Identification of special structure constraints in linear programs

In large LPs it may be that a major fraction of the constraints has a particularly simple form. Frequently this simple form may be exploited by either decomposition or implicit representation methods. In these cases, the effective size and computational difficulty are more closely related to the number of remaining nonspecial constraints than to the number of special constraints. A unified computational procedure is presented for mechanically identifying near maximal sets of special structure constraints for most of the kinds of special structures which have been identified thus far as being exploitable.