Surface of PSNR values as a function of p and q after deconvolution using the regularizer ${\phi }_q^p(\mathit{x})$. The data file is a MHD simulation output as described in section 4. In the experiment, the beam effect H is the beam kernel defined in section 4. The lambda parameter is λ = 10⁻³. For a discrete set of (p, q) values in the range [0, 2] × [−1, 3] (50 equally spaced points on each axis)), we draw a color point representing the PSNR between the original (with no beam) image and the reconstructed one after applying the optimization process eq. 8. The higher the PSNR, the better the reconstruction. Left and right panels show respectively a 2D (orthographic) top view and a 3D view of the PSNR surface.