A hyperplane in N dimensions is invisible. But its representation in parallel coordinates leaves a structured signature: an envelope that reveals the surface's geometry without ever showing the surface itself.
The duality sends points to polylines and lines to convergence points. The same machinery extends to curves, surfaces, and hyperplanes via the envelope.
A smooth curve is the object tangent to a family of lines. Each tangent maps to a convergence point in parallel coordinates. As the tangent rolls along the curve, the convergence point traces a new curve — the envelope.
A hyperplane in is defined by the equation:
Every point on this hyperplane maps to a polyline in parallel coordinates. Unlike random scatter, the constraint forces these polylines into a structured pattern. The pattern depends on the coefficients and the constant , and changing them changes the pattern in predictable ways.
Figure 1. A hyperplane in 3D represented as polylines in parallel coordinates. All 50 sample points satisfy the constraint, so their polylines form a structured family rather than filling the space. Adjust coefficients and constant to see the pattern change.
When all coefficients are equal, the polylines form a symmetric fan. When one coefficient is much larger than the others, the corresponding axis dominates and the polylines pinch through a narrow band on that axis. When a coefficient is negative, the slope of the polyline segments reverses across that axis.
The hyperplane equation is a single linear constraint on N variables — one degree of freedom removed. In parallel coordinates the constraint restricts where polylines can go, producing visible structure where unconstrained data would scatter.
A curve in is more than a set of points. At each point, the curve has a tangent line. Each tangent maps to a convergence point in parallel coordinates. As the tangent point sweeps along the curve, the convergence point traces a dual curve. The original curve is represented by the envelope of this family of convergence points.
Consider a circle of radius centered at the origin. At the point , the tangent line has slope and the equation:
This tangent line maps to a convergence point in parallel coordinates. As sweeps from 0 to , the convergence point traces a closed curve: the envelope representation of the circle.
Figure 2. A circle (left) and its envelope representation in parallel coordinates (right). Drag the red point around the circle to move the tangent line. Its convergence point traces the envelope curve (orange). Press Animate for a full sweep.
One of Inselberg's results: cusps and inflections exchange roles. A cusp in Cartesian space (sharp point) corresponds to an inflection in the envelope, and an inflection (curvature sign change) produces a cusp.
The duality is orientation-independent. At a cusp, the tangent reverses direction and produces a smooth turn in the dual; at an inflection, the tangent sweeps smoothly through a critical angle and produces a cusp.
Figure 3. Cusp-inflection duality. Left: a curve in Cartesian space. Right: its envelope representation in parallel coordinates. A cuspidal cubic (top) produces a smooth inflection in the envelope. A cubic with an inflection (bottom) produces a cusp in the envelope.
Cusp-inflection duality: A cusp in Cartesian space maps to an inflection in the envelope, and an inflection maps to a cusp. This is a theorem, not a visual coincidence. It holds for any smooth curve regardless of position or orientation.
A convex curve has no inflection points, so its envelope has no cusps. A non-convex shape must have inflections somewhere, and they appear as cusps in the envelope.
Parallel coordinates detects convexity visually: smooth envelope → convex; cusps → non-convex.
Figure 4. Convexity detection via envelopes. Left: the shape in Cartesian space. Right: its envelope in parallel coordinates. The ellipse (convex) produces a smooth envelope. The star (non-convex) produces cusps wherever the boundary has inflection points.
The parabola offers the cleanest demonstration of the envelope construction. At the point , the tangent line is:
With unit axis spacing, this tangent line (slope , intercept ) maps to the convergence point:
As varies, the convergence point traces a rational curve. This is the dual of the parabola.
Figure 5. The parabola and its envelope representation. Left: the parabola with one highlighted tangent line. Right: the family of convergence points tracing the dual curve, with the current convergence point highlighted. Drag the red point on the parabola, the convergence point in PC, or the slider.
Everything above generalizes. A smooth surface in has a tangent plane at each point. Each tangent plane is a hyperplane, and hyperplanes map to structured polyline families in parallel coordinates. The envelope of these families represents the surface.
A smooth hypersurface in has a tangent hyperplane at each point. The family of tangent hyperplanes, represented in parallel coordinates, produces an envelope that encodes the hypersurface's geometry. Curvature, convexity, and singularities all have signatures in this envelope.
The hierarchy: points → polylines, lines → convergence points, curves → envelopes of convergence points, surfaces → envelopes of polyline families. At every level the duality preserves geometric information while changing the representation.
The envelope construction extends the parallel-coordinates representation from discrete data to continuous geometry. The next explainer, Axis Order Is Everything, covers how dimension ordering determines which relationships are visible.