We propose and study the optimal estimates for a beta-Lipschitz periodic frontier of a 2D-set of points, for a given parameter smoothness. They are defined as kernel estimates being sufficiently regular, covering data points, and whose associated support is of smallest surface. The estimates are written as linear combinations of kernel functions applied to the points of the sample. The coefficients of the linear combination are then computed by solving related linear programming problem. The error of the frontier function estimation is shown to be almost surely converging to zero, with the optimal rate. The behavior of the estimates on finite sample situations is illustrated by some simulations. Finally, we discuss possibilities of adaptive version of the proposed estimator, when parameter smoothness beta is unknown.
