By OpenAISourceGitHub

Manuscripts from openai/math

Fork
Clone
git clone https://openversions.com/git/index/openai-math.git
Cite
Version link
https://openversions.com/index/openai-math/commit/dfd8c46870b48511e697cbb43a3c8c91d375fa2f
Version identifier (commit hash)
dfd8c46870b48511e697cbb43a3c8c91d375fa2f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

Latest version

Checks complete

Sync OpenAI Math manuscripts from 4eaee87548250b2682affda7d05d947c05554bae

Committed

All papers

Claims

Findings

Uniform exclusion of Landau–Siegel zeros

There is an absolute constant c>0c > 0 such that every real zero β∈(0,1)\beta \in (0, 1) of every primitive nonprincipal real Dirichlet LL-function of conductor q≥3q \geq 3 satisfies (1−β)log⁡q≥c(1 - \beta)\log q \geq c.

For a primitive nonprincipal real Dirichlet character χ\chi of conductor q≥3q \geq 3 with a real zero β∈(0,1)\beta \in (0, 1), writing ℓ=log⁡q\ell = \log q and δ=(1−β)ℓ\delta = (1-\beta)\ell, one has ∑p≤X,χ(p)=1log⁡pp≪ℓ+δ(log⁡X)2ℓ\sum_{p \leq X, \chi(p)=1} \frac{\log p}{p} \ll \ell + \frac{\delta(\log X)^2}{\ell} for X≥3X \geq 3, and consequently ∑H<p≤X,p∤2q,χ(p)=−1log⁡pp≥log⁡X−Cℓ−Cδ(log⁡X)2ℓ−CH\sum_{H < p \leq X, p \nmid 2q, \chi(p)=-1} \frac{\log p}{p} \geq \log X - C\ell - C\frac{\delta(\log X)^2}{\ell} - C_H for each integer H≥2H \geq 2, where CC is absolute and CHC_H depends only on HH.

Let A:C4→C3A : \mathbb{C}^4 \to \mathbb{C}^3 be a surjective linear map with ker⁡A=Cc\ker A = \mathbb{C}c, where the four coordinates of cc are linearly independent over Q\mathbb{Q}. For N≥1N \geq 1 and integers t1,t2,t3t_1, t_2, t_3 satisfying tj≥3(N−1)t_j \geq 3(N-1) and ∏j=13(tj−3(N−1)+1)>(4N−3)4\prod_{j=1}^3 (t_j - 3(N-1) + 1) > (4N-3)^4, the evaluation map B↦(B(An))n∈ENB \mapsto (B(An))_{n \in E_N} from Pt={B∈C[z1,z2,z3]:deg⁡zjB≤tj for 1≤j≤3}\mathcal{P}_t = \{B \in \mathbb{C}[z_1, z_2, z_3] : \deg_{z_j} B \leq t_j \text{ for } 1 \leq j \leq 3\} to CEN\mathbb{C}^{E_N} (where EN={0,…,N−1}4E_N = \{0, \dots, N-1\}^4) is surjective.

There are absolute constants c0,C0>0c_0, C_0 > 0 such that for every fixed integer H≥1H \geq 1 and all sufficiently large NN in terms of HH alone, the exponent sums S1=∑α∈Pα1S_1 = \sum_{\alpha \in \mathcal{P}} \alpha_1 and S2=∑α∈P(α2+α3)S_2 = \sum_{\alpha \in \mathcal{P}} (\alpha_2 + \alpha_3) of the greedily retained rows satisfy S1≥c0MH2/3US_1 \geq c_0 M H^{2/3} U and S2≤C0MH−1/3US_2 \leq C_0 M H^{-1/3} U (where U=N4/3U = N^{4/3} and M=N4M = N^4), and in particular S2/S1≤(C0/c0)/HS_2/S_1 \leq (C_0/c_0)/H.

For every admissible prime pp (i.e., p>Hp > H, p∤2qp \nmid 2q, and χ(p)=−1\chi(p) = -1), the determinant Δ=det⁡(Rα)α∈P∈R∖{0}\Delta = \det(R_\alpha)_{\alpha \in \mathcal{P}} \in \mathcal{R} \setminus \{0\} satisfies Δ∈pEpR\Delta \in p^{E_p}\mathcal{R} where Ep=∑α∈P⌊α1/p⌋E_p = \sum_{\alpha \in \mathcal{P}} \lfloor \alpha_1 / p \rfloor.

Methods

Logarithmic-derivative prime-sum estimation using the Hadamard product and functional equation of Λχ(s)\Lambda_\chi(s) evaluated at s=1+1/log⁡Xs = 1 + 1/\log X.

Source excerpts

“For the completed function Λχ(s) = (q/π) (s+ε)/2Γ ((s + ε)/2) L(s, χ), the Hadamard product and functional equation give, for real s > 1, (2.3) − L ′ L (s, χ) = 1 2 log q π + 1 2 Γ ′ Γ ((s + ε)/2) − X ρ Re 1 s − ρ .” · Show in paper

Polynomial interpolation on a projected integer box in C3\mathbb{C}^3 via a dimension count, nearest-point supporting-face selection, and Lagrange interpolation on a hyperplane.

Source excerpts

“We construct the test polynomial as a product. A translated polynomial will vanish at An for every n ∈ S outside one supporting face of P. A second polynomial will select a single vertex on that face, with total degree at most 3(N − 1).” · Show in paper

Weighted greedy row selection for an interpolation determinant Δ=det⁡(Rα)α∈P\Delta = \det(R_\alpha)_{\alpha \in \mathcal{P}} in the biquadratic ring R=Z[d,2]\mathcal{R} = \mathbb{Z}[\sqrt{d}, \sqrt{2}], ordered by w(α)=α1+Hα2+Hα3w(\alpha) = \alpha_1 + H\alpha_2 + H\alpha_3.

Source excerpts

“Order all indices in Z 3 ≥0 by increasing w(α) = α1 + Hα2 + Hα3, with any fixed rule to break ties. Retain a row precisely when it increases the span of the earlier rows over K.” · Show in paper

Archimedean Hadamard inequality and Frobenius-relation prime-divisibility bounds on the norm N(Δ)\mathrm{N}(\Delta) of the interpolation determinant.

Source excerpts

“Hadamard’s inequality and the four embeddings give (5.1) 1 4 log | N(∆)| ≤ M 2 log M + (S1 + S2) log N + 1 2 ℓ + log 8 .” · Show in paper