A new method for computing a \(p\)-solution to parametric interval linear systems with affine-linear and nonlinear dependencies (Q1689318)

A revised affine interval analysis is used, which includes all errors, including the rounding errors. This means a representation of the form \(\hat{x}=x_0+\sum_{i=1}^n x_i\varepsilon_i + x_r[-1,1]\), \(x_0\) is the central value, \(x_i\) are called partial derivatives, \(\varepsilon_i\in[-1,1]\) are uncertainties, and \(x_r\geq0\) is the radius of the accumulated error. After introducing the related arithmetic, it is used to solve a parametric interval linear system \(A(p)x=b(p)\), where \(p\) is an uncertain parameter from some affine interval. The problem of finding all solutions \(S\) is generically NP hard, so that an outer interval containing \(S\) or an estimate of the inner hull contained in \(S\) is computed in practice. This is obtained by first finding an affine transformation of the system. The resulting system is preconditioned and solved by a Gauss-Seidel type iteration. The convergence of the method is analysed and extensively illustrated with numerical results.