Dynamic system methods for solving mixed linear matrix inequalities and linear vector inequalities and equalities (Q972166): Difference between revisions

The author determines numerically a solution of the matrix inequality system \[ Ax \geq b, Cx = d, N(x)\geq 0\tag{1} \] where \(N(x)\) is a linear combination of symmetric \((n,n)\)-matrices with real coefficients \(x_i, i=1,\dots,m\). The first inequality is given with respect to the coordinatewise order in \(\mathbb R^m\) and the last inequality holds with respect to the cone of all positive semidefinite matrices well-known from semi-definite programming. Using orthogonal projections to these cones the problem (1) is reformulated as least square problem and solved by a continuous gradient flow algorithm (in the paper is used a Runge-Kutta method, \texttt{Matlab} ODE 45) and by a gradient algorithm with fixed step size naturally with a lot of iterations. Convergence statements are proven straight forward under the existence assumption of a solution of (1) and illustrative numerical examples are given. This approach also works for non strict inequalities in the contrary to interior point methods. The given approach approximates solutions at the boundary of the solution set.