Stationary second-degree iterative methods (Q1344332)

Consider the linear system \(Ax= b\). Many iterative solution methods construct approximating vectors \(x^{(n)}\) for \(x\), such that the errors \(e^{(n)}= x^{(n)}- x\) satisfy \(e^{(n)}= P_ n(G) e^{(0)}\), where \(P_ n\) is a polynomial with \(P_ n(1)= 1\) and \(G\) is the so-called splitting matrix. The convergence depends on the spectral radius of \(P_ n(G)\). If the eigenvalues \(\mu\) of \(G\) are known to be in an interval \([\alpha, \beta]\), with \(\beta< 1\), the virtual spectral radius \(S(P_ n(G))= \max_{\mu\in [\alpha, \beta]}| P_ n(\mu)|\) can serve as a substitute for the spectral radius. In this paper, expressions are derived for the virtual spectral radius and its asymptotic average \(\limsup_{n\to\infty} [S(P_ n(G))]^{1/n}\) in the case where the polynomials \(P_ n\) satisfy a three term recurrence relation with constant coefficients. Explicit forms of such polynomials in terms of Chebyshev polynomials are derived, which allow to find the required estimates.