High order numerical derivatives for one-dimensional scattered noisy data (Q2493742)

The authors investigate a classical ill-posed problem, numerical differentiation, by using smoothing splines and the Tikhonov regularization technique. The regularization method proposed here for computing high order derivatives from one dimensional noisy data uses a semi-norm in the global space as a stabilizing functional. This allows to represent the minimizer of the smoothing functional as a linear combination of high order splines and polynomials. For the choice of the regularization parameter, an a priori choice rule and Morozov's discrepancy principle, are applied. Convergence results for high order derivatives, under two different choices of the regularization parameter, are given. Numerical examples are performed and the results show that the proposed algorithm is effective and stable.