Numerical analysis of integer ABS methods (Q2915554)

The authors provide the first detailed numerical analysis of integer Abaffy-Broyden-Spedicato (iABS) algorithms. The different iABS variants were implemented using a new Java numerical package (JiABS) and their performance in terms of speed and the magnitude of the intermediate values was tested. All implemented algorithms were able to solve pre-defined and randomly generated linear Diophantine systems of equations. Analysis of a large set of randomly generated systems of equations identified one sub-variant (called W2 implementation) of the scaled non-symmetric integer ABS algorithm (snsiABS) algorithm to be the fastest of the algorithms tested while also generating low intermediate values. This subvariant also proved to be comparable to, or even faster than, the Linsolve algorithm of the commercially available Maple software under most conditions tested.