Algorithm Selection Guide

Algorithm Selection Guide

There is no single universally best public strategy. The appropriate choice depends on the cost structure, horizon size, frequency of repeated solves and whether optimality is required.

Use case	Recommended starting point	Why
General exact ULS	WagelmansGeneralSolver or FedergruenTzurSolver	Strong asymptotic performance for general costs
General exact reference / auditing	WagnerWhitinEvansSolver or ZangwillNetworkSolver	Simple independent architectures
Wagner–Whitin costs / no speculative motive	WagnerWhitinSolver	Linear-time specialization
Very large restricted instances	Linear specialized methods	O(T) when assumptions hold
Many-core experiment	LyuLeeParallelSolver	Parallel predecessor evaluation
Low-memory quadratic reference	WagnerWhitinEvansSolver	O(T) working memory
Memory-for-speed historical WW	SaydamMcKnewFastWagnerWhitinSolver	Precomputed triangular costs
Fast classical heuristic	SilverMealSolver, GroffSolver, PPB family	O(T) heuristics
Baseline MRP policy	LotForLotSolver	Transparent reference policy

Important rule

Applicability precedes speed.

A theoretically faster method can be invalid if its structural assumptions do not match the instance. Public restricted solvers expose applicability checks where appropriate and tests explicitly cover rejection cases.

See Complexity and Applicability for the complete matrix.