Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Remote preparation of quantum states

Definition
QAlgorithms_Papers_Independent_Intakenext100_P54

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

prove2quantum

The objects the formalized statements of Remote preparation of quantum states (arXiv:quant-ph/0307100) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • RSPVector: Normalized vectors in a specified finite register (PDF pp.1–3).
  • bar: Entrywise conjugation in the entanglement basis (PDF pp.2–3).
  • randomise: Averaging unitary conjugations (PDF p.3, equation (1)).
  • Randomising: The operator-order assertion of equation (1), for every density operator (PDF p.3).
  • randomisingBound: The printed randomisation cardinality bound (PDF p.3), with logarithm base two.
  • piGood: Good effects of protocol Pi. Conjugation is on the target alone (PDF p.3).
  • piEffect: Complete POVM alphabet, including the common failure flag (PDF pp.3–4).
  • piBranch: Unnormalized receiver branch after steering and the announced correction (PDF p.4). On a maximally entangled pair the steered operator is the effect's transpose divided by D.
  • RSPCode: One-way finite RSP code with input-independent maximally entangled rank S resource, input-dependent Alice POVM, and fixed CPTP Bob decoder per message (PDF pp.4–5). The source's support costs are log2 K and log2 S.
  • RSPCode.branch: Unnormalized decoded message branch, without conditioning on its probability (PDF p.5).
  • RSPCode.output: Averaged output of a finite one-way code (PDF p.5).
  • RSPCode.ExactWithError: Probabilistic exact preparation: one input-independent success set, exact conditional good outputs, and a uniform failure bound (PDF p.2).
  • pureWordVector: Product target with the source's ordered tensor factors (PDF p.5).
  • pureAverageOverlap: Global average squared fidelity, equation (2), rather than the local criterion (3) (PDF p.5).
  • RSPPureAchievable: One-way achievable rates with both slacks before the blocklength threshold (PDF p.5).
  • pureOperational: Extended nonnegative operational cost: an infeasible rate has value infinity (PDF pp.5–6).
  • AuxChannel: Finite but arbitrarily large classical auxiliary alphabet (PDF p.6, equation (5)).
  • auxWeight: Auxiliary marginal probability (PDF p.6).
  • conditionalBob: Conditional Bob state; a zero-probability auxiliary uses the maximally mixed state. Its entropy is multiplied by zero, so this convention does not affect any information (PDF p.6).
  • auxiliaryBobEntropy: S(B|C), averaged over the classical auxiliary (PDF pp.6,10).
  • labelAuxInfo: I(X;C) for the source's stochastic auxiliary channel (PDF p.6).
  • labelBobAuxInfo: I(X;BC) = I(X;C)+S(B|C)-S(B|X), evaluated from the specified cq state. For pure Bob targets the last term vanishes; for entangled targets it must be retained (PDF pp.6,10).
  • rspSingleLetter: Single-letter optimisation with no unsupported m+1 cardinality restriction. The empty feasible set has infimum infinity (PDF pp.6,10).
  • rspMinimumAttained: Actual attainment whenever the single-letter feasible value is finite; this records "min" rather than merely replacing it with an infimum (PDF pp.6,10).
  • BipartiteEnsemble: Pure bipartite ensemble: the full chosen purification is the target (PDF p.8).
  • bobStates: Receiver marginals of the fixed purifications (PDF pp.8–10).
  • BipartiteCode: General one-way entangled RSP: input-dependent Alice CP instrument, with only its SUM trace preserving; Bob's decoder is arbitrary CPTP (PDF p.8). Padding permits a common finite number L of Kraus operators per branch.
  • BipartiteCode.output: Full joint output, sum_j (M_j^I tensor T_j)(Phi_S); no Alice output is discarded (PDF p.8).
  • bipWord: IID bipartite target, with Alice factors grouped together and Bob factors grouped together (PDF p.8).
  • bipAverage: Global average overlap with the chosen bipartite purification (PDF pp.8–9).
  • BipartiteAchievable: One-way entangled achievable rates with the source's global criterion (PDF pp.8–9).
  • bipOperational: Entangled operational curve, infinity when no finite rate pair is achievable (PDF p.9).
  • bobAverageEntropy: Entropy of the receiver's average state, not the average of the individual entropies (PDF p.9).
  • bobHolevo: Holevo information of the receiver marginal ensemble (PDF p.9).
  • piRecord: Complete receiver record after replacing Pi's failure message uniformly by a good label. The failure flag stays private to Alice (PDF pp.12–13).
  • piSimulator: Linear CPTP simulator with one ACTUAL receiver output specimen as input. Conjugating U, rather than the unknown specimen, makes this a linear channel (PDF p.13).
  • piApproxOutput: Actual averaged output decoded from the uniformly replaced failure record (PDF p.13).
  • marginalAverageFidelity: Bob's global average mixed-state squared fidelity after discarding Alice's output (PDF p.13).
  • SymmetricGaussian: Independent centered real and imaginary coordinates of a symmetric complex Gaussian. Each real coordinate has variance s2/(2D), so the total expected norm square is s2. Definition 22's printed mu_j is read as zero by its immediately preceding prose (PDF p.14).
  • momentLog: Logarithm of a nonnegative exponential moment, including divergence and zero. It never turns a divergent Bochner integral into zero (PDF p.14).
  • squaredGaussianRate: Cramer transform of a squared real centered Gaussian, defined from the moment integral and supremum, not from the asserted closed form (PDF p.14).
  • elementaryRate: Extended endpoint of A3: ln(0) is minus infinity, not Lean's Real.log 0 = 0 (PDF p.14).
  • VisibleCode: Universal visible quantum-classical compression with a pure quantum codeword and a fixed CPTP decoder per classical message (PDF pp.15–16). All code sizes are fixed support sizes.
  • VisibleCode.output: Decoding the jointly visible quantum and classical description (PDF p.15).
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_Channel
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Defs_QuantumInformation
import Definitions.Def_QAlgorithms_Defs_FiniteRegisters
import Definitions.Def_QAlgorithms_Defs_SchurUnitaryEnsembles
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P03
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P53
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P07
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P08
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P38

/-!
Statements from Bennett, Hayden, Leung, Shor and Winter,
Remote preparation of quantum states, quant-ph/0307100v3.
All logarithms in resource bounds are base two. New definitions are paper-local.
Fidelity, iid probabilities, finite pure ensembles and classical entropy reuse pinned accepted
modules whose exact source bodies were hash-verified before reuse.
Theorem 2 and Lemma 4 use the small-error domains justified by their proofs.
Theorem 5 charges the failure message as well as all good messages.
-/

noncomputable section

open scoped BigOperators Matrix ComplexConjugate ComplexOrder NNReal ENNReal
open MeasureTheory ProbabilityTheory

namespace QAlgorithms.Papers.Independent.Intakenext100.P54

open QAlgorithms.Papers.Independent.Information100.P03 (PureEnsemble)
open QAlgorithms.Papers.Independent.Information100.P53 (fidelity)
open QAlgorithms.Papers.Independent.Dependency50.P07 (wordProb)
open QAlgorithms.Papers.Independent.Claimed50.P08 (classicalEntropy)
open QAlgorithms.Papers.Independent.Claimed50.P38 (tensorSuperop)

/-- Normalized vectors in a specified finite register (PDF pp.1–3). -/
abbrev RSPVector (O : Type*) [Fintype O] := {v : EuclideanSpace ℂ O // ‖v‖ = 1}

/-- Entrywise conjugation in the entanglement basis (PDF pp.2–3). -/
def bar {O : Type*} (M : Matrix O O ℂ) : Matrix O O ℂ := fun a b => conj (M a b)

/-- Averaging unitary conjugations (PDF p.3, equation (1)). -/
def randomise {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (rho : Matrix (Fin D) (Fin D) ℂ) : Matrix (Fin D) (Fin D) ℂ :=
  ((K : ℂ)⁻¹) • ∑ k, (U k : Matrix (Fin D) (Fin D) ℂ) * rho *
    (U k : Matrix (Fin D) (Fin D) ℂ)ᴴ

/-- The operator-order assertion of equation (1), for every density operator (PDF p.3). -/
def Randomising {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ) (eps : ℝ) : Prop :=
  ∀ rho : Matrix (Fin D) (Fin D) ℂ, IsDensityMatrix rho →
    (randomise U rho - (((1 - eps) / D : ℝ) : ℂ) • 1).PosSemidef ∧
    ((((1 + eps) / D : ℝ) : ℂ) • 1 - randomise U rho).PosSemidef

/-- The printed randomisation cardinality bound (PDF p.3), with logarithm base two. -/
def randomisingBound (D : ℕ) (eps : ℝ) : ℝ :=
  (10 / eps) ^ 2 * D * Real.logb 2 (20 * D / eps)

/-- Good effects of protocol Pi. Conjugation is on the target alone (PDF p.3). -/
def piGood {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) (k : Fin K) : Matrix (Fin D) (Fin D) ℂ :=
  ((D : ℂ) / (K * (1 + (eps : ℂ)))) •
    ((U k : Matrix (Fin D) (Fin D) ℂ) * bar (pureDensity psi.val) *
      (U k : Matrix (Fin D) (Fin D) ℂ)ᴴ)

/-- Complete POVM alphabet, including the common failure flag (PDF pp.3–4). -/
def piEffect {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) : Option (Fin K) → Matrix (Fin D) (Fin D) ℂ
  | some k => piGood U eps psi k
  | none => 1 - ∑ k, piGood U eps psi k

/-- Unnormalized receiver branch after steering and the announced correction (PDF p.4).
On a maximally entangled pair the steered operator is the effect's transpose divided by D. -/
def piBranch {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) : Option (Fin K) → Matrix (Fin D) (Fin D) ℂ
  | some k => (U k : Matrix (Fin D) (Fin D) ℂ).transpose *
      (((D : ℂ)⁻¹) • (piGood U eps psi k).transpose) *
        ((U k : Matrix (Fin D) (Fin D) ℂ).transpose)ᴴ
  | none => ((D : ℂ)⁻¹) • (piEffect U eps psi none).transpose

/-- Sanity: the failure effect completes precisely the family of good effects. -/
example {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) :
    piEffect U eps psi none + ∑ k, piEffect U eps psi (some k) = 1 := by
  simp only [piEffect]
  exact sub_add_cancel _ _

/-- One-way finite RSP code with input-independent maximally entangled rank S resource,
input-dependent Alice POVM, and fixed CPTP Bob decoder per message (PDF pp.4–5).
The source's support costs are log2 K and log2 S. -/
structure RSPCode (Input O : Type*) [Fintype O] [DecidableEq O] where
  S : ℕ
  Spos : 0 < S
  K : ℕ
  Kpos : 0 < K
  effect : Input → Fin K → Matrix (Fin S) (Fin S) ℂ
  povm : ∀ i, IsPOVM (effect i)
  decoder : Fin K → Superop (Fin S) O
  channel : ∀ j, IsChannel (decoder j)

/-- Unnormalized decoded message branch, without conditioning on its probability (PDF p.5). -/
def RSPCode.branch {Input O : Type*} [Fintype O] [DecidableEq O]
    (P : RSPCode Input O) (i : Input) (j : Fin P.K) : Matrix O O ℂ :=
  P.decoder j (((P.S : ℂ)⁻¹) • (P.effect i j).transpose)

/-- Averaged output of a finite one-way code (PDF p.5). -/
def RSPCode.output {Input O : Type*} [Fintype O] [DecidableEq O]
    (P : RSPCode Input O) (i : Input) : Matrix O O ℂ := ∑ j, P.branch i j

/-- Probabilistic exact preparation: one input-independent success set, exact conditional good
outputs, and a uniform failure bound (PDF p.2). -/
def RSPCode.ExactWithError {O : Type*} [Fintype O] [DecidableEq O]
    (P : RSPCode (RSPVector O) O) (eps : ℝ) : Prop :=
  ∃ good : Finset (Fin P.K), ∀ psi : RSPVector O,
    (∀ j ∈ good, P.branch psi j = ((P.branch psi j).trace.re : ℂ) • pureDensity psi.val) ∧
    1 - eps ≤ ∑ j ∈ good, (P.branch psi j).trace.re

/-- Product target with the source's ordered tensor factors (PDF p.5). -/
def pureWordVector {d : ℕ} (E : PureEnsemble (Fin d)) (n : ℕ) (w : Fin n → Fin E.count) :
    EuclideanSpace ℂ (Fin n → Fin d) :=
  WithLp.toLp 2 fun x => ∏ t, E.vector (w t) (x t)

/-- Global average squared fidelity, equation (2), rather than the local criterion (3) (PDF p.5). -/
def pureAverageOverlap {d : ℕ} (E : PureEnsemble (Fin d)) (n : ℕ)
    (P : RSPCode (Fin n → Fin E.count) (Fin n → Fin d)) : ℝ :=
  ∑ w, wordProb E.weight n w * (pureDensity (pureWordVector E n w) * P.output w).trace.re

/-- One-way achievable rates with both slacks before the blocklength threshold (PDF p.5). -/
def RSPPureAchievable {d : ℕ} (E : PureEnsemble (Fin d)) (R ent : ℝ) : Prop :=
  0 ≤ R ∧ 0 ≤ ent ∧ ∀ eps : ℝ, 0 < eps → ∀ delta : ℝ, 0 < delta →
    ∃ nzero : ℕ, 0 < nzero ∧ ∀ n ≥ nzero,
      ∃ P : RSPCode (Fin n → Fin E.count) (Fin n → Fin d),
        1 - eps ≤ pureAverageOverlap E n P ∧
        Real.logb 2 P.K / n ≤ R + delta ∧ Real.logb 2 P.S / n ≤ ent + delta

/-- Extended nonnegative operational cost: an infeasible rate has value infinity (PDF pp.5–6). -/
def pureOperational {d : ℕ} (E : PureEnsemble (Fin d)) (R : ℝ) : ℝ≥0∞ :=
  sInf {v | ∃ ent : ℝ, RSPPureAchievable E R ent ∧ v = ENNReal.ofReal ent}

/-- Finite but arbitrarily large classical auxiliary alphabet (PDF p.6, equation (5)). -/
structure AuxChannel (m k : ℕ) where
  prob : Fin m → Fin k → ℝ
  nonneg : ∀ i j, 0 ≤ prob i j
  total : ∀ i, ∑ j, prob i j = 1

/-- Auxiliary marginal probability (PDF p.6). -/
def auxWeight {m k : ℕ} (p : Fin m → ℝ) (c : AuxChannel m k) (j : Fin k) : ℝ :=
  ∑ i, p i * c.prob i j

/-- Conditional Bob state; a zero-probability auxiliary uses the maximally mixed state.
Its entropy is multiplied by zero, so this convention does not affect any information (PDF p.6). -/
def conditionalBob {m k d : ℕ} (p : Fin m → ℝ)
    (rho : Fin m → Matrix (Fin d) (Fin d) ℂ) (c : AuxChannel m k) (j : Fin k) :
    Matrix (Fin d) (Fin d) ℂ :=
  if auxWeight p c j = 0 then ((d : ℂ)⁻¹) • 1
  else ∑ i, ((p i * c.prob i j / auxWeight p c j : ℝ) : ℂ) • rho i

/-- S(B|C), averaged over the classical auxiliary (PDF pp.6,10). -/
def auxiliaryBobEntropy {m k d : ℕ} (p : Fin m → ℝ)
    (rho : Fin m → Matrix (Fin d) (Fin d) ℂ) (c : AuxChannel m k) : ℝ :=
  ∑ j, auxWeight p c j * vnEntropy (conditionalBob p rho c j)

/-- I(X;C) for the source's stochastic auxiliary channel (PDF p.6). -/
def labelAuxInfo {m k : ℕ} (p : Fin m → ℝ) (c : AuxChannel m k) : ℝ :=
  classicalEntropy p - ∑ j, auxWeight p c j *
    classicalEntropy (fun i => p i * c.prob i j / auxWeight p c j)

/-- I(X;BC) = I(X;C)+S(B|C)-S(B|X), evaluated from the specified cq state.
For pure Bob targets the last term vanishes; for entangled targets it must be retained (PDF pp.6,10). -/
def labelBobAuxInfo {m k d : ℕ} (p : Fin m → ℝ)
    (rho : Fin m → Matrix (Fin d) (Fin d) ℂ) (c : AuxChannel m k) : ℝ :=
  labelAuxInfo p c + auxiliaryBobEntropy p rho c - ∑ i, p i * vnEntropy (rho i)

/-- Single-letter optimisation with no unsupported m+1 cardinality restriction.
The empty feasible set has infimum infinity (PDF pp.6,10). -/
def rspSingleLetter {m d : ℕ} (p : Fin m → ℝ)
    (rho : Fin m → Matrix (Fin d) (Fin d) ℂ) (R : ℝ) : ℝ≥0∞ :=
  sInf {v | ∃ k : ℕ, ∃ c : AuxChannel m k,
    labelBobAuxInfo p rho c ≤ R ∧ v = ENNReal.ofReal (auxiliaryBobEntropy p rho c)}

/-- Actual attainment whenever the single-letter feasible value is finite; this records
"min" rather than merely replacing it with an infimum (PDF pp.6,10). -/
def rspMinimumAttained {m d : ℕ} (p : Fin m → ℝ)
    (rho : Fin m → Matrix (Fin d) (Fin d) ℂ) (R : ℝ) : Prop :=
  rspSingleLetter p rho R ≠ ⊤ → ∃ k : ℕ, ∃ c : AuxChannel m k,
    labelBobAuxInfo p rho c ≤ R ∧
    ENNReal.ofReal (auxiliaryBobEntropy p rho c) = rspSingleLetter p rho R

/-- Pure bipartite ensemble: the full chosen purification is the target (PDF p.8). -/
structure BipartiteEnsemble (m a b : ℕ) where
  prob : Fin m → ℝ
  nonneg : ∀ i, 0 ≤ prob i
  total : ∑ i, prob i = 1
  state : Fin m → RSPVector (Fin a × Fin b)

/-- Receiver marginals of the fixed purifications (PDF pp.8–10). -/
def bobStates {m a b : ℕ} (E : BipartiteEnsemble m a b) (i : Fin m) :
    Matrix (Fin b) (Fin b) ℂ := traceLeft (pureDensity (E.state i).val)

/-- General one-way entangled RSP: input-dependent Alice CP instrument, with only its SUM
trace preserving; Bob's decoder is arbitrary CPTP (PDF p.8). Padding permits a common finite
number L of Kraus operators per branch. -/
structure BipartiteCode (Input A B : Type*) [Fintype A] [Fintype B] [DecidableEq B] where
  S : ℕ
  Spos : 0 < S
  K : ℕ
  Kpos : 0 < K
  L : ℕ
  alice : Input → Fin K → Fin L → Matrix A (Fin S) ℂ
  complete : ∀ i, ∑ j, ∑ l, (alice i j l)ᴴ * alice i j l = 1
  bob : Fin K → Superop (Fin S) B
  channel : ∀ j, IsChannel (bob j)

/-- Full joint output, sum_j (M_j^I tensor T_j)(Phi_S); no Alice output is discarded (PDF p.8). -/
def BipartiteCode.output {Input A B : Type*} [Fintype A] [Fintype B] [DecidableEq B]
    (P : BipartiteCode Input A B) (i : Input) : Matrix (A × B) (A × B) ℂ :=
  letI : Nonempty (Fin P.S) := ⟨⟨0, P.Spos⟩⟩
  let s : Fin P.S := ⟨0, P.Spos⟩
  -- The frozen tensor operation has square registers. Extract each rectangular output
  -- coefficient through constant-entry square maps; its matrix-unit expansion is unchanged.
  ∑ j, fun u v =>
    tensorSuperop
      (fun X (_ _ : Fin P.S) => krausApply (P.alice i j) X u.1 v.1)
      (fun X (_ _ : Fin P.S) => P.bob j X u.2 v.2)
      (pureDensity (finiteMaxEntangled (Fin P.S))) (s, s) (s, s)

/-- Coefficient extraction through the imported square tensor operation retains every
matrix unit, including off-diagonal entries of entangled inputs (PDF p.8). -/
example {S A B : Type*} [Fintype S] [DecidableEq S]
    (left : Superop S A) (right : Superop S B)
    (X : Matrix (S × S) (S × S) ℂ) (s : S) (u v : A × B) :
    tensorSuperop (fun Y (_ _ : S) => left Y u.1 v.1)
      (fun Y (_ _ : S) => right Y u.2 v.2) X (s, s) (s, s) =
      ∑ x : S × S, ∑ y : S × S, X x y *
        left (Matrix.single x.1 y.1 1) u.1 v.1 *
        right (Matrix.single x.2 y.2 1) u.2 v.2 := by
  rfl

/-- IID bipartite target, with Alice factors grouped together and Bob factors grouped together
(PDF p.8). -/
def bipWord {m a b : ℕ} (E : BipartiteEnsemble m a b) (n : ℕ) (w : Fin n → Fin m) :
    EuclideanSpace ℂ ((Fin n → Fin a) × (Fin n → Fin b)) :=
  WithLp.toLp 2 fun x => ∏ t, (E.state (w t)).val (x.1 t, x.2 t)

/-- Global average overlap with the chosen bipartite purification (PDF pp.8–9). -/
def bipAverage {m a b : ℕ} (E : BipartiteEnsemble m a b) (n : ℕ)
    (P : BipartiteCode (Fin n → Fin m) (Fin n → Fin a) (Fin n → Fin b)) : ℝ :=
  ∑ w, wordProb E.prob n w * (pureDensity (bipWord E n w) * P.output w).trace.re

/-- One-way entangled achievable rates with the source's global criterion (PDF pp.8–9). -/
def BipartiteAchievable {m a b : ℕ} (E : BipartiteEnsemble m a b) (R ent : ℝ) : Prop :=
  0 ≤ R ∧ 0 ≤ ent ∧ ∀ eps : ℝ, 0 < eps → ∀ delta : ℝ, 0 < delta →
    ∃ nzero : ℕ, 0 < nzero ∧ ∀ n ≥ nzero,
      ∃ P : BipartiteCode (Fin n → Fin m) (Fin n → Fin a) (Fin n → Fin b),
        1 - eps ≤ bipAverage E n P ∧
        Real.logb 2 P.K / n ≤ R + delta ∧ Real.logb 2 P.S / n ≤ ent + delta

/-- Entangled operational curve, infinity when no finite rate pair is achievable (PDF p.9). -/
def bipOperational {m a b : ℕ} (E : BipartiteEnsemble m a b) (R : ℝ) : ℝ≥0∞ :=
  sInf {v | ∃ ent : ℝ, BipartiteAchievable E R ent ∧ v = ENNReal.ofReal ent}

/-- Entropy of the receiver's average state, not the average of the individual entropies (PDF p.9). -/
def bobAverageEntropy {m a b : ℕ} (E : BipartiteEnsemble m a b) : ℝ :=
  vnEntropy (∑ i, (E.prob i : ℂ) • bobStates E i)

/-- Holevo information of the receiver marginal ensemble (PDF p.9). -/
def bobHolevo {m a b : ℕ} (E : BipartiteEnsemble m a b) : ℝ :=
  bobAverageEntropy E - ∑ i, E.prob i * vnEntropy (bobStates E i)

/-- Complete receiver record after replacing Pi's failure message uniformly by a good label.
The failure flag stays private to Alice (PDF pp.12–13). -/
def piRecord {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) : Matrix (Fin K × Fin D) (Fin K × Fin D) ℂ :=
  fun x y => if x.1 = y.1 then
    (((D : ℂ)⁻¹) • (piGood U eps psi x.1).transpose +
      (((K : ℂ) * D)⁻¹) • (piEffect U eps psi none).transpose) x.2 y.2
    else 0

/-- Linear CPTP simulator with one ACTUAL receiver output specimen as input.
Conjugating U, rather than the unknown specimen, makes this a linear channel (PDF p.13). -/
def piSimulator {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ) :
    Superop (Fin D) (Fin K × Fin D) :=
  fun rho x y => if x.1 = y.1 then
    (((K : ℂ)⁻¹) • (bar (U x.1 : Matrix (Fin D) (Fin D) ℂ) * rho *
      (U x.1 : Matrix (Fin D) (Fin D) ℂ).transpose)) x.2 y.2 else 0

/-- Actual averaged output decoded from the uniformly replaced failure record (PDF p.13). -/
def piApproxOutput {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ)
    (eps : ℝ) (psi : RSPVector (Fin D)) : Matrix (Fin D) (Fin D) ℂ :=
  ∑ k, (U k : Matrix (Fin D) (Fin D) ℂ).transpose *
    (Matrix.of fun a b => piRecord U eps psi (k, a) (k, b)) *
    ((U k : Matrix (Fin D) (Fin D) ℂ).transpose)ᴴ

/-- Bob's global average mixed-state squared fidelity after discarding Alice's output (PDF p.13). -/
def marginalAverageFidelity {m a b : ℕ} (E : BipartiteEnsemble m a b) (n : ℕ)
    (P : BipartiteCode (Fin n → Fin m) (Fin n → Fin a) (Fin n → Fin b)) : ℝ :=
  ∑ w, wordProb E.prob n w * fidelity (traceLeft (pureDensity (bipWord E n w)))
    (traceLeft (P.output w))

/-- Independent centered real and imaginary coordinates of a symmetric complex Gaussian.
Each real coordinate has variance s2/(2D), so the total expected norm square is s2.
Definition 22's printed mu_j is read as zero by its immediately preceding prose (PDF p.14). -/
def SymmetricGaussian {Omega : Type*} [MeasurableSpace Omega] (mu : Measure Omega)
    (D : ℕ) (s2 : ℝ≥0) (G : Omega → EuclideanSpace ℂ (Fin D)) : Prop :=
  Measurable G ∧
    let coord : (Fin D × Bool) → Omega → ℝ := fun j w =>
      if j.2 then (G w j.1).im else (G w j.1).re
    iIndepFun coord mu ∧
      ∀ j, Measure.map (coord j) mu = gaussianReal 0 (s2 / (2 * D))

/-- Logarithm of a nonnegative exponential moment, including divergence and zero.
It never turns a divergent Bochner integral into zero (PDF p.14). -/
def momentLog (z : ℝ≥0∞) : EReal :=
  if z = ⊤ then ⊤ else if z = 0 then ⊥ else (Real.log z.toReal : EReal)

/-- Cramer transform of a squared real centered Gaussian, defined from the moment integral
and supremum, not from the asserted closed form (PDF p.14). -/
def squaredGaussianRate (s2 : ℝ≥0) (x : ℝ) : EReal :=
  ⨆ y : ℝ, ((y * x : ℝ) : EReal) - momentLog
    (∫⁻ t : ℝ, ENNReal.ofReal (Real.exp (y * t ^ 2)) ∂gaussianReal 0 s2)

/-- Sanity: a divergent moment has infinite logarithm. -/
example : momentLog ⊤ = ⊤ := by simp [momentLog]

/-- Sanity: the unit moment has zero logarithm. -/
example : momentLog 1 = 0 := by simp [momentLog]

/-- Extended endpoint of A3: ln(0) is minus infinity, not Lean's Real.log 0 = 0 (PDF p.14). -/
def elementaryRate (xi : ℝ) : EReal :=
  if xi = -1 then ⊤ else (((xi - Real.log (1 + xi)) / (2 * Real.log 2) : ℝ) : EReal)

/-- Sanity: the left endpoint is infinite. -/
example : elementaryRate (-1) = ⊤ := by simp [elementaryRate]

/-- Universal visible quantum-classical compression with a pure quantum codeword and a fixed
CPTP decoder per classical message (PDF pp.15–16). All code sizes are fixed support sizes. -/
structure VisibleCode (D : ℕ) where
  S : ℕ
  Spos : 0 < S
  K : ℕ
  Kpos : 0 < K
  quantum : RSPVector (Fin D) → RSPVector (Fin S)
  message : RSPVector (Fin D) → Fin K
  decoder : Fin K → Superop (Fin S) (Fin D)
  channel : ∀ j, IsChannel (decoder j)

/-- Decoding the jointly visible quantum and classical description (PDF p.15). -/
def VisibleCode.output {D : ℕ} (P : VisibleCode D) (psi : RSPVector (Fin D)) :
    Matrix (Fin D) (Fin D) ℂ := P.decoder (P.message psi) (pureDensity (P.quantum psi).val)

/-- Sanity: tensor-product word probability on the empty word is one. -/
example {m : ℕ} (p : Fin m → ℝ) (w : Fin 0 → Fin m) : wordProb p 0 w = 1 := by
  simp [wordProb]

/-- Sanity: the auxiliary register's impossible outcome contributes zero conditional entropy. -/
example {m k d : ℕ} (p : Fin m → ℝ) (rho : Fin m → Matrix (Fin d) (Fin d) ℂ)
    (c : AuxChannel m k) (j : Fin k) (h : auxWeight p c j = 0) :
    auxWeight p c j * vnEntropy (conditionalBob p rho c j) = 0 := by rw [h, zero_mul]

/-- Sanity: entrywise conjugation is an involution, including on non-real targets. -/
example {O : Type*} (M : Matrix O O ℂ) : bar (bar M) = M := by
  ext a b
  simp [bar]

/-- Sanity: the invalid empty randomising family is zero; the theorem excludes K = 0. -/
example {D : ℕ} (U : Fin 0 → Matrix.unitaryGroup (Fin D) ℂ)
    (rho : Matrix (Fin D) (Fin D) ℂ) : randomise U rho = 0 := by
  simp [randomise]

/-- Sanity: the record simulator is linear at zero, rather than conjugating its unknown input. -/
example {D K : ℕ} (U : Fin K → Matrix.unitaryGroup (Fin D) ℂ) : piSimulator U 0 = 0 := by
  ext x y
  simp [piSimulator]

/-- Sanity: the empty product target is the unit scalar. -/
example {d : ℕ} (E : PureEnsemble (Fin d)) (w : Fin 0 → Fin E.count) (x : Fin 0 → Fin d) :
    pureWordVector E 0 w x = 1 := by simp [pureWordVector]

/-- Sanity: the rate estimate at the mean is zero. -/
example : elementaryRate 0 = 0 := by simp [elementaryRate]

end QAlgorithms.Papers.Independent.Intakenext100.P54
Source
Remote preparation of quantum states, arXiv:quant-ph/0307100v3, https://arxiv.org/abs/quant-ph/0307100v3
Read-back

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

QAlgorithms.Papers.Independent.Intakenext100.P54.AuxChannel

For arbitrary natural numbers m,km,km,k, an element of this structure is an array of real numbers ci,jc_{i,j}ci,j​, indexed by 0≤i<m0\le i<m0≤i<m and 0≤j<k0\le j<k0≤j<k, together with the requirements

ci,j≥0for all i,j,∑j=0k−1ci,j=1for every i.c_{i,j}\ge0\quad\text{for all }i,j, \qquad \sum_{j=0}^{k-1}c_{i,j}=1\quad\text{for every }i.ci,j​≥0for all i,j,j=0∑k−1​ci,j​=1for every i.

There is no explicit positivity assumption on mmm or kkk. If m=0m=0m=0, all requirements are vacuous, including when k=0k=0k=0; if m>0m>0m>0 and k=0k=0k=0, the row-sum requirement would be 0=10=10=1, so no such array exists.

QAlgorithms.Papers.Independent.Intakenext100.P54.BipartiteAchievable

For natural numbers m,a,bm,a,bm,a,b, let EEE consist of probabilities pi≥0p_i\ge0pi​≥0 with ∑i<mpi=1\sum_{i<m}p_i=1∑i<m​pi​=1 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^a\otimes\mathbb C^bψi​∈Ca⊗Cb, and let R,e∈RR,e\in\mathbb RR,e∈R. The predicate asserts R≥0R\ge0R≥0, e≥0e\ge0e≥0, and the following: for every ε>0\varepsilon>0ε>0 and δ>0\delta>0δ>0, there is a positive integer n0n_0n0​ such that for every natural number n≥n0n\ge n_0n≥n0​ there is a code with positive integers S,KS,KS,K, a natural number LLL, matrices Aw,j,ℓ:CS→CanA_{w,j,\ell}:\mathbb C^S\to\mathbb C^{a^n}Aw,j,ℓ​:CS→Can satisfying ∑j<K,ℓ<LAw,j,ℓ†Aw,j,ℓ=IS\sum_{j<K,\ell<L}A_{w,j,\ell}^{\dagger}A_{w,j,\ell}=I_S∑j<K,ℓ<L​Aw,j,ℓ†​Aw,j,ℓ​=IS​ for each length-nnn word www over mmm labels, and maps Bj(X)=∑hBj,hXBj,h†\mathcal B_j(X)=\sum_h B_{j,h}XB_{j,h}^{\dagger}Bj​(X)=∑h​Bj,h​XBj,h†​ with finitely many Kraus matrices satisfying ∑hBj,h†Bj,h=IS\sum_hB_{j,h}^{\dagger}B_{j,h}=I_S∑h​Bj,h†​Bj,h​=IS​, such that

1−ε≤∑w(∏t<npwt)Re⁡Tr⁡(∣Ψw⟩⟨Ψw∣Ωw),log⁡2Kn≤R+δ,log⁡2Sn≤e+δ.1-\varepsilon\le \sum_w\left(\prod_{t<n}p_{w_t}\right) \operatorname{Re}\operatorname{Tr} \bigl(|\Psi_w\rangle\langle\Psi_w|\Omega_w\bigr), \qquad \frac{\log_2K}{n}\le R+\delta, \qquad \frac{\log_2S}{n}\le e+\delta.1−ε≤w∑​(t<n∏​pwt​​)ReTr(∣Ψw​⟩⟨Ψw​∣Ωw​),nlog2​K​≤R+δ,nlog2​S​≤e+δ.

Here Ψw(x,y)=∏t<nψwt(xt,yt)\Psi_w(x,y)=\prod_{t<n}\psi_{w_t}(x_t,y_t)Ψw​(x,y)=∏t<n​ψwt​​(xt​,yt​) and

Ωw((x,y),(x′,y′))=1S∑j,ℓ∑u,v<SAw,j,ℓ(x,u)Aw,j,ℓ(x′,v)‾ Bj(∣u⟩⟨v∣)(y,y′).\Omega_w((x,y),(x',y'))= \frac1S\sum_{j,\ell}\sum_{u,v<S} A_{w,j,\ell}(x,u)\overline{A_{w,j,\ell}(x',v)} \,\mathcal B_j(|u\rangle\langle v|)(y,y').Ωw​((x,y),(x′,y′))=S1​j,ℓ∑​u,v<S∑​Aw,j,ℓ​(x,u)Aw,j,ℓ​(x′,v)​Bj​(∣u⟩⟨v∣)(y,y′).

The code may depend on ε,δ,n\varepsilon,\delta,nε,δ,n. No upper bound on ε\varepsilonε or δ\deltaδ is imposed, and only positive block lengths are required. Zero label or local dimensions cannot satisfy the ensemble requirements.

QAlgorithms.Papers.Independent.Intakenext100.P54.BipartiteCode

For an arbitrary input type III, finite types A,BA,BA,B, and a specified decidable equality on BBB, this structure consists of positive natural numbers S,KS,KS,K, an arbitrary natural number LLL, complex matrices Ai,j,ℓA_{i,j,\ell}Ai,j,ℓ​ with rows indexed by AAA and columns indexed by {0,…,S−1}\{0,\ldots,S-1\}{0,…,S−1}, for every i∈Ii\in Ii∈I, j<Kj<Kj<K, and ℓ<L\ell<Lℓ<L, satisfying

∑j<K∑ℓ<LAi,j,ℓ†Ai,j,ℓ=ISfor every i∈I,\sum_{j<K}\sum_{\ell<L}A_{i,j,\ell}^{\dagger}A_{i,j,\ell}=I_S \quad\text{for every }i\in I,j<K∑​ℓ<L∑​Ai,j,ℓ†​Ai,j,ℓ​=IS​for every i∈I,

and maps Bj\mathcal B_jBj​ from complex S×SS\times SS×S matrices to complex matrices indexed by B×BB\times BB×B. For every j<Kj<Kj<K, the map Bj\mathcal B_jBj​ must admit a finite Kraus representation: there exist a natural number rjr_jrj​ and matrices Ej,hE_{j,h}Ej,h​ with rows indexed by BBB and SSS columns such that

∑h<rjEj,h†Ej,h=IS,Bj(X)=∑h<rjEj,hXEj,h†for every X.\sum_{h<r_j}E_{j,h}^{\dagger}E_{j,h}=I_S, \qquad \mathcal B_j(X)=\sum_{h<r_j}E_{j,h}XE_{j,h}^{\dagger} \quad\text{for every }X.h<rj​∑​Ej,h†​Ej,h​=IS​,Bj​(X)=h<rj​∑​Ej,h​XEj,h†​for every X.

No finiteness or nonemptiness is assumed for III, and nonemptiness of AAA or BBB is not an explicit assumption. The stated conditions force BBB to be nonempty; if III is nonempty, they also force AAA to be nonempty and L>0L>0L>0. When III is empty, the conditions indexed by iii are vacuous.

QAlgorithms.Papers.Independent.Intakenext100.P54.BipartiteCode.output

For an arbitrary input type III, finite types A,BA,BA,B with decidable equality on BBB, a code PPP, and i∈Ii\in Ii∈I, write S,K>0S,K>0S,K>0 and L∈NL\in\mathbb NL∈N for the code’s parameters, Ai,j,ℓA_{i,j,\ell}Ai,j,ℓ​ for its matrices from CS\mathbb C^SCS to CA\mathbb C^ACA, and Bj\mathcal B_jBj​ for its maps from S×SS\times SS×S matrices to matrices on BBB. These satisfy ∑j,ℓAi′,j,ℓ†Ai′,j,ℓ=IS\sum_{j,\ell}A_{i',j,\ell}^{\dagger}A_{i',j,\ell}=I_S∑j,ℓ​Ai′,j,ℓ†​Ai′,j,ℓ​=IS​ for every input i′i'i′, and each Bj\mathcal B_jBj​ has a finite Kraus representation Bj(X)=∑hEj,hXEj,h†\mathcal B_j(X)=\sum_hE_{j,h}XE_{j,h}^{\dagger}Bj​(X)=∑h​Ej,h​XEj,h†​ with ∑hEj,h†Ej,h=IS\sum_hE_{j,h}^{\dagger}E_{j,h}=I_S∑h​Ej,h†​Ej,h​=IS​. The definition returns the matrix Ωi\Omega_iΩi​ on A×BA\times BA×B whose entries, after expanding the tensor operation, Kraus application, and maximally entangled vector, are

Ωi((a,b),(a′,b′))=1S∑j<K∑r<S∑t<S∑ℓ<LAi,j,ℓ(a,r) Ai,j,ℓ(a′,t)‾ [Bj(Ert)]bb′,\Omega_i\bigl((a,b),(a',b')\bigr) =\frac1S \sum_{j<K}\sum_{r<S}\sum_{t<S}\sum_{\ell<L} A_{i,j,\ell}(a,r)\, \overline{A_{i,j,\ell}(a',t)}\, \bigl[\mathcal B_j(E_{rt})\bigr]_{bb'},Ωi​((a,b),(a′,b′))=S1​j<K∑​r<S∑​t<S∑​ℓ<L∑​Ai,j,ℓ​(a,r)Ai,j,ℓ​(a′,t)​[Bj​(Ert​)]bb′​,

where ErtE_{rt}Ert​ is the S×SS\times SS×S matrix with a single 111 in position (r,t)(r,t)(r,t). The maximally entangled vector used in the definition has amplitude S−1/2S^{-1/2}S−1/2 at (r,r)(r,r)(r,r) and zero elsewhere; the evaluation of the constant-entry square maps at the distinguished index 0<S0<S0<S gives the displayed coefficients. There is no distribution on inputs and no averaging over iii. The selected input and the code’s conditions force A,BA,BA,B to be nonempty and L>0L>0L>0.

QAlgorithms.Papers.Independent.Intakenext100.P54.BipartiteEnsemble

For natural numbers m,a,bm,a,bm,a,b, this structure consists of real numbers pip_ipi​ indexed by i<mi<mi<m, proofs that every pi≥0p_i\ge0pi​≥0 and ∑i<mpi=1\sum_{i<m}p_i=1∑i<m​pi​=1, and, for each i<mi<mi<m, a vector ψi∈Ca×b\psi_i\in\mathbb C^{a\times b}ψi​∈Ca×b satisfying ∥ψi∥=1\|\psi_i\|=1∥ψi​∥=1. Zero probabilities are permitted, and the unit-vector requirement applies even at indices of zero probability. Although the parameters are arbitrary natural numbers, such a structure cannot exist when m=0m=0m=0, or when either a=0a=0a=0 or b=0b=0b=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPCode.ExactWithError

Let OOO be a finite type with decidable equality, let ε∈R\varepsilon\in\mathbb Rε∈R, and let PPP be a code whose inputs are all vectors ψ∈CO\psi\in\mathbb C^Oψ∈CO with Euclidean norm 111. Thus PPP has positive integers S,KS,KS,K, positive semidefinite matrices Eψ,jE_{\psi,j}Eψ,j​ summing to ISI_SIS​ for each unit vector ψ\psiψ, and decoder maps Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​ with finite Kraus families satisfying ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​. Write Bψ,j=Dj(S−1Eψ,jT)B_{\psi,j}=\mathcal D_j(S^{-1}E_{\psi,j}^{\mathsf T})Bψ,j​=Dj​(S−1Eψ,jT​). The defined property holds exactly when there exists a single subset G⊆{0,…,K−1}G\subseteq\{0,\ldots,K-1\}G⊆{0,…,K−1}, chosen independently of ψ\psiψ, such that for every unit vector ψ\psiψ,

∀j∈G,Bψ,j=Re⁡Tr⁡(Bψ,j) ψψ†,1−ε≤∑j∈GRe⁡Tr⁡(Bψ,j).\forall j\in G,\quad B_{\psi,j}=\operatorname{Re}\operatorname{Tr}(B_{\psi,j})\,\psi\psi^\dagger, \qquad 1-\varepsilon\le\sum_{j\in G}\operatorname{Re}\operatorname{Tr}(B_{\psi,j}).∀j∈G,Bψ,j​=ReTr(Bψ,j​)ψψ†,1−ε≤j∈G∑​ReTr(Bψ,j​).

The matrix ψψ†\psi\psi^\daggerψψ† has entries ψaψb‾\psi_a\overline{\psi_b}ψa​ψb​​. There is no imposed range restriction on ε\varepsilonε, and GGG may be empty; in particular, every such code satisfies this property with G=∅G=\varnothingG=∅ when ε≥1\varepsilon\ge1ε≥1.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPCode.branch

For any input type III, finite output type OOO with decidable equality, and code PPP with positive dimensions S,KS,KS,K, input-dependent positive semidefinite matrices Ei,jE_{i,j}Ei,j​ summing to ISI_SIS​, and decoder maps Dj\mathcal D_jDj​ admitting finite representations Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​ with ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​, the branch matrix for an input i∈Ii\in Ii∈I and an index 0≤j<K0\le j<K0≤j<K is defined by

Bi,j=Dj ⁣(1SEi,jT).B_{i,j}=\mathcal D_j\!\left(\frac{1}{S}E_{i,j}^{\mathsf T}\right).Bi,j​=Dj​(S1​Ei,jT​).

Here T{}^{\mathsf T}T denotes ordinary transpose, without complex conjugation; the result is an O×OO\times OO×O complex matrix and is not divided by its trace.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPCode

For an arbitrary input type III and a finite output type OOO with decidable equality, an element of this structure consists of positive integers S,KS,KS,K, matrices Ei,j∈CS×SE_{i,j}\in\mathbb C^{S\times S}Ei,j​∈CS×S for every i∈Ii\in Ii∈I and 0≤j<K0\le j<K0≤j<K, and functions Dj:CS×S→CO×O\mathcal D_j:\mathbb C^{S\times S}\to\mathbb C^{O\times O}Dj​:CS×S→CO×O. For each input iii, every Ei,jE_{i,j}Ei,j​ is positive semidefinite and ∑j=0K−1Ei,j=IS\sum_{j=0}^{K-1}E_{i,j}=I_S∑j=0K−1​Ei,j​=IS​. For each jjj, the function Dj\mathcal D_jDj​ has a finite Kraus representation: there are a natural number mjm_jmj​ and matrices Aj,ℓ∈CO×SA_{j,\ell}\in\mathbb C^{O\times S}Aj,ℓ​∈CO×S such that ∑ℓ=0mj−1Aj,ℓ†Aj,ℓ=IS\sum_{\ell=0}^{m_j-1}A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ=0mj​−1​Aj,ℓ†​Aj,ℓ​=IS​ and Dj(X)=∑ℓ=0mj−1Aj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_{\ell=0}^{m_j-1}A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ=0mj​−1​Aj,ℓ​XAj,ℓ†​ for every matrix XXX. No finiteness or nonemptiness assumption is imposed on III; its requirements are vacuous when III is empty. Although OOO is not explicitly assumed nonempty, an empty OOO cannot satisfy the decoder requirements because S>0S>0S>0.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPCode.output

For any input type III, finite output type OOO with decidable equality, and code PPP consisting of positive integers S,KS,KS,K, positive semidefinite matrices Ei,jE_{i,j}Ei,j​ satisfying ∑j=0K−1Ei,j=IS\sum_{j=0}^{K-1}E_{i,j}=I_S∑j=0K−1​Ei,j​=IS​ for each iii, and decoder maps Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​ with finite Kraus families satisfying ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​, the output matrix at i∈Ii\in Ii∈I is defined to be

σi=∑j=0K−1Dj ⁣(1SEi,jT),\sigma_i=\sum_{j=0}^{K-1}\mathcal D_j\!\left(\frac{1}{S}E_{i,j}^{\mathsf T}\right),σi​=j=0∑K−1​Dj​(S1​Ei,jT​),

where T{}^{\mathsf T}T is ordinary transpose.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPPureAchievable

For a natural number ddd, an ensemble EEE of LLL unit vectors vℓ∈Cdv_\ell\in\mathbb C^dvℓ​∈Cd with nonnegative weights pℓp_\ellpℓ​ summing to 111, and real numbers R,eR,eR,e, this property means that R≥0R\ge0R≥0, e≥0e\ge0e≥0, and for every real ε>0\varepsilon>0ε>0 and δ>0\delta>0δ>0, there exists a natural number n0>0n_0>0n0​>0 such that for every natural number n≥n0n\ge n_0n≥n0​ there is a code with positive integers S,KS,KS,K, input words w∈{0,…,L−1}nw\in\{0,\ldots,L-1\}^nw∈{0,…,L−1}n, positive semidefinite effects Ew,jE_{w,j}Ew,j​ summing to ISI_SIS​, and decoders into matrices indexed by {0,…,d−1}n\{0,\ldots,d-1\}^n{0,…,d−1}n having finite Kraus representations Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​, ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​, for which

1−ε≤∑w(∏t=0n−1pw(t))Re⁡Tr⁡ ⁣(vwvw†∑j=0K−1Dj(S−1Ew,jT)),log⁡2Kn≤R+δ,log⁡2Sn≤e+δ,1-\varepsilon\le \sum_w\left(\prod_{t=0}^{n-1}p_{w(t)}\right) \operatorname{Re}\operatorname{Tr}\!\left( v_wv_w^\dagger\sum_{j=0}^{K-1}\mathcal D_j(S^{-1}E_{w,j}^{\mathsf T}) \right), \qquad \frac{\log_2K}{n}\le R+\delta, \qquad \frac{\log_2S}{n}\le e+\delta,1−ε≤w∑​(t=0∏n−1​pw(t)​)ReTr(vw​vw†​j=0∑K−1​Dj​(S−1Ew,jT​)),nlog2​K​≤R+δ,nlog2​S​≤e+δ,

where vw(x)=∏t=0n−1vw(t)(x(t))v_w(x)=\prod_{t=0}^{n-1}v_{w(t)}(x(t))vw​(x)=∏t=0n−1​vw(t)​(x(t)). The same code must satisfy both rate inequalities and the overlap inequality, but may depend on n,ε,δn,\varepsilon,\deltan,ε,δ; no uniform construction is required. The quantified block lengths are positive, so these divisions never use n=0n=0n=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.RSPVector

For any finite type OOO, this denotes the set of complex vectors v=(vo)o∈Ov=(v_o)_{o\in O}v=(vo​)o∈O​ with Euclidean norm ∥v∥2=1\|v\|_2=1∥v∥2​=1, equivalently ∑o∈O∣vo∣2=1\sum_{o\in O}|v_o|^2=1∑o∈O​∣vo​∣2=1. If OOO is empty, this set is empty.

QAlgorithms.Papers.Independent.Intakenext100.P54.Randomising

For any natural numbers D,KD,KD,K, any family U0,…,UK−1U_0,\ldots,U_{K-1}U0​,…,UK−1​ of unitary complex D×DD\times DD×D matrices, and any real ε\varepsilonε, this predicate means that every positive semidefinite complex D×DD\times DD×D matrix ρ\rhoρ with tr⁡ρ=1\operatorname{tr}\rho=1trρ=1 satisfies

1−εDID⪯K−1∑k=0K−1UkρUk†⪯1+εDID,\frac{1-\varepsilon}{D}I_D \preceq K^{-1}\sum_{k=0}^{K-1}U_k\rho U_k^\dagger \preceq \frac{1+\varepsilon}{D}I_D,D1−ε​ID​⪯K−1k=0∑K−1​Uk​ρUk†​⪯D1+ε​ID​,

where A⪯BA\preceq BA⪯B means that B−AB-AB−A is positive semidefinite. The predicate itself imposes no positivity assumptions on D,K,εD,K,\varepsilonD,K,ε. Division by zero is interpreted as zero. For D=0D=0D=0, there is no trace-one matrix, so the universal condition is vacuous; for K=0K=0K=0, the middle matrix is zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.SymmetricGaussian

For an arbitrary measurable space Ω\OmegaΩ, a measure μ\muμ on it, a natural number DDD, a nonnegative real number v=s2v=s2v=s2, and a function G:Ω→CDG:\Omega\to\mathbb C^DG:Ω→CD with its Euclidean structure, this predicate means that GGG is measurable and its 2D2D2D real coordinate functions

cj,0(ω)=Re⁡G(ω)j,cj,1(ω)=Im⁡G(ω)jc_{j,0}(\omega)=\operatorname{Re}G(\omega)_j, \qquad c_{j,1}(\omega)=\operatorname{Im}G(\omega)_jcj,0​(ω)=ReG(ω)j​,cj,1​(ω)=ImG(ω)j​

are jointly independent with respect to μ\muμ, with each pushforward measure (cj,e)∗μ(c_{j,e})_*\mu(cj,e​)∗​μ equal to the real Gaussian measure of mean 000 and variance v/(2D)v/(2D)v/(2D). The definition does not assume that μ\muμ is a probability measure and permits v=0v=0v=0, giving zero-variance Gaussian coordinate laws. It also permits D=0D=0D=0, when there are no coordinate functions and the independence and coordinate-law conditions are vacuous.

QAlgorithms.Papers.Independent.Intakenext100.P54.VisibleCode

For a natural number DDD, a value of this structure consists of positive natural numbers S,KS,KS,K, a function assigning to every unit vector ψ∈CD\psi\in\mathbb C^Dψ∈CD a unit vector q(ψ)∈CSq(\psi)\in\mathbb C^Sq(ψ)∈CS, a function assigning to every such ψ\psiψ a message j(ψ)∈{0,…,K−1}j(\psi)\in\{0,\ldots,K-1\}j(ψ)∈{0,…,K−1}, and a family of maps Dj\mathcal D_jDj​ from complex S×SS\times SS×S matrices to complex D×DD\times DD×D matrices. Each Dj\mathcal D_jDj​ must admit a finite Kraus representation

Dj(X)=∑h=0rj−1Bj,hXBj,h†for every X,∑h=0rj−1Bj,h†Bj,h=IS.\mathcal D_j(X)=\sum_{h=0}^{r_j-1}B_{j,h}XB_{j,h}^\dagger \quad\text{for every }X, \qquad \sum_{h=0}^{r_j-1}B_{j,h}^\dagger B_{j,h}=I_S.Dj​(X)=h=0∑rj​−1​Bj,h​XBj,h†​for every X,h=0∑rj​−1​Bj,h†​Bj,h​=IS​.

No measurability, continuity, or accuracy condition is imposed on the two functions of ψ\psiψ. The parameter DDD is not explicitly required to be positive; when D=0D=0D=0, its unit-vector input set is empty, but the required channels from positive dimension SSS to dimension zero cannot exist.

QAlgorithms.Papers.Independent.Intakenext100.P54.VisibleCode.output

For a natural number DDD, let PPP consist of positive integers S,KS,KS,K, functions qqq and jjj assigning each unit vector of CD\mathbb C^DCD respectively a unit vector of CS\mathbb C^SCS and a message in {0,…,K−1}\{0,\ldots,K-1\}{0,…,K−1}, and maps Dj\mathcal D_jDj​ admitting finite Kraus representations Dj(X)=∑hBj,hXBj,h†\mathcal D_j(X)=\sum_hB_{j,h}XB_{j,h}^\daggerDj​(X)=∑h​Bj,h​XBj,h†​ with ∑hBj,h†Bj,h=IS\sum_hB_{j,h}^\dagger B_{j,h}=I_S∑h​Bj,h†​Bj,h​=IS​. For every unit input vector ψ∈CD\psi\in\mathbb C^Dψ∈CD, this definition returns the complex D×DD\times DD×D matrix

Dj(ψ)(∣q(ψ)⟩⟨q(ψ)∣),\mathcal D_{j(\psi)}\bigl(|q(\psi)\rangle\langle q(\psi)|\bigr),Dj(ψ)​(∣q(ψ)⟩⟨q(ψ)∣),

where the input matrix to the selected map has entries q(ψ)s q(ψ)t‾q(\psi)_s\,\overline{q(\psi)_t}q(ψ)s​q(ψ)t​​.

QAlgorithms.Papers.Independent.Intakenext100.P54.auxWeight

For any natural numbers m,km,km,k, any real-valued function p:{0,…,m−1}→Rp:\{0,\ldots,m-1\}\to\mathbb Rp:{0,…,m−1}→R, any array ci,j≥0c_{i,j}\ge0ci,j​≥0 satisfying ∑j=0k−1ci,j=1\sum_{j=0}^{k-1}c_{i,j}=1∑j=0k−1​ci,j​=1 for every iii, and any index 0≤j<k0\le j<k0≤j<k, the defined real number is

qj=∑i=0m−1pi ci,j.q_j=\sum_{i=0}^{m-1}p_i\,c_{i,j}.qj​=i=0∑m−1​pi​ci,j​.

No nonnegativity or normalization requirement is imposed on ppp. If m=0m=0m=0, the sum equals 000; if k=0k=0k=0, there is no index jjj to which this definition can be applied.

QAlgorithms.Papers.Independent.Intakenext100.P54.auxiliaryBobEntropy

For arbitrary natural numbers m,k,dm,k,dm,k,d, a real vector p=(pi)i<mp=(p_i)_{i<m}p=(pi​)i<m​, complex d×dd\times dd×d matrices ρi\rho_iρi​, and a real array cij≥0c_{ij}\ge0cij​≥0 with ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1 for every i<mi<mi<m, set qj=∑i<mpicijq_j=\sum_{i<m}p_i c_{ij}qj​=∑i<m​pi​cij​, and let σj=d−1Id\sigma_j=d^{-1}I_dσj​=d−1Id​ if qj=0q_j=0qj​=0, and σj=∑i<m(picij/qj)ρi\sigma_j=\sum_{i<m}(p_i c_{ij}/q_j)\rho_iσj​=∑i<m​(pi​cij​/qj​)ρi​ otherwise. Writing

H(X)=−Re⁡tr⁡ ⁣[f(X)],f(x)=xlog⁡2x,H(X)=-\operatorname{Re}\operatorname{tr}\!\left[f(X)\right], \qquad f(x)=x\log_2 x,H(X)=−Retr[f(X)],f(x)=xlog2​x,

where f(X)f(X)f(X) is evaluated by continuous functional calculus, the definition returns

∑j<kqjH(σj).\sum_{j<k}q_jH(\sigma_j).j<k∑​qj​H(σj​).

Neither normalization or nonnegativity of ppp, nor positivity or trace normalization of ρi\rho_iρi​, is required. A zero qjq_jqj​ contributes zero. Empty sums are zero, d=0d=0d=0 is permitted, and all inverses use the convention 0−1=00^{-1}=00−1=0; if m>0m>0m>0, the assumptions on ccc exclude k=0k=0k=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.bar

For any type OOO and any complex matrix MMM with rows and columns indexed by OOO, this defines its entrywise complex conjugate:

M‾ab=Mab‾(a,b∈O).\overline M_{ab}=\overline{M_{ab}}\qquad(a,b\in O).Mab​=Mab​​(a,b∈O).

There is no finiteness assumption on OOO.

QAlgorithms.Papers.Independent.Intakenext100.P54.bipAverage

Let m,a,b,nm,a,b,nm,a,b,n be natural numbers. An ensemble EEE consists of nonnegative weights pip_ipi​ summing to 111 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^{a}\otimes\mathbb C^{b}ψi​∈Ca⊗Cb, indexed by i<mi<mi<m. For a word w∈{0,…,m−1}nw\in\{0,\ldots,m-1\}^nw∈{0,…,m−1}n, put pw=∏t<npwtp_w=\prod_{t<n}p_{w_t}pw​=∏t<n​pwt​​ and Ψw(x,y)=∏t<nψwt(xt,yt)\Psi_w(x,y)=\prod_{t<n}\psi_{w_t}(x_t,y_t)Ψw​(x,y)=∏t<n​ψwt​​(xt​,yt​). The code PPP consists of positive integers S,KS,KS,K, a natural number LLL, matrices Aw,j,ℓ:CS→CanA_{w,j,\ell}:\mathbb C^S\to\mathbb C^{a^n}Aw,j,ℓ​:CS→Can satisfying ∑j<K,ℓ<LAw,j,ℓ†Aw,j,ℓ=IS\sum_{j<K,\ell<L}A_{w,j,\ell}^{\dagger}A_{w,j,\ell}=I_S∑j<K,ℓ<L​Aw,j,ℓ†​Aw,j,ℓ​=IS​ for every www, and maps Bj\mathcal B_jBj​ from S×SS\times SS×S matrices to matrices on Cbn\mathbb C^{b^n}Cbn, each admitting a finite Kraus representation Bj(X)=∑hBj,hXBj,h†\mathcal B_j(X)=\sum_h B_{j,h}XB_{j,h}^{\dagger}Bj​(X)=∑h​Bj,h​XBj,h†​ with ∑hBj,h†Bj,h=IS\sum_hB_{j,h}^{\dagger}B_{j,h}=I_S∑h​Bj,h†​Bj,h​=IS​. Its output Ωw\Omega_wΩw​ has entries

Ωw((x,y),(x′,y′))=1S∑j<K∑ℓ<L∑u,v<SAw,j,ℓ(x,u)Aw,j,ℓ(x′,v)‾ Bj(∣u⟩⟨v∣)(y,y′).\Omega_w((x,y),(x',y'))= \frac1S\sum_{j<K}\sum_{\ell<L}\sum_{u,v<S} A_{w,j,\ell}(x,u)\overline{A_{w,j,\ell}(x',v)} \,\mathcal B_j(|u\rangle\langle v|)(y,y').Ωw​((x,y),(x′,y′))=S1​j<K∑​ℓ<L∑​u,v<S∑​Aw,j,ℓ​(x,u)Aw,j,ℓ​(x′,v)​Bj​(∣u⟩⟨v∣)(y,y′).

The definition is the real number

bipAverage⁡(E,n,P)=∑wpw Re⁡Tr⁡(∣Ψw⟩⟨Ψw∣Ωw).\operatorname{bipAverage}(E,n,P) =\sum_w p_w\,\operatorname{Re}\operatorname{Tr} \bigl(|\Psi_w\rangle\langle\Psi_w|\Omega_w\bigr).bipAverage(E,n,P)=w∑​pw​ReTr(∣Ψw​⟩⟨Ψw​∣Ωw​).

Here ∣v⟩⟨v∣|v\rangle\langle v|∣v⟩⟨v∣ has entries v(x)v(y)‾v(x)\overline{v(y)}v(x)v(y)​. The definition includes n=0n=0n=0: there is one empty word, its weight and vector amplitude are both 111, and the output registers are one dimensional. Although m,a,bm,a,bm,a,b are not explicitly required to be positive, the ensemble’s weight and normalization requirements exclude their being zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.bipOperational

For natural numbers m,a,bm,a,bm,a,b, an ensemble EEE consists of probabilities pi≥0p_i\ge0pi​≥0 summing to 111 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^a\otimes\mathbb C^bψi​∈Ca⊗Cb, indexed by i<mi<mi<m. For any real RRR, this definition takes the infimum in [0,∞][0,\infty][0,∞] of max⁡(e,0)\max(e,0)max(e,0) over real numbers eee satisfying the following condition: R,e≥0R,e\ge0R,e≥0, and for every ε,δ>0\varepsilon,\delta>0ε,δ>0, all sufficiently large positive integers nnn admit positive integers S,KS,KS,K, a natural number LLL, matrices Aw,j,ℓ:CS→CanA_{w,j,\ell}:\mathbb C^S\to\mathbb C^{a^n}Aw,j,ℓ​:CS→Can with ∑j,ℓAw,j,ℓ†Aw,j,ℓ=IS\sum_{j,\ell}A_{w,j,\ell}^{\dagger}A_{w,j,\ell}=I_S∑j,ℓ​Aw,j,ℓ†​Aw,j,ℓ​=IS​ for every word www, and maps Bj(X)=∑hBj,hXBj,h†\mathcal B_j(X)=\sum_hB_{j,h}XB_{j,h}^{\dagger}Bj​(X)=∑h​Bj,h​XBj,h†​ with finite Kraus families satisfying ∑hBj,h†Bj,h=IS\sum_hB_{j,h}^{\dagger}B_{j,h}=I_S∑h​Bj,h†​Bj,h​=IS​, for which

∑w(∏t<npwt)Re⁡Tr⁡(∣Ψw⟩⟨Ψw∣Ωw)≥1−ε,log⁡2Kn≤R+δ,log⁡2Sn≤e+δ.\sum_w\left(\prod_{t<n}p_{w_t}\right) \operatorname{Re}\operatorname{Tr} \bigl(|\Psi_w\rangle\langle\Psi_w|\Omega_w\bigr)\ge1-\varepsilon, \qquad \frac{\log_2K}{n}\le R+\delta, \qquad \frac{\log_2S}{n}\le e+\delta.w∑​(t<n∏​pwt​​)ReTr(∣Ψw​⟩⟨Ψw​∣Ωw​)≥1−ε,nlog2​K​≤R+δ,nlog2​S​≤e+δ.

Here Ψw(x,y)=∏t<nψwt(xt,yt)\Psi_w(x,y)=\prod_{t<n}\psi_{w_t}(x_t,y_t)Ψw​(x,y)=∏t<n​ψwt​​(xt​,yt​) and

Ωw((x,y),(x′,y′))=1S∑j,ℓ,u,vAw,j,ℓ(x,u)Aw,j,ℓ(x′,v)‾ Bj(∣u⟩⟨v∣)(y,y′).\Omega_w((x,y),(x',y'))= \frac1S\sum_{j,\ell,u,v} A_{w,j,\ell}(x,u)\overline{A_{w,j,\ell}(x',v)} \,\mathcal B_j(|u\rangle\langle v|)(y,y').Ωw​((x,y),(x′,y′))=S1​j,ℓ,u,v∑​Aw,j,ℓ​(x,u)Aw,j,ℓ​(x′,v)​Bj​(∣u⟩⟨v∣)(y,y′).

Because eligible eee are nonnegative, their conversion to [0,∞][0,\infty][0,∞] simply preserves their value. An empty feasible set has infimum +∞+\infty+∞; in particular, the value is +∞+\infty+∞ when R<0R<0R<0. The infimum’s definition does not itself assert attainment.

QAlgorithms.Papers.Independent.Intakenext100.P54.bipWord

For arbitrary natural numbers m,a,bm,a,bm,a,b, an ensemble EEE consisting of real probabilities pi≥0p_i\ge0pi​≥0 with ∑i<mpi=1\sum_{i<m}p_i=1∑i<m​pi​=1 and unit vectors ψi∈Ca×b\psi_i\in\mathbb C^{a\times b}ψi​∈Ca×b, any natural number nnn, and any word w:{0,…,n−1}→{0,…,m−1}w:\{0,\ldots,n-1\}\to\{0,\ldots,m-1\}w:{0,…,n−1}→{0,…,m−1}, the definition returns the vector on pairs of words x∈{0,…,a−1}nx\in\{0,\ldots,a-1\}^nx∈{0,…,a−1}n, y∈{0,…,b−1}ny\in\{0,\ldots,b-1\}^ny∈{0,…,b−1}n with amplitudes

Ψw(x,y)=∏t<nψwt(xt,yt).\Psi_w(x,y)=\prod_{t<n}\psi_{w_t}(x_t,y_t).Ψw​(x,y)=t<n∏​ψwt​​(xt​,yt​).

The probabilities do not appear in these amplitudes, and www may use indices of zero probability. When n=0n=0n=0, both word registers have one empty word and the empty product gives amplitude 111. The existence of EEE forces m,a,b>0m,a,b>0m,a,b>0.

QAlgorithms.Papers.Independent.Intakenext100.P54.bobAverageEntropy

For natural numbers m,a,bm,a,bm,a,b, let EEE consist of probabilities pi≥0p_i\ge0pi​≥0 summing to 111 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^a\otimes\mathbb C^bψi​∈Ca⊗Cb, indexed by i<mi<mi<m. Define the b×bb\times bb×b matrices

σi(y,y′)=∑x<aψi(x,y)ψi(x,y′)‾.\sigma_i(y,y')=\sum_{x<a}\psi_i(x,y)\overline{\psi_i(x,y')}.σi​(y,y′)=x<a∑​ψi​(x,y)ψi​(x,y′)​.

Then bobAverageEntropy⁡(E)=H(∑i<mpiσi)\operatorname{bobAverageEntropy}(E)=H(\sum_{i<m}p_i\sigma_i)bobAverageEntropy(E)=H(∑i<m​pi​σi​), where

H(M)=−Re⁡Tr⁡(f(M)),f(t)=tlog⁡2t,H(M)=-\operatorname{Re}\operatorname{Tr}\bigl(f(M)\bigr), \qquad f(t)=t\log_2t,H(M)=−ReTr(f(M)),f(t)=tlog2​t,

and f(M)f(M)f(M) is evaluated by continuous functional calculus. In particular, the scalar expression at t=0t=0t=0 is 000. The ensemble requirements exclude m=0m=0m=0, a=0a=0a=0, or b=0b=0b=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.bobHolevo

For natural numbers m,a,bm,a,bm,a,b, let EEE consist of probabilities pi≥0p_i\ge0pi​≥0 summing to 111 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^a\otimes\mathbb C^bψi​∈Ca⊗Cb, indexed by i<mi<mi<m. With

σi(y,y′)=∑x<aψi(x,y)ψi(x,y′)‾,H(M)=−Re⁡Tr⁡((t↦tlog⁡2t)(M)),\sigma_i(y,y')=\sum_{x<a}\psi_i(x,y)\overline{\psi_i(x,y')}, \qquad H(M)=-\operatorname{Re}\operatorname{Tr}\bigl((t\mapsto t\log_2t)(M)\bigr),σi​(y,y′)=x<a∑​ψi​(x,y)ψi​(x,y′)​,H(M)=−ReTr((t↦tlog2​t)(M)),

where the matrix function uses continuous functional calculus and 0log⁡20=00\log_2 0=00log2​0=0, the definition is

bobHolevo⁡(E)=H ⁣(∑i<mpiσi)−∑i<mpiH(σi).\operatorname{bobHolevo}(E) =H\!\left(\sum_{i<m}p_i\sigma_i\right) -\sum_{i<m}p_iH(\sigma_i).bobHolevo(E)=H(i<m∑​pi​σi​)−i<m∑​pi​H(σi​).

Zero probabilities are allowed. Zero label or local dimensions cannot satisfy the ensemble requirements.

QAlgorithms.Papers.Independent.Intakenext100.P54.bobStates

For arbitrary natural numbers m,a,bm,a,bm,a,b, an ensemble EEE consisting of pi≥0p_i\ge0pi​≥0 with ∑i<mpi=1\sum_{i<m}p_i=1∑i<m​pi​=1 and unit vectors ψi∈Ca×b\psi_i\in\mathbb C^{a\times b}ψi​∈Ca×b, and an index i<mi<mi<m, the definition returns the complex b×bb\times bb×b matrix

(ρi)uv=∑α<aψi(α,u) ψi(α,v)‾(u,v<b).(\rho_i)_{uv}=\sum_{\alpha<a}\psi_i(\alpha,u)\, \overline{\psi_i(\alpha,v)} \qquad(u,v<b).(ρi​)uv​=α<a∑​ψi​(α,u)ψi​(α,v)​(u,v<b).

Thus it takes the matrix ∣ψi⟩⟨ψi∣|\psi_i\rangle\langle\psi_i|∣ψi​⟩⟨ψi​∣ and sums its entries over equal first-register indices. The probabilities pip_ipi​ do not enter this formula, and the selected index need not have positive probability. The existence of EEE forces m,a,b>0m,a,b>0m,a,b>0.

QAlgorithms.Papers.Independent.Intakenext100.P54.conditionalBob

For arbitrary natural numbers m,k,dm,k,dm,k,d, arbitrary real numbers pip_ipi​, arbitrary complex d×dd\times dd×d matrices ρi\rho_iρi​, and a real array cijc_{ij}cij​ satisfying cij≥0c_{ij}\ge0cij​≥0 and ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1 for every i<mi<mi<m, define qj=∑i<mpicijq_j=\sum_{i<m}p_i c_{ij}qj​=∑i<m​pi​cij​. For each j<kj<kj<k, the definition returns

σj={d−1Id,qj=0,∑i<mpicijqjρi,qj≠0.\sigma_j= \begin{cases} d^{-1}I_d,&q_j=0,\\[2mm] \displaystyle\sum_{i<m}\frac{p_i c_{ij}}{q_j}\rho_i,&q_j\ne0. \end{cases}σj​=⎩⎨⎧​d−1Id​,i<m∑​qj​pi​cij​​ρi​,​qj​=0,qj​=0.​

Real coefficients are regarded as complex scalars. There is no assumption that ppp is a probability distribution or that the ρi\rho_iρi​ are density matrices; in particular, a nonzero qjq_jqj​ may be negative. Zero dimensions are allowed, with 0−1=00^{-1}=00−1=0; when m>0m>0m>0, the row normalization of ccc makes k=0k=0k=0 impossible. When m=0m=0m=0, every qjq_jqj​ is zero and the first branch applies.

QAlgorithms.Papers.Independent.Intakenext100.P54.elementaryRate

For every real number ξ\xiξ, this definition returns the extended real value

{+∞,ξ=−1,ξ−ln⁡(1+ξ)2ln⁡2,ξ≠−1.\begin{cases} +\infty,&\xi=-1,\\[2pt] \displaystyle\frac{\xi-\ln(1+\xi)}{2\ln2},&\xi\ne-1. \end{cases}⎩⎨⎧​+∞,2ln2ξ−ln(1+ξ)​,​ξ=−1,ξ=−1.​

There is no domain restriction on ξ\xiξ: for ξ<−1\xi<-1ξ<−1, the logarithm is the total real logarithm, which equals ln⁡∣1+ξ∣\ln|1+\xi|ln∣1+ξ∣ at the negative argument 1+ξ1+\xi1+ξ.

QAlgorithms.Papers.Independent.Intakenext100.P54.labelAuxInfo

For arbitrary natural numbers m,km,km,k, arbitrary pi∈Rp_i\in\mathbb Rpi​∈R, and a real array cij≥0c_{ij}\ge0cij​≥0 whose rows satisfy ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1, put qj=∑i<mpicijq_j=\sum_{i<m}p_i c_{ij}qj​=∑i<m​pi​cij​. Define H(X)=−Re⁡tr⁡[f(X)]H(X)=-\operatorname{Re}\operatorname{tr}[f(X)]H(X)=−Retr[f(X)], using continuous functional calculus with f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x, and define Hdiag(v)=H(diag⁡(v))H_{\mathrm{diag}}(v)=H(\operatorname{diag}(v))Hdiag​(v)=H(diag(v)) for a real vector vvv. The definition returns

Hdiag(p)−∑j<kqj Hdiag ⁣((picijqj)i<m).H_{\mathrm{diag}}(p) -\sum_{j<k}q_j\,H_{\mathrm{diag}}\!\left( \left(\frac{p_i c_{ij}}{q_j}\right)_{i<m}\right).Hdiag​(p)−j<k∑​qj​Hdiag​((qj​pi​cij​​)i<m​).

There is no probability-distribution assumption on ppp. Division by zero is zero, so when qj=0q_j=0qj​=0, the vector inside that term is the zero vector and the term contributes zero. Empty index sets are allowed; row normalization excludes k=0k=0k=0 when m>0m>0m>0.

QAlgorithms.Papers.Independent.Intakenext100.P54.labelBobAuxInfo

For arbitrary natural numbers m,k,dm,k,dm,k,d, real pip_ipi​, complex d×dd\times dd×d matrices ρi\rho_iρi​, and cij≥0c_{ij}\ge0cij​≥0 satisfying ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1 for every i<mi<mi<m, set qj=∑i<mpicijq_j=\sum_{i<m}p_i c_{ij}qj​=∑i<m​pi​cij​ and vi(j)=picij/qjv^{(j)}_i=p_i c_{ij}/q_jvi(j)​=pi​cij​/qj​. Let σj=d−1Id\sigma_j=d^{-1}I_dσj​=d−1Id​ when qj=0q_j=0qj​=0, and σj=∑i<mvi(j)ρi\sigma_j=\sum_{i<m}v^{(j)}_i\rho_iσj​=∑i<m​vi(j)​ρi​ otherwise. With H(X)=−Re⁡tr⁡[f(X)]H(X)=-\operatorname{Re}\operatorname{tr}[f(X)]H(X)=−Retr[f(X)] defined through continuous functional calculus for f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x, and Hdiag(v)=H(diag⁡(v))H_{\mathrm{diag}}(v)=H(\operatorname{diag}(v))Hdiag​(v)=H(diag(v)), the definition returns

Hdiag(p)−∑j<kqjHdiag(v(j))+∑j<kqjH(σj)−∑i<mpiH(ρi).H_{\mathrm{diag}}(p) -\sum_{j<k}q_jH_{\mathrm{diag}}(v^{(j)}) +\sum_{j<k}q_jH(\sigma_j) -\sum_{i<m}p_iH(\rho_i).Hdiag​(p)−j<k∑​qj​Hdiag​(v(j))+j<k∑​qj​H(σj​)−i<m∑​pi​H(ρi​).

No distribution assumption on ppp or density-matrix assumption on ρi\rho_iρi​ is imposed. Zero dimensions and empty sums are included; division by zero is zero, and terms multiplied by qj=0q_j=0qj​=0 vanish. The row conditions on ccc cannot hold with m>0m>0m>0 and k=0k=0k=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.marginalAverageFidelity

For natural numbers m,a,bm,a,bm,a,b, let EEE consist of probabilities pi≥0p_i\ge 0pi​≥0 with ∑i=0m−1pi=1\sum_{i=0}^{m-1}p_i=1∑i=0m−1​pi​=1 and unit vectors ψi∈Ca⊗Cb\psi_i\in\mathbb C^a\otimes\mathbb C^bψi​∈Ca⊗Cb. For any natural number nnn, write pw=∏t=0n−1pwtp_w=\prod_{t=0}^{n-1}p_{w_t}pw​=∏t=0n−1​pwt​​ and Ψw=⨂t=0n−1ψwt\Psi_w=\bigotimes_{t=0}^{n-1}\psi_{w_t}Ψw​=⨂t=0n−1​ψwt​​ for each word w∈{0,…,m−1}nw\in\{0,\ldots,m-1\}^nw∈{0,…,m−1}n. Let PPP be a code consisting of positive integers S,KS,KS,K, a natural number LLL, matrices Aw,j,ℓ:CS→(Ca)⊗nA_{w,j,\ell}:\mathbb C^S\to(\mathbb C^a)^{\otimes n}Aw,j,ℓ​:CS→(Ca)⊗n satisfying ∑j,ℓAw,j,ℓ†Aw,j,ℓ=IS\sum_{j,\ell}A_{w,j,\ell}^\dagger A_{w,j,\ell}=I_S∑j,ℓ​Aw,j,ℓ†​Aw,j,ℓ​=IS​ for every www, and channels Dj\mathcal D_jDj​ from SSS-dimensional matrices to matrices on (Cb)⊗n(\mathbb C^b)^{\otimes n}(Cb)⊗n. Here a channel means a map admitting a finite representation X↦∑hBhXBh†X\mapsto\sum_h B_hXB_h^\daggerX↦∑h​Bh​XBh†​ with ∑hBh†Bh=I\sum_hB_h^\dagger B_h=I∑h​Bh†​Bh​=I. Define its output Ωw\Omega_wΩw​ by

(Ωw)(α,β),(α′,β′)=1S∑j=0K−1∑ℓ=0L−1∑s,t=0S−1(Aw,j,ℓ)α,s (Aw,j,ℓ)α′,t‾ [Dj(∣s⟩⟨t∣)]β,β′.(\Omega_w)_{(\alpha,\beta),(\alpha',\beta')} =\frac1S\sum_{j=0}^{K-1}\sum_{\ell=0}^{L-1}\sum_{s,t=0}^{S-1} (A_{w,j,\ell})_{\alpha,s}\, \overline{(A_{w,j,\ell})_{\alpha',t}}\, \bigl[\mathcal D_j(|s\rangle\langle t|)\bigr]_{\beta,\beta'}.(Ωw​)(α,β),(α′,β′)​=S1​j=0∑K−1​ℓ=0∑L−1​s,t=0∑S−1​(Aw,j,ℓ​)α,s​(Aw,j,ℓ​)α′,t​​[Dj​(∣s⟩⟨t∣)]β,β′​.

The declared quantity is

∑wpw F ⁣(Tr⁡A∣Ψw⟩⟨Ψw∣,Tr⁡AΩw),F(X,Y)=(Re⁡tr⁡X Y X)2,\sum_w p_w\,F\!\left(\operatorname{Tr}_A|\Psi_w\rangle\langle\Psi_w|, \operatorname{Tr}_A\Omega_w\right), \qquad F(X,Y)=\left(\operatorname{Re}\operatorname{tr} \sqrt{\sqrt X\,Y\,\sqrt X}\right)^2,w∑​pw​F(TrA​∣Ψw​⟩⟨Ψw​∣,TrA​Ωw​),F(X,Y)=(RetrX​YX​​)2,

where Tr⁡A\operatorname{Tr}_ATrA​ sums over the first tensor factor and the matrix square roots use the continuous functional calculus. The definition includes n=0n=0n=0, when there is one empty word, its weight is 111, and its tensor vector has its sole coordinate equal to 111. Although m,a,bm,a,bm,a,b are not explicitly required to be positive, the conditions on EEE cannot be satisfied if any of them is zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.momentLog

For any extended nonnegative real number zzz, this definition returns the extended real value

momentLog⁡(z)={+∞,z=+∞,−∞,z=0,ln⁡z,0<z<+∞,\operatorname{momentLog}(z)= \begin{cases} +\infty,&z=+\infty,\\ -\infty,&z=0,\\ \ln z,&0<z<+\infty, \end{cases}momentLog(z)=⎩⎨⎧​+∞,−∞,lnz,​z=+∞,z=0,0<z<+∞,​

where the last case uses the ordinary finite real value of zzz and the natural logarithm.

QAlgorithms.Papers.Independent.Intakenext100.P54.piApproxOutput

For natural numbers D,KD,KD,K, any family of unitary D×DD\times DD×D matrices (Uk)k<K(U_k)_{k<K}(Uk​)k<K​, any real ε\varepsilonε, and any unit vector ψ∈CD\psi\in\mathbb C^Dψ∈CD, put

Gk=DK(1+ε)Ukψψ†‾Uk†,F=ID−∑k<KGk.G_k=\frac{D}{K(1+\varepsilon)} U_k\overline{\psi\psi^\dagger}U_k^\dagger, \qquad F=I_D-\sum_{k<K}G_k.Gk​=K(1+ε)D​Uk​ψψ†​Uk†​,F=ID​−k<K∑​Gk​.

The definition is the D×DD\times DD×D matrix

piApproxOutput⁡=∑k<KUkT(D−1GkT+(KD)−1FT)(UkT)†.\operatorname{piApproxOutput} =\sum_{k<K}U_k^T \left(D^{-1}G_k^T+(KD)^{-1}F^T\right) (U_k^T)^\dagger.piApproxOutput=k<K∑​UkT​(D−1GkT​+(KD)−1FT)(UkT​)†.

Overlines mean entrywise complex conjugation, TTT means ordinary transpose, and †\dagger† means conjugate transpose. No restriction on ε\varepsilonε or randomising assumption is imposed, and no positivity or normalization is asserted here. Inverses of zero are zero, so the expression remains defined at ε=−1\varepsilon=-1ε=−1 and K=0K=0K=0; at K=0K=0K=0 the sum is the zero matrix. A unit vector ψ\psiψ cannot exist when D=0D=0D=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.piBranch

For any natural numbers D,KD,KD,K, any family of unitary complex D×DD\times DD×D matrices UkU_kUk​, any real ε\varepsilonε, and any unit vector ψ∈CD\psi\in\mathbb C^Dψ∈CD, put

Gk=DK(1+ε)Uk ψψ†‾ Uk†.G_k=\frac{D}{K(1+\varepsilon)} U_k\,\overline{\psi\psi^\dagger}\,U_k^\dagger.Gk​=K(1+ε)D​Uk​ψψ†​Uk†​.

This defines matrices indexed by the labels k=0,…,K−1k=0,\ldots,K-1k=0,…,K−1 and none\mathrm{none}none as

Bk=UkT(D−1GkT)(UkT)†,Bnone=D−1(ID−∑k=0K−1Gk)T.B_k=U_k^{\mathsf T}\bigl(D^{-1}G_k^{\mathsf T}\bigr) (U_k^{\mathsf T})^\dagger, \qquad B_{\mathrm{none}}= D^{-1}\left(I_D-\sum_{k=0}^{K-1}G_k\right)^{\mathsf T}.Bk​=UkT​(D−1GkT​)(UkT​)†,Bnone​=D−1(ID​−k=0∑K−1​Gk​)T.

Here T\mathsf TT is transpose, †\dagger† is conjugate transpose, the overline is entrywise conjugation, and (ψψ†)ab=ψaψb‾(\psi\psi^\dagger)_{ab}=\psi_a\overline{\psi_b}(ψψ†)ab​=ψa​ψb​​. No randomising or positivity assumption is imposed. All scalar inverses are total, with 0−1=00^{-1}=00−1=0; if K=0K=0K=0, there are no indexed branches and Bnone=D−1IDB_{\mathrm{none}}=D^{-1}I_DBnone​=D−1ID​. The unit-vector argument excludes D=0D=0D=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.piEffect

For any natural numbers D,KD,KD,K, any family of unitary complex D×DD\times DD×D matrices UkU_kUk​, any real ε\varepsilonε, and any unit vector ψ∈CD\psi\in\mathbb C^Dψ∈CD, this defines a matrix for each label in {0,…,K−1}∪{none}\{0,\ldots,K-1\}\cup\{\mathrm{none}\}{0,…,K−1}∪{none} by

Ek=DK(1+ε)Uk ψψ†‾ Uk†,Enone=ID−∑k=0K−1Ek.E_k=\frac{D}{K(1+\varepsilon)} U_k\,\overline{\psi\psi^\dagger}\,U_k^\dagger, \qquad E_{\mathrm{none}}=I_D-\sum_{k=0}^{K-1}E_k.Ek​=K(1+ε)D​Uk​ψψ†​Uk†​,Enone​=ID​−k=0∑K−1​Ek​.

Here †\dagger† denotes conjugate transpose, the overline denotes entrywise conjugation, and (ψψ†)ab=ψaψb‾(\psi\psi^\dagger)_{ab}=\psi_a\overline{\psi_b}(ψψ†)ab​=ψa​ψb​​. This definition assumes neither positivity of these matrices nor a randomising property. Division by zero is zero; if K=0K=0K=0, the only label is none\mathrm{none}none and its matrix is IDI_DID​. A unit vector is unavailable when D=0D=0D=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.piGood

For any natural numbers D,KD,KD,K, any unitary complex D×DD\times DD×D matrices U0,…,UK−1U_0,\ldots,U_{K-1}U0​,…,UK−1​, any real ε\varepsilonε, any unit vector ψ∈CD\psi\in\mathbb C^Dψ∈CD, and any index 0≤k<K0\le k<K0≤k<K, this defines

Gk=DK(1+ε)Uk ψψ†‾ Uk†,G_k= \frac{D}{K(1+\varepsilon)} U_k\,\overline{\psi\psi^\dagger}\,U_k^\dagger,Gk​=K(1+ε)D​Uk​ψψ†​Uk†​,

where the scalar division takes place in C\mathbb CC, (ψψ†)ab=ψaψb‾(\psi\psi^\dagger)_{ab}=\psi_a\overline{\psi_b}(ψψ†)ab​=ψa​ψb​​, and the overline means entrywise complex conjugation. No restriction on ε\varepsilonε or randomising property of the family is assumed. If ε=−1\varepsilon=-1ε=−1, the scalar is zero by the convention for division by zero. The required index cannot exist when K=0K=0K=0, and the required unit vector cannot exist when D=0D=0D=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.piRecord

For natural numbers D,KD,KD,K, any family of unitary D×DD\times DD×D matrices (Uk)k<K(U_k)_{k<K}(Uk​)k<K​, any real ε\varepsilonε, and any vector ψ∈CD\psi\in\mathbb C^Dψ∈CD with ∥ψ∥=1\|\psi\|=1∥ψ∥=1, write Pψ=ψψ†P_\psi=\psi\psi^\daggerPψ​=ψψ†, let an overline mean entrywise complex conjugation, and define

Gk=DK(1+ε) UkPψ‾Uk†,F=ID−∑k<KGk.G_k=\frac{D}{K(1+\varepsilon)}\,U_k\overline{P_\psi}U_k^\dagger, \qquad F=I_D-\sum_{k<K}G_k.Gk​=K(1+ε)D​Uk​Pψ​​Uk†​,F=ID​−k<K∑​Gk​.

The definition is the matrix on {0,…,K−1}×{0,…,D−1}\{0,\ldots,K-1\}\times\{0,\ldots,D-1\}{0,…,K−1}×{0,…,D−1} given by

piRecord⁡((k,u),(ℓ,v))={(D−1GkT+(KD)−1FT)uv,k=ℓ,0,k≠ℓ.\operatorname{piRecord}((k,u),(\ell,v))= \begin{cases} \left(D^{-1}G_k^{T}+(KD)^{-1}F^{T}\right)_{uv},&k=\ell,\\ 0,&k\ne\ell. \end{cases}piRecord((k,u),(ℓ,v))={(D−1GkT​+(KD)−1FT)uv​,0,​k=ℓ,k=ℓ.​

The transpose is ordinary transpose. No bound on ε\varepsilonε, randomising condition, or positivity of FFF is assumed. All divisions are total, with inverse zero equal to zero, including when ε=−1\varepsilon=-1ε=−1 or K=0K=0K=0; for K=0K=0K=0 the resulting matrix has an empty index set. When D=0D=0D=0, no unit vector ψ\psiψ exists.

QAlgorithms.Papers.Independent.Intakenext100.P54.piSimulator

For natural numbers D,KD,KD,K and any family of unitary D×DD\times DD×D matrices (Uk)k<K(U_k)_{k<K}(Uk​)k<K​, this definition gives a function from arbitrary complex D×DD\times DD×D matrices ρ\rhoρ to matrices indexed by (k,u)(k,u)(k,u) with k<Kk<Kk<K and u<Du<Du<D:

piSimulator⁡(ρ)((k,u),(ℓ,v))={K−1(Uk‾ρUkT)uv,k=ℓ,0,k≠ℓ.\operatorname{piSimulator}(\rho)((k,u),(\ell,v))= \begin{cases} K^{-1}\bigl(\overline{U_k}\rho U_k^T\bigr)_{uv},&k=\ell,\\ 0,&k\ne\ell. \end{cases}piSimulator(ρ)((k,u),(ℓ,v))={K−1(Uk​​ρUkT​)uv​,0,​k=ℓ,k=ℓ.​

Here Uk‾\overline{U_k}Uk​​ is entrywise complex conjugation and UkTU_k^TUkT​ is ordinary transpose. The input need not be a density matrix. Neither DDD nor KKK is required to be positive; when K=0K=0K=0 the output index set is empty, and the scalar inverse convention is 0−1=00^{-1}=00−1=0. This definition alone makes no channel assertion.

QAlgorithms.Papers.Independent.Intakenext100.P54.pureAverageOverlap

For any natural numbers d,nd,nd,n, let EEE consist of LLL unit vectors vℓ∈Cdv_\ell\in\mathbb C^dvℓ​∈Cd and nonnegative weights pℓp_\ellpℓ​ summing to 111. Let PPP be a code with positive integers S,KS,KS,K, input words w∈{0,…,L−1}nw\in\{0,\ldots,L-1\}^nw∈{0,…,L−1}n, and output space indexed by x∈{0,…,d−1}nx\in\{0,\ldots,d-1\}^nx∈{0,…,d−1}n; for each www, its effects Ew,jE_{w,j}Ew,j​ are positive semidefinite and sum to ISI_SIS​, and each decoder has a finite representation Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​ with ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​. Define vw(x)=∏t=0n−1vw(t)(x(t))v_w(x)=\prod_{t=0}^{n-1}v_{w(t)}(x(t))vw​(x)=∏t=0n−1​vw(t)​(x(t)) and σw=∑j=0K−1Dj(S−1Ew,jT)\sigma_w=\sum_{j=0}^{K-1}\mathcal D_j(S^{-1}E_{w,j}^{\mathsf T})σw​=∑j=0K−1​Dj​(S−1Ew,jT​). The defined real number is

∑w∈{0,…,L−1}n(∏t=0n−1pw(t))Re⁡Tr⁡(vwvw†σw).\sum_{w\in\{0,\ldots,L-1\}^n} \left(\prod_{t=0}^{n-1}p_{w(t)}\right) \operatorname{Re}\operatorname{Tr}(v_wv_w^\dagger\sigma_w).w∈{0,…,L−1}n∑​(t=0∏n−1​pw(t)​)ReTr(vw​vw†​σw​).

When n=0n=0n=0, both word sets are singletons, all displayed products are empty products equal to 111, and the expression is the real trace of the output matrix for the empty input word.

QAlgorithms.Papers.Independent.Intakenext100.P54.pureOperational

For any natural number ddd, an ensemble EEE of LLL unit vectors vℓ∈Cdv_\ell\in\mathbb C^dvℓ​∈Cd with nonnegative weights pℓp_\ellpℓ​ summing to 111, and any real number RRR, this definition takes the infimum in [0,∞][0,\infty][0,∞] of all nonnegative real numbers eee with the following property: R≥0R\ge0R≥0, and for every real ε>0\varepsilon>0ε>0 and δ>0\delta>0δ>0, there is a natural number n0>0n_0>0n0​>0 such that every natural number n≥n0n\ge n_0n≥n0​ admits positive integers S,KS,KS,K, positive semidefinite effects Ew,jE_{w,j}Ew,j​ summing to ISI_SIS​ for each word w∈{0,…,L−1}nw\in\{0,\ldots,L-1\}^nw∈{0,…,L−1}n, and decoder maps into matrices indexed by {0,…,d−1}n\{0,\ldots,d-1\}^n{0,…,d−1}n of the form Dj(X)=∑ℓAj,ℓXAj,ℓ†\mathcal D_j(X)=\sum_\ell A_{j,\ell}XA_{j,\ell}^{\dagger}Dj​(X)=∑ℓ​Aj,ℓ​XAj,ℓ†​ for finite families satisfying ∑ℓAj,ℓ†Aj,ℓ=IS\sum_\ell A_{j,\ell}^{\dagger}A_{j,\ell}=I_S∑ℓ​Aj,ℓ†​Aj,ℓ​=IS​, such that, with vw(x)=∏t=0n−1vw(t)(x(t))v_w(x)=\prod_{t=0}^{n-1}v_{w(t)}(x(t))vw​(x)=∏t=0n−1​vw(t)​(x(t)) and σw=∑j=0K−1Dj(S−1Ew,jT)\sigma_w=\sum_{j=0}^{K-1}\mathcal D_j(S^{-1}E_{w,j}^{\mathsf T})σw​=∑j=0K−1​Dj​(S−1Ew,jT​),

∑w(∏t=0n−1pw(t))Re⁡Tr⁡(vwvw†σw)≥1−ε,log⁡2Kn≤R+δ,log⁡2Sn≤e+δ.\sum_w\left(\prod_{t=0}^{n-1}p_{w(t)}\right) \operatorname{Re}\operatorname{Tr}(v_wv_w^\dagger\sigma_w)\ge1-\varepsilon, \qquad \frac{\log_2K}{n}\le R+\delta, \qquad \frac{\log_2S}{n}\le e+\delta.w∑​(t=0∏n−1​pw(t)​)ReTr(vw​vw†​σw​)≥1−ε,nlog2​K​≤R+δ,nlog2​S​≤e+δ.

Each admissible eee is embedded into the extended nonnegative reals before taking the infimum. If no such eee exists, the result is +∞+\infty+∞; in particular this happens when R<0R<0R<0. The definition does not assert that the infimum is attained.

QAlgorithms.Papers.Independent.Intakenext100.P54.pureWordVector

For any natural number ddd, let EEE be a finite ensemble consisting of a natural number LLL, weights pℓ≥0p_\ell\ge0pℓ​≥0 with ∑ℓ=0L−1pℓ=1\sum_{\ell=0}^{L-1}p_\ell=1∑ℓ=0L−1​pℓ​=1, and vectors vℓ∈Cdv_\ell\in\mathbb C^dvℓ​∈Cd with ∥vℓ∥=1\|v_\ell\|=1∥vℓ​∥=1. For any natural number nnn and word w:{0,…,n−1}→{0,…,L−1}w:\{0,\ldots,n-1\}\to\{0,\ldots,L-1\}w:{0,…,n−1}→{0,…,L−1}, the defined vector belongs to C{0,…,d−1}n\mathbb C^{\{0,\ldots,d-1\}^n}C{0,…,d−1}n and has coordinate

vw(x)=∏t=0n−1vw(t)(x(t))(x:{0,…,n−1}→{0,…,d−1}).v_w(x)=\prod_{t=0}^{n-1}v_{w(t)}(x(t)) \qquad \left(x:\{0,\ldots,n-1\}\to\{0,\ldots,d-1\}\right).vw​(x)=t=0∏n−1​vw(t)​(x(t))(x:{0,…,n−1}→{0,…,d−1}).

The weights do not enter this formula. When n=0n=0n=0, there is one empty coordinate word and the empty product gives its amplitude as 111. Although d,Ld,Ld,L are natural numbers without explicit positivity hypotheses, the ensemble requirements cannot be satisfied if either is zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.randomise

For any natural numbers D,KD,KD,K, any family of unitary complex D×DD\times DD×D matrices U0,…,UK−1U_0,\ldots,U_{K-1}U0​,…,UK−1​, and any complex D×DD\times DD×D matrix ρ\rhoρ, this defines

RU(ρ)=K−1∑k=0K−1UkρUk†,\mathcal R_U(\rho)=K^{-1}\sum_{k=0}^{K-1}U_k\rho U_k^\dagger,RU​(ρ)=K−1k=0∑K−1​Uk​ρUk†​,

where †\dagger† denotes conjugate transpose and the scalar is interpreted in C\mathbb CC. No positivity or normalization of ρ\rhoρ is required. The definition includes K=0K=0K=0, when the empty sum and the inverse of zero are both zero, so the result is the zero matrix; it also includes D=0D=0D=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.randomisingBound

For any natural number DDD and any real ε\varepsilonε, this defines the real number

B(D,ε)=(10ε)2Dlog⁡2 ⁣(20Dε).B(D,\varepsilon)= \left(\frac{10}{\varepsilon}\right)^2D \log_2\!\left(\frac{20D}{\varepsilon}\right).B(D,ε)=(ε10​)2Dlog2​(ε20D​).

There are no restrictions on these arguments. The real division and logarithm are total functions, with division by zero and log⁡20\log_2 0log2​0 interpreted as zero; in particular, the defined value is zero if D=0D=0D=0 or ε=0\varepsilon=0ε=0.

QAlgorithms.Papers.Independent.Intakenext100.P54.rspMinimumAttained

For arbitrary natural numbers m,dm,dm,d, real pip_ipi​, complex d×dd\times dd×d matrices ρi\rho_iρi​, and real RRR, define H(X)=−Re⁡tr⁡[f(X)]H(X)=-\operatorname{Re}\operatorname{tr}[f(X)]H(X)=−Retr[f(X)] by continuous functional calculus for f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x, and Hdiag(v)=H(diag⁡(v))H_{\mathrm{diag}}(v)=H(\operatorname{diag}(v))Hdiag​(v)=H(diag(v)). For each natural kkk and real array cij≥0c_{ij}\ge0cij​≥0 with ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1, set qj=∑ipicijq_j=\sum_i p_i c_{ij}qj​=∑i​pi​cij​, vi(j)=picij/qjv^{(j)}_i=p_i c_{ij}/q_jvi(j)​=pi​cij​/qj​, and σj=d−1Id\sigma_j=d^{-1}I_dσj​=d−1Id​ if qj=0q_j=0qj​=0, otherwise σj=∑ivi(j)ρi\sigma_j=\sum_i v^{(j)}_i\rho_iσj​=∑i​vi(j)​ρi​. Put

B(c)=∑jqjH(σj),C(c)=Hdiag(p)−∑jqjHdiag(v(j))+B(c)−∑ipiH(ρi),B(c)=\sum_jq_jH(\sigma_j),\qquad C(c)=H_{\mathrm{diag}}(p)-\sum_jq_jH_{\mathrm{diag}}(v^{(j)}) +B(c)-\sum_ip_iH(\rho_i),B(c)=j∑​qj​H(σj​),C(c)=Hdiag​(p)−j∑​qj​Hdiag​(v(j))+B(c)−i∑​pi​H(ρi​),

and let VVV be the infimum in [0,∞][0,\infty][0,∞] of max⁡(B(c),0)\max(B(c),0)max(B(c),0) over these arrays satisfying C(c)≤RC(c)\le RC(c)≤R. The defined proposition says: if V≠+∞V\ne+\inftyV=+∞, then there exist such a finite kkk and array ccc with C(c)≤RC(c)\le RC(c)≤R and max⁡(B(c),0)=V\max(B(c),0)=Vmax(B(c),0)=V. It imposes no requirement when V=+∞V=+\inftyV=+∞, including when the feasible set is empty. There are no distribution or density-matrix hypotheses; zero dimensions are allowed and division by zero is zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.rspSingleLetter

For arbitrary natural numbers m,dm,dm,d, real pip_ipi​, complex d×dd\times dd×d matrices ρi\rho_iρi​, and R∈RR\in\mathbb RR∈R, consider every natural number kkk and every real array cij≥0c_{ij}\ge0cij​≥0 satisfying ∑j<kcij=1\sum_{j<k}c_{ij}=1∑j<k​cij​=1 for every i<mi<mi<m. For each such array, set qj=∑i<mpicijq_j=\sum_{i<m}p_i c_{ij}qj​=∑i<m​pi​cij​, vi(j)=picij/qjv^{(j)}_i=p_i c_{ij}/q_jvi(j)​=pi​cij​/qj​, and σj=d−1Id\sigma_j=d^{-1}I_dσj​=d−1Id​ if qj=0q_j=0qj​=0, otherwise σj=∑ivi(j)ρi\sigma_j=\sum_i v^{(j)}_i\rho_iσj​=∑i​vi(j)​ρi​. Define H(X)=−Re⁡tr⁡[f(X)]H(X)=-\operatorname{Re}\operatorname{tr}[f(X)]H(X)=−Retr[f(X)] by continuous functional calculus for f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x, and Hdiag(v)=H(diag⁡(v))H_{\mathrm{diag}}(v)=H(\operatorname{diag}(v))Hdiag​(v)=H(diag(v)). Put

B(c)=∑jqjH(σj),C(c)=Hdiag(p)−∑jqjHdiag(v(j))+B(c)−∑ipiH(ρi).B(c)=\sum_jq_jH(\sigma_j),\qquad C(c)=H_{\mathrm{diag}}(p)-\sum_jq_jH_{\mathrm{diag}}(v^{(j)}) +B(c)-\sum_ip_iH(\rho_i).B(c)=j∑​qj​H(σj​),C(c)=Hdiag​(p)−j∑​qj​Hdiag​(v(j))+B(c)−i∑​pi​H(ρi​).

The definition is the infimum, in [0,∞][0,\infty][0,∞], of

{max⁡(B(c),0):k∈N, c satisfies the stated row conditions, C(c)≤R}.\bigl\{\max(B(c),0): k\in\mathbb N,\ c\text{ satisfies the stated row conditions},\ C(c)\le R\bigr\}.{max(B(c),0):k∈N, c satisfies the stated row conditions, C(c)≤R}.

The infimum of an empty feasible set is +∞+\infty+∞. No normalization or positivity conditions are imposed on ppp or ρi\rho_iρi​, and RRR may be negative. Zero dimensions and k=0k=0k=0 are included wherever the row conditions permit them, and division by zero is zero.

QAlgorithms.Papers.Independent.Intakenext100.P54.squaredGaussianRate

For every nonnegative real variance parameter v=s2v=s2v=s2 and every real number xxx, this definition is the extended real supremum

sup⁡y∈R{yx−L ⁣(∫R−eyt2 dγ0,v(t))},\sup_{y\in\mathbb R} \left\{ yx-\mathcal L\!\left( \int_{\mathbb R}^{-} e^{yt^2}\,d\gamma_{0,v}(t) \right) \right\},y∈Rsup​{yx−L(∫R−​eyt2dγ0,v​(t))},

where γ0,v\gamma_{0,v}γ0,v​ is the real Gaussian measure of mean 000 and variance vvv, the integral is the extended nonnegative integral, and L(z)\mathcal L(z)L(z) equals +∞+\infty+∞ at z=+∞z=+\inftyz=+∞, −∞-\infty−∞ at z=0z=0z=0, and ln⁡z\ln zlnz at finite positive zzz. The exponential integrand is converted to an extended nonnegative real before integration. The supremum includes every real yyy, including those for which the integral is infinite; v=0v=0v=0 and arbitrary, possibly nonpositive, xxx are allowed.

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