On geometric programming problems having negative degrees of difficulty

The dual of a geometric programming problem with negative degree of difficult is often infeasible. It has been suggested that such problems be solved by finding a dual ‘approximate’ solution which minimizes a measure of the infeasibility, e.g., the summed squares of the infeasibilities in the dual constraints. The authors point out the shortcomings in that approach, and suggest a simple technique to ensure dual feasibility, namely the addition of a constant term to the primal objective.