betaFeedback

WorksMathematics

Reflexive polytopes can have doubly exponentially many vertices

Abstract

Reflexive polytopes are lattice polytopes containing the origin in their interior such that also the dual polytope is a lattice polytope. On the basis of low-dimensional observations and of bounds known for special classes of reflexive polytopes, the second author conjectured in 2004 that the number of vertices of dd-dimensional reflexive polytopes is bounded from above by 6d/26^{d/2}. In this note, we present counterexamples showing that this conjecture is false in every dimension d≥8d \ge 8. In fact, rather surprisingly, in sufficiently large dimensions dd the number of vertices (equivalently, by duality, the number of facets) can be as large as theoretically possible, namely doubly exponential in dd. These results were found using GPT-6.1 Sol.

The original conjecture

A polytope P⊂RdP \subset\mathbb{R}^d is called lattice polytope if the vertices of PP are elements of Zd\mathbb{Z}^d. In 1994 Batyrev [4] defined a reflexive polytope to be a full-dimensional lattice polytope P⊂RdP \subset\mathbb{R}^d containing the origin in its interior such that its polar polytope

P∗:={y∈(Rd)∗:⟨y,x⟩≥−1 for all x∈P}P^* := \{y \in(\mathbb{R}^d)^* : \langle y,x\rangle\ge-1 \text{ for all } x \in P\}

is also a lattice polytope. Hence, reflexivity is invariant under polar duality. In the toric dictionary, reflexive polytopes correspond to Gorenstein toric Fano varieties and have been used to construct Calabi-Yau varieties relevant for mirror symmetry. We refer to [3] for more background on this.

Recall that two lattice polytopes are considered isomorphic if there is an affine-unimodular transformation mapping them onto each other. As a reflexive polytope contains only the origin as an interior lattice point, there are only finitely many reflexive polytopes of each dimension up to isomorphisms [4, 18, 22].

In his PhD thesis the second author studied reflexive polytopes and posed the following conjecture, which was published as Conjecture 5.2 in [25] and discussed in more detail in [6]. For this, we need to define

H:=conv⁡([0,1]2,−[0,1]2),\mathcal{H} := \operatorname{conv}([0,1]^2,-[0,1]^2),

which is the (up to isomorphism unique) reflexive hexagon. Note that H∗\mathcal{H}^* is isomorphic to H\mathcal{H}.

Conjecture 1.1. Let PP be a dd-dimensional reflexive polytope. Then its number of vertices is bounded from above by 6d/26^{d/2} with equality if and only if dd is even and PP is isomorphic to Hd/2\mathcal{H}^{d/2}.

Equivalently, by duality, the conjecture can be reformulated for the number of facets (as in Conjecture 1.1 in [23]) on which we will focus from now on. Recall that the free sum Q1∘Q2Q_1 \circ Q_2 of two lattice polytopes Q1⊂Rn1Q_1 \subset\mathbb{R}^{n_1} and Q2⊂Rn2Q_2 \subset\mathbb{R}^{n_2} is defined as the convex hull in Rn1+n2\mathbb{R}^{n_1+n_2} of Q1×{0}Q_1 \times\{0\} and {0}×Q2\{0\} \times Q_2. If Q1Q_1 and Q2Q_2 are full-dimensional and contain the origins in their interiors, then (Q1∘Q2)∗=Q1∗×Q2∗(Q_1 \circ Q_2)^* = Q_1^* \times Q_2^*. Using this notion, Conjecture 1.1 can be restated as follows:

Conjecture 1.2 (Dual version). Let PP be a dd-dimensional reflexive polytope. Then its number of facets is bounded from above by 6d/26^{d/2} with equality if and only if dd is even and PP is isomorphic to the free sum of d/2d/2 reflexive hexagons H\mathcal{H}.

Evidence for the conjecture came from the following facts. It holds up to dimension 4 by the classification results of Kreuzer and Skarke [21]. It holds for simple reflexive polytopes [12]. It holds for simplicial reflexive polytopes with a centrally-symmetric pair of facets [26]. Recently, it was also verified for some classes of reflexive polytopes associated to graphs and posets [23, 24].

The counterexamples in dimensions 8 and 9

Using GPT-6.1 Sol we found counterexamples in dimensions as low as 8 and 9. Let us describe these examples. Their properties can easily be checked using standard polyhedral software and consumer hardware.

Define a three-dimensional lattice polytope Q3Q_3 as the convex hull of the following 21 lattice points in R3\mathbb{R}^3:

(6−6040−231−230030−321−12−2211−11011−1200−30−120−330−600−6−6−100−10−1−1−22−2−1−1−2−2−2−3−30)\begin{pmatrix} 6 & -6 & 0 \\ 4 & 0 & -2 \\ 3 & 1 & -2 \\ 3 & 0 & 0 \\ 3 & 0 & -3 \\ 2 & 1 & -1 \\ 2 & -2 & 2 \\ 1 & 1 & -1 \\ 1 & 0 & 1 \\ 1 & -1 & 2 \\ 0 & 0 & -3 \\ 0 & -1 & 2 \\ 0 & -3 & 3 \\ 0 & -6 & 0 \\ 0 & -6 & -6 \\ -1 & 0 & 0 \\ -1 & 0 & -1 \\ -1 & -2 & 2 \\ -2 & -1 & -1 \\ -2 & -2 & -2 \\ -3 & -3 & 0 \end{pmatrix}

Define in R3×R3×R2=R8\mathbb{R}^3 \times\mathbb{R}^3 \times\mathbb{R}^2 = \mathbb{R}^8 the lattice polytope P8P_8 as the convex hull of

((Q3∘Q3)×{(−1,−1)})∪({(0,0,0,0,0,0)}×{(−1,2),(1,−1)}).\left((Q_3 \circ Q_3) \times\{(-1,-1)\}\right) \cup\left(\{(0,0,0,0,0,0)\} \times\{(-1,2),(1,-1)\}\right).

The reader may look at the end of the paper where we list the vertices on the left side.

Theorem 2.1. P8P_8 is an 8-dimensional reflexive polytope with 44=21+21+244 = 21 + 21 + 2 vertices and 1371>1296=641371 > 1296 = 6^4 facets. Hence, Conjecture 1.2 fails in dimension 8.

For dimension 9 we define Q2Q_2 as the convex hull of the following 15 lattice points in R2\mathbb{R}^2:

(−42,168),  (−21,63),  (−14,35),  (−10,22),  (−7,14),(-42,168),\;(-21,63),\;(-14,35),\;(-10,22),\;(-7,14),
(−2,1),  (−1,−1),  (0,−2),  (1,−2),  (2,−1),(-2,1),\;(-1,-1),\;(0,-2),\;(1,-2),\;(2,-1),
(2,0),  (0,14),  (−6,54),  (−14,98),  (−42,210).(2,0),\;(0,14),\;(-6,54),\;(-14,98),\;(-42,210).

Now, we define in R2×R2×R2×R3=R9\mathbb{R}^2 \times\mathbb{R}^2 \times\mathbb{R}^2 \times\mathbb{R}^3 = \mathbb{R}^9 the lattice polytope P9P_9 as the convex hull of

((Q2∘Q2∘Q2)×{(−1,−1,−1)})∪((Q_2 \circ Q_2 \circ Q_2) \times\{(-1,-1,-1)\}) \cup
({(0,0,0,0,0,0)}×{(−1,−1,6),(−1,2,−1),(1,−1,−1)}).(\{(0,0,0,0,0,0)\} \times\{(-1,-1,6),(-1,2,-1),(1,-1,-1)\}).

Also here we list the vertices at the end of the paper on the right side.

Theorem 2.2. P9P_9 is a 9-dimensional reflexive polytope with 48=15+15+15+348 = 15 + 15 + 15 + 3 vertices and 3378>69/23378 > 6^{9/2} facets. Hence, Conjecture 1.2 fails in dimension 9.

In particular, it fails also in any dimension at least 10 by just taking sufficiently many free sums of P8P_8 or P9P_9 with H\mathcal{H}.

The construction behind the counterexamples

Consider the reflexive polytope P8P_8 from Theorem 2.1. Combinatorially it is a two-fold pyramid over the free sum Q3∘Q3Q_3 \circ Q_3. Here are two important observations:

  • Q3Q_3 has the origin in its interior, however, it is not reflexive, but its 37 facets have lattice distances1^1 1, 3 or 6 from the origin. By considering free sums (or dually products) of non-reflexive lattice polytopes as the starting point one clearly has a lot of flexibility compared to previous approaches starting with low-dimensional reflexive polytopes.

  • Next, choosing the apices of the two-fold pyramids in this specific way, all lattice distances of the facets become just 1. This is essentially the same idea as in Haase and Melnikov [16] where they show that any lattice polytope can be embedded as the face of a high-dimensional reflexive polytope. However, while they use a wedge construction which decreases the lattice distance a priori just for one facet, choosing pyramids affects all the facets at once, which makes the operation efficient enough to get counterexamples to the conjecture. And also this idea is already contained in the paper of Haase and Melnikov in their Proposition 4.2! Namely, they use precisely this approach together with number-theoretic considerations to bound the increase of dimensions needed to embed a lattice interval as the edge of a reflexive polytope (their Corollary 4.3).

In what follows, we adopt the very same approach by Haase and Melnikov in order to generalize and optimize this construction to higher-dimensions.

First, let us make the two-fold pyramid construction for P8P_8 more precise. Consider the reflexive triangle S2S_2 with vertices (−1,−1)(-1,-1), (−1,2)(-1,2), (1,−1)(1,-1). Notice that P8P_8 can be described by keeping the last two vertices of S2S_2 and replacing the vertex (−1,−1)(-1,-1) with (Q3∘Q3)×(−1,−1)(Q_3 \circ Q_3) \times(-1,-1).

In order to generalize this let us recall that for every reflexive mm-simplex v0,…,vmv_0,\ldots,v_m there is an associated unit partition:

1k0+⋯+1km=1\frac{1}{k_0}+\cdots+\frac{1}{k_m}=1

with positive integers k0,…,kmk_0,\ldots,k_m such that

v0k0+⋯+vmkm=0.\frac{v_0}{k_0}+\cdots+\frac{v_m}{k_m}=0.

If all k0,…,kmk_0,\ldots,k_m are pairwise distinct, then this is called an Egyptian fraction. The converse also holds (see [27 Theorem 4.3], and originally [14 Proposition 5.1] and [4 Theorem 5.4.3]).

Lemma 3.1. For each unit partition 1k0+⋯+1km=1\frac{1}{k_0}+\cdots+\frac{1}{k_m}=1 with integers 2≤k0≤⋯≤km2 \le k_0 \le\cdots\le k_m there exists a reflexive mm-dimensional simplex S⊂ZmS \subset\mathbb{Z}^m with vertices v0,…,vm∈Zmv_0,\ldots,v_m \in\mathbb{Z}^m such that v0k0+⋯+vmkm=0\frac{v_0}{k_0}+\cdots+\frac{v_m}{k_m}=0.

For instance, in our 8-dimensional counterexample with m=2m=2 we have (k0,k1,k2)=(6,3,2)(k_0,k_1,k_2)=(6,3,2):

16+13+12=1\frac{1}{6}+\frac{1}{3}+\frac{1}{2}=1

and

(−1,−1)6+(−1,2)3+(1,−1)2=(0,0).\frac{(-1,-1)}{6}+\frac{(-1,2)}{3}+\frac{(1,-1)}{2}=(0,0).

Next, let us generalize the Haase-Melnikov pyramid construction. Here, F(P)F(P) denotes the set of facets of a polytope PP.

Lemma 3.2. Let B⊂RnB \subset\mathbb{R}^n be an nn-dimensional lattice polytope with 00 in its relative interior. Let S⊂RmS \subset\mathbb{R}^m be an mm-dimensional reflexive simplex with vertices v0,…,vmv_0,\ldots,v_m, and associated unit partition with denominators k0,…,kmk_0,\ldots,k_m. If the lattice distance of each facet of BB from the origin divides k0k_0, then the convex hull of

B×{v0}∪{(0,v1),…,(0,vm)}⊂Rn+mB \times\{v_0\} \cup\{(0,v_1),\ldots,(0,v_m)\} \subset\mathbb{R}^{n+m}

is an n+mn+m-dimensional reflexive polytope PP. This polytope is combinatorially an mm-fold pyramid over the base face BB. In particular, its number of facets is ∣F(P)∣=∣F(B)∣+m|F(P)|=|F(B)|+m.

Proof. For i=0,…,mi=0,\ldots,m let uiu_i be the integral inward facet normal of the facet of SS with vertices {v0,…,vm}∖{vi}\{v_0,\ldots,v_m\} \setminus\{v_i\}. There are integers k0,…,km∈Z≥2k_0,\ldots,k_m \in\mathbb{Z}_{\ge2} such that we have for i,j∈{0,…,m}i,j \in\{0,\ldots,m\} the equalities ⟨ui,vj⟩=kiδij−1\langle u_i,v_j\rangle=k_i\delta_{ij}-1. Next, we write the facet inequalities (say, rr many) of BB as ⟨al,⋅⟩≥−hl\langle a_l,\cdot\rangle\ge-h_l with ala_l a primitive integral functional and integers hl>0h_l>0 for l=1,…,rl=1,\ldots,r.

Combinatorially, the polytope PP is an mm-fold pyramid over BB. Hence, we know its facets: they are given by the convex hull of all vertices except for one of the apices v1,…,vmv_1,\ldots,v_m, or by the convex hull of a facet of BB together with all apices v1,…,vmv_1,\ldots,v_m. Thus, the facet inequalities of PP are given as

⟨ui,y2⟩≥−1(for i=1,…,m),\langle u_i,y_2\rangle\ge-1 \quad(\text{for } i=1,\ldots,m),
k0hl⟨al,y1⟩+⟨u0,y2⟩≥−1(for l=1,…,r)\frac{k_0}{h_l}\langle a_l,y_1\rangle+\langle u_0,y_2\rangle\ge-1 \quad(\text{for } l=1,\ldots,r)

for all y=(y1,y2)∈Rn+my=(y_1,y_2)\in\mathbb{R}^{n+m}. Note that these linear functions are integral by our assumption as hlh_l (for l=1,…,rl=1,\ldots,r) are the lattice distances of the facets of BB from the origin. Thus, PP is reflexive. □

As a reality check the reader can verify that in our 9-dimensional counterexample the facet lattice distances of the lattice polygon Q2Q_2 (and thus of Q2∘Q2∘Q2Q_2\circ Q_2\circ Q_2) are all divisors of 42, while the 3-dimensional reflexive simplex with vertices (−1,−1,−1),(−1,−1,6),(−1,2,−1),(1,−1,−1)(-1,-1,-1),(-1,-1,6),(-1,2,-1),(1,-1,-1) has denominators 42, 7, 3, 2 in its unit partition.

Remark 3.3. Lemma 3.2 is a direct generalization of Proposition 4.2 in [16] due to Haase and Melnikov, where BB was assumed to be a suitable dilate of a reflexive polytope. In fact, another way to look at Lemma 3.2 is that this construction is just the free sum of BB and SS where the point of intersection is the vertex of SS. Free sums of lattice polytopes that intersect on the boundary appear at several places in the literature, e.g., [8, 5, 9] among others, and also the Haase-Melnikov wedge construction [16] can be seen as a special dual instance of this. The examples of hollow lattice polytopes by Codenotti and Santos [13] are free sums with intersection on the boundary. Even the spectacular recent counterexamples by Ferroni to the IDP unimodality conjecture [15] are Chimney polytopes [17], whose polars are consecutive free sums. It seems to us that the full potential of these boundary-free-sums constructions and their dual wedge constructions for finding interesting (counter)examples is just about to be completely realized by our community.

The general result

Now, the strategy to optimize the bounds in higher dimensions becomes clear. We need to find lattice polytopes BB containing the origin in their interiors that have as many facets as possible while the least common multiple HH of their facet lattice distances from the origin should be as small as possible, so that the reflexive simplex one needs to get a reflexive polytope should hopefully add as few dimensions as possible.

As it turns out, it is enough to consider this problem in dimension two and then just take sufficiently many free sums of a good candidate polygon with itself. In order to define these lattice polygons, let h∈Z≥2h\in\mathbb{Z}_{\ge2} and let τ(h)\tau(h) be the number of positive divisors of hh. We define the lattice polygon

Bh:=conv⁡({(0,−1)}∪{(a,a2),(−a,a2)∣a∈Z≥1,a∣h}).B_h:=\operatorname{conv}\left(\{(0,-1)\}\cup\{(a,a^2),(-a,a^2)\mid a\in\mathbb{Z}_{\ge1},a\mid h\}\right).

In the following V(P)V(P) denotes the set of vertices of a polytope PP.

Figure 1. The lattice polygons Bh⊂R2B_h \subset\mathbb{R}^2.

Lemma 4.1. The lattice polygon BhB_h satisfies the following properties:

  • 0∈int⁡(Bh)0 \in\operatorname{int}(B_h),

  • V(Bh)={(a,a2)∣a∈Z,a∣h}∪{(0,−1)}V(B_h) = \{(a,a^2) \mid a \in\mathbb{Z}, a \mid h\} \cup\{(0,-1)\}, and thus, ∣F(Bh)∣=2τ(h)+1|F(B_h)| = 2\tau(h) + 1,

  • the lattice distance of each facet from the origin divides h2h^2.

Proof. The triangle conv⁡((0,−1),(−1,1),(1,1))\operatorname{conv}((0,-1),(-1,1),(1,1)) is contained in BhB_h and does contain the origin in its interior. Hence, BhB_h contains the origin in its interior.

It is straightforward to verify that the following defines a supporting hyperplane for BhB_h that cuts out the vertex (a,a2)(a,a^2) for a divisor a∣ha \mid h:

y≥(2a+sgn⁡(a)2)x−a2−∣a∣2.y \ge\left(2a + \frac{\operatorname{sgn}(a)}{2}\right)x - a^2 - \frac{|a|}{2}.

Clearly, (0,−1)(0,-1) is a vertex of BhB_h too. From this it follows immediately that V(Bh)={(a,a2)∣a∈Z,a∣h}V(B_h) = \{(a,a^2) \mid a \in\mathbb{Z}, a \mid h\}. Order the set of the positive divisors of hh as

1=d1<d2<⋯<dk=h,1 = d_1 < d_2 < \cdots< d_k = h,

where k:=τ(h)k := \tau(h). Then it is again an easy exercise to show that the facets of BhB_h are defined by the following supporting hyperplanes:

y≤h2,y \le h^2,
y≥±2x−1,y \ge\pm2x - 1,
y≥±(di+di+1)x−didi+1,for 1≤i<k.y \ge\pm(d_i + d_{i+1})x - d_i d_{i+1}, \qquad\text{for } 1 \le i < k.

Hence, the lattice distances are h2h^2, 11, and didi+1d_i d_{i+1} for 1≤i<k1 \le i < k, that is, they divide h2h^2. □

Following Lemma 3.2 and the general ideas in Haase-Melnikov, the final step is to find for a choice of a base polytope BB with least common multiple of facet lattice distances HH a unit partition

1H+1k1+⋯+1km=1,\frac{1}{H} + \frac{1}{k_1} + \cdots+ \frac{1}{k_m} = 1,

where mm is as small as possible. This is a highly non-trivial number-theoretic problem for which ChatGPT originally uses the recent OpenAI solution of Erdős’s conjecture on short Egyptian fractions [28]. Recently, a very nice simplified and humanly written paper on this topic was uploaded on Hexagon by Thomas Bloom and Wouter van Doorn [7].

Theorem 4.2 ([7 Theorem 1 and Section 5]). For every ε>0\varepsilon> 0, there exists an integer b0(ε)b_0(\varepsilon) such that, for every integer b≥b0(ε)b \ge b_0(\varepsilon) and every integer 1≤a<b1 \le a < b, there exist k≥1k \ge1 and integers 2≤a1<⋯<ak2 \le a_1 < \cdots< a_k satisfying

ab=1a1+⋯+1akandk≤(16log⁡2+ε)log⁡log⁡b.\frac{a}{b} = \frac{1}{a_1} + \cdots+ \frac{1}{a_k} \qquad\text{and}\qquad k \le\left(\frac{16}{\log2} + \varepsilon\right)\log\log b.

Here, log⁡\log denotes the natural logarithm. By choosing, for example, ε=12\varepsilon= \frac{1}{2}, we get:

Corollary 4.3. There exists an integer H0H_0 such that for every integer H≥H0H \ge H_0 there exist an integer m≥1m \ge1 and integers 2≤k1<⋯<km2 \le k_1 < \cdots< k_m satisfying

1=1H+1k1+⋯+1km,m≤25log⁡log⁡H.1 = \frac{1}{H} + \frac{1}{k_1} + \cdots+ \frac{1}{k_m}, \qquad m \le25\log\log H.

Proof. Apply Theorem 4.2 with a=H−1a = H - 1 and b=Hb = H where H≥b0(12)H \ge b_0\left(\frac{1}{2}\right) and then add 1H\frac{1}{H} on both sides of the unit partition equation. □

Remark 4.4. The authors know that the constant in Corollary 4.3 is not optimal. Indeed, the authors have an argument where they can improve it to 10log⁡2\frac{10}{\log2}. However, for proving the doubly exponential growth of the vertices this constant is irrelevant.

Theorem 4.5. Let 0<κ<1250 < \kappa< \frac{1}{25} and set δ:=1−25κ4>0\delta:= \frac{1-25\kappa}{4} > 0. Then for every sufficiently large integer dd, depending on κ\kappa, there exist reflexive polytopes of dimension dd whose number of facets is at least

exp⁡(δlog⁡224κexp⁡(κd)).\exp\left(\frac{\delta\log2}{24\kappa}\exp(\kappa d)\right).

Proof. Let dd be the sufficiently large target dimension of the polytope in the statement. Fix 0<κ<1250<\kappa<\frac{1}{25} and set

N:=⌊eκd⌋,δ:=1−25κ4,r:=⌊δd⌋.N:=\left\lfloor e^{\kappa d}\right\rfloor,\qquad\delta:=\frac{1-25\kappa}{4},\qquad r:=\lfloor\delta d\rfloor.

Let hN:=∏p≤Nph_N:=\prod_{p\le N}p be the product of all prime numbers pp at most NN. Set HN:=hN2H_N:=h_N^2. By Corollary 4.3, for dd (and thus NN and HNH_N) sufficiently large there exist m≥1m\ge1 integers 2≤k1<⋯<km2\le k_1<\cdots<k_m satisfying

1=1HN+1k1+⋯+1kmandm≤25log⁡log⁡HN.1=\frac{1}{H_N}+\frac{1}{k_1}+\cdots+\frac{1}{k_m}\qquad\text{and}\qquad m\le25\log\log H_N.

By Lemma 3.1, there is a reflexive simplex S⊂ZmS\subset\mathbb{Z}^m with vertices v0,…,vmv_0,\ldots,v_m such that

v0HN+v1k1+⋯+vmkm=0.\frac{v_0}{H_N}+\frac{v_1}{k_1}+\cdots+\frac{v_m}{k_m}=0.

By Lemma 4.1, the lattice distance of each facet of BhNB_{h_N} from the origin divides HNH_N. Hence, this is also true for the lattice polytope BB which we define as the free sum of rr copies of BhNB_{h_N}. Therefore, by Lemma 3.2, the convex hull P′⊂Rm+2rP'\subset\mathbb{R}^{m+2r} of

B×{v0}∪{(0,v1),…,(0,vm)},B\times\{v_0\}\cup\{(0,v_1),\ldots,(0,v_m)\},

is a (m+2r)(m+2r)-dimensional reflexive polytope.

Let us convince ourselves that m+2r<dm+2r<d for dd sufficiently large. To that end, let us recall the following elementary version of the prime number theorem [1 Theorem 4.6] which holds for any integer N≥2N\ge2:

16Nlog⁡N<π(N)<6Nlog⁡N,(1)\frac{1}{6}\frac{N}{\log N}<\pi(N)<6\frac{N}{\log N}, \tag*{(1)}

where π(N)\pi(N) denotes the prime-counting function, i.e., the number of primes that are less than or equal to NN. This implies that

log⁡HN=2∑p≤Nlog⁡p<2π(N)log⁡N<12N.\log H_N=2\sum_{p\le N}\log p<2\pi(N)\log N<12N.

Hence,

m≤25log⁡log⁡HN<25log⁡(12N)=25log⁡(12)+25log⁡N≤25log⁡(12)+25κd.m\le25\log\log H_N<25\log(12N)=25\log(12)+25\log N\le25\log(12)+25\kappa d.

Putting everything together we get

m+2r<25log⁡(12)+25κd+1−25κ2d=25log⁡(12)+1+25κ2d<dm+2r<25\log(12)+25\kappa d+\frac{1-25\kappa}{2}d=25\log(12)+\frac{1+25\kappa}{2}d<d

for dd again sufficiently large which shows the claim.

As m+2r<dm+2r<d, we can take the free sum of P′P' with an appropriate number of copies of the reflexive interval [−1,1][-1,1] to obtain a reflexive polytope PP of dimension exactly dd. By Lemmas 3.2 and 4.1, PP has at least (2τ(hN)+1)r(2^{\tau(h_N)}+1)^r facets. All that remains is to bound this number from below in terms of dd. Clearly τ(hN)=2π(N)\tau(h_N)=2^{\pi(N)}, so that

log⁡((2τ(hN)+1)r)=rlog⁡(22π(N)+1)>rπ(N)log⁡2>r16Nlog⁡Nlog⁡2(2)\log\left((2^{\tau(h_N)}+1)^r\right)=r\log\left(2^{2^{\pi(N)}+1}\right)>r\pi(N)\log2>r\frac{1}{6}\frac{N}{\log N}\log2 \tag*{(2)}

where the last inequality follows again from (1). It is straightforward to verify that

N≥12eκdandlog⁡N≤κd.N \ge\frac{1}{2}e^{\kappa d} \qquad\text{and} \qquad\log N \le\kappa d.

Substituting this into Equation (2), we get

log⁡((2τ(hN)+1)r)>r16Nlog⁡Nlog⁡2≥rlog⁡212eκdκd≥δd2log⁡212eκdκd=δlog⁡224κeκd,\log\left((2\tau(h_N)+1)^r\right)>r\frac{1}{6}\frac{N}{\log N}\log2\ge\frac{r\log2}{12}\frac{e^{\kappa d}}{\kappa d}\ge\frac{\delta d}{2}\frac{\log2}{12}\frac{e^{\kappa d}}{\kappa d}=\frac{\delta\log2}{24\kappa}e^{\kappa d},

where in the second to last inequality we used r≥δd/2r \ge\delta d/2. This finishes the proof. □

We will add a more explicit bound in a later version.

Note that this result is rather stunning as it shows that the maximal number of facets and thus of vertices of reflexive dd-polytopes grows doubly exponential in dd. And this is essentially optimal as the number of lattice points is easily bounded3 in terms of the volume for which a sharp doubly exponential upper bound was proven in [2]. Previously, it was only known that the number of lattice points of reflexive polytopes can grow doubly exponential in the dimension. However, the previously known reflexive polytopes with this behaviour were just reflexive simplices, namely the Zaks-Perles-Wills simplices [29], also called Sylvester simplices [11].

Acknowledgments and AI declaration

The example, the lemmata and the general result were found in middle of October 2026 using GPT-6.1 Sol based upon prompts with ideas by the first author originally aimed at proving rather than disproving Conjecture 1.1. This paper was humanly written based upon the ChatGPT output.

When writing up this note, it was noticed that this problem was listed as Problem 610 in the internet repository [10] and a solution had already been posted by Kaiwen Jin just a few days after we had found our first 14-dimensional counterexample. Jin’s very nice counterexample is 45-dimensional [20] and its polar consists of cutting [−1,1]40[-1,1]^{40} such that all facet lattice distances from the origin divide 32 and then adding 5-times pyramids of height 2 (which corresponds in Lemma 3.2 to m=5m=5 with (k0,…,k5)=(32,32,16,8,4,2)(k_0,\ldots,k_5)=(32,32,16,8,4,2)). It does not seem to give rise to doubly exponential behaviour though.

This work is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 539867500 as part of the research priority program Combinatorial Synergies.

List of vertices of P8P_8 and P9P_9

(6−60000−1−140−2000−1−131−2000−1−1300000−1−130−3000−1−121−1000−1−12−22000−1−111−1000−1−1101000−1−11−12000−1−100−3000−1−10−12000−1−10−33000−1−10−60000−1−10−66000−1−1−100000−1−1−10−1000−1−1−1−22000−1−1−2−1−1000−1−1−2−22000−1−1−3−30000−1−10006−60−1−100040−2−1−100031−2−1−1000300−1−100030−3−1−10002−22−1−100011−1−1−1000101−1−10000−12−1−10000−12−1−10000−33−1−10000−60−1−10000−66−1−1000−100−1−1000−1−10−1−1000−1−22−1−1000−2−22−1−1000−2−30−1−1000000−120000001−1)(−421680000−1−1−21630000−1−1−14350000−1−1−10220000−1−1−7140000−1−1−210000−1−1−1−10000−1−10−20000−1−12−10000−1−1200000−1−10140000−1−1−6540000−1−1−14980000−1−1−422100000−1−100−4216800−1−100−216300−1−100−143500−1−100−102200−1−100−71400−1−100−2100−1−100−1−100−1−1001−200−1−1002−100−1−1002000−1−10001400−1−100−65400−1−100−149800−1−100−4221000−1−10000−42168−1−10000−2163−1−10000−1435−1−10000−1022−1−10000−714−1−10000−21−1−10000−1−1−1−100001−2−1−100002−1−1−1000020−1−10000014−1−10000−654−1−10000−1498−1−10000−42210−1−1000000−160000001−1).\left( \begin{array}{rrrrrrrr} 6&-6&0&0&0&0&-1&-1\\ 4&0&-2&0&0&0&-1&-1\\ 3&1&-2&0&0&0&-1&-1\\ 3&0&0&0&0&0&-1&-1\\ 3&0&-3&0&0&0&-1&-1\\ 2&1&-1&0&0&0&-1&-1\\ 2&-2&2&0&0&0&-1&-1\\ 1&1&-1&0&0&0&-1&-1\\ 1&0&1&0&0&0&-1&-1\\ 1&-1&2&0&0&0&-1&-1\\ 0&0&-3&0&0&0&-1&-1\\ 0&-1&2&0&0&0&-1&-1\\ 0&-3&3&0&0&0&-1&-1\\ 0&-6&0&0&0&0&-1&-1\\ 0&-6&6&0&0&0&-1&-1\\ -1&0&0&0&0&0&-1&-1\\ -1&0&-1&0&0&0&-1&-1\\ -1&-2&2&0&0&0&-1&-1\\ -2&-1&-1&0&0&0&-1&-1\\ -2&-2&2&0&0&0&-1&-1\\ -3&-3&0&0&0&0&-1&-1\\ 0&0&0&6&-6&0&-1&-1\\ 0&0&0&4&0&-2&-1&-1\\ 0&0&0&3&1&-2&-1&-1\\ 0&0&0&3&0&0&-1&-1\\ 0&0&0&3&0&-3&-1&-1\\ 0&0&0&2&-2&2&-1&-1\\ 0&0&0&1&1&-1&-1&-1\\ 0&0&0&1&0&1&-1&-1\\ 0&0&0&0&-1&2&-1&-1\\ 0&0&0&0&-1&2&-1&-1\\ 0&0&0&0&-3&3&-1&-1\\ 0&0&0&0&-6&0&-1&-1\\ 0&0&0&0&-6&6&-1&-1\\ 0&0&0&-1&0&0&-1&-1\\ 0&0&0&-1&-1&0&-1&-1\\ 0&0&0&-1&-2&2&-1&-1\\ 0&0&0&-2&-2&2&-1&-1\\ 0&0&0&-2&-3&0&-1&-1\\ 0&0&0&0&0&0&-1&2\\ 0&0&0&0&0&0&1&-1 \end{array} \right) \qquad \left( \begin{array}{rrrrrrrr} -42&168&0&0&0&0&-1&-1\\ -21&63&0&0&0&0&-1&-1\\ -14&35&0&0&0&0&-1&-1\\ -10&22&0&0&0&0&-1&-1\\ -7&14&0&0&0&0&-1&-1\\ -2&1&0&0&0&0&-1&-1\\ -1&-1&0&0&0&0&-1&-1\\ 0&-2&0&0&0&0&-1&-1\\ 2&-1&0&0&0&0&-1&-1\\ 2&0&0&0&0&0&-1&-1\\ 0&14&0&0&0&0&-1&-1\\ -6&54&0&0&0&0&-1&-1\\ -14&98&0&0&0&0&-1&-1\\ -42&210&0&0&0&0&-1&-1\\ 0&0&-42&168&0&0&-1&-1\\ 0&0&-21&63&0&0&-1&-1\\ 0&0&-14&35&0&0&-1&-1\\ 0&0&-10&22&0&0&-1&-1\\ 0&0&-7&14&0&0&-1&-1\\ 0&0&-2&1&0&0&-1&-1\\ 0&0&-1&-1&0&0&-1&-1\\ 0&0&1&-2&0&0&-1&-1\\ 0&0&2&-1&0&0&-1&-1\\ 0&0&2&0&0&0&-1&-1\\ 0&0&0&14&0&0&-1&-1\\ 0&0&-6&54&0&0&-1&-1\\ 0&0&-14&98&0&0&-1&-1\\ 0&0&-42&210&0&0&-1&-1\\ 0&0&0&0&-42&168&-1&-1\\ 0&0&0&0&-21&63&-1&-1\\ 0&0&0&0&-14&35&-1&-1\\ 0&0&0&0&-10&22&-1&-1\\ 0&0&0&0&-7&14&-1&-1\\ 0&0&0&0&-2&1&-1&-1\\ 0&0&0&0&-1&-1&-1&-1\\ 0&0&0&0&1&-2&-1&-1\\ 0&0&0&0&2&-1&-1&-1\\ 0&0&0&0&2&0&-1&-1\\ 0&0&0&0&0&14&-1&-1\\ 0&0&0&0&-6&54&-1&-1\\ 0&0&0&0&-14&98&-1&-1\\ 0&0&0&0&-42&210&-1&-1\\ 0&0&0&0&0&0&-1&6\\ 0&0&0&0&0&0&1&-1 \end{array} \right).

(J. Hofscheier) School of Mathematical Sciences, University of Nottingham, Nottingham, NG7 2RD, UK

Email address: [email protected]

(B. Nill) Faculty of Mathematics, Otto-von-Guericke-Universität Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany

Email address: [email protected]

References

  1. [1]Tom M. Apostol. Introduction to analytic number theory. Undergraduate Texts in Mathematics. Springer-Verlag, New York-Heidelberg, January 1976.
  2. [2]Gabriele Balletti, Alexander M. Kasprzyk, and Benjamin Nill. On the maximum dual volume of a canonical Fano polytope. Forum of Mathematics, Sigma, 10:e109, 2022.arxiv.org/abs/1611.02455
  3. [3]Victor Batyrev and Benjamin Nill. Combinatorial aspects of mirror symmetry. In Integer points in polyhedra—geometry, number theory, representation theory, algebra, optimization, statistics, volume 452 of Contemp. Math., pages 35–66. Amer. Math. Soc., Providence, RI, 2008.arxiv.org/abs/math/0703456
  4. [4]Victor V. Batyrev. Dual polyhedra and mirror symmetry for Calabi–Yau hypersurfaces in toric varieties. J. Algebra. Geom., 3(3):493–535, 1994.arxiv.org/abs/alg-geom/9310003
  5. [5]Matthias Beck, Pallavi Jayawant, and Tyrrell B. McAllister. Lattice-point generating functions for free sums of convex sets. Journal of Combinatorial Theory, Series A, 120(6):1246–1262, 2013.arxiv.org/abs/1207.0164
  6. [6]Matthias Beck, Benjamin Nill, Bruce Reznick, Carla Savage, Ivan Soprunov, and Zhiqiang Xu. Let me tell you my favorite lattice-point problem . . . In Integer points in polyhedra—geometry, number theory, representation theory, algebra, optimization, statistics, volume 452 of Contemp. Math., pages 179–187. Amer. Math. Soc., Providence, RI, 2008.DOI
  7. [7]Thomas Bloom and Wouter van Doorn. Optimal bounds for the length and number of unit fraction representations. Hexagon preprint, 2026. hexagon:2610.00407v1.
  8. [8]Benjamin Braun. An Ehrhart series formula for reflexive polytopes. The Electronic Journal of Combinatorics, 13(1):N15, 2006.DOI
  9. [9]Benjamin Braun and Robert Davis. Ehrhart series, unimodality, and integrally closed reflexive polytopes. Annals of Combinatorics, 20(4):705–717, 2016.arxiv.org/abs/1403.5378
  10. [10]Steve Brunton, Matthew J. Colbrook, Maarten V. de Hoop, George Stepaniants, Alex Townsend, and Rachel Ward. Nill’s facet bound for reflexive lattice polytopes. AIM—Open Applied Problems, Problem 610, GitHub, 2026. Last checked October 15, 2026. Accessed October 29, 2026.
  11. [11]Jhon B. Caicedo, Federico Castillo, Martina Juhnke, and Germain Poullot. Sylvester simplices: Triangulations and Ehrhart-theoretic aspects. arXiv preprint, 2026. arXiv:2609.23710.arxiv.org/abs/2609.23710
  12. [12]Cinzia Casagrande. The number of vertices of a Fano polytope. Annales de l’Institut Fourier, 56(1):121–130, 2006.DOI
  13. [13]Giulia Codenotti and Francisco Santos. Hollow polytopes of large width. Proceedings of the American Mathematical Society, 148(2):835–850, 2020.DOI
  14. [14]Heinke Conrads. Weighted projective spaces and reflexive simplices. Manuscripta Mathematica, 107(2):215–227, 2002.DOI
  15. [15]Luis Ferroni. Unimodality shenanigans in Ehrhart theory. arXiv preprint, 2026. arXiv:2609.10513.arxiv.org/abs/2609.10513
  16. [16]Christian Haase and Ilarion V. Melnikov. The reflexive dimension of a lattice polytope. Annals of Combinatorics, 10(2):211–217, 2006.DOI
  17. [17]Christian Haase, Andreas Paffenholz, Lindsay C. Piechnik, and Francisco Santos. Existence of unimodular triangulations - positive results. Memoirs of the American Mathematical Society, 270(1321), March 2021. arXiv:1405.1687.
  18. [18]Douglas Hensley. Lattice vertex polytopes with interior lattice points. Pac. J. Math., 105:183–191, 1983.DOI
  19. [19]Takayuki Hibi. A lower bound theorem for Ehrhart polynomials of convex polytopes. Advances in Mathematics, 105(2):162–165, 1994.
  20. [20]Kaiwen Jin. A counterexample to the facet bound for reflexive polytopes. Unpublished manuscript, GitHub revision b8c005e, October 14, 2026.
  21. [21]Maximilian Kreuzer and Harald Skarke. Complete classification of reflexive polyhedra in four dimensions. Adv. Theor. Math. Phys., 4(6):1209–1230, 2000.
  22. [22]Jeffrey C. Lagarias and Günter M. Ziegler. Bounds for lattice polytopes containing a fixed number of interior points in a sublattice. Can. J. Math., 43(5):1022–1035, 1991.DOI
  23. [23]Aki Mori, Kenta Mori, and Hidefumi Ohsugi. Number of facets of symmetric edge polytopes arising from join graphs. European Journal of Combinatorics, 127:104165, 2025.arxiv.org/abs/2312.11287
  24. [24]Aki Mori, Kenta Mori, and Hidefumi Ohsugi. Facet numbers of non-centrally symmetric reflexive polytopes arising from posets. Discrete & Computational Geometry, 2026.arxiv.org/abs/2511.22981
  25. [25]Benjamin Nill. Gorenstein toric Fano varieties. Manuscripta Mathematica, 116(2):183–210, 2005.DOI
  26. [26]Benjamin Nill. Classification of pseudo-symmetric simplicial reflexive polytopes. In Algebraic and Geometric Combinatorics, volume 423 of Contemporary Mathematics, pages 269–282. American Mathematical Society, Providence, RI, 2006.arxiv.org/abs/math/0511294
  27. [27]Benjamin Nill. Volume and lattice points of reflexive simplices. Discrete & Computational Geometry, 37(2):301–320, 2007.
  28. [28]OpenAI. Short Egyptian fractions. OpenAI Math Release preprint, September 2026. OAI:Short-Egyptian-fractions-September-25-2026.hub.valency.io/works/W_83ugjxhd
  29. [29]Joseph Zaks, Micha A. Perles, and Jörg M. Wills. On lattice polytopes having interior lattice points. Elemente der Mathematik, 37(2):44–46, 1982.

Paper details

vlnc:W_75kfchbm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

GPT-6.1 Sol

View statement
Looking for related papers…
Contents