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

Local smoothing for the Euclidean wave equation

For every spatial dimension n≥2n \ge 2, 2≤p<∞2 \le p < \infty, and ε>0\varepsilon > 0, the half-wave propagator U(T)=e−iT∣D∣U(T) = e^{-iT|D|} satisfies ∥U(T)f∥Lp(Rn×I)≤Cn,p,ε∥Js+(p)+εf∥Lp(Rn)\|U(T)f\|_{L^p(\mathbb{R}^n \times I)} \le C_{n,p,\varepsilon} \|J^{s_+(p)+\varepsilon}f\|_{L^p(\mathbb{R}^n)} on I=[1,2]I = [1,2], where J=(1−Δ)1/2J = (1-\Delta)^{1/2} and s+(p)=max⁡{0,(n−1)/2−n/p}s_+(p) = \max\{0, (n-1)/2 - n/p\}, with ε=0\varepsilon = 0 allowed at p=2p=2.

For n≥2n \ge 2, p=pc=2n/(n−1)p = p_c = 2n/(n-1), φ∈Cc∞(Rn∖{0})\varphi \in C_c^\infty(\mathbb{R}^n \setminus \{0\}), and N≥2N \ge 2 with PN=φ(D/N)P_N = \varphi(D/N), ∥U(T)PNf∥Lp(Rn×I)≤Cn,φ[log⁡(2+N)]2/p∥f∥Lp(Rn)\|U(T)P_N f\|_{L^p(\mathbb{R}^n \times I)} \le C_{n,\varphi}[\log(2+N)]^{2/p}\|f\|_{L^p(\mathbb{R}^n)}.

For n≥2n \ge 2, R≥2R \ge 2, z∈Rnz \in \mathbb{R}^n, and 0<δ≤10 < \delta \le 1, the spherical extension operator EE and Kakeya maximal operator KδK_\delta satisfy ∥Eg∥Lpc(B(z,R))≤Cn[log⁡(2+R)]2/pc∥g∥Lpc(Sn−1)\|Eg\|_{L^{p_c}(B(z,R))} \le C_n[\log(2+R)]^{2/p_c}\|g\|_{L^{p_c}(S^{n-1})} and ∥Kδf∥Ln(Sn−1)≤Cn[log⁡(2/δ)]4/pc∥f∥Ln(Rn)\|K_\delta f\|_{L^n(S^{n-1})} \le C_n[\log(2/\delta)]^{4/p_c}\|f\|_{L^n(\mathbb{R}^n)}, and every subset of Rn\mathbb{R}^n containing a unit line segment in each direction has Hausdorff dimension nn.

For 1<a≤∞1 < a \le \infty and b<∞b < \infty satisfying b>pcb > p_c and b>n+1n−1a′b > \frac{n+1}{n-1}a', the global Fourier extension operators for the sphere Sn−1S^{n-1} and the truncated cone Σ\Sigma satisfy ∥Eg∥Lb(Rn)≲n,a,b∥g∥La(Sn−1)\|Eg\|_{L^b(\mathbb{R}^n)} \lesssim_{n,a,b} \|g\|_{L^a(S^{n-1})} and ∥EΣG∥Lb(Rn+1)≲n,a,b∥G∥La(Σ)\|E_\Sigma G\|_{L^b(\mathbb{R}^{n+1})} \lesssim_{n,a,b} \|G\|_{L^a(\Sigma)}.

For n≥2n \ge 2 and 1≤p≤∞1 \le p \le \infty, the Bochner–Riesz means BrαB_r^\alpha satisfy sup⁡r>0∥Brαf∥Lp(Rn)≤Cn,p,α∥f∥Lp(Rn)\sup_{r>0} \|B_r^\alpha f\|_{L^p(\mathbb{R}^n)} \le C_{n,p,\alpha}\|f\|_{L^p(\mathbb{R}^n)} whenever α>δ(p)=max⁡{0,n∣1/p−1/2∣−1/2}\alpha > \delta(p) = \max\{0, n|1/p-1/2|-1/2\} (with LpL^p convergence for p<∞p < \infty), and for 2≤p<∞2 \le p < \infty and α>s+(p)\alpha > s_+(p) the maximal operator satisfies ∥B∗αf∥Lp(Rn)≤Cn,p,α∥f∥Lp(Rn)\|B_*^\alpha f\|_{L^p(\mathbb{R}^n)} \le C_{n,p,\alpha}\|f\|_{L^p(\mathbb{R}^n)} with Brαf→fB_r^\alpha f \to f almost everywhere.

For n≥2n \ge 2, 2≤p<∞2 \le p < \infty, and α>max⁡{12,n(12−1p)}\alpha > \max\{\frac{1}{2}, n(\frac{1}{2}-\frac{1}{p})\}, Stein's square function Gα\mathcal{G}_\alpha satisfies ∥Gαf∥p≤Cn,p,α∥f∥p\|\mathcal{G}_\alpha f\|_p \le C_{n,p,\alpha}\|f\|_p for f∈S(Rn)f \in \mathcal{S}(\mathbb{R}^n) and extends to Lp(Rn)L^p(\mathbb{R}^n).

For d≥1d \ge 1 and pc=2(d+1)/dp_c = 2(d+1)/d, the compact paraboloid extension operator EE satisfies ∥Eg∥Lb(Rd+1)≲d,a,b∥g∥La(Bd)\|Eg\|_{L^b(\mathbb{R}^{d+1})} \lesssim_{d,a,b} \|g\|_{L^a(B_d)} for 1<a≤∞1 < a \le \infty, pc<b<∞p_c < b < \infty, b>(d+2)a′/db > (d+2)a'/d; the Schrödinger propagator satisfies ∥eitΔf∥Lpc(Rd×I)≲d,I[log⁡(2+N)]2/pc∥f∥pc\|e^{it\Delta}f\|_{L^{p_c}(\mathbb{R}^d \times I)} \lesssim_{d,I} [\log(2+N)]^{2/p_c}\|f\|_{p_c} for supp⁡f^⊂B(0,N)\operatorname{supp}\widehat{f} \subset B(0,N); and for pc<p<∞p_c < p < \infty, p≤r≤∞p \le r \le \infty, ∥eitΔf∥Lp(Rd;Lr(I))≲d,p,r,I∥f∥Bp,ps(Rd)≲d,p,s∥(1−Δ)s/2f∥p\|e^{it\Delta}f\|_{L^p(\mathbb{R}^d;L^r(I))} \lesssim_{d,p,r,I} \|f\|_{B^s_{p,p}(\mathbb{R}^d)} \lesssim_{d,p,s} \|(1-\Delta)^{s/2}f\|_p with s=d(1−2/p)−2/rs = d(1-2/p)-2/r.

For real symmetric n×nn \times n matrices C1,…,CnC_1,\dots,C_n (n≥2n \ge 2) and C(ξ)=∑j=1nξjCjC(\xi) = \sum_{j=1}^n \xi_j C_j, if ϕ(ξ)=λmax⁡(C(ξ))\phi(\xi) = \lambda_{\max}(C(\xi)) is simple on an open cone Γ⊂Rn∖{0}\Gamma \subset \mathbb{R}^n \setminus \{0\} with D2ϕ(ξ)D^2\phi(\xi) positive definite on ξ⊥\xi^\perp, then for b∈Cc∞(Γ)b \in C_c^\infty(\Gamma), 0<h≤10 < h \le 1, and Schwartz ff, ∥e−iTϕ(D)b(hD)f∥Lpc(Rn×[1,2])≤Cb[log⁡(2/h)]2/pc∥f∥Lpc(Rn)\|e^{-iT\phi(D)}b(hD)f\|_{L^{p_c}(\mathbb{R}^n \times [1,2])} \le C_b[\log(2/h)]^{2/p_c}\|f\|_{L^{p_c}(\mathbb{R}^n)} (and similarly for a simple smallest eigenvalue with negative definite transverse Hessian).

Methods

Construct a positive semidefinite divergence-free quaternionic wave current A0=χχ∗A^0 = \chi\chi^* for a superposition of waves via analytic continuation along two auxiliary flows and a holomorphic flux map Θ\Theta.

Source excerpts

“Lemma 2.1. Let P = P µ Cµ∂Xµ have constant real symmetric coefficients. For a holomorphic solution Pfsys = 0 and the real displacement (2.3), put, wherever defined, G = fsys(X − iq), Vℓa = X µ (Cµwℓ)a ∂Xµ , Dℓa = ∂wℓa + iVℓa, Zℓ = X a GaDℓa. In the real coordinates (X, w1, w2), the tensor B = Re X 2 ℓ=1 Zℓ ⊗ Zℓ (2.5) is symmetric and positive semidefinite, and each of its columns has zero divergence.” · Show in paper

“Proposition 2.2 (The wave current). For any finite superposition (2.1), set χ = X κ,v V v ιe(uκ,v) mκ,v, A0 = χχ∗. (2.14) On Y ̸= 0 the matrix A0 is smooth and positive semidefinite, (∂Φ)A0 = 0, and each column of (A0)R has zero real divergence.” · Show in paper

Pair the regularized wave current with a quaternionic Monge–Ampère Dirichlet potential uu solving det⁡Hfib(λ∣Φ∣2+u)=λμ\det H_{\mathrm{fib}}(\lambda|\Phi|^2 + u) = \lambda\mu on moving spatial fibers, using an Aubin–Mabuchi-type segment energy F(z)=Jz(u)\mathcal{F}(z) = \mathcal{J}_z(u) whose Laplacian is bounded by the boundary wave flux.

Source excerpts

“On each spatial fiber there is a unique smooth real u satisfying Du = Hfib(B0 + u) > 0, det Du = λµ, u|∂Ωz = 0. (4.8)” · Show in paper

“Proposition 4.2 (Energy balance). The energy F is smooth on a < |τ | < 4. Let ρ denote the boundary value of u/r, which extends smoothly from the interior. In the three variables z = (T, τ ) one has ∆zF(z) + Z Ωz A : Σ dζ = Z ρ G(r)∗AwG(r) δ(r) dζ. (4.12)” · Show in paper

Propagate the energy estimate across a thin cylinder b≤∣τ∣≤3b \le |\tau| \le 3 using radial comparison solutions of (−ΔT,τ+V)H≥0(-\Delta_{T,\tau} + V)H \ge 0 and the maximum principle.

Source excerpts

“Lemma 6.1. Let 0 < b < 1/4 and D = log(3/b). Suppose a continuous potential V : [b, 3] → [0, ∞) satisfies Z 3 b s log(s/b)V (s) ds ≤ M, sup 1/2≤s≤3/2 V (s) ≤ M. (6.1) Let E ≥ 0 be smooth on RT × {τ ∈ R2 : b ≤ |τ | ≤ 3}, with ∆T,τE ≤ V (|τ |)E in the interior. Its circular mean H(T, s) = 1 2π Z 2π 0 E ( T, s(cos θ,sin θ) ) dθ satisfies, for b ≤ r ≤ 1, H(0, r) ≥ c(M) log(r/b) D2 Z 2 1 H(T, b) dT. (6.2)” · Show in paper

Remove the spatial period by tiling Rn\mathbb{R}^n with unit cubes, periodizing localized data on concentric cubes, and bounding far-field contributions via integration by parts on the wave kernel.

Source excerpts

“Tile Rn by unit cubes Q, and write 10Q for the concentric cube of side ten. On Q the contribution of f1(10Q)c is bounded pointwise by CMhM((1 + | · |)−M ∗ |f|)(X), since |X − Y | ≥ 9/2 for X ∈ Q and Y /∈ 10Q. This single function bounds the far contributions on every cube.” · Show in paper