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

Asymptotically minimal maxima of real Littlewood polynomials

For Littlewood polynomials with real coefficients in {−1,1}\{-1, 1\}, lim⁡N→∞mN=1\lim_{N\to\infty} m_N = 1, meaning that for every η>0\eta > 0 there exists N0N_0 such that for every integer N≥N0N \ge N_0 there are signs ε0,…,εN−1∈{−1,1}\varepsilon_0, \dots, \varepsilon_{N-1} \in \{-1, 1\} with max⁡∣z∣=1∣∑k=0N−1εkzk∣≤(1+η)N\max_{|z|=1} |\sum_{k=0}^{N-1} \varepsilon_k z^k| \le (1+\eta)\sqrt{N}, disproving the real-sign analogue of Erdős's fixed relative-gap conjecture.

For every 0<δ<1/200 < \delta < 1/20, there exist vectors XN=(XN,0,…,XN,N−1)∈[−1,1]NX_N = (X_{N,0}, \dots, X_{N,N-1}) \in [-1, 1]^N for each integer N≥1N \ge 1 such that lim sup⁡N→∞∥QXN∥∞/N≤Kδ=(1+δ)3/(1−δ)\limsup_{N\to\infty} \|Q_{X_N}\|_\infty / \sqrt{N} \le K_\delta = \sqrt{(1+\delta)^3/(1-\delta)} and lim inf⁡N→∞1N∑k=0N−1XN,k2≥1−7δ\liminf_{N\to\infty} \frac{1}{N} \sum_{k=0}^{N-1} X_{N,k}^2 \ge 1 - 7\delta.

There is an absolute constant CC such that for every N≥1N \ge 1 and every X∈[−1,1]NX \in [-1, 1]^N, some signs ε0,…,εN−1∈{−1,1}\varepsilon_0, \dots, \varepsilon_{N-1} \in \{-1, 1\} satisfy max⁡t∈T∣∑k=0N−1(εk−Xk)e(kt)∣≤C(1+μ(X)log⁡80Nμ(X))\max_{t\in\mathbb{T}} \left|\sum_{k=0}^{N-1} (\varepsilon_k - X_k)e(kt)\right| \le C\left(1 + \sqrt{\mu(X)\log\frac{80N}{\mu(X)}}\right).

There are Littlewood polynomials PNP_N of every length NN such that, for each fixed 0<p<∞0 < p < \infty, ∫T∣∣PN(e(t))∣N−1∣pdt→0\int_{\mathbb{T}} \left|\frac{|P_N(e(t))|}{\sqrt{N}} - 1\right|^p dt \to 0 as N→∞N \to \infty.

For FN=max⁡A∈{−1,1}NF(A)\mathcal{F}_N = \max_{A \in \{-1,1\}^N} F(A), lim⁡N→∞FN=+∞\lim_{N\to\infty} \mathcal{F}_N = +\infty where the limit runs through all integer lengths N≥2N \ge 2, disproving Turyn's conjecture (equivalently Erdős's L4L^4-norm conjecture).

There is a binary Morse shift (X,S)(X, S) with a unique SS-invariant Borel probability measure μ\mu for which the Koopman operator USg=g∘SU_S g = g \circ S on L2(X,μ)L^2(X, \mu) has simple spectrum, and the spectral measure σf\sigma_f of the zero-coordinate function f(x)=x(0)f(x) = x(0) satisfies dσf=h dtd\sigma_f = h\,dt for some h∈L2(T)h \in L^2(\mathbb{T}) with h≥0h \ge 0 almost everywhere and ∫Th dt=1\int_{\mathbb{T}} h\,dt = 1.

Methods

Signed interval packing for pairwise nonparallel integer linear forms using a randomly constructed finite-field hypergraph and the Pippenger–Spencer edge-coloring theorem.

Source excerpts

“We construct such a hypergraph whose vertices are small disjoint slots in a half-circle. An edge records one compatible center of each type. A large matching will provide many compatible choices without reusing any slot.” · Show in paper

Spreading Fourier coefficients of a real trigonometric polynomial by introducing auxiliary torus variables with quadratic oscillations and smooth cutoffs, followed by finite Fourier truncation.

Source excerpts

“The construction starts with a polynomial of almost unit mean square. Quadratic oscillations in new variables spread its coefficients; finite Fourier truncation then gives the required polynomial.” · Show in paper

Sampling an auxiliary trigonometric polynomial along quadratic phases on disjoint blocks with smooth cutoffs, evaluated uniformly on the unit circle via Poisson summation and stationary phase.

Source excerpts

“Bombieri and Bourgain [5, Section 7] use Poisson summation to evaluate smoothed quadratic phases in their unimodular construction. The next lemma isolates the uniformity needed here for the maximum over the whole circle.” · Show in paper

Defect-sensitive dyadic rounding of real vectors in [-1, 1]^N to signs {-1, 1}^N using Lovett–Meka partial coloring with simultaneous sign reversal at each dyadic scale, extended from a discrete grid to the unit circle via Cauchy's estimate.

Source excerpts

“The discrepancy estimate will be applied to real and imaginary Fourier evaluations on a grid. At each dyadic scale we can reverse all the chosen signs without changing their discrepancy. This permits rounding without increasing the total defect mass, which is the source of the improved error bound.” · Show in paper