Data completion via combined and optimized small scale high dimensional model representation.

This work deals with the data completion for a multivariate function whose values are given on the nodes, except a few ones, of an orthonormal hyperprismatic grid. The method we develop here uses a recently developed scheme we call "Combined Small Scale High Dimensional Model Representation (CSSHDMR)". This scheme is preferred to be used mostly at constant or perhaps univariate level bringing some discontinuities at the borders of subregions in the function values or its derivatives or both. In the case where certain function values or derivatives are unknown, these discontinuities can be suppressed to get those unknown values. This is done via optimisation. We try to present some details of this issue in this extended abstract. [ABSTRACT FROM AUTHOR]