betaFeedback

WorksMathematics

Euclidean view-obstruction in dimensions eight to ten and the asymptotic deficit

Abstract

The Euclidean view-obstruction problem asks for the critical radius of balls centered at the half-integer lattice that meet every positive ray from the origin. Writing ΔnΔ_n for its square, we determine Δ8=8/17Δ_8=8/17, Δ9=1087/1980Δ_9=1087/1980 and Δ10=7/11Δ_{10}=7/11 by computer-assisted proofs using Kahlon's seven-dimensional theorem.The upper bounds combine full-support subspace averaging, projected lattice coverings and exact finite certificates; dimension ten also requires a finite search through short integer relations. Explicit directions attain all three values, including a ten-dimensional direction previously listed by Gritsenko, Skoruppa and Zagier. In arbitrary dimension, we express their theta-block invariant in geometric normalization and derive the known bounds clog⁡(n+1)≤n/12−Δn≤Clog⁡3(n+1)c\log(n+1) \le n/12 - Δ_n \le C\log^3(n+1), with repeated coordinates allowed throughout. Thus Δn∼n/12Δ_n \sim n/12, while the deficit tends to infinity more slowly than every positive power of nn. The supplementary archive contains the executable verifiers, exact certificates and archived verification records.

2020 Mathematics Subject Classification. Primary 11H31; Secondary 11H06, 11H55, 42A05.

Keywords: view obstruction; Euclidean balls; lattice covering radius; Korkin–Zolotarev reduction; computer-assisted proof; theta blocks; trigonometric polynomials.

Introduction

The geometric question

Consider a periodic array of identical opaque balls, with one ball centered at every point of

cn+Zn,cn=(1/2,…,1/2).c_n + \mathbb{Z}^n,\qquad c_n = (1/2,\ldots,1/2).

An observer is at the origin. A viewing direction u∈R0nu \in\mathbb{R}^n_0 is blocked if the ray {tu:t>0}\{tu : t > 0\} meets at least one of the balls; tangency counts as intersection. The problem asks for the threshold radius at which all such directions are blocked. The half-lattice translation avoids placing an obstacle at the observer and makes the coordinate symmetries explicit. Only positive directions are considered, but repeated coordinates are allowed throughout.

For a precise definition, let rnr_n be the infimum of the radii that block every positive ray, and put Δn=rn2\Delta_n = r_n^2. If the reference ball has diameter one and its dilation factor is denoted by ν(n)\nu(n), then

rn=Δn,ν(n)=2rn,ν(n)2=4Δn.(1)r_n = \sqrt{\Delta_n},\qquad\nu(n) = 2r_n,\qquad\nu(n)^2 = 4\Delta_n. \tag*{(1)}

This conversion is essential when comparing the older literature with squared-distance computations.

The geometry has an equivalent periodic formulation. For v=(v1,…,vn)∈Z>0nv=(v_1,\ldots,v_n)\in\mathbb{Z}_{>0}^n, define

g(t)=dist⁡(t,1/2+Z)2=({t}−1/2)2,Fv(t)=∑i=1ng(vit),d(v)=min⁡t∈TFv(t),(2)g(t)=\operatorname{dist}(t,1/2+\mathbb{Z})^2=(\{t\}-1/2)^2,\qquad F_v(t)=\sum_{i=1}^{n}g(v_i t),\qquad d(v)=\min_{t\in\mathbb{T}}F_v(t), \tag*{(2)}

where T=R/Z\mathbb{T}=\mathbb{R}/\mathbb{Z} and {t}\{t\} denotes the fractional part of tt. Then

Δn=sup⁡v∈Z>0nd(v).(3)\Delta_n=\sup_{v\in\mathbb{Z}_{>0}^n}d(v). \tag*{(3)}

The reduction, including the distinction between an infimum and actual intersection at the critical radius, is proved in § 2. Multiplication of vv by a positive integer does not change d(v)d(v). One may therefore restrict the supremum to primitive vectors, although the finite censuses deliberately include all their multiples.

The integral of gg is 1/121/12. Consequently Δn≤n/12\Delta_n\le n/12. Determining the exact constant requires a direction that remains as far as possible from the shifted lattice over an entire period, together with an upper bound valid for infinitely many directions. The second requirement is the central obstacle to a finite computation.

Historical and analytic context

The problem originates in Cusick’s papers of 1972–1974 [7, 8, 9], which formulate view obstruction in the geometry of numbers. The replacement of balls by axis-parallel cubes connects the problem with simultaneous Diophantine approximation and the lonely runner problem [22, 10]. For a cube of side α\alpha centered at a half-integer point, intersection means that, at some time tt, all coordinates satisfy

dist⁡(tvi,Z)≥(1−α)/2.\operatorname{dist}(tv_i,\mathbb{Z})\ge(1-\alpha)/2.

For a ball the condition instead bounds the sum of the squared deviations from half-integers. This coupling of coordinates leads naturally to projected lattices and covering radii.

For spheres, Dumir and Hans-Gill determined the three-dimensional constant in 1986 [11]. Chen studied dimensions three and four in 1994 [4, 5]; Dumir, Hans-Gill and Wilker treated dimension four in 1994 and dimension five in 1996 [13, 14]. The general theory of rational subspaces, critical parameters and related isolation phenomena was developed in [12, 15]; Chen also studied the connection with Kronecker’s theorem [6]. Kahlon’s preprint [17] proves the values in dimensions six and seven using exact lattice searches and transference bounds, including Banaszczyk’s theorem [1]. Our use of that preprint is confined to its seven-dimensional upper bound and to the context of the method. The computations in dimensions eight, nine and ten are separate from those of Kahlon and are supplied with this paper.

A second connection is to theta blocks. Gritsenko, Skoruppa and Zagier [16 Sections 5–6] study the same minimum of a sum of periodic quadratics, with a factor-of-two change of normalization. Their Theorem 5.3 already establishes logarithmic and cubed-logarithmic bounds for the corresponding deficit. We give a geometric formulation of their asymptotic theorem and a detailed derivation that includes repeated coordinates. The analytic inputs are Littlewood’s inequality [20, 18] and the integer-coefficient cosine-polynomial construction of Belov and Konjagin [2]. The experimental table in [16 Table 2] also supplies the lower bounds 37/8037/80 and 89/16889/168 in dimensions eight and nine. The corresponding values considered here are 8/178/17 and 1087/19801087/1980. The direction (1,1,2,3,4,5,6,7,8,9)(1,1,2,3,4,5,6,7,8,9) and its ten-dimensional lower bound 7/117/11 already appear in [16 Table 2], after normalization. For this direction, the further finite-dimensional task is to prove the matching global upper bound.

Main results

The finite-dimensional contribution consists of the exact constants in dimensions eight through ten, the finite reductions proving their upper bounds, and explicit directions attaining them. The computations are organized into three components: plane classification and a bounded direction census in dimension eight; an enlarged catalogue, fourth-moment constraints and a bounded census in dimension nine; and a finite relation-lattice search in dimension ten. Appendix § 13 relates these proof components to the supplementary archive.

Theorem 1.1 (Exact constants). The squared critical radii in dimensions eight, nine and ten are

Δ8=817,Δ9=10871980,Δ10=711.\Delta_8 = \frac{8}{17}, \qquad\Delta_9 = \frac{1087}{1980}, \qquad\Delta_{10} = \frac{7}{11}.

The respective primitive attaining directions can be chosen as

v(8)=(1,2,3,4,5,7,9,11),v^{(8)} = (1,2,3,4,5,7,9,11),
v(9)=(2,3,5,7,8,11,13,15,18),v^{(9)} = (2,3,5,7,8,11,13,15,18),
v(10)=(1,1,2,3,4,5,6,7,8,9).v^{(10)} = (1,1,2,3,4,5,6,7,8,9).

For each of these three dimensions, balls of radius Δn\sqrt{\Delta_n} itself block every positive real ray.

The theorem supplies one attaining direction in each dimension. The repeated entry in v(10)v^{(10)} is intentional; equality directions need not have distinct coordinates. The dependency order of the upper bounds is

Δ7⟶Δ8⟶Δ9⟶Δ10.\Delta_7 \longrightarrow\Delta_8 \longrightarrow\Delta_9 \longrightarrow\Delta_{10}.

The dimension-ten proof also uses the published KZ inequalities of Pendavingh and van Z&#xwam [21], with their positive error terms retained.

Theorem 1.2 (Known asymptotic bounds in geometric normalization). Set En=n/12−ΔnE_n = n/12 - \Delta_n. There are absolute constants c,C>0c, C > 0 such that for every n≥1n \ge1,

clog⁡(n+1)≤En≤Clog⁡3(n+1).(4)c\log(n+1) \le E_n \le C\log^3(n+1). \tag*{(4)}

In particular,

Δn∼n12,En⟶∞,En=o(nε)(ε>0).\Delta_n \sim\frac{n}{12}, \qquad E_n \longrightarrow\infty, \qquad E_n = o(n^{\varepsilon}) \quad(\varepsilon> 0).

All logarithms are natural. The bounds in (4) are the bounds of [16 Theorem 5.3] after normalization. § 7 gives a derivation from the two stated analytic inputs for arbitrary positive integer multisets. Determining the precise order of EnE_n remains open.

Proof strategy

The proof has three stages. First, the closure of a real ray on the torus reduces the geometric problem to rational subspaces. An averaging lemma bounds a full-support subspace of dimension dd by (n−d)/12(n-d)/12, and a finite quadratic calculation evaluates any fixed integer direction.

Second, projected lattices transfer the remaining upper-bound problem to Korkin–Zolotarev heights. Covering estimates and plane bounds constrain their product. In dimensions eight and nine this produces a finite norm bound for every possible counterexample. Dimension ten additionally uses bounded short relations and a finite sequence of Gram-matrix extensions. Exact finite certificates close these reductions; their verification and the recorded coverage checks are described in § 8 and Appendix § 13.

nnΔn\Delta_nAn attaining directionSource
10(1)Immediate
21/20(1, 2)[9]
33/28(1, 2, 3)[11]
411/60(1, 1, 2, 3)[5, 13]
541/168(2, 3, 5, 7, 9)[14]
6107/312(2, 3, 5, 7, 9, 12)[17]
773/180(1, 3, 4, 7, 8, 10, 11)[17]
88/17(1, 2, 3, 4, 5, 7, 9, 11)Section 4
91087/1980(2, 3, 5, 7, 8, 11, 13, 15, 18)Section 5
107/11(1, 1, 2, 3, 4, 5, 6, 7, 8, 9)Section 6

Table 1. Squared critical radii and one attaining direction. Rows 8–10 give the exact constants established in this paper.

Finally, in arbitrary dimension a Fourier expansion connects the distance function to cosine sums. The McGehee–Pigno–Smith inequality gives the logarithmic lower bound for the deficit, including repeated speeds. A nonnegative integer-coefficient construction of Belov–Konyagin gives the cubed-logarithmic upper bound. This argument is independent of the finite certificates.

Geometric reductions and exact distance evaluation

This section relates positive real rays to rational subspaces and gives an exact finite calculation for a fixed integer direction. These two reductions separate the global upper-bound problem from the verification of explicit lower bounds.

Rational subspaces and real rays

For a rational subspace W⊆RnW \subseteq\mathbb{R}^n, write

β(W)=dist⁡(W,cn+Zn)2.\beta(W) = \operatorname{dist}(W, c_n + \mathbb{Z}^n)^2.

Call WW full-support if it is contained in no coordinate hyperplane. A subspace containing a positive vector has this property; the converse need not hold. We use full-support statements because they remain invariant under independent changes of coordinate signs.

Proposition 2.1. The identity (3) holds for the critical radius defined as an infimum. For a positive real vector uu, let WW be the smallest rational linear subspace containing uu. Then

inf⁡t>0dist⁡(tu,cn+Zn)2=β(W).\inf_{t>0} \operatorname{dist}(tu, c_n + \mathbb{Z}^n)^2 = \beta(W).

Proof. The continuous version of Kronecker’s theorem identifies the closure of the positive orbit tu mod Zntu \bmod\mathbb{Z}^n with (W+Zn)/Zn(W + \mathbb{Z}^n)/\mathbb{Z}^n; see the view-obstruction formulation in [12, 17]. Every tail of this positive orbit has the same closure. The squared distance to cn+Znc_n + \mathbb{Z}^n is continuous on the torus, so its infimum on the orbit is its minimum on that subtorus. This minimum equals β(W)\beta(W).

Since WW is rational and W∩R>0nW \cap\mathbb{R}^n_{>0} is a nonempty relatively open set, it contains a positive rational vector, hence a positive integer vector vv. Thus β(W)≤d(v)\beta(W) \le d(v). Integer directions are themselves allowed real directions, giving equality of the two suprema. A radius strictly above their square root blocks every ray, and a radius strictly below it misses some ray. This proves the assertion about the infimum. ∎

At the critical radius, an orbit with an unattained distance infimum requires separate care. For dimensions eight through ten the strict subspace margins established below settle this point.

A full-support averaging lemma

Lemma 2.2. If W⊆RnW \subseteq\mathbb{R}^{n} is a full-support rational subspace of dimension dd, then β(W)≤(n−d)/12\beta(W) \leq(n-d)/12.

Proof. Let BB be an integral basis matrix for W∩ZnW \cap\mathbb{Z}^{n}, with rows bib_i, and let P=B(BTB)−1BTP = B(B^{T}B)^{-1}B^{T}. For uniform u∈Tdu \in\mathbb{T}^{d}, set Xi={biu}−1/2X_i = \{b_i u\} - 1/2. Each bib_i is nonzero, so XiX_i has mean zero and second moment 1/121/12. If bi,bjb_i,b_j are linearly independent, their joint character map is onto T2\mathbb{T}^{2}, and EXiXj=0\mathbb{E}X_iX_j = 0. Otherwise write bi=pab_i = pa, bj=qab_j = qa for a primitive integral row aa and nonzero integers p,qp,q. The Fourier series of the periodic sawtooth, or direct integration over a common period, gives

EXiXj=gcd⁡(p,q)212pq,Pij=pq a(BTB)−1aT.\mathbb{E}X_iX_j = \frac{\gcd(p,q)^2}{12pq}, \qquad P_{ij} = pq\,a(B^{T}B)^{-1}a^{T}.

Their product is nonnegative, including when pq<0pq < 0. Hence E∥PX∥2≥tr⁡(P)/12=d/12\mathbb{E}\lVert PX\rVert^{2} \geq\operatorname{tr}(P)/12 = d/12, and E∥(I−P)X∥2≤(n−d)/12\mathbb{E}\lVert(I-P)X\rVert^{2} \leq(n-d)/12. Since Bu∈WBu \in W and X=Bu−z−cnX = Bu-z-c_n for an integer vector zz, (I−P)X(I-P)X is the negative of the orthogonal residual of a half-integer center. Every such squared residual is at least β(W)\beta(W).

Odd-center witnesses and complete periods

For v∈Z>0nv \in\mathbb{Z}^{n}_{>0}, put A=∥v∥2A = \lVert v\rVert^{2}. Every odd integer vector z∈(2Z+1)nz \in(2\mathbb{Z}+1)^{n} gives the valid upper bound

d(v)≤dist⁡(z/2,Rv)2=A∥z∥2−(v⋅z)24A.(5)d(v) \leq\operatorname{dist}(z/2,\mathbb{R}v)^{2} = \frac{A\lVert z\rVert^{2}-(v\cdot z)^{2}}{4A}. \tag*{(5)}

This inequality needs no assumption that z/2z/2 is the nearest center at the orthogonal projection time. To establish a lower bound, however, one must minimize over all possible times, not merely exhibit a witness.

On [0,1/2][0,1/2], form the ordered union of 00, 1/21/2 and all crossing times k/vik/v_i in that interval. Inside each resulting cell [a,b][a,b] the nearest odd center is fixed, say zz, and

Fv(t)=At2−(v⋅z)t+∥z∥2/4.(6)F_v(t) = At^2 - (v\cdot z)t + \lVert z\rVert^{2}/4. \tag*{(6)}

Its minimum on the closed cell is obtained by clipping t0=(v⋅z)/(2A)t_0 = (v\cdot z)/(2A) to [a,b][a,b]. The minimum of these cell minima is d(v)d(v), since Fv(1−t)=Fv(t)F_v(1-t) = F_v(t). All quantities are rational. This is also an independent way to audit a fast nearest-center sweep, whose projection values are useful as upper witnesses.

For the three displayed directions, the numbers of half-period cells are 16, 32 and 14. Complete clipped calculations give the minima in Theorem 1.1. The squared norms are respectively 306, 990 and 286. The last direction reproduces the lower bound in [16 Table 2]. Equality times in [0,1/2][0,1/2] are

88/51, 25/51949/495, 94/495, 104/495, 149/495109/22.(7)\begin{aligned} 8 &\quad8/51,\ 25/51 \\ 9 &\quad49/495,\ 94/495,\ 104/495,\ 149/495 \\ 10 &\quad9/22. \tag*{(7)} \end{aligned}

The supplementary verifier regenerates all cells, odd centers, clipped minimizers and exact values using rational arithmetic. These calculations establish the three lower bounds independently of the upper-bound censuses and relation searches.

Lattice coverings and finite product bounds

We next bound the distance to a rational subspace through covering radii of projected lattices. The resulting inequalities reduce a global upper bound to a finite product calculation. Background on reduced bases and reciprocal lattices can be found in [3, 19]. For a rational dd-dimensional subspace WW, let D=covol⁡(W∩Zn)2D = \operatorname{covol}(W \cap\mathbb{Z}^{n})^{2}. The lattice Γ=PW⊥Zn\Gamma= P_{W^\perp}\mathbb{Z}^{n} has rank k=n−dk = n-d and squared covolume 1/D1/D. Indeed its dual is W⊥∩ZnW^\perp\cap\mathbb{Z}^{n}, and complementary primitive subgroups of the unimodular lattice Zn\mathbb{Z}^{n} have equal covolumes. Choose a Korkin–Zolotarev (KZ) basis of Γ\Gamma, and let h1,…,hkh_1,\ldots,h_k be its squared Gram–Schmidt lengths. Then

∏i=1khi=1/D,hi+1≥34hi.(8)\prod_{i=1}^{k} h_i = 1/D,\qquad h_{i+1} \ge\frac{3}{4}h_i. \tag*{(8)}

The second assertion follows from the shortest-vector property in each projection and a size-reduction coefficient of absolute value at most 1/21/2.

Lemma 3.1 (Fiber covering). Let WrW_r be obtained by adjoining integer lifts of the first rr projected basis vectors to WW. If WW is full-support, then

β(W)≤β(Wr)+ρr2≤k−r12+ρr2.(9)\beta(W) \le\beta(W_r) + \rho_r^2 \le\frac{k-r}{12} + \rho_r^2. \tag*{(9)}

where ρr\rho_r is the covering radius of the lattice generated by those first rr projected vectors in their span.

Proof. Choose a half-integer center minimizing the normal distance to WrW_r. Adding an integer combination of the chosen lifts leaves this normal component unchanged and changes the component in Wr∩W⊥W_r \cap W^\perp by an element of the prefix lattice. Choose such a combination with remaining component of norm at most ρr\rho_r. The two errors are orthogonal. The second inequality is Lemma 2.2.

Lemma 3.2 (Adjacent two-dimensional blocks). For adjacent KZ heights h,kh,k, the corresponding projected rank-two lattice has covering radius squared at most

h+k4−512(h−k)+.(10)\frac{h+k}{4} - \frac{5}{12}(h-k)_+. \tag*{(10)}

For a matching MM of disjoint adjacent pairs in {1,…,r}\{1,\ldots,r\}, define

Ur,M(h)=14∑i=1rhi−512∑i∈M(hi−hi+1).(11)U_{r,M}(h) = \frac{1}{4}\sum_{i=1}^{r}h_i - \frac{5}{12}\sum_{i\in M}(h_i-h_{i+1}). \tag*{(11)}

Then ρr2≤Ur,M(h)\rho_r^2 \le U_{r,M}(h).

Proof. Write the two-dimensional basis as (h,0)(\sqrt{h},0) and (μh,k)(\mu\sqrt{h},\sqrt{k}), with 0≤μ≤1/20 \le\mu\le1/2 and k+μ2h≥hk+\mu^2h \ge h. Triangulating a fundamental parallelogram gives the squared circumradius bound

ρ2≤(k+μ2h)(k+(1−μ)2h)4k.\rho^2 \le\frac{(k+\mu^2h)(k+(1-\mu)^2h)}{4k}.

This expression decreases with μ\mu on [0,1/2][0,1/2]: after scaling h=1h=1, its derivative has the sign of (2μ−1)(k−μ(1−μ))(2\mu-1)(k-\mu(1-\mu)), which is nonpositive because k≥3/4k \ge3/4. If k≥hk \ge h, the value at μ=0\mu=0 suffices. If k<hk<h, set z=1−k/h≤1/2z=\sqrt{1-k/h}\le1/2 and use μ≥z\mu\ge z to obtain ρ2≤h/[2(1+z)]\rho^2 \le h/[2(1+z)]. The required comparison reduces to z(1−2z)(3+2z)≥0z(1-2z)(3+2z)\ge0. Round successive blocks from the end of a prefix, adding the orthogonal squared errors. Finally replace (hi−hi+1)+(h_i-h_{i+1})_+ by hi−hi+1h_i-h_{i+1}, which only enlarges the upper bound.

Why a finite vertex certificate is sufficient

The inequalities (8) and (9)–(11) produce rational polyhedra in the positive heights. The objective is a lower bound on their product, hence an upper bound on DD or AA. For the unbounded polyhedra used here, the following detail is needed. Every coordinate has a positive lower bound, and every recession direction dd satisfies d1≥0d_1 \ge0, di+1≥3di/4d_{i+1} \ge3d_i/4. Thus all its coordinates are nonnegative. By the vertex–recession decomposition of a pointed polyhedron [23], any feasible point is a convex combination of vertices plus a nonnegative recession vector. Increasing coordinates cannot decrease their product. Concavity of ∑ilog⁡hi\sum_i \log h_i then shows that the product is bounded below by the smallest vertex product. This argument does not presume compactness.

The vertex-traversal programs first relax each right-hand side by a fixed, tiny positive rational number. The feasible region expands, so a lower bound on the expanded region remains valid for the original one. Exact inverse-matrix and slack checks establish simplicity of each visited vertex. Releasing each active constraint identifies every outgoing edge by rational ratio comparisons, or certifies it as unbounded. Closure of the queue, together with connectivity of the vertex graph, establishes exhaustiveness. The perturbation is not trusted generically: nondegeneracy is checked at every vertex. All products and comparisons use arbitrary-precision integers.

Dimension eight

Put T8=8/17T_8 = 8/17. We first reduce a classification of full-support planes to finite checks, then bound the norm of a putative counterexample. The supplementary archive contains the exact certificates for the plane classification, the norm bound and the final direction census.

Proposition 4.1 (Eight-dimensional plane certificate). Every full-support rational plane W⊂R8W \subset\mathbb{R}^{8} satisfies β(W)≤43/100\beta(W) \le43/100, unless, up to signed coordinate permutations, its saturated integer lattice has basis matrix

B=(031−21−111122−12130).(12)B = \begin{pmatrix} 0 & 3 \\ 1 & -2 \\ 1 & -1 \\ 1 & 1 \\ 1 & 2 \\ 2 & -1 \\ 2 & 1 \\ 3 & 0 \end{pmatrix}. \tag*{(12)}

For this exceptional plane, BTB=21I2B^{T}B = 21I_2, D=441D = 441, and β(W)≤19/42\beta(W) \le19/42.

Proof and finite certificate. Suppose β(W)>43/100\beta(W) > 43/100. There are six projected KZ heights. The ordinary nearest-plane bound in (9), with xi=3hix_i = 3h_i, implies

∑i=1rxi>r−21/25,xi+1≥3xi/4.\sum_{i=1}^{r} x_i > r - 21/25,\qquad x_{i+1} \ge3x_i/4.

Relax the strict inequalities. At a tight prefix jj, subtraction of the preceding prefix inequality gives xj≤1x_j \le1, whereas the next prefix gives xj+1≥1x_{j+1} \ge1. Thus the ratio constraint immediately following a tight prefix cannot also be tight. At a vertex, active ratio edges consequently form geometric blocks, each terminated by a tight prefix. The blocks are precisely the 32 compositions of 6; the first block has sum equal to its length minus 21/25, and each subsequent block has sum equal to its length. The positive coordinate bounds and the recession argument of § 3 reduce the product minimum to these vertices. Their exact products give D<6843D < 6843, hence D≤6842D \le6842.

Choose a Gauss-reduced basis a,ba,b of W∩Z8W \cap\mathbb{Z}^{8}, and write

s=∥a∥2,t=a⋅b,u=∥b∥2,∣2t∣≤s≤u,D=su−t2.s=\lVert a\rVert^{2},\qquad t=a\cdot b,\qquad u=\lVert b\rVert^{2},\qquad|2t|\le s\le u,\qquad D=su-t^{2}.

Then 3s2≤4D3s^{2}\le4D, so s≤95s\le95. If s=1s=1, WW contains a coordinate axis, and deletion of that coordinate, followed by (3) in dimension seven, gives β(W)≤Δ7<43/100\beta(W)\le\Delta_{7}<43/100.

The remaining finite classification is implemented by verify_planes.js. It first covers all 174 047 114 sorted positive eight-tuples of squared norm at most 4000, retaining only 195 not certified below 43/100. Let HH be their norm set. No norm s≤95s\le95 is in HH, so aa cannot have full support. For every kk with ∥b+ka∥2≤4000\lVert b+ka\rVert^{2}\le4000 whose norm is not in HH, b+kab+ka must have a zero coordinate. Distinct kk require distinct nonzero coordinates of aa; there are at most min⁡(7,s)\min(7,s) of them. All relevant kk lie in [−46,46][-46,46]. This leaves ten Gram matrices.

For each forced zero at kk, a row of the basis matrix is (q,−kq)(q,-kq) with q≠0q\ne0. Enumerating these rows and all completions with the prescribed Gram matrix yields 63 row multisets, of which 45 are saturated. Saturation is checked by the gcd of all 2×22\times2 minors. Forty-four have explicit odd-center witnesses at most 43/100; the remaining matrix is (12). For it, the all-ones odd vector gives [8−(112+32)/21]/4=19/42[8-(11^{2}+3^{2})/21]/4=19/42. These finite row and witness checks are retained in planes_certificate.json.

Suppose now that a primitive vv has d(v)>T8d(v)>T_{8}, and put A=∥v∥2A=\lVert v\rVert^{2}. The first projected KZ vector lifts to a plane W⊃RvW\supset\mathbb{R}v. Since this vector is primitive in the projected lattice, the lift together with vv is a saturated basis of W∩Z8W\cap\mathbb{Z}^{8}, so D=Ah1D=Ah_{1}. If WW is exceptional, the one-dimensional fiber bound gives

T8<d(v)≤19/42+441/(4A),A<157437/26<6056.T_{8}<d(v)\le19/42+441/(4A),\qquad A<157437/26<6056.

Otherwise, h1>4(T8−43/100)=69/425h_{1}>4(T_{8}-43/100)=69/425. For every prefix and adjacent matching, (9) implies

∑i≤rhi−53∑i∈M(hi−hi+1)≥17r−2351.(13)\sum_{i\le r}h_i-\frac{5}{3}\sum_{i\in M}(h_i-h_{i+1})\ge\frac{17r-23}{51}. \tag*{(13)}

Together with the six adjacent KZ constraints and the h1h_{1} bound, these are 60 inequalities. The exact outward-perturbed vertex check visits 2933 vertices, 19 830 directed bounded edges and 701 unbounded edges, proving ∏hi>1/13700\prod h_i>1/13700. Thus every strict counterexample must have A≤13699A\le13699.

The complete direction census at this bound contains 18 766 118 836 sorted positive eight-tuples. It certifies

17[A∥z∥2−(v⋅z)2]≤32A(14)17[A\lVert z\rVert^{2}-(v\cdot z)^{2}]\le32A \tag*{(14)}

for odd-center witnesses, either individually or for whole families by the extension bounds in Appendix § 11. The recorded disjoint counts are

Deletion using the seven-dimensional theorem507 096 272
Whole tails from a seven-coordinate center18 246 848 326
Adaptive whole-tail bounds10 229 385
Individual additional centers1 944 853

Their sum agrees with an independent generating-function count; the 790 944 734 extendable seven-coordinate prefixes also agree. No counterexample remains. The exact period evaluation of ϑ(8)\vartheta^{(8)} supplies the matching lower bound, proving Δ8=T8\Delta_{8}=T_{8}.

Dimension nine

Put T9=1087/1980T_9 = 1087/1980. Here a larger low-norm catalogue and fourth-moment constraints make the plane classification finite and manageable. The supplementary archive records the catalogue, the plane certificates and the complete partitioned census used below.

High directions and exceptional planes

The complete census of sorted positive nine-tuples with A≤10000A \le10000 contains 29 044 309 143 tuples. Retaining every tuple not proved at most 51/10051/100 gives 2368 directions and 1365 distinct squared norms. Each retained distance is checked by the clipped rational cell method. Their maximum is T9T_9; within this catalogue its attainers are v(9),2v(9),3v(9)v^{(9)}, 2v^{(9)}, 3v^{(9)}. This local observation is not a global uniqueness assertion.

Proposition 5.1 (Nine-dimensional plane certificate). A full-support rational plane in R9\mathbb{R}^9 either has β(W)≤51/100\beta(W) \le51/100 or admits, after signed coordinate permutations, one of the seven bases in Table 2. In particular, every such plane has β(W)≤31/60\beta(W) \le31/60.

(s,t,u)(s,t,u)DDRows (ai,bi)(a_i,b_i)β(W)≤\beta(W) \le
(10,−5,10)(10,-5,10)75(0,1),(0,1),(1,−2),(1,−1),(1,−1),(1,0),(1,0),(1,1),(2,−1)(0,1),(0,1),(1,-2),(1,-1),(1,-1),(1,0),(1,0),(1,1),(2,-1)31/6031/60
(10,5,10)(10,5,10)75(0,1),(0,1),(1,−1),(1,0),(1,0),(1,1),(1,1),(1,2),(2,1)(0,1),(0,1),(1,-1),(1,0),(1,0),(1,1),(1,1),(1,2),(2,1)31/6031/60
(10,0,45)(10,0,45)450(0,1),(0,4),(1,−3),(1,−2),(1,−1),(1,1),(1,2),(1,3),(2,0)(0,1),(0,4),(1,-3),(1,-2),(1,-1),(1,1),(1,2),(1,3),(2,0)23/4523/45
(16,−8,40)(16,-8,40)576(0,1),(0,4),(1,−3),(1,−2),(1,1),(1,2),(2,−2),(2,−1),(2,0)(0,1),(0,4),(1,-3),(1,-2),(1,1),(1,2),(2,-2),(2,-1),(2,0)37/7237/72
(16,8,40)(16,8,40)576(0,1),(0,4),(1,−2),(1,−1),(1,2),(1,3),(2,0),(2,1),(2,2)(0,1),(0,4),(1,-2),(1,-1),(1,2),(1,3),(2,0),(2,1),(2,2)37/7237/72
(56,−28,56)(56,-28,56)2352(0,3),(0,4),(1,−2),(1,1),(2,−1),(3,−3),(3,0),(4,−4),(4,0)(0,3),(0,4),(1,-2),(1,1),(2,-1),(3,-3),(3,0),(4,-4),(4,0)43/8443/84
(56,28,56)(56,28,56)2352(0,3),(0,4),(1,−1),(1,2),(2,1),(3,0),(3,3),(4,0),(4,4)(0,3),(0,4),(1,-1),(1,2),(2,1),(3,0),(3,3),(4,0),(4,4)43/8443/84

Table 2. Exceptional nine-dimensional plane bases, written as ordered lists of rows. The last column is an upper bound, not an asserted exact offset.

Table 2: Exceptional nine-dimensional plane bases, written as ordered lists of rows. The last column is an upper bound, not an asserted exact offset.

Proof and finite certificate. If β(W)>51/100\beta(W) > 51/100, the seven projected heights satisfy

12∑i≤rhi+20∑i∈M(hi+1−hi)≥4r−28+48(51/100).12\sum_{i \le r} h_i + 20\sum_{i \in M}(h_{i+1}-h_i) \ge4r - 28 + 48(51/100).

With the six KZ chain constraints this gives 59 inequalities. Exact traversal of 3181 vertices, 21 480 directed bounded edges and 787 unbounded edges proves D<18226D < 18226. Thus a reduced Gram matrix has D≤18225D \le18225 and s≤155s \le155. The case s=1s = 1 is excluded using Δ8<51/100\Delta_8 < 51/100.

For the 1 735 940 remaining Gram triples, set q(k)=sk2+2tk+uq(k) = sk^2 + 2tk + u. If q(k)≤10000q(k) \le10000 but its norm is absent from the high catalogue, b+kab + ka must have a zero coordinate. The relevant range is ∣k∣≤71|k| \le71. The maximum number of such zeros is min⁡(8,s)\min(8,s) if ss is absent from the catalogue, and min⁡(9,s)\min(9,s) otherwise. Moreover the residual Gram matrix after subtracting the forced matrices (1−k−kk2)\begin{pmatrix}1&-k\\-k&k^2\end{pmatrix} must be positive semidefinite. These tests leave 713 triples. Let H4(q)H_4(q) be the set of fourth moments of retained vectors of norm qq. The polynomial P(k)=∑i(bi+kai)4P(k) = \sum_i(b_i+ka_i)^4 must have degree at most four, and except at the allowed zero-coordinate positions must belong to H4(q(k))H_4(q(k)). If the zero-coordinate allowance is cc and mm positions have empty H4H_4, then any c−m+5c-m+5 nonempty positions contain five genuine full-support positions. The program checks that enough positions exist, enumerates every such five-point selection and value choice, and interpolates exactly. Integer coefficient, moment and compatibility tests leave 282 polynomials on 261 Gram triples.

Writing P(k)=∑j=04pjkjP(k) = \sum_{j=0}^{4}p_jk^j, complete enumeration of the second and fourth moments of (ai)(a_i) and (∣bi∣)(|b_i|), followed by all pairings and signs, enforces

4∑iaibi3=p1,6∑iai2bi2=p2,4∑iai3bi=p3.4\sum_i a_i b_i^3=p_1,\qquad6\sum_i a_i^2b_i^2=p_2,\qquad4\sum_i a_i^3b_i=p_3.

The Cauchy, convexity and rearrangement bounds used to shorten this search are necessary conditions only. There are 449 row matrices, 119 saturated. For an odd vector zz, put Q=∥z∥2Q=\lVert z\rVert^2, P=a⋅zP=a\cdot z, R=b⋅zR=b\cdot z. Its squared residual is

DQ−uP2+2tPR−sR24D.(15)\frac{DQ-uP^2+2tPR-sR^2}{4D}. \tag*{(15)}

Saved witnesses bound 112 matrices by 51/100; the seven remaining matrices have the bounds in Table 2, using the all-minus-ones vector. Saturation, Gram data and all witnesses are rechecked with arbitrary-precision integers. □\square

Exceptional directions, a norm bound, and the full census

If vv is primitive in an exceptional plane with offset upper bound bWb_W, the one-dimensional fiber inequality gives

d(v)≤bW+D4A.(16)d(v)\le b_W+\frac{D}{4A}. \tag*{(16)}

Thus d(v)>T9d(v)>T_9 forces the bounds

D754505762352A<37125/642970570240/1394074840/257A≤5802969410215855\begin{array}{c|cccc} D & 75 & 450 & 576 & 2352\\ A< & 37125/64 & 2970 & 570240/139 & 4074840/257\\ A\le& 580 & 2969 & 4102 & 15855 \end{array}

Write v=ma+nbv=ma+nb. Positive definiteness gives m2≤Au/Dm^2\le Au/D, n2≤As/Dn^2\le As/D, so a finite integer rectangle suffices. Saturation makes gcd⁡(m,n)=1\gcd(m,n)=1 equivalent to primitivity of vv. Discard zero coordinates and sort the absolute values. The exact check covers 1186 Gram-labelled cases modulo overall sign and 76 136 cells, all at most T9T_9; for D=450D=450 it also includes the safe extra boundary A=2970A=2970.

For a strict counterexample outside these planes, the first projected KZ height satisfies

h1>4(T9−51/100)=386/2475.h_1>4(T_9-51/100)=386/2475.

The eight-height product is 1/A1/A. The prefix constraints are

12∑i≤rhi+20∑i∈M(hi+1−hi)≥48T9+4r−32.(17)12\sum_{i\le r}h_i+20\sum_{i\in M}(h_{i+1}-h_i)\ge48T_9+4r-32. \tag*{(17)}

There are 87 prefix matchings, seven chain constraints and the additional first-height constraint: 95 in total. The exact vertex check has 17 812 vertices, 138 862 directed bounded edges and 3634 unbounded edges. It bounds the reciprocal product strictly below 45960, so A≤45959A\le45959.

The full census at this bound partitions by the first three coordinates into 79 075 disjoint jobs. It covers 21 484 571 760 845 sorted positive nine-tuples and finds none above T9T_9. The disjoint coverage categories are

Seven-coordinate whole intervals 18 999 514 214 339 Eight-coordinate intervals 2 017 779 983 325 Deletion using the eight-dimensional bound 16 453 487 495 Uniform single tails 450 702 840 824 Adaptive tails 104 693 510 Individual directions 16 541 352.

Each job satisfies the sum identity, and independent generating-function counts agree job by job. The census uses the proved two-tail and convex interval inequalities of Appendix § 11; logical coverage does not mean separate numerical minimization for every tuple. Together with the exact lower bound from v(9)v^{(9)}, this proves Δ9=T9\Delta_9=T_9.

Dimension ten

Put T=7/11T=7/11. In this dimension, a bounded direction census is combined with a finite search on short integer relations. The upper bound uses the complete branch coverage specified below, with exact certificates and independent set audits in the supplementary archive.

The base census and the plane bound

The complete census of sorted positive ten-tuples with A≤20000A \le20000 contains 3 519 937 718 862 tuples, split into 19 048 completed jobs, and certifies d(v)≤Td(v) \le T throughout. A separate complete catalogue at A≤15000A \le15000 and threshold 3/53/5 contains 895 707 071 005 tuples; exactly 111 directions on 96 norms exceed 3/53/5. Every retained value has a rational clipped-cell audit. The stronger three-tail bound in Appendix § 12 is part of the census proof.

Proposition 6.1. Every full-support rational plane W⊂R10W \subset\mathbb{R}^{10} satisfies β(W)≤3/5\beta(W) \le3/5.

Proof and finite certificate. Assuming β(W)>3/5\beta(W)>3/5, the eight projected KZ heights and the block inequalities yield D<32806D<32806, by an exact traversal of 19 124 vertices. Thus D≤32805D \le32805 and a reduced Gram matrix satisfies s≤209s \le209. If s=1s=1, delete the coordinate axis and use Δ9<3/5\Delta_9<3/5. If s=2s=2, the short vector has two entries ±1\pm1. At a fixed value of the other parameter, these two coordinates can be adjusted with remaining squared error at most 1/81/8. The other eight coordinates give

β(W)≤Δ8+1/8=81/136<3/5.\beta(W) \le\Delta_8 + 1/8 = 81/136 < 3/5.

For s≥3s \ge3, there is no high-catalogue norm at most 209, so aa must have a zero coordinate. Each kk for which sk2+2tk+u≤15000sk^2+2tk+u \le15000 is not a high norm forces a distinct nonzero coordinate of aa to vanish in b+kab+ka. There can be at most min⁡(s,9)\min(s,9) such kk. All 4 136 388 admissible Gram triples violate this necessary condition, by at least four. A direct enumeration and an independent discriminant-based reverse enumeration agree. This excludes every possible plane.

For a primitive strict counterexample vv, the projected lattice Γ=Pv⊥Z10\Gamma=P_{v^\perp}\mathbb{Z}^{10} has nine heights with product 1/A1/A. The plane bound gives

h1>4(T−3/5)=8/55.(18)h_1 > 4(T-3/5)=8/55. \tag*{(18)}

Additional KZ inputs and short relations

Besides the classical inequalities, the following hold for every valid consecutive index range:

3hi≤4hi+1,2hi≤3hi+2,3h_i \le4h_{i+1}, \qquad2h_i \le3h_{i+2},
25hi+36hi+1≤48hi+2+(40+7/106)hi+3,(19)25h_i + 36h_{i+1} \le48h_{i+2} + (40 + 7/10^{6})h_{i+3}, \tag*{(19)}
5hi≤2hi+3+(8+3/104)hi+4,5h_i \le2h_{i+3} + (8 + 3/10^{4})h_{i+4},
4hi+3hi+2≤4hi+3+(8+1/20000)hi+4.4h_i + 3h_{i+2} \le4h_{i+3} + (8 + 1/20000)h_{i+4}.

The last three are precisely the proved versions of [21 Theorems 6.1–6.2]. Their error-free analogues are not assumed. Nonnegative rational linear combinations of these inequalities and (11) give ρr2≤Crhr\rho_r^2 \le C_rh_r, with

(C1,…,C9)=(14,12,56,2924,505003300000,265003120000,261654239000000,132802613600000,4020502145513855000000000).(20)(C_1,\ldots,C_9) = \left(\frac{1}{4},\frac{1}{2},\frac{5}{6},\frac{29}{24},\frac{505003}{300000},\frac{265003}{120000}, \frac{26165423}{9000000},\frac{13280261}{3600000},\frac{4020502145513}{855000000000}\right). \tag*{(20)}

The saved multipliers are checked by direct rational vector addition; no numerical linear-programming output is accepted without this check.

Reverse the dual basis of Γ\Gamma to obtain a basis a1,…,a9a_1,\ldots,a_9 of Λ=v⊥∩Z10\Lambda= v^\perp\cap\mathbb{Z}^{10}. Successively reduce each new vector to a nearest representative modulo its predecessors. This operation corresponds in the primal basis to adding earlier vectors to later ones, preserving every projected KZ lattice and height. The relation prefix Mk=Za1+⋯+ZakM_k = \mathbb{Z}a_1+\cdots+\mathbb{Z}a_k is primitive in Z10\mathbb{Z}^{10}. Write GkG_k for its Gram matrix and Dk=det⁡GkD_k = \det G_k; set D0=1D_0 = 1. Its squared Gram–Schmidt lengths satisfy

gj=1/h10−j,Dk=∏j≤kgj,A=Dkh1⋯h9−k.(21)g_j = 1/h_{10-j}, \qquad D_k = \prod_{j\le k} g_j, \qquad A = \frac{D_k}{h_1\cdots h_{9-k}}. \tag*{(21)}

Primitivity makes the integral pairing map onto Zk\mathbb{Z}^k. Since ai⋅1≡∥ai∥2(mod2)a_i \cdot1 \equiv\lVert a_i\rVert^2 \pmod2, the minimum squared normal offset of the half-integer centers is

Rk=14min⁡x∈Zkxi≡(Gk)ii(mod2)xTGk−1x.(22)R_k = \frac{1}{4}\min_{\substack{x\in\mathbb{Z}^k\\x_i\equiv(G_k)_{ii}\pmod2}} x^{T}G_k^{-1}x. \tag*{(22)}

The complementary subspace contains vv and is full-support, so Rk≤k/12R_k \le k/12 for k≤7k \le7 and R8≤3/5R_8 \le3/5. More importantly, the fiber bound is

d(v)≤Rk+U9−k,M(h).(23)d(v) \le R_k + U_{9-k,M}(h). \tag*{(23)}

A particular parity-compatible xx gives an upper bound on RkR_k and suffices in this inequality. It does not certify a lower bound on RkR_k. The final proof has no exclusions based on a claim that the characteristic minimum exceeds the subspace mean; all recorded “mean-exclusion” lists are empty.

Since T<d(v)≤C9h9T < d(v) \le C_9h_9, one has ∥a1∥2<C9/T<8\lVert a_1\rVert^2 < C_9/T < 8. Norm one is impossible because vv has full support. If m=∥a1∥2∈{2,…,7}m = \lVert a_1\rVert^2 \in\{2,\ldots,7\}, then

∥a2∥2<C8T−1/(4m)+m4<8.\lVert a_2\rVert^2 < \frac{C_8}{T - 1/(4m)} + \frac{m}{4} < 8.

Thus the first two relations both have squared norm at most seven. For a general prefix, putting r=9−kr = 9-k and using a covering-radius upper bound ρ(Mk)2\rho(M_k)^2 yields

∥ak+1∥2<CrT−Rk+ρ(Mk)2.(24)\lVert a_{k+1}\rVert^2 < \frac{C_r}{T-R_k}+\rho(M_k)^2. \tag*{(24)}

If a rational height certificate also gives hr≥L>0h_r \ge L > 0, then ∥ak+1∥2≤1/L+ρ(Mk)2\lVert a_{k+1}\rVert^2 \le1/L+\rho(M_k)^2. The strict bound uses ⌈b⌉−1\lceil b\rceil- 1 as the integer upper limit; the weak bound uses ⌊b⌋\lfloor b\rfloor.

Complete Gram extension and exact exclusions

Let wi=ai⋅ak+1w_i=a_i\cdot a_{k+1} and q=∥ak+1∥2q=\lVert a_{k+1}\rVert^2. Nearest reduction implies

2∣x⋅w∣≤xTGkx(x∈Zk),∣wi∣≤⌊(Gk)ii/2⌋.(25)2|x\cdot w|\le x^{T}G_kx\quad(x\in\mathbb{Z}^k),\qquad|w_i|\le\left\lfloor(G_k)_{ii}/2\right\rfloor. \tag*{(25)}

Enumerating the whole latter box and every allowed qq is therefore complete; tests with x∈{−1,0,1}kx\in\{-1,0,1\}^k only prune by necessary conditions. Negating the new vector permits the first nonzero entry of ww to be positive. No additional symmetry deletion is needed for completeness. A further necessary dual-KZ test is

xTadj⁡(Gk)x≥Dk−1(0≠x∈Zk).(26)x^{T}\operatorname{adj}(G_k)x\ge D_{k-1}\quad(0\ne x\in\mathbb{Z}^k). \tag*{(26)}

If a violating vector exists, it lies in the finite box

∣xi∣≤⌈Dk−1(Gk)ii−1Dk⌉.|x_i|\le\left\lceil\sqrt{\frac{D_{k-1}(G_k)_{ii}-1}{D_k}}\right\rceil.

This follows from xi2≤(Gk)iixTGk−1xx_i^2\le(G_k)_{ii}x^TG_k^{-1}x and integrality of the adjugate quadratic form. The square-root endpoints are checked by integer comparisons.

Each surviving Gram prefix fixes the last kk primal heights. Equations (19), (18), the mean-fiber bounds and (23) define a rational polyhedron in the remaining rr heights. The following exclusions are used.

  1. A rational Farkas certificate proves infeasibility by a nonnegative combination of the original inequalities. A numerical solver may propose multipliers, but exact arithmetic accepts them.

  1. A product certificate proves A≤20000A\le20000, reducing the branch to the completed base census. All free heights have positive lower bounds and finite upper bounds from the fixed tail, so the product minimum is attained at a vertex. For small rr, every rr-element active set is tested with exact Cramer arithmetic, including degenerate vertices. Earlier layers also use complete box trees and lower chords for logarithms, with rational series remainder bounds. Both children of every split must be covered.

  1. A necessary integral embedding condition fails. A realizable Gram matrix must equal ∑j=110cjcjT\sum_{j=1}^{10}c_jc_j^T for integral columns cjc_j. Full support requires cjTGk−1cj<1c_j^TG_k^{-1}c_j<1 for every jj, and primitivity requires the gcd of all maximal minors to equal one. Complete finite searches with positive-semidefinite residuals certify negative instances. Timeouts remain unresolved, never excluded.

The covering bounds for the relation lattice may use one-dimensional rounding or exact adjacent Gauss-reduced blocks: a planar Gram (pttq)\begin{pmatrix}p&t\\t&q\end{pmatrix} with p≤qp\le q, 2∣t∣≤p2|t|\le p has squared covering radius pq(p+q−2∣t∣)/(4(pq−t2))pq(p+q-2|t|)/(4(pq-t^2)). Taking the minimum over all partitions into singleton and adjacent pair blocks is valid even though the dual basis need not itself be KZ-reduced.

Local embedding obstructions provide an additional inexpensive filter. A primitive kk-row integer matrix has row rank kk over every Fp\mathbb{F}_p, so

rank⁡(Gk mod p)≥2k−10.(27)\operatorname{rank}(G_k\bmod p)\ge2k-10. \tag*{(27)}

If equality holds at an odd prime, the determinant square class of the nonsingular quotient must be (−1)10−k(-1)^{10-k}. Indeed the row space UU then contains U⊥U^\perp as its radical of dimension 10−k10-k. Splitting off that many hyperbolic planes in the ambient determinant-one quadratic space leaves U/U⊥U/U^\perp with the stated determinant class. These are necessary conditions, not sufficient embedding criteria.

The special family and the final branch ledger

The exact product computations through rank five give, for the remaining branches,

A≤⌊449212556⌋=80216.(28)A \le\left\lfloor\frac{4492125}{56} \right\rfloor= 80216. \tag*{(28)}

One rank-six Gram matrix has exactly 24 primitive full-support embeddings in the finite classification. They are all of G2⊕B2G_2 \oplus B_2 type. Signed coordinate permutations and Weyl symmetries reduce their full-support integer directions to

(a,b,a+b,2a+b,3a+b,3a+2b,x,y,x+y,x−y),a,b≥1,x>y≥1,(a,b,a+b,2a+b,3a+b,3a+2b,x,y,x+y,x-y), \qquad a,b \ge1,\quad x>y\ge1,

with norm A=8(3a2+3ab+b2)+3(x2+y2)A=8(3a^2+3ab+b^2)+3(x^2+y^2). All 15 093 137 parameter tuples with A≤80216A\le80216 have an odd-center witness satisfying

11[A∥z∥2−(v⋅z)2]≤28A.11\left[A\lVert z\rVert^2-(v\cdot z)^2\right]\le28A.

The family closes the whole Gram branch because all its embeddings were classified, not just one representative.

The initial complete tables have 63 rank-two Gram matrices and 1943 rank-three candidates. After exclusions, 290 rank-three parents generate 21 839 rank-four candidates. A safe superset of 794 rank-four parents generates 164 085 rank-five candidates. Exact certificate and embedding checks leave 1717 rank-five parents for the final stages.

Rank 6Rank 7Rank 8
Parents1717220361
Candidates passing dual KZ1 126 9813 970 866295 391
Local embedding exclusions2032113 64837 572
Height certificates1 088 6643 811 705257 819
Exact product bounds34 05545 4150
Complete embedding negatives26370
Complete special-family closure100
Remaining for next rank2203610

Table 3. Complete high-rank relation ledger in dimension ten. Each row is a disjoint classification of the candidate set at that rank.

The archived set audits reconcile the actual sets, in addition to these cardinalities. Rank seven uses parts 0, 1, 2, 3, 4, 5, 6; rank eight uses parts 5, 40, 41, 42, 43, 44, 45. Their parent sets are disjoint and have unions equal to the preceding full survivor selections. Each excluded Gram is linked to its exact certificate, and every survivor is passed to the next layer. The final gate and independent set audit both record zero unresolved branches. Any actual counterexample would produce a prefix in every one of these complete tables, contradicting the empty rank-eight remainder. The exact period calculation for v(10)v^{(10)} supplies the matching lower bound, proving Δ10=7/11\Delta_{10}=7/11.

Intersection at the critical radius

For an irrational positive direction, the rational orbit closure from Proposition 2.1 has dimension at least two. The established plane bounds and Lemma 2.2 give

nndim⁡W=2\dim W=2dim⁡W≥3\dim W\ge3Δn\Delta_n
819/4219/425/125/128/178/17
931/6031/601/21/21087/19801087/1980
103/53/57/127/127/117/11

Table 4.

Both subspace bounds are strictly smaller than Δn\Delta_n. Density of the positive orbit therefore gives an actual intersection with the interior of a critical-radius ball. For rational directions the minimum occurs in a period and can be shifted to a positive time. Every orbit tail is dense in the same subtorus, so intersections can also be taken arbitrarily far along the ray. If only positive half-integer centers are retained in the geometric formulation, this last observation ensures that the same critical radii apply. This completes Theorem 1.1 for real rays, including its boundary assertion.

The deficit in general dimension

For a positive integer multiset vv, define

e(v)=n12−d(v),En=inf⁡∣v∣=ne(v).e(v)=\frac{n}{12}-d(v), \qquad E_n=\inf_{|v|=n}e(v).

The averaging bound makes these quantities nonnegative. The results in this section do not depend on any finite-dimensional census.

Normalization of the theta-block invariant

The function BB used in [16] is g/2g/2. Accordingly its extremal quantity max⁡(n)\max(n) is Δn/2\Delta_n/2, and its definition (5.1) gives the exact identity

wt⁡(n)=n/2−12max⁡(n)=6En.(29)\operatorname{wt}(n)=n/2-12\max(n)=6E_n. \tag*{(29)}

Here wt⁡(n)\operatorname{wt}(n) is the extremal quantity as defined there, allowing fractional powers of the eta function; no additional integral-weight rounding is imposed. This is why their logarithmic bounds transfer directly to the sphere problem. Their experimental theta-block tables are useful constructions but are not substitutes for the global upper bounds of § 4–§ 6.

Fourier expansion and a uniform comparison

Put

β2(t)={t}2−{t}+16=g(t)−112,Gv(t)=∑iβ2(vit),\beta_2(t)=\{t\}^2-\{t\}+\frac{1}{6}=g(t)-\frac{1}{12}, \qquad G_v(t)=\sum_i\beta_2(v_i t),

and define Cv(t)=∑icos⁡(2πvit)C_v(t)=\sum_i\cos(2\pi v_i t) and K(v)=−min⁡TCvK(v)=-\min_{\mathbb{T}}C_v. All integrals on T\mathbb{T} use probability Haar measure. Twice integrating by parts gives

β2(t)=1π2∑k≥1cos⁡(2πkt)k2,Gv(t)=1π2∑k≥1Cv(kt)k2.(30)\beta_2(t)=\frac{1}{\pi^2}\sum_{k\geq1}\frac{\cos(2\pi k t)}{k^2}, \qquad G_v(t)=\frac{1}{\pi^2}\sum_{k\geq1}\frac{C_v(k t)}{k^2}. \tag*{(30)}

Both series converge absolutely and uniformly. In particular,

e(v)≤K(v)/6.(31)e(v)\leq K(v)/6. \tag*{(31)}

Writing μ\mu for the Möbius function, inversion also gives

Cv(t)=π2∑k≥1μ(k)k2Gv(kt).(32)C_v(t)=\pi^2\sum_{k\geq1}\frac{\mu(k)}{k^2}G_v(k t). \tag*{(32)}

Substitution of (30) proves this formula: the coefficient of Cv(rt)C_v(rt) is r−2∑k∣rμ(k)r^{-2}\sum_{k\mid r}\mu(k), which vanishes except at r=1r=1. Absolute convergence justifies regrouping the double sum.

Proposition 7.1. For every multiset v∈Z>0nv \in\mathbb{Z}_{>0}^{n}, ∥Cv∥L1(T)≤21e(v)\|C_v\|_{L^1(\mathbb{T})} \le21e(v).

Proof. Write e=e(v)e=e(v) and H=Gv+e≥0H=G_v+e\ge0, so ∫H=e\int H=e. Separate the positive and negative Möbius coefficients:

P(t)=π2∑k≥1μ(k)=1H(kt)k2,N(t)=π2∑k≥1μ(k)=−1H(kt)k2.P(t)=\pi^2\sum_{\substack{k\ge1\\ \mu(k)=1}}\frac{H(kt)}{k^2},\qquad N(t)=\pi^2\sum_{\substack{k\ge1\\ \mu(k)=-1}}\frac{H(kt)}{k^2}.

The Euler products give

∑k≥1μ(k)k2=6π2,∑k≥1∣μ(k)∣k2=ζ(2)ζ(4)=15π2.\sum_{k\ge1}\frac{\mu(k)}{k^2}=\frac{6}{\pi^2},\qquad\sum_{k\ge1}\frac{|\mu(k)|}{k^2}=\frac{\zeta(2)}{\zeta(4)}=\frac{15}{\pi^2}.

Thus Cv=P−N−6eC_v=P-N-6e, while ∫P=21e/2\int P=21e/2 and ∫N=9e/2\int N=9e/2. Multiplication by an integer preserves Haar measure, which justifies these integrals. Since (Cv)+≤P(C_v)_+\le P and ∫Cv=0\int C_v=0, ∥Cv∥1=2∫(Cv)+≤21e\|C_v\|_1=2\int(C_v)_+\le21e.

This convenient comparison supplies all that is needed below. Its role is to transfer an L1L^1 lower bound to the geometric deficit.

The lower bound, including repeated speeds

We use the following form of the McGehee–Pigno–Smith inequality [20 Theorem 2]: there is an absolute b>0b>0 such that for distinct integers q1<⋯<qNq_1<\cdots<q_N and complex coefficients aja_j,

∫T∣∑j=1Naje2πiqjt∣ dt≥b∑j=1N∣aj∣j.(33)\int_{\mathbb{T}}\left|\sum_{j=1}^{N}a_j e^{2\pi i q_jt}\right|\,dt\ge b\sum_{j=1}^{N}\frac{|a_j|}{j}. \tag*{(33)}

Konyagin independently resolved Littlewood’s problem [18]; the weighted formulation (33) is the input used in this proof.

Suppose vv has ss distinct positive speeds, of multiplicities m1,…,msm_1,\ldots,m_s summing to nn. The exponential form of CvC_v has 2s2s distinct frequencies, with coefficients mj/2m_j/2. Consequently

∥Cv∥1≥b2∑j=12s1j≫log⁡(2s+1).\|C_v\|_1\ge\frac{b}{2}\sum_{j=1}^{2s}\frac{1}{j}\gg\log(2s+1).

Moreover every Fourier coefficient is bounded in absolute value by the L1L^1 norm, giving

∥Cv∥1≥12max⁡jmj≥n/(2s).\|C_v\|_1\ge\frac{1}{2}\max_j m_j\ge n/(2s).

If s≥ns\ge\sqrt{n}, the logarithmic bound suffices; otherwise the multiplicity bound is larger than n/2\sqrt{n}/2 and also dominates a fixed positive multiple of log⁡(n+1)\log(n+1). Thus uniformly over all multisets, ∥Cv∥1≫log⁡(n+1)\|C_v\|_1\gg\log(n+1). Proposition 7.1, followed by the infimum over vv, gives En≫log⁡(n+1)E_n\gg\log(n+1).

The upper bound with prescribed total multiplicity

The needed consequence of Belov–Konyagin [2 Theorem 0.5(2) and Corollary 5.4] is that, for every integer n≥3n\ge3, there are nonnegative integers a1≥⋯≥ana_1\ge\cdots\ge a_n, with ∑j=1naj=n\sum_{j=1}^{n}a_j=n, and a real a0≪log⁡3na_0\ll\log^3 n such that

a0+∑j=1najcos⁡(jx)≥0(x∈R).(34)a_0+\sum_{j=1}^{n}a_j\cos(jx)\ge0\qquad(x\in\mathbb{R}). \tag*{(34)}

This is the bound for KZ12(n)K_{\mathbb{Z}}^{\frac{1}{2}}(n), whose coefficient sum is exactly nn; a bound merely along a subsequence of lengths would not suffice. Choose speed jj with multiplicity aja_j, omitting zero multiplicities. This gives a vector of exactly nn coordinates and Cv≥−a0C v \ge-a_0. Equation (30) therefore gives Gv≥−a0/6G_v \ge-a_0/6, so

En≤e(v)≤a0/6≪log⁡3n.E_n \le e(v) \le a_0/6 \ll\log^3 n.

Enlarging the absolute constants handles n=1,2n=1,2 and proves Theorem 1.2. The deep analytic inputs (33) and (34) are invoked as external theorems; their original proofs are not reproduced here.

Consequences and the limits of the bounds

The exact identity

n/12−rn=Enn/12+Δn(35)\sqrt{n/12}-r_n=\frac{E_n}{\sqrt{n/12}+\sqrt{\Delta_n}} \tag*{(35)}

and Δn∼n/12\Delta_n\sim n/12 imply, for sufficiently large nn,

c′log⁡nn≤n/12−rn≤C′log⁡3nn⟶0,rn/n⟶1/12.c'\frac{\log n}{\sqrt{n}}\le\sqrt{n/12}-r_n\le C'\frac{\log^3 n}{\sqrt{n}}\longrightarrow0,\qquad r_n/\sqrt{n}\longrightarrow1/\sqrt{12}.

In particular, En=o(n)E_n=o(\sqrt{n}), which excludes a positive square-root term in the deficit. The exact deficits from the three computations are

E8=10/51,E9=199/990,E10=13/66.E_8=10/51,\qquad E_9=199/990,\qquad E_{10}=13/66.

Their variation is compatible with the general asymptotic bounds and does not by itself determine the growth rate of EnE_n.

There is also a structural limit to the L1L^1 argument. Put A(n)=inf⁡∣v∣=n∥Cv∥1A(n)=\inf_{|v|=n}\|Cv\|_1. The previous proof gives A(n)≫log⁡(n+1)A(n)\gg\log(n+1). Conversely, v=(1,…,n)v=(1,\ldots,n) gives Cv=(Dn−1)/2Cv=(D_n-1)/2, where Dn(t)=sin⁡((2n+1)πt)/sin⁡(πt)D_n(t)=\sin((2n+1)\pi t)/\sin(\pi t) is the Dirichlet kernel. For 0<∣t∣≤1/20<|t|\le1/2, ∣Dn(t)∣≤min⁡(2n+1,1/(2∣t∣))|D_n(t)|\le\min(2n+1,1/(2|t|)), whence

∥Dn∥1≤1+log⁡(2n+1),A(n)≤1+12log⁡(2n+1).\|D_n\|_1\le1+\log(2n+1),\qquad A(n)\le1+\frac{1}{2}\log(2n+1).

Thus A(n)≍log⁡(n+1)A(n)\asymp\log(n+1), and improving only the constant in En≥cA(n)E_n\ge cA(n) cannot yield a higher-order deficit lower bound. Nor does a lower bound for a restricted monotone-coefficient cosine problem transfer backwards through the one-sided inequality (31). The precise order of EnE_n remains beyond the conclusions proved here.

Verification and further questions

The computer-assisted proof combines mathematical reductions with exact finite calculations. This section describes the role of the supplementary archive (which is available at https://zenodo.org/records/24037416) and distinguishes its verification levels; the logical map is given in Appendix § 13.

Exact certificates and verification levels

The archive contains executable sources, exact witnesses, rational certificates, partition records and verification logs. Its main entry point uses Python’s standard library. The archived implementations also use JavaScript, Node.js and checked C# arithmetic. Python integers and rational fractions, JavaScript arbitrary-precision integers and integer cross-multiplication implement the decisive comparisons. In the JavaScript censuses that use ordinary numbers, integer intermediate values are bounded below 2532^{53}. Floating-point calculations may propose an endpoint or a certificate, but acceptance uses an exact check.

The verification levels have distinct mathematical scopes. An integrity check validates the file manifest, the source-archive hash and nested archive checksums. The quick check also recomputes every cell for the three lower-bound directions. The selected replay checks saved ledgers, the nine-dimensional two-tail identities and plane witnesses, and the ten-dimensional gate, short-relation bounds and three-tail identities. The terminal replay additionally recomputes all rank-eight local and height exclusions, using the archived candidate sets. Full eight-dimensional verification regenerates its plane classification, norm bound and direction census. Complete regeneration in dimensions nine and ten additionally requires the large censuses and relation generation procedures described in the archive’s reproduction guide.

These levels should be interpreted through the corresponding proof obligations. File identity establishes which evidence is being checked. Population counts check the sizes of finite search domains. The distance witnesses, exact exclusions and parent–child set reconciliations establish the geometric bounds and the completeness of the finite search. Neither hashes nor cardinalities alone provide those conclusions.

Archived verification records

The archive contains records of a complete eight-dimensional rerun, selected nine-dimensional arithmetic and ledger audits, and ten-dimensional independent candidate-set audits and terminal replays. In particular, its terminal replay records all 37 572 local and 257 819 height exclusions at rank eight. That replay verifies the exclusion arithmetic for the archived candidate population; the separate set-generation audits establish the population to which it applies. The full nine-dimensional census and ten-dimensional base censuses and relation generations have their own archived records.

The present revision aligns the mathematical exposition with those sources and records. The large computations were not regenerated during this revision. The analytic argument in § 7 is independent of the finite censuses. Kahlon’s seven-dimensional theorem and the inequalities of Pendavingh and van Zwam remain external inputs.

Further questions

The finite method raises two further problems: classifying all attaining directions and constructing smaller independent verifiers for the large relation tables. In general dimension, the main analytic question is to determine the order of EnE_n between the logarithmic and cubed-logarithmic bounds. A sharper estimate would need information beyond improving only the constant in the L1L^1 comparison.

AI assistance declaration

OpenAI GPT-6 Astra was used in the determination of the exact view-obstruction constants in dimensions eight to ten. OpenAI Codex assisted with the preparation of the manuscript, including expository editing, literature checks, local exact-arithmetic checks and the comparison of the manuscript with the supplementary archive. The authors are responsible for the mathematical statements, proofs and computational evidence.

Data and code availability

The source code and exact verification data accompany this paper in a separate archive (https://zenodo.org/records/24037416). It includes the verification entry point, reproduction instructions, the original dimensional packages and the later audit records. The mathematical paper is supplied separately as a standalone LaTeX source; the code archive contains no LaTeX source files. Original computational evidence is preserved byte for byte inside the archive.

Single-tail, two-tail and interval certificates

A single appended coordinate

For a positive prefix ww and an odd center zz, write S=∥w∥2S = \lVert w\rVert^2, D=w⋅zD = w \cdot z, Q=∥z∥2Q = \lVert z\rVert^2, N=SQ−D2N = SQ-D^2. Appending a coordinate MM and odd center coordinate qq gives

dist⁡((z,q)/2,R(w,M))2=N4S+S4(S+M2)(q−MDS)2\operatorname{dist}((z,q)/2,\mathbb{R}(w,M))^2 = \frac{N}{4S} + \frac{S}{4(S+M^2)}\left(q-\frac{MD}{S}\right)^2

The nearest odd integer to MD/SMD/S is at distance at most one. Hence

d(w,M)≤N4S+S4(S+M2).(36)d(w,M) \le\frac{N}{4S}+\frac{S}{4(S+M^2)}. \tag*{(36)}

Taking an optimal center for ww yields the deletion bound d(w,M)≤Δn−1+S/[4(S+M2)]d(w,M) \le\Delta_{n-1} + S/[4(S+M^2)]. For n=8n = 8, T=8/17T = 8/17 and Δ7=73/180\Delta_7 = 73/180, a strict counterexample requires 199M2<566S199M^2 < 566S; retaining equality in a census is a harmless enlargement of its candidate set.

Two appended coordinates

Lemma A.1. Let S=∥w∥2>0S = \lVert w\rVert^2 > 0, L=max⁡iwiL = \max_i w_i, and L≤x≤yL \le x \le y. Then

d(w,x,y)≤d(w)+M(S/L2),M(s)=max⁡(14,(s+1)22(s+2)2).d(w,x,y) \le d(w) + M(S/L^2), \qquad M(s) = \max\left(\frac{1}{4},\frac{(s+1)^2}{2(s+2)^2}\right).

An arbitrary odd-center upper bound N/(4S)N/(4S) may replace d(w)d(w).

Proof. The normal prefix residual and the tail fiber are orthogonal. The latter has Gram matrix

1S+x2+y2(S+y2−xy−xyS+x2).\frac{1}{S+x^2+y^2} \begin{pmatrix} S+y^2 & -xy \\ -xy & S+x^2 \end{pmatrix}.

Normalize s=S/x2s = S/x^2, t=y/x≥1t = y/x \ge1. If t≥s+1t \ge s+1, nearest-plane rounding in a suitable order gives

ρ2≤14(s+1s+1+t2+ss+1)≤14(1s+2+ss+1)<14.\rho^2 \le\frac{1}{4}\left(\frac{s+1}{s+1+t^2}+\frac{s}{s+1}\right) \le\frac{1}{4}\left(\frac{1}{s+2}+\frac{s}{s+1}\right) < \frac{1}{4}.

For 1≤t≤s+11 \le t \le s+1, the standard two-triangle tiling gives

ρ2≤f(s,t)=(s+1)(s+t2)(2s+(t−1)2)4s(s+1+t2)2.\rho^2 \le f(s,t) = \frac{(s+1)(s+t^2)(2s+(t-1)^2)}{4s(s+1+t^2)^2}.

Put u=t−1∈[0,s]u=t-1\in[0,s]. If s≥2s\ge\sqrt{2}, the numerator of f(s,1)−f(s,t)f(s,1)-f(s,t) after multiplication by the positive factor 4s(s+1+t2)2(s+2)2/(s+1)4s(s+1+t^2)^2(s+2)^2/(s+1) is

u{4s2(s+2)+(s3+7s2−4)u+(6s2−8)u2+(s2−2s−4)u3}.u\left\{4s^2(s+2)+(s^3+7s^2-4)u+(6s^2-8)u^2+(s^2-2s-4)u^3\right\}.

If the final coefficient is negative, combine its term with the linear term inside the braces, using u2≤s2u^{2} \le s^{2}. The result is at least u(s4−s3+3s2−4)u(s^{4}-s^{3}+3s^{2}-4), which is positive for s≥2s \ge\sqrt{2}: its derivative is s(4s2−3s+6)>0s(4s^{2}-3s+6)>0, and its value at 2\sqrt{2} is 6−226-2\sqrt{2}. The other terms are nonnegative.

If 0<s≤20<s\le\sqrt{2}, the numerator of 1/4−f(s,t)1/4-f(s,t) after multiplication by 4s(s+1+t2)24s(s+1+t^{2})^{2} is

s(2−s2)+4su+(−s2+4s−1)u2+(2s−2)u3−u4.s(2-s^{2})+4su+(-s^{2}+4s-1)u^{2}+(2s-2)u^{3}-u^{4}.

For s≤1s\le1, the last three terms are at least (2s−1)u2(2s-1)u^{2}. This is nonnegative when s≥1/2s\ge1/2; for smaller ss use 4su−u2≥3su≥04su-u^{2}\ge3su\ge0. For 1≤s≤21\le s\le\sqrt{2}, the cubic term is nonnegative and the quadratic and quartic terms are at least (−2s2+4s−1)u2≥0(-2s^{2}+4s-1)u^{2}\ge0. Therefore f(s,t)≤M(s)f(s,t)\le M(s). Finally, MM is nondecreasing and S/x2≤S/L2S/x^{2}\le S/L^{2}.

Exact whole-interval skipping

For the nine-dimensional census, another valid two-tail bound is the successive single-tail bound

N4S+S4(S+L2)+S+L24(S+2L2).\frac{N}{4S}+\frac{S}{4(S+L^{2})}+\frac{S+L^{2}}{4(S+2L^{2})}.

With T=1087/1980T=1087/1980 and K=220K=2^{20}, define

rold=⌊4TK⌋−⌊KSS+L2⌋−⌊K(S+L2)S+2L2⌋,r_{\mathrm{old}}=\left\lfloor4TK\right\rfloor-\left\lfloor\frac{KS}{S+L^{2}}\right\rfloor-\left\lfloor\frac{K(S+L^{2})}{S+2L^{2}}\right\rfloor,
rnew=⌊4TK⌋−max⁡(K,⌈2K(S+L2)(S+2L2)2⌉),r=max⁡(rold,rnew).r_{\mathrm{new}}=\left\lfloor4TK\right\rfloor-\max\left(K,\left\lceil\frac{2K(S+L^{2})}{(S+2L^{2})^{2}}\right\rceil\right),\qquad r=\max(r_{\mathrm{old}},r_{\mathrm{new}}).

Then KN≤rSKN\le rS certifies every allowed two-tail extension. When the last prefix coordinate increases, with preceding norm S0S_{0} fixed, S/L2=1+S0/L2S/L^{2}=1+S_{0}/L^{2} decreases and both tail bounds improve. Freezing the current rr is therefore conservative.

For a fixed odd center (z0,q)(z_{0},q) and variable last prefix coordinate yy, its projection numerator is

N(y)=N0+S0q2−2D0qy+Q0y2.N(y)=N_{0}+S_{0}q^{2}-2D_{0}qy+Q_{0}y^{2}.

The inequality KN(y)≤r(S0+y2)KN(y)\le r(S_{0}+y^{2}) has a convex quadratic left-minus-right side, since KQ0−r>0KQ_{0}-r>0 in the implemented seven-coordinate case (Q0≥6Q_{0}\ge6, r<3Kr<3K). Exact verification at two endpoints proves the entire interval. Eight-coordinate intervals use the same reasoning with the single-tail bound. Every proposed endpoint is checked by integer arithmetic; a floating-point quadratic root is only a proposal. If no witness succeeds, enumeration continues down to the complete period test.

The three-tail covering certificate

The ten-dimensional census uses the following finite polynomial certificate for all real tail parameters, not a numerical grid.

Lemma B.1. Let S>0S>0, 0<L≤x≤y≤z0<L\le x\le y\le z and 0<S/L2≤70<S/L^{2}\le7. For a=(x,y,z)a=(x,y,z) and A=S+x2+y2+z2A=S+x^{2}+y^{2}+z^{2}, the lattice with Gram G=I3−aaT/AG=I_{3}-aa^{T}/A has covering radius satisfying

ρ(G)2≤H(S/L2),H(s)=max⁡(12,3s2+9s+84(s+3)2).\rho(G)^{2}\le H(S/L^{2}),\qquad H(s)=\max\left(\frac{1}{2},\frac{3s^{2}+9s+8}{4(s+3)^{2}}\right).

Proof with explicit polynomial verification. Normalize the parameters by setting s=S/x2s = S/x^2, t=y/xt = y/x and u=z/xu = z/x. Thus the normalized parameters are s,1,t,us,1,t,u. Since s≤S/L2s \le S/L^2 and HH is nondecreasing, it suffices to prove the bound for 0<s≤70 < s \le7, 1≤t≤u1 \le t \le u. In the following calculation write zz for the normalized variable uu.

If z(1+t)>s+1+t2z(1+t) > s+1+t^2, first quotient by the third basis vector and then round in its direction. The squared covering radius is at most

ρ22(s;1,t)+s+1+t24(s+1+t2+z2)\rho_2^2(s;1,t) + \frac{s+1+t^2}{4(s+1+t^2+z^2)}

For t≥s+1t \ge s+1 this is less than 1/4+1/6<1/21/4+1/6 < 1/2. Otherwise use f(s,t)f(s,t) from Lemma A.1 and bound the second term by (1+t)2/[4(s+1+t2+(1+t)2)](1+t)^2/[4(s+1+t^2+(1+t)^2)]. Let C0=s+1+t2C_0=s+1+t^2, R=(1+t)2R=(1+t)^2. Positivity of the difference from 1/21/2 reduces to positivity of

P=2sC02(C0+R)−(s+1)(s+t2)(2s+(t−1)2)(C0+R)−sC02R.P=2sC_0^2(C_0+R)-(s+1)(s+t^2)(2s+(t-1)^2)(C_0+R)-sC_0^2R.

With s=7us=7u, t=1+7uvt=1+7uv, the polynomial P/uP/u has bidegree (6,6)(6,6) on [0,1]2[0,1]^2. All 49 tensor-Bernstein coefficients are positive; their minimum is 140. The exact verifier constructs the polynomial, computes these coefficients and re-expands them to recover the original identity.

In the complementary region z(1+t)≤s+1+t2z(1+t)\le s+1+t^2, the entire domain has the parametrization

t=1+sv,z=t+s(1−v)w1+t,0≤v,w≤1.t=1+sv,\qquad z=t+\frac{s(1-v)w}{1+t},\qquad0\le v,w\le1.

The six coordinate-order path simplices triangulate a fundamental cube. The maximum of their circumradii bounds the lattice covering radius: for barycentric coordinates λj\lambda_j in a simplex of circumcenter qq and radius RπR_\pi,

∑jλj∥p−vj∥G2=Rπ2−∥p−q∥G2≤Rπ2.\sum_j\lambda_j\lVert p-v_j\rVert_G^2=R_\pi^2-\lVert p-q\rVert_G^2\le R_\pi^2.

Reversal of the path gives the same radius, so three orders π=(1,2,3),(1,3,2),(2,1,3)\pi=(1,2,3),(1,3,2),(2,1,3) suffice. Put

H0=1+t,b=(H0,tH0,tH0+s(1−v)w),S0=sH02,A0=S0+∑ibi2,H_0=1+t,\qquad b=(H_0,tH_0,tH_0+s(1-v)w),\qquad S_0=sH_0^2,\qquad A_0=S_0+\sum_i b_i^2,

and in order π\pi set Ei=A0−bπi(2∑j<ibπj+bπi)E_i=A_0-b_{\pi_i}(2\sum_{j<i}b_{\pi_j}+b_{\pi_i}). Solving the circumcenter equations with G−1=I+bbT/S0G^{-1}=I+bb^T/S_0 gives

Rπ2=N4D,N=S0∑iEi2+(∑ibπiEi)2,D=S0A02.R_\pi^2=\frac{N}{4D},\qquad N=S_0\sum_i E_i^2+\left(\sum_i b_{\pi_i}E_i\right)^2,\qquad D=S_0A_0^2.

For 0<s≤50<s\le5, substitute s=5us=5u. The inequality Rπ2≤1/2R_\pi^2\le1/2 is 2D−N≥02D-N\ge0; after removing a factor uu, each of the three polynomials has tridegree (10,10,4)(10,10,4) and all 605 tensor-Bernstein coefficients are nonnegative. For 5≤s≤75\le s\le7, substitute s=5+2us=5+2u. The inequality becomes

(3s2+9s+8)D−(s+3)2N≥0.(3s^2+9s+8)D-(s+3)^2N\ge0.

Each polynomial has tridegree (13,10,4)(13,10,4) and all 770 tensor-Bernstein coefficients are nonnegative. In total the obtuse-region check has 3(605+770)=41253(605+770)=4125 coefficients, with independent monomial re-expansion for every identity.

For clarity, these checks use the exact basis-conversion identity

xk=∑i=kd(ik)(dk)(di)xi(1−x)d−i.x^k=\sum_{i=k}^{d}\frac{\binom{i}{k}}{\binom{d}{k}}\binom{d}{i}x^i(1-x)^{d-i}.

in each variable. All arithmetic is rational, and every Bernstein basis polynomial is nonnegative on the unit interval. Thus the 4125+49=41744125+49=4174 coefficient signs prove the continuum inequalities. ▫

For a sorted seven-coordinate prefix, S/L2≤7S/L^{2} \le7 automatically. The normal-prefix residual is orthogonal to the three-tail fiber, so d(w,x,y,z)≤N/(4S)+H(S/L2)d(w,x,y,z) \le N/(4S) + H(S/L^{2}). With X=L2X = L^{2}, K=220K = 2^{20}, the integer certificate

KN≤S[⌊4TK⌋−max⁡(2K,⌈K(3S2+9SX+8X2)(S+3X)2⌉)]KN \le S\left[\lfloor4TK\rfloor-\max\left(2K,\left\lceil\frac{K(3S^{2}+9SX+8X^{2})}{(S+3X)^{2}}\right\rceil\right)\right]

therefore proves an entire three-coordinate tail safe. This is the envelope required by the dimension-ten census. The polynomial identities and coefficient signs above establish its continuum validity once checked in exact arithmetic.

Verification map for the supplementary archive

The archive follows the same dependency order as the proof. Its eight-, nine- and ten-dimensional packages contain the original sources and exact data. A later ten-dimensional review package contains independent candidate-set generation and exclusion checks. The following map describes the evidence needed at each stage without tying the mathematical argument to individual implementation file names.

Proof componentVerification evidence and acceptance condition
Candidate lower boundsComplete rational half-period cells for the three displayed directions, with clipped minimizers and the values 8/17, 1087/1980 and 7/11.
Dimension eightExact plane witnesses, the height-product certificate giving A≤13699A \le13699, and a complete census of 18 766 118 836 sorted positive directions. The seven-prefix count is 790 944 734; no counterexample remains.
Dimension nineThe high-direction catalogue, fourth-moment and plane certificates, exceptional-plane checks, and the height-product bound A≤45959A \le45959. The final census has 79 075 partitions and covers 21 484 571 760 845 directions, with zero counterexamples.
Dimension-ten baseThe census at A≤20000A \le20000 covers 3 519 937 718 862 directions in 19 048 partitions. The high catalogue at A≤15000A \le15000 and the reverse Gram enumeration support the plane bound.
Covering and norm boundsExact rational combinations of the KZ inequalities, parity-compatible witnesses and the 4174 Bernstein coefficient checks in Appendix § 12.
Relation generationComplete bounded Gram extensions and parent–child set audits. Ranks six, seven and eight contain respectively 1 126 981, 3 970 866 and 295 391 candidate Grams after the stated dual-KZ tests.
Terminal exclusionsEvery candidate is linked to an exact exclusion certificate or to the next-rank survivor set. Rank seven uses parts 0, 1, 2, 3, 4, 5, 6; rank eight uses parts 5, 40, 41, 42, 43, 44, 45. The terminal split is 37 572 local and 257 819 height exclusions, with zero unresolved branches.
Special familyAll 24 primitive full-support embeddings are accounted for, followed by the complete check of 15 093 137 parameter tuples at A≤80216A \le80216.

Table 5.

The executable guide specifies commands, parameters and expected outputs for each component. The independent set audits compare the mathematical sets passed between stages, while the saved logs document the original completed computations. New runs use fresh working directories so that the original evidence remains available for comparison.

References

  1. [1]W. Banaszczyk, New bounds in some transference theorems in the geometry of numbers, Math. Ann. 296 (1993), 625–635. 10.1007/BF01445125.
  2. [2]A. S. Belov and S. V. Konyagin, An estimate of the free term of a non-negative trigonometric polynomial with integer coefficients, Izv. Math. 60 (1996), no. 6, 1123–1182. 10.1070/IM1996v060n06ABEH000095.DOI
  3. [3]J. W. S. Cassels, An Introduction to the Geometry of Numbers, Grundlehren der mathematischen Wissenschaften, vol. 99, Springer-Verlag, Berlin, 1959.DOI
  4. [4]Y.-G. Chen, The view-obstruction problem for 3-dimensional spheres, Acta Math. Sinica (Engl. Ser.) 10 (1994), no. 2, 158–167. 10.1007/BF02580423.DOI
  5. [5]Y.-G. Chen, The view-obstruction problem for 4-dimensional spheres, Amer. J. Math. 116 (1994), no. 6, 1381–1419. 10.2307/2375051.DOI
  6. [6]Y.-G. Chen, View-obstruction problems and Kronecker’s theorem, Proc. Amer. Math. Soc. 123 (1995), no. 11, 3279–3284. 10.1090/S0002-9939-1995-1277102-1.
  7. [7]T. W. Cusick, View-obstruction problems, Aequationes Math. 8 (1972), no. 1–2, 197–198. 10.1007/BF01832753.
  8. [8]T. W. Cusick, View-obstruction problems, Aequationes Math. 9 (1973), 165–170. 10.1007/BF01832623.
  9. [9]T. W. Cusick, View-obstruction problems in n-dimensional geometry, J. Combin. Theory Ser. A 16 (1974), no. 1, 1–11. 10.1016/0097-3165(74)90066-1.DOI
  10. [10]T. W. Cusick and C. Pomerance, View-obstruction problems, III, J. Number Theory 19 (1984), no. 2, 131–139. 10.1016/0022-314X(84)90097-0.
  11. [11]V. C. Dumir and R. J. Hans-Gill, View-obstruction problem for 3-dimensional spheres, Monatsh. Math. 101 (1986), 279–289. 10.1007/BF01559391.DOI
  12. [12]V. C. Dumir, R. J. Hans-Gill and J. B. Wilker, Contributions to a general theory of view-obstruction problems, Canad. J. Math. 45 (1993), no. 3, 517–536. 10.4153/CJM-1993-027-9.DOI
  13. [13]V. C. Dumir, R. J. Hans-Gill and J. B. Wilker, A Markoff-type chain for the view-obstruction problem for spheres in ℝ⁴, Monatsh. Math. 118 (1994), 205–217. 10.1007/BF01301689.DOI
  14. [14]V. C. Dumir, R. J. Hans-Gill and J. B. Wilker, The view-obstruction problem for spheres in ℝ⁵, Monatsh. Math. 122 (1996), 21–34. 10.1007/BF01298453.DOI
  15. [15]V. C. Dumir, R. J. Hans-Gill and J. B. Wilker, Contributions to a general theory of view-obstruction problems, II, J. Number Theory 59 (1996), no. 2, 352–373. 10.1006/jnth.1996.0102.DOI
  16. [16]V. Gritsenko, N.-P. Skoruppa and D. Zagier, Theta blocks, J. Eur. Math. Soc. 28 (2026), no. 1, 113–169. Published online 13 June 2024. 10.4171/JEMS/1471.
  17. [17]V. Kahlon, The view-obstruction problem for spheres in dimensions six and seven, preprint, arXiv:2610.37711v1, 26 October 2026. https://arxiv.org/abs/2610.37711v1. Accompanying data: 10.5281/zenodo.23130200.
  18. [18]S. V. Konyagin, On a problem of Littlewood, Math. USSR-Izv. 18 (1982), no. 2, 205–225; Russian original in Izv. Akad. Nauk SSSR Ser. Mat. 45 (1981), no. 2, 243–265. 10.1070/IM1982v018n02ABEH001386.
  19. [19]J. C. Lagarias, H. W. Lenstra, Jr., and C. P. Schnorr, Korkin–Zolotarev bases and successive minima of a lattice and its reciprocal lattice, Combinatorica 10 (1990), no. 4, 333–348. 10.1007/BF02128669.DOI
  20. [20]O. C. McGehee, L. Pigno and B. Smith, Hardy’s inequality and the L¹ norm of exponential sums, Ann. of Math. (2) 113 (1981), no. 3, 613–618. 10.2307/2007000.DOI
  21. [21]R. A. Pendavingh and S. H. M. van Zwam, New Korkin–Zolotarev inequalities, SIAM J. Optim. 18 (2007), no. 1, 364–378. 10.1137/060658795.DOI
  22. [22]J. M. Wills, Zwei Sätze über inhomogene diophantische Approximation von Irrationalzahlen, Monatsh. Math. 71 (1967), 263–269. 10.1007/BF01298332.
  23. [23]G. M. Ziegler, Lectures on Polytopes, Graduate Texts in Mathematics, vol. 152, Springer-Verlag, New York, 1995. 10.1007/978-1-4613-8431-1.

Paper details

vlnc:W_gh8tmt4e
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

Not recorded for this paper

View statement
Looking for related papers…
Contents