Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Capacity Theorems for Quantum Multiple Access Channels: Classical-Quantum and Quantum-Quantum Capacity Regions

Definition
QAlgorithms_Papers_Independent_Support50_P02

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

prove2quantum

The objects the formalized statements of Capacity Theorems for Quantum Multiple Access Channels: Classical-Quantum and Quantum-Quantum Capacity Regions (arXiv:quant-ph/0501045) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • independentPairs: Reindex two independent reference/input pairs as (references, inputs), pp.5–7, (8).
  • macPower: Memoryless MAC power with each sender's n inputs grouped together, pp.5–7.
  • bellPair: Product of two Bell targets, regrouped as (references, receiver outputs), p.7.
  • cqInput: Alice's input and Bob's independent reference/input state in (7), p.8.
  • cqMessageTarget: Joint classical label/pure quantum target, with reference first, pp.6–7,23.
  • dephaseClassicalRegister: Dephase the decoder's classical register, making its output an instrument, pp.5–6.
  • pureEnsembleAverage: The average state, p.4.
  • CQMacData: Data of (7), p.8: one Alice ensemble and one independent pure Bob pair.
  • CQMacData.output: Conditional BC states from that same input data, (7), p.8.
  • CQMacData.averageOutput: Average BC output of (7), p.8.
  • CQMacData.classical: I(X;C) of (7), pp.4,8.
  • CQMacData.conditionalCoherent: I_c(B coherent-to CX), evaluated on the SAME state (7), pp.4,8.
  • CQMacData.sideClassical: I(X;BC) of (7), p.30.
  • CQMacData.coherent: I_c(B coherent-to C) of the averaged BC state, p.30.
  • QQData: Data of (8), p.8: independent pure reference/input pairs, with arbitrary references.
  • QQData.output: The actual ABC output in (8), reference pair grouped first, p.8.
  • QQData.bobOutput: BC marginal computed by tracing Alice's reference from that output, p.8.
  • QQData.aliceOutput: AC marginal computed by tracing Bob's reference from that output, p.8.
  • QQData.aliceConditional: I_c(A coherent-to BC), p.8.
  • QQData.bobConditional: I_c(B coherent-to AC), p.8.
  • QQData.sumCoherent: I_c(AB coherent-to C), p.8.
  • QQData.aliceCoherent: I_c(A coherent-to C), p.30.
  • QQData.bobCoherent: I_c(B coherent-to C), p.30.
  • Nonnegative: Nonnegative operational rate pairs, pp.5–8.
  • CQMacData.Rectangle: Boundary-inclusive rectangle from a fixed state (7), pp.7–8,13. Correction: replace the printed strict signs BEFORE closure; the replacer edge otherwise excludes (0,0). This does not claim equality of the strict and closed unions.
  • cqOne: CQ one-use region; optional cap on the alphabet, pp.8,36.
  • QQData.Pentagon: One-state qq pentagon with ALL three constraints, pp.8,18. The same disclosed non-strict boundary correction as for Theorem 1 is used.
  • qqOne: QQ one-use pentagon union, p.8.
  • qqRectangles: Earlier qq rectangle union, on the SAME product inputs (8), p.30.
  • cqPentagons: Alternative cq pentagons on a single state (7), p.30.
  • regularized: Positive-block regularization, with multiplication avoiding division at k=0, pp.7–8,30.
  • CQGeneration: CQ generation code, pp.5–6, (1)–(2): separate pure preparations and joint decoder. A final dephasing in cqDecoded makes the classical output an instrument.
  • cqDecoded: Apply the one n-use joint channel and decoder while leaving Bob's reference untouched, then read the classical decoder output, pp.5–7.
  • CQGeneration.Good: Uniform-message average of a JOINT classical/Bell fidelity, (1)–(2), p.6.
  • CQTransmission: CQ entanglement transmission, pp.6–7: local CPTP encoder on a Bell half only.
  • CQTransmission.Good: Transmission success (3)–(4), pp.6–7; reference is never acted on by the encoder.
  • CQStrong: CQ strong transmission, p.23: local stochastic classical preparations, one fixed quantum encoder and one fixed joint decoder. Mixed message states represent finite private random pure preparations after discarding Alice's private seed. No seed is available to Charlie except through the charged n channel uses.
  • CQStrong.Good: Maximal JOINT error for every classical message and every reference/input pure state, (36), p.23. Correction: test the ORIGINAL message, as the inverse cyclic relabeling on p.25 does. The printed arbitrary f(m) target would permit a constant f.
  • QQGeneration: QQ generation code, p.7, (5); the sender preparations are independent.
  • QQGeneration.Good: One fidelity against the PRODUCT Bell target, (5), p.7.
  • QQTransmission: QQ transmission code, (6), p.7; two individually local CPTP encoders, no reference access or pre-shared sender/decoder resources.
  • QQTransmission.decoded: Encoders act only on the two input halves; channel and decoder act jointly, p.7.
  • QQTransmission.Good: One simultaneous fidelity, not separate marginal fidelities, (6), p.7.
  • QQTransmission.Strong: Strong qq transmission, (37), pp.23–24: all independent external references, allowing arbitrary entanglement WITHIN each pair, and one product-target fidelity.
  • operationalRegion: Closed region from actual integer-rank codes with vanishing error and arbitrary asymptotic rate slack, pp.6–7. Positive n,m,k guard all logs, divisions, and targets. The code predicate contains its existential code BEFORE its universal input tests.
  • MACScenario: The three operational tasks, pp.5–7,23–24.
  • cqMacCapacity: Closed operational CQ region for the chosen task, pp.5–7,23.
  • qqMacCapacity: Closed operational Q region for the chosen task, pp.5–7,23–24.
  • optimizedSum: Optimized sum information over exactly product purifications (8), p.30. For finite positive physical dimensions and CPTP N, this set is nonempty and bounded; reference compression bounds their dimensions and gives attainment.
  • controlledErasure: Corrected controlled-erasure channel of §7.1, PDF p.32. Printed N0=sum_j |flag><0,j| and Ni=|i><1,i| are NOT TP and dephase Bob. The prose and N(tau tensor rho)=tau00|flag><flag|+tau11 rho instead specify K_j=|flag><0,j| and ONE L=sum_j |j><1,j|. This entry formula implements those Kraus operators, retaining transfer-sector coherences.
  • collectivePhaseOperator: Collective phase flip Z tensor Z, §7.2, PDF p.34.
  • collectivePhase: The convex mixture (1-p)rho+p(Z tensor Z)rho(Z tensor Z), p.34.
  • inputCoherent: Channel coherent information using the SAME isometric extension for both outputs, §7.3, PDF p.35. For density rho this equals the purification definition.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_Channel
import Definitions.Def_QAlgorithms_Defs_Ensemble
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Defs_QuantumInformation
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P67
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P21
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P03
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P19

/-!
Yard–Devetak–Hayden, arXiv:quant-ph/0501045v3. Page numbers are PDF pages.
All registers are finite, dimensions in statements are positive, entropy is in bits,
fidelity is squared Uhlmann fidelity, and trace distance is the FULL trace norm.
The operational code definitions below are independent of the information regions.
Operator matrices, normalized pure ensembles, and Bell densities reuse accepted primitives.
Paper-specific MAC definitions have names distinct from unrelated frozen concepts.
Compilation is pending the explicitly requested batch handoff.
-/

noncomputable section

open scoped BigOperators Matrix ComplexConjugate

namespace QAlgorithms.Papers.Independent.Support50.P02

open QAlgorithms.Papers.Independent.Information100.P67
  (squaredFidelity receiverChannel complementaryChannel IsDegradable)

open QAlgorithms.Papers.Independent.Information100.P21 (Op)
open QAlgorithms.Papers.Independent.Information100.P03 (PureEnsemble)
open QAlgorithms.Papers.Independent.Dependency50.P19 (epr)

attribute [local instance] Classical.propDecidable Classical.decEq

variable {A B C E R S I J : Type*}
variable [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
variable [Fintype C] [DecidableEq C] [Fintype E] [DecidableEq E]
variable [Fintype R] [DecidableEq R] [Fintype S] [DecidableEq S]
variable [Fintype I] [DecidableEq I] [Fintype J] [DecidableEq J]

/-- Reindex two independent reference/input pairs as (references, inputs), pp.5–7, (8). -/
def independentPairs (rho : Op (R × A)) (tau : Op (S × B)) : Op ((R × S) × (A × B)) :=
  fun i j => rho (i.1.1, i.2.1) (j.1.1, j.2.1) * tau (i.1.2, i.2.2) (j.1.2, j.2.2)

/-- Memoryless MAC power with each sender's n inputs grouped together, pp.5–7. -/
def macPower (N : Superop (A × B) C) (n : ℕ) :
    Superop ((Fin n → A) × (Fin n → B)) (Fin n → C) :=
  fun rho => piSuperop (fun _ : Fin n => N)
    (fun x y => rho ((fun k => (x k).1), (fun k => (x k).2))
      ((fun k => (y k).1), (fun k => (y k).2)))

/-- Product of two Bell targets, regrouped as (references, receiver outputs), p.7. -/
def bellPair (a b : ℕ) : Op ((Fin a × Fin b) × (Fin a × Fin b)) :=
  independentPairs (epr a) (epr b)

/-- Alice's input and Bob's independent reference/input state in (7), p.8. -/
def cqInput (rho : Op A) (tau : Op (R × B)) : Op (R × (A × B)) :=
  fun i j => rho i.2.1 j.2.1 * tau (i.1,i.2.2) (j.1,j.2.2)

/-- Joint classical label/pure quantum target, with reference first, pp.6–7,23. -/
def cqMessageTarget {m : ℕ} (s : Fin m) (rho : Op (R × I)) : Op (R × (Fin m × I)) :=
  fun i j => if i.2.1 = s ∧ j.2.1 = s then rho (i.1,i.2.2) (j.1,j.2.2) else 0

/-- Dephase the decoder's classical register, making its output an instrument, pp.5–6. -/
def dephaseClassicalRegister {m : ℕ} (rho : Op (R × (Fin m × I))) : Op (R × (Fin m × I)) :=
  fun i j => if i.2.1 = j.2.1 then rho i j else 0

/-- The average state, p.4. -/
def pureEnsembleAverage (P : PureEnsemble A) : Op A :=
  ∑ x, (P.weight x : ℂ) • pureDensity (P.vector x)

/-- Data of (7), p.8: one Alice ensemble and one independent pure Bob pair. -/
structure CQMacData (A B : Type*) [Fintype A] [Fintype B] where
  ensemble : PureEnsemble A
  bobRank : ℕ
  bobPositive : 0 < bobRank
  bob : EuclideanSpace ℂ (Fin bobRank × B)
  bobUnit : IsState bob

/-- Conditional BC states from that same input data, (7), p.8. -/
def CQMacData.output (D : CQMacData A B) (N : Superop (A × B) C)
    (x : Fin D.ensemble.count) : Op (Fin D.bobRank × C) :=
  idTensorSuperop N (cqInput (pureDensity (D.ensemble.vector x)) (pureDensity D.bob))

/-- Average BC output of (7), p.8. -/
def CQMacData.averageOutput (D : CQMacData A B) (N : Superop (A × B) C) : Op (Fin D.bobRank × C) :=
  ∑ x, (D.ensemble.weight x : ℂ) • D.output N x

/-- I(X;C) of (7), pp.4,8. -/
def CQMacData.classical (D : CQMacData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (traceLeft (D.averageOutput N)) -
    ∑ x, D.ensemble.weight x * vnEntropy (traceLeft (D.output N x))

/-- I_c(B coherent-to CX), evaluated on the SAME state (7), pp.4,8. -/
def CQMacData.conditionalCoherent (D : CQMacData A B) (N : Superop (A × B) C) : ℝ :=
  ∑ x, D.ensemble.weight x *
    (vnEntropy (traceLeft (D.output N x)) - vnEntropy (D.output N x))

/-- I(X;BC) of (7), p.30. -/
def CQMacData.sideClassical (D : CQMacData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (D.averageOutput N) - ∑ x, D.ensemble.weight x * vnEntropy (D.output N x)

/-- I_c(B coherent-to C) of the averaged BC state, p.30. -/
def CQMacData.coherent (D : CQMacData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (traceLeft (D.averageOutput N)) - vnEntropy (D.averageOutput N)

/-- Data of (8), p.8: independent pure reference/input pairs, with arbitrary references. -/
structure QQData (A B : Type*) [Fintype A] [Fintype B] where
  aliceRank : ℕ
  alicePositive : 0 < aliceRank
  bobRank : ℕ
  bobPositive : 0 < bobRank
  alice : EuclideanSpace ℂ (Fin aliceRank × A)
  aliceUnit : IsState alice
  bob : EuclideanSpace ℂ (Fin bobRank × B)
  bobUnit : IsState bob

/-- The actual ABC output in (8), reference pair grouped first, p.8. -/
def QQData.output (D : QQData A B) (N : Superop (A × B) C) :
    Op ((Fin D.aliceRank × Fin D.bobRank) × C) :=
  idTensorSuperop N (independentPairs (pureDensity D.alice) (pureDensity D.bob))

/-- BC marginal computed by tracing Alice's reference from that output, p.8. -/
def QQData.bobOutput (D : QQData A B) (N : Superop (A × B) C) : Op (Fin D.bobRank × C) :=
  fun i j => ∑ a, D.output N ((a,i.1),i.2) ((a,j.1),j.2)

/-- AC marginal computed by tracing Bob's reference from that output, p.8. -/
def QQData.aliceOutput (D : QQData A B) (N : Superop (A × B) C) : Op (Fin D.aliceRank × C) :=
  fun i j => ∑ b, D.output N ((i.1,b),i.2) ((j.1,b),j.2)

/-- I_c(A coherent-to BC), p.8. -/
def QQData.aliceConditional (D : QQData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (D.bobOutput N) - vnEntropy (D.output N)

/-- I_c(B coherent-to AC), p.8. -/
def QQData.bobConditional (D : QQData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (D.aliceOutput N) - vnEntropy (D.output N)

/-- I_c(AB coherent-to C), p.8. -/
def QQData.sumCoherent (D : QQData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (traceLeft (D.output N)) - vnEntropy (D.output N)

/-- I_c(A coherent-to C), p.30. -/
def QQData.aliceCoherent (D : QQData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (traceLeft (D.output N)) - vnEntropy (D.aliceOutput N)

/-- I_c(B coherent-to C), p.30. -/
def QQData.bobCoherent (D : QQData A B) (N : Superop (A × B) C) : ℝ :=
  vnEntropy (traceLeft (D.output N)) - vnEntropy (D.bobOutput N)

/-- Nonnegative operational rate pairs, pp.5–8. -/
def Nonnegative (r : ℝ × ℝ) : Prop := 0 ≤ r.1 ∧ 0 ≤ r.2

/-- Boundary-inclusive rectangle from a fixed state (7), pp.7–8,13.
Correction: replace the printed strict signs BEFORE closure; the replacer edge
otherwise excludes (0,0). This does not claim equality of the strict and closed unions. -/
def CQMacData.Rectangle (D : CQMacData A B) (N : Superop (A × B) C) (r : ℝ × ℝ) : Prop :=
  Nonnegative r ∧ r.1 ≤ D.classical N ∧ r.2 ≤ D.conditionalCoherent N

/-- CQ one-use region; optional cap on the alphabet, pp.8,36. -/
def cqOne (N : Superop (A × B) C) (cap : Option ℕ := none) : Set (ℝ × ℝ) :=
  {r | ∃ D : CQMacData A B, (∀ b ∈ cap, D.ensemble.count ≤ b) ∧ D.Rectangle N r}

/-- One-state qq pentagon with ALL three constraints, pp.8,18.
The same disclosed non-strict boundary correction as for Theorem 1 is used. -/
def QQData.Pentagon (D : QQData A B) (N : Superop (A × B) C) (r : ℝ × ℝ) : Prop :=
  Nonnegative r ∧ r.1 ≤ D.aliceConditional N ∧ r.2 ≤ D.bobConditional N ∧
    r.1 + r.2 ≤ D.sumCoherent N

/-- QQ one-use pentagon union, p.8. -/
def qqOne (N : Superop (A × B) C) : Set (ℝ × ℝ) :=
  {r | ∃ D : QQData A B, D.Pentagon N r}

/-- Earlier qq rectangle union, on the SAME product inputs (8), p.30. -/
def qqRectangles (N : Superop (A × B) C) : Set (ℝ × ℝ) :=
  {r | Nonnegative r ∧ ∃ D : QQData A B,
    r.1 ≤ D.aliceCoherent N ∧ r.2 ≤ D.bobCoherent N}

/-- Alternative cq pentagons on a single state (7), p.30. -/
def cqPentagons (N : Superop (A × B) C) : Set (ℝ × ℝ) :=
  {r | Nonnegative r ∧ ∃ D : CQMacData A B, r.1 ≤ D.sideClassical N ∧
    r.2 ≤ D.conditionalCoherent N ∧ r.1 + r.2 ≤ D.classical N + D.conditionalCoherent N}

/-- Positive-block regularization, with multiplication avoiding division at k=0, pp.7–8,30. -/
def regularized (F : (n : ℕ) → Set (ℝ × ℝ)) : Set (ℝ × ℝ) :=
  {r | ∃ n : ℕ, 0 < n ∧ ((n : ℝ) * r.1, (n : ℝ) * r.2) ∈ F n}

/-- CQ generation code, pp.5–6, (1)–(2): separate pure preparations and joint decoder.
A final dephasing in `cqDecoded` makes the classical output an instrument. -/
structure CQGeneration (A B C : Type*) [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq C] (n m k : ℕ) where
  alice : Fin m → EuclideanSpace ℂ (Fin n → A)
  aliceUnit : ∀ s, IsState (alice s)
  bob : EuclideanSpace ℂ (Fin k × (Fin n → B))
  bobUnit : IsState bob
  decoder : Superop (Fin n → C) (Fin m × Fin k)
  decoderChannel : IsChannel decoder

/-- Apply the one n-use joint channel and decoder while leaving Bob's reference untouched,
then read the classical decoder output, pp.5–7. -/
def cqDecoded {n m k : ℕ} (N : Superop (A × B) C)
    (D : Superop (Fin n → C) (Fin m × Fin k))
    (rho : Op (Fin n → A)) (tau : Op (R × (Fin n → B))) :
    Op (R × (Fin m × Fin k)) :=
  dephaseClassicalRegister (idTensorSuperop D (idTensorSuperop (macPower N n) (cqInput rho tau)))

/-- Uniform-message average of a JOINT classical/Bell fidelity, (1)–(2), p.6. -/
def CQGeneration.Good {n m k : ℕ} (K : CQGeneration A B C n m k)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  1 - epsilon ≤ (m : ℝ)⁻¹ * ∑ s, squaredFidelity (cqMessageTarget s (epr k))
    (cqDecoded N K.decoder (pureDensity (K.alice s)) (pureDensity K.bob))

/-- CQ entanglement transmission, pp.6–7: local CPTP encoder on a Bell half only. -/
structure CQTransmission (A B C : Type*) [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq C] (n m k : ℕ) where
  alice : Fin m → EuclideanSpace ℂ (Fin n → A)
  aliceUnit : ∀ s, IsState (alice s)
  encoder : Superop (Fin k) (Fin n → B)
  encoderChannel : IsChannel encoder
  decoder : Superop (Fin n → C) (Fin m × Fin k)
  decoderChannel : IsChannel decoder

/-- Transmission success (3)–(4), pp.6–7; reference is never acted on by the encoder. -/
def CQTransmission.Good {n m k : ℕ} (K : CQTransmission A B C n m k)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  1 - epsilon ≤ (m : ℝ)⁻¹ * ∑ s, squaredFidelity (cqMessageTarget s (epr k))
    (cqDecoded N K.decoder (pureDensity (K.alice s)) (idTensorSuperop K.encoder (epr k)))

/-- CQ strong transmission, p.23: local stochastic classical preparations, one fixed
quantum encoder and one fixed joint decoder. Mixed message states represent finite
private random pure preparations after discarding Alice's private seed. No seed is
available to Charlie except through the charged n channel uses. -/
structure CQStrong (A B C : Type*) [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq C] (n m k : ℕ) where
  alice : Fin m → Op (Fin n → A)
  aliceDensity : ∀ s, IsDensityMatrix (alice s)
  encoder : Superop (Fin k) (Fin n → B)
  encoderChannel : IsChannel encoder
  decoder : Superop (Fin n → C) (Fin m × Fin k)
  decoderChannel : IsChannel decoder

/-- Maximal JOINT error for every classical message and every reference/input pure state,
(36), p.23. Correction: test the ORIGINAL message, as the inverse cyclic relabeling
on p.25 does. The printed arbitrary f(m) target would permit a constant f. -/
def CQStrong.Good {n m k : ℕ} (K : CQStrong A B C n m k)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  ∀ s, ∀ t : ℕ, 0 < t → ∀ psi : EuclideanSpace ℂ (Fin t × Fin k), IsState psi →
    1 - epsilon ≤ squaredFidelity (cqMessageTarget s (pureDensity psi))
      (cqDecoded N K.decoder (K.alice s) (idTensorSuperop K.encoder (pureDensity psi)))

/-- QQ generation code, p.7, (5); the sender preparations are independent. -/
structure QQGeneration (A B C : Type*) [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq C] (n a b : ℕ) where
  alice : EuclideanSpace ℂ (Fin a × (Fin n → A))
  aliceUnit : IsState alice
  bob : EuclideanSpace ℂ (Fin b × (Fin n → B))
  bobUnit : IsState bob
  decoder : Superop (Fin n → C) (Fin a × Fin b)
  decoderChannel : IsChannel decoder

/-- One fidelity against the PRODUCT Bell target, (5), p.7. -/
def QQGeneration.Good {n a b : ℕ} (K : QQGeneration A B C n a b)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  1 - epsilon ≤ squaredFidelity (bellPair a b)
    (idTensorSuperop K.decoder (idTensorSuperop (macPower N n)
      (independentPairs (pureDensity K.alice) (pureDensity K.bob))))

/-- QQ transmission code, (6), p.7; two individually local CPTP encoders,
no reference access or pre-shared sender/decoder resources. -/
structure QQTransmission (A B C : Type*) [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq C] (n a b : ℕ) where
  alice : Superop (Fin a) (Fin n → A)
  aliceChannel : IsChannel alice
  bob : Superop (Fin b) (Fin n → B)
  bobChannel : IsChannel bob
  decoder : Superop (Fin n → C) (Fin a × Fin b)
  decoderChannel : IsChannel decoder

/-- Encoders act only on the two input halves; channel and decoder act jointly, p.7. -/
def QQTransmission.decoded {n a b : ℕ} (K : QQTransmission A B C n a b)
    (N : Superop (A × B) C) (rho : Op (R × Fin a)) (tau : Op (S × Fin b)) :
    Op ((R × S) × (Fin a × Fin b)) :=
  idTensorSuperop K.decoder (idTensorSuperop (macPower N n)
    (independentPairs (idTensorSuperop K.alice rho) (idTensorSuperop K.bob tau)))

/-- One simultaneous fidelity, not separate marginal fidelities, (6), p.7. -/
def QQTransmission.Good {n a b : ℕ} (K : QQTransmission A B C n a b)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  1 - epsilon ≤ squaredFidelity (bellPair a b) (K.decoded N (epr a) (epr b))

/-- Strong qq transmission, (37), pp.23–24: all independent external references,
allowing arbitrary entanglement WITHIN each pair, and one product-target fidelity. -/
def QQTransmission.Strong {n a b : ℕ} (K : QQTransmission A B C n a b)
    (N : Superop (A × B) C) (epsilon : ℝ) : Prop :=
  ∀ q t : ℕ, 0 < q → 0 < t →
    ∀ psi : EuclideanSpace ℂ (Fin q × Fin a), IsState psi →
    ∀ phi : EuclideanSpace ℂ (Fin t × Fin b), IsState phi →
      1 - epsilon ≤ squaredFidelity (independentPairs (pureDensity psi) (pureDensity phi))
        (K.decoded N (pureDensity psi) (pureDensity phi))

/-- Closed region from actual integer-rank codes with vanishing error and arbitrary
asymptotic rate slack, pp.6–7. Positive n,m,k guard all logs, divisions, and targets.
The code predicate contains its existential code BEFORE its universal input tests. -/
def operationalRegion (good : ℕ → ℕ → ℕ → ℝ → Prop) : Set (ℝ × ℝ) :=
  closure {r | Nonnegative r ∧ ∀ epsilon delta : ℝ, 0 < epsilon → 0 < delta →
    ∃ n0 : ℕ, 0 < n0 ∧ ∀ n ≥ n0, ∃ a b : ℕ, 0 < a ∧ 0 < b ∧
      r.1 - delta ≤ Real.logb 2 a / n ∧ r.2 - delta ≤ Real.logb 2 b / n ∧
      good n a b epsilon}

/-- The three operational tasks, pp.5–7,23–24. -/
inductive MACScenario where
  | generation
  | transmission
  | strong

/-- Closed operational CQ region for the chosen task, pp.5–7,23. -/
def cqMacCapacity (N : Superop (A × B) C) (scenario : MACScenario := .generation) : Set (ℝ × ℝ) :=
  operationalRegion fun n m k epsilon => match scenario with
    | .generation => ∃ K : CQGeneration A B C n m k, K.Good N epsilon
    | .transmission => ∃ K : CQTransmission A B C n m k, K.Good N epsilon
    | .strong => ∃ K : CQStrong A B C n m k, K.Good N epsilon

/-- Closed operational Q region for the chosen task, pp.5–7,23–24. -/
def qqMacCapacity (N : Superop (A × B) C) (scenario : MACScenario := .generation) : Set (ℝ × ℝ) :=
  operationalRegion fun n a b epsilon => match scenario with
    | .generation => ∃ K : QQGeneration A B C n a b, K.Good N epsilon
    | .transmission => ∃ K : QQTransmission A B C n a b, K.Good N epsilon
    | .strong => ∃ K : QQTransmission A B C n a b, K.Strong N epsilon

/-- Optimized sum information over exactly product purifications (8), p.30.
For finite positive physical dimensions and CPTP N, this set is nonempty and
bounded; reference compression bounds their dimensions and gives attainment. -/
def optimizedSum (N : Superop (A × B) C) : ℝ :=
  sSup {z | ∃ D : QQData A B, z = D.sumCoherent N}

/-- Corrected controlled-erasure channel of §7.1, PDF p.32.
Printed N0=sum_j |flag><0,j| and Ni=|i><1,i| are NOT TP and dephase Bob.
The prose and N(tau tensor rho)=tau00|flag><flag|+tau11 rho instead specify
K_j=|flag><0,j| and ONE L=sum_j |j><1,j|. This entry formula implements those
Kraus operators, retaining transfer-sector coherences. -/
def controlledErasure (d : ℕ) : Superop (Bool × Fin d) (Option (Fin d)) :=
  fun rho i j => match i, j with
    | none, none => ∑ a, rho (false,a) (false,a)
    | some a, some b => rho (true,a) (true,b)
    | _, _ => 0

/-- Collective phase flip Z tensor Z, §7.2, PDF p.34. -/
def collectivePhaseOperator : Op (Bool × Bool) :=
  Matrix.diagonal fun z => (if z.1 then (-1 : ℂ) else 1) * (if z.2 then (-1 : ℂ) else 1)

/-- The convex mixture (1-p)rho+p(Z tensor Z)rho(Z tensor Z), p.34. -/
def collectivePhase (p : ℝ) : Superop (Bool × Bool) (Bool × Bool) :=
  fun rho => ((1 - p : ℝ) : ℂ) • rho + (p : ℂ) • (collectivePhaseOperator * rho * collectivePhaseOperator)

/-- Channel coherent information using the SAME isometric extension for both
outputs, §7.3, PDF p.35. For density rho this equals the purification definition. -/
def inputCoherent (U : Matrix (B × E) A ℂ) (rho : Op A) : ℝ :=
  vnEntropy (receiverChannel U rho) - vnEntropy (complementaryChannel U rho)

/-- Sanity checks for the local definitions: scalar Bell normalization, independent
references, classical labels, and erasure coherences. All proofs are complete. -/
example : epr 1 (0,0) (0,0) = 1 := by simp [epr]

example (a b : ℕ) (i j : (Fin a × Fin b) × (Fin a × Fin b)) :
    bellPair a b i j = epr a (i.1.1,i.2.1) (j.1.1,j.2.1) *
      epr b (i.1.2,i.2.2) (j.1.2,j.2.2) := rfl

example (rho : Op (R × A)) (tau : Op (S × B))
    (r rr : R) (s ss : S) (a aa : A) (b bb : B) :
    independentPairs rho tau ((r,s),(a,b)) ((rr,ss),(aa,bb)) =
      rho (r,a) (rr,aa) * tau (s,b) (ss,bb) := rfl

example {m : ℕ} (s : Fin m) (rho : Op (R × I)) (r t : R) (i j : I) :
    cqMessageTarget s rho (r,(s,i)) (t,(s,j)) = rho (r,i) (t,j) := by simp [cqMessageTarget]

example (d : ℕ) (rho : Op (Bool × Fin d)) (a b : Fin d) :
    controlledErasure d rho (some a) (some b) = rho (true,a) (true,b) := rfl

example (d : ℕ) (rho : Op (Bool × Fin d)) :
    controlledErasure d rho none none = ∑ a, rho (false,a) (false,a) := rfl

example (rho : Op (Bool × Bool)) : collectivePhase 0 rho = rho := by
  simp [collectivePhase]

example : collectivePhaseOperator (false,true) (false,true) = -1 := by
  simp [collectivePhaseOperator]

example : collectivePhaseOperator (true,true) (true,true) = 1 := by
  simp [collectivePhaseOperator]

example (D : CQMacData A B) (N : Superop (A × B) C) :
    D.conditionalCoherent N = ∑ x, D.ensemble.weight x *
      (vnEntropy (traceLeft (D.output N x)) - vnEntropy (D.output N x)) := rfl

example (D : QQData A B) (N : Superop (A × B) C) :
    D.sumCoherent N = vnEntropy (traceLeft (D.output N)) - vnEntropy (D.output N) := rfl

example (F : (n : ℕ) → Set (ℝ × ℝ)) (r : ℝ × ℝ)
    (h : r ∈ regularized F) : ∃ n : ℕ, 0 < n ∧
      ((n : ℝ) * r.1, (n : ℝ) * r.2) ∈ F n := h

end QAlgorithms.Papers.Independent.Support50.P02
Source
Capacity Theorems for Quantum Multiple Access Channels: Classical-Quantum and Quantum-Quantum Capacity Regions, arXiv:quant-ph/0501045v3, https://arxiv.org/abs/quant-ph/0501045v3
Read-back

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

QAlgorithms.Papers.Independent.Support50.P02.CQGeneration.Good

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,m,kn,m,kn,m,k, a generation datum KKK consists of unit vectors as∈CAna_s\in\mathbb C^{A^n}as​∈CAn for every s∈Fin ms\in\mathrm{Fin}\,ms∈Finm, a unit vector b∈CFin k×Bnb\in\mathbb C^{\mathrm{Fin}\,k\times B^n}b∈CFink×Bn, and a decoder D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink) admitting a finite Kraus representation D(X)=∑hLhXLh†D(X)=\sum_h L_hXL_h^\daggerD(X)=∑h​Lh​XLh†​ with ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I. For any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ, the predicate asserts

1−ϵ≤1m∑s∈Fin mF(Ts,Os).1-\epsilon\le \frac1m\sum_{s\in\mathrm{Fin}\,m}\mathcal F(T_s,O_s).1−ϵ≤m1​s∈Finm∑​F(Ts​,Os​).

Here TsT_sTs​ has entries

(Ts)(r,(j,ℓ)),(r′,(j′,ℓ′))=1j=s1j′=s1r=ℓ1r′=ℓ′k,(T_s)_{(r,(j,\ell)),(r',(j',\ell'))} =\mathbf1_{j=s}\mathbf1_{j'=s}\frac{\mathbf1_{r=\ell}\mathbf1_{r'=\ell'}}{k},(Ts​)(r,(j,ℓ)),(r′,(j′,ℓ′))​=1j=s​1j′=s​k1r=ℓ​1r′=ℓ′​​,

and OsO_sOs​ has entries 1j=j′D(Hsr,r′)(j,ℓ),(j′,ℓ′)\mathbf1_{j=j'}D(H_s^{r,r'})_{(j,\ell),(j',\ell')}1j=j′​D(Hsr,r′​)(j,ℓ),(j′,ℓ′)​, where

(Hsr,r′)c,c′=∑x,y∈(A×B)n(as)xA(as)yA‾ b(r,xB)b(r′,yB)‾∏iN(Exi,yi)ci,ci′.(H_s^{r,r'})_{c,c'}= \sum_{x,y\in(A\times B)^n} (a_s)_{x_A}\overline{(a_s)_{y_A}}\, b_{(r,x_B)}\overline{b_{(r',y_B)}} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.(Hsr,r′​)c,c′​=x,y∈(A×B)n∑​(as​)xA​​(as​)yA​​​b(r,xB​)​b(r′,yB​)​​i∏​N(Exi​,yi​​)ci​,ci′​​.

For arbitrary square matrices, the squared fidelity used here is

F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2,S(X)=cfc⁡(⋅,X).\mathcal F(X,Y)= \left(\operatorname{Re}\operatorname{tr} \mathcal S\bigl(\mathcal S(X)Y\mathcal S(X)\bigr)\right)^2, \qquad \mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X).F(X,Y)=(RetrS(S(X)YS(X)))2,S(X)=cfc(⋅​,X).

These are the total functional-calculus expressions even when an argument is not positive. No channel assumption on NNN or restriction on ϵ\epsilonϵ is included. The definition permits n=0n=0n=0, when the product does not use NNN. If m=0m=0m=0, the inequality is 1−ϵ≤01-\epsilon\le01−ϵ≤0, because the sum and the totalized inverse of zero are both zero; k=0k=0k=0 prevents such a KKK from existing because its vector bbb could not have norm 111.

QAlgorithms.Papers.Independent.Support50.P02.CQGeneration

For any finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, and natural numbers n,m,kn,m,kn,m,k, this structure consists of a vector αs∈CAn\alpha_s\in\mathbb C^{A^n}αs​∈CAn of norm 111 for every s∈{0,…,m−1}s\in\{0,\ldots,m-1\}s∈{0,…,m−1}, a vector β∈C{0,…,k−1}×Bn\beta\in\mathbb C^{\{0,\ldots,k-1\}\times B^n}β∈C{0,…,k−1}×Bn of norm 111, and a function D\mathcal DD from complex matrices on CnC^nCn to complex matrices on {0,…,m−1}×{0,…,k−1}\{0,\ldots,m-1\}\times\{0,\ldots,k-1\}{0,…,m−1}×{0,…,k−1} that admits a finite Kraus representation: there exist a natural number qqq and matrices EjE_jEj​, with rows indexed by the latter product and columns indexed by CnC^nCn, such that

∑j<qEj†Ej=I,D(X)=∑j<qEjXEj†for every complex matrix X on Cn.\sum_{j<q}E_j^\dagger E_j=I, \qquad \mathcal D(X)=\sum_{j<q}E_jXE_j^\dagger \quad\text{for every complex matrix }X\text{ on }C^n.j<q∑​Ej†​Ej​=I,D(X)=j<q∑​Ej​XEj†​for every complex matrix X on Cn.

Here An,Bn,CnA^n,B^n,C^nAn,Bn,Cn mean functions from {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1} to the respective types. No explicit positivity assumptions on n,m,kn,m,kn,m,k, or nonemptiness assumptions on A,B,CA,B,CA,B,C, are included; at n=0n=0n=0 each function type is a singleton, at m=0m=0m=0 the family of Alice vectors is empty, and at k=0k=0k=0 the required norm-one Bob vector cannot exist.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.Rectangle

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, let NNN be a function from complex matrices indexed by A×BA\times BA×B to complex matrices indexed by CCC; no channel assumption on NNN is imposed. The data DDD consist of a finite ensemble (px,αx)x<K(p_x,\alpha_x)_{x<K}(px​,αx​)x<K​, where px≥0p_x\ge0px​≥0, ∑xpx=1\sum_xp_x=1∑x​px​=1, and each αx∈CA\alpha_x\in\mathbb C^Aαx​∈CA has norm 111, together with an integer k>0k>0k>0 and a norm-one vector β∈C{0,…,k−1}×B\beta\in\mathbb C^{\{0,\ldots,k-1\}\times B}β∈C{0,…,k−1}×B. Define matrices ωx\omega_xωx​ by

(ωx)(u,c),(v,c′)=N ⁣([αx(a)αx(a′)‾ β(u,b)β(v,b′)‾](a,b),(a′,b′))c,c′,ωˉ=∑xpxωx.(\omega_x)_{(u,c),(v,c')} = N\!\left( \left[ \alpha_x(a)\overline{\alpha_x(a')}\, \beta(u,b)\overline{\beta(v,b')} \right]_{(a,b),(a',b')} \right)_{c,c'}, \qquad \bar\omega=\sum_xp_x\omega_x.(ωx​)(u,c),(v,c′)​=N([αx​(a)αx​(a′)​β(u,b)β(v,b′)​](a,b),(a′,b′)​)c,c′​,ωˉ=x∑​px​ωx​.

Write H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), where f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t is applied by continuous functional calculus, and let Tr⁡K\operatorname{Tr}_KTrK​ denote summation over the diagonal of the kkk-dimensional register. For r=(r1,r2)∈R2r=(r_1,r_2)\in\mathbb R^2r=(r1​,r2​)∈R2, the predicate holds exactly when

r1≥0,r2≥0,r1≤H(Tr⁡Kωˉ)−∑xpxH(Tr⁡Kωx),r2≤∑xpx[H(Tr⁡Kωx)−H(ωx)].r_1\ge0,\qquad r_2\ge0,\qquad r_1\le H(\operatorname{Tr}_K\bar\omega)-\sum_xp_xH(\operatorname{Tr}_K\omega_x), \qquad r_2\le\sum_xp_x\bigl[H(\operatorname{Tr}_K\omega_x)-H(\omega_x)\bigr].r1​≥0,r2​≥0,r1​≤H(TrK​ωˉ)−x∑​px​H(TrK​ωx​),r2​≤x∑​px​[H(TrK​ωx​)−H(ωx​)].

The ensemble normalization precludes K=0K=0K=0, but zero ensemble weights are permitted.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.averageOutput

For finite index types A,BA,BA,B, an output index type CCC, data comprising nonnegative real weights pxp_xpx​ summing to 111, unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA for x∈Fin⁡Kx\in\operatorname{Fin}Kx∈FinK, a positive integer rrr, and a unit vector β∈CFin⁡r×B\beta\in\mathbb C^{\operatorname{Fin}r\times B}β∈CFinr×B, and any matrix-valued function NNN from matrices on A×BA\times BA×B to matrices on CCC, this defines

Ω‾=∑x=0K−1pxΩx,\overline\Omega=\sum_{x=0}^{K-1}p_x\Omega_x,Ω=x=0∑K−1​px​Ωx​,

where

(Ωx)(j,c),(j′,c′)=[N(Mx,j,j′)]c,c′,(Mx,j,j′)(a,b),(a′,b′)=ϕx(a)ϕx(a′)‾ β(j,b)β(j′,b′)‾.(\Omega_x)_{(j,c),(j',c')} =\bigl[N(M^{x,j,j'})\bigr]_{c,c'}, \qquad (M^{x,j,j'})_{(a,b),(a',b')} =\phi_x(a)\overline{\phi_x(a')}\, \beta(j,b)\overline{\beta(j',b')}.(Ωx​)(j,c),(j′,c′)​=[N(Mx,j,j′)]c,c′​,(Mx,j,j′)(a,b),(a′,b′)​=ϕx​(a)ϕx​(a′)​β(j,b)β(j′,b′)​.

Real weights act as complex scalars. No linearity, positivity, or channel assumption is imposed on NNN.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.classical

For finite index types A,B,CA,B,CA,B,C, with decidable equality where required by the matrix functional calculus, take data consisting of nonnegative real weights pxp_xpx​ summing to 111, unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA indexed by Fin⁡K\operatorname{Fin}KFinK, a positive integer rrr, and a unit vector β∈CFin⁡r×B\beta\in\mathbb C^{\operatorname{Fin}r\times B}β∈CFinr×B, and let NNN be any function from matrices on A×BA\times BA×B to matrices on CCC. Define matrices Ωx\Omega_xΩx​ by

(Ωx)(j,c),(j′,c′)=[N(Mx,j,j′)]c,c′,(Mx,j,j′)(a,b),(a′,b′)=ϕx(a)ϕx(a′)‾β(j,b)β(j′,b′)‾,(\Omega_x)_{(j,c),(j',c')} =\bigl[N(M^{x,j,j'})\bigr]_{c,c'}, \qquad (M^{x,j,j'})_{(a,b),(a',b')} =\phi_x(a)\overline{\phi_x(a')}\beta(j,b)\overline{\beta(j',b')},(Ωx​)(j,c),(j′,c′)​=[N(Mx,j,j′)]c,c′​,(Mx,j,j′)(a,b),(a′,b′)​=ϕx​(a)ϕx​(a′)​β(j,b)β(j′,b′)​,

and set Ω‾=∑xpxΩx\overline\Omega=\sum_xp_x\Omega_xΩ=∑x​px​Ωx​ and T(Q)c,c′=∑j=0r−1Q(j,c),(j,c′)T(Q)_{c,c'}=\sum_{j=0}^{r-1}Q_{(j,c),(j,c')}T(Q)c,c′​=∑j=0r−1​Q(j,c),(j,c′)​. The declaration defines the real number

H(T(Ω‾))−∑xpxH(T(Ωx)),H(Q)=−Re⁡tr⁡ ⁣(cfc⁡[t↦tlog⁡2t](Q)),H(T(\overline\Omega))-\sum_xp_xH(T(\Omega_x)), \qquad H(Q)=-\operatorname{Re}\operatorname{tr}\!\left(\operatorname{cfc}[t\mapsto t\log_2t](Q)\right),H(T(Ω))−x∑​px​H(T(Ωx​)),H(Q)=−Retr(cfc[t↦tlog2​t](Q)),

where cfc⁡\operatorname{cfc}cfc is the total continuous functional calculus used in the code. No channel assumption on NNN, or density-matrix assumption on these outputs, is required; the displayed expression remains the definition in those cases.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.coherent

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD specify unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA indexed by x∈{0,…,K−1}x\in\{0,\ldots,K-1\}x∈{0,…,K−1}, nonnegative real weights pxp_xpx​ with ∑xpx=1\sum_xp_x=1∑x​px​=1, a positive integer bbb, and a unit vector β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B. In particular, K=0K=0K=0 is excluded by the weight condition. For any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, set

Ω(s,c),(s′,c′)=∑xpx[N(Xxs,s′)]c,c′,(Xxs,s′)(u,v),(u′,v′)=ϕx(u)ϕx(u′)‾ β(s,v)β(s′,v′)‾.\Omega_{(s,c),(s',c')}= \sum_xp_x\bigl[N(X_x^{s,s'})\bigr]_{c,c'}, \qquad (X_x^{s,s'})_{(u,v),(u',v')}= \phi_x(u)\overline{\phi_x(u')}\,\beta(s,v)\overline{\beta(s',v')}.Ω(s,c),(s′,c′)​=x∑​px​[N(Xxs,s′​)]c,c′​,(Xxs,s′​)(u,v),(u′,v′)​=ϕx​(u)ϕx​(u′)​β(s,v)β(s′,v′)​.

Writing (ΩC)c,c′=∑s=0b−1Ω(s,c),(s,c′)(\Omega_C)_{c,c'}=\sum_{s=0}^{b-1}\Omega_{(s,c),(s,c')}(ΩC​)c,c′​=∑s=0b−1​Ω(s,c),(s,c′)​, the definition assigns H(ΩC)−H(Ω)H(\Omega_C)-H(\Omega)H(ΩC​)−H(Ω), where H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)], f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, and f(Z)f(Z)f(Z) is evaluated by the total continuous functional calculus. No linearity, channel, or density-matrix condition on the outputs of NNN is assumed.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.conditionalCoherent

For finite index types A,B,CA,B,CA,B,C, with decidable equality where required by the matrix functional calculus, take data consisting of nonnegative real weights pxp_xpx​ summing to 111, unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA indexed by Fin⁡K\operatorname{Fin}KFinK, a positive integer rrr, and a unit vector β∈CFin⁡r×B\beta\in\mathbb C^{\operatorname{Fin}r\times B}β∈CFinr×B, and any function NNN from matrices on A×BA\times BA×B to matrices on CCC. Define

(Ωx)(j,c),(j′,c′)=[N(Mx,j,j′)]c,c′,(Mx,j,j′)(a,b),(a′,b′)=ϕx(a)ϕx(a′)‾β(j,b)β(j′,b′)‾,(\Omega_x)_{(j,c),(j',c')} =\bigl[N(M^{x,j,j'})\bigr]_{c,c'}, \qquad (M^{x,j,j'})_{(a,b),(a',b')} =\phi_x(a)\overline{\phi_x(a')}\beta(j,b)\overline{\beta(j',b')},(Ωx​)(j,c),(j′,c′)​=[N(Mx,j,j′)]c,c′​,(Mx,j,j′)(a,b),(a′,b′)​=ϕx​(a)ϕx​(a′)​β(j,b)β(j′,b′)​,

and T(Q)c,c′=∑j=0r−1Q(j,c),(j,c′)T(Q)_{c,c'}=\sum_{j=0}^{r-1}Q_{(j,c),(j,c')}T(Q)c,c′​=∑j=0r−1​Q(j,c),(j,c′)​. This declaration defines the real number

∑xpx(H(T(Ωx))−H(Ωx)),H(Q)=−Re⁡tr⁡ ⁣(cfc⁡[t↦tlog⁡2t](Q)).\sum_xp_x\bigl(H(T(\Omega_x))-H(\Omega_x)\bigr), \qquad H(Q)=-\operatorname{Re}\operatorname{tr}\!\left(\operatorname{cfc}[t\mapsto t\log_2t](Q)\right).x∑​px​(H(T(Ωx​))−H(Ωx​)),H(Q)=−Retr(cfc[t↦tlog2​t](Q)).

Here cfc⁡\operatorname{cfc}cfc is the total continuous functional calculus used in the code. There is no assumption that NNN is linear or a channel, or that the matrices to which HHH is applied are density matrices.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData

For finite index types A,BA,BA,B, this defines a structure containing an ensemble with a natural-number count KKK, real weights px≥0p_x\ge0px​≥0 summing to 111, and unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA, together with a natural number rrr, a proof that r>0r>0r>0, and a vector β∈CFin⁡r×B\beta\in\mathbb C^{\operatorname{Fin}r\times B}β∈CFinr×B satisfying ∥β∥=1\|\beta\|=1∥β∥=1. The weight condition forces K>0K>0K>0. The normalization requirements make this structure uninhabited if either AAA or BBB is empty.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.output

For finite index types A,BA,BA,B, an output index type CCC, data DDD consisting of weights py≥0p_y\ge0py​≥0 summing to 111, unit vectors ϕy∈CA\phi_y\in\mathbb C^Aϕy​∈CA indexed by Fin⁡K\operatorname{Fin}KFinK, a positive integer rrr, and a unit vector β∈CFin⁡r×B\beta\in\mathbb C^{\operatorname{Fin}r\times B}β∈CFinr×B, let NNN be any function from matrices on A×BA\times BA×B to matrices on CCC, and choose x∈Fin⁡Kx\in\operatorname{Fin}Kx∈FinK. This defines a matrix Ωx\Omega_xΩx​ on Fin⁡r×C\operatorname{Fin}r\times CFinr×C by

(Ωx)(j,c),(j′,c′)=[N(Mx,j,j′)]c,c′,(Mx,j,j′)(a,b),(a′,b′)=ϕx(a)ϕx(a′)‾ β(j,b)β(j′,b′)‾.(\Omega_x)_{(j,c),(j',c')} =\bigl[N(M^{x,j,j'})\bigr]_{c,c'}, \qquad (M^{x,j,j'})_{(a,b),(a',b')} =\phi_x(a)\overline{\phi_x(a')}\, \beta(j,b)\overline{\beta(j',b')}.(Ωx​)(j,c),(j′,c′)​=[N(Mx,j,j′)]c,c′​,(Mx,j,j′)(a,b),(a′,b′)​=ϕx​(a)ϕx​(a′)​β(j,b)β(j′,b′)​.

The weights do not enter this formula. No linearity or channel condition is imposed on NNN, and the definition imposes no separate density-matrix condition on its output.

QAlgorithms.Papers.Independent.Support50.P02.CQMacData.sideClassical

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD consist of an ensemble of KKK unit vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA, nonnegative real weights pxp_xpx​ satisfying ∑xpx=1\sum_xp_x=1∑x​px​=1, and a unit vector β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B for a positive integer bbb. The weight condition excludes K=0K=0K=0. Let NNN be any function from complex matrices on A×BA\times BA×B to complex matrices on CCC; no channel assumption is imposed. Define matrices Ωx\Omega_xΩx​ on {0,…,b−1}×C\{0,\ldots,b-1\}\times C{0,…,b−1}×C by

(Ωx)(s,c),(s′,c′)=[N(Xxs,s′)]c,c′,(Xxs,s′)(u,v),(u′,v′)=ϕx(u)ϕx(u′)‾ β(s,v)β(s′,v′)‾,(\Omega_x)_{(s,c),(s',c')}= \bigl[N(X_x^{s,s'})\bigr]_{c,c'}, \qquad (X_x^{s,s'})_{(u,v),(u',v')}= \phi_x(u)\overline{\phi_x(u')}\,\beta(s,v)\overline{\beta(s',v')},(Ωx​)(s,c),(s′,c′)​=[N(Xxs,s′​)]c,c′​,(Xxs,s′​)(u,v),(u′,v′)​=ϕx​(u)ϕx​(u′)​β(s,v)β(s′,v′)​,

and put Ω‾=∑xpxΩx\overline\Omega=\sum_xp_x\Omega_xΩ=∑x​px​Ωx​. The definition assigns the real number

H(Ω‾)−∑xpxH(Ωx),H(Z)=−Re⁡tr⁡ ⁣[f(Z)],f(t)=tlog⁡2t,H(\overline\Omega)-\sum_xp_xH(\Omega_x), \qquad H(Z)=-\operatorname{Re}\operatorname{tr}\!\left[f(Z)\right], \quad f(t)=t\log_2t,H(Ω)−x∑​px​H(Ωx​),H(Z)=−Retr[f(Z)],f(t)=tlog2​t,

where f(Z)f(Z)f(Z) uses the total continuous functional calculus. The matrices produced by NNN are not required to be density matrices.

QAlgorithms.Papers.Independent.Support50.P02.CQStrong.Good

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,m,kn,m,kn,m,k, and a datum KKK containing positive semidefinite trace-one matrices ρs\rho_sρs​ on AnA^nAn, an encoder E:Mat⁡(Fin k)→Mat⁡(Bn)\mathcal E:\operatorname{Mat}(\mathrm{Fin}\,k)\to\operatorname{Mat}(B^n)E:Mat(Fink)→Mat(Bn), and a decoder D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink), where both maps admit finite Kraus representations Φ(X)=∑hLhXLh†\Phi(X)=\sum_hL_hXL_h^\daggerΦ(X)=∑h​Lh​XLh†​ with ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I, this predicate says the following for any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ: for every s∈Fin ms\in\mathrm{Fin}\,ms∈Finm, every natural number t>0t>0t>0, and every unit vector ψ∈CFin t×Fin k\psi\in\mathbb C^{\mathrm{Fin}\,t\times\mathrm{Fin}\,k}ψ∈CFint×Fink,

1−ϵ≤F(Ts,Os).1-\epsilon\le\mathcal F(T_s,O_s).1−ϵ≤F(Ts​,Os​).

The encoded reference–input matrix has entries

τ(r,β),(r′,β′)=E ⁣([ψ(r,u)ψ(r′,v)‾]u,v)β,β′,\tau_{(r,\beta),(r',\beta')}= \mathcal E\!\left([\psi_{(r,u)}\overline{\psi_{(r',v)}}]_{u,v}\right)_{\beta,\beta'},τ(r,β),(r′,β′)​=E([ψ(r,u)​ψ(r′,v)​​]u,v​)β,β′​,

the target has entries (Ts)(r,(j,ℓ)),(r′,(j′,ℓ′))=1j=s1j′=sψ(r,ℓ)ψ(r′,ℓ′)‾(T_s)_{(r,(j,\ell)),(r',(j',\ell'))}=\mathbf1_{j=s}\mathbf1_{j'=s}\psi_{(r,\ell)}\overline{\psi_{(r',\ell')}}(Ts​)(r,(j,ℓ)),(r′,(j′,ℓ′))​=1j=s​1j′=s​ψ(r,ℓ)​ψ(r′,ℓ′)​​, and the output has entries

(Os)(r,(j,ℓ)),(r′,(j′,ℓ′))=1j=j′D(Hsr,r′)(j,ℓ),(j′,ℓ′),(Hsr,r′)c,c′=∑x,y∈(A×B)n(ρs)xA,yAτ(r,xB),(r′,yB)∏iN(Exi,yi)ci,ci′.(O_s)_{(r,(j,\ell)),(r',(j',\ell'))} =\mathbf1_{j=j'}D(H_s^{r,r'})_{(j,\ell),(j',\ell')}, \quad (H_s^{r,r'})_{c,c'}= \sum_{x,y\in(A\times B)^n} (\rho_s)_{x_A,y_A}\tau_{(r,x_B),(r',y_B)} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.(Os​)(r,(j,ℓ)),(r′,(j′,ℓ′))​=1j=j′​D(Hsr,r′​)(j,ℓ),(j′,ℓ′)​,(Hsr,r′​)c,c′​=x,y∈(A×B)n∑​(ρs​)xA​,yA​​τ(r,xB​),(r′,yB​)​i∏​N(Exi​,yi​​)ci​,ci′​​.

Here Ex,yE_{x,y}Ex,y​ is a matrix unit and F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2\mathcal F(X,Y)=(\operatorname{Re}\operatorname{tr}\mathcal S(\mathcal S(X)Y\mathcal S(X)))^2F(X,Y)=(RetrS(S(X)YS(X)))2, where S(X)=cfc⁡(⋅,X)\mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X)S(X)=cfc(⋅​,X). There is no channel assumption on NNN or restriction on ϵ\epsilonϵ. If m=0m=0m=0, there are no labels sss; if k=0k=0k=0, there are no such unit vectors ψ\psiψ, so in either case the predicate is vacuous whenever KKK exists. At n=0n=0n=0, the empty products equal 111 and NNN is not evaluated.

QAlgorithms.Papers.Independent.Support50.P02.CQStrong

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, and arbitrary natural numbers n,m,kn,m,kn,m,k, this structure consists of complex matrices ρs\rho_sρs​ on AnA^nAn, one for each s∈Fin ms\in\mathrm{Fin}\,ms∈Finm, each positive semidefinite with complex trace exactly 111; an encoder E:Mat⁡(Fin k)→Mat⁡(Bn)\mathcal E:\operatorname{Mat}(\mathrm{Fin}\,k)\to\operatorname{Mat}(B^n)E:Mat(Fink)→Mat(Bn); and a decoder D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)\mathcal D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink). Each map must have, for some natural number qqq, a finite Kraus representation

Φ(X)=∑h∈Fin qLhXLh†for every X,∑hLh†Lh=I.\Phi(X)=\sum_{h\in\mathrm{Fin}\,q}L_hXL_h^\dagger \quad\text{for every }X, \qquad \sum_hL_h^\dagger L_h=I.Φ(X)=h∈Finq∑​Lh​XLh†​for every X,h∑​Lh†​Lh​=I.

No communication map or success requirement is part of the structure. The parameters n,m,kn,m,kn,m,k need not be positive: m=0m=0m=0 makes the conditions on ρs\rho_sρs​ vacuous, but a density matrix on an empty index set cannot have trace 111, and a channel from a nonzero-dimensional space to a zero-dimensional space cannot satisfy the stated Kraus identity.

QAlgorithms.Papers.Independent.Support50.P02.CQTransmission.Good

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,m,kn,m,kn,m,k, and a transmission datum KKK containing unit vectors as∈CAna_s\in\mathbb C^{A^n}as​∈CAn, an encoder E:Mat⁡(Fin k)→Mat⁡(Bn)\mathcal E:\operatorname{Mat}(\mathrm{Fin}\,k)\to\operatorname{Mat}(B^n)E:Mat(Fink)→Mat(Bn), and a decoder D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink), where both maps have finite Kraus representations Φ(X)=∑hLhXLh†\Phi(X)=\sum_hL_hXL_h^\daggerΦ(X)=∑h​Lh​XLh†​ satisfying ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I, this predicate is defined for any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ by

1−ϵ≤1m∑s∈Fin mF(Ts,Os).1-\epsilon\le\frac1m\sum_{s\in\mathrm{Fin}\,m}\mathcal F(T_s,O_s).1−ϵ≤m1​s∈Finm∑​F(Ts​,Os​).

Define Ωk\Omega_kΩk​ on Fin k×Fin k\mathrm{Fin}\,k\times\mathrm{Fin}\,kFink×Fink by (Ωk)(r,u),(r′,v)=k−11r=u1r′=v(\Omega_k)_{(r,u),(r',v)}=k^{-1}\mathbf1_{r=u}\mathbf1_{r'=v}(Ωk​)(r,u),(r′,v)​=k−11r=u​1r′=v​, and define the encoder output τ\tauτ blockwise by

τ(r,β),(r′,β′)=E ⁣([(Ωk)(r,u),(r′,v)]u,v)β,β′.\tau_{(r,\beta),(r',\beta')}= \mathcal E\!\left([(\Omega_k)_{(r,u),(r',v)}]_{u,v}\right)_{\beta,\beta'}.τ(r,β),(r′,β′)​=E([(Ωk​)(r,u),(r′,v)​]u,v​)β,β′​.

The target is (Ts)(r,(j,ℓ)),(r′,(j′,ℓ′))=1j=s1j′=s(Ωk)(r,ℓ),(r′,ℓ′)(T_s)_{(r,(j,\ell)),(r',(j',\ell'))}=\mathbf1_{j=s}\mathbf1_{j'=s}(\Omega_k)_{(r,\ell),(r',\ell')}(Ts​)(r,(j,ℓ)),(r′,(j′,ℓ′))​=1j=s​1j′=s​(Ωk​)(r,ℓ),(r′,ℓ′)​, and the actual output is

(Os)(r,(j,ℓ)),(r′,(j′,ℓ′))=1j=j′D(Hsr,r′)(j,ℓ),(j′,ℓ′),(Hsr,r′)c,c′=∑x,y∈(A×B)n(as)xA(as)yA‾ τ(r,xB),(r′,yB)∏iN(Exi,yi)ci,ci′.(O_s)_{(r,(j,\ell)),(r',(j',\ell'))} =\mathbf1_{j=j'}D(H_s^{r,r'})_{(j,\ell),(j',\ell')}, \quad (H_s^{r,r'})_{c,c'}= \sum_{x,y\in(A\times B)^n} (a_s)_{x_A}\overline{(a_s)_{y_A}}\, \tau_{(r,x_B),(r',y_B)} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.(Os​)(r,(j,ℓ)),(r′,(j′,ℓ′))​=1j=j′​D(Hsr,r′​)(j,ℓ),(j′,ℓ′)​,(Hsr,r′​)c,c′​=x,y∈(A×B)n∑​(as​)xA​​(as​)yA​​​τ(r,xB​),(r′,yB​)​i∏​N(Exi​,yi​​)ci​,ci′​​.

Here F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2\mathcal F(X,Y)=(\operatorname{Re}\operatorname{tr}\mathcal S(\mathcal S(X)Y\mathcal S(X)))^2F(X,Y)=(RetrS(S(X)YS(X)))2, with S(X)=cfc⁡(⋅,X)\mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X)S(X)=cfc(⋅​,X), and Ex,yE_{x,y}Ex,y​ denotes a matrix unit. No channel condition on NNN or range restriction on ϵ\epsilonϵ is imposed. Zero parameters are included wherever KKK exists: n=0n=0n=0 uses empty products and does not evaluate NNN, while m=0m=0m=0 or k=0k=0k=0 makes the displayed fidelity average zero, so the condition becomes ϵ≥1\epsilon\ge1ϵ≥1.

QAlgorithms.Papers.Independent.Support50.P02.CQTransmission

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, and arbitrary natural numbers n,m,kn,m,kn,m,k, this structure consists of a family as∈CAna_s\in\mathbb C^{A^n}as​∈CAn, indexed by s∈Fin ms\in\mathrm{Fin}\,ms∈Finm, satisfying ∥as∥=1\|a_s\|=1∥as​∥=1 for every sss; an encoder E:Mat⁡(Fin k)→Mat⁡(Bn)\mathcal E:\operatorname{Mat}(\mathrm{Fin}\,k)\to\operatorname{Mat}(B^n)E:Mat(Fink)→Mat(Bn); and a decoder D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)\mathcal D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink). Each of the two maps is required to admit a finite Kraus representation

Φ(X)=∑h∈Fin qLhXLh†,∑hLh†Lh=I\Phi(X)=\sum_{h\in\mathrm{Fin}\,q}L_hXL_h^\dagger, \qquad \sum_hL_h^\dagger L_h=IΦ(X)=h∈Finq∑​Lh​XLh†​,h∑​Lh†​Lh​=I

for some natural number qqq, with appropriately sized complex matrices LhL_hLh​, and the representation must hold for every input matrix. There is no communication map or success condition among the fields. All three natural-number parameters may be zero; when m=0m=0m=0, the vector conditions are vacuous, while a required unit vector on an empty index set, or a channel from a nonzero-dimensional space into a zero-dimensional space, makes the corresponding structure impossible to instantiate.

QAlgorithms.Papers.Independent.Support50.P02.MACScenario

This defines a type with exactly three distinct possibilities, named generation, transmission, and strong. None carries any additional data.

QAlgorithms.Papers.Independent.Support50.P02.Nonnegative

For every ordered pair of real numbers r=(r1,r2)r=(r_1,r_2)r=(r1​,r2​), this predicate means exactly

0≤r1and0≤r2.0\le r_1\quad\text{and}\quad 0\le r_2.0≤r1​and0≤r2​.

Either coordinate, or both coordinates, may equal zero.

QAlgorithms.Papers.Independent.Support50.P02.QQData.Pentagon

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, let NNN be an arbitrary function from complex matrices on A×BA\times BA×B to complex matrices on CCC. The data DDD consist of positive integers a,ba,ba,b and norm-one vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A and β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B. Define

ω((u,v),c),((u′,v′),c′)=N ⁣([α(u,i)α(u′,i′)‾ β(v,j)β(v′,j′)‾](i,j),(i′,j′))c,c′.\omega_{((u,v),c),((u',v'),c')} = N\!\left( \left[ \alpha(u,i)\overline{\alpha(u',i')}\, \beta(v,j)\overline{\beta(v',j')} \right]_{(i,j),(i',j')} \right)_{c,c'}.ω((u,v),c),((u′,v′),c′)​=N([α(u,i)α(u′,i′)​β(v,j)β(v′,j′)​](i,j),(i′,j′)​)c,c′​.

Let ωBC=Tr⁡aω\omega_{BC}=\operatorname{Tr}_{a}\omegaωBC​=Tra​ω, ωAC=Tr⁡bω\omega_{AC}=\operatorname{Tr}_{b}\omegaωAC​=Trb​ω, and ωC=Tr⁡a,bω\omega_C=\operatorname{Tr}_{a,b}\omegaωC​=Tra,b​ω, where these traces sum over the corresponding register's diagonal entries. With H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), where f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t is applied by continuous functional calculus, the predicate for r=(r1,r2)∈R2r=(r_1,r_2)\in\mathbb R^2r=(r1​,r2​)∈R2 means exactly

r1≥0,r2≥0,r1≤H(ωBC)−H(ω),r2≤H(ωAC)−H(ω),r1+r2≤H(ωC)−H(ω).r_1\ge0,\qquad r_2\ge0,\qquad r_1\le H(\omega_{BC})-H(\omega),\qquad r_2\le H(\omega_{AC})-H(\omega),\qquad r_1+r_2\le H(\omega_C)-H(\omega).r1​≥0,r2​≥0,r1​≤H(ωBC​)−H(ω),r2​≤H(ωAC​)−H(ω),r1​+r2​≤H(ωC​)−H(ω).

No channel assumption on NNN is included.

QAlgorithms.Papers.Independent.Support50.P02.QQData.aliceCoherent

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD contain positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B, and let NNN be any function from complex matrices on A×BA\times BA×B to complex matrices on CCC. Form

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾β(s,v)β(s′,v′)‾.\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \quad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\beta(s,v)\overline{\beta(s',v')}.Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​.

Set (ΩAlice)(r,c),(r′,c′)=∑sΩ((r,s),c),((r′,s),c′)(\Omega_{\mathrm{Alice}})_{(r,c),(r',c')}=\sum_s\Omega_{((r,s),c),((r',s),c')}(ΩAlice​)(r,c),(r′,c′)​=∑s​Ω((r,s),c),((r′,s),c′)​ and (ΩC)c,c′=∑r∑sΩ((r,s),c),((r,s),c′)(\Omega_C)_{c,c'}=\sum_r\sum_s\Omega_{((r,s),c),((r,s),c')}(ΩC​)c,c′​=∑r​∑s​Ω((r,s),c),((r,s),c′)​. The defined value is H(ΩC)−H(ΩAlice)H(\Omega_C)-H(\Omega_{\mathrm{Alice}})H(ΩC​)−H(ΩAlice​), with H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)], f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, interpreted through the total continuous functional calculus. No channel or output-density condition is required of NNN.

QAlgorithms.Papers.Independent.Support50.P02.QQData.aliceConditional

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD supply positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B, and let NNN be any function from complex matrices on A×BA\times BA×B to complex matrices on CCC. Define

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾β(s,v)β(s′,v′)‾,\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \quad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\beta(s,v)\overline{\beta(s',v')},Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​,

and (ΩBob)(s,c),(s′,c′)=∑rΩ((r,s),c),((r,s′),c′)(\Omega_{\mathrm{Bob}})_{(s,c),(s',c')}=\sum_r\Omega_{((r,s),c),((r,s'),c')}(ΩBob​)(s,c),(s′,c′)​=∑r​Ω((r,s),c),((r,s′),c′)​. The definition assigns H(ΩBob)−H(Ω)H(\Omega_{\mathrm{Bob}})-H(\Omega)H(ΩBob​)−H(Ω), where H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)], f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, using the total continuous functional calculus. No channel assumption on NNN or density-matrix assumption on these output matrices is imposed.

QAlgorithms.Papers.Independent.Support50.P02.QQData.aliceOutput

For finite types A,BA,BA,B, an arbitrary type CCC, data DDD supplying positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B, and any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this definition returns the matrix on {0,…,a−1}×C\{0,\ldots,a-1\}\times C{0,…,a−1}×C with entries

(ΩAlice)(r,c),(r′,c′)=∑s=0b−1[N(Xr,s;r′,s)]c,c′,X(u,v),(u′,v′)r,s;r′,s=α(r,u)α(r′,u′)‾ β(s,v)β(s,v′)‾.(\Omega_{\mathrm{Alice}})_{(r,c),(r',c')}= \sum_{s=0}^{b-1}\bigl[N(X^{r,s;r',s})\bigr]_{c,c'}, \qquad X^{r,s;r',s}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\, \beta(s,v)\overline{\beta(s,v')}.(ΩAlice​)(r,c),(r′,c′)​=s=0∑b−1​[N(Xr,s;r′,s)]c,c′​,X(u,v),(u′,v′)r,s;r′,s​=α(r,u)α(r′,u′)​β(s,v)β(s,v′)​.

The sum is taken after applying NNN to each block. The definition assumes no linearity or channel property of NNN and does not require its result to be a density matrix.

QAlgorithms.Papers.Independent.Support50.P02.QQData.bobCoherent

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD supply positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B. For any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, define

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾β(s,v)β(s′,v′)‾.\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \quad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\beta(s,v)\overline{\beta(s',v')}.Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​.

Let (ΩBob)(s,c),(s′,c′)=∑rΩ((r,s),c),((r,s′),c′)(\Omega_{\mathrm{Bob}})_{(s,c),(s',c')}=\sum_r\Omega_{((r,s),c),((r,s'),c')}(ΩBob​)(s,c),(s′,c′)​=∑r​Ω((r,s),c),((r,s′),c′)​ and (ΩC)c,c′=∑r∑sΩ((r,s),c),((r,s),c′)(\Omega_C)_{c,c'}=\sum_r\sum_s\Omega_{((r,s),c),((r,s),c')}(ΩC​)c,c′​=∑r​∑s​Ω((r,s),c),((r,s),c′)​. This definition assigns H(ΩC)−H(ΩBob)H(\Omega_C)-H(\Omega_{\mathrm{Bob}})H(ΩC​)−H(ΩBob​), where H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)], f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, evaluated by the total continuous functional calculus. No linearity, channel, or output-density assumption on NNN is imposed.

QAlgorithms.Papers.Independent.Support50.P02.QQData.bobConditional

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD contain positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B. For any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, define

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾β(s,v)β(s′,v′)‾,\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \quad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\beta(s,v)\overline{\beta(s',v')},Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​,

and (ΩAlice)(r,c),(r′,c′)=∑sΩ((r,s),c),((r′,s),c′)(\Omega_{\mathrm{Alice}})_{(r,c),(r',c')}=\sum_s\Omega_{((r,s),c),((r',s),c')}(ΩAlice​)(r,c),(r′,c′)​=∑s​Ω((r,s),c),((r′,s),c′)​. The defined real number is H(ΩAlice)−H(Ω)H(\Omega_{\mathrm{Alice}})-H(\Omega)H(ΩAlice​)−H(Ω), where H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)] for f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, evaluated by the total continuous functional calculus. Neither a channel property of NNN nor positivity or normalization of its outputs is assumed.

QAlgorithms.Papers.Independent.Support50.P02.QQData.bobOutput

For finite types A,BA,BA,B, an arbitrary type CCC, positive integers a,ba,ba,b, unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B supplied by DDD, and any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this definition returns the matrix on {0,…,b−1}×C\{0,\ldots,b-1\}\times C{0,…,b−1}×C given by

(ΩBob)(s,c),(s′,c′)=∑r=0a−1[N(Xr,s;r,s′)]c,c′,X(u,v),(u′,v′)r,s;r,s′=α(r,u)α(r,u′)‾ β(s,v)β(s′,v′)‾.(\Omega_{\mathrm{Bob}})_{(s,c),(s',c')}= \sum_{r=0}^{a-1}\bigl[N(X^{r,s;r,s'})\bigr]_{c,c'}, \qquad X^{r,s;r,s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r,u')}\, \beta(s,v)\overline{\beta(s',v')}.(ΩBob​)(s,c),(s′,c′)​=r=0∑a−1​[N(Xr,s;r,s′)]c,c′​,X(u,v),(u′,v′)r,s;r,s′​=α(r,u)α(r,u′)​β(s,v)β(s′,v′)​.

The sum is outside the application of NNN. No linearity, channel, or output-normalization assumption is imposed on NNN.

QAlgorithms.Papers.Independent.Support50.P02.QQData

For any finite types AAA and BBB, this structure consists of natural numbers a,ba,ba,b, proofs that a>0a>0a>0 and b>0b>0b>0, and vectors

α∈C{0,…,a−1}×A,β∈C{0,…,b−1}×B\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}, \qquad \beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}α∈C{0,…,a−1}×A,β∈C{0,…,b−1}×B

with their Euclidean norms equal to 111. These are the fields named aliceRank, bobRank, alice, and bob, together with their positivity and normalization proofs. The two rank parameters cannot be zero, and the normalization requirements make such data impossible if either AAA or BBB is empty.

QAlgorithms.Papers.Independent.Support50.P02.QQData.output

For finite types A,BA,BA,B, an arbitrary type CCC, data DDD containing positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B, and any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this definition returns the matrix Ω\OmegaΩ indexed by ({0,…,a−1}×{0,…,b−1})×C(\{0,\ldots,a-1\}\times\{0,\ldots,b-1\})\times C({0,…,a−1}×{0,…,b−1})×C whose entries are

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾ β(s,v)β(s′,v′)‾.\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \qquad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\, \beta(s,v)\overline{\beta(s',v')}.Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​.

Thus NNN is applied separately to each matrix block specified by the two reference indices. No linearity or channel property of NNN, and no density-matrix property of the result, is required.

QAlgorithms.Papers.Independent.Support50.P02.QQData.sumCoherent

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, let DDD specify positive integers a,ba,ba,b and unit vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B. Given any function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, set

Ω((r,s),c),((r′,s′),c′)=[N(Xr,s;r′,s′)]c,c′,X(u,v),(u′,v′)r,s;r′,s′=α(r,u)α(r′,u′)‾β(s,v)β(s′,v′)‾,\Omega_{((r,s),c),((r',s'),c')}= \bigl[N(X^{r,s;r',s'})\bigr]_{c,c'}, \quad X^{r,s;r',s'}_{(u,v),(u',v')}= \alpha(r,u)\overline{\alpha(r',u')}\beta(s,v)\overline{\beta(s',v')},Ω((r,s),c),((r′,s′),c′)​=[N(Xr,s;r′,s′)]c,c′​,X(u,v),(u′,v′)r,s;r′,s′​=α(r,u)α(r′,u′)​β(s,v)β(s′,v′)​,

and (ΩC)c,c′=∑r∑sΩ((r,s),c),((r,s),c′)(\Omega_C)_{c,c'}=\sum_r\sum_s\Omega_{((r,s),c),((r,s),c')}(ΩC​)c,c′​=∑r​∑s​Ω((r,s),c),((r,s),c′)​. This definition assigns H(ΩC)−H(Ω)H(\Omega_C)-H(\Omega)H(ΩC​)−H(Ω), where H(Z)=−Re⁡tr⁡[f(Z)]H(Z)=-\operatorname{Re}\operatorname{tr}[f(Z)]H(Z)=−Retr[f(Z)], f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t, using the total continuous functional calculus. The function NNN need not be a channel, and its output matrices need not be density matrices.

QAlgorithms.Papers.Independent.Support50.P02.QQGeneration.Good

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,a,bn,a,bn,a,b, and a generation datum KKK containing unit vectors α∈CFin a×An\alpha\in\mathbb C^{\mathrm{Fin}\,a\times A^n}α∈CFina×An, β∈CFin b×Bn\beta\in\mathbb C^{\mathrm{Fin}\,b\times B^n}β∈CFinb×Bn, and a decoder D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb) with a finite Kraus representation D(X)=∑hLhXLh†D(X)=\sum_hL_hXL_h^\daggerD(X)=∑h​Lh​XLh†​, ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I, this predicate is defined for any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ by

1−ϵ≤F(T,O).1-\epsilon\le\mathcal F(T,O).1−ϵ≤F(T,O).

For reference indices r,r′∈Fin ar,r'\in\mathrm{Fin}\,ar,r′∈Fina and s,s′∈Fin bs,s'\in\mathrm{Fin}\,bs,s′∈Finb, define

Hc,c′(r,s),(r′,s′)=∑x,y∈(A×B)nα(r,xA)α(r′,yA)‾ β(s,xB)β(s′,yB)‾∏iN(Exi,yi)ci,ci′.H^{(r,s),(r',s')}_{c,c'}= \sum_{x,y\in(A\times B)^n} \alpha_{(r,x_A)}\overline{\alpha_{(r',y_A)}}\, \beta_{(s,x_B)}\overline{\beta_{(s',y_B)}} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.Hc,c′(r,s),(r′,s′)​=x,y∈(A×B)n∑​α(r,xA​)​α(r′,yA​)​​β(s,xB​)​β(s′,yB​)​​i∏​N(Exi​,yi​​)ci​,ci′​​.

Then O((r,s),(u,v)),((r′,s′),(u′,v′))=D(H(r,s),(r′,s′))(u,v),(u′,v′)O_{((r,s),(u,v)),((r',s'),(u',v'))}=D(H^{(r,s),(r',s')})_{(u,v),(u',v')}O((r,s),(u,v)),((r′,s′),(u′,v′))​=D(H(r,s),(r′,s′))(u,v),(u′,v′)​, while the target is

T((r,s),(u,v)),((r′,s′),(u′,v′))=1r=u1r′=u′1s=v1s′=v′ab.T_{((r,s),(u,v)),((r',s'),(u',v'))} =\frac{\mathbf1_{r=u}\mathbf1_{r'=u'}\mathbf1_{s=v}\mathbf1_{s'=v'}}{ab}.T((r,s),(u,v)),((r′,s′),(u′,v′))​=ab1r=u​1r′=u′​1s=v​1s′=v′​​.

Here Ex,yE_{x,y}Ex,y​ is a matrix unit and F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2\mathcal F(X,Y)=(\operatorname{Re}\operatorname{tr}\mathcal S(\mathcal S(X)Y\mathcal S(X)))^2F(X,Y)=(RetrS(S(X)YS(X)))2, with S(X)=cfc⁡(⋅,X)\mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X)S(X)=cfc(⋅​,X). There is no dephasing of the decoded output, no channel assumption on NNN, and no restriction on ϵ\epsilonϵ. Although zero parameters are syntactically allowed, the unit-vector fields of KKK exclude a=0a=0a=0 and b=0b=0b=0; n=0n=0n=0 remains included and makes the product empty, so NNN is not evaluated.

QAlgorithms.Papers.Independent.Support50.P02.QQGeneration

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, and arbitrary natural numbers n,a,bn,a,bn,a,b, this structure consists of a unit vector α∈CFin a×An\alpha\in\mathbb C^{\mathrm{Fin}\,a\times A^n}α∈CFina×An, a unit vector β∈CFin b×Bn\beta\in\mathbb C^{\mathrm{Fin}\,b\times B^n}β∈CFinb×Bn, and a decoder D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)\mathcal D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb). The decoder must admit, for some natural number qqq, complex matrices LhL_hLh​ such that

D(X)=∑h∈Fin qLhXLh†for every X,∑hLh†Lh=I.\mathcal D(X)=\sum_{h\in\mathrm{Fin}\,q}L_hXL_h^\dagger \quad\text{for every }X, \qquad \sum_hL_h^\dagger L_h=I.D(X)=h∈Finq∑​Lh​XLh†​for every X,h∑​Lh†​Lh​=I.

No communication map or success condition is included. The parameters are not explicitly required to be positive, but a=0a=0a=0 or b=0b=0b=0 makes the corresponding required unit vector impossible; likewise, a required channel into a zero-dimensional space from a nonzero-dimensional space cannot exist. The value n=0n=0n=0 is included, with A0A^0A0 and B0B^0B0 each having one element.

QAlgorithms.Papers.Independent.Support50.P02.QQTransmission.Good

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,a,bn,a,bn,a,b, and transmission data KKK consisting of encoders EA:Mat⁡(Fin a)→Mat⁡(An)\mathcal E_A:\operatorname{Mat}(\mathrm{Fin}\,a)\to\operatorname{Mat}(A^n)EA​:Mat(Fina)→Mat(An), EB:Mat⁡(Fin b)→Mat⁡(Bn)\mathcal E_B:\operatorname{Mat}(\mathrm{Fin}\,b)\to\operatorname{Mat}(B^n)EB​:Mat(Finb)→Mat(Bn), and a decoder D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb), each admitting a finite Kraus representation Φ(X)=∑hLhXLh†\Phi(X)=\sum_hL_hXL_h^\daggerΦ(X)=∑h​Lh​XLh†​ with ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I, this predicate says, for any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ,

1−ϵ≤F(T,O).1-\epsilon\le\mathcal F(T,O).1−ϵ≤F(T,O).

For d=a,bd=a,bd=a,b, let (Ωd)(r,u),(r′,v)=d−11r=u1r′=v(\Omega_d)_{(r,u),(r',v)}=d^{-1}\mathbf1_{r=u}\mathbf1_{r'=v}(Ωd​)(r,u),(r′,v)​=d−11r=u​1r′=v​. Define

Hc,c′(r,s),(r′,s′)=∑x,y∈(A×B)nEA ⁣([(Ωa)(r,u),(r′,v)]u,v)xA,yAEB ⁣([(Ωb)(s,u),(s′,v)]u,v)xB,yB∏iN(Exi,yi)ci,ci′.H^{(r,s),(r',s')}_{c,c'}= \sum_{x,y\in(A\times B)^n} \mathcal E_A\!\left([(\Omega_a)_{(r,u),(r',v)}]_{u,v}\right)_{x_A,y_A} \mathcal E_B\!\left([(\Omega_b)_{(s,u),(s',v)}]_{u,v}\right)_{x_B,y_B} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.Hc,c′(r,s),(r′,s′)​=x,y∈(A×B)n∑​EA​([(Ωa​)(r,u),(r′,v)​]u,v​)xA​,yA​​EB​([(Ωb​)(s,u),(s′,v)​]u,v​)xB​,yB​​i∏​N(Exi​,yi​​)ci​,ci′​​.

The output is O((r,s),(u,v)),((r′,s′),(u′,v′))=D(H(r,s),(r′,s′))(u,v),(u′,v′)O_{((r,s),(u,v)),((r',s'),(u',v'))}=D(H^{(r,s),(r',s')})_{(u,v),(u',v')}O((r,s),(u,v)),((r′,s′),(u′,v′))​=D(H(r,s),(r′,s′))(u,v),(u′,v′)​, and the target is

T((r,s),(u,v)),((r′,s′),(u′,v′))=(Ωa)(r,u),(r′,u′)(Ωb)(s,v),(s′,v′).T_{((r,s),(u,v)),((r',s'),(u',v'))} =(\Omega_a)_{(r,u),(r',u')}(\Omega_b)_{(s,v),(s',v')}.T((r,s),(u,v)),((r′,s′),(u′,v′))​=(Ωa​)(r,u),(r′,u′)​(Ωb​)(s,v),(s′,v′)​.

Here F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2\mathcal F(X,Y)=(\operatorname{Re}\operatorname{tr}\mathcal S(\mathcal S(X)Y\mathcal S(X)))^2F(X,Y)=(RetrS(S(X)YS(X)))2, with S(X)=cfc⁡(⋅,X)\mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X)S(X)=cfc(⋅​,X), and Ex,yE_{x,y}Ex,y​ denotes a matrix unit. No channel assumption on NNN or restriction on ϵ\epsilonϵ is included. If a=0a=0a=0 or b=0b=0b=0, the fidelity is zero and the condition becomes ϵ≥1\epsilon\ge1ϵ≥1, whenever KKK exists. At n=0n=0n=0, the product is empty and the construction does not evaluate NNN.

QAlgorithms.Papers.Independent.Support50.P02.QQTransmission.Strong

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, natural numbers n,a,bn,a,bn,a,b, and transmission data KKK containing encoders EA:Mat⁡(Fin a)→Mat⁡(An)\mathcal E_A:\operatorname{Mat}(\mathrm{Fin}\,a)\to\operatorname{Mat}(A^n)EA​:Mat(Fina)→Mat(An), EB:Mat⁡(Fin b)→Mat⁡(Bn)\mathcal E_B:\operatorname{Mat}(\mathrm{Fin}\,b)\to\operatorname{Mat}(B^n)EB​:Mat(Finb)→Mat(Bn), and a decoder D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb), each with a finite Kraus representation Φ(X)=∑hLhXLh†\Phi(X)=\sum_hL_hXL_h^\daggerΦ(X)=∑h​Lh​XLh†​, ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I, this predicate is defined for any matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and any real ϵ\epsilonϵ. It requires that for every pair of positive natural numbers q,tq,tq,t, every unit vector ψ∈CFin q×Fin a\psi\in\mathbb C^{\mathrm{Fin}\,q\times\mathrm{Fin}\,a}ψ∈CFinq×Fina, and every unit vector ϕ∈CFin t×Fin b\phi\in\mathbb C^{\mathrm{Fin}\,t\times\mathrm{Fin}\,b}ϕ∈CFint×Finb,

1−ϵ≤F(T,O),T((r,s),(u,v)),((r′,s′),(u′,v′))=ψ(r,u)ψ(r′,u′)‾ ϕ(s,v)ϕ(s′,v′)‾.1-\epsilon\le\mathcal F(T,O), \qquad T_{((r,s),(u,v)),((r',s'),(u',v'))} =\psi_{(r,u)}\overline{\psi_{(r',u')}}\, \phi_{(s,v)}\overline{\phi_{(s',v')}}.1−ϵ≤F(T,O),T((r,s),(u,v)),((r′,s′),(u′,v′))​=ψ(r,u)​ψ(r′,u′)​​ϕ(s,v)​ϕ(s′,v′)​​.

For each pair of reference indices, form

Hc,c′(r,s),(r′,s′)=∑x,y∈(A×B)nEA ⁣([ψ(r,u)ψ(r′,v)‾]u,v)xA,yAEB ⁣([ϕ(s,u)ϕ(s′,v)‾]u,v)xB,yB∏iN(Exi,yi)ci,ci′,H^{(r,s),(r',s')}_{c,c'}= \sum_{x,y\in(A\times B)^n} \mathcal E_A\!\left([\psi_{(r,u)}\overline{\psi_{(r',v)}}]_{u,v}\right)_{x_A,y_A} \mathcal E_B\!\left([\phi_{(s,u)}\overline{\phi_{(s',v)}}]_{u,v}\right)_{x_B,y_B} \prod_iN(E_{x_i,y_i})_{c_i,c'_i},Hc,c′(r,s),(r′,s′)​=x,y∈(A×B)n∑​EA​([ψ(r,u)​ψ(r′,v)​​]u,v​)xA​,yA​​EB​([ϕ(s,u)​ϕ(s′,v)​​]u,v​)xB​,yB​​i∏​N(Exi​,yi​​)ci​,ci′​​,

and set O((r,s),(u,v)),((r′,s′),(u′,v′))=D(H(r,s),(r′,s′))(u,v),(u′,v′)O_{((r,s),(u,v)),((r',s'),(u',v'))}=D(H^{(r,s),(r',s')})_{(u,v),(u',v')}O((r,s),(u,v)),((r′,s′),(u′,v′))​=D(H(r,s),(r′,s′))(u,v),(u′,v′)​. Here Ex,yE_{x,y}Ex,y​ is a matrix unit and F(X,Y)=(Re⁡tr⁡S(S(X)YS(X)))2\mathcal F(X,Y)=(\operatorname{Re}\operatorname{tr}\mathcal S(\mathcal S(X)Y\mathcal S(X)))^2F(X,Y)=(RetrS(S(X)YS(X)))2, with S(X)=cfc⁡(⋅,X)\mathcal S(X)=\operatorname{cfc}(\sqrt{\cdot},X)S(X)=cfc(⋅​,X). Thus the quantified inputs are independent pure states of the two reference–input pairs. No channel assumption on NNN or range restriction on ϵ\epsilonϵ is imposed. If a=0a=0a=0 or b=0b=0b=0, the corresponding unit vectors do not exist and the predicate is vacuous whenever KKK exists; n=0n=0n=0 is included and uses empty products without evaluating NNN.

QAlgorithms.Papers.Independent.Support50.P02.QQTransmission.decoded

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, reference index types R,SR,SR,S, natural numbers n,a,bn,a,bn,a,b, and transmission data KKK comprising channels EA:Mat⁡(Fin a)→Mat⁡(An)\mathcal E_A:\operatorname{Mat}(\mathrm{Fin}\,a)\to\operatorname{Mat}(A^n)EA​:Mat(Fina)→Mat(An), EB:Mat⁡(Fin b)→Mat⁡(Bn)\mathcal E_B:\operatorname{Mat}(\mathrm{Fin}\,b)\to\operatorname{Mat}(B^n)EB​:Mat(Finb)→Mat(Bn), and D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb), this defines an output matrix from an arbitrary matrix-valued function N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and arbitrary complex matrices ρ\rhoρ on R×Fin aR\times\mathrm{Fin}\,aR×Fina and τ\tauτ on S×Fin bS\times\mathrm{Fin}\,bS×Finb. Each channel condition means a finite Kraus representation Φ(X)=∑hLhXLh†\Phi(X)=\sum_hL_hXL_h^\daggerΦ(X)=∑h​Lh​XLh†​ with ∑hLh†Lh=I\sum_hL_h^\dagger L_h=I∑h​Lh†​Lh​=I. Define the reference blocks ρr,r′=[ρ(r,u),(r′,v)]u,v\rho^{r,r'}=[\rho_{(r,u),(r',v)}]_{u,v}ρr,r′=[ρ(r,u),(r′,v)​]u,v​ and τs,s′=[τ(s,u),(s′,v)]u,v\tau^{s,s'}=[\tau_{(s,u),(s',v)}]_{u,v}τs,s′=[τ(s,u),(s′,v)​]u,v​, and then

Hc,c′(r,s),(r′,s′)=∑x,y∈(A×B)nEA(ρr,r′)xA,yA EB(τs,s′)xB,yB∏iN(Exi,yi)ci,ci′.H^{(r,s),(r',s')}_{c,c'}= \sum_{x,y\in(A\times B)^n} \mathcal E_A(\rho^{r,r'})_{x_A,y_A}\, \mathcal E_B(\tau^{s,s'})_{x_B,y_B} \prod_iN(E_{x_i,y_i})_{c_i,c'_i}.Hc,c′(r,s),(r′,s′)​=x,y∈(A×B)n∑​EA​(ρr,r′)xA​,yA​​EB​(τs,s′)xB​,yB​​i∏​N(Exi​,yi​​)ci​,ci′​​.

The defined output is the matrix on (R×S)×(Fin a×Fin b)(R\times S)\times(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)(R×S)×(Fina×Finb) with entries

O((r,s),(u,v)),((r′,s′),(u′,v′))=D(H(r,s),(r′,s′))(u,v),(u′,v′).O_{((r,s),(u,v)),((r',s'),(u',v'))} =D(H^{(r,s),(r',s')})_{(u,v),(u',v')}.O((r,s),(u,v)),((r′,s′),(u′,v′))​=D(H(r,s),(r′,s′))(u,v),(u′,v′)​.

No positivity or normalization is required of ρ,τ\rho,\tauρ,τ, and no channel or linearity condition is imposed on NNN. The construction includes zero natural-number parameters wherever KKK exists; at n=0n=0n=0, the empty product is 111 and NNN is not evaluated.

QAlgorithms.Papers.Independent.Support50.P02.QQTransmission

For finite types A,B,CA,B,CA,B,C, with decidable equality on CCC, and arbitrary natural numbers n,a,bn,a,bn,a,b, this structure consists of three matrix-valued maps: an Alice encoder EA:Mat⁡(Fin a)→Mat⁡(An)\mathcal E_A:\operatorname{Mat}(\mathrm{Fin}\,a)\to\operatorname{Mat}(A^n)EA​:Mat(Fina)→Mat(An), a Bob encoder EB:Mat⁡(Fin b)→Mat⁡(Bn)\mathcal E_B:\operatorname{Mat}(\mathrm{Fin}\,b)\to\operatorname{Mat}(B^n)EB​:Mat(Finb)→Mat(Bn), and a decoder D:Mat⁡(Cn)→Mat⁡(Fin a×Fin b)\mathcal D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,a\times\mathrm{Fin}\,b)D:Mat(Cn)→Mat(Fina×Finb). Each map must admit a finite Kraus representation

Φ(X)=∑h∈Fin qLhXLh†for every input matrix X,∑hLh†Lh=I\Phi(X)=\sum_{h\in\mathrm{Fin}\,q}L_hXL_h^\dagger \quad\text{for every input matrix }X, \qquad \sum_hL_h^\dagger L_h=IΦ(X)=h∈Finq∑​Lh​XLh†​for every input matrix X,h∑​Lh†​Lh​=I

for some natural number qqq, separately for each map. No communication map or success requirement is included. Zero values of n,a,bn,a,bn,a,b are allowed by the parameters, although the Kraus conditions can rule out particular choices; in particular, no such channel can map a nonzero-dimensional input space into a zero-dimensional output space. A zero-dimensional input space is not itself excluded.

QAlgorithms.Papers.Independent.Support50.P02.bellPair

For any natural numbers a,ba,ba,b, this defines a complex matrix indexed by (Fin⁡a×Fin⁡b)×(Fin⁡a×Fin⁡b)(\operatorname{Fin}a\times\operatorname{Fin}b)\times(\operatorname{Fin}a\times\operatorname{Fin}b)(Fina×Finb)×(Fina×Finb), whose entries are

[bellPair⁡(a,b)]((r,s),(u,v)), ((r′,s′),(u′,v′))={a−1b−1,r=u, r′=u′, s=v, s′=v′,0,otherwise.\bigl[\operatorname{bellPair}(a,b)\bigr]_{((r,s),(u,v)),\,((r',s'),(u',v'))} = \begin{cases} a^{-1}b^{-1},&r=u,\ r'=u',\ s=v,\ s'=v',\\ 0,&\text{otherwise}. \end{cases}[bellPair(a,b)]((r,s),(u,v)),((r′,s′),(u′,v′))​={a−1b−1,0,​r=u, r′=u′, s=v, s′=v′,otherwise.​

The reciprocals are taken in C\mathbb CC. Thus each constituent matrix has constant entry a−1a^{-1}a−1, respectively b−1b^{-1}b−1, whenever both its row and column pairs have equal components, and zero otherwise. Zero values of aaa or bbb are allowed; in those cases the resulting matrix has an empty index set.

QAlgorithms.Papers.Independent.Support50.P02.collectivePhase

For every real number ppp, this defines a function on all complex matrices indexed by Boolean pairs by

Pp(ρ)=(1−p)ρ+pZρZ,Z=diag⁡(1,−1,−1,1),\mathcal P_p(\rho)=(1-p)\rho+pZ\rho Z, \qquad Z=\operatorname{diag}(1,-1,-1,1),Pp​(ρ)=(1−p)ρ+pZρZ,Z=diag(1,−1,−1,1),

using the Boolean-pair order (false,false)(\mathrm{false},\mathrm{false})(false,false), (false,true)(\mathrm{false},\mathrm{true})(false,true), (true,false)(\mathrm{true},\mathrm{false})(true,false), (true,true)(\mathrm{true},\mathrm{true})(true,true). The definition imposes no positivity or normalization condition on ρ\rhoρ, and no restriction 0≤p≤10\le p\le10≤p≤1. In particular, p=0p=0p=0 gives the identity map and p=1p=1p=1 gives conjugation by ZZZ.

QAlgorithms.Papers.Independent.Support50.P02.collectivePhaseOperator

This defines the complex matrix ZZZ indexed by pairs of Booleans to be diagonal, with

Z(a,b),(a,b)={−1a=true1a=false{−1b=true1b=false.Z_{(a,b),(a,b)} =\begin{cases}-1& a=\mathrm{true}\\1&a=\mathrm{false}\end{cases} \begin{cases}-1& b=\mathrm{true}\\1&b=\mathrm{false}\end{cases}.Z(a,b),(a,b)​={−11​a=truea=false​{−11​b=trueb=false​.

Thus its diagonal entries in the order (false,false)(\mathrm{false},\mathrm{false})(false,false), (false,true)(\mathrm{false},\mathrm{true})(false,true), (true,false)(\mathrm{true},\mathrm{false})(true,false), (true,true)(\mathrm{true},\mathrm{true})(true,true) are 1,−1,−1,11,-1,-1,11,−1,−1,1, and all off-diagonal entries are zero.

QAlgorithms.Papers.Independent.Support50.P02.controlledErasure

For every natural number ddd, this defines a function from complex matrices indexed by {false,true}×{0,…,d−1}\{\mathrm{false},\mathrm{true}\}\times\{0,\ldots,d-1\}{false,true}×{0,…,d−1} to complex matrices indexed by {⊥}∪{0,…,d−1}\{\bot\}\cup\{0,\ldots,d-1\}{⊥}∪{0,…,d−1}, where ⊥\bot⊥ denotes the extra “none” index. Its value on an arbitrary matrix ρ\rhoρ is

Ed(ρ)⊥,⊥=∑a=0d−1ρ(false,a),(false,a),Ed(ρ)a,b=ρ(true,a),(true,b),\mathcal E_d(\rho)_{\bot,\bot}=\sum_{a=0}^{d-1}\rho_{(\mathrm{false},a),(\mathrm{false},a)},\qquad \mathcal E_d(\rho)_{a,b}=\rho_{(\mathrm{true},a),(\mathrm{true},b)},Ed​(ρ)⊥,⊥​=a=0∑d−1​ρ(false,a),(false,a)​,Ed​(ρ)a,b​=ρ(true,a),(true,b)​,

and every entry with exactly one index equal to ⊥\bot⊥ is zero. No positivity or trace condition on ρ\rhoρ is required. The definition includes d=0d=0d=0; then the input index set is empty and the output is the zero matrix on the singleton index set {⊥}\{\bot\}{⊥}.

QAlgorithms.Papers.Independent.Support50.P02.cqDecoded

For finite index types A,B,CA,B,CA,B,C, with decidable equality on CCC, a reference index type RRR, and arbitrary natural numbers n,m,kn,m,kn,m,k, this defines a matrix on R×(Fin m×Fin k)R\times(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)R×(Finm×Fink) from arbitrary matrix-valued functions N:Mat⁡(A×B)→Mat⁡(C)N:\operatorname{Mat}(A\times B)\to\operatorname{Mat}(C)N:Mat(A×B)→Mat(C) and D:Mat⁡(Cn)→Mat⁡(Fin m×Fin k)D:\operatorname{Mat}(C^n)\to\operatorname{Mat}(\mathrm{Fin}\,m\times\mathrm{Fin}\,k)D:Mat(Cn)→Mat(Finm×Fink), a complex matrix ρ\rhoρ on AnA^nAn, and a complex matrix τ\tauτ on R×BnR\times B^nR×Bn, where AnA^nAn means functions Fin n→A\mathrm{Fin}\,n\to AFinn→A. Writing Ex,yE_{x,y}Ex,y​ for a matrix unit, define the matrix Hr,r′H^{r,r'}Hr,r′ on CnC^nCn by

Hc,c′r,r′=∑x,y∈(A×B)nρxA,yA τ(r,xB),(r′,yB)∏i∈Fin nN(Exi,yi)ci,ci′.H^{r,r'}_{c,c'}= \sum_{x,y\in(A\times B)^n} \rho_{x_A,y_A}\, \tau_{(r,x_B),(r',y_B)} \prod_{i\in\mathrm{Fin}\,n} N(E_{x_i,y_i})_{c_i,c'_i}.Hc,c′r,r′​=x,y∈(A×B)n∑​ρxA​,yA​​τ(r,xB​),(r′,yB​)​i∈Finn∏​N(Exi​,yi​​)ci​,ci′​​.

The defined output has entries

(cqDecoded⁡(N,D,ρ,τ))(r,(j,ℓ)),(r′,(j′,ℓ′))={D(Hr,r′)(j,ℓ),(j′,ℓ′),j=j′,0,j≠j′.\bigl(\operatorname{cqDecoded}(N,D,\rho,\tau)\bigr)_{(r,(j,\ell)),(r',(j',\ell'))} = \begin{cases} D(H^{r,r'})_{(j,\ell),(j',\ell')},&j=j',\\ 0,&j\ne j'. \end{cases}(cqDecoded(N,D,ρ,τ))(r,(j,ℓ)),(r′,(j′,ℓ′))​={D(Hr,r′)(j,ℓ),(j′,ℓ′)​,0,​j=j′,j=j′.​

Thus only entries between different classical labels j,j′j,j'j,j′ are removed. No linearity, channel, positivity, normalization, or error assumption is imposed on these arguments. Zero values of n,m,kn,m,kn,m,k are included: at n=0n=0n=0, the products are empty products equal to 111, so the construction does not evaluate NNN; if m=0m=0m=0 or k=0k=0k=0, the output index set is empty.

QAlgorithms.Papers.Independent.Support50.P02.cqInput

For index types A,R,BA,R,BA,R,B, an arbitrary complex matrix ρ\rhoρ on AAA, and an arbitrary complex matrix τ\tauτ on R×BR\times BR×B, this defines a matrix on R×(A×B)R\times(A\times B)R×(A×B) by

[cqInput⁡(ρ,τ)](r,(a,b)), (r′,(a′,b′))=ρa,a′ τ(r,b),(r′,b′).\bigl[\operatorname{cqInput}(\rho,\tau)\bigr]_{(r,(a,b)),\,(r',(a',b'))} =\rho_{a,a'}\,\tau_{(r,b),(r',b')}.[cqInput(ρ,τ)](r,(a,b)),(r′,(a′,b′))​=ρa,a′​τ(r,b),(r′,b′)​.

There are no positivity, trace, or normalization assumptions, and empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P02.cqMacCapacity

For finite types A,B,CA,B,CA,B,C, each with decidable equality, an arbitrary map NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, and a scenario sss defaulting to generation, this defines the closure of the nonnegative pairs (r1,r2)(r_1,r_2)(r1​,r2​) satisfying

∀ϵ,δ>0 ∃n0>0 ∀n≥n0 ∃m,k>0:r1−δ≤log⁡2mn,r2−δ≤log⁡2kn,\forall\epsilon,\delta>0\ \exists n_0>0\ \forall n\ge n_0\ \exists m,k>0:\quad r_1-\delta\le\frac{\log_2 m}{n},\qquad r_2-\delta\le\frac{\log_2 k}{n},∀ϵ,δ>0 ∃n0​>0 ∀n≥n0​ ∃m,k>0:r1​−δ≤nlog2​m​,r2​−δ≤nlog2​k​,

with a code of the selected kind satisfying the condition below. Write AnA^nAn for the functions {0,…,n−1}→A\{0,\ldots,n-1\}\to A{0,…,n−1}→A, and similarly for other types. The block map NnN_nNn​ acts on a matrix XXX on An×BnA^n\times B^nAn×Bn by grouping these coordinates into (A×B)n(A\times B)^n(A×B)n and setting

(NnX)z,z′=∑u,v∈(A×B)nXu,v∏i=0n−1(N(Eui,vi))zi,zi′,(N_nX)_{z,z'}= \sum_{u,v\in(A\times B)^n}X_{u,v} \prod_{i=0}^{n-1}\bigl(N(E_{u_i,v_i})\bigr)_{z_i,z_i'},(Nn​X)z,z′​=u,v∈(A×B)n∑​Xu,v​i=0∏n−1​(N(Eui​,vi​​))zi​,zi′​​,

where Eui,viE_{u_i,v_i}Eui​,vi​​ is a matrix unit. An extension id⁡⊗F\operatorname{id}\otimes Fid⊗F applies FFF separately to every matrix block indexed by the untouched reference. Put [v]=vv†[v]=vv^\dagger[v]=vv†, let EkE_kEk​ have entries 1/k1/k1/k between (i,i)(i,i)(i,i) and (j,j)(j,j)(j,j) and zero otherwise, and let

F(X,Y)=(Re⁡tr⁡X Y X)2,F(X,Y)=\left(\operatorname{Re}\operatorname{tr} \sqrt{\sqrt X\,Y\,\sqrt X}\right)^2,F(X,Y)=(RetrX​YX​​)2,

where square roots use the supplied continuous functional calculus. A channel means a map X↦∑jLjXLj†X\mapsto\sum_jL_jXL_j^\daggerX↦∑j​Lj​XLj†​ for some finite family with ∑jLj†Lj=I\sum_jL_j^\dagger L_j=I∑j​Lj†​Lj​=I; a density matrix is positive semidefinite with trace one. In generation, the code consists of unit vectors αs∈CAn\alpha_s\in\mathbb C^{A^n}αs​∈CAn for s<ms<ms<m, a unit vector β∈Ck×Bn\beta\in\mathbb C^{k\times B^n}β∈Ck×Bn, and a decoding channel D:Cn→m×kD:C^n\to m\times kD:Cn→m×k, and must satisfy

1−ϵ≤1m∑s<mF ⁣(∣s⟩⟨s∣⊗Ek, Δ ⁣[(id⁡k⊗(D∘Nn))([αs]⊗[β])]).1-\epsilon\le\frac1m\sum_{s<m} F\!\left(|s\rangle\langle s|\otimes E_k,\, \Delta\!\left[(\operatorname{id}_k\otimes(D\circ N_n)) ([\alpha_s]\otimes[\beta])\right]\right).1−ϵ≤m1​s<m∑​F(∣s⟩⟨s∣⊗Ek​,Δ[(idk​⊗(D∘Nn​))([αs​]⊗[β])]).

Here inputs are reordered to k×(An×Bn)k\times(A^n\times B^n)k×(An×Bn), and Δ\DeltaΔ sets to zero all output entries whose two message indices differ. In transmission, β\betaβ is replaced by (id⁡k⊗E)(Ek)(\operatorname{id}_k\otimes E)(E_k)(idk​⊗E)(Ek​) for an encoding channel E:k→BnE:k\to B^nE:k→Bn, with the same average condition. In strong, Alice instead supplies density matrices ρs\rho_sρs​ on AnA^nAn, and there are encoding and decoding channels of those same types such that, for every s<ms<ms<m, every positive integer ttt, and every unit vector ψ∈Ct×k\psi\in\mathbb C^{t\times k}ψ∈Ct×k,

1−ϵ≤F ⁣(∣s⟩⟨s∣⊗[ψ], Δ ⁣[(id⁡t⊗(D∘Nn))(ρs⊗(id⁡t⊗E)([ψ]))]).1-\epsilon\le F\!\left(|s\rangle\langle s|\otimes[\psi],\, \Delta\!\left[(\operatorname{id}_t\otimes(D\circ N_n)) \bigl(\rho_s\otimes(\operatorname{id}_t\otimes E)([\psi])\bigr)\right]\right).1−ϵ≤F(∣s⟩⟨s∣⊗[ψ],Δ[(idt​⊗(D∘Nn​))(ρs​⊗(idt​⊗E)([ψ]))]).

The definition assumes neither that NNN is a channel nor that the finite types are nonempty; existence of the required vectors and channels remains part of the rate condition.

QAlgorithms.Papers.Independent.Support50.P02.cqMessageTarget

For index types R,IR,IR,I, a natural number mmm, a chosen label s∈Fin⁡ms\in\operatorname{Fin}ms∈Finm, and any complex matrix ρ\rhoρ on R×IR\times IR×I, this defines a matrix on R×(Fin⁡m×I)R\times(\operatorname{Fin}m\times I)R×(Finm×I) by

[cqMessageTarget⁡(s,ρ)](r,(j,i)), (r′,(j′,i′))={ρ(r,i),(r′,i′),j=s and j′=s,0,otherwise.\bigl[\operatorname{cqMessageTarget}(s,\rho)\bigr]_{(r,(j,i)),\,(r',(j',i'))} = \begin{cases} \rho_{(r,i),(r',i')},&j=s\ \text{and}\ j'=s,\\ 0,&\text{otherwise}. \end{cases}[cqMessageTarget(s,ρ)](r,(j,i)),(r′,(j′,i′))​={ρ(r,i),(r′,i′)​,0,​j=s and j′=s,otherwise.​

No density-matrix assumption is imposed on ρ\rhoρ. Although mmm ranges over all natural numbers, supplying sss is impossible when m=0m=0m=0.

QAlgorithms.Papers.Independent.Support50.P02.cqOne

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, an arbitrary function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, and an optional natural-number bound cap\mathrm{cap}cap, this is the set of pairs (r1,r2)(r_1,r_2)(r1​,r2​) for which there exist a finite ensemble of norm-one vectors αx∈CA\alpha_x\in\mathbb C^Aαx​∈CA, indexed by x<Kx<Kx<K, nonnegative weights pxp_xpx​ summing to 111, an integer k>0k>0k>0, and a norm-one vector β∈C{0,…,k−1}×B\beta\in\mathbb C^{\{0,\ldots,k-1\}\times B}β∈C{0,…,k−1}×B, satisfying the following conditions. Define

(ωx)(u,c),(v,c′)=N ⁣([αx(a)αx(a′)‾ β(u,b)β(v,b′)‾](a,b),(a′,b′))c,c′,ωˉ=∑xpxωx,(\omega_x)_{(u,c),(v,c')} = N\!\left( \left[ \alpha_x(a)\overline{\alpha_x(a')}\, \beta(u,b)\overline{\beta(v,b')} \right]_{(a,b),(a',b')} \right)_{c,c'}, \qquad \bar\omega=\sum_xp_x\omega_x,(ωx​)(u,c),(v,c′)​=N([αx​(a)αx​(a′)​β(u,b)β(v,b′)​](a,b),(a′,b′)​)c,c′​,ωˉ=x∑​px​ωx​,

and H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), with f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t applied by continuous functional calculus. Then

0≤r1≤H(Tr⁡Kωˉ)−∑xpxH(Tr⁡Kωx),0≤r2≤∑xpx[H(Tr⁡Kωx)−H(ωx)],0\le r_1\le H(\operatorname{Tr}_K\bar\omega)-\sum_xp_xH(\operatorname{Tr}_K\omega_x), \qquad 0\le r_2\le\sum_xp_x\bigl[H(\operatorname{Tr}_K\omega_x)-H(\omega_x)\bigr],0≤r1​≤H(TrK​ωˉ)−x∑​px​H(TrK​ωx​),0≤r2​≤x∑​px​[H(TrK​ωx​)−H(ωx​)],

where Tr⁡K\operatorname{Tr}_KTrK​ traces out the kkk-dimensional register. If cap\mathrm{cap}cap contains bbb, one additionally requires K≤bK\le bK≤b; if it contains no value, as in the default argument, this restriction is vacuous. Normalization forces K>0K>0K>0, so a bound of 000 admits no such data. No channel property of NNN is required.

QAlgorithms.Papers.Independent.Support50.P02.cqPentagons

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, and an arbitrary function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this is the set of nonnegative pairs (r1,r2)(r_1,r_2)(r1​,r2​) admitting a finite ensemble of norm-one vectors αx∈CA\alpha_x\in\mathbb C^Aαx​∈CA, indexed by x<Kx<Kx<K, nonnegative weights pxp_xpx​ with ∑xpx=1\sum_xp_x=1∑x​px​=1, a positive integer kkk, and a norm-one vector β∈C{0,…,k−1}×B\beta\in\mathbb C^{\{0,\ldots,k-1\}\times B}β∈C{0,…,k−1}×B, such that

(ωx)(u,c),(v,c′)=N ⁣([αx(a)αx(a′)‾ β(u,b)β(v,b′)‾](a,b),(a′,b′))c,c′,ωˉ=∑xpxωx(\omega_x)_{(u,c),(v,c')} = N\!\left( \left[ \alpha_x(a)\overline{\alpha_x(a')}\, \beta(u,b)\overline{\beta(v,b')} \right]_{(a,b),(a',b')} \right)_{c,c'}, \qquad \bar\omega=\sum_xp_x\omega_x(ωx​)(u,c),(v,c′)​=N([αx​(a)αx​(a′)​β(u,b)β(v,b′)​](a,b),(a′,b′)​)c,c′​,ωˉ=x∑​px​ωx​

satisfy

r1≤H(ωˉ)−∑xpxH(ωx),r2≤∑xpx[H(Tr⁡Kωx)−H(ωx)],r1+r2≤H(Tr⁡Kωˉ)−∑xpxH(Tr⁡Kωx)+∑xpx[H(Tr⁡Kωx)−H(ωx)].\begin{aligned} r_1&\le H(\bar\omega)-\sum_xp_xH(\omega_x),\\ r_2&\le\sum_xp_x\bigl[H(\operatorname{Tr}_K\omega_x)-H(\omega_x)\bigr],\\ r_1+r_2&\le H(\operatorname{Tr}_K\bar\omega)-\sum_xp_xH(\operatorname{Tr}_K\omega_x) +\sum_xp_x\bigl[H(\operatorname{Tr}_K\omega_x)-H(\omega_x)\bigr]. \end{aligned}r1​r2​r1​+r2​​≤H(ωˉ)−x∑​px​H(ωx​),≤x∑​px​[H(TrK​ωx​)−H(ωx​)],≤H(TrK​ωˉ)−x∑​px​H(TrK​ωx​)+x∑​px​[H(TrK​ωx​)−H(ωx​)].​

Here Tr⁡K\operatorname{Tr}_KTrK​ traces out the kkk-dimensional register and H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), with f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t applied by continuous functional calculus. Ensemble normalization forces K>0K>0K>0, while zero weights are allowed; NNN need not be a channel.

QAlgorithms.Papers.Independent.Support50.P02.dephaseClassicalRegister

For index types R,IR,IR,I, any natural number mmm, and any complex matrix ρ\rhoρ on R×(Fin⁡m×I)R\times(\operatorname{Fin}m\times I)R×(Finm×I), this defines a matrix on the same index set by

[dephaseClassicalRegister⁡(ρ)](r,(j,i)), (r′,(j′,i′))={ρ(r,(j,i)),(r′,(j′,i′)),j=j′,0,j≠j′.\bigl[\operatorname{dephaseClassicalRegister}(\rho)\bigr]_{(r,(j,i)),\,(r',(j',i'))} = \begin{cases} \rho_{(r,(j,i)),(r',(j',i'))},&j=j',\\ 0,&j\ne j'. \end{cases}[dephaseClassicalRegister(ρ)](r,(j,i)),(r′,(j′,i′))​={ρ(r,(j,i)),(r′,(j′,i′))​,0,​j=j′,j=j′.​

Thus entries are retained precisely when their label coordinates agree; no restriction is placed on their other coordinates. No positivity or normalization assumption is required. The case m=0m=0m=0 gives a matrix on an empty index set.

QAlgorithms.Papers.Independent.Support50.P02.independentPairs

For index types R,A,S,BR,A,S,BR,A,S,B, and arbitrary complex matrices ρ\rhoρ indexed by R×AR\times AR×A and τ\tauτ indexed by S×BS\times BS×B, this defines a complex matrix indexed by (R×S)×(A×B)(R\times S)\times(A\times B)(R×S)×(A×B), with entries

[independentPairs⁡(ρ,τ)]((r,s),(a,b)), ((r′,s′),(a′,b′))=ρ(r,a),(r′,a′) τ(s,b),(s′,b′).\bigl[\operatorname{independentPairs}(\rho,\tau)\bigr]_{((r,s),(a,b)),\,((r',s'),(a',b'))} =\rho_{(r,a),(r',a')}\,\tau_{(s,b),(s',b')}.[independentPairs(ρ,τ)]((r,s),(a,b)),((r′,s′),(a′,b′))​=ρ(r,a),(r′,a′)​τ(s,b),(s′,b′)​.

Neither input is required to be positive semidefinite or normalized. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P02.inputCoherent

For arbitrary finite types A,B,EA,B,EA,B,E, each equipped with decidable equality, an arbitrary complex matrix UUU with rows indexed by B×EB\times EB×E and columns indexed by AAA, and an arbitrary complex matrix ρ\rhoρ indexed by AAA, this defines

IU(ρ)=S ⁣(tr⁡E(UρU†))−S ⁣(tr⁡B(UρU†)),S(X)=−Re⁡tr⁡[f(X)],f(t)=tlog⁡2t,I_U(\rho)=S\!\left(\operatorname{tr}_E(U\rho U^\dagger)\right) -S\!\left(\operatorname{tr}_B(U\rho U^\dagger)\right), \qquad S(X)=-\operatorname{Re}\operatorname{tr}[f(X)],\quad f(t)=t\log_2t,IU​(ρ)=S(trE​(UρU†))−S(trB​(UρU†)),S(X)=−Retr[f(X)],f(t)=tlog2​t,

where f(X)f(X)f(X) uses continuous functional calculus. Explicitly, the two partial traces have entries ∑e(UρU†)(b,e),(b′,e)\sum_e(U\rho U^\dagger)_{(b,e),(b',e)}∑e​(UρU†)(b,e),(b′,e)​ and ∑b(UρU†)(b,e),(b,e′)\sum_b(U\rho U^\dagger)_{(b,e),(b,e')}∑b​(UρU†)(b,e),(b,e′)​, respectively. No isometry condition on UUU, density-matrix condition on ρ\rhoρ, or nonemptiness condition on the finite types is imposed; the expression uses the total functional calculus and logarithm also outside the density-matrix setting.

QAlgorithms.Papers.Independent.Support50.P02.macPower

For finite index types A,BA,BA,B with decidable equality, an index type CCC, an arbitrary function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, and any natural number nnn, this defines a function from matrices on An×BnA^n\times B^nAn×Bn to matrices on CnC^nCn. Explicitly, for an input matrix ρ\rhoρ and z,z′∈Cnz,z'\in C^nz,z′∈Cn, its output entry is

∑u,v∈(A×B)nρ(uA,uB),(vA,vB)∏k=0n−1[N(Euk,vk)]zk,zk′,\sum_{u,v\in(A\times B)^n} \rho_{(u_A,u_B),(v_A,v_B)} \prod_{k=0}^{n-1} \bigl[N(E_{u_k,v_k})\bigr]_{z_k,z'_k},u,v∈(A×B)n∑​ρ(uA​,uB​),(vA​,vB​)​k=0∏n−1​[N(Euk​,vk​​)]zk​,zk′​​,

where uA(k)u_A(k)uA​(k) and uB(k)u_B(k)uB​(k) are the two components of uku_kuk​, similarly for vvv, and Ea,bE_{a,b}Ea,b​ is the matrix with entry 111 at (a,b)(a,b)(a,b) and zero elsewhere. No linearity, positivity, or channel condition is imposed on NNN. The case n=0n=0n=0 is included: all sequence index sets are singletons, the empty product is 111, and the construction preserves the sole entry of ρ\rhoρ, independently of NNN.

QAlgorithms.Papers.Independent.Support50.P02.operationalRegion

For any predicate G(n,a,b,ϵ)G(n,a,b,\epsilon)G(n,a,b,ϵ), where n,a,bn,a,bn,a,b are natural numbers and ϵ\epsilonϵ is real, this defines the following subset of R2\mathbb R^2R2, with closure taken in the usual topology:

{(r1,r2) | r1≥0,r2≥0,∀ϵ>0 ∀δ>0 ∃n0∈N, n0>0,∀n≥n0 ∃a,b∈N, a>0, b>0,r1−δ≤log⁡2an,r2−δ≤log⁡2bn,G(n,a,b,ϵ)}‾.\overline{\left\{(r_1,r_2)\ \middle|\ \begin{gathered} r_1\ge0,\quad r_2\ge0,\quad \forall\epsilon>0\ \forall\delta>0\ \exists n_0\in\mathbb N,\ n_0>0,\\ \forall n\ge n_0\ \exists a,b\in\mathbb N,\ a>0,\ b>0,\\ r_1-\delta\le\frac{\log_2 a}{n},\qquad r_2-\delta\le\frac{\log_2 b}{n},\qquad G(n,a,b,\epsilon) \end{gathered}\right\}}.⎩⎨⎧​(r1​,r2​) ​ r1​≥0,r2​≥0,∀ϵ>0 ∀δ>0 ∃n0​∈N, n0​>0,∀n≥n0​ ∃a,b∈N, a>0, b>0,r1​−δ≤nlog2​a​,r2​−δ≤nlog2​b​,G(n,a,b,ϵ)​⎭⎬⎫​​.

The threshold n0n_0n0​ may depend on both tolerances, and a,ba,ba,b may additionally depend on nnn. The predicate GGG is arbitrary; no monotonicity or other property is required. Although its arguments allow zero natural numbers and arbitrary real errors, the condition defining this region tests it only with n,a,b>0n,a,b>0n,a,b>0 and ϵ>0\epsilon>0ϵ>0. Taking the closure also includes limits of the displayed rate pairs.

QAlgorithms.Papers.Independent.Support50.P02.optimizedSum

For finite types A,B,CA,B,CA,B,C with decidable equality and an arbitrary map NNN from matrices on A×BA\times BA×B to matrices on CCC, this defines a real number as follows. For every pair of positive integers a,ba,ba,b and unit vectors α∈Ca×A\alpha\in\mathbb C^{a\times A}α∈Ca×A, β∈Cb×B\beta\in\mathbb C^{b\times B}β∈Cb×B, form the matrix ω\omegaω on (a×b)×C(a\times b)\times C(a×b)×C by

ω((r,s),c),((r′,s′),c′)=[N ⁣((αr,iαr′,i′‾ βs,jβs′,j′‾)(i,j),(i′,j′))]c,c′.\omega_{((r,s),c),((r',s'),c')} = \left[ N\!\left( \bigl(\alpha_{r,i}\overline{\alpha_{r',i'}} \,\beta_{s,j}\overline{\beta_{s',j'}}\bigr)_{(i,j),(i',j')} \right)\right]_{c,c'}.ω((r,s),c),((r′,s′),c′)​=[N((αr,i​αr′,i′​​βs,j​βs′,j′​​)(i,j),(i′,j′)​)]c,c′​.

Its CCC-marginal is (ωC)c,c′=∑r<a∑s<bω((r,s),c),((r,s),c′)(\omega_C)_{c,c'}=\sum_{r<a}\sum_{s<b}\omega_{((r,s),c),((r,s),c')}(ωC​)c,c′​=∑r<a​∑s<b​ω((r,s),c),((r,s),c′)​. The defined number is

sup⁡a,b>0∥α∥=∥β∥=1(S(ωC)−S(ω)),S(X)=−Re⁡tr⁡f(X),f(x)=xlog⁡2x,\sup_{\substack{a,b>0\\ \|\alpha\|=\|\beta\|=1}} \bigl(S(\omega_C)-S(\omega)\bigr), \qquad S(X)=-\operatorname{Re}\operatorname{tr}f(X),\quad f(x)=x\log_2x,a,b>0∥α∥=∥β∥=1​sup​(S(ωC​)−S(ω)),S(X)=−Retrf(X),f(x)=xlog2​x,

where fff is applied through continuous functional calculus and the supremum is taken in R\mathbb RR. No channel property of NNN, nonemptiness of the types, or separate boundedness condition is assumed. If no such source vectors exist, the underlying set is empty and its real supremum is the totalized value 000.

QAlgorithms.Papers.Independent.Support50.P02.pureEnsembleAverage

For a finite index type AAA, let PPP consist of a natural number KKK, real weights px≥0p_x\ge0px​≥0 for x∈Fin⁡Kx\in\operatorname{Fin}Kx∈FinK satisfying ∑xpx=1\sum_xp_x=1∑x​px​=1, and vectors ϕx∈CA\phi_x\in\mathbb C^Aϕx​∈CA satisfying ∥ϕx∥=1\|\phi_x\|=1∥ϕx​∥=1. This defines the matrix

ρ‾a,a′=∑x=0K−1px ϕx(a) ϕx(a′)‾.\overline\rho_{a,a'} =\sum_{x=0}^{K-1}p_x\,\phi_x(a)\,\overline{\phi_x(a')}.ρ​a,a′​=x=0∑K−1​px​ϕx​(a)ϕx​(a′)​.

Equivalently, it is the weighted sum of the matrices whose entries are ϕx(a)ϕx(a′)‾\phi_x(a)\overline{\phi_x(a')}ϕx​(a)ϕx​(a′)​. The structure permits a natural-number count syntactically, but its weight condition rules out K=0K=0K=0; together with vector normalization, these conditions also rule out such an ensemble when AAA is empty.

QAlgorithms.Papers.Independent.Support50.P02.qqMacCapacity

For finite types A,B,CA,B,CA,B,C with decidable equality, an arbitrary matrix-valued map N:A×B→CN:A\times B\to CN:A×B→C, and a scenario defaulting to generation, this defines the closure of the nonnegative pairs (r1,r2)(r_1,r_2)(r1​,r2​) such that, for every ϵ,δ>0\epsilon,\delta>0ϵ,δ>0, there is a positive integer n0n_0n0​ such that every n≥n0n\ge n_0n≥n0​ admits positive integers a,ba,ba,b with

r1−δ≤log⁡2an,r2−δ≤log⁡2bn,r_1-\delta\le\frac{\log_2 a}{n},\qquad r_2-\delta\le\frac{\log_2 b}{n},r1​−δ≤nlog2​a​,r2​−δ≤nlog2​b​,

and a code of the selected kind. Here AnA^nAn denotes the functions from {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1} to AAA; NnN_nNn​ is defined, after grouping the two input strings into a string of pairs, by

(NnX)z,z′=∑u,v∈(A×B)nXu,v∏i<n(N(Eui,vi))zi,zi′.(N_nX)_{z,z'}=\sum_{u,v\in(A\times B)^n}X_{u,v} \prod_{i<n}\bigl(N(E_{u_i,v_i})\bigr)_{z_i,z_i'}.(Nn​X)z,z′​=u,v∈(A×B)n∑​Xu,v​i<n∏​(N(Eui​,vi​​))zi​,zi′​​.

Reference extensions apply a map to each reference-indexed matrix block. Write [v]=vv†[v]=vv^\dagger[v]=vv†, let EdE_dEd​ be the matrix with entry 1/d1/d1/d between (i,i)(i,i)(i,i) and (j,j)(j,j)(j,j) and zero elsewhere, and use

F(X,Y)=(Re⁡tr⁡X Y X)2F(X,Y)=\left(\operatorname{Re}\operatorname{tr} \sqrt{\sqrt X\,Y\,\sqrt X}\right)^2F(X,Y)=(RetrX​YX​​)2

with continuous functional calculus square roots. Every channel required below has a finite Kraus representation X↦∑jLjXLj†X\mapsto\sum_jL_jXL_j^\daggerX↦∑j​Lj​XLj†​ with ∑jLj†Lj=I\sum_jL_j^\dagger L_j=I∑j​Lj†​Lj​=I. In generation, a code consists of unit vectors α∈Ca×An\alpha\in\mathbb C^{a\times A^n}α∈Ca×An, β∈Cb×Bn\beta\in\mathbb C^{b\times B^n}β∈Cb×Bn, and a decoding channel D:Cn→a×bD:C^n\to a\times bD:Cn→a×b, satisfying

F ⁣(Ea⊗Eb, (id⁡a×b⊗(D∘Nn))([α]⊗[β]))≥1−ϵ,F\!\left(E_a\otimes E_b,\, (\operatorname{id}_{a\times b}\otimes(D\circ N_n)) ([\alpha]\otimes[\beta])\right)\ge1-\epsilon,F(Ea​⊗Eb​,(ida×b​⊗(D∘Nn​))([α]⊗[β]))≥1−ϵ,

with coordinates reordered so that the two references precede the two inputs or outputs. In transmission, there are encoding channels EA:a→AnE_A:a\to A^nEA​:a→An, EB:b→BnE_B:b\to B^nEB​:b→Bn, and such a decoding channel, and the same condition holds with the source replaced by

(id⁡a⊗EA)(Ea)⊗(id⁡b⊗EB)(Eb).(\operatorname{id}_a\otimes E_A)(E_a)\otimes (\operatorname{id}_b\otimes E_B)(E_b).(ida​⊗EA​)(Ea​)⊗(idb​⊗EB​)(Eb​).

In strong, the same kinds of encoding and decoding channels must, for all positive integers q,tq,tq,t and all unit vectors ψ∈Cq×a\psi\in\mathbb C^{q\times a}ψ∈Cq×a, ϕ∈Ct×b\phi\in\mathbb C^{t\times b}ϕ∈Ct×b, satisfy

F ⁣([ψ]⊗[ϕ], (id⁡q×t⊗(D∘Nn))((id⁡q⊗EA)([ψ])⊗(id⁡t⊗EB)([ϕ])))≥1−ϵ,F\!\left([\psi]\otimes[\phi],\, (\operatorname{id}_{q\times t}\otimes(D\circ N_n)) \left((\operatorname{id}_q\otimes E_A)([\psi]) \otimes(\operatorname{id}_t\otimes E_B)([\phi])\right)\right) \ge1-\epsilon,F([ψ]⊗[ϕ],(idq×t​⊗(D∘Nn​))((idq​⊗EA​)([ψ])⊗(idt​⊗EB​)([ϕ])))≥1−ϵ,

again grouping references before inputs and outputs. The definition imposes no channel assumption on NNN and no nonemptiness assumption on A,B,CA,B,CA,B,C.

QAlgorithms.Papers.Independent.Support50.P02.qqOne

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, and an arbitrary function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this is the set of pairs (r1,r2)(r_1,r_2)(r1​,r2​) for which there exist positive integers a,ba,ba,b and norm-one vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B such that, defining

ω((u,v),c),((u′,v′),c′)=N ⁣([α(u,i)α(u′,i′)‾ β(v,j)β(v′,j′)‾](i,j),(i′,j′))c,c′,\omega_{((u,v),c),((u',v'),c')} = N\!\left( \left[ \alpha(u,i)\overline{\alpha(u',i')}\, \beta(v,j)\overline{\beta(v',j')} \right]_{(i,j),(i',j')} \right)_{c,c'},ω((u,v),c),((u′,v′),c′)​=N([α(u,i)α(u′,i′)​β(v,j)β(v′,j′)​](i,j),(i′,j′)​)c,c′​,

one has

r1≥0,r2≥0,r1≤H(Tr⁡aω)−H(ω),r2≤H(Tr⁡bω)−H(ω),r1+r2≤H(Tr⁡a,bω)−H(ω).r_1\ge0,\qquad r_2\ge0,\qquad r_1\le H(\operatorname{Tr}_{a}\omega)-H(\omega),\qquad r_2\le H(\operatorname{Tr}_{b}\omega)-H(\omega),\qquad r_1+r_2\le H(\operatorname{Tr}_{a,b}\omega)-H(\omega).r1​≥0,r2​≥0,r1​≤H(Tra​ω)−H(ω),r2​≤H(Trb​ω)−H(ω),r1​+r2​≤H(Tra,b​ω)−H(ω).

Here partial traces sum over the diagonal entries of the indicated reference registers, and H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), with f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t applied by continuous functional calculus. The witnessing data may depend on the pair; no channel property of NNN is required.

QAlgorithms.Papers.Independent.Support50.P02.qqRectangles

For finite index types A,B,CA,B,CA,B,C, with the required decidable equalities, and an arbitrary function NNN from complex matrices on A×BA\times BA×B to complex matrices on CCC, this is the set of nonnegative pairs (r1,r2)(r_1,r_2)(r1​,r2​) for which there exist positive integers a,ba,ba,b and norm-one vectors α∈C{0,…,a−1}×A\alpha\in\mathbb C^{\{0,\ldots,a-1\}\times A}α∈C{0,…,a−1}×A, β∈C{0,…,b−1}×B\beta\in\mathbb C^{\{0,\ldots,b-1\}\times B}β∈C{0,…,b−1}×B satisfying the following bounds. Define

ω((u,v),c),((u′,v′),c′)=N ⁣([α(u,i)α(u′,i′)‾ β(v,j)β(v′,j′)‾](i,j),(i′,j′))c,c′.\omega_{((u,v),c),((u',v'),c')} = N\!\left( \left[ \alpha(u,i)\overline{\alpha(u',i')}\, \beta(v,j)\overline{\beta(v',j')} \right]_{(i,j),(i',j')} \right)_{c,c'}.ω((u,v),c),((u′,v′),c′)​=N([α(u,i)α(u′,i′)​β(v,j)β(v′,j′)​](i,j),(i′,j′)​)c,c′​.

Then the required inequalities are

r1≤H(Tr⁡a,bω)−H(Tr⁡bω),r2≤H(Tr⁡a,bω)−H(Tr⁡aω).r_1\le H(\operatorname{Tr}_{a,b}\omega)-H(\operatorname{Tr}_{b}\omega), \qquad r_2\le H(\operatorname{Tr}_{a,b}\omega)-H(\operatorname{Tr}_{a}\omega).r1​≤H(Tra,b​ω)−H(Trb​ω),r2​≤H(Tra,b​ω)−H(Tra​ω).

Here Tr⁡a\operatorname{Tr}_{a}Tra​ and Tr⁡b\operatorname{Tr}_{b}Trb​ trace out the respective reference registers, and H(X)=−Re⁡tr⁡f(X)H(X)=-\operatorname{Re}\operatorname{tr}f(X)H(X)=−Retrf(X), where f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t is applied by continuous functional calculus. No additional bound on r1+r2r_1+r_2r1​+r2​, and no channel hypothesis on NNN, appears.

QAlgorithms.Papers.Independent.Support50.P02.regularized

For any family FFF assigning a subset Fn⊆R2F_n\subseteq\mathbb R^2Fn​⊆R2 to each natural number nnn, this is the set

{(r1,r2)∈R2: ∃n∈N, n>0 and (nr1,nr2)∈Fn}.\{(r_1,r_2)\in\mathbb R^2:\ \exists n\in\mathbb N,\ n>0 \ \text{and}\ (nr_1,nr_2)\in F_n\}.{(r1​,r2​)∈R2: ∃n∈N, n>0 and (nr1​,nr2​)∈Fn​}.

Thus membership requires one positive integer witness; F0F_0F0​ contributes nothing. No nonnegativity condition on the pair, closure operation, or limiting condition is imposed.

Human review
  • Endorsed by Shuze Chen · Oct 23, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 23, 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