As this discretisation method works by finding an optimal solution to the local wave vector amplitudes via a Moore–Penrose References pseudo-inverse (computed here via a singular value decomposition), it is important that the size of the matrix H is kept to a minimum so as to limit the assembly requirements for the overall stiffness matrix. For the present 2D computation, the number of neighbours on the regular grid is M = 8, wave vectors be used for internal points and N > 9 for boundary points, which have an additional row. Fig. 13 shows that the accuracy of the solution with N = 10 used throughout the domain is almost identical to that with a larger number of wave vectors used. The size of the pseudo-inverse should therefore not be adversely large.