WorksMathematics
Hadamard configurations and sharp cubic-period bounds for stealth in dimensions 10, 11, and 12
Abstract
We construct explicit, nonlattice periodic point configurations in dimensions , and with stealth center density and unit-density stealth radius . These values exceed those of duals of the current record lattice packings, without claiming to exceed all possible lattices. The construction uses the sign rows of an order- Hadamard matrix and their antipodes, projected onto coordinates. For an arbitrary, not necessarily binary, equally weighted motif with marked period , we prove a sharp -point lower bound for a raw stealth radius at least one and classify equality by antipodal Hadamard configurations. We deduce bounded-complexity optimality, sharp optimality among binary motifs in dimensions –, and local maximality under simultaneous perturbations of the motif and the marked period shape. A divisibility argument excludes ordinary equally weighted periodic formal duals. Statements are confined to their indicated classes; no global optimality among arbitrary-period configurations is asserted.
Introduction and main results
For a periodic point set, stealthiness is the radius of the largest open ball about the origin free of nonforward diffraction peaks. Cohen, Cohn, and Viazovska [1] recently established sharp global theorems in dimensions 8 and 24 and constructed a dimension-20 nonlattice point set outperforming all known lattices. The problem of whether the globally optimal stealth configuration must be a lattice remains open in many dimensions. The distinction between a periodic set and a single lattice is essential: a basis form factor can cancel diffraction shells.
Our constructions arise from classical Hadamard orthogonal arrays, but their interpretation as stealth configurations and the accompanying marked-cubic-period rigidity statements are the focus here. Let denote the volume of the Euclidean unit ball. For a periodic set of point density and raw stealth radius , define the scale-invariant stealth center density
In the normalization of unit particle density this gives . On lattices the quantity is the center density of the dual lattice; for general periodic configurations it need not be a sphere-packing density.
Theorem 1.1 (Explicit examples). For each there exists a periodic point configuration with 24 representatives modulo such that
The configurations are not translates of lattices. At unit particle density their stealth radii are , exceeding the stealth radii of duals of the current record lattice packings in those dimensions.
Theorem 1.2 (Sharp cubic-torus cardinality and equality). Let and let , with distinct equally weighted representatives . If , then . If , then, after a common translation, consists of the binary sign rows associated with and for a Hadamard matrix of order . Conversely, every such antipodal Hadamard array achieves .
Theorem 1.3 (Local maximality in a marked period model). Let and let be a strength-three binary orthogonal array of distinct rows with . Then is locally maximal for against simultaneous sufficiently small changes in the positions of its labelled representatives and in its marked period matrix. Every sufficiently small nonsimilarity change of period shape strictly decreases . For and , the equality case is locally isolated modulo translations, similarities, and point relabelling.
The marked period is not necessarily primitive: for our arrays the extra translation is a period, and the primitive cell has 12, rather than 24, representatives. Neither Theorem 1.2 nor Theorem 1.3 is a global theorem over arbitrary period shapes or arbitrary supercells.
Diffraction and its normalization
Fix the Fourier kernel . If is a full-rank lattice and has distinct representatives modulo , then the support of its diffraction away from the forward peak consists precisely of
The diffraction weights are nonnegative multiples of . Thus
For a cubic period , put
Then , and . When , the amplitude depends only on mod .
More generally, for and the density is . Set
The matrix has determinant one, and direct substitution into (1) gives
These statements do not require to be the primitive period lattice.
Orthogonal arrays and the Hadamard construction
For a binary motif , let
Let denote the minimum Hamming weight of a nonzero character for which , with the convention if none exists. A binary orthogonal array of strength is a set with all -coordinate projections uniform; equivalently, the character sums of weights vanish.
Proposition 3.1 (Exact binary Fourier gap). For a nonempty distinct binary motif ,
Proof. The shortest integer vector in a nonzero parity class has squared length equal to its Hamming weight. A nonzero vector in the zero parity class has squared length at least four; the vectors realize this and have amplitude . Formula (2) and (1) conclude the proof. □
Let be a Hadamard matrix of order 12, so has entries and . Convert the rows of the matrix to binary by mapping a sign to . Let denote any projection onto columns, .
Proof of Theorem 1.1. Antipodal pairing kills the character sum for every odd number of distinct columns. Hadamard column orthogonality kills the weight-two sums. Thus is a strength-three orthogonal array. Distinct nonantipodal sign rows of are at Hamming distance 6 (the antipodal pairs have distance 12); deletion of at most two columns cannot identify them. Therefore and Proposition 3.1 yields and .
If were a translate of a lattice, translate one point to the origin. The resulting lattice would contain and lie in , so its quotient by would be a subgroup of of order 24. Every subgroup of has power-of-two order, a contradiction. The numerical packing comparisons are recorded in § 7.
An explicit integral certificate
Let be the quadratic character of , with . Index a matrix by and define , , , and . Then , , and satisfies . This is the classical Paley construction [4]. The accompanying script verifies these identities, all projections, and all character cancellations by exact integer arithmetic.
The sharp cardinality theorem
Proof of Theorem 1.2. Fix and take the integer-frequency indices
Every nonzero difference of two such indices has squared Euclidean norm at most three. Since , the evaluation columns
are mutually orthogonal and have norm . Hence .
If , these columns form a scaled unitary matrix, and the rows are orthogonal. Take distinct and set , so . Factoring the row inner product using the pivot gives
Put . Each is either or . If all are , then is the all-ones binary vector modulo 2. Otherwise let be the number of indices with ; their common value has modulus one. Counting in gives
For this forces , excluded by assumption. Hence is real and ; because , it equals 1, and therefore . Thus every difference is binary modulo 2 and has Hamming weight or .
Translate a selected representative to zero; all representatives then lie in modulo 2. Their sign vectors have pairwise inner products either 0 or . A sign line contains at most two distinct vectors and there are at most mutually orthogonal lines in . Therefore there are exactly lines, each represented with both signs. Choosing a representative from each line yields an order- Hadamard matrix. The converse follows from Proposition 3.1. □
Lemma 4.1 (A stronger rank obstruction). If has , then .
Proof. Use the frequencies for . Their nonzero pairwise differences have squared norms at most four. Under the strict inequality all resulting evaluation columns are mutually orthogonal. There are such columns. □
Two exact optimality consequences
Theorem 5.1 (Bounded-complexity cubic optimality). Let and let have distinct representatives at arbitrary real positions. Then
Equality is possible only for , when Theorem 1.2 classifies up to common translation as an antipodal Hadamard array.
Proof. By Lemma 4.1, . Let . For , trivially . For , orthogonality of the columns for yields . For , the characters give . For , Theorem 1.2 gives . Consequently,
For the first three terms are strictly smaller than (the third follows by checking and observing that decreases). Equality therefore requires and , as asserted. □
Corollary 5.2 (The dimension-12 cubic-cell threshold). Among all equally weighted periodic configurations with marked period and fewer than representatives in that cell, the largest possible stealth center density is . Equality holds precisely for translated 24-point antipodal Hadamard configurations.
Theorem 5.3 (All binary motifs in dimensions 10–12). For ,
Equality holds exactly for the binary orthogonal arrays of strength three with 24 distinct rows.
Proof. Proposition 3.1 reduces the minimum surviving squared norm to . In the cases the first three bounds in (5) are strictly below for . If , all binary character sums of weights one, two, and three vanish, so is a strength-three orthogonal array. Theorem 1.2 gives , and uniformity on any three coordinates forces . Thus . Proposition 3.1 yields the upper bound and equality characterization, and our projected Paley arrays attain it.
Local optimality of the marked configuration
Put
Lemma 6.1 (Strict reciprocal-Gram maximum). For , the function on positive-definite determinant-one symmetric matrices has a strict local maximum at , with .
Proof. Let be any nonzero trace-free real symmetric matrix. If some , then . If no diagonal entry is negative, all diagonal entries vanish. Choose a four-element set containing a nonzero off-diagonal entry. Over the 16 choices of signs, the quadratic forms have mean zero and are not all zero because the degree-two sign characters are linearly independent. Hence at least one is strictly negative. By compactness on the unit sphere of trace-free symmetric matrices, there is with
Write with and put . The expansion yields , and all vectors in have squared norm four. Thus
for sufficiently small nonzero . This is a strict local maximum.
Proof of Theorem 1.3. On any set of four distinct coordinates, a strength-three array has vanishing binary Fourier coefficients of weights one through three. If its weight-four coefficient also vanished, Fourier inversion would make that four-coordinate marginal uniform, hence , contrary to . Therefore every four-coordinate coefficient is nonzero. Every has amplitude , and changing the signs of the four nonzero coordinates leaves the relevant binary residue class unchanged. Thus for all .
This finite set of amplitudes remains nonzero under all sufficiently small perturbations of the motif. The formula (3) then implies
Lemma 6.1 makes the right-hand side strictly less than for a small nonsimilarity deformation of , while at , the value is . The condition is equivalent to being a similarity.
For and , after removing this similarity, equality forces the radius to equal one. By Theorem 1.2, the motif is a common translate of an antipodal binary Hadamard array. There are only finitely many labelled binary arrays of the specified size; in a sufficiently small neighborhood only the original remains after a chosen representative is aligned. Thus equality is locally isolated modulo the stated symmetries. □
Remark 6.2. The proof does not cover arbitrary supercell perturbations or global optimization of period shape. In dimensions 10 and 11 the theorem asserts local maximality but not isolated equality in motif coordinates.
Formal duality and numerical comparisons
Proposition 7.1 (Absence of an ordinary periodic formal dual). None of the configurations , , has an ordinary periodic formal dual among simple equally weighted periodic configurations in the sense of [3].
Proof. Suppose were formally dual to . Formal duality converts the normalized pair correlation of into the normalized Fourier transform of that of . The latter is supported on , so the former is supported on . Since each difference of two points of a simple periodic configuration contributes positively to its pair measure, after translation . For each , the Fourier-side pair coefficient equals one, since . On the side this coefficient is the fraction of points translated to another point by ; equality to one forces to be a period. Hence is a union of an integer number of cosets of and its particle density is an integer. Formal duality requires , hence , which is not an integer in the three stated dimensions. Contradiction. □
The following recorded lattice center densities are those of the duals of the best known lattices in the respective dimensions [2]:
| Lattice center density | Ratio | Unit-density | ||
| 10 | ||||
| 11 | ||||
| 12 |
Table 1.
Cohn’s table lists overall known packing center densities , , and in dimensions 10, 11, and 12, respectively. These are comparisons of the dimensionless normalized quantities, not constructions of better sphere packings. The optimal packing density among all lattices in these dimensions is not known, so this paper establishes no separation from every lattice.
Reproducibility, provenance, and scope
The accompanying standard-library Python script constructs the Paley order-12 matrix, verifies exactly, checks the 24 distinct projected rows and the weight-1,2,3 character cancellations, and enumerates weight-four values. It is a finite-instance verification of the construction, not a machine proof of Theorems 1.2–1.3. The statements of the rank and local results are given with full mathematical proofs above. The verification script verify_hadamard.py and its output verification.txt are available among the arXiv submission’s ancillary files.
The underlying Hadamard arrays are classical and should not be claimed as a new construction. Establishing publication priority for the stealth application, the arbitrary-position equality classification, and local marked-period rigidity requires specialist comparison with existing cubature, design, and hyperuniformity literature. This manuscript is an AI-assisted draft (GPT 6-Astra) requiring final author approval.
In dimension two, a recent AI-generated manuscript released by OpenAI [7] claims that the triangular lattice is universally optimal among planar configurations of density one. That result concerns energy minimization rather than stealth, and we do not use it here. Whether it yields a corresponding statement that a lattice is stealth-optimal in the plane, and whether such claims can be verified with AI-assisted formal proof, remain open.
Acknowledgments
The author thanks xAI for providing a SuperGrok subscription at no cost, and the Turing Fellowship for providing a Claude Max subscription for free.
References
- [1]A. Cohen, H. Cohn, and M. Viazovska, The stealthiest particle configurations in 8 and 24 dimensions, arXiv:2609.03121 (2026). https://arxiv.org/abs/2609.03121.
- [2]H. Cohn, Sphere packing: tables of best known densities and bounds, https://cohn.mit.edu/sphere-packing/, accessed November 2026.
- [3]H. Cohn, A. Kumar, C. Reier, and A. Schürmann, Formal duality and generalizations of the Poisson summation formula, Contemporary Mathematics 625 (2014), 123–140. https://arxiv.org/abs/1306.6796.
- [4]R. E. A. C. Paley, On orthogonal matrices, Journal of Mathematics and Physics 12 (1933), 311–320.DOI
- [7]OpenAI, Universal optimality of the triangular lattice, manuscript dated 23 September 2026, released 6 October 2026. https://github.com/openai/math.