Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Multiplicativity of maximal output purities of Gaussian channels under Gaussian inputs

Definition
QAlgorithms_Papers_Independent_Support50_P28

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

prove2quantum

The objects the formalized statements of Multiplicativity of maximal output purities of Gaussian channels under Gaussian inputs (arXiv:quant-ph/0406065) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • OscillatorPhase: PDF p.2: the source interleaved coordinate order.
  • OscillatorCovariance: PDF p.2: real covariance arrays, not density matrices.
  • OscillatorCoeff: PDF pp.2--3: arrays of actual oscillator-operator oscillatorCoefficients.
  • interleavedSigma: PDF p.2: canonical interleavedSigma with sigma_xp=+1, sigma_px=-1.
  • groupedIndex: PDF p.2: permutation to the accepted grouped Weyl coordinates; no rescaling.
  • groupedVector: PDF p.2: source phase vector in the accepted grouped coordinates.
  • groupedCovariance: PDF p.2: covariance permutation used only for the accepted Weyl formula.
  • oscillatorCharacteristic: PDF p.2, (4): oscillatorCharacteristic of the actual density, with the fixed Fock Weyl kernel.
  • WeylTraceSummable: PDF p.2, (4): convergence of both sums in the trace/Weyl pairing. Each row is an l2 pairing; the outer series is a trace-class/bounded trace.
  • PhysicalOscillatorCovariance: PDF p.2, corrected (6)--(8): physical symmetric doubled covariance.
  • HasCenteredGaussianCovariance: PDF pp.2--3: centered Gaussian ACTUAL density with its covariance. Normalization and nuclear trace class are carried by State. The corrected oscillatorCharacteristic exponent is -xi^T interleavedSigma^T G interleavedSigma xi/4, not printed (6).
  • IsCenteredGaussianDensity: PDF p.2: all centered Gaussian densities, without an energy bound.
  • IsPureOscillatorDensity: PDF pp.2,4: purity is a rank-one density of a unit vector; no infinite squeezing.
  • OscillatorOperator: PDF p.3: operators on the completed oscillator space, never a photon truncation.
  • oscillatorNumberBasis: PDF p.3: complete number basis of l2.
  • oscillatorCoefficients: PDF p.3: row=bra, column=ket, with conjugate-linearity on the bra.
  • realizeOscillatorCoeff: PDF p.3: unique bounded realization of an operator array when it exists. All arrays used as states are nuclear (State or CPTP output), hence have a unique realization on the completed number basis. The off-domain zero is not a state.
  • positivePower: PDF p.3, (24): genuine positive spectral functional calculus, independent of G.
  • tracePower: PDF pp.3--4, (24),(27): positive spectral trace, retaining infinity rather than a default-zero nonsummable real trace. On a density and p>1 it is finite.
  • pNorm: PDF p.3, (24): Schatten p-norm of an actual density, extended off its domain.
  • gaussianCovarianceMap: PDF p.3, (18): fixed covariance action; the transpose is on the left.
  • GaussianCPCondition: PDF p.3, (16): full Hermitian CP condition, including singular noise.
  • gaussianWeylArgument: PDF p.3: the Weyl-coordinate pullback compatible with (18).
  • noiseMultiplier: PDF p.3: source quadratic noise in the same Weyl coordinates, with /4.
  • CenteredOscillatorChannel: PDF p.3, (15)--(19), interpreted consistently with (18): an actual CPTP map on ALL trace class. Its defining Weyl law holds for every density, not just Gaussian inputs. No output-purity or multiplicativity assertion is a field.
  • maximalOutputPurity: PDF p.4, (31): supremum on actual centered Gaussian input densities. ENNReal suprema avoid empty/unbounded real defaults; the physical set is nonempty and <=1.
  • pureMaximalOutputPurity: PDF p.4, (34): the same supremum restricted to pure Gaussian densities.
  • replaceOccupation: PDF pp.3--4: replace one whole channel's occupation string, keeping all other modes.
  • applyOscillatorBlock: PDF pp.3--4, (36): apply a local map to every off-diagonal block. The other factors form an arbitrary completed oscillator ancilla.
  • tensorOscillatorChannel: PDF pp.3--4, (36): actual finite tensor product of the full channel maps. The fold applies each factor once; it is not a rule only on product inputs.
  • GaussianPurityMultiplicative: PDF p.4, (37): global Gaussian inputs may be entangled across channel blocks.
  • scalarX: PDF p.5, (40): canonical X computed from its factor, including det X=0.
  • scalarY: PDF p.5, (40): canonical Y computed from its factor, including det Y=0.
  • EqualSigns: PDF p.5: equal determinant signs, including the common zero sign.
  • PreservesInterleavedSigma: PDF p.2, (9): source convention S^T interleavedSigma S=interleavedSigma.
  • WilliamsonSpectrum: PDF p.2, (12): Williamson spectrum with multiplicity, not singular values of the generally nonnormal i interleavedSigma G. Ordering is immaterial to products.
  • fp: PDF p.4, (28): reuse the exact accepted f_p.
  • spectralFp: PDF p.4, (29): F_p from an explicitly identified symplectic spectrum.
  • kyFan: PDF p.4, (30): intrinsic Ky Fan sum over all r-dimensional orthonormal frames. For nonempty mode sets this equals the sum of the r largest density eigenvalues with zero padding. ENNReal prevents an unbounded-real default.
  • OscillatorMajorizes: PDF p.4, (30): spectral majorization of actual positive trace-one densities.
  • productInput: PDF pp.3--5: oscillatorCoefficients of an actual tensor product of factor states.
  • sourceBeta: PDF p.8, (A1): the one-mode sourceBeta matrix.
  • sourceDelta: PDF p.8, (A1): the one-mode sourceDelta matrix.
  • oneModeGenerator: PDF p.8, (A2): sigma2, sourceBeta and sourceDelta. Ordering does not affect a basis.
  • GeneratorIndex: PDF p.8, (A2)--(A5): three single-mode and four unordered-pair generators.
  • sourceSymplecticGenerator: PDF p.8, (A2)--(A4): the actual embedded one/two-mode matrices. Pair labels 0,1 are the two skew generators in (A3); 2,3 are the symmetric ones in (A4).
  • InSymplecticAlgebra: PDF p.8: infinitesimal symplecticity, not a subgroup of symmetric matrices.
  • CompactGeneratorIndex: PDF p.8: skew infinitesimal generators of the compact passive subgroup.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P33
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P30
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P47

/-!
Serafini--Eisert--Wolf, arXiv:quant-ph/0406065v1.
Full occupation-number Hilbert spaces, no cutoff or energy constraint.
The paper's proof convention is vacuum-I doubled covariance and hbar=1.
Printed (6),(7),(15),(38),(69),(71),(B6) have the discrepancies recorded in
MODERATION_NOTES.md. No frozen definition is changed. The defective
P47.HasBipartiteMoments predicate is never used. Spectral powers are powers of
actual bounded density operators; the Gaussian spectral formula is a theorem.
Compilation and independent moderation are pending the separate batch handoff.
-/

noncomputable section

open scoped BigOperators ENNReal ComplexOrder Matrix

namespace QAlgorithms.Papers.Independent.Support50.P28

attribute [local instance] Classical.propDecidable Classical.decEq

open QAlgorithms.RealRegisters (Hilbert)
open QAlgorithms.Papers.Independent.Dependency50.P33
  (State IsDensity IsTraceClassChannel rankOne)

open QAlgorithms.Papers.Independent.Claimed50.P47
  (Occupation gaussianStateCharacteristic singleCharacteristic weylKernel)
open QAlgorithms.Papers.Independent.Claimed50.P30 (phiMinus)

/-- PDF p.2: the source interleaved coordinate order. -/
abbrev OscillatorPhase (I : Type) := I × Fin 2

/-- PDF p.2: real covariance arrays, not density matrices. -/
abbrev OscillatorCovariance (I : Type) := Matrix (OscillatorPhase I) (OscillatorPhase I) ℝ

/-- PDF pp.2--3: arrays of actual oscillator-operator oscillatorCoefficients. -/
abbrev OscillatorCoeff (I : Type) := Matrix (Occupation I) (Occupation I) ℂ

/-- PDF p.2: canonical interleavedSigma with sigma_xp=+1, sigma_px=-1. -/
def interleavedSigma (I : Type) : OscillatorCovariance I :=
  fun i j => if i.1 = j.1 then
    if i.2 = j.2 then 0 else if i.2 = 0 then 1 else -1
  else 0

/-- PDF p.2: permutation to the accepted grouped Weyl coordinates; no rescaling. -/
def groupedIndex {I : Type} : I ⊕ I → OscillatorPhase I
  | Sum.inl i => (i, 0)
  | Sum.inr i => (i, 1)

/-- PDF p.2: source phase vector in the accepted grouped coordinates. -/
def groupedVector {I : Type} (xi : OscillatorPhase I → ℝ) : I ⊕ I → ℝ := xi ∘ groupedIndex

/-- PDF p.2: covariance permutation used only for the accepted Weyl formula. -/
def groupedCovariance {I : Type} (G : OscillatorCovariance I) : Matrix (I ⊕ I) (I ⊕ I) ℝ :=
  G.submatrix groupedIndex groupedIndex

/-- PDF p.2, (4): oscillatorCharacteristic of the actual density, with the fixed Fock Weyl kernel. -/
def oscillatorCharacteristic {I : Type} [Fintype I] (A : OscillatorCoeff I) (xi : OscillatorPhase I → ℝ) : ℂ :=
  singleCharacteristic A (groupedVector xi)

/-- PDF p.2, (4): convergence of both sums in the trace/Weyl pairing.
Each row is an l2 pairing; the outer series is a trace-class/bounded trace. -/
def WeylTraceSummable {I : Type} [Fintype I] (A : OscillatorCoeff I) (xi : OscillatorPhase I → ℝ) : Prop :=
  (∀ u, Summable (fun v => A u v * weylKernel (groupedVector xi) v u)) ∧
    Summable (fun u => ∑' v, A u v * weylKernel (groupedVector xi) v u)

/-- PDF p.2, corrected (6)--(8): physical symmetric doubled covariance. -/
def PhysicalOscillatorCovariance {I : Type} [Fintype I] (G : OscillatorCovariance I) : Prop :=
  G.transpose = G ∧
    (G.map (fun x : ℝ => (x : ℂ)) +
      Complex.I • (interleavedSigma I).map (fun x : ℝ => (x : ℂ))).PosSemidef

/-- PDF pp.2--3: centered Gaussian ACTUAL density with its covariance.
Normalization and nuclear trace class are carried by State. The corrected
oscillatorCharacteristic exponent is -xi^T interleavedSigma^T G interleavedSigma xi/4, not printed (6). -/
def HasCenteredGaussianCovariance {I : Type} [Fintype I]
    (rho : State (Occupation I)) (G : OscillatorCovariance I) : Prop :=
  PhysicalOscillatorCovariance G ∧ ∀ xi,
    WeylTraceSummable rho.coefficients xi ∧
    oscillatorCharacteristic rho.coefficients xi =
      gaussianStateCharacteristic 0 (groupedCovariance G) (groupedVector xi)

/-- PDF p.2: all centered Gaussian densities, without an energy bound. -/
def IsCenteredGaussianDensity {I : Type} [Fintype I] (rho : State (Occupation I)) : Prop :=
  ∃ G, HasCenteredGaussianCovariance rho G

/-- PDF pp.2,4: purity is a rank-one density of a unit vector; no infinite squeezing. -/
def IsPureOscillatorDensity {I : Type} (rho : State (Occupation I)) : Prop :=
  ∃ u : Hilbert (Occupation I), ‖u‖ = 1 ∧ rho.coefficients = rankOne u

/-- PDF p.3: operators on the completed oscillator space, never a photon truncation. -/
abbrev OscillatorOperator (I : Type) := Hilbert (Occupation I) →L[ℂ] Hilbert (Occupation I)

/-- PDF p.3: complete number basis of l2. -/
def oscillatorNumberBasis (I : Type) : HilbertBasis (Occupation I) ℂ (Hilbert (Occupation I)) :=
  ⟨LinearIsometryEquiv.refl ℂ (Hilbert (Occupation I))⟩

/-- PDF p.3: row=bra, column=ket, with conjugate-linearity on the bra. -/
def oscillatorCoefficients {I : Type} (T : OscillatorOperator I) : OscillatorCoeff I :=
  fun u v => inner ℂ (oscillatorNumberBasis I u) (T (oscillatorNumberBasis I v))

/-- PDF p.3: unique bounded realization of an operator array when it exists.
All arrays used as states are nuclear (State or CPTP output), hence have a unique
realization on the completed number basis. The off-domain zero is not a state. -/
def realizeOscillatorCoeff {I : Type} (A : OscillatorCoeff I) : OscillatorOperator I :=
  if h : ∃ T : OscillatorOperator I, oscillatorCoefficients T = A then Classical.choose h else 0

/-- PDF p.3, (24): genuine positive spectral functional calculus, independent of G. -/
def positivePower {I : Type} (A : OscillatorCoeff I) (p : ℝ) : OscillatorOperator I := by
  letI : ContinuousFunctionalCalculus ℂ (OscillatorOperator I) IsStarNormal :=
    IsStarNormal.instContinuousFunctionalCalculus (A := OscillatorOperator I)
  letI : ContinuousFunctionalCalculus ℝ (OscillatorOperator I) IsSelfAdjoint :=
    IsSelfAdjoint.instContinuousFunctionalCalculus (A := OscillatorOperator I)
  exact CFC.rpow (realizeOscillatorCoeff A) p

/-- PDF pp.3--4, (24),(27): positive spectral trace, retaining infinity rather
than a default-zero nonsummable real trace. On a density and p>1 it is finite. -/
def tracePower {I : Type} (p : ℝ) (A : OscillatorCoeff I) : ℝ≥0∞ :=
  ∑' u, ENNReal.ofReal
    (inner ℂ (oscillatorNumberBasis I u) (positivePower A p (oscillatorNumberBasis I u))).re

/-- PDF p.3, (24): Schatten p-norm of an actual density, extended off its domain. -/
def pNorm {I : Type} (p : ℝ) (A : OscillatorCoeff I) : ℝ≥0∞ :=
  (tracePower p A).rpow (1 / p)

/-- PDF p.3, (18): fixed covariance action; the transpose is on the left. -/
def gaussianCovarianceMap {I : Type} [Fintype I] (X Y G : OscillatorCovariance I) : OscillatorCovariance I :=
  X.transpose * G * X + Y

/-- PDF p.3, (16): full Hermitian CP condition, including singular noise. -/
def GaussianCPCondition {I : Type} [Fintype I] (X Y : OscillatorCovariance I) : Prop :=
  Y.transpose = Y ∧
    (Y.map (fun x : ℝ => (x : ℂ)) + Complex.I •
      (interleavedSigma I - X.transpose * interleavedSigma I * X).map (fun x : ℝ => (x : ℂ))).PosSemidef

/-- PDF p.3: the Weyl-coordinate pullback compatible with (18). -/
def gaussianWeylArgument {I : Type} [Fintype I] (X : OscillatorCovariance I) (xi : OscillatorPhase I → ℝ) : OscillatorPhase I → ℝ :=
  ((interleavedSigma I).transpose * X * interleavedSigma I).mulVec xi

/-- PDF p.3: source quadratic noise in the same Weyl coordinates, with /4. -/
def noiseMultiplier {I : Type} [Fintype I] (Y : OscillatorCovariance I) (xi : OscillatorPhase I → ℝ) : ℂ :=
  gaussianStateCharacteristic 0 (groupedCovariance Y) (groupedVector xi)

/-- PDF p.3, (15)--(19), interpreted consistently with (18): an actual CPTP
map on ALL trace class. Its defining Weyl law holds for every density, not just
Gaussian inputs. No output-purity or multiplicativity assertion is a field. -/
structure CenteredOscillatorChannel (I : Type) [Fintype I] where
  X : OscillatorCovariance I
  Y : OscillatorCovariance I
  admissible : GaussianCPCondition X Y
  map : OscillatorCoeff I → OscillatorCoeff I
  cptp : IsTraceClassChannel map
  output : State (Occupation I) → State (Occupation I)
  output_coefficients : ∀ rho, (output rho).coefficients = map rho.coefficients
  characteristic_law : ∀ rho xi,
    WeylTraceSummable (output rho).coefficients xi ∧
    oscillatorCharacteristic (output rho).coefficients xi =
      oscillatorCharacteristic rho.coefficients (gaussianWeylArgument X xi) * noiseMultiplier Y xi

/-- PDF p.4, (31): supremum on actual centered Gaussian input densities. ENNReal
suprema avoid empty/unbounded real defaults; the physical set is nonempty and <=1. -/
def maximalOutputPurity {I : Type} [Fintype I] (p : ℝ) (E : OscillatorCoeff I → OscillatorCoeff I) : ℝ≥0∞ :=
  ⨆ rho : {rho : State (Occupation I) // IsCenteredGaussianDensity rho}, pNorm p (E rho.val.coefficients)

/-- PDF p.4, (34): the same supremum restricted to pure Gaussian densities. -/
def pureMaximalOutputPurity {I : Type} [Fintype I] (p : ℝ) (E : OscillatorCoeff I → OscillatorCoeff I) : ℝ≥0∞ :=
  ⨆ rho : {rho : State (Occupation I) // IsCenteredGaussianDensity rho ∧ IsPureOscillatorDensity rho},
    pNorm p (E rho.val.coefficients)

/-- PDF pp.3--4: replace one whole channel's occupation string, keeping all other modes. -/
def replaceOccupation {r : ℕ} {m : Fin r → ℕ} (i : Fin r)
    (u : Occupation (Σ j, Fin (m j))) (a : Occupation (Fin (m i))) :
    Occupation (Σ j, Fin (m j)) :=
  fun j => if h : j.1 = i then a (h ▸ j.2) else u j

/-- PDF pp.3--4, (36): apply a local map to every off-diagonal block.
The other factors form an arbitrary completed oscillator ancilla. -/
def applyOscillatorBlock {r : ℕ} {m : Fin r → ℕ} (i : Fin r)
    (E : OscillatorCoeff (Fin (m i)) → OscillatorCoeff (Fin (m i)))
    (A : OscillatorCoeff (Σ j, Fin (m j))) : OscillatorCoeff (Σ j, Fin (m j)) :=
  fun u v => E (fun a b => A (replaceOccupation i u a) (replaceOccupation i v b))
    (fun k => u ⟨i, k⟩) (fun k => v ⟨i, k⟩)

/-- PDF pp.3--4, (36): actual finite tensor product of the full channel maps.
The fold applies each factor once; it is not a rule only on product inputs. -/
def tensorOscillatorChannel {r : ℕ} {m : Fin r → ℕ}
    (E : (i : Fin r) → CenteredOscillatorChannel (Fin (m i)))
    (A : OscillatorCoeff (Σ j, Fin (m j))) : OscillatorCoeff (Σ j, Fin (m j)) :=
  (List.finRange r).foldl (fun B i => applyOscillatorBlock i (E i).map B) A

/-- PDF p.4, (37): global Gaussian inputs may be entangled across channel blocks. -/
def GaussianPurityMultiplicative {r : ℕ} {m : Fin r → ℕ} (p : ℝ)
    (E : (i : Fin r) → CenteredOscillatorChannel (Fin (m i))) : Prop :=
  maximalOutputPurity p (tensorOscillatorChannel E) = ∏ i, maximalOutputPurity p (E i).map

/-- PDF p.5, (40): canonical X computed from its factor, including det X=0. -/
def scalarX (E : CenteredOscillatorChannel (Fin 1)) : OscillatorCovariance (Fin 1) :=
  Real.sqrt |E.X.det| • 1

/-- PDF p.5, (40): canonical Y computed from its factor, including det Y=0. -/
def scalarY (E : CenteredOscillatorChannel (Fin 1)) : OscillatorCovariance (Fin 1) :=
  Real.sqrt E.Y.det • 1

/-- PDF p.5: equal determinant signs, including the common zero sign. -/
def EqualSigns {r : ℕ} (E : Fin r → CenteredOscillatorChannel (Fin 1)) : Prop :=
  ∀ i j, (if (E i).X.det < 0 then (-1 : ℤ) else if (E i).X.det = 0 then 0 else 1) =
    (if (E j).X.det < 0 then (-1 : ℤ) else if (E j).X.det = 0 then 0 else 1)

/-- PDF p.2, (9): source convention S^T interleavedSigma S=interleavedSigma. -/
def PreservesInterleavedSigma {I : Type} [Fintype I] (S : OscillatorCovariance I) : Prop :=
  S.transpose * interleavedSigma I * S = interleavedSigma I

/-- PDF p.2, (12): Williamson spectrum with multiplicity, not singular values
of the generally nonnormal i interleavedSigma G. Ordering is immaterial to products. -/
def WilliamsonSpectrum {I : Type} [Fintype I] (G : OscillatorCovariance I) (nu : I → ℝ) : Prop :=
  (∀ i, 1 ≤ nu i) ∧ ∃ S : OscillatorCovariance I, PreservesInterleavedSigma S ∧
    S.transpose * G * S = Matrix.diagonal (fun i => nu i.1)

/-- PDF p.4, (28): reuse the exact accepted f_p. -/
def fp (p x : ℝ) : ℝ := phiMinus p x

/-- PDF p.4, (29): F_p from an explicitly identified symplectic spectrum. -/
def spectralFp {I : Type} [Fintype I] (p : ℝ) (nu : I → ℝ) : ℝ := ∏ i, fp p (nu i)

/-- PDF p.4, (30): intrinsic Ky Fan sum over all r-dimensional orthonormal
frames. For nonempty mode sets this equals the sum of the r largest density
eigenvalues with zero padding. ENNReal prevents an unbounded-real default. -/
def kyFan {I : Type} (A : OscillatorCoeff I) (r : ℕ) : ℝ≥0∞ :=
  ⨆ v : {v : Fin r → Hilbert (Occupation I) // Orthonormal ℂ v},
    ∑ j, ENNReal.ofReal (inner ℂ (v.val j) (realizeOscillatorCoeff A (v.val j))).re

/-- PDF p.4, (30): spectral majorization of actual positive trace-one densities. -/
def OscillatorMajorizes {I : Type} (A B : OscillatorCoeff I) : Prop := ∀ r : ℕ, kyFan B r ≤ kyFan A r

/-- PDF pp.3--5: oscillatorCoefficients of an actual tensor product of factor states. -/
def productInput {r : ℕ} {m : Fin r → ℕ}
    (rho : (i : Fin r) → State (Occupation (Fin (m i)))) : OscillatorCoeff (Σ j, Fin (m j)) :=
  fun u v => ∏ i, (rho i).coefficients (fun k => u ⟨i, k⟩) (fun k => v ⟨i, k⟩)

/-- PDF p.8, (A1): the one-mode sourceBeta matrix. -/
def sourceBeta : Matrix (Fin 2) (Fin 2) ℝ := !![1, 0; 0, -1]

/-- PDF p.8, (A1): the one-mode sourceDelta matrix. -/
def sourceDelta : Matrix (Fin 2) (Fin 2) ℝ := !![0, 1; 1, 0]

/-- PDF p.8, (A2): sigma2, sourceBeta and sourceDelta. Ordering does not affect a basis. -/
def oneModeGenerator (a : Fin 3) : Matrix (Fin 2) (Fin 2) ℝ :=
  if a = 0 then !![0, 1; -1, 0] else if a = 1 then sourceBeta else sourceDelta

/-- PDF p.8, (A2)--(A5): three single-mode and four unordered-pair generators. -/
abbrev GeneratorIndex (n : ℕ) :=
  (Fin n × Fin 3) ⊕ ({ij : Fin n × Fin n // ij.1 < ij.2} × Fin 4)

/-- PDF p.8, (A2)--(A4): the actual embedded one/two-mode matrices. Pair labels
0,1 are the two skew generators in (A3); 2,3 are the symmetric ones in (A4). -/
def sourceSymplecticGenerator {n : ℕ} (a : GeneratorIndex n) : OscillatorCovariance (Fin n) :=
  match a with
  | Sum.inl a => fun i j =>
      if i.1 = a.1 ∧ j.1 = a.1 then oneModeGenerator a.2 i.2 j.2 else 0
  | Sum.inr a => fun i j =>
      if i.1 = a.1.val.1 ∧ j.1 = a.1.val.2 then
        if a.2 = 0 then !![0, 1; -1, 0] i.2 j.2
        else if a.2 = 1 then -(1 : Matrix (Fin 2) (Fin 2) ℝ) i.2 j.2
        else if a.2 = 2 then sourceDelta i.2 j.2 else sourceBeta i.2 j.2
      else if i.1 = a.1.val.2 ∧ j.1 = a.1.val.1 then
        if a.2 = 0 then !![0, 1; -1, 0] i.2 j.2
        else if a.2 = 1 then (1 : Matrix (Fin 2) (Fin 2) ℝ) i.2 j.2
        else if a.2 = 2 then sourceDelta i.2 j.2 else sourceBeta i.2 j.2
      else 0

/-- PDF p.8: infinitesimal symplecticity, not a subgroup of symmetric matrices. -/
def InSymplecticAlgebra {n : ℕ} (K : OscillatorCovariance (Fin n)) : Prop :=
  K.transpose * interleavedSigma (Fin n) + interleavedSigma (Fin n) * K = 0

/-- PDF p.8: skew infinitesimal generators of the compact passive subgroup. -/
abbrev CompactGeneratorIndex (n : ℕ) :=
  {a : GeneratorIndex n // (sourceSymplecticGenerator a).transpose = -sourceSymplecticGenerator a}

example : interleavedSigma (Fin 1) (0, 0) (0, 1) = 1 := by simp [interleavedSigma]

example : interleavedSigma (Fin 1) (0, 1) (0, 0) = -1 := by simp [interleavedSigma]

example (X G : OscillatorCovariance (Fin 1)) : gaussianCovarianceMap X 0 G = X.transpose * G * X := by
  simp [gaussianCovarianceMap]

example {I : Type} (T : OscillatorOperator I) : oscillatorCoefficients (realizeOscillatorCoeff (oscillatorCoefficients T)) = oscillatorCoefficients T := by
  unfold realizeOscillatorCoeff
  split_ifs with h
  · exact Classical.choose_spec h
  · exact (h ⟨T, rfl⟩).elim

example : fp 2 1 = 4 := by
  norm_num [fp, phiMinus, Real.rpow_natCast]

example : sourceSymplecticGenerator (Sum.inl ((0 : Fin 1), (0 : Fin 3))) (0, 0) (0, 1) = 1 := by
  simp [sourceSymplecticGenerator, oneModeGenerator]

example : sourceSymplecticGenerator (Sum.inl ((0 : Fin 1), (1 : Fin 3))) (0, 1) (0, 1) = -1 := by
  norm_num [sourceSymplecticGenerator, oneModeGenerator, sourceBeta]

example (m : Fin 0 → ℕ) (E : (i : Fin 0) → CenteredOscillatorChannel (Fin (m i)))
    (A : OscillatorCoeff (Σ j, Fin (m j))) : tensorOscillatorChannel E A = A := by
  simp [tensorOscillatorChannel]

example : sourceSymplecticGenerator (Sum.inr (⟨((0 : Fin 2), (1 : Fin 2)), by decide⟩, (1 : Fin 4)))
    (0, 0) (1, 0) = -1 := by
  norm_num [sourceSymplecticGenerator]

example : sourceSymplecticGenerator (Sum.inr (⟨((0 : Fin 2), (1 : Fin 2)), by decide⟩, (1 : Fin 4)))
    (1, 0) (0, 0) = 1 := by
  norm_num [sourceSymplecticGenerator]

end QAlgorithms.Papers.Independent.Support50.P28
Source
Multiplicativity of maximal output purities of Gaussian channels under Gaussian inputs, arXiv:quant-ph/0406065v1, https://arxiv.org/abs/quant-ph/0406065v1
Read-back

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

QAlgorithms.Papers.Independent.Support50.P28.CenteredOscillatorChannel

For any finite type III, which may be empty, put L=NIL=\mathbb N^IL=NI, Q=I×{0,1}Q=I\times\{0,1\}Q=I×{0,1}, and H=ℓ2(L;C)\mathcal H=\ell^2(L;\mathbb C)H=ℓ2(L;C). A centered oscillator channel is the following collection of data and conditions. It has real matrices X,YX,YX,Y indexed by QQQ, with

YT=Y,Y+i(J−XTJX)⪰0,Y^{\mathsf T}=Y,\qquad Y+\mathrm i(J-X^{\mathsf T}JX)\succeq0,YT=Y,Y+i(J−XTJX)⪰0,

where J(i,0),(i,1)=1J_{(i,0),(i,1)}=1J(i,0),(i,1)​=1, J(i,1),(i,0)=−1J_{(i,1),(i,0)}=-1J(i,1),(i,0)​=−1, and all other entries are zero. It has a function Φ:CL×L→CL×L\Phi:\mathbb C^{L\times L}\to\mathbb C^{L\times L}Φ:CL×L→CL×L satisfying these requirements: it preserves matrices admitting a representation Ruv=∑a=0∞fa(u)ga(v)‾R_{uv}=\sum_{a=0}^{\infty}f_a(u)\overline{g_a(v)}Ruv​=∑a=0∞​fa​(u)ga​(v)​ with fa,ga∈Hf_a,g_a\in\mathcal Hfa​,ga​∈H and ∑a∥fa∥∥ga∥<∞\sum_a\|f_a\|\|g_a\|<\infty∑a​∥fa​∥∥ga​∥<∞; it is additive and complex homogeneous on such matrices; and, for every natural number ddd and every density matrix ZZZ indexed by L×{0,…,d−1}L\times\{0,\ldots,d-1\}L×{0,…,d−1}, the matrix

Φ~(Z)(u,a),(v,b)=Φ((Z(x,a),(y,b))x,y∈L)uv\widetilde\Phi(Z)_{(u,a),(v,b)} =\Phi\bigl((Z_{(x,a),(y,b)})_{x,y\in L}\bigr)_{uv}Φ(Z)(u,a),(v,b)​=Φ((Z(x,a),(y,b)​)x,y∈L​)uv​

is a density matrix. Here, on any label set, a density matrix means a matrix admitting the preceding summable representation, whose quadratic form on every finite set of labels is real and nonnegative, and whose real diagonal entries are summable with sum 111; in particular, no density matrix exists on an empty label set, so the d=0d=0d=0 requirement has no instances. The channel also has a function OOO from density states on LLL to density states on LLL, agreeing with Φ\PhiΦ on their coefficient matrices. Finally, for every input density state with coefficient matrix RRR and every ξ∈RQ\xi\in\mathbb R^Qξ∈RQ, writing R′R'R′ for the coefficients of O(R)O(R)O(R), it requires

χR′(ξ)=χR(JTXJξ)exp⁡ ⁣(−14 ξTJTYJξ).\chi_{R'}(\xi) =\chi_R(J^{\mathsf T}XJ\xi) \exp\!\left(-\frac14\,\xi^{\mathsf T}J^{\mathsf T}YJ\xi\right).χR′​(ξ)=χR​(JTXJξ)exp(−41​ξTJTYJξ).

The characteristic function here is the particular iterated sum

χR(ξ)=∑u∈L∑v∈LRuv∏i∈IDvi,ui ⁣(−ξi,0+iξi,12),\chi_R(\xi)= \sum_{u\in L}\sum_{v\in L}R_{uv} \prod_{i\in I}D_{v_i,u_i}\!\left( -\frac{\xi_{i,0}+\mathrm i\xi_{i,1}}{\sqrt2}\right),χR​(ξ)=u∈L∑​v∈L∑​Ruv​i∈I∏​Dvi​,ui​​(−2​ξi,0​+iξi,1​​),

where

Dmn(z)=e−∣z∣2/2m! n!∑a=0min⁡(m,n)zm−a(−z‾)n−aa!(m−a)!(n−a)!.D_{mn}(z)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{a=0}^{\min(m,n)} \frac{z^{m-a}(-\overline z)^{n-a}} {a!(m-a)!(n-a)!}.Dmn​(z)=e−∣z∣2/2m!n!​a=0∑min(m,n)​a!(m−a)!(n−a)!zm−a(−z)n−a​.

For R′R'R′, each inner sum over vvv must be summable, and the resulting outer sum over uuu must be summable. This summability condition is imposed on the output in the characteristic law; the input characteristic function is defined by the displayed total sums. When III is empty, LLL is a singleton, the quadrature index set is empty, and the products over III are 111.

QAlgorithms.Papers.Independent.Support50.P28.CompactGeneratorIndex

For every natural number nnn, define the compact generator index type as the subtype of generator indices whose associated real matrix is skew-symmetric, meaning KaT=−KaK_a^{\mathsf T}=-K_aKaT​=−Ka​. Explicitly, it retains the single-mode indices (k,0)(k,0)(k,0), 0≤k<n0\le k<n0≤k<n, whose matrix has the block (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​) at (k,k)(k,k)(k,k), and the pair indices (k,ℓ,t)(k,\ell,t)(k,ℓ,t), 0≤k<ℓ<n0\le k<\ell<n0≤k<ℓ<n, with t=0t=0t=0 or t=1t=1t=1: label 000 places that same block at both (k,ℓ)(k,\ell)(k,ℓ) and (ℓ,k)(\ell,k)(ℓ,k), while label 111 places −I2-I_2−I2​ at (k,ℓ)(k,\ell)(k,ℓ) and I2I_2I2​ at (ℓ,k)(\ell,k)(ℓ,k); all unlisted blocks are zero. The subtype includes the proof of the skew-symmetry condition and is empty when n=0n=0n=0.

QAlgorithms.Papers.Independent.Support50.P28.EqualSigns

For any natural number rrr and any family EiE_iEi​, indexed by i<ri<ri<r, of centered oscillator channels on one mode, this predicate requires

sgn⁡(det⁡Xi)=sgn⁡(det⁡Xj)for every i,j<r,\operatorname{sgn}(\det X_i)=\operatorname{sgn}(\det X_j) \qquad\text{for every }i,j<r,sgn(detXi​)=sgn(detXj​)for every i,j<r,

where XiX_iXi​ is the real 2×22\times22×2 matrix XXX supplied by EiE_iEi​, and the integer-valued sign is −1-1−1 for a negative determinant, 000 for determinant zero, and 111 for a positive determinant. Thus zero determinants are included. The condition is vacuous for r=0r=0r=0 and automatically holds for r=1r=1r=1.

QAlgorithms.Papers.Independent.Support50.P28.GaussianCPCondition

For any finite type III and real matrices X,YX,YX,Y indexed by I×{0,1}I\times\{0,1\}I×{0,1}, this predicate requires YT=YY^{\mathsf T}=YYT=Y and

YC+i(Σ−XTΣX)C⪰0,Y_{\mathbb C}+i\bigl(\Sigma-X^{\mathsf T}\Sigma X\bigr)_{\mathbb C}\succeq0,YC​+i(Σ−XTΣX)C​⪰0,

where complexification embeds each real entry into C\mathbb CC, and Σ\SigmaΣ has entries Σ(a,0),(a,1)=1\Sigma_{(a,0),(a,1)}=1Σ(a,0),(a,1)​=1, Σ(a,1),(a,0)=−1\Sigma_{(a,1),(a,0)}=-1Σ(a,1),(a,0)​=−1, with all other entries zero. Positive semidefiniteness means that the displayed complex matrix is Hermitian and every complex-vector quadratic form is nonnegative. There is no invertibility requirement on XXX. Empty III is allowed, and then the condition holds for the empty matrices.

QAlgorithms.Papers.Independent.Support50.P28.GaussianPurityMultiplicative

For every natural number rrr, every family mi∈Nm_i\in\mathbb Nmi​∈N indexed by i<ri<ri<r, every real ppp, and every family of centered oscillator channels EiE_iEi​ on mim_imi​ modes, this predicate means exactly

sup⁡R∈GLNp(Φseq(R))=∏i=0r−1sup⁡Ri∈GmiNp(Φi(Ri))in [0,∞],\sup_{R\in\mathcal G_L}N_p(\Phi_{\mathrm{seq}}(R)) = \prod_{i=0}^{r-1} \sup_{R_i\in\mathcal G_{m_i}}N_p(\Phi_i(R_i)) \quad\text{in }[0,\infty],R∈GL​sup​Np​(Φseq​(R))=i=0∏r−1​Ri​∈Gmi​​sup​Np​(Φi​(Ri​))in [0,∞],

where Φi\Phi_iΦi​ is the coefficient map supplied by EiE_iEi​, L={(i,k):i<r, k<mi}L=\{(i,k):i<r,\ k<m_i\}L={(i,k):i<r, k<mi​}, and Φseq\Phi_{\mathrm{seq}}Φseq​ applies the maps in increasing block order: starting from B=AB=AB=A, its iii-th update replaces BuvB_{uv}Buv​ by

Φi ⁣((Bu[i←a], v[i←b])a,b)ui,vi.\Phi_i\!\left( \bigl(B_{u[i\leftarrow a],\,v[i\leftarrow b]}\bigr)_{a,b} \right)_{u_i,v_i}.Φi​((Bu[i←a],v[i←b]​)a,b​)ui​,vi​​.

Replacing block iii means replacing its occupation coordinates by aaa or bbb and retaining all other coordinates. For any finite mode set KKK, GK\mathcal G_KGK​ consists of coefficient matrices RRR on NK\mathbb N^KNK admitting a summable nuclear representation Ruv=∑afa(u)ga(v)‾R_{uv}=\sum_a f_a(u)\overline{g_a(v)}Ruv​=∑a​fa​(u)ga​(v)​, with ∑a∥fa∥∥ga∥<∞\sum_a\|f_a\|\|g_a\|<\infty∑a​∥fa​∥∥ga​∥<∞, real nonnegative finite quadratic forms, summable real diagonal of sum 111, and a symmetric real GGG with G+iJ⪰0G+\mathrm iJ\succeq0G+iJ⪰0 such that

χR(ξ)=e−ξTJTGJξ/4.\chi_R(\xi)=e^{-\xi^{\mathsf T}J^{\mathsf T}GJ\xi/4}.χR​(ξ)=e−ξTJTGJξ/4.

Here JJJ has blocks (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​), and χR\chi_RχR​ is the summable iterated series

χR(ξ)=∑u∑vRuv∏k∈KDvk,uk ⁣(−ξk,0+iξk,12),Dmn(z)=e−∣z∣2/2m! n!∑a=0min⁡(m,n)zm−a(−z‾)n−aa!(m−a)!(n−a)!,\chi_R(\xi)=\sum_u\sum_v R_{uv} \prod_{k\in K}D_{v_k,u_k}\!\left( -\frac{\xi_{k,0}+\mathrm i\xi_{k,1}}{\sqrt2}\right), \quad D_{mn}(z)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{a=0}^{\min(m,n)} \frac{z^{m-a}(-\overline z)^{n-a}} {a!(m-a)!(n-a)!},χR​(ξ)=u∑​v∑​Ruv​k∈K∏​Dvk​,uk​​(−2​ξk,0​+iξk,1​​),Dmn​(z)=e−∣z∣2/2m!n!​a=0∑min(m,n)​a!(m−a)!(n−a)!zm−a(−z)n−a​,

with every inner sum and the outer sum summable for every ξ\xiξ. The quantity being optimized is

Np(B)=(∑u[Re⁡⟨eu,TB peu⟩]+)1/p,N_p(B)= \left(\sum_u [\operatorname{Re}\langle e_u,T_B^{\,p}e_u\rangle]_+ \right)^{1/p},Np​(B)=(u∑​[Re⟨eu​,TBp​eu​⟩]+​)1/p,

where TBT_BTB​ is a selected bounded operator with coefficient matrix BBB, or zero if none exists, and its power is defined by real continuous functional calculus. The superscript 1/p1/p1/p denotes extended nonnegative real exponentiation, and [x]+=max⁡(x,0)[x]_+=\max(x,0)[x]+​=max(x,0). There is no restriction on ppp, so zero and negative values are included; at p=0p=0p=0, 1/p=01/p=01/p=0 under total division. The supremum on the left includes centered Gaussian inputs across all blocks without a product-input restriction. Zero-sized blocks are allowed. For r=0r=0r=0, the sequential map is the identity and the product on the right is 111.

QAlgorithms.Papers.Independent.Support50.P28.GeneratorIndex

For every natural number nnn, define the generator index type to be the disjoint union

({0,…,n−1}×{0,1,2})  ⊔  ({(k,ℓ):0≤k<ℓ<n}×{0,1,2,3}).\bigl(\{0,\ldots,n-1\}\times\{0,1,2\}\bigr) \;\sqcup\; \bigl(\{(k,\ell):0\le k<\ell<n\}\times\{0,1,2,3\}\bigr).({0,…,n−1}×{0,1,2})⊔({(k,ℓ):0≤k<ℓ<n}×{0,1,2,3}).

Thus an index carries either a single mode and one of three labels, or a strictly ordered pair of modes and one of four labels, with the two kinds distinguished by their disjoint-union tags. For n=0n=0n=0, this type is empty.

QAlgorithms.Papers.Independent.Support50.P28.HasCenteredGaussianCovariance

For any finite type III, write N=I→N\mathcal N=I\to\mathbb NN=I→N, and let ρ\rhoρ be a state with coefficient matrix RRR on N\mathcal NN: explicitly, there are sequences aℓ,bℓ∈ℓ2(N;C)a_\ell,b_\ell\in\ell^2(\mathcal N;\mathbb C)aℓ​,bℓ​∈ℓ2(N;C) with ∑ℓ≥0∥aℓ∥∥bℓ∥<∞\sum_{\ell\ge0}\|a_\ell\|\|b_\ell\|<\infty∑ℓ≥0​∥aℓ​∥∥bℓ​∥<∞ and Ru,v=∑ℓ≥0aℓ(u)bℓ(v)‾R_{u,v}=\sum_{\ell\ge0}a_\ell(u)\overline{b_\ell(v)}Ru,v​=∑ℓ≥0​aℓ​(u)bℓ​(v)​; for every finite s⊆Ns\subseteq\mathcal Ns⊆N and every c:N→Cc:\mathcal N\to\mathbb Cc:N→C, the sum ∑u,v∈scu‾Ru,vcv\sum_{u,v\in s}\overline{c_u}R_{u,v}c_v∑u,v∈s​cu​​Ru,v​cv​ is real and nonnegative; and the family (Re⁡Ru,u)u∈N(\operatorname{Re}R_{u,u})_{u\in\mathcal N}(ReRu,u​)u∈N​ is summable with sum 111. For a real matrix GGG on I×{0,1}I\times\{0,1\}I×{0,1}, the defined property requires GT=GG^{\mathsf T}=GGT=G and G+iΣIG+\mathrm i\Sigma_IG+iΣI​ positive semidefinite, where ΣI\Sigma_IΣI​ has blocks (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​) for each label and zero entries between different labels; it further requires, for every ξ:I×{0,1}→R\xi:I\times\{0,1\}\to\mathbb Rξ:I×{0,1}→R,

∑u∈N∑v∈NRu,v∏i∈IDzi(vi,ui)=exp⁡ ⁣(−14∑r,s∈I×{0,1}ξr(ΣITGΣI)r,sξs),zi=−ξ(i,0)+i ξ(i,1)2,\sum_{u\in\mathcal N}\sum_{v\in\mathcal N} R_{u,v}\prod_{i\in I}D_{z_i}(v_i,u_i) = \exp\!\left(-\frac14 \sum_{r,s\in I\times\{0,1\}} \xi_r(\Sigma_I^{\mathsf T}G\Sigma_I)_{r,s}\xi_s\right), \qquad z_i=-\frac{\xi(i,0)+\mathrm i\,\xi(i,1)}{\sqrt2},u∈N∑​v∈N∑​Ru,v​i∈I∏​Dzi​​(vi​,ui​)=exp​−41​r,s∈I×{0,1}∑​ξr​(ΣIT​GΣI​)r,s​ξs​​,zi​=−2​ξ(i,0)+iξ(i,1)​,

together with summability of the inner complex family over vvv for each uuu, and summability over uuu of those inner sums. Here

Dz(m,n)=e−∣z∣2/2m! n!∑j=0min⁡(m,n)zm−j(−z‾)n−jj! (m−j)! (n−j)!.D_z(m,n)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{j=0}^{\min(m,n)} \frac{z^{m-j}(-\overline z)^{n-j}}{j!\,(m-j)!\,(n-j)!}.Dz​(m,n)=e−∣z∣2/2m!n!​j=0∑min(m,n)​j!(m−j)!(n−j)!zm−j(−z)n−j​.

Empty III is allowed: N\mathcal NN is then a singleton, all phase sums are empty, and the displayed equality requires the sole coefficient of RRR to equal 111.

QAlgorithms.Papers.Independent.Support50.P28.InSymplecticAlgebra

For any natural number nnn and any real matrix KKK indexed by {0,…,n−1}×{0,1}\{0,\ldots,n-1\}\times\{0,1\}{0,…,n−1}×{0,1}, define membership in the symplectic algebra by the equation

KTΩ+ΩK=0,Ω(k,b),(ℓ,c)={1,k=ℓ, b=0, c=1,−1,k=ℓ, b=1, c=0,0,otherwise.K^{\mathsf T}\Omega+\Omega K=0, \qquad \Omega_{(k,b),(\ell,c)} = \begin{cases} 1,&k=\ell,\ b=0,\ c=1,\\ -1,&k=\ell,\ b=1,\ c=0,\\ 0,&\text{otherwise}. \end{cases}KTΩ+ΩK=0,Ω(k,b),(ℓ,c)​=⎩⎨⎧​1,−1,0,​k=ℓ, b=0, c=1,k=ℓ, b=1, c=0,otherwise.​

No further condition on KKK is included. For n=0n=0n=0, the equation holds for the unique matrix with empty index set.

QAlgorithms.Papers.Independent.Support50.P28.IsCenteredGaussianDensity

For any finite type III, write N=I→N\mathcal N=I\to\mathbb NN=I→N, and let ρ\rhoρ be a state with coefficient matrix RRR on N\mathcal NN: this means that Ru,v=∑ℓ≥0aℓ(u)bℓ(v)‾R_{u,v}=\sum_{\ell\ge0}a_\ell(u)\overline{b_\ell(v)}Ru,v​=∑ℓ≥0​aℓ​(u)bℓ​(v)​ for sequences aℓ,bℓ∈ℓ2(N;C)a_\ell,b_\ell\in\ell^2(\mathcal N;\mathbb C)aℓ​,bℓ​∈ℓ2(N;C) satisfying ∑ℓ≥0∥aℓ∥∥bℓ∥<∞\sum_{\ell\ge0}\|a_\ell\|\|b_\ell\|<\infty∑ℓ≥0​∥aℓ​∥∥bℓ​∥<∞, that every finite quadratic form ∑u,v∈scu‾Ru,vcv\sum_{u,v\in s}\overline{c_u}R_{u,v}c_v∑u,v∈s​cu​​Ru,v​cv​ is real and nonnegative for every finite s⊆Ns\subseteq\mathcal Ns⊆N and every c:N→Cc:\mathcal N\to\mathbb Cc:N→C, and that the real diagonal entries are summable with sum 111. The defined property says that there exists a real matrix GGG on I×{0,1}I\times\{0,1\}I×{0,1} such that GT=GG^{\mathsf T}=GGT=G, G+iΣIG+\mathrm i\Sigma_IG+iΣI​ is positive semidefinite, and, for every real function ξ\xiξ on I×{0,1}I\times\{0,1\}I×{0,1},

∑u∈N∑v∈NRu,v∏i∈IDzi(vi,ui)=exp⁡ ⁣(−14∑r,s∈I×{0,1}ξr(ΣITGΣI)r,sξs).\sum_{u\in\mathcal N}\sum_{v\in\mathcal N} R_{u,v}\prod_{i\in I}D_{z_i}(v_i,u_i) = \exp\!\left(-\frac14 \sum_{r,s\in I\times\{0,1\}} \xi_r(\Sigma_I^{\mathsf T}G\Sigma_I)_{r,s}\xi_s\right).u∈N∑​v∈N∑​Ru,v​i∈I∏​Dzi​​(vi​,ui​)=exp​−41​r,s∈I×{0,1}∑​ξr​(ΣIT​GΣI​)r,s​ξs​​.

Here ΣI\Sigma_IΣI​ has block (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​) at each label and zero entries between labels, zi=−(ξ(i,0)+i ξ(i,1))/2z_i=-(\xi(i,0)+\mathrm i\,\xi(i,1))/\sqrt2zi​=−(ξ(i,0)+iξ(i,1))/2​, and

Dz(m,n)=e−∣z∣2/2m! n!∑j=0min⁡(m,n)zm−j(−z‾)n−jj! (m−j)! (n−j)!.D_z(m,n)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{j=0}^{\min(m,n)} \frac{z^{m-j}(-\overline z)^{n-j}}{j!\,(m-j)!\,(n-j)!}.Dz​(m,n)=e−∣z∣2/2m!n!​j=0∑min(m,n)​j!(m−j)!(n−j)!zm−j(−z)n−j​.

For each ξ\xiξ, the inner complex family over vvv must be summable for every uuu, and the resulting inner sums must be summable over uuu. The existence of GGG is asserted without uniqueness. Empty III is included, with empty phase matrices and a singleton occupation index set.

QAlgorithms.Papers.Independent.Support50.P28.IsPureOscillatorDensity

For any type III, put Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N} and H=ℓ2(Ω;C)H=\ell^2(\Omega;\mathbb C)H=ℓ2(Ω;C). Let ρ\rhoρ be a state: its coefficient matrix RRR admits a representation Rab=∑r=0∞xr(a)yr(b)‾R_{ab}=\sum_{r=0}^\infty x_r(a)\overline{y_r(b)}Rab​=∑r=0∞​xr​(a)yr​(b)​, where xr,yr∈Hx_r,y_r\in Hxr​,yr​∈H and ∑r∥xr∥∥yr∥<∞\sum_r\|x_r\|\|y_r\|<\infty∑r​∥xr​∥∥yr​∥<∞; every finite quadratic form ∑a,b∈sza‾Rabzb\sum_{a,b\in s}\overline{z_a}R_{ab}z_b∑a,b∈s​za​​Rab​zb​ is real and nonnegative; and the real diagonal entries are summable with sum 111. The predicate defined here holds precisely when there exists u∈Hu\in Hu∈H with ∥u∥=1\|u\|=1∥u∥=1 such that Rab=u(a)u(b)‾R_{ab}=u(a)\overline{u(b)}Rab​=u(a)u(b)​ for every a,b∈Ωa,b\in\Omegaa,b∈Ω. No finiteness or countability assumption is imposed on III, and III may be empty.

QAlgorithms.Papers.Independent.Support50.P28.OscillatorCoeff

For any type III, an oscillator coefficient matrix is an arbitrary complex matrix whose row and column labels are functions I→NI\to\mathbb NI→N. No boundedness, summability, positivity, or normalization condition is imposed. If III is empty, there is exactly one function I→NI\to\mathbb NI→N, so these are 1×11\times11×1 matrices.

QAlgorithms.Papers.Independent.Support50.P28.OscillatorCovariance

For any type III, an oscillator covariance is a real matrix indexed in both directions by I×{0,1}I\times\{0,1\}I×{0,1}, equivalently an arbitrary function (I×{0,1})×(I×{0,1})→R(I\times\{0,1\})\times(I\times\{0,1\})\to\mathbb R(I×{0,1})×(I×{0,1})→R. This abbreviation imposes no symmetry or positivity condition and requires neither finiteness nor nonemptiness of III.

QAlgorithms.Papers.Independent.Support50.P28.OscillatorMajorizes

For any type III and any two complex matrices A,BA,BA,B indexed by O=NI\mathcal O=\mathbb N^IO=NI, say that AAA oscillator-majorizes BBB if, for every natural number rrr,

sup⁡v0,…,vr−1∈ℓ2(O;C)⟨vj,vk⟩=δjk∑j=0r−1max⁡{0,Re⁡⟨vj,TBvj⟩}  ≤  sup⁡v0,…,vr−1∈ℓ2(O;C)⟨vj,vk⟩=δjk∑j=0r−1max⁡{0,Re⁡⟨vj,TAvj⟩}.\sup_{\substack{v_0,\ldots,v_{r-1}\in\ell^2(\mathcal O;\mathbb C)\\ \langle v_j,v_k\rangle=\delta_{jk}}} \sum_{j=0}^{r-1}\max\{0,\operatorname{Re}\langle v_j,T_Bv_j\rangle\} \;\le\; \sup_{\substack{v_0,\ldots,v_{r-1}\in\ell^2(\mathcal O;\mathbb C)\\ \langle v_j,v_k\rangle=\delta_{jk}}} \sum_{j=0}^{r-1}\max\{0,\operatorname{Re}\langle v_j,T_Av_j\rangle\}.v0​,…,vr−1​∈ℓ2(O;C)⟨vj​,vk​⟩=δjk​​sup​j=0∑r−1​max{0,Re⟨vj​,TB​vj​⟩}≤v0​,…,vr−1​∈ℓ2(O;C)⟨vj​,vk​⟩=δjk​​sup​j=0∑r−1​max{0,Re⟨vj​,TA​vj​⟩}.

Here, for C=A,BC=A,BC=A,B, TCT_CTC​ is a chosen bounded complex-linear operator whose coordinate matrix is CCC, or the zero operator if none exists. Suprema and sums take values in [0,∞][0,\infty][0,∞], and an empty supremum is 000. The quantifier includes r=0r=0r=0, when both sides are 000; neither matrix is required to be a density matrix.

QAlgorithms.Papers.Independent.Support50.P28.OscillatorOperator

For any type III, an oscillator operator is a bounded complex-linear map from ℓ2(Ω;C)\ell^2(\Omega;\mathbb C)ℓ2(Ω;C) to itself, where Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}. There are no positivity, trace-class, or normalization requirements. The type III may be infinite or empty; when III is empty, Ω\OmegaΩ has one element.

QAlgorithms.Papers.Independent.Support50.P28.OscillatorPhase

For any type III, the oscillator phase index set is I×{0,1}I\times\{0,1\}I×{0,1}. Thus each element consists of a label i∈Ii\in Ii∈I and one of two indices, 000 or 111. No finiteness or nonemptiness assumption is imposed on III; if III is empty, this index set is empty.

QAlgorithms.Papers.Independent.Support50.P28.PhysicalOscillatorCovariance

For any finite type III and any real matrix GGG indexed by I×{0,1}I\times\{0,1\}I×{0,1}, this property means that GT=GG^{\mathsf T}=GGT=G and that the complex matrix G+iΣIG+\mathrm i\Sigma_IG+iΣI​, with GGG viewed entrywise as complex, is positive semidefinite. Here (ΣI)(a,0),(a,1)=1(\Sigma_I)_{(a,0),(a,1)}=1(ΣI​)(a,0),(a,1)​=1, (ΣI)(a,1),(a,0)=−1(\Sigma_I)_{(a,1),(a,0)}=-1(ΣI​)(a,1),(a,0)​=−1, and all other entries are zero. Positive semidefiniteness means that this complex matrix is Hermitian and its quadratic form is nonnegative on every complex vector. No strict positive definiteness is required. Empty III is included, and both conditions then hold for the unique empty matrix.

QAlgorithms.Papers.Independent.Support50.P28.PreservesInterleavedSigma

For any finite type III and any real matrix SSS indexed by I×{0,1}I\times\{0,1\}I×{0,1}, this predicate is the equality

STJS=J,S^{\mathsf T}JS=J,STJS=J,

where J(i,0),(i,1)=1J_{(i,0),(i,1)}=1J(i,0),(i,1)​=1, J(i,1),(i,0)=−1J_{(i,1),(i,0)}=-1J(i,1),(i,0)​=−1, and every other entry is zero. No separate hypothesis on SSS is imposed. Empty III is allowed, and then this is an equality of empty matrices and holds automatically.

QAlgorithms.Papers.Independent.Support50.P28.WeylTraceSummable

For any finite type III, any complex matrix AAA indexed by functions I→NI\to\mathbb NI→N, and any real function ξ\xiξ on I×{0,1}I\times\{0,1\}I×{0,1}, put zi=−(ξ(i,0)+i ξ(i,1))/2z_i=-(\xi(i,0)+\mathrm i\,\xi(i,1))/\sqrt2zi​=−(ξ(i,0)+iξ(i,1))/2​ and

Dz(m,n)=e−∣z∣2/2m! n!∑j=0min⁡(m,n)zm−j(−z‾)n−jj! (m−j)! (n−j)!.D_z(m,n)= e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{j=0}^{\min(m,n)} \frac{z^{m-j}(-\overline z)^{n-j}} {j!\,(m-j)!\,(n-j)!}.Dz​(m,n)=e−∣z∣2/2m!n!​j=0∑min(m,n)​j!(m−j)!(n−j)!zm−j(−z)n−j​.

The defined property says both that, for every occupation function uuu, the complex family

(Au,v∏i∈IDzi(vi,ui))v:I→N\left(A_{u,v}\prod_{i\in I}D_{z_i}(v_i,u_i)\right)_{v:I\to\mathbb N}(Au,v​i∈I∏​Dzi​​(vi​,ui​))v:I→N​

is summable, and that the family of its inner sums, indexed by u:I→Nu:I\to\mathbb Nu:I→N, is summable. These are two successive summability requirements; no joint summability requirement is stated. Empty III is allowed, in which case both families have one term and the property always holds.

QAlgorithms.Papers.Independent.Support50.P28.WilliamsonSpectrum

For any finite type III, any real matrix GGG indexed by I×{0,1}I\times\{0,1\}I×{0,1}, and any function ν:I→R\nu:I\to\mathbb Rν:I→R, this predicate requires νi≥1\nu_i\ge1νi​≥1 for every i∈Ii\in Ii∈I and the existence of a real matrix SSS with the same indices such that

STJS=J,STGS=diag⁡(i,b)∈I×{0,1}(νi).S^{\mathsf T}JS=J,\qquad S^{\mathsf T}GS=\operatorname{diag}_{(i,b)\in I\times\{0,1\}}(\nu_i).STJS=J,STGS=diag(i,b)∈I×{0,1}​(νi​).

Here J(i,0),(i,1)=1J_{(i,0),(i,1)}=1J(i,0),(i,1)​=1, J(i,1),(i,0)=−1J_{(i,1),(i,0)}=-1J(i,1),(i,0)​=−1, and all other entries are zero; each νi\nu_iνi​ therefore appears twice on the displayed diagonal. There is no ordering requirement on ν\nuν and no separately stated symmetry or physical-covariance hypothesis on GGG. If III is empty, the inequalities and matrix equations are vacuous, and the predicate holds.

QAlgorithms.Papers.Independent.Support50.P28.applyOscillatorBlock

For every natural number rrr, family mj∈Nm_j\in\mathbb Nmj​∈N indexed by j<rj<rj<r, index i<ri<ri<r, arbitrary function EEE from complex matrices indexed by Nmi\mathbb N^{m_i}Nmi​ to matrices with the same indices, and complex matrix AAA indexed by occupation assignments on L={(j,k):j<r, k<mj}L=\{(j,k):j<r,\ k<m_j\}L={(j,k):j<r, k<mj​}, this defines a matrix BBB on those full occupation assignments by

Buv=E ⁣((Au[i←a], v[i←b])a,b∈Nmi)ui,vi.B_{uv} = E\!\left( \bigl(A_{u[i\leftarrow a],\,v[i\leftarrow b]}\bigr)_{a,b\in\mathbb N^{m_i}} \right)_{u_i,v_i}.Buv​=E((Au[i←a],v[i←b]​)a,b∈Nmi​​)ui​,vi​​.

Here ui(k)=u(i,k)u_i(k)=u(i,k)ui​(k)=u(i,k), and u[i←a](j,k)u[i\leftarrow a](j,k)u[i←a](j,k) equals a(k)a(k)a(k) when j=ij=ij=i and u(j,k)u(j,k)u(j,k) otherwise; the replacement in vvv is defined identically. No linearity, positivity, channel, or summability hypothesis is imposed on EEE or AAA. A block with mi=0m_i=0mi​=0 still has one occupation assignment, so EEE acts on a 1×11\times11×1 matrix and is not required to act trivially. When r=0r=0r=0, no index iii exists.

QAlgorithms.Papers.Independent.Support50.P28.fp

For arbitrary real numbers ppp and xxx, define

fp(x)=(x+1)p−(x−1)p.f_p(x)=(x+1)^p-(x-1)^p.fp​(x)=(x+1)p−(x−1)p.

Both powers use the total real-power operation, so this definition includes zero or negative bases and imposes no restriction on ppp or xxx.

QAlgorithms.Papers.Independent.Support50.P28.gaussianCovarianceMap

For any finite type III and any three real matrices X,Y,GX,Y,GX,Y,G indexed by I×{0,1}I\times\{0,1\}I×{0,1}, this defines the real matrix

XTGX+Y.X^{\mathsf T}GX+Y.XTGX+Y.

The matrices need not be symmetric, positive, or satisfy any covariance or channel condition. The empty type III is included, in which case the result is the empty matrix.

QAlgorithms.Papers.Independent.Support50.P28.gaussianWeylArgument

For any finite type III, real matrix XXX indexed by I×{0,1}I\times\{0,1\}I×{0,1}, and real vector ξ:I×{0,1}→R\xi:I\times\{0,1\}\to\mathbb Rξ:I×{0,1}→R, this defines the vector

ΣTXΣ ξ,\Sigma^{\mathsf T}X\Sigma\,\xi,ΣTXΣξ,

where Σ(a,0),(a,1)=1\Sigma_{(a,0),(a,1)}=1Σ(a,0),(a,1)​=1, Σ(a,1),(a,0)=−1\Sigma_{(a,1),(a,0)}=-1Σ(a,1),(a,0)​=−1, and every other entry of Σ\SigmaΣ is zero. Matrix-vector multiplication uses finite sums. No symmetry, positivity, or channel condition is imposed on XXX, and empty III is included.

QAlgorithms.Papers.Independent.Support50.P28.groupedCovariance

For any type III and any real matrix GGG indexed by I×{0,1}I\times\{0,1\}I×{0,1}, this defines a matrix on I⊔II\sqcup II⊔I by Ga,bgrp=Gg(a),g(b)G^{\mathrm{grp}}_{a,b}=G_{g(a),g(b)}Ga,bgrp​=Gg(a),g(b)​, where ggg sends the first copy of iii to (i,0)(i,0)(i,0) and the second copy to (i,1)(i,1)(i,1). Equivalently, its four blocks have entries G(i,0),(j,0)G_{(i,0),(j,0)}G(i,0),(j,0)​, G(i,0),(j,1)G_{(i,0),(j,1)}G(i,0),(j,1)​, G(i,1),(j,0)G_{(i,1),(j,0)}G(i,1),(j,0)​, and G(i,1),(j,1)G_{(i,1),(j,1)}G(i,1),(j,1)​, respectively. No finiteness, symmetry, or positivity is assumed.

QAlgorithms.Papers.Independent.Support50.P28.groupedIndex

For any type III, this defines a map g:I⊔I→I×{0,1}g:I\sqcup I\to I\times\{0,1\}g:I⊔I→I×{0,1}, where the domain consists of two distinguished copies of III: an element iii in the first copy maps to (i,0)(i,0)(i,0), and an element iii in the second copy maps to (i,1)(i,1)(i,1). The definition also applies when III is empty.

QAlgorithms.Papers.Independent.Support50.P28.groupedVector

For any type III and any real function ξ:I×{0,1}→R\xi:I\times\{0,1\}\to\mathbb Rξ:I×{0,1}→R, this defines a real function on two distinguished copies of III by assigning ξ(i,0)\xi(i,0)ξ(i,0) to iii in the first copy and ξ(i,1)\xi(i,1)ξ(i,1) to iii in the second copy. No finiteness or nonemptiness assumption is imposed.

QAlgorithms.Papers.Independent.Support50.P28.interleavedSigma

For any type III, this defines the real matrix ΣI\Sigma_IΣI​ on I×{0,1}I\times\{0,1\}I×{0,1} by

(ΣI)(a,r),(b,s)={1,a=b, r=0, s=1,−1,a=b, r=1, s=0,0,otherwise.(\Sigma_I)_{(a,r),(b,s)} = \begin{cases} 1,&a=b,\ r=0,\ s=1,\\ -1,&a=b,\ r=1,\ s=0,\\ 0,&\text{otherwise}. \end{cases}(ΣI​)(a,r),(b,s)​=⎩⎨⎧​1,−1,0,​a=b, r=0, s=1,a=b, r=1, s=0,otherwise.​

Thus different labels have zero entries between them, and each label has the block (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​). No finiteness or nonemptiness assumption is required; for empty III, this is the empty matrix.

QAlgorithms.Papers.Independent.Support50.P28.kyFan

For any type III, let O=NI\mathcal O=\mathbb N^IO=NI be the set of occupation functions and let H=ℓ2(O;C)H=\ell^2(\mathcal O;\mathbb C)H=ℓ2(O;C), with coordinate unit vectors eue_ueu​. Given any complex matrix A=(Auv)u,v∈OA=(A_{uv})_{u,v\in\mathcal O}A=(Auv​)u,v∈O​ and any natural number rrr, let TA:H→HT_A:H\to HTA​:H→H be a chosen bounded complex-linear operator satisfying ⟨eu,TAev⟩=Auv\langle e_u,T_Ae_v\rangle=A_{uv}⟨eu​,TA​ev​⟩=Auv​ for every u,vu,vu,v, if such an operator exists; otherwise let TA=0T_A=0TA​=0. Define

kyFan⁡(A,r)=sup⁡v0,…,vr−1∈H⟨vj,vk⟩=δjk∑j=0r−1max⁡ ⁣{0,Re⁡⟨vj,TAvj⟩},\operatorname{kyFan}(A,r) =\sup_{\substack{v_0,\ldots,v_{r-1}\in H\\ \langle v_j,v_k\rangle=\delta_{jk}}} \sum_{j=0}^{r-1} \max\!\left\{0,\operatorname{Re}\langle v_j,T_Av_j\rangle\right\},kyFan(A,r)=v0​,…,vr−1​∈H⟨vj​,vk​⟩=δjk​​sup​j=0∑r−1​max{0,Re⟨vj​,TA​vj​⟩},

where each nonnegative summand and the supremum are taken in [0,∞][0,\infty][0,∞]. No positivity, self-adjointness, trace condition, or finiteness assumption on III is required. For r=0r=0r=0, the value is 000; if no orthonormal rrr-tuple exists, the supremum is also 000.

QAlgorithms.Papers.Independent.Support50.P28.maximalOutputPurity

For any finite type III, any real ppp, and any function E:CL×L→CL×LE:\mathbb C^{L\times L}\to\mathbb C^{L\times L}E:CL×L→CL×L, where L=NIL=\mathbb N^IL=NI, this defines the extended nonnegative number

sup⁡R(∑u∈L[Re⁡⟨eu,TE(R) peu⟩]+)1/p.\sup_R \left( \sum_{u\in L} \left[\operatorname{Re} \langle e_u,T_{E(R)}^{\,p}e_u\rangle\right]_+ \right)^{1/p}.Rsup​(u∈L∑​[Re⟨eu​,TE(R)p​eu​⟩]+​)1/p.

Here [x]+=max⁡(x,0)[x]_+=\max(x,0)[x]+​=max(x,0), the sum takes values in [0,∞][0,\infty][0,∞], and the outer power is the extended nonnegative real power. For any coefficient matrix BBB, TBT_BTB​ is a selected bounded complex linear operator on ℓ2(L;C)\ell^2(L;\mathbb C)ℓ2(L;C) with matrix coefficients BBB, if such an operator exists, and is the zero operator otherwise; TB pT_B^{\,p}TBp​ is its real continuous functional-calculus power, and eue_ueu​ is the coordinate unit vector. The supremum ranges over precisely the following matrices RRR: they admit a representation Ruv=∑afa(u)ga(v)‾R_{uv}=\sum_a f_a(u)\overline{g_a(v)}Ruv​=∑a​fa​(u)ga​(v)​ with ∑a∥fa∥∥ga∥<∞\sum_a\|f_a\|\|g_a\|<\infty∑a​∥fa​∥∥ga​∥<∞; every finitely supported quadratic form of RRR is real and nonnegative; the real diagonal is summable with sum 111; and there exists a real matrix GGG indexed by Q=I×{0,1}Q=I\times\{0,1\}Q=I×{0,1} such that

GT=G,G+iJ⪰0,χR(ξ)=exp⁡ ⁣(−14ξTJTGJξ)for every ξ∈RQ.G^{\mathsf T}=G,\qquad G+\mathrm iJ\succeq0,\qquad \chi_R(\xi)= \exp\!\left(-\frac14\xi^{\mathsf T}J^{\mathsf T}GJ\xi\right) \quad\text{for every }\xi\in\mathbb R^Q.GT=G,G+iJ⪰0,χR​(ξ)=exp(−41​ξTJTGJξ)for every ξ∈RQ.

The matrix JJJ has blocks (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​), and the characteristic function is

χR(ξ)=∑u∑vRuv∏i∈IDvi,ui ⁣(−ξi,0+iξi,12),Dmn(z)=e−∣z∣2/2m! n!∑a=0min⁡(m,n)zm−a(−z‾)n−aa!(m−a)!(n−a)!;\chi_R(\xi)=\sum_u\sum_v R_{uv} \prod_{i\in I}D_{v_i,u_i}\!\left( -\frac{\xi_{i,0}+\mathrm i\xi_{i,1}}{\sqrt2}\right), \quad D_{mn}(z)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{a=0}^{\min(m,n)} \frac{z^{m-a}(-\overline z)^{n-a}} {a!(m-a)!(n-a)!};χR​(ξ)=u∑​v∑​Ruv​i∈I∏​Dvi​,ui​​(−2​ξi,0​+iξi,1​​),Dmn​(z)=e−∣z∣2/2m!n!​a=0∑min(m,n)​a!(m−a)!(n−a)!zm−a(−z)n−a​;

for every ξ\xiξ, all inner sums and the resulting outer sum must be summable. No channel, linearity, positivity, or bounded-realization condition is assumed for EEE, and no restriction such as p>0p>0p>0 is imposed. In particular, p=0p=0p=0 is included, with 1/p=01/p=01/p=0 under total real division, making the displayed extended power 111 for each admissible input. An empty supremum is 000. The type III may be empty, in which case LLL is a singleton.

QAlgorithms.Papers.Independent.Support50.P28.noiseMultiplier

For any finite type III, real matrix YYY indexed by I×{0,1}I\times\{0,1\}I×{0,1}, and real vector ξ:I×{0,1}→R\xi:I\times\{0,1\}\to\mathbb Rξ:I×{0,1}→R, this defines the complex number

exp⁡ ⁣(−14∑a,b∈I×{0,1}ξa(ΣTYΣ)abξb),\exp\!\left( -\frac14\sum_{a,b\in I\times\{0,1\}} \xi_a(\Sigma^{\mathsf T}Y\Sigma)_{ab}\xi_b \right),exp​−41​a,b∈I×{0,1}∑​ξa​(ΣTYΣ)ab​ξb​​,

with the real exponent embedded into C\mathbb CC, where Σ(i,0),(i,1)=1\Sigma_{(i,0),(i,1)}=1Σ(i,0),(i,1)​=1, Σ(i,1),(i,0)=−1\Sigma_{(i,1),(i,0)}=-1Σ(i,1),(i,0)​=−1, and all other entries are zero. This is the specified Gaussian characteristic expression with zero mean, after regrouping the two quadrature coordinates into I⊕II\oplus II⊕I. No symmetry or positivity condition is required of YYY, so no bound by 111 is part of the definition. For empty III, the sum is zero and the value is 111.

QAlgorithms.Papers.Independent.Support50.P28.oneModeGenerator

For any index a∈{0,1,2}a\in\{0,1,2\}a∈{0,1,2}, define the real 2×22\times22×2 matrix

oneModeGenerator⁡(a)={(01−10),a=0,(100−1),a=1,(0110),a=2.\operatorname{oneModeGenerator}(a)= \begin{cases} \begin{pmatrix}0&1\\-1&0\end{pmatrix},&a=0,\\[4pt] \begin{pmatrix}1&0\\0&-1\end{pmatrix},&a=1,\\[4pt] \begin{pmatrix}0&1\\1&0\end{pmatrix},&a=2. \end{cases}oneModeGenerator(a)=⎩⎨⎧​(0−1​10​),(10​0−1​),(01​10​),​a=0,a=1,a=2.​

QAlgorithms.Papers.Independent.Support50.P28.oscillatorCharacteristic

For any finite type III, any complex matrix AAA indexed by occupation functions u:I→Nu:I\to\mathbb Nu:I→N, and any ξ:I×{0,1}→R\xi:I\times\{0,1\}\to\mathbb Rξ:I×{0,1}→R, this defines

χA(ξ)=∑u:I→N ∑v:I→NAu,v∏i∈IDzi(vi,ui),zi=−ξ(i,0)+i ξ(i,1)2,\chi_A(\xi)= \sum_{u:I\to\mathbb N}\ \sum_{v:I\to\mathbb N} A_{u,v}\prod_{i\in I}D_{z_i}(v_i,u_i), \qquad z_i=-\frac{\xi(i,0)+\mathrm i\,\xi(i,1)}{\sqrt2},χA​(ξ)=u:I→N∑​ v:I→N∑​Au,v​i∈I∏​Dzi​​(vi​,ui​),zi​=−2​ξ(i,0)+iξ(i,1)​,

where, for z∈Cz\in\mathbb Cz∈C and m,n∈Nm,n\in\mathbb Nm,n∈N,

Dz(m,n)=e−∣z∣2/2m! n!∑j=0min⁡(m,n)zm−j(−z‾)n−jj! (m−j)! (n−j)!.D_z(m,n)= e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{j=0}^{\min(m,n)} \frac{z^{m-j}(-\overline z)^{n-j}} {j!\,(m-j)!\,(n-j)!}.Dz​(m,n)=e−∣z∣2/2m!n!​j=0∑min(m,n)​j!(m−j)!(n−j)!zm−j(−z)n−j​.

No summability or density assumption is required: each infinite complex sum is the total sum operation, which takes the value zero when its family is not summable. Empty III is included; then the product is 111, there is one occupation function, and χA(ξ)\chi_A(\xi)χA​(ξ) is the sole entry of AAA.

QAlgorithms.Papers.Independent.Support50.P28.oscillatorCoefficients

For any type III and any bounded complex-linear operator TTT on H=ℓ2(Ω;C)H=\ell^2(\Omega;\mathbb C)H=ℓ2(Ω;C), where Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, this defines the complex matrix indexed by Ω×Ω\Omega\times\OmegaΩ×Ω whose entries are

Auv=⟨eu,Tev⟩,A_{uv}=\langle e_u,T e_v\rangle,Auv​=⟨eu​,Tev​⟩,

where eue_ueu​ is 111 at occupation uuu and 000 elsewhere, and the inner product is conjugate-linear in its first argument. Thus AuvA_{uv}Auv​ is the uuu-coordinate of TevT e_vTev​. No positivity or trace-class assumption is imposed, and III may be empty or infinite.

QAlgorithms.Papers.Independent.Support50.P28.oscillatorNumberBasis

For any type III, this defines the canonical orthonormal Hilbert basis of H=ℓ2(Ω;C)H=\ell^2(\Omega;\mathbb C)H=ℓ2(Ω;C), indexed by Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, using the identity linear isometry of HHH. Its vector eue_ueu​, for u∈Ωu\in\Omegau∈Ω, has coordinate 111 at uuu and 000 at every other occupation. No finiteness or countability assumption is required; for empty III, this basis has one vector.

QAlgorithms.Papers.Independent.Support50.P28.pNorm

For any type III, real number ppp, and complex matrix AAA indexed by Ω×Ω\Omega\times\OmegaΩ×Ω, where Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, this defines

(∑u∈Ωmax⁡ ⁣{Re⁡⟨eu,Tpeu⟩,0})1/pin [0,+∞].\left(\sum_{u\in\Omega} \max\!\left\{\operatorname{Re}\langle e_u,T^p e_u\rangle,0\right\}\right)^{1/p} \quad\text{in }[0,+\infty].(u∈Ω∑​max{Re⟨eu​,Tpeu​⟩,0})1/pin [0,+∞].

Here eue_ueu​ is the coordinate unit vector of ℓ2(Ω;C)\ell^2(\Omega;\mathbb C)ℓ2(Ω;C), TTT is a classically selected bounded complex-linear realization of AAA if one exists and is zero otherwise, and TpT^pTp uses the total continuous-functional-calculus real power. The outer power is the extended-nonnegative-real power. There are no positivity, summability, or realizability assumptions, and ppp may be negative or zero; at p=0p=0p=0, real division gives 1/p=01/p=01/p=0, so the defined value is 111, including when the inner sum is zero or infinite.

QAlgorithms.Papers.Independent.Support50.P28.positivePower

For any type III, any complex matrix AAA indexed by Ω×Ω\Omega\times\OmegaΩ×Ω, with Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, and any real exponent ppp, this defines the continuous-functional-calculus real power TpT^pTp on H=ℓ2(Ω;C)H=\ell^2(\Omega;\mathbb C)H=ℓ2(Ω;C). Here TTT is a classically selected bounded complex-linear operator satisfying ⟨eu,Tev⟩=Auv\langle e_u,T e_v\rangle=A_{uv}⟨eu​,Tev​⟩=Auv​ for all coordinate unit vectors eu,eve_u,e_veu​,ev​, if such an operator exists, and is the zero operator otherwise. The definition installs the complex functional calculus for normal operators and the real functional calculus for self-adjoint operators, then uses the total operation for real powers. It imposes no positivity, self-adjointness, realizability, or trace-class hypothesis on AAA, and allows zero and negative ppp.

QAlgorithms.Papers.Independent.Support50.P28.productInput

For any natural number rrr, any family of natural numbers mim_imi​ indexed by 0≤i<r0\le i<r0≤i<r, and any family of states ρi\rho_iρi​ with coefficient matrices XiX_iXi​ indexed by Oi=N{0,…,mi−1}\mathcal O_i=\mathbb N^{\{0,\ldots,m_i-1\}}Oi​=N{0,…,mi​−1}, define a complex matrix on occupation functions of the disjoint union J={(i,k):0≤i<r, 0≤k<mi}J=\{(i,k):0\le i<r,\ 0\le k<m_i\}J={(i,k):0≤i<r, 0≤k<mi​} by

productInput⁡(ρ)uv=∏i=0r−1Xi(ui,vi),ui(k)=u(i,k),vi(k)=v(i,k).\operatorname{productInput}(\rho)_{uv} =\prod_{i=0}^{r-1}X_i(u_i,v_i), \qquad u_i(k)=u(i,k),\quad v_i(k)=v(i,k).productInput(ρ)uv​=i=0∏r−1​Xi​(ui​,vi​),ui​(k)=u(i,k),vi​(k)=v(i,k).

Being a state here means that, for each iii, there exist sequences ai,n,bi,n∈ℓ2(Oi;C)a_{i,n},b_{i,n}\in\ell^2(\mathcal O_i;\mathbb C)ai,n​,bi,n​∈ℓ2(Oi​;C) with ∑n∥ai,n∥∥bi,n∥<∞\sum_n\|a_{i,n}\|\|b_{i,n}\|<\infty∑n​∥ai,n​∥∥bi,n​∥<∞ and Xi(s,t)=∑nai,n(s)bi,n(t)‾X_i(s,t)=\sum_n a_{i,n}(s)\overline{b_{i,n}(t)}Xi​(s,t)=∑n​ai,n​(s)bi,n​(t)​; every finite quadratic form ∑s,t∈Fzs‾Xi(s,t)zt\sum_{s,t\in F}\overline{z_s}X_i(s,t)z_t∑s,t∈F​zs​​Xi​(s,t)zt​, for finite F⊆OiF\subseteq\mathcal O_iF⊆Oi​ and arbitrary z:Oi→Cz:\mathcal O_i\to\mathbb Cz:Oi​→C, is real and nonnegative; and the real diagonal entries are summable with sum 111. Zero values of mim_imi​ are allowed. If r=0r=0r=0, JJJ is empty and the resulting matrix has its single entry equal to the empty product 111.

QAlgorithms.Papers.Independent.Support50.P28.pureMaximalOutputPurity

For any finite type III, any real ppp, and any function E:CL×L→CL×LE:\mathbb C^{L\times L}\to\mathbb C^{L\times L}E:CL×L→CL×L, where L=NIL=\mathbb N^IL=NI, this defines

sup⁡R(∑u∈L[Re⁡⟨eu,TE(R) peu⟩]+)1/p∈[0,∞].\sup_R \left( \sum_{u\in L} \left[\operatorname{Re} \langle e_u,T_{E(R)}^{\,p}e_u\rangle\right]_+ \right)^{1/p}\in[0,\infty].Rsup​(u∈L∑​[Re⟨eu​,TE(R)p​eu​⟩]+​)1/p∈[0,∞].

Here [x]+=max⁡(x,0)[x]_+=\max(x,0)[x]+​=max(x,0); eue_ueu​ is the coordinate unit vector in ℓ2(L;C)\ell^2(L;\mathbb C)ℓ2(L;C); TBT_BTB​ is a selected bounded complex linear operator with coefficient matrix BBB, if one exists, and zero otherwise; its power is taken by real continuous functional calculus; and the outer power is extended nonnegative real exponentiation. The inputs RRR must satisfy all of these conditions: Ruv=∑afa(u)ga(v)‾R_{uv}=\sum_a f_a(u)\overline{g_a(v)}Ruv​=∑a​fa​(u)ga​(v)​ for square-summable vectors with ∑a∥fa∥∥ga∥<∞\sum_a\|f_a\|\|g_a\|<\infty∑a​∥fa​∥∥ga​∥<∞; all its finite quadratic forms are real and nonnegative; its real diagonal is summable with sum 111; there is a unit vector h∈ℓ2(L;C)h\in\ell^2(L;\mathbb C)h∈ℓ2(L;C) with Ruv=h(u)h(v)‾R_{uv}=h(u)\overline{h(v)}Ruv​=h(u)h(v)​; and there is a real matrix GGG, indexed by Q=I×{0,1}Q=I\times\{0,1\}Q=I×{0,1}, satisfying

GT=G,G+iJ⪰0,χR(ξ)=exp⁡ ⁣(−14ξTJTGJξ)(ξ∈RQ).G^{\mathsf T}=G,\qquad G+\mathrm iJ\succeq0,\qquad \chi_R(\xi)= \exp\!\left(-\frac14\xi^{\mathsf T}J^{\mathsf T}GJ\xi\right) \quad(\xi\in\mathbb R^Q).GT=G,G+iJ⪰0,χR​(ξ)=exp(−41​ξTJTGJξ)(ξ∈RQ).

Here JJJ has blocks (01−10)\begin{pmatrix}0&1\\-1&0\end{pmatrix}(0−1​10​), and

χR(ξ)=∑u∑vRuv∏i∈IDvi,ui ⁣(−ξi,0+iξi,12),Dmn(z)=e−∣z∣2/2m! n!∑a=0min⁡(m,n)zm−a(−z‾)n−aa!(m−a)!(n−a)!.\chi_R(\xi)=\sum_u\sum_v R_{uv} \prod_{i\in I}D_{v_i,u_i}\!\left( -\frac{\xi_{i,0}+\mathrm i\xi_{i,1}}{\sqrt2}\right), \quad D_{mn}(z)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{a=0}^{\min(m,n)} \frac{z^{m-a}(-\overline z)^{n-a}} {a!(m-a)!(n-a)!}.χR​(ξ)=u∑​v∑​Ruv​i∈I∏​Dvi​,ui​​(−2​ξi,0​+iξi,1​​),Dmn​(z)=e−∣z∣2/2m!n!​a=0∑min(m,n)​a!(m−a)!(n−a)!zm−a(−z)n−a​.

Every inner sum and the resulting outer sum must be summable for every ξ\xiξ. The function EEE is otherwise arbitrary, and ppp may be zero or negative. At p=0p=0p=0, total real division gives 1/p=01/p=01/p=0, so each admissible input contributes 111. An empty supremum is 000. Empty III is allowed and gives a singleton occupation label set.

QAlgorithms.Papers.Independent.Support50.P28.realizeOscillatorCoeff

For any type III and any complex matrix AAA indexed by Ω×Ω\Omega\times\OmegaΩ×Ω, where Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, this defines an operator on H=ℓ2(Ω;C)H=\ell^2(\Omega;\mathbb C)H=ℓ2(Ω;C) as follows: if there exists a bounded complex-linear operator T:H→HT:H\to HT:H→H satisfying ⟨eu,Tev⟩=Auv\langle e_u,T e_v\rangle=A_{uv}⟨eu​,Tev​⟩=Auv​ for every u,v∈Ωu,v\in\Omegau,v∈Ω, it selects such an operator by classical choice; otherwise it returns the zero operator. Here eue_ueu​ is the coordinate unit vector at uuu. The definition accepts arbitrary matrices, including those with no bounded operator realization, and requires no finiteness or countability assumption on III.

QAlgorithms.Papers.Independent.Support50.P28.replaceOccupation

For every natural number rrr, every family of natural numbers m0,…,mr−1m_0,\ldots,m_{r-1}m0​,…,mr−1​, every index i<ri<ri<r, every occupation assignment uuu on the disjoint union

L={(j,k):0≤j<r, 0≤k<mj},L=\{(j,k):0\le j<r,\ 0\le k<m_j\},L={(j,k):0≤j<r, 0≤k<mj​},

and every occupation assignment a:{0,…,mi−1}→Na:\{0,\ldots,m_i-1\}\to\mathbb Na:{0,…,mi​−1}→N, this defines the assignment

u[i←a](j,k)={a(k),j=i,u(j,k),j≠i.u[i\leftarrow a](j,k)= \begin{cases} a(k),&j=i,\\ u(j,k),&j\ne i. \end{cases}u[i←a](j,k)={a(k),u(j,k),​j=i,j=i.​

Thus it replaces exactly block iii, retaining every other value. If mi=0m_i=0mi​=0, that block has no coordinates and the result is uuu. If r=0r=0r=0, there is no possible argument iii.

QAlgorithms.Papers.Independent.Support50.P28.scalarX

For any centered oscillator channel EEE on the singleton mode set {0}\{0\}{0}, let XEX_EXE​ be its real matrix indexed by {0}×{0,1}\{0\}\times\{0,1\}{0}×{0,1}, hence a 2×22\times22×2 matrix. This defines the real covariance-indexed matrix

∣det⁡XE∣ I2.\sqrt{|\det X_E|}\,I_2.∣detXE​∣​I2​.

The determinant is first replaced by its absolute value, so zero, positive, and negative determinants are all covered, and determinant zero gives the zero matrix.

QAlgorithms.Papers.Independent.Support50.P28.scalarY

For any centered oscillator channel EEE on the singleton mode set {0}\{0\}{0}, let YEY_EYE​ be its real matrix indexed by {0}×{0,1}\{0\}\times\{0,1\}{0}×{0,1}, hence a 2×22\times22×2 matrix. This defines

det⁡YE I2.\sqrt{\det Y_E}\,I_2.detYE​​I2​.

There is no absolute value around the determinant. The square root is the total real square-root function, which is zero for nonpositive arguments; in particular, determinant zero gives the zero matrix.

QAlgorithms.Papers.Independent.Support50.P28.sourceBeta

Define the real matrix indexed by 0,10,10,1, in that order, by

sourceBeta⁡=(100−1).\operatorname{sourceBeta} =\begin{pmatrix}1&0\\0&-1\end{pmatrix}.sourceBeta=(10​0−1​).

QAlgorithms.Papers.Independent.Support50.P28.sourceDelta

Define the real matrix indexed by 0,10,10,1, in that order, by

sourceDelta⁡=(0110).\operatorname{sourceDelta} =\begin{pmatrix}0&1\\1&0\end{pmatrix}.sourceDelta=(01​10​).

QAlgorithms.Papers.Independent.Support50.P28.sourceSymplecticGenerator

For any natural number nnn and any generator index aaa, define a real matrix KaK_aKa​ indexed by pairs (k,b)(k,b)(k,b) with 0≤k<n0\le k<n0≤k<n and b∈{0,1}b\in\{0,1\}b∈{0,1}, viewing it as an n×nn\times nn×n array of 2×22\times22×2 blocks. Write

J=(01−10),B=(100−1),D=(0110).J=\begin{pmatrix}0&1\\-1&0\end{pmatrix}, \qquad B=\begin{pmatrix}1&0\\0&-1\end{pmatrix}, \qquad D=\begin{pmatrix}0&1\\1&0\end{pmatrix}.J=(0−1​10​),B=(10​0−1​),D=(01​10​).

If aaa is a single-mode index (k,t)(k,t)(k,t), where t∈{0,1,2}t\in\{0,1,2\}t∈{0,1,2}, its only possibly nonzero block is (k,k)(k,k)(k,k), equal respectively to J,B,DJ,B,DJ,B,D. If aaa is a pair index (k,ℓ,t)(k,\ell,t)(k,ℓ,t), where 0≤k<ℓ<n0\le k<\ell<n0≤k<ℓ<n and t∈{0,1,2,3}t\in\{0,1,2,3\}t∈{0,1,2,3}, all blocks vanish except the following:

t(Ka)kℓ(Ka)ℓk0JJ1−I2I22DD3BB\begin{array}{c|cc} t&(K_a)_{k\ell}&(K_a)_{\ell k}\\ \hline 0&J&J\\ 1&-I_2&I_2\\ 2&D&D\\ 3&B&B \end{array}t0123​(Ka​)kℓ​J−I2​DB​(Ka​)ℓk​JI2​DB​​

For n=0n=0n=0, there is no generator index to supply as an argument.

QAlgorithms.Papers.Independent.Support50.P28.spectralFp

For any finite type III, any real number ppp, and any function ν:I→R\nu:I\to\mathbb Rν:I→R, define

spectralFp⁡(p,ν)=∏i∈I((νi+1)p−(νi−1)p),\operatorname{spectralFp}(p,\nu) =\prod_{i\in I}\bigl((\nu_i+1)^p-(\nu_i-1)^p\bigr),spectralFp(p,ν)=i∈I∏​((νi​+1)p−(νi​−1)p),

using total real powers without assuming positivity of the bases or exponent. The type III may be empty, in which case the product is 111.

QAlgorithms.Papers.Independent.Support50.P28.tensorOscillatorChannel

For every natural number rrr, every family mj∈Nm_j\in\mathbb Nmj​∈N indexed by j<rj<rj<r, every family of centered oscillator channel data EjE_jEj​ on the mjm_jmj​ modes, and every complex coefficient matrix AAA indexed by occupation assignments on L={(j,k):j<r, k<mj}L=\{(j,k):j<r,\ k<m_j\}L={(j,k):j<r, k<mj​}, this defines the result of applying the coefficient maps Φj=(Ej).map\Phi_j=(E_j).\mathrm{map}Φj​=(Ej​).map successively in the order 0,1,…,r−10,1,\ldots,r-10,1,…,r−1. Explicitly, set B(0)=AB^{(0)}=AB(0)=A and, for i<ri<ri<r, set

Buv(i+1)=Φi ⁣((Bu[i←a], v[i←b](i))a,b∈Nmi)ui,vi;B^{(i+1)}_{uv} = \Phi_i\!\left( \bigl(B^{(i)}_{u[i\leftarrow a],\,v[i\leftarrow b]}\bigr)_ {a,b\in\mathbb N^{m_i}} \right)_{u_i,v_i};Buv(i+1)​=Φi​((Bu[i←a],v[i←b](i)​)a,b∈Nmi​​)ui​,vi​​;

the defined output is B(r)B^{(r)}B(r). Here ui(k)=u(i,k)u_i(k)=u(i,k)ui​(k)=u(i,k), and replacing block iii changes u(i,k)u(i,k)u(i,k) to a(k)a(k)a(k) while retaining all coordinates in other blocks. Each EiE_iEi​ supplies its Gaussian complete-positivity data, its map preserving summable nuclear representations and linear on those matrices, its density-state output agreeing with that map, finite-ancilla density preservation, and its summable output characteristic law; this construction uses its coefficient map. The argument AAA itself need not be a density matrix or admit a nuclear representation. If r=0r=0r=0, the function is the identity. Blocks with mi=0m_i=0mi​=0 are allowed and still involve their supplied maps on matrices indexed by the singleton empty occupation assignment.

QAlgorithms.Papers.Independent.Support50.P28.tracePower

For any type III, real number ppp, and complex matrix AAA indexed by Ω×Ω\Omega\times\OmegaΩ×Ω, where Ω={u:I→N}\Omega=\{u:I\to\mathbb N\}Ω={u:I→N}, this defines an extended nonnegative real number

∑u∈Ωmax⁡ ⁣{Re⁡⟨eu,Tpeu⟩,0}.\sum_{u\in\Omega}\max\!\left\{\operatorname{Re}\langle e_u,T^p e_u\rangle,0\right\}.u∈Ω∑​max{Re⟨eu​,Tpeu​⟩,0}.

Here eue_ueu​ is the coordinate unit vector in ℓ2(Ω;C)\ell^2(\Omega;\mathbb C)ℓ2(Ω;C), TTT is a bounded complex-linear operator realizing the entries of AAA, selected by classical choice if one exists and set to zero otherwise, and TpT^pTp is its total continuous-functional-calculus real power. Each real summand is converted to an extended nonnegative real before summation, so negative real diagonal values contribute zero and the sum may be +∞+\infty+∞. No summability, positivity, realizability, or restriction on ppp is assumed, and the index type III need not be finite or countable.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 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