Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Remarks on additivity of the Holevo channel capacity and of the entanglement of formation

Definition
QAlgorithms_Papers_Independent_Intakenext100_P60

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

prove2quantum

The objects the formalized statements of Remarks on additivity of the Holevo channel capacity and of the entanglement of formation (arXiv:quant-ph/0206148) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • pureEntropyBits: PDF p.3: entropy in bits of the first-party marginal of a pure vector. Physical vector arguments are normalized in the theorem or imported ensemble.
  • finiteBasisState: PDF p.3: change only the local basis labels to the accepted finite-dimensional operational cost interface. The same bijection is used in both matrix indices.
  • formationCostBits: PDF p.3: operational entanglement cost in singlets per copy, with half trace norm error. Reuse the accepted actual finite adaptive two-way LOCC definition; local basis relabelling accommodates all finite registers, including direct sums.
  • HasRangeIn: PDF p.4: support means every matrix column belongs to the stated linear subspace.
  • dilationSpace: PDF pp.4,9--13: range of the actual displayed embedding, as the span of its columns.
  • SignalEnsemble: PDF pp.2,5--6: a finite ensemble on an arbitrary input Hilbert space.
  • FiniteOutputDualChannel: PDF pp.2,5--6: unital completely positive dual channel, with no finite input assumption. Complete positivity uses all positive block matrices and all test vectors.
  • pureOutput: PDF p.6: actual pure output; Tr(rho E_ji)=rho_ij fixes the transposition.
  • signalBarycenter: PDF pp.2,6: output barycenter of an input ensemble.
  • signalHolevo: PDF p.2, (1): Holevo objective, in bits. Only normalized physical signals are used.
  • signalCapacity: PDF p.2, (1): the actual finite-ensemble supremum. With positive finite output and an attaining ensemble it is nonempty and bounded between zero and log2(output dim).
  • completedOperator: Canonical bounded-operator extension to the Hilbert completion; its embedding is isometric and dense, so ContinuousLinearMap.extend never takes its junk branch.
  • tensorDualOutput: PDF p.6, (9): tensor channel on the completed Hilbert tensor product. This formula uses the actual tensor product of each pair of dual matrix units, and includes arbitrary infinite-Schmidt-rank inputs through completion, not merely algebraic simple tensors.
  • PauliAdmissible: PDF p.9, (13): the closed weight simplex and both axis-order inequalities.
  • pauliDilation: PDF p.9: columns are exactly psi_T and psi_T_perp, output:environment order. The y-column phases are +i and -i, and the second z coefficient is negative.
  • pauliMixture: PDF p.9, (14): mixtures of the two specified projectors; endpoints are allowed.
  • pauliDilationWord: PDF p.9: a named tensor word, regrouped across the party cut.
  • pauliWordMixture: PDF p.9: an arbitrary convex mixture of the named tensor-word projectors.
  • gapPolynomial: PDF p.14: the exact quartic, with no total negative-root selector.
  • GapCriterion: PDF p.14, (20): the entire numerator 2^h-1 is negated before division by two.
  • DepolarizingData: PDF p.10: physical parameter and the source's orthogonal unitary error basis.
  • depolarizingWeight: PDF p.10: p0=lambda+(1-lambda)/d^2, all other weights equal.
  • depolarizingDilation: PDF p.10: the actual Kraus-block embedding, in output:environment order.
  • depolarizingEntropyBits: PDF p.10: entropy of the printed spectrum (a,b,...,b), including zero eigenvalues.
  • productMatrices: Grouped tensor product of a list of local density matrices.
  • productMixture: PDF p.10: finite convex mixtures of supported product states, not arbitrary entangled states in the tensor-power subspace.
  • jointDepolarizing: PDF p.8: tensor of the specified dilation maps, allowing different local dimensions.
  • pairMarginal: PDF p.8: the actual ith pair marginal, summing the remaining coordinates once. The update changes only coordinate i in the column; all other registers are traced.
  • IrreducibleMatrix: PDF p.11: irreducibility on exactly the represented space.
  • CovariantData: PDF p.11, (15): compact continuous unitary actions on K and H2, both irreducible. W is the actual isometry into environment:output. Covariance is imposed on all input operators; no twirling identity, optimum, or product-case action is a hypothesis.
  • projectorState: PDF p.11: P/Tr(P), constructed from W, whose rank and trace are k>0.
  • embedded: PDF p.11: embed a source vector using the actual isometry.
  • minimumEntanglement: PDF p.11, (17): minimum-entropy value over unit vectors, not all vectors of K. The result below also asserts attainment, so the printed minimum is retained.
  • dilationHolevo: PDF p.11: finite-input Holevo objective for the channel defined by W and traceLeft.
  • dilationCapacity: PDF p.11, (16): actual channel capacity (1), with unit normalized inputs.
  • minimizingProjectorHull: PDF pp.11--12: convex hull of the orbit of the minimizing pure projector. This is a convex hull of density matrices, not their linear support span.
  • PureTransitive: PDF p.12: transitivity on pure projectors, thus insensitive to global phase.
  • exampleSevenDilation: PDF p.13, Example 7: the three displayed orthonormal columns. j+3 is an ordinary environment coordinate (3,4,5), not a residue modulo three.
  • exampleSevenMixtureDilation: PDF p.13: coherent superposition in C3 tensor (C6 direct-sum (C3 tensor C3)). The right branch is |Phi_3> tensor |j>; no incoherent flag replaces the direct sum.
  • exampleSevenMixtureEntropy: PDF p.13: explicit constant from the spectrum of the printed channel. a=(1-|c|^2)/3+|c|^2/2, b=(1-|c|^2)/3+|c|^2/4, spectrum (a,b,b).
  • antisymmetricDilation: PDF p.13, Example 8: the specified normalized antisymmetric basis wedges.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_QuantumOutputMetrics
import Definitions.Def_QAlgorithms_Defs_BipartiteSeparability
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P08
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P21
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P80
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P42
import Definitions.Def_QAlgorithms_Papers_Independent_Gap48_P32

/-!
Matsumoto--Shimono--Winter, arXiv:quant-ph/0206148v4.
Accepted interfaces are imported; paper-specific adapters have complete bodies.
Only result statements have sorry.
Entropy and operational rates are in bits. Compilation is pending batch handoff.
-/

namespace QAlgorithms.Papers.Independent.Intakenext100.P60

open scoped Matrix ComplexConjugate ComplexOrder TensorProduct

noncomputable section

attribute [local instance] Classical.propDecidable Classical.decEq

-- Exact accepted matrix, normalized ensemble, convex roof, grouped copies and LOCC.
open QAlgorithms.Papers.Independent.Information100.P21 (Op FiniteLOCC)
open QAlgorithms.Papers.Independent.Information100.P08 (PureEnsemble pureAverage formation)
open QAlgorithms.Papers.Independent.Information100.P80 (groupedPower)
open QAlgorithms.Papers.Independent.Gap48.P32 (twoWayDistillable)

/-- PDF p.3: entropy in bits of the first-party marginal of a pure vector.
Physical vector arguments are normalized in the theorem or imported ensemble. -/
def pureEntropyBits {A B : Type*} [Fintype A] [Fintype B] [DecidableEq A]
    (v : EuclideanSpace ℂ (A × B)) : ℝ := vnEntropy (traceRight (pureDensity v))

/-- PDF p.3: change only the local basis labels to the accepted finite-dimensional
operational cost interface. The same bijection is used in both matrix indices. -/
def finiteBasisState {A B : Type*} [Fintype A] [Fintype B]
    (rho : Op (A × B)) : Op (Fin (Fintype.card A) × Fin (Fintype.card B)) :=
  fun p q => rho ((Fintype.equivFin A).symm p.1, (Fintype.equivFin B).symm p.2)
    ((Fintype.equivFin A).symm q.1, (Fintype.equivFin B).symm q.2)

/-- PDF p.3: operational entanglement cost in singlets per copy, with half trace
norm error. Reuse the accepted actual finite adaptive two-way LOCC definition;
local basis relabelling accommodates all finite registers, including direct sums. -/
def formationCostBits {A B : Type*} [Fintype A] [Fintype B]
    (rho : Op (A × B)) : ℝ :=
  QAlgorithms.Papers.Independent.Dependency50.P42.entanglementCost
    QAlgorithms.Papers.Independent.Dependency50.P42.halfTraceDistance (finiteBasisState rho)

/-- The cost adapter changes basis labels without changing any matrix entry. -/
example {A B : Type*} [Fintype A] [Fintype B] (rho : Op (A × B))
    (p q : Fin (Fintype.card A) × Fin (Fintype.card B)) :
    finiteBasisState rho p q =
      rho ((Fintype.equivFin A).symm p.1, (Fintype.equivFin B).symm p.2)
        ((Fintype.equivFin A).symm q.1, (Fintype.equivFin B).symm q.2) := rfl

/-- PDF p.4: support means every matrix column belongs to the stated linear subspace. -/
def HasRangeIn {I : Type*} (K : Submodule ℂ (I → ℂ)) (rho : Op I) : Prop :=
  ∀ j, (fun i => rho i j) ∈ K

/-- PDF pp.4,9--13: range of the actual displayed embedding, as the span of its columns. -/
def dilationSpace {I J : Type*} (W : Matrix I J ℂ) : Submodule ℂ (I → ℂ) :=
  Submodule.span ℂ (Set.range fun j => fun i => W i j)

/-- Normalization sanity for the finite roof, not a proof of a source theorem. -/
example {I : Type} [Fintype I] (v : EuclideanSpace ℂ I) (hv : ‖v‖ = 1) :
    ∃ e : PureEnsemble I, pureAverage e = pureDensity v := by
  refine ⟨⟨1, fun _ => 1, fun _ => v, fun _ => by norm_num, by simp,
    fun _ => hv⟩, ?_⟩
  simp [pureAverage]

/-- Zero copies are the scalar trace-one operator, without a division by zero. -/
example {A B : Type*} (rho : Op (A × B))
    (x y : (Fin 0 → A) × (Fin 0 → B)) : groupedPower rho 0 x y = 1 := by
  simp [groupedPower]

/-- The protocol class includes the identity without spending entanglement. -/
example (a b : ℕ) : FiniteLOCC (fun X : Op (Fin a × Fin b) => X) :=
  FiniteLOCC.identity a b

/-! General Hilbert input, finite output: Heisenberg-picture normal channels.
The domain algebra is finite dimensional, so positivity at all matrix levels and
unitality specify precisely the adjoint of a trace-preserving completely positive
Schrodinger channel on all trace-class input states. No input dimension restriction. -/

/-- PDF pp.2,5--6: a finite ensemble on an arbitrary input Hilbert space. -/
structure SignalEnsemble (H : Type*) [NormedAddCommGroup H] where
  size : ℕ
  prob : Fin size → ℝ
  vector : Fin size → H
  nonneg : ∀ j, 0 ≤ prob j
  total : ∑ j, prob j = 1
  unit : ∀ j, ‖vector j‖ = 1

/-- PDF pp.2,5--6: unital completely positive dual channel, with no finite input
assumption. Complete positivity uses all positive block matrices and all test vectors. -/
structure FiniteOutputDualChannel (H : Type*) [NormedAddCommGroup H] [InnerProductSpace ℂ H]
    (d : ℕ) where
  dual : Op (Fin d) →ₗ[ℂ] (H →L[ℂ] H)
  unital : dual 1 = ContinuousLinearMap.id ℂ H
  completely_positive : ∀ k : ℕ, ∀ X : Op (Fin k × Fin d), X.PosSemidef →
    ∀ v : Fin k → H,
      0 ≤ ∑ r, ∑ s, inner ℂ (v r)
        (dual (fun i j => X (r, i) (s, j)) (v s))

/-- PDF p.6: actual pure output; Tr(rho E_ji)=rho_ij fixes the transposition. -/
def pureOutput {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℂ H]
    {d : ℕ} (T : FiniteOutputDualChannel H d) (v : H) : Op (Fin d) :=
  fun i j => inner ℂ v (T.dual (Matrix.single j i 1) v)

/-- PDF pp.2,6: output barycenter of an input ensemble. -/
def signalBarycenter {H I : Type*} [NormedAddCommGroup H]
    (signal : H → Op I) (e : SignalEnsemble H) : Op I :=
  ∑ j, (e.prob j : ℂ) • signal (e.vector j)

/-- PDF p.2, (1): Holevo objective, in bits. Only normalized physical signals are used. -/
def signalHolevo {H I : Type*} [NormedAddCommGroup H] [Fintype I] [DecidableEq I]
    (signal : H → Op I) (e : SignalEnsemble H) : ℝ :=
  vnEntropy (signalBarycenter signal e) - ∑ j, e.prob j * vnEntropy (signal (e.vector j))

/-- PDF p.2, (1): the actual finite-ensemble supremum. With positive finite output and
an attaining ensemble it is nonempty and bounded between zero and log2(output dim). -/
def signalCapacity {H I : Type*} [NormedAddCommGroup H] [Fintype I] [DecidableEq I]
    (signal : H → Op I) : ℝ := sSup (Set.range (signalHolevo signal))

/-- Canonical bounded-operator extension to the Hilbert completion; its embedding
is isometric and dense, so ContinuousLinearMap.extend never takes its junk branch. -/
def completedOperator {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℂ E]
    (f : E →L[ℂ] E) : UniformSpace.Completion E →L[ℂ] UniformSpace.Completion E :=
  ((UniformSpace.Completion.toComplL : E →L[ℂ] UniformSpace.Completion E).comp f).extend
    (UniformSpace.Completion.toComplL : E →L[ℂ] UniformSpace.Completion E)

/-- PDF p.6, (9): tensor channel on the completed Hilbert tensor product. This formula
uses the actual tensor product of each pair of dual matrix units, and includes arbitrary
infinite-Schmidt-rank inputs through completion, not merely algebraic simple tensors. -/
def tensorDualOutput {H K : Type*} [NormedAddCommGroup H] [InnerProductSpace ℂ H]
    [NormedAddCommGroup K] [InnerProductSpace ℂ K] {d e : ℕ}
    (T : FiniteOutputDualChannel H d) (S : FiniteOutputDualChannel K e)
    (v : UniformSpace.Completion (H ⊗[ℂ] K)) : Op (Fin d × Fin e) :=
  fun i j => inner ℂ v (completedOperator
    (TensorProduct.mapL (T.dual (Matrix.single j.1 i.1 1))
      (S.dual (Matrix.single j.2 i.2 1))) v)

/-- PDF p.9, (13): the closed weight simplex and both axis-order inequalities. -/
def PauliAdmissible (p : Fin 4 → ℝ) : Prop :=
  (∀ i, 0 ≤ p i) ∧ (∑ i, p i) = 1 ∧
    |p 0 + p 2 - p 1 - p 3| ≤ p 0 + p 3 - p 1 - p 2 ∧
    |p 0 + p 1 - p 2 - p 3| ≤ p 0 + p 3 - p 1 - p 2

/-- PDF p.9: columns are exactly psi_T and psi_T_perp, output:environment order.
The y-column phases are +i and -i, and the second z coefficient is negative. -/
def pauliDilation (p : Fin 4 → ℝ) : Matrix (Fin 2 × Fin 4) (Fin 2) ℂ :=
  fun q j =>
    if q.2 = 0 then if q.1 = j then (Real.sqrt (p 0) : ℂ) else 0
    else if q.2 = 1 then if q.1 ≠ j then (Real.sqrt (p 1) : ℂ) else 0
    else if q.2 = 2 then if q.1 ≠ j then
      (if j = 0 then Complex.I else -Complex.I) * (Real.sqrt (p 2) : ℂ) else 0
    else if q.1 = j then
      (if j = 0 then (1 : ℂ) else -1) * (Real.sqrt (p 3) : ℂ) else 0

/-- PDF p.9, (14): mixtures of the two specified projectors; endpoints are allowed. -/
def pauliMixture (p : Fin 4 → ℝ) (s : ℝ) : Op (Fin 2 × Fin 4) :=
  (s : ℂ) • pureDensity (WithLp.toLp 2 fun q => pauliDilation p q 0) +
    ((1 - s : ℝ) : ℂ) • pureDensity (WithLp.toLp 2 fun q => pauliDilation p q 1)

/-- PDF p.9: a named tensor word, regrouped across the party cut. -/
def pauliDilationWord (p : Fin 4 → ℝ) (n : ℕ) (w : Fin n → Fin 2) :
    EuclideanSpace ℂ ((Fin n → Fin 2) × (Fin n → Fin 4)) :=
  WithLp.toLp 2 fun q => ∏ i, pauliDilation p (q.1 i, q.2 i) (w i)

/-- PDF p.9: an arbitrary convex mixture of the named tensor-word projectors. -/
def pauliWordMixture (p : Fin 4 → ℝ) (n : ℕ) (mu : (Fin n → Fin 2) → ℝ) :
    Op ((Fin n → Fin 2) × (Fin n → Fin 4)) :=
  ∑ w, (mu w : ℂ) • pureDensity (pauliDilationWord p n w)

/-- PDF p.14: the exact quartic, with no total negative-root selector. -/
def gapPolynomial (p : Fin 4 → ℝ) (z : ℝ) : ℝ :=
  z ^ 4 - z ^ 3 + 4 * (p 0 * p 1 * p 2 + p 0 * p 1 * p 3 +
    p 0 * p 2 * p 3 + p 1 * p 2 * p 3) * z - 16 * p 0 * p 1 * p 2 * p 3

/-- PDF p.14, (20): the entire numerator 2^h-1 is negated before division by two. -/
def GapCriterion (p : Fin 4 → ℝ) : Prop :=
  0 < gapPolynomial p (-(((2 : ℝ) ^ binEntropy2 (p 0 + p 3) - 1) / 2))

/-- The y-phase and perpendicular z-sign are pinned directly at matrix entries. -/
example (p : Fin 4 → ℝ) :
    pauliDilation p (1, 2) 0 = Complex.I * (Real.sqrt (p 2) : ℂ) ∧
    pauliDilation p (0, 2) 1 = -Complex.I * (Real.sqrt (p 2) : ℂ) ∧
    pauliDilation p (1, 3) 1 = -(Real.sqrt (p 3) : ℂ) := by
  norm_num [pauliDilation, Fin.ext_iff]

/-- Quartic boundary check: the allowed identity channel has no negative root. -/
example (z : ℝ) : gapPolynomial ![1, 0, 0, 0] z = z ^ 4 - z ^ 3 := by
  change z ^ 4 - z ^ 3 +
    4 * (1 * 0 * 0 + 1 * 0 * 0 + 1 * 0 * 0 + 0 * 0 * 0) * z -
    16 * 1 * 0 * 0 * 0 = z ^ 4 - z ^ 3
  ring

/-! General depolarizing subspaces. -/

/-- PDF p.10: physical parameter and the source's orthogonal unitary error basis. -/
structure DepolarizingData (d : ℕ) where
  dimension : 2 ≤ d
  parameter : ℝ
  lower : -1 / ((d : ℝ) ^ 2 - 1) ≤ parameter
  upper : parameter ≤ 1
  sigma : Fin (d ^ 2) → Op (Fin d)
  identity : ∀ i, i.val = 0 → sigma i = 1
  unitary : ∀ i, (sigma i)ᴴ * sigma i = 1
  orthogonal : ∀ i j, ((sigma i)ᴴ * sigma j).trace = if i = j then (d : ℂ) else 0

/-- PDF p.10: p0=lambda+(1-lambda)/d^2, all other weights equal. -/
def depolarizingWeight {d : ℕ} (D : DepolarizingData d) (i : Fin (d ^ 2)) : ℝ :=
  let p0 := D.parameter + (1 - D.parameter) / (d : ℝ) ^ 2
  if i.val = 0 then p0 else (1 - p0) / ((d : ℝ) ^ 2 - 1)

/-- PDF p.10: the actual Kraus-block embedding, in output:environment order. -/
def depolarizingDilation {d : ℕ} (D : DepolarizingData d) :
    Matrix (Fin d × Fin (d ^ 2)) (Fin d) ℂ :=
  fun q j => (Real.sqrt (depolarizingWeight D q.2) : ℂ) * D.sigma q.2 q.1 j

/-- PDF p.10: entropy of the printed spectrum (a,b,...,b), including zero eigenvalues. -/
def depolarizingEntropyBits {d : ℕ} (D : DepolarizingData d) : ℝ :=
  let b := (1 - D.parameter) / (d : ℝ)
  let a := D.parameter + b;
  -(a * Real.logb 2 a + (d - 1 : ℕ) * b * Real.logb 2 b)

/-- Grouped tensor product of a list of local density matrices. -/
def productMatrices {A B : Type*} (n : ℕ) (rho : Fin n → Op (A × B)) :
    Op ((Fin n → A) × (Fin n → B)) :=
  fun p q => ∏ i, rho i (p.1 i, p.2 i) (q.1 i, q.2 i)

/-- PDF p.10: finite convex mixtures of supported product states, not arbitrary
entangled states in the tensor-power subspace. -/
def productMixture {A B : Type*} (n m : ℕ) (mu : Fin m → ℝ)
    (rho : Fin m → Fin n → Op (A × B)) : Op ((Fin n → A) × (Fin n → B)) :=
  ∑ j, (mu j : ℂ) • productMatrices n (rho j)

/-- PDF p.8: tensor of the specified dilation maps, allowing different local dimensions. -/
def jointDepolarizing {n : ℕ} {d : Fin n → ℕ} (D : ∀ i, DepolarizingData (d i)) :
    Matrix ((∀ i, Fin (d i)) × (∀ i, Fin (d i ^ 2))) (∀ i, Fin (d i)) ℂ :=
  fun q j => ∏ i, depolarizingDilation (D i) (q.1 i, q.2 i) (j i)

/-- PDF p.8: the actual ith pair marginal, summing the remaining coordinates once.
The update changes only coordinate i in the column; all other registers are traced. -/
def pairMarginal {n : ℕ} {a b : Fin n → ℕ}
    (rho : Op ((∀ i, Fin (a i)) × (∀ i, Fin (b i)))) (i : Fin n) :
    Op (Fin (a i) × Fin (b i)) :=
  fun p q => ∑ x : {x : ∀ j, Fin (a j) // x i = p.1},
    ∑ y : {y : ∀ j, Fin (b j) // y i = p.2},
      rho (x.val, y.val) (Function.update x.val i q.1, Function.update y.val i q.2)

/-- At the identity endpoint the nonidentity Kraus weights vanish, without 0/0. -/
example {d : ℕ} (D : DepolarizingData d) (h : D.parameter = 1)
    (i : Fin (d ^ 2)) (hi : i.val ≠ 0) : depolarizingWeight D i = 0 := by
  simp [depolarizingWeight, h, hi]

/-! Compact symmetry, on a nonzero finite-dimensional effective dilation.
An arbitrary ambient environment can be compressed to the component span of the finite
range, preserving every supported matrix, its marginal spectrum, and operational rates. -/

/-- PDF p.11: irreducibility on exactly the represented space. -/
def IrreducibleMatrix {G : Type*} {k : ℕ} (V : G → Op (Fin k)) : Prop :=
  ∀ L : Submodule ℂ (Fin k → ℂ),
    (∀ g x, x ∈ L → (V g).mulVec x ∈ L) → L = ⊥ ∨ L = ⊤

/-- PDF p.11, (15): compact continuous unitary actions on K and H2, both irreducible.
W is the actual isometry into environment:output. Covariance is imposed on all input
operators; no twirling identity, optimum, or product-case action is a hypothesis. -/
structure CovariantData (G : Type*) [Group G] [TopologicalSpace G]
    (a b k : ℕ) where
  environment_positive : 0 < a
  output_positive : 0 < b
  input_positive : 0 < k
  W : Matrix (Fin a × Fin b) (Fin k) ℂ
  isometry : Wᴴ * W = 1
  V : G → Op (Fin k)
  U : G → Op (Fin b)
  V_one : V 1 = 1
  U_one : U 1 = 1
  V_mul : ∀ g h, V (g * h) = V g * V h
  U_mul : ∀ g h, U (g * h) = U g * U h
  V_unitary : ∀ g, (V g)ᴴ * V g = 1
  U_unitary : ∀ g, (U g)ᴴ * U g = 1
  V_continuous : ∀ i j, Continuous (fun g => V g i j)
  U_continuous : ∀ i j, Continuous (fun g => U g i j)
  V_irreducible : IrreducibleMatrix V
  U_irreducible : IrreducibleMatrix U
  covariance : ∀ g X, traceLeft (W * (V g * X * (V g)ᴴ) * Wᴴ) =
    U g * traceLeft (W * X * Wᴴ) * (U g)ᴴ

/-- PDF p.11: P/Tr(P), constructed from W, whose rank and trace are k>0. -/
def projectorState {G : Type*} [Group G] [TopologicalSpace G] {a b k : ℕ}
    (D : CovariantData G a b k) : Op (Fin a × Fin b) :=
  (k : ℂ)⁻¹ • (D.W * D.Wᴴ)

/-- PDF p.11: embed a source vector using the actual isometry. -/
def embedded {a b k : ℕ} (W : Matrix (Fin a × Fin b) (Fin k) ℂ)
    (v : EuclideanSpace ℂ (Fin k)) : EuclideanSpace ℂ (Fin a × Fin b) :=
  WithLp.toLp 2 (W.mulVec (WithLp.ofLp v))

/-- PDF p.11, (17): minimum-entropy value over unit vectors, not all vectors of K.
The result below also asserts attainment, so the printed minimum is retained. -/
def minimumEntanglement {G : Type*} [Group G] [TopologicalSpace G] {a b k : ℕ}
    (D : CovariantData G a b k) : ℝ :=
  sInf {r | ∃ v : EuclideanSpace ℂ (Fin k), ‖v‖ = 1 ∧
    r = pureEntropyBits (embedded D.W v)}

/-- PDF p.11: finite-input Holevo objective for the channel defined by W and traceLeft. -/
def dilationHolevo {a b k : ℕ} (W : Matrix (Fin a × Fin b) (Fin k) ℂ)
    (e : PureEnsemble (Fin k)) : ℝ :=
  vnEntropy (traceLeft (W * pureAverage e * Wᴴ)) -
    ∑ j, e.prob j * vnEntropy (traceLeft (pureDensity (embedded W (e.vector j))))

/-- PDF p.11, (16): actual channel capacity (1), with unit normalized inputs. -/
def dilationCapacity {a b k : ℕ} (W : Matrix (Fin a × Fin b) (Fin k) ℂ) : ℝ :=
  sSup (Set.range (dilationHolevo W))

/-- PDF pp.11--12: convex hull of the orbit of the minimizing pure projector.
This is a convex hull of density matrices, not their linear support span. -/
def minimizingProjectorHull {G : Type*} [Group G] [TopologicalSpace G] {a b k : ℕ}
    (D : CovariantData G a b k) (v : EuclideanSpace ℂ (Fin k)) : Set (Op (Fin a × Fin b)) :=
  convexHull ℝ {rho | ∃ g : G, rho = pureDensity (embedded D.W (act (D.V g) v))}

/-- PDF p.12: transitivity on pure projectors, thus insensitive to global phase. -/
def PureTransitive {G : Type*} [Group G] [TopologicalSpace G] {a b k : ℕ}
    (D : CovariantData G a b k) : Prop :=
  ∀ v w : EuclideanSpace ℂ (Fin k), ‖v‖ = 1 → ‖w‖ = 1 →
    ∃ g : G, pureDensity (act (D.V g) v) = pureDensity w

/-- PDF p.13, Example 7: the three displayed orthonormal columns. j+3 is an
ordinary environment coordinate (3,4,5), not a residue modulo three. -/
def exampleSevenDilation : Matrix (Fin 3 × Fin 6) (Fin 3) ℂ :=
  fun q j =>
    if q.1 = j ∧ q.2.val = j.val + 3 then (Real.sqrt 2 / 2 : ℝ)
    else if (q.1.val = (j.val + 1) % 3 ∧ q.2.val = (j.val + 2) % 3) ∨
      (q.1.val = (j.val + 2) % 3 ∧ q.2.val = (j.val + 1) % 3) then (1 / 2 : ℂ) else 0

/-- PDF p.13: coherent superposition in C3 tensor (C6 direct-sum (C3 tensor C3)).
The right branch is |Phi_3> tensor |j>; no incoherent flag replaces the direct sum. -/
def exampleSevenMixtureDilation (c s : ℂ) :
    Matrix (Fin 3 × (Fin 6 ⊕ (Fin 3 × Fin 3))) (Fin 3) ℂ :=
  fun q j => match q.2 with
    | Sum.inl e => c * exampleSevenDilation (q.1, e) j
    | Sum.inr e => if q.1 = e.1 ∧ e.2 = j then s * ((Real.sqrt 3)⁻¹ : ℂ) else 0

/-- PDF p.13: explicit constant from the spectrum of the printed channel.
a=(1-|c|^2)/3+|c|^2/2, b=(1-|c|^2)/3+|c|^2/4, spectrum (a,b,b). -/
def exampleSevenMixtureEntropy (c : ℂ) : ℝ :=
  let t := Complex.normSq c
  let a := (1 - t) / 3 + t / 2
  let b := (1 - t) / 3 + t / 4;
  -(a * Real.logb 2 a + 2 * b * Real.logb 2 b)

/-- PDF p.13, Example 8: the specified normalized antisymmetric basis wedges. -/
def antisymmetricDilation : Matrix (Fin 3 × Fin 3) (Fin 3) ℂ :=
  fun q j =>
    if q.1.val = (j.val + 1) % 3 ∧ q.2.val = (j.val + 2) % 3 then ((Real.sqrt 2)⁻¹ : ℂ)
    else if q.1.val = (j.val + 2) % 3 ∧ q.2.val = (j.val + 1) % 3 then -((Real.sqrt 2)⁻¹ : ℂ)
    else 0

/-- Sanity: the noncyclic environment coordinate and the first cross coefficient. -/
example : exampleSevenDilation (0, 3) 0 = ((Real.sqrt 2 / 2 : ℝ) : ℂ) ∧
    exampleSevenDilation (1, 2) 0 = (1 / 2 : ℂ) := by
  norm_num [exampleSevenDilation]

/-- The direct-sum zero branch is exactly the scaled original vector. -/
example (c s : ℂ) (i j : Fin 3) (e : Fin 6) :
    exampleSevenMixtureDilation c s (i, Sum.inl e) j = c * exampleSevenDilation (i, e) j := rfl

/-- The antisymmetric sign is fixed by the printed order of the wedge. -/
example : antisymmetricDilation (2, 1) 0 = -((Real.sqrt 2)⁻¹ : ℂ) := by
  norm_num [antisymmetricDilation]

end

end QAlgorithms.Papers.Independent.Intakenext100.P60
Source
Remarks on additivity of the Holevo channel capacity and of the entanglement of formation, arXiv:quant-ph/0206148v4, https://arxiv.org/abs/quant-ph/0206148v4
Read-back

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

QAlgorithms.Papers.Independent.Intakenext100.P60.CovariantData

For a group GGG equipped with a topology and natural numbers a,b,ka,b,ka,b,k, this structure consists of proofs that a,b,k>0a,b,k>0a,b,k>0; a complex matrix W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb satisfying W†W=IkW^\dagger W=I_kW†W=Ik​; and matrix families Vg∈Mk(C)V_g\in M_k(\mathbb C)Vg​∈Mk​(C) and Ug∈Mb(C)U_g\in M_b(\mathbb C)Ug​∈Mb​(C) satisfying V1=IkV_1=I_kV1​=Ik​, U1=IbU_1=I_bU1​=Ib​, Vgh=VgVhV_{gh}=V_gV_hVgh​=Vg​Vh​, Ugh=UgUhU_{gh}=U_gU_hUgh​=Ug​Uh​, Vg†Vg=IkV_g^\dagger V_g=I_kVg†​Vg​=Ik​, and Ug†Ug=IbU_g^\dagger U_g=I_bUg†​Ug​=Ib​ for every g,h∈Gg,h\in Gg,h∈G. Every matrix entry of each family is continuous in ggg, and each family's only common invariant complex linear subspaces are the zero and full spaces. Finally, for every g∈Gg\in Gg∈G and every complex k×kk\times kk×k matrix XXX, it requires

Tr⁡Ca ⁣(WVgXVg†W†)=Ug Tr⁡Ca(WXW†) Ug†,\operatorname{Tr}_{\mathbb C^a}\!\left(WV_gXV_g^\dagger W^\dagger\right) = U_g\,\operatorname{Tr}_{\mathbb C^a}(WXW^\dagger)\,U_g^\dagger,TrCa​(WVg​XVg†​W†)=Ug​TrCa​(WXW†)Ug†​,

where (Tr⁡CaY)r,s=∑u=0a−1Y(u,r),(u,s)(\operatorname{Tr}_{\mathbb C^a}Y)_{r,s}=\sum_{u=0}^{a-1}Y_{(u,r),(u,s)}(TrCa​Y)r,s​=∑u=0a−1​Y(u,r),(u,s)​. The topology on GGG is an argument of the structure; continuity of the group operations is not a separate requirement here.

QAlgorithms.Papers.Independent.Intakenext100.P60.DepolarizingData

For a natural number ddd, an element of this structure consists of a proof that d≥2d\ge2d≥2, a real parameter λ\lambdaλ satisfying

−1d2−1≤λ≤1,-\frac1{d^2-1}\le\lambda\le1,−d2−11​≤λ≤1,

and d2d^2d2 complex d×dd\times dd×d matrices σi\sigma_iσi​, indexed by 0≤i<d20\le i<d^20≤i<d2, satisfying σi=Id\sigma_i=I_dσi​=Id​ whenever i=0i=0i=0, σi†σi=Id\sigma_i^\dagger\sigma_i=I_dσi†​σi​=Id​ for every iii, and

Tr⁡(σi†σj)={d,i=j,0,i≠j.\operatorname{Tr}(\sigma_i^\dagger\sigma_j)= \begin{cases}d,&i=j,\\0,&i\ne j.\end{cases}Tr(σi†​σj​)={d,0,​i=j,i=j.​

The dimension condition excludes d=0,1d=0,1d=0,1, and the permitted parameter interval includes negative values.

QAlgorithms.Papers.Independent.Intakenext100.P60.FiniteOutputDualChannel

For any complex inner-product space HHH with its normed additive commutative group structure, and any natural number ddd, this structure consists of a complex-linear map

T:Md(C)⟶LC(H,H),T:M_d(\mathbb C)\longrightarrow\mathcal L_{\mathbb C}(H,H),T:Md​(C)⟶LC​(H,H),

where the codomain consists of continuous complex-linear operators, such that T(Id)=IHT(I_d)=I_HT(Id​)=IH​ and, for every natural number kkk, every positive semidefinite matrix XXX indexed by Fin⁡(k)×Fin⁡(d)\operatorname{Fin}(k)\times\operatorname{Fin}(d)Fin(k)×Fin(d), and every family v0,…,vk−1∈Hv_0,\ldots,v_{k-1}\in Hv0​,…,vk−1​∈H,

∑r,s=0k−1⟨vr, T ⁣((X(r,i),(s,j))i,j)vs⟩\sum_{r,s=0}^{k-1} \left\langle v_r,\, T\!\left((X_{(r,i),(s,j)})_{i,j}\right)v_s \right\rangler,s=0∑k−1​⟨vr​,T((X(r,i),(s,j)​)i,j​)vs​⟩

is real and nonnegative. Completeness of HHH is not assumed. The test includes k=0k=0k=0, when the sum is zero; d=0d=0d=0 is also allowed, in which case unitality forces IH=0I_H=0IH​=0.

QAlgorithms.Papers.Independent.Intakenext100.P60.GapCriterion

For every real four-tuple p=(p0,p1,p2,p3)p=(p_0,p_1,p_2,p_3)p=(p0​,p1​,p2​,p3​), put

h=binEntropy⁡(p0+p3)log⁡2,t=−2h−12.h=\frac{\operatorname{binEntropy}(p_0+p_3)}{\log2}, \qquad t=-\frac{2^h-1}{2}.h=log2binEntropy(p0​+p3​)​,t=−22h−1​.

The predicate means exactly

0<t4−t3+4(p0p1p2+p0p1p3+p0p2p3+p1p2p3)t−16p0p1p2p3.0<t^4-t^3 +4(p_0p_1p_2+p_0p_1p_3+p_0p_2p_3+p_1p_2p_3)t -16p_0p_1p_2p_3.0<t4−t3+4(p0​p1​p2​+p0​p1​p3​+p0​p2​p3​+p1​p2​p3​)t−16p0​p1​p2​p3​.

The inequality is strict. No admissibility condition or restriction p0+p3∈[0,1]p_0+p_3\in[0,1]p0​+p3​∈[0,1] is part of this definition; the real binary-entropy function is used with its total definition for all real inputs.

QAlgorithms.Papers.Independent.Intakenext100.P60.HasRangeIn

For any type III, any complex linear subspace K⊆CIK\subseteq\mathbb C^IK⊆CI, and any complex matrix ρ\rhoρ indexed by III, this predicate means that every column of ρ\rhoρ belongs to KKK:

∀j∈I,(i↦ρi,j)∈K.\forall j\in I,\qquad (i\mapsto\rho_{i,j})\in K.∀j∈I,(i↦ρi,j​)∈K.

No finiteness, boundedness, positivity, or normalization is required. If III is empty, the condition is vacuous.

QAlgorithms.Papers.Independent.Intakenext100.P60.IrreducibleMatrix

For any type GGG, natural number kkk, and family V=(Vg)g∈GV=(V_g)_{g\in G}V=(Vg​)g∈G​ of complex k×kk\times kk×k matrices, this property says that every complex linear subspace L⊆CkL\subseteq\mathbb C^kL⊆Ck satisfying Vgx∈LV_gx\in LVg​x∈L for every g∈Gg\in Gg∈G and every x∈Lx\in Lx∈L is either {0}\{0\}{0} or all of Ck\mathbb C^kCk. No group structure on GGG is assumed. The definition includes empty GGG, when every subspace satisfies the invariance hypothesis, and k=0k=0k=0, when the zero and full subspaces coincide.

QAlgorithms.Papers.Independent.Intakenext100.P60.PauliAdmissible

For an arbitrary real four-tuple p=(p0,p1,p2,p3)p=(p_0,p_1,p_2,p_3)p=(p0​,p1​,p2​,p3​), this predicate means exactly

pi≥0(0≤i<4),p0+p1+p2+p3=1,p_i\ge0\quad(0\le i<4),\qquad p_0+p_1+p_2+p_3=1,pi​≥0(0≤i<4),p0​+p1​+p2​+p3​=1,

together with

∣p0+p2−p1−p3∣≤p0+p3−p1−p2,∣p0+p1−p2−p3∣≤p0+p3−p1−p2.|p_0+p_2-p_1-p_3|\le p_0+p_3-p_1-p_2,\qquad |p_0+p_1-p_2-p_3|\le p_0+p_3-p_1-p_2.∣p0​+p2​−p1​−p3​∣≤p0​+p3​−p1​−p2​,∣p0​+p1​−p2​−p3​∣≤p0​+p3​−p1​−p2​.

All inequalities are non-strict, so zero coordinates and equality cases are included.

QAlgorithms.Papers.Independent.Intakenext100.P60.PureTransitive

For a group GGG equipped with a topology and data DDD comprising positive dimensions a,b,ka,b,ka,b,k, an isometry W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, and entrywise continuous irreducible unitary representations V,UV,UV,U on Ck,Cb\mathbb C^k,\mathbb C^bCk,Cb, respectively, satisfying Tr⁡Ca(WVgXVg†W†)=UgTr⁡Ca(WXW†)Ug†\operatorname{Tr}_{\mathbb C^a}(WV_gXV_g^\dagger W^\dagger)=U_g\operatorname{Tr}_{\mathbb C^a}(WXW^\dagger)U_g^\daggerTrCa​(WVg​XVg†​W†)=Ug​TrCa​(WXW†)Ug†​ for every g,Xg,Xg,X, this property says that for every pair v,w∈Ckv,w\in\mathbb C^kv,w∈Ck with ∥v∥=∥w∥=1\|v\|=\|w\|=1∥v∥=∥w∥=1, there exists g∈Gg\in Gg∈G such that

(Vgv)(Vgv)†=ww†.(V_gv)(V_gv)^\dagger=ww^\dagger.(Vg​v)(Vg​v)†=ww†.

Here irreducibility means that the only common invariant complex subspaces are zero and the whole space, and the partial trace sums over the aaa-dimensional factor. The required equality is equality of rank-one matrices, so it allows the vectors VgvV_gvVg​v and www to differ by a complex scalar of modulus one.

QAlgorithms.Papers.Independent.Intakenext100.P60.SignalEnsemble

For any normed additive commutative group HHH, this structure consists of a natural number mmm, real weights pjp_jpj​, and vectors vj∈Hv_j\in Hvj​∈H, indexed by 0≤j<m0\le j<m0≤j<m, together with the requirements

pj≥0,∑j=0m−1pj=1,∥vj∥=1 for every j.p_j\ge0,\qquad \sum_{j=0}^{m-1}p_j=1,\qquad \|v_j\|=1\ \text{for every }j.pj​≥0,j=0∑m−1​pj​=1,∥vj​∥=1 for every j.

Weights may be zero and vectors may repeat; even a vector with zero weight must have norm 111. The choice m=0m=0m=0 cannot satisfy the required total weight.

QAlgorithms.Papers.Independent.Intakenext100.P60.antisymmetricDilation

This defines a complex matrix W−W_-W−​ with rows indexed by (r,e)∈{0,1,2}2(r,e)\in\{0,1,2\}^2(r,e)∈{0,1,2}2 and columns indexed by j∈{0,1,2}j\in\{0,1,2\}j∈{0,1,2}, with

(W−)(r,e),j={1/2,r=(j+1) mod 3, e=(j+2) mod 3,−1/2,r=(j+2) mod 3, e=(j+1) mod 3,0,otherwise.(W_-)_{(r,e),j}= \begin{cases} 1/\sqrt2,&r=(j+1)\bmod3,\ e=(j+2)\bmod3,\\ -1/\sqrt2,&r=(j+2)\bmod3,\ e=(j+1)\bmod3,\\ 0,&\text{otherwise}. \end{cases}(W−​)(r,e),j​=⎩⎨⎧​1/2​,−1/2​,0,​r=(j+1)mod3, e=(j+2)mod3,r=(j+2)mod3, e=(j+1)mod3,otherwise.​

Thus column jjj is (∣(j+1) mod 3,(j+2) mod 3⟩−∣(j+2) mod 3,(j+1) mod 3⟩)/2\bigl(|(j+1)\bmod3,(j+2)\bmod3\rangle-|(j+2)\bmod3,(j+1)\bmod3\rangle\bigr)/\sqrt2(∣(j+1)mod3,(j+2)mod3⟩−∣(j+2)mod3,(j+1)mod3⟩)/2​, where each ket is a coordinate unit vector.

QAlgorithms.Papers.Independent.Intakenext100.P60.completedOperator

For any complex inner-product space EEE with its normed additive commutative group structure, and any continuous complex-linear operator f:E→Ef:E\to Ef:E→E, this defines its continuous complex-linear extension f^:E^→E^\widehat f:\widehat E\to\widehat Ef​:E→E to the uniform-space completion. Writing ι:E→E^\iota:E\to\widehat Eι:E→E for the canonical completion map, the construction extends ι∘f\iota\circ fι∘f along ι\iotaι, so that

f^(ιx)=ι(fx)(x∈E).\widehat f(\iota x)=\iota(fx)\qquad(x\in E).f​(ιx)=ι(fx)(x∈E).

Completeness or nontriviality of EEE is not assumed.

QAlgorithms.Papers.Independent.Intakenext100.P60.depolarizingDilation

For any natural number ddd and data DDD comprising d≥2d\ge2d≥2, −1/(d2−1)≤λ≤1-1/(d^2-1)\le\lambda\le1−1/(d2−1)≤λ≤1, and matrices σ0,…,σd2−1∈Cd×d\sigma_0,\ldots,\sigma_{d^2-1}\in\mathbb C^{d\times d}σ0​,…,σd2−1​∈Cd×d satisfying σ0=Id\sigma_0=I_dσ0​=Id​, σi†σi=Id\sigma_i^\dagger\sigma_i=I_dσi†​σi​=Id​, and Tr⁡(σi†σj)=d δij\operatorname{Tr}(\sigma_i^\dagger\sigma_j)=d\,\delta_{ij}Tr(σi†​σj​)=dδij​, this defines the complex matrix WDW_DWD​ with rows indexed by pairs (a,i)(a,i)(a,i), 0≤a<d0\le a<d0≤a<d, 0≤i<d20\le i<d^20≤i<d2, and columns indexed by 0≤j<d0\le j<d0≤j<d, by

(WD)(a,i),j=wD(i) (σi)a,j,wD(i)={p0,i=0,(1−p0)/(d2−1),i≠0,p0=λ+1−λd2.(W_D)_{(a,i),j}=\sqrt{w_D(i)}\,(\sigma_i)_{a,j}, \qquad w_D(i)= \begin{cases} p_0,&i=0,\\ (1-p_0)/(d^2-1),&i\ne0, \end{cases} \qquad p_0=\lambda+\frac{1-\lambda}{d^2}.(WD​)(a,i),j​=wD​(i)​(σi​)a,j​,wD​(i)={p0​,(1−p0​)/(d2−1),​i=0,i=0,​p0​=λ+d21−λ​.

The square root is the real square root, embedded into C\mathbb CC; zero weights, including those at parameter endpoints, give zero entries in the corresponding row blocks.

QAlgorithms.Papers.Independent.Intakenext100.P60.depolarizingEntropyBits

For any natural number ddd and data DDD comprising d≥2d\ge2d≥2, a parameter −1/(d2−1)≤λ≤1-1/(d^2-1)\le\lambda\le1−1/(d2−1)≤λ≤1, and matrices σi∈Cd×d\sigma_i\in\mathbb C^{d\times d}σi​∈Cd×d, 0≤i<d20\le i<d^20≤i<d2, with σ0=Id\sigma_0=I_dσ0​=Id​, σi†σi=Id\sigma_i^\dagger\sigma_i=I_dσi†​σi​=Id​, and Tr⁡(σi†σj)=d δij\operatorname{Tr}(\sigma_i^\dagger\sigma_j)=d\,\delta_{ij}Tr(σi†​σj​)=dδij​, this defines the real number

HD=−(alog⁡2a+(d−1)blog⁡2b),b=1−λd,a=λ+b.H_D=-\left(a\log_2a+(d-1)b\log_2b\right), \qquad b=\frac{1-\lambda}{d},\qquad a=\lambda+b.HD​=−(alog2​a+(d−1)blog2​b),b=d1−λ​,a=λ+b.

Here log⁡2x=log⁡x/log⁡2\log_2x=\log x/\log2log2​x=logx/log2 uses the total real logarithm, with log⁡0=0\log0=0log0=0, so a term 0log⁡200\log_200log2​0 is zero. The subtraction d−1d-1d−1 is natural-number subtraction, which equals ordinary subtraction because d≥2d\ge2d≥2; the matrices σi\sigma_iσi​ do not enter this formula.

QAlgorithms.Papers.Independent.Intakenext100.P60.depolarizingWeight

For any natural number ddd and data DDD consisting of d≥2d\ge2d≥2, a real parameter −1/(d2−1)≤λ≤1-1/(d^2-1)\le\lambda\le1−1/(d2−1)≤λ≤1, and complex d×dd\times dd×d matrices σi\sigma_iσi​, 0≤i<d20\le i<d^20≤i<d2, with σ0=Id\sigma_0=I_dσ0​=Id​, σi†σi=Id\sigma_i^\dagger\sigma_i=I_dσi†​σi​=Id​, and Tr⁡(σi†σj)=d\operatorname{Tr}(\sigma_i^\dagger\sigma_j)=dTr(σi†​σj​)=d for i=ji=ji=j and zero otherwise, define

p0=λ+1−λd2,wD(i)={p0,i=0,1−p0d2−1,i≠0.p_0=\lambda+\frac{1-\lambda}{d^2},\qquad w_D(i)= \begin{cases} p_0,&i=0,\\[2pt] \dfrac{1-p_0}{d^2-1},&i\ne0. \end{cases}p0​=λ+d21−λ​,wD​(i)=⎩⎨⎧​p0​,d2−11−p0​​,​i=0,i=0.​

This defines a real number for every index i∈{0,…,d2−1}i\in\{0,\ldots,d^2-1\}i∈{0,…,d2−1}; the matrices in DDD do not enter the formula. Both endpoints of the parameter interval are allowed, so weights may vanish.

QAlgorithms.Papers.Independent.Intakenext100.P60.dilationCapacity

For arbitrary natural numbers a,b,ka,b,ka,b,k and any complex matrix W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, this definition is the real supremum, over all finite ensembles (pj,vj)j=0m−1(p_j,v_j)_{j=0}^{m-1}(pj​,vj​)j=0m−1​ with pj≥0p_j\ge0pj​≥0, ∑jpj=1\sum_jp_j=1∑j​pj​=1, and ∥vj∥=1\|v_j\|=1∥vj​∥=1, of

S ⁣(Tr⁡Ca(W(∑jpjvjvj†)W†))−∑jpjS ⁣(Tr⁡Ca((Wvj)(Wvj)†)).S\!\left(\operatorname{Tr}_{\mathbb C^a} \left(W\left(\sum_jp_jv_jv_j^\dagger\right)W^\dagger\right)\right) - \sum_jp_jS\!\left(\operatorname{Tr}_{\mathbb C^a} \bigl((Wv_j)(Wv_j)^\dagger\bigr)\right).S(TrCa​(W(j∑​pj​vj​vj†​)W†))−j∑​pj​S(TrCa​((Wvj​)(Wvj​)†)).

Here mmm ranges over natural numbers, the partial trace sums over the aaa-dimensional factor, and S(X)=−Re⁡tr⁡f(X)S(X)=-\operatorname{Re}\operatorname{tr}f(X)S(X)=−Retrf(X), where f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x is applied by continuous functional calculus. No isometry condition is imposed on WWW. If k=0k=0k=0, there are no admissible ensembles, and the real supremum of the empty set used here is 000; if a=0a=0a=0 or b=0b=0b=0, all the displayed entropy terms are zero.

QAlgorithms.Papers.Independent.Intakenext100.P60.dilationHolevo

For arbitrary natural numbers a,b,ka,b,ka,b,k, a complex matrix W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, and a finite pure ensemble eee consisting of a natural number mmm, probabilities pj≥0p_j\ge0pj​≥0 with ∑j=0m−1pj=1\sum_{j=0}^{m-1}p_j=1∑j=0m−1​pj​=1, and vectors vj∈Ckv_j\in\mathbb C^kvj​∈Ck with ∥vj∥=1\|v_j\|=1∥vj​∥=1 for every jjj, this definition returns

S ⁣(Tr⁡Ca(W(∑j=0m−1pjvjvj†)W†))−∑j=0m−1pjS ⁣(Tr⁡Ca((Wvj)(Wvj)†)).S\!\left(\operatorname{Tr}_{\mathbb C^a} \left(W\left(\sum_{j=0}^{m-1}p_jv_jv_j^\dagger\right)W^\dagger\right)\right) - \sum_{j=0}^{m-1}p_j S\!\left(\operatorname{Tr}_{\mathbb C^a}\bigl((Wv_j)(Wv_j)^\dagger\bigr)\right).S(TrCa​(W(j=0∑m−1​pj​vj​vj†​)W†))−j=0∑m−1​pj​S(TrCa​((Wvj​)(Wvj​)†)).

The partial trace sums over the first, aaa-dimensional factor, and S(X)=−Re⁡tr⁡f(X)S(X)=-\operatorname{Re}\operatorname{tr}f(X)S(X)=−Retrf(X), with f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x applied by continuous functional calculus. The matrix WWW is arbitrary, so the definition imposes no trace-one requirement on its output matrices. Although m=0m=0m=0 and k=0k=0k=0 are natural-number possibilities, no such ensemble exists in either case: empty probabilities cannot sum to one, and the zero-dimensional vector space has no unit vector.

QAlgorithms.Papers.Independent.Intakenext100.P60.dilationSpace

For arbitrary types I,JI,JI,J and any complex matrix WWW with rows indexed by III and columns indexed by JJJ, this defines the complex linear subspace of CI\mathbb C^ICI spanned by its columns:

dilationSpace⁡(W)=span⁡C{ i↦Wi,j:j∈J }.\operatorname{dilationSpace}(W) =\operatorname{span}_{\mathbb C}\{\,i\mapsto W_{i,j}:j\in J\,\}.dilationSpace(W)=spanC​{i↦Wi,j​:j∈J}.

The span consists of finite linear combinations, with no topological closure. Neither index type must be finite, and if JJJ is empty the result is the zero subspace.

QAlgorithms.Papers.Independent.Intakenext100.P60.embedded

For arbitrary natural numbers a,b,ka,b,ka,b,k, any complex matrix WWW with rows indexed by {0,…,a−1}×{0,…,b−1}\{0,\ldots,a-1\}\times\{0,\ldots,b-1\}{0,…,a−1}×{0,…,b−1} and columns indexed by {0,…,k−1}\{0,\ldots,k-1\}{0,…,k−1}, and any vector v∈Ckv\in\mathbb C^kv∈Ck, this definition returns the vector Wv∈Ca×bWv\in\mathbb C^{a\times b}Wv∈Ca×b, with coordinates

(Wv)r,s=∑j=0k−1W(r,s),jvj.(Wv)_{r,s}=\sum_{j=0}^{k-1}W_{(r,s),j}v_j.(Wv)r,s​=j=0∑k−1​W(r,s),j​vj​.

Neither normalization of vvv nor an isometry condition on WWW is assumed. Zero dimensions are allowed: k=0k=0k=0 makes every coordinate an empty sum, hence zero, and a=0a=0a=0 or b=0b=0b=0 gives a vector on an empty coordinate set.

QAlgorithms.Papers.Independent.Intakenext100.P60.exampleSevenDilation

This defines a complex matrix WWW with rows indexed by (r,e)∈{0,1,2}×{0,1,2,3,4,5}(r,e)\in\{0,1,2\}\times\{0,1,2,3,4,5\}(r,e)∈{0,1,2}×{0,1,2,3,4,5} and columns indexed by j∈{0,1,2}j\in\{0,1,2\}j∈{0,1,2}, whose entries are

W(r,e),j={2/2,r=j and e=j+3,1/2,(r,e)=((j+1) mod 3,(j+2) mod 3) or (r,e)=((j+2) mod 3,(j+1) mod 3),0,otherwise.W_{(r,e),j}= \begin{cases} \sqrt2/2,&r=j\text{ and }e=j+3,\\ 1/2,&(r,e)=((j+1)\bmod3,(j+2)\bmod3) \text{ or }(r,e)=((j+2)\bmod3,(j+1)\bmod3),\\ 0,&\text{otherwise}. \end{cases}W(r,e),j​=⎩⎨⎧​2​/2,1/2,0,​r=j and e=j+3,(r,e)=((j+1)mod3,(j+2)mod3) or (r,e)=((j+2)mod3,(j+1)mod3),otherwise.​

The first case takes priority. Thus, writing ∣r,e⟩|r,e\rangle∣r,e⟩ for the coordinate vector that is 111 at (r,e)(r,e)(r,e) and 000 elsewhere, column jjj is (2/2)∣j,j+3⟩+12∣(j+1) mod 3,(j+2) mod 3⟩+12∣(j+2) mod 3,(j+1) mod 3⟩(\sqrt2/2)|j,j+3\rangle+\frac12|(j+1)\bmod3,(j+2)\bmod3\rangle+\frac12|(j+2)\bmod3,(j+1)\bmod3\rangle(2​/2)∣j,j+3⟩+21​∣(j+1)mod3,(j+2)mod3⟩+21​∣(j+2)mod3,(j+1)mod3⟩.

QAlgorithms.Papers.Independent.Intakenext100.P60.exampleSevenMixtureDilation

For arbitrary complex numbers c,sc,sc,s, this defines a complex matrix V(c,s)V(c,s)V(c,s) with columns j∈{0,1,2}j\in\{0,1,2\}j∈{0,1,2} and rows indexed by (r,ξ)(r,\xi)(r,ξ), where r∈{0,1,2}r\in\{0,1,2\}r∈{0,1,2} and ξ\xiξ belongs to the disjoint union of {0,…,5}\{0,\ldots,5\}{0,…,5} and {0,1,2}2\{0,1,2\}^2{0,1,2}2. On the first part of that disjoint union,

V(c,s)(r,inl⁡e),j={c2/2,r=j, e=j+3,c/2,(r,e)=((j+1) mod 3,(j+2) mod 3) or (r,e)=((j+2) mod 3,(j+1) mod 3),0,otherwise,V(c,s)_{(r,\operatorname{inl}e),j}= \begin{cases} c\sqrt2/2,&r=j,\ e=j+3,\\ c/2,&(r,e)=((j+1)\bmod3,(j+2)\bmod3) \text{ or }(r,e)=((j+2)\bmod3,(j+1)\bmod3),\\ 0,&\text{otherwise}, \end{cases}V(c,s)(r,inle),j​=⎩⎨⎧​c2​/2,c/2,0,​r=j, e=j+3,(r,e)=((j+1)mod3,(j+2)mod3) or (r,e)=((j+2)mod3,(j+1)mod3),otherwise,​

with the first case taking priority. On the second part,

V(c,s)(r,inr⁡(u,v)),j={s/3,r=u and v=j,0,otherwise.V(c,s)_{(r,\operatorname{inr}(u,v)),j}= \begin{cases} s/\sqrt3,&r=u\text{ and }v=j,\\ 0,&\text{otherwise}. \end{cases}V(c,s)(r,inr(u,v)),j​={s/3​,0,​r=u and v=j,otherwise.​

The two parts of the row label remain distinct even when their numerical components coincide. There is no normalization or nonzero assumption on ccc or sss.

QAlgorithms.Papers.Independent.Intakenext100.P60.exampleSevenMixtureEntropy

For every complex number ccc, this defines the real number

E(c)=−alog⁡2a−2blog⁡2b,t=∣c∣2,a=1−t3+t2,b=1−t3+t4.E(c)=-a\log_2 a-2b\log_2 b,\qquad t=|c|^2,\quad a=\frac{1-t}{3}+\frac t2,\quad b=\frac{1-t}{3}+\frac t4.E(c)=−alog2​a−2blog2​b,t=∣c∣2,a=31−t​+2t​,b=31−t​+4t​.

There is no assumption that ∣c∣≤1|c|\le1∣c∣≤1, so this formula also applies when bbb is zero or negative. The logarithm is the total real base-222 logarithm used by the code: log⁡20=0\log_2 0=0log2​0=0, and for nonzero real xxx, log⁡2x=log⁡∣x∣/log⁡2\log_2 x=\log|x|/\log2log2​x=log∣x∣/log2.

QAlgorithms.Papers.Independent.Intakenext100.P60.finiteBasisState

For any finite types A,BA,BA,B and any complex matrix ρ\rhoρ indexed by A×BA\times BA×B, let a=∣A∣a=|A|a=∣A∣, b=∣B∣b=|B|b=∣B∣, and let α:A→{0,…,a−1}\alpha:A\to\{0,\ldots,a-1\}α:A→{0,…,a−1} and β:B→{0,…,b−1}\beta:B\to\{0,\ldots,b-1\}β:B→{0,…,b−1} be the chosen finite-type equivalences. This defines the matrix indexed by {0,…,a−1}×{0,…,b−1}\{0,\ldots,a-1\}\times\{0,\ldots,b-1\}{0,…,a−1}×{0,…,b−1} whose entries are

finiteBasisState⁡(ρ)(i,j),(i′,j′)=ρ(α−1(i),β−1(j)),(α−1(i′),β−1(j′)).\operatorname{finiteBasisState}(\rho)_{(i,j),(i',j')} =\rho_{(\alpha^{-1}(i),\beta^{-1}(j)),(\alpha^{-1}(i'),\beta^{-1}(j'))}.finiteBasisState(ρ)(i,j),(i′,j′)​=ρ(α−1(i),β−1(j)),(α−1(i′),β−1(j′))​.

Thus both indices are relabelled by these chosen equivalences. The types may be empty, and ρ\rhoρ need not be positive or normalized.

QAlgorithms.Papers.Independent.Intakenext100.P60.formationCostBits

For finite types A,BA,BA,B and any complex matrix ρ\rhoρ indexed by A×BA\times BA×B, first relabel ρ\rhoρ as a matrix ρ^\widehat\rhoρ​ on Fin⁡(a)×Fin⁡(b)\operatorname{Fin}(a)\times\operatorname{Fin}(b)Fin(a)×Fin(b), where a=∣A∣a=|A|a=∣A∣, b=∣B∣b=|B|b=∣B∣, using the chosen equivalences with these finite index sets. This defines the real infimum of all E∈RE\in\mathbb RE∈R such that, for every ϵ>0\epsilon>0ϵ>0 and δ>0\delta>0δ>0, there exist natural numbers m,nm,nm,n with n>0n>0n>0 and a finite local-operations-and-classical-communication map LLL satisfying

∣E−mn∣≤δ,12∥L(Φ−⊗m)−ρ^⊗n∥1≤ϵ.\left|E-\frac mn\right|\le\delta, \qquad \frac12\left\|L(\Phi_-^{\otimes m})-\widehat\rho^{\otimes n}\right\|_1\le\epsilon.​E−nm​​≤δ,21​​L(Φ−⊗m​)−ρ​⊗n​1​≤ϵ.

Here Φ−=∣s⟩⟨s∣\Phi_-=|s\rangle\langle s|Φ−​=∣s⟩⟨s∣, with s=(∣0,1⟩−∣1,0⟩)/2s=(|0,1\rangle-|1,0\rangle)/\sqrt2s=(∣0,1⟩−∣1,0⟩)/2​; each tensor power is grouped into its two parties and relabelled by the chosen equivalences from functions on the copy indices to Fin⁡(2m)\operatorname{Fin}(2^m)Fin(2m), Fin⁡(an)\operatorname{Fin}(a^n)Fin(an), or Fin⁡(bn)\operatorname{Fin}(b^n)Fin(bn). The zeroth singlet power is the one-dimensional identity. A finite local-operations-and-classical-communication map belongs to the recursively generated class containing identity maps and maps obtained by either party applying finitely many local Kraus operators KiK_iKi​ with ∑iKi†Ki=I\sum_iK_i^\dagger K_i=I∑i​Ki†​Ki​=I, following each outcome by another map in the class, and summing the branch outputs; intermediate register dimensions may be zero. The trace norm is ∥X∥1=Re⁡Tr⁡(cfc⁡[⋅](X†X))\|X\|_1=\operatorname{Re}\operatorname{Tr}(\operatorname{cfc}[\sqrt{\cdot}](X^\dagger X))∥X∥1​=ReTr(cfc[⋅​](X†X)). No density-matrix or nonemptiness hypothesis is imposed on the original input; the infimum uses the total real-infimum convention, which gives 000 for an empty set or a set unbounded below.

QAlgorithms.Papers.Independent.Intakenext100.P60.gapPolynomial

For every real four-tuple p=(p0,p1,p2,p3)p=(p_0,p_1,p_2,p_3)p=(p0​,p1​,p2​,p3​) and every real number zzz, this definition assigns the real value

gp(z)=z4−z3+4(p0p1p2+p0p1p3+p0p2p3+p1p2p3)z−16p0p1p2p3.g_p(z)=z^4-z^3 +4(p_0p_1p_2+p_0p_1p_3+p_0p_2p_3+p_1p_2p_3)z -16p_0p_1p_2p_3.gp​(z)=z4−z3+4(p0​p1​p2​+p0​p1​p3​+p0​p2​p3​+p1​p2​p3​)z−16p0​p1​p2​p3​.

No sign, normalization, or admissibility conditions on ppp are required.

QAlgorithms.Papers.Independent.Intakenext100.P60.jointDepolarizing

For any natural number nnn, any family of natural dimensions did_idi​, 0≤i<n0\le i<n0≤i<n, and data DiD_iDi​ consisting of di≥2d_i\ge2di​≥2, parameters −1/(di2−1)≤λi≤1-1/(d_i^2-1)\le\lambda_i\le1−1/(di2​−1)≤λi​≤1, and matrices σi,t∈Cdi×di\sigma_{i,t}\in\mathbb C^{d_i\times d_i}σi,t​∈Cdi​×di​, 0≤t<di20\le t<d_i^20≤t<di2​, satisfying σi,0=Idi\sigma_{i,0}=I_{d_i}σi,0​=Idi​​, σi,t†σi,t=Idi\sigma_{i,t}^\dagger\sigma_{i,t}=I_{d_i}σi,t†​σi,t​=Idi​​, and Tr⁡(σi,t†σi,u)=diδtu\operatorname{Tr}(\sigma_{i,t}^\dagger\sigma_{i,u})=d_i\delta_{tu}Tr(σi,t†​σi,u​)=di​δtu​, this defines a complex matrix with rows indexed by (x,y)∈(∏i{0,…,di−1})×(∏i{0,…,di2−1})(x,y)\in(\prod_i\{0,\ldots,d_i-1\})\times(\prod_i\{0,\ldots,d_i^2-1\})(x,y)∈(∏i​{0,…,di​−1})×(∏i​{0,…,di2​−1}) and columns indexed by j∈∏i{0,…,di−1}j\in\prod_i\{0,\ldots,d_i-1\}j∈∏i​{0,…,di​−1}, by

W(x,y),j=∏i<nwi,yi (σi,yi)xi,ji,wi,t={pi,0,t=0,(1−pi,0)/(di2−1),t≠0,pi,0=λi+1−λidi2.W_{(x,y),j}=\prod_{i<n}\sqrt{w_{i,y_i}}\,(\sigma_{i,y_i})_{x_i,j_i}, \qquad w_{i,t}= \begin{cases} p_{i,0},&t=0,\\ (1-p_{i,0})/(d_i^2-1),&t\ne0, \end{cases} \quad p_{i,0}=\lambda_i+\frac{1-\lambda_i}{d_i^2}.W(x,y),j​=i<n∏​wi,yi​​​(σi,yi​​)xi​,ji​​,wi,t​={pi,0​,(1−pi,0​)/(di2​−1),​t=0,t=0,​pi,0​=λi​+di2​1−λi​​.

All square roots are real square roots embedded into C\mathbb CC. The case n=0n=0n=0 is allowed: all indexing products are singletons, no component data are required, and the sole matrix entry is the empty product 111.

QAlgorithms.Papers.Independent.Intakenext100.P60.minimizingProjectorHull

For a group GGG equipped with a topology, data DDD comprising positive dimensions a,b,ka,b,ka,b,k, an isometry W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, and entrywise continuous irreducible unitary representations VVV and UUU satisfying Tr⁡Ca(WVgXVg†W†)=UgTr⁡Ca(WXW†)Ug†\operatorname{Tr}_{\mathbb C^a}(WV_gXV_g^\dagger W^\dagger)=U_g\operatorname{Tr}_{\mathbb C^a}(WXW^\dagger)U_g^\daggerTrCa​(WVg​XVg†​W†)=Ug​TrCa​(WXW†)Ug†​ for all g,Xg,Xg,X, and any vector v∈Ckv\in\mathbb C^kv∈Ck, this definition is the real convex hull

conv⁡R{(WVgv)(WVgv)†:g∈G}.\operatorname{conv}_{\mathbb R} \left\{(WV_gv)(WV_gv)^\dagger:g\in G\right\}.convR​{(WVg​v)(WVg​v)†:g∈G}.

Thus its elements are finite convex combinations of these matrices, with nonnegative real coefficients summing to one. Here W†W=IkW^\dagger W=I_kW†W=Ik​, irreducibility means that the only common invariant complex subspaces are zero and the whole space, and the partial trace sums over the first factor. The argument vvv is arbitrary; this definition requires neither unit norm nor an entropy-minimizing property, and it takes a convex hull without adding a closure.

QAlgorithms.Papers.Independent.Intakenext100.P60.minimumEntanglement

For a group GGG with a topology and data DDD consisting of positive dimensions a,b,ka,b,ka,b,k, an isometry W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, and entrywise continuous irreducible unitary representations VVV on Ck\mathbb C^kCk and UUU on Cb\mathbb C^bCb satisfying Tr⁡Ca(WVgXVg†W†)=UgTr⁡Ca(WXW†)Ug†\operatorname{Tr}_{\mathbb C^a}(WV_gXV_g^\dagger W^\dagger)=U_g\operatorname{Tr}_{\mathbb C^a}(WXW^\dagger)U_g^\daggerTrCa​(WVg​XVg†​W†)=Ug​TrCa​(WXW†)Ug†​ for every g,Xg,Xg,X, this definition is the real infimum

inf⁡v∈Ck∥v∥=1S ⁣(Tr⁡Cb((Wv)(Wv)†)).\inf_{\substack{v\in\mathbb C^k\\ \|v\|=1}} S\!\left(\operatorname{Tr}_{\mathbb C^b}\bigl((Wv)(Wv)^\dagger\bigr)\right).v∈Ck∥v∥=1​inf​S(TrCb​((Wv)(Wv)†)).

Here irreducibility means that only the zero and full complex subspaces are invariant under every representation matrix, WvWvWv is ordinary matrix multiplication, (Tr⁡CbY)r,s=∑tY(r,t),(s,t)(\operatorname{Tr}_{\mathbb C^b}Y)_{r,s}=\sum_tY_{(r,t),(s,t)}(TrCb​Y)r,s​=∑t​Y(r,t),(s,t)​, and S(X)=−Re⁡tr⁡f(X)S(X)=-\operatorname{Re}\operatorname{tr}f(X)S(X)=−Retrf(X), where f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x is applied by continuous functional calculus. The unit-vector indexing set is nonempty because k>0k>0k>0; the definition itself asserts an infimum, without asserting attainment.

QAlgorithms.Papers.Independent.Intakenext100.P60.pairMarginal

For any natural number nnn, natural-number families ai,bia_i,b_iai​,bi​, 0≤i<n0\le i<n0≤i<n, any complex matrix ρ\rhoρ indexed on (∏i{0,…,ai−1})×(∏i{0,…,bi−1})(\prod_i\{0,\ldots,a_i-1\})\times(\prod_i\{0,\ldots,b_i-1\})(∏i​{0,…,ai​−1})×(∏i​{0,…,bi​−1}), and any index i<ni<ni<n, this defines a matrix RiR_iRi​ indexed on {0,…,ai−1}×{0,…,bi−1}\{0,\ldots,a_i-1\}\times\{0,\ldots,b_i-1\}{0,…,ai​−1}×{0,…,bi​−1} by

(Ri)(u,v),(u′,v′)=∑x∈∏j{0,…,aj−1}xi=u ∑y∈∏j{0,…,bj−1}yi=vρ(x,y),(x[i←u′], y[i←v′]).(R_i)_{(u,v),(u',v')}= \sum_{\substack{x\in\prod_j\{0,\ldots,a_j-1\}\\x_i=u}} \ \sum_{\substack{y\in\prod_j\{0,\ldots,b_j-1\}\\y_i=v}} \rho_{(x,y),(x[i\leftarrow u'],\,y[i\leftarrow v'])}.(Ri​)(u,v),(u′,v′)​=x∈∏j​{0,…,aj​−1}xi​=u​∑​ y∈∏j​{0,…,bj​−1}yi​=v​∑​ρ(x,y),(x[i←u′],y[i←v′])​.

The notation x[i←u′]x[i\leftarrow u']x[i←u′] means replacing only coordinate iii by u′u'u′. No positivity or trace assumption on ρ\rhoρ, and no positivity assumption on the dimensions, is imposed. Empty summation sets contribute zero; when n=0n=0n=0, there is no index iii at which to apply the definition.

QAlgorithms.Papers.Independent.Intakenext100.P60.pauliDilation

For every real four-tuple p=(p0,p1,p2,p3)p=(p_0,p_1,p_2,p_3)p=(p0​,p1​,p2​,p3​), the definition gives a complex matrix DpD_pDp​ with rows indexed by (a,r)∈{0,1}×{0,1,2,3}(a,r)\in\{0,1\}\times\{0,1,2,3\}(a,r)∈{0,1}×{0,1,2,3} and columns indexed by j∈{0,1}j\in\{0,1\}j∈{0,1}, namely

(Dp)(a,r),j=pr (Pr)aj,P0=(1001),P1=(0110),P2=(0−ii0),P3=(100−1).(D_p)_{(a,r),j}=\sqrt{p_r}\,(P_r)_{aj},\qquad P_0=\begin{pmatrix}1&0\\0&1\end{pmatrix},\quad P_1=\begin{pmatrix}0&1\\1&0\end{pmatrix},\quad P_2=\begin{pmatrix}0&-\mathrm i\\ \mathrm i&0\end{pmatrix},\quad P_3=\begin{pmatrix}1&0\\0&-1\end{pmatrix}.(Dp​)(a,r),j​=pr​​(Pr​)aj​,P0​=(10​01​),P1​=(01​10​),P2​=(0i​−i0​),P3​=(10​0−1​).

Here each square root is the real square root, embedded into C\mathbb CC. No nonnegativity, normalization, or admissibility condition is imposed on ppp; in particular, a negative prp_rpr​ has real square root zero under the total square-root convention used by the definition.

QAlgorithms.Papers.Independent.Intakenext100.P60.pauliDilationWord

For every real four-tuple ppp, natural number nnn, and word w:{0,…,n−1}→{0,1}w:\{0,\ldots,n-1\}\to\{0,1\}w:{0,…,n−1}→{0,1}, the definition gives a complex Euclidean vector indexed by pairs of words x:{0,…,n−1}→{0,1}x:\{0,\ldots,n-1\}\to\{0,1\}x:{0,…,n−1}→{0,1} and r:{0,…,n−1}→{0,1,2,3}r:\{0,\ldots,n-1\}\to\{0,1,2,3\}r:{0,…,n−1}→{0,1,2,3}, with amplitude

Vp,n,w(x,r)=∏i=0n−1pri (Pri)xi,wi,V_{p,n,w}(x,r)=\prod_{i=0}^{n-1}\sqrt{p_{r_i}}\,(P_{r_i})_{x_i,w_i},Vp,n,w​(x,r)=i=0∏n−1​pri​​​(Pri​​)xi​,wi​​,

where

P0=(1001),P1=(0110),P2=(0−ii0),P3=(100−1).P_0=\begin{pmatrix}1&0\\0&1\end{pmatrix},\quad P_1=\begin{pmatrix}0&1\\1&0\end{pmatrix},\quad P_2=\begin{pmatrix}0&-\mathrm i\\ \mathrm i&0\end{pmatrix},\quad P_3=\begin{pmatrix}1&0\\0&-1\end{pmatrix}.P0​=(10​01​),P1​=(01​10​),P2​=(0i​−i0​),P3​=(10​0−1​).

The square roots are real, with negative inputs assigned zero. No admissibility or normalization of ppp is assumed. The case n=0n=0n=0 is included: all three word types then have one empty word, and the sole amplitude is the empty product 111.

QAlgorithms.Papers.Independent.Intakenext100.P60.pauliMixture

For every real four-tuple p=(p0,p1,p2,p3)p=(p_0,p_1,p_2,p_3)p=(p0​,p1​,p2​,p3​) and every real number sss, define vectors in C{0,1}×{0,1,2,3}\mathbb C^{\{0,1\}\times\{0,1,2,3\}}C{0,1}×{0,1,2,3} by

v0=p0∣0,0⟩+p1∣1,1⟩+ip2∣1,2⟩+p3∣0,3⟩,v1=p0∣1,0⟩+p1∣0,1⟩−ip2∣0,2⟩−p3∣1,3⟩,v_0=\sqrt{p_0}|0,0\rangle+\sqrt{p_1}|1,1\rangle +\mathrm i\sqrt{p_2}|1,2\rangle+\sqrt{p_3}|0,3\rangle,\qquad v_1=\sqrt{p_0}|1,0\rangle+\sqrt{p_1}|0,1\rangle -\mathrm i\sqrt{p_2}|0,2\rangle-\sqrt{p_3}|1,3\rangle,v0​=p0​​∣0,0⟩+p1​​∣1,1⟩+ip2​​∣1,2⟩+p3​​∣0,3⟩,v1​=p0​​∣1,0⟩+p1​​∣0,1⟩−ip2​​∣0,2⟩−p3​​∣1,3⟩,

where the kets are coordinate unit vectors and the square roots are real square roots, equal to zero on negative inputs. The definition assigns the matrix

Rp,s=s v0v0†+(1−s) v1v1†,R_{p,s}=s\,v_0v_0^\dagger+(1-s)\,v_1v_1^\dagger,Rp,s​=sv0​v0†​+(1−s)v1​v1†​,

so its (q,q′)(q,q')(q,q′)-entry is s v0(q)v0(q′)‾+(1−s) v1(q)v1(q′)‾s\,v_0(q)\overline{v_0(q')}+(1-s)\,v_1(q)\overline{v_1(q')}sv0​(q)v0​(q′)​+(1−s)v1​(q)v1​(q′)​. Neither ppp nor sss is restricted, and the vectors are not required to have norm 111; the definition therefore does not itself require this matrix to be a density matrix.

QAlgorithms.Papers.Independent.Intakenext100.P60.pauliWordMixture

For every real four-tuple ppp, natural number nnn, and arbitrary real function μ\muμ on binary words of length nnn, define VwV_wVw​, indexed by a binary word xxx and a four-symbol word rrr, by

Vw(x,r)=∏i=0n−1pri (Pri)xi,wi,P0=(1001), P1=(0110), P2=(0−ii0), P3=(100−1).V_w(x,r)=\prod_{i=0}^{n-1}\sqrt{p_{r_i}}\,(P_{r_i})_{x_i,w_i}, \quad P_0=\begin{pmatrix}1&0\\0&1\end{pmatrix},\ P_1=\begin{pmatrix}0&1\\1&0\end{pmatrix},\ P_2=\begin{pmatrix}0&-\mathrm i\\ \mathrm i&0\end{pmatrix},\ P_3=\begin{pmatrix}1&0\\0&-1\end{pmatrix}.Vw​(x,r)=i=0∏n−1​pri​​​(Pri​​)xi​,wi​​,P0​=(10​01​), P1​=(01​10​), P2​=(0i​−i0​), P3​=(10​0−1​).

The definition assigns

Rp,n,μ=∑w∈{0,1}nμ(w) VwVw†.R_{p,n,\mu}=\sum_{w\in\{0,1\}^n}\mu(w)\,V_wV_w^\dagger.Rp,n,μ​=w∈{0,1}n∑​μ(w)Vw​Vw†​.

Real square roots are used, with negative inputs assigned zero. There are no assumptions that ppp or μ\muμ is nonnegative or normalized. For n=0n=0n=0, the matrix has one entry, equal to μ(∅)\mu(\varnothing)μ(∅), because each vector VwV_wVw​ has sole amplitude 111.

QAlgorithms.Papers.Independent.Intakenext100.P60.productMatrices

For arbitrary types A,BA,BA,B, any natural number nnn, and any family of complex matrices ρi\rho_iρi​ indexed on A×BA\times BA×B, 0≤i<n0\le i<n0≤i<n, this defines a complex matrix on An×BnA^n\times B^nAn×Bn by

P(x,y),(x′,y′)=∏i=0n−1(ρi)(xi,yi),(xi′,yi′),P_{(x,y),(x',y')}=\prod_{i=0}^{n-1}(\rho_i)_{(x_i,y_i),(x'_i,y'_i)},P(x,y),(x′,y′)​=i=0∏n−1​(ρi​)(xi​,yi​),(xi′​,yi′​)​,

where AnA^nAn and BnB^nBn mean functions from {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1} into AAA and BBB, respectively. There are no finiteness, nonemptiness, positivity, trace, or normalization assumptions on A,B,ρiA,B,\rho_iA,B,ρi​. When n=0n=0n=0, both function spaces are singletons and the empty product makes the sole matrix entry equal to 111.

QAlgorithms.Papers.Independent.Intakenext100.P60.productMixture

For arbitrary types A,BA,BA,B, natural numbers n,mn,mn,m, arbitrary real coefficients μj\mu_jμj​, 0≤j<m0\le j<m0≤j<m, and arbitrary complex matrices ρj,i\rho_{j,i}ρj,i​ indexed on A×BA\times BA×B, 0≤i<n0\le i<n0≤i<n, this defines the matrix on An×BnA^n\times B^nAn×Bn with entries

X(x,y),(x′,y′)=∑j=0m−1μj∏i=0n−1(ρj,i)(xi,yi),(xi′,yi′).X_{(x,y),(x',y')}= \sum_{j=0}^{m-1}\mu_j \prod_{i=0}^{n-1}(\rho_{j,i})_{(x_i,y_i),(x'_i,y'_i)}.X(x,y),(x′,y′)​=j=0∑m−1​μj​i=0∏n−1​(ρj,i​)(xi​,yi​),(xi′​,yi′​)​.

The coefficients are embedded into C\mathbb CC, and no nonnegativity, unit-sum, density, or finiteness assumptions on the types are imposed. If m=0m=0m=0, the result is the zero matrix; if n=0n=0n=0, the indexing space is a singleton and its sole entry is ∑j<mμj\sum_{j<m}\mu_j∑j<m​μj​, including zero when also m=0m=0m=0.

QAlgorithms.Papers.Independent.Intakenext100.P60.projectorState

For a group GGG equipped with a topology, natural numbers a,b,ka,b,ka,b,k, and data DDD consisting of positive a,b,ka,b,ka,b,k, an isometry W:Ck→Ca⊗CbW:\mathbb C^k\to\mathbb C^a\otimes\mathbb C^bW:Ck→Ca⊗Cb, and entrywise continuous unitary representations V:G→Mk(C)V:G\to M_k(\mathbb C)V:G→Mk​(C), U:G→Mb(C)U:G\to M_b(\mathbb C)U:G→Mb​(C), each having only zero and full invariant complex subspaces and satisfying Tr⁡Ca(WVgXVg†W†)=UgTr⁡Ca(WXW†)Ug†\operatorname{Tr}_{\mathbb C^a}(WV_gXV_g^\dagger W^\dagger)=U_g\operatorname{Tr}_{\mathbb C^a}(WXW^\dagger)U_g^\daggerTrCa​(WVg​XVg†​W†)=Ug​TrCa​(WXW†)Ug†​ for all g,Xg,Xg,X, this definition returns the matrix

1k WW†\frac1k\,WW^\daggerk1​WW†

on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb. Here W†W=IkW^\dagger W=I_kW†W=Ik​, and the partial trace in the data sums over the Ca\mathbb C^aCa index. The data require k>0k>0k>0, so the reciprocal in this definition has a nonzero denominator.

QAlgorithms.Papers.Independent.Intakenext100.P60.pureEntropyBits

For finite types A,BA,BA,B, with decidable equality on AAA, and any vector v∈CA×Bv\in\mathbb C^{A\times B}v∈CA×B equipped with its Euclidean norm, this defines

pureEntropyBits⁡(v)=−Re⁡Tr⁡ ⁣(cfc⁡[x↦xlog⁡2x](R)),Ra,a′=∑b∈Bva,bva′,b‾.\operatorname{pureEntropyBits}(v) =-\operatorname{Re}\operatorname{Tr}\!\left(\operatorname{cfc}[x\mapsto x\log_2 x](R)\right), \qquad R_{a,a'}=\sum_{b\in B}v_{a,b}\overline{v_{a',b}}.pureEntropyBits(v)=−ReTr(cfc[x↦xlog2​x](R)),Ra,a′​=b∈B∑​va,b​va′,b​​.

Here cfc⁡\operatorname{cfc}cfc denotes continuous functional calculus, and log⁡2\log_2log2​ is the total real logarithm to base 222. No normalization of vvv or nonemptiness of either type is assumed; zero vectors and empty index types are included.

QAlgorithms.Papers.Independent.Intakenext100.P60.pureOutput

For any complex inner-product space HHH with its normed additive commutative group structure, any natural number ddd, any complex-linear map T:Md(C)→LC(H,H)T:M_d(\mathbb C)\to\mathcal L_{\mathbb C}(H,H)T:Md​(C)→LC​(H,H) belonging to the specified structure, and any v∈Hv\in Hv∈H, this defines a complex d×dd\times dd×d matrix by

pureOutput⁡(T,v)i,j=⟨v,T(Ej,i)v⟩,\operatorname{pureOutput}(T,v)_{i,j} =\langle v,T(E_{j,i})v\rangle,pureOutput(T,v)i,j​=⟨v,T(Ej,i​)v⟩,

where Ej,iE_{j,i}Ej,i​ has entry 111 in row jjj, column iii, and zero elsewhere. The structure requires T(Id)=IHT(I_d)=I_HT(Id​)=IH​ and, for every kkk, positive semidefinite block matrix X=(Xr,s)X=(X_{r,s})X=(Xr,s​), and vectors wr∈Hw_r\in Hwr​∈H, that ∑r,s⟨wr,T(Xr,s)ws⟩\sum_{r,s}\langle w_r,T(X_{r,s})w_s\rangle∑r,s​⟨wr​,T(Xr,s​)ws​⟩ be real and nonnegative. The vector vvv need not be normalized, HHH need not be complete, and d=0d=0d=0 is included whenever such a structure exists.

QAlgorithms.Papers.Independent.Intakenext100.P60.signalBarycenter

For any normed additive commutative group HHH, any type III, any function S:H→MI(C)S:H\to M_I(\mathbb C)S:H→MI​(C), and any finite ensemble consisting of real weights pj≥0p_j\ge0pj​≥0 summing to 111 and vectors vj∈Hv_j\in Hvj​∈H of norm 111, this defines

signalBarycenter⁡(S,e)=∑j=0m−1pjS(vj),\operatorname{signalBarycenter}(S,e)=\sum_{j=0}^{m-1}p_jS(v_j),signalBarycenter(S,e)=j=0∑m−1​pj​S(vj​),

where mmm is the ensemble size and the real weights act as complex scalars. No finiteness assumption on III, or linearity, positivity, or normalization assumption on SSS, is imposed. Zero weights and repeated vectors are allowed; an ensemble of size zero cannot satisfy the weight requirement.

QAlgorithms.Papers.Independent.Intakenext100.P60.signalCapacity

For any normed additive commutative group HHH, any finite type III with decidable equality, and any function S:H→MI(C)S:H\to M_I(\mathbb C)S:H→MI​(C), this defines the real supremum

signalCapacity⁡(S)=sup⁡e[S ⁣(∑j<mpjS(vj))−∑j<mpjS(S(vj))],S(X)=−Re⁡Tr⁡ ⁣(cfc⁡[x↦xlog⁡2x](X)).\operatorname{signalCapacity}(S) =\sup_e\left[ \mathcal S\!\left(\sum_{j<m}p_jS(v_j)\right) -\sum_{j<m}p_j\mathcal S(S(v_j)) \right], \qquad \mathcal S(X)=-\operatorname{Re}\operatorname{Tr} \!\left(\operatorname{cfc}[x\mapsto x\log_2x](X)\right).signalCapacity(S)=esup​[S(j<m∑​pj​S(vj​))−j<m∑​pj​S(S(vj​))],S(X)=−ReTr(cfc[x↦xlog2​x](X)).

The supremum ranges over every finite ensemble size mmm, real weights pj≥0p_j\ge0pj​≥0 summing to 111, and vectors vj∈Hv_j\in Hvj​∈H with ∥vj∥=1\|v_j\|=1∥vj​∥=1 for every jjj; zero weights and repeated vectors are allowed. No linearity, positivity, or normalization of SSS is assumed, and III may be empty. This is the total real supremum, giving 000 if the set is empty—for example, when HHH has no unit vector—or unbounded above.

QAlgorithms.Papers.Independent.Intakenext100.P60.signalHolevo

For any normed additive commutative group HHH, any finite type III with decidable equality, any function S:H→MI(C)S:H\to M_I(\mathbb C)S:H→MI​(C), and any finite ensemble e=(pj,vj)j<me=(p_j,v_j)_{j<m}e=(pj​,vj​)j<m​ with pj≥0p_j\ge0pj​≥0, ∑jpj=1\sum_jp_j=1∑j​pj​=1, and ∥vj∥=1\|v_j\|=1∥vj​∥=1 for every jjj, this defines

signalHolevo⁡(S,e)=S ⁣(∑jpjS(vj))−∑jpjS(S(vj)),S(X)=−Re⁡Tr⁡ ⁣(cfc⁡[x↦xlog⁡2x](X)).\operatorname{signalHolevo}(S,e) =\mathcal S\!\left(\sum_jp_jS(v_j)\right)-\sum_jp_j\mathcal S(S(v_j)), \qquad \mathcal S(X)=-\operatorname{Re}\operatorname{Tr} \!\left(\operatorname{cfc}[x\mapsto x\log_2x](X)\right).signalHolevo(S,e)=S(j∑​pj​S(vj​))−j∑​pj​S(S(vj​)),S(X)=−ReTr(cfc[x↦xlog2​x](X)).

The functional calculus and real logarithm are the total functions appearing in the definitions. There is no requirement that SSS be linear or that its values be density matrices, and III may be empty. Zero ensemble weights are allowed, although their associated vectors must still have norm 111.

QAlgorithms.Papers.Independent.Intakenext100.P60.tensorDualOutput

Let HHH and KKK be complex inner product spaces with their normed additive group structures, without any completeness assumption, and let d,ed,ed,e be arbitrary natural numbers, including zero. Let TTT and SSS have complex linear maps T∗:Md(C)→L(H)T^*:M_d(\mathbb C)\to\mathcal L(H)T∗:Md​(C)→L(H) and S∗:Me(C)→L(K)S^*:M_e(\mathbb C)\to\mathcal L(K)S∗:Me​(C)→L(K), respectively, which send the identity matrix to the identity operator and satisfy the following positivity condition: for every natural number rrr, every positive semidefinite block matrix XXX indexed by {0,…,r−1}×{0,…,d−1}\{0,\ldots,r-1\}\times\{0,\ldots,d-1\}{0,…,r−1}×{0,…,d−1}, and every family u0,…,ur−1∈Hu_0,\ldots,u_{r-1}\in Hu0​,…,ur−1​∈H, the complex number ∑a,b⟨ua,T∗(Xab)ub⟩\sum_{a,b}\langle u_a,T^*(X_{ab})u_b\rangle∑a,b​⟨ua​,T∗(Xab​)ub​⟩ is nonnegative; impose the corresponding condition on S∗S^*S∗. For any vector vvv in the completion of H⊗CKH\otimes_{\mathbb C}KH⊗C​K, the definition assigns the matrix indexed by pairs (a,b)∈{0,…,d−1}×{0,…,e−1}(a,b)\in\{0,\ldots,d-1\}\times\{0,\ldots,e-1\}(a,b)∈{0,…,d−1}×{0,…,e−1} whose entries are

Q(a,b),(a′,b′)=⟨v,  T∗(Ea′,a)⊗S∗(Eb′,b) ^ v⟩,Q_{(a,b),(a',b')}= \left\langle v,\, \widehat{\,T^*(E_{a',a})\otimes S^*(E_{b',b})\,}\,v \right\rangle,Q(a,b),(a′,b′)​=⟨v,T∗(Ea′,a​)⊗S∗(Eb′,b​)​v⟩,

where Ej,iE_{j,i}Ej,i​ has entry 111 at row jjj, column iii, and zero elsewhere, and the hat denotes the continuous operator on the completion obtained by extending the tensor product operator through the canonical map into the completion. No normalization of vvv is required. If either output dimension is zero, the resulting matrix has an empty index set.

Human review
  • Endorsed by Shuze Chen · Oct 31, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 31, 2026

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

View graph

Get started

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

About Prove2Me

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

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me