MICROCOMPUTER ACCURACY IN SOLVING LINEAR PROGRAMMING PROBLEMS WITH REDUNDANT CONSTRAINTS

This study reports on how different microcomputer systems performed in the solution of two linear programming models purposely specified with redundant vectors. Comparisons were made to a Cyber 720 that used both a Fortran and Basic version of the same primal-dual algorithm. Results are mixed. But Microsoft Basic with double precision under CP/M on a Z80A processor performed at least equally well to the Cyber 720 provided that an appropriate essential zero value was specified. Different coefficient scaling schemes were also tested. The results should be of interest to all users of matrix inversion schemes on microcomputers. Extensions of the study to new hardware and software systems are encouraged.