Projective Geometry and Homogeneous Coordinates

Projective geometry is the study of incidence relations (which points lie on which lines) under the operation of central projection, the process by which a point of light casts shadows. Its central innovation is to complete affine space by adding a "line at infinity" (or hyperplane at infinity) where parallel families of lines are declared to meet, eliminating the special-case clutter of Euclidean geometry. Homogeneous coordinates provide the algebraic engine: a point of the projective plane ${\mathbb{P}}^2$ is an equivalence class $[x:y:z]$ of nonzero triples, where $[x:y:z]=[\lambda x:\lambda y:\lambda z]$ for any $\lambda \ne 0$. The surprising fact is that this enlargement makes geometry simpler, not more complicated: every two distinct lines meet in exactly one point, every two distinct points lie on exactly one line, and a perfect duality swaps the roles of points and lines, turning theorems into pairs of theorems for free.

The Core Insight

A point in the affine plane $(x,y)$ becomes the line through the origin in 3-space passing through $(x,y,1)$. Two triples represent the same projective point iff they are scalar multiples — i.e., they lie on the same line through the origin of ${\mathbb{R}}^3$. Points "at infinity" correspond to lines through the origin lying in the plane $z=0$, which the affine embedding never reaches. Projective geometry is the geometry of lines through the origin of a vector space.

Formal Statement

Projective Space

Let $V$ be an $(n+1)$-dimensional vector space over a field $\mathbb{F}$ (usually $\mathbb{R}$ or $\mathbb{C}$). The projective space $\mathbb{P}(V)={\mathbb{P}}^n(\mathbb{F})$ is the set of one-dimensional linear subspaces of $V$, equivalently:

The equivalence class of $(x_0,x_1,\dots ,x_n)$ is written $[x_0:x_1:\cdots :x_n]$, its homogeneous coordinates. At least one coordinate must be nonzero, and only ratios are meaningful.

Affine Embedding and Points at Infinity

Fix the standard embedding $\iota :{\mathbb{A}}^n\hookrightarrow {\mathbb{P}}^n$ via

The image is the affine chart $\{x_n\ne 0\}$. Its complement, the hyperplane at infinity, is

A direction $(d_1,\dots ,d_n)$ in ${\mathbb{A}}^n$ corresponds to the point at infinity $[d_1:\cdots :d_n:0]$; all parallel lines share this single point.

Lines in the Projective Plane

In ${\mathbb{P}}^2$, a line is the zero set of a nonzero homogeneous linear form:

The triple $[a:b:c]$ is the line's homogeneous coordinates in the dual projective plane $({\mathbb{P}}^2{)}^{\ast }$.

Incidence Duality of the Projective Plane

In ${\mathbb{P}}^2$:

1. Any two distinct points lie on exactly one line.
2. Any two distinct lines meet in exactly one point.

Both are computed by the cross product of homogeneous coordinate vectors:

1. A point $\mathbf{p}=[x:y:z]$ lies on both lines iff it satisfies the linear system

1. This is a homogeneous $2\times 3$ system. Since ${\ell }_1\ne {\ell }_2$ as projective points, the coefficient vectors are linearly independent, so the coefficient matrix has rank $2$.

1. By rank–nullity, the solution space has dimension $3-2=1$: a single line through the origin in ${\mathbb{F}}^3$.

1. A one-dimensional subspace is exactly one point of ${\mathbb{P}}^2$. Explicitly, the solution is spanned by ${\ell }_1\times {\ell }_2$, since the cross product is orthogonal to both coefficient vectors. $■$

Jean-Victor Poncelet | 1788–1867

A French engineer and officer in Napoleon's army, Poncelet was captured during the 1812 Russian campaign and, while imprisoned at Saratov on the Volga, reconstructed projective geometry from memory without books. His Traité des propriétés projectives des figures (1822) revived and systematized Desargues's neglected ideas, established the principle of continuity and duality as guiding methods, and introduced imaginary points (the circular points at infinity) and the line at infinity as full citizens of geometry. He insisted that projective properties — those preserved under projection — were the truly fundamental ones, decoupling geometry from measurement of distance and angle.

Building Intuition

Imagine standing on a long straight road. The two parallel edges of the road visibly converge to a single point on the horizon — the "vanishing point." Projective geometry takes this perceptual fact and makes it literally true: parallel lines meet at a point at infinity, and the horizon is the line at infinity.

The homogeneous coordinate trick is a "lift and project" picture. Place a screen at height $z=1$ above the origin of 3-space. Every point on the screen corresponds to the ray from the origin through it. As you tilt a ray toward horizontal ($z\to 0$), its piercing point on the screen flies off to infinity. The horizontal rays (the plane $z=0$) are the limiting "points at infinity" the screen could never display — but in projective space they are perfectly ordinary points. There is nothing special about which rays are "at infinity"; choosing a different screen orientation relabels which points are finite. This is the deep lesson: infinity is not absolute, it is a choice of chart.

Central projection: affine points are rays through the origin piercing the plane $z=1$; horizontal rays are points at infinity

Worked Examples

Intersection of Two Lines via Cross Product

1. Homogenize (replace the constant term with a $z$-term): ${\ell }_1=[2:3:-6]$, ${\ell }_2=[1:-1:1]$.
2. Compute the cross product:

1. Components:

1. So $\mathbf{p}=[-3:-8:-5]$. Since $z=-5\ne 0$, this is a finite point.
2. Dehomogenize: $x=-3/-5=3/5$, $y=-8/-5=8/5$.

Detecting a Point at Infinity

1. Rewrite: $2x-y+1=0$ and $2x-y-4=0$. Homogenize: ${\ell }_1=[2:-1:1]$, ${\ell }_2=[2:-1:-4]$.
2. Cross product ${\ell }_1\times {\ell }_2$:

1. So $\mathbf{p}=[5:10:0]$. The third coordinate is $0$ — a point at infinity.
2. Simplify direction: $[5:10:0]=[1:2:0]$, the direction $(1,2)$, which is exactly the common slope ($\Delta y/\Delta x=2/1$).

Common Misconceptions

Connections

Projective geometry sits at a crossroads. Algebraically, ${\mathbb{P}}^n$ is the simplest nontrivial object of Algebraic Geometry, where homogeneous polynomials and the Proj construction generalize this entire framework. The map ${\mathbb{F}}^{n+1}\setminus \{0\}\to {\mathbb{P}}^n$ is the prototype of a Quotient Space by a group action, and over $\mathbb{C}$ it underlies Complex Manifolds (${\mathbb{CP}}^n$ is a compact complex manifold). The transformation group ${\mathrm{PGL}}_{n+1}$ links it to Group Theory and the Erlangen Program, in which Felix Klein classified geometries by their symmetry groups. Affine and Euclidean geometry are recovered by restricting projective transformations that fix the hyperplane at infinity, making projective geometry the parent of Affine Geometry. The duality of points and lines anticipates Dual Vector Spaces in linear algebra.

Real-World Applications

1. Computer graphics (3D rendering pipelines). The OpenGL/Vulkan/DirectX rasterization pipeline represents every vertex in homogeneous coordinates $(x,y,z,w)$, so that perspective projection, translation, rotation, and scaling are all single $4\times 4$ matrix multiplications. The perspective divide by $w$ produces the vanishing-point foreshortening of realistic scenes.

1. Computer vision — camera models. The pinhole camera is exactly a projection ${\mathbb{P}}^3\to {\mathbb{P}}^2$ encoded by a $3\times 4$ camera matrix $P=K[R|t]$. Structure-from-motion and 3D reconstruction rely on this.

1. Epipolar geometry and the fundamental matrix. Two-view geometry is governed by ${\mathbf{x}}^{\prime \top }F\mathbf{x}=0$, a bilinear relation in homogeneous image coordinates used in stereo vision and SLAM (robot navigation).

1. Homography / image rectification. A planar scene viewed from two angles relates by a $3\times 3$ homography $H$; used in panorama stitching (e.g., phone "panorama" mode) and document scanners that "deskew" photographed pages.

1. CAD and projective splines. NURBS (Non-Uniform Rational B-Splines), the industry standard for CAD surfaces, use homogeneous (weighted) control points so that conic sections like circles can be represented exactly.

1. Error-correcting codes and combinatorics. Finite projective planes ${\mathbb{P}}^2({\mathbb{F}}_q)$ yield optimal combinatorial designs and perfect difference sets used in coding theory.

1. Robotics and pose estimation. Rigid-body transformations are handled with homogeneous transformation matrices in $SE(3)$, expressing rotation and translation uniformly.

Historical Development

• c. 300 BCE — Euclid & Apollonius. Conics studied as plane sections of a cone; the seed of "projection," though incidence-at-infinity is absent.
• 1435 — Alberti, Della Pittura. Renaissance perspective: the geometry of vanishing points formalized for painters; the first practical projective thinking.
• 1639 — Girard Desargues. Brouillon project introduces points at infinity and proves Desargues's theorem on perspective triangles; largely ignored and nearly lost.
• 1640 — Blaise Pascal. At sixteen, proves the "mystic hexagon" theorem (Pascal's theorem) on conics, a projective result.
• 1822 — Jean-Victor Poncelet. Traité des propriétés projectives des figures founds projective geometry as an independent discipline; principles of duality and continuity.
• 1827 — Möbius. Introduces barycentric (a form of homogeneous) coordinates in Der barycentrische Calcul.
• 1830s — Plücker & Steiner. Plücker develops the analytic/coordinate approach and line coordinates; Steiner the synthetic approach. The duality principle is fully clarified.
• 1871 — Felix Klein. Shows projective geometry contains Euclidean and non-Euclidean geometries as special cases (the Cayley–Klein metrics); 1872 Erlangen Program organizes all geometry by transformation groups.
• 20th century — Algebraic geometry. Hilbert, Zariski, Weil, Serre, and Grothendieck recast projective space as Proj of a graded ring; ${\mathbb{P}}^n$ becomes a foundational scheme.

Subtleties & Limits

Projective space is conceptually clean but harbors genuine depth at its edges. Orientation and topology: ${\mathbb{RP}}^n$ is non-orientable for even $n$ and is the quotient of the $n$-sphere by the antipodal map; ${\mathbb{RP}}^2$ cannot be embedded in ${\mathbb{R}}^3$ without self-intersection (Boy's surface), so visual intuition fails. Metric structure: projective geometry by itself has no notion of distance or angle — these require choosing a quadric "absolute" (the Cayley–Klein construction), and which choice you make selects Euclidean, hyperbolic, or elliptic geometry. Finite fields: the existence and uniqueness of finite projective planes is deeply open — it is a famous theorem (Bruck–Ryser) that no projective plane of order $6$ exists, and whether a projective plane of order $12$ exists is unknown; the nonexistence of order $10$ was settled only by massive computer search in 1989. Higher-dimensional invariants: classifying subvarieties of ${\mathbb{P}}^n$ (the substance of modern algebraic geometry) connects to Hodge theory and motivic questions that remain wide open.

See Also