An iterative algorithm for solving a finite-dimensional linear operator equation \(T(x)=f\) with applications (Q847208)

The authors introduce a new two stage iterative algorithm to solve linear equations over finite dimensional vector spaces \(H_i\) in finitely many steps. If \(T : H_1 \to H_2\) is linear and \(T^* : H_2 \to H_1\) is its adjoint with \(\langle y,T^*z\rangle = \langle Ty,z\rangle\) for all \(y\) and \(z\) and \(Tx = f\) is to be solved, the algorithm works on a sequence of \(T\) residuals and simultaneously on the adjoint residuals in a way reminiscent of non-orthogonal projection methods. The method is adapted to solve different matrix equations such as \(AXB = C\), min\(\|AXB-C\|\), or \(AXB + CYD = E\) and inverse problems such as \(XA = B\). Numerical tests conclude the paper.