VibeMathedMath problems solved with AI

The Ideal Generation Conjecture for general points in projective space

Let ZZ be a general set of pp distinct points in Pn−1\mathbb{P}^{n-1} over an algebraically closed field kk, with homogeneous vanishing ideal IZ⊂S=k[x1,…,xn]I_Z \subset S = k[x_1,\ldots,x_n], and let DD be the initial degree of IZI_Z. The Hilbert function of S/IZS/I_Z is known, and IZI_Z is generated in degrees DD and D+1D+1. The Ideal Generation Conjecture (Geramita and Maroscia, 1984) predicts the number of minimal generators: the multiplication map μZ:(IZ)D⊗kS1→(IZ)D+1\mu_Z : (I_Z)_D \otimes_k S_1 \to (I_Z)_{D+1} has maximal rank, that is, it is injective or surjective. Before this work the conjecture was known in projective dimension at most four and for sufficiently many points. Does it hold for every number of points in every dimension?

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI co-developed
Method
Argument
Field
Commutative algebra; ideals of general points
Posed by
A. V. Geramita and P. Maroscia (1984), as attributed by Gesmundo, Kayser and Telen (2025)
Year posed
1984
Years open
42y
Solved
2026-11-11
Model
OpenAI Astra agents (GPT-6, Codex)
Vendor
OpenAI
Collaborators
Vlado Vojdanovski
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
32 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 1.1: let kk be algebraically closed of characteristic zero, n≥2n \ge 2, D≥2D \ge 2 and Nj(n)=(n+j−1j)N_j(n) = \binom{n+j-1}{j}. For every pp with ND−1(n)≤p<ND(n)N_{D-1}(n) \le p < N_D(n), a general reduced set ZZ of pp points in Pn−1\mathbb{P}^{n-1} has maximal-rank multiplication (IZ)D⊗kS1→(IZ)D+1(I_Z)_D \otimes_k S_1 \to (I_Z)_{D+1}. So with r=ND(n)−pr = N_D(n) - p the ideal has rr minimal generators in degree DD, max⁡{0, (n−1)(ND(n)D+1−r)}\max\{0,\ (n-1)(\tfrac{N_D(n)}{D+1} - r)\} in degree D+1D+1, and none in higher degrees. The proof treats the statement as interpolation for ΩPn−1(D+1)\Omega_{\mathbb{P}^{n-1}}(D+1) (vectorial Horace method), inducts on the degree for D≥4D \ge 4, and rests on explicit quadratic and cubic constructions with finite determinant calculations. Positive characteristic is not covered, and nothing is claimed about the higher syzygies of the Minimal Resolution Conjecture.

What the AI did

The Hexagon provenance statement reads: this work was produced as a guided exploration by Open AI Astra agents. Hexagon labels the paper as primarily AI-generated text with human understanding of some parts; the PDF itself carries no AI statement. The Lean repository's provenance file names Vlado Vojdanovski as author and responsible maintainer and says OpenAI Codex agents (GPT-6) assisted with the Lean development, the exact finite certificates and the internal reviews; its metadata records the review status as agent-reviewed.

Verification

No independent mathematician has checked this yet. Lean-checked on this site's ladder, not Lean-verified. Checked here on 19 November 2026: Hexagon page 2611.00274v1 with its provenance label, the paper's introduction and Theorem 1.1, and the conjecture as stated by Gesmundo, Kayser and Telen (arXiv 2307.02560, Conjecture 3). Palomar record PALOMAR-2026-11-10-000007 v1 was read: source Vlado-V/ideal-generation-projective-points at commit 9782452, Lean v4.35.0-rc3, Comparator theorem IdealGenerationPaper.main_result, the three standard axioms only, NanoDa kernel replay by Palomar, trust level high. Challenge.lean (169 lines, Mathlib imports only) was read in full. For projective dimension n≥1n \ge 1, degree d≥2d \ge 2 and (n+d−1d−1)≤p<(n+dd)\binom{n+d-1}{d-1} \le p < \binom{n+d}{d}, it asserts a polynomial Δ\Delta on ordered affine-chart configurations, nonzero somewhere, such that wherever Δ≠0\Delta \ne 0 the points are distinct, the ideal is radical, the Hilbert function is min⁡(N,p)\min(N, p), the map Id⊗S1→Id+1I_d \otimes S_1 \to I_{d+1} is injective or surjective, and the ideal has exactly the predicted minimal generators in degrees dd and d+1d+1. That is Theorem 1.1 stated faithfully, with general meaning a nonempty Zariski open set. A grep of the 613 Lean files (about 172,000 lines) found no sorry outside the challenge stub and no axiom, native_decide, implemented_by, extern or unsafe. Not rebuilt here. The statement was written by the author's agents and is not independently anchored.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion