Argonne National Laboratory

Grey is better than black

October 13, 2016

What does a computational scientist do when faced with a nonlinear optimization problem in which derivatives of the objective or constraint functions are not available? A common approach is to pose the problem as a black-box optimization problem.

But at this year’s gathering of the Society for Industrial and Applied Mathematics – attended by a record number of 1,700 participants – Stefan Wild argued that rarely is such a black-box approach necessary.

A computational mathematician in the Mathematics and Computer Science Division at Argonne National Laboratory, Wild presented an innovative approach that involves using families of local surrogates to infer additional information on problems consisting of black-box components. The result: new, globally convergent grey-box optimization methods.

Wild’s work is reported in the October issue of SIAM News.

A video of the invited presentation is on the web: