Kernel Approximation on Manifolds I: Bounding the Lebesgue Constant

Abstract

The purpose of this paper is to establish that for any compact, connected C^{\infty} Riemannian manifold there exists a robust family of kernels of increasing smoothness that are well suited for interpolation. They generate Lagrange functions that are uniformly bounded and decay away from their center at an exponential rate. An immediate corollary is that the corresponding Lebesgue constant will be uniformly bounded with a constant whose only dependence on the set of data sites is reflected in the mesh ratio, which measures the uniformity of the data. The analysis needed for these results was inspired by some fundamental work of Matveev where the Sobolev decay of Lagrange functions associated with certain kernels on \Omega \subset R^d was obtained. With a bit more work, one establishes the following: Lebesgue constants associated with surface splines and Sobolev splines are uniformly bounded on R^d provided the data sites \Xi are quasi-uniformly distributed. The non-Euclidean case is more involved as the geometry of the underlying surface comes into play. In addition to establishing bounded Lebesgue constants in this setting, a "zeros lemma" for compact Riemannian manifolds is established.Comment: 33 pages, 2 figures, new title, accepted for publication in SIAM J. on Math. Ana

A Galerkin Radial Basis Function Method for Nonlocal Diffusion

“…The thin plate spline was chosen over other RBFs due to its approximation properties and because the corresponding Lagrange basis functions enjoy an exponential decay. Recent results have noted the Lagrange function of the thin plate spline decays exponentially away from its center, which is not known for other RBFs [10,11]. There is evidence that the Lagrange functions for the thin plate splines can be replaced with local Lagrange functions, which are cheaper to construct than the Lagrange functions.…”

“…11 The log of h versus the log of the L 2 error for the discontinuous solution u 2 with the smooth kernel functions using non-uniformly spaced centers is displayed Fig. 12 The log of h versus the log of the L 2 error for the discontinuous solution u 2 with the triangular kernel functions using non-uniformly spaced centers is displayed Table 2 For 1D experiments, the mesh norm h, number of rows n of the stiffness matrix (11), and the reciprocal condition number for (11) Fig. 13 The log of h versus the log of the L 2 error for the 2D continuous solution w 1 with the smooth kernel functions using non-uniformly spaced centers is displayed For both w 1 and w 2 , we solve the Galerkin RBF problem for both the smooth kernel and the triangular kernel.…”

“…16 The log of h versus the log of the L 2 error for the 2D discontinuous solution w 2 with the triangular kernel functions using non-uniformly spaced centers is displayed Table 4 For 2D experiments, the mesh norm h, number of rows n of the stiffness matrix (11), and the reciprocal condition number for (11) …”

Gaussian processes for computer experiments

“…In particular, when considering the setting where D is a fixed compact space and n goes to infinity, it is not clear, to the best of the authors' knowledge, if E [Y (x)|y], where y = f (X) + , converges to f (x) as n → ∞, for any fixed continuous function f . Some partial results are nevertheless given in [24,63]. In contrasts, it is clear that other popular methods, like nearest-neighboor regression or kernel smoothing, can asymptotically recover continuous functions as n → ∞.…”

Optimal order Jackson type inequality for scaled Shepard approximation

“…This is a standard technique in many branches of analysis. Notably, the technique has recently been applied by authors of [8] and [9] in bounding the L ∞ -norms of interpolation operators and the least square operators associated with radial basis functions. The following question arises naturally: how many well-separated points can be put inside a specified annulus?…”