An improved μ lower bound via adaptive power iteration

Many robustness analysis problems can be reduced to that of computing the structured singular value, μ. The general mixed μ problem (i.e., including real uncertainty) is now known to be NP complete. This emphasizes the importance of computing good bounds. It is shown that the mixed μ lower bound can be addressed by means of a power algorithm, which is computationally inexpensive. A detailed study of this power algorithm is presented, and several modifications are developed, including a mixed power/inverse-power iteration. This leads to an adaptive power iteration that significantly enhances performance over the original scheme