On polyharmonic interpolation (Q878454): Difference between revisions

The authors introduce a new approach to multivariate interpolation by employing polyharmonic functions as interpolants, a.e. by solutions of higher order elliptic equations. They assume that the data arise from \(C^{\infty}\) or analytic functions in the ball \(B_R\). They prove two main results on the interpolation of \(C^{\infty}\) or analytic functions \(f\) in the ball \(B_R=\{x\in\mathbb R^n:| x| <R\}\) by polyharmonic functions \(h\) of a given order of polyharmonicity \(p\). Specifically they assume to have a basis of the space of harmonic homogeneous polynomials of degree \(k\) which are denoted as \(Y_{kl}(x)\) for \(k=0,1,\dots \), and \(l=1,2,\dots ,d_k\), where \[ d_k={1\over (n-2)!}(n+2k-2)(n+k-3)\cdots (k+1), \] and are assumed to be orthonormalized with respect to the scalar product \[ {1\over \omega _{n-1}}\int_{S^{n-1}}u(\theta)v(\theta)\,d\theta \] on the unit sphere, where \(\omega _{n-1}\) is the area of the unit sphere \(S^{n-1}\) in \(\mathbb R^n\); it has been put \(x=r\theta\), \(r=| x| \) For \(f\in C^{\infty }(\overline{B_R})\) (the set of \(C^{\infty }\) functions on a neighborhood of \(\overline{B_R}\)) one has the expansion in spherical harmonics \[ f(x)=\sum_{k=0}^{\infty }\sum_{l=1}^{d_k} \widetilde{f}_{k,l}(r)Y_{k,l}(\theta). \] They use the following representation of \(C^{\infty }\) and of analytic functions in the ball: Proposition: Let \(f\in C^{\infty }(\overline{B_R})\). Then we have the following expansion \[ f(x)=\sum_{k=0}^{\infty }\sum_{l=1}^{d_k} f_{k,l}(r^2)r^kY_{k,l}(\theta), \] where the function \(f_{k,l}\in C^{\infty }([0,R^2])\). The function \(f\) is analitic in some neighborhood of \(0\) in \(\mathbb R^n\) if and only if there exists \(t_0>0\) and \(M>0\) such that, for all indices \(k\geq 0\), \(1\leq l\leq d_k\), and \(j\geq 0\), we have \[ \sup_{0\leq t\leq t_0}\left| \left({d^j\over dt^j}f_{k,l}(t)\right)\right| \leq M^{k+j+1}j! \] On the other hand if \(h\) is a function polyharmonic of order \(N\) in the ball \(B_R\), then we have the expansion \[ h(x)=\sum_{k=0}^{\infty }\sum_{l=1}^{d_k} h_{k,l}(r^2)r^kY_{k,l}(\theta), \] and it is well known that the coefficients \(h_{k,l}(\cdot )\) are polynomial of degreee \(N-1\). Thus the polynomials \(h_{k,l}\) are put into correspondence with the functions \(f_{k,l}\). The polyharmonic interpolation problem is then natural to formulate: assume that for every fixed \((k,l)\) with \(k=0,1,\dots \) and \(l=1,2,\dots , d_k\) we have interpolation points which we assume to be pairwise different \(0\leq r_{k,l,1}<r_{k,l,2}<\cdots <r_{k,l,N}\leq R \). Then for every \((k,l)\) we find the polynomials \(h_{k,l}\) of degree \(\leq N-1\) from the one dimensional interpolation problems \[ h_{kl}(r^2_{k,l,j})=f_{kl}(r^2_{k,l,j})\;\;\;j=1,2,\dots , N. \] Now the question is for which distribution of the points \(\{r_{k,l,j}\}\) and for which functions \(f\) is the series in \[ h(x)=\sum_{k=0}^{\infty }\sum_{l=1}^{d_k} h_{k,l}(r^2)r^kY_{k,l}(\theta) \] convergent. The first result of the paper says that for every distribution of the points \(\{r_{k,l,j}\}\) and for a wide class of \(C^{\infty}\) functions \(f\) we have convergence. In the second theorem, the special case of the knots which are lying on \(N\) concentric spheres in \(\mathbb R^n\) is considered.