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.

Latest version

Checks complete

Sync OpenAI Math manuscripts from 4eaee87548250b2682affda7d05d947c05554bae

Committed

All papers

Claims

Findings

Catalan's constant is irrational

Catalan's constant G=∑j≥0(−1)j/(2j+1)2G = \sum_{j \ge 0} (-1)^j / (2j+1)^2 is irrational.

The minimum volume of an orientable complete finite-volume hyperbolic three-manifold with exactly two cusps (attained by the Whitehead-link and (−2,3,8)(-2,3,8)-pretzel-link complements) is irrational, and more generally every orientable arithmetic hyperbolic three-orbifold defined over Q(i)\mathbb{Q}(i) has irrational volume.

Each raw entry Fj(r)F_j(r) lies in Q+QG\mathbb{Q} + \mathbb{Q}G, and consequently if G∈QG \in \mathbb{Q}, then ΔN∈Q\Delta_N \in \mathbb{Q} for every positive integer NN.

Assuming G∈QG \in \mathbb{Q}, along any sequence of positive integers N→∞N \to \infty for which ΔN≠0\Delta_N \neq 0, lim inf⁡N→∞(log⁡∣ΔN∣n2−12log⁡2)≥−86094608−(12+5054608)log⁡2>−2.29084\liminf_{N \to \infty} \left(\frac{\log |\Delta_N|}{n^2} - \frac{1}{2}\log 2\right) \ge -\frac{8609}{4608} - \left(\frac{1}{2} + \frac{505}{4608}\right)\log 2 > -2.29084.

With δ=(q+g+h)/n=5/24\delta = (q+g+h)/n = 5/24, for every positive integer NN and assuming G∈QG \in \mathbb{Q}, v2(ΔN)≥−(δ2+δ28)n2−OG(nlog⁡(n+2))=−5054608n2−OG(nlog⁡(n+2))v_2(\Delta_N) \ge -\left(\frac{\delta}{2} + \frac{\delta^2}{8}\right)n^2 - O_G(n \log(n+2)) = -\frac{505}{4608}n^2 - O_G(n \log(n+2)).

Assuming G∈QG \in \mathbb{Q}, for every sufficiently large prime pp, the determinant with scale parameter N=pN = p satisfies vp(Δp)=−96pv_p(\Delta_p) = -96p, and in particular Δp≠0\Delta_p \neq 0.

Unconditionally, for the determinants ΔN\Delta_N of Equation (7), lim sup⁡N→∞(log⁡∣ΔN∣n2−12log⁡2)≤−2.290939875<−2.2909\limsup_{N \to \infty} \left(\frac{\log |\Delta_N|}{n^2} - \frac{1}{2}\log 2\right) \le -2.290939875 < -2.2909.

For any admissible trial sequences p,vp, v and λ=0\lambda = 0 (when κ=2\kappa = 2) or λ≥0\lambda \ge 0 (when κ=1\kappa = 1), lim sup⁡N→∞sup⁡case κ(log⁡Iκ,nn2−12log⁡2)\limsup_{N\to\infty} \sup_{\text{case }\kappa} \left(\frac{\log \mathcal{I}_{\kappa,n}}{n^2} - \frac{1}{2}\log 2\right) is bounded above by (−1+α+γ)log⁡2+κ∥p∥∗2+12∥v∥∗2+sup⁡−1≤x<1,x≠0{Wκ(x)+λD(x)−2κT(p,x)−S(v,x)}+sup⁡0<s<1{Ws(s)+2T(v,s)}(-1 + \alpha + \gamma)\log 2 + \kappa\|p\|_*^2 + \frac{1}{2}\|v\|_*^2 + \sup_{-1\le x<1, x\neq 0}\{W_\kappa(x) + \lambda D(x) - 2\kappa T(p,x) - S(v,x)\} + \sup_{0<s<1}\{W_s(s) + 2T(v,s)\}.

Methods

Construct determinants Delta_N of size n = 48N combining moments M(P_r, sb+ks^{b+k}(1-s)^q) - (3/2)Z(D_r, sb+ks^{b+k}(1-s)^q) with integral Chebyshev polynomial rows P_r, D_r having high Taylor contact tP_r/f - D_r = O(t^L) to cancel zeta(2).

Source excerpts

“Pr = (1 − t)^h tC−1t^{C−1} Td(1/t), Dr = sgn(r − g)(1 − t)^h tC−1t^{C−1} Ud−1(1/t).” · Show in paper

“∆N = det_{0≤r,k<n} (M(Pr, sb+ks^{b+k}(1 − s)^q) − 3/2 Z(Dr, sb+ks^{b+k}(1 − s)^q)).” · Show in paper

Apply p-adic digit reductions to rational moments M^0(i,j) and Z^0(i,j) and perform invertible column eliminations over Z_p with Cauchy-Binet transfer to bound v_p(Delta_N) for odd primes p <= H.

Source excerpts

“p^2 M^0(i, j) ≡ Ed M^0(i', j') (mod p), p^2 Z^0(i, j) ≡ {Z^0(⌊i/p⌋, j'), i ≡ j (mod p); 0, i ̸≡ j (mod p).” · Show in paper

“If Q ∈ GL_{n+q}(Zp) and every full minor of TQ has valuation at least −D, the same holds for every full minor of T = (TQ)Q−1Q^{−1} by Cauchy–Binet.” · Show in paper

Use Frobenius and two palindromic bases in F_p[w] at prime scale N = p to factor det(p^2 F_p) mod p into determinants of three fixed 48x48 rational matrices B_0, B_0 + B_1, and B_0 - B_1, certified invertible modulo 101.

Source excerpts

“det Lp = (det a)^{48} det B0 · det(B0 + B1)^{(p−1)/2} det(B0 − B1)^{(p−1)/2}.” · Show in paper

Apply Andréief's determinant integration identity and Cauchy's double alternant, then bound the mixed-sheet evaluation determinant R_n(x) uniformly using a finite-dimensional Hardy-space (H^2) interpolation operator on one domain and Hadamard's inequality on the complement.

Source excerpts

“We apply Andréief’s determinant integration identity twice [10, Equation (1.7) and Section 2.2].” · Show in paper

“J = ΠQ Mh∗M_{h∗}|_{HQ}, L = I + JR satisfy ∥L∥ ≤ 1 + K0n.” · Show in paper

Regularize logarithmic Vandermonde and cross-kernel interactions using a damped Chebyshev/Haagerup expansion to complete squares and bound the many-variable energy by one-variable suprema.

Source excerpts

“Fix 0 < ε < 1/8 and define τ = 1 − ε, ρ(x) = 1 − ε(1 − x), σ(s) = 1 − ε√(1 − s).” · Show in paper

Certify global suprema of the one-variable barrier functions using Descartes' rule of signs to isolate all stationary points into disjoint rational brackets and rational Taylor-series interval arithmetic for logarithms and arctangents.

Source excerpts

“The sign-variation bound is Descartes’ rule of signs [7, Book III, p. 373].” · Show in paper

“This fully specifies a rational interval calculation; no numerical logarithms are needed.” · Show in paper