“…Algorithm 1 -Edge-by-edge matrix-vector product It is known that iterative Krylov solvers such as CG are very efficient at minimizing errors associated with higher frequency modes but can be rather slow for lower frequency ones (Aubry et al, 2008;Frank and Vuik, 2001;Löhner et al, 2011;Nicolaides, 1987;Vermolen et al, 2002), which in other words translates as the smaller the spectral condition number, that is, the ratio between the largest and smallest eigenvalues of the matrix, the faster convergence will be. The use of deflation aims to accelerate solution convergence by reducing the spectral condition number in comparison with the original system Mut, 2008), with Nicolaides (1987) presenting its mathematical proof, later discussed again by Frank and Vuik (2001). For a generic system of linear equations A·x ¼ b, given a deflation space W, of order n × m, rank m and m ≪ n, the projector P, of order n × n, mapping A −1 orthogonally onto W is:…”