betaFeedback

WorksMathematics

A Universal Square-Root-of-Six Bound for Kissing Numbers

Abstract

The least positive number α\alpha for which the Euclidean kissing numbers satisfy τn≤αn\tau_n\leq\alpha^n in every positive dimension is 6\sqrt{6}. We show how this follows from a classical spherical-cap estimate and known low-dimensional bounds, then give two elementary proofs that require no tables of kissing numbers. A recurrence for cap areas gives strict inequality in every dimension except two. A second proof packs explicit bodies of revolution into the unit ball. The Lean 4/Mathlib certificate registered in Palomar uses this volume construction to prove the universal bound, the planar value τ2=6\tau_2=6, and the optimality of the base.

The universal base

A kissing configuration in Rn\mathbb{R}^n is a finite set XX satisfying

∥x∥=1(x∈X),⟨x,z⟩≤12(x,z∈X, x≠z).(1)\lVert x\rVert= 1 \quad(x \in X), \qquad\langle x,z\rangle\le\frac{1}{2} \quad(x,z \in X,\ x \ne z). \tag*{(1)}

Multiplying its vectors by two gives the centers of nonoverlapping unit balls touching the unit ball at the origin. We write τn\tau_n for the largest cardinality of a kissing configuration in Rn\mathbb{R}^n. The bounds below also establish the existence of this maximum.

Theorem 1. For every integer n≥1n \ge1,

τn≤(6)n.\tau_n \le\left(\sqrt{6}\right)^n.

Corollary 2. The least real number α>0\alpha> 0 such that τn≤αn\tau_n \le\alpha^n for every n≥1n \ge1 is 6\sqrt{6}. Equivalently,

sup⁡n≥1τn1/n=6.\sup_{n \ge1} \tau_n^{1/n} = \sqrt{6}.

The planar regular hexagon forces every such α\alpha to be at least 6\sqrt{6}. The issue is whether that same base works with prefactor one in every dimension. The question was raised in 2017 [5], motivated by a result of Bakshi, Das, Liu, and Ren [2 Theorem 4.1]. For a finite-index irreducible inclusion N⊂MN \subset M of type II1\mathrm{II}_1 factors, the number of minimal intermediate subfactors is at most τd\tau_d, where d=dim⁡C(N′∩M1)d = \dim_{\mathbb{C}}(N' \cap M_1) and M1M_1 is the basic construction. Theorem 1 thus gives the explicit bound 6d/26^{d/2}. Cohn [6] anticipated the value 6\sqrt{6} and outlined an approach using the finite-dimensional bound of Kabatiansky and Levenshtein.

Throughout, put s=3s = \sqrt{3} and a=s/2a = s/2. For a unit vector xx, let

Cx={y∈Rn:⟨x,y⟩>a∥y∥}.C_x = \{y \in\mathbb{R}^n : \langle x,y\rangle> a\lVert y\rVert\}.

This is the open cone of half-angle π/6\pi/6 about xx. Its intersection with the unit sphere is the corresponding open spherical cap.

Lemma 3. For distinct members x,zx,z of a kissing configuration, Cx∩Cz=∅C_x \cap C_z = \varnothing.

Proof. A vector in the intersection could be normalized to a unit vector uu satisfying

s<⟨x+z,u⟩≤∥x+z∥=2+2⟨x,z⟩≤3=s,s < \langle x+z,u\rangle\le\lVert x+z\rVert= \sqrt{2+2\langle x,z\rangle} \le\sqrt{3}=s,

a contradiction.

In dimension one the only unit vectors are 11 and −1-1, so τ1=2<6\tau_1=2<\sqrt{6}.

Deduction from published bounds. For n≥2n\ge2, write Ωn(θ)\Omega_n(\theta) for the fraction of the area of Sn−1S^{n-1} covered by a cap of angular radius θ\theta. For 0<θ<π/20<\theta<\pi/2, the estimate of Böröczky and Wintsche [3 Corollary 3.2(i)], also stated in [4 Lemma 3.4, (7)], gives

Ωn(θ)≥(sin⁡θ)n−12πn.\Omega_n(\theta)\ge\frac{(\sin\theta)^{n-1}}{\sqrt{2\pi n}}.

By Lemma 3, the open caps of angular radius π/6\pi/6 about the points of XX are disjoint. Hence ∣X∣≤2n−12πn|X|\le2^{n-1}\sqrt{2\pi n}. For n≥6n\ge6 this is strictly smaller than 6n/26^{n/2}. Squaring reduces the comparison to nπ/2<(3/2)nn\pi/2<(3/2)^n. It holds at n=6n=6 since 3π<(3/2)63\pi<(3/2)^6, and (3/2)n/n(3/2)^n/n increases for n>2n>2. In dimensions one and two the elementary bounds are 2 and 6 (in the plane the angles between consecutive directions are at least π/3\pi/3). For dimensions three through five, the upper bounds 12, 24, and 45 in Bachoc and Vallentin [1 Table 5.1] suffice. Thus results already available before the 2017 question imply Theorem 1. The proofs below derive all the needed low-dimensional estimates directly.

First proof: a spherical-cap recurrence

For n≥2n\ge2 and 0<θ<π/20<\theta<\pi/2, polar coordinates about the cap center give

Ωn(θ)=In−2(θ)Jn−2,Im(θ)=∫0θsin⁡mt dt,Jm=∫0πsin⁡mt dt.(2)\Omega_n(\theta)=\frac{I_{n-2}(\theta)}{J_{n-2}},\qquad I_m(\theta)=\int_0^\theta\sin^m t\,\mathrm{d}t,\qquad J_m=\int_0^\pi\sin^m t\,\mathrm{d}t. \tag*{(2)}

Indeed, the polar area element has a factor sin⁡n−2t\sin^{n-2}t, and the transverse area cancels. Put cn=Ωn(π/6)c_n=\Omega_n(\pi/6). Disjointness gives

∣X∣cn≤1.(3)|X|c_n\le1. \tag*{(3)}

Lemma 4. For n≥2n\ge2 and 0<θ<π/20<\theta<\pi/2,

Ωn+2(θ)Ωn(θ)>nn+1sin⁡2θ.\frac{\Omega_{n+2}(\theta)}{\Omega_n(\theta)}>\frac{n}{n+1}\sin^2\theta.

In particular, cn+2>n4(n+1)cn≥16cnc_{n+2}>\frac{n}{4(n+1)}c_n\ge\frac{1}{6}c_n.

Proof. Put m=n−2m=n-2 and h=sin⁡θh=\sin\theta. Substituting u=sin⁡tu=\sin t gives Im(θ)=∫0humw(u) duI_m(\theta)=\int_0^h u^m w(u)\,\mathrm{d}u, with w(u)=(1−u2)−1/2w(u)=(1-u^2)^{-1/2}. Give [0,h][0,h] the probability measure dμ(u)=(m+1)umdu/hm+1\mathrm{d}\mu(u)=(m+1)u^m\mathrm{d}u/h^{m+1} and denote its expectation by EE. Both u2u^2 and w(u)w(u) are strictly increasing. Thus

2(E[u2w]−E[u2]E[w])=∬(u2−v2)(w(u)−w(v)) dμ(u) dμ(v)>0,2\bigl(E[u^2w]-E[u^2]E[w]\bigr)=\iint(u^2-v^2)(w(u)-w(v))\,\mathrm{d}\mu(u)\,\mathrm{d}\mu(v)>0,

since the integrand is positive off the diagonal. It follows that

Im+2(θ)Im(θ)=E[u2w]E[w]>E[u2]=h2m+1m+3.\frac{I_{m+2}(\theta)}{I_m(\theta)}=\frac{E[u^2w]}{E[w]}>E[u^2]=h^2\frac{m+1}{m+3}.

Integration by parts gives Jm+2/Jm=(m+1)/(m+2)J_{m+2}/J_m=(m+1)/(m+2). Dividing the preceding inequality by this ratio gives the result.

First proof of Theorem 1. Direct integration gives c2=1/6c_2 = 1/6 and c3=(2−s)/4>1/15c_3 = (2-s)/4 > 1/15, since s<26/15s < 26/15 by squaring. From c2c_2 the strict recurrence gives cn>6−n/2c_n > 6^{-n/2} in every even dimension n≥4n \ge4. In dimension three, (3) gives ∣X∣<15|X| < 15. Integrality yields ∣X∣≤14<66|X| \le14 < 6\sqrt{6}. To start the odd induction we use the stronger coefficient at n=3n = 3:

c5>316c3>180>6−5/2,c_5 > \frac{3}{16}c_3 > \frac{1}{80} > 6^{-5/2},

where the last inequality follows from 802<6580^2 < 6^5. Thus cn>6−n/2c_n > 6^{-n/2} for every odd n≥5n \ge5 as well. Equation (3) proves the result, with strict inequality for n≠2n \ne2. In each dimension the set of realizable cardinalities is nonempty and bounded in the nonnegative integers, so it has a greatest element.

Second proof: explicit Euclidean volumes

The second proof uses the bodies in the registered Lean certificate. For any unit direction ee, radial integration identifies the fraction of the unit ball in CeC_e with the cap fraction cnc_n of § 2. The common radial factor ∫01rn−1 dr\int_0^1 r^{n-1}\,dr cancels. To obtain a packing bound it suffices to inscribe a body of sufficiently large volume in that sector. We now do this using only Euclidean integration and Lemma 3.

Let BnB_n be the open unit ball and Vn=vol⁡(Bn)V_n = \operatorname{vol}(B_n). Fix e=(1,0)e = (1,0) in orthogonal coordinates (t,v)∈R×Rn−1(t,v) \in\mathbb{R} \times\mathbb{R}^{n-1}. Rotating any measurable body K⊂Ce∩BnK \subset C_e \cap B_n toward the points of XX gives disjoint bodies of equal volume by Lemma 3. Thus

∣X∣vol⁡(K)≤Vn.(4)|X| \operatorname{vol}(K) \le V_n. \tag*{(4)}

For a continuous nonnegative profile gg on [b,c][b,c], Fubini’s theorem gives

vol⁡{(t,v):b<t<c, ∥v∥<g(t)}=Vn−1∫bcg(t)n−1 dt.(5)\operatorname{vol}\{(t,v): b < t < c,\ \lVert v\rVert< g(t)\} = V_{n-1}\int_b^c g(t)^{n-1}\,dt. \tag*{(5)}

The coordinate conditions for containment are

(t,v)∈Ce⟺t>0 and 3∥v∥2<t2,(t,v)∈Bn⟺t2+∥v∥2<1.(6)(t,v) \in C_e \Longleftrightarrow t > 0\ \text{and}\ 3\lVert v\rVert^2 < t^2,\qquad (t,v) \in B_n \Longleftrightarrow t^2 + \lVert v\rVert^2 < 1. \tag*{(6)}

A bicone and the dimensions n≥5n \ge5

Consider the union Kn=Ln∪UnK_n = L_n \cup U_n, where

Ln={(t,v):0<t<a, ∥v∥<t/s},L_n = \{(t,v): 0 < t < a,\ \lVert v\rVert< t/s\},
Un={(t,v):a<t<1, ∥v∥<(2+s)(1−t)}.U_n = \{(t,v): a < t < 1,\ \lVert v\rVert< (2+s)(1-t)\}.

It is a bicone of total height one and common base radius 1/21/2, with the base section omitted. More precisely, each point is a strict convex combination of a point of the disk

D={(a,v):∥v∥<1/2}D = \{(a,v): \lVert v\rVert< 1/2\}

with either 00 or ee. The disk lies in Ce∩BnC_e \cap B_n, whereas 0,e0,e satisfy the weak inequalities defining their closures. The triangle inequality therefore gives Kn⊂Ce∩BnK_n \subset C_e \cap B_n.

By (5), the two cones have volumes aVn−1/(n2n−1)aV_{n-1}/(n2^{n-1}) and (1−a)Vn−1/(n2n−1)(1-a)V_{n-1}/(n2^{n-1}). Hence

vol⁡(Kn)=Vn−1n2n−1,∣X∣≤An:=n2n−1VnVn−1.(7)\operatorname{vol}(K_n) = \frac{V_{n-1}}{n2^{n-1}},\qquad |X| \le A_n := n2^{n-1}\frac{V_n}{V_{n-1}}. \tag*{(7)}

The standard ball-volume formula gives

Vd=πd/2Γ(d/2+1),Vd+2=2πd+2Vd.(8)V_d = \frac{\pi^{d/2}}{\Gamma(d/2+1)},\qquad V_{d+2} = \frac{2\pi}{d+2}V_d. \tag*{(8)}

and consequently

An+2=4(n+1)nAn≤6An(n≥2).(9)A_{n+2}=\frac{4(n+1)}{n}A_n\le6A_n \qquad(n\ge2). \tag*{(9)}

The initial estimates

A5=2563<366=(6)5,A6=60π<216=(6)6A_5=\frac{256}{3}<36\sqrt{6}=(\sqrt{6})^5,\qquad A_6=60\pi<216=(\sqrt{6})^6

follow by squaring the first comparison and using π<22/7\pi<22/7 for the second. Induction on each parity proves ∣X∣<(6)n|X|<(\sqrt{6})^n for all n≥5n\ge5. In dimension two, the same bicone gives ∣X∣≤A2=2π<7|X|\le A_2=2\pi<7, so integrality gives ∣X∣≤6|X|\le6.

Dimension three

Retain L3L_3 and replace the upper cone by

W3={(t,v):a<t<1, ∥v∥<1−t2}.W_3=\{(t,v):a<t<1,\ \|v\|<\sqrt{1-t^2}\}.

Together with L3L_3 this fills the entire sector apart from the joining slice t=at=a, which has volume zero. Indeed, below aa the cone gives the smaller cross-section, and above aa the ball does. In particular W3W_3 lies in the cone because t>at>a implies 3(1−t2)<t23(1-t^2)<t^2. Since V2=πV_2=\pi, formula (5) gives

vol⁡(L3∪W3)=π(s24+∫a1(1−t2) dt)=(2−s)π3>4π45=V315.\operatorname{vol}(L_3\cup W_3)=\pi\left(\frac{s}{24}+\int_a^1(1-t^2)\,\mathrm{d}t\right)=\frac{(2-s)\pi}{3}>\frac{4\pi}{45}=\frac{V_3}{15}.

Here V3=4π/3V_3=4\pi/3, and the strict inequality uses s<26/15s<26/15. Thus (4) yields ∣X∣<15|X|<15, hence ∣X∣≤14<66|X|\le14<6\sqrt{6}.

Dimension four

The bicone alone gives A4=12π>37A_4=12\pi>37. To bring the packing bound below 37 while keeping the volume integral polynomial, retain L4L_4 and raise the upper chord q(t)=(2+s)(1−t)q(t)=(2+s)(1-t) by a quadratic that vanishes at its endpoints. More precisely, set

p(t)=q(t)+14(1−t)(t−a)=(1−t)(14t+2−6s),W4={(t,v):a<t<1, ∥v∥<p(t)}.p(t)=q(t)+14(1-t)(t-a)=(1-t)(14t+2-6s),\qquad W_4=\{(t,v):a<t<1,\ \|v\|<p(t)\}.

The coefficient 14 can be seen from the tangent at aa. There qq meets the circular profile 1−t2\sqrt{1-t^2} at height 1/21/2. For the raised chord with coefficient kk to stay below the circle on the right, its derivative must satisfy −(2+s)+k(1−a)≤−s-(2+s)+k(1-a)\le-s, or k≤2/(1−a)=8+4sk\le2/(1-a)=8+4s. The largest integral candidate is 14. The following calculation checks containment on the entire interval. For a≤t≤1a\le t\le1 we have p(t)≥0p(t)\ge0, since 14t+2−6s≥s+2>014t+2-6s\ge s+2>0. Writing u=t−au=t-a, a factorization gives

1−t2−p(t)2=(1−t)(t−a)(4s−6+(126s−140)u+196u2)≥0.(10)1-t^2-p(t)^2=(1-t)(t-a)\left(4s-6+(126s-140)u+196u^2\right)\ge0. \tag*{(10)}

Indeed, u≥0u\ge0 and the constant and linear coefficients in the last factor are positive because s>3/2s>3/2. Thus W4W_4 lies in the ball, and t>at>a puts it in the cone as in dimension three.

A lower bound for the volume avoids cancellation in the exact integral. Write δ=1−a\delta=1-a, x=(t−a)/δx=(t-a)/\delta, and b=14δ2b=14\delta^2. Then p(t)=(1−x)(1/2+bx)p(t)=(1-x)(1/2+bx). The lower cone contributes a/32a/32 to the profile integral in (5). For the upper piece use (1/2+bx)3≥1/8+3bx/4(1/2+bx)^3\ge1/8+3bx/4 on [0,1][0,1]. Integrating gives

T:=vol⁡(L4∪W4)V3≥a32+δ∫01(1−x)3(18+3b4x) dx=a+δ32+3bδ80=132+2140δ3.T:=\frac{\operatorname{vol}(L_4\cup W_4)}{V_3}\ge\frac{a}{32}+\delta\int_0^1(1-x)^3\left(\frac{1}{8}+\frac{3b}{4}x\right)\,\mathrm{d}x =\frac{a+\delta}{32}+\frac{3b\delta}{80}=\frac{1}{32}+\frac{21}{40}\delta^3.

Since s<7/4s < 7/4, we have δ>1/8\delta> 1/8, so T>1/32+12(1/8)3=33/1024T > 1/32 + \frac{1}{2}(1/8)^3 = 33/1024. Also V4/V3=3π/8<33/28V_4/V_3 = 3\pi/8 < 33/28 by π<22/7\pi< 22/7. The packing inequality (4) therefore gives

∣X∣≤V4/V3T<33/2833/1024=2567<37.|X| \le\frac{V_4/V_3}{T} < \frac{33/28}{33/1024} = \frac{256}{7} < 37.

Thus ∣X∣≤36=(6)4|X| \le36 = (\sqrt{6})^4.

Completion of the second proof of Theorem 1. The volume estimates cover every n≥2n \ge2, and the dimension-one observation covers n=1n = 1. Thus every kissing configuration satisfies the claimed bound, independently of the first proof. As before, the finite nonempty set of realizable cardinalities has an attained maximum, which is τn\tau_n.

Sharpness and the equality case

Proof of Corollary 2. The vertices of a regular hexagon are

(1,0),(12,32),(−12,32),(−1,0),(−12,−32),(12,−32).(1,0),\quad\left(\frac{1}{2},\frac{\sqrt{3}}{2}\right),\quad\left(-\frac{1}{2},\frac{\sqrt{3}}{2}\right),\quad(-1,0),\quad\left(-\frac{1}{2},-\frac{\sqrt{3}}{2}\right),\quad\left(\frac{1}{2},-\frac{\sqrt{3}}{2}\right).

with every inner product between distinct vertices at most 1/21/2. These six unit vectors form a kissing configuration. Theorem 1 gives the reverse bound, so τ2=6\tau_2 = 6.

The base 6\sqrt{6} works by the theorem. Conversely, any positive universal base α\alpha must satisfy 6=τ2≤α26 = \tau_2 \le\alpha^2, hence α≥6\alpha\ge\sqrt{6}. Likewise, τn1/n≤6\tau_n^{1/n} \le\sqrt{6} in every dimension, with equality at n=2n = 2, proving the supremum statement.

Corollary 5. Dimension two is the unique maximizer of τn1/n\tau_n^{1/n} among positive integers nn.

Proof. The regular hexagon gives equality in dimension two, and the first proof of Theorem 1 gives τn<6n/2\tau_n < 6^{n/2} for every n≠2n \ne2.

Formal statements and provenance

The Palomar registration PALOMAR-2026-09-05-000002, version 1 of September 5, 2026, certifies eight statements in Lean 4/Mathlib. They bound every finite kissing configuration, establish the existence of a greatest realizable cardinality, and give the planar value and equivalent formulations of the optimal base. The proof uses the Euclidean bodies of Section 3, including the exceptional profiles in dimensions three and four. For the four-dimensional body the certificate evaluates the integral exactly, while the lower estimate in Section 3.3 suffices here. The cap recurrence and its strict consequence, Corollary 5, are additional informal results.

The registered sources contain the formal statements, proofs, and instructions for reproducing the checks. Comparator matched the challenge statements to the solution. The recorded verification includes checking by Lean’s kernel and independent replay by NanoDa.

The author’s provenance correction of September 5, 2026 attributes the original argument, manuscript, formalization, and registration preparation to OpenAI’s GPT-6 Astra, working from the author’s prompts. The archived metadata records the earlier credits to GPT-5.6 Pro and ChatGPT Work for the proof and to GPT-6 Pro for packaging. OpenAI’s GPT-6.2 Astra assisted with the present manuscript revision. The Lean development and Palomar record are those of the registered version.

Beijing Institute of Mathematical Sciences and Applications (BIMSA), Huairou District, Beijing 101408, China

[email protected]

References

  1. [1]C. Bahoc and F. Vallentin, New upper bounds for kissing numbers from semidefinite programming, J. Amer. Math. Soc. 21 (2008), 909–924.
  2. [2]K. C. Bakshi, S. Das, Z. Liu, and Y. Ren, An angle between intermediate subfactors and its rigidity, Trans. Amer. Math. Soc. 371 (2019), 5973–5991.arxiv.org/abs/1710.00285
  3. [3]K. Böröczky, Jr., and G. Wintsche, Covering the sphere by equal spherical balls, in Discrete and Computational Geometry: The Goodman–Pollack Festschrift, Algorithms Combin. 25, Springer, 2003, 235–251.DOI
  4. [4]M. Naszódi, On some covering problems in geometry, Proc. Amer. Math. Soc. 144 (2016), 3555–3562.arxiv.org/abs/1404.1691
  5. [5]S. Palcoux, Upper bound of the kissing number in n dimensions, MathOverflow question 282644 (2017), https://mathoverflow.net/questions/282644/.
  6. [6]H. Cohn, Answer to “Upper bound of the kissing number in n dimensions”, MathOverflow answer 282685 (2017), https://mathoverflow.net/a/282685/.

Paper details

vlnc:W_u39vknpc
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 3

downloads · 0

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

  • Mathematics

OpenAI ChatGPT GPT-6 Astra, OpenAI ChatGPT GPT-6.2 Astra

View statement
Looking for related papers…
Contents