Duality in optimal impulse control (Q2069787)

This paper builds on previous works of the author and colleagues [\textit{F. Dufour} et al., Int. J. Control 95, No. 1, 173--186 (2022; Zbl 1482.49038); \textit{A. Piunovskiy} et al., SIAM J. Control Optim. 57, No. 4, 2720--2752 (2019; Zbl 1420.49038); \textit{A. Piunovskiy} and \textit{Y. Zhang}, J. Math. Anal. Appl. 496, No. 2, Article ID 124817, 17 p. (2021; Zbl 1458.49029); \textit{A. Piunovskiy} and \textit{Y. Zhang}, J. Math. Anal. Appl. 499, No. 2, Article ID 125070, 46 p. (2021; Zbl 1461.93225)] on Lagrangian duality for optimal impulse control problems (i.e., optimal control with singular or measure-valued controls). Formulating this problem in terms of the impulse times and vectors, a dynamic programming principle can be applied to derive a Bellman equation, which is connected to a dual linear program. In this work, two such dual linear programs are introduced in an abstract setting and their solvability is shown. It is verified that the Bellman function provides such a solution in either case and that there is no primal-dual gap. Finally, the construction of an optimal control strategy from a solution of the primal-dual pairs is discussed. The theoretical results are then illustrated via a benchmark SIR model.