GROSVENOR, D.D and H.O. Hartley. IBM 650 Program for Linear Programming. Statistical Laboratory, Iowa State University; Ames, Iowa: 1960. 11x8.5 inches, 134pp (printed on one side of the sheet, only) offset mimeo, staple bound. Fine condition. $500 There are no citings for this work in WorldCat/OCLC.
Notes on this paper as they appear in "Linear Programming on High Speed Computers", Author(s): J. Rudolf : Journal of Farm Economics, Vol. 42, No. 5, Proceedings of the Annual Meeting of the American Farm Economic Association (Dec., 1960), pp. 1439-1444:
"1. Sorting and selection. This problem has to some degree been alleviated by the new linear programming code for the IBM 650 as developed by Hartley and Grosvenor at Iowa State University. In this program the operator has a choice of allowing the 650 to do an iteration for every positive element appearing in the simplex vector. This produces a larger number of total iterations, but the saving in time afforded by not having to look for the largest in an entire set of vectors can actually reduce the total computation time. Also, this program allows the operator to preselect certain iterations, thereby allowing a human to enter himself into the program. This is, however, done at the beginning of the program and therefore the program need not be interrupted, which, as indicated, loses a great amount of time."
"2. Storage. It is in the storage requirements that a great step forward can be made. It is possible to construct a vector (the "eta vector") of numbers which allows the transition from one iteration to the next by a matrix vector multiplication. This has been taken advantage of, again in the Hartley-Grosvenor program from Iowa State, to reduce storage requirements. In these programs the original matrix is on cards or tape and only the eta vectors are stored on the drum. Since the size of these vectors corresponds to the number of restrictions in the problem, one can store a large number of vectors, and thereby record a large number of iterations in the storage capacity of most computers. This advantage is somewhat tempered by the fact that the entire matrix must be fed into the machine through an input-output device for every iteration. Since input-output is very often a limiting factor in high speed computers this presents somewhat of a disadvantage. Therefore, this approach may not be very suitable for the very fast, very large computers."