“…In particular, we shall extend most of the solution techniques in [24] for usual PDEs to fractional PDEs. More precisely, for separable fractional PDEs, we shall apply the matrix diagonalization methods (i.e., discrete separation of variables) [15,8,24] to reduce a (discrete) multi-dimensional problem to a sequence of (discrete ) one-dimensional problems or to a diagonal system with a total cost of just a few N d+1 flops (d is the space dimension); for non-separable fractional PDEs, we shall apply a preconditioned BICGSTAB iterative method using (i) a related fractional separable problem with constant-coefficients as preconditioned, and (ii) a fast matrix-free algorithm for the matrix-vector multiplication, so that the total cost is still O(N d+1 ). Thus, the cost of a spectral method for fractional PDEs is essentially the same as that for usual PDEs.…”