=0, \end{cases} \tag{P} \] where the functions appearing in this problem satisfy essentially the same assumptions as in the continuous-time counterpart of (P). Using the mathematical programming approach, second order necessary and sufficient optimality conditions for (P) are derived. These conditions are formulated via nonnegativity and positivity of discrete quadratic functionals associated with (P). Then, conditions for positivity and nonnegativity of these quadratic functionals are formulated in terms of the nonexistence of conjugate intervals and of the existence of a symmetric solution of the associated Riccati matrix equation. Open problems and direction for further investigation are also discussed.