OctoInformBeta

OpenAI preprint, September 2026 · Section 5. The finite scale induction

Section 5: Finite scale bootstrap: Validity above one extends to all real pairs s,ts,t with 0≤s<t0\le s<t under a backward-step hypothesis

Backward implications close a finite-scale bootstrap.

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

Uniform circuit localization bounds the passage from a product projection’s zero-mismatch sector to a high-mismatch sector through a conjugated product projection. The mismatch cutoff and the decay rate have separate exponents. The finite-scale bootstrap concerns assertions indexed by these two exponents.

A classical bit is an element of {0,1}\{0,1\}✦, and a bit string of length NN assigns a classical bit to each of NN wire positions.✦ A quantum operator on NN total qubits is a complex matrix whose rows and columns are indexed by bit strings of length NN.✦ A single-qubit operator is a complex matrix indexed by classical bits.✦

For a one-qubit vector vv, the outer product |v⟩⟨v||v\rangle\langle v| has entry v(x)v(y)¯v(x)\overline{v(y)} in row xx and column yy.✦ The tensor product of single-qubit operators aia_i is the operator whose entry from a bit string yy to a bit string xx is ∏i=1Nai(xi,yi)\prod_{i=1}^{N}a_i(x_i,y_i).✦

Product Projections and Circuit Scales

Product projection data M=(S,(vi)i=1N)M=(S,(v_i)_{i=1}^{N}) on NN qubits consist of a support S⊆{1,…,N}S\subseteq\{1,\ldots,N\} and local vectors vi∈ℂ{0,1}v_i\in\mathbb C^{\{0,1\}} satisfying (1) on the support: ∑b∈{0,1}|vi(b)|2=1(i∈S).\sum_{b\in\{0,1\}}|v_i(b)|^2=1\qquad(i\in S).(1)✦ For product projection data MM on NN qubits and a set FF of wire positions, the mismatch pattern projection PFMP_F^M is the tensor product in (2): PFM=⨂i=1N{I−|vi⟩⟨vi|,i∈S∩F,|vi⟩⟨vi|,i∈S\F,I,i∉S.P_F^M=\bigotimes_{i=1}^{N} \begin{cases} I-|v_i\rangle\langle v_i|,&i\in S\cap F,\\ |v_i\rangle\langle v_i|,&i\in S\setminus F,\\ I,&i\notin S. \end{cases}(2) For product projection data MM on NN qubits and a real cutoff rr, the zero-mismatch product projection [M=0][M=0] and the high spectral projection [M≥r][M\ge r] are the operators defined by (3); the high cutoff includes equality: [M=0]=⨂i=1N{|vi⟩⟨vi|,i∈S,I,i∉S,[M≥r]=∑F⊆S|F|≥rPFM.[M=0]=\bigotimes_{i=1}^{N} \begin{cases} |v_i\rangle\langle v_i|,&i\in S,\\ I,&i\notin S, \end{cases} \qquad [M\ge r]=\sum_{\substack{F\subseteq S\\|F|\ge r}}P_F^M.(3)

For a natural depth dd and NN total qubits, the enlarged reflection-circuit class Ud(N)U_d(N) is the recursively defined class of operators of depth at most dd in (4). Here LL denotes the matrix of a layer of single-qubit unitaries and RR the matrix of a layer of reflections about product projections with pairwise disjoint supports; identity layers are allowed, and the rightmost matrix acts first. $$\begin{aligned} U_0(N)&=\{L:L\text{ is a local-layer matrix}\},\\ U_{d+1}(N)&=\{VRL:V\in U_d(N),\ R\text{ is a reflection-layer matrix},\\ &\hspace{42mm}L\text{ is a local-layer matrix}\}. \end{aligned}$$(4)✦

For quantum operators AA and UU on NN qubits, operator conjugation of AA by UU is the operator UAU†UAU^\dagger, where the dagger denotes the adjoint.✦ For a natural depth dd and real exponents s,ts,t, uniform localization Bd(s,t)B_d(s,t) is the assertion that (5) holds for all sufficiently large NN, simultaneously for every U∈Ud(N)U\in U_d(N) and all product projection data M,DM,D on NN qubits: ∥[M≥Nt]U[D=0]U†[M=0]∥≤exp⁡(−Ns).\left\|[M\ge N^t]\,U[D=0]U^\dagger[M=0]\right\| \le \exp(-N^s).(5)✦ The norm is the Euclidean operator norm.

The Claim

Theorem 1 (Finite-scale bootstrap). Let B(s,t)B(s,t) be an assertion for pairs of real numbers. Suppose B(s,t)B(s,t) holds whenever 0≤s<t0\le s<t and 1<t1<t. Suppose also that for every choice of real numbers satisfying 0≤s<a<u<w<t<b0\le s<a<u<w<t<b and (a+b)/2<u(a+b)/2<u, validity of B(t,b)B(t,b) implies B(s,t)B(s,t). Then B(s,t)for all real s,t with 0≤s<t.B(s,t)\qquad\text{for all real }s,t\text{ with }0\le s<t.(6)✦

Theorem 2 (Localization above one). For every natural depth dd and all real exponents s,ts,t with t>1t>1, Bd(s,t).B_d(s,t).(7)✦

What Is Checked

The finite-scale bootstrap (6) is stated over arbitrary assertions on pairs of real numbers with the large-threshold and backward-step hypotheses, while (7) concerns uniform localization over the enlarged reflection-circuit classes and all normalized product projection data.