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 Kakeya maximal conjecture in three dimensions

For every ε>0\varepsilon > 0 there exists a finite constant CεC_\varepsilon such that, for every 0<δ<10 < \delta < 1 and every f∈L3(R3)f \in L^3(\mathbb{R}^3), ∥Kδf∥L3(S2)≤Cεδ−ε∥f∥L3(R3)\|K_\delta f\|_{L^3(S^2)} \leq C_\varepsilon \delta^{-\varepsilon} \|f\|_{L^3(\mathbb{R}^3)}, resolving the three-dimensional Kakeya maximal conjecture affirmatively.

There are values p>2p > 2 arbitrarily close to 22 for which the critical exponent h(p,0)=0h(p, 0) = 0.

If p>2p > 2 is in the parameter range of Definition 2.3 and h(p,0)=0h(p, 0) = 0, then for every κ>0\kappa > 0 there is a finite constant Cp,κC_{p,\kappa} such that, for all 0<δ<10 < \delta < 1 and all complex-valued f∈L3(R3)f \in L^3(\mathbb{R}^3), ∥Kδf∥L3(S2,σ)≤Cp,κδ−(p−2)/3−κ∥f∥L3(R3)\|K_\delta f\|_{L^3(S^2,\sigma)} \leq C_{p,\kappa} \delta^{-(p-2)/3 - \kappa} \|f\|_{L^3(\mathbb{R}^3)}.

For every 1≤p≤31 \leq p \leq 3 and ε>0\varepsilon > 0, with q=2p′q = 2p' (q=∞q = \infty at p=1p=1), ∥NδEucf∥Lq(R3)≤Cp,εδ1−3/p−ε∥f∥Lp(R3)\|N^{\mathrm{Euc}}_\delta f\|_{L^q(\mathbb{R}^3)} \leq C_{p,\varepsilon} \delta^{1-3/p-\varepsilon} \|f\|_{L^p(\mathbb{R}^3)} for 0<δ<10 < \delta < 1, and for each fixed local geometry U⋐VU \Subset V on a constant-curvature 3-manifold (M,g)(M, g) and 0<δ<δ00 < \delta < \delta_0, ∥Nδ,Ugf∥Lq(U,dVg)≤Cp,ε,U,V,gδ1−3/p−ε∥f∥Lp(V,dVg)\|N^g_{\delta,U} f\|_{L^q(U, dV_g)} \leq C_{p,\varepsilon,U,V,g} \delta^{1-3/p-\varepsilon} \|f\|_{L^p(V, dV_g)}.

For a fixed smooth real-valued translation-invariant phase ϕ(x,t;y)=x⋅y+ψ(t;y)\phi(x,t;y) = x \cdot y + \psi(t;y) satisfying Hörmander's conditions (H1)-(H2) and Bourgain's condition on Ω×Y\Omega \times Y, there is 0<δcurv≤ρ0 < \delta_{\mathrm{curv}} \leq \rho such that for every 1≤p≤31 \leq p \leq 3, ε>0\varepsilon > 0, q=2p′q = 2p' (q=∞q = \infty at p=1p=1), 0<δ<δcurv0 < \delta < \delta_{\mathrm{curv}}, and f∈Lp(Ω)f \in L^p(\Omega), ∥Kδϕf∥Lq(Y)≤Cp,ε,ϕ,Ω,Yδ1−3/p−ε∥f∥Lp(Ω)\|\mathcal{K}^\phi_\delta f\|_{L^q(Y)} \leq C_{p,\varepsilon,\phi,\Omega,Y} \delta^{1-3/p-\varepsilon} \|f\|_{L^p(\Omega)}.

There is an exact equality whose profile is F(κ)=β(κ−τ)+F(\kappa) = \beta(\kappa - \tau)_+, where 0<τ<10 < \tau < 1, ℓ=1−τ\ell = 1 - \tau, β=v/ℓ≤1\beta = v/\ell \leq 1, its deficit is v=−hpv = -h_p, and τ\tau is the least initial delay among all equalities with deficit vv.

For the canonical equality, there is a realizing sequence of canonical equalities and compatible packet assignments with regular branching exponent s0=1−β>0s_0 = 1 - \beta > 0, assigned packets BcB_c of raw active index mass at least nc=ΔhN(B∗+τ)c+dw(1−c)n_c = \Delta_h N^{(B_*+\tau)c + dw(1-c)} in exponent, and width bounds LX(c),LY(c),LU(c),LV(c)L_X(c), L_Y(c), L_U(c), L_V(c) in every used event frame.

At almost every interior point (c,a)∈U(c, a) \in \mathcal{U}, the conditional entropy rates satisfy a1≥y1a_1 \geq y_1, u1≥v1u_1 \geq v_1, v1>0  ⟹  y1≥min⁡{1,s0+v1}v_1 > 0 \implies y_1 \geq \min\{1, s_0 + v_1\}, and a1<1  ⟹  u1≤(a1−s0)+a_1 < 1 \implies u_1 \leq (a_1 - s_0)_+ when τ≠ℓ\tau \neq \ell, and a1<1  ⟹  m1≤1+(a1−s0)+a_1 < 1 \implies m_1 \leq 1 + (a_1 - s_0)_+ and v1≤(a1−s0)+v_1 \leq (a_1 - s_0)_+ when τ=ℓ\tau = \ell.

Under uniform ss-regularity and marginal Frostman bounds for μi\mu_i, for every ϵ>0\epsilon > 0 there are a subsequence, a fixed rational ζ∈(0,ζmax⁡]\zeta \in (0, \zeta_{\max}], and a probability νD\nu_D on triples (i,ξ,υ)(i, \xi, \upsilon) whose marginals (i,ξ)(i,\xi) and (i,υ)(i,\upsilon) are dominated by DϵζπD(di)μiD^{\epsilon\zeta}\pi_D(di)\mu_i and whose conditional product-slope law pξ(υ)=(tυ−tξ)(ϑυ−ϑξ)\mathfrak{p}_\xi(\upsilon) = (t_\upsilon - t_\xi)(\vartheta_\upsilon - \vartheta_\xi) satisfies νD{pξ(υ)∈Jr∣i,ξ}≤CDϵζrs−ϵ\nu_D\{\mathfrak{p}_\xi(\upsilon) \in J_r \mid i, \xi\} \leq C D^{\epsilon\zeta} r^{s-\epsilon} for D−ζ≤r≤1D^{-\zeta} \leq r \leq 1.

Methods

Weighted indexed line-configuration model with deformed critical exponent h(p,z)h(p, z) and smoothed temporal cost ΠF(1)=∫01Pp,q(F′(u)) du\Pi_F(1) = \int_0^1 P_{p,q}(F'(u))\,du.

Source excerpts

“An index i consists of a positive weight ωi, a matrix Mi = (bi, ui) in a fixed bounded subset of R2×2, and a nonempty subset Si of the N dyadic time bins in [0, 1].” · Show in paper

“Definition 2.3 (Critical exponent). For fixed p, z, η, let h(p, z) be the least A ∈ [0, d − 1] such that for every sequence with a uniform profile F, m ≤exp NA+ΠF (1)∆A.” · Show in paper

Finite cleaning and slow-diagonal profile regularization preserving uniform temporal profiles and active-weight lower bounds.

Source excerpts

“Lemma 2.5 (Finite cleaning). Let Ω be a finite set with positive measure µ, let A ⊆ Ω satisfy µ(A) ≥ ρµ(Ω), and let P1, . . . , PL be partitions of Ω.” · Show in paper

“Lemma 2.6 (Regularization and inherited branching). The following assertions hold. (i) Every sequence of indexed configurations has, after passage to a subsequence, an extensive refinement with a uniform temporal profile.” · Show in paper

Isotropic scale cut and anisotropic normalization of saturated plates to rescale local packets into full-time critical equalities.

Source excerpts

“If χ < 1, the isotropic scale cut described below gives an ensemble at resolution N′ = N b/(Ka) = N1−χ with matching sharp upper and lower bounds.” · Show in paper

“Lemma 4.8 (Anisotropic normalization). Suppose that on an extensive regular graph, over larger blocks of K1 = Nκ1 bins, there are disjoint indexed-block assignments to saturated plates on a flat scale K = Nκ ≤ K1.” · Show in paper

Derivative capture and compatible finite chains to construct canonical stationary equalities with self-reproducing outer models.

Source excerpts

“Lemma 5.1 (Derivative capture). Every equality profile has deficit α ≤ v = −hp. In particular, v > 0.” · Show in paper

“Lemma 5.2 (Compatible finite cuts). Fix a nontrivial equality and a finite increasing mesh τ = κ0 < κ1 < · · · < κJ < 1. There is an extensive refinement with simultaneous saturated assignments at these physical prefixes.” · Show in paper

Moving-frame conditional entropy rates and two-frame projection/product-slope pinning comparison against product laws.

Source excerpts

“Proposition 7.4 (Entropy demand). At almost every point of U, ∂cH(c, a) ≥ s0ℓ + wg − M.” · Show in paper

“Lemma 8.7 (Transfer of a projection bound to rates). Fix the original single-event law µ and one of the same-index pair laws above, with the conditional marginal domination already established.” · Show in paper