Hi Dave, Im happy to hear from you. I hope that all is going well with you. It appears that you have an interesting application. Ill do my best to help. For disk problems |k - h| < r the best thing is the power method which you used in the original hinf code. But the way of implementing should not be the same. I explain. In your code hinf, the power method was used, this involved storing a matrix (n^2 slots) as well as matrix times vector operations. (order n^2 ). One can get around both things! The trick is to use the fft. The algorithm in DualExtremal.m is an implementation of this. The result: no matrices are stored, and the iteration is order n log(n). Another thing: the original spectral factorization (Bauer method) was also inefficient (similar problems with storage and speed). The function WilsonSF.m does the scalar spectral factorization it only takes fft basically. Also there is HarrisSF.m, which is slightly more robust (deals better with zeros on the circle) and can be generalized to matrix spectral factorization if needed (based on Newtons method). ======================== There is another algorithm (newtons) for solving Hinfinity problems, but they require software capable of handling symbolic calculations. Although tremendously powerful, it has 2 drawbacks: it requires storing and inverting large matrices, and, it is sensible to the initial guess (convergence is local). To run problems with large grids (1024 +) or with vector valued functions, I recommend the power method as the practical choice. What is the level of complication in the performance you have in mind? is it always disk-type? =========================== Some time ago a couple of people in GBritain asked me for advice to port the power method etc to matlab (we lost contact with them). Luckily I kept some files that should help with the conversion. (they are a subset of our package, but enough to get the ball rolling). Try anonymous ftp to osiris, go to: cd pub/helton/Schwartz there you will find the following files: SampleMathematicaRun.m An example of a session for solving disk problem. Utility.dec92.m Utility mathematica functions (spectral facto etc) Dave.feb22 Your message README This message DualExtremal.dec92.m The power method for disk problems SampleConversion An example of conversion matlab - mathematica The converted code is Threfethen's CFapproximation algo.