Why Methods for Optimization Problems with Time-Consuming Function Evaluations and Integer Variables Should Use Global Approximation Models

This paper advocates the use of methods based on global approximation models for optimization problems with time-consuming function evaluations and integer variables.We show that methods based on local approximations may lead to the integer rounding of the optimal solution of the continuous problem, and even to worse solutions.Then we discuss a method based on global approximations.Test results show that such a method performs well, both for theoretical and practical examples, without suffering the disadvantages of methods based on local approximations.