Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Quantum f-divergences and error correction

Definition
QAlgorithms_Papers_Independent_Support50_P12

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

prove2quantum

The objects the formalized statements of Quantum f-divergences and error correction (arXiv:1008.2529) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • Alg
  • supp: Section 2, p.4: support projection, including support(0)=0.
  • AlgebraDensity: Section 2, p.4: density operators in the stated concrete algebra.
  • spec: Section 2, p.5: distinct spectral values, including zero when it is an eigenvalue.
  • cpow: Section 2, p.5: support-zero complex powers, including z=0.
  • modular: Equation (2.1), p.5: functional calculus of L_A R_(B^-1), acting on X.
  • quasi: Definition 2.1, p.5: the finite Hilbert–Schmidt expression.
  • recessionSlope: Definition 2.1, p.5: recession slope in signed extended reals.
  • DivergenceGenerator: Definition 2.1, p.5: continuity only at strictly positive arguments and existence of the signed extended recession slope. Values at negative arguments are unused.
  • div: Definition 2.1, p.5: supported expression, and one-sided regularization otherwise.
  • supportScalarPower: Example 2.7, pp.6–7: scalar support powers, also at exponent zero.
  • petzRenyi: Example 2.7, pp.6–7: unnormalized Petz Rényi relative entropy.
  • AlgebraExtension: Section 3, p.9: canonical extensions of maps between concrete algebras. The range conditions mean Phi=i_B ◦ phi ◦ E_A; E_A is the trace-orthogonal projection.
  • PositiveOnConcreteAlgebra: Section 3, p.9: positivity only on the stated algebra.
  • AlgebraTracePreserving: Section 3, p.9: trace preservation on the stated algebra.
  • AlgebraTraceNonincreasing: Section 3, p.9: trace non-increase on positive elements.
  • SchwarzBound: Section 3, p.12: Schwarz bound, including positivity at coefficient zero.
  • AlgebraSchwarzMap: Section 3, p.12: finite Schwarz norm means existence of a nonnegative bound.
  • snorm: Section 3, p.12: extended norm, +infinity if there is no Schwarz bound.
  • pnorm: Section 3, p.12: operator norm, using the source equality for positive maps.
  • Substochastic: Section 3, p.12: the adjoint Schwarz condition, not a CP condition.
  • AlgebraStochastic: Section 3, p.12: stochastic = trace preserving and substochastic.
  • modularContraction: Equation (4.1), p.16: the actual Hilbert–Schmidt contraction V.
  • CornerTP: Section 3, p.10: trace preservation on a support corner.
  • amp: Section 3, p.9: n×n block amplification of a map, acting on genuine block entries.
  • AlgebraKPositive: Section 3, p.9: n-positivity on all positive block matrices over the domain algebra.
  • AlgebraCompletelyPositive: Section 3, p.9: complete positivity at every finite amplification size.
  • CoBound: Remark 4.8, p.20: co-Schwarz condition, with reversed product in the argument.
  • CoSubstochastic: Remark 4.8, p.20: co-substochastic maps use the adjoint co-Schwarz contraction.
  • CoStochastic: Remark 4.8, p.20: co-stochastic maps.
  • KCoPositive: Remark 5.7, p.26: n-copositivity is positivity after transposing each input block.
  • SemidefiniteOperatorConvex: Section 4, p.17: all-size operator convexity INCLUDING semidefinite matrices.
  • PosReal: Theorem 8.1, pp.37–38: measures live on the OPEN positive half-line.
  • representationKernel: Equation (8.2), p.37: the combined kernel is integrated as one function.
  • HalfLineRepresentation: Theorem 8.1, p.37: a genuine integral representation, with integrability guards.
  • continuousPart: Equation (5.1), p.21: remove the possible jump at zero before representing f.
  • representingMeasure: Section 5, p.21: the unique representing measure, computed from the continuous part. Only operator convex functions are used in measure-support criteria.
  • Rich: Sections 5 and 7: actual cardinal support, including infinite support.
  • resolventGenerator: Equation (4.9), p.17.
  • modularSpec: Section 5, p.21: spectrum of the relative modular operator on the stated algebra.
  • logOverlap: Definition 6.1, p.27: raw log-overlap, no normalization by trace.
  • chernoff: Definition 6.1, p.27: Chernoff supremum on [0,1).
  • hoeffding: Definition 6.1, p.27: Hoeffding distance as a signed extended supremum.
  • psiReal: Remark 6.2, pp.27–28: finite log-overlap representative, used only on nonorthogonal pairs.
  • logOverlapPerspective: Remark 6.2, p.27: perspective, including +infinity on the negative half-line.
  • polar: Remark 6.2, p.27: Legendre–Fenchel transform in extended reals.
  • Reversible: Sections 6–7: exact recoverability of a positive pair by one stochastic map.
  • RepresentsStateDivergence: Section 6, p.30: representability on the state space of one algebra.
  • TwoProjections: Section 6, p.30: nontrivial algebras admit two nonzero orthogonal projections.
  • Noncommutative: Sections 6–7: noncommutativity is a property of the actual observable algebra.
  • signedSchatten: Proposition A.6, p.46: norm-like functional for ALL nonzero real exponents. The zero-matrix branch is explicitly zero, including at negative q.
  • SpectralAlphabet: Lemma 2.9, p.7: the exact product spectral alphabet.
  • law1: Lemma 2.9, p.7: the first classical distribution a Tr(P_a Q_b).
  • law2: Lemma 2.9, p.7: the second classical distribution b Tr(P_a Q_b).
  • isometricConjugation: Corollary 2.5, p.6: linear isometries, or anti-linear isometries via transpose.
  • pairConditions: Section 5, p.21: ten conditions, computed from the input pair. The Boolean selects Remark 5.7's co-variant, changing precisely (i),(ii),(vi)–(viii).
  • PairChain: Theorem 5.1, pp.21–25: asymmetric implication chain, never a blanket equivalence.
  • RecoveryRefinement: Sections 5 and 7: preserve the positivity level when a stochastic recovery exists.
  • hull: Section 7, p.33: CLOSED convex hull in the finite-dimensional matrix topology.
  • Covers: Section 7, p.33: maximal support without choosing an unrelated projection.
  • CodeSupportEquals: Section 7, p.33: equality of a reference support with the join of code supports.
  • RecoveryBlocks: Section 7, p.34: shared direct-sum/tensor data. Algebraic action and cone equality are deliberately NOT fields; they are conclusions in the theorems below.
  • inputBlock: Shared block embedding on the input side, equation (3.8), p.15.
  • outputBlock: Shared block action on the output side, equation (3.8), p.15.
  • BlockSupports: Section 7, p.34: support joins are exactly the direct sum of all the shared blocks.
  • BlockState: Section 7, p.34: a code state has variable left states and probabilities, while the noise states, embeddings, and unitaries are shared across the whole code.
  • bayesianTestingDistance: Equation (6.8), p.29: optimal Bayesian error normalized by min(p,1-p).
  • codeConditions: Theorem 7.1, pp.33–34: its ten uniform criteria.
  • MonotoneRepresentation: Section 6, p.30: monotone functions are strictly increasing on the ATTAINED extended divergence range, rather than on an arbitrary real domain.
  • transposeNoise: Example 3.6, p.12: transpose mixed with complete depolarization.
  • identityNoise: Example 3.6, pp.12–13: identity mixed with complete depolarization.
  • transposeMix: Example 3.6, p.13: transpose mixed with identity.
  • transposeDefect: Example 3.6, p.12: the full six-term Schwarz defect expansion.
  • MonotoneMap: Remark 4.8, pp.20–21: maps monotone for a specified f.
  • renyiOne: Remark 6.2, p.28: limiting order-one entropy, before identifying it on states.
  • exampleRho: Example 6.6, p.31: the stated diagonal state.
  • exampleSigma: Example 6.6, p.31: the stated reference state.
  • merge: Example 6.6, p.31: canonical extension of the diagonal block-merging map.
  • diagonalAlgebra: Section 6, p.31: the concrete diagonal observable algebra.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_Learning
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P04
import Definitions.Def_QAlgorithms_Defs_SpectralFilter
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P30
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P60
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P21
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P28
import Definitions.Def_QAlgorithms_Papers_Independent_Frontier18_P17
import Definitions.Def_QAlgorithms_Papers_Independent_Next46_P23
import Definitions.Def_QAlgorithms_Papers_Independent_Support21_P03
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Defs_QuantumOutputMetrics
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P71

/-! Hiai–Mosonyi–Petz–Beny, arXiv:1008.2529v6.
Accepted primitives supply matrices, linear maps, trace, Hilbert–Schmidt product,
spectral calculus, support inclusion, adjoints, Petz recovery and tensor-state powers.
Remaining paper-specific definitions use actual representation traces and support-zero powers.
An intrinsic algebra map is represented by its canonical full-matrix extension:
inclusion after the map after the trace-orthogonal projection onto its domain.
`AlgebraExtension` characterizes exactly these extensions by the ranges of the map and
its computed Hilbert–Schmidt adjoint. It adds no positivity or trace assumption.
Compilation is pending the authorized batch handoff. -/

noncomputable section

namespace QAlgorithms.Papers.Independent.Support50.P12

open Filter MeasureTheory Set
open QAlgorithms.Papers.Independent.Information100.P60 (Mat copies)
open QAlgorithms.Papers.Independent.Information100.P30 (MatrixMap)
open QAlgorithms.Papers.Independent.Dependency50.P28 (tr LE logNonnegativeExtended pinch)
open QAlgorithms.Papers.Independent.Dependency50.P21 (channelAdjoint spectralFunction petz relativeEntropy)
open QAlgorithms.Papers.Independent.Frontier18.P17 (hs)
open QAlgorithms.Papers.Independent.Next46.P23 (Supports)
open QAlgorithms.Papers.Independent.Support21.P03 (ExtendedConvex)
open QAlgorithms (spectralProjector)
open QAlgorithms.Learning (IsDistribution)
open QAlgorithms.Papers.Independent.Information100.P04 (absOp)
open scoped BigOperators Matrix ComplexConjugate ComplexOrder ENNReal Topology Kronecker

attribute [local instance] Classical.propDecidable Classical.decEq

abbrev Alg (I : Type) [Fintype I] [DecidableEq I] := StarSubalgebra ℂ (Mat I)

open QAlgorithms.Papers.Independent.Information100.P71 (supportPower)

variable {I J K : Type} [Fintype I] [DecidableEq I] [Fintype J] [DecidableEq J]
  [Fintype K] [DecidableEq K]

/-- Section 2, p.4: support projection, including support(0)=0. -/
def supp (X : Mat I) : Mat I := supportPower X 0

/-- Section 2, p.4: density operators in the stated concrete algebra. -/
def AlgebraDensity (A : Alg I) (rho : Mat I) : Prop := rho ∈ A ∧ rho.PosSemidef ∧ tr rho = 1

/-- Section 2, p.5: distinct spectral values, including zero when it is an eigenvalue. -/
def spec (X : Mat I) : Finset ℝ :=
  if h : X.IsHermitian then Finset.univ.image h.eigenvalues else ∅

/-- Section 2, p.5: support-zero complex powers, including z=0. -/
def cpow (X : Mat I) (z : ℂ) : Mat I :=
  spectralFunction (fun a => if a = 0 then 0 else Complex.exp (z * (Real.log a : ℂ))) X

/-- Equation (2.1), p.5: functional calculus of L_A R_(B^-1), acting on X. -/
def modular (f : ℝ → ℝ) (A B X : Mat I) : Mat I :=
  ∑ a ∈ spec A, ∑ b ∈ spec B, (f (a / b) : ℂ) • (spectralProjector A a * X * spectralProjector B b)

/-- Definition 2.1, p.5: the finite Hilbert–Schmidt expression. -/
def quasi (f : ℝ → ℝ) (A B : Mat I) : ℝ :=
  (hs (supportPower B (1 / 2)) (modular f A B (supportPower B (1 / 2)))).re

/-- Definition 2.1, p.5: recession slope in signed extended reals. -/
def recessionSlope (f : ℝ → ℝ) : EReal := limUnder atTop (fun x : ℝ => ((f x / x : ℝ) : EReal))

/-- Definition 2.1, p.5: continuity only at strictly positive arguments and existence
of the signed extended recession slope. Values at negative arguments are unused. -/
def DivergenceGenerator (f : ℝ → ℝ) : Prop :=
  ContinuousOn f (Ioi 0) ∧ Tendsto (fun x : ℝ => ((f x / x : ℝ) : EReal)) atTop (𝓝 (recessionSlope f))

/-- Definition 2.1, p.5: supported expression, and one-sided regularization otherwise. -/
def div (f : ℝ → ℝ) (A B : Mat I) : EReal :=
  if Supports A B then (quasi f A B : EReal) else
    limUnder (𝓝[>] (0 : ℝ)) (fun e => (quasi f A (B + (e : ℂ) • 1) : EReal))

/-- Example 2.7, pp.6–7: scalar support powers, also at exponent zero. -/
def supportScalarPower (a x : ℝ) : ℝ := if x = 0 then 0 else Real.rpow x a

/-- Example 2.7, pp.6–7: unnormalized Petz Rényi relative entropy. -/
def petzRenyi (a : ℝ) (A B : Mat I) : EReal :=
  if Supports A B ∨ a < 1 then
    (((a - 1)⁻¹ : ℝ) : EReal) * logNonnegativeExtended (tr (supportPower A a * supportPower B (1 - a)))
  else ⊤

/-- Section 3, p.9: canonical extensions of maps between concrete algebras.
The range conditions mean Phi=i_B ◦ phi ◦ E_A; E_A is the trace-orthogonal projection. -/
def AlgebraExtension (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) : Prop :=
  (∀ X, Phi X ∈ B) ∧ (∀ Y, channelAdjoint Phi Y ∈ A)

/-- Section 3, p.9: positivity only on the stated algebra. -/
def PositiveOnConcreteAlgebra (A : Alg I) (Phi : Mat I → Mat J) : Prop :=
  ∀ X ∈ A, X.PosSemidef → (Phi X).PosSemidef

/-- Section 3, p.9: trace preservation on the stated algebra. -/
def AlgebraTracePreserving (A : Alg I) (Phi : Mat I → Mat J) : Prop := ∀ X ∈ A, (Phi X).trace = X.trace

/-- Section 3, p.9: trace non-increase on positive elements. -/
def AlgebraTraceNonincreasing (A : Alg I) (Phi : Mat I → Mat J) : Prop :=
  ∀ X ∈ A, X.PosSemidef → tr (Phi X) ≤ tr X

/-- Section 3, p.12: Schwarz bound, including positivity at coefficient zero. -/
def SchwarzBound (A : Alg I) (Phi : Mat I → Mat J) (c : ℝ) : Prop :=
  0 ≤ c ∧ PositiveOnConcreteAlgebra A Phi ∧ ∀ X ∈ A, LE ((Phi X)ᴴ * Phi X) ((c : ℂ) • Phi (Xᴴ * X))

/-- Section 3, p.12: finite Schwarz norm means existence of a nonnegative bound. -/
def AlgebraSchwarzMap (A : Alg I) (Phi : Mat I → Mat J) : Prop := ∃ c, SchwarzBound A Phi c

/-- Section 3, p.12: extended norm, +infinity if there is no Schwarz bound. -/
def snorm (A : Alg I) (Phi : Mat I → Mat J) : ℝ≥0∞ :=
  sInf {c | ∃ r : ℝ, SchwarzBound A Phi r ∧ c = ENNReal.ofReal r}

/-- Section 3, p.12: operator norm, using the source equality for positive maps. -/
def pnorm (Phi : Mat I → Mat J) : ℝ :=
  Real.sqrt (sSup {x : ℝ | ∃ v : J → ℂ, ∑ j, ‖v j‖ ^ 2 = 1 ∧
    x = (∑ j, conj (v j) * (((Phi 1)ᴴ * Phi 1) *ᵥ v) j).re})

/-- Section 3, p.12: the adjoint Schwarz condition, not a CP condition. -/
def Substochastic (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) : Prop :=
  AlgebraExtension A B Phi ∧ SchwarzBound B (channelAdjoint Phi) 1

/-- Section 3, p.12: stochastic = trace preserving and substochastic. -/
def AlgebraStochastic (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) : Prop :=
  Substochastic A B Phi ∧ AlgebraTracePreserving A Phi

/-- Equation (4.1), p.16: the actual Hilbert–Schmidt contraction V. -/
def modularContraction (Phi : MatrixMap I J) (B : Mat I) (X : Mat J) : Mat I :=
  channelAdjoint Phi (X * supportPower (Phi B) (-(1 / 2))) * supportPower B (1 / 2)

/-- Section 3, p.10: trace preservation on a support corner. -/
def CornerTP (A : Alg I) (Phi : Mat I → Mat J) (P : Mat I) : Prop :=
  ∀ X ∈ A, (Phi (P * X * P)).trace = (P * X * P).trace

/-- Section 3, p.9: n×n block amplification of a map, acting on genuine block entries. -/
def amp (Phi : Mat I → Mat J) (n : ℕ) (X : Mat (Fin n × I)) : Mat (Fin n × J) :=
  fun u v => Phi (fun i j => X (u.1, i) (v.1, j)) u.2 v.2

/-- Section 3, p.9: n-positivity on all positive block matrices over the domain algebra. -/
def AlgebraKPositive (A : Alg I) (Phi : Mat I → Mat J) (n : ℕ) : Prop :=
  ∀ X : Mat (Fin n × I), (∀ a b, (fun i j => X (a, i) (b, j)) ∈ A) →
    X.PosSemidef → (amp Phi n X).PosSemidef

/-- Section 3, p.9: complete positivity at every finite amplification size. -/
def AlgebraCompletelyPositive (A : Alg I) (Phi : Mat I → Mat J) : Prop := ∀ n, AlgebraKPositive A Phi n

/-- Remark 4.8, p.20: co-Schwarz condition, with reversed product in the argument. -/
def CoBound (A : Alg I) (Phi : Mat I → Mat J) (c : ℝ) : Prop :=
  0 ≤ c ∧ PositiveOnConcreteAlgebra A Phi ∧ ∀ X ∈ A, LE ((Phi X)ᴴ * Phi X) ((c : ℂ) • Phi (X * Xᴴ))

/-- Remark 4.8, p.20: co-substochastic maps use the adjoint co-Schwarz contraction. -/
def CoSubstochastic (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) : Prop :=
  AlgebraExtension A B Phi ∧ CoBound B (channelAdjoint Phi) 1

/-- Remark 4.8, p.20: co-stochastic maps. -/
def CoStochastic (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) : Prop :=
  CoSubstochastic A B Phi ∧ AlgebraTracePreserving A Phi

/-- Remark 5.7, p.26: n-copositivity is positivity after transposing each input block. -/
def KCoPositive (A : Alg I) (Phi : Mat I → Mat J) (n : ℕ) : Prop :=
  ∀ X : Mat (Fin n × I), (∀ a b, (fun i j => X (a, j) (b, i)) ∈ A) →
    X.PosSemidef → (amp (fun Y => Phi (Yᵀ)) n X).PosSemidef

/-- Section 4, p.17: all-size operator convexity INCLUDING semidefinite matrices. -/
def SemidefiniteOperatorConvex (f : ℝ → ℝ) : Prop :=
  ∀ (n : ℕ) (A B : Mat (Fin n)), A.PosSemidef → B.PosSemidef →
    ∀ t ∈ Icc (0 : ℝ) 1,
      LE (cfc f ((t : ℂ) • A + ((1 - t : ℝ) : ℂ) • B))
        ((t : ℂ) • cfc f A + ((1 - t : ℝ) : ℂ) • cfc f B)

/-- Theorem 8.1, pp.37–38: measures live on the OPEN positive half-line. -/
abbrev PosReal := {t : ℝ // 0 < t}

/-- Equation (8.2), p.37: the combined kernel is integrated as one function. -/
def representationKernel (x : ℝ) (t : PosReal) : ℝ := x / (1 + t.1) - x / (x + t.1)

/-- Theorem 8.1, p.37: a genuine integral representation, with integrability guards. -/
def HalfLineRepresentation (f : ℝ → ℝ) (a b : ℝ) (mu : Measure PosReal) : Prop :=
  0 ≤ b ∧ Integrable (fun t : PosReal => (1 + t.1)⁻¹ ^ 2) mu ∧
  ∀ x ≥ 0, Integrable (representationKernel x) mu ∧
    f x = f 0 + a * x + b * x ^ 2 + ∫ t, representationKernel x t ∂mu

/-- Equation (5.1), p.21: remove the possible jump at zero before representing f. -/
def continuousPart (f : ℝ → ℝ) (x : ℝ) : ℝ :=
  if x = 0 then limUnder (𝓝[>] (0 : ℝ)) f else f x

/-- Section 5, p.21: the unique representing measure, computed from the continuous part.
Only operator convex functions are used in measure-support criteria. -/
def representingMeasure (f : ℝ → ℝ) : Measure PosReal :=
  Classical.epsilon (fun mu => ∃ a b, HalfLineRepresentation (continuousPart f) a b mu)

/-- Sections 5 and 7: actual cardinal support, including infinite support. -/
def Rich (f : ℝ → ℝ) (n : ℕ) : Prop :=
  (n : Cardinal) ≤ Cardinal.mk (representingMeasure f).support

/-- Equation (4.9), p.17. -/
def resolventGenerator (t x : ℝ) : ℝ := -x / (x + t)

/-- Section 5, p.21: spectrum of the relative modular operator on the stated algebra. -/
def modularSpec (A : Alg I) (X Y : Mat I) : Finset ℝ :=
  ((spec X).product (spec Y)).image (fun p => p.1 / p.2) |>.filter
    (fun a => ∃ Z ∈ A, Z ≠ 0 ∧ X * Z * supportPower Y (-1) = (a : ℂ) • Z)

/-- Definition 6.1, p.27: raw log-overlap, no normalization by trace. -/
def logOverlap (A B : Mat I) (a : ℝ) : EReal :=
  logNonnegativeExtended (tr (supportPower A a * supportPower B (1 - a)))

/-- Definition 6.1, p.27: Chernoff supremum on [0,1). -/
def chernoff (A B : Mat I) : EReal :=
  sSup {v | ∃ a : ℝ, 0 ≤ a ∧ a < 1 ∧ v = -logOverlap A B a}

/-- Definition 6.1, p.27: Hoeffding distance as a signed extended supremum. -/
def hoeffding (A B : Mat I) (r : ℝ) : EReal :=
  sSup {v | ∃ a : ℝ, 0 ≤ a ∧ a < 1 ∧
    v = (((1 - a)⁻¹ : ℝ) : EReal) * ((-a * r : ℝ) - logOverlap A B a)}

/-- Remark 6.2, pp.27–28: finite log-overlap representative, used only on nonorthogonal pairs. -/
def psiReal (A B : Mat I) (a : ℝ) : ℝ := (logOverlap A B a).toReal

/-- Remark 6.2, p.27: perspective, including +infinity on the negative half-line. -/
def logOverlapPerspective (A B : Mat I) (s : ℝ) : EReal :=
  if 0 ≤ s then ((1 + s : ℝ) : EReal) * logOverlap A B (s / (1 + s)) else ⊤

/-- Remark 6.2, p.27: Legendre–Fenchel transform in extended reals. -/
def polar (f : ℝ → EReal) (r : ℝ) : EReal :=
  sSup {v | ∃ s : ℝ, v = ((s * r : ℝ) : EReal) - f s}

/-- Sections 6–7: exact recoverability of a positive pair by one stochastic map. -/
def Reversible (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) (X Y : Mat I) : Prop :=
  ∃ Psi : MatrixMap J I, AlgebraStochastic B A Psi ∧ Psi (Phi X) = X ∧ Psi (Phi Y) = Y

/-- Section 6, p.30: representability on the state space of one algebra. -/
def RepresentsStateDivergence (A : Alg I) (D : Mat I → Mat I → EReal) (f : ℝ → ℝ) : Prop :=
  DivergenceGenerator f ∧ ∀ rho sigma, AlgebraDensity A rho → AlgebraDensity A sigma → D rho sigma = div f rho sigma

/-- Section 6, p.30: nontrivial algebras admit two nonzero orthogonal projections. -/
def TwoProjections (A : Alg I) : Prop :=
  ∃ P Q : Mat I, P ∈ A ∧ Q ∈ A ∧ P.IsHermitian ∧ Q.IsHermitian ∧
    P * P = P ∧ Q * Q = Q ∧ P ≠ 0 ∧ Q ≠ 0 ∧ P * Q = 0

/-- Sections 6–7: noncommutativity is a property of the actual observable algebra. -/
def Noncommutative (A : Alg I) : Prop := ∃ X ∈ A, ∃ Y ∈ A, X * Y ≠ Y * X

/-- Proposition A.6, p.46: norm-like functional for ALL nonzero real exponents.
The zero-matrix branch is explicitly zero, including at negative q. -/
def signedSchatten (p : ℝ) (X : Mat I) : ℝ :=
  if X = 0 then 0 else Real.rpow (tr (supportPower (absOp X) p)) (1 / p)

/-- Section 2 convention at the zero eigenvalue and zero exponent. -/
example : supportScalarPower 0 0 = 0 := by simp [supportScalarPower]

/-- Section 2 convention for the zero overlap. -/
example : logNonnegativeExtended 0 = ⊥ := by simp [logNonnegativeExtended]

/-- The zero-map adjoint is zero, independently of the algebra representation. -/
example (Y : Mat J) : channelAdjoint (0 : MatrixMap I J) Y = 0 := by ext i j; simp [channelAdjoint]

/-- The Appendix's negative-exponent zero branch. -/
example (p : ℝ) : signedSchatten p (0 : Mat I) = 0 := by simp [signedSchatten]

/-- Lemma 2.9, p.7: the exact product spectral alphabet. -/
abbrev SpectralAlphabet (X Y : Mat I) := ↥(spec X) × ↥(spec Y)

/-- Lemma 2.9, p.7: the first classical distribution a Tr(P_a Q_b). -/
def law1 (X Y : Mat I) : SpectralAlphabet X Y → ℝ :=
  fun p => p.1.1 * tr (spectralProjector X p.1.1 * spectralProjector Y p.2.1)

/-- Lemma 2.9, p.7: the second classical distribution b Tr(P_a Q_b). -/
def law2 (X Y : Mat I) : SpectralAlphabet X Y → ℝ :=
  fun p => p.2.1 * tr (spectralProjector X p.1.1 * spectralProjector Y p.2.1)

/-- Corollary 2.5, p.6: linear isometries, or anti-linear isometries via transpose. -/
def isometricConjugation (anti : Bool) (W : Matrix J I ℂ) (X : Mat I) : Mat J :=
  W * (if anti then Xᵀ else X) * Wᴴ

/-- Section 5, p.21: ten conditions, computed from the input pair. The Boolean
selects Remark 5.7's co-variant, changing precisely (i),(ii),(vi)–(viii). -/
def pairConditions (co : Bool) (A : Alg I) (B : Alg J) (Phi : MatrixMap I J)
    (X Y : Mat I) : Fin 10 → Prop :=
  let stochastic := if co then CoStochastic else AlgebraStochastic
  let substochastic := if co then CoSubstochastic else Substochastic
  let cocycle := fun z : ℂ =>
    supp Y * channelAdjoint Phi (if co then cpow (Phi X) z * cpow (Phi Y) (-z)
      else cpow (Phi Y) (-z) * cpow (Phi X) z) = cpow Y (-z) * cpow X z
  ![∃ Psi : MatrixMap J I, stochastic B A Psi ∧ Psi (Phi X) = X ∧ Psi (Phi Y) = Y,
    ∃ Psi : MatrixMap J I, substochastic B A Psi ∧ Psi (Phi X) = X ∧ Psi (Phi Y) = Y,
    ∀ f, DivergenceGenerator f → SemidefiniteOperatorConvex f → div f (Phi X) (Phi Y) = div f X Y,
    ∃ f, DivergenceGenerator f ∧ SemidefiniteOperatorConvex f ∧
      Rich f ((modularSpec A X Y ∪ modularSpec B (Phi X) (Phi Y)).card) ∧
      div f (Phi X) (Phi Y) = div f X Y,
    ∃ T : Set ℝ, T ⊆ Ioi 0 ∧
      (((modularSpec A X Y ∪ modularSpec B (Phi X) (Phi Y)).card : ℕ) : Cardinal) ≤
        Cardinal.mk T ∧ ∀ t ∈ T, div (resolventGenerator t) (Phi X) (Phi Y) = div (resolventGenerator t) X Y,
    ∀ z : ℂ, cocycle z,
    ∃ a : ℝ, 0 < a ∧ a < 2 ∧ a ≠ 1 ∧ cocycle (a : ℂ),
    ∀ t : ℝ, cocycle (Complex.I * (t : ℂ)),
    supp Y * channelAdjoint Phi (cfc Real.log (Phi X) - cfc Real.log (Phi Y) * supp (Phi X)) =
      cfc Real.log X - cfc Real.log Y * supp X,
    petz Y Phi (Phi X) = X]

/-- Theorem 5.1, pp.21–25: asymmetric implication chain, never a blanket equivalence. -/
def PairChain (c : Fin 10 → Prop) : Prop :=
  (c 0 → c 1) ∧ (c 1 → c 2) ∧ (c 2 → c 3) ∧ (c 3 ↔ c 4) ∧
  (c 4 ↔ c 5) ∧ (c 5 ↔ c 6) ∧ (c 6 ↔ c 7) ∧ (c 7 ↔ c 8) ∧ (c 8 → c 9)

/-- Sections 5 and 7: preserve the positivity level when a stochastic recovery exists. -/
def RecoveryRefinement (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) (C : Set (Mat I)) : Prop :=
  (∀ n, AlgebraKPositive A Phi n → ∃ Psi : MatrixMap J I,
    AlgebraStochastic B A Psi ∧ AlgebraKPositive B Psi n ∧ ∀ rho ∈ C, Psi (Phi rho) = rho) ∧
  (AlgebraCompletelyPositive A Phi → ∃ Psi : MatrixMap J I,
    AlgebraStochastic B A Psi ∧ AlgebraCompletelyPositive B Psi ∧ ∀ rho ∈ C, Psi (Phi rho) = rho)

/-- Section 7, p.33: CLOSED convex hull in the finite-dimensional matrix topology. -/
def hull (C : Set (Mat I)) : Set (Mat I) := closure (convexHull ℝ C)

/-- Section 7, p.33: maximal support without choosing an unrelated projection. -/
def Covers (C : Set (Mat I)) (sigma : Mat I) : Prop := ∀ rho ∈ C, Supports rho sigma

/-- Section 7, p.33: equality of a reference support with the join of code supports. -/
def CodeSupportEquals (C : Set (Mat I)) (sigma : Mat I) : Prop :=
  ∀ v : I → ℂ, sigma *ᵥ v = 0 ↔ ∀ rho ∈ C, rho *ᵥ v = 0

/-- Section 7, p.34: shared direct-sum/tensor data. Algebraic action and cone equality
are deliberately NOT fields; they are conclusions in the theorems below. -/
structure RecoveryBlocks (I J : Type) [Fintype I] [DecidableEq I] [Fintype J] [DecidableEq J] where
  count : ℕ
  left1 : Fin count → ℕ
  right1 : Fin count → ℕ
  left2 : Fin count → ℕ
  right2 : Fin count → ℕ
  left1_pos : ∀ k, 0 < left1 k
  right1_pos : ∀ k, 0 < right1 k
  left2_pos : ∀ k, 0 < left2 k
  right2_pos : ∀ k, 0 < right2 k
  embed1 : ∀ k, Matrix I (Fin (left1 k) × Fin (right1 k)) ℂ
  embed2 : ∀ k, Matrix J (Fin (left2 k) × Fin (right2 k)) ℂ
  isometry1 : ∀ k, (embed1 k)ᴴ * embed1 k = 1
  isometry2 : ∀ k, (embed2 k)ᴴ * embed2 k = 1
  orthogonal1 : ∀ k l, k ≠ l → (embed1 k)ᴴ * embed1 l = 0
  orthogonal2 : ∀ k l, k ≠ l → (embed2 k)ᴴ * embed2 l = 0
  noise1 : ∀ k, Mat (Fin (right1 k))
  noise2 : ∀ k, Mat (Fin (right2 k))
  noise1_pos : ∀ k, (noise1 k).PosDef
  noise2_pos : ∀ k, (noise2 k).PosDef
  noise1_trace : ∀ k, tr (noise1 k) = 1
  noise2_trace : ∀ k, tr (noise2 k) = 1
  unitary : ∀ k, Matrix (Fin (left2 k)) (Fin (left1 k)) ℂ
  unitary_left : ∀ k, (unitary k)ᴴ * unitary k = 1
  unitary_right : ∀ k, unitary k * (unitary k)ᴴ = 1

/-- Shared block embedding on the input side, equation (3.8), p.15. -/
def inputBlock (D : RecoveryBlocks I J) (k : Fin D.count) (X : Mat (Fin (D.left1 k))) : Mat I :=
  D.embed1 k * Matrix.kronecker X (D.noise1 k) * (D.embed1 k)ᴴ

/-- Shared block action on the output side, equation (3.8), p.15. -/
def outputBlock (D : RecoveryBlocks I J) (k : Fin D.count) (X : Mat (Fin (D.left1 k))) : Mat J :=
  D.embed2 k * Matrix.kronecker (D.unitary k * X * (D.unitary k)ᴴ) (D.noise2 k) * (D.embed2 k)ᴴ

/-- Section 7, p.34: support joins are exactly the direct sum of all the shared blocks. -/
def BlockSupports (D : RecoveryBlocks I J) (C : Set (Mat I)) (E : Set (Mat J)) : Prop :=
  CodeSupportEquals C (∑ k, D.embed1 k * (D.embed1 k)ᴴ) ∧
  CodeSupportEquals E (∑ k, D.embed2 k * (D.embed2 k)ᴴ)

/-- Section 7, p.34: a code state has variable left states and probabilities,
while the noise states, embeddings, and unitaries are shared across the whole code. -/
def BlockState (D : RecoveryBlocks I J) (rho : Mat I) : Prop :=
  ∃ (p : Fin D.count → ℝ) (X : ∀ k, Mat (Fin (D.left1 k))),
    IsDistribution p ∧ (∀ k, (X k).PosSemidef ∧ tr (X k) = 1) ∧
    rho = ∑ k, (p k : ℂ) • inputBlock D k (X k)

/-- Equation (6.8), p.29: optimal Bayesian error normalized by min(p,1-p). -/
def bayesianTestingDistance (p : ℝ) (rho sigma : Mat I) : EReal :=
  -logNonnegativeExtended ((1 - QAlgorithms.traceNorm ((p : ℂ) • rho - ((1 - p : ℝ) : ℂ) • sigma)) /
    (2 * min p (1 - p)))

/-- Theorem 7.1, pp.33–34: its ten uniform criteria. -/
def codeConditions (A : Alg I) (B : Alg J) (Phi : MatrixMap I J) (C : Set (Mat I)) : Fin 10 → Prop :=
  let d := Fintype.card I ^ 2 + Fintype.card J ^ 2
  ![∃ Psi : MatrixMap J I, AlgebraStochastic B A Psi ∧ ∀ rho ∈ hull C, Psi (Phi rho) = rho,
    ∀ f, DivergenceGenerator f → SemidefiniteOperatorConvex f → ∀ rho ∈ hull C, ∀ sigma ∈ hull C,
      div f (Phi rho) (Phi sigma) = div f rho sigma,
    ∃ sigma f, AlgebraDensity A sigma ∧ Covers C sigma ∧ DivergenceGenerator f ∧ SemidefiniteOperatorConvex f ∧ Rich f d ∧
      ∀ rho ∈ C, div f (Phi rho) (Phi sigma) = div f rho sigma,
    ∃ sigma, AlgebraDensity A sigma ∧ Covers C sigma ∧ ∃ T : Set ℝ, T ⊆ Ioi 0 ∧
      (d : Cardinal) ≤ Cardinal.mk T ∧ ∀ rho ∈ C, ∀ t ∈ T,
        div (resolventGenerator t) (Phi rho) (Phi sigma) = div (resolventGenerator t) rho sigma,
    ∀ rho ∈ hull C, ∀ sigma ∈ hull C, ∀ r : ℝ,
      hoeffding (Phi rho) (Phi sigma) r = hoeffding rho sigma r,
    ∃ sigma, AlgebraDensity A sigma ∧ Covers C sigma ∧ ∃ delta : ℝ, 0 < delta ∧
      ∀ rho ∈ C, ∀ r ∈ Ioo 0 delta, hoeffding (Phi rho) (Phi sigma) r = hoeffding rho sigma r,
    ∀ rho ∈ hull C, ∀ sigma ∈ hull C, CodeSupportEquals C sigma → petz sigma Phi (Phi rho) = rho,
    ∃ sigma, AlgebraDensity A sigma ∧ ∀ rho ∈ C, petz sigma Phi (Phi rho) = rho,
    ∃ D : RecoveryBlocks I J, BlockSupports D C (Phi '' C) ∧ (∀ rho ∈ C, BlockState D rho) ∧
      ∀ k X, inputBlock D k X ∈ A ∧ Phi (inputBlock D k X) = outputBlock D k X,
    ∃ sigma, AlgebraDensity A sigma ∧ Covers C sigma ∧ ∀ rho ∈ C, ∀ p ∈ Ioo (0 : ℝ) 1, ∀ n : ℕ,
      bayesianTestingDistance p ((QAlgorithms.piSuperop (fun _ : Fin n => (Phi : Mat I → Mat J))) (copies n rho)) ((QAlgorithms.piSuperop (fun _ : Fin n => (Phi : Mat I → Mat J))) (copies n sigma)) =
        bayesianTestingDistance p (copies n rho) (copies n sigma)]

/-- Section 6, p.30: monotone functions are strictly increasing on the ATTAINED
extended divergence range, rather than on an arbitrary real domain. -/
def MonotoneRepresentation (A : Alg I) (D : Mat I → Mat I → EReal) (f : ℝ → ℝ) : Prop :=
  let R := {v : EReal | ∃ rho sigma, AlgebraDensity A rho ∧ AlgebraDensity A sigma ∧ v = div f rho sigma}
  ∃ g : R → EReal, StrictMono g ∧ ∀ rho sigma (hr : AlgebraDensity A rho) (hs : AlgebraDensity A sigma),
    D rho sigma = g ⟨div f rho sigma, ⟨rho, sigma, hr, hs, rfl⟩⟩

/-- The zero-copy product is the scalar identity. -/
example (X : Mat I) (a b : Fin 0 → I) : copies 0 X a b = 1 := by simp [copies]

/-- Resolvent kernels do not divide by zero on their intended parameter domain. -/
example (t : PosReal) : representationKernel 0 t = 0 := by simp [representationKernel]

/-- Example 3.6, p.12: transpose mixed with complete depolarization. -/
def transposeNoise (e : ℝ) (X : Mat I) : Mat I :=
  ((1 - e : ℝ) : ℂ) • Xᵀ + ((e : ℂ) * X.trace / (Fintype.card I : ℂ)) • 1

/-- Example 3.6, pp.12–13: identity mixed with complete depolarization. -/
def identityNoise (e : ℝ) (X : Mat I) : Mat I :=
  ((1 - e : ℝ) : ℂ) • X + ((e : ℂ) * X.trace / (Fintype.card I : ℂ)) • 1

/-- Example 3.6, p.13: transpose mixed with identity. -/
def transposeMix (e : ℝ) (X : Mat I) : Mat I :=
  ((1 - e : ℝ) : ℂ) • Xᵀ + (e : ℂ) • X

/-- The source's identity endpoint. -/
example (X : Mat I) : transposeMix 1 X = X := by simp [transposeMix]

/-- Example 3.6, p.12: the full six-term Schwarz defect expansion. -/
def transposeDefect (e c : ℝ) (X : Mat I) : Mat I :=
  ((c * (1 - e) : ℝ) : ℂ) • (Xᴴ * X)ᵀ +
  (((c * e : ℝ) : ℂ) * (Xᴴ * X).trace / (Fintype.card I : ℂ)) • 1 -
  (((1 - e) ^ 2 : ℝ) : ℂ) • ((Xᴴ)ᵀ * Xᵀ) -
  (((e * (1 - e) : ℝ) : ℂ) * X.trace / (Fintype.card I : ℂ)) • (Xᴴ)ᵀ -
  (((e * (1 - e) : ℝ) : ℂ) * (Xᴴ).trace / (Fintype.card I : ℂ)) • Xᵀ -
  (((e ^ 2 : ℝ) : ℂ) * (Complex.normSq X.trace : ℂ) / (Fintype.card I : ℂ) ^ 2) • 1

/-- Remark 4.8, pp.20–21: maps monotone for a specified f. -/
def MonotoneMap (A : Alg I) (f : ℝ → ℝ) (Phi : Mat I → Mat J) : Prop :=
  PositiveOnConcreteAlgebra A Phi ∧ ∀ X ∈ A, ∀ Y ∈ A, X.PosSemidef → Y.PosSemidef →
    div f (Phi X) (Phi Y) ≤ div f X Y

/-- Remark 6.2, p.28: limiting order-one entropy, before identifying it on states. -/
def renyiOne (X Y : Mat I) : EReal := limUnder (𝓝[<] (1 : ℝ)) (fun a => petzRenyi a X Y)

/-- Example 6.6, p.31: the stated diagonal state. -/
def exampleRho : Mat (Fin 3) := Matrix.diagonal ![(2 / 3 : ℂ), 1 / 3, 0]

/-- Example 6.6, p.31: the stated reference state. -/
def exampleSigma : Mat (Fin 3) := Matrix.diagonal ![(1 / 6 : ℂ), 1 / 3, 1 / 2]

/-- Example 6.6, p.31: canonical extension of the diagonal block-merging map. -/
def merge : MatrixMap (Fin 3) (Fin 3) where
  toFun X := Matrix.diagonal (fun i => if i = 0 then X 0 0 + X 1 1 else if i = 2 then X 2 2 else 0)
  map_add' X Y := by
    ext i j
    simp only [Matrix.diagonal_apply, Matrix.add_apply]
    split_ifs <;> simp_all <;> ring
  map_smul' c X := by
    ext i j
    simp only [Matrix.diagonal_apply, Matrix.smul_apply, smul_eq_mul]
    split_ifs <;> simp_all <;> ring

/-- Section 6, p.31: the concrete diagonal observable algebra. -/
def diagonalAlgebra (I : Type) [Fintype I] [DecidableEq I] : Alg I where
  carrier := {X | ∀ i j, i ≠ j → X i j = 0}
  zero_mem' := by simp
  one_mem' := by intro i j h; simp [Matrix.one_apply, h]
  add_mem' := by intro X Y hX hY i j h; simp [hX i j h, hY i j h]
  mul_mem' := by
    intro X Y hX hY i j h
    apply Finset.sum_eq_zero
    intro k hk
    by_cases hik : i = k
    · subst k; simp [hY i j h]
    · simp [hX i k hik]
  algebraMap_mem' := by intro c i j h; simp [Algebra.algebraMap_eq_smul_one, Matrix.one_apply, h]
  star_mem' := by intro X hX i j h; simpa using congrArg conj (hX j i (Ne.symm h))

/-- The merge map preserves the last basis projection exactly. -/
example : merge (Matrix.single 2 2 1) = (Matrix.single 2 2 1 : Mat (Fin 3)) := by
  change Matrix.diagonal (fun k : Fin 3 =>
    if k = 0 then
      (Matrix.single 2 2 (1 : ℂ) : Mat (Fin 3)) 0 0 +
        (Matrix.single 2 2 (1 : ℂ) : Mat (Fin 3)) 1 1
    else if k = 2 then (Matrix.single 2 2 (1 : ℂ) : Mat (Fin 3)) 2 2 else 0) =
      (Matrix.single 2 2 1 : Mat (Fin 3))
  ext i j
  by_cases hi : i = 2
  · subst i
    simp only [Matrix.diagonal_apply, Matrix.single_apply,
      show (2 : Fin 3) ≠ 0 from by decide, if_false, and_self,
      true_and, eq_self_iff_true, if_true]
  · by_cases hij : i = j
    · subst j
      simp only [Matrix.diagonal_apply, Matrix.single_apply,
        show (2 : Fin 3) ≠ 0 from by decide,
        show (2 : Fin 3) ≠ 1 from by decide,
        hi, Ne.symm hi, eq_self_iff_true, and_self,
        if_false, if_true, zero_add, ite_self]
    · have hsingle : ¬ (2 = i ∧ 2 = j) := by
        intro h
        exact hij (h.1.symm.trans h.2)
      simp only [Matrix.diagonal_apply, Matrix.single_apply, hij, hsingle, if_false]

/-- Support inclusion includes the zero and reflexive edge cases. -/
example (X : Mat I) : Supports X X := by intro v hv; exact hv

/-- The support-zero convention applies to the zero matrix even at exponent zero. -/
example : supp (0 : Mat I) = 0 := by simp [supp, supportPower, cfc_zero]

/-- The source identity channel has no hidden trace normalization. -/
example : AlgebraTracePreserving (⊤ : Alg I) (LinearMap.id : MatrixMap I I) := by intro X hX; rfl

/-- The actual amplification of identity acts on every block entry. -/
example (n : ℕ) (X : Mat (Fin n × I)) : amp (fun Y : Mat I => Y) n X = X := rfl

/-- The combined representing kernel vanishes at x=1, pinning a=f(1)-f(0)-b. -/
example (t : PosReal) : representationKernel 1 t = 0 := by simp [representationKernel]

end QAlgorithms.Papers.Independent.Support50.P12
Source
Quantum f-divergences and error correction, arXiv:1008.2529v6, https://arxiv.org/abs/1008.2529v6
Read-back

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

QAlgorithms.Papers.Independent.Support50.P12.Alg

For any finite type III with decidable equality, Alg⁡(I)\operatorname{Alg}(I)Alg(I) abbreviates the type of complex star subalgebras of the algebra of complex matrices indexed by I×II\times II×I. Such a subalgebra contains the identity and is closed under addition, multiplication, complex scalar multiplication, and conjugate transpose. The index type III may be empty.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraCompletelyPositive

For finite index types I,JI,JI,J, with decidable equality on III, a complex star subalgebra A\mathcal AA of the complex I×II\times II×I matrices, and an arbitrary map Φ\PhiΦ from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, this property means that, for every natural number nnn, every positive semidefinite block matrix X=(Xab)0≤a,b<nX=(X_{ab})_{0\le a,b<n}X=(Xab​)0≤a,b<n​ whose blocks lie in A\mathcal AA yields a positive semidefinite block matrix (Φ(Xab))0≤a,b<n(\Phi(X_{ab}))_{0\le a,b<n}(Φ(Xab​))0≤a,b<n​. This includes n=0n=0n=0, where the condition is automatic; linearity of Φ\PhiΦ is not part of the definition.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraDensity

For any finite type III with decidable equality, any complex star subalgebra A\mathcal AA of the complex I×II\times II×I matrices, and any such matrix ρ\rhoρ, AlgebraDensity⁡(A,ρ)\operatorname{AlgebraDensity}(\mathcal A,\rho)AlgebraDensity(A,ρ) means that ρ∈A\rho\in\mathcal Aρ∈A, ρ\rhoρ is positive semidefinite, and Re⁡Tr⁡ρ=1\operatorname{Re}\operatorname{Tr}\rho=1ReTrρ=1. If III is empty, this condition cannot hold because the trace is zero.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraExtension

For finite index types I,JI,JI,J with decidable equality, unital complex star subalgebras A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), B⊆MJ(C)B\subseteq M_J(\mathbb C)B⊆MJ​(C), and a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property means that Φ(X)∈B\Phi(X)\in BΦ(X)∈B for every matrix X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C), and Φ†(Y)∈A\Phi^\dagger(Y)\in AΦ†(Y)∈A for every matrix Y∈MJ(C)Y\in M_J(\mathbb C)Y∈MJ​(C), where, writing EijE_{ij}Eij​ for the standard matrix unit,

(Φ†(Y))ij=∑k,l∈J(Φ(Eij))kl‾ Ykl.\bigl(\Phi^\dagger(Y)\bigr)_{ij} =\sum_{k,l\in J}\overline{\bigl(\Phi(E_{ij})\bigr)_{kl}}\,Y_{kl}.(Φ†(Y))ij​=k,l∈J∑​(Φ(Eij​))kl​​Ykl​.

Both conditions quantify over the entire matrix spaces, including matrices outside the specified subalgebras. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraKPositive

For finite index types I,JI,JI,J, with decidable equality on III, a complex star subalgebra A\mathcal AA of the complex I×II\times II×I matrices, an arbitrary map Φ\PhiΦ from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, and a natural number nnn, this property means that every positive semidefinite matrix XXX indexed by {0,…,n−1}×I\{0,\ldots,n-1\}\times I{0,…,n−1}×I, all of whose blocks Xab=(X(a,i),(b,j))i,j∈IX_{ab}=(X_{(a,i),(b,j)})_{i,j\in I}Xab​=(X(a,i),(b,j)​)i,j∈I​ belong to A\mathcal AA, produces a positive semidefinite matrix with entries [Φ(Xab)]u,v[\Phi(X_{ab})]_{u,v}[Φ(Xab​)]u,v​ at ((a,u),(b,v))((a,u),(b,v))((a,u),(b,v)). No linearity of Φ\PhiΦ is assumed. The value n=0n=0n=0 is included, and its condition is automatically satisfied.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraSchwarzMap

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property means that there exists a real c≥0c\ge0c≥0 such that Φ\PhiΦ sends every positive semidefinite matrix in AAA to a positive semidefinite matrix and

c Φ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A.c\,\Phi(X^*X)-\Phi(X)^*\Phi(X) \quad\text{is positive semidefinite for every }X\in A.cΦ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A.

The constant need not be unique or strictly positive, and Φ\PhiΦ need not be linear. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraStochastic

For finite index types I,JI,JI,J with decidable equality, unital complex star subalgebras A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), B⊆MJ(C)B\subseteq M_J(\mathbb C)B⊆MJ​(C), and a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property requires that the entire range of Φ\PhiΦ lie in BBB, that the entire range of the map

(Φ†(Y))ij=∑k,l∈J(Φ(Eij))kl‾ Ykl\bigl(\Phi^\dagger(Y)\bigr)_{ij} =\sum_{k,l\in J}\overline{\bigl(\Phi(E_{ij})\bigr)_{kl}}\,Y_{kl}(Φ†(Y))ij​=k,l∈J∑​(Φ(Eij​))kl​​Ykl​

lie in AAA, that Φ†\Phi^\daggerΦ† send positive semidefinite matrices in BBB to positive semidefinite matrices, and that Φ†(Y∗Y)−Φ†(Y)∗Φ†(Y)\Phi^\dagger(Y^*Y)-\Phi^\dagger(Y)^*\Phi^\dagger(Y)Φ†(Y∗Y)−Φ†(Y)∗Φ†(Y) be positive semidefinite for every Y∈BY\in BY∈B. Additionally, Tr⁡Φ(X)=Tr⁡X\operatorname{Tr}\Phi(X)=\operatorname{Tr}XTrΦ(X)=TrX as complex numbers for every X∈AX\in AX∈A. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraTraceNonincreasing

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property means that

Re⁡Tr⁡Φ(X)≤Re⁡Tr⁡X\operatorname{Re}\operatorname{Tr}\Phi(X) \le \operatorname{Re}\operatorname{Tr}XReTrΦ(X)≤ReTrX

for every positive semidefinite X∈AX\in AX∈A. The definition does not require Φ(X)\Phi(X)Φ(X) to be positive semidefinite or Φ\PhiΦ to be linear; only the real parts of the traces are compared. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.AlgebraTracePreserving

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property means that

Tr⁡Φ(X)=Tr⁡Xfor every X∈A.\operatorname{Tr}\Phi(X)=\operatorname{Tr}X \qquad\text{for every }X\in A.TrΦ(X)=TrXfor every X∈A.

This is equality of complex traces for all matrices in AAA, without restricting them to positive or Hermitian matrices and without assuming that Φ\PhiΦ is linear. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.BlockState

For finite types I,JI,JI,J with decidable equality, let DDD consist of r≥0r\ge0r≥0 blocks with positive dimensions ℓ1,k,r1,k,ℓ2,k,r2,k\ell_{1,k},r_{1,k},\ell_{2,k},r_{2,k}ℓ1,k​,r1,k​,ℓ2,k​,r2,k​, mutually orthogonal isometric embeddings E1,kE_{1,k}E1,k​ and E2,kE_{2,k}E2,k​ into CI\mathbb C^ICI and CJ\mathbb C^JCJ, positive definite right-factor matrices τ1,k,τ2,k\tau_{1,k},\tau_{2,k}τ1,k​,τ2,k​ of real trace 111, and matrices UkU_kUk​ between the left factors satisfying both unitary identities. For an arbitrary complex I×II\times II×I matrix ρ\rhoρ, this property means that there exist real numbers pkp_kpk​ and complex ℓ1,k×ℓ1,k\ell_{1,k}\times\ell_{1,k}ℓ1,k​×ℓ1,k​ matrices XkX_kXk​ such that pk≥0p_k\ge0pk​≥0 for every k<rk<rk<r, ∑k<rpk=1\sum_{k<r}p_k=1∑k<r​pk​=1, every XkX_kXk​ is positive semidefinite with Re⁡Tr⁡Xk=1\operatorname{Re}\operatorname{Tr}X_k=1ReTrXk​=1, and

ρ=∑k<rpk E1,k(Xk⊗τ1,k)E1,k†.\rho=\sum_{k<r}p_k\,E_{1,k}(X_k\otimes\tau_{1,k})E_{1,k}^\dagger.ρ=k<r∑​pk​E1,k​(Xk​⊗τ1,k​)E1,k†​.

The requirements on XkX_kXk​ also apply when pk=0p_k=0pk​=0. If r=0r=0r=0, the required probability weights would have empty sum 111, so this property cannot hold.

QAlgorithms.Papers.Independent.Support50.P12.BlockSupports

For finite types I,JI,JI,J with decidable equality, let DDD consist of r≥0r\ge0r≥0 blocks with positive left and right dimensions, mutually orthogonal isometric embeddings E1,kE_{1,k}E1,k​ into CI\mathbb C^ICI and E2,kE_{2,k}E2,k​ into CJ\mathbb C^JCJ, positive definite right-factor matrices of real trace 111, and unitary identifications between the corresponding left factors. For arbitrary sets C⊆MI(C)C\subseteq M_I(\mathbb C)C⊆MI​(C) and E⊆MJ(C)E\subseteq M_J(\mathbb C)E⊆MJ​(C), the defined property is the conjunction

(∑k<rE1,kE1,k†)v=0  ⟺  ∀ρ∈C, ρv=0for every v∈CI,(∑k<rE2,kE2,k†)w=0  ⟺  ∀η∈E, ηw=0for every w∈CJ.\begin{aligned} \left(\sum_{k<r}E_{1,k}E_{1,k}^\dagger\right)v=0 &\iff \forall\rho\in C,\ \rho v=0 &&\text{for every }v\in\mathbb C^I,\\ \left(\sum_{k<r}E_{2,k}E_{2,k}^\dagger\right)w=0 &\iff \forall\eta\in E,\ \eta w=0 &&\text{for every }w\in\mathbb C^J. \end{aligned}(k<r∑​E1,k​E1,k†​)v=0(k<r∑​E2,k​E2,k†​)w=0​⟺∀ρ∈C, ρv=0⟺∀η∈E, ηw=0​​for every v∈CI,for every w∈CJ.​

There are no positivity or normalization assumptions on members of either set. An empty set on either side makes its universal annihilation condition vacuous. If r=0r=0r=0, both sums are zero, and the property requires every matrix in CCC and every matrix in EEE to be zero.

QAlgorithms.Papers.Independent.Support50.P12.CoBound

For finite index types I,JI,JI,J, with decidable equality on III, a complex star subalgebra A\mathcal AA of the complex I×II\times II×I matrices, an arbitrary map Φ\PhiΦ to complex J×JJ\times JJ×J matrices, and a real number ccc, this property means that c≥0c\ge0c≥0, that Φ(X)\Phi(X)Φ(X) is positive semidefinite whenever X∈AX\in\mathcal AX∈A is positive semidefinite, and that, for every X∈AX\in\mathcal AX∈A,

c Φ(XX∗)−Φ(X)∗Φ(X)is positive semidefinite.c\,\Phi(XX^*)-\Phi(X)^*\Phi(X) \quad\text{is positive semidefinite}.cΦ(XX∗)−Φ(X)∗Φ(X)is positive semidefinite.

Here X∗X^*X∗ is the conjugate transpose. The inequality applies to all X∈AX\in\mathcal AX∈A, without requiring XXX to be Hermitian or positive semidefinite, and c=0c=0c=0 is allowed.

QAlgorithms.Papers.Independent.Support50.P12.CoStochastic

For finite index types I,JI,JI,J with decidable equality, complex star subalgebras A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C) and B⊆MJ(C)\mathcal B\subseteq M_J(\mathbb C)B⊆MJ​(C), and a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), let

H(Y)i,j=∑k,l∈JΦ(Eij)k,l‾ Yk,l,H(Y)_{i,j}=\sum_{k,l\in J}\overline{\Phi(E_{ij})_{k,l}}\,Y_{k,l},H(Y)i,j​=k,l∈J∑​Φ(Eij​)k,l​​Yk,l​,

where EijE_{ij}Eij​ is the standard matrix unit. This property means that Φ(X)∈B\Phi(X)\in\mathcal BΦ(X)∈B for every input matrix XXX, that H(Y)∈AH(Y)\in\mathcal AH(Y)∈A for every matrix YYY, that HHH sends positive semidefinite elements of B\mathcal BB to positive semidefinite matrices, that H(YY∗)−H(Y)∗H(Y)H(YY^*)-H(Y)^*H(Y)H(YY∗)−H(Y)∗H(Y) is positive semidefinite for every Y∈BY\in\mathcal BY∈B, and additionally that tr⁡Φ(X)=tr⁡X\operatorname{tr}\Phi(X)=\operatorname{tr}XtrΦ(X)=trX for every X∈AX\in\mathcal AX∈A.

QAlgorithms.Papers.Independent.Support50.P12.CoSubstochastic

For finite index types I,JI,JI,J with decidable equality, complex star subalgebras A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C) and B⊆MJ(C)\mathcal B\subseteq M_J(\mathbb C)B⊆MJ​(C), and a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), define

H(Y)i,j=∑k,l∈JΦ(Eij)k,l‾ Yk,l,H(Y)_{i,j}=\sum_{k,l\in J}\overline{\Phi(E_{ij})_{k,l}}\,Y_{k,l},H(Y)i,j​=k,l∈J∑​Φ(Eij​)k,l​​Yk,l​,

where EijE_{ij}Eij​ has entry 111 at (i,j)(i,j)(i,j) and zero elsewhere. This property means that Φ(X)∈B\Phi(X)\in\mathcal BΦ(X)∈B for every X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C), that H(Y)∈AH(Y)\in\mathcal AH(Y)∈A for every Y∈MJ(C)Y\in M_J(\mathbb C)Y∈MJ​(C), that H(Y)H(Y)H(Y) is positive semidefinite for every positive semidefinite Y∈BY\in\mathcal BY∈B, and that H(YY∗)−H(Y)∗H(Y)H(YY^*)-H(Y)^*H(Y)H(YY∗)−H(Y)∗H(Y) is positive semidefinite for every Y∈BY\in\mathcal BY∈B.

QAlgorithms.Papers.Independent.Support50.P12.CodeSupportEquals

For a finite type III with decidable equality, a set CCC of complex I×II\times II×I matrices, and a complex I×II\times II×I matrix σ\sigmaσ, this property means that, for every vector v∈CIv\in\mathbb C^Iv∈CI, σv=0\sigma v=0σv=0 if and only if ρv=0\rho v=0ρv=0 for every ρ∈C\rho\in Cρ∈C; equivalently, ker⁡σ=⋂ρ∈Cker⁡ρ\ker\sigma=\bigcap_{\rho\in C}\ker\rhokerσ=⋂ρ∈C​kerρ. No positivity or normalization is required. If CCC is empty, the right-hand condition holds for every vector, so the property requires σ=0\sigma=0σ=0; empty III is also allowed.

QAlgorithms.Papers.Independent.Support50.P12.CornerTP

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and any matrix P∈MI(C)P\in M_I(\mathbb C)P∈MI​(C), this property means that

Tr⁡Φ(PXP)=Tr⁡(PXP)for every X∈A.\operatorname{Tr}\Phi(PXP)=\operatorname{Tr}(PXP) \qquad\text{for every }X\in A.TrΦ(PXP)=Tr(PXP)for every X∈A.

The traces are complex traces. There is no assumption that PPP is a projection, Hermitian, or a member of AAA, and no linearity assumption on Φ\PhiΦ. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.Covers

For a finite index type III, any set C⊆MI(C)C\subseteq M_I(\mathbb C)C⊆MI​(C), and any σ∈MI(C)\sigma\in M_I(\mathbb C)σ∈MI​(C), this predicate means that

∀ρ∈C, ∀v:I→C,σv=0 ⟹ ρv=0.\forall\rho\in C,\ \forall v:I\to\mathbb C,\qquad \sigma v=0\ \Longrightarrow\ \rho v=0.∀ρ∈C, ∀v:I→C,σv=0 ⟹ ρv=0.

Equivalently, ker⁡σ⊆ker⁡ρ\ker\sigma\subseteq\ker\rhokerσ⊆kerρ for every ρ∈C\rho\in Cρ∈C. No Hermiticity, positivity, or normalization is required. It holds vacuously when CCC is empty, and for σ=0\sigma=0σ=0 it requires every member of CCC to be the zero matrix.

QAlgorithms.Papers.Independent.Support50.P12.DivergenceGenerator

For every function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, DivergenceGenerator⁡(f)\operatorname{DivergenceGenerator}(f)DivergenceGenerator(f) means that fff is continuous on (0,∞)(0,\infty)(0,∞) and that f(x)/xf(x)/xf(x)/x, viewed in the extended real numbers, converges as x→+∞x\to+\inftyx→+∞ to the value selected by the total limit operation for that same function. Thus an extended-real limit is required, and it may be infinite. No condition is imposed on f(0)f(0)f(0), on its values at negative arguments, or on convexity.

QAlgorithms.Papers.Independent.Support50.P12.HalfLineRepresentation

For a function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, real numbers a,ba,ba,b, and a measure μ\muμ on the positive real numbers, this property means that b≥0b\ge0b≥0, that t↦(1+t)−2t\mapsto(1+t)^{-2}t↦(1+t)−2 is integrable with respect to μ\muμ, and that, for every real x≥0x\ge0x≥0, the function t↦x/(1+t)−x/(x+t)t\mapsto x/(1+t)-x/(x+t)t↦x/(1+t)−x/(x+t) is integrable with respect to μ\muμ and

f(x)=f(0)+ax+bx2+∫(0,∞)(x1+t−xx+t) dμ(t).f(x)=f(0)+ax+bx^2+ \int_{(0,\infty)} \left(\frac{x}{1+t}-\frac{x}{x+t}\right)\,d\mu(t).f(x)=f(0)+ax+bx2+∫(0,∞)​(1+tx​−x+tx​)dμ(t).

The coefficient aaa is unrestricted, and μ\muμ need not have finite total mass. The case x=0x=0x=0 is included; its kernel vanishes everywhere.

QAlgorithms.Papers.Independent.Support50.P12.KCoPositive

For finite index types I,JI,JI,J, with decidable equality on III, a complex star subalgebra A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C), an arbitrary map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and a natural number nnn, this property means that every positive semidefinite block matrix X=(Xab)0≤a,b<nX=(X_{ab})_{0\le a,b<n}X=(Xab​)0≤a,b<n​ satisfying XabT∈AX_{ab}^{\mathsf T}\in\mathcal AXabT​∈A for every a,ba,ba,b yields a positive semidefinite block matrix (Φ(XabT))0≤a,b<n(\Phi(X_{ab}^{\mathsf T}))_{0\le a,b<n}(Φ(XabT​))0≤a,b<n​. The transpose here does not conjugate entries. No linearity of Φ\PhiΦ is assumed, and n=0n=0n=0 is included with an automatically satisfied condition.

QAlgorithms.Papers.Independent.Support50.P12.MonotoneMap

Let I,JI,JI,J be finite index types with decidable equality, possibly empty, let A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C) be a unital complex star subalgebra, let f:R→Rf:\mathbb R\to\mathbb Rf:R→R be arbitrary, and let Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C) be an arbitrary function, without a linearity assumption. Define sp⁡(Z)\operatorname{sp}(Z)sp(Z) to be its distinct eigenvalues if ZZZ is Hermitian and the empty set otherwise, let Pλ(Z)P_\lambda(Z)Pλ​(Z) be the orthogonal projection onto its λ\lambdaλ-eigenspace, and define W01/2W_0^{1/2}W01/2​ by spectral functional calculus with value zero at zero. Put

Qf(Z,W)=Re⁡Tr⁡ ⁣[(W01/2)∗∑λ∈sp⁡(Z)∑μ∈sp⁡(W)f(λ/μ)Pλ(Z)W01/2Pμ(W)],Df(Z,W)={Qf(Z,W),ker⁡W⊆ker⁡Z,lim⁡e↓0selectedQf(Z,W+e1),otherwise,Q_f(Z,W)=\operatorname{Re}\operatorname{Tr}\!\left[(W_0^{1/2})^* \sum_{\lambda\in\operatorname{sp}(Z)}\sum_{\mu\in\operatorname{sp}(W)} f(\lambda/\mu)P_\lambda(Z)W_0^{1/2}P_\mu(W)\right], \quad D_f(Z,W)= \begin{cases} Q_f(Z,W),&\ker W\subseteq\ker Z,\\ \displaystyle\operatorname{lim}_{e\downarrow0}^{\mathrm{selected}}Q_f(Z,W+e1),&\text{otherwise}, \end{cases}Qf​(Z,W)=ReTr​(W01/2​)∗λ∈sp(Z)∑​μ∈sp(W)∑​f(λ/μ)Pλ​(Z)W01/2​Pμ​(W)​,Df​(Z,W)={Qf​(Z,W),lime↓0selected​Qf​(Z,W+e1),​kerW⊆kerZ,otherwise,​

with values embedded in the extended real line, division by zero equal to zero, and a total selected-limit operation in the second branch. The predicate MonotoneMap(A,f,Φ)\mathrm{MonotoneMap}(A,f,\Phi)MonotoneMap(A,f,Φ) means that Φ(X)⪰0\Phi(X)\succeq0Φ(X)⪰0 whenever X∈AX\in AX∈A and X⪰0X\succeq0X⪰0, and that Df(ΦX,ΦY)≤Df(X,Y)D_f(\Phi X,\Phi Y)\le D_f(X,Y)Df​(ΦX,ΦY)≤Df​(X,Y) for every positive semidefinite X,Y∈AX,Y\in AX,Y∈A. There are no normalization, support-inclusion, trace-preservation, continuity, or convexity requirements in this definition.

QAlgorithms.Papers.Independent.Support50.P12.MonotoneRepresentation

Let III be a finite type with decidable equality, let A\mathcal AA be a complex star subalgebra of MI(C)M_I(\mathbb C)MI​(C), let D:MI(C)×MI(C)→[−∞,+∞]D:M_I(\mathbb C)\times M_I(\mathbb C)\to[-\infty,+\infty]D:MI​(C)×MI​(C)→[−∞,+∞] be any function, and let f:R→Rf:\mathbb R\to\mathbb Rf:R→R. An algebra density means a positive semidefinite member of A\mathcal AA with real trace 111. For matrices X,YX,YX,Y, write X[a]=cfc⁡(x↦0 if x=0, xa otherwise)(X)X_{[a]}=\operatorname{cfc}(x\mapsto 0\text{ if }x=0,\ x^a\text{ otherwise})(X)X[a]​=cfc(x↦0 if x=0, xa otherwise)(X), let sp⁡X\operatorname{sp}XspX be its distinct eigenvalues if Hermitian and the empty set otherwise, and let PaXP^X_aPaX​ denote its eigenspace projection. Define

qf(X,Y)=Re⁡Tr⁡ ⁣[(Y[1/2])†∑a∈sp⁡X∑b∈sp⁡Yf(a/b)PaXY[1/2]PbY],q_f(X,Y)=\operatorname{Re}\operatorname{Tr}\!\left[ (Y_{[1/2]})^\dagger \sum_{a\in\operatorname{sp}X}\sum_{b\in\operatorname{sp}Y} f(a/b)P^X_aY_{[1/2]}P^Y_b \right],qf​(X,Y)=ReTr​(Y[1/2]​)†a∈spX∑​b∈spY∑​f(a/b)PaX​Y[1/2]​PbY​​,

and define Df(X,Y)D_f(X,Y)Df​(X,Y) to be this real number embedded in the extended reals if ker⁡Y⊆ker⁡X\ker Y\subseteq\ker XkerY⊆kerX, and otherwise the total selected right limit of qf(X,Y+ϵ1)q_f(X,Y+\epsilon1)qf​(X,Y+ϵ1) as ϵ↓0\epsilon\downarrow0ϵ↓0. Let RRR be exactly the set of values Df(ρ,σ)D_f(\rho,\sigma)Df​(ρ,σ) attained by pairs of algebra densities. The defined property says that there exists a function g:R→[−∞,+∞]g:R\to[-\infty,+\infty]g:R→[−∞,+∞] such that u<vu<vu<v in RRR implies g(u)<g(v)g(u)<g(v)g(u)<g(v), and D(ρ,σ)=g(Df(ρ,σ))D(\rho,\sigma)=g(D_f(\rho,\sigma))D(ρ,σ)=g(Df​(ρ,σ)) for every pair of algebra densities. No continuity or generator condition is imposed on fff, and there is no restriction on DDD outside such pairs. If there are no algebra densities, then RRR is empty and the property holds vacuously.

QAlgorithms.Papers.Independent.Support50.P12.Noncommutative

For a finite index type III with decidable equality and a unital complex ∗*∗-subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), this predicate means that there exist X,Y∈AX,Y\in AX,Y∈A with XY≠YXXY\ne YXXY=YX. There are no positivity, Hermiticity, or normalization requirements on these witnesses. The predicate cannot hold when III is empty or has one element.

QAlgorithms.Papers.Independent.Support50.P12.PairChain

For an arbitrary family of propositions c0,…,c9c_0,\ldots,c_9c0​,…,c9​, this predicate is exactly the conjunction

(c0⇒c1)∧(c1⇒c2)∧(c2⇒c3)∧(c3⇔c4)∧(c4⇔c5)∧(c5⇔c6)∧(c6⇔c7)∧(c7⇔c8)∧(c8⇒c9).(c_0\Rightarrow c_1)\land (c_1\Rightarrow c_2)\land (c_2\Rightarrow c_3)\land (c_3\Leftrightarrow c_4)\land (c_4\Leftrightarrow c_5)\land (c_5\Leftrightarrow c_6)\land (c_6\Leftrightarrow c_7)\land (c_7\Leftrightarrow c_8)\land (c_8\Rightarrow c_9).(c0​⇒c1​)∧(c1​⇒c2​)∧(c2​⇒c3​)∧(c3​⇔c4​)∧(c4​⇔c5​)∧(c5​⇔c6​)∧(c6​⇔c7​)∧(c7​⇔c8​)∧(c8​⇒c9​).

It does not assert that any cic_ici​ holds, and the one-way implications do not include their converses.

QAlgorithms.Papers.Independent.Support50.P12.PosReal

This is the type of real numbers ttt equipped with the condition 0<t0<t0<t; zero and negative real numbers are excluded.

QAlgorithms.Papers.Independent.Support50.P12.PositiveOnConcreteAlgebra

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property means that every positive semidefinite matrix X∈AX\in AX∈A has positive semidefinite image Φ(X)\Phi(X)Φ(X). No linearity assumption is imposed on Φ\PhiΦ, and no condition is imposed on its values at matrices outside the positive semidefinite part of AAA. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.RecoveryBlocks

For finite types I,JI,JI,J, both with decidable equality, this structure consists of a natural number rrr of blocks and, for each 0≤k<r0\le k<r0≤k<r, four positive natural numbers ℓ1,k,r1,k,ℓ2,k,r2,k\ell_{1,k},r_{1,k},\ell_{2,k},r_{2,k}ℓ1,k​,r1,k​,ℓ2,k​,r2,k​, complex matrices E1,kE_{1,k}E1,k​ and E2,kE_{2,k}E2,k​ mapping respectively Cℓ1,k⊗Cr1,k\mathbb C^{\ell_{1,k}}\otimes\mathbb C^{r_{1,k}}Cℓ1,k​⊗Cr1,k​ into CI\mathbb C^ICI and Cℓ2,k⊗Cr2,k\mathbb C^{\ell_{2,k}}\otimes\mathbb C^{r_{2,k}}Cℓ2,k​⊗Cr2,k​ into CJ\mathbb C^JCJ, positive definite matrices τ1,k∈Mr1,k(C)\tau_{1,k}\in M_{r_{1,k}}(\mathbb C)τ1,k​∈Mr1,k​​(C) and τ2,k∈Mr2,k(C)\tau_{2,k}\in M_{r_{2,k}}(\mathbb C)τ2,k​∈Mr2,k​​(C) with Re⁡Tr⁡τ1,k=Re⁡Tr⁡τ2,k=1\operatorname{Re}\operatorname{Tr}\tau_{1,k}=\operatorname{Re}\operatorname{Tr}\tau_{2,k}=1ReTrτ1,k​=ReTrτ2,k​=1, and a complex ℓ2,k×ℓ1,k\ell_{2,k}\times\ell_{1,k}ℓ2,k​×ℓ1,k​ matrix UkU_kUk​. The required identities are E1,k†E1,k=1E_{1,k}^{\dagger}E_{1,k}=1E1,k†​E1,k​=1, E2,k†E2,k=1E_{2,k}^{\dagger}E_{2,k}=1E2,k†​E2,k​=1, E1,k†E1,l=0E_{1,k}^{\dagger}E_{1,l}=0E1,k†​E1,l​=0 and E2,k†E2,l=0E_{2,k}^{\dagger}E_{2,l}=0E2,k†​E2,l​=0 whenever k≠lk\ne lk=l, and Uk†Uk=1U_k^\dagger U_k=1Uk†​Uk​=1, UkUk†=1U_kU_k^\dagger=1Uk​Uk†​=1. Thus the embeddings have mutually orthogonal ranges within each ambient space, while the two identities for UkU_kUk​ require a unitary identification of its two coordinate spaces. There is no requirement that the embedding ranges exhaust the ambient spaces. The case r=0r=0r=0 is allowed, with all block-indexed requirements vacuous; neither III nor JJJ is assumed nonempty.

QAlgorithms.Papers.Independent.Support50.P12.RecoveryRefinement

For finite index types I,JI,JI,J with decidable equality, possibly empty, unital complex ∗*∗-subalgebras A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), B⊆MJ(C)B\subseteq M_J(\mathbb C)B⊆MJ​(C), a complex-linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and any set C⊆MI(C)C\subseteq M_I(\mathbb C)C⊆MI​(C), this predicate asserts two recovery-existence statements. To specify them, a map FFF is nnn-positive on an algebra DDD when every positive semidefinite block matrix Z=(Zab)a,b<nZ=(Z_{ab})_{a,b<n}Z=(Zab​)a,b<n​ whose blocks all belong to DDD remains positive semidefinite after replacing each block ZabZ_{ab}Zab​ by F(Zab)F(Z_{ab})F(Zab​); complete positivity on DDD means this for every natural nnn. A reverse map Ψ:MJ(C)→MI(C)\Psi:M_J(\mathbb C)\to M_I(\mathbb C)Ψ:MJ​(C)→MI​(C) has the required stochastic property when every ambient matrix has image in AAA, every ambient matrix has adjoint image in BBB, Ψ†\Psi^\daggerΨ† is positive on positive semidefinite members of AAA,

(Ψ†Z)†(Ψ†Z)⪯Ψ†(Z†Z)(Z∈A),Tr⁡Ψ(H)=Tr⁡H(H∈B),(\Psi^\dagger Z)^\dagger(\Psi^\dagger Z)\preceq\Psi^\dagger(Z^\dagger Z) \quad(Z\in A), \qquad \operatorname{Tr}\Psi(H)=\operatorname{Tr}H\quad(H\in B),(Ψ†Z)†(Ψ†Z)⪯Ψ†(Z†Z)(Z∈A),TrΨ(H)=TrH(H∈B),

where U⪯VU\preceq VU⪯V means V−UV-UV−U is positive semidefinite and

(Ψ†Z)ij=∑k,lΨ(Eij)kl‾Zkl.(\Psi^\dagger Z)_{ij}=\sum_{k,l}\overline{\Psi(E_{ij})_{kl}}Z_{kl}.(Ψ†Z)ij​=k,l∑​Ψ(Eij​)kl​​Zkl​.

The first statement is that, for every natural nnn, if Φ\PhiΦ is nnn-positive on AAA, there exists a complex-linear reverse map Ψ\PsiΨ with this stochastic property that is nnn-positive on BBB and satisfies Ψ(Φ(ρ))=ρ\Psi(\Phi(\rho))=\rhoΨ(Φ(ρ))=ρ for every ρ∈C\rho\in Cρ∈C. The second statement is that, if Φ\PhiΦ is completely positive on AAA, there exists such a stochastic reverse map that is completely positive on BBB and recovers every ρ∈C\rho\in Cρ∈C. The first witness may depend on nnn. The quantifier includes n=0n=0n=0, where the positivity antecedent is vacuous, so the first statement still requires a stochastic recovering map. No density or algebra-membership assumptions are imposed on members of CCC; when C=∅C=\varnothingC=∅, only the recovery equalities are vacuous.

QAlgorithms.Papers.Independent.Support50.P12.RepresentsStateDivergence

For any finite index type III with decidable equality, unital complex star subalgebra A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C), function D:MI(C)×MI(C)→[−∞,+∞]D:M_I(\mathbb C)\times M_I(\mathbb C)\to[-\infty,+\infty]D:MI​(C)×MI​(C)→[−∞,+∞], and function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, this predicate requires fff to be continuous on (0,∞)(0,\infty)(0,∞) and f(x)/xf(x)/xf(x)/x, embedded in the extended reals, to converge as x→+∞x\to+\inftyx→+∞ to its selected extended-real limit. It also requires D(ρ,σ)=Df(ρ,σ)D(\rho,\sigma)=\mathcal D_f(\rho,\sigma)D(ρ,σ)=Df​(ρ,σ) for every ρ,σ∈A\rho,\sigma\in\mathcal Aρ,σ∈A that are positive semidefinite with real trace 111. To define Df\mathcal D_fDf​, write

Qf(R,S)=Re⁡Tr⁡ ⁣[(S[1/2])†∑u∈Σ(R)∑v∈Σ(S)f(u/v) Pu(R)S[1/2]Pv(S)],Q_f(R,S)=\operatorname{Re}\operatorname{Tr}\!\left[ (S^{[1/2]})^\dagger \sum_{u\in\Sigma(R)}\sum_{v\in\Sigma(S)} f(u/v)\,P_u(R)S^{[1/2]}P_v(S)\right],Qf​(R,S)=ReTr​(S[1/2])†u∈Σ(R)∑​v∈Σ(S)∑​f(u/v)Pu​(R)S[1/2]Pv​(S)​,

where Σ(H)\Sigma(H)Σ(H) is the finite set of distinct real eigenvalues when HHH is Hermitian and is empty otherwise, Pu(H)P_u(H)Pu​(H) is the orthogonal projection onto its uuu-eigenspace, and H[t]H^{[t]}H[t] is real functional calculus with value 000 at scalar zero and xtx^txt elsewhere. If ker⁡σ⊆ker⁡ρ\ker\sigma\subseteq\ker\rhokerσ⊆kerρ, then Df(ρ,σ)=Qf(ρ,σ)\mathcal D_f(\rho,\sigma)=Q_f(\rho,\sigma)Df​(ρ,σ)=Qf​(ρ,σ), embedded in the extended reals; otherwise it is the totalized selected limit of Qf(ρ,σ+εI)Q_f(\rho,\sigma+\varepsilon I)Qf​(ρ,σ+εI) as ε→0+\varepsilon\to0^+ε→0+, with existence of this latter limit not separately required. The ratios u/0u/0u/0 in the sums equal 000 by totalized division. No condition is imposed on DDD outside the specified density matrices; if III is empty, there are no such density matrices, so that equality requirement is vacuous.

QAlgorithms.Papers.Independent.Support50.P12.Reversible

For any finite index types I,JI,JI,J with decidable equality, unital complex star subalgebras A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C) and B⊆MJ(C)\mathcal B\subseteq M_J(\mathbb C)B⊆MJ​(C), complex-linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and arbitrary X,Y∈MI(C)X,Y\in M_I(\mathbb C)X,Y∈MI​(C), this predicate says that there exists a complex-linear map Ψ:MJ(C)→MI(C)\Psi:M_J(\mathbb C)\to M_I(\mathbb C)Ψ:MJ​(C)→MI​(C) satisfying Ψ(Φ(X))=X\Psi(\Phi(X))=XΨ(Φ(X))=X and Ψ(Φ(Y))=Y\Psi(\Phi(Y))=YΨ(Φ(Y))=Y, together with these conditions: Ψ(Z)∈A\Psi(Z)\in\mathcal AΨ(Z)∈A for every matrix ZZZ, its adjoint Ψ†(H)∈B\Psi^\dagger(H)\in\mathcal BΨ†(H)∈B for every matrix HHH, it preserves the trace of every Z∈BZ\in\mathcal BZ∈B, and for every H∈AH\in\mathcal AH∈A,

H≥0⟹Ψ†(H)≥0,Ψ†(H†H)−Ψ†(H)†Ψ†(H)≥0.H\ge0\Longrightarrow\Psi^\dagger(H)\ge0, \qquad \Psi^\dagger(H^\dagger H)-\Psi^\dagger(H)^\dagger\Psi^\dagger(H)\ge0.H≥0⟹Ψ†(H)≥0,Ψ†(H†H)−Ψ†(H)†Ψ†(H)≥0.

The adjoint here is explicitly Ψ†(H)jk=∑i,lΨ(Ejk)il‾Hil\Psi^\dagger(H)_{jk}=\sum_{i,l}\overline{\Psi(E_{jk})_{il}}H_{il}Ψ†(H)jk​=∑i,l​Ψ(Ejk​)il​​Hil​, where EjkE_{jk}Ejk​ is a matrix unit; matrix inequalities mean positive semidefiniteness. Neither X,YX,YX,Y nor their images are required to be density matrices or algebra members, and no additional condition is imposed on Φ\PhiΦ. Empty index types are permitted.

QAlgorithms.Papers.Independent.Support50.P12.Rich

For any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R and natural number nnn, this predicate says that the support of a selected measure μf\mu_fμf​ on (0,∞)(0,\infty)(0,∞) has cardinality at least nnn. The measure μf\mu_fμf​ is chosen, whenever possible, to admit real coefficients a,ba,ba,b, with b≥0b\ge0b≥0, for which (1+t)−2(1+t)^{-2}(1+t)−2 is integrable and, for every x≥0x\ge0x≥0, the integrable kernel x/(1+t)−x/(x+t)x/(1+t)-x/(x+t)x/(1+t)−x/(x+t) gives

fc(x)=fc(0)+ax+bx2+∫(0,∞)(x1+t−xx+t) dμf(t),f_c(x)=f_c(0)+ax+bx^2+ \int_{(0,\infty)}\left(\frac{x}{1+t}-\frac{x}{x+t}\right)\,d\mu_f(t),fc​(x)=fc​(0)+ax+bx2+∫(0,∞)​(1+tx​−x+tx​)dμf​(t),

where fc=ff_c=ffc​=f away from zero and fc(0)f_c(0)fc​(0) is the totalized selected right limit of fff at zero. If no such representation exists, μf\mu_fμf​ is still an arbitrarily selected measure, and the predicate still tests its support cardinality. The comparison is a cardinal comparison, so infinite supports are allowed; when n=0n=0n=0, the predicate always holds.

QAlgorithms.Papers.Independent.Support50.P12.SchwarzBound

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and a real number ccc, this property means that c≥0c\ge0c≥0, that Φ(X)\Phi(X)Φ(X) is positive semidefinite whenever X∈AX\in AX∈A is positive semidefinite, and that

c Φ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A,c\,\Phi(X^*X)-\Phi(X)^*\Phi(X) \quad\text{is positive semidefinite for every }X\in A,cΦ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A,

where X∗X^*X∗ denotes conjugate transpose. The value c=0c=0c=0 is included, no linearity is assumed, and empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.SemidefiniteOperatorConvex

For a function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, this property means that, for every natural number nnn, every pair of positive semidefinite complex n×nn\times nn×n matrices A,BA,BA,B, and every t∈[0,1]t\in[0,1]t∈[0,1], the matrix

t f(A)+(1−t) f(B)−f(tA+(1−t)B)t\,f(A)+(1-t)\,f(B)-f\bigl(tA+(1-t)B\bigr)tf(A)+(1−t)f(B)−f(tA+(1−t)B)

is positive semidefinite, where each application of fff to a matrix uses the real continuous functional calculus. The definition includes n=0n=0n=0 and both endpoints t=0,1t=0,1t=0,1; it imposes no separate continuity hypothesis on fff.

QAlgorithms.Papers.Independent.Support50.P12.SpectralAlphabet

For a finite index type III with decidable equality and arbitrary matrices X,Y∈MI(C)X,Y\in M_I(\mathbb C)X,Y∈MI​(C), this is the type of pairs (a,b)(a,b)(a,b) where aaa belongs to the finite set assigned to XXX and bbb belongs to the finite set assigned to YYY. For a Hermitian matrix, its assigned set consists of its distinct real eigenvalues; for a matrix that is not Hermitian, the assigned set is empty. Thus the alphabet is the Cartesian product of these two sets, with membership included in each coordinate. It is empty if either matrix is not Hermitian, and also when III is empty.

QAlgorithms.Papers.Independent.Support50.P12.Substochastic

For finite index types I,JI,JI,J with decidable equality, unital complex star subalgebras A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), B⊆MJ(C)B\subseteq M_J(\mathbb C)B⊆MJ​(C), and a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this property requires that Φ(X)∈B\Phi(X)\in BΦ(X)∈B for every X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C) and Φ†(Y)∈A\Phi^\dagger(Y)\in AΦ†(Y)∈A for every Y∈MJ(C)Y\in M_J(\mathbb C)Y∈MJ​(C), where

(Φ†(Y))ij=∑k,l∈J(Φ(Eij))kl‾ Ykl.\bigl(\Phi^\dagger(Y)\bigr)_{ij} =\sum_{k,l\in J}\overline{\bigl(\Phi(E_{ij})\bigr)_{kl}}\,Y_{kl}.(Φ†(Y))ij​=k,l∈J∑​(Φ(Eij​))kl​​Ykl​.

It also requires that Φ†(Y)\Phi^\dagger(Y)Φ†(Y) be positive semidefinite for every positive semidefinite Y∈BY\in BY∈B, and that Φ†(Y∗Y)−Φ†(Y)∗Φ†(Y)\Phi^\dagger(Y^*Y)-\Phi^\dagger(Y)^*\Phi^\dagger(Y)Φ†(Y∗Y)−Φ†(Y)∗Φ†(Y) be positive semidefinite for every Y∈BY\in BY∈B. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.TwoProjections

For a finite index type III with decidable equality and a unital complex ∗*∗-subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), this predicate means that there exist matrices P,Q∈AP,Q\in AP,Q∈A such that

P†=P,Q†=Q,P2=P,Q2=Q,P≠0,Q≠0,PQ=0.P^\dagger=P,\qquad Q^\dagger=Q,\qquad P^2=P,\qquad Q^2=Q,\qquad P\ne0,\qquad Q\ne0,\qquad PQ=0.P†=P,Q†=Q,P2=P,Q2=Q,P=0,Q=0,PQ=0.

Thus it requires two nonzero orthogonal Hermitian projections belonging to AAA. It cannot hold when III has fewer than two elements.

QAlgorithms.Papers.Independent.Support50.P12.amp

For any index types I,JI,JI,J, any map Φ\PhiΦ from complex I×II\times II×I matrices to complex J×JJ\times JJ×J matrices, any natural number nnn, and any complex matrix XXX indexed by {0,…,n−1}×I\{0,\ldots,n-1\}\times I{0,…,n−1}×I, the matrix amp⁡n(Φ,X)\operatorname{amp}_n(\Phi,X)ampn​(Φ,X) is defined by

amp⁡n(Φ,X)(a,u),(b,v)=[Φ(Xab)]u,v,(Xab)i,j=X(a,i),(b,j).\operatorname{amp}_n(\Phi,X)_{(a,u),(b,v)} =\bigl[\Phi(X_{ab})\bigr]_{u,v}, \qquad (X_{ab})_{i,j}=X_{(a,i),(b,j)}.ampn​(Φ,X)(a,u),(b,v)​=[Φ(Xab​)]u,v​,(Xab​)i,j​=X(a,i),(b,j)​.

Thus Φ\PhiΦ is applied separately to each block. No linearity or positivity of Φ\PhiΦ is required. For n=0n=0n=0, the result is the matrix with empty index set.

QAlgorithms.Papers.Independent.Support50.P12.bayesianTestingDistance

For a finite type III with decidable equality, any real ppp, and arbitrary complex I×II\times II×I matrices ρ,σ\rho,\sigmaρ,σ, define Z=pρ−(1−p)σZ=p\rho-(1-p)\sigmaZ=pρ−(1−p)σ and

x=1−Re⁡Tr⁡Z†Z2min⁡(p,1−p).x=\frac{1-\operatorname{Re}\operatorname{Tr}\sqrt{Z^\dagger Z}} {2\min(p,1-p)}.x=2min(p,1−p)1−ReTrZ†Z​​.

The defined extended-real value is −log⁡x-\log x−logx when x>0x>0x>0, and +∞+\infty+∞ when x≤0x\le0x≤0. The square root is given by real functional calculus, so the numerator uses exactly the trace norm Re⁡Tr⁡Z†Z\operatorname{Re}\operatorname{Tr}\sqrt{Z^\dagger Z}ReTrZ†Z​. No restriction 0<p<10<p<10<p<1, positivity, or normalization is imposed. Division by zero is the total real operation returning zero; consequently p=0p=0p=0 and p=1p=1p=1 both give +∞+\infty+∞. Empty III is allowed.

QAlgorithms.Papers.Independent.Support50.P12.chernoff

For any finite index type III with decidable equality and arbitrary complex I×II\times II×I matrices A,BA,BA,B, this definition is the extended-real supremum

sup⁡0≤a<1[−ℓA,B(a)],ℓA,B(a)={log⁡qa,qa>0,−∞,qa≤0,qa=Re⁡Tr⁡(A[a]B[1−a]).\sup_{0\le a<1}\bigl[-\ell_{A,B}(a)\bigr], \qquad \ell_{A,B}(a)= \begin{cases} \log q_a,&q_a>0,\\ -\infty,&q_a\le0, \end{cases} \quad q_a=\operatorname{Re}\operatorname{Tr}(A^{[a]}B^{[1-a]}).0≤a<1sup​[−ℓA,B​(a)],ℓA,B​(a)={logqa​,−∞,​qa​>0,qa​≤0,​qa​=ReTr(A[a]B[1−a]).

Here H[t]H^{[t]}H[t] is real functional calculus for the function x↦xtx\mapsto x^tx↦xt away from zero, assigned value 000 at zero, and log⁡\loglog is natural. The endpoint a=0a=0a=0 is included and a=1a=1a=1 is excluded. No positivity, normalization, or support condition is assumed, and no maximizing parameter is asserted to exist. If qa≤0q_a\le0qa​≤0 for any allowed parameter, the supremum is +∞+\infty+∞; this includes the case of an empty index type.

QAlgorithms.Papers.Independent.Support50.P12.codeConditions

Let I,JI,JI,J be finite types with decidable equality, let A⊆MI(C)\mathcal A\subseteq M_I(\mathbb C)A⊆MI​(C) and B⊆MJ(C)\mathcal B\subseteq M_J(\mathbb C)B⊆MJ​(C) be complex star subalgebras, let Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C) be complex-linear, and let C⊆MI(C)C\subseteq M_I(\mathbb C)C⊆MI​(C) be arbitrary. This definition assigns ten propositions, indexed 0,…,90,\ldots,90,…,9; it does not assert their conjunction. The following notation specifies their meanings. Write tr⁡X=Re⁡Tr⁡X\operatorname{tr}X=\operatorname{Re}\operatorname{Tr}XtrX=ReTrX, H=conv⁡RC‾H=\overline{\operatorname{conv}_{\mathbb R}C}H=convR​C​, and d=∣I∣2+∣J∣2d=|I|^2+|J|^2d=∣I∣2+∣J∣2. An algebra density is a matrix belonging to the indicated algebra that is positive semidefinite and has real trace 111. Say X≼sYX\preccurlyeq_sYX≼s​Y when Yv=0Yv=0Yv=0 implies Xv=0Xv=0Xv=0 for every vector vvv. Let X[a]X_{[a]}X[a]​ denote functional calculus with the scalar function that is zero at zero and equals xax^axa otherwise. Let sp⁡X\operatorname{sp}XspX be the finite set of distinct eigenvalues when XXX is Hermitian, and the empty set otherwise; let PaXP^X_aPaX​ be the orthogonal projection onto the eigenspace of XXX for eigenvalue aaa. For a real function fff, put

qf(X,Y)=Re⁡Tr⁡ ⁣[(Y[1/2])†∑a∈sp⁡X∑b∈sp⁡Yf(a/b) PaXY[1/2]PbY].q_f(X,Y)=\operatorname{Re}\operatorname{Tr}\!\left[ (Y_{[1/2]})^\dagger \sum_{a\in\operatorname{sp}X}\sum_{b\in\operatorname{sp}Y} f(a/b)\,P^X_aY_{[1/2]}P^Y_b \right].qf​(X,Y)=ReTr​(Y[1/2]​)†a∈spX∑​b∈spY∑​f(a/b)PaX​Y[1/2]​PbY​​.

Define Df(X,Y)=qf(X,Y)D_f(X,Y)=q_f(X,Y)Df​(X,Y)=qf​(X,Y), embedded in the extended reals, when X≼sYX\preccurlyeq_sYX≼s​Y; otherwise define it by the total right-limit selection at ϵ=0\epsilon=0ϵ=0 of qf(X,Y+ϵ1)q_f(X,Y+\epsilon1)qf​(X,Y+ϵ1). A generator fff means that fff is continuous on (0,∞)(0,\infty)(0,∞) and that f(x)/xf(x)/xf(x)/x, embedded in the extended reals, converges as x→+∞x\to+\inftyx→+∞ to its selected limit LfL_fLf​. Semidefinite operator convexity means that, in every finite matrix dimension, including zero, for positive semidefinite X,YX,YX,Y and t∈[0,1]t\in[0,1]t∈[0,1], the difference

tf(X)+(1−t)f(Y)−f(tX+(1−t)Y)t f(X)+(1-t)f(Y)-f(tX+(1-t)Y)tf(X)+(1−t)f(Y)−f(tX+(1−t)Y)

is positive semidefinite, with functions evaluated by functional calculus. To say that fff is rich for ddd means that the support of its selected representing measure has cardinality at least ddd: this measure on (0,∞)(0,\infty)(0,∞) is selected to admit real a,ba,ba,b, with b≥0b\ge0b≥0, such that (1+t)−2(1+t)^{-2}(1+t)−2 is integrable and, for every x≥0x\ge0x≥0, the kernel x/(1+t)−x/(x+t)x/(1+t)-x/(x+t)x/(1+t)−x/(x+t) is integrable and

fc(x)=fc(0)+ax+bx2+∫(0,∞)(x1+t−xx+t)dμ(t),f_c(x)=f_c(0)+ax+bx^2+ \int_{(0,\infty)}\left(\frac{x}{1+t}-\frac{x}{x+t}\right)d\mu(t),fc​(x)=fc​(0)+ax+bx2+∫(0,∞)​(1+tx​−x+tx​)dμ(t),

where fc=ff_c=ffc​=f away from zero and fc(0)f_c(0)fc​(0) is the selected right limit of fff; the measure-selection definition itself does not require such a representation to exist. Write ℓ(x)=log⁡x\ell(x)=\log xℓ(x)=logx for x>0x>0x>0 and ℓ(x)=−∞\ell(x)=-\inftyℓ(x)=−∞ otherwise, and define

H(X,Y;r)=sup⁡0≤a<1−ar−ℓ ⁣(tr⁡(X[a]Y[1−a]))1−a.\mathcal H(X,Y;r)= \sup_{0\le a<1}\frac{-ar-\ell\!\left(\operatorname{tr}(X_{[a]}Y_{[1-a]})\right)}{1-a}.H(X,Y;r)=0≤a<1sup​1−a−ar−ℓ(tr(X[a]​Y[1−a]​))​.

For a complex-linear map LLL, its adjoint here is explicitly

L∗(Z)ij=∑u,vL(Eij)uv‾ Zuv,L^\ast(Z)_{ij}=\sum_{u,v}\overline{L(E_{ij})_{uv}}\,Z_{uv},L∗(Z)ij​=u,v∑​L(Eij​)uv​​Zuv​,

where EijE_{ij}Eij​ is the matrix unit. Define the recovery expression

Pσ,L(Z)=S L∗(TZT) S,S=cfc⁡(⋅)(σ),T=cfc⁡ ⁣(x↦(x)−1)(Lσ).\mathcal P_{\sigma,L}(Z)=S\,L^\ast(TZT)\,S,\qquad S=\operatorname{cfc}(\sqrt{\cdot})(\sigma),\quad T=\operatorname{cfc}\!\left(x\mapsto(\sqrt{x})^{-1}\right)(L\sigma).Pσ,L​(Z)=SL∗(TZT)S,S=cfc(⋅​)(σ),T=cfc(x↦(x​)−1)(Lσ).

A map L:B→AL:\mathcal B\to\mathcal AL:B→A, understood as a map on the full ambient matrix spaces, is algebra-stochastic when its entire range lies in A\mathcal AA, the entire range of L∗L^\astL∗ lies in B\mathcal BB, L∗L^\astL∗ is positive on positive semidefinite members of A\mathcal AA, (L∗X)†L∗X≤L∗(X†X)(L^\ast X)^\dagger L^\ast X\le L^\ast(X^\dagger X)(L∗X)†L∗X≤L∗(X†X) for every X∈AX\in\mathcal AX∈A, and Tr⁡L(Y)=Tr⁡Y\operatorname{Tr}L(Y)=\operatorname{Tr}YTrL(Y)=TrY for every Y∈BY\in\mathcal BY∈B; matrix inequalities mean positive semidefinite differences. With this notation, proposition 000 says that there exists an algebra-stochastic Ψ:MJ(C)→MI(C)\Psi:M_J(\mathbb C)\to M_I(\mathbb C)Ψ:MJ​(C)→MI​(C) with Ψ(Φρ)=ρ\Psi(\Phi\rho)=\rhoΨ(Φρ)=ρ for every ρ∈H\rho\in Hρ∈H. Proposition 111 says that Df(Φρ,Φσ)=Df(ρ,σ)D_f(\Phi\rho,\Phi\sigma)=D_f(\rho,\sigma)Df​(Φρ,Φσ)=Df​(ρ,σ) for every generator fff that is semidefinite operator convex and every ρ,σ∈H\rho,\sigma\in Hρ,σ∈H. Proposition 222 says that there exist an algebra density σ∈A\sigma\in\mathcal Aσ∈A and a generator fff that is semidefinite operator convex and rich for ddd, such that every ρ∈C\rho\in Cρ∈C satisfies ρ≼sσ\rho\preccurlyeq_s\sigmaρ≼s​σ and Df(Φρ,Φσ)=Df(ρ,σ)D_f(\Phi\rho,\Phi\sigma)=D_f(\rho,\sigma)Df​(Φρ,Φσ)=Df​(ρ,σ). Proposition 333 says that there exist an algebra density σ∈A\sigma\in\mathcal Aσ∈A supporting every ρ∈C\rho\in Cρ∈C in that same sense, and a set T⊆(0,∞)T\subseteq(0,\infty)T⊆(0,∞) of cardinality at least ddd, such that Dft(Φρ,Φσ)=Dft(ρ,σ)D_{f_t}(\Phi\rho,\Phi\sigma)=D_{f_t}(\rho,\sigma)Dft​​(Φρ,Φσ)=Dft​​(ρ,σ) for all ρ∈C\rho\in Cρ∈C, t∈Tt\in Tt∈T, where ft(x)=−x/(x+t)f_t(x)=-x/(x+t)ft​(x)=−x/(x+t). Proposition 444 says that H(Φρ,Φσ;r)=H(ρ,σ;r)\mathcal H(\Phi\rho,\Phi\sigma;r)=\mathcal H(\rho,\sigma;r)H(Φρ,Φσ;r)=H(ρ,σ;r) for every ρ,σ∈H\rho,\sigma\in Hρ,σ∈H and every real rrr. Proposition 555 says that there exist an algebra density σ∈A\sigma\in\mathcal Aσ∈A supporting CCC and a real δ>0\delta>0δ>0 such that this same equality holds for every ρ∈C\rho\in Cρ∈C and 0<r<δ0<r<\delta0<r<δ. Proposition 666 says that, for every ρ,σ∈H\rho,\sigma\in Hρ,σ∈H, if ker⁡σ=⋂η∈Cker⁡η\ker\sigma=\bigcap_{\eta\in C}\ker\etakerσ=⋂η∈C​kerη, then Pσ,Φ(Φρ)=ρ\mathcal P_{\sigma,\Phi}(\Phi\rho)=\rhoPσ,Φ​(Φρ)=ρ. Proposition 777 says that there exists an algebra density σ∈A\sigma\in\mathcal Aσ∈A such that Pσ,Φ(Φρ)=ρ\mathcal P_{\sigma,\Phi}(\Phi\rho)=\rhoPσ,Φ​(Φρ)=ρ for every ρ∈C\rho\in Cρ∈C, without an additional support requirement. Proposition 888 says that there exist r≥0r\ge0r≥0 blocks with positive left and right dimensions, mutually orthogonal isometric embeddings E1,kE_{1,k}E1,k​ into CI\mathbb C^ICI and E2,kE_{2,k}E2,k​ into CJ\mathbb C^JCJ, positive definite right-factor matrices τ1,k,τ2,k\tau_{1,k},\tau_{2,k}τ1,k​,τ2,k​ of real trace 111, and left-factor matrices UkU_kUk​ satisfying both unitary identities, such that ker⁡∑kE1,kE1,k†=⋂ρ∈Cker⁡ρ\ker\sum_kE_{1,k}E_{1,k}^\dagger=\bigcap_{\rho\in C}\ker\rhoker∑k​E1,k​E1,k†​=⋂ρ∈C​kerρ and ker⁡∑kE2,kE2,k†=⋂ρ∈Cker⁡(Φρ)\ker\sum_kE_{2,k}E_{2,k}^\dagger=\bigcap_{\rho\in C}\ker(\Phi\rho)ker∑k​E2,k​E2,k†​=⋂ρ∈C​ker(Φρ); every ρ∈C\rho\in Cρ∈C has a representation ρ=∑kpkE1,k(Xk⊗τ1,k)E1,k†\rho=\sum_kp_kE_{1,k}(X_k\otimes\tau_{1,k})E_{1,k}^\daggerρ=∑k​pk​E1,k​(Xk​⊗τ1,k​)E1,k†​ with pk≥0p_k\ge0pk​≥0, ∑kpk=1\sum_kp_k=1∑k​pk​=1, and positive semidefinite XkX_kXk​ of real trace 111; and, for every block kkk and every complex matrix XXX on its first left factor,

E1,k(X⊗τ1,k)E1,k†∈A,Φ ⁣(E1,k(X⊗τ1,k)E1,k†)=E2,k((UkXUk†)⊗τ2,k)E2,k†.E_{1,k}(X\otimes\tau_{1,k})E_{1,k}^\dagger\in\mathcal A,\qquad \Phi\!\left(E_{1,k}(X\otimes\tau_{1,k})E_{1,k}^\dagger\right) =E_{2,k}\bigl((U_kXU_k^\dagger)\otimes\tau_{2,k}\bigr)E_{2,k}^\dagger.E1,k​(X⊗τ1,k​)E1,k†​∈A,Φ(E1,k​(X⊗τ1,k​)E1,k†​)=E2,k​((Uk​XUk†​)⊗τ2,k​)E2,k†​.

Proposition 999 says that there exists an algebra density σ∈A\sigma\in\mathcal Aσ∈A supporting CCC such that, for every ρ∈C\rho\in Cρ∈C, every 0<p<10<p<10<p<1, and every natural nnn, the Bayesian testing value of Φ[n](ρ⊗n)\Phi^{[n]}(\rho^{\otimes n})Φ[n](ρ⊗n) and Φ[n](σ⊗n)\Phi^{[n]}(\sigma^{\otimes n})Φ[n](σ⊗n) equals that of ρ⊗n\rho^{\otimes n}ρ⊗n and σ⊗n\sigma^{\otimes n}σ⊗n. Here (X⊗n)x,y=∏i<nXxi,yi(X^{\otimes n})_{x,y}=\prod_{i<n}X_{x_i,y_i}(X⊗n)x,y​=∏i<n​Xxi​,yi​​, the map on nnn factors is defined by

Φ[n](Z)u,v=∑x,y∈InZx,y∏i<nΦ(Exiyi)ui,vi,\Phi^{[n]}(Z)_{u,v} =\sum_{x,y\in I^n}Z_{x,y}\prod_{i<n}\Phi(E_{x_i y_i})_{u_i,v_i},Φ[n](Z)u,v​=x,y∈In∑​Zx,y​i<n∏​Φ(Exi​yi​​)ui​,vi​​,

and the Bayesian value of X,YX,YX,Y is −ℓ((1−Re⁡Tr⁡Z†Z)/(2min⁡(p,1−p)))-\ell((1-\operatorname{Re}\operatorname{Tr}\sqrt{Z^\dagger Z})/(2\min(p,1-p)))−ℓ((1−ReTrZ†Z​)/(2min(p,1−p))), with Z=pX−(1−p)YZ=pX-(1-p)YZ=pX−(1−p)Y. All divisions and extended-real operations have their total meanings, including division by zero; n=0n=0n=0 is included and its products are empty products equal to 111. No assumption that Φ\PhiΦ is a channel, that CCC consists of densities, or that C,I,JC,I,JC,I,J are nonempty is part of this definition.

QAlgorithms.Papers.Independent.Support50.P12.continuousPart

For any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R and any real number xxx, this defines a real-valued function equal to f(x)f(x)f(x) when x≠0x\ne0x=0, and equal at x=0x=0x=0 to the library’s total limit selector for f(t)f(t)f(t) as t→0t\to0t→0 through strictly positive real numbers. No existence of a right-hand limit is assumed: when a finite right-hand limit exists, that is the selected value; otherwise the definition still returns a real value without asserting convergence.

QAlgorithms.Papers.Independent.Support50.P12.cpow

For any finite type III with decidable equality, any complex I×II\times II×I matrix XXX, and any z∈Cz\in\mathbb Cz∈C, cpow⁡(X,z)\operatorname{cpow}(X,z)cpow(X,z) is zero if XXX is not Hermitian. Otherwise, writing the chosen Hermitian spectral decomposition as X=Udiag⁡(λj)U†X=U\operatorname{diag}(\lambda_j)U^\daggerX=Udiag(λj​)U†, it is Udiag⁡(gz(λj))U†U\operatorname{diag}(g_z(\lambda_j))U^\daggerUdiag(gz​(λj​))U†, where gz(0)=0g_z(0)=0gz​(0)=0 and gz(a)=exp⁡(zlog⁡a)g_z(a)=\exp(z\log a)gz​(a)=exp(zloga) for nonzero real aaa. Here log⁡\loglog is the total real logarithm, so for negative aaa it equals log⁡∣a∣\log|a|log∣a∣. In particular, zero eigenvalues remain zero even when z=0z=0z=0; neither positivity nor invertibility of XXX is assumed.

QAlgorithms.Papers.Independent.Support50.P12.diagonalAlgebra

For every finite index type III with decidable equality, diagonalAlgebra(I)\mathrm{diagonalAlgebra}(I)diagonalAlgebra(I) is the unital complex star subalgebra of matrices X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C) satisfying Xij=0X_{ij}=0Xij​=0 whenever i≠ji\ne ji=j, with the inherited matrix addition, multiplication, complex scalar action and conjugate transpose. The definition also applies when III is empty.

QAlgorithms.Papers.Independent.Support50.P12.div

For any finite type III with decidable equality, any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, and any complex I×II\times II×I matrices A,BA,BA,B, define qf(A,B)=Re⁡Tr⁡(SB†∑a∈SA∑b∈SBeigf(a/b)PA,aSBPB,b)q_f(A,B)=\operatorname{Re}\operatorname{Tr}\bigl(S_B^\dagger\sum_{a\in S_A}\sum_{b\in S_B^{\mathrm{eig}}}f(a/b)P_{A,a}S_BP_{B,b}\bigr)qf​(A,B)=ReTr(SB†​∑a∈SA​​∑b∈SBeig​​f(a/b)PA,a​SB​PB,b​), where SBS_BSB​ is the real functional calculus of BBB with g(0)=0g(0)=0g(0)=0, g(x)=x1/2g(x)=x^{1/2}g(x)=x1/2 for x≠0x\ne0x=0, the eigenvalue sets contain distinct Hermitian eigenvalues and are empty for non-Hermitian matrices, and the PPP's are orthogonal eigenspace projections. Powers and division use their total real operations, including a/0=0a/0=0a/0=0. Then div⁡(f,A,B)\operatorname{div}(f,A,B)div(f,A,B) equals the finite extended-real value qf(A,B)q_f(A,B)qf​(A,B) if every vector annihilated by BBB is also annihilated by AAA. Otherwise it is the total limit-selection value of qf(A,B+εI)q_f(A,B+\varepsilon I)qf​(A,B+εI) as ε↓0\varepsilon\downarrow0ε↓0, without asserting that this limit exists. No positivity, normalization, continuity, or generator condition is assumed.

QAlgorithms.Papers.Independent.Support50.P12.exampleRho

The matrix exampleRho\mathrm{exampleRho}exampleRho is the complex 3×33\times33×3 diagonal matrix

(2/30001/30000),\begin{pmatrix} 2/3&0&0\\ 0&1/3&0\\ 0&0&0 \end{pmatrix},​2/300​01/30​000​​,

with row and column indices 0,1,20,1,20,1,2.

QAlgorithms.Papers.Independent.Support50.P12.exampleSigma

The matrix exampleSigma\mathrm{exampleSigma}exampleSigma is the complex 3×33\times33×3 diagonal matrix

(1/60001/30001/2),\begin{pmatrix} 1/6&0&0\\ 0&1/3&0\\ 0&0&1/2 \end{pmatrix},​1/600​01/30​001/2​​,

with row and column indices 0,1,20,1,20,1,2.

QAlgorithms.Papers.Independent.Support50.P12.hoeffding

For any finite index type III with decidable equality, arbitrary complex I×II\times II×I matrices A,BA,BA,B, and real number rrr, this definition is the extended-real supremum

sup⁡0≤a<1−ar−ℓA,B(a)1−a,ℓA,B(a)={log⁡qa,qa>0,−∞,qa≤0,qa=Re⁡Tr⁡(A[a]B[1−a]).\sup_{0\le a<1}\frac{-ar-\ell_{A,B}(a)}{1-a}, \qquad \ell_{A,B}(a)= \begin{cases} \log q_a,&q_a>0,\\ -\infty,&q_a\le0, \end{cases} \quad q_a=\operatorname{Re}\operatorname{Tr}(A^{[a]}B^{[1-a]}).0≤a<1sup​1−a−ar−ℓA,B​(a)​,ℓA,B​(a)={logqa​,−∞,​qa​>0,qa​≤0,​qa​=ReTr(A[a]B[1−a]).

Here H[t]H^{[t]}H[t] denotes real functional calculus with scalar value 000 at zero and xtx^txt elsewhere, and log⁡\loglog is natural. The factor (1−a)−1(1-a)^{-1}(1−a)−1 is a positive finite real number throughout the parameter range. There is no restriction on rrr, no positivity or normalization assumption on A,BA,BA,B, and no assertion of attainment. A nonpositive qaq_aqa​ at any allowed parameter makes the value +∞+\infty+∞, including when III is empty.

QAlgorithms.Papers.Independent.Support50.P12.hull

For an index type III and any set CCC of complex matrices indexed by III, this is the topological closure of the real convex hull of CCC: the closure, in the usual matrix topology, of all finite convex combinations of elements of CCC with nonnegative real coefficients summing to one. No positivity, normalization, or nonemptiness assumption is imposed on CCC; the hull of the empty set is empty.

QAlgorithms.Papers.Independent.Support50.P12.identityNoise

For a finite index type III with decidable equality, any real number eee, and any complex matrix XXX indexed by I×II\times II×I, this defines

Re(X)=(1−e)X+e Tr⁡X∣I∣1.R_e(X)=(1-e)X+\frac{e\,\operatorname{Tr}X}{|I|}\mathbf 1.Re​(X)=(1−e)X+∣I∣eTrX​1.

Here 1\mathbf 11 is the identity matrix. No restriction on eee, positivity, or normalization of XXX is imposed. Empty III is allowed, with division by zero interpreted as 000.

QAlgorithms.Papers.Independent.Support50.P12.inputBlock

Let I,JI,JI,J be finite types with decidable equality, and let DDD be block data consisting of r≥0r\ge0r≥0 blocks with positive dimensions ℓ1,k,r1,k,ℓ2,k,r2,k\ell_{1,k},r_{1,k},\ell_{2,k},r_{2,k}ℓ1,k​,r1,k​,ℓ2,k​,r2,k​, isometric embeddings E1,k:Cℓ1,k⊗Cr1,k→CIE_{1,k}:\mathbb C^{\ell_{1,k}}\otimes\mathbb C^{r_{1,k}}\to\mathbb C^IE1,k​:Cℓ1,k​⊗Cr1,k​→CI and E2,k:Cℓ2,k⊗Cr2,k→CJE_{2,k}:\mathbb C^{\ell_{2,k}}\otimes\mathbb C^{r_{2,k}}\to\mathbb C^JE2,k​:Cℓ2,k​⊗Cr2,k​→CJ whose ranges are pairwise orthogonal within each family, positive definite matrices τ1,k,τ2,k\tau_{1,k},\tau_{2,k}τ1,k​,τ2,k​ on the corresponding right factors with real trace 111, and matrices Uk:Cℓ1,k→Cℓ2,kU_k:\mathbb C^{\ell_{1,k}}\to\mathbb C^{\ell_{2,k}}Uk​:Cℓ1,k​→Cℓ2,k​ satisfying Uk†Uk=1U_k^\dagger U_k=1Uk†​Uk​=1 and UkUk†=1U_kU_k^\dagger=1Uk​Uk†​=1. For every block index k<rk<rk<r and every complex ℓ1,k×ℓ1,k\ell_{1,k}\times\ell_{1,k}ℓ1,k​×ℓ1,k​ matrix XXX, the defined input block is the I×II\times II×I matrix

E1,k(X⊗τ1,k)E1,k†.E_{1,k}(X\otimes\tau_{1,k})E_{1,k}^{\dagger}.E1,k​(X⊗τ1,k​)E1,k†​.

Here ⊗\otimes⊗ is the Kronecker product. No positivity or trace condition is imposed on XXX; when r=0r=0r=0, there is no permissible block index kkk.

QAlgorithms.Papers.Independent.Support50.P12.isometricConjugation

For a finite input index type III, an output index type JJJ, a Boolean parameter anti\mathrm{anti}anti, a rectangular complex matrix WWW with rows indexed by JJJ and columns indexed by III, and X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C), this function returns

{WXTW†,anti=true,WXW†,anti=false.\begin{cases} WX^{\mathsf T}W^\dagger,&\mathrm{anti}=\mathrm{true},\\ WXW^\dagger,&\mathrm{anti}=\mathrm{false}. \end{cases}{WXTW†,WXW†,​anti=true,anti=false.​

The transpose in the first branch is the ordinary transpose. No isometry, unitarity, positivity, or normalization hypothesis is imposed on WWW or XXX; empty index types are permitted.

QAlgorithms.Papers.Independent.Support50.P12.law1

For a finite index type III with decidable equality and arbitrary X,Y∈MI(C)X,Y\in M_I(\mathbb C)X,Y∈MI​(C), this function assigns to each pair (a,b)(a,b)(a,b) of distinct-eigenvalue labels from XXX and YYY the real number

a Re⁡Tr⁡ ⁣(PX(a)PY(b)),a\,\operatorname{Re}\operatorname{Tr}\!\left(P_X(a)P_Y(b)\right),aReTr(PX​(a)PY​(b)),

where PH(λ)P_H(\lambda)PH​(λ) is the orthogonal projection onto the λ\lambdaλ-eigenspace of the operator given by multiplication by HHH. The label set of a Hermitian matrix is its set of distinct real eigenvalues, and that of a non-Hermitian matrix is empty. No positivity or trace-one assumptions are imposed on XXX or YYY, so the definition does not require these values to be nonnegative or to sum to one. If either label set is empty, the function has empty domain.

QAlgorithms.Papers.Independent.Support50.P12.law2

For a finite index type III with decidable equality and arbitrary X,Y∈MI(C)X,Y\in M_I(\mathbb C)X,Y∈MI​(C), this function assigns to each pair (a,b)(a,b)(a,b) of distinct-eigenvalue labels from XXX and YYY the real number

b Re⁡Tr⁡ ⁣(PX(a)PY(b)),b\,\operatorname{Re}\operatorname{Tr}\!\left(P_X(a)P_Y(b)\right),bReTr(PX​(a)PY​(b)),

where PH(λ)P_H(\lambda)PH​(λ) is the orthogonal projection onto the λ\lambdaλ-eigenspace of the operator given by multiplication by HHH. The label set of a Hermitian matrix is its set of distinct real eigenvalues, and that of a non-Hermitian matrix is empty. Neither positivity nor trace-one normalization of XXX or YYY is assumed; consequently nonnegativity and normalization of this function are not part of the definition. If either label set is empty, its domain is empty.

QAlgorithms.Papers.Independent.Support50.P12.logOverlap

For any finite index type III with decidable equality, arbitrary complex I×II\times II×I matrices A,BA,BA,B, and real number aaa, define H[t]H^{[t]}H[t] by real functional calculus using the scalar function that equals 000 at zero and xtx^txt elsewhere, and put q=Re⁡Tr⁡(A[a]B[1−a])q=\operatorname{Re}\operatorname{Tr}(A^{[a]}B^{[1-a]})q=ReTr(A[a]B[1−a]). This definition returns log⁡q\log qlogq, embedded in the extended real numbers, if q>0q>0q>0, and returns −∞-\infty−∞ if q≤0q\le0q≤0. The logarithm is natural. No density, Hermitian, positivity, or support assumption is imposed; in particular, zero eigenvalues remain zero even at exponent 000. An empty index type gives q=0q=0q=0 and hence value −∞-\infty−∞.

QAlgorithms.Papers.Independent.Support50.P12.logOverlapPerspective

For any finite index type III with decidable equality, arbitrary complex I×II\times II×I matrices A,BA,BA,B, and real number sss, this definition returns +∞+\infty+∞ if s<0s<0s<0. If s≥0s\ge0s≥0, it returns

(1+s) ℓA,B ⁣(s1+s),(1+s)\,\ell_{A,B}\!\left(\frac{s}{1+s}\right),(1+s)ℓA,B​(1+ss​),

in extended-real arithmetic, where ℓA,B(a)\ell_{A,B}(a)ℓA,B​(a) is the natural logarithm of Re⁡Tr⁡(A[a]B[1−a])\operatorname{Re}\operatorname{Tr}(A^{[a]}B^{[1-a]})ReTr(A[a]B[1−a]) when that number is positive, and is −∞-\infty−∞ otherwise. The matrix power H[t]H^{[t]}H[t] is defined by real functional calculus with scalar value 000 at zero and xtx^txt elsewhere. For s≥0s\ge0s≥0, the factor 1+s1+s1+s is positive and s/(1+s)∈[0,1)s/(1+s)\in[0,1)s/(1+s)∈[0,1), including parameter 000 when s=0s=0s=0. No matrix positivity or normalization is required; for empty III, the value is −∞-\infty−∞ for s≥0s\ge0s≥0.

QAlgorithms.Papers.Independent.Support50.P12.merge

The declaration defines the complex-linear map merge:M3(C)→M3(C)\mathrm{merge}:M_3(\mathbb C)\to M_3(\mathbb C)merge:M3​(C)→M3​(C) by

merge(X)=(X00+X110000000X22)\mathrm{merge}(X)= \begin{pmatrix} X_{00}+X_{11}&0&0\\ 0&0&0\\ 0&0&X_{22} \end{pmatrix}merge(X)=​X00​+X11​00​000​00X22​​​

for every complex matrix XXX, without any Hermiticity, positivity or normalization requirement.

QAlgorithms.Papers.Independent.Support50.P12.modular

For any finite type III with decidable equality, any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, and any complex I×II\times II×I matrices A,B,XA,B,XA,B,X, the matrix modular⁡(f,A,B,X)\operatorname{modular}(f,A,B,X)modular(f,A,B,X) is

∑a∈SA∑b∈SBf(a/b) PA,aXPB,b,\sum_{a\in S_A}\sum_{b\in S_B} f(a/b)\,P_{A,a}XP_{B,b},a∈SA​∑​b∈SB​∑​f(a/b)PA,a​XPB,b​,

where SAS_ASA​ and SBS_BSB​ are the sets of distinct real eigenvalues if the respective matrices are Hermitian, and are empty otherwise, and PA,aP_{A,a}PA,a​ is the orthogonal projection onto ker⁡(A−aI)\ker(A-aI)ker(A−aI), with PB,bP_{B,b}PB,b​ defined similarly. Scalars are embedded into C\mathbb CC. Division is total real division, so a/0=0a/0=0a/0=0. If either matrix is non-Hermitian, or III is empty, the sum is zero; there are no positivity or regularity assumptions.

QAlgorithms.Papers.Independent.Support50.P12.modularContraction

For finite index types I,JI,JI,J with decidable equality, a complex linear map Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), and arbitrary matrices B∈MI(C)B\in M_I(\mathbb C)B∈MI​(C), X∈MJ(C)X\in M_J(\mathbb C)X∈MJ​(C), this defines the matrix

Φ† ⁣(X P−1/2(Φ(B)))P1/2(B),\Phi^\dagger\!\left(X\,P_{-1/2}(\Phi(B))\right)P_{1/2}(B),Φ†(XP−1/2​(Φ(B)))P1/2​(B),

where Pt(Y)P_t(Y)Pt​(Y) is the real continuous functional calculus of YYY for the scalar function taking 000 to 000 and each nonzero real xxx to its total real power xtx^txt, and

(Φ†(Y))ij=∑k,l∈J(Φ(Eij))kl‾ Ykl.\bigl(\Phi^\dagger(Y)\bigr)_{ij} =\sum_{k,l\in J}\overline{\bigl(\Phi(E_{ij})\bigr)_{kl}}\,Y_{kl}.(Φ†(Y))ij​=k,l∈J∑​(Φ(Eij​))kl​​Ykl​.

No positivity, Hermiticity, or invertibility assumption is imposed on BBB, XXX, or their images; the functional calculus is total, returning zero on non-Hermitian matrices. Zero eigenvalues contribute zero to both indicated powers. Empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.modularSpec

For any finite index type III with decidable equality, any unital complex star subalgebra A\mathcal AA of the complex I×II\times II×I matrices, and arbitrary matrices X,YX,YX,Y of that size, this definition forms a finite set of real numbers. For a Hermitian matrix HHH, let Σ(H)\Sigma(H)Σ(H) be its set of distinct real eigenvalues; for a non-Hermitian matrix, put Σ(H)=∅\Sigma(H)=\varnothingΣ(H)=∅. The resulting set consists exactly of those ratios a=u/va=u/va=u/v, with u∈Σ(X)u\in\Sigma(X)u∈Σ(X) and v∈Σ(Y)v\in\Sigma(Y)v∈Σ(Y), for which there exists a nonzero Z∈AZ\in\mathcal AZ∈A satisfying

XZ Y[−1]=aZ.XZ\,Y^{[-1]}=aZ.XZY[−1]=aZ.

Here Y[−1]Y^{[-1]}Y[−1] is obtained through real functional calculus from the scalar function that is 000 at zero and y−1y^{-1}y−1 elsewhere. Ratios with v=0v=0v=0 are included in the initial construction and equal 000 by totalized division. No positivity or invertibility is assumed. If either matrix is non-Hermitian, or if III is empty, the resulting finite set is empty.

QAlgorithms.Papers.Independent.Support50.P12.outputBlock

Let I,JI,JI,J be finite types with decidable equality, and let DDD consist of r≥0r\ge0r≥0 blocks with positive dimensions ℓ1,k,r1,k,ℓ2,k,r2,k\ell_{1,k},r_{1,k},\ell_{2,k},r_{2,k}ℓ1,k​,r1,k​,ℓ2,k​,r2,k​, isometric embeddings E1,k:Cℓ1,k⊗Cr1,k→CIE_{1,k}:\mathbb C^{\ell_{1,k}}\otimes\mathbb C^{r_{1,k}}\to\mathbb C^IE1,k​:Cℓ1,k​⊗Cr1,k​→CI and E2,k:Cℓ2,k⊗Cr2,k→CJE_{2,k}:\mathbb C^{\ell_{2,k}}\otimes\mathbb C^{r_{2,k}}\to\mathbb C^JE2,k​:Cℓ2,k​⊗Cr2,k​→CJ with pairwise orthogonal ranges within each family, positive definite right-factor matrices τ1,k,τ2,k\tau_{1,k},\tau_{2,k}τ1,k​,τ2,k​ with real trace 111, and matrices Uk:Cℓ1,k→Cℓ2,kU_k:\mathbb C^{\ell_{1,k}}\to\mathbb C^{\ell_{2,k}}Uk​:Cℓ1,k​→Cℓ2,k​ satisfying both Uk†Uk=1U_k^\dagger U_k=1Uk†​Uk​=1 and UkUk†=1U_kU_k^\dagger=1Uk​Uk†​=1. For every k<rk<rk<r and every complex ℓ1,k×ℓ1,k\ell_{1,k}\times\ell_{1,k}ℓ1,k​×ℓ1,k​ matrix XXX, the defined output block is the J×JJ\times JJ×J matrix

E2,k((UkXUk†)⊗τ2,k)E2,k†.E_{2,k}\bigl((U_kXU_k^\dagger)\otimes\tau_{2,k}\bigr)E_{2,k}^{\dagger}.E2,k​((Uk​XUk†​)⊗τ2,k​)E2,k†​.

No positivity or normalization is imposed on XXX, and there is no such block index when r=0r=0r=0.

QAlgorithms.Papers.Independent.Support50.P12.pairConditions

Let I,JI,JI,J be finite index types with decidable equality, possibly empty; let A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C) and B⊆MJ(C)B\subseteq M_J(\mathbb C)B⊆MJ​(C) be unital complex ∗*∗-subalgebras, let Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C) be complex-linear, let X,Y∈MI(C)X,Y\in M_I(\mathbb C)X,Y∈MI​(C) be arbitrary, and let co\mathrm{co}co be a Boolean. This definition returns a family of ten propositions indexed by 0,…,90,\ldots,90,…,9, as specified below. The notation needed to state them is as follows: H⪯KH\preceq KH⪯K means K−HK-HK−H is positive semidefinite; tr⁡RH=Re⁡Tr⁡H\operatorname{tr}_{\mathbb R}H=\operatorname{Re}\operatorname{Tr}HtrR​H=ReTrH; and the adjoint used for any linear map FFF is defined entrywise by

(F†Z)ij=∑k,lF(Eij)kl‾ Zkl,(F^\dagger Z)_{ij}=\sum_{k,l}\overline{F(E_{ij})_{kl}}\,Z_{kl},(F†Z)ij​=k,l∑​F(Eij​)kl​​Zkl​,

where EijE_{ij}Eij​ is the matrix unit. For algebras D,ED,ED,E and a map FFF from the ambient matrices of DDD to those of EEE, write Sco−(D,E;F)\mathsf S^-_{\mathrm{co}}(D,E;F)Sco−​(D,E;F) for the conjunction that every ambient input has image in EEE, every ambient output matrix has adjoint image in DDD, F†F^\daggerF† sends positive semidefinite members of EEE to positive semidefinite matrices, and, for every Z∈EZ\in EZ∈E,

(F†Z)†(F†Z)⪯{F†(ZZ†),co=true,F†(Z†Z),co=false.(F^\dagger Z)^\dagger(F^\dagger Z)\preceq \begin{cases} F^\dagger(ZZ^\dagger),&\mathrm{co}=\mathrm{true},\\ F^\dagger(Z^\dagger Z),&\mathrm{co}=\mathrm{false}. \end{cases}(F†Z)†(F†Z)⪯{F†(ZZ†),F†(Z†Z),​co=true,co=false.​

Write Sco(D,E;F)\mathsf S_{\mathrm{co}}(D,E;F)Sco​(D,E;F) for this conjunction together with Tr⁡F(Z)=Tr⁡Z\operatorname{Tr}F(Z)=\operatorname{Tr}ZTrF(Z)=TrZ for every Z∈DZ\in DZ∈D. Define H[s]=hs[H]H^{[s]}=h_s[H]H[s]=hs​[H], where hs(x)=0h_s(x)=0hs​(x)=0 at x=0x=0x=0 and hs(x)=xsh_s(x)=x^shs​(x)=xs otherwise, using real functional calculus; set SH=H[0]S_H=H^{[0]}SH​=H[0] and LH=(log⁡)[H]L_H=(\log)[H]LH​=(log)[H]. Let ΣH\Sigma_HΣH​ be the set of distinct real eigenvalues if HHH is Hermitian and the empty set otherwise, and let PH(a)P_H(a)PH​(a) be the orthogonal projection onto its aaa-eigenspace. Define CH(z)=0C_H(z)=0CH​(z)=0 for non-Hermitian HHH; for Hermitian HHH, define it in the chosen orthonormal eigenbasis by replacing each eigenvalue aaa with 000 when a=0a=0a=0, and with ezlog⁡ae^{z\log a}ezloga otherwise. For a real function fff, define

qf(H,K)=Re⁡Tr⁡ ⁣[(K[1/2])†∑a∈ΣH∑b∈ΣKf(a/b) PH(a)K[1/2]PK(b)],Df(H,K)={qf(H,K),ker⁡K⊆ker⁡H,limUnder⁡e→0+qf(H,K+e1),otherwise,\begin{aligned} q_f(H,K) &=\operatorname{Re}\operatorname{Tr}\!\left[ (K^{[1/2]})^\dagger \sum_{a\in\Sigma_H}\sum_{b\in\Sigma_K} f(a/b)\,P_H(a)K^{[1/2]}P_K(b)\right],\\ D_f(H,K) &=\begin{cases} q_f(H,K),&\ker K\subseteq\ker H,\\ \displaystyle\operatorname{limUnder}_{e\to0^+}q_f(H,K+e1),&\text{otherwise}, \end{cases} \end{aligned}qf​(H,K)Df​(H,K)​=ReTr[(K[1/2])†a∈ΣH​∑​b∈ΣK​∑​f(a/b)PH​(a)K[1/2]PK​(b)],={qf​(H,K),limUndere→0+​qf​(H,K+e1),​kerK⊆kerH,otherwise,​​

where real values are embedded in the extended reals and limUnder⁡\operatorname{limUnder}limUnder denotes the total limit-selection operation, without an additional assertion that this limit exists. Say fff satisfies G\mathcal GG when it is continuous on (0,∞)(0,\infty)(0,∞) and f(x)/xf(x)/xf(x)/x, viewed in the extended reals, converges as x→+∞x\to+\inftyx→+∞ to its limit-selection value. Say fff satisfies O\mathcal OO when, in every natural matrix dimension, including zero, all positive semidefinite H,KH,KH,K and u∈[0,1]u\in[0,1]u∈[0,1] satisfy

f[uH+(1−u)K]⪯uf[H]+(1−u)f[K].f[uH+(1-u)K]\preceq u f[H]+(1-u)f[K].f[uH+(1−u)K]⪯uf[H]+(1−u)f[K].

For the richness requirement, let f~(0)=limUnder⁡x→0+f(x)\widetilde f(0)=\operatorname{limUnder}_{x\to0^+}f(x)f​(0)=limUnderx→0+​f(x) and f~(x)=f(x)\widetilde f(x)=f(x)f​(x)=f(x) for x≠0x\ne0x=0, and let μf\mu_fμf​ be the measure on (0,∞)(0,\infty)(0,∞) selected by choice to admit real a,ba,ba,b with b≥0b\ge0b≥0, integrable (1+t)−2(1+t)^{-2}(1+t)−2, and, for every x≥0x\ge0x≥0, an integrable kernel and representation

f~(x)=f~(0)+ax+bx2+∫(0,∞)(x1+t−xx+t) dμf(t).\widetilde f(x)=\widetilde f(0)+ax+bx^2+ \int_{(0,\infty)}\left(\frac{x}{1+t}-\frac{x}{x+t}\right)\,d\mu_f(t).f​(x)=f​(0)+ax+bx2+∫(0,∞)​(1+tx​−x+tx​)dμf​(t).

If no such measure exists, the choice still supplies a measure. Write Rm(f)\mathcal R_m(f)Rm​(f) when the cardinality of the topological support of μf\mu_fμf​ is at least mmm. Define

MD(H,K)={ab: a∈ΣH, b∈ΣK, ∃Z∈D, Z≠0, HZK[−1]=(a/b)Z},m=∣MA(X,Y)∪MB(ΦX,ΦY)∣,\mathcal M_D(H,K)= \left\{\frac ab:\ a\in\Sigma_H,\ b\in\Sigma_K,\ \exists Z\in D,\ Z\ne0,\ HZK^{[-1]}=(a/b)Z\right\}, \quad m=\left|\mathcal M_A(X,Y)\cup\mathcal M_B(\Phi X,\Phi Y)\right|,MD​(H,K)={ba​: a∈ΣH​, b∈ΣK​, ∃Z∈D, Z=0, HZK[−1]=(a/b)Z},m=∣MA​(X,Y)∪MB​(ΦX,ΦY)∣,

and define the proposition C(z)\mathcal C(z)C(z) by

SYΦ† ⁣({CΦX(z)CΦY(−z),co=true,CΦY(−z)CΦX(z),co=false)=CY(−z)CX(z).S_Y\Phi^\dagger\!\left( \begin{cases} C_{\Phi X}(z)C_{\Phi Y}(-z),&\mathrm{co}=\mathrm{true},\\ C_{\Phi Y}(-z)C_{\Phi X}(z),&\mathrm{co}=\mathrm{false} \end{cases}\right) =C_Y(-z)C_X(z).SY​Φ†({CΦX​(z)CΦY​(−z),CΦY​(−z)CΦX​(z),​co=true,co=false​)=CY​(−z)CX​(z).

Then the ten returned propositions are: (0) there exists a complex-linear Ψ:MJ(C)→MI(C)\Psi:M_J(\mathbb C)\to M_I(\mathbb C)Ψ:MJ​(C)→MI​(C) satisfying Sco(B,A;Ψ)\mathsf S_{\mathrm{co}}(B,A;\Psi)Sco​(B,A;Ψ), Ψ(ΦX)=X\Psi(\Phi X)=XΨ(ΦX)=X, and Ψ(ΦY)=Y\Psi(\Phi Y)=YΨ(ΦY)=Y; (1) such a Ψ\PsiΨ exists with Sco−\mathsf S^-_{\mathrm{co}}Sco−​ in place of Sco\mathsf S_{\mathrm{co}}Sco​; (2) every real function satisfying G\mathcal GG and O\mathcal OO obeys Df(ΦX,ΦY)=Df(X,Y)D_f(\Phi X,\Phi Y)=D_f(X,Y)Df​(ΦX,ΦY)=Df​(X,Y); (3) some real function satisfying G\mathcal GG, O\mathcal OO, and Rm\mathcal R_mRm​ obeys that equality; (4) there is a set T⊆(0,∞)T\subseteq(0,\infty)T⊆(0,∞) of cardinality at least mmm such that Dft(ΦX,ΦY)=Dft(X,Y)D_{f_t}(\Phi X,\Phi Y)=D_{f_t}(X,Y)Dft​​(ΦX,ΦY)=Dft​​(X,Y) for every t∈Tt\in Tt∈T, where ft(x)=−x/(x+t)f_t(x)=-x/(x+t)ft​(x)=−x/(x+t); (5) C(z)\mathcal C(z)C(z) holds for every complex zzz; (6) C(a)\mathcal C(a)C(a) holds for some real 0<a<20<a<20<a<2 with a≠1a\ne1a=1; (7) C(it)\mathcal C(it)C(it) holds for every real ttt; (8) the identity

SYΦ† ⁣(LΦX−LΦYSΦX)=LX−LYSXS_Y\Phi^\dagger\!\left(L_{\Phi X}-L_{\Phi Y}S_{\Phi X}\right) =L_X-L_YS_XSY​Φ†(LΦX​−LΦY​SΦX​)=LX​−LY​SX​

holds; (9) writing RH=(⋅)[H]R_H=(\sqrt{\cdot})[H]RH​=(⋅​)[H] and QH=(x↦(x)−1)[H]Q_H=(x\mapsto(\sqrt x)^{-1})[H]QH​=(x↦(x​)−1)[H], the identity

RY Φ† ⁣(QΦY(ΦX)QΦY)RY=XR_Y\,\Phi^\dagger\!\left(Q_{\Phi Y}(\Phi X)Q_{\Phi Y}\right)R_Y=XRY​Φ†(QΦY​(ΦX)QΦY​)RY​=X

holds. No density, positivity, support, algebra-membership, or stochasticity assumptions on X,Y,ΦX,Y,\PhiX,Y,Φ are imposed by this definition itself. All displayed scalar operations are total: division by zero is zero, real logarithms are used also at nonpositive inputs, and zero spectral values are explicitly assigned zero in H[s]H^{[s]}H[s] and CH(z)C_H(z)CH​(z), including at exponent zero.

QAlgorithms.Papers.Independent.Support50.P12.petzRenyi

For any finite type III with decidable equality, any a∈Ra\in\mathbb Ra∈R, and any complex I×II\times II×I matrices A,BA,BA,B, let Pt(X)P_t(X)Pt​(X) denote the real continuous functional calculus of XXX with gt(0)=0g_t(0)=0gt​(0)=0 and gt(x)=xtg_t(x)=x^tgt​(x)=xt for x≠0x\ne0x=0, using total real powers. If ker⁡B⊆ker⁡A\ker B\subseteq\ker AkerB⊆kerA or a<1a<1a<1, then

petzRenyi⁡(a,A,B)=(a−1)−1 ℓ ⁣(Re⁡Tr⁡(Pa(A)P1−a(B))),\operatorname{petzRenyi}(a,A,B) =(a-1)^{-1}\,\ell\!\left(\operatorname{Re}\operatorname{Tr}\bigl(P_a(A)P_{1-a}(B)\bigr)\right),petzRenyi(a,A,B)=(a−1)−1ℓ(ReTr(Pa​(A)P1−a​(B))),

where ℓ(x)=log⁡x\ell(x)=\log xℓ(x)=logx for x>0x>0x>0 and ℓ(x)=−∞\ell(x)=-\inftyℓ(x)=−∞ for x≤0x\le0x≤0, and multiplication takes place in the extended real numbers. Otherwise the value is +∞+\infty+∞. No positivity or normalization is assumed, and a=1a=1a=1 is included: the real inverse of a−1a-1a−1 is then zero, so the supported branch evaluates to zero under extended-real multiplication, while failure of support gives +∞+\infty+∞.

QAlgorithms.Papers.Independent.Support50.P12.pnorm

For finite index types I,JI,JI,J, with decidable equality on III, and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this defines the real number

sup⁡{Re⁡∑j∈Jvj‾(Φ(1)∗Φ(1)v)j: v∈CJ, ∑j∈J∣vj∣2=1},\sqrt{\sup\left\{ \operatorname{Re}\sum_{j\in J}\overline{v_j} \bigl(\Phi(1)^*\Phi(1)v\bigr)_j :\ v\in\mathbb C^J,\ \sum_{j\in J}|v_j|^2=1 \right\}},sup⎩⎨⎧​Rej∈J∑​vj​​(Φ(1)∗Φ(1)v)j​: v∈CJ, j∈J∑​∣vj​∣2=1⎭⎬⎫​​,

where 111 is the identity matrix on III. The function is evaluated only at that identity, and no linearity or positivity is assumed. If JJJ is empty, there is no vector satisfying the normalization condition; the real supremum of the empty set used here is 000, so the defined value is 000. The type III may also be empty.

QAlgorithms.Papers.Independent.Support50.P12.polar

For any function f:R→[−∞,+∞]f:\mathbb R\to[-\infty,+\infty]f:R→[−∞,+∞] and any real number rrr, this definition returns the extended-real supremum

sup⁡s∈R(sr−f(s)),\sup_{s\in\mathbb R}\bigl(sr-f(s)\bigr),s∈Rsup​(sr−f(s)),

using extended-real subtraction after embedding the finite real number srsrsr. There are no convexity, continuity, or finiteness assumptions on fff, and no maximizing sss is asserted to exist. A value f(s)=−∞f(s)=-\inftyf(s)=−∞ contributes +∞+\infty+∞, while f(s)=+∞f(s)=+\inftyf(s)=+∞ contributes −∞-\infty−∞.

QAlgorithms.Papers.Independent.Support50.P12.psiReal

For any finite index type III with decidable equality, arbitrary complex I×II\times II×I matrices A,BA,BA,B, and real number aaa, let q=Re⁡Tr⁡(A[a]B[1−a])q=\operatorname{Re}\operatorname{Tr}(A^{[a]}B^{[1-a]})q=ReTr(A[a]B[1−a]), where H[t]H^{[t]}H[t] is real functional calculus for the scalar function equal to 000 at zero and xtx^txt elsewhere. This definition returns the natural logarithm log⁡q\log qlogq when q>0q>0q>0, and returns 000 when q≤0q\le0q≤0: in the latter case the intermediate extended-real logarithm is −∞-\infty−∞, whose conversion to a real number is 000. No positivity or normalization is assumed, and the empty index case also returns 000.

QAlgorithms.Papers.Independent.Support50.P12.quasi

For any finite type III with decidable equality, any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, and any complex I×II\times II×I matrices A,BA,BA,B, let SSS be the real continuous functional calculus of BBB with the function g(0)=0g(0)=0g(0)=0 and g(x)=x1/2g(x)=x^{1/2}g(x)=x1/2 for x≠0x\ne0x=0, using the total real power operation. Then quasi⁡(f,A,B)\operatorname{quasi}(f,A,B)quasi(f,A,B) is

Re⁡Tr⁡ ⁣(S†∑a∈SA∑b∈SBf(a/b) PA,aSPB,b),\operatorname{Re}\operatorname{Tr}\!\left( S^\dagger\sum_{a\in S_A}\sum_{b\in S_B} f(a/b)\,P_{A,a}SP_{B,b} \right),ReTr(S†a∈SA​∑​b∈SB​∑​f(a/b)PA,a​SPB,b​),

where SA,SBS_A,S_BSA​,SB​ are the distinct Hermitian eigenvalue sets, taken to be empty for non-Hermitian matrices, and PA,a,PB,bP_{A,a},P_{B,b}PA,a​,PB,b​ project orthogonally onto the corresponding eigenspaces. This uses the Hilbert–Schmidt inner product, conjugate-linear in its first argument. Zero denominators give a/0=0a/0=0a/0=0; no positivity or normalization is required. The value is zero if either matrix is non-Hermitian or III is empty.

QAlgorithms.Papers.Independent.Support50.P12.recessionSlope

For every function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, recessionSlope⁡(f)\operatorname{recessionSlope}(f)recessionSlope(f) is the value selected by the total limit operation for the extended-real-valued function x↦f(x)/xx\mapsto f(x)/xx↦f(x)/x as x→+∞x\to+\inftyx→+∞. If this limit exists, including a limit of +∞+\infty+∞ or −∞-\infty−∞, the definition returns it; otherwise the total operation supplies a fallback value. The definition itself asserts no convergence.

QAlgorithms.Papers.Independent.Support50.P12.renyiOne

For every finite index type III with decidable equality, possibly empty, and arbitrary complex square matrices X,YX,YX,Y indexed by III, define Z0aZ_0^aZ0a​ by real functional calculus with scalar value zero at zero and xax^axa otherwise, and define log⁡+(u)\log_+(u)log+​(u) to be the extended real value log⁡u\log ulogu when u>0u>0u>0, and −∞-\infty−∞ otherwise. With tr⁡Z=Re⁡Tr⁡Z\operatorname{tr}Z=\operatorname{Re}\operatorname{Tr}ZtrZ=ReTrZ, set

DaP(X,Y)={(a−1)−1log⁡+ ⁣(tr⁡(X0aY01−a)),ker⁡Y⊆ker⁡X or a<1,+∞,otherwise.D_a^{\mathrm P}(X,Y)= \begin{cases} (a-1)^{-1}\log_+\!\bigl(\operatorname{tr}(X_0^aY_0^{1-a})\bigr), &\ker Y\subseteq\ker X\ \text{or}\ a<1,\\ +\infty,&\text{otherwise}. \end{cases}DaP​(X,Y)={(a−1)−1log+​(tr(X0a​Y01−a​)),+∞,​kerY⊆kerX or a<1,otherwise.​

Then renyiOne(X,Y)\mathrm{renyiOne}(X,Y)renyiOne(X,Y) is the value selected by the total limit operation for DaP(X,Y)D_a^{\mathrm P}(X,Y)DaP​(X,Y) as a→1a\to1a→1 strictly from below. On this approach the first branch always applies, regardless of support inclusion. This is a definition, with no assertion that a limit exists and no positivity, Hermiticity, normalization, or nonzero assumptions on X,YX,YX,Y; powers at zero remain zero, scalar inversion at zero is zero, and multiplication is extended real arithmetic.

QAlgorithms.Papers.Independent.Support50.P12.representationKernel

For every real number xxx and every positive real number ttt, the representation kernel is

K(x,t)=x1+t−xx+t.K(x,t)=\frac{x}{1+t}-\frac{x}{x+t}.K(x,t)=1+tx​−x+tx​.

There is no restriction on xxx. In particular, x=−tx=-tx=−t is allowed, and the second quotient is then 000, according to the total real division convention used here.

QAlgorithms.Papers.Independent.Support50.P12.representingMeasure

For any function f:R→Rf:\mathbb R\to\mathbb Rf:R→R, this definition chooses a measure μ\muμ on (0,∞)(0,\infty)(0,∞) satisfying the following property, if such a measure exists: there are real numbers a,ba,ba,b, with b≥0b\ge0b≥0, such that (1+t)−2(1+t)^{-2}(1+t)−2 is integrable with respect to μ\muμ, and for every x≥0x\ge0x≥0, the function t↦x/(1+t)−x/(x+t)t\mapsto x/(1+t)-x/(x+t)t↦x/(1+t)−x/(x+t) is integrable and

fc(x)=fc(0)+ax+bx2+∫(0,∞)(x1+t−xx+t) dμ(t).f_c(x)=f_c(0)+ax+bx^2+\int_{(0,\infty)} \left(\frac{x}{1+t}-\frac{x}{x+t}\right)\,d\mu(t).fc​(x)=fc​(0)+ax+bx2+∫(0,∞)​(1+tx​−x+tx​)dμ(t).

Here fc(x)=f(x)f_c(x)=f(x)fc​(x)=f(x) for x≠0x\ne0x=0, while fc(0)f_c(0)fc​(0) is the totalized limit selected for f(x)f(x)f(x) as x→0+x\to0^+x→0+; existence of that limit is not assumed. The choice of μ\muμ need not be unique. If no measure has the stated property, the definition still selects a measure, without guaranteeing that property.

QAlgorithms.Papers.Independent.Support50.P12.resolventGenerator

For every pair of real numbers t,xt,xt,x, this definition assigns the real number −x/(x+t)-x/(x+t)−x/(x+t). Neither variable is restricted to be positive. Division is totalized, so when x+t=0x+t=0x+t=0, the value is 000, including at t=x=0t=x=0t=x=0.

QAlgorithms.Papers.Independent.Support50.P12.signedSchatten

For a finite index type III with decidable equality, any real number ppp, and any matrix X∈MI(C)X\in M_I(\mathbb C)X∈MI​(C), this real-valued function is

{0,X=0,(Re⁡Tr⁡ ⁣(hp[ ∣X∣ ]))1/p,X≠0,∣X∣=X†X,hp(x)={0,x=0,xp,x≠0.\begin{cases} 0,&X=0,\\[2pt] \left(\operatorname{Re}\operatorname{Tr}\!\left(h_p[\,|X|\,]\right)\right)^{1/p},&X\ne0, \end{cases} \qquad |X|=\sqrt{X^\dagger X},\qquad h_p(x)= \begin{cases} 0,&x=0,\\ x^p,&x\ne0. \end{cases}{0,(ReTr(hp​[∣X∣]))1/p,​X=0,X=0,​∣X∣=X†X​,hp​(x)={0,xp,​x=0,x=0.​

Here brackets denote the real functional calculus, and both real powers are the total real-power operation. No restriction such as p>0p>0p>0 is imposed. In particular, 1/0=01/0=01/0=0, so at p=0p=0p=0 the value is 111 for every nonzero XXX, while the zero matrix receives the separately specified value 000. Zero spectral values contribute zero to hp[∣X∣]h_p[|X|]hp​[∣X∣], including when p=0p=0p=0 or p<0p<0p<0.

QAlgorithms.Papers.Independent.Support50.P12.snorm

For finite index types I,JI,JI,J, with decidable equality on III, a unital complex star subalgebra A⊆MI(C)A\subseteq M_I(\mathbb C)A⊆MI​(C), and any function Φ:MI(C)→MJ(C)\Phi:M_I(\mathbb C)\to M_J(\mathbb C)Φ:MI​(C)→MJ​(C), this defines an extended nonnegative real number as the infimum of all real constants r≥0r\ge0r≥0 for which Φ\PhiΦ sends positive semidefinite matrices in AAA to positive semidefinite matrices and

r Φ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A.r\,\Phi(X^*X)-\Phi(X)^*\Phi(X) \quad\text{is positive semidefinite for every }X\in A.rΦ(X∗X)−Φ(X)∗Φ(X)is positive semidefinite for every X∈A.

Each admissible rrr is regarded as an element of [0,∞][0,\infty][0,∞]. If no such constant exists, the infimum is +∞+\infty+∞; attainment is not required. No linearity is assumed, and empty index types are allowed.

QAlgorithms.Papers.Independent.Support50.P12.spec

For any finite type III with decidable equality and any complex I×II\times II×I matrix XXX, spec⁡(X)\operatorname{spec}(X)spec(X) is the finite set of real eigenvalues supplied by a Hermitian spectral decomposition when XXX is Hermitian; repeated eigenvalues occur only once in this set. If XXX is not Hermitian, the set is empty. It is also empty when III is empty.

QAlgorithms.Papers.Independent.Support50.P12.supp

For any finite type III with decidable equality and any complex matrix XXX indexed by I×II\times II×I, supp⁡(X)\operatorname{supp}(X)supp(X) is the real continuous functional calculus applied to XXX with the function that is 000 at 000 and 111 at every nonzero real number. Thus, for Hermitian XXX, it is the orthogonal projection onto the sum of all nonzero eigenspaces, including eigenspaces with negative eigenvalues. For non-Hermitian XXX, this functional calculus returns the zero matrix. No positivity assumption is imposed.

QAlgorithms.Papers.Independent.Support50.P12.supportScalarPower

For all real a,xa,xa,x, supportScalarPower⁡(a,x)\operatorname{supportScalarPower}(a,x)supportScalarPower(a,x) is 000 when x=0x=0x=0, and is the total real power xax^axa otherwise. In particular, its value at x=0x=0x=0 is zero even for a=0a=0a=0 or negative aaa. Negative bases and arbitrary real exponents are included through the library’s total real power operation.

QAlgorithms.Papers.Independent.Support50.P12.transposeDefect

For a finite index type III with decidable equality, arbitrary real numbers e,ce,ce,c, and an arbitrary complex matrix XXX indexed by I×II\times II×I, this defines the matrix

Δe,c(X)=c(1−e)(X†X)T+ce Tr⁡(X†X)∣I∣1−(1−e)2(X†)TXT−e(1−e)Tr⁡X∣I∣(X†)T−e(1−e)Tr⁡(X†)∣I∣XT−e2∣Tr⁡X∣2∣I∣21.\begin{aligned} \Delta_{e,c}(X)={}&c(1-e)(X^\dagger X)^{\mathsf T} +\frac{ce\,\operatorname{Tr}(X^\dagger X)}{|I|}\mathbf 1 -(1-e)^2(X^\dagger)^{\mathsf T}X^{\mathsf T}\\ &-\frac{e(1-e)\operatorname{Tr}X}{|I|}(X^\dagger)^{\mathsf T} -\frac{e(1-e)\operatorname{Tr}(X^\dagger)}{|I|}X^{\mathsf T} -\frac{e^2|\operatorname{Tr}X|^2}{|I|^2}\mathbf 1. \end{aligned}Δe,c​(X)=​c(1−e)(X†X)T+∣I∣ceTr(X†X)​1−(1−e)2(X†)TXT−∣I∣e(1−e)TrX​(X†)T−∣I∣e(1−e)Tr(X†)​XT−∣I∣2e2∣TrX∣2​1.​

Here †\dagger† denotes conjugate transpose, T\mathsf TT ordinary transpose, and 1\mathbf 11 the identity. No positivity or normalization assumptions are imposed; empty III is included, and division by zero is 000.

QAlgorithms.Papers.Independent.Support50.P12.transposeMix

For a finite index type III with decidable equality, any real number eee, and any complex matrix XXX indexed by I×II\times II×I, this defines

Me(X)=(1−e)XT+eX,M_e(X)=(1-e)X^{\mathsf T}+eX,Me​(X)=(1−e)XT+eX,

where XTX^{\mathsf T}XT is the ordinary transpose. The parameter eee is unrestricted, including values outside [0,1][0,1][0,1], and III may be empty.

QAlgorithms.Papers.Independent.Support50.P12.transposeNoise

For a finite index type III with decidable equality, any real number eee, and any complex matrix XXX indexed by I×II\times II×I, this defines

Te(X)=(1−e)XT+e Tr⁡X∣I∣1,T_e(X)=(1-e)X^{\mathsf T}+\frac{e\,\operatorname{Tr}X}{|I|}\mathbf 1,Te​(X)=(1−e)XT+∣I∣eTrX​1,

where XTX^{\mathsf T}XT is the ordinary transpose and 1\mathbf 11 is the identity matrix. No restriction on eee, positivity, or normalization of XXX is imposed. The definition also includes empty III; division by zero then has the totalized value 000.

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).

  • Endorsed by Gohan · Oct 29, 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