Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Quantum states representing perfectly secure bits are always distillable

Definition
QAlgorithms_Papers_Independent_Gap48_P42

by Goku · Oct 22, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

The objects the formalized statements of Quantum states representing perfectly secure bits are always distillable (arXiv:quant-ph/0602176) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • PartyBasis: PDF pp.2--3: party k owns a register of dimension D k.
  • localIndex: Fixed local basis relabeling; used identically on bras and kets.
  • partyPower: PDF pp.2--4: independent copies, grouped by the original owners. The local copy enumeration is the accepted fixed enumeration.
  • indexedPrivateDit: PDF p.2, multipartite private dit, after a fixed local basis relabeling. The accepted constructor includes the full joint-shield twists.
  • CanConvert: PDF pp.2--3: deterministic conversion by a complete finite adaptive LOCC tree, including arbitrary local ancillary dimensions and disposal. Density proofs are existentially certified, never assumptions deleting a result.
  • CanSelect: PDF pp.2--3,4: a positive-probability selected outcome of a complete LOCC tree; its other outcomes are retained in the underlying tree.
  • twoLabelGHZVector: PDF p.3: normalized GHZ plus/minus kets on distinct local labels i,j. All uses have N>=2 and i<j.
  • twoPhaseGHZMixture: PDF p.3, Eq.(2.6): the two-phase GHZ mixture, with p in [0,1] in uses.
  • ghzCopies: PDF pp.2--3: m N-party qubit GHZ states, keeping all m qubits per party.
  • GHZAchievable: PDF pp.2--3 and note [19], p.5: operational asymptotic GHZ yield. For every error/rate tolerance, every sufficiently large positive input length has a deterministic LOCC output of full trace-norm error at most epsilon. This is the vanishing-error, liminf-yield formulation; preprocessing success probabilities cannot be discarded.
  • distillableGHZ: PDF pp.2--3: Bell-pair rate for N=2, GHZ rate for N>=2. Extended suprema avoid default values of unbounded real suprema.
  • ProductVectors: PDF pp.2--3: one vector in each local shield, with norms checked separately.
  • UnitFactors: PDF p.2, Eq.(2.2), carried to Eq.(2.5): all local factors are unit vectors.
  • productElement: PDF pp.2--3: conjugate-linear bra, linear ket, on the whole shield.
  • productOverlap: PDF pp.2--3, Eqs.(2.2),(2.5): supremum over exactly normalized product vectors. For positive finite shields this set is nonempty, compact and bounded; the rate theorems assert attainment instead of presupposing a maximizing pair.
  • Maximizes: A maximizing normalized pair; existence is a conclusion below.
  • shieldBlock: PDF pp.2--3: the computed joint-shield block U_i rho U_j^dagger.
  • privateDiagonalWeight: PDF pp.2--3: a_s is computed from the same maximizing vector as eta.
  • filterSuccess: Corrected branch probability and normalized phase parameter, PDF pp.2--3. Theorems establish a,b>0 and 0<eta<=sqrt(a*b) before using these quantities.
  • phaseParameter: PDF pp.2--3: probability of the plus-phase GHZ component.
  • filterYield: Corrected probability-weighted hashing yield, not a conditional branch rate.
  • localFilter: Explicit balanced success operator, PDF pp.2--3. Party 0 balances the weights and removes the coherence phase; other parties project their shields. This rectangular operator alone is trace decreasing, not a TP protocol.
  • jointFilter: Tensor product of precisely the local success operators.
  • filteredBranch: The actual unnormalized all-success output, not an arbitrary output state.
  • PairConclusion: Data and exact success identity used by both quantitative conclusions. Every positivity, probability domain and Cauchy--Schwarz bound is a conclusion.
  • KeyPair: Distinct ordered key pairs; the domain is nonempty at d>=2.
  • bestPairYield: PDF p.3: the finite maximum over all key pairs for the chosen maximizing vectors. The finite nonempty domain at d>=2 guards the real supremum.
  • bipartitePrivateRaw: Bipartite private constructor, with the accepted exact block formula. Only the parentheses and the two fixed local bases are changed; each shield stays with its owner. Raw twists allow the explicitly computed tensor twists below to be checked for unitarity in the tensor-closure conclusion.
  • bipartitePrivate: PDF pp.1,4: the physical bipartite private state; rho is normalized in uses.
  • asTwoParties: Fixed bipartite-to-partywise adapter, with party 0 Alice and party 1 Bob.
  • bipartiteDistillable: PDF pp.2,4: ordinary Bell-pair distillability, the two-party specialization.
  • KeyAchievable: PDF p.4: private-key rate, using deterministic accepted finite LOCC maps, full trace norm and arbitrary positive finite output shields at every n. The denominator counts input copies and the numerator counts only key bits. All sufficiently large positive n are covered; no optimal sequence is assumed.
  • distillableKey: PDF pp.3--4: operational distillable key, including rate zero.
  • regroupIndex: PDF p.4: regroup key and shield copy tuples without changing either owner.
  • tensorTwists: PDF p.4: the computed twist for a key tuple, not an existential new twist.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Defs_Information
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P33
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P42
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P67

/-!
Horodecki--Augusiak, arXiv:quant-ph/0602176v2, PDF pp.1--5.
All new objects are paper-local adapters or explicit operational definitions.
Rates are nonnegative extended suprema, in bits per input copy. LOCC uses
complete adaptive instruments and unconditional outputs. No optimal protocol
or attainment of an asymptotic rate is assumed. Compilation is pending.
-/

namespace QAlgorithms.Papers.Independent.Gap48.P42

open scoped BigOperators Matrix ComplexConjugate ComplexOrder ENNReal

noncomputable section

attribute [local instance] Classical.propDecidable Classical.decEq

/-- PDF pp.2--3: party k owns a register of dimension D k. -/
abbrev PartyBasis {N : ℕ} (D : Fin N → ℕ) :=
  Claimed50.P33.RegisterBasis (fun k => Fin (D k))

/-- Fixed local basis relabeling; used identically on bras and kets. -/
def localIndex (a b : ℕ) : (Fin a × Fin b) ≃ Fin (a * b) :=
  Fintype.equivFinOfCardEq (by simp)

/-- PDF pp.2--4: independent copies, grouped by the original owners.
The local copy enumeration is the accepted fixed enumeration. -/
def partyPower {N : ℕ} {D : Fin N → ℕ}
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ) (n : ℕ) :
    Matrix (PartyBasis (fun k => D k ^ n)) (PartyBasis (fun k => D k ^ n)) ℂ :=
  fun x y => ∏ t : Fin n, rho
    (fun k => (Dependency50.P42.copyIndex (D k) n).symm (x k) t)
    (fun k => (Dependency50.P42.copyIndex (D k) n).symm (y k) t)

/-- PDF p.2, multipartite private dit, after a fixed local basis relabeling.
The accepted constructor includes the full joint-shield twists. -/
def indexedPrivateDit {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ) :
    Matrix (PartyBasis (fun k => d * S k)) (PartyBasis (fun k => d * S k)) ℂ :=
  fun x y => Claimed50.P33.privateDit (m := N) (H := fun k => Fin (S k)) d rho
    (fun i => ⟨(U i : Matrix (PartyBasis S) (PartyBasis S) ℂ), by
      constructor
      · ext a b
        have hab := congrFun (congrFun (U i).property.1 a) b
        by_cases heq : a = b <;>
          simpa only [Matrix.one_apply, heq, if_true, if_false] using hab
      · ext a b
        have hab := congrFun (congrFun (U i).property.2 a) b
        by_cases heq : a = b <;>
          simpa only [Matrix.one_apply, heq, if_true, if_false] using hab⟩)
    (fun k => (localIndex d (S k)).symm (x k))
    (fun k => (localIndex d (S k)).symm (y k))

/-- PDF pp.2--3: deterministic conversion by a complete finite adaptive
LOCC tree, including arbitrary local ancillary dimensions and disposal.
Density proofs are existentially certified, never assumptions deleting a result. -/
def CanConvert {N : ℕ} (D E : Fin N → ℕ)
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ)
    (sigma : Matrix (PartyBasis E) (PartyBasis E) ℂ) : Prop :=
  ∃ hD : ∀ k, 0 < D k, ∃ hE : ∀ k, 0 < E k,
  ∃ hrho : IsDensityMatrix rho, ∃ hsigma : IsDensityMatrix sigma,
    Claimed50.P33.LOCC ⟨D, hD, rho, hrho⟩ [(1, ⟨E, hE, sigma, hsigma⟩)]

/-- PDF pp.2--3,4: a positive-probability selected outcome of a complete
LOCC tree; its other outcomes are retained in the underlying tree. -/
def CanSelect {N : ℕ} (D E : Fin N → ℕ)
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ)
    (sigma : Matrix (PartyBasis E) (PartyBasis E) ℂ) : Prop :=
  ∃ hD : ∀ k, 0 < D k, ∃ hE : ∀ k, 0 < E k,
  ∃ hrho : IsDensityMatrix rho, ∃ hsigma : IsDensityMatrix sigma,
  ∃ out : List (ℝ × Claimed50.P33.FiniteState N),
    Claimed50.P33.LOCC ⟨D, hD, rho, hrho⟩ out ∧
    ∃ p : ℝ, 0 < p ∧ (p, ⟨E, hE, sigma, hsigma⟩) ∈ out

/-- PDF p.3: normalized GHZ plus/minus kets on distinct local labels i,j.
All uses have N>=2 and i<j. -/
def twoLabelGHZVector (N d : ℕ) (i j : Fin d) (minus : Bool) :
    EuclideanSpace ℂ (Fin N → Fin d) :=
  WithLp.toLp 2 fun x => (Real.sqrt 2 : ℂ)⁻¹ *
    ((if ∀ k, x k = i then 1 else 0) +
      (if ∀ k, x k = j then (if minus then -1 else 1) else 0))

/-- PDF p.3, Eq.(2.6): the two-phase GHZ mixture, with p in [0,1] in uses. -/
def twoPhaseGHZMixture (N d : ℕ) (i j : Fin d) (p : ℝ) :
    Matrix (Fin N → Fin d) (Fin N → Fin d) ℂ :=
  (p : ℂ) • pureDensity (twoLabelGHZVector N d i j false) +
    ((1 - p : ℝ) : ℂ) • pureDensity (twoLabelGHZVector N d i j true)

/-- PDF pp.2--3: m N-party qubit GHZ states, keeping all m qubits per party. -/
def ghzCopies (N m : ℕ) :
    Matrix (Fin N → Fin (2 ^ m)) (Fin N → Fin (2 ^ m)) ℂ :=
  partyPower (pureDensity (twoLabelGHZVector N 2 0 1 false)) m

/-- PDF pp.2--3 and note [19], p.5: operational asymptotic GHZ yield.
For every error/rate tolerance, every sufficiently large positive input length
has a deterministic LOCC output of full trace-norm error at most epsilon.
This is the vanishing-error, liminf-yield formulation; preprocessing success
probabilities cannot be discarded. -/
def GHZAchievable {N : ℕ} (D : Fin N → ℕ)
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ) (r : ℝ) : Prop :=
  0 ≤ r ∧ ∀ epsilon > (0 : ℝ), ∀ delta > (0 : ℝ), ∃ n0 : ℕ,
    ∀ n : ℕ, n0 ≤ n → 0 < n → ∃ m : ℕ,
    ∃ sigma : Matrix (Fin N → Fin (2 ^ m)) (Fin N → Fin (2 ^ m)) ℂ,
      CanConvert (fun k => D k ^ n) (fun _ => 2 ^ m) (partyPower rho n) sigma ∧
      traceNorm (sigma - ghzCopies N m) ≤ epsilon ∧ r - delta ≤ (m : ℝ) / n

/-- PDF pp.2--3: Bell-pair rate for N=2, GHZ rate for N>=2. Extended
suprema avoid default values of unbounded real suprema. -/
def distillableGHZ {N : ℕ} (D : Fin N → ℕ)
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ) : ℝ≥0∞ :=
  ⨆ r : {r : ℝ // GHZAchievable D rho r}, ENNReal.ofReal r.val

/-- PDF pp.2--3: one vector in each local shield, with norms checked separately. -/
abbrev ProductVectors {N : ℕ} (S : Fin N → ℕ) :=
  (k : Fin N) → EuclideanSpace ℂ (Fin (S k))

/-- PDF p.2, Eq.(2.2), carried to Eq.(2.5): all local factors are unit vectors. -/
def UnitFactors {N : ℕ} {S : Fin N → ℕ} (f : ProductVectors S) : Prop :=
  ∀ k, ‖f k‖ = 1

/-- PDF pp.2--3: conjugate-linear bra, linear ket, on the whole shield. -/
def productElement {N : ℕ} {S : Fin N → ℕ}
    (X : Matrix (PartyBasis S) (PartyBasis S) ℂ) (f g : ProductVectors S) : ℂ :=
  ∑ x : PartyBasis S, ∑ y : PartyBasis S,
    conj (∏ k, f k (x k)) * X x y * (∏ k, g k (y k))

/-- PDF pp.2--3, Eqs.(2.2),(2.5): supremum over exactly normalized product
vectors. For positive finite shields this set is nonempty, compact and bounded;
the rate theorems assert attainment instead of presupposing a maximizing pair. -/
def productOverlap {N : ℕ} {S : Fin N → ℕ}
    (X : Matrix (PartyBasis S) (PartyBasis S) ℂ) : ℝ :=
  sSup {r : ℝ | ∃ f g : ProductVectors S,
    UnitFactors f ∧ UnitFactors g ∧ r = ‖productElement X f g‖}

/-- A maximizing normalized pair; existence is a conclusion below. -/
def Maximizes {N : ℕ} {S : Fin N → ℕ}
    (X : Matrix (PartyBasis S) (PartyBasis S) ℂ) (f g : ProductVectors S) : Prop :=
  UnitFactors f ∧ UnitFactors g ∧ ‖productElement X f g‖ = productOverlap X

/-- PDF pp.2--3: the computed joint-shield block U_i rho U_j^dagger. -/
def shieldBlock {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ) (i j : Fin d) :
    Matrix (PartyBasis S) (PartyBasis S) ℂ :=
  (U i : Matrix (PartyBasis S) (PartyBasis S) ℂ) * rho *
    (U j : Matrix (PartyBasis S) (PartyBasis S) ℂ)ᴴ

/-- PDF pp.2--3: a_s is computed from the same maximizing vector as eta. -/
def privateDiagonalWeight {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ) (i : Fin d)
    (f : ProductVectors S) : ℝ := (productElement (shieldBlock rho U i i) f f).re

/-- Corrected branch probability and normalized phase parameter, PDF pp.2--3.
Theorems establish a,b>0 and 0<eta<=sqrt(a*b) before using these quantities. -/
def filterSuccess (d : ℕ) (a b : ℝ) : ℝ := 2 * min a b / d

/-- PDF pp.2--3: probability of the plus-phase GHZ component. -/
def phaseParameter (eta a b : ℝ) : ℝ := 1 / 2 + eta / (2 * Real.sqrt (a * b))

/-- Corrected probability-weighted hashing yield, not a conditional branch rate. -/
def filterYield (d : ℕ) (eta a b : ℝ) : ℝ :=
  filterSuccess d a b * (1 - QAlgorithms.binEntropy2 (phaseParameter eta a b))

/-- Explicit balanced success operator, PDF pp.2--3. Party 0 balances the
weights and removes the coherence phase; other parties project their shields.
This rectangular operator alone is trace decreasing, not a TP protocol. -/
def localFilter {N d : ℕ} {S : Fin N → ℕ} (i j : Fin d)
    (f g : ProductVectors S) (a b : ℝ) (c : ℂ) (k : Fin N) :
    Matrix (Fin d) (Fin (d * S k)) ℂ :=
  fun x y =>
    let z := (localIndex d (S k)).symm y
    (if x = i ∧ z.1 = i then
      (if k.val = 0 then (Real.sqrt (min a b / a) : ℂ) else 1) * conj (f k z.2)
      else 0) +
    (if x = j ∧ z.1 = j then
      (if k.val = 0 then (Real.sqrt (min a b / b) : ℂ) * (c / (‖c‖ : ℂ)) else 1) *
        conj (g k z.2) else 0)

/-- Tensor product of precisely the local success operators. -/
def jointFilter {N d : ℕ} {S : Fin N → ℕ} (i j : Fin d)
    (f g : ProductVectors S) (a b : ℝ) (c : ℂ) :
    Matrix (Fin N → Fin d) (PartyBasis (fun k => d * S k)) ℂ :=
  fun x y => ∏ k, localFilter i j f g a b c k (x k) (y k)

/-- The actual unnormalized all-success output, not an arbitrary output state. -/
def filteredBranch {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ) (i j : Fin d)
    (f g : ProductVectors S) : Matrix (Fin N → Fin d) (Fin N → Fin d) ℂ :=
  let K := jointFilter i j f g (privateDiagonalWeight rho U i f) (privateDiagonalWeight rho U j g)
    (productElement (shieldBlock rho U i j) f g)
  K * indexedPrivateDit rho U * Kᴴ

/-- Data and exact success identity used by both quantitative conclusions.
Every positivity, probability domain and Cauchy--Schwarz bound is a conclusion. -/
def PairConclusion {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ) (i j : Fin d)
    (f g : ProductVectors S) : Prop :=
  let eta := productOverlap (shieldBlock rho U i j)
  let a := privateDiagonalWeight rho U i f
  let b := privateDiagonalWeight rho U j g
  let p := phaseParameter eta a b
  let q := filterSuccess d a b
  0 < eta ∧ 0 < a ∧ 0 < b ∧ eta ≤ Real.sqrt (a * b) ∧
  0 < q ∧ q ≤ 1 ∧ 1 / 2 < p ∧ p ≤ 1 ∧
  (∀ k, (1 - (localFilter i j f g a b
    (productElement (shieldBlock rho U i j) f g) k)ᴴ *
      localFilter i j f g a b (productElement (shieldBlock rho U i j) f g) k).PosSemidef) ∧
  filteredBranch rho U i j f g = (q : ℂ) • twoPhaseGHZMixture N d i j p ∧
  (filteredBranch rho U i j f g).trace.re = q ∧
  0 < filterYield d eta a b ∧
  ENNReal.ofReal (filterYield d eta a b) ≤
    distillableGHZ (fun k => d * S k) (indexedPrivateDit rho U)

/-- Distinct ordered key pairs; the domain is nonempty at d>=2. -/
abbrev KeyPair (d : ℕ) := {p : Fin d × Fin d // p.1 < p.2}

/-- PDF p.3: the finite maximum over all key pairs for the chosen maximizing
vectors. The finite nonempty domain at d>=2 guards the real supremum. -/
def bestPairYield {N d : ℕ} {S : Fin N → ℕ}
    (rho : Matrix (PartyBasis S) (PartyBasis S) ℂ)
    (U : Fin d → Matrix.unitaryGroup (PartyBasis S) ℂ)
    (v : KeyPair d → ProductVectors S × ProductVectors S) : ℝ :=
  sSup (Set.range fun p : KeyPair d => filterYield d
    (productOverlap (shieldBlock rho U p.val.1 p.val.2))
    (privateDiagonalWeight rho U p.val.1 (v p).1)
    (privateDiagonalWeight rho U p.val.2 (v p).2))

/-- Bipartite private constructor, with the accepted exact block formula.
Only the parentheses and the two fixed local bases are changed; each shield
stays with its owner. Raw twists allow the explicitly computed tensor twists
below to be checked for unitarity in the tensor-closure conclusion. -/
def bipartitePrivateRaw {d a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ)
    (U : Fin d → Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ) :
    Matrix (Fin (d * a) × Fin (d * b)) (Fin (d * a) × Fin (d * b)) ℂ :=
  fun x y =>
    let xa := (localIndex d a).symm x.1
    let xb := (localIndex d b).symm x.2
    let ya := (localIndex d a).symm y.1
    let yb := (localIndex d b).symm y.2
    Information100.P67.privateState U rho
      (xa.1, (xa.2, (xb.1, xb.2))) (ya.1, (ya.2, (yb.1, yb.2)))

/-- PDF pp.1,4: the physical bipartite private state; rho is normalized in uses. -/
def bipartitePrivate {d a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ)
    (U : Fin d → Matrix.unitaryGroup (Fin a × Fin b) ℂ) :=
  bipartitePrivateRaw rho (fun i => (U i : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ))

/-- Fixed bipartite-to-partywise adapter, with party 0 Alice and party 1 Bob. -/
def asTwoParties {a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ) :
    Matrix (PartyBasis ![a, b]) (PartyBasis ![a, b]) ℂ :=
  fun x y => rho (x 0, x 1) (y 0, y 1)

/-- PDF pp.2,4: ordinary Bell-pair distillability, the two-party specialization. -/
def bipartiteDistillable {a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ) : ℝ≥0∞ :=
  distillableGHZ ![a, b] (asTwoParties rho)

/-- PDF p.4: private-key rate, using deterministic accepted finite LOCC maps,
full trace norm and arbitrary positive finite output shields at every n.
The denominator counts input copies and the numerator counts only key bits.
All sufficiently large positive n are covered; no optimal sequence is assumed. -/
def KeyAchievable {a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ) (r : ℝ) : Prop :=
  0 ≤ r ∧ ∀ epsilon > (0 : ℝ), ∀ delta > (0 : ℝ), ∃ n0 : ℕ,
  ∀ n : ℕ, n0 ≤ n → 0 < n →
  ∃ d s t : ℕ, 0 < d ∧ 0 < s ∧ 0 < t ∧
  ∃ tau : Matrix (Fin s × Fin t) (Fin s × Fin t) ℂ, IsDensityMatrix tau ∧
  ∃ U : Fin d → Matrix.unitaryGroup (Fin s × Fin t) ℂ,
  ∃ L : Superop (Fin (a ^ n) × Fin (b ^ n)) (Fin (d * s) × Fin (d * t)),
    Information100.P21.FiniteLOCC L ∧
    traceNorm (L (Dependency50.P42.dimensionedPower rho n) - bipartitePrivate tau U) ≤ epsilon ∧
    r - delta ≤ Real.logb 2 d / n

/-- PDF pp.3--4: operational distillable key, including rate zero. -/
def distillableKey {a b : ℕ}
    (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ) : ℝ≥0∞ :=
  ⨆ r : {r : ℝ // KeyAchievable rho r}, ENNReal.ofReal r.val

/-- PDF p.4: regroup key and shield copy tuples without changing either owner. -/
def regroupIndex (d a m : ℕ) : Fin (d ^ m * a ^ m) → Fin ((d * a) ^ m) :=
  fun x =>
    let z := (localIndex (d ^ m) (a ^ m)).symm x
    Dependency50.P42.copyIndex (d * a) m (fun t => localIndex d a
      ((Dependency50.P42.copyIndex d m).symm z.1 t,
       (Dependency50.P42.copyIndex a m).symm z.2 t))

/-- PDF p.4: the computed twist for a key tuple, not an existential new twist. -/
def tensorTwists {d a b : ℕ}
    (U : Fin d → Matrix.unitaryGroup (Fin a × Fin b) ℂ) (m : ℕ) :
    Fin (d ^ m) → Matrix (Fin (a ^ m) × Fin (b ^ m)) (Fin (a ^ m) × Fin (b ^ m)) ℂ :=
  fun key x y => ∏ t : Fin m,
    (U ((Dependency50.P42.copyIndex d m).symm key t) :
      Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ)
      ((Dependency50.P42.copyIndex a m).symm x.1 t,
       (Dependency50.P42.copyIndex b m).symm x.2 t)
      ((Dependency50.P42.copyIndex a m).symm y.1 t,
       (Dependency50.P42.copyIndex b m).symm y.2 t)

/-- Sanity: no-copy party power is the scalar-one entry. -/
example {N : ℕ} {D : Fin N → ℕ} (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ)
    (x y : PartyBasis (fun k => D k ^ 0)) : partyPower rho 0 x y = 1 := by
  simp [partyPower]

/-- Sanity: the bipartite adapter keeps the off-diagonal entries on the same cut. -/
example {a b : ℕ} (rho : Matrix (Fin a × Fin b) (Fin a × Fin b) ℂ)
    (x y : PartyBasis ![a, b]) : asTwoParties rho x y = rho (x 0, x 1) (y 0, y 1) := rfl

/-- Sanity: vanishing coherence gives the unbiased phase mixture. -/
example (a b : ℝ) : phaseParameter 0 a b = 1 / 2 := by
  simp [phaseParameter]

/-- Sanity: unit diagonal weights yield success 2/d, including d=2. -/
example (d : ℕ) : filterSuccess d 1 1 = 2 / (d : ℝ) := by
  simp [filterSuccess]

/-- Sanity: the plus endpoint is the pure GHZ target. -/
example (N d : ℕ) (i j : Fin d) :
    twoPhaseGHZMixture N d i j 1 = pureDensity (twoLabelGHZVector N d i j false) := by
  simp [twoPhaseGHZMixture]

/-- Sanity: the computed tensor twist is the empty product at zero copies. -/
example {d a b : ℕ} (U : Fin d → Matrix.unitaryGroup (Fin a × Fin b) ℂ)
    (key : Fin (d ^ 0)) (x y : Fin (a ^ 0) × Fin (b ^ 0)) :
    tensorTwists U 0 key x y = 1 := by
  simp [tensorTwists]

/-- Sanity: a deterministic LOCC conversion is already a positive selected branch. -/
example {N : ℕ} (D E : Fin N → ℕ)
    (rho : Matrix (PartyBasis D) (PartyBasis D) ℂ)
    (sigma : Matrix (PartyBasis E) (PartyBasis E) ℂ) (h : CanConvert D E rho sigma) :
    CanSelect D E rho sigma := by
  rcases h with ⟨hD, hE, hrho, hsigma, hL⟩
  exact ⟨hD, hE, hrho, hsigma, [(1, ⟨E, hE, sigma, hsigma⟩)], hL,
    1, by norm_num, by simp⟩

end

end QAlgorithms.Papers.Independent.Gap48.P42
Source
Quantum states representing perfectly secure bits are always distillable, arXiv:quant-ph/0602176v2, https://arxiv.org/abs/quant-ph/0602176v2
Read-back

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

QAlgorithms.Papers.Independent.Gap48.P42.CanConvert

For any natural number NNN, dimension assignments D,E:{0,…,N−1}→ND,E:\{0,\ldots,N-1\}\to\mathbb ND,E:{0,…,N−1}→N, and complex matrices ρ,σ\rho,\sigmaρ,σ indexed respectively by tuples with local dimensions Dk,EkD_k,E_kDk​,Ek​, CanConvert⁡(D,E,ρ,σ)\operatorname{CanConvert}(D,E,\rho,\sigma)CanConvert(D,E,ρ,σ) means that every DkD_kDk​ and EkE_kEk​ is strictly positive, both matrices are positive semidefinite with trace 111, and the following finitely generated protocol relation takes the bundled state (D,ρ)(D,\rho)(D,ρ) to exactly the singleton output list [(1,(E,σ))][(1,(E,\sigma))][(1,(E,σ))]. The relation permits stopping at any state sss, yielding [(1,s)][(1,s)][(1,s)]; it permits a local instrument followed by a separate such protocol on every outcome, multiplying the instrument weight by each subsequent output weight and concatenating the resulting lists; and it permits replacing an output list by its weighted mixture when all output states have the same local dimensions. More explicitly, a local instrument consists of finitely many nonnegative weights pjp_jpj​ summing to 111, density matrices ρj\rho_jρj​ with positive local dimensions, a selected party, and finitely many Kraus matrices Aj,zA_{j,z}Aj,z​ acting on that party, with every other party’s dimension unchanged. Their full-system lifts Kj,zK_{j,z}Kj,z​ have entry Aj,z(xparty,yparty)A_{j,z}(x_{\mathrm{party}},y_{\mathrm{party}})Aj,z​(xparty​,yparty​) when every other numerical coordinate of x,yx,yx,y agrees, and entry 000 otherwise; they must satisfy ∑j,zKj,z†Kj,z=I\sum_{j,z}K_{j,z}^\dagger K_{j,z}=I∑j,z​Kj,z†​Kj,z​=I and pjρj=∑zKj,zρKj,z†p_j\rho_j=\sum_zK_{j,z}\rho K_{j,z}^\daggerpj​ρj​=∑z​Kj,z​ρKj,z†​. The mixture rule requires the target matrix entries to equal the weighted sum of the output entries, with entries outside a state’s coordinate bounds interpreted as zero. No lower bound on NNN is required; if any specified local dimension is zero, the predicate is false.

QAlgorithms.Papers.Independent.Gap48.P42.CanSelect

For any natural number NNN, natural dimension assignments D,ED,ED,E, and complex matrices ρ,σ\rho,\sigmaρ,σ on the corresponding tuple-indexed spaces, CanSelect⁡(D,E,ρ,σ)\operatorname{CanSelect}(D,E,\rho,\sigma)CanSelect(D,E,ρ,σ) means that all local dimensions are strictly positive, both matrices are positive semidefinite with trace 111, and there exists a finite protocol output list containing the exact pair (p,(E,σ))(p,(E,\sigma))(p,(E,σ)) for some real p>0p>0p>0. The protocol relation is the smallest relation generated by stopping with the single output (1,s)(1,s)(1,s), applying a local instrument and then outcome-dependent protocols with their weights multiplied and lists concatenated, and forgetting an output list by replacing it with its weighted mixture when all output dimensions agree. A local instrument has finitely many nonnegative outcome weights pjp_jpj​ summing to 111, normalized positive semidefinite branch matrices ρj\rho_jρj​, positive branch dimensions, one selected party, and finitely many Kraus matrices Aj,zA_{j,z}Aj,z​ on that party; other parties’ dimensions remain unchanged. The full-system lift Kj,zK_{j,z}Kj,z​ equals Aj,zA_{j,z}Aj,z​ on the selected coordinates and enforces equality of every other numerical coordinate. It satisfies ∑j,zKj,z†Kj,z=I\sum_{j,z}K_{j,z}^\dagger K_{j,z}=I∑j,z​Kj,z†​Kj,z​=I and pjρj=∑zKj,zρKj,z†p_j\rho_j=\sum_zK_{j,z}\rho K_{j,z}^\daggerpj​ρj​=∑z​Kj,z​ρKj,z†​. Forgetting requires equality of the target entries with the weighted sum of the output entries, extending each state’s entries by zero outside its dimension bounds. The selected pair need only occur in the list; it need not be its sole entry, and other occurrences of the same state are allowed. The definition includes N=0N=0N=0; any zero dimension in DDD or EEE makes it false.

QAlgorithms.Papers.Independent.Gap48.P42.GHZAchievable

For any natural number NNN, natural dimensions DkD_kDk​, arbitrary complex matrix ρ\rhoρ on ∏k<N{0,…,Dk−1}\prod_{k<N}\{0,\ldots,D_k-1\}∏k<N​{0,…,Dk​−1}, and real rrr, this predicate requires r≥0r\ge0r≥0 and the following: for every ϵ>0\epsilon>0ϵ>0 and δ>0\delta>0δ>0, there exists a natural n0n_0n0​ such that every natural n≥n0n\ge n_0n≥n0​ with n>0n>0n>0 admits a natural mmm and a density matrix σ\sigmaσ on ∏k<N{0,…,2m−1}\prod_{k<N}\{0,\ldots,2^m-1\}∏k<N​{0,…,2m−1} satisfying

∥σ−GN,m∥1≤ϵ,r−δ≤mn,\|\sigma-G_{N,m}\|_1\le\epsilon, \qquad r-\delta\le\frac mn,∥σ−GN,m​∥1​≤ϵ,r−δ≤nm​,

together with an exact finite protocol conversion from ρ[n]\rho^{[n]}ρ[n] to σ\sigmaσ. Here ρ[n]\rho^{[n]}ρ[n] is the grouped tensor power: decode each local index in {0,…,Dkn−1}\{0,\ldots,D_k^n-1\}{0,…,Dkn​−1} into a length-nnn tuple by the fixed cardinality-chosen bijection, and multiply the corresponding nnn entries of ρ\rhoρ. The target GN,mG_{N,m}GN,m​ is the similarly grouped mmm-fold tensor power of gg†gg^\daggergg†, where g(z)=(1{∀k, zk=0}+1{∀k, zk=1})/2g(z)=\bigl(\mathbf1_{\{\forall k,\ z_k=0\}}+\mathbf1_{\{\forall k,\ z_k=1\}}\bigr)/\sqrt2g(z)=(1{∀k, zk​=0}​+1{∀k, zk​=1}​)/2​. The trace norm is ∥X∥1=Re⁡tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{tr}\sqrt{X^\dagger X}∥X∥1​=RetrX†X​. Exact conversion requires positive input and output local dimensions, positive semidefinite trace-one input and output matrices, and a protocol ending in the singleton list [(1,σ)][(1,\sigma)][(1,σ)]. Such protocols are generated by stopping, finite local instruments with outcome-dependent continuation, and forgetting outcomes by taking their weighted mixture when dimensions agree. Each instrument has nonnegative weights summing to 111, density-matrix branch states, and Kraus operators acting on one party with identities on the others, satisfying ∑j,zKj,z†Kj,z=I\sum_{j,z}K_{j,z}^\dagger K_{j,z}=I∑j,z​Kj,z†​Kj,z​=I and pjρj=∑zKj,zρinKj,z†p_j\rho_j=\sum_zK_{j,z}\rho_{\mathrm{in}}K_{j,z}^\daggerpj​ρj​=∑z​Kj,z​ρin​Kj,z†​; continuation weights multiply, and forgetting sums weighted matrix entries. There is no separate density assumption on ρ\rhoρ, but every required ρ[n]\rho^{[n]}ρ[n] must be a density matrix. Zero parties and m=0m=0m=0 are permitted; any zero DkD_kDk​ makes the predicate false, since n>0n>0n>0 is required.

QAlgorithms.Papers.Independent.Gap48.P42.KeyAchievable

For arbitrary natural numbers a,ba,ba,b, a complex matrix ρ\rhoρ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), and a real number rrr, this predicate means r≥0r\ge0r≥0 and the following: for every ε,δ>0\varepsilon,\delta>0ε,δ>0, there exists n0∈Nn_0\in\mathbb Nn0​∈N such that, for every n≥n0n\ge n_0n≥n0​ with n>0n>0n>0, there exist positive integers d,s,td,s,td,s,t, a positive semidefinite matrix τ\tauτ of trace one on Fin⁡(s)×Fin⁡(t)\operatorname{Fin}(s)\times\operatorname{Fin}(t)Fin(s)×Fin(t), unitaries UαU_\alphaUα​ on that space for α<d\alpha<dα<d, and a map LLL from matrices on Fin⁡(an)×Fin⁡(bn)\operatorname{Fin}(a^n)\times\operatorname{Fin}(b^n)Fin(an)×Fin(bn) to matrices on Fin⁡(ds)×Fin⁡(dt)\operatorname{Fin}(ds)\times\operatorname{Fin}(dt)Fin(ds)×Fin(dt), such that

∥L(ρ⊗n)−Γτ,U∥1≤ε,r−δ≤log⁡2dn.\|L(\rho^{\otimes n})-\Gamma_{\tau,U}\|_1\le\varepsilon, \qquad r-\delta\le\frac{\log_2d}{n}.∥L(ρ⊗n)−Γτ,U​∥1​≤ε,r−δ≤nlog2​d​.

Here LLL belongs to the recursively generated class consisting of identity maps and finite complete local Kraus measurements by either party, followed by an outcome-dependent map in that same class, with the outcome branches summed. The source tensor power has entries ∏ℓ<nρ((xℓ,yℓ),(xℓ′,yℓ′))\prod_{\ell<n}\rho((x_\ell,y_\ell),(x'_\ell,y'_\ell))∏ℓ<n​ρ((xℓ​,yℓ​),(xℓ′​,yℓ′​)), regrouped through chosen finite-cardinality equivalences into dimensions an,bna^n,b^nan,bn. Using chosen equivalences Fin⁡(d)×Fin⁡(s)≃Fin⁡(ds)\operatorname{Fin}(d)\times\operatorname{Fin}(s)\simeq\operatorname{Fin}(ds)Fin(d)×Fin(s)≃Fin(ds) and similarly for ttt, the target has entries d−1(UατUα′†)(u,v),(u′,v′)d^{-1}(U_\alpha\tau U_{\alpha'}^\dagger)_{(u,v),(u',v')}d−1(Uα​τUα′†​)(u,v),(u′,v′)​ when the two parties’ row key labels both equal α\alphaα and their column key labels both equal α′\alpha'α′, and zero otherwise. The norm is ∥X∥1=Re⁡Tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{Tr}\sqrt{X^\dagger X}∥X∥1​=ReTrX†X​, without a factor of one half. The input ρ\rhoρ is not assumed to be a density matrix, and a,ba,ba,b may be zero; d=1d=1d=1 is allowed, whereas n=0n=0n=0 is excluded from the requirements.

QAlgorithms.Papers.Independent.Gap48.P42.KeyPair

For every natural number ddd, this is the type of ordered pairs (i,j)(i,j)(i,j) of integers with 0≤i<j<d0\le i<j<d0≤i<j<d. It is empty when d=0d=0d=0 or d=1d=1d=1.

QAlgorithms.Papers.Independent.Gap48.P42.Maximizes

For any N∈NN\in\mathbb NN∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, complex matrix XXX indexed by BS=∏k∈[N][Sk]\mathcal B_S=\prod_{k\in[N]}[S_k]BS​=∏k∈[N]​[Sk​], and families fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, this predicate requires ∥fk∥=∥gk∥=1\|f_k\|=\|g_k\|=1∥fk​∥=∥gk​∥=1 for every kkk, and requires

∣∑x,y∈BS∏kfk(xk)‾ Xx,y∏kgk(yk)∣=sup⁡uk,vk∈CSk∥uk∥=∥vk∥=1 for every k∣∑x,y∈BS∏kuk(xk)‾ Xx,y∏kvk(yk)∣.\left|\sum_{x,y\in\mathcal B_S} \overline{\prod_k f_k(x_k)}\,X_{x,y}\prod_k g_k(y_k)\right| = \sup_{\substack{u_k,v_k\in\mathbb C^{S_k}\\ \|u_k\|=\|v_k\|=1\ \text{for every }k}} \left|\sum_{x,y\in\mathcal B_S} \overline{\prod_k u_k(x_k)}\,X_{x,y}\prod_k v_k(y_k)\right|.​x,y∈BS​∑​k∏​fk​(xk​)​Xx,y​k∏​gk​(yk​)​=uk​,vk​∈CSk​∥uk​∥=∥vk​∥=1 for every k​sup​​x,y∈BS​∑​k∏​uk​(xk​)​Xx,y​k∏​vk​(yk​)​.

Thus it asserts attainment by the supplied families, without asserting their uniqueness. If any Sk=0S_k=0Sk​=0, the predicate is impossible; for N=0N=0N=0, it holds for the unique empty families.

QAlgorithms.Papers.Independent.Gap48.P42.PairConclusion

For arbitrary natural numbers N,dN,dN,d, local dimensions Sk∈NS_k\in\mathbb NSk​∈N for k<Nk<Nk<N, a complex matrix ρ\rhoρ indexed by BS=∏k<NFin⁡(Sk)\mathcal B_S=\prod_{k<N}\operatorname{Fin}(S_k)BS​=∏k<N​Fin(Sk​), unitaries UiU_iUi​ on this space for i<di<di<d, labels i,j<di,j<di,j<d, and families of local vectors fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, this defines the following conjunction; there is no hypothesis here that ρ\rhoρ is a density matrix, that the local vectors are normalized, that i≠ji\ne ji=j, or that N≥2N\ge2N≥2. Write F(x)=∏kfk(xk)F(x)=\prod_k f_k(x_k)F(x)=∏k​fk​(xk​), G(x)=∏kgk(xk)G(x)=\prod_k g_k(x_k)G(x)=∏k​gk​(xk​), Xij=UiρUj†X_{ij}=U_i\rho U_j^\daggerXij​=Ui​ρUj†​, c=∑x,yF(x)‾Xij(x,y)G(y)c=\sum_{x,y}\overline{F(x)}X_{ij}(x,y)G(y)c=∑x,y​F(x)​Xij​(x,y)G(y), a=Re⁡∑x,yF(x)‾Xii(x,y)F(y)a=\operatorname{Re}\sum_{x,y}\overline{F(x)}X_{ii}(x,y)F(y)a=Re∑x,y​F(x)​Xii​(x,y)F(y), and b=Re⁡∑x,yG(x)‾Xjj(x,y)G(y)b=\operatorname{Re}\sum_{x,y}\overline{G(x)}X_{jj}(x,y)G(y)b=Re∑x,y​G(x)​Xjj​(x,y)G(y). Let η\etaη be the real supremum of ∣∑x,y∏khk(xk)‾Xij(x,y)∏kℓk(yk)∣\left|\sum_{x,y}\overline{\prod_k h_k(x_k)}X_{ij}(x,y)\prod_k \ell_k(y_k)\right|​∑x,y​∏k​hk​(xk​)​Xij​(x,y)∏k​ℓk​(yk​)​ over all families h,ℓh,\ellh,ℓ with every local factor of norm one; put p=12+η/(2ab)p=\frac12+\eta/(2\sqrt{ab})p=21​+η/(2ab​), q=2min⁡(a,b)/dq=2\min(a,b)/dq=2min(a,b)/d, and Y=q[1−H2(p)]Y=q[1-H_2(p)]Y=q[1−H2​(p)], where H2H_2H2​ is binary entropy in bits. For each party, use the chosen finite-cardinality equivalence λk:Fin⁡(d)×Fin⁡(Sk)≃Fin⁡(dSk)\lambda_k:\operatorname{Fin}(d)\times\operatorname{Fin}(S_k)\simeq\operatorname{Fin}(dS_k)λk​:Fin(d)×Fin(Sk​)≃Fin(dSk​), and define

Kk(x,λk(r,z))=1x=i, r=i αkfk(z)‾+1x=j, r=j βkgk(z)‾,(αk,βk)={(min⁡(a,b)/a,min⁡(a,b)/b c/∣c∣),k=0,(1,1),k≠0.K_k(x,\lambda_k(r,z))= \mathbf1_{x=i,\ r=i}\,\alpha_k\overline{f_k(z)} +\mathbf1_{x=j,\ r=j}\,\beta_k\overline{g_k(z)},\qquad (\alpha_k,\beta_k)= \begin{cases} \bigl(\sqrt{\min(a,b)/a},\sqrt{\min(a,b)/b}\,c/|c|\bigr),&k=0,\\ (1,1),&k\ne0. \end{cases}Kk​(x,λk​(r,z))=1x=i, r=i​αk​fk​(z)​+1x=j, r=j​βk​gk​(z)​,(αk​,βk​)={(min(a,b)/a​,min(a,b)/b​c/∣c∣),(1,1),​k=0,k=0.​

Let K(x,y)=∏kKk(xk,yk)K(x,y)=\prod_k K_k(x_k,y_k)K(x,y)=∏k​Kk​(xk​,yk​), and let Γ\GammaΓ be the matrix whose entry between configurations (rk,zk)k(r_k,z_k)_k(rk​,zk​)k​ and (sk,wk)k(s_k,w_k)_k(sk​,wk​)k​, expressed through the λk\lambda_kλk​, is

Γ=1d∑α,β<d1∀k, rk=α, sk=β (UαρUβ†)(z,w).\Gamma=\frac1d\sum_{\alpha,\beta<d} \mathbf1_{\forall k,\ r_k=\alpha,\ s_k=\beta}\, (U_\alpha\rho U_\beta^\dagger)(z,w).Γ=d1​α,β<d∑​1∀k, rk​=α, sk​=β​(Uα​ρUβ†​)(z,w).

Finally, set v±(x)=2−1/2(1∀k, xk=i±1∀k, xk=j)v_\pm(x)=2^{-1/2}\bigl(\mathbf1_{\forall k,\ x_k=i}\pm\mathbf1_{\forall k,\ x_k=j}\bigr)v±​(x)=2−1/2(1∀k, xk​=i​±1∀k, xk​=j​) and Ω=pv+v+†+(1−p)v−v−†\Omega=pv_+v_+^\dagger+(1-p)v_-v_-^\daggerΩ=pv+​v+†​+(1−p)v−​v−†​. The conjunction is

η>0,a>0,b>0,η≤ab,0<q≤1,12<p≤1,I−Kk†Kk⪰0 for every k,KΓK†=qΩ,Re⁡Tr⁡(KΓK†)=q,Y>0,ofReal⁡(Y)≤DGHZ(Γ).\eta>0,\quad a>0,\quad b>0,\quad \eta\le\sqrt{ab},\quad 0<q\le1,\quad \tfrac12<p\le1,\quad I-K_k^\dagger K_k\succeq0\ \text{for every }k,\quad K\Gamma K^\dagger=q\Omega,\quad \operatorname{Re}\operatorname{Tr}(K\Gamma K^\dagger)=q,\quad Y>0,\quad \operatorname{ofReal}(Y)\le D_{\rm GHZ}(\Gamma).η>0,a>0,b>0,η≤ab​,0<q≤1,21​<p≤1,I−Kk†​Kk​⪰0 for every k,KΓK†=qΩ,ReTr(KΓK†)=q,Y>0,ofReal(Y)≤DGHZ​(Γ).

Here DGHZ(Γ)∈[0,∞]D_{\rm GHZ}(\Gamma)\in[0,\infty]DGHZ​(Γ)∈[0,∞] is the supremum of max⁡(r,0)\max(r,0)max(r,0) over real r≥0r\ge0r≥0 such that, for every ε,δ>0\varepsilon,\delta>0ε,δ>0, all sufficiently large positive integers nnn admit an integer m≥0m\ge0m≥0 and a density matrix σ\sigmaσ on local dimensions 2m2^m2m, obtained deterministically from Γ⊗n\Gamma^{\otimes n}Γ⊗n by the following finite local-operation relation, with ∥σ−GHZN⊗m∥1≤ε\|\sigma-\mathrm{GHZ}_N^{\otimes m}\|_1\le\varepsilon∥σ−GHZN⊗m​∥1​≤ε and r−δ≤m/nr-\delta\le m/nr−δ≤m/n. That relation starts with the unchanged state, permits finite local Kraus instruments whose operators satisfy global completeness and whose branches satisfy pjρj=∑zAjzρAjz†p_j\rho_j=\sum_z A_{jz}\rho A_{jz}^\daggerpj​ρj​=∑z​Ajz​ρAjz†​, permits recursive branch-dependent operations with multiplied probabilities, and permits forgetting a finite output list by replacing it with its weighted mixture when all output dimensions agree. The target uses GHZN=vv†\mathrm{GHZ}_N=vv^\daggerGHZN​=vv†, v(x)=2−1/2(1∀k,xk=0+1∀k,xk=1)v(x)=2^{-1/2}(\mathbf1_{\forall k,x_k=0}+\mathbf1_{\forall k,x_k=1})v(x)=2−1/2(1∀k,xk​=0​+1∀k,xk​=1​); tensor powers are regrouped separately at each party through chosen finite-cardinality equivalences, and ∥M∥1=Re⁡Tr⁡M†M\|M\|_1=\operatorname{Re}\operatorname{Tr}\sqrt{M^\dagger M}∥M∥1​=ReTrM†M​. Zero parties, zero local dimensions, and equal labels are included wherever the supplied arguments exist; empty products equal one, and the conjunction itself imposes the displayed positivity conditions.

QAlgorithms.Papers.Independent.Gap48.P42.PartyBasis

For any natural number NNN and any assignment of natural numbers DkD_kDk​ to indices k∈{0,…,N−1}k\in\{0,\ldots,N-1\}k∈{0,…,N−1}, PartyBasis⁡(D)\operatorname{PartyBasis}(D)PartyBasis(D) is the set of tuples x=(xk)kx=(x_k)_kx=(xk​)k​ with xk∈{0,…,Dk−1}x_k\in\{0,\ldots,D_k-1\}xk​∈{0,…,Dk​−1}. Zero dimensions are permitted: this set is empty if some Dk=0D_k=0Dk​=0. When N=0N=0N=0, it consists of the unique empty tuple.

QAlgorithms.Papers.Independent.Gap48.P42.ProductVectors

For any natural number NNN and natural local dimensions SkS_kSk​, ProductVectors⁡(S)\operatorname{ProductVectors}(S)ProductVectors(S) is the type of families f=(fk)k<Nf=(f_k)_{k<N}f=(fk​)k<N​, where each fkf_kfk​ is a complex Euclidean vector indexed by {0,…,Sk−1}\{0,\ldots,S_k-1\}{0,…,Sk​−1}. This abbreviation imposes no normalization or nonzero condition and does not itself form a tensor product. If Sk=0S_k=0Sk​=0, that component is the unique zero vector in the zero-dimensional space; if N=0N=0N=0, there is exactly one such family, the empty family.

QAlgorithms.Papers.Independent.Gap48.P42.UnitFactors

For any natural number NNN, any dimensions Sk∈NS_k\in\mathbb NSk​∈N indexed by k∈[N]={0,…,N−1}k\in[N]=\{0,\ldots,N-1\}k∈[N]={0,…,N−1}, and any family f=(fk)k∈[N]f=(f_k)_{k\in[N]}f=(fk​)k∈[N]​ with fk∈CSkf_k\in\mathbb C^{S_k}fk​∈CSk​, this predicate means that every factor has Euclidean norm one:

∀k∈[N],∥fk∥=1.\forall k\in[N],\qquad \|f_k\|=1.∀k∈[N],∥fk​∥=1.

For N=0N=0N=0 the condition is vacuous. If some Sk=0S_k=0Sk​=0, it cannot hold, since the corresponding vector space contains only the zero vector.

QAlgorithms.Papers.Independent.Gap48.P42.asTwoParties

For all natural numbers a,ba,ba,b and every complex matrix ρ\rhoρ indexed by Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), this defines the same entries using configurations of two indexed parties: a configuration xxx has x0∈Fin⁡(a)x_0\in\operatorname{Fin}(a)x0​∈Fin(a) and x1∈Fin⁡(b)x_1\in\operatorname{Fin}(b)x1​∈Fin(b), and the resulting matrix satisfies ρ~(x,y)=ρ((x0,x1),(y0,y1))\widetilde\rho(x,y)=\rho((x_0,x_1),(y_0,y_1))ρ​(x,y)=ρ((x0​,x1​),(y0​,y1​)). No positivity or normalization is required, and a=0a=0a=0 or b=0b=0b=0 is allowed.

QAlgorithms.Papers.Independent.Gap48.P42.bestPairYield

For arbitrary natural numbers N,dN,dN,d, dimensions Sk∈NS_k\in\mathbb NSk​∈N, a complex matrix ρ\rhoρ on BS=∏k<NFin⁡(Sk)\mathcal B_S=\prod_{k<N}\operatorname{Fin}(S_k)BS​=∏k<N​Fin(Sk​), unitaries UiU_iUi​ on that space for i<di<di<d, and an assignment vvv choosing two local-vector families fkij,gkij∈CSkf^{ij}_k,g^{ij}_k\in\mathbb C^{S_k}fkij​,gkij​∈CSk​ for every 0≤i<j<d0\le i<j<d0≤i<j<d, this defines the real supremum, over those pairs, of

2min⁡(aij,bij)d[1−H2(12+ηij2aijbij)].\frac{2\min(a_{ij},b_{ij})}{d} \left[1-H_2\left(\frac12+ \frac{\eta_{ij}}{2\sqrt{a_{ij}b_{ij}}}\right)\right].d2min(aij​,bij​)​[1−H2​(21​+2aij​bij​​ηij​​)].

Here aij=Re⁡⟨Fij,UiρUi†Fij⟩a_{ij}=\operatorname{Re}\langle F^{ij},U_i\rho U_i^\dagger F^{ij}\rangleaij​=Re⟨Fij,Ui​ρUi†​Fij⟩, bij=Re⁡⟨Gij,UjρUj†Gij⟩b_{ij}=\operatorname{Re}\langle G^{ij},U_j\rho U_j^\dagger G^{ij}\ranglebij​=Re⟨Gij,Uj​ρUj†​Gij⟩, Fij(x)=∏kfkij(xk)F^{ij}(x)=\prod_k f^{ij}_k(x_k)Fij(x)=∏k​fkij​(xk​), Gij(x)=∏kgkij(xk)G^{ij}(x)=\prod_k g^{ij}_k(x_k)Gij(x)=∏k​gkij​(xk​), and ηij\eta_{ij}ηij​ is the real supremum of ∣⟨H,UiρUj†L⟩∣|\langle H,U_i\rho U_j^\dagger L\rangle|∣⟨H,Ui​ρUj†​L⟩∣ over product vectors H(x)=∏khk(xk)H(x)=\prod_k h_k(x_k)H(x)=∏k​hk​(xk​), L(x)=∏kℓk(xk)L(x)=\prod_k\ell_k(x_k)L(x)=∏k​ℓk​(xk​) with every local factor of norm one. The chosen families fij,gijf^{ij},g^{ij}fij,gij need not be normalized or attain that supremum, and ρ\rhoρ need not be positive or have trace one. The entropy function is H2(t)=[−tlog⁡t−(1−t)log⁡(1−t)]/log⁡2H_2(t)=[-t\log t-(1-t)\log(1-t)]/\log2H2​(t)=[−tlogt−(1−t)log(1−t)]/log2, with the total real-logarithm convention log⁡0=0\log0=0log0=0 and log⁡t=log⁡∣t∣\log t=\log|t|logt=log∣t∣ for negative ttt; real square roots of negative numbers and division by zero are zero. Thus no admissibility or positivity conditions are imposed on these yield expressions. The supremum is zero for d<2d<2d<2, when there are no pairs; an empty set of unit local-factor families likewise gives zero for the corresponding real supremum.

QAlgorithms.Papers.Independent.Gap48.P42.bipartiteDistillable

For natural numbers a,ba,ba,b and any complex matrix ρ\rhoρ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), this defines an extended nonnegative real number as the supremum of max⁡(r,0)\max(r,0)max(r,0) over real rates r≥0r\ge0r≥0 satisfying the following condition: for every ε,δ>0\varepsilon,\delta>0ε,δ>0, there is n0∈Nn_0\in\mathbb Nn0​∈N such that every n≥n0n\ge n_0n≥n0​ with n>0n>0n>0 admits m∈Nm\in\mathbb Nm∈N and a density matrix σ\sigmaσ on two local spaces of dimension 2m2^m2m, deterministically obtainable from the regrouped tensor power ρ⊗n\rho^{\otimes n}ρ⊗n, such that ∥σ−Φ⊗m∥1≤ε\|\sigma-\Phi^{\otimes m}\|_1\le\varepsilon∥σ−Φ⊗m∥1​≤ε and r−δ≤m/nr-\delta\le m/nr−δ≤m/n, where Φ=vv†\Phi=vv^\daggerΦ=vv† and v=(∣00⟩+∣11⟩)/2v=(|00\rangle+|11\rangle)/\sqrt2v=(∣00⟩+∣11⟩)/2​. The source is represented as two indexed parties by replacing ρ((x0,x1),(y0,y1))\rho((x_0,x_1),(y_0,y_1))ρ((x0​,x1​),(y0​,y1​)) with the entry indexed by configurations x,yx,yx,y. Tensor copies are regrouped at each party through chosen equivalences (Fin⁡(a))n≃Fin⁡(an)(\operatorname{Fin}(a))^n\simeq\operatorname{Fin}(a^n)(Fin(a))n≃Fin(an) and similarly for bbb, and ∥X∥1=Re⁡Tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{Tr}\sqrt{X^\dagger X}∥X∥1​=ReTrX†X​. Deterministic obtainability means the finite local-operation relation generated by leaving a state unchanged, finite local Kraus instruments with complete Kraus families and branch identities pjρj=∑zAjzρAjz†p_j\rho_j=\sum_z A_{jz}\rho A_{jz}^\daggerpj​ρj​=∑z​Ajz​ρAjz†​, recursive branch-dependent operations with multiplied probabilities, and forgetting an output list by taking its weighted mixture when all output dimensions agree; the final list must consist solely of (1,σ)(1,\sigma)(1,σ). Every state appearing in this relation has positive local dimensions and is positive semidefinite with trace one. No such assumptions are imposed on the argument ρ\rhoρ itself; if no rate satisfies the condition, the supremum is zero. The choices m=0m=0m=0 and a=0a=0a=0 or b=0b=0b=0 are included, although zero source dimensions cannot satisfy the conversion requirement for positive nnn.

QAlgorithms.Papers.Independent.Gap48.P42.bipartitePrivate

For arbitrary natural numbers d,a,bd,a,bd,a,b, any complex matrix ρ\rhoρ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), and unitaries UαU_\alphaUα​ on that space for α<d\alpha<dα<d, this defines the matrix on Fin⁡(da)×Fin⁡(db)\operatorname{Fin}(da)\times\operatorname{Fin}(db)Fin(da)×Fin(db) with entries

Γ(α,u;β,v),(α′,u′;β′,v′)={d−1(UαρUα′†)(u,v),(u′,v′),α=β and α′=β′,0,otherwise.\Gamma_{(\alpha,u;\beta,v),(\alpha',u';\beta',v')}= \begin{cases} d^{-1}(U_\alpha\rho U_{\alpha'}^\dagger)_{(u,v),(u',v')}, &\alpha=\beta\ \text{and }\alpha'=\beta',\\ 0,&\text{otherwise}. \end{cases}Γ(α,u;β,v),(α′,u′;β′,v′)​={d−1(Uα​ρUα′†​)(u,v),(u′,v′)​,0,​α=β and α′=β′,otherwise.​

The displayed key-and-shield coordinates are converted to the actual indices using the chosen finite-cardinality equivalences Fin⁡(d)×Fin⁡(a)≃Fin⁡(da)\operatorname{Fin}(d)\times\operatorname{Fin}(a)\simeq\operatorname{Fin}(da)Fin(d)×Fin(a)≃Fin(da) and Fin⁡(d)×Fin⁡(b)≃Fin⁡(db)\operatorname{Fin}(d)\times\operatorname{Fin}(b)\simeq\operatorname{Fin}(db)Fin(d)×Fin(b)≃Fin(db). The input matrix is not assumed to be a density matrix. Zero dimensions are permitted and produce an empty output index set.

QAlgorithms.Papers.Independent.Gap48.P42.bipartitePrivateRaw

For arbitrary natural numbers d,a,bd,a,bd,a,b, any complex matrix ρ\rhoρ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), and any family of complex matrices UαU_\alphaUα​ on that same space indexed by α<d\alpha<dα<d, this defines a matrix Γ\GammaΓ on Fin⁡(da)×Fin⁡(db)\operatorname{Fin}(da)\times\operatorname{Fin}(db)Fin(da)×Fin(db). Let λd,a:Fin⁡(d)×Fin⁡(a)≃Fin⁡(da)\lambda_{d,a}:\operatorname{Fin}(d)\times\operatorname{Fin}(a)\simeq\operatorname{Fin}(da)λd,a​:Fin(d)×Fin(a)≃Fin(da) and λd,b\lambda_{d,b}λd,b​ be the chosen equivalences obtained from equality of finite cardinalities. Decode its row as (λd,a(α,u),λd,b(β,v))(\lambda_{d,a}(\alpha,u),\lambda_{d,b}(\beta,v))(λd,a​(α,u),λd,b​(β,v)) and its column as (λd,a(α′,u′),λd,b(β′,v′))(\lambda_{d,a}(\alpha',u'),\lambda_{d,b}(\beta',v'))(λd,a​(α′,u′),λd,b​(β′,v′)). Then

Γ(α,u;β,v),(α′,u′;β′,v′)={d−1(UαρUα′†)(u,v),(u′,v′),α=β and α′=β′,0,otherwise.\Gamma_{(\alpha,u;\beta,v),(\alpha',u';\beta',v')}= \begin{cases} d^{-1}(U_\alpha\rho U_{\alpha'}^\dagger)_{(u,v),(u',v')}, &\alpha=\beta\ \text{and }\alpha'=\beta',\\ 0,&\text{otherwise}. \end{cases}Γ(α,u;β,v),(α′,u′;β′,v′)​={d−1(Uα​ρUα′†​)(u,v),(u′,v′)​,0,​α=β and α′=β′,otherwise.​

There is no positivity, trace, or unitarity assumption. The equivalences are chosen by finite cardinality rather than specified arithmetic encodings. Zero values of d,a,bd,a,bd,a,b are allowed and give an empty output index set.

QAlgorithms.Papers.Independent.Gap48.P42.distillableGHZ

For any natural number NNN, natural dimension assignment DkD_kDk​, and arbitrary complex matrix ρ\rhoρ on ∏k<N{0,…,Dk−1}\prod_{k<N}\{0,\ldots,D_k-1\}∏k<N​{0,…,Dk​−1}, this definition is an extended nonnegative real number: the supremum of max⁡(r,0)\max(r,0)max(r,0) over real rates rrr satisfying the following property. The rate is nonnegative, and for every ϵ,δ>0\epsilon,\delta>0ϵ,δ>0 there is a natural n0n_0n0​ such that, for every natural n≥n0n\ge n_0n≥n0​ with n>0n>0n>0, some natural mmm and density matrix σ\sigmaσ on NNN local spaces of dimension 2m2^m2m obey r−δ≤m/nr-\delta\le m/nr−δ≤m/n and Re⁡tr⁡(σ−GN,m)†(σ−GN,m)≤ϵ\operatorname{Re}\operatorname{tr}\sqrt{(\sigma-G_{N,m})^\dagger(\sigma-G_{N,m})}\le\epsilonRetr(σ−GN,m​)†(σ−GN,m​)​≤ϵ, and admit an exact finite protocol conversion from ρ[n]\rho^{[n]}ρ[n] to σ\sigmaσ. Here ρ[n]\rho^{[n]}ρ[n] has entries given by products of nnn entries of ρ\rhoρ, with each local index decoded through the fixed cardinality-chosen bijection between {0,…,Dkn−1}\{0,\ldots,D_k^n-1\}{0,…,Dkn​−1} and length-nnn local tuples; GN,mG_{N,m}GN,m​ is the grouped mmm-fold tensor power of gg†gg^\daggergg†, where g(z)=(1{∀k, zk=0}+1{∀k, zk=1})/2g(z)=\bigl(\mathbf1_{\{\forall k,\ z_k=0\}}+\mathbf1_{\{\forall k,\ z_k=1\}}\bigr)/\sqrt2g(z)=(1{∀k, zk​=0}​+1{∀k, zk​=1}​)/2​. The exact conversion requires positive local dimensions and density matrices at its endpoints and ends in the singleton weighted output [(1,σ)][(1,\sigma)][(1,σ)]. Its finite protocol relation is generated by stopping, finite local instruments followed by outcome-dependent protocols, and forgetting an output list through its weighted mixture when all dimensions agree. Instruments have nonnegative weights pjp_jpj​ summing to 111, density-matrix branches, and one-party Kraus operators lifted by identities on the other parties, satisfying ∑j,zKj,z†Kj,z=I\sum_{j,z}K_{j,z}^\dagger K_{j,z}=I∑j,z​Kj,z†​Kj,z​=I and pjρj=∑zKj,zρinKj,z†p_j\rho_j=\sum_zK_{j,z}\rho_{\mathrm{in}}K_{j,z}^\daggerpj​ρj​=∑z​Kj,z​ρin​Kj,z†​; successive weights multiply. No positivity or normalization is separately assumed for ρ\rhoρ. The supremum is 000 if there are no eligible rates and may be +∞+\infty+∞. Zero parties and zero output-copy counts are included; if some Dk=0D_k=0Dk​=0, no rate is eligible and the value is 000.

QAlgorithms.Papers.Independent.Gap48.P42.distillableKey

For all natural numbers a,ba,ba,b and every complex matrix ρ\rhoρ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), this defines the supremum in [0,∞][0,\infty][0,∞] of max⁡(r,0)\max(r,0)max(r,0) over all real r≥0r\ge0r≥0 with this property: for every ε,δ>0\varepsilon,\delta>0ε,δ>0, all sufficiently large positive integers nnn admit positive integers d,s,td,s,td,s,t, a density matrix τ\tauτ on local dimensions s,ts,ts,t, unitaries UαU_\alphaUα​ for α<d\alpha<dα<d, and a finite local-operation map LLL satisfying ∥L(ρ⊗n)−Γτ,U∥1≤ε\|L(\rho^{\otimes n})-\Gamma_{\tau,U}\|_1\le\varepsilon∥L(ρ⊗n)−Γτ,U​∥1​≤ε and r−δ≤log⁡2(d)/nr-\delta\le\log_2(d)/nr−δ≤log2​(d)/n. A density matrix means a positive semidefinite matrix of trace one. The permitted maps are generated recursively from identity maps by finite complete local Kraus measurements on either party and outcome-dependent permitted maps, summing the output branches. The target Γτ,U\Gamma_{\tau,U}Γτ,U​, on dimensions ds,dtds,dtds,dt, has entries d−1(UατUα′†)(u,v),(u′,v′)d^{-1}(U_\alpha\tau U_{\alpha'}^\dagger)_{(u,v),(u',v')}d−1(Uα​τUα′†​)(u,v),(u′,v′)​ when both row key labels equal α\alphaα and both column key labels equal α′\alpha'α′, and zero otherwise; key-and-shield coordinates are identified with the output indices by chosen finite-cardinality equivalences. The source tensor power is likewise regrouped separately at the two parties using chosen equivalences, and ∥X∥1=Re⁡Tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{Tr}\sqrt{X^\dagger X}∥X∥1​=ReTrX†X​. There is no density-matrix assumption on ρ\rhoρ, zero values of a,ba,ba,b are allowed, and the supremum is zero if there are no qualifying rates; it may otherwise be infinite.

QAlgorithms.Papers.Independent.Gap48.P42.filterSuccess

For every natural number ddd and arbitrary real numbers a,ba,ba,b, this defines

2min⁡(a,b)d.\frac{2\min(a,b)}{d}.d2min(a,b)​.

Division is total, with division by zero equal to zero, so the value is 000 when d=0d=0d=0. There are no sign or magnitude restrictions on a,ba,ba,b, and the definition does not require its value to lie in [0,1][0,1][0,1].

QAlgorithms.Papers.Independent.Gap48.P42.filterYield

For every d∈Nd\in\mathbb Nd∈N and arbitrary real numbers η,a,b\eta,a,bη,a,b, this defines

2min⁡(a,b)d[1−H2 ⁣(12+η2ab)],H2(p)=−plog⁡p−(1−p)log⁡(1−p)log⁡2.\frac{2\min(a,b)}{d}\left[ 1-H_2\!\left(\frac12+\frac{\eta}{2\sqrt{ab}}\right) \right], \qquad H_2(p)=\frac{-p\log p-(1-p)\log(1-p)}{\log 2}.d2min(a,b)​[1−H2​(21​+2ab​η​)],H2​(p)=log2−plogp−(1−p)log(1−p)​.

These are total real operations: square roots of negative numbers and divisions by zero are zero; the real logarithm is defined at zero as zero and at negative arguments using their absolute values. There are no positivity or probability-range assumptions. In particular, d=0d=0d=0 gives value 000, and ab≤0ab\le0ab≤0 gives entropy argument 1/21/21/2 and hence value 000.

QAlgorithms.Papers.Independent.Gap48.P42.filteredBranch

For any N,d∈NN,d\in\mathbb NN,d∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, arbitrary complex matrix ρ\rhoρ indexed by BS=∏k[Sk]\mathcal B_S=\prod_k[S_k]BS​=∏k​[Sk​], unitary matrices UrU_rUr​ on this index set for r∈[d]r\in[d]r∈[d], indices i,j∈[d]i,j\in[d]i,j∈[d], and families fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, put Br,s=UrρUs†B_{r,s}=U_r\rho U_s^\daggerBr,s​=Ur​ρUs†​ and

a=Re⁡∑u,v∈BS∏kfk(uk)‾(Bi,i)u,v∏kfk(vk),b=Re⁡∑u,v∈BS∏kgk(uk)‾(Bj,j)u,v∏kgk(vk),c=∑u,v∈BS∏kfk(uk)‾(Bi,j)u,v∏kgk(vk).a=\operatorname{Re}\sum_{u,v\in\mathcal B_S} \overline{\prod_k f_k(u_k)}(B_{i,i})_{u,v}\prod_k f_k(v_k), \quad b=\operatorname{Re}\sum_{u,v\in\mathcal B_S} \overline{\prod_k g_k(u_k)}(B_{j,j})_{u,v}\prod_k g_k(v_k), \quad c=\sum_{u,v\in\mathcal B_S} \overline{\prod_k f_k(u_k)}(B_{i,j})_{u,v}\prod_k g_k(v_k).a=Reu,v∈BS​∑​k∏​fk​(uk​)​(Bi,i​)u,v​k∏​fk​(vk​),b=Reu,v∈BS​∑​k∏​gk​(uk​)​(Bj,j​)u,v​k∏​gk​(vk​),c=u,v∈BS​∑​k∏​fk​(uk​)​(Bi,j​)u,v​k∏​gk​(vk​).

For a configuration y∈∏k[dSk]y\in\prod_k[dS_k]y∈∏k​[dSk​], write (rk,uk)=λd,Sk−1(yk)(r_k,u_k)=\lambda_{d,S_k}^{-1}(y_k)(rk​,uk​)=λd,Sk​−1​(yk​), where λd,Sk\lambda_{d,S_k}λd,Sk​​ is the chosen bijection from [d]×[Sk][d]\times[S_k][d]×[Sk​] to [dSk][dS_k][dSk​]. Define a matrix HHH on these configurations by

Hy,y′=1d∑r,s∈[d]1{∀k, rk=r, rk′=s} (Br,s)u,u′,H_{y,y'}=\frac1d\sum_{r,s\in[d]} \mathbf1_{\{\forall k,\ r_k=r,\ r'_k=s\}}\, (B_{r,s})_{u,u'},Hy,y′​=d1​r,s∈[d]∑​1{∀k, rk​=r, rk′​=s}​(Br,s​)u,u′​,

and define the matrix KKK, with rows x∈[d]Nx\in[d]^Nx∈[d]N, by

Kx,y=∏k(1{xk=i, rk=i}Akfk(uk)‾+1{xk=j, rk=j}Ckgk(uk)‾),Ak={min⁡(a,b)/a,k=0,1,k≠0,Ck={min⁡(a,b)/b  c/∣c∣,k=0,1,k≠0.K_{x,y}=\prod_k \left( \mathbf1_{\{x_k=i,\ r_k=i\}}A_k\overline{f_k(u_k)} + \mathbf1_{\{x_k=j,\ r_k=j\}}C_k\overline{g_k(u_k)} \right), \quad A_k=\begin{cases}\sqrt{\min(a,b)/a},&k=0,\\1,&k\ne0,\end{cases} \quad C_k=\begin{cases}\sqrt{\min(a,b)/b}\;c/|c|,&k=0,\\1,&k\ne0.\end{cases}Kx,y​=k∏​(1{xk​=i, rk​=i}​Ak​fk​(uk​)​+1{xk​=j, rk​=j}​Ck​gk​(uk​)​),Ak​={min(a,b)/a​,1,​k=0,k=0,​Ck​={min(a,b)/b​c/∣c∣,1,​k=0,k=0.​

The declaration defines the output matrix to be KHK†KHK^\daggerKHK†, indexed by [d]N[d]^N[d]N. No density-matrix condition on ρ\rhoρ, unit-norm condition on the factors, or distinctness of i,ji,ji,j is assumed. The indices require d>0d>0d>0; N=0N=0N=0 and zero local dimensions remain permitted, with empty products equal to 111, empty sums equal to 000, and universal conditions over no parties vacuous. Division by zero and square roots of negative real arguments are zero.

QAlgorithms.Papers.Independent.Gap48.P42.ghzCopies

For any natural numbers N,mN,mN,m, this definition gives a complex matrix indexed by NNN-tuples of elements of {0,…,2m−1}\{0,\ldots,2^m-1\}{0,…,2m−1}. Let cmc_mcm​ be the cardinality-chosen bijection from length-mmm tuples over {0,1}\{0,1\}{0,1} to {0,…,2m−1}\{0,\ldots,2^m-1\}{0,…,2m−1}, and put g(z)=(1{∀k, zk=0}+1{∀k, zk=1})/2g(z)=\bigl(\mathbf1_{\{\forall k,\ z_k=0\}}+\mathbf1_{\{\forall k,\ z_k=1\}}\bigr)/\sqrt2g(z)=(1{∀k, zk​=0}​+1{∀k, zk​=1}​)/2​ for binary NNN-tuples zzz. The entries are

GN,m(x,y)=∏t=0m−1g ⁣((cm−1(xk)t)k<N)g ⁣((cm−1(yk)t)k<N)‾.G_{N,m}(x,y)= \prod_{t=0}^{m-1} g\!\left((c_m^{-1}(x_k)_t)_{k<N}\right) \overline{g\!\left((c_m^{-1}(y_k)_t)_{k<N}\right)}.GN,m​(x,y)=t=0∏m−1​g((cm−1​(xk​)t​)k<N​)g((cm−1​(yk​)t​)k<N​)​.

Thus the construction takes the entrywise grouped mmm-fold tensor power of the matrix g(z)g(w)‾g(z)\overline{g(w)}g(z)g(w)​, using those chosen index bijections. Both N=0N=0N=0 and m=0m=0m=0 are included: for m=0m=0m=0 the singleton-indexed matrix has entry 111, while for N=0N=0N=0 its sole entry is 2m2^m2m.

QAlgorithms.Papers.Independent.Gap48.P42.indexedPrivateDit

For any natural numbers N,dN,dN,d, natural dimensions SkS_kSk​, arbitrary complex matrix ρ\rhoρ indexed by ∏k<N{0,…,Sk−1}\prod_{k<N}\{0,\ldots,S_k-1\}∏k<N​{0,…,Sk​−1}, and family of unitary matrices UiU_iUi​ on that same space indexed by 0≤i<d0\le i<d0≤i<d, this definition gives a matrix indexed by ∏k<N{0,…,dSk−1}\prod_{k<N}\{0,\ldots,dS_k-1\}∏k<N​{0,…,dSk​−1}. Decode each row coordinate xkx_kxk​ and column coordinate yky_kyk​ through the inverses of the cardinality-chosen bijections localIndex⁡(d,Sk)\operatorname{localIndex}(d,S_k)localIndex(d,Sk​), writing the decoded pairs as (αk,sk)(\alpha_k,s_k)(αk​,sk​) and (βk,tk)(\beta_k,t_k)(βk​,tk​). The matrix entry is

∑a=0d−1∑b=0d−1{d−1(UaρUb†)s,t,if αk=a and βk=b for every k<N,0,otherwise.\sum_{a=0}^{d-1}\sum_{b=0}^{d-1} \begin{cases} d^{-1}(U_a\rho U_b^\dagger)_{s,t}, &\text{if }\alpha_k=a\text{ and }\beta_k=b\text{ for every }k<N,\\ 0,&\text{otherwise}. \end{cases}a=0∑d−1​b=0∑d−1​{d−1(Ua​ρUb†​)s,t​,0,​if αk​=a and βk​=b for every k<N,otherwise.​

Here d−1d^{-1}d−1 is interpreted in C\mathbb CC, with 0−1=00^{-1}=00−1=0, and Ui†U_i^\daggerUi†​ denotes conjugate transpose. No density-matrix assumption is imposed on ρ\rhoρ, and no positivity assumption is imposed on ddd or the dimensions SkS_kSk​. If d=0d=0d=0, the defining sum is empty and the result is the zero matrix. If N=0N=0N=0, the conditions on the key coordinates are vacuous, so the sole entry is d−1∑a,b(UaρUb†)d^{-1}\sum_{a,b}(U_a\rho U_b^\dagger)d−1∑a,b​(Ua​ρUb†​).

QAlgorithms.Papers.Independent.Gap48.P42.jointFilter

For any N,d∈NN,d\in\mathbb NN,d∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, indices i,j∈[d]i,j\in[d]i,j∈[d], families fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, real numbers a,ba,ba,b, and complex number ccc, this defines a matrix with rows x∈[d]Nx\in[d]^Nx∈[d]N and columns y∈∏k[dSk]y\in\prod_k[dS_k]y∈∏k​[dSk​]. For each kkk, write (rk,sk)=λd,Sk−1(yk)(r_k,s_k)=\lambda_{d,S_k}^{-1}(y_k)(rk​,sk​)=λd,Sk​−1​(yk​), using the chosen finite-cardinality bijection λd,Sk:[d]×[Sk]→[dSk]\lambda_{d,S_k}:[d]\times[S_k]\to[dS_k]λd,Sk​​:[d]×[Sk​]→[dSk​]. The entry is

Kx,y=∏k∈[N](1{xk=i, rk=i}Akfk(sk)‾+1{xk=j, rk=j}Bkgk(sk)‾),Ak={min⁡(a,b)/a,k=0,1,k≠0,Bk={min⁡(a,b)/b  c/∣c∣,k=0,1,k≠0.K_{x,y}= \prod_{k\in[N]} \left( \mathbf1_{\{x_k=i,\ r_k=i\}}A_k\overline{f_k(s_k)} + \mathbf1_{\{x_k=j,\ r_k=j\}}B_k\overline{g_k(s_k)} \right), \quad A_k= \begin{cases}\sqrt{\min(a,b)/a},&k=0,\\1,&k\ne0,\end{cases} \quad B_k= \begin{cases}\sqrt{\min(a,b)/b}\;c/|c|,&k=0,\\1,&k\ne0.\end{cases}Kx,y​=k∈[N]∏​(1{xk​=i, rk​=i}​Ak​fk​(sk​)​+1{xk​=j, rk​=j}​Bk​gk​(sk​)​),Ak​={min(a,b)/a​,1,​k=0,k=0,​Bk​={min(a,b)/b​c/∣c∣,1,​k=0,k=0.​

No normalization, positivity, or distinctness of i,ji,ji,j is required. All divisions and square roots are total, with division by zero and square roots of negative arguments equal to zero. The indices require d>0d>0d>0, but N=0N=0N=0 is allowed: both matrix index sets then have one element and the empty product gives entry 111. For N>0N>0N>0, any zero local dimension makes the column index set empty.

QAlgorithms.Papers.Independent.Gap48.P42.localFilter

For any N,d∈NN,d\in\mathbb NN,d∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, indices i,j∈[d]i,j\in[d]i,j∈[d], vector families fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, real numbers a,ba,ba,b, complex number ccc, and party k∈[N]k\in[N]k∈[N], this defines a complex matrix with rows x∈[d]x\in[d]x∈[d] and columns y∈[dSk]y\in[dS_k]y∈[dSk​]. Let (r,s)=λd,Sk−1(y)(r,s)=\lambda_{d,S_k}^{-1}(y)(r,s)=λd,Sk​−1​(y), where λd,Sk:[d]×[Sk]→[dSk]\lambda_{d,S_k}:[d]\times[S_k]\to[dS_k]λd,Sk​​:[d]×[Sk​]→[dSk​] is the chosen bijection supplied by equality of the finite cardinalities. Its entry is

1{x=i, r=i}Ak fk(s)‾+1{x=j, r=j}Bk gk(s)‾,Ak={min⁡(a,b)/a,k=0,1,k≠0,Bk={min⁡(a,b)/b  c/∣c∣,k=0,1,k≠0.\mathbf1_{\{x=i,\ r=i\}}A_k\,\overline{f_k(s)} + \mathbf1_{\{x=j,\ r=j\}}B_k\,\overline{g_k(s)}, \quad A_k= \begin{cases} \sqrt{\min(a,b)/a},&k=0,\\ 1,&k\ne0, \end{cases} \quad B_k= \begin{cases} \sqrt{\min(a,b)/b}\;c/|c|,&k=0,\\ 1,&k\ne0. \end{cases}1{x=i, r=i}​Ak​fk​(s)​+1{x=j, r=j}​Bk​gk​(s)​,Ak​={min(a,b)/a​,1,​k=0,k=0,​Bk​={min(a,b)/b​c/∣c∣,1,​k=0,k=0.​

Here indicators are 111 when their conditions hold and 000 otherwise. No normalization, positivity, or condition i≠ji\ne ji=j is imposed; if i=ji=ji=j, both terms can contribute to the same entry. Square roots of negative arguments and divisions by zero are zero, including c/∣c∣=0c/|c|=0c/∣c∣=0 when c=0c=0c=0. The supplied indices require N,d>0N,d>0N,d>0; Sk=0S_k=0Sk​=0 gives no columns.

QAlgorithms.Papers.Independent.Gap48.P42.localIndex

For every pair of natural numbers a,ba,ba,b, localIndex⁡(a,b)\operatorname{localIndex}(a,b)localIndex(a,b) is a bijection from {0,…,a−1}×{0,…,b−1}\{0,\ldots,a-1\}\times\{0,\ldots,b-1\}{0,…,a−1}×{0,…,b−1} to {0,…,ab−1}\{0,\ldots,ab-1\}{0,…,ab−1}, chosen using equality of their finite cardinalities. The definition specifies no explicit arithmetic formula for this bijection. If a=0a=0a=0 or b=0b=0b=0, both its domain and codomain are empty.

QAlgorithms.Papers.Independent.Gap48.P42.partyPower

For any natural number NNN, natural local dimensions DkD_kDk​, complex matrix ρ\rhoρ indexed by tuples in ∏k<N{0,…,Dk−1}\prod_{k<N}\{0,\ldots,D_k-1\}∏k<N​{0,…,Dk​−1}, and natural number nnn, this definition produces a complex matrix indexed by ∏k<N{0,…,Dkn−1}\prod_{k<N}\{0,\ldots,D_k^n-1\}∏k<N​{0,…,Dkn​−1}. For each kkk, let cDk,nc_{D_k,n}cDk​,n​ be the cardinality-chosen bijection from length-nnn tuples over {0,…,Dk−1}\{0,\ldots,D_k-1\}{0,…,Dk​−1} to {0,…,Dkn−1}\{0,\ldots,D_k^n-1\}{0,…,Dkn​−1}. Its entries are

partyPower⁡(ρ,n)x,y=∏t=0n−1ρ(cDk,n−1(xk)t)k<N,(cDk,n−1(yk)t)k<N.\operatorname{partyPower}(\rho,n)_{x,y} =\prod_{t=0}^{n-1} \rho_{\left(c_{D_k,n}^{-1}(x_k)_t\right)_{k<N}, \left(c_{D_k,n}^{-1}(y_k)_t\right)_{k<N}}.partyPower(ρ,n)x,y​=t=0∏n−1​ρ(cDk​,n−1​(xk​)t​)k<N​,(cDk​,n−1​(yk​)t​)k<N​​.

There is no positivity, Hermiticity, or normalization assumption on ρ\rhoρ. When n=0n=0n=0, every local dimension Dk0D_k^0Dk0​ is 111, including when Dk=0D_k=0Dk​=0, and the resulting singleton-indexed matrix has entry 111. When N=0N=0N=0, ρ\rhoρ has one entry and the result is that entry raised to the power nnn.

QAlgorithms.Papers.Independent.Gap48.P42.phaseParameter

For arbitrary real numbers η,a,b\eta,a,bη,a,b, this defines

12+η2ab.\frac12+\frac{\eta}{2\sqrt{ab}}.21​+2ab​η​.

The real square root is zero for nonpositive arguments, and division by zero is defined as zero. Consequently, when ab≤0ab\le0ab≤0, the value is 1/21/21/2. No assumptions restrict this expression to [0,1][0,1][0,1].

QAlgorithms.Papers.Independent.Gap48.P42.privateDiagonalWeight

For any N,d∈NN,d\in\mathbb NN,d∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, arbitrary complex matrix ρ\rhoρ indexed by BS=∏k∈[N][Sk]\mathcal B_S=\prod_{k\in[N]}[S_k]BS​=∏k∈[N]​[Sk​], family of unitary matrices UrU_rUr​ on this index set, index i∈[d]i\in[d]i∈[d], and vectors fk∈CSkf_k\in\mathbb C^{S_k}fk​∈CSk​, this defines the real number

Re⁡ ⁣(∑x,y∈BS∏kfk(xk)‾ (UiρUi†)x,y ∏kfk(yk)).\operatorname{Re}\!\left( \sum_{x,y\in\mathcal B_S} \overline{\prod_k f_k(x_k)}\, (U_i\rho U_i^\dagger)_{x,y}\, \prod_k f_k(y_k) \right).Re​x,y∈BS​∑​k∏​fk​(xk​)​(Ui​ρUi†​)x,y​k∏​fk​(yk​)​.

Neither normalization of the factors nor any density-matrix condition on ρ\rhoρ is required. The index iii requires d>0d>0d>0. If any local dimension is zero, the sums are empty and the value is 000; for N=0N=0N=0, the products are 111 and the value is the real part of the sole entry of UiρUi†U_i\rho U_i^\daggerUi​ρUi†​.

QAlgorithms.Papers.Independent.Gap48.P42.productElement

For any N∈NN\in\mathbb NN∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, complex matrix XXX indexed by the configurations BS=∏k∈[N][Sk]\mathcal B_S=\prod_{k\in[N]}[S_k]BS​=∏k∈[N]​[Sk​], and families fk,gk∈CSkf_k,g_k\in\mathbb C^{S_k}fk​,gk​∈CSk​, this defines the complex number

∑x∈BS∑y∈BS∏k∈[N]fk(xk)‾  Xx,y  ∏k∈[N]gk(yk).\sum_{x\in\mathcal B_S}\sum_{y\in\mathcal B_S} \overline{\prod_{k\in[N]}f_k(x_k)}\;X_{x,y}\; \prod_{k\in[N]}g_k(y_k).x∈BS​∑​y∈BS​∑​k∈[N]∏​fk​(xk​)​Xx,y​k∈[N]∏​gk​(yk​).

No normalization or positivity is required. When N=0N=0N=0, the configuration set has one element and both products are 111, so the result is the sole entry of XXX. If any Sk=0S_k=0Sk​=0, the configuration set is empty and the result is 000.

QAlgorithms.Papers.Independent.Gap48.P42.productOverlap

For any N∈NN\in\mathbb NN∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, and complex matrix XXX indexed by BS=∏k∈[N][Sk]\mathcal B_S=\prod_{k\in[N]}[S_k]BS​=∏k∈[N]​[Sk​], this defines the real supremum

sup⁡fk,gk∈CSk∥fk∥=∥gk∥=1 for every k∣∑x,y∈BS∏kfk(xk)‾ Xx,y∏kgk(yk)∣.\sup_{\substack{f_k,g_k\in\mathbb C^{S_k}\\ \|f_k\|=\|g_k\|=1\ \text{for every }k}} \left| \sum_{x,y\in\mathcal B_S} \overline{\prod_k f_k(x_k)}\,X_{x,y}\prod_k g_k(y_k) \right|.fk​,gk​∈CSk​∥fk​∥=∥gk​∥=1 for every k​sup​​x,y∈BS​∑​k∏​fk​(xk​)​Xx,y​k∏​gk​(yk​)​.

There is no assumption on XXX. If some Sk=0S_k=0Sk​=0, no admissible factor families exist, and the real supremum of this empty set is defined here as 000. For N=0N=0N=0, empty products are 111, and the value is the absolute value of the sole entry of XXX.

QAlgorithms.Papers.Independent.Gap48.P42.regroupIndex

For every d,a,m∈Nd,a,m\in\mathbb Nd,a,m∈N, this defines a function Fin⁡(dmam)→Fin⁡((da)m)\operatorname{Fin}(d^m a^m)\to\operatorname{Fin}((da)^m)Fin(dmam)→Fin((da)m) as follows. First decode the input through the chosen equivalence Fin⁡(dm)×Fin⁡(am)≃Fin⁡(dmam)\operatorname{Fin}(d^m)\times\operatorname{Fin}(a^m)\simeq\operatorname{Fin}(d^m a^m)Fin(dm)×Fin(am)≃Fin(dmam), obtaining (u,v)(u,v)(u,v). Decode uuu and vvv through the chosen finite-cardinality equivalences (Fin⁡(d))m≃Fin⁡(dm)(\operatorname{Fin}(d))^m\simeq\operatorname{Fin}(d^m)(Fin(d))m≃Fin(dm) and (Fin⁡(a))m≃Fin⁡(am)(\operatorname{Fin}(a))^m\simeq\operatorname{Fin}(a^m)(Fin(a))m≃Fin(am), obtaining sequences (uℓ)ℓ<m(u_\ell)_{\ell<m}(uℓ​)ℓ<m​ and (vℓ)ℓ<m(v_\ell)_{\ell<m}(vℓ​)ℓ<m​. At each coordinate, encode (uℓ,vℓ)(u_\ell,v_\ell)(uℓ​,vℓ​) through the chosen equivalence Fin⁡(d)×Fin⁡(a)≃Fin⁡(da)\operatorname{Fin}(d)\times\operatorname{Fin}(a)\simeq\operatorname{Fin}(da)Fin(d)×Fin(a)≃Fin(da), then encode the resulting length-mmm sequence through (Fin⁡(da))m≃Fin⁡((da)m)(\operatorname{Fin}(da))^m\simeq\operatorname{Fin}((da)^m)(Fin(da))m≃Fin((da)m). These equivalences are chosen from cardinality equalities, with no specified arithmetic ordering. When m=0m=0m=0, both the domain and codomain are singletons, including when d=0d=0d=0 or a=0a=0a=0; if m>0m>0m>0 and either is zero, the domain is empty.

QAlgorithms.Papers.Independent.Gap48.P42.shieldBlock

For any N,d∈NN,d\in\mathbb NN,d∈N, dimensions Sk∈NS_k\in\mathbb NSk​∈N, arbitrary complex matrix ρ\rhoρ indexed by BS=∏k∈[N][Sk]\mathcal B_S=\prod_{k\in[N]}[S_k]BS​=∏k∈[N]​[Sk​], family of unitary matrices (Ur)r∈[d](U_r)_{r\in[d]}(Ur​)r∈[d]​ on CBS\mathbb C^{\mathcal B_S}CBS​, and indices i,j∈[d]i,j\in[d]i,j∈[d], this defines the matrix

UiρUj†.U_i\rho U_j^\dagger.Ui​ρUj†​.

The indices iii and jjj need not differ, and ρ\rhoρ need not be a density matrix. The supplied indices require d>0d>0d>0. Zero local dimensions are permitted, yielding an empty matrix index set; N=0N=0N=0 instead yields a one-element matrix index set.

QAlgorithms.Papers.Independent.Gap48.P42.tensorTwists

For arbitrary natural numbers d,a,bd,a,bd,a,b, unitaries UαU_\alphaUα​ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b) for α<d\alpha<dα<d, and m∈Nm\in\mathbb Nm∈N, this defines a family of complex matrices VκV_\kappaVκ​, indexed by κ<dm\kappa<d^mκ<dm, on Fin⁡(am)×Fin⁡(bm)\operatorname{Fin}(a^m)\times\operatorname{Fin}(b^m)Fin(am)×Fin(bm). Decode κ\kappaκ as a sequence (κℓ)ℓ<m(\kappa_\ell)_{\ell<m}(κℓ​)ℓ<m​ through the chosen equivalence (Fin⁡(d))m≃Fin⁡(dm)(\operatorname{Fin}(d))^m\simeq\operatorname{Fin}(d^m)(Fin(d))m≃Fin(dm), and decode a row (xA,xB)(x_A,x_B)(xA​,xB​) and column (yA,yB)(y_A,y_B)(yA​,yB​) through the analogous chosen equivalences for aaa and bbb. The entry is

Vκ((xA,xB),(yA,yB))=∏ℓ<m(Uκℓ)(xA,ℓ,xB,ℓ),(yA,ℓ,yB,ℓ).V_\kappa((x_A,x_B),(y_A,y_B)) =\prod_{\ell<m} (U_{\kappa_\ell})_{(x_{A,\ell},x_{B,\ell}),(y_{A,\ell},y_{B,\ell})}.Vκ​((xA​,xB​),(yA​,yB​))=ℓ<m∏​(Uκℓ​​)(xA,ℓ​,xB,ℓ​),(yA,ℓ​,yB,ℓ​)​.

The output is defined as a family of matrices, without attaching a unitarity assertion to it. For m=0m=0m=0, there is one family label and a single matrix entry, equal to the empty product 111, even if some of d,a,bd,a,bd,a,b are zero. For m>0m>0m>0, d=0d=0d=0 gives no family labels, and a=0a=0a=0 or b=0b=0b=0 gives empty matrix indices.

QAlgorithms.Papers.Independent.Gap48.P42.twoLabelGHZVector

For any natural numbers N,dN,dN,d, labels i,j∈{0,…,d−1}i,j\in\{0,\ldots,d-1\}i,j∈{0,…,d−1}, and Boolean parameter minus\mathit{minus}minus, this definition gives the complex Euclidean vector on configurations x:{0,…,N−1}→{0,…,d−1}x:\{0,\ldots,N-1\}\to\{0,\ldots,d-1\}x:{0,…,N−1}→{0,…,d−1} with amplitude

v(x)=12(1{∀k, xk=i}+ε 1{∀k, xk=j}),ε={−1,minus=true,1,minus=false.v(x)=\frac{1}{\sqrt2} \left(\mathbf1_{\{\forall k,\ x_k=i\}} +\varepsilon\,\mathbf1_{\{\forall k,\ x_k=j\}}\right), \qquad \varepsilon= \begin{cases} -1,&\mathit{minus}=\mathrm{true},\\ 1,&\mathit{minus}=\mathrm{false}. \end{cases}v(x)=2​1​(1{∀k, xk​=i}​+ε1{∀k, xk​=j}​),ε={−1,1,​minus=true,minus=false.​

There is no requirement that i≠ji\ne ji=j or that N>0N>0N>0. If i=ji=ji=j, the vector is zero for the minus choice and is 2\sqrt22​ times the basis vector at the constant configuration for the plus choice. When N=0N=0N=0, both conditions are vacuous, so the unique amplitude is respectively 000 or 2\sqrt22​, regardless of the labels. The supplied labels can exist only when d>0d>0d>0.

QAlgorithms.Papers.Independent.Gap48.P42.twoPhaseGHZMixture

For any natural numbers N,dN,dN,d, labels i,j∈{0,…,d−1}i,j\in\{0,\ldots,d-1\}i,j∈{0,…,d−1}, and arbitrary real ppp, this definition gives the matrix

Mx,y=p v+(x)v+(y)‾+(1−p) v−(x)v−(y)‾,v±(x)=1{∀k, xk=i}±1{∀k, xk=j}2,M_{x,y}=p\,v_+(x)\overline{v_+(y)} +(1-p)\,v_-(x)\overline{v_-(y)}, \qquad v_\pm(x)=\frac{\mathbf1_{\{\forall k,\ x_k=i\}} \pm\mathbf1_{\{\forall k,\ x_k=j\}}}{\sqrt2},Mx,y​=pv+​(x)v+​(y)​+(1−p)v−​(x)v−​(y)​,v±​(x)=2​1{∀k, xk​=i}​±1{∀k, xk​=j}​​,

indexed by configurations x,y:{0,…,N−1}→{0,…,d−1}x,y:\{0,\ldots,N-1\}\to\{0,\ldots,d-1\}x,y:{0,…,N−1}→{0,…,d−1}. Neither 0≤p≤10\le p\le10≤p≤1, distinctness of i,ji,ji,j, nor N>0N>0N>0 is assumed. If i=ji=ji=j, this is 2p2p2p times the projector onto the constant configuration. If N=0N=0N=0, its sole entry is 2p2p2p. The labels require d>0d>0d>0.

Human review
  • Endorsed by Shuze Chen · Oct 22, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 22, 2026

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

  • Endorsed by Gohan · Oct 31, 2026

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

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me