VibeMathedMath problems solved with AI

Optimality of the D4 lattice quantizer in dimension four

Is the root lattice D4D_4 the optimal lattice quantizer in dimension four, that is, does it minimize the normalized second moment G(L)=14 covol(L)3/2∫V(L)∥x∥2 dxG(L)=\frac{1}{4\,\mathrm{covol}(L)^{3/2}}\int_{V(L)}\|x\|^2\,dx of the Voronoi cell V(L)V(L) among all lattices L⊂R4L\subset\mathbb{R}^4? Gersho (1979) proposed the D4D_4 structure; Barnes and Sloane (1983) proved the body-centred cubic lattice optimal in dimension three, and no optimal lattice quantizer had been proved in any higher dimension.

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI-assisted
Method
Computation
Field
Discrete geometry; lattice quantization
Posed by
Allen Gersho, Asymptotically optimal block quantization, IEEE Trans. Inform. Theory 25 (1979), who proposed the D4 structure
Year posed
1979
Years open
47y
Solved
2026-10-26
Model
Undisclosed generative AI tools
Vendor
—
Collaborators
Tianle Liu
Verification
Site-confirmed
Publication
Preprint
Significance
32 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem A: every full-rank lattice L⊂R4L\subset\mathbb{R}^4 has G(L)≥G(D4)=131202G(L)\ge G(D_4)=\frac{13}{120\sqrt2}, with equality if and only if LL is similar to D4D_4. Proof: Voronoi's reduction to three secondary cones, an explicit degree-five moment polynomial on each, six polynomial certificates (symmetrised weighted squares and Schur cubics with nonnegative remainders) and one-variable endgames. Theorem B adds a facet-count lower bound in every dimension d≥2d\ge2, proving the lattice case of Ajjanagadde's Conjectures 2 and 4. Single human author; AI helped construct and verify the certificates.

What the AI did

Acknowledgments (arXiv v2): 'Generative AI tools were used to discuss research ideas, construct and verify the polynomial certificates, assist in checking the correctness of the proofs, and proofread the manuscript. The author is responsible for the correctness and content of the paper.' No model is named.

Verification

No independent mathematician has checked this yet. Checked here on 15 November 2026 against the arXiv v2 TeX source. All six polynomial certificates were re-expanded exactly in rational arithmetic (Python and sympy, with our own parser of the appendix tables, symmetrised over the groups H1,H2,H3H_1,H_2,H_3 of orders 120, 12, 36, whose invariance of DD, NN and tt was confirmed). Every remainder has nonnegative coefficients on all 2002 degree-five monomials, and the zero counts and least positive coefficients (1/10000, 1/60000, 1/192000, 73/30000, 859/7200000, 41/50000) match the paper. The moment polynomials SS, N2N_2, N3N_3 and the determinants agree with a Monte Carlo integral of the Voronoi second moment at a random form in each cone (within 0.1%), M(A)=13/30M(A)=13/30 and G(D4)≈0.07660G(D_4)\approx0.07660 were confirmed numerically, and the closing rational endpoint comparisons were rechecked exactly. Not checked: that Zimmermann's three secondary cones cover all forms, the exact derivation of the moment polynomials, and the equality argument.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion