Regularized least squares approximations on the sphere using spherical designs (Q2903048)

In the paper by \textit{M. Gräf} and \textit{D. Potts} [Numer. Math. 119, No. 4, 699--724 (2011; Zbl 1232.65045)] a method is presented for constructing optimal \(t\)-designs on the unit sphere \(\mathbb{S}^2\subset\mathbb{R}^3\). That is a set of nodes for a quadrature formula that is exact for all spherical polynomials up to degree \(t\) (i.e., \(\forall p\in\mathbb{P}_t\)). Here \(t\)-designs are used to approximate a function \(f\) on the sphere. The idea is to solve a regularized least squares problem NEWLINE\[NEWLINE\sum_{j=1}^N (p(x_j)-f(x_j))^2+\lambda\sum_{j=1}^N(\mathcal{R}_L p(x_j))^2,NEWLINE\]NEWLINE where \(f\) is a continuous function, \(p\in\mathbb{P}_L\), \(\mathcal{X}_N=\{x_j\}_{j=1}^N\) is a set of samples, \(N\geq \mathrm{dim}\;\mathbb{P}_t=(t+1)^2\), \(\lambda>0\) and \(\mathcal{R}_L\) is a regularization operator. The polynomial \(p\) and the regularized \(\mathcal{R}_Lp\) is expanded in terms of spherical harmonics. The corresponding coefficients for \(p\) are the parameters in the discrete least squares problem. The coefficients in \(\mathcal{R}_Lp\) are chosen so as to correspond to a discretized version of, e.g., the Laplace-Beltrami regularization or filtered polynomial approximation. Numerical aspects that are discussed include a condition number for the least squares problem, theoretical error bounds and several other issues.