On multivariable matrix spectral factorization method (Q2673018): Difference between revisions

The algorithm proposed in [\textit{G. Janashia} and \textit{E. Lagvilava}, Stud. Math. 137, No. 1, 93--100 (1999; Zbl 0960.47013)] for spectral factorization of a positive define matrix is generalized to the multivariate case. Appropriate Hardy spaces \(H^p(\mathbb{T}^N)\) are defined by recursion on the number of variables. This is tailored to the algorithm so that also the outer spectral factor can be computed recursively. An iterative procedure is proposed to find an outer matrix valued spectral factor for the first variable keeping the other variables as parameters. By a similar iteration, the other variables are introduced one by one, which results in an update of the previous spectral factor by right multiplying it with unitary matrix functions that depend on the remaining variables until all the variables are taken into account. Convergence for the iterative procedure is proved, so that in practice a finite number of iterations can be used for each variable. There are several applications, one of them is Granger's causality estimation in neuroscience, which is explained in an appendix (cf. [\textit{C. W. J. Granger}, Econometrica 37, 424--438 (1969; Zbl 1366.91115)]).