Generalized column generation for linear programming

Abstract

Column generation is a well-known and widely practiced technique for solving linear programs with too many variables or constraints to include in the initial formulation explicitly. Instead, the required column information is generated at each iteration of the simplex algorithm. This paper shows that, even if the number of variables is low enough for explicit inclusion in the model with the available technology, it may still be more efficient to resort to column generation for some class of problems.