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

A sharp Fourier certificate for planar circle packing

There is a real radial Schwartz function f:R2→Rf : \mathbb{R}^2 \to \mathbb{R} such that f^(0)=1\widehat{f}(0) = 1, f(0)=23f(0) = \frac{2}{\sqrt{3}}, f^(ξ)≥0\widehat{f}(\xi) \ge 0 for all ξ∈R2\xi \in \mathbb{R}^2, and f(x)≤0f(x) \le 0 for all ∣x∣≥1|x| \ge 1, resolving the two-dimensional case of Cohn and Elkies's Conjecture 7.3.

If a periodic packing of disks of radius 1/21/2 has covered-area density π/(23)\pi/(2\sqrt{3}), then its center set is the image of Z(1,0)+Z(1/2,3/2)\mathbb{Z}(1, 0) + \mathbb{Z}(1/2, \sqrt{3}/2) under a Euclidean isometry.

There are unique real lists x1,x2∈X=ℓ1(N;R2)x_1, x_2 \in X = \ell^1(\mathcal{N}; \mathbb{R}^2) satisfying the coupled interpolation equations (2.25) with fixed data (2.22), and each satisfies ∥xi−x~i∥1<.00003\|x_i - \widetilde{x}_i\|_1 < .00003 and total norm less than 2.32.3.

The exact functions H1,H2H_1, H_2 have zero sets {s≥1:H1(s)=0}=N\{s \ge 1 : H_1(s) = 0\} = \mathcal{N} and {s≥0:H2(s)=0}=N\{s \ge 0 : H_2(s) = 0\} = \mathcal{N}, satisfy H1(s)<0H_1(s) < 0 (s≥1,s∉Ns \ge 1, s \notin \mathcal{N}) and H2(s)>0H_2(s) > 0 (s≥0,s∉Ns \ge 0, s \notin \mathcal{N}), with a simple zero of H1H_1 at 11 and double zeros at all other listed nodes.

The exact Fourier pair satisfies F(0)=F^(0)=6Q1e−πh>0F(0) = \widehat{F}(0) = 6 Q_1 e^{-\pi h} > 0.

For a cardinal function pp from Lemma 2.1, qp(x)=e−πhsp(s)q_p(x) = e^{-\pi h s} p(s) (s=b∣x∣2s = b|x|^2) is real, radial, and Schwartz on R2\mathbb{R}^2, and qp^(ξ)=e−πhsKp(s)\widehat{q_p}(\xi) = e^{-\pi h s} K_p(s) (s=b∣ξ∣2s = b|\xi|^2), where KpK_p is entire, real on R\mathbb{R}, and every derivative decays exponentially as s→+∞s \to +\infty.

For each η∈{1,−1}\eta \in \{1, -1\}, I−ηSI - \eta S has a bounded inverse on XX, and ∥(I−ηS)−1r∥1≤91∥rf∥1+2540∥rt∥1\|(I - \eta S)^{-1} r\|_1 \le 91 \|r_f\|_1 + 2540 \|r_t\|_1 for r=rf+rtr = r_f + r_t supported on IfI_f and N∩(100,∞)\mathcal{N} \cap (100, \infty) respectively.

The N=256N = 256 first Fejér quadrature rule (A.6), with exact evaluations at its nodes and at the atoms, has error less than 10−2210^{-22} in each moment Up,l(m,y)U_{p,l}(m,y) and Xp,l(m,y)X_{p,l}(m,y) for ∥(c,d)∥1≤3\|(c,d)\|_1 \le 3, ∣C∣≤1|C| \le 1, 0≤m≤1000 \le m \le 100, ∣y∣≤3/2|y| \le 3/2, and l∈{0,…,30}l \in \{0,\dots,30\}.

Methods

Gaussian-damped trigonometric cardinal family qp(x)=e−πhsp(s)q_p(x) = e^{-\pi h s} p(s) with p(s)=P(s)R(s)p(s) = P(s)R(s) vanishing doubly on the periodic node set N=(12Z+{0,1,3,4,7,9})∩(0,∞)\mathcal{N} = (12\mathbb{Z} + \{0,1,3,4,7,9\}) \cap (0,\infty).

Source excerpts

“We impose interpolation conditions at the positive nodes of a periodic set: A = {0, 1, 3, 4, 7, 9}, N = (12Z + A) ∩ (0, ∞).” · Show in paper

“For a cardinal function p, let qp(x) = e−πhsp(s), s = b |x|2.” · Show in paper

Two-stage Schur complement and Neumann series inversion of I−ηSI - \eta S on ℓ1(N;R2)\ell^1(\mathcal{N}; \mathbb{R}^2).

Source excerpts

“A finite inverse handles the coordinates on J. A first Schur complement then incorporates E, using separate bounds for its rows through 16 and beyond 16; a second incorporates the infinite tail beyond 100.” · Show in paper

Taylor quotient range enclosure on half-gaps using degree-28 Bernstein polynomial coefficients, combined with logarithmic concavity of P(s)P(s) and second-derivative remainder bounds on [41.5,∞)[41.5, \infty).

Source excerpts

“After writing u = yv, the first 29 terms of the tabulated series in (4.1) form a polynomial of degree at most 28 in v. We bound it on [0, 1] using the standard range enclosure by Bernstein coefficients [9, Section 3], retaining its elementary proof below.” · Show in paper

First Fejér quadrature rule (N=256N = 256) with analytic Cauchy remainder bounds (10−2210^{-22}) and 192-bit Arb interval arithmetic.

Source excerpts

“We use the first Fejér quadrature rule, whose cosine-sum weights are described in [11, Section 2, formula (2.3)].” · Show in paper

“The supplied implementation uses Arb through python-flint at 192 bits of precision; the citation does not by itself certify this local environment, its call domains, or a run.” · Show in paper

Alternative rational certification methods, including fixed rational evaluation with rounded exponentials, an N=64N=64 exact rational first-block check, rational-center Taylor integration, and direct node-Taylor representation.

Source excerpts

“The first gives a fixed rational evaluation of every N = 256 moment and proves a uniform error theorem. A separate exact-rational program checks the first 6 × 6 block with an N = 64 rule. Rational-center Taylor integration gives a different, arbitrarily refinable enclosure method, and the final proposition gives a local formula for the direct cardinal derivatives.” · Show in paper