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

A positive resolution of De Giorgi's conjecture in dimension eight

Every entire C2C^2 solution u:R8→(−1,1)u : \mathbb{R}^8 \to (-1, 1) of Δu=u3−u\Delta u = u^3 - u with ∂8u>0\partial_8 u > 0 in R8\mathbb{R}^8 has the form u(x)=tanh⁡(e⋅x−c2)u(x) = \tanh\left(\frac{e \cdot x - c}{\sqrt{2}}\right) for some e∈S7e \in \mathbb{S}^7 with e8>0e_8 > 0 and c∈Rc \in \mathbb{R}.

Every stable solution v∈C2(R7,[−1,1])v \in C^2(\mathbb{R}^7, [-1, 1]) of Δv=v3−v\Delta v = v^3 - v is either v≡1v \equiv 1, v≡−1v \equiv -1, or v(x)=tanh⁡(a⋅x−b2)v(x) = \tanh\left(\frac{a \cdot x - b}{\sqrt{2}}\right) for some a∈S6a \in \mathbb{S}^6 and b∈Rb \in \mathbb{R}, with no energy-growth assumption.

For 4≤n≤74 \le n \le 7, if v∈C2(Rn,[−1,1])v \in C^2(\mathbb{R}^n, [-1, 1]) solves Δv=v3−v\Delta v = v^3 - v, is stable, and satisfies M∞<∞M_\infty < \infty, then v≡1v \equiv 1, v≡−1v \equiv -1, or v(x)=g(a⋅x−b)v(x) = g(a \cdot x - b) for some a∈Sn−1a \in \mathbb{S}^{n-1} and b∈Rb \in \mathbb{R}.

For every c>1/(22)c > 1/(2\sqrt{2}) there exist P=P(c)<∞P = P(c) < \infty and δ=δ(n,c)∈(0,1/2]\delta = \delta(n, c) \in (0, 1/2] such that every entire solution and every zero zz with 0<D(z)<δ0 < D(z) < \delta satisfies ∣H(z)∣≤cσ∫−PPq(z+sν(z))3D(z+sν(z)) ds|H(z)| \le \frac{c}{\sigma} \int_{-P}^P q(z + s\nu(z))^3 D(z + s\nu(z)) \, ds.

In the seven-dimensional stable setting with unbounded energy density, the penalized reach dd is locally semiconcave on Σreg\Sigma_{\mathrm{reg}} with nonpositive singular surface Laplacian, satisfies $| \nabla_\Sigma d| \le C_\Gamma \exp\{C_\Gamma \sqrt{\log(K(d)+1)}\}$ almost everywhere, and obeys the weak reach inequalities (7.22)–(7.23) with multiplicity-at-most-one positive-foot transfer (7.24).

In the seven-dimensional stable setting with unbounded energy density, for sufficiently close penalty parameters and appropriate fixed P,δ,R0,CP, \delta, R_0, C, the fourth-order growth bound YR:=∫Σgχ2(a4+D2+E2)+NR≤CK(R)Rm−4\mathcal{Y}_R := \int_{\Sigma_g} \chi^2(a^4 + D^2 + E^2) + N_R \le C K(R) R^{m-4} holds for every R≥R0R \ge R_0.

Methods

Inverse-profile formulation (q=1−v2q = 1 - v^2, U=g−1(v)U = g^{-1}(v), D=1−∣∇U∣2D = 1 - |\nabla U|^2) and the exact stability identity 0≤Qv(qη)=∫Rn(q2∣∇η∣2−q3Dη2) dx0 \le Q_v(q\eta) = \int_{\mathbb{R}^n} (q^2 |\nabla \eta|^2 - q^3 D \eta^2) \, dx.

Source excerpts

“Define the inverse-profile variables q = 1 − v^2, U = g−1g^{−1}(v), p = ∇U, T = ∇^2U, D = 1 − |p|^2. ... For a stable solution, use qη in Qv and integrate by parts with (5.6): 0 ≤ Qv(qη) = ∫_{ℝ^n} (q^2|∇η|^2 − q^3Dη^2) dx.” · Show in paper

Penalized reach d(z)=inf⁡y∈Σ∖{z}ℓΓ(s)d(z) = \inf_{y \in \Sigma \setminus \{z\}} \ell \Gamma(s) with a smooth angular penalty Γ(s)\Gamma(s) equal to 1/s1/s away from s=1s = 1 and constant G1G_1 near s=1s = 1.

Source excerpts

“On Σreg define d(z) = inf_{y∈Σ\{z}} ℓΓ(s), ℓ = |y − z|, e = (y − z)/ℓ, s = |ν(z) · e|.” · Show in paper

Pushing lifted surface tests to physical space by taking the square root of the sum of squares ξ=(∑jξj2)1/2\xi = (\sum_j \xi_j^2)^{1/2} of local inverse branches and retaining the nonnegative gradient variance V=min⁡b∈Rn∑j∣∇ξj−ξjb∣2\mathcal{V} = \min_{b \in \mathbb{R}^n} \sum_j |\nabla \xi_j - \xi_j b|^2.

Source excerpts

“For a compactly supported Lipschitz amplitude on the lifted tube, write ξj for its local inverse branches and put ξ = (∑_j ξ_j^2)^{1/2}. Then ξ is an admissible physical test. Its quadratic form is the lifted form minus ∫ 𝒱 dx, where the nonnegative variance density is 𝒱 = min_{b∈ℝ^n} ∑_j |∇ξj − ξj b|^2.” · Show in paper

Nearly linear gap test w=min⁡{T0,l+log⁡l}w = \min\{T_0, l + \log l\} integrated against an inward radial Green kernel ϕR\phi_R to compare sheet separation against mean curvature and Hardy defect bounds.

Source excerpts

“Fix a small tolerance δ > 0 and use ψ(l) = l + log l, w = min{T0, ψ(l)}.” · Show in paper

Radial ambient interpolation test η~(x)=U(t)ψ(zi(x))+(1−U(t))ψ(zj(x))\tilde{\eta}(x) = \mathcal{U}(t)\psi(z_i(x)) + (1 - \mathcal{U}(t))\psi(z_j(x)) across fixed-width middle bands using U(t)\mathcal{U}(t) built from sech⁡2(2u)\operatorname{sech}^2(\sqrt{2}u) to match the phase tail.

Source excerpts

“Fixing the half-width at Asw = 3, we therefore define 𝒰(t) = (∫_t^3 sech^2(√2u) du) / (∫_{-3}^3 sech^2(√2u) du), −3 ≤ t ≤ 3. ... In that band set η˜(x) = 𝒰(t)ψ(zi(x)) + (1 − 𝒰(t))ψ(zj(x)), t = (si − sj)/2.” · Show in paper