An elementary proof of the Pontryagin maximum principle (Q2022434)

The subject is the standard control problem for systems of ODE \begin{gather*} \begin{aligned} \text{minimize} & \quad \ell_0(x(0), x(T)) \\ \text{subject to} & \quad x'(t) = f(t, x(t), u(t)) \quad (u(t) \in U) \end{aligned} \\ \ell_j(x(0), x(T)) \le 0 \quad j = 1,\dots ,l \, , \quad \ell_j(x(0), x(T)) = 0 \quad j = l+1,\dots ,r \, . \end{gather*} If \(H(t, x, p, u) = \langle p, f(t, x, u) \rangle\) and \((\bar x(t), \bar y(t))\) is an optimal trajectory/control then there exist numbers \(\lambda_j\) not all zero such that \(\lambda_j \ge 0\), \(\lambda_j \ell_j(\bar x(0), \bar x(T)) = 0\) for \(j \le l\) and a function \(\bar p(t)\) satisfying the adjoint equation and transversality conditions \[ \begin{aligned} - \bar p'(t) & = H'_x(t, \bar x(t), \bar p(t), \bar u(t)) \\ (\bar p(0), - \bar p(T)) & = \lambda_0 \ell'_0(\bar x(0), \bar x(T)) + \dots + \lambda_r \ell'_r(\bar x(0), \bar x(T)) \end{aligned} \] and such that the maximum principle \[ H(t, \bar x(t), \bar p(t), \bar u(t)) = \max_{u \in U} H(t, \bar x(t), \bar p(t), u) \] is satisfied. The proof is based on the author's \textit{optimality alternative,} where \(f(x)\) is a function defined in a complete metric space \(X\) and \(\bar x \in M \subseteq X\) is a local minimum constrained by \(M\). If \(\varphi(x) \ge 0\), \(\varphi(\bar x) = 0\) and \(f, \varphi\) are suitably restricted then either there exists \(\rho > 0\) such that \(\rho f(x) + \varphi(x)\) has an unconditional local minimum at \(\bar x\) or there exists a sequence \(x_m \to \bar x\) such that \(\varphi(x) + d(x, x_m)/m\) has a global minimum at \(x_m\). Application of the alternative is rather technical and requires various modifications and approximations of the original problem, one of which involves a discretization of the control space where \(U\) is replaced by \(U_k(t) = \{ \bar u(t), u_1(t), \dots, u_k(t) \}\).