OctoInformBeta

OpenAI preprint, September 2026 · Section 2. From conditioning to a threshold bound

Section 2 (setting): Finite probability laws, local measurements and the events used to count wins or condition on selected coordinates

Repeated quantum experiments and their win statistics.

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 threshold event in a repeated nonlocal game concerns how many coordinates the players win. Conditioning on success at selected coordinates concerns a different event in the same experiment. The question and answer transcript provides the outcome space for both statistics.

The Definition

Definition 1 (Repeated quantum experiments and win statistics). A finite probability law pp on a finite set SS is a function assigning nonnegative real weights to its elements, with total weight one, as in (1). p:S→ℝ,p(s)≥0(s∈S),∑s∈Sp(s)=1.p:S\longrightarrow\mathbb R,\qquad p(s)\geq 0\quad(s\in S),\qquad \sum_{s\in S}p(s)=1.(1)✦ The expectation 𝔼p[f]\mathbb E_p[f] of a function f:S→Ef:S\to E, where EE is a real vector space, is the weighted sum in (2). 𝔼p[f]=∑s∈Sp(s)f(s).\mathbb E_p[f]=\sum_{s\in S}p(s)f(s).(2)✦ For a finite index set JJ and laws pjp_j on SS, their product law p⊗p_{\otimes} on SJS^J assigns weights as in (3). The uniform law uRu_R on a nonempty finite set RR assigns the equal weights given there. p⊗(s)=∏j∈Jpj(sj)(s∈SJ),uR(r)=1|R|(r∈R).p_{\otimes}(s)=\prod_{j\in J}p_j(s_j)\quad(s\in S^J), \qquad u_R(r)=\frac{1}{|R|}\quad(r\in R).(3)

For local dimensions m,nm,n, a rectangular coefficient matrix Γ∈ℂm×n\Gamma\in\mathbb C^{m\times n} determines the bipartite vector |Γ⟩|\Gamma\rangle and has squared Hilbert-Schmidt norm given in (4). |Γ⟩=∑a=1m∑b=1nΓab|a⟩⊗|b⟩,∥Γ∥HS2=∑a=1m∑b=1n|Γab|2.|\Gamma\rangle=\sum_{a=1}^{m}\sum_{b=1}^{n}\Gamma_{ab}|a\rangle\otimes|b\rangle, \qquad \|\Gamma\|_{\mathrm{HS}}^2 =\sum_{a=1}^{m}\sum_{b=1}^{n}|\Gamma_{ab}|^2.(4) A positive operator-valued measurement (POVM) 𝖤=(Eu)u∈U\mathsf E=(E_u)_{u\in U} on a finite-dimensional Hilbert space HH, with finite outcome set UU, is a family of operators satisfying (5). Eu≽0(u∈U),∑u∈UEu=IH.E_u\succeq 0\quad(u\in U),\qquad \sum_{u\in U}E_u=I_H.(5)✦ For local operators LL on ℂm\mathbb C^m and NN on ℂn\mathbb C^n, the bipartite operator expectation eΓ(L,N)e_\Gamma(L,N) and the real-valued Born expression bΓ(L,N)b_\Gamma(L,N) are defined by (6). eΓ(L,N)=⟨Γ|(L⊗N)|Γ⟩,bΓ(L,N)=Re⁡eΓ(L,N).e_\Gamma(L,N)=\langle\Gamma|(L\otimes N)|\Gamma\rangle, \qquad b_\Gamma(L,N)=\operatorname{Re}e_\Gamma(L,N).(6) For a normalized matrix Γ\Gamma and local POVMs 𝖤=(Eu)u∈U\mathsf E=(E_u)_{u\in U} and 𝖥=(Fz)z∈Z\mathsf F=(F_z)_{z\in Z}, the joint measurement law qΓ,𝖤,𝖥q_{\Gamma,\mathsf E,\mathsf F} on U×ZU\times Z assigns the Born probabilities in (7). qΓ,𝖤,𝖥(u,z)=bΓ(Eu,Fz),∥Γ∥HS2=1.q_{\Gamma,\mathsf E,\mathsf F}(u,z)=b_\Gamma(E_u,F_z), \qquad \|\Gamma\|_{\mathrm{HS}}^2=1.(7)✦

Let X,YX,Y be finite question alphabets, A,BA,B finite answer alphabets and II a finite coordinate set. A repeated quantum experiment ℳ\mathcal M is a question law μ\mu on X×YX\times Y, a normalized rectangular state matrix and local POVMs indexed by each player’s full question string, together with a winning predicate VV, as specified in (8). ℳ=(μ,Γ,P,Q,V),Γ∈ℂm×n,∥Γ∥HS2=1,P𝒙=(P𝒙𝒂)𝒂∈AI is a POVM on ℂm(𝒙∈XI),Q𝒚=(Q𝒚𝒃)𝒃∈BI is a POVM on ℂn(𝒚∈YI),V:X×Y×A×B→{0,1}.\begin{gathered} \mathcal M=(\mu,\Gamma,P,Q,V),\qquad \Gamma\in\mathbb C^{m\times n},\quad \|\Gamma\|_{\mathrm{HS}}^2=1,\\ P_{\mathbf x}=(P_{\mathbf x}^{\mathbf a})_{\mathbf a\in A^I} \text{ is a POVM on }\mathbb C^m\quad(\mathbf x\in X^I),\\ Q_{\mathbf y}=(Q_{\mathbf y}^{\mathbf b})_{\mathbf b\in B^I} \text{ is a POVM on }\mathbb C^n\quad(\mathbf y\in Y^I),\\ V:X\times Y\times A\times B\longrightarrow\{0,1\}. \end{gathered}(8)✦ A question profile z:I→X×Yz:I\to X\times Y is a string of question pairs indexed by the coordinates.✦ Writing zi=(xi,yi)z_i=(x_i,y_i), let 𝒙=(xi)i∈I\mathbf x=(x_i)_{i\in I} and 𝒚=(yi)i∈I\mathbf y=(y_i)_{i\in I}. A question and answer transcript tt is a profile together with answer strings 𝒂∈AI\mathbf a\in A^I and 𝒃∈BI\mathbf b\in B^I; its outcome space 𝒯\mathcal T is given in (9). 𝒯=(X×Y)I×(AI×BI),t=(z,(𝒂,𝒃))∈𝒯.\mathcal T=(X\times Y)^I\times(A^I\times B^I), \qquad t=(z,(\mathbf a,\mathbf b))\in\mathcal T.(9)✦ For a question profile z:I→X×Yz:I\to X\times Y with full question strings 𝒙,𝒚\mathbf x,\mathbf y, the conditional answer law qzq_z on AI×BIA^I\times B^I is the joint measurement law for the state matrix Γ\Gamma and the local POVMs P𝒙,Q𝒚P_{\mathbf x},Q_{\mathbf y}. The joint transcript law ρ\rho on 𝒯\mathcal T is the probability law obtained by combining the conditional answer laws qzq_z with the product question law induced by μ\mu, as in (10). qz(𝒂,𝒃)=qΓ,P𝒙,Q𝒚(𝒂,𝒃),ρ(z,(𝒂,𝒃))=(∏i∈Iμ(xi,yi))qz(𝒂,𝒃).\begin{aligned} q_z(\mathbf a,\mathbf b) &=q_{\Gamma,P_{\mathbf x},Q_{\mathbf y}}(\mathbf a,\mathbf b),\\ \rho(z,(\mathbf a,\mathbf b)) &=\left(\prod_{i\in I}\mu(x_i,y_i)\right)q_z(\mathbf a,\mathbf b). \end{aligned}(10) The coordinate winning predicate wi:𝒯→{0,1}w_i:\mathcal T\to\{0,1\} evaluates the game’s predicate at coordinate ii, as in (11). wi(t)=V(xi,yi,ai,bi).w_i(t)=V(x_i,y_i,a_i,b_i).(11)✦ For a tested set C⊆IC\subseteq I, the event EC⊆𝒯E_C\subseteq\mathcal T consists of transcripts winning every tested coordinate. The real-valued indicator 𝟏EC\mathbf 1_{E_C} of the tested-set event EC⊆𝒯E_C\subseteq\mathcal T is the function on 𝒯\mathcal T equal to one on ECE_C and zero elsewhere, as in (12). EC={t∈𝒯:wi(t)=1 for every i∈C},𝟏EC(t)={1,t∈EC,0,t∉EC.\begin{gathered} E_C=\{t\in\mathcal T:w_i(t)=1\text{ for every }i\in C\},\\ \mathbf 1_{E_C}(t)= \begin{cases} 1,&t\in E_C,\\ 0,&t\notin E_C. \end{cases} \end{gathered}(12)✦ For a tested set C⊆IC\subseteq I, the tested-set probability pCp_C is the probability of the event ECE_C under the joint transcript law ρ\rho, expressed as the expectation of 𝟏EC\mathbf 1_{E_C} in (13). pC=𝔼ρ[𝟏EC].p_C=\mathbb E_\rho[\mathbf 1_{E_C}].(13)✦

For a finite outcome space Ω\Omega and coordinate winning predicates ui:Ω→{0,1}u_i:\Omega\to\{0,1\} indexed by a nonempty II, set k=|I|k=|I| and let Wk(ω)W_k(\omega) denote the total number of wins. For coordinate winning predicates ui:Ω→{0,1}u_i:\Omega\to\{0,1\} indexed by a nonempty finite set II, the winning fraction rr on Ω\Omega is the real-valued function given by their uniform coordinate average in (14); a repeated quantum experiment uses Ω=𝒯\Omega=\mathcal T and ui=wiu_i=w_i. r(ω)=𝔼uI[i↦ui(ω)]=1k∑i∈Iui(ω)=Wk(ω)k.r(\omega)=\mathbb E_{u_I}[i\mapsto u_i(\omega)] =\frac{1}{k}\sum_{i\in I}u_i(\omega) =\frac{W_k(\omega)}{k}.(14)✦

What Is Checked

The probability laws in (1) and the repeated quantum experiments in (8) are stated over finite sets and finite-dimensional local systems; the winning fraction in (14) requires a nonempty coordinate set. Product laws of finite probability laws have nonnegative weights summing to one. The same holds for the joint measurement law of a normalized state and local POVMs and for the resulting joint transcript law.