A linearisation of the maximum entropy formalism using separable programming

The maximum entropy principle is a standard tool for the calibration of non‐linear programming models which are frequently used for policy analysis. The information entropy function is concave and separable. In this paper, we derive a linear approximation of the entropy using separable programming. As we demonstrate, our linear entropy formulation is useful for the calibration of separable non‐linear models of very large scale. To demonstrate, we solve both an ill‐posed and a well‐posed inverse problem and we analyse the sensitivity of the results on the number of breakpoints of the piecewise linear approximation.