Dynamic Programming State-Space Relaxation for Single-Machine Scheduling

TLDR: The problem of sequencing jobs on a single machine to minimize total cost is considered, and it is shown that the dynamic programming formulation can be relaxed by mapping the state-space onto a smaller state- space and performing the recursion on this smallerstate-space, thereby giving a lower bound.

Abstract: The problem of sequencing jobs on a single machine to minimize total cost is considered. Machine capacity constraints require that, at any time, at most one job is processed. Also, no machine idle-time between processing jobs is allowed. In contrast to most research, it is not assumed that the cost is a non-decreasing function of completion time. A dynamic programming formulation of the problem is presented. Since the number of states required by this formulation is prohibitively large, the possibilities for branch and bound algorithms are explored. It is shown that the dynamic programming formulation can be relaxed by mapping the state-space onto a smaller state-space and performing the recursion on this smaller state-space, thereby giving a lower bound. Techniques for improving this lower bound through the use of penalties and through the use of state-space modifiers are discussed. Computational results are presented for the problem in which each job has a due date, and the objective is to minimize the sum of holding costs for jobs completed before their due date and tardiness costs for jobs completed after their due date.