betaFeedback

WorksMathematics

Infinitely many convex polytopes with dihedral angles bounded above by π/2+δπ/2+δ

Abstract

For every δ>0δ>0, we construct infinitely many pairwise combinatorially distinct convex three-dimensional polytopes whose dihedral angles are all less than π/2+δπ/2+δ.

Introduction

Throughout this paper, dihedral angles are understood to be interior dihedral angles. As reported by Adiprasito [1], Gromov asked the following finiteness question:

Question 1.1. For a fixed dimension d≥3d \ge3 and ϵ>0\epsilon> 0, are there only finitely many combinatorial types of convex dd-polytopes whose dihedral angles are all less than π−ϵ\pi-\epsilon?

Classical results imply finiteness for ϵ≥π/2\epsilon\ge\pi/2; see [3 Theorem 2]. Gromov discusses infinite families for small ϵ>0\epsilon> 0 in [4 Section 1.1], attributing the initial observation to Adiprasito. The corresponding question about the threshold for finiteness appears as Problem 18 in [2 Page 196], attributed to Adiprasito.

We prove that π/2\pi/2 is the optimal dihedral angle upper bound for every d≥3d \ge3. More precisely, we establish the following results.

Theorem 1.2. For every δ>0\delta> 0 and every positive integer NN, there exists a convex three-dimensional polytope with at least NN two-dimensional faces whose dihedral angles are all less than π/2+δ\pi/2+\delta.

The following finiteness theorem (which directly answers Question 1.1 with the optimal ϵ\epsilon) follows from Theorem 1.2 and [3 Theorem 2].

Theorem 1.3. For any d≥3d \ge3 and β>0\beta> 0, let Pd,βP_{d,\beta} be the collection of convex dd-dimensional polytopes with dihedral angles at most β\beta, modulo combinatorial equivalence. Then

(1) ∣Pd,π/2∣<∞|P_{d,\pi/2}| < \infty, and

(2) ∣Pd,π/2+δ∣=∞|P_{d,\pi/2+\delta}| = \infty for any δ>0\delta> 0.

Proof. (1) By [3 Theorem 2], any member of Pd,π/2P_{d,\pi/2} is a direct product of simplices. Thus Pd,π/2P_{d,\pi/2} is finite. In particular, ∣Pd,π/2∣|P_{d,\pi/2}| equals the partition number of dd, number of ways of writing dd as a sum of positive integers.

(2) For any δ>0\delta> 0, ∣Pd,π/2+δ∣=∞|P_{d,\pi/2+\delta}| = \infty is witnessed by the direct product of the examples in Theorem 1.2 with [0,1]d−3[0,1]^{d-3}.

We prove Theorem 1.2 in § 2.

AI disclosure. The authors used GPT-6 Astra to help develop the proof. The authors checked and rewrote the proof, and take full responsibility for its final form and correctness.

Acknowledgement. Jinmin Wang is partially supported by NSFC 12501169.

The construction in dimension three

In this section, we will construct infinitely many combinatorially distinct polytopes with uniformly bounded dihedral angles.

Below we will describe a special family {v⃗i}\{\vec{v}_i\} of normal vectors that supplies our construction.

Fix R>1R > 1 a large transcendental real number and M>0M > 0 a large integer. Define

θ=arccos⁡(−1/R)∈(π2,π).(1)\theta= \arccos(-1/R) \in\left(\frac{\pi}{2},\pi\right). \tag*{(1)}

For j=0,1,⋯ ,M−1j = 0,1,\cdots,M-1, define

v⃗j=(cos⁡(jθ),sin⁡(jθ),Rj−M),n⃗j=v⃗j∣v⃗j∣.(2)\vec{v}_j = \left(\cos(j\theta),\sin(j\theta),R^{j-M}\right),\qquad\vec{n}_j = \frac{\vec{v}_j}{\lvert\vec{v}_j\rvert}. \tag*{(2)}

For suitable positive real values hjh_j for j=0,⋯ ,M−1j=0,\cdots,M-1, we construct a polytope PkP_k, for k=2,⋯ ,M−1k=2,\cdots,M-1, exactly enclosed bottom face B=Pk∩{x3=0}B=P_k\cap\{x_3=0\}, with unit outer normal vector n=(0,0,−1)\boldsymbol{n}=(0,0,-1), and faces given by

Hj={x⃗∈R3:⟨x⃗,v⃗j⟩=hj}.H_j=\{\vec{x}\in\mathbb{R}^3:\langle\vec{x},\vec{v}_j\rangle=h_j\}.

More precisely,

Pk={x⃗∈R3:⟨x⃗,n⟩≤0}∩(⋂j=0k{x⃗∈R3:⟨v⃗j,x⃗⟩≤hj}).(3)P_k=\{\vec{x}\in\mathbb{R}^3:\langle\vec{x},\boldsymbol{n}\rangle\le0\}\cap\left(\bigcap_{j=0}^{k}\{\vec{x}\in\mathbb{R}^3:\langle\vec{v}_j,\vec{x}\rangle\le h_j\}\right). \tag*{(3)}

For a fixed PkP_k, write

Fj=Pk∩Hj(0≤j≤k).F_j=P_k\cap H_j\qquad(0\le j\le k).

suppressing the dependence of these faces on kk. We will ensure that PkP_k is bounded and that B,F0,…,FkB,F_0,\ldots,F_k are exactly its faces, with outward unit normals n,n⃗0,…,n⃗k\boldsymbol{n},\vec{n}_0,\ldots,\vec{n}_k, respectively.

The basic idea is an induction: we start from a tetrahedron P2P_2 enclosed by BB and F0,F1,F2F_0,F_1,F_2. At each induction step, we cut Pk−1P_{k-1} with the plane HkH_k near its “top vertex” incident with Fk−1,Fk−2,Fk−3F_{k-1},F_{k-2},F_{k-3}, and get FkF_k. We then control all newly created dihedral angles and show that they all stay bounded. To proceed, we need some linear algebra preparations.

Lemma 2.1. Let n=(0,0,−1)\boldsymbol{n}=(0,0,-1) and v⃗j\vec{v}_j be defined as in (2). Then the following statements hold.

  1. For distinct nonnegative integers ii and jj, the vectors n,v⃗i,v⃗j\boldsymbol{n},\vec{v}_i,\vec{v}_j are linearly independent.

(2) For every integer j≥0j \ge0, the vectors v⃗j,v⃗j+1,v⃗j+2\vec{v}_j,\vec{v}_{j+1},\vec{v}_{j+2} are linearly independent.

Proof. (1) A direct computation shows that det⁡[n,v⃗i,v⃗j]=sin⁡((i−j)θ)\det[\mathbf{n},\vec{v}_i,\vec{v}_j]=\sin((i-j)\theta). This is a non-zero transcendental number by (1).

(2) We also have direct computation

det⁡[v⃗j,v⃗j+1,v⃗j+2]=Rj−M(sin⁡θ−Rsin⁡(2θ)+R2sin⁡θ)=Rj−M(R2+3)sin⁡θ.\det[\vec{v}_j,\vec{v}_{j+1},\vec{v}_{j+2}] =R^{j-M}\left(\sin\theta-R\sin(2\theta)+R^2\sin\theta\right) =R^{j-M}\left(R^2+3\right)\sin\theta.

which is non-zero. □\square

A core ingredient of the construction is the following observation of a positive linear combination.

Lemma 2.2. For every integer j≥0j \ge0,

v⃗j+3=Rv⃗j+v⃗j+1+(R−2R)v⃗j+2.\vec{v}_{j+3}=R\vec{v}_j+\vec{v}_{j+1}+\left(R-\frac{2}{R}\right)\vec{v}_{j+2}.

Proof. By (1), cos⁡θ=−1/R\cos\theta=-1/R, thus we have

cos⁡((j+2)θ)+cos⁡(jθ)=2cos⁡((j+1)θ)cos⁡θ=−2Rcos⁡((j+1)θ),\cos((j+2)\theta)+\cos(j\theta)=2\cos((j+1)\theta)\cos\theta=-\frac{2}{R}\cos((j+1)\theta),
sin⁡((j+2)θ)+sin⁡(jθ)=2sin⁡((j+1)θ)cos⁡θ=−2Rsin⁡((j+1)θ).\sin((j+2)\theta)+\sin(j\theta)=2\sin((j+1)\theta)\cos\theta=-\frac{2}{R}\sin((j+1)\theta).

Recall that

v⃗j=(cos⁡(jθ),sin⁡(jθ),Rj−M).\vec{v}_j=(\cos(j\theta),\sin(j\theta),R^{j-M}).

It follows that

v⃗j+2+v⃗j+2Rv⃗j+1=(0,0,Rj−M(R2+3)).\vec{v}_{j+2}+\vec{v}_j+\frac{2}{R}\vec{v}_{j+1}=(0,0,R^{j-M}(R^2+3)).

Therefore,

R(v⃗j+2+v⃗j+2Rv⃗j+1)=v⃗j+3+v⃗j+1+2Rv⃗j+2.R\left(\vec{v}_{j+2}+\vec{v}_j+\frac{2}{R}\vec{v}_{j+1}\right)=\vec{v}_{j+3}+\vec{v}_{j+1}+\frac{2}{R}\vec{v}_{j+2}.

Rearranging gives the required recurrence. □\square

The following lemma will provide the desired dihedral angle bounds.

Lemma 2.3. For integers 0≤i<j≤M−10 \le i < j \le M-1 with j−i≤3j-i \le3, we have

⟨n⃗i,n⃗j⟩=⟨v⃗i,v⃗j⟩∣v⃗i∣∣v⃗j∣<4R.\langle\vec{n}_i,\vec{n}_j\rangle=\frac{\langle\vec{v}_i,\vec{v}_j\rangle}{|\vec{v}_i||\vec{v}_j|}<\frac{4}{R}.

Proof. A direct computation shows

⟨v⃗i,v⃗j⟩∣v⃗i∣∣v⃗j∣=cos⁡((j−i)θ)+Ri+j−2M(1+R2i−2M)(1+R2j−2M).\frac{\langle\vec{v}_i,\vec{v}_j\rangle}{|\vec{v}_i||\vec{v}_j|} =\frac{\cos((j-i)\theta)+R^{i+j-2M}}{\sqrt{(1+R^{2i-2M})(1+R^{2j-2M})}}.

As cos⁡θ=−1/R\cos\theta=-1/R and 1≤j−i≤31 \le j-i \le3, the term cos⁡((j−i)θ)\cos((j-i)\theta) only takes one of the three values:

cos⁡θ=−1R,cos⁡(2θ)=2R2−1,cos⁡(3θ)=3R−4R3.\cos\theta=-\frac{1}{R},\qquad\cos(2\theta)=\frac{2}{R^2}-1,\qquad\cos(3\theta)=\frac{3}{R}-\frac{4}{R^3}.

In particular, we have

cos⁡((j−i)θ)<3R\cos((j-i)\theta)<\frac{3}{R}

Furthermore, as i,j≤M−1i,j\le M-1, we have

Ri+j−2M(1+R2i−2M)(1+R2j−2M)=Ri+j(R2M+R2i)(R2M+R2j)<1R\frac{R^{i+j-2M}}{\sqrt{(1+R^{2i-2M})(1+R^{2j-2M})}}=\frac{R^{i+j}}{\sqrt{(R^{2M}+R^{2i})(R^{2M}+R^{2j})}}<\frac{1}{R}

As a consequence,

⟨v⃗i,v⃗j⟩∣v⃗i∣∣v⃗j∣<3R+1R=4R.\frac{\langle\vec{v}_i,\vec{v}_j\rangle}{|\vec{v}_i||\vec{v}_j|}<\frac{3}{R}+\frac{1}{R}=\frac{4}{R}.

This concludes the proof. □

Proof of Theorem 1.2. We now present the construction. For two faces F,GF,G sharing an edge, write ∠(F,G)\angle(F,G) for their interior dihedral angle.

Base case. When k=2k=2, taking h0=R−Mh_0=R^{-M}, h1=R1−Mh_1=R^{1-M}, and h2=R2−Mh_2=R^{2-M}, we introduce P2P_2 as in (2.3). Thus P2P_2 is a tetrahedron, with a unique vertex above the xyxy-plane. Its bottom face is BB and side faces are F0,F1,F2F_0,F_1,F_2. The side dihedral angles are given by

cos⁡(π−∠(Fi,Fj))=⟨n⃗i,n⃗j⟩\cos(\pi-\angle(F_i,F_j))=\langle\vec{n}_i,\vec{n}_j\rangle

for 0≤i<j≤20\le i<j\le2. By Lemma 2.3,

∠(Fi,Fj)=arccos⁡(−⟨n⃗i,n⃗j⟩)<arccos⁡(−4R)=π2+arcsin⁡(4R).\angle(F_i,F_j)=\arccos(-\langle\vec{n}_i,\vec{n}_j\rangle)<\arccos\left(-\frac{4}{R}\right)=\frac{\pi}{2}+\arcsin\left(\frac{4}{R}\right).

For bottom dihedral angles, we have

cos⁡(π−∠(B,Fj))=⟨n,n⃗j⟩\cos(\pi-\angle(B,F_j))=\langle n,\vec{n}_j\rangle

for j=0,1,2j=0,1,2. A direct computation shows

⟨n,n⃗j⟩=−Rj−M1+R2j−2M<0.\langle n,\vec{n}_j\rangle=-\frac{R^{j-M}}{\sqrt{1+R^{2j-2M}}}<0.

Hence ∠(B,Fj)\angle(B,F_j) are all acute.

Induction hypotheses. We will construct inductively the polytopes PkP_k for k=2,3,…,M−1k=2,3,\ldots,M-1 such that the following properties hold.

  1. PkP_k has exactly (k+2)(k+2) faces, namely

B=Pk∩{x3=0}andFj=Pk∩Hj(0≤j≤k).B=P_k\cap\{x_3=0\}\quad\text{and}\quad F_j=P_k\cap H_j\quad(0\le j\le k).
  1. The degree of each vertex of PkP_k is three.

  1. For 0≤j≤k0\le j\le k, BB and FjF_j share an edge if and only if j=0,1,2j=0,1,2.

  1. For 0≤i<j≤k0\le i<j\le k, FiF_i and FjF_j share an edge if and only if 1≤∣i−j∣≤31\le|i-j|\le3.

  1. All dihedral angles of PkP_k are smaller than (π/2+arcsin⁡(4R−1))(\pi/2+\arcsin(4R^{-1})).

  1. PkP_k has 2k2k vertices and 3k3k edges.

The base case verifies all these properties for P2P_2.

Induction step. Suppose that 3≤k≤M−13 \le k \le M - 1 and that Pk−1P_{k-1} has been constructed. We now construct PkP_k out of Pk−1P_{k-1}. Let pp be the vertex of Pk−1P_{k-1} meeting three faces Fk−1,Fk−2,Fk−3F_{k-1}, F_{k-2}, F_{k-3}, guaranteed by Lemma 2.1. We cut the vertex pp away by a plane with normal vector v⃗k\vec{v}_k.

Claim. The linear functional x↦⟨v⃗k,x⟩x \mapsto\langle\vec{v}_k, x \rangle attains its maximum on Pk−1P_{k-1} uniquely at pp.

Indeed, by Lemma 2.2, v⃗k\vec{v}_k is a positive linear combination of v⃗k−1,v⃗k−2,v⃗k−3\vec{v}_{k-1}, \vec{v}_{k-2}, \vec{v}_{k-3}. Therefore, for any x∈Pk−1x \in P_{k-1},

⟨v⃗k,x⟩≤(R−2R)⟨v⃗k−1,x⟩+⟨v⃗k−2,x⟩+R⟨v⃗k−3,x⟩≤(R−2R)hk−1+hk−2+Rhk−3.\begin{aligned} \langle\vec{v}_k, x \rangle&\le\left(R-\frac{2}{R}\right)\langle\vec{v}_{k-1},x\rangle+\langle\vec{v}_{k-2},x\rangle+R\langle\vec{v}_{k-3},x\rangle\\ &\le\left(R-\frac{2}{R}\right)h_{k-1}+h_{k-2}+Rh_{k-3}. \end{aligned}

Equality holds if and only if

⟨v⃗k−1,x⟩=hk−1,⟨v⃗k−2,x⟩=hk−2,⟨v⃗k−3,x⟩=hk−3,\langle\vec{v}_{k-1},x\rangle=h_{k-1},\quad\langle\vec{v}_{k-2},x\rangle=h_{k-2},\quad\langle\vec{v}_{k-3},x\rangle=h_{k-3},

that is, x=px=p. This finishes the proof of the claim.

Set

hk=12(⟨v⃗k,p⟩+max⁡w≠p∣⟨v⃗k,w⟩∣).(4)h_k=\frac{1}{2}\left(\langle\vec{v}_k,p\rangle+\max_{w\ne p}|\langle\vec{v}_k,w\rangle|\right). \tag*{(4)}

where the maximum is taken for all vertices ww of Pk−1P_{k-1} that is not pp. Consequently, the plane HkH_k strictly separates pp from every other vertex of Pk−1P_{k-1}. We define

Pk=Pk−1∩{x∈R3:⟨v⃗k,x⟩≤hk},P_k=P_{k-1}\cap\{x\in\mathbb{R}^3:\langle\vec{v}_k,x\rangle\le h_k\},

in agreement with (3).

Every old edge survives, possibly shortened, and FkF_k is adjacent precisely to Fk−1,Fk−2,Fk−3F_{k-1}, F_{k-2}, F_{k-3}. All old faces remain faces, and all old face adjacencies are preserved. Since p∉Bp\notin B, the bottom face is unchanged. This verifies (1)–(4).

The truncation removes one vertex and creates three vertices, adds three edges, and adds one face. Hence PkP_k has 2k2k vertices, 3k3k edges, and k+2k+2 faces, verifying (6).

The dihedral angles along surviving edges are unchanged. Each new edge lies in Fk∩FjF_k\cap F_j for some j∈{k−3,k−2,k−1}j\in\{k-3,k-2,k-1\}. By Lemma 2.3, its dihedral angle satisfies

∠(Fk,Fj)=arccos⁡(−⟨n⃗k,n⃗j⟩)<π2+arcsin⁡(4/R).\angle(F_k,F_j)=\arccos\left(-\langle\vec{n}_k,\vec{n}_j\rangle\right)<\frac{\pi}{2}+\arcsin(4/R).

Together with the induction hypothesis, this verifies (5) and completes the induction. □

References

(Zhenkun Li) Institute of Mathematics, Chinese Academy of Sciences Email address: [email protected]

(Jinmin Wang) Institute of Mathematics, Chinese Academy of Sciences Email address: [email protected]

(Jing Yu) Shanghai Center for Mathematical Sciences, Fudan University Email address: [email protected]

References

  1. [1]Karim Adiprasito. Scalar curvature and angles of polyhedra: test your intuition, 2022. Published July 11, 2022; available at https://kaguprasetya.com/2022/07/11/scalar-curvature-and-angles-of-polyhedra-test-your-intuition/.
  2. [2]Karim Adiprasito, Xavier Goaoc, and Zuzana Patáková. Discrete geometry. Oberwolfach Reports, 21(1):137–202, 2024.
  3. [3]H. S. M. Coxeter. Discrete groups generated by reflections. Ann. of Math. (2), 35(3):588–621, 1934.DOI
  4. [4]Misha Gromov. Convex polytopes, dihedral angles, mean curvature and scalar curvature. Discrete Comput. Geom., 72(2):849–875, 2024.arxiv.org/abs/2207.13346

Paper details

vlnc:W_267swzja
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents