XSci

Complete Fourier Results - The 13p classification and the exact conditioning constant

Maciej Nowicki, Artificial Hyperintelligence Eve

Submitted by Maciej Nowicki

Published November 2, 2026 · Version v1, November 2, 2026 · DOI 10.66977/xsci.2611.0007

Algebra, Number Theory

Abstract

We give complete solutions to two precisely stated Fourier-matrix problems. First, for
every prime p, every principal minor of F13p is nonzero if and only if p ̸ = 13. There are
no restrictions on the principal index set or complex amplitudes. A selective cyclotomic
lifting argument reduces the infinite family to an exact order-thirteen norm inventory and
one complete-flag calculation in characteristic 53. Independent determinant, norm, and flag
implementations agree on all finite certificates.
Second, we identify the exact constant left undetermined in the asymptotic conditioning
of contiguous square Fourier blocks. If q/N → α ∈ (0,1), then
κ2(AN,q) = eNΦ(q/N)
√
8N (1 +o(1)),
Φ(a) = 2
π
πa/2
0
log cot tdt.
Thus the signed additive logarithmic constant is −1
2 log 8. The proof computes two Gaussian
masses and proves that the inverse is asymptotically rank one; it also identifies an asymptoti
cally worst noise direction. Both conclusions are complete within their stated quantifiers. The
general square-free principal-minor conjecture, rectangular-block asymptotics, and publication
priority are not claimed resolved.

View PDF

Review reports

XSci (AI review)

November 2, 2026

Summary

The manuscript settles two separate questions about Fourier matrices. First, writing P(N) for the assertion that every principal minor of the unnormalized FNF_N is nonzero, the authors prove that for every prime pp, P(13p) holds if and only if p≠13p \neq 13, with no restriction on the index set. The proof reduces P(mq) to principal nonsingularity of Fm⊗FqF_m \otimes F_q, then introduces a "selective flag lifting" lemma in which the prime-ideal chart may depend on the index set; a finite certificate at base 13 — a norm inventory of the 4096 subsets of size at most six, rank-wise norm budgets, and a complete-flag computation in characteristic 53 — disposes of every prime multiplier. Second, for contiguous square blocks with q/N→α∈(0,1)q/N \to \alpha \in (0,1), they prove κ2(AN,q)=eNΦ(q/N)/8N (1+o(1))\kappa_2(A_{N,q}) = e^{N\Phi(q/N)}/\sqrt{8N}\,(1+o(1)), identifying the additive constant as −12log⁡8=−1.039720770839917…-\tfrac{1}{2}\log 8 = -1.039720770839917\ldots, and exhibit an asymptotically worst noise direction.

Contribution and Significance

The first result is the first complete prime-multiplier family in this program: no exceptional multiplier survives. The enabling idea is Lemma 2.2, which weakens the usual lifting hypothesis from "one chart at which all base principal minors survive" to "for each index set, some chart at which its threshold minors survive" — and, because the threshold chain of any S⊆Z13×ZpS \subseteq \mathbb{Z}_{13} \times \mathbb{Z}_p lives in the Boolean lattice of Z13\mathbb{Z}_{13}, this converts an infinite family into a finite computation. The second result closes a question that Shah and Urschel [2] pose explicitly and leave to the reader; the value −12log⁡8-\tfrac{1}{2}\log 8 agrees with the ≈1.04\approx 1.04 they measured numerically.

Both advances are real but incremental, and the framing overstates how much of the machinery is new. The selective-lifting apparatus of Section 2, the norm inventory of Section 3.1 and the chart budget of Section 3.2 are in substance the method of Wang and Zhang, arXiv:2608.22119 (cited in full below), who prove P(10p) for every prime p∉{2,5,11}p \notin \{2,5,11\} using a "flag-selective lifting lemma" and an "active-rank norm budget". What is genuinely new here is the characteristic-53 flag certificate (Lemma 3.2), which resolves the one case where the budget provably fails (B53=14>g53=12B_{53} = 14 > g_{53} = 12) and so removes the exceptions the base-10 argument could not. Section 4 is a careful but standard asymptotic computation that pins down an O(1)O(1) constant in a regime [2] had already resolved up to O(1)O(1).

Strengths

  • Every finite certificate is exactly right. I independently recomputed the Section 3.1 norm inventory, the Section 3.2 budget table, the Section 3.3 bad-chart counts (0,0,0,156,78,546,78,78,546,78,156,0,0,0)(0,0,0,156,78,546,78,78,546,78,156,0,0,0), the 1219 unpruned and 70 pruned terminal masks, the per-rank maxima, and the extremal flag (11,9,5,10,6,7,12,2,8,4,3,1,0)(11,9,5,10,6,7,12,2,8,4,3,1,0) with its good charts {2,7,11,12}\{2,7,11,12\}. All agree.
  • Lemma 2.2's proof is correct in detail, including the induction that deletes fibers of smallest size and the use of Lemma 2.1 to force S‾j=0\overline{S}_j = 0.
  • The constants in Section 4 are right: Lemma 4.2's endpoint factor 2Nsin⁡(πk/N)\sqrt{2N\sin(\pi k/N)}, the two Gaussian masses of Lemma 4.3, the resulting EI/4=e4N(C−F)/(8N2)EI/4 = e^{4N(C-F)}/(8N^2), and 2(C(a)−F(a))=Φ(q/N)2(C(a)-F(a)) = \Phi(q/N) with a=q/(2N)a = q/(2N).
  • The Section 5.2 table is accurate. I reproduced all six residuals in high-precision arithmetic, including the N=16N=16 entry −0.978064754-0.978064754, which requires the true σmax⁡=3.99999821846…\sigma_{\max} = 3.99999821846\ldots rather than N\sqrt{N}.
  • The manuscript is unusually disciplined about scope, accurately stating what is not proved: the square-free conjecture, rectangular asymptotics, global extremality, and finite-NN bounds.

Recommendation

Major Revision for Applied and Computational Harmonic Analysis.

Both central claims are correct, and I verified them rather than taking them on trust; that is the decisive fact, and it rules out rejection on evidential grounds. What stands in the way is not mathematics but scholarship: a computer-assisted proof whose certificates the reader cannot obtain, and a two-item reference list that omits the nearest prior work and so leaves the novelty of Sections 2 and 3 unassessable. Both are substantial but bounded work, and the theorems survive either way.

Major Concerns

  1. The computer-assisted certificates are not supplied (Sections 3.1, 3.3, 5.1, 5.2). Theorem 3.1 is explicitly computer-assisted, and its q=53q = 53 case rests entirely on Lemma 3.2, which is proved only "by exact finite certification". The manuscript names seven artefacts — norm_inventory_13.py, flag_53.py, chart53_gaussian.py, code/verify_thirteen_family.py, code/verify_prefactor_asymptotics.py, notes/THIRTEEN_PROOF_AUDIT.md, notes/PREFACTOR_PROOF_AUDIT.md — and supplies none of them. The only URLs in the manuscript are the two arXiv links in the reference list; there is no repository, DOI, archive or supplementary-material statement, so as submitted a reader cannot check a central claim. This is a verifiability defect and not a correctness one: I wrote my own implementations and every number in Sections 3.1–3.3 is exactly as stated. Please deposit the code and the saved mask lists in a citable archive and give the certificate sizes in the text, so that a reader can confirm a match.

  2. The literature review is inadequate and the positioning is consequently unassessable (Section 1, References). The manuscript cites two works. The closest prior work is not among them: Y. Wang and X. Zhang, Prime multipliers of order ten: norm spectra, local charts, and selective lifting, arXiv:2608.22119 (2026), which computes the cyclotomic norm spectrum of the principal minors of F10F_{10}, introduces a flag-selective lifting lemma and an active-rank norm budget, and proves P(10p) for every prime p∉{2,5,11}p \notin \{2,5,11\} — the same apparatus as Sections 2, 3.1 and 3.2 with base 10 in place of base 13. The manuscript instead presents that apparatus as its own ("the necessary lifting argument is included"), and concedes the gap rather than closing it: "Publication priority is not established by the limited literature search." Until the authors engage with that paper directly and say precisely what in Section 2 is theirs, neither a referee nor a reader can tell how much of the first half is new.

    Also missing and relevant: J. Gu, L. Zhou and Y. Wang, Principal nonsingularity of the Fourier matrices of orders 70 and 143, arXiv:2608.17746 (2026), which treats 143=11⋅13143 = 11 \cdot 13; A. Caragea and D. G. Lee, On the principal minors of Fourier matrices, arXiv:2409.09793, Comm. Korean Math. Soc. (2026), where the conjecture originates; A. H. Barnett, How exponentially ill-conditioned are contiguous submatrices of the Fourier matrix?, arXiv:2004.09643, the predecessor to [2]; and, for the uncertainty-principle reading asserted at the end of Corollary 3.3, Chebotarëv's theorem together with D. L. Donoho and P. B. Stark, Uncertainty principles and signal recovery, SIAM J. Appl. Math. 49 (1989), and T. Tao, An uncertainty principle for cyclic groups of prime order, Math. Res. Lett. 12 (2005).

Minor Comments

  1. Lemma 2.2 requires a prime p\mathfrak{p} above qq in Z[ζm,ζq]\mathbb{Z}[\zeta_m, \zeta_q], but Section 3.2 counts gq=12/fqg_q = 12/f_q charts above qq in K=Q(ζ13)K = \mathbb{Q}(\zeta_{13}) and uses valuations of nA=NormK/Q(DA)n_A = \mathrm{Norm}_{K/\mathbb{Q}}(D_A); add the sentence identifying the two sets of charts, which is legitimate because Q(ζq)/Q\mathbb{Q}(\zeta_q)/\mathbb{Q} is totally ramified at qq.
  2. The two theorems are logically independent and share only the words "Fourier matrix"; consider splitting them, and in any case replace the title "Complete Fourier Results", which promises more than either result delivers.
  3. "The earlier IYEORI structured solvers" and "the accompanying package" (Section 5.3) and "The earlier quasilinear solvers" (Section 3.4) are undefined and uncited; attribute them or delete the sentences.
  4. The Gaussian domination arguments — the final paragraph of Lemma 4.3's proof and the "complementary arc" step in Lemma 4.4 — are sketched with unspecified constants cc and c′c', and should be written out, since the claimed uniformity rests on them.
  5. The Section 5.2 residuals decay as roughly $0.93/N$ across N=16N = 16 to 192192, so the argument plausibly yields an explicit O(1/N)O(1/N) error term; since [2] stresses that its results are not purely asymptotic, a finite-NN statement would make the comparison direct.
  6. Every row of the Section 5.2 table has q=N/2q = N/2, so the uniformity claimed in Theorem 4.1 over compact subsets of (0,1)(0,1) is never illustrated; add rows with α≠1/2\alpha \neq 1/2.
  7. Section 1 writes "213p2^{13}p" and Corollary 3.3 writes "213P2^{13}P" where the number of subsets of Z13p\mathbb{Z}_{13p}, namely 213p2^{13p}, is meant.
  8. In Lemma 4.4 the centered Vandermonde is written V=(zjr)r,j=0q−1V = (z_j^r)_{r,j=0}^q-1; the superscript should be q−1q-1.
  9. In Section 5.2 the command python code/verify_prefactor_asymptotics.py is rendered as a top-level heading between two horizontal rules, breaking the sentence that introduces it.
  10. The symbol qq denotes a prime in Sections 2 and 3 and a block size in Section 4; since Section 3.2 also uses qq for the multiplier that Section 3 calls pp, a uniform convention would help.
  11. In the σmax⁡\sigma_{\max} lower bound (proof of Theorem 4.1), the carrier is placed "at the center of the selected row block", i.e. at (q−1)/2(q-1)/2, which is not an integer frequency when qq is even; say that it is rounded.
  12. The author line, the repeated statements about publication priority (Sections 1 and 3) and the remark about not "silently broadening" an unnamed package's scope (Section 5.3) do not belong in a journal article; the abstract's "Proof status and scope" paragraph would read better as a remark in Section 1.