Polar Vertices Overview

• Polar vertices are defined via polarity and duality, serving as key points in convex polytopes, discrete spherical polarization, and quantitative geometric theorems.
• In optics and computer vision, they denote unique geometric features—such as polarization-singularity C-points or polar coordinate vertices—that enable precise pattern characterization.
• Across finite geometry, planetary science, and ferroelectric studies, polar vertices represent observable and manipulable structures critical for understanding system behaviors.

The cited literature suggests that polar vertices is a polysemous research term rather than a single standardized object. In convex geometry it refers to vertices of a shifted polar polytope and to a polarity construction used in a quantitative Steinitz theorem (Ivanov, 2024). In discrete polarization on the sphere it denotes the points of $\pm w_*$ at which the $f$-potential of a regular simplex attains its minima (Borodachov, 2020). In optics, a narrative exposition applies the label to polarization-singularity C-points in tailored beams (Galvez et al., 2014). In computer vision, it designates polygon vertices represented by angle–distance pairs in polar coordinates (Zheng et al., 2023). In finite geometry, the vertices of a polar graph are isotropic one-dimensional subspaces of a vector space with respect to a non-degenerate form (Pantangi et al., 2017). In oxide ferroelectrics, isolated three-fold polar vertices are electrically manipulable topological structures (Li et al., 2021). Related planetary-atmospheric usage concerns the six vertices of Saturn’s north-polar hexagon (Fletcher et al., 2014).

1. Terminological scope and unifying motifs

Across these literatures, the common element is not a single ontology but a recurring reliance on polarity, polar coordinates, or polar topology. The convex-geometric usage is duality-theoretic: given a polytope $P$ and an interior point $c$, one studies the vertices of $(P-c)^\circ$ (Ivanov, 2024). The spherical-polarization usage is potential-theoretic: for a regular simplex $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$, the minima of the $f$-potential occur at $\pm w_*$, called “polar vertices” (Borodachov, 2020). The computer-vision usage is coordinate-parametric: a polygon is represented by an origin $O=(o_x,o_y)$ and vertices $(r_i,\theta_i)$ in Polar form (Zheng et al., 2023).

A different cluster of meanings is topological or structural. In the optics exposition, polarization-singularity C-points are described in terms of local Stokes fields, half-integer index, and lemon/star/monstar morphology (Galvez et al., 2014). In Pantangi–Sin’s graph-theoretic setting, “polar vertices” are the vertices of polar graphs, namely isotropic lines in finite vector spaces endowed with symplectic, orthogonal, or Hermitian forms (Pantangi et al., 2017). In Li et al., isolated three-fold polar vertices arise at the PbTiO$f$0/SrRuO$f$1 interface under incomplete screening and are characterized by a winding number $f$2 (Li et al., 2021). This suggests that the phrase functions as a domain-specific shorthand whose precise meaning is determined by the ambient formalism.

2. Convex-geometric polarity and the quantitative Steinitz theorem

In Ivanov’s formulation, let $f$3 be a convex polytope in $f$4 with nonempty interior and let $f$5. The shifted polytope is

$f$6

and for a convex set $f$7 containing the origin, the polar is

$f$8

Hence

$f$9

The central statement is Theorem 3.1: if $P$0 has facets $P$1 and $P$2, then there exists a unique point

$P$3

where $P$4 is the vertex of $P$5 dual to the facet $P$6 of $P$7. Setting all $P$8 yields the corollary

$P$9

The proof is based on the strictly concave maximization problem

$c$0

with

$c$1

and on the gradient identity

$c$2

At the unique maximizer $c$3,

$c$4

and the facet–vertex correspondence gives

$c$5

These are the “balanced vertices” of the shifted polar polytope (Ivanov, 2024).

The polarity construction is then combined with an “Atlantis” lemma. If $c$6, $c$7, $c$8, $c$9, and $(P-c)^\circ$0, then for any subset of vertices $(P-c)^\circ$1 satisfying

$(P-c)^\circ$2

the corresponding vertices $(P-c)^\circ$3 satisfy

$(P-c)^\circ$4

and since $(P-c)^\circ$5 one takes $(P-c)^\circ$6 (Ivanov, 2024).

This machinery yields the quantitative Steinitz bound. If $(P-c)^\circ$7 is finite, $(P-c)^\circ$8, and $(P-c)^\circ$9, then there are at most $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$0 points of $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$1 whose convex hull contains

$w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$2

If $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$3, then

$w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$4

so one can remove all but $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$5 vertices from $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$6 while still containing $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$7. In the abstract, the corresponding formulation is: if the number of vertices of $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$8 scales linearly with the dimension, i.e. $w_*=\{v_0,\dots,v_d\}\subset S^{d-1}$9, then one can select $f$0 vertices such that $f$1 (Ivanov, 2024). The significance of the result lies in how polarity turns a sparsification problem for $f$2 into a centroid-type balancing problem on $f$3, extending the classical Steinitz theorem and the earlier quantitative result of Bárány, Katchalski, and Pach.

3. Polar vertices in maximal discrete polarization on the sphere

For the maximal discrete polarization problem, let $f$4, let $f$5, and let

$f$6

where $f$7 is continuous in the extended sense on $f$8, monotone non-increasing on $f$9, and convex on $\pm w_*$0, with either concave or convex derivative $\pm w_*$1. The quantity of interest is

$\pm w_*$2

and the $\pm w_*$3-point $\pm w_*$4-polarization constant is

$\pm w_*$5

Borodachov proves that the configuration of the vertices of a regular $\pm w_*$6-simplex inscribed in $\pm w_*$7, denoted $\pm w_*$8, is optimal (Borodachov, 2020).

The two main formulas depend on the curvature of $\pm w_*$9. If $O=(o_x,o_y)$0 is concave on $O=(o_x,o_y)$1, then for every $O=(o_x,o_y)$2-point $O=(o_x,o_y)$3,

$O=(o_x,o_y)$4

If $O=(o_x,o_y)$5 is convex on $O=(o_x,o_y)$6, then

$O=(o_x,o_y)$7

Strict convexity of $O=(o_x,o_y)$8 on $O=(o_x,o_y)$9 forces uniqueness at $(r_i,\theta_i)$0. The proof introduces the auxiliary function

$(r_i,\theta_i)$1

and shows that $(r_i,\theta_i)$2 increases on $(r_i,\theta_i)$3, decreases on $(r_i,\theta_i)$4, and satisfies

$(r_i,\theta_i)$5

The terminology “polar vertices” enters through the location of the minima of the simplex potential. Proposition 6.1 and Theorem 2.4 show that the $(r_i,\theta_i)$6-potential of $(r_i,\theta_i)$7 attains its minima exactly at the points of $(r_i,\theta_i)$8, called “polar vertices,” with

$(r_i,\theta_i)$9

or

$f$00

The comparison with an arbitrary $f$01 splits into the cases where $f$02 is contained in a closed hemisphere and where $f$03 contains $f$04 (Borodachov, 2020).

Low-dimensional instances recover known extremals. For $f$05, the optimal configuration on $f$06 is the equilateral triangle, and

$f$07

Its polar vertices are the three antipodes of the triangle’s vertices. For $f$08, the optimal configuration on $f$09 is the regular tetrahedron, with

$f$10

As a byproduct, the same geometric lemmas imply that among all $f$11-point subsets of $f$12, the regular simplex uniquely minimizes the covering radius

$f$13

and

$f$14

Here the “polar” designation is antipodal and extremal rather than duality-theoretic.

4. Polarization-singularity C-points in optics

In the optics exposition of Galvez et al., polarization-singularity C-points are treated as “polar vertices” of a polarization-spatial field (Galvez et al., 2014). At each point in an optical beam the state of polarization is represented by the Stokes vector

$f$15

with

$f$16

Near a C-point, a very general local expansion for the normalized Jones vector in polar coordinates $f$17 is

$f$18

To first order in $f$19,

$f$20

$f$21

The local ellipse-orientation angle and ellipticity angle are

$f$22

In practice one plots the “Stokes-field” $f$23.

Classification is by half-integer topological index

$f$24

Lemons have $f$25 and exactly one angle $f$26 at which $f$27; stars have $f$28 and three such radial lines; monstars have $f$29 like lemons, but three radial lines like stars. The radial-line condition produces a cubic in $f$30, whose number of real roots is $f$31 or $f$32. At $f$33, the cusp boundary between lemon and monstar occurs at

$f$34

and more generally the lemon–monstar boundary satisfies

$f$35

A convenient global representation is the “C-point sphere,” with polar coordinates $f$36 on a unit sphere, whose three regions are lemon, monstar, and star (Galvez et al., 2014).

Galvez et al. describe two complementary generation methods. The first is interferometric creation by superposing two circularly polarized Laguerre–Gaussian modes in a Mach–Zehnder-type set-up, with one arm encoded on a spatial light-modulator with

$f$37

and the relative phase $f$38 set by a Pancharatnam–Berry phase shifter. The second is an in-line fiber-mode converter obtained by skew-launching into a few-mode optical fiber, so that at the output

$f$39

In both cases imaging polarimetry records six intensity frames through polarizers at $f$40, right- and left-circular, from which the full Stokes vector is reconstructed pixel by pixel. In the LG-based experiment, monstars at $f$41 with radial lines at $f$42 and at $f$43 with $f$44 were distinguished from the adjacent lemon and star; agreement with theory was within a few degrees. The reported applications include high-sensitivity interferometry and sensing, vector-beam microscopy and imaging, optical trapping and micromanipulation, quantum information, and fundamental studies of natural polarization singularities (Galvez et al., 2014).

5. Polar vertices as polygon parameters in computer vision

In DPPD, “polar vertices” are the sparse set of flexible vertices used to construct a polygon representation of an object (Zheng et al., 2023). Each object is represented by a center, or polygon “origin,”

$f$45

and by $f$46 vertices in Polar form

$f$47

The Cartesian coordinates are

$f$48

and conversely,

$f$49

The network is anchor-free, NMS-free, and set-prediction style: it produces $f$50 candidate detections, matches them to $f$51 ground truths via Hungarian matching, and applies classification and regression loss on the assigned ones (Zheng et al., 2023).

The regression head outputs a $f$52-dimensional vector

$f$53

Origin offsets relative to the grid-cell center $f$54 are decoded as

$f$55

where $f$56 are the cell stride and $f$57 is sigmoid. Radial distances are

$f$58

where $f$59 is a scale prior. Angles are predicted by un-normalized “angle deltas”:

$f$60

then

$f$61

This ensures

$f$62

and enforces a valid polygon ordering. The paper explicitly contrasts this with PolarMask, which fixes $f$63 and only regresses $f$64 (Zheng et al., 2023).

Training uses differentiable dense resampling. One chooses a large $f$65 such as $f$66 of equally spaced rays at angles

$f$67

and samples both the ground-truth polygon and the predicted polygon to obtain two $f$68-length radial-distance vectors. For each ray $f$69, if $f$70 and $f$71 are neighboring vertices with $f$72, then with

$f$73

the interpolated radius is

$f$74

The computation is fully differentiable, so gradients from any loss on $f$75 flow back into both the predicted $f$76 and $f$77. The regression objective is

$f$78

with origin loss

$f$79

polar-IoU loss

$f$80

and smoothness loss

$f$81

$f$82

At inference, the dense resampling machinery is dropped. For each surviving candidate, DPPD decodes $f$83, converts each vertex to Cartesian coordinates, and assembles the $f$84-vertex polygon in ascending $f$85 order. The paper states that these $f$86 vertices, “typically as few as 12 or 24,” already approximate the object shape very accurately. DPPD is reported on traffic-sign, crosswalk, vehicle, and pedestrian detection for autonomous driving (Zheng et al., 2023). Here the adjective “polar” is strictly coordinate-theoretic.

6. Vertices of polar graphs in finite geometry

Pantangi and Sin use “polar vertices” in the literal graph-theoretic sense: the vertices of a polar graph are the isotropic one-dimensional subspaces of a finite vector space with respect to a non-degenerate form (Pantangi et al., 2017). Let $f$87 be a finite vector space over $f$88 or $f$89 endowed with a symplectic, orthogonal, or Hermitian form $f$90. A one-dimensional subspace $f$91 is isotropic, or singular, if

$f$92

The vertex set is

$f$93

and the notation $f$94 is used for the vertex $f$95. Two distinct vertices $f$96 are adjacent precisely when

$f$97

The number of vertices is expressed through the Witt index $f$98, a shift parameter $f$99, and $P$00 in the orthogonal and symplectic cases or $P$01 in the Hermitian case:

$P$02

Hence

$P$03

The six classical families are listed as $P$04, $P$05, $P$06, $P$07, $P$08, and $P$09, with the corresponding values of $P$10 and $P$11 specified in the paper (Pantangi et al., 2017).

Each $P$12 is a strongly regular graph with parameters

$P$13

$P$14

$P$15

$P$16

Writing $P$17 for the adjacency matrix and $P$18 for the Laplacian, the spectra are

$P$19

$P$20

with

$P$21

$P$22

and

$P$23

These data feed into the computation of the Smith group

$P$24

and the critical group

$P$25

By Kirchhoff’s Matrix-Tree theorem,

$P$26

The analysis separates the non-nilpotent case, where a prime $P$27 divides at most two of the relevant eigenvalues and rank arguments apply, from the nilpotent case, where $P$28 or $P$29 is nilpotent and the submodule structure of the permutation module $P$30 under the form-preserving group $P$31 is used. The paper states that the end result is twelve uniform tables—six for $P$32 and six for $P$33—giving a complete determination of the Smith normal form of $P$34 and $P$35 in each family (Pantangi et al., 2017). In this setting, “polar vertices” are the combinatorial carriers of orthogonality geometry.

7. Planetary and ferroelectric topological structures

In Saturn studies, the relevant objects are the six vertices of the north-polar hexagon rather than a formally defined class called “polar vertices.” Cassini/CIRS thermal-infrared maps show a warm belt encircling the pole at planetographic latitude $P$36, and the six nearly equally spaced vertices of the hexagon appear as local maxima in brightness temperature $P$37 when one steps around System III longitude (Fletcher et al., 2014). Their longitudes are extracted by sampling $P$38 at fixed latitude $P$39 and fitting

$P$40

A representative time series from $P$41 to $P$42 gives vertex longitudes shifting from $P$43 to $P$44, corresponding to a westward drift of $P$45 over $P$46:

$P$47

The conversion from observed longitude to System III longitude is

$P$48

with

$P$49

The paper interprets the drift as evidence that the hexagon is a nearly stationary barotropic Rossby wave trapped in the prograde jet at $P$50 rather than a feature rigidly locked to System III (Fletcher et al., 2014).

In ferroelectric oxide thin films, Li et al. study isolated three-fold polar vertices at the PbTiO$P$51/SrRuO$P$52 interface (Li et al., 2021). The film stack comprises a $P$53-DyScO$P$54 substrate, a $P$55 SrRuO$P$56 bottom electrode, a $P$57 PbTiO$P$58 layer, and a diffused $P$59 SrTiO$P$60 “transition” layer of approximately $P$61 unit cells at the interface. Atomic-resolution EDS/HAADF indicates that Ti diffuses into SrRuO$P$62, substituting Ru and spontaneously forming an insulating SrTiO$P$63 slab; within approximately $P$64 unit cells of the interface the out-of-plane Ti–Pb displacements are strongly suppressed, and the transition layer itself shows approximately $P$65 residual displacement, indicating incomplete screening. Phase-field comparison shows that only with the $P$66 SrTiO$P$67 does a three-fold vertex spontaneously form at the intersection of an incoming $P$68 domain wall and the interface (Li et al., 2021).

Under in situ biasing, the experimental sequence is: initial $P$69-domains with downward out-of-plane polarization, application of negative tip bias such as $P$70, nucleation of an upward-polarized domain under the tip, propagation of a $P$71 domain wall toward the SrTiO$P$72 layer, insertion of a triangular $P$73 sector when the wall reaches the imperfectly screened interface, and creation of two three-fold vertices. Reversing the bias drives the wall and its attached vertices back and eventually annihilates them. The local free-energy density is written in Landau–Ginzburg–Devonshire form as

$P$74

and the onset of three-fold vertex nucleation occurs at approximately $P$75 across $P$76, giving

$P$77

The maximum observed velocity is approximately $P$78 under $P$79, and over the accessible field range the velocity is approximately linear in field:

$P$80

with

$P$81

The phase-field dynamics obey

$P$82

and the winding number around an isolated three-fold vertex is

$P$83

The reported implications include movable topological bits, analogy to racetrack memory, and potential for ultrahigh-density nonvolatile memories if integrated with nanoscale electrodes (Li et al., 2021).

Taken together, these planetary and ferroelectric examples show that “vertex” language at a pole or in a polar medium may denote either geometrically identifiable extrema of a large-scale pattern or nanoscale topological defects whose nucleation and propagation are field controlled.