Data error analysis in unconstrained optimization problems with the CESTAC method

The CESTAC method—also known as the Permutation-Perturbation method—was first conceived to analyse the propagation of the round-off error when a numerical algorithm is run on a computer. Moreover, for iterative optimization methods it leads to the definition of an optimal termination criterion. In this paper we show how the same mecanisms can be used when data errors interfere with unconstrained optimization problems, and what the interests of this method are compared to the most frequently used ones: the crude Monte Carlo method and the linearisation of the optimization function.