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

Endpoint convergence for the planar Schrodinger equation

For every f∈H1/3(R2)f \in H^{1/3}(\mathbb{R}^2) there is a set Ef⊂R2E_f \subset \mathbb{R}^2 of full Lebesgue measure such that, for each x∈Efx \in E_f, the finite limit uf(x,t)=lim⁡a↓0Uaf(x,t)u_f(x, t) = \lim_{a \downarrow 0} U_a f(x, t) exists for every real 0<t<10 < t < 1, lim⁡t↓0uf(x,t)=f∗(x)\lim_{t \downarrow 0} u_f(x, t) = f^*(x), uf(x,⋅)u_f(x, \cdot) extends continuously to [0,1)[0, 1) with value f∗(x)f^*(x) at zero, and the Gaussian limit is uniform over 0<t<10 < t < 1.

There is a constant CC such that for every f∈H1/3(R2)f \in H^{1/3}(\mathbb{R}^2) whose Fourier transform has compact support and every x0∈R2x_0 \in \mathbb{R}^2, ∫B(x0,1)sup⁡0≤t<1∣S(t)f(x)∣ dx≤C∥f∥H1/3\int_{B(x_0, 1)} \sup_{0 \le t < 1} |S(t)f(x)| \, dx \le C \|f\|_{H^{1/3}}, and for arbitrary f∈H1/3(R2)f \in H^{1/3}(\mathbb{R}^2) there is a jointly measurable representative vf(x,t)v_f(x, t) continuous in tt for almost every xx satisfying the same bound.

There are fixed p>2p > 2 and σ>0\sigma > 0 such that for any finite sum UU of compact atoms of packet size L≥1L \ge 1 with ∑α∣cα∣2≤1\sum_\alpha |c_\alpha|^2 \le 1, any measure μ\mu supported in a spatial box of side O(L2)O(L^2) and ∣t∣≤L2|t| \le L^2 with μ(Br)≤mr2\mu(B_r) \le m r^2 for r≥1r \ge 1, and satisfying the sparsity condition (2.4) with 1≤D≤L1 \le D \le L, μ({z:∣U(z)∣≥gL−4/3})≤CmL4D−σg−p\mu(\{z : |U(z)| \ge g L^{-4/3}\}) \le C m L^4 D^{-\sigma} g^{-p} for every g>0g > 0.

For finite packet sums V1,V2V_1, V_2 at scale h≥1h \ge 1 with frequencies in intervals of length δ\delta (h−1/2≤δ≲1h^{-1/2} \le \delta \lesssim 1) separated by at least a fixed positive multiple of δ\delta, if αi=Ai/Bi≥C\alpha_i = A_i/B_i \ge C, the number of unit lattice squares containing a point where simultaneously ∣Vi∣≥Ai|V_i| \ge A_i and Si≤BiS_i \le B_i for i=1,2i = 1, 2 satisfies #≤C′δ−C′α1−2α2−2(H1∗/A12+H2∗/A22)\# \le C' \delta^{-C'} \alpha_1^{-2} \alpha_2^{-2} (H_1^*/A_1^2 + H_2^*/A_2^2) uniformly in hh, the number of packets, and the number of scales.

There is an absolute constant cs>0c_s > 0 such that for τ0>0\tau_0 > 0, Lτ0≤D≤LL^{\tau_0} \le D \le L, a finite packet array UU of the form (2.2) with coefficient vector cc, and a measure μ\mu satisfying μ(B(z,r))≤mr2\mu(B(z, r)) \le m r^2 (r≥1r \ge 1), supp⁡μ⊂{∣t∣≤CL2}\operatorname{supp} \mu \subset \{|t| \le C L^2\}, and the sparsity condition (2.4) with M=100M = 100, ∑qμ(q)2/3sup⁡q∣U∣2≲τ0m2/3L4/3D−cs∥c∥ℓ22\sum_q \mu(q)^{2/3} \sup_q |U|^2 \lesssim_{\tau_0} m^{2/3} L^{4/3} D^{-c_s} \|c\|_{\ell^2}^2.

For a fixed bounded frequency region, there exist c>0c > 0 and L0<∞L_0 < \infty such that for L≥L0L \ge L_0, a collection Q\mathcal{Q} of unit lattice cubes in a ball of radius L2L^2 with #Q≥L4−c\#\mathcal{Q} \ge L^{4-c} and #{Q∈Q:Q∩B(z,r)≠∅}≤Lcr2\#\{Q \in \mathcal{Q} : Q \cap B(z, r) \neq \emptyset\} \le L^c r^2 (r≥1r \ge 1), and frequency-supported v1,v2,v3v_1, v_2, v_3 with ∥vi∥2≤1\|v_i\|_2 \le 1 and sup⁡Q∣Evi∣≥L−4/3−c\sup_Q |Ev_i| \ge L^{-4/3-c} on Q∈QQ \in \mathcal{Q}, it is impossible that every triple of unit frequency directions has absolute determinant at least L−cL^{-c}.

Under the hypotheses of Proposition 7.1, every tube in a finite family T\mathcal{T} that is rich against a measure ν\nu (satisfying a two-dimensional ball bound) and supports a pointwise angularly nonconcentrated probability law at ν\nu-almost every point can be assigned to one of at most XC′X^{C'} plates of thickness XC′X^{C'} such that ∣n⋅(2ωT,1)∣≤XC′/H|n \cdot (2\omega_T, 1)| \le X^{C'}/H.

Methods

Limiting information calculus on a subsequence of counterexamples, tracking realized conditional information IL(Y∣V)=−log⁡LP(Y=Y(ω)∣V)I_L(Y | V) = -\log_L \mathbb{P}(Y = Y(\omega) | V) across spatial scales and applying Bourgain-He's discretized projection theorem and polynomial partitioning to obtain a power saving for transverse inputs.

Source excerpts

“For a discrete variable Y and arbitrary conditioning V , write IL(Y | V ) = − logL P(Y = Y (ω) | V ) for its information evaluated at the sampled labels.” · Show in paper

“This is the n = 2, m = 1, α = 2d case of [17, Theorem 1], a general form of the planar Bourgain discretized projection theorem.” · Show in paper

Positive frame iteration with smooth real-variable deletion masks ρa\rho_a and a Gaussian quartic potential Pj(f)\mathcal{P}_j(f) to control low-frequency flux and smoothing-scale changes uniformly in the number of scales.

Source excerpts

“The masks will obey 0 ≤ ρa ≤ 1. Analysis by the Pa is an isometry and synthesis is its adjoint, hence a contraction.” · Show in paper

“For a spatial Gaussian multiplier qj,a of width ∆j centered at a, define the normalized potential Pj (f) = R R R R2 |qj,af(z)|4 dz da ∆j .” · Show in paper

Common dyadic time tree summation exploiting the convexity deficit (a+b)α−aα−bα≥(2α−2)bα(a+b)^\alpha - a^\alpha - b^\alpha \ge (2^\alpha - 2)b^\alpha for α=3/2−1/p>1\alpha = 3/2 - 1/p > 1 (from p>2p > 2) to sum dualized short-time maximal estimates across dyadic frequency scales without a logarithmic loss.

Source excerpts

“The next step uses the same mass assignment I 7→ mI for every j. Choosing a different graph separately at each frequency would destroy the telescoping argument.” · Show in paper

“If a parent has child masses a ≥ b ≥ 0, then (a + b)α − aα − bα ≥ (2α − 2)bα. (9.15)” · Show in paper

Constructing a jointly measurable, continuous-in-time representative vf(x,t)v_f(x, t) via smooth compactly supported approximants and identifying (Ka∗v(⋅,t))(x)=Uaf(x,t)(K_a * v(\cdot, t))(x) = U_a f(x, t) via Plancherel for each fixed tt to recover the Gaussian-first, all-real-time limit without intersecting uncountably many time-dependent exceptional sets.

Source excerpts

“Plancherel’s identity therefore gives (Ka ∗ v(·, t))(x) = 1 (2π)2 Z eix·ξ−it|ξ|2−a|ξ|2 bf(ξ) dξ = Uaf(x, t) (9.23) for every x ∈ R2, every a > 0, and every 0 < t < 1.” · Show in paper