Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Weak locking capacity of quantum channels can be much larger than private capacity

Definition
QAlgorithms_Papers_Independent_Intakenext100_P63

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

prove2quantum

The objects the formalized statements of Weak locking capacity of quantum channels can be much larger than private capacity (arXiv:1403.6361) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • jointConditionalBits: PDF p.5: H(I|J); zero-probability columns contribute zero.
  • measuredEnsembleJoint: PDF p.5: the joint law p_i Tr(rho_i Q_j).
  • ensembleConditionalBits: PDF p.8: conditional Shannon entropy of an ensemble measured with Q.
  • accessibleEquivocation: PDF p.5: optimization over every finite POVM, including joint measurements. For normalized data the set is nonempty and contained in [0,log |I|].
  • optimizedEquivocation: PDF pp.5,7: optimization over all priors, including zero weights.
  • measurementWorstPrior: PDF p.8: opposite optimization order in the minimax formula.
  • taggedEffects: PDF p.8: random measurement mixing RETAINS the branch label.
  • averageState: PDF pp.7,9: ensemble average state.
  • supportInvSqrt: PDF pp.7,9: support inverse square root, with zero eigenvalues mapped to zero.
  • associatedEffects: PDF pp.7,9: associated effects, summing to the average state's support projector.
  • povmMinimumShannon: PDF p.8: minimum output entropy of an ordinary normalized POVM.
  • associatedMinimum: PDF pp.7,9–10: associated POVM minimum on supp(rho-bar), excluding the kernel. No invertibility hypothesis or arbitrary kernel completion is introduced.
  • stateMinBits: PDF p.9, Definition 7: min-entropy of a normalized density.
  • traceSmoothStateMinBits: PDF p.9: smoothing over NORMALIZED states at HALF trace norm distance.
  • conditionalMinBits: PDF p.9: conditional min-entropy, optimizing a normalized conditioning density. The accepted extended exponent optimization is finite on normalized states; toReal keeps its value. Zero/subnormalized off-domain values are not asserted here.
  • marginalMinEntropy: PDF p.9: the actual marginal reference, with its support inverse.
  • traceSmoothConditionalMinBits: PDF p.9: normalized bipartite smoothing, without fixing the marginal.
  • qcBranchDensity: PDF p.9: qc-state with the classical register SECOND.
  • classicalJoint: PDF p.9: a classical joint law represented by its diagonal density.
  • tensorEffects: PDF p.8: product measurement on arbitrary, possibly entangled, input states.
  • measuredState: PDF p.8: classical output density of a product measurement.
  • iidJoint: PDF p.9: IID labels, with an unrestricted collective measurement on E^n.
  • environment: PDF p.2: environment's partial trace of the SAME dilation.
  • KeyedBlockCode: PDF pp.2–3: keyed code, normalized mixed encodings and a POVM for each key. Both message and independent key are uniform. The key count is positive, and each accepted branch enforces positive message count, density signals and a POVM.
  • keyedAverageError: PDF p.2, (1): decoding error averaged over messages and independent secret key.
  • averagedSignal: PDF p.3: Eve sees key-averaged signal states.
  • WeakSecure: PDF p.2, (3): full l1 norm; EVERY POVM has ONE message-independent law.
  • PrivateSecure: PDF p.2, (2): full trace norm, with NO one-half factor.
  • messageRateBits: PDF p.3: rate per channel use, indexing positive lengths by n+1.
  • keyRate: PDF p.3: secret bits per channel use.
  • KeyedRateAtLeast: PDF p.3: limiting inferior rate at least R, allowing arbitrary rate slack.
  • ReliableKeyedSequence: PDF p.3: reliability and a sublinear preshared uniform secret.
  • KeyedAchievable: PDF p.3: operational nonnegative achievable rates for the selected security criterion.
  • keyedCapacity: PDF p.3: supremum of achievable nonnegative rates. For valid finite channels it is a nonempty bounded set; no feedback or free discussion is present in these codes.
  • CqComplement: PDF p.5, (5): cq complement in the specified computational input basis.
  • twoBasisVector: PDF p.11: computational and discrete Fourier basis vectors.
  • twoBasisDilation: PDF p.11, Example 12: 2d-dimensional input and receiver, d-dimensional environment.
  • upperInformation: PDF pp.5,14, (8): L_W^(u), over arbitrary finite mixed input ensembles. I_acc = H(p) - S_acc; this uses the same measurement optimization as above.
  • activatedReceiver: PDF p.15: N tensor id_k, retaining off-diagonal auxiliary blocks.
  • activatedEnvironment: PDF p.15: the noiseless factor leaks nothing; Eve receives only N^c.
  • activatedCapacity: PDF p.15: amortized capacity, with positive noiseless dimension and log k charge.
  • regularizedUpper: PDF p.15: regularized upper information, over positive block lengths.
  • arimotoOutput: PDF p.14: Arimoto conditional alpha entropy of measured ensemble branches.
  • constrainedRenyi: PDF p.14: constrained minimum over all finite decompositions with average sigma.
  • ConstrainedRenyiAdditivity: PDF p.14, (16): the GLOBAL conjecture, not an asserted fact or scalar placeholder. Fin enumerations cover every finite-dimensional system and outcome space.
  • IIDUncertainty: PDF p.14, (17): global uniform eventual IID uncertainty, with arbitrary rate slack. The threshold precedes Eve's collective POVM. Arbitrarily small fixed epsilon permits vanishing errors by diagonal choice. This predicate is a conjectural premise.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_Channel
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Defs_Ensemble
import Definitions.Def_QAlgorithms_Defs_Learning
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P04
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P27
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P63
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P67
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P71
import Definitions.Def_QAlgorithms_Papers_Independent_Bridge50_P47

/-! Winter, arXiv:1403.6361v3. Imported accepted primitives and paper-local operational
and entropy interfaces. Names distinguish Winter's keyed and normalized-trace conventions.
All logarithms are binary. Compilation is pending the batch mechanical handoff.
The three plan deferrals are documented in HARD.md. -/

noncomputable section

namespace QAlgorithms.Papers.Independent.Intakenext100.P63

open QAlgorithms Filter
open QAlgorithms.Papers.Independent.Information100.P63 (Mat)
open QAlgorithms.Papers.Independent.Information100.P04 (Measurement)
open QAlgorithms.Papers.Independent.Information100.P27 (shannon labelInformation)
open QAlgorithms.Papers.Independent.Information100.P71 (channelPower dilationMap SupportIncluded)
open QAlgorithms.Papers.Independent.Bridge50.P47 (SymmetricIndex universalSymmetricDilation)
open QAlgorithms.Learning (IsDistribution)
open scoped Matrix ComplexConjugate ComplexOrder

attribute [local instance] Classical.propDecidable Classical.decEq

/-- PDF p.5: H(I|J); zero-probability columns contribute zero. -/
def jointConditionalBits {I J : Type} [Fintype I] [Fintype J] (P : I → J → ℝ) : ℝ :=
  -∑ j, ∑ i, P i j * Real.logb 2 (P i j / ∑ k, P k j)

/-- PDF p.5: the joint law p_i Tr(rho_i Q_j). -/
def measuredEnsembleJoint {I E J : Type} [Fintype E] (p : I → ℝ) (rho : I → Mat E)
    (Q : J → Mat E) : I → J → ℝ :=
  fun i j => p i * effectProb (Q j) (rho i)

/-- PDF p.8: conditional Shannon entropy of an ensemble measured with Q. -/
def ensembleConditionalBits {I E J : Type} [Fintype I] [Fintype E] [Fintype J]
    (p : I → ℝ) (rho : I → Mat E) (Q : J → Mat E) : ℝ :=
  jointConditionalBits (measuredEnsembleJoint p rho Q)

/-- PDF p.5: optimization over every finite POVM, including joint measurements.
For normalized data the set is nonempty and contained in [0,log |I|]. -/
def accessibleEquivocation {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (p : I → ℝ) (rho : I → Mat E) : ℝ :=
  sInf {h | ∃ Q : Measurement E, h = ensembleConditionalBits p rho Q.effect}

/-- PDF pp.5,7: optimization over all priors, including zero weights. -/
def optimizedEquivocation {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (rho : I → Mat E) : ℝ :=
  sSup {h | ∃ p : I → ℝ, IsDistribution p ∧ h = accessibleEquivocation p rho}

/-- PDF p.8: opposite optimization order in the minimax formula. -/
def measurementWorstPrior {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (rho : I → Mat E) (Q : Measurement E) : ℝ :=
  sSup {h | ∃ p : I → ℝ, IsDistribution p ∧ h = ensembleConditionalBits p rho Q.effect}

/-- PDF p.8: random measurement mixing RETAINS the branch label. -/
def taggedEffects {E J K : Type} (lambda : ℝ) (Q : J → Mat E) (R : K → Mat E) :
    J ⊕ K → Mat E :=
  Sum.elim (fun j => (lambda : ℂ) • Q j) (fun k => ((1 - lambda : ℝ) : ℂ) • R k)

/-- PDF pp.7,9: ensemble average state. -/
def averageState {I E : Type} [Fintype I] (p : I → ℝ) (rho : I → Mat E) : Mat E :=
  ∑ i, (p i : ℂ) • rho i

/-- PDF pp.7,9: support inverse square root, with zero eigenvalues mapped to zero. -/
def supportInvSqrt {E : Type} [Fintype E] [DecidableEq E] (rho : Mat E) : Mat E :=
  cfc (fun x : ℝ => (Real.sqrt x)⁻¹) rho

/-- PDF pp.7,9: associated effects, summing to the average state's support projector. -/
def associatedEffects {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (p : I → ℝ) (rho : I → Mat E) (i : I) : Mat E :=
  supportInvSqrt (averageState p rho) * ((p i : ℂ) • rho i) *
    supportInvSqrt (averageState p rho)

/-- PDF p.8: minimum output entropy of an ordinary normalized POVM. -/
def povmMinimumShannon {E I : Type} [Fintype E] [Fintype I] (M : I → Mat E) : ℝ :=
  sInf {h | ∃ psi : Mat E, IsDensityMatrix psi ∧
    h = shannon (fun i => effectProb (M i) psi)}

/-- PDF pp.7,9–10: associated POVM minimum on supp(rho-bar), excluding the kernel.
No invertibility hypothesis or arbitrary kernel completion is introduced. -/
def associatedMinimum {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (p : I → ℝ) (rho : I → Mat E) : ℝ :=
  sInf {h | ∃ psi : Mat E, IsDensityMatrix psi ∧ SupportIncluded psi (averageState p rho) ∧
    h = shannon (fun i => effectProb (associatedEffects p rho i) psi)}

/-- PDF p.9, Definition 7: min-entropy of a normalized density. -/
def stateMinBits {A : Type} [Fintype A] [DecidableEq A] (rho : Mat A) : ℝ :=
  -Real.logb 2 (specNorm rho)

/-- PDF p.9: smoothing over NORMALIZED states at HALF trace norm distance. -/
def traceSmoothStateMinBits {A : Type} [Fintype A] [DecidableEq A]
    (eps : ℝ) (rho : Mat A) : ℝ :=
  sSup {h | ∃ r : Mat A, IsDensityMatrix r ∧ traceNorm (rho - r) / 2 ≤ eps ∧
    h = stateMinBits r}

/-- PDF p.9: conditional min-entropy, optimizing a normalized conditioning density.
The accepted extended exponent optimization is finite on normalized states; toReal
keeps its value. Zero/subnormalized off-domain values are not asserted here. -/
def conditionalMinBits {A B : Type} [Fintype A] [DecidableEq A]
    [Fintype B] [DecidableEq B] (rho : Mat (A × B)) : ℝ :=
  (QAlgorithms.Papers.Independent.Information100.P67.conditionalMinEntropy rho).toReal

/-- PDF p.9: the actual marginal reference, with its support inverse. -/
def marginalMinEntropy {A B : Type} [Fintype A] [DecidableEq A]
    [Fintype B] [DecidableEq B] (rho : Mat (A × B)) : ℝ :=
  let S := Matrix.kronecker (1 : Mat A) (supportInvSqrt (traceLeft rho));
  -Real.logb 2 (specNorm (S * rho * S))

/-- PDF p.9: normalized bipartite smoothing, without fixing the marginal. -/
def traceSmoothConditionalMinBits {A B : Type} [Fintype A] [DecidableEq A]
    [Fintype B] [DecidableEq B] (eps : ℝ) (rho : Mat (A × B)) : ℝ :=
  sSup {h | ∃ r : Mat (A × B), IsDensityMatrix r ∧ traceNorm (rho - r) / 2 ≤ eps ∧
    h = conditionalMinBits r}

/-- PDF p.9: qc-state with the classical register SECOND. -/
def qcBranchDensity {A J : Type} [DecidableEq J] (q : J → ℝ) (rho : J → Mat A) : Mat (A × J) :=
  fun x y => if x.2 = y.2 then (q x.2 : ℂ) * rho x.2 x.1 y.1 else 0

/-- PDF p.9: a classical joint law represented by its diagonal density. -/
def classicalJoint {I J : Type} [DecidableEq I] [DecidableEq J]
    (P : I → J → ℝ) : Mat (I × J) :=
  Matrix.diagonal (fun x => (P x.1 x.2 : ℂ))

/-- PDF p.8: product measurement on arbitrary, possibly entangled, input states. -/
def tensorEffects {E I : Type} (M : I → Mat E) (n : ℕ) :
    (Fin n → I) → Mat (Fin n → E) :=
  fun i x y => ∏ t, M (i t) (x t) (y t)

/-- PDF p.8: classical output density of a product measurement. -/
def measuredState {E I : Type} [Fintype E] [DecidableEq I]
    (M : I → Mat E) (n : ℕ) (psi : Mat (Fin n → E)) : Mat (Fin n → I) :=
  Matrix.diagonal (fun i => (effectProb (tensorEffects M n i) psi : ℂ))

/-- PDF p.9: IID labels, with an unrestricted collective measurement on E^n. -/
def iidJoint {I E : Type} [Fintype E] [DecidableEq E]
    (p : I → ℝ) (rho : I → Mat E) (n : ℕ) (Q : Measurement (Fin n → E)) :
    (Fin n → I) → Fin Q.outcomes → ℝ :=
  fun i j => (∏ t, p (i t)) * effectProb (Q.effect j) (fun x y => ∏ t, rho (i t) (x t) (y t))

/-- PDF p.2: environment's partial trace of the SAME dilation. -/
def environment {A B E : Type} [Fintype A] [Fintype B]
    (V : Matrix (B × E) A ℂ) : Superop A E :=
  fun rho => traceLeft (V * rho * Vᴴ)

/-- PDF pp.2–3: keyed code, normalized mixed encodings and a POVM for each key.
Both message and independent key are uniform. The key count is positive, and
each accepted branch enforces positive message count, density signals and a POVM. -/
structure KeyedBlockCode (A B : Type) [Fintype A] [Fintype B] [DecidableEq B] (n : ℕ) where
  messages : ℕ
  keys : ℕ
  keys_pos : 0 < keys
  branch : Fin keys →
    QAlgorithms.Papers.Independent.Information100.P63.Code (Fin n → A) (Fin n → B) messages

/-- PDF p.2, (1): decoding error averaged over messages and independent secret key. -/
def keyedAverageError {A B : Type} [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
    (N : Superop A B) {n : ℕ} (C : KeyedBlockCode A B n) : ℝ :=
  ((C.keys : ℝ) * C.messages)⁻¹ *
    ∑ k, ∑ m, (1 - effectProb ((C.branch k).measurement m) (channelPower N n ((C.branch k).word m)))

/-- PDF p.3: Eve sees key-averaged signal states. -/
def averagedSignal {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    {n : ℕ} (C : KeyedBlockCode A B n) (m : Fin C.messages) : Mat (Fin n → A) :=
  ((C.keys : ℂ)⁻¹) • ∑ k, (C.branch k).word m

/-- PDF p.2, (3): full l1 norm; EVERY POVM has ONE message-independent law. -/
def WeakSecure {A B E : Type} [Fintype A] [DecidableEq A] [Fintype B]
    [DecidableEq B] [Fintype E] [DecidableEq E]
    (Nc : Superop A E) {n : ℕ} (C : KeyedBlockCode A B n) (delta : ℝ) : Prop :=
  ∀ Q : Measurement (Fin n → E), ∃ omega : Fin Q.outcomes → ℝ, IsDistribution omega ∧
    (C.messages : ℝ)⁻¹ * ∑ m, ∑ j,
      |effectProb (Q.effect j) (channelPower Nc n (averagedSignal C m)) - omega j| ≤ delta

/-- PDF p.2, (2): full trace norm, with NO one-half factor. -/
def PrivateSecure {A B E : Type} [Fintype A] [DecidableEq A] [Fintype B]
    [DecidableEq B] [Fintype E] [DecidableEq E]
    (Nc : Superop A E) {n : ℕ} (C : KeyedBlockCode A B n) (delta : ℝ) : Prop :=
  ∃ omega : Mat (Fin n → E), IsDensityMatrix omega ∧
    (C.messages : ℝ)⁻¹ * ∑ m,
      traceNorm (channelPower Nc n (averagedSignal C m) - omega) ≤ delta

/-- PDF p.3: rate per channel use, indexing positive lengths by n+1. -/
def messageRateBits {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    {n : ℕ} (C : KeyedBlockCode A B (n + 1)) : ℝ :=
  Real.logb 2 C.messages / (n + 1)

/-- PDF p.3: secret bits per channel use. -/
def keyRate {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    {n : ℕ} (C : KeyedBlockCode A B (n + 1)) : ℝ :=
  Real.logb 2 C.keys / (n + 1)

/-- PDF p.3: limiting inferior rate at least R, allowing arbitrary rate slack. -/
def KeyedRateAtLeast {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    (C : (n : ℕ) → KeyedBlockCode A B (n + 1)) (R : ℝ) : Prop :=
  ∀ slack > (0 : ℝ), ∀ᶠ n in atTop, R - slack ≤ messageRateBits (C n)

/-- PDF p.3: reliability and a sublinear preshared uniform secret. -/
def ReliableKeyedSequence {A B : Type} [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
    (N : Superop A B) (C : (n : ℕ) → KeyedBlockCode A B (n + 1)) : Prop :=
  Tendsto (fun n => keyRate (C n)) atTop (nhds 0) ∧
  Tendsto (fun n => keyedAverageError N (C n)) atTop (nhds 0)

/-- PDF p.3: operational nonnegative achievable rates for the selected security criterion. -/
def KeyedAchievable {A B E : Type} [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
    [Fintype E] [DecidableEq E] (N : Superop A B) (Nc : Superop A E)
    (weak : Bool) (R : ℝ) : Prop :=
  0 ≤ R ∧ ∃ C : (n : ℕ) → KeyedBlockCode A B (n + 1), ReliableKeyedSequence N C ∧ KeyedRateAtLeast C R ∧
    ∀ eps > (0 : ℝ), ∀ᶠ n in atTop,
      if weak then WeakSecure Nc (C n) eps else PrivateSecure Nc (C n) eps

/-- PDF p.3: supremum of achievable nonnegative rates. For valid finite channels it
is a nonempty bounded set; no feedback or free discussion is present in these codes. -/
def keyedCapacity {A B E : Type} [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
    [Fintype E] [DecidableEq E] (N : Superop A B) (Nc : Superop A E) (weak : Bool) : ℝ :=
  sSup {R | KeyedAchievable N Nc weak R}

/-- PDF p.5, (5): cq complement in the specified computational input basis. -/
def CqComplement {A E : Type} [Fintype A] [DecidableEq A] [Fintype E]
    (Nc : Superop A E) (rho : A → Mat E) : Prop :=
  (∀ i, IsDensityMatrix (rho i)) ∧
  ∀ i j, Nc (Matrix.single i j 1) = if i = j then rho i else 0

/-- PDF p.11: computational and discrete Fourier basis vectors. -/
def twoBasisVector (d : ℕ) (b : Bool) (i x : Fin d) : ℂ :=
  if b then (Real.sqrt d : ℂ)⁻¹ *
    Complex.exp (2 * (Real.pi : ℂ) * Complex.I * (i.val : ℂ) * (x.val : ℂ) / (d : ℂ))
  else if x = i then 1 else 0

/-- PDF p.11, Example 12: 2d-dimensional input and receiver, d-dimensional environment. -/
def twoBasisDilation (d : ℕ) : Matrix ((Bool × Fin d) × Fin d) (Bool × Fin d) ℂ :=
  fun out inp => if out.1 = inp then twoBasisVector d inp.1 inp.2 out.2 else 0

/-- PDF pp.5,14, (8): L_W^(u), over arbitrary finite mixed input ensembles.
I_acc = H(p) - S_acc; this uses the same measurement optimization as above. -/
def upperInformation {A B E : Type} [Fintype A] [Fintype B] [DecidableEq B]
    [Fintype E] [DecidableEq E] (N : Superop A B) (Nc : Superop A E) : ℝ :=
  sSup {h | ∃ m : ℕ, ∃ p : Fin m → ℝ, ∃ rho : Fin m → Mat A,
    IsDistribution p ∧ (∀ i, IsDensityMatrix (rho i)) ∧
    h = labelInformation p (fun i => N (rho i)) - shannon p +
      accessibleEquivocation p (fun i => Nc (rho i))}

/-- PDF p.15: N tensor id_k, retaining off-diagonal auxiliary blocks. -/
def activatedReceiver {A B : Type} (N : Superop A B) (k : ℕ) :
    Superop (A × Fin k) (B × Fin k) :=
  fun rho x y => N (fun a a' => rho (a, x.2) (a', y.2)) x.1 y.1

/-- PDF p.15: the noiseless factor leaks nothing; Eve receives only N^c. -/
def activatedEnvironment {A E : Type} (Nc : Superop A E) (k : ℕ) :
    Superop (A × Fin k) E :=
  fun rho => Nc (traceRight rho)

/-- PDF p.15: amortized capacity, with positive noiseless dimension and log k charge. -/
def activatedCapacity {A B E : Type} [Fintype A] [DecidableEq A]
    [Fintype B] [DecidableEq B] [Fintype E] [DecidableEq E]
    (N : Superop A B) (Nc : Superop A E) : ℝ :=
  sSup {h | ∃ k : ℕ, 0 < k ∧ h =
    keyedCapacity (activatedReceiver N k) (activatedEnvironment Nc k) true - Real.logb 2 k}

/-- PDF p.15: regularized upper information, over positive block lengths. -/
def regularizedUpper {A B E : Type} [Fintype A] [DecidableEq A]
    [Fintype B] [DecidableEq B] [Fintype E] [DecidableEq E]
    (N : Superop A B) (Nc : Superop A E) : ℝ :=
  sSup {h | ∃ n : ℕ, 0 < n ∧ h = upperInformation (channelPower N n) (channelPower Nc n) / n}

/-- PDF p.14: Arimoto conditional alpha entropy of measured ensemble branches. -/
def arimotoOutput {E I J : Type} [Fintype E] [Fintype I] [Fintype J]
    (alpha : ℝ) (M : I → Mat E) (q : J → ℝ) (psi : J → Mat E) : ℝ :=
  -alpha / (alpha - 1) * Real.logb 2
    (∑ j, q j * (∑ i, (effectProb (M i) (psi j)) ^ alpha) ^ (1 / alpha))

/-- PDF p.14: constrained minimum over all finite decompositions with average sigma. -/
def constrainedRenyi {E I : Type} [Fintype E] [Fintype I]
    (alpha : ℝ) (M : I → Mat E) (sigma : Mat E) : ℝ :=
  sInf {h | ∃ r : ℕ, ∃ q : Fin r → ℝ, ∃ psi : Fin r → Mat E,
    IsDistribution q ∧ (∀ j, IsDensityMatrix (psi j)) ∧ averageState q psi = sigma ∧
    h = arimotoOutput alpha M q psi}

/-- PDF p.14, (16): the GLOBAL conjecture, not an asserted fact or scalar placeholder.
Fin enumerations cover every finite-dimensional system and outcome space. -/
def ConstrainedRenyiAdditivity : Prop :=
  ∀ a b i j : ℕ, 0 < a → 0 < b →
  ∀ M : Fin i → Mat (Fin a), IsPOVM M →
  ∀ R : Fin j → Mat (Fin b), IsPOVM R →
  ∀ sigma : Mat (Fin a), IsDensityMatrix sigma →
  ∀ tau : Mat (Fin b), IsDensityMatrix tau → ∀ alpha : ℝ, 1 < alpha →
    constrainedRenyi alpha (fun x : Fin i × Fin j => Matrix.kronecker (M x.1) (R x.2))
      (Matrix.kronecker sigma tau) =
    constrainedRenyi alpha M sigma + constrainedRenyi alpha R tau

/-- PDF p.14, (17): global uniform eventual IID uncertainty, with arbitrary rate slack.
The threshold precedes Eve's collective POVM. Arbitrarily small fixed epsilon permits
vanishing errors by diagonal choice. This predicate is a conjectural premise. -/
def IIDUncertainty : Prop :=
  ∀ a e : ℕ, 0 < a → 0 < e → ∀ p : Fin a → ℝ, IsDistribution p →
  ∀ rho : Fin a → Mat (Fin e), (∀ i, IsDensityMatrix (rho i)) →
  ∀ eps : ℝ, 0 < eps → eps < 1 → ∀ slack : ℝ, 0 < slack →
  ∃ n0 : ℕ, ∀ n ≥ n0, ∀ Q : Measurement (Fin n → Fin e),
    (n : ℝ) * (accessibleEquivocation p rho - slack) ≤
      traceSmoothConditionalMinBits eps (classicalJoint (iidJoint p rho n Q))

/-! Sanity examples pin deterministic entropy, a discarded tagged branch,
the empty tensor unit, and the diagonal symmetric normalization. -/

example : IsDistribution (fun _ : Fin 1 => (1 : ℝ)) := by
  constructor
  · intro i; norm_num
  · simp

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

example {E J K : Type} (Q : J → Mat E) (R : K → Mat E) (k : K) :
    taggedEffects 1 Q R (Sum.inr k) = 0 := by
  simp [taggedEffects]

example {E I : Type} (M : I → Mat E) (i : Fin 0 → I) (x y : Fin 0 → E) :
    tensorEffects M 0 i x y = 1 := by
  simp [tensorEffects]

example (d : ℕ) (i : Fin d) : universalSymmetricDilation d (i, i) ⟨(i, i), le_refl i⟩ = 1 := by
  simp [universalSymmetricDilation]

example {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    (n : ℕ) (C : KeyedBlockCode A B (n + 1)) (hm : C.messages = 1) : messageRateBits C = 0 := by
  simp [messageRateBits, hm]

/-- Reusing accepted code branches retains positive message count for every key. -/
example {A B : Type} [Fintype A] [Fintype B] [DecidableEq B]
    (n : ℕ) (C : KeyedBlockCode A B n) : 0 < C.messages := by
  exact (C.branch ⟨0, C.keys_pos⟩).positive

example {E : Type} [Fintype E] [DecidableEq E] :
    IsPOVM (fun _ : Fin 1 => (1 : Mat E)) := by
  exact ⟨fun _ => Matrix.PosSemidef.one, by simp⟩

example {E : Type} [Fintype E] [DecidableEq E] : supportInvSqrt (0 : Mat E) = 0 := by
  simp [supportInvSqrt, cfc_zero]

example {I E : Type} [Fintype I] [Fintype E] [DecidableEq E]
    (p : I → ℝ) (rho : I → Mat E) (i : I) (hi : p i = 0) :
    associatedEffects p rho i = 0 := by
  simp [associatedEffects, hi]

example {A J : Type} [DecidableEq J] (q : J → ℝ) (rho : J → Mat A)
    (a b : A) (j k : J) (hjk : j ≠ k) : qcBranchDensity q rho (a, j) (b, k) = 0 := by
  simp [qcBranchDensity, hjk]

example {A : Type} (k : ℕ) (rho : Mat (A × Fin k)) :
    activatedReceiver (fun X : Mat A => X) k rho = rho := by
  ext x y
  simp [activatedReceiver]

example (alpha : ℝ) :
    arimotoOutput alpha (fun _ : Fin 1 => (1 : Mat (Fin 1)))
      (fun _ : Fin 1 => (1 : ℝ)) (fun _ : Fin 1 => (1 : Mat (Fin 1))) = 0 := by
  simp [arimotoOutput, effectProb, Matrix.trace]

end QAlgorithms.Papers.Independent.Intakenext100.P63
Source
Weak locking capacity of quantum channels can be much larger than private capacity, arXiv:1403.6361v3, https://arxiv.org/abs/1403.6361v3
Read-back

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

QAlgorithms.Papers.Independent.Intakenext100.P63.ConstrainedRenyiAdditivity

This proposition asserts that for every pair of positive natural numbers a,ba,ba,b, every pair of natural numbers i,ji,ji,j, every family M1,…,MiM_1,\ldots,M_iM1​,…,Mi​ of positive semidefinite complex a×aa\times aa×a matrices summing to IaI_aIa​, every family R1,…,RjR_1,\ldots,R_jR1​,…,Rj​ of positive semidefinite complex b×bb\times bb×b matrices summing to IbI_bIb​, every positive semidefinite trace-one matrices σ\sigmaσ on Ca\mathbb C^aCa and τ\tauτ on Cb\mathbb C^bCb, and every real α>1\alpha>1α>1,

Rα((Ms⊗Rt)s,t,σ⊗τ)=Rα(M,σ)+Rα(R,τ).\mathcal R_\alpha((M_s\otimes R_t)_{s,t},\sigma\otimes\tau) = \mathcal R_\alpha(M,\sigma)+\mathcal R_\alpha(R,\tau).Rα​((Ms​⊗Rt​)s,t​,σ⊗τ)=Rα​(M,σ)+Rα​(R,τ).

For any finite family F=(Fs)F=(F_s)F=(Fs​) of matrices and matrix ω\omegaω, the symbol Rα(F,ω)\mathcal R_\alpha(F,\omega)Rα​(F,ω) here means the real infimum, over all natural numbers rrr, probability weights qh≥0q_h\ge0qh​≥0 summing to 111, and positive semidefinite trace-one matrices ψh\psi_hψh​ satisfying ∑hqhψh=ω\sum_hq_h\psi_h=\omega∑h​qh​ψh​=ω, of

−αα−1log⁡2 ⁣[∑h<rqh(∑s(Re⁡tr⁡(Fsψh))α)1/α].-\frac{\alpha}{\alpha-1}\log_2\!\left[ \sum_{h<r}q_h \left(\sum_s\bigl(\operatorname{Re}\operatorname{tr}(F_s\psi_h)\bigr)^\alpha\right)^{1/\alpha} \right].−α−1α​log2​​h<r∑​qh​(s∑​(Retr(Fs​ψh​))α)1/α​.

The tensor product has entries (X⊗Y)(u,v),(u′,v′)=Xu,u′Yv,v′(X\otimes Y)_{(u,v),(u',v')}=X_{u,u'}Y_{v,v'}(X⊗Y)(u,v),(u′,v′)​=Xu,u′​Yv,v′​. The outcome counts i,ji,ji,j include zero, although a zero-outcome family cannot sum to identity in either positive dimension, so those hypotheses are then impossible. Zero component weights are allowed, and all powers and logarithms use their total real definitions.

QAlgorithms.Papers.Independent.Intakenext100.P63.CqComplement

For finite types A,EA,EA,E, with decidable equality on AAA, an arbitrary map NcN_cNc​ from complex matrices indexed by AAA to complex matrices indexed by EEE, and a family (ρi)i∈A(\rho_i)_{i\in A}(ρi​)i∈A​ of complex matrices on EEE, this property means that every ρi\rho_iρi​ is positive semidefinite with trace 111, and

Nc(Ei,j)={ρi,i=j,0,i≠jfor every i,j∈A,N_c(E_{i,j})= \begin{cases} \rho_i,&i=j,\\ 0,&i\ne j \end{cases} \qquad\text{for every }i,j\in A,Nc​(Ei,j​)={ρi​,0,​i=j,i=j​for every i,j∈A,

where Ei,jE_{i,j}Ei,j​ has its sole nonzero entry, equal to 111, at (i,j)(i,j)(i,j). No channel or linearity assumption is imposed on NcN_cNc​, so the condition directly specifies its values only on these matrix units. Empty types are permitted: if AAA is empty, both universally quantified conditions are vacuous.

QAlgorithms.Papers.Independent.Intakenext100.P63.IIDUncertainty

This proposition asserts that for every positive natural numbers a,ea,ea,e, every probability vector p=(pi)i<ap=(p_i)_{i<a}p=(pi​)i<a​, every family of positive semidefinite trace-one matrices ρi\rho_iρi​ on Ce\mathbb C^eCe, every 0<ε<10<\varepsilon<10<ε<1, and every s>0s>0s>0, there exists a natural number n0n_0n0​ such that for every n≥n0n\ge n_0n≥n0​ and every measurement QQQ on (Ce)⊗n(\mathbb C^e)^{\otimes n}(Ce)⊗n with a finite, strictly positive number qqq of positive semidefinite effects summing to identity,

n(E(p,ρ)−s)≤sup⁡r⪰0, tr⁡r=1∥DP−r∥1/2≤εreal⁡ ⁣[sup⁡{λ∈R:∃ω⪰0, tr⁡ω=1, 2−λI⊗ω−r⪰0}].n\bigl(\mathcal E(p,\rho)-s\bigr) \le \sup_{\substack{r\succeq0,\ \operatorname{tr}r=1\\ \|D_P-r\|_1/2\le\varepsilon}} \operatorname{real}\!\left[ \sup\left\{\lambda\in\mathbb R: \exists \omega\succeq0,\ \operatorname{tr}\omega=1,\ 2^{-\lambda}I\otimes\omega-r\succeq0 \right\}\right].n(E(p,ρ)−s)≤r⪰0, trr=1∥DP​−r∥1​/2≤ε​sup​real[sup{λ∈R:∃ω⪰0, trω=1, 2−λI⊗ω−r⪰0}].

Here Pi,j=(∏t<npit)Re⁡tr⁡(Qj⨂t<nρit)P_{\mathbf i,j}=(\prod_{t<n}p_{i_t})\operatorname{Re}\operatorname{tr}(Q_j\bigotimes_{t<n}\rho_{i_t})Pi,j​=(∏t<n​pit​​)Retr(Qj​⨂t<n​ρit​​), and DPD_PDP​ is the diagonal matrix with diagonal Pi,jP_{\mathbf i,j}Pi,j​, indexed by {0,…,a−1}n×{0,…,q−1}\{0,\ldots,a-1\}^n\times\{0,\ldots,q-1\}{0,…,a−1}n×{0,…,q−1}. The matrix rrr has that same indexing set, ω\omegaω is on the outcome register, and ∥X∥1=Re⁡tr⁡X∗X\|X\|_1=\operatorname{Re}\operatorname{tr}\sqrt{X^*X}∥X∥1​=RetrX∗X​, with the square root taken by continuous functional calculus. The inner supremum is taken in the extended real numbers, and real⁡\operatorname{real}real sends either infinite value to zero; the outer supremum is a total real supremum. Finally,

E(p,ρ)=inf⁡F[−∑j,ipiRe⁡tr⁡(Fjρi)log⁡2 ⁣(piRe⁡tr⁡(Fjρi)∑hphRe⁡tr⁡(Fjρh))],\mathcal E(p,\rho)= \inf_F\left[-\sum_{j,i}p_i\operatorname{Re}\operatorname{tr}(F_j\rho_i) \log_2\!\left( \frac{p_i\operatorname{Re}\operatorname{tr}(F_j\rho_i)} {\sum_h p_h\operatorname{Re}\operatorname{tr}(F_j\rho_h)} \right)\right],E(p,ρ)=Finf​[−j,i∑​pi​Retr(Fj​ρi​)log2​(∑h​ph​Retr(Fj​ρh​)pi​Retr(Fj​ρi​)​)],

where FFF ranges over all measurements on Ce\mathbb C^eCe with a finite, strictly positive number of outcomes. No strict positivity of the entries of ppp is assumed. The threshold n0n_0n0​ is not required to be positive; if it is zero, the assertion also includes n=0n=0n=0, with empty tensor products and products equal to their units. Logarithms and division are total, including at zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.KeyedAchievable

For finite types A,B,EA,B,EA,B,E, all with decidable equality, arbitrary matrix maps N:A→BN:A\to BN:A→B and Nc:A→EN_c:A\to ENc​:A→E, a Boolean flag www, and a real number RRR, this property means that R≥0R\ge0R≥0 and there exists a sequence of keyed block codes of lengths ℓ=n+1\ell=n+1ℓ=n+1 satisfying the following conditions. Each code has Kn,Mn>0K_n,M_n>0Kn​,Mn​>0, density matrices ρn,k,m\rho_{n,k,m}ρn,k,m​ on AℓA^\ellAℓ, and effects Qn,k,mQ_{n,k,m}Qn,k,m​ on BℓB^\ellBℓ which are positive semidefinite and sum over mmm to the identity for each key. Writing ρ‾n,m=Kn−1∑kρn,k,m\overline{\rho}_{n,m}=K_n^{-1}\sum_k\rho_{n,k,m}ρ​n,m​=Kn−1​∑k​ρn,k,m​, the required limits and rate condition are

log⁡2Knn+1⟶0,1KnMn∑k,m(1−Re⁡tr⁡[Qn,k,mTN(ℓ)(ρn,k,m)])⟶0,∀s>0  ∃n0  ∀n≥n0,R−s≤log⁡2Mnn+1.\frac{\log_2 K_n}{n+1}\longrightarrow0,\qquad \frac1{K_nM_n}\sum_{k,m} \left(1-\operatorname{Re}\operatorname{tr}[Q_{n,k,m}\mathcal T_N^{(\ell)}(\rho_{n,k,m})]\right) \longrightarrow0, \qquad \forall s>0\;\exists n_0\;\forall n\ge n_0,\quad R-s\le\frac{\log_2 M_n}{n+1}.n+1log2​Kn​​⟶0,Kn​Mn​1​k,m∑​(1−Retr[Qn,k,m​TN(ℓ)​(ρn,k,m​)])⟶0,∀s>0∃n0​∀n≥n0​,R−s≤n+1log2​Mn​​.

Moreover, for every ε>0\varepsilon>0ε>0, all sufficiently large nnn satisfy the security condition selected by www: if www is true, every finite measurement (Fj)j<J(F_j)_{j<J}(Fj​)j<J​ on EℓE^\ellEℓ, with J>0J>0J>0, positive semidefinite effects, and ∑jFj=I\sum_jF_j=I∑j​Fj​=I, admits a probability distribution (ωj)j<J(\omega_j)_{j<J}(ωj​)j<J​ such that

1Mn∑m,j∣Re⁡tr⁡[FjTNc(ℓ)(ρ‾n,m)]−ωj∣≤ε;\frac1{M_n}\sum_{m,j} \left|\operatorname{Re}\operatorname{tr}[F_j\mathcal T_{N_c}^{(\ell)}(\overline{\rho}_{n,m})]-\omega_j\right| \le\varepsilon;Mn​1​m,j∑​​Retr[Fj​TNc​(ℓ)​(ρ​n,m​)]−ωj​​≤ε;

if www is false, there exists a positive semidefinite trace-one matrix ω\omegaω on EℓE^\ellEℓ such that

1Mn∑m∥TNc(ℓ)(ρ‾n,m)−ω∥1≤ε.\frac1{M_n}\sum_m\|\mathcal T_{N_c}^{(\ell)}(\overline{\rho}_{n,m})-\omega\|_1\le\varepsilon.Mn​1​m∑​∥TNc​(ℓ)​(ρ​n,m​)−ω∥1​≤ε.

Here a probability distribution has nonnegative real entries summing to 111, ∥Z∥1=Re⁡tr⁡Z†Z\|Z\|_1=\operatorname{Re}\operatorname{tr}\sqrt{Z^\dagger Z}∥Z∥1​=RetrZ†Z​, and for either map FFF its length-ℓ\ellℓ action is defined by

TF(ℓ)(X)y,y′=∑x,x′∈AℓXx,x′∏i=0ℓ−1F(Exi,xi′)yi,yi′,\mathcal T_F^{(\ell)}(X)_{y,y'} =\sum_{x,x'\in A^\ell}X_{x,x'}\prod_{i=0}^{\ell-1} F(E_{x_i,x'_i})_{y_i,y'_i},TF(ℓ)​(X)y,y′​=x,x′∈Aℓ∑​Xx,x′​i=0∏ℓ−1​F(Exi​,xi′​​)yi​,yi′​​,

with Ea,a′E_{a,a'}Ea,a′​ a matrix unit. Neither map is assumed to be a channel or linear, and no relation between the two maps is required.

QAlgorithms.Papers.Independent.Intakenext100.P63.KeyedBlockCode

For finite types A,BA,BA,B, with decidable equality on BBB, and any natural number nnn, this structure consists of a natural number MMM of messages, a natural number KKK of keys, a proof that K>0K>0K>0, and one branch code for each k∈{0,…,K−1}k\in\{0,\ldots,K-1\}k∈{0,…,K−1}. Each branch contains a proof that M>0M>0M>0, matrices ρk,m\rho_{k,m}ρk,m​ on AnA^nAn for every message mmm, each positive semidefinite with trace one, and matrices Qk,mQ_{k,m}Qk,m​ on BnB^nBn, each positive semidefinite and satisfying ∑m=0M−1Qk,m=IBn\sum_{m=0}^{M-1}Q_{k,m}=I_{B^n}∑m=0M−1​Qk,m​=IBn​. No decoding-error, security, or relation between different branches is included. The types A,BA,BA,B may be empty and nnn may be zero; at n=0n=0n=0, the sequence spaces are singletons even when the underlying types are empty.

QAlgorithms.Papers.Independent.Intakenext100.P63.KeyedRateAtLeast

For finite types A,BA,BA,B, with decidable equality on BBB, a sequence CnC_nCn​ of keyed block codes of lengths n+1n+1n+1, and any real number RRR, this property means

∀s>0  ∃n0∈N  ∀n≥n0,R−s≤log⁡2Mnn+1.\forall s>0\;\exists n_0\in\mathbb N\;\forall n\ge n_0,\qquad R-s\le\frac{\log_2 M_n}{n+1}.∀s>0∃n0​∈N∀n≥n0​,R−s≤n+1log2​Mn​​.

Here each CnC_nCn​ has Kn>0K_n>0Kn​>0 keys and Mn>0M_n>0Mn​>0 messages, density matrices on An+1A^{n+1}An+1 for each key and message, and, for each key, MnM_nMn​ positive semidefinite measurement effects on Bn+1B^{n+1}Bn+1 summing to the identity. The threshold n0n_0n0​ may depend on sss, and RRR is not required to be nonnegative.

QAlgorithms.Papers.Independent.Intakenext100.P63.PrivateSecure

For finite types A,B,EA,B,EA,B,E, all with decidable equality, an arbitrary matrix map Nc:A→EN_c:A\to ENc​:A→E, a natural number nnn, a keyed block code CCC, and a real number δ\deltaδ, this property means that there exists a positive semidefinite matrix ω\omegaω of trace 111 on EnE^nEn such that

1M∑m=0M−1∥TNc(n)(ρ‾m)−ω∥1≤δ.\frac1M\sum_{m=0}^{M-1} \left\|\mathcal T_{N_c}^{(n)}(\overline{\rho}_m)-\omega\right\|_1\le\delta.M1​m=0∑M−1​​TNc​(n)​(ρ​m​)−ω​1​≤δ.

Here the code has K>0K>0K>0 keys and M>0M>0M>0 messages, density matrices ρk,m\rho_{k,m}ρk,m​ on AnA^nAn, and, for each key, MMM positive semidefinite measurement effects on BnB^nBn summing to the identity; ρ‾m=K−1∑kρk,m\overline{\rho}_m=K^{-1}\sum_k\rho_{k,m}ρ​m​=K−1∑k​ρk,m​. The map appearing here is defined by

TNc(n)(X)y,y′=∑x,x′∈AnXx,x′∏i=0n−1Nc(Exi,xi′)yi,yi′,\mathcal T_{N_c}^{(n)}(X)_{y,y'} =\sum_{x,x'\in A^n}X_{x,x'}\prod_{i=0}^{n-1} N_c(E_{x_i,x'_i})_{y_i,y'_i},TNc​(n)​(X)y,y′​=x,x′∈An∑​Xx,x′​i=0∏n−1​Nc​(Exi​,xi′​​)yi​,yi′​​,

with Ea,a′E_{a,a'}Ea,a′​ a matrix unit, and the norm means ∥Z∥1=Re⁡tr⁡Z†Z\|Z\|_1=\operatorname{Re}\operatorname{tr}\sqrt{Z^\dagger Z}∥Z∥1​=RetrZ†Z​, with the square root given by continuous functional calculus. The same ω\omegaω is used for all messages. No channel or linearity assumption on NcN_cNc​, nonnegativity assumption on δ\deltaδ, or factor 1/21/21/2 is present. The definition includes n=0n=0n=0, using singleton register index sets and empty products equal to 111.

QAlgorithms.Papers.Independent.Intakenext100.P63.ReliableKeyedSequence

For finite types A,BA,BA,B, both with decidable equality, an arbitrary matrix map N:A→BN:A\to BN:A→B, and a sequence CnC_nCn​ of keyed block codes of lengths n+1n+1n+1, this property means that both real sequences

log⁡2Knn+1and1KnMn∑k=0Kn−1∑m=0Mn−1(1−Re⁡tr⁡ ⁣[Qn,k,mTN(n+1)(ρρ‾n,k,m)])\frac{\log_2 K_n}{n+1} \quad\text{and}\quad \frac1{K_nM_n}\sum_{k=0}^{K_n-1}\sum_{m=0}^{M_n-1} \left(1-\operatorname{Re}\operatorname{tr}\!\left[ Q_{n,k,m}\mathcal T_N^{(n+1)}(\overline{\vphantom{\rho}\smash{\rho}}_{n,k,m}) \right]\right)n+1log2​Kn​​andKn​Mn​1​k=0∑Kn​−1​m=0∑Mn​−1​(1−Retr[Qn,k,m​TN(n+1)​(ρρ​n,k,m​)])

converge to 000 as n→∞n\to\inftyn→∞, where the symbol ρρ‾n,k,m\overline{\vphantom{\rho}\smash{\rho}}_{n,k,m}ρρ​n,k,m​ here denotes the branch message matrix itself. More explicitly, Kn,Mn>0K_n,M_n>0Kn​,Mn​>0, those branch matrices are positive semidefinite of trace 111 on An+1A^{n+1}An+1, and, for each n,kn,kn,k, the effects Qn,k,mQ_{n,k,m}Qn,k,m​ on Bn+1B^{n+1}Bn+1 are positive semidefinite and sum over mmm to the identity. For any length ℓ\ellℓ,

TN(ℓ)(X)y,y′=∑x,x′∈AℓXx,x′∏i=0ℓ−1N(Exi,xi′)yi,yi′,\mathcal T_N^{(\ell)}(X)_{y,y'} =\sum_{x,x'\in A^\ell}X_{x,x'}\prod_{i=0}^{\ell-1} N(E_{x_i,x'_i})_{y_i,y'_i},TN(ℓ)​(X)y,y′​=x,x′∈Aℓ∑​Xx,x′​i=0∏ℓ−1​N(Exi​,xi′​​)yi​,yi′​​,

where Ea,a′E_{a,a'}Ea,a′​ is a matrix unit. No channel or linearity assumption is made on NNN, and no condition on the limiting message rate is part of this property.

QAlgorithms.Papers.Independent.Intakenext100.P63.WeakSecure

For finite types A,B,EA,B,EA,B,E, all with decidable equality, an arbitrary matrix map Nc:A→EN_c:A\to ENc​:A→E, a natural number nnn, a keyed block code CCC, and a real number δ\deltaδ, this property means that every measurement (Qj)j=0J−1(Q_j)_{j=0}^{J-1}(Qj​)j=0J−1​ on EnE^nEn, with J>0J>0J>0, QjQ_jQj​ positive semidefinite, and ∑jQj=I\sum_jQ_j=I∑j​Qj​=I, admits real numbers ωj≥0\omega_j\ge0ωj​≥0 with ∑jωj=1\sum_j\omega_j=1∑j​ωj​=1 such that

1M∑m=0M−1∑j=0J−1∣Re⁡tr⁡ ⁣[QjTNc(n)(ρ‾m)]−ωj∣≤δ.\frac1M\sum_{m=0}^{M-1}\sum_{j=0}^{J-1} \left|\operatorname{Re}\operatorname{tr}\!\left[Q_j\mathcal T_{N_c}^{(n)}(\overline{\rho}_m)\right]-\omega_j\right| \le\delta.M1​m=0∑M−1​j=0∑J−1​​Retr[Qj​TNc​(n)​(ρ​m​)]−ωj​​≤δ.

Here the code has K>0K>0K>0 keys and M>0M>0M>0 messages, density matrices ρk,m\rho_{k,m}ρk,m​ on AnA^nAn, and a measurement with MMM positive semidefinite effects summing to the identity on BnB^nBn for each key; ρ‾m=K−1∑kρk,m\overline{\rho}_m=K^{-1}\sum_k\rho_{k,m}ρ​m​=K−1∑k​ρk,m​. For any matrix XXX, the map in the inequality is

TNc(n)(X)y,y′=∑x,x′∈AnXx,x′∏i=0n−1Nc(Exi,xi′)yi,yi′,\mathcal T_{N_c}^{(n)}(X)_{y,y'} =\sum_{x,x'\in A^n}X_{x,x'}\prod_{i=0}^{n-1} N_c(E_{x_i,x'_i})_{y_i,y'_i},TNc​(n)​(X)y,y′​=x,x′∈An∑​Xx,x′​i=0∏n−1​Nc​(Exi​,xi′​​)yi​,yi′​​,

where Ea,a′E_{a,a'}Ea,a′​ is a matrix unit. The distribution ω\omegaω may depend on the measurement but is shared by all messages. There is no channel or linearity assumption on NcN_cNc​, no restriction on the sign of δ\deltaδ, and no factor 1/21/21/2 in the displayed discrepancy. The case n=0n=0n=0 uses singleton register index sets and empty products equal to 111.

QAlgorithms.Papers.Independent.Intakenext100.P63.accessibleEquivocation

For arbitrary finite types I,EI,EI,E, with decidable equality on EEE, real weights p:I→Rp:I\to\mathbb Rp:I→R, and complex square matrices (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ indexed by EEE, this defines the real infimum over all positive integers qqq and all families Q0,…,Qq−1Q_0,\ldots,Q_{q-1}Q0​,…,Qq−1​ of positive semidefinite matrices satisfying ∑jQj=1E\sum_jQ_j=1_E∑j​Qj​=1E​, of

−∑j=0q−1∑i∈Iaijlog⁡2 ⁣(aij∑k∈Iakj),aij=piRe⁡tr⁡(Qjρi).-\sum_{j=0}^{q-1}\sum_{i\in I}a_{ij}\log_2\!\left(\frac{a_{ij}}{\sum_{k\in I}a_{kj}}\right), \qquad a_{ij}=p_i\operatorname{Re}\operatorname{tr}(Q_j\rho_i).−j=0∑q−1​i∈I∑​aij​log2​(∑k∈I​akj​aij​​),aij​=pi​Retr(Qj​ρi​).

Neither normalization nor nonnegativity of ppp, nor any density-matrix condition on ρi\rho_iρi​, is required. Zero denominators give zero quotients; the real logarithm is zero at zero and uses the absolute value at negative arguments. The infimum is real-valued, with value zero for an empty set or a set unbounded below. Empty III or EEE is allowed and gives value zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.activatedCapacity

For arbitrary finite types A,B,EA,B,EA,B,E, all with decidable equality, and arbitrary matrix-valued functions N:A→BN:A\to BN:A→B and Nc:A→EN_c:A\to ENc​:A→E in the sense of maps between their complex square-matrix spaces, this defines

sup⁡k≥1(Cweak(N~k,N~c,k)−log⁡2k),\sup_{k\ge1}\left(C_{\mathrm{weak}}(\widetilde N_k,\widetilde N_{c,k})-\log_2k\right),k≥1sup​(Cweak​(Nk​,Nc,k​)−log2​k),

where (N~k(X))(b,s),(b′,t)=N((X(a,s),(a′,t))a,a′)b,b′(\widetilde N_k(X))_{(b,s),(b',t)}=N((X_{(a,s),(a',t)})_{a,a'})_{b,b'}(Nk​(X))(b,s),(b′,t)​=N((X(a,s),(a′,t)​)a,a′​)b,b′​ and N~c,k(X)=Nc((∑sX(a,s),(a′,s))a,a′)\widetilde N_{c,k}(X)=N_c((\sum_sX_{(a,s),(a',s)})_{a,a'})Nc,k​(X)=Nc​((∑s​X(a,s),(a′,s)​)a,a′​). For any finite-register maps F,GF,GF,G, Cweak(F,G)C_{\mathrm{weak}}(F,G)Cweak​(F,G) means the real supremum of nonnegative rates RRR for which there is a sequence of codes of lengths ℓ=n+1\ell=n+1ℓ=n+1: each code has positive integers Mn,KnM_n,K_nMn​,Kn​, and for every key u<Knu<K_nu<Kn​, density matrices ρu,m\rho_{u,m}ρu,m​ for m<Mnm<M_nm<Mn​ and positive semidefinite decoding effects Du,mD_{u,m}Du,m​ summing to the output identity. A density matrix means a positive semidefinite matrix of trace 111. The requirements are

log⁡2Knℓ⟶0,1KnMn∑u,m(1−Re⁡tr⁡ ⁣(Du,mF[ℓ](ρu,m)))⟶0,\frac{\log_2K_n}{\ell}\longrightarrow0,\qquad \frac1{K_nM_n}\sum_{u,m}\left(1-\operatorname{Re}\operatorname{tr}\!\left(D_{u,m}F^{[\ell]}(\rho_{u,m})\right)\right)\longrightarrow0,ℓlog2​Kn​​⟶0,Kn​Mn​1​u,m∑​(1−Retr(Du,m​F[ℓ](ρu,m​)))⟶0,

and, for every s>0s>0s>0, eventually log⁡2Mn/ℓ≥R−s\log_2M_n/\ell\ge R-slog2​Mn​/ℓ≥R−s. Writing ρˉm=Kn−1∑uρu,m\bar\rho_m=K_n^{-1}\sum_u\rho_{u,m}ρˉ​m​=Kn−1​∑u​ρu,m​, security requires that for every ε>0\varepsilon>0ε>0, eventually every finite measurement QQQ with a strictly positive number of outcomes admits weights ωj≥0\omega_j\ge0ωj​≥0, ∑jωj=1\sum_j\omega_j=1∑j​ωj​=1, such that

1Mn∑m,j∣Re⁡tr⁡ ⁣(QjG[ℓ](ρˉm))−ωj∣≤ε.\frac1{M_n}\sum_{m,j} \left|\operatorname{Re}\operatorname{tr}\!\left(Q_jG^{[\ell]}(\bar\rho_m)\right)-\omega_j\right| \le\varepsilon.Mn​1​m,j∑​​Retr(Qj​G[ℓ](ρˉ​m​))−ωj​​≤ε.

Here a measurement has positive semidefinite effects summing to identity, and the power of any map is defined on all matrices by

F[ℓ](X)y,y′=∑x,x′Xx,x′∏t<ℓF(Ext,xt′)yt,yt′,F^{[\ell]}(X)_{y,y'}=\sum_{x,x'}X_{x,x'}\prod_{t<\ell}F(E_{x_t,x'_t})_{y_t,y'_t},F[ℓ](X)y,y′​=x,x′∑​Xx,x′​t<ℓ∏​F(Ext​,xt′​​)yt​,yt′​​,

with Ea,a′E_{a,a'}Ea,a′​ the matrix unit. No channel assumption is imposed on N,NcN,N_cN,Nc​, and the finite types may be empty. All suprema are total real suprema, equal to zero on empty or unbounded-above sets.

QAlgorithms.Papers.Independent.Intakenext100.P63.activatedEnvironment

For arbitrary types A,EA,EA,E, an arbitrary function Nc:CA×A→CE×EN_c:\mathbb C^{A\times A}\to\mathbb C^{E\times E}Nc​:CA×A→CE×E, and every natural number kkk, this defines a function N~c,k\widetilde N_{c,k}Nc,k​ on matrices indexed by A×{0,…,k−1}A\times\{0,\ldots,k-1\}A×{0,…,k−1} by

N~c,k(ρ)=Nc ⁣((∑s=0k−1ρ(a,s),(a′,s))a,a′∈A).\widetilde N_{c,k}(\rho) = N_c\!\left(\left(\sum_{s=0}^{k-1}\rho_{(a,s),(a',s)}\right)_{a,a'\in A}\right).Nc,k​(ρ)=Nc​​(s=0∑k−1​ρ(a,s),(a′,s)​)a,a′∈A​​.

The sum traces out the auxiliary index before applying NcN_cNc​. No finiteness assumption on A,EA,EA,E, or linearity, positivity, or channel assumption on NcN_cNc​, is required. In particular, for k=0k=0k=0 the summed matrix is zero and the result is Nc(0)N_c(0)Nc​(0), which is not required to be zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.activatedReceiver

For arbitrary types A,BA,BA,B, an arbitrary function N:CA×A→CB×BN:\mathbb C^{A\times A}\to\mathbb C^{B\times B}N:CA×A→CB×B, and every natural number kkk, this defines a function N~k\widetilde N_kNk​ from matrices on A×{0,…,k−1}A\times\{0,\ldots,k-1\}A×{0,…,k−1} to matrices on B×{0,…,k−1}B\times\{0,\ldots,k-1\}B×{0,…,k−1} by

(N~k(ρ))(b,s),(b′,t)=[N ⁣((ρ(a,s),(a′,t))a,a′∈A)]b,b′.\bigl(\widetilde N_k(\rho)\bigr)_{(b,s),(b',t)} = \left[N\!\left((\rho_{(a,s),(a',t)})_{a,a'\in A}\right)\right]_{b,b'}.(Nk​(ρ))(b,s),(b′,t)​=[N((ρ(a,s),(a′,t)​)a,a′∈A​)]b,b′​.

Thus NNN is applied separately to each block determined by the two auxiliary indices s,ts,ts,t. No finiteness, linearity, positivity, or channel assumption is imposed on A,B,NA,B,NA,B,N. The definition includes k=0k=0k=0, when the input and output matrices have empty indexing sets.

QAlgorithms.Papers.Independent.Intakenext100.P63.arimotoOutput

For arbitrary finite types E,I,JE,I,JE,I,J, a real number α\alphaα, arbitrary complex square matrices MiM_iMi​ on EEE indexed by i∈Ii\in Ii∈I, arbitrary real weights qjq_jqj​ indexed by j∈Jj\in Jj∈J, and arbitrary complex square matrices ψj\psi_jψj​ on EEE, this defines the real number

−αα−1log⁡2 ⁣[∑j∈Jqj(∑i∈I(Re⁡tr⁡(Miψj))α)1/α].-\frac{\alpha}{\alpha-1}\log_2\!\left[ \sum_{j\in J}q_j \left(\sum_{i\in I} \bigl(\operatorname{Re}\operatorname{tr}(M_i\psi_j)\bigr)^\alpha \right)^{1/\alpha} \right].−α−1α​log2​​j∈J∑​qj​(i∈I∑​(Retr(Mi​ψj​))α)1/α​.

There is no assumption that MMM is a measurement, that qqq is a probability distribution, that the ψj\psi_jψj​ are density matrices, or that α\alphaα lies in any particular range. All three finite types may be empty, and the powers, divisions, and logarithm use total real operations even at zero or outside their usual positive-argument domains. In particular, the definition includes α=0\alpha=0α=0 and α=1\alpha=1α=1, and gives zero at either value because the displayed prefactor is zero under total division.

QAlgorithms.Papers.Independent.Intakenext100.P63.associatedEffects

For arbitrary finite types I,EI,EI,E, with decidable equality on EEE, real weights p:I→Rp:I\to\mathbb Rp:I→R, a family (ρj)j∈I(\rho_j)_{j\in I}(ρj​)j∈I​ of complex square matrices indexed by EEE, and an index i∈Ii\in Ii∈I, this defines

S(piρi)S,ρˉ=∑j∈Ipjρj,S=f(ρˉ),f(x)=(x)−1,S(p_i\rho_i)S, \qquad \bar\rho=\sum_{j\in I}p_j\rho_j,\qquad S=f(\bar\rho),\quad f(x)=(\sqrt{x})^{-1},S(pi​ρi​)S,ρˉ​=j∈I∑​pj​ρj​,S=f(ρˉ​),f(x)=(x​)−1,

using real continuous functional calculus. If ρˉ\bar\rhoρˉ​ is Hermitian, SSS has eigenvalue 1/x1/\sqrt{x}1/x​ on each positive-eigenvalue subspace and zero on each nonpositive-eigenvalue subspace; if ρˉ\bar\rhoρˉ​ is non-Hermitian, S=0S=0S=0. No distribution, density-matrix, or measurement assumptions are imposed. Empty EEE is allowed; empty III supplies no index iii.

QAlgorithms.Papers.Independent.Intakenext100.P63.associatedMinimum

For arbitrary finite types I,EI,EI,E, with decidable equality on EEE, real weights p:I→Rp:I\to\mathbb Rp:I→R, and complex square matrices (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ indexed by EEE, set ρˉ=∑ipiρi\bar\rho=\sum_i p_i\rho_iρˉ​=∑i​pi​ρi​, S=f(ρˉ)S=f(\bar\rho)S=f(ρˉ​) for f(x)=(x)−1f(x)=(\sqrt{x})^{-1}f(x)=(x​)−1 under real continuous functional calculus, and Mi=S(piρi)SM_i=S(p_i\rho_i)SMi​=S(pi​ρi​)S. This defines

inf⁡ψ⪰0, tr⁡ψ=1ker⁡ρˉ⊆ker⁡ψ[−∑i∈Irilog⁡2ri],ri=Re⁡tr⁡(Miψ).\inf_{\substack{\psi\succeq0,\ \operatorname{tr}\psi=1\\ \ker\bar\rho\subseteq\ker\psi}} \left[-\sum_{i\in I}r_i\log_2 r_i\right], \qquad r_i=\operatorname{Re}\operatorname{tr}(M_i\psi).ψ⪰0, trψ=1kerρˉ​⊆kerψ​inf​[−i∈I∑​ri​log2​ri​],ri​=Retr(Mi​ψ).

Here the kernels are those of matrix multiplication on vectors. For Hermitian ρˉ\bar\rhoρˉ​, SSS replaces positive eigenvalues by their inverse square roots and nonpositive eigenvalues by zero; for non-Hermitian ρˉ\bar\rhoρˉ​, S=0S=0S=0. No distribution or density-matrix assumptions are imposed on p,ρp,\rhop,ρ. The total real logarithm is zero at zero and uses absolute values otherwise. The real infimum is zero for an empty or unbounded-below defining set; in particular, if EEE is empty or ρˉ=0\bar\rho=0ρˉ​=0, no admissible trace-one ψ\psiψ exists and the value is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.averageState

For any finite type III, arbitrary type EEE, real weights p:I→Rp:I\to\mathbb Rp:I→R, and family (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ of complex square matrices indexed by EEE, this defines the matrix

∑i∈Ipiρi,\sum_{i\in I}p_i\rho_i,i∈I∑​pi​ρi​,

where the weights act as complex scalars. No probability-distribution or density-matrix hypotheses are imposed. If III is empty, this is the zero matrix; EEE may also be empty.

QAlgorithms.Papers.Independent.Intakenext100.P63.averagedSignal

For finite types A,BA,BA,B, with decidable equality on BBB, a natural number nnn, a keyed block code CCC, and any message index m∈{0,…,M−1}m\in\{0,\ldots,M-1\}m∈{0,…,M−1}, this defines the complex matrix on AnA^nAn

ρ‾m=1K∑k=0K−1ρk,m.\overline{\rho}_m=\frac1K\sum_{k=0}^{K-1}\rho_{k,m}.ρ​m​=K1​k=0∑K−1​ρk,m​.

Here K=C.keys>0K=C.\mathrm{keys}>0K=C.keys>0, M=C.messages>0M=C.\mathrm{messages}>0M=C.messages>0, and ρk,m\rho_{k,m}ρk,m​ is the message matrix in branch kkk; every such matrix is positive semidefinite with trace 111. Each branch also supplies MMM positive semidefinite measurement matrices on BnB^nBn whose sum is the identity. The scalar 1/K1/K1/K is taken in C\mathbb CC. The case n=0n=0n=0 is included, with A0A^0A0 a singleton; no decidable equality assumption on AAA is required.

QAlgorithms.Papers.Independent.Intakenext100.P63.classicalJoint

For arbitrary types I,JI,JI,J, both with decidable equality, and any function P:I×J→RP:I\times J\to\mathbb RP:I×J→R, this defines the complex diagonal matrix on I×JI\times JI×J with diagonal entry P(i,j)P(i,j)P(i,j) at (i,j)(i,j)(i,j) and all off-diagonal entries zero. The types need not be finite or nonempty, and PPP is not required to be nonnegative or normalized.

QAlgorithms.Papers.Independent.Intakenext100.P63.conditionalMinBits

For any finite types A,BA,BA,B with decidable equality and any complex matrix ρ\rhoρ indexed by (A×B)×(A×B)(A\times B)\times(A\times B)(A×B)×(A×B), this defines the real conversion of the extended-real supremum

sup⁡{λ∈R:∃ σ∈CB×B, σ⪰0, tr⁡σ=1, 2−λ(IA⊗σ)−ρ⪰0}.\sup\left\{\lambda\in\mathbb R: \exists\,\sigma\in\mathbb C^{B\times B},\ \sigma\succeq0,\ \operatorname{tr}\sigma=1,\ 2^{-\lambda}(I_A\otimes\sigma)-\rho\succeq0\right\}.sup{λ∈R:∃σ∈CB×B, σ⪰0, trσ=1, 2−λ(IA​⊗σ)−ρ⪰0}.

The supremum is formed in the extended real numbers; its conversion to R\mathbb RR preserves finite values and sends both +∞+\infty+∞ and −∞-\infty−∞ to zero. No positivity or normalization condition is imposed on ρ\rhoρ, and neither index type is required to be nonempty. If BBB is empty there is no admissible σ\sigmaσ, so the extended supremum is −∞-\infty−∞ and the defined real value is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.constrainedRenyi

For arbitrary finite types E,IE,IE,I, a real number α\alphaα, arbitrary complex square matrices MiM_iMi​ on EEE, and an arbitrary complex square matrix σ\sigmaσ on EEE, this defines the real infimum

inf⁡r∈N, qj≥0, ∑j<rqj=1ψj⪰0, tr⁡ψj=1∑j<rqjψj=σ{−αα−1log⁡2 ⁣[∑j<rqj(∑i∈I(Re⁡tr⁡(Miψj))α)1/α]}.\inf_{\substack{r\in\mathbb N,\ q_j\ge0,\ \sum_{j<r}q_j=1\\ \psi_j\succeq0,\ \operatorname{tr}\psi_j=1\\ \sum_{j<r}q_j\psi_j=\sigma}} \left\{ -\frac{\alpha}{\alpha-1}\log_2\!\left[ \sum_{j<r}q_j \left(\sum_{i\in I} \bigl(\operatorname{Re}\operatorname{tr}(M_i\psi_j)\bigr)^\alpha \right)^{1/\alpha} \right]\right\}.r∈N, qj​≥0, ∑j<r​qj​=1ψj​⪰0, trψj​=1∑j<r​qj​ψj​=σ​inf​⎩⎨⎧​−α−1α​log2​​j<r∑​qj​(i∈I∑​(Retr(Mi​ψj​))α)1/α​⎭⎬⎫​.

Thus the optimization ranges over finite probability-weighted decompositions of σ\sigmaσ into density matrices; the component matrices are not required to be pure. No measurement condition on MMM, density-matrix hypothesis on σ\sigmaσ, or restriction on α\alphaα is imposed. The index types may be empty, and r=0r=0r=0 is allowed syntactically but cannot satisfy ∑jqj=1\sum_jq_j=1∑j​qj​=1. If no decomposition exists, the real infimum is zero; it is also zero if the set of values is unbounded below. Powers, logarithms, and division are total, including at α=0,1\alpha=0,1α=0,1.

QAlgorithms.Papers.Independent.Intakenext100.P63.ensembleConditionalBits

For arbitrary finite types I,E,JI,E,JI,E,J, real weights p:I→Rp:I\to\mathbb Rp:I→R, and families (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ and (Qj)j∈J(Q_j)_{j\in J}(Qj​)j∈J​ of complex square matrices indexed by EEE, set aij=piRe⁡tr⁡(Qjρi)a_{ij}=p_i\operatorname{Re}\operatorname{tr}(Q_j\rho_i)aij​=pi​Retr(Qj​ρi​). This defines

−∑j∈J∑i∈Iaijlog⁡2 ⁣(aij∑k∈Iakj).-\sum_{j\in J}\sum_{i\in I}a_{ij}\log_2\!\left(\frac{a_{ij}}{\sum_{k\in I}a_{kj}}\right).−j∈J∑​i∈I∑​aij​log2​(∑k∈I​akj​aij​​).

There are no probability, density-matrix, or measurement assumptions. Division by zero is zero, and the total real logarithm is zero at zero and equals log⁡∣x∣/log⁡2\log|x|/\log 2log∣x∣/log2 otherwise. Empty index types are permitted; if any of I,E,JI,E,JI,E,J is empty, the value is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.environment

For finite types A,BA,BA,B, an arbitrary type EEE, and any complex rectangular matrix VVV with rows indexed by B×EB\times EB×E and columns indexed by AAA, this defines the function sending a complex matrix ρ\rhoρ on AAA to the matrix on EEE given by

E(ρ)e,e′=∑b∈B(VρV†)(b,e),(b,e′)=∑b∈B∑a,a′∈AV(b,e),aρa,a′V(b,e′),a′‾.\mathcal E(\rho)_{e,e'} =\sum_{b\in B}(V\rho V^\dagger)_{(b,e),(b,e')} =\sum_{b\in B}\sum_{a,a'\in A} V_{(b,e),a}\rho_{a,a'}\overline{V_{(b,e'),a'}}.E(ρ)e,e′​=b∈B∑​(VρV†)(b,e),(b,e′)​=b∈B∑​a,a′∈A∑​V(b,e),a​ρa,a′​V(b,e′),a′​​.

No isometry condition is imposed on VVV, no density condition is imposed on ρ\rhoρ, and EEE need not be finite. Empty AAA or empty BBB makes the resulting matrix zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.iidJoint

For any type III, finite type EEE with decidable equality, function p:I→Rp:I\to\mathbb Rp:I→R, family of complex matrices ρi\rho_iρi​ on EEE, natural number nnn, and measurement QQQ on the sequence space EnE^nEn, this defines

P(i,j)=(∏t=0n−1pi(t))Re⁡tr⁡ ⁣(QjRi),(Ri)x,y=∏t=0n−1(ρi(t))x(t),y(t).P(i,j)= \left(\prod_{t=0}^{n-1}p_{i(t)}\right) \operatorname{Re}\operatorname{tr}\!\left(Q_jR_i\right), \qquad (R_i)_{x,y}=\prod_{t=0}^{n-1}(\rho_{i(t)})_{x(t),y(t)}.P(i,j)=(t=0∏n−1​pi(t)​)Retr(Qj​Ri​),(Ri​)x,y​=t=0∏n−1​(ρi(t)​)x(t),y(t)​.

Here i∈Ini\in I^ni∈In, and jjj ranges over the measurement’s kkk outcomes, where kkk is a positive natural number, each QjQ_jQj​ is positive semidefinite, and ∑j=0k−1Qj=IEn\sum_{j=0}^{k-1}Q_j=I_{E^n}∑j=0k−1​Qj​=IEn​. No probability-distribution assumption is imposed on ppp, no density-matrix assumption is imposed on the ρi\rho_iρi​, and III need not be finite or nonempty. When n=0n=0n=0, both products equal one, so P(i,j)=Re⁡tr⁡QjP(i,j)=\operatorname{Re}\operatorname{tr}Q_jP(i,j)=RetrQj​ for the unique empty sequence iii.

QAlgorithms.Papers.Independent.Intakenext100.P63.jointConditionalBits

For arbitrary finite types I,JI,JI,J and any real array P:I×J→RP:I\times J\to\mathbb RP:I×J→R, this defines

−∑j∈J∑i∈IPijlog⁡2 ⁣(Pij∑k∈IPkj).-\sum_{j\in J}\sum_{i\in I}P_{ij}\log_2\!\left(\frac{P_{ij}}{\sum_{k\in I}P_{kj}}\right).−j∈J∑​i∈I∑​Pij​log2​(∑k∈I​Pkj​Pij​​).

No nonnegativity or normalization of PPP is assumed. Division by zero is defined to be zero; the total real logarithm is zero at zero and equals log⁡∣x∣/log⁡2\log |x|/\log 2log∣x∣/log2 at nonzero xxx. Either index type may be empty, in which case the expression is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.keyRate

For finite types A,BA,BA,B, with decidable equality on BBB, any natural number nnn, and a keyed block code CCC of block length n+1n+1n+1, this defines

log⁡2Kn+1,K=C.keys>0.\frac{\log_2 K}{n+1}, \qquad K=C.\mathrm{keys}>0.n+1log2​K​,K=C.keys>0.

The code also has a positive number MMM of messages; for each key it supplies density matrices on An+1A^{n+1}An+1 for those messages and MMM positive semidefinite measurement effects on Bn+1B^{n+1}Bn+1 summing to the identity. The denominator is positive, and n=0n=0n=0 gives block length 111.

QAlgorithms.Papers.Independent.Intakenext100.P63.keyedAverageError

For finite types A,BA,BA,B, both with decidable equality, an arbitrary map NNN from complex matrices indexed by AAA to complex matrices indexed by BBB, a natural number nnn, and a keyed block code CCC, this defines the real number

1KM∑k=0K−1∑m=0M−1(1−Re⁡tr⁡ ⁣[Qk,m TN(n)(ρk,m)]).\frac{1}{KM}\sum_{k=0}^{K-1}\sum_{m=0}^{M-1} \left(1-\operatorname{Re}\operatorname{tr}\!\left[Q_{k,m}\,\mathcal T_N^{(n)}(\rho_{k,m})\right]\right).KM1​k=0∑K−1​m=0∑M−1​(1−Retr[Qk,m​TN(n)​(ρk,m​)]).

Here K=C.keys>0K=C.\mathrm{keys}>0K=C.keys>0, M=C.messages>0M=C.\mathrm{messages}>0M=C.messages>0, each ρk,m\rho_{k,m}ρk,m​ is a positive semidefinite matrix of trace 111 on AnA^nAn, and, for each key kkk, the matrices Qk,mQ_{k,m}Qk,m​ on BnB^nBn are positive semidefinite and sum over mmm to the identity. Positivity of MMM follows from the branch code supplied for every key. The map used in the formula is defined entrywise by

TN(n)(X)y,y′=∑x,x′∈AnXx,x′∏i=0n−1N(Exi,xi′)yi,yi′,\mathcal T_N^{(n)}(X)_{y,y'} =\sum_{x,x'\in A^n}X_{x,x'}\prod_{i=0}^{n-1} N(E_{x_i,x'_i})_{y_i,y'_i},TN(n)​(X)y,y′​=x,x′∈An∑​Xx,x′​i=0∏n−1​N(Exi​,xi′​​)yi​,yi′​​,

where Ea,a′E_{a,a'}Ea,a′​ has its sole nonzero entry, equal to 111, at (a,a′)(a,a')(a,a′). No channel or linearity assumption is imposed on NNN. The definition includes n=0n=0n=0, when the register index sets are singletons and the empty product is 111.

QAlgorithms.Papers.Independent.Intakenext100.P63.keyedCapacity

For finite types A,B,EA,B,EA,B,E, all with decidable equality, arbitrary matrix maps N:A→BN:A\to BN:A→B and Nc:A→EN_c:A\to ENc​:A→E, and a Boolean flag www, this defines the real supremum of all R≥0R\ge0R≥0 for which there exists a sequence of keyed block codes of lengths ℓ=n+1\ell=n+1ℓ=n+1 with the following properties: each code has Kn,Mn>0K_n,M_n>0Kn​,Mn​>0, positive semidefinite trace-one message matrices ρn,k,m\rho_{n,k,m}ρn,k,m​ on AℓA^\ellAℓ, and, for each key, positive semidefinite effects Qn,k,mQ_{n,k,m}Qn,k,m​ on BℓB^\ellBℓ summing over messages to the identity; the quantities

log⁡2Knn+1and1KnMn∑k,m(1−Re⁡tr⁡[Qn,k,mTN(ℓ)(ρn,k,m)])\frac{\log_2 K_n}{n+1} \quad\text{and}\quad \frac1{K_nM_n}\sum_{k,m} \left(1-\operatorname{Re}\operatorname{tr}[Q_{n,k,m}\mathcal T_N^{(\ell)}(\rho_{n,k,m})]\right)n+1log2​Kn​​andKn​Mn​1​k,m∑​(1−Retr[Qn,k,m​TN(ℓ)​(ρn,k,m​)])

converge to 000; and for every s>0s>0s>0, eventually log⁡2Mn/(n+1)≥R−s\log_2 M_n/(n+1)\ge R-slog2​Mn​/(n+1)≥R−s. In addition, for every ε>0\varepsilon>0ε>0, eventually the following selected condition holds, with ρ‾n,m=Kn−1∑kρn,k,m\overline{\rho}_{n,m}=K_n^{-1}\sum_k\rho_{n,k,m}ρ​n,m​=Kn−1​∑k​ρn,k,m​: if www is true, every measurement (Fj)j<J(F_j)_{j<J}(Fj​)j<J​ on EℓE^\ellEℓ with J>0J>0J>0, positive semidefinite effects, and sum III, admits nonnegative real ωj\omega_jωj​ summing to 111 such that

1Mn∑m,j∣Re⁡tr⁡[FjTNc(ℓ)(ρ‾n,m)]−ωj∣≤ε;\frac1{M_n}\sum_{m,j} \left|\operatorname{Re}\operatorname{tr}[F_j\mathcal T_{N_c}^{(\ell)}(\overline{\rho}_{n,m})]-\omega_j\right| \le\varepsilon;Mn​1​m,j∑​​Retr[Fj​TNc​(ℓ)​(ρ​n,m​)]−ωj​​≤ε;

if www is false, some positive semidefinite trace-one matrix ω\omegaω on EℓE^\ellEℓ satisfies

1Mn∑m∥TNc(ℓ)(ρ‾n,m)−ω∥1≤ε.\frac1{M_n}\sum_m\|\mathcal T_{N_c}^{(\ell)}(\overline{\rho}_{n,m})-\omega\|_1\le\varepsilon.Mn​1​m∑​∥TNc​(ℓ)​(ρ​n,m​)−ω∥1​≤ε.

Here ∥Z∥1=Re⁡tr⁡Z†Z\|Z\|_1=\operatorname{Re}\operatorname{tr}\sqrt{Z^\dagger Z}∥Z∥1​=RetrZ†Z​, and

TF(ℓ)(X)y,y′=∑x,x′∈AℓXx,x′∏i=0ℓ−1F(Exi,xi′)yi,yi′\mathcal T_F^{(\ell)}(X)_{y,y'} =\sum_{x,x'\in A^\ell}X_{x,x'}\prod_{i=0}^{\ell-1} F(E_{x_i,x'_i})_{y_i,y'_i}TF(ℓ)​(X)y,y′​=x,x′∈Aℓ∑​Xx,x′​i=0∏ℓ−1​F(Exi​,xi′​​)yi​,yi′​​

for either map FFF, with Ea,a′E_{a,a'}Ea,a′​ a matrix unit. No channel, linearity, or complementary-channel relationship is assumed. This is a supremum in R\mathbb RR, whose total supremum operation returns 000 for an empty or unbounded-above set, rather than an extended-real supremum.

QAlgorithms.Papers.Independent.Intakenext100.P63.marginalMinEntropy

For any finite types A,BA,BA,B with decidable equality and any complex matrix ρ\rhoρ on A×BA\times BA×B, let ρB\rho_BρB​ be the matrix with entries (ρB)bb′=∑a∈Aρ(a,b),(a,b′)(\rho_B)_{bb'}=\sum_{a\in A}\rho_{(a,b),(a,b')}(ρB​)bb′​=∑a∈A​ρ(a,b),(a,b′)​, let f(x)=(x)−1f(x)=(\sqrt{x})^{-1}f(x)=(x​)−1, and set S=IA⊗f(ρB)S=I_A\otimes f(\rho_B)S=IA​⊗f(ρB​), with f(ρB)f(\rho_B)f(ρB​) defined by continuous functional calculus. This defines

−log⁡2∥SρS∥op.-\log_2\|S\rho S\|_{\mathrm{op}}.−log2​∥SρS∥op​.

No density-matrix or Hermitian assumption is imposed on ρ\rhoρ, and A,BA,BA,B may be empty. The real square root and inverse are total, with f(0)=0f(0)=0f(0)=0; for a positive semidefinite ρB\rho_BρB​, zero eigenvalues therefore receive value zero. The logarithm is also total and gives zero when its argument is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.measuredEnsembleJoint

For arbitrary types I,E,JI,E,JI,E,J, with EEE finite, any real weights p:I→Rp:I\to\mathbb Rp:I→R, any family (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ of complex square matrices indexed by EEE, and any family (Qj)j∈J(Q_j)_{j\in J}(Qj​)j∈J​ of such matrices, this defines the real array

Pij=pi Re⁡tr⁡(Qjρi).P_{ij}=p_i\,\operatorname{Re}\operatorname{tr}(Q_j\rho_i).Pij​=pi​Retr(Qj​ρi​).

The weights need not form a probability distribution, the matrices ρi\rho_iρi​ need not be density matrices, and the QjQ_jQj​ need not form a measurement. The types may be empty; if EEE is empty, every value of this array is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.measuredState

For a finite type EEE, a type III with decidable equality, any family of complex matrices MiM_iMi​ on EEE, any natural number nnn, and any complex matrix ψ\psiψ indexed by sequences of length nnn in EEE, this defines a diagonal matrix indexed by sequences iii of length nnn in III. Its diagonal entry at iii is the real number, embedded in C\mathbb CC,

Re⁡tr⁡(Tiψ),(Ti)x,y=∏t=0n−1(Mi(t))x(t),y(t),\operatorname{Re}\operatorname{tr}(T_i\psi), \qquad (T_i)_{x,y}=\prod_{t=0}^{n-1}(M_{i(t)})_{x(t),y(t)},Retr(Ti​ψ),(Ti​)x,y​=t=0∏n−1​(Mi(t)​)x(t),y(t)​,

and every off-diagonal entry is zero. There are no positivity, normalization, or completeness assumptions on MMM or ψ\psiψ, and III need not be finite. For n=0n=0n=0, both sequence index sets are singletons and the unique output entry is Re⁡tr⁡ψ\operatorname{Re}\operatorname{tr}\psiRetrψ.

QAlgorithms.Papers.Independent.Intakenext100.P63.measurementWorstPrior

For arbitrary finite types I,EI,EI,E, with decidable equality on EEE, any family (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ of complex square matrices indexed by EEE, and a measurement consisting of a positive number qqq of positive semidefinite matrices QjQ_jQj​ with ∑jQj=1E\sum_jQ_j=1_E∑j​Qj​=1E​, this defines

sup⁡p:I→Rpi≥0, ∑ipi=1[−∑j=0q−1∑i∈Iaijlog⁡2 ⁣(aij∑k∈Iakj)],aij=piRe⁡tr⁡(Qjρi).\sup_{\substack{p:I\to\mathbb R\\p_i\ge0,\ \sum_i p_i=1}} \left[-\sum_{j=0}^{q-1}\sum_{i\in I} a_{ij}\log_2\!\left(\frac{a_{ij}}{\sum_{k\in I}a_{kj}}\right)\right], \qquad a_{ij}=p_i\operatorname{Re}\operatorname{tr}(Q_j\rho_i).p:I→Rpi​≥0, ∑i​pi​=1​sup​[−j=0∑q−1​i∈I∑​aij​log2​(∑k∈I​akj​aij​​)],aij​=pi​Retr(Qj​ρi​).

The ρi\rho_iρi​ need not be density matrices. Division by zero is zero; the total real logarithm is zero at zero and uses absolute values at negative arguments. This is a real supremum, equal to zero if its defining set is empty or unbounded above; in particular, it is zero when III is empty.

QAlgorithms.Papers.Independent.Intakenext100.P63.messageRateBits

For finite types A,BA,BA,B, with decidable equality on BBB, any natural number nnn, and a keyed block code CCC of block length n+1n+1n+1, this defines

log⁡2Mn+1,M=C.messages.\frac{\log_2 M}{n+1}, \qquad M=C.\mathrm{messages}.n+1log2​M​,M=C.messages.

Such a code has a positive number KKK of keys, and each key supplies a branch code with a positive number MMM of messages, density matrices on An+1A^{n+1}An+1, and MMM positive semidefinite measurement effects on Bn+1B^{n+1}Bn+1 summing to the identity. Thus M>0M>0M>0 and the denominator is positive. The case n=0n=0n=0 is the rate for block length 111.

QAlgorithms.Papers.Independent.Intakenext100.P63.optimizedEquivocation

For arbitrary finite types I,EI,EI,E, with decidable equality on EEE, and any family (ρi)i∈I(\rho_i)_{i\in I}(ρi​)i∈I​ of complex square matrices indexed by EEE, this defines

sup⁡p:I→Rpi≥0, ∑ipi=1 inf⁡q≥1, Qj⪰0∑j=0q−1Qj=1E[−∑j=0q−1∑i∈Iaijlog⁡2 ⁣(aij∑k∈Iakj)],aij=piRe⁡tr⁡(Qjρi).\sup_{\substack{p:I\to\mathbb R\\p_i\ge0,\ \sum_i p_i=1}} \ \inf_{\substack{q\ge1,\ Q_j\succeq0\\\sum_{j=0}^{q-1}Q_j=1_E}} \left[-\sum_{j=0}^{q-1}\sum_{i\in I} a_{ij}\log_2\!\left(\frac{a_{ij}}{\sum_{k\in I}a_{kj}}\right)\right], \qquad a_{ij}=p_i\operatorname{Re}\operatorname{tr}(Q_j\rho_i).p:I→Rpi​≥0, ∑i​pi​=1​sup​ q≥1, Qj​⪰0∑j=0q−1​Qj​=1E​​inf​[−j=0∑q−1​i∈I∑​aij​log2​(∑k∈I​akj​aij​​)],aij​=pi​Retr(Qj​ρi​).

No density-matrix assumption is imposed on the ρi\rho_iρi​. All extrema are taken in R\mathbb RR: an empty or unbounded-below inner set has infimum zero, and an empty or unbounded-above outer set has supremum zero. Division by zero is zero; the real logarithm is zero at zero and uses absolute values for negative arguments. In particular, if III is empty, there is no admissible ppp, and the value is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.povmMinimumShannon

For arbitrary finite types E,IE,IE,I and any family (Mi)i∈I(M_i)_{i\in I}(Mi​)i∈I​ of complex square matrices indexed by EEE, this defines

inf⁡ψ⪰0tr⁡ψ=1[−∑i∈Irilog⁡2ri],ri=Re⁡tr⁡(Miψ).\inf_{\substack{\psi\succeq0\\\operatorname{tr}\psi=1}} \left[-\sum_{i\in I}r_i\log_2 r_i\right], \qquad r_i=\operatorname{Re}\operatorname{tr}(M_i\psi).ψ⪰0trψ=1​inf​[−i∈I∑​ri​log2​ri​],ri​=Retr(Mi​ψ).

No positivity or completeness condition is imposed on the MiM_iMi​, so the rir_iri​ need not be probabilities. The real logarithm is zero at zero and equals log⁡∣ri∣/log⁡2\log|r_i|/\log2log∣ri​∣/log2 otherwise. The infimum is taken in R\mathbb RR, giving zero if the defining set is empty or unbounded below. Empty EEE admits no trace-one matrix and therefore gives zero; empty III also gives zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.qcBranchDensity

For arbitrary types A,JA,JA,J, with decidable equality on JJJ, any function q:J→Rq:J\to\mathbb Rq:J→R, and any family of complex matrices ρj\rho_jρj​ indexed by A×AA\times AA×A, this defines the matrix on A×JA\times JA×J whose entries are

X(a,j),(a′,j′)={qj (ρj)a,a′,j=j′,0,j≠j′.X_{(a,j),(a',j')}= \begin{cases} q_j\,(\rho_j)_{a,a'},&j=j',\\ 0,&j\ne j'. \end{cases}X(a,j),(a′,j′)​={qj​(ρj​)a,a′​,0,​j=j′,j=j′.​

Neither type is required to be finite or nonempty. There are no assumptions that the weights are nonnegative or sum to one, or that the matrices are positive semidefinite or normalized.

QAlgorithms.Papers.Independent.Intakenext100.P63.regularizedUpper

For arbitrary finite types A,B,EA,B,EA,B,E, each with decidable equality, and arbitrary maps N:CA×A→CB×BN:\mathbb C^{A\times A}\to\mathbb C^{B\times B}N:CA×A→CB×B and Nc:CA×A→CE×EN_c:\mathbb C^{A\times A}\to\mathbb C^{E\times E}Nc​:CA×A→CE×E, this defines sup⁡n≥1U(N[n],Nc[n])/n\sup_{n\ge1}U(N^{[n]},N_c^{[n]})/nsupn≥1​U(N[n],Nc[n]​)/n. For any map FFF, its power is defined by

F[n](X)y,y′=∑x,x′Xx,x′∏t<nF(Ext,xt′)yt,yt′,F^{[n]}(X)_{y,y'}=\sum_{x,x'}X_{x,x'}\prod_{t<n}F(E_{x_t,x'_t})_{y_t,y'_t},F[n](X)y,y′​=x,x′∑​Xx,x′​t<n∏​F(Ext​,xt′​​)yt​,yt′​​,

where x,x′x,x'x,x′ are nnn-tuples of input indices, y,y′y,y'y,y′ are nnn-tuples of output indices, and Ea,a′E_{a,a'}Ea,a′​ is a matrix unit. For finite-register maps F,GF,GF,G, U(F,G)U(F,G)U(F,G) is the real supremum over all finite ensembles (pi,ρi)i<m(p_i,\rho_i)_{i<m}(pi​,ρi​)i<m​, with m∈Nm\in\mathbb Nm∈N, pi≥0p_i\ge0pi​≥0, ∑ipi=1\sum_i p_i=1∑i​pi​=1, and each ρi\rho_iρi​ positive semidefinite of trace 111, of

S ⁣(∑ipiF(ρi))−∑ipiS(F(ρi))+∑ipilog⁡2pi+inf⁡Q[−∑j,iPijlog⁡2 ⁣(Pij∑hPhj)],Pij=piRe⁡tr⁡(QjG(ρi)).S\!\left(\sum_i p_iF(\rho_i)\right)-\sum_i p_iS(F(\rho_i)) +\sum_i p_i\log_2p_i +\inf_Q\left[-\sum_{j,i}P_{ij}\log_2\!\left(\frac{P_{ij}}{\sum_hP_{hj}}\right)\right], \quad P_{ij}=p_i\operatorname{Re}\operatorname{tr}(Q_jG(\rho_i)).S(i∑​pi​F(ρi​))−i∑​pi​S(F(ρi​))+i∑​pi​log2​pi​+Qinf​[−j,i∑​Pij​log2​(∑h​Phj​Pij​​)],Pij​=pi​Retr(Qj​G(ρi​)).

The infimum ranges over measurements with a finite, strictly positive number of positive semidefinite effects summing to identity. Here S(X)=−Re⁡tr⁡(f(X))S(X)=-\operatorname{Re}\operatorname{tr}(f(X))S(X)=−Retr(f(X)), with f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x applied by continuous functional calculus. No channel assumptions are imposed; the finite types may be empty, whereas n=0n=0n=0 is excluded and m=0m=0m=0 cannot satisfy normalization. Logarithms, divisions, and real suprema and infima are total; in particular, the relevant supremum or infimum is zero for an empty set or for a set unbounded in the corresponding direction.

QAlgorithms.Papers.Independent.Intakenext100.P63.stateMinBits

For any finite type AAA with decidable equality and any complex matrix ρ\rhoρ indexed by A×AA\times AA×A, this defines the real number

−log⁡2∥ρ∥op,-\log_2\|\rho\|_{\mathrm{op}},−log2​∥ρ∥op​,

where ∥⋅∥op\|\cdot\|_{\mathrm{op}}∥⋅∥op​ is the operator norm for the Euclidean norm. No positivity or trace condition is imposed on ρ\rhoρ, and AAA may be empty. The logarithm is total: its value at zero is zero, so this definition gives zero when ρ=0\rho=0ρ=0, including on an empty index type.

QAlgorithms.Papers.Independent.Intakenext100.P63.supportInvSqrt

For any finite type EEE with decidable equality and any complex square matrix ρ\rhoρ indexed by EEE, this defines f(ρ)f(\rho)f(ρ) by the real continuous functional calculus, where f(x)=(x)−1f(x)=(\sqrt{x})^{-1}f(x)=(x​)−1. For Hermitian ρ\rhoρ, this replaces every positive eigenvalue xxx by 1/x1/\sqrt{x}1/x​ and every nonpositive eigenvalue by zero, since the real square root of a negative number and the inverse of zero are both defined to be zero. For non-Hermitian ρ\rhoρ, this functional-calculus expression is the zero matrix. No positivity, invertibility, or trace condition is assumed; empty EEE is permitted.

QAlgorithms.Papers.Independent.Intakenext100.P63.taggedEffects

For arbitrary types E,J,KE,J,KE,J,K, any real number λ\lambdaλ, and families QjQ_jQj​ and RkR_kRk​ of complex square matrices indexed by EEE, this defines a matrix family indexed by the disjoint union J⊔KJ\sqcup KJ⊔K: its value on a left-tagged j∈Jj\in Jj∈J is λQj\lambda Q_jλQj​, and its value on a right-tagged k∈Kk\in Kk∈K is (1−λ)Rk(1-\lambda)R_k(1−λ)Rk​, with real scalars embedded in C\mathbb CC. No finiteness, positivity, measurement condition, or restriction 0≤λ≤10\le\lambda\le10≤λ≤1 is assumed. Empty types are permitted; at λ=0\lambda=0λ=0 the left branch is zero, and at λ=1\lambda=1λ=1 the right branch is zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.tensorEffects

For arbitrary types E,IE,IE,I, any family Mi∈CE×EM_i\in\mathbb C^{E\times E}Mi​∈CE×E, and any natural number nnn, this defines a family of matrices indexed by label sequences i:{0,…,n−1}→Ii:\{0,\ldots,n-1\}\to Ii:{0,…,n−1}→I, with entries

Ti(x,y)=∏t=0n−1(Mi(t))x(t),y(t)(x,y:{0,…,n−1}→E).T_i(x,y)=\prod_{t=0}^{n-1}(M_{i(t)})_{x(t),y(t)} \qquad \bigl(x,y:\{0,\ldots,n-1\}\to E\bigr).Ti​(x,y)=t=0∏n−1​(Mi(t)​)x(t),y(t)​(x,y:{0,…,n−1}→E).

No finiteness, nonemptiness, positivity, or measurement-completeness assumptions are imposed on E,I,ME,I,ME,I,M. When n=0n=0n=0, the product is empty and equals one, so the matrix for the unique empty label sequence is the 1×11\times11×1 matrix with entry one.

QAlgorithms.Papers.Independent.Intakenext100.P63.traceSmoothConditionalMinBits

For any finite types A,BA,BA,B with decidable equality, real number ε\varepsilonε, and complex matrix ρ\rhoρ on A×BA\times BA×B, this defines the real supremum of the numbers H(r)H(r)H(r) over all matrices r⪰0r\succeq0r⪰0 with tr⁡r=1\operatorname{tr}r=1trr=1 and

12Re⁡tr⁡(ρ−r)†(ρ−r)≤ε,\frac12\operatorname{Re}\operatorname{tr} \sqrt{(\rho-r)^\dagger(\rho-r)}\le\varepsilon,21​Retr(ρ−r)†(ρ−r)​≤ε,

where H(r)H(r)H(r) is the real conversion of the extended-real supremum of all real λ\lambdaλ for which there exists a positive semidefinite matrix σ\sigmaσ on BBB, of trace one, satisfying 2−λ(IA⊗σ)−r⪰02^{-\lambda}(I_A\otimes\sigma)-r\succeq02−λ(IA​⊗σ)−r⪰0. Matrix square roots use continuous functional calculus, and the real conversion sends either extended infinity to zero. No density condition is imposed on ρ\rhoρ, and ε\varepsilonε need not be nonnegative. The outer supremum is in R\mathbb RR, with value zero for an empty set or a set unbounded above; the feasible set is empty if either index type is empty or ε<0\varepsilon<0ε<0.

QAlgorithms.Papers.Independent.Intakenext100.P63.traceSmoothStateMinBits

For any finite type AAA with decidable equality, real number ε\varepsilonε, and complex matrix ρ\rhoρ indexed by A×AA\times AA×A, this defines

sup⁡{−log⁡2∥r∥op:r⪰0, tr⁡r=1, 12Re⁡tr⁡(ρ−r)†(ρ−r)≤ε},\sup\left\{-\log_2\|r\|_{\mathrm{op}}: r\succeq0,\ \operatorname{tr}r=1,\ \frac12\operatorname{Re}\operatorname{tr} \sqrt{(\rho-r)^\dagger(\rho-r)}\le\varepsilon\right\},sup{−log2​∥r∥op​:r⪰0, trr=1, 21​Retr(ρ−r)†(ρ−r)​≤ε},

where the matrix square root is given by continuous functional calculus and the norm is the Euclidean operator norm. There is no density-matrix assumption on ρ\rhoρ and no nonnegativity assumption on ε\varepsilonε. The supremum is taken in R\mathbb RR, whose specified supremum is zero for an empty set or a set unbounded above; in particular, the feasible set is empty if AAA is empty or ε<0\varepsilon<0ε<0. The total logarithm assigns zero at zero.

QAlgorithms.Papers.Independent.Intakenext100.P63.twoBasisDilation

For every natural number ddd, this defines a complex matrix VdV_dVd​ whose columns are indexed by pairs (b,i)∈{false,true}×{0,…,d−1}(b,i)\in\{\mathrm{false},\mathrm{true}\}\times\{0,\ldots,d-1\}(b,i)∈{false,true}×{0,…,d−1} and whose rows are indexed by ((c,j),x)((c,j),x)((c,j),x) with the same kind of pair and x∈{0,…,d−1}x\in\{0,\ldots,d-1\}x∈{0,…,d−1}. Its entries are

(Vd)((c,j),x),(b,i)={0,(c,j)≠(b,i),1{x=i},(c,j)=(b,i), b=false,d−1/2exp⁡(2πi ix/d),(c,j)=(b,i), b=true.(V_d)_{((c,j),x),(b,i)} = \begin{cases} 0,&(c,j)\ne(b,i),\\ \mathbf 1_{\{x=i\}},&(c,j)=(b,i),\ b=\mathrm{false},\\ d^{-1/2}\exp(2\pi\mathrm{i}\,ix/d),&(c,j)=(b,i),\ b=\mathrm{true}. \end{cases}(Vd​)((c,j),x),(b,i)​=⎩⎨⎧​0,1{x=i}​,d−1/2exp(2πiix/d),​(c,j)=(b,i),(c,j)=(b,i), b=false,(c,j)=(b,i), b=true.​

Here i,xi,xi,x in the exponential denote the natural-number values of the indices. The definition includes d=0d=0d=0, when both indexing sets are empty; it does not itself assert any normalization or isometry property.

QAlgorithms.Papers.Independent.Intakenext100.P63.twoBasisVector

For a natural number ddd, a Boolean bbb, and indices i,x∈{0,…,d−1}i,x\in\{0,\ldots,d-1\}i,x∈{0,…,d−1}, this defines the complex number

vb,i(x)={1dexp⁡ ⁣(2πi ixd),b=true,1,b=false and x=i,0,b=false and x≠i,v_{b,i}(x)= \begin{cases} \dfrac1{\sqrt d}\exp\!\left(\dfrac{2\pi\mathrm{i}\,ix}{d}\right),&b=\mathrm{true},\\[4pt] 1,&b=\mathrm{false}\text{ and }x=i,\\ 0,&b=\mathrm{false}\text{ and }x\ne i, \end{cases}vb,i​(x)=⎩⎨⎧​d​1​exp(d2πiix​),1,0,​b=true,b=false and x=i,b=false and x=i,​

where i\mathrm{i}i is the imaginary unit and the index values i,xi,xi,x, the square root, and ddd are coerced into C\mathbb CC in the formula. The parameter d=0d=0d=0 is allowed, but then there are no possible indices i,xi,xi,x, so there are no entries to evaluate.

QAlgorithms.Papers.Independent.Intakenext100.P63.upperInformation

For arbitrary finite types A,B,EA,B,EA,B,E, with decidable equality on B,EB,EB,E, and arbitrary functions N:CA×A→CB×BN:\mathbb C^{A\times A}\to\mathbb C^{B\times B}N:CA×A→CB×B and Nc:CA×A→CE×EN_c:\mathbb C^{A\times A}\to\mathbb C^{E\times E}Nc​:CA×A→CE×E, this defines the real supremum, over natural numbers mmm, weights p1,…,pm≥0p_1,\ldots,p_m\ge0p1​,…,pm​≥0 summing to 111, and positive semidefinite matrices ρ1,…,ρm\rho_1,\ldots,\rho_mρ1​,…,ρm​ of trace 111 on AAA, of

S ⁣(∑ipiN(ρi))−∑ipiS(N(ρi))−H(p)+inf⁡Q[−∑j∑iPijlog⁡2 ⁣(Pij∑kPkj)],Pij=pi Re⁡tr⁡ ⁣(QjNc(ρi)).S\!\left(\sum_i p_iN(\rho_i)\right)-\sum_i p_iS(N(\rho_i)) -H(p) +\inf_Q\left[-\sum_j\sum_i P_{ij}\log_2\!\left(\frac{P_{ij}}{\sum_kP_{kj}}\right)\right], \qquad P_{ij}=p_i\,\operatorname{Re}\operatorname{tr}\!\left(Q_jN_c(\rho_i)\right).S(i∑​pi​N(ρi​))−i∑​pi​S(N(ρi​))−H(p)+Qinf​[−j∑​i∑​Pij​log2​(∑k​Pkj​Pij​​)],Pij​=pi​Retr(Qj​Nc​(ρi​)).

The infimum ranges over all measurements QQQ on EEE with a finite, strictly positive number of outcomes, positive semidefinite effects QjQ_jQj​, and ∑jQj=I\sum_jQ_j=I∑j​Qj​=I. Here H(p)=−∑ipilog⁡2piH(p)=-\sum_i p_i\log_2p_iH(p)=−∑i​pi​log2​pi​, and S(X)=−Re⁡tr⁡(f(X))S(X)=-\operatorname{Re}\operatorname{tr}(f(X))S(X)=−Retr(f(X)), where f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x is applied using the continuous functional calculus. Neither map is assumed linear or a channel, so their outputs are not assumed to be density matrices. The finite types may be empty; m=0m=0m=0 is quantified over but cannot satisfy the weight-sum condition. All logarithms and divisions use their total real-valued definitions, including log⁡20=0\log_2 0=0log2​0=0 and division by zero equal to zero; real suprema and infima here are also total, with value zero for an empty set or, respectively, a set unbounded above or below.

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