Introduction

Abstract

DECOMP is a Fortran code of the Dantzig-Wolfe (D-W) decomposition algorithm for solving block-angular linear programs. Originally coded in 1973 by Carlos Winkler at the Systems Optimization Laboratory (SOL) at Stanford University, DECOMP was built around John Tomlin’s LPM1 (Tomlin [1973]), an all-in-core implementation of the revised simplex method. Since then James Ho and his European collaborators, notably Etienne Loute at the Center for Operations Research and Econometrics in Belgium, had expanded and improved upon the code as well as adapting it to run on various machines, including IBM’s 370 series, CDC’s Cyber series and DATA General’s MV8000. It was the prototype for subsequent implementations based on commercial software (e.g. DECOMPSX with IBM’s MPSX/370 in Ho & Loute [1981]) that provided significant benchmark results in LP decomposition (Ho & Loute [1983]). More recently, R.P. Sundarraj adapted the code for DEC’s VAX computers in both the UNIX and VMS environment. DECOMP, as documented herein, is dimensioned to solve problems with up to 4000 rows, 10,000 columns and 55,000 non-zero elements and is intended primarily to be an experimental tool for research on computational aspects of large scale linear programming. Also, it has proven to be robust and relatively portable and may actually be useful for routine applications in certain computing environments.