Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Strong converse for the classical capacity of entanglement-breaking and Hadamard channels via a sandwiched Renyi relative entropy

Definition
QAlgorithms_Papers_Independent_Support50_P19

by Goku · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

The objects the formalized statements of Strong converse for the classical capacity of entanglement-breaking and Hadamard channels via a sandwiched Renyi relative entropy (arXiv:1306.1586) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • classicalRenyiBits: PDF p.12: unnormalized classical Renyi divergence in bits, for alpha>1. Dependency50.P28.classicalD uses natural logs and a different order interface; its value cannot be substituted here. Test support before a negative power.
  • filtered: PDF p.7, Remark 1: the positive filter is not normalized.
  • renyiD: PDF pp.17,19: true selects sandwiched and false selects traditional Petz entropy.
  • renyiEnsembleInformation: PDF p.17, Definition 12: Renyi cq information, before using the block formula. Dependency50.P07.ensembleInformation instead denotes ordinary Holevo information.
  • channelInformation: PDF p.17, Definition 12: supremum over all finite mixed-state input ensembles.
  • fixedRadius: PDF p.24, Definition 21: fixed-center radius, permitting infinite radius.
  • informationRadius: PDF p.17, Definition 13: minimize over output states AFTER maximizing over inputs.
  • blockInformation: PDF p.17, (14): zero-weight unsupported blocks contribute zero in ENNReal.
  • InformationExtrema: PDF pp.17–19: witnesses for the maxima and minima in Definitions 12 and 13. The semantic extrema are not replaced by unattained suprema or infima.
  • outputWeightedNorm: PDF p.19: alpha-norm of a positive output with weight sigma^((1-alpha)/alpha). On the compatible support domain this is the weighted Schatten norm in the source.
  • CompatibleCenter: PDF pp.19,25: admissible centers for the FINITE weighted-norm expression. Singular centers containing every output support are included.
  • normRadiusBefore: PDF p.19, (16): min max log norm, with the inverse-support domain explicit.
  • normRadiusAfter: PDF p.19, (16): min log max norm; finite-dimensional compatible centers give nonempty bounded norm sets, with positive supremum on channels.
  • NormExtrema: PDF p.19: attainment of the finite-domain norm minima and maxima in (16).
  • expTwo: PDF pp.20,25: base-two exponential on extended reals, retaining BOTH endpoints.
  • codingBound: PDF p.20, (17): success upper bound at one specified normalized radius.
  • renyiNu: PDF p.22: the Renyi comparison constant nu in (20), expressed without converting possibly infinite divergences. Claimed50.P48.nu is an output norm, not this two-state constant. On supported states, these traces are 2^(D_3/2 / 2) and 2^(-D_1/2 / 2).
  • CenterControl: PDF p.22: an attained ordinary information-radius center and a uniform finite renyiNu bound. These are witnesses in the conclusions below, never assumed approximation theorems.
  • ExponentialAt: PDF p.24: exponential strong-converse THRESHOLD, without identifying it with capacity.
  • InteriorEB: PDF p.25: interior relative to the finite-dimensional affine TP/HP map space. Entrywise coefficient balls give its usual topology; no CP constraint is placed on candidate perturbations before testing membership in the EB-channel set.
  • InteriorHadamard: PDF p.25: the EB interior condition is on a complement of this SAME channel.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_MatrixMapPositivity
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P09
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P56
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P27
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P48
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P35
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P26

/-! Wilde–Winter–Yang, arXiv:1306.1586v4. All citations are PDF pages; logs are in bits.
The imported paper modules are pinned accepted libraries. In particular, their spectral
calculus, physical codes, tensor channels, and covariance interface are reused. The
unnormalized divergences of Dependency50.P09 and Q of Information100.P56 implement
Definition 4; the trace-normalized divergence of Dependency50.P35 is not substituted.
All new definitions below are local to this paper. Compilation is pending batch handoff. -/

noncomputable section

open scoped Matrix ComplexOrder ComplexConjugate ENNReal
open Filter Topology

namespace QAlgorithms.Papers.Independent.Support50.P19

attribute [local instance] Classical.propDecidable Classical.decEq

open QAlgorithms.Papers.Independent.Dependency50.P35
  (tr Dominates spectralPower elog elogNN rawTrace cq)
open QAlgorithms.Papers.Independent.Dependency50.P09 (sandwichedD petzD relativeD)
open QAlgorithms.Papers.Independent.Information100.P56 (sandwichedTrace)
open QAlgorithms.Papers.Independent.Information100.P27 (average)
open QAlgorithms.Papers.Independent.Information100.P67
  (IsIsometry receiverChannel complementaryChannel)
open QAlgorithms.Papers.Independent.Claimed50.P48 (FiniteKrausCP)
open QAlgorithms.Papers.Independent.Dependency50.P23
  (UACode uaCodeSuccess ordinaryClassicalCapacity IsEB IsHadamard)
open QAlgorithms.Papers.Independent.Information100.P08 (Ensemble)
open QAlgorithms.Papers.Independent.Information100.P71 (channelPower)
open QAlgorithms.Papers.Independent.Information100.P62 (tensorChannel)
open QAlgorithms.Papers.Independent.Information100.P63 (tensorPower)
open QAlgorithms.Papers.Independent.Claimed50.P26
  (CovariantIrreducible minOutput UnassistedExponentialConverse)

/-- PDF p.12: unnormalized classical Renyi divergence in bits, for alpha>1.
Dependency50.P28.classicalD uses natural logs and a different order interface;
its value cannot be substituted here. Test support before a negative power. -/
def classicalRenyiBits {I : Type} [Fintype I]
    (a : ℝ) (p q : I → ℝ) : EReal :=
  if ∀ i, q i = 0 → p i = 0 then
    ((1 / (a - 1) : ℝ) : EReal) *
      elog (∑ i, if p i = 0 then 0 else p i ^ a * q i ^ (1 - a))
  else ⊤

/-- PDF p.7, Remark 1: the positive filter is not normalized. -/
def filtered {I J : Type} [Fintype J] (Y : Matrix J J ℂ) (N : Superop I J) : Superop I J :=
  fun X => Y * N X * Y

/-- PDF pp.17,19: true selects sandwiched and false selects traditional Petz entropy. -/
def renyiD {I : Type} [Fintype I] [DecidableEq I]
    (sandwiched : Bool) (a : ℝ) (A B : Matrix I I ℂ) : EReal :=
  if sandwiched then sandwichedD a A B else petzD a A B

/-- PDF p.17, Definition 12: Renyi cq information, before using the block formula.
Dependency50.P07.ensembleInformation instead denotes ordinary Holevo information. -/
def renyiEnsembleInformation {I : Type} [Fintype I] [DecidableEq I] {m : ℕ}
    (s : Bool) (a : ℝ) (p : Fin m → ℝ) (rho : Fin m → Matrix I I ℂ) : EReal :=
  ⨅ sigma : {X : Matrix I I ℂ // IsDensityMatrix X}, renyiD s a (cq p rho) (cq p (fun _ => sigma.1))

/-- PDF p.17, Definition 12: supremum over all finite mixed-state input ensembles. -/
def channelInformation {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (s : Bool) (a : ℝ) (N : Superop I J) : EReal :=
  ⨆ e : Ensemble I, renyiEnsembleInformation s a e.prob (fun x => N (e.state x))

/-- PDF p.24, Definition 21: fixed-center radius, permitting infinite radius. -/
def fixedRadius {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (s : Bool) (a : ℝ) (N : Superop I J) (sigma : Matrix J J ℂ) : EReal :=
  ⨆ rho : {X : Matrix I I ℂ // IsDensityMatrix X}, renyiD s a (N rho.1) sigma

/-- PDF p.17, Definition 13: minimize over output states AFTER maximizing over inputs. -/
def informationRadius {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (s : Bool) (a : ℝ) (N : Superop I J) : EReal :=
  ⨅ sigma : {X : Matrix J J ℂ // IsDensityMatrix X}, fixedRadius s a N sigma.1

/-- PDF p.17, (14): zero-weight unsupported blocks contribute zero in ENNReal. -/
def blockInformation {I : Type} [Fintype I] [DecidableEq I] {m : ℕ}
    (a : ℝ) (p : Fin m → ℝ) (rho : Fin m → Matrix I I ℂ) : EReal :=
  ⨅ sigma : {X : Matrix I I ℂ // IsDensityMatrix X}, ((1 / (a - 1) : ℝ) : EReal) *
    elogNN (∑ x, ENNReal.ofReal (p x) * sandwichedTrace a (rho x) sigma.1)

/-- PDF pp.17–19: witnesses for the maxima and minima in Definitions 12 and 13.
The semantic extrema are not replaced by unattained suprema or infima. -/
def InformationExtrema {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (s : Bool) (a : ℝ) (N : Superop I J) : Prop :=
  (∃ e : Ensemble I, channelInformation s a N =
    renyiEnsembleInformation s a e.prob (fun x => N (e.state x))) ∧
  (∃ sigma : {X : Matrix J J ℂ // IsDensityMatrix X}, informationRadius s a N = fixedRadius s a N sigma.1) ∧
  (∀ sigma : {X : Matrix J J ℂ // IsDensityMatrix X}, ∃ rho : {X : Matrix I I ℂ // IsDensityMatrix X},
    fixedRadius s a N sigma.1 = renyiD s a (N rho.1) sigma.1)

/-- PDF p.19: alpha-norm of a positive output with weight sigma^((1-alpha)/alpha).
On the compatible support domain this is the weighted Schatten norm in the source. -/
def outputWeightedNorm {I : Type} [Fintype I] [DecidableEq I]
    (a : ℝ) (A sigma : Matrix I I ℂ) : ℝ :=
  (rawTrace a A sigma) ^ (1 / a)

/-- PDF pp.19,25: admissible centers for the FINITE weighted-norm expression.
Singular centers containing every output support are included. -/
def CompatibleCenter {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (N : Superop I J) (sigma : Matrix J J ℂ) : Prop :=
  IsDensityMatrix sigma ∧ ∀ rho : {X : Matrix I I ℂ // IsDensityMatrix X}, Dominates sigma (N rho.1)

/-- PDF p.19, (16): min max log norm, with the inverse-support domain explicit. -/
def normRadiusBefore {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (a : ℝ) (N : Superop I J) : EReal :=
  ⨅ sigma : {S : Matrix J J ℂ // CompatibleCenter N S},
    ⨆ rho : {X : Matrix I I ℂ // IsDensityMatrix X}, ((a / (a - 1) : ℝ) : EReal) *
      elog (outputWeightedNorm a (N rho.1) sigma.1)

/-- PDF p.19, (16): min log max norm; finite-dimensional compatible centers give
nonempty bounded norm sets, with positive supremum on channels. -/
def normRadiusAfter {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (a : ℝ) (N : Superop I J) : EReal :=
  ⨅ sigma : {S : Matrix J J ℂ // CompatibleCenter N S},
    ((a / (a - 1) : ℝ) : EReal) *
      elog (sSup (Set.range (fun rho : {X : Matrix I I ℂ // IsDensityMatrix X} => outputWeightedNorm a (N rho.1) sigma.1)))

/-- PDF p.19: attainment of the finite-domain norm minima and maxima in (16). -/
def NormExtrema {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (a : ℝ) (N : Superop I J) : Prop :=
  (∃ sigma : {S : Matrix J J ℂ // CompatibleCenter N S},
    normRadiusAfter a N = ((a / (a - 1) : ℝ) : EReal) *
      elog (sSup (Set.range (fun rho : {X : Matrix I I ℂ // IsDensityMatrix X} => outputWeightedNorm a (N rho.1) sigma.1)))) ∧
  (∀ sigma : {S : Matrix J J ℂ // CompatibleCenter N S}, ∃ rho : {X : Matrix I I ℂ // IsDensityMatrix X},
    sSup (Set.range (fun r : {X : Matrix I I ℂ // IsDensityMatrix X} => outputWeightedNorm a (N r.1) sigma.1)) =
      outputWeightedNorm a (N rho.1) sigma.1)

/-- PDF pp.20,25: base-two exponential on extended reals, retaining BOTH endpoints. -/
def expTwo (x : EReal) : ℝ≥0∞ :=
  if x = ⊤ then ⊤ else if x = ⊥ then 0 else ENNReal.ofReal ((2 : ℝ) ^ x.toReal)

/-- PDF p.20, (17): success upper bound at one specified normalized radius. -/
def codingBound (n : ℕ) (a R : ℝ) (v : EReal) : ℝ≥0∞ :=
  expTwo (-(((n : ℝ) * (a - 1) / a : ℝ) : EReal) * ((R : EReal) - v))

/-- PDF p.22: the Renyi comparison constant nu in (20), expressed without
converting possibly infinite divergences. Claimed50.P48.nu is an output norm,
not this two-state constant.
On supported states, these traces are 2^(D_3/2 / 2) and 2^(-D_1/2 / 2). -/
def renyiNu {I : Type} [Fintype I] [DecidableEq I] (rho sigma : Matrix I I ℂ) : ℝ :=
  tr (spectralPower rho (3 / 2 : ℝ) * spectralPower sigma (-1 / 2 : ℝ)) +
    tr (spectralPower rho (1 / 2 : ℝ) * spectralPower sigma (1 / 2 : ℝ)) + 1

/-- PDF p.22: an attained ordinary information-radius center and a uniform finite renyiNu bound.
These are witnesses in the conclusions below, never assumed approximation theorems. -/
def CenterControl {I J : Type} [Fintype I] [Fintype J] [DecidableEq J]
    (N : Superop I J) (sigma : Matrix J J ℂ) (c : ℝ) : Prop :=
  CompatibleCenter N sigma ∧ 1 < c ∧
  (⨆ rho : {X : Matrix I I ℂ // IsDensityMatrix X}, relativeD (N rho.1) sigma) = ((QAlgorithms.Papers.Independent.Claimed50.P26.channelChi 1 N) : EReal) ∧
  ∀ rho : {X : Matrix I I ℂ // IsDensityMatrix X}, renyiNu (N rho.1) sigma ≤ c

/-- PDF p.24: exponential strong-converse THRESHOLD, without identifying it with capacity. -/
def ExponentialAt {I J : Type} [Fintype I] [Fintype J] [DecidableEq I] [DecidableEq J]
    (N : Superop I J) (threshold : ℝ) : Prop :=
  ∀ R : ℝ, threshold < R → ∃ c : ℝ, 0 < c ∧ ∃ n0 : ℕ, 0 < n0 ∧
    ∀ n : ℕ, n0 ≤ n → ∀ M : ℕ, 0 < M → R ≤ (Real.logb 2 (M : ℝ) / (n : ℝ)) →
      ∀ K : UACode (Fin n → I) (Fin n → J) M,
        uaCodeSuccess (channelPower N n) K ≤ (2 : ℝ) ^ (-(c * (n : ℝ)))

/-- PDF p.25: interior relative to the finite-dimensional affine TP/HP map space.
Entrywise coefficient balls give its usual topology; no CP constraint is placed on
candidate perturbations before testing membership in the EB-channel set. -/
def InteriorEB {I J : Type} [Fintype I] [Fintype J] [DecidableEq I] [DecidableEq J]
    (N : Superop I J) : Prop :=
  IsChannel N ∧ IsEB N ∧
  ∃ e : ℝ, 0 < e ∧ ∀ L : Matrix I I ℂ →ₗ[ℂ] Matrix J J ℂ,
    (∀ X, (L X).trace = X.trace) →
    (∀ X, X.IsHermitian → (L X).IsHermitian) →
    (∀ i j k l, ‖L (Matrix.single i j 1) k l - N (Matrix.single i j 1) k l‖ < e) →
    Entanglement.IsCompletelyPositive L ∧ IsEB L

/-- PDF p.25: the EB interior condition is on a complement of this SAME channel. -/
def InteriorHadamard {I J : Type} [Fintype I] [Fintype J] [DecidableEq I] [DecidableEq J]
    (N : Superop I J) : Prop :=
  ∃ k : ℕ, 0 < k ∧ ∃ V : Matrix (J × Fin k) I ℂ,
    IsIsometry V ∧ receiverChannel V = N ∧ InteriorEB (complementaryChannel V)

-- Small sanity checks pin support failure, zero operators, zero weights and the code boundary.
example {I : Type} [Fintype I] [DecidableEq I] (a : ℝ) (A B : Matrix I I ℂ)
    (h : ¬ Dominates B A) : sandwichedTrace a A B = ⊤ := by
  exact if_neg h

example {I J : Type} [Fintype I] [Fintype J] [DecidableEq I] [DecidableEq J] :
    FiniteKrausCP (fun _ : Matrix I I ℂ => (0 : Matrix J J ℂ)) := by
  refine ⟨0, Fin.elim0, ?_⟩
  funext X
  simp [krausApply]

example {I J : Type} [Fintype J] (N : Superop I J) (X : Matrix I I ℂ) :
    filtered 0 N X = 0 := by simp [filtered]

example : classicalRenyiBits 2 (fun _ : Fin 1 => (1 : ℝ)) (fun _ => 0) = ⊤ := by
  simp [classicalRenyiBits]

example : Real.logb 2 (1 : ℝ) / (1 : ℝ) = 0 := by simp [Real.logb]

example : binEntropy2 0 = 0 := by simp [binEntropy2]

example (x : ℝ≥0∞) : ENNReal.ofReal 0 * x = 0 := by simp

example : expTwo ⊤ = ⊤ ∧ expTwo ⊥ = 0 := by simp [expTwo]

variable {I J K L : Type} [Fintype I] [Fintype J] [Fintype K] [Fintype L]
variable [DecidableEq I] [DecidableEq J] [DecidableEq K] [DecidableEq L]

end QAlgorithms.Papers.Independent.Support50.P19
Source
Strong converse for the classical capacity of entanglement-breaking and Hadamard channels via a sandwiched Renyi relative entropy, arXiv:1306.1586v4, https://arxiv.org/abs/1306.1586v4
Read-back

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

QAlgorithms.Papers.Independent.Support50.P19.CenterControl

For finite types I,JI,JI,J, with decidable equality on JJJ, any function NNN from complex matrices on III to complex matrices on JJJ, a matrix σ\sigmaσ on JJJ, and a real ccc, this predicate requires that σ\sigmaσ be positive semidefinite of trace 111, that ker⁡σ⊆ker⁡N(ρ)\ker\sigma\subseteq\ker N(\rho)kerσ⊆kerN(ρ) for every positive semidefinite trace-one input ρ\rhoρ, and that c>1c>1c>1. It additionally requires

sup⁡ρ≥0, Tr⁡ρ=1D(N(ρ),σ)=χ1(N),Re⁡Tr⁡(P3/2(N(ρ))P−1/2(σ))+Re⁡Tr⁡(P1/2(N(ρ))P1/2(σ))+1≤c\sup_{\rho\ge0,\ \operatorname{Tr}\rho=1}D(N(\rho),\sigma)=\chi_1(N), \qquad \operatorname{Re}\operatorname{Tr} \bigl(P_{3/2}(N(\rho))P_{-1/2}(\sigma)\bigr) + \operatorname{Re}\operatorname{Tr} \bigl(P_{1/2}(N(\rho))P_{1/2}(\sigma)\bigr)+1\le cρ≥0, Trρ=1sup​D(N(ρ),σ)=χ1​(N),ReTr(P3/2​(N(ρ))P−1/2​(σ))+ReTr(P1/2​(N(ρ))P1/2​(σ))+1≤c

for every such ρ\rhoρ. Here D(X,Y)=+∞D(X,Y)=+\inftyD(X,Y)=+∞ if ker⁡Y⊈ker⁡X\ker Y\not\subseteq\ker XkerY⊆kerX, and otherwise

D(X,Y)=Re⁡Tr⁡(cfc⁡[x↦xlog⁡2x](X)−X cfc⁡[log⁡2](Y)),D(X,Y)=\operatorname{Re}\operatorname{Tr} \left(\operatorname{cfc}[x\mapsto x\log_2x](X) -X\,\operatorname{cfc}[\log_2](Y)\right),D(X,Y)=ReTr(cfc[x↦xlog2​x](X)−Xcfc[log2​](Y)),

regarded as an extended real number. The quantity χ1(N)\chi_1(N)χ1​(N) is the real supremum over all finite ensembles (px,ρx)(p_x,\rho_x)(px​,ρx​), with px≥0p_x\ge0px​≥0, ∑xpx=1\sum_xp_x=1∑x​px​=1, and every ρx\rho_xρx​ positive semidefinite of trace 111, of

H ⁣(N ⁣(∑xpxρx))−∑xpxH(N(ρx)),H(X)=−Re⁡Tr⁡cfc⁡[x↦xlog⁡2x](X).H\!\left(N\!\left(\sum_xp_x\rho_x\right)\right) -\sum_xp_xH(N(\rho_x)), \qquad H(X)=-\operatorname{Re}\operatorname{Tr} \operatorname{cfc}[x\mapsto x\log_2x](X).H(N(x∑​px​ρx​))−x∑​px​H(N(ρx​)),H(X)=−ReTrcfc[x↦xlog2​x](X).

For the powers in the inequality, PtP_tPt​ uses functional calculus with zero mapped to zero and nonzero xxx mapped to the total real power xtx^txt. The supremum on the left is in the extended reals, whereas χ1(N)\chi_1(N)χ1​(N) is embedded from R\mathbb RR. No channel assumption on NNN or nonemptiness assumption on I,JI,JI,J is included.

QAlgorithms.Papers.Independent.Support50.P19.CompatibleCenter

For finite types I,JI,JI,J, with decidable equality on JJJ, an arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, and a complex J×JJ\times JJ×J matrix σ\sigmaσ, this predicate means that σ\sigmaσ is positive semidefinite with trace one and that

ker⁡σ⊆ker⁡N(ρ)for every positive semidefinite I×I matrix ρ of trace one.\ker\sigma\subseteq\ker N(\rho) \qquad\text{for every positive semidefinite $I\times I$ matrix $\rho$ of trace one.}kerσ⊆kerN(ρ)for every positive semidefinite I×I matrix ρ of trace one.

No positivity, trace-preservation, or linearity assumption is made on NNN. If III is empty, the universally quantified condition is vacuous, so the predicate requires only that σ\sigmaσ be a density matrix. If JJJ is empty, no such σ\sigmaσ exists.

QAlgorithms.Papers.Independent.Support50.P19.ExponentialAt

For finite types I,JI,JI,J with decidable equality, any function NNN from complex matrices on III to complex matrices on JJJ, and any real threshold ttt, this predicate says that for every real R>tR>tR>t, there exist a real c>0c>0c>0 and a natural number n0>0n_0>0n0​>0 such that, for every natural n≥n0n\ge n_0n≥n0​, every natural M>0M>0M>0 satisfying R≤log⁡2M/nR\le\log_2M/nR≤log2​M/n, and every collection of positive semidefinite trace-one matrices ρ0,…,ρM−1\rho_0,\ldots,\rho_{M-1}ρ0​,…,ρM−1​ on InI^nIn together with positive semidefinite matrices D0,…,DM−1D_0,\ldots,D_{M-1}D0​,…,DM−1​ on JnJ^nJn satisfying ∑mDm=1\sum_mD_m=1∑m​Dm​=1,

1M∑m=0M−1Re⁡Tr⁡ ⁣(DmN[n](ρm))≤2−cn.\frac1M\sum_{m=0}^{M-1} \operatorname{Re}\operatorname{Tr}\!\left(D_mN^{[n]}(\rho_m)\right) \le 2^{-cn}.M1​m=0∑M−1​ReTr(Dm​N[n](ρm​))≤2−cn.

Here InI^nIn denotes functions from {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1} to III, and the function N[n]N^{[n]}N[n] is defined on all matrices by

N[n](X)y,z=∑x,x′∈InXx,x′∏i=0n−1N(Exi,xi′)yi,zi,N^{[n]}(X)_{y,z} =\sum_{x,x'\in I^n}X_{x,x'} \prod_{i=0}^{n-1} N(E_{x_i,x'_i})_{y_i,z_i},N[n](X)y,z​=x,x′∈In∑​Xx,x′​i=0∏n−1​N(Exi​,xi′​​)yi​,zi​​,

where Eu,vE_{u,v}Eu,v​ has its sole nonzero entry, equal to 111, at (u,v)(u,v)(u,v). The threshold need not be nonnegative, and no channel assumption on NNN or nonemptiness assumption on I,JI,JI,J is imposed. Since n0>0n_0>0n0​>0, the asserted bound only concerns positive block lengths; if the required collections of states and measurement matrices do not exist, the corresponding universal assertion is vacuous.

QAlgorithms.Papers.Independent.Support50.P19.InformationExtrema

For finite types I,JI,JI,J, with decidable equality on JJJ, Boolean sss, real aaa, and an arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, this predicate is the conjunction of three attainment statements. Let DI,DJ\mathcal D_I,\mathcal D_JDI​,DJ​ denote the positive semidefinite matrices of trace one on the indicated types. Define Ft(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)F_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Ft​(X)=cfc(x↦0 if x=0, otherwise xt)(X), and define Dsa(U,V)D_s^a(U,V)Dsa​(U,V) to be +∞+\infty+∞ if a>1a>1a>1 and ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU, and otherwise 1a−1L(Q)\frac1{a-1}L(Q)a−11​L(Q), where Q=Re⁡tr⁡Fa(F1−a2a(V)UF1−a2a(V))Q=\operatorname{Re}\operatorname{tr}F_a(F_{\frac{1-a}{2a}}(V)UF_{\frac{1-a}{2a}}(V))Q=RetrFa​(F2a1−a​​(V)UF2a1−a​​(V)) for s=trues=\mathrm{true}s=true, or Q=Re⁡tr⁡(Fa(U)F1−a(V))Q=\operatorname{Re}\operatorname{tr}(F_a(U)F_{1-a}(V))Q=Retr(Fa​(U)F1−a​(V)) for s=falses=\mathrm{false}s=false; LLL is the extended logarithm, equal to −∞-\infty−∞ on nonpositive arguments, log⁡2\log_2log2​ on finite positive arguments, and +∞+\infty+∞ at +∞+\infty+∞. For a finite ensemble e=(px,ρx)e=(p_x,\rho_x)e=(px​,ρx​), with px≥0p_x\ge0px​≥0, ∑xpx=1\sum_xp_x=1∑x​px​=1, and ρx∈DI\rho_x\in\mathcal D_Iρx​∈DI​, put C(e)=inf⁡σ∈DJDsa(diag⁡x(pxN(ρx)),diag⁡x(pxσ))C(e)=\inf_{\sigma\in\mathcal D_J}D_s^a(\operatorname{diag}_x(p_xN(\rho_x)),\operatorname{diag}_x(p_x\sigma))C(e)=infσ∈DJ​​Dsa​(diagx​(px​N(ρx​)),diagx​(px​σ)). The three assertions are

∃e: C(e)=sup⁡e′C(e′),∃σ∈DJ: inf⁡τ∈DJsup⁡ρ∈DIDsa(N(ρ),τ)=sup⁡ρ∈DIDsa(N(ρ),σ),\exists e:\ C(e)=\sup_{e'}C(e'),\qquad \exists\sigma\in\mathcal D_J:\ \inf_{\tau\in\mathcal D_J}\sup_{\rho\in\mathcal D_I}D_s^a(N(\rho),\tau)=\sup_{\rho\in\mathcal D_I}D_s^a(N(\rho),\sigma),∃e: C(e)=e′sup​C(e′),∃σ∈DJ​: τ∈DJ​inf​ρ∈DI​sup​Dsa​(N(ρ),τ)=ρ∈DI​sup​Dsa​(N(ρ),σ),

and

∀σ∈DJ ∃ρ∈DI: sup⁡τ∈DIDsa(N(τ),σ)=Dsa(N(ρ),σ).\forall\sigma\in\mathcal D_J\ \exists\rho\in\mathcal D_I:\ \sup_{\tau\in\mathcal D_I}D_s^a(N(\tau),\sigma)=D_s^a(N(\rho),\sigma).∀σ∈DJ​ ∃ρ∈DI​: τ∈DI​sup​Dsa​(N(τ),σ)=Dsa​(N(ρ),σ).

No uniqueness is asserted, NNN need not be linear or a channel, and aaa is unrestricted with total arithmetic. Empty types are allowed in the parameters, but the predicate cannot hold if either III or JJJ is empty, because its existential ensemble or density-matrix witness would be unavailable.

QAlgorithms.Papers.Independent.Support50.P19.InteriorEB

For finite types I,JI,JI,J with decidable equality and a function NNN from complex matrices on III to complex matrices on JJJ, this predicate requires a finite family EjE_jEj​ of matrices from CI\mathbb C^ICI to CJ\mathbb C^JCJ such that

∑jEj†Ej=1,N(X)=∑jEjXEj†for every X.\sum_jE_j^\dagger E_j=1,\qquad N(X)=\sum_jE_jXE_j^\dagger \quad\text{for every }X.j∑​Ej†​Ej​=1,N(X)=j∑​Ej​XEj†​for every X.

It also requires that, for every natural r>0r>0r>0 and every positive semidefinite trace-one matrix XXX on {0,…,r−1}×I\{0,\ldots,r-1\}\times I{0,…,r−1}×I, the matrix (idr⊗N)(X)(\mathrm{id}_r\otimes N)(X)(idr​⊗N)(X) be expressible as a finite sum ∑jwjAj⊗Bj\sum_jw_jA_j\otimes B_j∑j​wj​Aj​⊗Bj​, where wj≥0w_j\ge0wj​≥0 and Aj,BjA_j,B_jAj​,Bj​ are positive semidefinite trace-one matrices on the reference and output spaces, respectively; no condition ∑jwj=1\sum_jw_j=1∑j​wj​=1 is part of this requirement. Finally, there must exist a real ε>0\varepsilon>0ε>0 such that every complex-linear map LLL on the same matrix spaces which preserves the complex trace of every matrix, maps every Hermitian matrix to a Hermitian matrix, and satisfies

∣L(Ei,j)k,l−N(Ei,j)k,l∣<εfor all i,j∈I, k,l∈J\left|L(E_{i,j})_{k,l}-N(E_{i,j})_{k,l}\right|<\varepsilon \quad\text{for all }i,j\in I,\ k,l\in J∣L(Ei,j​)k,l​−N(Ei,j​)k,l​∣<εfor all i,j∈I, k,l∈J

has both of the following properties: (idr⊗L)(X)(\mathrm{id}_r\otimes L)(X)(idr​⊗L)(X) is positive semidefinite for every natural rrr, including r=0r=0r=0, and every positive semidefinite XXX; and for every r>0r>0r>0 and positive semidefinite trace-one XXX, it has a finite decomposition of the preceding form. Here Ei,jE_{i,j}Ei,j​ is the matrix unit, and the amplification acts on reference-indexed blocks by (idr⊗L)(X)(u,k),(v,l)=L(X(u,⋅),(v,⋅))k,l(\mathrm{id}_r\otimes L)(X)_{(u,k),(v,l)}=L(X_{(u,\cdot),(v,\cdot)})_{k,l}(idr​⊗L)(X)(u,k),(v,l)​=L(X(u,⋅),(v,⋅)​)k,l​. Empty finite types are not excluded.

QAlgorithms.Papers.Independent.Support50.P19.InteriorHadamard

For finite types I,JI,JI,J with decidable equality and any function NNN from complex matrices on III to complex matrices on JJJ, this predicate asserts the existence of a natural k>0k>0k>0 and a matrix V:CI→CJ×{0,…,k−1}V:\mathbb C^I\to\mathbb C^{J\times\{0,\ldots,k-1\}}V:CI→CJ×{0,…,k−1} such that V†V=1V^\dagger V=1V†V=1 and

N(X)j,j′=∑e=0k−1(VXV†)(j,e),(j′,e)for every X.N(X)_{j,j'}=\sum_{e=0}^{k-1}(VXV^\dagger)_{(j,e),(j',e)} \quad\text{for every }X.N(X)j,j′​=e=0∑k−1​(VXV†)(j,e),(j′,e)​for every X.

Define the other marginal map by

E(X)e,e′=∑j∈J(VXV†)(j,e),(j,e′).\mathcal E(X)_{e,e'}=\sum_{j\in J}(VXV^\dagger)_{(j,e),(j,e')}.E(X)e,e′​=j∈J∑​(VXV†)(j,e),(j,e′)​.

The predicate further requires that E\mathcal EE admit a finite representation E(X)=∑sKsXKs†\mathcal E(X)=\sum_sK_sXK_s^\daggerE(X)=∑s​Ks​XKs†​ with ∑sKs†Ks=1\sum_sK_s^\dagger K_s=1∑s​Ks†​Ks​=1, and that for every natural r>0r>0r>0 and every positive semidefinite trace-one XXX on the reference space times III, (idr⊗E)(X)(\mathrm{id}_r\otimes\mathcal E)(X)(idr​⊗E)(X) be a finite sum ∑swsAs⊗Bs\sum_sw_sA_s\otimes B_s∑s​ws​As​⊗Bs​ with nonnegative weights and positive semidefinite trace-one factors, without requiring the weights to sum to 111. Finally, there must be an ε>0\varepsilon>0ε>0 such that every complex-linear map LLL from matrices on III to matrices on {0,…,k−1}\{0,\ldots,k-1\}{0,…,k−1}, preserving trace and Hermitian matrices and satisfying ∣L(Ei,j)e,e′−E(Ei,j)e,e′∣<ε\left|L(E_{i,j})_{e,e'}-\mathcal E(E_{i,j})_{e,e'}\right|<\varepsilon∣L(Ei,j​)e,e′​−E(Ei,j​)e,e′​∣<ε for every choice of indices, is completely positive and has that same finite-decomposition property on every reference-extended density matrix. Complete positivity here means positivity of every finite reference amplification, including reference dimension zero; amplification applies the map separately to reference-indexed blocks. No nonemptiness assumption on I,JI,JI,J is included.

QAlgorithms.Papers.Independent.Support50.P19.NormExtrema

For finite types I,JI,JI,J, with decidable equality on JJJ, any real aaa, and any function NNN from complex matrices on III to complex matrices on JJJ, let CN\mathcal C_NCN​ be the set of positive semidefinite trace-one matrices σ\sigmaσ satisfying ker⁡σ⊆ker⁡N(ρ)\ker\sigma\subseteq\ker N(\rho)kerσ⊆kerN(ρ) for every positive semidefinite trace-one input ρ\rhoρ, and put

wa(ρ,σ)=[Re⁡Tr⁡Pa ⁣(P(1−a)/(2a)(σ) N(ρ) P(1−a)/(2a)(σ))]1/a,Sa(σ)=sup⁡ρ≥0, Tr⁡ρ=1wa(ρ,σ).w_a(\rho,\sigma)= \left[\operatorname{Re}\operatorname{Tr} P_a\!\left(P_{(1-a)/(2a)}(\sigma)\,N(\rho)\, P_{(1-a)/(2a)}(\sigma)\right)\right]^{1/a}, \qquad S_a(\sigma)=\sup_{\rho\ge0,\ \operatorname{Tr}\rho=1}w_a(\rho,\sigma).wa​(ρ,σ)=[ReTrPa​(P(1−a)/(2a)​(σ)N(ρ)P(1−a)/(2a)​(σ))]1/a,Sa​(σ)=ρ≥0, Trρ=1sup​wa​(ρ,σ).

Here P0(X)=1P_0(X)=1P0​(X)=1, while Pt(X)P_t(X)Pt​(X) for t≠0t\ne0t=0 is functional calculus with scalar function x↦0x\mapsto0x↦0 at x=0x=0x=0 and x↦xtx\mapsto x^tx↦xt elsewhere, using total real powers. This predicate asserts both that there exists σ∈CN\sigma\in\mathcal C_Nσ∈CN​ with

inf⁡τ∈CNaa−1elog⁡(Sa(τ))=aa−1elog⁡(Sa(σ)),\inf_{\tau\in\mathcal C_N} \frac{a}{a-1}\operatorname{elog}(S_a(\tau)) = \frac{a}{a-1}\operatorname{elog}(S_a(\sigma)),τ∈CN​inf​a−1a​elog(Sa​(τ))=a−1a​elog(Sa​(σ)),

and that, for every σ∈CN\sigma\in\mathcal C_Nσ∈CN​, there exists a positive semidefinite trace-one input ρ\rhoρ such that Sa(σ)=wa(ρ,σ)S_a(\sigma)=w_a(\rho,\sigma)Sa​(σ)=wa​(ρ,σ). The infimum and displayed products use extended-real arithmetic, with elog⁡(x)=log⁡2x\operatorname{elog}(x)=\log_2xelog(x)=log2​x for x>0x>0x>0 and −∞-\infty−∞ otherwise; the suprema defining SaS_aSa​ are total real suprema. Neither optimizer is asserted to be unique. There is no channel assumption on NNN or restriction on aaa; zero denominators use total division. Empty index types are included: an empty set of centers makes the first existence assertion false, and a center without any density-matrix inputs cannot satisfy the second existence assertion.

QAlgorithms.Papers.Independent.Support50.P19.blockInformation

For any finite type III with decidable equality, natural number mmm, real number aaa, arbitrary real weights pxp_xpx​, and arbitrary complex I×II\times II×I matrices ρx\rho_xρx​, indexed by 0≤x<m0\le x<m0≤x<m, this defines

inf⁡σ≥0, tr⁡σ=11a−1 L+ ⁣(∑x=0m−1max⁡(px,0) Ta(ρx,σ))\inf_{\sigma\ge0,\ \operatorname{tr}\sigma=1} \frac1{a-1}\,L_+\!\left(\sum_{x=0}^{m-1}\max(p_x,0)\,T_a(\rho_x,\sigma)\right)σ≥0, trσ=1inf​a−11​L+​(x=0∑m−1​max(px​,0)Ta​(ρx​,σ))

in the extended reals. The sum inside L+L_+L+​ is in [0,+∞][0,+\infty][0,+∞], and L+(0)=−∞L_+(0)=-\inftyL+​(0)=−∞, L+(+∞)=+∞L_+(+\infty)=+\inftyL+​(+∞)=+∞, while L+(z)=log⁡2zL_+(z)=\log_2zL+​(z)=log2​z for finite positive zzz. To define Ta(U,V)T_a(U,V)Ta​(U,V), return +∞+\infty+∞ if ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU, regardless of aaa; otherwise put S=cfc⁡(x↦0 if x=0, otherwise x(1−a)/(2a))(V)S=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^{(1-a)/(2a)})(V)S=cfc(x↦0 if x=0, otherwise x(1−a)/(2a))(V) and set

Ta(U,V)=max⁡ ⁣(Re⁡tr⁡(cfc⁡(x↦xa)(SUS)),0),T_a(U,V)=\max\!\left(\operatorname{Re}\operatorname{tr}\bigl(\operatorname{cfc}(x\mapsto x^a)(SUS)\bigr),0\right),Ta​(U,V)=max(Retr(cfc(x↦xa)(SUS)),0),

viewed in [0,+∞][0,+\infty][0,+∞]. There is no requirement that the weights sum to one or that the ρx\rho_xρx​ are density matrices; negative weights are replaced by zero. Zero times +∞+\infty+∞ in the inner sum is zero. The cases m=0m=0m=0, empty III, and arbitrary aaa, including 000 and 111, are included, with total powers and division; empty III makes the outer infimum +∞+\infty+∞.

QAlgorithms.Papers.Independent.Support50.P19.channelInformation

For finite types I,JI,JI,J, with decidable equality on JJJ, any Boolean sss, real aaa, and arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, this defines

sup⁡e inf⁡σ≥0, tr⁡σ=1Dsa(Re,Se,σ).\sup_e\ \inf_{\sigma\ge0,\ \operatorname{tr}\sigma=1}D_s^a(R_e,S_{e,\sigma}).esup​ σ≥0, trσ=1inf​Dsa​(Re​,Se,σ​).

Here eee ranges over all finite lists of density matrices ρx\rho_xρx​ on III, each positive semidefinite with trace one, with nonnegative real weights pxp_xpx​ summing to one; ReR_eRe​ and Se,σS_{e,\sigma}Se,σ​ are block diagonal, with respective blocks pxN(ρx)p_xN(\rho_x)px​N(ρx​) and pxσp_x\sigmapx​σ. To expand the divergence, put Ft(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)F_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Ft​(X)=cfc(x↦0 if x=0, otherwise xt)(X). For arbitrary U,VU,VU,V, Dsa(U,V)=+∞D_s^a(U,V)=+\inftyDsa​(U,V)=+∞ when a>1a>1a>1 and ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU; otherwise it is 1a−1L(Q)\frac1{a-1}L(Q)a−11​L(Q), where Q=Re⁡tr⁡Fa(F1−a2a(V)UF1−a2a(V))Q=\operatorname{Re}\operatorname{tr}F_a(F_{\frac{1-a}{2a}}(V)UF_{\frac{1-a}{2a}}(V))Q=RetrFa​(F2a1−a​​(V)UF2a1−a​​(V)) if s=trues=\mathrm{true}s=true, and Q=Re⁡tr⁡(Fa(U)F1−a(V))Q=\operatorname{Re}\operatorname{tr}(F_a(U)F_{1-a}(V))Q=Retr(Fa​(U)F1−a​(V)) if s=falses=\mathrm{false}s=false. The extended logarithm LLL is log⁡2\log_2log2​ on finite positive arguments, −∞-\infty−∞ on nonpositive arguments, and +∞+\infty+∞ at +∞+\infty+∞. No channel or linearity assumption on NNN, and no restriction on aaa, is imposed. An ensemble cannot have zero entries because its weights sum to one. If III is empty there are no such ensembles and the supremum is −∞-\infty−∞; if III is nonempty and JJJ empty, each inner infimum is over an empty set and is +∞+\infty+∞. All arithmetic and functional calculus are total.

QAlgorithms.Papers.Independent.Support50.P19.classicalRenyiBits

For any finite type III, any real number aaa, and any real-valued functions p,q:I→Rp,q:I\to\mathbb Rp,q:I→R, this defines an extended-real number as follows. If qi=0q_i=0qi​=0 implies pi=0p_i=0pi​=0 for every iii, put

Z=∑i∈I{0,pi=0,piaqi 1−a,pi≠0,Z=\sum_{i\in I}\begin{cases}0,&p_i=0,\\ p_i^a q_i^{\,1-a},&p_i\ne0,\end{cases}Z=i∈I∑​{0,pia​qi1−a​,​pi​=0,pi​=0,​

and return 1a−1 ℓ(Z)\frac1{a-1}\,\ell(Z)a−11​ℓ(Z), where ℓ(x)=log⁡2x\ell(x)=\log_2 xℓ(x)=log2​x when x>0x>0x>0 and ℓ(x)=−∞\ell(x)=-\inftyℓ(x)=−∞ otherwise. If that implication fails at any index, return +∞+\infty+∞. Neither ppp nor qqq is required to be nonnegative or normalized, and aaa is unrestricted; powers and division use the total real operations. In particular, 1/0=01/0=01/0=0, so at a=1a=1a=1 the value is 000 when the implication holds and +∞+\infty+∞ when it fails, using extended-real multiplication with 0⋅(±∞)=00\cdot(\pm\infty)=00⋅(±∞)=0. The empty type III is allowed: the implication then holds vacuously and the sum is zero.

QAlgorithms.Papers.Independent.Support50.P19.codingBound

For any natural number nnn, real numbers a,Ra,Ra,R, and extended real number vvv, form

x=−(n(a−1)a)(R−v)x=-\left(\frac{n(a-1)}a\right)(R-v)x=−(an(a−1)​)(R−v)

using real arithmetic for the coefficient n(a−1)/an(a-1)/an(a−1)/a, followed by extended-real arithmetic for the product. The definition returns +∞+\infty+∞ if x=+∞x=+\inftyx=+∞, 000 if x=−∞x=-\inftyx=−∞, and 2x2^x2x, embedded in [0,+∞][0,+\infty][0,+∞], if xxx is finite. No sign or finiteness restriction is imposed on a,R,va,R,va,R,v, and n=0n=0n=0 is included. Real division by zero is zero; consequently, n=0n=0n=0, a=0a=0a=0, or a=1a=1a=1 gives value 111, including when vvv is infinite, under the specified extended-real arithmetic.

QAlgorithms.Papers.Independent.Support50.P19.expTwo

For any extended real number xxx, this defines a value in [0,+∞][0,+\infty][0,+∞]: it is +∞+\infty+∞ when x=+∞x=+\inftyx=+∞, it is 000 when x=−∞x=-\inftyx=−∞, and it is the nonnegative extended-real embedding of the real number 2t2^t2t when xxx is the finite real number ttt.

QAlgorithms.Papers.Independent.Support50.P19.filtered

For any type III, any finite type JJJ, any complex matrix YYY indexed by J×JJ\times JJ×J, and any function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, this defines the function

X⟼Y N(X) Y.X\longmapsto Y\,N(X)\,Y.X⟼YN(X)Y.

There are no assumptions that NNN is linear or a channel, or that YYY is Hermitian, positive, or invertible. Both outside factors are YYY itself. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P19.fixedRadius

For finite types I,JI,JI,J, with decidable equality on JJJ, any Boolean sss, real aaa, arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, and arbitrary complex J×JJ\times JJ×J matrix σ\sigmaσ, this defines

sup⁡ρ≥0, tr⁡ρ=1Dsa(N(ρ),σ)\sup_{\rho\ge0,\ \operatorname{tr}\rho=1}D_s^a(N(\rho),\sigma)ρ≥0, trρ=1sup​Dsa​(N(ρ),σ)

in the extended reals. Explicitly, put Ft(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)F_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Ft​(X)=cfc(x↦0 if x=0, otherwise xt)(X). The value Dsa(U,V)D_s^a(U,V)Dsa​(U,V) is +∞+\infty+∞ if a>1a>1a>1 and ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU; otherwise it is 1a−1L(Q)\frac1{a-1}L(Q)a−11​L(Q), where Q=Re⁡tr⁡Fa(F1−a2a(V)UF1−a2a(V))Q=\operatorname{Re}\operatorname{tr}F_a(F_{\frac{1-a}{2a}}(V)UF_{\frac{1-a}{2a}}(V))Q=RetrFa​(F2a1−a​​(V)UF2a1−a​​(V)) for s=trues=\mathrm{true}s=true, and Q=Re⁡tr⁡(Fa(U)F1−a(V))Q=\operatorname{Re}\operatorname{tr}(F_a(U)F_{1-a}(V))Q=Retr(Fa​(U)F1−a​(V)) for s=falses=\mathrm{false}s=false. Here LLL is log⁡2\log_2log2​ on finite positive arguments, −∞-\infty−∞ on nonpositive arguments, and +∞+\infty+∞ at +∞+\infty+∞. The center σ\sigmaσ is not required to be positive or normalized, and NNN is not required to be linear or a channel. The order aaa is unrestricted and all operations are total. Empty III gives an empty supremum, hence −∞-\infty−∞; empty JJJ is also allowed.

QAlgorithms.Papers.Independent.Support50.P19.informationRadius

For finite types I,JI,JI,J, with decidable equality on JJJ, any Boolean sss, real aaa, and arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, this defines

inf⁡σ≥0, tr⁡σ=1 sup⁡ρ≥0, tr⁡ρ=1Dsa(N(ρ),σ).\inf_{\sigma\ge0,\ \operatorname{tr}\sigma=1}\ \sup_{\rho\ge0,\ \operatorname{tr}\rho=1}D_s^a(N(\rho),\sigma).σ≥0, trσ=1inf​ ρ≥0, trρ=1sup​Dsa​(N(ρ),σ).

Both extrema are taken in the extended reals. To specify the divergence, let Ft(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)F_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Ft​(X)=cfc(x↦0 if x=0, otherwise xt)(X). Set Dsa(U,V)=+∞D_s^a(U,V)=+\inftyDsa​(U,V)=+∞ if a>1a>1a>1 and ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU; otherwise set Dsa(U,V)=1a−1L(Q)D_s^a(U,V)=\frac1{a-1}L(Q)Dsa​(U,V)=a−11​L(Q), where Q=Re⁡tr⁡Fa(F1−a2a(V)UF1−a2a(V))Q=\operatorname{Re}\operatorname{tr}F_a(F_{\frac{1-a}{2a}}(V)UF_{\frac{1-a}{2a}}(V))Q=RetrFa​(F2a1−a​​(V)UF2a1−a​​(V)) for s=trues=\mathrm{true}s=true, and Q=Re⁡tr⁡(Fa(U)F1−a(V))Q=\operatorname{Re}\operatorname{tr}(F_a(U)F_{1-a}(V))Q=Retr(Fa​(U)F1−a​(V)) for s=falses=\mathrm{false}s=false. The function LLL equals log⁡2\log_2log2​ on finite positive arguments, −∞-\infty−∞ on nonpositive arguments, and +∞+\infty+∞ at +∞+\infty+∞. There is no assumption that NNN is linear or a channel, and aaa is unrestricted with total arithmetic. If JJJ is empty, the outer infimum is empty and equals +∞+\infty+∞; if JJJ is nonempty and III empty, every inner supremum is empty and the result is −∞-\infty−∞.

QAlgorithms.Papers.Independent.Support50.P19.normRadiusAfter

For finite types I,JI,JI,J, with decidable equality on JJJ, any real aaa, and any function NNN from complex matrices on III to complex matrices on JJJ, this defines the extended real number

inf⁡σ∈CN[(aa−1)elog⁡ ⁣(sup⁡ρ≥0, Tr⁡ρ=1[Re⁡Tr⁡Pa ⁣(P(1−a)/(2a)(σ) N(ρ) P(1−a)/(2a)(σ))]1/a)].\inf_{\sigma\in\mathcal C_N} \left[ \left(\frac{a}{a-1}\right) \operatorname{elog}\!\left( \sup_{\rho\ge0,\ \operatorname{Tr}\rho=1} \left[\operatorname{Re}\operatorname{Tr} P_a\!\left(P_{(1-a)/(2a)}(\sigma)\,N(\rho)\, P_{(1-a)/(2a)}(\sigma)\right)\right]^{1/a} \right)\right].σ∈CN​inf​[(a−1a​)elog(ρ≥0, Trρ=1sup​[ReTrPa​(P(1−a)/(2a)​(σ)N(ρ)P(1−a)/(2a)​(σ))]1/a)].

Here CN\mathcal C_NCN​ consists of positive semidefinite matrices σ\sigmaσ of trace 111 such that ker⁡σ⊆ker⁡N(ρ)\ker\sigma\subseteq\ker N(\rho)kerσ⊆kerN(ρ) for every positive semidefinite trace-one input ρ\rhoρ. The operation Pt(X)P_t(X)Pt​(X) is the identity matrix when t=0t=0t=0; otherwise it is the real continuous functional calculus applied to the function taking 000 to 000 and every nonzero xxx to the total real power xtx^txt. Also, elog⁡(x)=log⁡2x\operatorname{elog}(x)=\log_2xelog(x)=log2​x for x>0x>0x>0, and −∞-\infty−∞ otherwise. The inner supremum is taken in R\mathbb RR, using the total real supremum, which is 000 for an empty or unbounded-above set; the outer infimum and multiplication are in the extended reals. No channel assumption on NNN, restriction on aaa, or nonemptiness assumption on I,JI,JI,J is imposed. In particular, division by zero is evaluated as zero, and an empty set CN\mathcal C_NCN​ gives outer infimum +∞+\infty+∞.

QAlgorithms.Papers.Independent.Support50.P19.normRadiusBefore

For finite types I,JI,JI,J, with decidable equality on JJJ, a real number aaa, and an arbitrary function NNN from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, let CN\mathcal C_NCN​ consist of all positive semidefinite J×JJ\times JJ×J matrices σ\sigmaσ of trace one such that ker⁡σ⊆ker⁡N(ρ)\ker\sigma\subseteq\ker N(\rho)kerσ⊆kerN(ρ) for every positive semidefinite I×II\times II×I matrix ρ\rhoρ of trace one. Define G0(X)=1G_0(X)=1G0​(X)=1, and for t≠0t\ne0t=0 define Gt(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)G_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Gt​(X)=cfc(x↦0 if x=0, otherwise xt)(X). For each ρ,σ\rho,\sigmaρ,σ, put S=G1−a2a(σ)S=G_{\frac{1-a}{2a}}(\sigma)S=G2a1−a​​(σ). This definition returns

inf⁡σ∈CN sup⁡ρ≥0, tr⁡ρ=1aa−1 ℓ ⁣((Re⁡tr⁡Ga(SN(ρ)S))1/a),\inf_{\sigma\in\mathcal C_N}\ \sup_{\rho\ge0,\ \operatorname{tr}\rho=1} \frac{a}{a-1}\, \ell\!\left(\left(\operatorname{Re}\operatorname{tr}G_a(SN(\rho)S)\right)^{1/a}\right),σ∈CN​inf​ ρ≥0, trρ=1sup​a−1a​ℓ((RetrGa​(SN(ρ)S))1/a),

where ℓ(x)=log⁡2x\ell(x)=\log_2xℓ(x)=log2​x for x>0x>0x>0 and ℓ(x)=−∞\ell(x)=-\inftyℓ(x)=−∞ otherwise, and the extrema and multiplication are in the extended reals. The supremum is taken after multiplying each logarithm by the displayed coefficient. No channel or linearity assumption on NNN, and no restriction on aaa, is imposed; all real operations are total, including division by zero. If no compatible center exists, the outer infimum is +∞+\infty+∞. If III is empty and JJJ nonempty, compatibility is vacuous and the inner supremum is empty, giving −∞-\infty−∞; if JJJ is empty, there are no compatible centers and the value is +∞+\infty+∞.

QAlgorithms.Papers.Independent.Support50.P19.outputWeightedNorm

For any finite type III with decidable equality, any real number aaa, and arbitrary complex I×II\times II×I matrices A,σA,\sigmaA,σ, put

Gt(X)={1,t=0,cfc⁡(x↦0 if x=0, otherwise xt)(X),t≠0,S=G1−a2a(σ).G_t(X)= \begin{cases} 1,&t=0,\\ \operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X),&t\ne0, \end{cases} \qquad S=G_{\frac{1-a}{2a}}(\sigma).Gt​(X)={1,cfc(x↦0 if x=0, otherwise xt)(X),​t=0,t=0,​S=G2a1−a​​(σ).

This defines the real number

(Re⁡tr⁡Ga(SAS))1/a.\left(\operatorname{Re}\operatorname{tr}G_a(SAS)\right)^{1/a}.(RetrGa​(SAS))1/a.

The symbol 111 in G0G_0G0​ denotes the identity matrix, including on the kernel. There are no positivity, normalization, support-inclusion, or order restrictions. Powers, division, and functional calculus are total; in particular a=0a=0a=0 is permitted and gives value 111, since the final exponent is 1/0=01/0=01/0=0. Empty III is also permitted.

QAlgorithms.Papers.Independent.Support50.P19.renyiD

For any finite type III with decidable equality, a Boolean flag sss, a real number aaa, and arbitrary complex I×II\times II×I matrices A,BA,BA,B, this defines an extended-real number by selecting the following formulas. Write

Ft(X)=cfc⁡ ⁣(x↦{0,x=0,xt,x≠0)(X),Qs={Re⁡tr⁡Fa ⁣(F1−a2a(B) A F1−a2a(B)),s=true,Re⁡tr⁡ ⁣(Fa(A)F1−a(B)),s=false.F_t(X)=\operatorname{cfc}\!\left(x\mapsto \begin{cases}0,&x=0,\\x^t,&x\ne0\end{cases}\right)(X), \qquad Q_s=\begin{cases} \operatorname{Re}\operatorname{tr}F_a\!\left(F_{\frac{1-a}{2a}}(B)\,A\,F_{\frac{1-a}{2a}}(B)\right),&s=\mathrm{true},\\ \operatorname{Re}\operatorname{tr}\!\left(F_a(A)F_{1-a}(B)\right),&s=\mathrm{false}. \end{cases}Ft​(X)=cfc(x↦{0,xt,​x=0,x=0​)(X),Qs​={RetrFa​(F2a1−a​​(B)AF2a1−a​​(B)),Retr(Fa​(A)F1−a​(B)),​s=true,s=false.​

Here cfc⁡\operatorname{cfc}cfc is the total real continuous functional calculus. If a>1a>1a>1 and ker⁡B⊈ker⁡A\ker B\not\subseteq\ker AkerB⊆kerA, the value is +∞+\infty+∞; otherwise it is 1a−1L(Qs)\frac1{a-1}L(Q_s)a−11​L(Qs​), where L(+∞)=+∞L(+\infty)=+\inftyL(+∞)=+∞, L(x)=−∞L(x)=-\inftyL(x)=−∞ for x≤0x\le0x≤0, and L(x)=log⁡2xL(x)=\log_2xL(x)=log2​x for finite positive xxx. No positivity or normalization of the matrices, and no restriction on aaa, is imposed. All real divisions and powers are total; F0F_0F0​ assigns zero to the zero spectral value rather than assigning one there. At a=1a=1a=1 the result is 000, using 1/0=01/0=01/0=0 and 0⋅(±∞)=00\cdot(\pm\infty)=00⋅(±∞)=0. Empty III is included.

QAlgorithms.Papers.Independent.Support50.P19.renyiEnsembleInformation

For any finite type III with decidable equality, any natural number mmm, Boolean sss, real aaa, real weights p0,…,pm−1p_0,\ldots,p_{m-1}p0​,…,pm−1​, and arbitrary complex I×II\times II×I matrices ρ0,…,ρm−1\rho_0,\ldots,\rho_{m-1}ρ0​,…,ρm−1​, define block matrices R,SσR,S_\sigmaR,Sσ​ indexed by {0,…,m−1}×I\{0,\ldots,m-1\}\times I{0,…,m−1}×I through

R(x,i),(y,j)={px(ρx)ij,x=y,0,x≠y,(Sσ)(x,i),(y,j)={pxσij,x=y,0,x≠y.R_{(x,i),(y,j)}=\begin{cases}p_x(\rho_x)_{ij},&x=y,\\0,&x\ne y,\end{cases} \qquad (S_\sigma)_{(x,i),(y,j)}=\begin{cases}p_x\sigma_{ij},&x=y,\\0,&x\ne y.\end{cases}R(x,i),(y,j)​={px​(ρx​)ij​,0,​x=y,x=y,​(Sσ​)(x,i),(y,j)​={px​σij​,0,​x=y,x=y.​

The definition returns inf⁡σ≥0, tr⁡σ=1Dsa(R,Sσ)\inf_{\sigma\ge0,\ \operatorname{tr}\sigma=1}D_s^a(R,S_\sigma)infσ≥0, trσ=1​Dsa​(R,Sσ​) in the extended reals. Explicitly, let Ft(X)=cfc⁡(x↦0 if x=0, otherwise xt)(X)F_t(X)=\operatorname{cfc}(x\mapsto0\text{ if }x=0,\text{ otherwise }x^t)(X)Ft​(X)=cfc(x↦0 if x=0, otherwise xt)(X), let LLL be the extended logarithm with L(+∞)=+∞L(+\infty)=+\inftyL(+∞)=+∞, L(x)=−∞L(x)=-\inftyL(x)=−∞ for x≤0x\le0x≤0, and L(x)=log⁡2xL(x)=\log_2xL(x)=log2​x for finite positive xxx. Then Dsa(U,V)=+∞D_s^a(U,V)=+\inftyDsa​(U,V)=+∞ if a>1a>1a>1 and ker⁡V⊈ker⁡U\ker V\not\subseteq\ker UkerV⊆kerU; otherwise it is 1a−1L(Q)\frac1{a-1}L(Q)a−11​L(Q), with Q=Re⁡tr⁡Fa(F1−a2a(V)UF1−a2a(V))Q=\operatorname{Re}\operatorname{tr}F_a(F_{\frac{1-a}{2a}}(V)UF_{\frac{1-a}{2a}}(V))Q=RetrFa​(F2a1−a​​(V)UF2a1−a​​(V)) for s=trues=\mathrm{true}s=true, and Q=Re⁡tr⁡(Fa(U)F1−a(V))Q=\operatorname{Re}\operatorname{tr}(F_a(U)F_{1-a}(V))Q=Retr(Fa​(U)F1−a​(V)) for s=falses=\mathrm{false}s=false. The supplied weights need not be probabilities and the supplied matrices need not be density matrices. Both m=0m=0m=0 and empty III are allowed; for empty III there is no density matrix σ\sigmaσ, so the infimum is +∞+\infty+∞. The order aaa is unrestricted, with total arithmetic, including division by zero.

QAlgorithms.Papers.Independent.Support50.P19.renyiNu

For any finite type III with decidable equality and any two complex matrices ρ,σ\rho,\sigmaρ,σ indexed by III, this defines the real number

Re⁡Tr⁡(P3/2(ρ)P−1/2(σ))+Re⁡Tr⁡(P1/2(ρ)P1/2(σ))+1.\operatorname{Re}\operatorname{Tr} \bigl(P_{3/2}(\rho)P_{-1/2}(\sigma)\bigr) + \operatorname{Re}\operatorname{Tr} \bigl(P_{1/2}(\rho)P_{1/2}(\sigma)\bigr)+1.ReTr(P3/2​(ρ)P−1/2​(σ))+ReTr(P1/2​(ρ)P1/2​(σ))+1.

For each of these nonzero exponents ttt, Pt(X)P_t(X)Pt​(X) is real continuous functional calculus with scalar function equal to 000 at zero and to the total real power xtx^txt elsewhere. Thus the negative half-power also assigns zero at a zero spectral value. Neither matrix is assumed Hermitian, positive semidefinite, normalized, or nonzero; the total functional-calculus operation is used without those hypotheses. If III is empty, both traces are zero and the value is 111.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 2026

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

  • Endorsed by Gohan · Oct 29, 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