Smoothing noisy data under monotonicity constraints existence, characterization and convergence rates (Q1084831): Difference between revisions

Some properties about the study of interpolation procedures preserving some shape defined in the form of a convex constraint (monotonicity, positivity and convexity) are extended to the case of constrained smoothing splines. If we call \(S_{n,\lambda,C}\) to the smoothing spline where \(\lambda\) is a parameter such that \(\lambda >0\) and C is some closed convex subset of the Sobolev space of absolutely continuous functions having their first m-1 derivatives absolutely continuous and the m-st derivative square integrable, i.e. \(H^ m[0,1]\); let us assume that \(f\in C\). The author proves the existence and uniqueness of \(S_{n,\lambda,C}\) and gives a general characterization theorem that will allow him to study the case of monotone splines. Finally, if \(\sigma^ 2\) is the variance of the data, it is proved that \[ E(| f-S_{n,\lambda,C}|^ 2_ k)=O(\lambda^{(m-k)/m})+\sigma^ 2.O(n^{-1}.\lambda^{-(2k+1)/2m}). \] This result shows that in the case of smoothing splines the attainable convergence rates are the same in the unconstrained and constrained cases.