OctoInformBeta

OpenAI preprint, September 2026 · §5. All normalized scales and all centers

§5 (completion of Lemma 2.1): open enlargement, expectation, and repair: Periodic enlargement and exceptional centers give density at most 6p6p for a positive parameter pp and all normalized scales

Repairing random blockers into open hitting sets.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

A periodic blocker is tested against the points x+t2−nx+t2^{-n}, where the center xx is arbitrary, the normalized scale tt belongs to [1,2][1,2], and nn is a natural number with n≥1n\geq1. Exceptional centers are repaired by an open periodic set.

A unit-periodic set A⊆ℝA\subseteq\mathbb R is a set satisfying A+1=AA+1=A. The density ρ(A)\rho(A) of a unit-periodic set A⊆ℝA\subseteq\mathbb R is its Lebesgue measure in one period, as in (1). ρ(A)=|A∩[0,1]|.\rho(A)=\lvert A\cap[0,1]\rvert .(1)

For B⊆ℝB\subseteq\mathbb R, N⊆ℕN\subseteq\mathbb N and a real sequence aa, the exceptional set R(B;N,a)R(B;N,a) consists of the centers at which some normalized scale misses BB at every index in NN.✦ The dyadic notation RN(B)R_N(B) is specified in (2). R(B;N,a)={x∈ℝ:∃t∈[1,2]x+tan∉B for every n∈N},RN(B)=R(B;N,(2−n)n≥0).\begin{split} R(B;N,a)&=\{x\in\mathbb R:\exists t\in[1,2]\quad x+t a_n\notin B\text{ for every }n\in N\},\\ R_N(B)&=R(B;N,(2^{-n})_{n\geq0}). \end{split}(2)✦

For p>0p>0, let ℋp\mathcal H_p be the class of open unit-periodic sets with the hitting and density conditions in (3). ℋp={H⊆ℝ:H is open,H+1=H,ρ(H)≤6p,for every x∈ℝ and t∈[1,2], there is n∈ℕ,n≥1 with x+t2−n∈H}.\mathcal H_p=\left\{H\subseteq\mathbb R: \begin{array}{l} H\text{ is open},\quad H+1=H,\quad\rho(H)\leq6p,\\ \text{for every }x\in\mathbb R\text{ and }t\in[1,2],\text{ there is }n\in\mathbb N,\ n\geq1 \text{ with }x+t2^{-n}\in H \end{array}\right\}.(3)

The Random Blocker

For natural numbers M,d,g,r0M,d,g,r_0, a router of depth dd has children 0,…,M0,\ldots,M, with MM the default child.✦ Thus MM counts the nondefault children, and the paper’s total number of children is M+1M+1. A decision node PP of a router of depth dd is a path with |P|<d|P|<d✦; a terminal path of the router is a path of length dd.✦ A window edge PP of a router of depth dd is a nonempty path of length at most dd.✦ For a decision node PP and a nondefault child i<Mi<M, the nondefault child edge PiPi is the path obtained by appending ii to PP.✦

Separated window data WW with gap gg and minimum length r0r_0 consist of a length function r:ℕ→ℕr:\mathbb N\to\mathbb N and natural-number assignments a,b,b*a,b,b^* to paths. Write rP=r(d+1−|P|)r_P=r(d+1-|P|). For every window edge PP they satisfy (4). rP≥max⁡{1,r0},a(P)≥3,b(P)+1=a(P)+rP,b*(P)+1≤a(P)+2rP.r_P\geq\max\{1,r_0\},\qquad a(P)\geq3,\qquad b(P)+1=a(P)+r_P,\qquad b^*(P)+1\leq a(P)+2r_P.(4) With P≼QP\preceq Q denoting a prefix, i,ji,j denoting child labels, TT denoting a path, and juxtaposition denoting concatenation, the separation and descendant conditions are (5); all paths appearing as arguments of a,b,b*a,b,b^* there are window edges. b(P)≤b(Q)≤b*(P)(P≼Q),b(P)+g<a(Q)(P≼Q,P≠Q),b(Pi)+g<a(PjT)(i<j).\begin{aligned} b(P)&\leq b(Q)\leq b^*(P) &&(P\preceq Q),\\ b(P)+g&<a(Q) &&(P\preceq Q,\ P\ne Q),\\ b(Pi)+g&<a(PjT) &&(i<j). \end{aligned}(5) The selector endpoint at (P,i)(P,i) is bS(P,i)=b(Pi)b_S(P,i)=b(Pi)✦, and the terminal endpoint at a terminal path LL is bT(L)=b(L)b_T(L)=b(L).✦

For a real mesh parameter qq and a natural number bb, the periodic mesh key κq(x)\kappa_q(x) and its dyadic specialization kb(x)k_b(x) are given by (6), where {x}\{x\} is the fractional part of xx. κq(x)=⌊q{x}⌋,kb(x)=κ2b+2(x)∈{0,…,2b+2−1}.\kappa_q(x)=\lfloor q\{x\}\rfloor,\qquad k_b(x)=\kappa_{2^{b+2}}(x)\in\{0,\ldots,2^{b+2}-1\}.(6) The dyadic cells are half-open and include their left endpoints.

The selector addresses SS and terminal addresses TT are the finite sets in (7). S={(P,i,k):|P|<d,0≤i<M,0≤k<2bS(P,i)+2},T={(L,k):|L|=d,0≤k<2bT(L)+2}.\begin{split} S&=\{(P,i,k):|P|<d,\ 0\leq i<M,\ 0\leq k<2^{b_S(P,i)+2}\},\\ T&=\{(L,k):|L|=d,\ 0\leq k<2^{b_T(L)+2}\}. \end{split}(7)

A selector table α:S→{0,1}\alpha:S\to\{0,1\} assigns a bit to each selector address✦; a terminal table β:T→{0,1}\beta:T\to\{0,1\} assigns a bit to each terminal address.✦ The queried selector address at (P,i,x)(P,i,x) is (P,i,kbS(P,i)(x))(P,i,k_{b_S(P,i)}(x)).✦ The selector value at (P,i,x)(P,i,x) for the table α\alpha is the table entry at the queried selector address when |P|<d|P|<d; the extended selector value at (P,i,x)(P,i,x) is zero when |P|≥d|P|\geq d.✦

The first successful child at a path PP and a point xx for a selector table α\alpha is the least nondefault child whose selector value is one, or the default child if none succeeds. Given a child set II, a state space XX and a rule c(Q,x)∈Ic(Q,x)\in I for finite paths QQ in II and states x∈Xx\in X, the iterated routing path from an initial path PP at xx after n∈ℕn\in\mathbb N steps is obtained by successively appending c(Q,x)c(Q,x) to the current path QQ, starting with PP.✦ Here the rule selects the first successful child for the selector table α\alpha. Starting from the empty path, dd steps give the routed leaf Lα(x)L_\alpha(x).✦ The queried terminal address at (L,x)(L,x) is (L,kbT(L)(x))(L,k_{b_T(L)}(x))✦, and the routed address uα(x)u_\alpha(x) is the queried terminal address with L=Lα(x)L=L_\alpha(x).✦

For any terminal-address map uu, the address-selected set Bu(α,β)B_u(\alpha,\beta) is the set defined in (8). The raw routed set uses the routed address in that formula. Bu(α,β)={x∈ℝ:β(uα(x))=1},Bα,β=Bu(α,β)(uα(x)=(Lα(x),kbT(Lα(x))(x))).B_u(\alpha,\beta)=\{x\in\mathbb R:\beta(u_\alpha(x))=1\},\qquad B_{\alpha,\beta}=B_u(\alpha,\beta)\quad \bigl(u_\alpha(x)=(L_\alpha(x),k_{b_T(L_\alpha(x))}(x))\bigr).(8)

For 0≤p≤10\leq p\leq1, the Bernoulli bit law νp\nu_p assigns masses pp and 1−p1-p to one and zero✦; the fair bit law is ν1/2\nu_{1/2}.✦ The independent fair selector law and independent Bernoulli terminal law have the joint outcome law μp\mu_p in (9). μp=ν1/2⊗S⊗νp⊗T.\mu_p=\nu_{1/2}^{\otimes S}\otimes\nu_p^{\otimes T}.(9)✦

The Claim

Theorem 1. Let R⊆ℝR\subseteq\mathbb R be unit-periodic, and let ε\varepsilon be a positive extended real number. There is an open periodic enlargement with ∃V⊆ℝ:R⊆V,V is open,V+1=V,ρ(V)≤ρ(R)+ε.\exists V\subseteq\mathbb R:\quad R\subseteq V,\quad V\text{ is open},\quad V+1=V,\quad \rho(V)\leq\rho(R)+\varepsilon.(10)✦

Let (Ω,μ)(\Omega,\mu) be a finite probability space, let p>0p>0, and let N⊆{1,2,…}N\subseteq\{1,2,\ldots\}. Let BωB_\omega be unit-periodic subsets of ℝ\mathbb R. If every BωB_\omega is open and 𝔼μρ(Bω)≤2p\mathbb E_\mu\rho(B_\omega)\leq2p and 𝔼μρ(RN(Bω))≤3p\mathbb E_\mu\rho(R_N(B_\omega))\leq3p, then ∃H∈ℋp.\exists H\in\mathcal H_p.(11)✦

Let G⊆ℝG\subseteq\mathbb R be measurable with ρ(Gc)≤p\rho(G^c)\leq p. If every BωB_\omega is open, 𝔼μρ(Bω)≤2p\mathbb E_\mu\rho(B_\omega)\leq2p, and μ{ω:x∈RN(Bω)}≤2p\mu\{\omega:x\in R_N(B_\omega)\}\leq2p for every x∈Gx\in G, then ∃H∈ℋp.\exists H\in\mathcal H_p.(12)✦

For arbitrary unit-periodic BωB_\omega, assume instead that 𝔼μρ(Bω)≤p\mathbb E_\mu\rho(B_\omega)\leq p and μ{ω:x∈RN(Bω)}≤2p\mu\{\omega:x\in R_N(B_\omega)\}\leq2p for every x∈Gx\in G. Then ∃H∈ℋp.\exists H\in\mathcal H_p.(13)✦

Let WW be separated window data as above, let 0<p<10<p<1, and let G⊆ℝG\subseteq\mathbb R be measurable with ρ(Gc)≤p\rho(G^c)\leq p. If μp{ω:x∈R{1,2,…}(Bω)}≤2p\mu_p\{\omega:x\in R_{\{1,2,\ldots\}}(B_\omega)\}\leq2p for every x∈Gx\in G, where BωB_\omega is the raw routed set associated with WW, then ∃H∈ℋp.\exists H\in\mathcal H_p.(14)✦

In the paper, the result reads: enlarge BB to an open periodic B+B^+ at added density at most pp, with 𝔼ρ(B+)≤2p\mathbb E\rho(B^+)\leq2p and 𝔼ρ(R)≤3p\mathbb E\rho(R)\leq3p for the exceptional set RR, select an outcome with ρ(B+)+ρ(R)≤5p\rho(B^+)+\rho(R)\leq5p, cover RR by an open periodic VV satisfying ρ(V)≤ρ(R)+p\rho(V)\leq\rho(R)+p, and set H=B+∪VH=B^+\cup V, obtaining density at most 6p6p and hitting at all centers.

What Is Checked

The enlargement in (10) is stated over arbitrary unit-periodic sets, and the hitting statements (11), (12) and (13) are stated over finite probability spaces, with (14) using the independent table law and the routed set associated with separated window data.