Canonical greedy algorithms and dynamic programming

There has been little work on how to construct greedy algorithms to solve new optimization problems efficiently. Instead, greedy algorithms have generally been designed on an ad hoc basis. On the other hand, dynamic programming has a long history of being a useful tool for solving optimization problems, but is often inefficient. We show how dynamic programming can be used to derive efficient greedy algorithms that are optimal for a wide variety of problems. This approach also provides a way to obtain less efficient but optimal solutions to problems where derived greedy algorithms are nonoptimal.