Figure 4

Predecessor and successor functions, and equivalence classes of columns. The predecessor and successor functions (f _P and f _S respectively) map from columns (edges) to nodes (circles) in the dynamic programming matrix. All columns mapping to a particular node under f _P share the same set of possible predecessor columns, and are grouped together in an equivalence class, denoted by E _P (shown in red). An analogous definition holds for E _S (blue).