Factorization of multivariate positive Laurent polynomials (Q2369539)

\textit{M. A. Dritschel} [``On factorization of trigonometric polynomials'', Integral Equ. Oper. Theory 49, No. 1, 11--42 (2004; Zbl 1070.47011)] proved that if \(P(z_1,\ldots,z_d)\) is a multivariate Laurent polynomial that is positive on the multi-torus \(| z_1| =\cdots=| z_d| =1\), then it can be written as \(P(z)=\sum_k Q_k(x)^*Q_k(z)\), with \(Q_k\) polynomials, and \({}^*\) denoting complex conjugate. This paper gives an alternative proof based on the factorization of univariate matrix-valued positive definite Laurent polynomials by recursively separating one of the indeterminate variables. A Bauer-type procedure [\textit{F. L. Bauer}, ``Beiträge zur Entwicklung numerischer Verfahren für programmgesteuerte Rechenanlagen. II. Direkte Faktorisierung eines Polynoms'', Sitzungsber. Math.-Naturw. Kl. Bayer. Akad. Wiss. München 1956, 163--203 (1957; Zbl 0078.30602)] is proposed for the latter for which exponentially fast convergence is proved if the polynomial coefficients decay exponentially. The method consists in considering the infinite banded block Toeplitz matrix whose symbol is the univariate positive definite Laurent polynomial. Factoring the finite central part, into its block Cholesky factors approximates the banded Cholesky factor of the infinite matrix, and hence the polynomial factor of the Laurent polynomial. Exponential convergence refers to the size of the central part going to infinity.