Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§4.2.2 (4.30)–(4.31): minimality bound

Open
QAlgorithms.Papers.Independent.Intakenext100.P07.sandwiched_minimality

by Gohan · Oct 31, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

All quantum systems below are finite-dimensional complex Hilbert spaces of positive dimension, and logarithms are binary. A normalized state is positive semidefinite with trace one; a subnormalized state has trace at most one. Marginals are partial traces. Tildes denote sandwiched Rényi quantities and bars denote Petz quantities. Their trace functionals are Q~α=Tr⁡(σ(1−α)/(2α)ρσ(1−α)/(2α))α\widetilde Q_\alpha=\operatorname{Tr}(\sigma^{(1-\alpha)/(2\alpha)}\rho\sigma^{(1-\alpha)/(2\alpha)})^\alphaQ​α​=Tr(σ(1−α)/(2α)ρσ(1−α)/(2α))α and Q‾α=Tr⁡(ρασ1−α)\overline Q_\alpha=\operatorname{Tr}(\rho^\alpha\sigma^{1-\alpha})Q​α​=Tr(ρασ1−α); the divergences include division by Tr⁡ρ\operatorname{Tr}\rhoTrρ inside the logarithm. Matrix powers vanish on the kernel, including power zero. The notation ρ≪σ\rho\ll\sigmaρ≪σ means support inclusion.

Every additive CPTP-monotone quantum extension agreeing with classical Dα on commuting inputs dominates the sandwiched divergence given by the normalized-trace version of (4.31), in the source order domain.

D~α(ρ∥σ)=lim⁡n1nDα(Pσ⊗n(ρ⊗n)∥σ⊗n)≤Dαextension(ρ∥σ).\widetilde D_\alpha(\rho\|\sigma)=\lim_n\frac1nD_\alpha(\mathcal P_{\sigma^{\otimes n}}(\rho^{\otimes n})\|\sigma^{\otimes n})\le D_\alpha^{\rm extension}(\rho\|\sigma).Dα​(ρ∥σ)=nlim​n1​Dα​(Pσ⊗n​(ρ⊗n)∥σ⊗n)≤Dαextension​(ρ∥σ).

This result records the source’s operational or structural relation, with the stated endpoint and attainment clauses retained.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P07
Formal statement
noncomputable section

open Filter

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

namespace QAlgorithms.Papers.Independent.Intakenext100.P07

attribute [local instance] Classical.propDecidable Classical.decEq

variable {I J A B C D X Y Z : Type*}

variable [Fintype I] [DecidableEq I] [Fintype J] [DecidableEq J]

variable [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]

variable [Fintype C] [DecidableEq C] [Fintype D] [DecidableEq D]

variable [Fintype X] [DecidableEq X] [Fintype Y] [DecidableEq Y]

variable [Fintype Z] [DecidableEq Z]

/-- NODE paper:1504.00233#4.minimalbound
Equations (4.30)-(4.31), PDF p.58: an additive CPTP-monotone extension
is bounded below by the normalized-trace sandwiched divergence. The pinched regularization
converges to that same quantity, including the endpoint orders. -/
theorem sandwiched_minimality (D : paperDivergenceFamily) (a : ENNReal) (ha : 0 < a)
    (hclassical : paperClassicalAgreement D a) (hadd : AdditiveDivergence D) (hdpi : paperDataProcessing D)
    {I : Type} [Fintype I] [DecidableEq I] [Nonempty I] (r s : paperMat I)
    (hr : r.PosSemidef) (hs : s.PosSemidef) (hne : r ≠ 0) (hc : paperSupported r s) :
    paperRenyi true r s a ≤ D.value I r s ∧
    Tendsto (fun n : ℕ => ((1/(n:ℝ):ℝ):EReal)*paperRenyi false
      (paperPinch (paperTensorPower s n) (paperTensorPower r n)) (paperTensorPower s n) a)
      atTop (nhds (paperRenyi true r s a)) := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P07

end
Source
Quantum Information Processing with Finite Resources -- Mathematical Foundations, arXiv:1504.00233v5, PDF p. 58, §4.2.2 (4.30)–(4.31): minimality bound
Read-back

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

Let DDD assign an extended real number DI(X,Y)D_I(X,Y)DI​(X,Y) to every pair of square complex matrices on each finite index type III equipped with decidable equality, and let a∈(0,∞]a\in(0,\infty]a∈(0,∞]. The divergence expressions used here have the following literal definitions. Write X⪯YX\preceq YX⪯Y when Y−XY-XY−X is positive semidefinite, and say that XXX is supported in YYY when ker⁡Y⊆ker⁡X\ker Y\subseteq\ker XkerY⊆kerX. Let Pt(X)P_t(X)Pt​(X) be continuous functional calculus applied to the scalar function that is 000 at 000 and xtx^txt elsewhere, and let L(X)L(X)L(X) be functional calculus applied to log⁡2x\log_2 xlog2​x. Define

QP(X,Y;t)=Re⁡tr⁡ ⁣(Pt(X)P1−t(Y)),QS(X,Y;t)=Re⁡tr⁡ ⁣(Pt ⁣(P1−t2t(Y)XP1−t2t(Y))).Q_{\mathrm P}(X,Y;t)=\operatorname{Re}\operatorname{tr}\!\left(P_t(X)P_{1-t}(Y)\right),\qquad Q_{\mathrm S}(X,Y;t)=\operatorname{Re}\operatorname{tr}\!\left( P_t\!\left(P_{\frac{1-t}{2t}}(Y)X P_{\frac{1-t}{2t}}(Y)\right)\right).QP​(X,Y;t)=Retr(Pt​(X)P1−t​(Y)),QS​(X,Y;t)=Retr(Pt​(P2t1−t​​(Y)XP2t1−t​​(Y))).

Put ℓ(z)=−∞\ell(z)=-\inftyℓ(z)=−∞ if z=0z=0z=0, and ℓ(z)=log⁡2z\ell(z)=\log_2zℓ(z)=log2​z, embedded in the extended reals, otherwise. For either b=Pb=\mathrm Pb=P or b=Sb=\mathrm Sb=S, the finite positive order expression Fb(X,Y;t)F_b(X,Y;t)Fb​(X,Y;t) is, at t=1t=1t=1, +∞+\infty+∞ if XXX is not supported in YYY, and otherwise

Fb(X,Y;1)=Re⁡tr⁡ ⁣(X(L(X)−L(Y)))Re⁡tr⁡X.F_b(X,Y;1)= \frac{\operatorname{Re}\operatorname{tr}\!\left(X(L(X)-L(Y))\right)} {\operatorname{Re}\operatorname{tr}X}.Fb​(X,Y;1)=RetrXRetr(X(L(X)−L(Y)))​.

For t≠1t\ne1t=1, it is

Fb(X,Y;t)=1t−1 ℓ ⁣(Qb(X,Y;t)Re⁡tr⁡X)F_b(X,Y;t)=\frac{1}{t-1}\, \ell\!\left(\frac{Q_b(X,Y;t)}{\operatorname{Re}\operatorname{tr}X}\right)Fb​(X,Y;t)=t−11​ℓ(RetrXQb​(X,Y;t)​)

if either t<1t<1t<1 and Qb(X,Y;t)>0Q_b(X,Y;t)>0Qb​(X,Y;t)>0, or XXX is supported in YYY; it is +∞+\infty+∞ otherwise. For finite positive aaa, set Rb(X,Y;a)=Fb(X,Y;a)\mathcal R_b(X,Y;a)=F_b(X,Y;a)Rb​(X,Y;a)=Fb​(X,Y;a); at infinite order set

RS(X,Y;∞)=inf⁡{l∈R:X⪯2lY},RP(X,Y;∞)=sup⁡t>0FP(X,Y;t),\mathcal R_{\mathrm S}(X,Y;\infty) =\inf\{l\in\mathbb R:X\preceq2^lY\},\qquad \mathcal R_{\mathrm P}(X,Y;\infty) =\sup_{t>0}F_{\mathrm P}(X,Y;t),RS​(X,Y;∞)=inf{l∈R:X⪯2lY},RP​(X,Y;∞)=t>0sup​FP​(X,Y;t),

with these infima and suprema taken in the extended reals. Assume that: for every finite nonempty index type with decidable equality and all nonnegative real vectors p,qp,qp,q with p≠0p\ne0p=0, D(diag⁡p,diag⁡q)=RP(diag⁡p,diag⁡q;a)D(\operatorname{diag}p,\operatorname{diag}q)=\mathcal R_{\mathrm P}(\operatorname{diag}p,\operatorname{diag}q;a)D(diagp,diagq)=RP​(diagp,diagq;a); for all such index types I,JI,JI,J, all positive semidefinite X,YX,YX,Y on III and Z,WZ,WZ,W on JJJ, with X≠0X\ne0X=0 and Z≠0Z\ne0Z=0, one has DI×J(X⊗Z,Y⊗W)=DI(X,Y)+DJ(Z,W)D_{I\times J}(X\otimes Z,Y\otimes W)=D_I(X,Y)+D_J(Z,W)DI×J​(X⊗Z,Y⊗W)=DI​(X,Y)+DJ​(Z,W); and for every complex linear map EEE between matrix spaces on such types that preserves the complex trace of every matrix and is completely positive, DJ(E(X),E(Y))≤DI(X,Y)D_J(E(X),E(Y))\le D_I(X,Y)DJ​(E(X),E(Y))≤DI​(X,Y) for positive semidefinite X,YX,YX,Y with X≠0X\ne0X=0. Here complete positivity means that, for every natural number kkk, including 000, the map acting as EEE on each matrix block indexed by a pair from Fin⁡k\operatorname{Fin}kFink preserves positive semidefiniteness. Then, for every finite nonempty III with decidable equality and positive semidefinite matrices r,sr,sr,s on III, with r≠0r\ne0r=0 and ker⁡s⊆ker⁡r\ker s\subseteq\ker rkers⊆kerr, both

RS(r,s;a)≤DI(r,s)andlim⁡n→∞1n RP ⁣(Πs⊗n(r⊗n),s⊗n;a)=RS(r,s;a)\mathcal R_{\mathrm S}(r,s;a)\le D_I(r,s) \quad\text{and}\quad \lim_{n\to\infty}\frac1n\, \mathcal R_{\mathrm P}\!\left(\Pi_{s^{\otimes n}}(r^{\otimes n}),s^{\otimes n};a\right) =\mathcal R_{\mathrm S}(r,s;a)RS​(r,s;a)≤DI​(r,s)andn→∞lim​n1​RP​(Πs⊗n​(r⊗n),s⊗n;a)=RS​(r,s;a)

hold, the limit being in the topology of the extended reals. The tensor power has entries (X⊗n)x,y=∏i=0n−1Xxi,yi(X^{\otimes n})_{x,y}=\prod_{i=0}^{n-1}X_{x_i,y_i}(X⊗n)x,y​=∏i=0n−1​Xxi​,yi​​, and ΠH(K)=∑λPλKPλ\Pi_H(K)=\sum_{\lambda}P_\lambda K P_\lambdaΠH​(K)=∑λ​Pλ​KPλ​, where the sum runs over the distinct eigenvalues of the Hermitian matrix HHH and PλP_\lambdaPλ​ is functional calculus applied to the indicator of {λ}\{\lambda\}{λ}. No trace normalization is assumed for r,sr,sr,s. Order a=∞a=\inftya=∞ is included; the sequence is defined also at n=0n=0n=0, using the empty tensor product and the real convention 1/0=01/0=01/0=0. All scalar divisions and functional calculus expressions use their total definitions, including inverse 0=00=00=0, while ℓ(0)\ell(0)ℓ(0) is explicitly −∞-\infty−∞.

Human review
  • Endorsed by Shuze Chen · Oct 31, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 31, 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