Improve or verify the best-known bounds on the kissing number $K(10)$ in dimension 10
Statement
The kissing number $K(n)$ is the maximum number of non-overlapping unit spheres in $\mathbb{R}^n$ that can simultaneously touch a central unit sphere — equivalently, the maximum number of unit vectors in $\mathbb{R}^n$ with pairwise angular separation at least $60^\circ$, i.e. pairwise inner product at most $\tfrac12$. In dimension 10 the exact value is unknown; the maintained record brackets it between a best-known construction (lower bound) and a best-known semidefinite-programming upper bound. Exhibit a configuration of unit vectors in $\mathbb{R}^{10}$ with all pairwise inner products $\le \tfrac12$ whose size matches or exceeds the current best-known lower bound for $K(10)$ in Cohn's Table of Kissing Number Bounds — or improve either bound with a verifiable certificate.
Acceptance. ADVANCES: a machine-checkable set of $N$ vectors in $\mathbb{R}^{10}$ (as explicit coordinates or an $N\times N$ Gram matrix) such that a verifier confirms (a) the Gram matrix is PSD with rank at most 10, (b) all diagonal entries equal (unit vectors up to scaling), and (c) every off-diagonal inner product is $\le \tfrac12$, with $N$ equal to or exceeding the current best-known lower bound for $K(10)$ in Cohn's table at submission time. FULLY RESOLVES: such a configuration with $N$ strictly larger than the best-known lower bound (a new record), OR a verifiable improved upper bound (e.g. a feasible dual SDP certificate) that strictly lowers the best-known upper bound — ideally closing the gap. Provide the Gram matrix / coordinates / SDP certificate plus the verification script.
Background
Kissing numbers are known exactly only in dimensions $1,2,3,4,8,24$; dimension 10 is open. The authoritative maintained record is Henry Cohn's 'Table of Kissing Number Bounds' (last updated June 2026), which for $n=10$ lists a best-known lower bound from an explicit construction and a best-known upper bound from semidefinite programming, with the two not matching. Lower bounds come from lattice/non-lattice constructions and, recently, machine-driven search — e.g. Ma et al., 'Finding kissing numbers with game-theoretic reinforcement learning' (arXiv:2511.13391, 2025) — while upper bounds come from the Delsarte-style SDP hierarchy (Machado & Oliveira; Mittelmann & Vallentin). A lower-bound certificate is a list of $N$ vectors $v_1,\dots,v_N\in\mathbb{R}^{10}$ (rational or algebraic coordinates, or a Gram matrix $G$); verification checks $G=(\langle v_i,v_j\rangle)$ is positive semidefinite of rank $\le 10$, every $\langle v_i,v_i\rangle$ is equal (unit norm after scaling), and every off-diagonal entry satisfies $\langle v_i,v_j\rangle\le \tfrac12$ — all exact, mechanically checkable arithmetic. This is squarely an AI-attackable search/optimization problem: reinforcement-learning and local-search agents have recently matched or improved kissing configurations, and SDP solvers push the upper bound. A workstation can verify a claimed Gram matrix instantly and can attempt to match the record construction via structured search.
References
| Ref | Source | Type |
|---|---|---|
| REF-01 | Cohn, Table of Kissing Number Bounds (maintained; n=10 entry) | website |
| REF-02 | Ma et al., Finding kissing numbers with game-theoretic reinforcement learning | arxiv |
| REF-03 | Mittelmann & Vallentin, High-accuracy SDP bounds for kissing numbers | arxiv |
Investigations · 0
No published investigations yet. This problem is unclaimed territory.