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.

All papers

Claims

Findings

The Quasi-Riemann Hypothesis (alternate 11/12 proof)

Every finite-order Hecke L-function over K = \mathbb{Q}(\sqrt{-3}) has no zeros in the half-plane \operatorname{Re} s > 11/12, and consequently every Dirichlet L-function, including \zeta(s), has no zeros for \operatorname{Re} s > 11/12 (which in particular rules out Landau-Siegel zeros).

For x \ge 2, \sup_{1 \le q \le x, (a,q)=1} |\pi(x; q, a) - \frac{1}{\varphi(q)} \int_2^x \frac{dt}{\log t}| \ll x11/12x^{11/12} \log x, with an absolute and effective implied constant.

For every fixed 0 < \vartheta \le 1/10 and \varepsilon > 0, there exists k = k(\vartheta, \varepsilon) \ge 1 such that for every compact interval I \subset (0, \infty), all smooth W supported in I, and D \ge 2, \sum_{0 < N_{K/\mathbb{Q}}(u) \le D1+ϑD^{1+\vartheta}} |A_u(D)|^2 \ll_{\nu,S,I,\vartheta,\varepsilon} (\max_{0\le j\le k} \|W(j)W^{(j)}\|_\infty)^2 D2+ϑ+εD^{2+\vartheta+\varepsilon}.

Fix \kappa > 0 and C_0 \ge 1. If \mathcal{H}, X, F \ge 1, \Sigma = XF \le DC0D^{C_0}, and \mathcal{H} \le \Sigma D−κD^{-\kappa}, then for every \varepsilon > 0 there is J = J(\kappa, C_0, \varepsilon) \ge 1 such that \mathcal{E}(\mathcal{H}, X, F; \xi, W) \ll_{I,\nu,S,\kappa,C_0,\varepsilon} \|W\|_{C^J(I)}^2 D^\varepsilon \Sigma.

For \varepsilon > 0, C_0 \ge 1, and a character \xi of the fixed ray class group, there is J = J(\varepsilon, C_0) \ge 1 such that \sum_{0 < N_{K/\mathbb{Q}}(k) \ll \mathcal{H}} |T(X; k, f)|^2 \ll_{I,\xi,S,\varepsilon,C_0} D^\varepsilon \|W\|_{C^J(I)}^2 (\mathcal{H} + \mathcal{H}^2 N_{K/\mathbb{Q}}(f)/X) whenever 1 \le \mathcal{H}, X, N_{K/\mathbb{Q}}(f) \le DC0D^{C_0}.

Under 1 \le \mathcal{H}, L, F, \Sigma \le DC0D^{C_0} and \max\{\mathcal{H}, LF\} \le \Sigma, for every m \ge 0 and \varepsilon > 0, \mathcal{A}(W) \ll D^\varepsilon \Sigma \|W\|_{C4m+12C^{4m+12}(I)}^2 (1 + \sup \mathcal{E}(\mathcal{H}', X', F'; \xi', U)/\Sigma'), where the supremum is over \mathcal{H}', X', F' \ge 1 with \Sigma' = X'F', \mathcal{H}' \le \mathcal{H}L/(\Sigma F), \mathcal{H}'/\Sigma' \le \mathcal{H}/\Sigma, and \Sigma' \le L.

Methods

Embedding a smoothed \nu-twisted Mobius sum A_1(D) into a family A_u(D) parametrized by sextic residue symbols \chi_n(u) = (u/n)_6 and recovering A_1(D) from a mean-square estimate over u by averaging over sixth powers p^6 of primary primes.

Source excerpts

“In order to estimate A_1(D), we embed it into a family of such sums, parametrized by u \in \mathcal{O}_K, by introducing a sextic twist.” · Show in paper

“For prime ideals Y/2 < N_{K/\mathbb{Q}}(\mathfrak{p}) \le Y, \mathfrak{p} \notin S, let p be their primary generators, chosen with p \equiv 1 (mod 3). Then \chi_n(p^6) = \mathbf{1}_{\mathfrak{p}\nmid n}” · Show in paper

Converting sextic Gauss sums arising from Poisson summation into cubic Gauss sums using Gauss-Jacobi identities and reciprocity, then realizing the resulting sums as Fourier coefficients of Kubota's cubic theta function \theta(z, v) and applying its automorphy law.

Source excerpts

“Upon application of the Gauss-Jacobi identities, these sextic Gauss sums absorb the factor of \mu and turn into cubic Gauss sums. We interpret the resulting cubic Gauss sums as coefficients of Kubota's cubic theta function.” · Show in paper

“The automorphy of \theta gives a summation formula that transforms the completed column sum T_h(X), with the row h held fixed, into dual sums of theta coefficients with new smooth weights and character twists.” · Show in paper

Applying Goldmakher and Louvel's quadratic large sieve over number fields to bound the mean square of the transformed theta-coefficient sums twisted by the quadratic character \chi_h^3.

Source excerpts

“Now a key point is that since \chi_p^3 is quadratic, Goldmakher and Louvel's quadratic large sieve [13, Theorem 1.1 and Corollary 1.2] (a generalization of Heath-Brown's quadratic large sieve [18] to number fields) bounds the mean square of the sums in (2.13) as h varies.” · Show in paper

Cube completion and Mobius inversion to pass between the completed theta sum T(X; k, f) and the squarefree-index sum, combined with a recursive transfer estimate via two Poisson summations with an enlarged intermediate row range.

Source excerpts

“We begin by undoing the addition of cube factors using Mobius inversion.” · Show in paper

“These two Poisson summations give the transfer estimate: a bound for the remaining mean square in terms of new mean squares of the same type.” · Show in paper

Multivariable Mellin inversion to separate variables in smooth kernels depending on the row index, extending mean-square bounds for common test functions to coupled kernels.

Source excerpts

“We use Mellin inversion to separate the variables of a smooth weight, with coefficient bounds controlled by finitely many derivatives.” · Show in paper

“The next lemma extends mean-square bounds for a common test function to smooth kernels depending on the row, with losses controlled by uniform bounds on their derivatives.” · Show in paper