Sequential linear-quadratic method for differential games with air combat applications (Q1810977)

The authors study the computation of saddle points of non linear-quadratic zero-sum differential games, and apply their proposed numerical method to air combat simulations. The main idea stems from the result, proved in the paper, that the zero variation of the saddle point of the original problem satisfies the first order necessary conditions for an associated linear-quadratic problem. Starting from an initial candidate saddle point, the authors then compute the solution to the associated problem; update the candidiate by adding a small multiple of the above solution; and go on sequentially until the norm of the solution to the updated associated problem is small enough. An important point is that if the solution to the Riccati equation associated to the linear-quadratic problem does not have a finite escape time, then the solution to that problem can be expressed in state feedback form. Thus the authors are able to deal with problems where the opponent does not possibly use the same model and/or solution as the first player (which is supposed to use the authors' algorithm). The proposed numerical method converges in the proposed applications, although as the authors stress out one cannot be sure that the limit point is the true solution to the original problem. The authors develop in great detail quite a general model of air combat, to which they apply their proposed method. The simulated solutions are interesting and nontrivial, and the algorithm performs well (although the authors remind again that convergence is still not formally established).