“…This is achieved by choosing a large value of the parameter h to minimize N h , and iteratively lowering it to increase N h to include more eigenfunctions until the noiseless spectrum s h is accurately reconstructed. This is in agreement with the properties of the eigenfunctions ψ ( f ), where it has been shown that the n th squared eigenfunction has n wells, implying that the first squared eigenfunction is represented by a single well function localized at its maximum, the second squared eigenfunction has two wells, and so on . By analogy, one can see that the squared eigenfunctions with low n h values represent, in a broad manner, the profiles of the peaks of the spectrum, whereas the functions with high n h values characterize more the fine details of these profiles.…”