Since G.E.P. Box introduced central composite designs in early fifties of 20th century, the classic design of experiments (DoE) utilizes response surface models (RSM), however usually limited to the simple form of low-degree polynomials. In the case of small size datasets, the conformity with the normal distribution has very weak reliability and it leads to very uncertain assessment of a parameter statistical significance. The bootstrap approach appears to be better solution than - theoretically proved but only asymptotically equal - t distribution based evaluation. The authors presents the comparison of the RSM model evaluated by a classic method and bootstrap approach.
The paper shows the adaptation of the cross validation approach, known from interpolation problems, for estimating the value of a shape parameter for radial basis functions. The latter are involved in two collocation techniques used on an unstructured grid to find approximate solution of differential equations. To obtain accurate results, the shape parameter should be chosen as a result of a trade-off between accuracy and conditioning of the system. The cross validation approach called “leave one out” takes these aspects into consideration. The numerical examples that summarize the investigations show the usefulness of the approach.
The paper shows the approach to the interpolation of scattered data which includes not only function values, but also values of derivatives of the function. To this end, an interpolant composed of radial basis functions is used and extended by terms possessing appropriate derivative terms. The latter match the given derivatives. Special attention is paid to the problem of choosing the value of the shape parameter, which is included in radial functions and influences the accuracy and stability of the solution. To validate the method, several numerical tests are carried out in the paper.