Approximating separable nonlinear functions via mixed zero–one programs

We discuss two models from the literature that have been developed to formulate piecewise linear approximation of separable nonlinear functons by way of mixed-integer programs. We show that the most commonly proposed method is computationally inferior to a lesser known technique by comparing analytically the linear programming relaxations of the two formulations. A third way of formulating the problem, that shares the advantages of the better of the two known methods, is also proposed.