Geometric duality and arrangements

A primal point lies on a primal line exactly when its coordinates satisfy that line equation. Substituting the dual coordinates yields the same scalar identity, so collinear primal points correspond to concurrent dual lines. The converse follows by reversing the calculation, provided all relevant objects remain in the chart. Parallel primal lines become dual points with the same first coordinate; attempting to dualize their common point at infinity is precisely where an affine-only treatment stops.

The same algebra determines order. If a primal point lies above a line, the line's dual point lies above the point's dual line under our convention. The objects have exchanged roles, so an informal slogan about preserving or reversing above and below can be misleading. Always write the inequality with the actual map. Euclidean distances and angles are generally not preserved, and a nearest-neighbor problem does not become an equivalent nearest-neighbor problem merely by dualizing its input.

The arrangement of lines consists of their intersection vertices, edge pieces, and complementary faces. For n distinct lines in general position, no two are parallel and no three are concurrent. Each new line meets all earlier lines at distinct points, dividing it into as many pieces as its insertion index. Each piece splits one existing face, giving the region-count recurrence below. The number of vertices is the number of unordered line pairs, so storing the full arrangement takes quadratic space in the worst case.

This quadratic complexity may be essential for an arrangement task but excessive for a particular query. A hull, envelope, or selected level can have much smaller relevant output than the complete decomposition. Duality helps reveal such a reduction, yet the algorithm must show that its reduced structure answers the original question. Constructing every dual-line intersection first can destroy an intended output-sensitive advantage. Count the features actually requested instead of assuming that a change of viewpoint makes a large intermediate object disappear.

Concurrent lines share one intersection rather than contributing a different vertex for each pair. Parallel lines contribute no finite intersection with each other. Both phenomena reduce the simple-arrangement count, although the worst-case quadratic bound remains valid for arbitrary distinct lines. Duplicate lines need a representation policy because two identical supporting lines do not define separate geometric edges. A perturbation may restore generic counts for a nearby arrangement, but it does not make those counts describe the original unperturbed one.

Exact comparisons also matter when duality is implemented. A dual coefficient copied from an input coordinate can be exact while the intersection of two dual lines requires a rational construction. Supporting-line and segment problems differ because a segment additionally imposes parameter bounds; dualizing its supporting line alone loses that finite extent. Preserve every constraint involved in the original problem and verify the reverse mapping of the answer. Otherwise a formally correct incidence transformation can still solve the wrong computational task.