Projective Polarity Formulation

• Projective polarity formulation is defined as a reciprocal correlation mapping points to hyperplanes via nondegenerate bilinear forms and symmetric matrices, establishing duality in projective spaces.
• The framework extends to quadratic, conic, and higher-order forms, enabling applications from algebraic geometry to computational imaging.
• Advanced matrix models and Clifford algebra approaches reveal that every quadratic polarity is a collineation of the canonical Legendre polarity, providing insights for geometric classification and imaging algorithms.

A projective polarity is a fundamental concept in projective geometry, establishing a reciprocal, often involutive, correspondence between points and hyperplanes (or more generally, between subspaces of complementary dimension), defined by quadratic or higher-degree forms, correlation, or incidence--geometric conditions. The algebraic, combinatorial, and geometric structure of projective polarities underlies a wide class of dualities, tangent constructions, divergence functions, and applications from algebraic geometry to computational imaging.

1. Core Definitions and Algebraic Framework

A projective polarity on a projective space $P^n$ over a field $K$ is classically defined as an involutive correlation, that is, a bijection $\Phi:P^n \to (P^n)^\vee$ mapping points to hyperplanes and reversing incidence: for any point $P$ and hyperplane $H$, $P \in H$ if and only if $\Phi(H) \in \Phi(P)$. Such a polarity arises canonically from a nondegenerate symmetric bilinear form $B:V \times V \to K$ on the vector space $V$ underlying $P^n$. Via a choice of basis, this is encoded by an invertible symmetric matrix $K$0. For any point $K$1, its polar hyperplane is the loci $K$2; for a hyperplane $K$3, its pole is $K$4 (Kloeckner, 2013, Piene, 2016, Nielsen et al., 5 Mar 2026). The polarity map is involutive up to scale.

The definition extends: quadratic forms yield conic polarities; higher-degree forms, e.g., cubics, induce higher polar maps; and the notion admits generalization to reciprocal polarities associated with arbitrary quadrics, harmonic polarities with respect to triangles, and convex duality (Kloeckner, 2013). In homogeneous coordinates, any symmetric $K$5 matrix $K$6 on $K$7 defines a quadratic polarity on $K$8 via the bilinear form $K$9 (Nielsen et al., 5 Mar 2026).

2. Reciprocal and Generalized Projective Polarities

Given a nondegenerate quadric $\Phi:P^n \to (P^n)^\vee$0, one obtains a projective reciprocal polarity (sometimes termed projective orthogonality or Euclidean polarity in projective space) (Piene, 2016). The reciprocal polarity associates to any point $\Phi:P^n \to (P^n)^\vee$1 its polar hyperplane $\Phi:P^n \to (P^n)^\vee$2, defined by $\Phi:P^n \to (P^n)^\vee$3, where $\Phi:P^n \to (P^n)^\vee$4 and $\Phi:P^n \to (P^n)^\vee$5 defines the quadric.

This structure extends to singular varieties and higher-order subspaces, leading to the theory of reciprocal polar varieties. The main features:

• The polarity defined by a quadric is an involutive isomorphism between points and hyperplanes.
• For an irreducible variety $\Phi:P^n \to (P^n)^\vee$6, reciprocal polar varieties $\Phi:P^n \to (P^n)^\vee$7 are defined by singularity/transversality conditions involving the polar subspaces $\Phi:P^n \to (P^n)^\vee$8 (Piene, 2016).
• Classical polarities are recovered as particular cases; e.g., the polarity with respect to a conic in $\Phi:P^n \to (P^n)^\vee$9.

Generalized polarities are formalized as linear (possibly partial) maps $P$0 from $P$1-subspaces to hyperplanes, satisfying linearity and reciprocity conditions. The “null property” ($P$2) ensures reciprocity: $P$3. Each generalized null polarity corresponds canonically to a linear complex of $P$4-subspaces, and vice versa, via a bijection (Havlicek et al., 2013). This perspective underlies the algebraic-geometric structure of line and plane complexes in higher-dimensional spaces.

3. Polarity, Polar Maps, and Gauss Maps of Hypersurfaces

For hypersurfaces $P$5, the classical polar map, or gradient map, is

$P$6

Its base locus corresponds to $P$7, the set of singular points. The polar degree $P$8 is the topological degree of this map. For reduced hypersurfaces with isolated singularities, $P$9, with $H$0 the Milnor number (Fassarella et al., 2011). There are also cohomological and Chern class formulas:

$H$1

and a celebrated Euler characteristic formula $H$2 for a generic hyperplane $H$3. These invariants control not only the geometry of the hypersurface but also enumerative quantities such as the Euclidean distance degree (EDdeg).

In the case of projective varieties of arbitrary dimension $H$4, classical polar varieties are subvarieties defined via the Gauss map and Schubert cycles. Their degrees (“ranks”) connect to Chern-Mather theory and singularity corrections, providing explicit counts of critical loci, dual/focal loci, and caustics (Piene, 2016).

4. Matrix Models and New Categorical Structures

Recent advances interpret polarities in terms of matrix manipulations in homogeneous and projective spaces. Any quadratic polarity on $H$5 is encoded by a symmetric invertible $H$6 matrix $H$7, defining the bilinear pairing $H$8 and the associated polar hyperplane $H$9 (Nielsen et al., 5 Mar 2026). Any $P \in H$0 can be decomposed as a congruence transformation $P \in H$1 of the canonical Legendre polarity $P \in H$2, yielding:

$P \in H$3

where $P \in H$4 is the Legendre polarity. Thus, all quadratic polarities are collineations of the canonical Legendre polarity, and their action on convex bodies, functions, or epigraphs can be expressed as deformed Legendre transform duality.

In addition, Clifford algebra models enable factorization of every projective collineation in, e.g., $P \in H$5 as a product of at most six null polarities, with explicit matrix correspondence between null vectors and skew-symmetric $P \in H$6 matrices generating the full collineation group (Klawitter, 2014).

5. Polarity in Projective Imaging and Computational Applications

In computational imaging, notably polarization imaging and shape-from-polarization (SfP), projective polarity formulations arise in modeling how light's polarization state is transformed by projective cameras. The geometric model for polarization imaging on projective cameras (Pistellato et al., 2022) establishes a per-pixel polarization transformation pipeline. Each pixel is assigned a local frame adapted to the incident ray direction; the transmission axes of micro-polarizers and effective orientation are adjusted via projective rotation matrices. In this model, the measurement process is described by a per-ray Mueller matrix, with Stokes vectors reconstructed via a Moore-Penrose pseudoinverse.

Critically, normal vectors estimated in the per-ray frame are rotated back to the camera frame, closing the projective loop. Any conventional SfP pipeline based on orthographic assumptions may thus be wrapped with pre/post processing to correctly handle perspective/projective effects, enforce physical constraints, and operate without bias on DoFP demosaiced data (Pistellato et al., 2022).

6. Unified Divergences, Dualities, and Information Geometry

Quadratic polarity models can be interpreted as generalizing the classical Legendre–Fenchel duality and its induced divergences to the projective setting (Nielsen et al., 5 Mar 2026). Given $P \in H$7, the associated polar Fenchel–Young divergence is

$P \in H$8

and for convex functions,

$P \in H$9

with $\Phi(H) \in \Phi(P)$0 the $\Phi(H) \in \Phi(P)$1-conjugate. Special choices of $\Phi(H) \in \Phi(P)$2 recover standard Fenchel-Young, Bregman, and total Bregman divergences. Reference duality relations and normalization (polar conformal factors) extend these divergences further, exposing deep projective dualities within information geometry (Nielsen et al., 5 Mar 2026). Remarkably, such constructions reduce to Legendre polarity under canonical $\Phi(H) \in \Phi(P)$3 and to Euclidean, spherical, or hyperbolic divergences under appropriate $\Phi(H) \in \Phi(P)$4.

7. Classification Results, Combinatorics, and Structural Insights

Polarity constructions provide structural insight into the classification of projective varieties. For instance, the degree of the polar map for plane curves (the polar degree) allows full classification of homaloidal plane curves (those with polar degree 1): three non-concurrent lines, a smooth conic, and a smooth conic plus a tangent line (Fassarella et al., 2011). The inclusion–exclusion principle, together with the Euler characteristic formula, extends this classification to reducible curves and arrangements of hyperplanes, where the polar degree encodes the combinatorics of intersections.

From a combinatorial viewpoint, generalized polarities with the null property correspond uniquely to linear complexes of subspaces and induce partitions (or spreads) of lines in projective space, with vivid implications for finite geometry and incidence structures (Havlicek et al., 2013). In the setting of Klein’s quadric, null polarities generate the entire projective group, with applications to transformation classification and algebraic generation of geometric symmetries (Klawitter, 2014).

---

References:

• (Pistellato et al., 2022): "A Geometric Model for Polarization Imaging on Projective Cameras"
• (Piene, 2016): "Polar varieties revisited"
• (Havlicek et al., 2013): "Incidence and Combinatorial Properties of Linear Complexes"
• (Nielsen et al., 5 Mar 2026): "Quadratic polarity and polar Fenchel-Young divergences from the canonical Legendre polarity"
• (Fassarella et al., 2011): "On the polar degree of projective hypersurfaces"
• (Kloeckner, 2013): "Polarités définies par un triangle"
• (Klawitter, 2014): "Null Polarities as Generators of the Projective Group"