For large-scale p-median problems, it is common to aggregate the demand points. This size reduction via aggregation makes the problem easier to solve, but introduces error. Doing this aggregation well is provably difficult. The authors present a median-row-column aggregation algorithm, MRC, with provable properties including an error bound, an (attainable) upper bound on the maximum objective function error. MRC adjusts spacing of individual rows and columns to exploit problem structure. For e demand points, r rows, and c columns, the algorithm has computational order e(c+r+loge), and order e storage requirements. The authors report encouraging computational experience.
