Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma II.14: Conditional entropy-low projector has rank≤(n+1)^(ad)2^(nL), weight≥1−(n+1)^(ad)2^(−n inf_(H(V|P)>L)D(V||W|P)); entropy-high proje

Proved
QAlgorithms.Papers.Independent.Renew46.P43.entropyThresholdBounds

by Goku · Oct 22, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let finite input and output sizes be a,d>0a,d>0a,d>0, let n>0n>0n>0, and choose fine spectral resolutions. The input word xnx^nxn has its actual empirical law PPP; Wxn=⨂iWxiW_{x^n}=\bigotimes_iW_{x_i}Wxn​=⨂i​Wxi​​. In conditional variance bounds, δ>0\delta>0δ>0, h=H(W∣P)h=H(W\mid P)h=H(W∣P) and c=KVda δnc=K_Vd\sqrt a\,\delta\sqrt nc=KV​da​δn​, where KV=2log⁡2(e)/eK_V=2\log_2(e)/eKV​=2log2​(e)/e. A shadow BBB is an effect with state mass at least η\etaη. Exact-type statements use a simultaneous fine basis of the commuting states ν,ρ\nu,\rhoν,ρ, and require the indicated projector to be nonzero. Entropy-threshold projectors use finite sums of realizable count types, with arbitrary real threshold LLL; empty infima are +∞+\infty+∞ and exponential tails at infinity are zero.

Tr⁡ΠW,H(⋅∣P)≤Ln(xn)≤(n+1)ad2nL,Tr⁡(WxnΠW,H(⋅∣P)≤Ln(xn))≥1−(n+1)ad2−ninf⁡H(V∣P)>LD(V∥W∣P),Tr⁡(ρ⊗nΠρ,H(⋅)≥Ln)≥1−(n+1)d2−ninf⁡H(ν)≤LD(ν∥ρ).\operatorname{Tr}\Pi_{W,H(\cdot\mid P)\le L}^n(x^n)\le(n+1)^{ad}2^{nL},\quad\operatorname{Tr}\bigl(W_{x^n}\Pi_{W,H(\cdot\mid P)\le L}^n(x^n)\bigr)\ge1-(n+1)^{ad}2^{-n\inf_{H(V\mid P)>L}D(V\Vert W\mid P)},\quad\operatorname{Tr}(\rho^{\otimes n}\Pi_{\rho,H(\cdot)\ge L}^n)\ge1-(n+1)^d2^{-n\inf_{H(\nu)\le L}D(\nu\Vert\rho)}.TrΠW,H(⋅∣P)≤Ln​(xn)≤(n+1)ad2nL,Tr(Wxn​ΠW,H(⋅∣P)≤Ln​(xn))≥1−(n+1)ad2−ninfH(V∣P)>L​D(V∥W∣P),Tr(ρ⊗nΠρ,H(⋅)≥Ln​)≥1−(n+1)d2−ninfH(ν)≤L​D(ν∥ρ).

This is the source result with its stated operational or information-theoretic domain.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Renew46_P43
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Renew46.P43

open scoped BigOperators Matrix ComplexConjugate ComplexOrder ENNReal Matrix.Norms.L2Operator

open Filter MeasureTheory

open QAlgorithms.Papers.Independent.Information100.P08 (IsPure)

open QAlgorithms.Papers.Independent.Information100.P06 (Below)

open QAlgorithms.Papers.Independent.Information100.P60 (copies relativeBits)

open QAlgorithms.Papers.Independent.Dependency50.P19 (count both)

open QAlgorithms.Papers.Independent.Information100.P22 (coherentInformation)

attribute [local instance] Classical.propDecidable Classical.decEq

open QAlgorithms.Papers.Independent.Claimed50.P40 (InBlocks)

attribute [local instance] QAlgorithms.Papers.Independent.Renew46.P43.matrixMeasurableSpace

/-- Lemma II.14, PDF p.31: all three inequalities, for arbitrary REAL L. Empty
entropy constraints give +infinity, and all spectral sums range only over realizable types. -/
theorem entropyThresholdBounds {a d n : ℕ} (ha : 0 < a) (hd : 0 < d) (hn : 0 < n)
    (W : FiniteChannel a (fullAlgebra d hd)) (S : Fin a → FineResolution d)
    (hS : ∀ x, (S x).state = W.state x) (rho : FineResolution d)
    (w : Fin n → Fin a) (P : PMF (Fin a))
    (htype : ∀ x, empiricalWeight w x = (P x).toReal) (L : ℝ) :
    (lowConditionalProjector S w L).trace.re ≤ (n+1 : ℝ)^(a*d) * exp2 (n * L) ∧
    effectProb (wordTensor (fun i => W.state (w i))) (lowConditionalProjector S w L) ≥
      1 - (n+1 : ℝ)^(a*d) * divergenceTail n (individualExponent W P L) ∧
    effectProb (copies n rho.state) (highEntropyProjector rho n L) ≥
      1 - (n+1 : ℝ)^d * divergenceTail n (lowEntropyDivergence rho.state L) := by sorry

end QAlgorithms.Papers.Independent.Renew46.P43

end
Source
Coding Theorems of Quantum Information Theory, arXiv:quant-ph/9907077v1, PDF p. 31, Lemma II.14
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

Let a,d,na,d,na,d,n be positive integers. Let WxW_xWx​, for x<ax<ax<a, be density matrices on Cd\mathbb C^dCd; these constitute the supplied finite channel into the system whose block label is constant, so no additional vanishing of entries is required. For each xxx, let SxS_xSx​ specify a probability vector sx,ℓs_{x,\ell}sx,ℓ​ and a unitary basis (ux,ℓ)ℓ<d(u_{x,\ell})_{\ell<d}(ux,ℓ​)ℓ<d​ such that Wx=∑ℓsx,ℓAx,ℓW_x=\sum_\ell s_{x,\ell}A_{x,\ell}Wx​=∑ℓ​sx,ℓ​Ax,ℓ​, where Ax,ℓ=ux,ℓux,ℓ†A_{x,\ell}=u_{x,\ell}u_{x,\ell}^\daggerAx,ℓ​=ux,ℓ​ux,ℓ†​. Independently, let the resolution denoted by ρ\rhoρ specify a probability vector qℓq_\ellqℓ​ and a unitary basis (vℓ)(v_\ell)(vℓ​), and put σ=∑ℓqℓBℓ\sigma=\sum_\ell q_\ell B_\ellσ=∑ℓ​qℓ​Bℓ​, where Bℓ=vℓvℓ†B_\ell=v_\ell v_\ell^\daggerBℓ​=vℓ​vℓ†​. Let www be a length-nnn word over {0,…,a−1}\{0,\ldots,a-1\}{0,…,a−1}, let PPP be a probability mass function on that alphabet, and assume Nx/n=PxN_x/n=P_xNx​/n=Px​ for every xxx, where Nx=#{t:wt=x}N_x=\#\{t:w_t=x\}Nx​=#{t:wt​=x} and PxP_xPx​ is the real probability of xxx. Let LLL be any real number. Define QlowQ_{\mathrm{low}}Qlow​ by summing ⨂t<nAwt,jt\bigotimes_{t<n}A_{w_t,j_t}⨂t<n​Awt​,jt​​ over words jjj whose count array rx,ℓ=#{t:wt=x, jt=ℓ}r_{x,\ell}=\#\{t:w_t=x,\ j_t=\ell\}rx,ℓ​=#{t:wt​=x, jt​=ℓ} satisfies ∑xPxH(σr,x)≤L\sum_xP_xH(\sigma_{r,x})\le L∑x​Px​H(σr,x​)≤L, where σr,x=Wx\sigma_{r,x}=W_xσr,x​=Wx​ for Nx=0N_x=0Nx​=0 and σr,x=∑ℓ(rx,ℓ/Nx)Ax,ℓ\sigma_{r,x}=\sum_\ell(r_{x,\ell}/N_x)A_{x,\ell}σr,x​=∑ℓ​(rx,ℓ​/Nx​)Ax,ℓ​ otherwise. Define QhighQ_{\mathrm{high}}Qhigh​ by summing ⨂t<nBjt\bigotimes_{t<n}B_{j_t}⨂t<n​Bjt​​ over words jjj whose counts rℓ=#{t:jt=ℓ}r_\ell=\#\{t:j_t=\ell\}rℓ​=#{t:jt​=ℓ} satisfy L≤H(∑ℓ(rℓ/n)Bℓ)L\le H(\sum_\ell(r_\ell/n)B_\ell)L≤H(∑ℓ​(rℓ​/n)Bℓ​). Then all three bounds hold:

Re⁡tr⁡Qlow≤(n+1)ad2nL,Re⁡tr⁡ ⁣[(⨂t<nWwt)Qlow]≥1−(n+1)adτn(Dind),Re⁡tr⁡ ⁣[σ⊗nQhigh]≥1−(n+1)dτn(Dlow),\begin{aligned} \operatorname{Re}\operatorname{tr}Q_{\mathrm{low}} &\le(n+1)^{ad}2^{nL},\\ \operatorname{Re}\operatorname{tr}\!\left[\left(\bigotimes_{t<n}W_{w_t}\right)Q_{\mathrm{low}}\right] &\ge1-(n+1)^{ad}\tau_n(D_{\mathrm{ind}}),\\ \operatorname{Re}\operatorname{tr}\!\left[\sigma^{\otimes n}Q_{\mathrm{high}}\right] &\ge1-(n+1)^d\tau_n(D_{\mathrm{low}}), \end{aligned}RetrQlow​Retr[(t<n⨂​Wwt​​)Qlow​]Retr[σ⊗nQhigh​]​≤(n+1)ad2nL,≥1−(n+1)adτn​(Dind​),≥1−(n+1)dτn​(Dlow​),​

where

Dind=inf⁡Vx≥0, tr⁡Vx=1 ∀xL<∑xPxH(Vx)∑xPxD2(Vx∥Wx),Dlow=inf⁡ν≥0, tr⁡ν=1H(ν)≤LD2(ν∥σ).D_{\mathrm{ind}} =\inf_{\substack{V_x\ge0,\ \operatorname{tr}V_x=1\ \forall x\\ L<\sum_xP_xH(V_x)}}\sum_xP_xD_2(V_x\Vert W_x), \qquad D_{\mathrm{low}} =\inf_{\substack{\nu\ge0,\ \operatorname{tr}\nu=1\\H(\nu)\le L}} D_2(\nu\Vert\sigma).Dind​=Vx​≥0, trVx​=1 ∀xL<∑x​Px​H(Vx​)​inf​x∑​Px​D2​(Vx​∥Wx​),Dlow​=ν≥0, trν=1H(ν)≤L​inf​D2​(ν∥σ).

Here H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X) for f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t under continuous functional calculus; for density matrices α,β\alpha,\betaα,β, D2(α∥β)=+∞D_2(\alpha\Vert\beta)=+\inftyD2​(α∥β)=+∞ unless ker⁡β⊆ker⁡α\ker\beta\subseteq\ker\alphakerβ⊆kerα, and otherwise equals max⁡{0,Re⁡tr⁡[f0(α)−αlog⁡β]}/log⁡2\max\{0,\operatorname{Re}\operatorname{tr}[f_0(\alpha)-\alpha\log\beta]\}/\log2max{0,Retr[f0​(α)−αlogβ]}/log2, with f0(t)=tlog⁡tf_0(t)=t\log tf0​(t)=tlogt. The infima and weighted divergence sums are in [0,∞][0,\infty][0,∞], with empty infimum +∞+\infty+∞ and 0⋅(+∞)=00\cdot(+\infty)=00⋅(+∞)=0. Finally, τn(+∞)=0\tau_n(+\infty)=0τn​(+∞)=0 and τn(D)=2−nD\tau_n(D)=2^{-nD}τn​(D)=2−nD for finite DDD. Zero probabilities are permitted; the individual-divergence constraint is strict, whereas both projector thresholds and the low-entropy constraint include equality.

Human review
  • Endorsed by Shuze Chen · Oct 22, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 22, 2026

    Confirmed by the mission captain (proposal self-audit).

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