Linear convergence for vector sequences and some applications (Q1891039)

Vector sequences \(S\) whose members \(s(n)\) \((n \geq 0)\) are in \(\mathbb{C}^k\) are considered. Assuming convergence of \(S\) to \(s\), set \(e(n) = s(n) - s\). Let \(|. |\) denote a suitable norm. Three types of linear convergence of \(S\) to \(s\) are defined: a) that in which \(\lim |e(n+1)|/ |e(n)|= r\) with \(0 < r < 1\), b) that in which for a \(k \times k\) matrix \(B\), \(I - B\) is nonsingular and \(\lim |e(n + 1) - Be(n)|/ |e(n)|= 0\) and c) that in which for the spectral radius \(\rho(B)\) of \(B\), \(\rho(B) < 1\), and for a \(k \times k\) matrix sequence \(B(n)\) \((n \geq 0)\), \(\lim B(n) = 0\) and \(e(n+1) = \{B + B(n)\} e(n)\), \((n \geq 0)\). Their equivalence is established. It is shown that for a nonsingular \(k \times k\) matrix \(B\) for which condition b) holds and either d) \(\lim e(n |i) / |e(n)|\) exists \((1 \leq i \leq k)\) \((e (n|i)\) being the \(i\)th component of \(e(n)\)) or e) \(B\) is similar to a diagonal matrix, \(B\) in b) may be replaced by a nonsingular diagonal matrix. \(E(n)\) \((n \geq 0)\) being a \(k \times k\) matrix sequence for which \(I - E(n)\) is nonsingular \((n \geq 0)\), transformations of the form \(T(E, s|n) = \{I - E(n)\}^{-1} \{s(n + 1) - E(n) s(n)\}\) are considered with reference to the preceding theory; convergence improvement results are derived. The special transformation in which, \(F(s,n)\) being the matrix whose columns are \(s(n + \omega + 1) - s(n + \omega)\), \((0 \leq \omega < k)\) and \(G(s,n)\) being derived from \(s(n + \omega + 2) - 2s(n + \omega + 1) + s(n + \omega)\) similarly, \(E(n) = F(s,n) G(s,n)^{-1}\), is investigated.