Column generation for linear and integer programming

Abstract

Column generation refers to linear programming (LP) algorithms designed to solve problems in which there are a huge number of variables compared to the number of constraints and the simplex algorithm step of determining whether the current basic solution is optimal or finding a variable to enter the basis is done by solving an optimization problem rather than by enumeration. To the best of my knowledge, the idea of using column generation to solve linear programs was first proposed by Ford and Fulkerson [16]. However, I couldn’t find the term column generation in that paper or the subsequent two seminal papers by Dantzig and Wolfe [8] and Gilmore and Gomory [17,18]. The first use of the term that I could find was in [3], a paper with the title “A column generation algorithm for a ship scheduling problem”. Ford and Fulkerson [16] gave a formulation for a multicommodity maximum flow problem in which the variables represented path flows for each commodity. The commodities represent distinct origin-destination pairs and integrality of the flows is not required. This formulation needs a number of variables exponential in the size of the underlying network since the number of paths in a graph is exponential in the size of the network. What motivated them to propose this formulation? A more natural and smaller formulation in terms of the number of constraints plus the numbers of variables is easily obtained by using arc variables rather than path variables. Ford and Fulkerson observed that even with an exponential number of variables in the path formulation, the minimum reduced cost for each commodity could be calculated by solving a shortest path problem, which was already known to be an easy problem. Moreover the number of constraints in the path formulation is the number of arcs, while in the arc formulation it is roughly the (number of nodes)x(number of commodities) + number of arcs. Therefore the size of the basis in the path formulation is independent of the number of commodities and is significantly