Super-optimization for a class of four-block problems

A super-optimization method applicable to a reasonably large class of genuine four-block problems is outlined via a dimension-reducing technique based on the so-called equalizer principle. It successively reduces the original problem to super-optimizations problems with steadily fewer singular values, until finally all freedom is exhausted. Each step amounts to an ordinary H¿ optimization of the largest remaining singular value, and a subsequent removal of this singular value. The proposed procedure allows computation of superoptimal solutions for a large class of model-matching problems, including all one-block problems and a significant portion of genuine two- and four-block problems