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 Mahler Conjecture for General Convex Bodies

For every integer n≥1n \ge 1 and every convex body K⊂RnK \subset \mathbb{R}^n, the volume product satisfies P(K)=∣K∣ ∣(K−s(K))∘∣≥(n+1)n+1(n!)2P(K) = |K|\,|(K - s(K))^\circ| \ge \frac{(n+1)^{n+1}}{(n!)^2}, with equality if and only if KK is a simplex.

For every integer n≥1n \ge 1 and every proper lower-semicontinuous convex function φ:Rn→R∪{+∞}\varphi : \mathbb{R}^n \to \mathbb{R} \cup \{+\infty\} such that 0<∫Rne−φ(x) dx<∞0 < \int_{\mathbb{R}^n} e^{-\varphi(x)}\,dx < \infty, (∫Rne−φ(x) dx)(∫Rne−φ∗(y) dy)≥en\left(\int_{\mathbb{R}^n} e^{-\varphi(x)}\,dx\right)\left(\int_{\mathbb{R}^n} e^{-\varphi^*(y)}\,dy\right) \ge e^n.

For every integer n≥1n \ge 1, if ηi(dx)=e−Vi(x) dx\eta_i(dx) = e^{-V_i(x)}\,dx (i=1,2i=1,2) are full-dimensional log-concave probability measures with proper lower-semicontinuous convex potentials ViV_i whose densities are essentially continuous, and their moment measures νi=(∇Vi)#ηi\nu_i = (\nabla V_i)_{\#}\eta_i have finite first moments with finite H(ηi)=−∫Vi dηiH(\eta_i) = -\int V_i\,d\eta_i, then H(η1)+H(η2)≤−3n+T(ν1,ν2)H(\eta_1) + H(\eta_2) \le -3n + \mathcal{T}(\nu_1, \nu_2).

A linear transformation of the cone data and a symmetric matrix TT in [t−I,t+I][t_- I, t_+ I] can be chosen so that EA=0\mathbb{E}A = 0, λT2+C∘T+K=0\lambda T^2 + \mathbf{C} \circ T + \mathbf{K} = 0, and H:=−C−2λT≥m∗I\mathbf{H} := -\mathbf{C} - 2\lambda T \ge m_* I, where C=EC(L)\mathbf{C} = \mathbb{E}\mathsf{C}(L) and K=EK(L)\mathbf{K} = \mathbb{E}\mathsf{K}(L).

For the cone data and fixed projection field constructed in Section 3, 1mlog⁡(χC(V)χD(U))≥τ(log⁡Σ−T)+τTHv−δ[d]+18N=−E\frac{1}{m}\log(\chi_C(V)\chi_D(U)) \ge \tau(\log\Sigma - T) + \tau_T H_v - \delta[d] + \frac{1}{8}N = -\mathcal{E}.

For the normalized field satisfying (8) with t−I≤T≤t+It_- I \le T \le t_+ I and the fixed profiles satisfying (2)–(4) and C≤−0.341\mathsf{C} \le -0.341, Δp−JΣ(u)+0.80∥[L,T]∥22≤−1.94τT2\Delta_{\mathrm{p}} - J_\Sigma(u) + 0.80\|[L, T]\|_2^2 \le -1.94\tau T^2, and consequently E≤−0.14τT2−∫(W−W0) dβ≤0\mathcal{E} \le -0.14\tau T^2 - \int (W - W_0)\,d\beta \le 0, with E=0\mathcal{E} = 0 implying T=0T = 0, ν\nu supported on the diagonal, and β=0\beta = 0.

The fixed scalar profiles and exact parameters satisfy the pointwise, two-variable, and strict segment inequalities in Proposition 2.3 (Equations (3) and (4)).

Methods

Cone/Laplace reduction expressing the volume product of a convex body K⊂RnK \subset \mathbb{R}^n as a product of Laplace integrals χC(V)χD(U)\chi_C(V)\chi_D(U) on dual cones in Rn+1\mathbb{R}^{n+1}.

Source excerpts

“Lemma 2.1 (Cone reduction). For U = (0, m) and V = (0, 1), χ_C(V) χ_D(U) = ((n!)^2 / m^m) |K| |(K - z_0)^◦|.” · Show in paper

Biased Gaussian nearest-point projections onto dual cones via Moreau's decomposition, paired with simultaneous normalization of linear coordinates and Gaussian covariance using Brouwer's fixed point theorem and the Löwner–Heinz inequality.

Source excerpts

“Lemma 3.1 (Biased projections). For every z ∈ R there is a unique vector ξ_z such that X_z = Π_C(Z + ξ_z), Y_z = Π_D(-Z - ξ_z), E X_z = a(z)U.” · Show in paper

“Proposition 3.4 (Simultaneous normalization). A linear transformation of the cone data and a symmetric T in the prescribed spectral box can be chosen so that E A = 0, λ T^2 + C ◦ T + K = 0, H := -C - 2λ T ≥ m_* I” · Show in paper

Truncated, Gaussian-smoothed cone-valued projection maps with change of variables and conditional Jensen's inequality to lower bound the log Laplace product without boundary regularity assumptions.

Source excerpts

“For integers j ≥ 1, define maps of G by TX,jT_{X,j}(G) = ∫_{-j}^j W_X(z) EZ′E_{Z'} X_z(\tilde{Z}) dz, TY,jT_{Y,j}(G) = ∫_{-j}^j W_Y(z) EZ′E_{Z'} Y_z(\tilde{Z}) dz.” · Show in paper

Gaussian Ornstein–Uhlenbeck inverse-generator covariance representation and Hermite degree separation comparing nonlinear projection-derivative fields to a linear Gaussian matrix L=∑i=1mGiMiL = \sum_{i=1}^m G_i M_i.

Source excerpts

“Lemma 5.1 (Gaussian covariance formula). Let X, Y be centered elements of the Gaussian Sobolev space W1,2W^{1,2}(R^m; R^m). Then E(X · Y) = E ∑_{i=1}^m (∂_{G_i} X) · K_G(∂_{G_i} Y).” · Show in paper

Daleckiĭ–Kreĭn divided-difference spectral averaging for matrix functions of a linear Gaussian matrix LL.

Source excerpts

“This is the Daleckiĭ–Kreĭn divided-difference derivative formula; see [23, Theorem 2.11].” · Show in paper

Exact rational interval arithmetic, analytic strip interpolation from second differences, uniform tail derivative bounds, and Bernstein-basis polynomial certificates to prove dimension-independent scalar inequalities.

Source excerpts

“The proof separates three tasks: exact arithmetic at finitely many nodes, analytic interpolation between those nodes, and estimates on the two unbounded tails.” · Show in paper

“Both polynomials are positive on [.16, .54], as the following finite rational certificate shows.” · Show in paper