Condition number complexity of an elementary algorithm for computing a reliable solution of a conic linear system (Q1587936): Difference between revisions

Let \(X,\) \(Y\) be two finite-dimensional normed linear spaces, \(A:X\rightarrow Y\) a linear operator, \(b\in Y\) and \(C_{X}\) a closed convex cone in \(X.\) For solving the conic linear system \[ Ax = b, \qquad x \in C_{X}, \] the authors consider a generalization of an algorithm due to von Neumann for solving ordinary linear inequality systems and analyze its complexity in terms of a suitably defined condition number. They stress that the advantage of their algorithm with respect to more efficient (in terms of number of iterations) interior point or ellipsoid type methods lies in the significantly less computational effort that each iteration requires.