On the relation between Fourier and Walsh-Rademacher spectra for random fields (Q6144891)

Many areas of medicine, computer vision, fluid flow and thermodynamics, have a statistical basis which makes important use of the spectral representation of certain random fields. On the other hand, modern computational methods, for various structural reasons are based on local discrete operators on unstructured meshes. Thus, in order to interpret computational results in the classical language of spectral analysis well, suitable postprocessing is important. In this interesting paper, the authors consider a certain discrete averaging operation on a hierarchy of nested grids. The inferred spectra can then be translated consistently to the classical Fourier setting in a statistical sense. The translation function is computed explicitly in the simple case of nested square grids. More generally, it is inferred computationally for complex grid geometries. To give more insight. Consider the relations between the expansion coefficients of a discrete random field when analyzed with respect to the following two different hierarchical bases, the Walsh-Rademacher basis and the trigonometric Fourier basis. The authors prove that, in a statistical sense, the rate of spectral decay computed in one basis can be translated to the other. They also give expressions for this translation on quadrilateral meshes. The results are illustrated with numerical examples for deterministic and random fields. The paper is well written with a good set of references.