Discrete curvature

Ivan Izmestiev · 2025

Read Discrete curvature by Ivan Izmestiev (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-09353.txt.

Introduction

Ivan Izmestiev surveys curvature for faceted shapes: polygons, polyhedral surfaces, and abstract polyhedral spaces of any dimension. Smooth curvature needs calculus, yet corner-based versions have existed for centuries in disguise, from Descartes onward. The survey shows how these discrete notions echo the classic theorems of smooth geometry, sometimes word for word, and where the analogy strains.

The Theorema Egregium, the Gauss-Bonnet theorem, and its Chern generalization all have discrete twins. Convergence results also appear: refine a polyhedral approximation under the right assumptions and its curvature tends to the smooth one. The survey ties a scattered field into one story and points to open ground.

Catalog record

CREATORIvan Izmestiev
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-09353

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Discrete curvature Ivan Izmestiev Abstract . The combination of words “discrete curvature” is only an apparent contradiction. In this survey article we describe curvature notions associated with polygons, polyhedral surfaces, and with abstract polyhedral manifolds. Several theorems about the discrete curvature are stated that repeat literally classical theorems of differential and Riemannian geometry: Theorema Egregium, Gauss–Bonnet theorem, and the Chern–Gauss–Bonnet theorem among the others. Some convergence results are also mentioned: under certain assumptions the discrete curvature tends to the smooth curvature as a smooth object is approximated by polyhedral ones. 1. Curvature of curves and polygons 1.1. The total curvature of planar curves The curvature of a smooth curve in the plane is the rotation speed of its tangent relative to the arc length. Let \gamma\colon[a,b]\to\mathbb{R}^{2} be an arc-length parametrized curve: \|\dot{\gamma}(t)\|=1 for all t\in[a,b] . Then the acceleration vector \ddot{\gamma} is orthogonal to \dot{\gamma} because of \langle\dot{\gamma},\ddot{\gamma}\rangle=\frac{1}{2}\frac{d}{dt}\langle\dot{\gamma},\dot{\gamma}\rangle=0. Thus the norm of \ddot{\gamma} is equal to the rotation speed of the tangent, and one has the following formula for the curvature: \kappa:=\|\ddot{\gamma}\|. Let us equip the curvature with a sign: plus if the curve turns left, and minus if the curve turns right. Formally, choose a unit normal \nu at every point of the curve so that the orthonormal basis (\dot{\gamma},\nu) is positively oriented. Then the signed curvature \kappa^{s} of \gamma is obtained from \ddot{\gamma}=\kappa^{s}\nu. Now consider a polygon C_{1}\ldots C_{n} in the plane. A point moving along a polygonal line changes the direction of its motion abruptly at the vertices. Let us assume that the direction is never changed to its opposite. Then at each vertex one has the turning angle \kappa^{s}_{i}=\angle(\overrightarrow{C_{i-1}C_{i}},\overrightarrow{C_{i}C_{i+1}})\in(-\pi,\pi), and we call this angle the signed curvature of the polygon at the vertex C_{i} . Figure 1 illustrates the definitions of the signed curvature in the smooth and the discrete situation. Figure 1. The signed curvature of curves and polygons. Theorem 1.1 . The total signed curvature of a closed plane curve is an integer multiple of 2\pi : \int_{a}^{b}\kappa_{s}\,dt=2\pi k,\quad\text{respectively}\quad\sum_{i=1}^{n}\kappa_{i}^{s}=2\pi k. Proof. For a smooth curve parametrized with a unit speed there is a continuous function \alpha\colon[a,b]\to\mathbb{R} such that \dot{\gamma}(t)=(\cos\alpha(t),\sin\alpha(t)) . One has \ddot{\gamma}=\dot{\alpha}(-\sin\alpha,\cos\alpha)=\dot{\alpha}\nu, which implies \kappa^{s}=\dot{\alpha} . Therefore \int_{a}^{b}\kappa^{s}\,dt=\int_{a}^{b}\dot{\alpha}\,dt=\alpha(b)-\alpha(a), which is an integer multiple of 2\pi because \dot{\gamma}(b)=\dot{\gamma}(a) . In the discrete case the sum of the signed curvatures is the total turning angle of the directed edges of the polygon as one makes a closed tour. The direction of the motion is at the end the same as in the beginning, but the angle which describes the direction is well-defined only modulo 2\pi . Hence the total signed curvature is an integer multiple of 2\pi . ∎ If the total curvature is equal to 2\pi k , then the number k is called the turning number or the rotation index of the curve. Figure 2 shows two examples. Figure 2. A smooth curve with the turning number \pm 2 and a polygon with the turning number 0 . The turning number of a simple (that is, non-self-intersecting) closed curve is \pm 1 , where the sign depends on the orientation of the curve. Intuitively this is clear; a proof in the smooth case was given by Hopf [ 15 ] , see also [ 16 ] . 1.2. The total curvature of spatial curves For an oriented plane curve it was possible to equip the curvature with a sign according to whether the velocity vector turns left or right as a point moves along the curve. For a space curve there is no left and right, therefore we define the curvature as a non-negative quantity in both smooth and the discrete setting: \displaystyle\kappa \displaystyle:=\|\ddot{\gamma}\|\ \text{ for a unit-speed curve,} \displaystyle\kappa_{i} \displaystyle:=\pi-\angle C_{i-1}C_{i}C_{i+1}\ \text{ for a polygon}. Theorem 1.2 (Fenchel [ 10 ] ) . The total curvature of a closed space curve is bigger or equal 2\pi : \int_{a}^{b}\kappa\,dt\geq 2\pi,\quad\text{respectively}\quad\sum_{i=1}^{n}\kappa_{i}\geq 2\pi. The equality takes place if and only if the curve is planar and convex. For the proof we will need a definition and several lemmas. Definition 1.3 . The tangent indicatrix of a smooth space curve \gamma\colon[a,b]\to\mathbb{R}^{3} is a curve in the unit sphere \mathbb{S}^{2}=\{x\in\mathbb{R}^{3}\mid\|x\|=1\} given by T_{\gamma}\colon[a,b]\to\mathbb{S}^{2},\quad T_{\gamma}(t)=\frac{\dot{\gamma}(t)}{\|\dot{\gamma}(t)\|}. The tangent indicatrix of a space polygon C_{1}\ldots C_{n} is the polygon in \mathbb{S}^{2} obtained by connecting consecutive points of the sequence T_{i}=\frac{C_{i+1}-C_{i}}{\|C_{i+1}-C_{i}\|},\quad i=1,\ldots,n-1 by arcs of great circles. If the polygon is closed, one considers the indices modulo n , introduces the point T_{n} and connects it with T_{1} and T_{n-1} . Figure 3. The tangent indicatrix of a space polygon. Lemma 1.4 . The total curvature of a space curve is equal to the length of its tangent indicatrix. Proof. In the discrete case this is true because the length of the arc T_{i-1}T_{i} is equal to the turning angle \kappa_{i} , see Figure 3 . In the smooth case assume without loss of generality that \gamma is arc-length parametrized. Then one has T_{\gamma}=\dot{\gamma}\ \Rightarrow\ \dot{T}_{\gamma}=\ddot{\gamma}. Hence the length of T_{\gamma} is the integral of \|\ddot{\gamma}\|=\kappa , and the lemma is proved. ∎ Lemma 1.5 . The tangent indicatrix of a closed space curve is not contained in any open hemisphere. Proof. Let us prove this in the discrete case. Assume the contrary: there is v\in\mathbb{R}^{3} such that \langle T_{i},v\rangle>0 for all i . Then due to \sum_{i=1}^{n}(C_{i+1}-C_{i})=0 one has 0=\sum_{i=1}^{n}\langle C_{i+1}-C_{i},v\rangle=\sum_{i=1}^{n}\|C_{i+1}-C_{i}\|\langle T_{i},v\rangle>0, which is a contradiction. The proof in the smooth case is similar. ∎ Lemma 1.5 implies that almost every great circle intersects the tangent indicatrix at least twice. Lemma 1.6 (Crofton) . The length of a spherical curve is \pi times the average number of its intersections with great circles. The average is taken with respect to the following measure on the space of great circles. A great circle is the intersection of \mathbb{S}^{2} with a 2 -dimensional linear subspace of \mathbb{R}^{3} ; the orthogonal complement defines a bijection between 2 -dimensional linear subspaces and 1 -dimensional linear subspaces; the latter form the projective plane which is doubly covered by the sphere. Thus the standard measure on the sphere can be pushed to a measure on the space of great circles. In other words, the lemma states that for any curve T\subset\mathbb{S}^{2} one has \operatorname{Length}(T)=\frac{1}{4}\int_{\mathbb{S}^{2}}|T\cap v^{\perp}|\,dv. (The total measure of \mathbb{S}^{2} is 4\pi .) Proof. Consider the discrete case. The number of intersections of a spherical polygon with a great circle is equal to the sum of the numbers of intersections of its edges with this circle (for almost all circles); the length of a polygon is the sum of the lengths of its edges. Therefore it suffices to prove the lemma for a single edge of the polygon, that is for any arc of a great circle. Since almost every great circle intersects an arc either once or not at all, one has to show that the length of a spherical arc is \pi times the probability that a random great circle intersects this arc. A great circle v^{\circ}:=v^{\perp}\cap\mathbb{S}^{2} intersects an arc pq if and only if the inner products \langle v,p\rangle and \langle v,q\rangle have opposite signs. This in turn is equivalent to v\in(H_{p}^{+}\cap H_{q}^{-})\cup(H_{p}^{-}\cap H_{q}^{+}), where H_{p}^{+}=\{x\in\mathbb{S}^{2}\mid\langle x,p\rangle\geq 0\} etc. It is not hard to

Read the complete text