Plenary Talk: Mo July 5, 14:30-15:30 (UTC+2)

Hub Networks

Morton E. O'Kelly

This paper reviews some salient aspects of the location of hub facilities and network design. Contrasts between idealized models and observed airline networks are shown. While early research emphasized what is known as a single allocation version of the problem, this paper shows that empirical airline networks are likely to have incomplete backbone connections; multiple allocations to hubs; and incidental direct connections between some of the nodes. Some airlines are close to theoretically ideal one-hub configurations, but the majority are much more complex.

The paper reviews and contrasts empirically observed cases and idealized models. In a detailed example, the multiple allocation un-capacitated hub location problem is solved using Benders decomposition. Sample solutions are shown with an emphasis on interpreting the results and extracting good understanding of the parameters that are consistent with the observed network structure. When a = 0, we have the special case of nearest hub assignment and p-median, single allocation, and multiple allocation models all give the same result. When a = 1, we have no discount, and therefore no incentive to move on the interhub network. Because of the triangle inequality, we expect that the linkages make one intermediate hub stop (at most).

Solving the multiple allocation problem with a large number of parameter variants yields a tradeoff curve between the interhub discount factor and the allocation count, (1,2,…,p). For example, a network with a very efficient and low-cost backbone would be expected to have strong leaning towards a median-like single allocation, while a network with virtually no saving in interhub flow would lean towards multiple allocation, and in the extreme case might have assignment to all p hubs.