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 power saving for planar unit distances

There are absolute constants 0<C<∞0 < C < \infty and 1≤β<4/31 \le \beta < 4/3 such that u(n)≤Cnβu(n) \le C n^{\beta} for every nonnegative integer nn (equivalently, u(n)=O(n4/3−δ)u(n) = O(n^{4/3 - \delta}) for an absolute δ>0\delta > 0).

From a sequence of n=t3n = t^3 real planar points with at least t4−o(1)t^{4-o(1)} ordered unit pairs, one can obtain real algebraic point and center sets P,QP, Q and a nonempty unit-incidence graph E⊆P×QE \subseteq P \times Q with parameter a=a(t)>0a = a(t) > 0 (a→0a \to 0, alog⁡t→∞a \log t \to \infty), sizes ∣P∣=t1+2a+o(a)|P| = t^{1+2a+o(a)}, ∣Q∣=t2+a+o(a)|Q| = t^{2+a+o(a)}, ∣E∣=t2+2a+o(a)|E| = t^{2+2a+o(a)}, uniform degrees d(p)=t1+o(a)d(p) = t^{1+o(a)}, d(q)=ta+o(a)d(q) = t^{a+o(a)}, lower marginal bounds μ(p)≥γ/∣P∣\mu(p) \ge \gamma/|P|, κ(q)≥γ/∣Q∣\kappa(q) \ge \gamma/|Q| (γ=27/50\gamma = 27/50), and uniform expansion PE(p∈F,q∈G)≤uv((uv)η+ξt)\mathbb{P}_E(p \in F, q \in G) \le \sqrt{uv}((uv)^{\eta} + \xi_t) with η=1/25\eta = 1/25 and ξt→0\xi_t \to 0.

Under the wedge law of Proposition 2.4, the mutual information of the point endpoints satisfies D:=DKL(L(p,r)∥μ⊗μ)=o(log⁡t)D := D_{\mathrm{KL}}(\mathcal{L}(p, r) \| \mu \otimes \mu) = o(\log t).

For a finite point-center law ω\omega with wedge mutual information DD, and K,M≥1K, M \ge 1 with independent tests T1,…,TKT_1, \dots, T_K having feature maps ∥fT(c)∥≤B\|f_T(c)\| \le B and coefficient maps ∥gT(q)∥≤B′\|g_T(q)\| \le B', choosing J∈[K]J \in [K] uniformly and retaining the strict prefix H(p)=(cT1(p),…,cTJ−1(p))H(p) = (c_{T_1}(p), \dots, c_{T_{J-1}}(p)) yields EJ,T1,…,TJ−1,T∣E(p,q)∼ω⟨fT(cT(p))−mT,H(p),gT(q)⟩∣≤4BB′(D/K)1/4\mathbb{E}_{J, T_1, \dots, T_{J-1}, T} |\mathbb{E}_{(p,q)\sim\omega} \langle f_T(c_T(p)) - m_{T, H(p)}, g_T(q) \rangle| \le 4BB'(D/K)^{1/4}, so a history with at most MK−1M^{K-1} values can be fixed before drawing the fresh test TT.

For a sequence of the incidence pieces in Proposition 2.4, with the level spaces and quantities SS in (4.8), no subsequence has SS bounded; consequently S→∞S \to \infty.

For the center-star and point-template-star experiments of Section 5.2, the common-level variations satisfy CQ+CPθ=o(S)C_Q + C_{P\theta} = o(S), and their rare exception means satisfy EκHq∗≤EμHp∗+o(S)\mathbb{E}_\kappa H_q^* \le \mathbb{E}_\mu H_p^* + o(S).

The good rare point exceptions satisfy lim inf⁡t→∞EμHp∗S>0\liminf_{t \to \infty} \frac{\mathbb{E}_\mu H_p^*}{S} > 0.

There are J→∞J \to \infty, ωt→0\omega_t \to 0, and a simple bipartite graph E\mathcal{E} between two copies of PP (each of size at most t1+o(1)t^{1+o(1)}) with at least t2−o(1)t^{2-o(1)} edges, where each edge (p,r)(p, r) has distinct endpoints and an assigned center q(p,r)∈Qq(p, r) \in Q with pq,rq∈Epq, rq \in E, no edge has both endpoints on any real unit circle containing at least t9/10t^{9/10} points of PP, and uniformly over edges and complete rectangles (zp−zr)(wp−wr)=2−λpr−λpr−1(z_p - z_r)(w_p - w_r) = 2 - \lambda_{pr} - \lambda_{pr}^{-1}, h(λpr)≥J/3h(\lambda_{pr}) \ge J/3, and h(λprλp′r′λpr′λp′r)≤ωtJh\left(\frac{\lambda_{pr}\lambda_{p'r'}}{\lambda_{pr'}\lambda_{p'r}}\right) \le \omega_t J.

There is no sequence of finite algebraic point sets Pn⊂R2P_n \subset \mathbb{R}^2 with ∣Pn∣≤tn1+o(1)|P_n| \le t_n^{1+o(1)}, bipartite graphs Gn⊂Pn×PnG_n \subset P_n \times P_n with at least tn2−o(1)t_n^{2-o(1)} edges, and nonzero algebraic numbers λpr\lambda_{pr} satisfying (zp−zr)(wp−wr)=2−λpr−λpr−1(z_p - z_r)(w_p - w_r) = 2 - \lambda_{pr} - \lambda_{pr}^{-1}, h(λpr)≥cJnh(\lambda_{pr}) \ge c J_n, h(λprλp′r′λpr′λp′r)=o(Jn)h\left(\frac{\lambda_{pr}\lambda_{p'r'}}{\lambda_{pr'}\lambda_{p'r}}\right) = o(J_n) uniformly over edges and complete rectangles, and no edge having both endpoints on a unit circle containing at least tn9/10t_n^{9/10} points of PnP_n.

Methods

Algebraic specialization of finite planar point configurations using the transfer principle for real closed fields to obtain real algebraic coordinates while preserving all specified unit-distance equations and point distinctness.

Source excerpts

“This finite system has rational coefficients and a real solution, namely the original configuration. The transfer principle for real closed fields [17, Theorems 3.1 and 4.1] gives a solution in the field of real algebraic numbers.” · Show in paper

Random circle cuttings (Clarkson-Shor framework) combined with degree trimming and maximizing w/(uv)γw/(uv)^\gamma over induced rectangles to obtain an unbalanced incidence graph with controlled degrees and uniform expansion.

Source excerpts

“The cutting proof uses the random-sampling principle for regions with bounded defining data developed by Clarkson [5, Section 4].” · Show in paper

“Among all induced rectangles with counting fractions u, v ≥ τ and captured edge fraction w, maximize w/(uv)γ , γ = 27/50.” · Show in paper

Short-history prediction from wedge mutual information using the relative entropy chain rule, Pinsker's inequality, and conditional-copy resampling.

Source excerpts

“We use the standard chain rule for relative entropy [8, Theorem 2.5.3]:” · Show in paper

“Everything proved so far holds after averaging the initially frozen tests. The residual at step j uses only T1, . . . , Tj .” · Show in paper

Weighted product formula and absolute height of algebraic numbers over finite normal number fields, combined with threshold level decomposition into common, damaged, and good rare levels.

Source excerpts

“Use the classification of number-field places and the product formula [19, Theorems 7.14–7.15].” · Show in paper

“Define the common, damaged, and good rare levels, respectively, by C = {f ≥ ϵ}, B = {f < ϵ, m > 1/k}, G = {f < ϵ, m ≤ 1/k}.” · Show in paper

Ultraproduct construction with a negligible-height subfield k0k_0, k0k_0-derivations, separate odd valuations, and quadratic Kummer theory to rule out multiplicative rectangles.

Source excerpts

“Fix a nonprincipal ultrafilter u on N: a collection of subsets closed under finite intersection and supersets, containing the cofinite sets and exactly one part of every finite partition.” · Show in paper

“We use the quadratic case of Kummer theory over the characteristic-zero field M.” · Show in paper