Application of Fredholm integral equations inverse theory to the radial basis function approximation problem I

This paper reveals and examines the relationship between the solution and stability of Fredholm integral equations and radial basis function approximation or interpolation. The underlying system (kernel) matrices are shown to have a smoothing property which is dependent on the choice of kernel. Instead of using the condition number to describe the ill-conditioning, hence only looking at the largest and smallest singular values of the matrix, techniques from inverse theory, particularly the Picard condition, show that it is understanding the exponential decay of the singular values which is critical for interpreting and mitigating instability. Results on the spectra of certain classes of kernel matrices are reviewed, verifying the exponential decay of the singular values. Numerical results illustrating the application of integral equation inverse theory are also provided and demonstrate that interpolation weights may be regarded as samplings of a weighted solution of an integral equation. This is then relevant for mapping from one set of radial basis function centers to another set. Techniques for the solution of integral equations can be further exploited in future studies to find stable solutions and to reduce the impact of errors in the data.

[1]  M. Bôcher,et al.  An Introduction to the Study of Integral Equations , 1909, Nature.

[2]  J. Mercer.  Functions of positive and negative type, and their connection with the theory of integral equations , 1909 .

[3]  H. Weyl.  Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen (mit einer Anwendung auf die Theorie der Hohlraumstrahlung) , 1912 .

[4]  E. Hille,et al.  On the characteristic values of linear integral equations , 1931, Proceedings of the National Academy of Sciences of the United States of America.

[5]  F. Smithies.  The Eigen‐Values and Singular Values of Integral Equations† , 1938 .

[6]  R. Courant,et al.  Methods of Mathematical Physics , 1962 .

[7]  Peter R. Winters,et al.  Forecasting Sales by Exponentially Weighted Moving Averages , 1960 .

[8]  J. Gillis,et al.  Matrix Iterative Analysis , 1961 .

[9]  J. A. Cochran.  The analysis of linear integral equations , 1973 .

[10]  Chris Chatfield,et al.  Introduction to Statistical Time Series. , 1976 .

[11]  J. Varah.  A Practical Examination of Some Numerical Methods for Linear Discrete Ill-Posed Problems , 1979 .

[12]  F. Hoog.  Review of Fredholm Equations of the First Kind , 1980 .

[13]  J. Varah.  Pitfalls in the Numerical Solution of Linear Ill-Posed Problems , 1981 .

[14]  Gene H. Golub,et al.  Matrix computations , 1983 .

[15]  J. Reade.  Eigenvalues of Positive Definite Kernels II , 1983 .

[16]  J. Reade,et al.  Eigenvalues of Analytic Kernels , 1984 .

[17]  J. Reade.  On the Sharpness of Weyl’s Estimates for Eigenvalues of Smooth Kernels, II , 1985 .

[18]  C. Micchelli.  Interpolation of scattered data: Distance matrices and conditionally positive definite functions , 1986 .

[19]  P. Hansen.  Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects of Linear Inversion , 1987 .

[20]  R. Kress.  Linear Integral Equations , 1989 .

[21]  P. Hansen.  The discrete picard condition for discrete ill-posed problems , 1990 .

[22]  Joseph D. Ward,et al.  Norms of Inverses for Matrices Associated with Scattered Data , 1991, Curves and Surfaces.

[23]  K. Ball.  Eigenvalues of Euclidean distance matrices , 1992 .

[24]  J. Ward,et al.  On the sensitivity of radial basis interpolation to minimal data separation distance , 1992 .

[25]  Xingping Sun.  Norm estimates for inverses of Euclidean distance matrices , 1992 .

[26]  F. J. Narcowich,et al.  Norm estimates for the inverse of a general class of scattered-data radial-function interpolation matrices , 1992 .

[27]  F. J. Narcowich,et al.  On Condition Numbers Associated with Radial-Function Interpolation , 1994 .

[28]  R. Schaback.  Lower Bounds for Norms of Inverses of Interpolation Matrices for Radial Basis Functions , 1994 .

[29]  B. Baxter.  Norm Estimates for Inverses of Toeplitz Distance Matrices , 1994 .

[30]  Robert Schaback,et al.  Error estimates and condition numbers for radial basis function interpolation , 1995, Adv. Comput. Math..

[31]  B. Rust.  Truncating the Singular Value Decomposition for Ill-Posed Problems , 1998 .

[32]  T. Driscoll,et al.  Interpolation in the limit of increasingly flat radial basis functions , 2002 .

[33]  B. Fornberg,et al.  Some observations regarding interpolants in the limit of flat radial basis functions , 2003 .

[34]  Holger Wendland,et al.  Scattered Data Approximation: Conditionally positive definite functions , 2004 .

[35]  C. Holt.  Author's retrospective on ‘Forecasting seasonals and trends by exponentially weighted moving averages’ , 2004 .

[36]  Per Christian Hansen,et al.  Computation of the singular value expansion , 1988, Computing.

[37]  P. Hansen,et al.  Exploiting Residual Information in the Parameter Choice for Discrete Ill-Posed Problems , 2006 .

[38]  Per Christian Hansen,et al.  Regularization Tools version 4.0 for Matlab 7.3 , 2007, Numerical Algorithms.

[39]  Dianne P. O'Leary,et al.  Residual periodograms for choosing regularization parameters for ill-posed problems , 2008 .

[40]  P. Hansen.  Discrete Inverse Problems: Insight and Algorithms , 2010 .

[41]  Elisabeth Larsson,et al.  Stable Computations with Gaussian Radial Basis Functions , 2011, SIAM J. Sci. Comput..

[42]  B. Fornberg,et al.  Radial basis functions: Developments and applications to planetary scale flows , 2011 .