Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 6.4 — LocalSearch with value estimates

Open
NuclearTensor.Membership.Supplement.lemma_6_4

by mikedeng1 · Nov 15, 2026 · Mathlib 0df444a (Lean v4.33.1)

computational-complexityp2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1quantum-informationseparability

Let ω\omegaω be an unknown state on (Cq)⊗m(\mathbb C^q)^{\otimes m}(Cq)⊗m, and let K=∑ℓ=1Tcℓ⨂jKℓ,jK=\sum_{\ell=1}^T c_\ell\bigotimes_jK_{\ell,j}K=∑ℓ=1T​cℓ​⨂j​Kℓ,j​ be explicit, with real cℓc_\ellcℓ​, Hermitian factors satisfying ∥Kℓ,j∥F≤1\|K_{\ell,j}\|_F\le1∥Kℓ,j​∥F​≤1, and G=1+∑ℓ∣cℓ∣G=1+\sum_\ell|c_\ell|G=1+∑ℓ​∣cℓ​∣. A unit product reference aaa has m−1m-1m−1 factors. Define

R(a,α)={x=⨂jxj:∥xj∥=1, ∣⟨a,x[m−1]⟩∣2≥α}.R(a,\alpha)=\{x=\bigotimes_jx_j:\|x_j\|=1,\ |\langle a,x_{[m-1]}\rangle|^2\ge\alpha\}.R(a,α)={x=j⨂​xj​:∥xj​∥=1, ∣⟨a,x[m−1]​⟩∣2≥α}.

For every α,ϵ,δ>0\alpha,\epsilon,\delta>0α,ϵ,δ>0, a quantum algorithm returns a unit product x^\widehat xx and real vvv. With probability at least 1−δ1-\delta1−δ, both clauses hold: if the regional maximum is at least ϵ\epsilonϵ, then its value at x^\widehat xx is within ϵ\epsilonϵ of that maximum, and

∣v−Tr⁡((ω−K)∣x^⟩⟨x^∣)∣≤ϵ.|v-\operatorname{Tr}((\omega-K)|\widehat x\rangle\langle\widehat x|)|\le\epsilon.∣v−Tr((ω−K)∣x⟩⟨x∣)∣≤ϵ.

Copy complexity has exponent O(log⁡(G/(αϵ)))O(\log(G/(\alpha\epsilon)))O(log(G/(αϵ))) and a logarithmic confidence factor; total time is poly⁡α,ϵ,G(m,q,T,L(δ))\operatorname{poly}_{\alpha,\epsilon,G}(m,q,T,L(\delta))polyα,ϵ,G​(m,q,T,L(δ)). This supplies the full LocalSearch contract used by the cover and signed optimizer.

Formalization Note The proof's truncation uses h=4elog⁡(32G/(αϵ))h=4e\log(32G/(\alpha\epsilon))h=4elog(32G/(αϵ)). The all-positive-parameter reading uses internal αˉ=min⁡(α,1)\bar\alpha=\min(\alpha,1)αˉ=min(α,1) and ϵˉ=min⁡(ϵ,1)\bar\epsilon=\min(\epsilon,1)ϵˉ=min(ϵ,1) and copy bound b(1+mq)clog⁡(32G/(αˉϵˉ))L(δ)b(1+mq)^{c\log(32G/(\bar\alpha\bar\epsilon))}L(\delta)b(1+mq)clog(32G/(αˉϵˉ))L(δ), with absolute positive b,cb,cb,c. External accuracies are unchanged. For α>1\alpha>1α>1 only the regional optimization clause is vacuous; the value clause remains required.

Differences from the source:

  • encoding: finite complex Hilbert spaces and Hilbert–Schmidt geometry → row-major complex matrices on Fin (d^k), explicit pure-product hull, and an extended-nonnegative Frobenius-distance infimum; reason: these coordinates cover every finite-dimensional source instance; no density or norm restriction is added to the classical Hermitian input
  • reading: algorithms with classical arithmetic and quantum copy access → finite arithmetic/spectral programs and elementary H,T,CX programs with Born-rule execution, charged fresh copies, and explicit classical outputs; reason: makes the computational model and all resource claims operational; neither global optimization nor a desired quantum subroutine is a primitive
  • encoding: O and polynomial resource notation → positive integer multiplicative constants and degrees quantified before varying instance parameters; reason: preserves the indicated parameter dependencies and ties each bound to actual executed charges
  • encoding: positive failure parameters and logarithmic confidence factors → all positive failure parameters, using L(x)=1+max(0,log(1/x)) in resource bounds; reason: agrees with logarithmic confidence scaling in the nontrivial range and accommodates the vacuous-success branch for x at least one
  • reading: the printed exponent O(log(G/(alpha epsilon))) with all positive parameters → copy exponent c log(32G/(min(alpha,1) min(epsilon,1))); reason: uses Algorithm 3 and Appendix A truncation scale and clamps only internal tolerances; the two external-accuracy clauses remain intact
Preamble
import Definitions.Def_NuclearTensor_Membership_Supplement_Tasks

set_option autoImplicit false
noncomputable section
Formal statement
namespace NuclearTensor.Membership.Supplement

theorem lemma_6_4 :
    ∃ b c : ℕ, 0 < b ∧ 0 < c ∧
      ∀ α ε G : ℝ, 0 < α → 0 < ε → 1 ≤ G → ∃ P : QProgram,
      ∀ (σ : Spectral) (A r : ℕ), 0 < A → 0 < r →
      ∃ B C : ℕ, 0 < B ∧ 0 < C ∧
      ∀ d m T : ℕ, 0 < d → 0 < m → ∀ δ : ℝ, 0 < δ →
      ∀ K : SignedOperator d m T, K.G = G →
      ∀ a : Fin (m - 1) → CVec d, (∀ j, normSq (a j) = 1) →
      ∀ ω : CMat (d ^ m), IsDensity ω →
        quantumAlgorithm ω σ A r P [d, m, T]
          ([α, ε, δ] ++ encodeSigned K ++ encodeProducts a)
          (b * Real.rpow (1 + (m : ℝ) * d)
            (c * Real.log (32 * G / (min α 1 * min ε 1))) * confidence δ)
          (B * (1 + m + d + T + confidence δ) ^ C) δ
          (localSearchOutput ω K a α ε) (returnsProductValue d m) := by sorry

end NuclearTensor.Membership.Supplement
Source
Martino Bernasconi, Giulio Malavolta, STOC 2027 submission #945 (October 2026), p. 23, Lemma 6.4; supporting proof in pp. 40–43.

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me