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

Universal optimality of the triangular lattice

For every smooth and completely monotone function g:(0,∞)→[0,∞)g : (0, \infty) \to [0, \infty) and every locally finite configuration C⊂R2\mathcal{C} \subset \mathbb{R}^2 of centered disk density one, the lower limit of centered-ball energy averages satisfies Eg(C)≥∑a∈A∖{0}g(∣a∣2)=Eg(A)E_g(\mathcal{C}) \ge \sum_{a \in A \setminus \{0\}} g(|a|^2) = E_g(A) in the extended nonnegative reals, where AA is the triangular lattice of covolume one.

For every finite real k≥2.36k \ge 2.36, there exist entire functions H1,H2H_1, H_2 real on the real axis and a real radial Schwartz function ff on R2\mathbb{R}^2 such that f(x)=e−πhb∣x∣2H1(b∣x∣2)f(x) = e^{-\pi h b |x|^2} H_1(b|x|^2), f^(ξ)=e−πhb∣ξ∣2H2(b∣ξ∣2)\widehat{f}(\xi) = e^{-\pi h b |\xi|^2} H_2(b|\xi|^2), H1(s)≤Tk(s)=k−1e−k(s−1)H_1(s) \le T_k(s) = k^{-1}e^{-k(s-1)} and H2(s)≥0H_2(s) \ge 0 for all s≥0s \ge 0, with exact value and derivative contacts H1(n)=Tk(n)H_1(n) = T_k(n), H1′(n)=Tk′(n)H_1'(n) = T_k'(n), and H2(n)=H2′(n)=0H_2(n) = H_2'(n) = 0 at every n∈Nn \in \mathcal{N}.

For every α>0\alpha > 0, there exists a real radial Schwartz function fαf_\alpha satisfying fα≤Gαf_\alpha \le G_\alpha, fα^≥0\widehat{f_\alpha} \ge 0, fα(a)=Gα(a)f_\alpha(a) = G_\alpha(a) for a∈A∖{0}a \in A \setminus \{0\}, and fα^(w)=0\widehat{f_\alpha}(w) = 0 for w∈A∗∖{0}w \in A^* \setminus \{0\}, where Gα(x)=e−πα∣x∣2G_\alpha(x) = e^{-\pi \alpha |x|^2}; consequently, every locally finite configuration C\mathcal{C} of centered disk density one satisfies lim inf⁡R→∞NR−1∑x,y∈CR,x≠ye−πα∣x−y∣2≥∑a∈A∖{0}e−πα∣a∣2\liminf_{R \to \infty} N_R^{-1} \sum_{x, y \in \mathcal{C}_R, x \neq y} e^{-\pi \alpha |x-y|^2} \ge \sum_{a \in A \setminus \{0\}} e^{-\pi \alpha |a|^2}.

For 0≤s<20 \le s < 2 and every positive integer nn, the n2n^2 classes of A/(nA)A/(nA) minimize the canonical periodic energy Ws,nAW_{s, nA} among all configurations of n2n^2 distinct points in R2/(nA)\mathbb{R}^2/(nA); moreover, AA minimizes the infinite-system field energy WsW_s among density-one planar configurations, and the thermodynamic jellium minimum is eJel(s)=κs−1Ws(A)e_{\mathrm{Jel}}^{(s)} = \kappa_s^{-1} W_s(A).

The exact quadrature arrays defined in Section 3 satisfy ∥D64∥<.004\|D^{64}\| < .004, ∥V+∥+∥V−∥<36\|V_+\| + \|V_-\| < 36, ∑i=12∥(Xi)j∥1<2.3\sum_{i=1}^2 \|(X_i)_j\|_1 < 2.3 for 1≤j≤51 \le j \le 5 and <5.6< 5.6 for 6≤j≤106 \le j \le 10, as well as the 37,310 Bernstein half-gap lower bounds (3.11) exceeding .001.001 and the 10 tail lower bounds (3.12) exceeding .0021.0021.

For every real k≥2.36k \ge 2.36, there is a unique pair of positive-node lists in E=ℓ1(N;R2)\mathcal{E} = \ell^1(\mathcal{N}; \mathbb{R}^2) with fixed extra coordinates ei=Eiq(k)e_i = E_i q(k) whose full lists ℓi\ell_i satisfy the exact interpolation equations (4.1), with ∑i=12∥ℓi−ℓiref∥1<2⋅10−8\sum_{i=1}^2 \|\ell_i - \ell_i^{\mathrm{ref}}\|_1 < 2 \cdot 10^{-8} and ∑i=12∥ℓi∥1<5\sum_{i=1}^2 \|\ell_i\|_1 < 5.

For any real even Schwartz function ff on R2\mathbb{R}^2 with f^≥0\widehat{f} \ge 0 and any locally finite configuration C⊂R2\mathcal{C} \subset \mathbb{R}^2 of centered disk density one, lim inf⁡R→∞NR−1∑x,y∈CR,x≠yf(x−y)≥f^(0)−f(0)\liminf_{R \to \infty} N_R^{-1} \sum_{x, y \in \mathcal{C}_R, x \neq y} f(x - y) \ge \widehat{f}(0) - f(0).

Methods

Cardinal interpolation on the periodic superset N0\mathcal{N}_0 using a degree-12 trigonometric sine product P(s)P(s) with double zeros, realized by a compactly supported spectral measure μa\mu_a on [−1,1][-1, 1] and its Gaussian-damped Fourier transform Kpa(s)K_{p_a}(s).

Source excerpts

“P(s) = \prod_{a \in L} \left(2\sin\frac{\pi(s-a)}{12}\right)^2 = \sum_{l=-6}^6 P_l eiπtlse^{i\pi t_l s}, \quad t_l = \frac{l}{6}.” · Show in paper

“There is a finite complex measure \mu_a on [-1, 1] such that p_a(s) = \int_{-1}^1 eiπtse^{i\pi t s} \, d\mu_a(t) (s \in \mathbb{C})” · Show in paper

First Fejér positive quadrature rule on the six frequency intervals ((j−1)/6,j/6)((j-1)/6, j/6) with N=384N = 384 nodes per interval, exact through polynomial degree 383, combined with finite-block inversion and Schur-complement tail elimination in ℓ1(N;R2)\ell^1(\mathcal{N}; \mathbb{R}^2).

Source excerpts

“The replacement of the density integral is the sum of masses o_v d_j(\tau_{jv})/6. Since 0 < \pi \alpha_v < \pi, every quadrature point lies inside its frequency interval. This is the first Fejér rule described by Waldvogel [32, Section 2], divided by two so that its weights on [-1, 1] approximate the average rather than the integral of mass two.” · Show in paper

“Substituting that solution into the tail equation leaves the Schur complement \mathcal{Q} = \mathcal{L}_{\mathcal{T}\mathcal{T}} - \mathcal{L}_{\mathcal{T}\mathcal{F}} \mathcal{L}_{\mathcal{F}\mathcal{F}}^{-1} \mathcal{L}_{\mathcal{F}\mathcal{T}}.” · Show in paper

Sign certification on bounded half-gaps [0,s0][0, s_0] using degree-40 Bernstein polynomial enclosures of second-order Taylor-subtracted quotients Dm,2u(s)\mathcal{D}_{m,2} u(s) and rectangular containing-box parameter bounds [l]K,J[l]_{K,J}, paired with a quadratic sine-product barrier P(s)≥.68(s−m)2P(s) \ge .68(s-m)^2 on the tail [s0,∞)[s_0, \infty).

Source excerpts

“The data vector q(k) lies on a one-dimensional curve. To obtain lower bounds uniform in k, we place each part of that curve in a rectangular box.” · Show in paper

“The rows Li,dL_{i,d} are exactly the degree-40 Bernstein coefficients of \sum_{r=0}^{40} li,rl_{i,r} x^r.” · Show in paper

“For every real s, let m \in \mathcal{Z} be any nearest point of the full periodic zero set, so |s-m| = \min_{n \in \mathcal{Z}} |s-n|. Then P(s) \ge .68(s-m)^2.” · Show in paper

Shifted-moment positive-mixture representation g(ε+t)=∫[0,1]vt dρε(v)g(\varepsilon + t) = \int_{[0,1]} v^t \, d\rho_\varepsilon(v) for completely monotone functions, combined with Fatou's lemma, Tonelli's theorem, and monotone convergence as ε↓0\varepsilon \downarrow 0.

Source excerpts

“Finally, Section 7 represents each positive shift of a completely monotone potential as g(\varepsilon + t) = \int_{[0,1]} v^t \, d\rho_\varepsilon(v) (t > 0), where \varepsilon > 0 and \rho_\varepsilon is positive with total mass g(\varepsilon).” · Show in paper

Heat-kernel comparison across arbitrary period lattices for 0≤s<20 \le s < 2 and cubic periodic approximation of the infinite-system field energy via stationary minimizing laws, rectangular shell screening without new close pairs, reflection, and canonical projection.

Source excerpts

“V_s(\mathcal{C}) - V_s(\mathcal{D}) = \frac{\kappa_s q_s}{\Gamma(1-s/2)} \int_0^\infty (H_{\mathcal{C}}(t) - H_{\mathcal{D}}(t)) t−s/2t^{-s/2} \, dt.” · Show in paper

“The proof has three inputs: cutoff control, a stationary minimizing law, and a screening construction. The following lemmas establish them before we reflect and project the screened fields to prove the proposition.” · Show in paper