A practical anti-cycling procedure for linearly constrained optimization (Q1264089)

An anti-cycling procedure in active-set methods for general linearly constrained problems, including the simplex method, is described in this paper. Two features of the procedure are crucial: controlled infeasibility of all variables (including nonbasic) and a ``working'' feasibility tolerance that increases slightly and consistently through an extended sequence of iterations. The additional computation per iteration is nominal, and ``stalling'' cannot occur with exact arithmetic. The paper reports the authors' computational results and the method appears to be reliable.