Saydam-McKnew fast Wagner-Whitin implementation

Public class:

SaydamMcKnewFastWagnerWhitinSolver

Reference:

C. Saydam and M. McKnew (1987), A Fast Microcomputer Program for Ordering Using the Wagner-Whitin Algorithm, Production and Inventory Management Journal 28(4), 15-19.

Publication identity

The author-uploaded article record describes the program as a very fast implementation of the full original Wagner-Whitin algorithm and explicitly contrasts it with Evans (1985): Evans uses less array storage and is recommended when storage is in severe shortage.

ULSAlgorithms preserves that architectural distinction.

Modern implementation

This class materializes every regeneration interval cost before the DP phase.

The triangular matrix is flattened into one contiguous pooled double[], avoiding row objects and preserving sequential memory access.

For period i and terminal period j, the precomputation incrementally updates:

• demand accumulated in the lot;
• delivered unit cost from period i;
• regeneration interval variable cost.

The second phase performs the original forward Wagner-Whitin recursion using only table lookups and additions.

Complexity

• precomputation: O(T²);
• forward DP: O(T²);
• total time: O(T²);
• working memory: O(T²).

This deliberate memory-for-speed trade-off is the principal distinction from WagnerWhitinEvansSolver, which uses O(T) memory.