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 Weak Inhomogeneous Duffin–Schaeffer Conjecture

For every fixed γ∈R\gamma \in \mathbb{R} and finite-valued ψ:N→[0,∞)\psi : \mathbb{N} \to [0, \infty), if ∑q=1∞ϕ(q)qψ(q)=∞\sum_{q=1}^{\infty} \frac{\phi(q)}{q} \psi(q) = \infty, then for Lebesgue-almost every x∈Rx \in \mathbb{R}, the inequality ∥qx−γ∥<ψ(q)\|qx - \gamma\| < \psi(q) holds for infinitely many positive integers qq.

For fixed γ∈R\gamma \in \mathbb{R}, finite-valued ψ:N→[0,∞)\psi : \mathbb{N} \to [0, \infty), and a continuous nondecreasing function f:(0,∞)→(0,∞)f : (0, \infty) \to (0, \infty) with f(r)→0f(r) \to 0 as r↓0r \downarrow 0, f(r)/rf(r)/r monotone, and f(0)=0f(0) = 0, if ∑q=1∞ϕ(q)f(ψ(q)/q)=∞\sum_{q=1}^{\infty} \phi(q) f(\psi(q)/q) = \infty, then Hf(I∩Wγ(ψ))=Hf(I)\mathcal{H}^f(I \cap W_\gamma(\psi)) = \mathcal{H}^f(I) for every bounded interval II.

Assuming the selections and birth prescriptions of Sections 3 and 4 and Propositions 5.2–5.4, no measurable set of positive measure can avoid the original approximation intervals in a whole tail, which implies Theorem 1.1.

At determinant scale 0<η<10 < \eta < 1, for any sufficiently small fixed ϑ>0\vartheta > 0, there are u∈(1/2,1)u \in (1/2, 1), χ>0\chi > 0, and η0>0\eta_0 > 0 such that for η<η0\eta < \eta_0, either the normalized edge weight satisfies E≤η1+χR−1/10(FG)u\mathcal{E} \le \eta^{1+\chi} \mathcal{R}^{-1/10} (FG)^u, or R≤η−ϑ\mathcal{R} \le \eta^{-\vartheta} and a subfamily of edges has the structured form v=NU0a/bv = N U_0 a/b, w=NT0c/d1w = N T_0 c/d_1, [v,w]=NU0T0ac[v,w] = N U_0 T_0 a c with captured normalized edge weight at least ηϑD1D2\eta^\vartheta D_1 D_2.

For the retained centres of one type and colour, good masks, and birth prescriptions of Sections 3 and 4, the broad direct-pair sum is O(1)O(1) and the same-width one-law sum is O(W)+Ok0(W1+c)O(W) + O_{k_0}(W^{1+c}) for some fixed c>0c > 0, with an additional static discard on medium groups losing at most δW\delta W first mass.

Uniformly in the allowed prior table history and slot cap N0N_0, the sparse-step alignment sums for a fixed centre pair and bin are at most CΘije−cYτ(UT)CC \Theta_{ij} e^{-cY} \tau(UT)^C, and accounting classes are joined by nonexceptional tight equal-width pairs covered by deterministic danger trains, proving Proposition 5.3.

The cells and accounting classes can be constructed chronologically for every finite block, preserving current-core equivariance, cell spans and padding, inherited accounting classes without splitting, and old-cell separation, with non-clipping first-mass loss made arbitrarily small relative to WW.

At each first sharing of a component, omitting simultaneous-birth comparisons, the broad pair sum is O(1)O(1) and the same-width, same-remaining-mask clipping count in common new cells is O(W)+Ok0(W1+c)O(W) + O_{k_0}(W^{1+c}) for some fixed c>0c > 0, establishing Proposition 5.2 together with Proposition 7.1.

Methods

Finite-block reduction and dyadic width discretization reducing the full-measure divergence problem to arbitrarily late finite blocks of rows with totient-weighted mass in [w,2w][w, 2w].

Source excerpts

“Lemma 2.2. For every fixed 0 < w ≤ 1/2 there are arbitrarily late, pairwise disjoint finite sets B of positive-width rows such that w ≤ W := X v∈B ϕ(v)r(v) ≤ 2w. (2.1)” · Show in paper

Adjusted host assignment and density purge, assigning nonraw rows to popular multiples minimizing a logarithmic cost function to achieve both small-scale density bounds and band separation.

Source excerpts

“For a nonraw row v of width r, minimize over all original popular centres (D, r0) in the block the cost b + 10a + 10c+ + (−c)+, b = − log(Dr0), a = log v (v, D) , c = log r r0 , (3.3)” · Show in paper

Core-equivariant projected tables with fractional prime-matching updates, cap clipping, and virtual product marginals.

Source excerpts

“Fractionally match the residual-scaled plus weights to the minus weights in that cell. Concretely, use a finite nonnegative matrix whose row and column sums are bounded by the endpoint weights and whose total mass is the smaller of the two list loads.” · Show in paper

“Proposition 5.7 (Virtual product marginals). For each retained good row, there is a nonnegative virtual weight on its true numerator grid with the following properties.” · Show in paper

Selberg upper-bound sieve on determinant cycles, integer/prime rotation alternative, weighted LCM bound, and small-scale extraction.

Source excerpts

“Lemma 6.1 (Small-prime exclusions). For every fixed C > 0 there is an absolute sufficiently small ξ > 0 with the following property.” · Show in paper

“Lemma 6.2 (Rotation alternative). There are absolute constants C1 ≫ C2 ≫ 1 such that, for all sufficiently large u, the following holds.” · Show in paper

Deterministic danger trains and chronological jittered partitions preserving inherited accounting classes and separating old cells.

Source excerpts

“Construction 8.2 (Deterministic danger trains). For a centre i, current prime q in bin Y , and processed deficit d such that dq is the processed portion of some good mask, put V ′ = Li/(dq) and use a list of residues y0 (mod 1) of size at most ZY = exp(exp(haY )).” · Show in paper

“Choose one independent uniform candidate boundary point, called a raw line, in each microinterval, and replicate the configuration with period 1/Q. Skip a line if it belongs to any closed inherited hull of the preliminary survivors.” · Show in paper