On a Hermite interpolation on the sphere (Q631927)

The authors study Hermite interpolation on the unit sphere using the space \(\Pi^3_n\) of trivariate polymomials with real coefficients and of total degree at most \(n\); the space of all these polynomials, restricted to \(S^2=\{(x,y,z)= (\sin\varphi\,\sin\theta,\,\sin\varphi\,\cos\theta,\,\cos\varphi)\in {\mathbb R}^3\mid x^2+y^2+z^2=1,\;\theta\in [0,2\pi],\varphi\in [0,\pi]\}\) has dimension \((n+1)^2\). First they look at Hermite interpolation on a set \({\mathcal C}=\{C_i,\,1\leq i\leq\kappa\}\) of coaxial circles on the sphere whose common axis is \(\ell\) and \(C_i\) is a circle with spherical coordinates \((\overline{\varphi},\overline{\theta})\). On \(C_i\) a set of \(m_i\) distinct nodes \(A_i^n=\{Q_{i,j},\,1\leq j\leq m_i\}\) is chosen and \(\delta_{i,j}\geq 1\) is the multiplicity of each point \(Q_{i,j}\in A_i^n\); furthermore \(\delta_i=\max_j\,\delta_{i,j}\) and \(\lambda=\sum_{i=1}^{\kappa} \,\delta_i\). Assume \[ \sum_{i=1}^{\kappa} \sum_{j=1}^{m_i} \delta_{i,j}=\lambda (2n+2-\lambda). \] (This number is the dimension of the interpolation space along the set of coaxial circles.) Define the set of interpolating functionals \(\overline{L}^{n,\lambda}=\{ \overline{L}_{i,j,k} \}\) with \[ \begin{multlined}\overline{L}_{i,j,k}= \left[\left({\partial\over \partial\overline{\varphi}}\right)^k+ \beta_1^{(i,j)}\left({\partial\over \partial\overline{\varphi}}\right)^{k-1}+ \cdots+ \beta_{k-1}^{(i,j)}\left({\partial\over \partial\overline{\varphi}}\right) +\beta_k^{(i,j)}\right]_{Q_{i,j}},\\ 0\leq k\leq \delta_{i,j}-1,\;1\leq j\leq m_i,\;1\leq i\leq \kappa,\end{multlined} \] where the coefficients are real. The problem is now: find for any set of data \(\{f_{i,j,k}\}\) a polynomial \(p_n(x,y,z)\in\Pi_n^3\) with \[ \overline{L}_{i,j,k}[p_n]=f_{i,j,l},\quad 0\leq k\leq \delta_{i,j}-1,\;1\leq j\leq m_i,\;1\leq i\leq \kappa.\tag \(*\) \] The authors then state a definition along the following line. If there always exists \textbf{a} polynomial solution such that \((*)\) is satisfied, then the problem is well posed. At this time the set of interpolation functionals is properly posed only. This reviewer thinks that the \textbf{a} should be replaced here by \textbf{a unique}. Furthermore, the authors define a set of interpolating functionals denoted by \(L^n\) on a set of mutually distinct circles \(C_i\) on the unit sphere with axes \(\ell_i\) and sets \(V_i\) of \(m_i\) distinct nodes. Here the condition on the multiplicities (defined as for the other interpolating polynomials) is \[ \sum_{i=1}^\sigma \sum_{j=1}^{m_i} \delta_{i,j}=(n+1)^2. \] The \(\overline{\varphi}\) in the partial derivatives is here replaced by the local coordinate \(\overline{\varphi}_i\) of the \(i\)-th circle. The main results of the paper are: Theorem 2.1 + 2.2 + 2.3. The interpolating functionals \(\overline{L}^{n,\lambda}\) lead to a well-posed problem. The first theorem deals with \(\lambda\) odd, the second with \(\lambda\) even and both covering equidistant interpolation nodes; the third with arbitrary nodes. However, in each of the theorems there are conditions on the circles and the multiplicities that are stated in the text of the sections and not in the formulation of the theorems themselves. Finally there is a result on superposition of interpolating processes (that only works for the concept `properly posed': there is \textbf{a} solution): Theorem 3.1. Let \(L^n\) be a properly posed set of interpolating functionals of degree \(n\) on the sphere with sets of nodes \(V_n\). Suppose a set of coaxial cicles \({\mathcal C}\) with toal multiplicity \(\lambda\) does not pass through any point of \(V_n\) and let \(\overline{L}^{n+\lambda,\lambda}\) be a properly posed set of interpolating functionals of degree \(n+\lambda\) along \({\mathcal C}\) with node sets \(A^{\lambda}_{n+\lambda}\). Then \(L^n\cup\overline{L}^{n+\lambda,\lambda}\) is a properly posed set of interpolating functionals of degree \(n+\lambda\) on the sphere with node sets \(V_n\cup A^{\lambda}_{n+\lambda}\).