Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Entropic characteristics of subsets of states

Definition
QAlgorithms_Papers_Independent_Gap48_P18

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

prove2quantum

The objects the formalized statements of Entropic characteristics of subsets of states (arXiv:quant-ph/0510073) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • stateEntropyNats: Normalized-state entropy in nats, PDF pp.4–5; infinity is retained.
  • stateRelativeEntropyNats: Umegaki relative entropy with its support guard, PDF p.4.
  • distanceSquared: Squared trace distance, PDF p.4, (1); not half trace distance.
  • meanCoefficients: Coefficients of a measure's average, PDF p.5. Only probability laws are used for barycenters; scalar integrability follows from bounded state entries.
  • measureHolevo: The integral definition of chi, PDF p.5. This extends the accepted probability-law functional to measures so finite mixtures can be constructed without extra normalization witnesses. Every theorem uses probability mass one.
  • probabilityAverage: The actual normalized average, selected from the accepted barycenter existence theorem; PDF p.5. No new existence or continuity assumption is made.
  • MeasureConcentratedOn: A law concentrated on A, PDF p.5. For closed A this is equivalent to its topological support being contained in A.
  • FiniteEnsembleOn: Conventional finite ensembles whose signals lie in A, PDF pp.5,24.
  • setChiCapacity: Per-signal set capacity, PDF p.24, (24), distinct from average constraints.
  • stateClosedConvexHull: The trace-norm closed real convex hull, PDF p.4.
  • CoefficientConvexStates: Convexity of actual operator coefficients, PDF p.4.
  • entropySup: Entropy supremum in the complete extended nonnegative reals, PDF p.7.
  • FiniteContinuousOn: Source continuity is finite-valued continuity, explicitly used on PDF p.18.
  • FiniteContinuousWithinAt: Finite-valued relative continuity at a point, PDF p.7.
  • HConverges: H-convergence has its fixed reference in the SECOND slot, PDF p.4.
  • RadiusCenter: The information-radius condition defining the unique optimal average, PDF pp.24,27. Theorem 1 establishes existence and uniqueness.
  • optimalAverage: Canonical choice by the source's defining radius condition, PDF p.27, Definition 1. On a nonempty finite-capacity set Theorem 1 makes this choice unique. Elsewhere this total helper has no claimed meaning as an optimal average: its fallback is an element of A, not an assumed radius minimizer.
  • radius: Radius about a reference state, PDF p.25, (25); the empty supremum is zero.
  • RegularityOne: First regularity alternative, PDF p.28: finite entropy at the center and entropy convergence along every H-convergent sequence in the CLOSED hull.
  • RegularityTwo: Second regularity alternative, PDF p.28; no conjunction with the first.
  • ChiRegular: Regular sets, PDF p.28, Definition 2, with the implicit nonempty domain.
  • OptimalMeasure: Optimal generalized measure, PDF p.39, Definition 3.
  • extremeClosure: Trace closure of extreme points, PDF p.40 (both overlines are essential).
  • StateChannel: A concrete CPTP map packaged with its normalized state action, PDF pp.3,35. The action is tied coefficient by coefficient to the accepted full trace-class map; no finite Kraus or finite-dimensional restriction is imposed.
  • PreservingChannels: Capacity-preserving channels mapping into the closed hull, PDF p.35.
  • CommonChannelInvariant: Common invariant states, PDF p.35; no commutativity is assumed.
  • countableLaw: Countable ensemble as its actual atomic probability law, PDF pp.39–40. Normalization of w is required at each use, and zero weights contribute zero.
  • jointMarginalConstraintSet: All joint states with marginal membership, including entangled states, PDF p.61. The two marginals are identified with actual normalized states.
  • couplingSet: Fixed-marginal coupling set C(rho,sigma), PDF p.62.
  • layer: A layer in a CHOSEN eigenbasis, PDF p.22 and footnote 10. The theorems explicitly require sigma diagonal in these coordinates, leaving repeated eigenvalues and their basis choice visible.
  • Diagonal: Chosen coordinates are an eigenbasis of sigma, PDF p.22.
  • InSimplex: Finite probability simplex, PDF p.36. The equation excludes n=0.
  • finiteLaw: The actual finite ensemble of specified states, PDF pp.5,36.
  • unionObjective: The objective whose maximum appears in Proposition 7, PDF p.36.
  • tailLayer: Layer with the first n-1 rows and columns diagonal, PDF p.59. Source indices start at one, so coordinate i corresponds to source i+1.
  • decreasingFamily: The concrete decreasing family of PDF p.61, not an arbitrary family satisfying capacity conclusions: L_(n+1)(sigma) intersect V_(sigma,c).
  • MinimalClosedEqualCapacity: Minimal closed equal-capacity subset, PDF p.30, footnote 19.
  • spectralPowerTrace: Actual spectral-power trace by finite-compression exhaustion, PDF p.13. For positive exponents finite-compression eigenvalue interlacing gives the extended trace of sigma^t. At t=0 this is the rank of the support projection, not the ambient dimension. The scalar zero-eigenvalue branch is explicit.
  • decreaseCoefficient: Decrease coefficient, PDF p.13. Extended infima avoid a real junk value when a convergence set is empty; state normalization makes t=1 admissible.
  • PositiveFiniteRankState: Finite-dimensional operator support, PDF p.13. A finite positive rank-one sum characterizes positive finite rank in any coordinate basis; this does NOT require finite coordinate support.
  • BoundedOrthogonalProjection: Any bounded orthogonal projection, PDF pp.71–73. Neither finite rank nor a monotone-sequence condition is built into this type.
  • projectionMass: Positive compression mass Tr(P rho), PDF p.71. On states the bounded conjugation is trace class, hence the real trace series is summable.
  • compressedCoefficients: The source's normalized coefficients, PDF p.71. Uses are restricted to positive mass, so inverse at zero is never asserted to produce a state.
  • projectionImage: Actual positive-domain image under Theta_P, PDF pp.71–72. Output states are identified by the formula, not supplied with assumed capacity properties.
  • projectionSuccessInfimum: Uniform retained mass, PDF p.72. The extended empty infimum is infinity; all theorem uses explicitly require A nonempty.
  • projectedAverage: Optimal average of a compression image on the eventual valid domain, PDF p.73. An empty early image uses the original set's average only to make the sequence total; Theorem 4 concludes that this branch eventually disappears.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Papers_Independent_Frontier18_P07
import Definitions.Def_QAlgorithms_Papers_Independent_Frontier18_P01

/-!
Shirokov, arXiv:quant-ph/0510073v2. Statements use countable orthonormal
coordinates for arbitrary separable complex Hilbert spaces, trace-norm topology,
and natural logarithms. All entropic quantities retain positive infinity.
Deferred full nodes are recorded in HARD.md; no partial substitute is stated.
Compilation is pending the separately authorized batch handoff.
-/

noncomputable section

open MeasureTheory Filter
open scoped ENNReal NNReal Topology Matrix.Norms.L2Operator
open QAlgorithms.Papers.Independent.Dependency50.P33
  (State PureState traceSize traceClosure IsTraceClassChannel productCoeff)
open QAlgorithms.Papers.Independent.Frontier18.P07
  (GeneralizedEnsemble HasFiniteSupport HasAverage averageCoefficients
   extendedEntropyNats extendedRelativeEntropyNats
   finiteEnsembleCapacity pureStates)
open QAlgorithms.Papers.Independent.Frontier18.P01 (traceFirst traceSecond)

namespace QAlgorithms.Papers.Independent.Gap48.P18

attribute [local instance] Classical.propDecidable Classical.decEq

/-- Normalized-state entropy in nats, PDF pp.4–5; infinity is retained. -/
def stateEntropyNats {I : Type} (rho : State I) : ℝ≥0∞ :=
  extendedEntropyNats rho.coefficients

/-- Umegaki relative entropy with its support guard, PDF p.4. -/
def stateRelativeEntropyNats {I : Type} (rho sigma : State I) : ℝ≥0∞ :=
  extendedRelativeEntropyNats rho.coefficients sigma.coefficients

/-- Squared trace distance, PDF p.4, (1); not half trace distance. -/
def distanceSquared {I : Type} (rho sigma : State I) : ℝ≥0∞ :=
  traceSize (rho.coefficients - sigma.coefficients) ^ 2

/-- Coefficients of a measure's average, PDF p.5. Only probability laws are
used for barycenters; scalar integrability follows from bounded state entries. -/
def meanCoefficients {I : Type} (mu : Measure (State I)) : Matrix I I ℂ :=
  fun i j => ∫ rho, rho.coefficients i j ∂mu

/-- The integral definition of chi, PDF p.5. This extends the accepted
probability-law functional to measures so finite mixtures can be constructed
without extra normalization witnesses. Every theorem uses probability mass one. -/
def measureHolevo {I : Type} (mu : Measure (State I)) : ℝ≥0∞ :=
  ∫⁻ rho, extendedRelativeEntropyNats rho.coefficients (meanCoefficients mu) ∂mu

/-- The actual normalized average, PDF p.5: the barycenter map whose existence is
the accepted theorem `Frontier18.P07.barycenter_continuous`, chosen under that
theorem's statement as a guard (a definition may import only proved theorems).
Whenever the statement holds this is the map that theorem selects; no new
existence or continuity assumption is made. -/
def probabilityAverage {I : Type} [Countable I] (mu : GeneralizedEnsemble I) : State I :=
  if h : ∃ average : GeneralizedEnsemble I → State I,
      (∀ mu, HasAverage mu (average mu)) ∧ Continuous average then
    Classical.choose h mu
  else
    Classical.choice (MeasureTheory.nonempty_of_isProbabilityMeasure (mu : Measure (State I)))

/-- A law concentrated on A, PDF p.5. For closed A this is equivalent to
its topological support being contained in A. -/
def MeasureConcentratedOn {I : Type} (mu : Measure (State I)) (A : Set (State I)) : Prop :=
  ∀ᵐ rho ∂mu, rho ∈ A

/-- Conventional finite ensembles whose signals lie in A, PDF pp.5,24. -/
def FiniteEnsembleOn {I : Type} (mu : GeneralizedEnsemble I)
    (A : Set (State I)) : Prop :=
  HasFiniteSupport mu ∧ (mu : Measure (State I)).support ⊆ A

/-- Per-signal set capacity, PDF p.24, (24), distinct from average constraints. -/
def setChiCapacity {I : Type} (A : Set (State I)) : ℝ≥0∞ :=
  ⨆ mu : {mu : GeneralizedEnsemble I // FiniteEnsembleOn mu A},
    measureHolevo (mu.val : Measure (State I))

/-- The trace-norm closed real convex hull, PDF p.4. -/
def stateClosedConvexHull {I : Type} (A : Set (State I)) : Set (State I) :=
  {rho | rho.coefficients ∈ traceClosure (convexHull ℝ (State.coefficients '' A))}

/-- Convexity of actual operator coefficients, PDF p.4. -/
def CoefficientConvexStates {I : Type} (A : Set (State I)) : Prop :=
  Convex ℝ (State.coefficients '' A)

/-- Entropy supremum in the complete extended nonnegative reals, PDF p.7. -/
def entropySup {I : Type} (A : Set (State I)) : ℝ≥0∞ :=
  ⨆ rho : A, stateEntropyNats rho.val

/-- Source continuity is finite-valued continuity, explicitly used on PDF p.18. -/
def FiniteContinuousOn {I : Type} (f : State I → ℝ≥0∞)
    (A : Set (State I)) : Prop :=
  (∀ rho ∈ A, f rho < ⊤) ∧ ContinuousOn f A

/-- Finite-valued relative continuity at a point, PDF p.7. -/
def FiniteContinuousWithinAt {I : Type} (f : State I → ℝ≥0∞)
    (A : Set (State I)) (rho : State I) : Prop :=
  f rho < ⊤ ∧ ContinuousWithinAt f A rho

/-- H-convergence has its fixed reference in the SECOND slot, PDF p.4. -/
def HConverges {I : Type} (seq : ℕ → State I) (rho : State I) : Prop :=
  Tendsto (fun n => stateRelativeEntropyNats (seq n) rho) atTop (𝓝 0)

/-- The information-radius condition defining the unique optimal average,
PDF pp.24,27. Theorem 1 establishes existence and uniqueness. -/
def RadiusCenter {I : Type} (A : Set (State I)) (omega : State I) : Prop :=
  ∀ rho ∈ A, stateRelativeEntropyNats rho omega ≤ setChiCapacity A

/-- Canonical choice by the source's defining radius condition, PDF p.27,
Definition 1. On a nonempty finite-capacity set Theorem 1 makes this choice
unique. Elsewhere this total helper has no claimed meaning as an optimal
average: its fallback is an element of A, not an assumed radius minimizer. -/
def optimalAverage {I : Type} (A : Set (State I)) (hA : A.Nonempty) : State I :=
  if h : ∃ omega, RadiusCenter A omega then Classical.choose h else hA.choose

/-- Radius about a reference state, PDF p.25, (25); the empty supremum is zero. -/
def radius {I : Type} (A : Set (State I)) (sigma : State I) : ℝ≥0∞ :=
  ⨆ rho : A, stateRelativeEntropyNats rho.val sigma

/-- First regularity alternative, PDF p.28: finite entropy at the center and
entropy convergence along every H-convergent sequence in the CLOSED hull. -/
def RegularityOne {I : Type} (A : Set (State I)) (omega : State I) : Prop :=
  stateEntropyNats omega < ⊤ ∧ ∀ seq : ℕ → State I,
    (∀ n, seq n ∈ stateClosedConvexHull A) → HConverges seq omega →
      Tendsto (fun n => stateEntropyNats (seq n)) atTop (𝓝 (stateEntropyNats omega))

/-- Second regularity alternative, PDF p.28; no conjunction with the first. -/
def RegularityTwo {I : Type} (A : Set (State I)) (omega : State I) : Prop :=
  FiniteContinuousOn (fun rho => stateRelativeEntropyNats rho omega) A

/-- Regular sets, PDF p.28, Definition 2, with the implicit nonempty domain. -/
def ChiRegular {I : Type} (A : Set (State I)) : Prop :=
  A.Nonempty ∧ setChiCapacity A < ⊤ ∧ ∃ omega,
    RadiusCenter A omega ∧ (RegularityOne A omega ∨ RegularityTwo A omega)

/-- Optimal generalized measure, PDF p.39, Definition 3. -/
def OptimalMeasure {I : Type} (A : Set (State I)) (mu : GeneralizedEnsemble I) : Prop :=
  MeasureConcentratedOn (mu : Measure (State I)) A ∧ measureHolevo (mu : Measure (State I)) = setChiCapacity A

/-- Trace closure of extreme points, PDF p.40 (both overlines are essential). -/
def extremeClosure {I : Type} (A : Set (State I)) : Set (State I) :=
  closure {rho | rho.coefficients ∈ (State.coefficients '' A).extremePoints ℝ}

/-- A concrete CPTP map packaged with its normalized state action, PDF pp.3,35.
The action is tied coefficient by coefficient to the accepted full trace-class
map; no finite Kraus or finite-dimensional restriction is imposed. -/
structure StateChannel (I J : Type) where
  map : Matrix I I ℂ → Matrix J J ℂ
  channel : IsTraceClassChannel map
  action : State I → State J
  action_eq : ∀ rho, (action rho).coefficients = map rho.coefficients

/-- Capacity-preserving channels mapping into the closed hull, PDF p.35. -/
def PreservingChannels {I : Type} (A : Set (State I)) : Set (StateChannel I I) :=
  {Phi | Phi.action '' A ⊆ stateClosedConvexHull A ∧ setChiCapacity (Phi.action '' A) = setChiCapacity A}

/-- Common invariant states, PDF p.35; no commutativity is assumed. -/
def CommonChannelInvariant {I : Type} (F : Set (StateChannel I I)) (rho : State I) : Prop :=
  ∀ Phi ∈ F, Phi.action rho = rho

/-- Countable ensemble as its actual atomic probability law, PDF pp.39–40.
Normalization of w is required at each use, and zero weights contribute zero. -/
def countableLaw {I : Type} (w : ℕ → ℝ≥0) (rho : ℕ → State I) : Measure (State I) :=
  Measure.sum (fun n => (w n : ℝ≥0∞) • Measure.dirac (rho n))

/-- All joint states with marginal membership, including entangled states,
PDF p.61. The two marginals are identified with actual normalized states. -/
def jointMarginalConstraintSet {I J : Type} (A : Set (State I)) (B : Set (State J)) :
    Set (State (I × J)) :=
  {omega | (∃ rho ∈ A, traceSecond omega.coefficients = rho.coefficients) ∧
    (∃ sigma ∈ B, traceFirst omega.coefficients = sigma.coefficients)}

/-- Fixed-marginal coupling set C(rho,sigma), PDF p.62. -/
def couplingSet {I J : Type} (rho : State I) (sigma : State J) :
    Set (State (I × J)) := jointMarginalConstraintSet {rho} {sigma}

/-- A layer in a CHOSEN eigenbasis, PDF p.22 and footnote 10. The theorems
explicitly require sigma diagonal in these coordinates, leaving repeated
eigenvalues and their basis choice visible. -/
def layer {I : Type} (sigma : State I) : Set (State I) :=
  {rho | ∀ i, rho.coefficients i i = sigma.coefficients i i}

/-- Chosen coordinates are an eigenbasis of sigma, PDF p.22. -/
def Diagonal {I : Type} (sigma : State I) : Prop :=
  ∀ i j, i ≠ j → sigma.coefficients i j = 0

/-- Finite probability simplex, PDF p.36. The equation excludes n=0. -/
def InSimplex {n : ℕ} (w : Fin n → ℝ≥0) : Prop := (∑ k, w k) = 1

/-- The actual finite ensemble of specified states, PDF pp.5,36. -/
def finiteLaw {I : Type} {n : ℕ} (w : Fin n → ℝ≥0) (rho : Fin n → State I) :
    Measure (State I) := ∑ k, (w k : ℝ≥0∞) • Measure.dirac (rho k)

/-- The objective whose maximum appears in Proposition 7, PDF p.36. -/
def unionObjective {I : Type} {n : ℕ} (A : Fin n → Set (State I))
    (hA : ∀ k, (A k).Nonempty) (w : Fin n → ℝ≥0) : ℝ≥0∞ :=
  (∑ k, (w k : ℝ≥0∞) * setChiCapacity (A k)) +
    measureHolevo (finiteLaw w (fun k => optimalAverage (A k) (hA k)))

/-- Layer with the first n-1 rows and columns diagonal, PDF p.59.
Source indices start at one, so coordinate i corresponds to source i+1. -/
def tailLayer (sigma : State ℕ) (n : ℕ) : Set (State ℕ) :=
  {rho | rho ∈ layer sigma ∧ ∀ i j, i ≠ j →
    (i + 1 < n ∨ j + 1 < n) → rho.coefficients i j = 0}

/-- The concrete decreasing family of PDF p.61, not an arbitrary family
satisfying capacity conclusions: L_(n+1)(sigma) intersect V_(sigma,c). -/
def decreasingFamily (sigma : State ℕ) (c : ℝ≥0) (n : ℕ) : Set (State ℕ) :=
  tailLayer sigma (n + 1) ∩ {rho | stateRelativeEntropyNats rho sigma ≤ (c : ℝ≥0∞)}

/-- Minimal closed equal-capacity subset, PDF p.30, footnote 19. -/
def MinimalClosedEqualCapacity {I : Type} (A B : Set (State I)) : Prop :=
  IsClosed B ∧ B ⊆ A ∧ setChiCapacity B = setChiCapacity A ∧
    ∀ C : Set (State I), IsClosed C → C ⊆ B → setChiCapacity C = setChiCapacity B → C = B

/-- Actual spectral-power trace by finite-compression exhaustion, PDF p.13.
For positive exponents finite-compression eigenvalue interlacing gives the
extended trace of sigma^t. At t=0 this is the rank of the support projection,
not the ambient dimension. The scalar zero-eigenvalue branch is explicit. -/
def spectralPowerTrace {I : Type} (sigma : State I) (t : ℝ) : ℝ≥0∞ :=
  ⨆ s : Finset I, ENNReal.ofReal
    ((cfc (fun x : ℝ => if x = 0 then 0 else x ^ t)
      (sigma.coefficients.submatrix (fun i : s => i.val) (fun i : s => i.val))).trace.re)

/-- Decrease coefficient, PDF p.13. Extended infima avoid a real junk value
when a convergence set is empty; state normalization makes t=1 admissible. -/
def decreaseCoefficient {I : Type} (sigma : State I) : ℝ≥0∞ :=
  ⨅ t : {t : ℝ // 0 < t ∧ spectralPowerTrace sigma t < ⊤}, ENNReal.ofReal t.val

/-- Finite-dimensional operator support, PDF p.13. A finite positive
rank-one sum characterizes positive finite rank in any coordinate basis;
this does NOT require finite coordinate support. -/
def PositiveFiniteRankState {I : Type} (rho : State I) : Prop :=
  ∃ (n : ℕ) (v : Fin n → QAlgorithms.RealRegisters.Hilbert I),
    ∀ i j, rho.coefficients i j = ∑ k, v k i * star (v k j)

/-- Any bounded orthogonal projection, PDF pp.71–73. Neither finite rank
nor a monotone-sequence condition is built into this type. -/
structure BoundedOrthogonalProjection (I : Type) where
  operator : QAlgorithms.RealRegisters.Hilbert I →L[ℂ] QAlgorithms.RealRegisters.Hilbert I
  idempotent : operator.comp operator = operator
  selfAdjoint : operator.adjoint = operator

/-- Positive compression mass Tr(P rho), PDF p.71. On states the bounded
conjugation is trace class, hence the real trace series is summable. -/
def projectionMass {I : Type} (P : BoundedOrthogonalProjection I) (rho : State I) : ℝ :=
  ∑' i, (QAlgorithms.Papers.Independent.Frontier18.P01.boundedConjugation
    P.operator rho.coefficients i i).re

/-- The source's normalized coefficients, PDF p.71. Uses are restricted to
positive mass, so inverse at zero is never asserted to produce a state. -/
def compressedCoefficients {I : Type} (P : BoundedOrthogonalProjection I)
    (rho : State I) : Matrix I I ℂ :=
  ((projectionMass P rho)⁻¹ : ℂ) •
    QAlgorithms.Papers.Independent.Frontier18.P01.boundedConjugation P.operator rho.coefficients

/-- Actual positive-domain image under Theta_P, PDF pp.71–72. Output states
are identified by the formula, not supplied with assumed capacity properties. -/
def projectionImage {I : Type} (P : BoundedOrthogonalProjection I) (A : Set (State I)) :
    Set (State I) :=
  {sigma | ∃ rho ∈ A, 0 < projectionMass P rho ∧
    sigma.coefficients = compressedCoefficients P rho}

/-- Uniform retained mass, PDF p.72. The extended empty infimum is infinity;
all theorem uses explicitly require A nonempty. -/
def projectionSuccessInfimum {I : Type} (P : BoundedOrthogonalProjection I)
    (A : Set (State I)) : ℝ≥0∞ :=
  ⨅ rho : A, ENNReal.ofReal (projectionMass P rho.val)

/-- Optimal average of a compression image on the eventual valid domain,
PDF p.73. An empty early image uses the original set's average only to make
the sequence total; Theorem 4 concludes that this branch eventually disappears. -/
def projectedAverage {I : Type} (P : BoundedOrthogonalProjection I) (A : Set (State I))
    (hA : A.Nonempty) : State I :=
  if h : (projectionImage P A).Nonempty then optimalAverage (projectionImage P A) h
    else optimalAverage A hA

/-! Sanity tests for the defining conventions; no source theorem is proved. -/

example {I : Type} : entropySup (∅ : Set (State I)) = 0 := by simp [entropySup]

example {I : Type} : measureHolevo (0 : Measure (State I)) = 0 := by simp [measureHolevo]

example {I : Type} (mu : Measure (State I)) :
    MeasureConcentratedOn mu Set.univ := by simp [MeasureConcentratedOn]

example {I : Type} (sigma : State I) : sigma ∈ layer sigma := by
  intro i
  rfl

example {I : Type} (X : Matrix I I ℂ) (i : I) : (Matrix.diagonal (Matrix.diag X)) i i = X i i := by
  simp

example {I : Type} (Phi : StateChannel I I) (rho : State I) :
    (Phi.action rho).coefficients = Phi.map rho.coefficients := Phi.action_eq rho

example {I J : Type} (rho : State I) (sigma : State J) (omega : State (I × J)) :
    omega ∈ couplingSet rho sigma ↔
      traceSecond omega.coefficients = rho.coefficients ∧
        traceFirst omega.coefficients = sigma.coefficients := by
  simp [couplingSet, jointMarginalConstraintSet]

example (t : ℝ) : (if (0 : ℝ) = 0 then 0 else (0 : ℝ) ^ t) = (0 : ℝ) := by
  simp

example {I : Type} (rho : PureState I) : PositiveFiniteRankState rho.val := by
  rcases rho.property with ⟨u, hu, heq⟩
  refine ⟨1, fun _ => u, ?_⟩
  intro i j
  simp [heq, QAlgorithms.Papers.Independent.Dependency50.P33.rankOne]

example (sigma : State ℕ) (hdiag : Diagonal sigma) (n : ℕ) :
    sigma ∈ tailLayer sigma n := by
  refine ⟨fun _ => rfl, ?_⟩
  intro i j hne hbelow
  exact hdiag i j hne

example {I : Type} [Countable I] (mu : GeneralizedEnsemble I)
    (hb : ∃ average : GeneralizedEnsemble I → State I,
      (∀ mu, HasAverage mu (average mu)) ∧ Continuous average) :
    HasAverage mu (probabilityAverage mu) := by
  unfold probabilityAverage
  rw [dif_pos hb]
  exact (Classical.choose_spec hb).1 mu

example {I : Type} [Countable I] (mu : GeneralizedEnsemble I)
    (hb : ∃ average : GeneralizedEnsemble I → State I,
      (∀ mu, HasAverage mu (average mu)) ∧ Continuous average) :
    meanCoefficients (mu : Measure (State I)) = (probabilityAverage mu).coefficients := by
  unfold probabilityAverage
  rw [dif_pos hb]
  funext i j
  exact ((Classical.choose_spec hb).1 mu i j).2.symm

example {I : Type} (A : Set (State I)) (hA : A.Nonempty)
    (h : ∃ omega, RadiusCenter A omega) : RadiusCenter A (optimalAverage A hA) := by
  unfold optimalAverage
  rw [dif_pos h]
  exact Classical.choose_spec h

example : ¬ InSimplex (fun _ : Fin 0 => (0 : ℝ≥0)) := by simp [InSimplex]

example : InSimplex (fun _ : Fin 1 => (1 : ℝ≥0)) := by simp [InSimplex]

example {I : Type} (rho : ℕ → State I) :
    countableLaw (fun _ => 0) rho = 0 := by simp [countableLaw]

example {I J : Type} (B : Set (State J)) :
    jointMarginalConstraintSet (∅ : Set (State I)) B = ∅ := by
  ext omega
  simp [jointMarginalConstraintSet]

/-! Source statement batch. -/

end QAlgorithms.Papers.Independent.Gap48.P18
Source
Entropic characteristics of subsets of states, arXiv:quant-ph/0510073v2, https://arxiv.org/abs/quant-ph/0510073v2
Read-back

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

QAlgorithms.Papers.Independent.Gap48.P18.BoundedOrthogonalProjection

For any index type III, including an empty or uncountable one, this structure consists of a bounded complex-linear operator P:ℓ2(I;C)→ℓ2(I;C)P:\ell^2(I;\mathbb C)\to\ell^2(I;\mathbb C)P:ℓ2(I;C)→ℓ2(I;C), together with the conditions P∘P=PP\circ P=PP∘P=P and P∗=PP^*=PP∗=P, where P∗P^*P∗ is its Hilbert-space adjoint. No nonzero, finite-rank, or proper-range condition is imposed; in particular, the zero and identity operators are permitted.

QAlgorithms.Papers.Independent.Gap48.P18.ChiRegular

For any index type III, let SI\mathcal S_ISI​ consist of complex arrays ρ\rhoρ with an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​, where un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, whose finite quadratic forms are real and nonnegative and whose real diagonal is summable with sum 111. Let τ(X)\tau(X)τ(X) be the infimum of these expansion costs, taking the empty infimum as ∞\infty∞, and give SI\mathcal S_ISI​ the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε}, ε>0\varepsilon>0ε>0, and its Borel measurable structure. For coordinate compressions to finite F⊆IF\subseteq IF⊆I, define

S(X)=sup⁡F finite[(Re⁡tr⁡XF)log⁡(Re⁡tr⁡XF)−Re⁡tr⁡(XFlog⁡XF)]+,D(X,Y)=sup⁡F finite{[Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+,ker⁡YF⊆ker⁡XF,∞,otherwise,S(X)=\sup_{F\text{ finite}} \bigl[(\operatorname{Re}\operatorname{tr}X_F)\log(\operatorname{Re}\operatorname{tr}X_F) -\operatorname{Re}\operatorname{tr}(X_F\log X_F)\bigr]_+, \qquad D(X,Y)=\sup_{F\text{ finite}} \begin{cases} [\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+, &\ker Y_F\subseteq\ker X_F,\\ \infty,&\text{otherwise}, \end{cases}S(X)=F finitesup​[(RetrXF​)log(RetrXF​)−Retr(XF​logXF​)]+​,D(X,Y)=F finitesup​{[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​,∞,​kerYF​⊆kerXF​,otherwise,​

using [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0), natural logarithms with log⁡0=0\log0=0log0=0, and continuous functional calculus for matrix logarithms. Set mμ,ij=∫σij dμ(σ)m_{\mu,ij}=\int\sigma_{ij}\,d\mu(\sigma)mμ,ij​=∫σij​dμ(σ) and

C(A)=sup⁡μ probability on SIsupp⁡μ finite, supp⁡μ⊆A∫SI−D(σ,mμ) dμ(σ),C(A)=\sup_{\substack{\mu\text{ probability on }\mathcal S_I\\ \operatorname{supp}\mu\text{ finite},\ \operatorname{supp}\mu\subseteq A}} \int_{\mathcal S_I}^{-}D(\sigma,m_\mu)\,d\mu(\sigma),C(A)=μ probability on SI​suppμ finite, suppμ⊆A​sup​∫SI​−​D(σ,mμ​)dμ(σ),

and let KAK_AKA​ consist of states η\etaη such that every ε>0\varepsilon>0ε>0 admits a real convex combination YYY of coefficient arrays from AAA with τ(η−Y)<ε\tau(\eta-Y)<\varepsilonτ(η−Y)<ε. The defined property of A⊆SIA\subseteq\mathcal S_IA⊆SI​ means that A≠∅A\ne\varnothingA=∅, C(A)<∞C(A)<\inftyC(A)<∞, and there exists a state ω\omegaω such that D(ρ,ω)≤C(A)D(\rho,\omega)\le C(A)D(ρ,ω)≤C(A) for every ρ∈A\rho\in Aρ∈A, together with at least one of the following two conditions: S(ω)<∞S(\omega)<\inftyS(ω)<∞ and every sequence ρn∈KA\rho_n\in K_Aρn​∈KA​ satisfying D(ρn,ω)→0D(\rho_n,\omega)\to0D(ρn​,ω)→0 also satisfies S(ρn)→S(ω)S(\rho_n)\to S(\omega)S(ρn​)→S(ω); or D(ρ,ω)<∞D(\rho,\omega)<\inftyD(ρ,ω)<∞ for every ρ∈A\rho\in Aρ∈A and ρ↦D(ρ,ω)\rho\mapsto D(\rho,\omega)ρ↦D(ρ,ω) is continuous on AAA. Limits and continuity of entropy quantities use [0,∞][0,\infty][0,∞]. The alternatives are joined by inclusive “or”; neither membership ω∈A\omega\in Aω∈A nor uniqueness is required.

QAlgorithms.Papers.Independent.Gap48.P18.CoefficientConvexStates

For any index type III and any subset AAA of its state space, this predicate says that the set of coefficient matrices of states in AAA is convex over R\mathbb RR: for every ρ,σ∈A\rho,\sigma\in Aρ,σ∈A and every pair of real numbers a,b≥0a,b\geq0a,b≥0 with a+b=1a+b=1a+b=1, there exists ω∈A\omega\in Aω∈A whose coefficient matrix is aR+bSaR+bSaR+bS, where R,SR,SR,S are the coefficient matrices of ρ,σ\rho,\sigmaρ,σ, and real scalars act through their inclusion in C\mathbb CC. A state matrix XXX here admits a representation Xij=∑nun(i)vn(j)‾X_{ij}=\sum_n u_n(i)\overline{v_n(j)}Xij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, has every finite quadratic form real and nonnegative, and has summable real diagonal with sum 111. No countability or nonemptiness hypothesis is imposed. In particular, the predicate holds vacuously for A=∅A=\varnothingA=∅, and it includes convex combinations with either coefficient equal to zero.

QAlgorithms.Papers.Independent.Gap48.P18.CommonChannelInvariant

For any type III, any set FFF of state channels on III, and any state ρ\rhoρ on III, this predicate means that Φ(ρ)=ρ\Phi(\rho)=\rhoΦ(ρ)=ρ for every Φ∈F\Phi\in FΦ∈F. Here a state on any index type TTT is a complex matrix XXX with an expansion Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, where am,bm∈ℓ2(T;C)a_m,b_m\in\ell^2(T;\mathbb C)am​,bm​∈ℓ2(T;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, with real nonnegative quadratic form ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ for every finite s⊆Ts\subseteq Ts⊆T and every z:T→Cz:T\to\mathbb Cz:T→C, and with summable real diagonal of sum 111. Each state channel has a matrix map LLL, additive and complex homogeneous on matrices admitting such expansions and preserving that class, together with a state map Φ\PhiΦ whose coefficient matrix at each state is its image under LLL; furthermore, for every natural number mmm and every state XXX on I×{0,…,m−1}I\times\{0,\ldots,m-1\}I×{0,…,m−1}, the matrix with entries L((X(i,a),(i′,b))i,i′)jj′L((X_{(i,a),(i',b)})_{i,i'})_{jj'}L((X(i,a),(i′,b)​)i,i′​)jj′​ is a state. The predicate is vacuously true when F=∅F=\varnothingF=∅; no countability assumption on III is made.

QAlgorithms.Papers.Independent.Gap48.P18.Diagonal

For any type III and any state σ\sigmaσ on III, this predicate means that σij=0\sigma_{ij}=0σij​=0 for every pair of distinct indices i,j∈Ii,j\in Ii,j∈I. Here a state is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​ with am,bm∈ℓ2(I;C)a_m,b_m\in\ell^2(I;\mathbb C)am​,bm​∈ℓ2(I;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, whose quadratic form ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ is real and nonnegative for every finite s⊆Is\subseteq Is⊆I and every z:I→Cz:I\to\mathbb Cz:I→C, and whose real diagonal is summable with sum 111. No countability assumption is made. The off-diagonal condition is vacuous when III has at most one element, although an empty III admits no state.

QAlgorithms.Papers.Independent.Gap48.P18.FiniteContinuousOn

For any index type III, let SI\mathcal S_ISI​ be the space of states: complex arrays ρ=(ρij)i,j∈I\rho=(\rho_{ij})_{i,j\in I}ρ=(ρij​)i,j∈I​ admitting an expansion ρij=∑n=0∞un(i)vn(j)‾\rho_{ij}=\sum_{n=0}^{\infty}u_n(i)\overline{v_n(j)}ρij​=∑n=0∞​un​(i)vn​(j)​, where un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, such that every finite quadratic form ∑i,j∈Fzi‾ρijzj\sum_{i,j\in F}\overline{z_i}\rho_{ij}z_j∑i,j∈F​zi​​ρij​zj​, for finite F⊆IF\subseteq IF⊆I and z:I→Cz:I\to\mathbb Cz:I→C, is real and nonnegative, and the real diagonal is summable with ∑iRe⁡ρii=1\sum_i\operatorname{Re}\rho_{ii}=1∑i​Reρii​=1. For any complex array XXX, define τ(X)∈[0,∞]\tau(X)\in[0,\infty]τ(X)∈[0,∞] as the infimum of ∑n∥un∥∥vn∥\sum_n\|u_n\|\|v_n\|∑n​∥un​∥∥vn​∥ over all expansions of this same kind representing XXX, with infimum ∞\infty∞ if there are none. Equip SI\mathcal S_ISI​ with the topology generated by the sets {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε}, for ρ∈SI\rho\in\mathcal S_Iρ∈SI​ and real ε>0\varepsilon>0ε>0. For a function f:SI→[0,∞]f:\mathcal S_I\to[0,\infty]f:SI​→[0,∞] and a subset A⊆SIA\subseteq\mathcal S_IA⊆SI​, the defined property means that f(ρ)<∞f(\rho)<\inftyf(ρ)<∞ for every ρ∈A\rho\in Aρ∈A and that fff is continuous on AAA, using the usual topology on [0,∞][0,\infty][0,∞]. If A=∅A=\varnothingA=∅, both requirements hold vacuously; no countability or nonemptiness assumption is imposed on III.

QAlgorithms.Papers.Independent.Gap48.P18.FiniteContinuousWithinAt

For any index type III, let SI\mathcal S_ISI​ consist of complex arrays σ\sigmaσ admitting an expansion σij=∑n=0∞un(i)vn(j)‾\sigma_{ij}=\sum_{n=0}^{\infty}u_n(i)\overline{v_n(j)}σij​=∑n=0∞​un​(i)vn​(j)​, with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, whose finite quadratic forms ∑i,j∈Fzi‾σijzj\sum_{i,j\in F}\overline{z_i}\sigma_{ij}z_j∑i,j∈F​zi​​σij​zj​ are real and nonnegative for every finite F⊆IF\subseteq IF⊆I and every z:I→Cz:I\to\mathbb Cz:I→C, and whose real diagonal is summable with sum 111. Write τ(X)\tau(X)τ(X) for the infimum of the costs ∑n∥un∥∥vn∥\sum_n\|u_n\|\|v_n\|∑n​∥un​∥∥vn​∥ of all such expansions of an array XXX, taking the empty infimum to be ∞\infty∞, and give SI\mathcal S_ISI​ the topology generated by {σ:τ(σ−η)<ε}\{\sigma:\tau(\sigma-\eta)<\varepsilon\}{σ:τ(σ−η)<ε}, with η∈SI\eta\in\mathcal S_Iη∈SI​ and ε>0\varepsilon>0ε>0. For f:SI→[0,∞]f:\mathcal S_I\to[0,\infty]f:SI​→[0,∞], A⊆SIA\subseteq\mathcal S_IA⊆SI​, and ρ∈SI\rho\in\mathcal S_Iρ∈SI​, the defined property is f(ρ)<∞f(\rho)<\inftyf(ρ)<∞ together with continuity of fff at ρ\rhoρ when its argument is restricted to AAA. Membership ρ∈A\rho\in Aρ∈A is not required, and no finiteness condition is imposed on other values of fff. For A=∅A=\varnothingA=∅, the continuity condition is vacuous, while f(ρ)<∞f(\rho)<\inftyf(ρ)<∞ remains required.

QAlgorithms.Papers.Independent.Gap48.P18.FiniteEnsembleOn

For any index type III, any Borel probability measure μ\muμ on its state space, and any subset AAA of that space, this predicate means that the topological support of μ\muμ is a finite set and is contained in AAA. States are complex matrices admitting expansions Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​, where an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, having real nonnegative finite quadratic forms and summable real diagonal of sum 111. The topology is generated by {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, for states QQQ and ε>0\varepsilon>0ε>0, where τ(X)\tau(X)τ(X) is the infimum of the expansion sums for XXX, with empty infimum +∞+\infty+∞. Support means the set of states every open neighborhood of which has positive measure. Thus finiteness concerns this topological support, and containment requires every point of that support to lie in AAA. Neither countability of III nor measurability of AAA is assumed; when A=∅A=\varnothingA=∅, the predicate requires empty support.

QAlgorithms.Papers.Independent.Gap48.P18.HConverges

For any index type III, a state is a complex array XXX admitting a countable expansion Xij=∑nun(i)vn(j)‾X_{ij}=\sum_nu_n(i)\overline{v_n(j)}Xij​=∑n​un​(i)vn​(j)​, with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, such that every finite quadratic form ∑i,j∈Fzi‾Xijzj\sum_{i,j\in F}\overline{z_i}X_{ij}z_j∑i,j∈F​zi​​Xij​zj​ is real and nonnegative and the real diagonal is summable with sum 111. For finite F⊆IF\subseteq IF⊆I, write XF=(Xij)i,j∈FX_F=(X_{ij})_{i,j\in F}XF​=(Xij​)i,j∈F​. Define

D(X,Y)=sup⁡F⊆IF finite{[Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+,ker⁡YF⊆ker⁡XF,∞,otherwise,D(X,Y)=\sup_{\substack{F\subseteq I\\F\text{ finite}}} \begin{cases} \bigl[\operatorname{Re}\operatorname{tr} (X_F\log X_F-X_F\log Y_F+Y_F-X_F)\bigr]_+, &\ker Y_F\subseteq\ker X_F,\\ \infty,&\text{otherwise}, \end{cases}D(X,Y)=F⊆IF finite​sup​{[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​,∞,​kerYF​⊆kerXF​,otherwise,​

where [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0) and the matrix logarithms are obtained by continuous functional calculus from the natural logarithm, with its total convention log⁡0=0\log 0=0log0=0. For a sequence of states (ρn)n∈N(\rho_n)_{n\in\mathbb N}(ρn​)n∈N​ and a state ρ\rhoρ, the defined property means precisely that D(ρn,ρ)→0D(\rho_n,\rho)\to0D(ρn​,ρ)→0 as n→∞n\to\inftyn→∞, in [0,∞][0,\infty][0,∞]. The order of the two arguments is as displayed, and no separate convergence of the coefficient arrays is included in the definition.

QAlgorithms.Papers.Independent.Gap48.P18.InSimplex

For every natural number nnn, including 000, and every family (wk)0≤k<n(w_k)_{0\le k<n}(wk​)0≤k<n​ of finite nonnegative real numbers, this predicate means exactly

∑k=0n−1wk=1.\sum_{k=0}^{n-1}w_k=1.k=0∑n−1​wk​=1.

Nonnegativity is already part of the weights’ type. When n=0n=0n=0, the sum is 000, so the predicate is false.

QAlgorithms.Papers.Independent.Gap48.P18.MeasureConcentratedOn

For any index type III, any measure μ\muμ on its state space, and any subset AAA of that state space, this predicate means

μ({R:R∉A})=0,\mu\bigl(\{R:R\notin A\}\bigr)=0,μ({R:R∈/A})=0,

equivalently, R∈AR\in AR∈A for μ\muμ-almost every state RRR. States here are matrices admitting expansions Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and summable ∥an∥∥bn∥\|a_n\|\|b_n\|∥an​∥∥bn​∥, having real nonnegative finite quadratic forms, and having summable real diagonal with sum 111. The measure is on the Borel space for the topology generated by {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, where QQQ is a state, ε>0\varepsilon>0ε>0, and τ\tauτ is the infimum of the expansion sums for its argument, with empty infimum +∞+\infty+∞. No measurability assumption on AAA, probability assumption on μ\muμ, or countability assumption on III is required. The zero measure satisfies the predicate for every AAA, including A=∅A=\varnothingA=∅; for A=∅A=\varnothingA=∅, the predicate requires that the whole state space have measure zero.

QAlgorithms.Papers.Independent.Gap48.P18.MinimalClosedEqualCapacity

For any index type III, possibly empty or uncountable, and sets A,BA,BA,B of states on III, this predicate says that BBB is closed, B⊆AB\subseteq AB⊆A, its capacity equals that of AAA, and every closed subset C⊆BC\subseteq BC⊆B with the same capacity as BBB is equal to BBB. Here a state has a complex coefficient matrix RRR admitting an expansion Rij=∑n=0∞an(i)bn(j)‾R_{ij}=\sum_{n=0}^{\infty}a_n(i)\overline{b_n(j)}Rij​=∑n=0∞​an​(i)bn​(j)​, with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞; every finite quadratic form ∑i,j∈szi‾Rijzj\sum_{i,j\in s}\overline{z_i}R_{ij}z_j∑i,j∈s​zi​​Rij​zj​ is real and nonnegative; and the real diagonal entries are summable with sum 111. Define τ(X)\tau(X)τ(X) to be the infimum, over all such expansions of XXX, of max⁡{∑n∥an∥∥bn∥,0}\max\{\sum_n\|a_n\|\|b_n\|,0\}max{∑n​∥an​∥∥bn​∥,0}, regarded in [0,∞][0,\infty][0,∞]; an empty infimum is ∞\infty∞. Closedness uses the topology generated by the sets {S:τ(S−R)<ε}\{S:\tau(S-R)<\varepsilon\}{S:τ(S−R)<ε}, for states RRR and real ε>0\varepsilon>0ε>0. To specify the capacity, for finite matrices X,YX,YX,Y put

d(X,Y)={max⁡{Re⁡tr⁡(Xlog⁡X−Xlog⁡Y+Y−X),0},ker⁡Y⊆ker⁡X,∞,otherwise,D(X,Y)=sup⁡s⊆I, s finited(X∣s×s,Y∣s×s),d(X,Y)= \begin{cases} \max\{\operatorname{Re}\operatorname{tr}(X\log X-X\log Y+Y-X),0\}, &\ker Y\subseteq\ker X,\\ \infty,&\text{otherwise}, \end{cases} \qquad D(X,Y)=\sup_{s\subseteq I,\ s\text{ finite}}d(X|_{s\times s},Y|_{s\times s}),d(X,Y)={max{Retr(XlogX−XlogY+Y−X),0},∞,​kerY⊆kerX,otherwise,​D(X,Y)=s⊆I, s finitesup​d(X∣s×s​,Y∣s×s​),

where matrix logarithms use the continuous functional calculus with the total real logarithm, for which log⁡0=0\log0=0log0=0. For a Borel probability measure μ\muμ on the state space, set (R‾μ)ij=∫Rij dμ(R)(\overline R_\mu)_{ij}=\int R_{ij}\,d\mu(R)(Rμ​)ij​=∫Rij​dμ(R) and χ(μ)=∫−D(R,R‾μ) dμ(R)\chi(\mu)=\int^-D(R,\overline R_\mu)\,d\mu(R)χ(μ)=∫−D(R,Rμ​)dμ(R); the coordinate integrals are the total complex integrals, equal to zero when the corresponding function is not integrable. The capacity of a set EEE is sup⁡μχ(μ)\sup_\mu\chi(\mu)supμ​χ(μ), over probability measures whose topological support is finite and contained in EEE. Thus the asserted minimality concerns all closed subsets CCC of BBB, including the empty set. No nonemptiness or finite-capacity condition is imposed; the capacity of the empty set is 000.

QAlgorithms.Papers.Independent.Gap48.P18.OptimalMeasure

For any index type III, let SI\mathcal S_ISI​ be the space of complex arrays ρ\rhoρ admitting an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, whose finite quadratic forms are real and nonnegative and whose real diagonal is summable with sum 111. Let τ(X)\tau(X)τ(X) be the infimum of these expansion costs for XXX, with empty infimum ∞\infty∞; use the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε}, ε>0\varepsilon>0ε>0, and its Borel measurable structure. For arrays X,YX,YX,Y, define D(X,Y)D(X,Y)D(X,Y) as the supremum over finite coordinate compressions F⊆IF\subseteq IF⊆I of [Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+[\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​ when ker⁡YF⊆ker⁡XF\ker Y_F\subseteq\ker X_FkerYF​⊆kerXF​, and ∞\infty∞ otherwise, using [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0) and natural logarithms through continuous functional calculus. For any measure ν\nuν on SI\mathcal S_ISI​, put

mν,ij=∫σij dν(σ),χ(ν)=∫SI−D(σ,mν) dν(σ),C(A)=sup⁡ν probabilitysupp⁡ν finite, supp⁡ν⊆Aχ(ν).m_{\nu,ij}=\int\sigma_{ij}\,d\nu(\sigma), \qquad \chi(\nu)=\int_{\mathcal S_I}^{-}D(\sigma,m_\nu)\,d\nu(\sigma), \qquad C(A)=\sup_{\substack{\nu\text{ probability}\\ \operatorname{supp}\nu\text{ finite},\ \operatorname{supp}\nu\subseteq A}}\chi(\nu).mν,ij​=∫σij​dν(σ),χ(ν)=∫SI​−​D(σ,mν​)dν(σ),C(A)=ν probabilitysuppν finite, suppν⊆A​sup​χ(ν).

For A⊆SIA\subseteq\mathcal S_IA⊆SI​ and a probability measure μ\muμ on SI\mathcal S_ISI​, the defined property means that ρ∈A\rho\in Aρ∈A for μ\muμ-almost every ρ\rhoρ and that χ(μ)=C(A)\chi(\mu)=C(A)χ(μ)=C(A). It requires neither finite support of μ\muμ nor containment of its topological support in AAA, and it imposes no finiteness requirement on the common value of this equality. No measurability assumption on AAA is explicitly included.

QAlgorithms.Papers.Independent.Gap48.P18.PositiveFiniteRankState

For any index type III and state ρ\rhoρ with coefficient matrix RRR, this predicate says that there exist a natural number nnn and vectors v0,…,vn−1∈ℓ2(I;C)v_0,\ldots,v_{n-1}\in\ell^2(I;\mathbb C)v0​,…,vn−1​∈ℓ2(I;C) such that

Rij=∑k=0n−1vk(i)vk(j)‾for every i,j∈I.R_{ij}=\sum_{k=0}^{n-1}v_k(i)\overline{v_k(j)} \quad\text{for every }i,j\in I.Rij​=k=0∑n−1​vk​(i)vk​(j)​for every i,j∈I.

The state argument already requires a countable expansion Rij=∑mam(i)bm(j)‾R_{ij}=\sum_m a_m(i)\overline{b_m(j)}Rij​=∑m​am​(i)bm​(j)​ with ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, real nonnegative quadratic forms on every finite coordinate subset, and summable real diagonal entries of total sum 111. The finitely many vectors vkv_kvk​ have no individual normalization, orthogonality, or nonzero requirement. The quantified value n=0n=0n=0 gives the zero matrix through the empty sum and therefore cannot represent a state with diagonal sum 111. No finiteness or countability assumption on III is made.

QAlgorithms.Papers.Independent.Gap48.P18.PreservingChannels

For any type III and set AAA of states on III, this defines the set of state channels Φ\PhiΦ such that every Φ(ρ)\Phi(\rho)Φ(ρ), for ρ∈A\rho\in Aρ∈A, belongs to the closure described below of the real convex hull of AAA, and C(Φ(A))=C(A)C(\Phi(A))=C(A)C(Φ(A))=C(A). Here a state on any index type TTT is a complex matrix XXX admitting an expansion Xij=∑m=0∞am(i)bm(j)‾X_{ij}=\sum_{m=0}^{\infty}a_m(i)\overline{b_m(j)}Xij​=∑m=0∞​am​(i)bm​(j)​, where am,bm∈ℓ2(T;C)a_m,b_m\in\ell^2(T;\mathbb C)am​,bm​∈ℓ2(T;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, whose quadratic form ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ is real and nonnegative for every finite s⊆Ts\subseteq Ts⊆T and every z:T→Cz:T\to\mathbb Cz:T→C, and whose real diagonal entries form a summable family with sum 111. A state channel consists of a matrix map LLL and a state map Φ\PhiΦ with coefficients Φ(ρ)=L(ρ)\Phi(\rho)=L(\rho)Φ(ρ)=L(ρ): LLL preserves matrices admitting the stated expansions, is additive and complex homogeneous on those matrices, and, for every natural number mmm, sends every state XXX on I×{0,…,m−1}I\times\{0,\ldots,m-1\}I×{0,…,m−1} to a state by the rule Y(j,a),(j′,b)=L((X(i,a),(i′,b))i,i′)jj′Y_{(j,a),(j',b)}=L((X_{(i,a),(i',b)})_{i,i'})_{jj'}Y(j,a),(j′,b)​=L((X(i,a),(i′,b)​)i,i′​)jj′​. The case m=0m=0m=0 has no such input states. Define τ(Z)∈[0,∞]\tau(Z)\in[0,\infty]τ(Z)∈[0,∞] as the infimum, over all stated expansions of ZZZ, of ∑m∥am∥∥bm∥\sum_m\|a_m\|\|b_m\|∑m​∥am​∥∥bm​∥, with infimum +∞+\infty+∞ if there is no expansion. The closure requirement means that for every ρ∈A\rho\in Aρ∈A and every ε>0\varepsilon>0ε>0, there is a finite real convex combination YYY of coefficient matrices of states in AAA such that τ(Φ(ρ)−Y)<ε\tau(\Phi(\rho)-Y)<\varepsilonτ(Φ(ρ)−Y)<ε. To define C(B)C(B)C(B), equip states with the Borel measurable structure of the topology generated by the sets {X:τ(X−R)<ε}\{X:\tau(X-R)<\varepsilon\}{X:τ(X−R)<ε}, for states RRR and ε>0\varepsilon>0ε>0; then C(B)C(B)C(B) is the supremum of

∫D(X,Mμ) dμ(X),(Mμ)ij=∫Xij dμ(X),\int D(X,M_\mu)\,d\mu(X), \qquad (M_\mu)_{ij}=\int X_{ij}\,d\mu(X),∫D(X,Mμ​)dμ(X),(Mμ​)ij​=∫Xij​dμ(X),

over probability measures μ\muμ with finite topological support contained in BBB. These outer integrals are nonnegative extended integrals, and the coordinate integrals defining MμM_\muMμ​ are complex Bochner integrals. For arbitrary matrices X,YX,YX,Y, D(X,Y)D(X,Y)D(X,Y) is the supremum over all finite s⊆Is\subseteq Is⊆I, including s=∅s=\varnothings=∅, of

d(Xs,Ys)={max⁡ ⁣{Re⁡tr⁡(Xslog⁡Xs−Xslog⁡Ys+Ys−Xs),0},ker⁡Ys⊆ker⁡Xs,+∞,otherwise,d(X_s,Y_s)= \begin{cases} \max\!\left\{\operatorname{Re}\operatorname{tr} \bigl(X_s\log X_s-X_s\log Y_s+Y_s-X_s\bigr),0\right\}, &\ker Y_s\subseteq\ker X_s,\\ +\infty,&\text{otherwise}, \end{cases}d(Xs​,Ys​)={max{Retr(Xs​logXs​−Xs​logYs​+Ys​−Xs​),0},+∞,​kerYs​⊆kerXs​,otherwise,​

where Xs,YsX_s,Y_sXs​,Ys​ are coordinate restrictions and matrix logarithms use continuous functional calculus with the total real logarithm, including log⁡0=0\log 0=0log0=0. No countability, nonemptiness, or finite-capacity assumption is imposed; an empty supremum defining CCC is 000, and capacities may be infinite.

QAlgorithms.Papers.Independent.Gap48.P18.RadiusCenter

For any index type III, let SI\mathcal S_ISI​ consist of complex arrays ρ\rhoρ admitting an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, with all finite quadratic forms real and nonnegative and with summable real diagonal of sum 111. Let τ(X)\tau(X)τ(X) be the infimum of these expansion costs for XXX, with empty infimum ∞\infty∞, and give SI\mathcal S_ISI​ the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε} for ε>0\varepsilon>0ε>0, together with its Borel measurable structure. For arrays X,YX,YX,Y, define D(X,Y)D(X,Y)D(X,Y) as the supremum over finite F⊆IF\subseteq IF⊆I of [Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+[\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​ when ker⁡YF⊆ker⁡XF\ker Y_F\subseteq\ker X_FkerYF​⊆kerXF​, and ∞\infty∞ otherwise; here XF,YFX_F,Y_FXF​,YF​ are coordinate compressions, [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0), and logarithms use continuous functional calculus of the natural logarithm. For a probability measure μ\muμ on SI\mathcal S_ISI​, put mμ,ij=∫σij dμ(σ)m_{\mu,ij}=\int\sigma_{ij}\,d\mu(\sigma)mμ,ij​=∫σij​dμ(σ) and

χ(μ)=∫SI−D(σ,mμ) dμ(σ),C(A)=sup⁡μ probabilitysupp⁡μ finitesupp⁡μ⊆Aχ(μ).\chi(\mu)=\int_{\mathcal S_I}^{-}D(\sigma,m_\mu)\,d\mu(\sigma), \qquad C(A)=\sup_{\substack{\mu\text{ probability}\\ \operatorname{supp}\mu\text{ finite}\\ \operatorname{supp}\mu\subseteq A}}\chi(\mu).χ(μ)=∫SI​−​D(σ,mμ​)dμ(σ),C(A)=μ probabilitysuppμ finitesuppμ⊆A​sup​χ(μ).

For A⊆SIA\subseteq\mathcal S_IA⊆SI​ and ω∈SI\omega\in\mathcal S_Iω∈SI​, the defined property is ∀ρ∈A, D(ρ,ω)≤C(A)\forall\rho\in A,\ D(\rho,\omega)\le C(A)∀ρ∈A, D(ρ,ω)≤C(A). It does not require ω∈A\omega\in Aω∈A or uniqueness of ω\omegaω. For empty AAA, C(A)=0C(A)=0C(A)=0 and the property is vacuous; if C(A)=∞C(A)=\inftyC(A)=∞, every state ω\omegaω satisfies the inequality.

QAlgorithms.Papers.Independent.Gap48.P18.RegularityOne

For any index type III, let SI\mathcal S_ISI​ consist of complex arrays ρ\rhoρ admitting an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​, with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, having real, nonnegative finite quadratic forms and a summable real diagonal of sum 111. Write τ(X)\tau(X)τ(X) for the infimum of such expansion costs for XXX, with empty infimum ∞\infty∞. For finite F⊆IF\subseteq IF⊆I, write XFX_FXF​ for coordinate compression and define

S(X)=sup⁡F finite[(Re⁡tr⁡XF)log⁡(Re⁡tr⁡XF)−Re⁡tr⁡(XFlog⁡XF)]+,D(X,Y)=sup⁡F finite{[Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+,ker⁡YF⊆ker⁡XF,∞,otherwise.S(X)=\sup_{F\text{ finite}} \bigl[(\operatorname{Re}\operatorname{tr}X_F)\log(\operatorname{Re}\operatorname{tr}X_F) -\operatorname{Re}\operatorname{tr}(X_F\log X_F)\bigr]_+, \quad D(X,Y)=\sup_{F\text{ finite}} \begin{cases} [\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+, &\ker Y_F\subseteq\ker X_F,\\ \infty,&\text{otherwise}. \end{cases}S(X)=F finitesup​[(RetrXF​)log(RetrXF​)−Retr(XF​logXF​)]+​,D(X,Y)=F finitesup​{[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​,∞,​kerYF​⊆kerXF​,otherwise.​

Here [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0), logarithms are natural, matrix logarithms use continuous functional calculus, and log⁡0=0\log0=0log0=0. For A⊆SIA\subseteq\mathcal S_IA⊆SI​, let KAK_AKA​ be the states η\etaη such that for every real ε>0\varepsilon>0ε>0 there exists YYY in the real convex hull of the coefficient arrays of members of AAA with τ(η−Y)<ε\tau(\eta-Y)<\varepsilonτ(η−Y)<ε. For a state ω\omegaω, the defined property means that S(ω)<∞S(\omega)<\inftyS(ω)<∞ and, for every sequence (ρn)n∈N(\rho_n)_{n\in\mathbb N}(ρn​)n∈N​ of states with ρn∈KA\rho_n\in K_Aρn​∈KA​ for all nnn, the implication D(ρn,ω)→0⇒S(ρn)→S(ω)D(\rho_n,\omega)\to0\Rightarrow S(\rho_n)\to S(\omega)D(ρn​,ω)→0⇒S(ρn​)→S(ω) holds, with both limits taken in [0,∞][0,\infty][0,∞]. It does not require ω∈A\omega\in Aω∈A or ω∈KA\omega\in K_Aω∈KA​. If A=∅A=\varnothingA=∅, then KA=∅K_A=\varnothingKA​=∅, so the sequence condition is vacuous and only S(ω)<∞S(\omega)<\inftyS(ω)<∞ remains.

QAlgorithms.Papers.Independent.Gap48.P18.RegularityTwo

For any index type III, states are complex arrays ρ\rhoρ admitting an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​, where un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, with every finite quadratic form real and nonnegative and with summable real diagonal of sum 111. Let τ(X)\tau(X)τ(X) be the infimum of these expansion costs for XXX, with empty infimum ∞\infty∞, and equip the state space SI\mathcal S_ISI​ with the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε} for ε>0\varepsilon>0ε>0. For finite coordinate compressions XF,YFX_F,Y_FXF​,YF​, define D(X,Y)D(X,Y)D(X,Y) as the supremum over finite F⊆IF\subseteq IF⊆I of [Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+[\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​ if ker⁡YF⊆ker⁡XF\ker Y_F\subseteq\ker X_FkerYF​⊆kerXF​, and ∞\infty∞ otherwise, where [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0) and matrix logarithms use continuous functional calculus of the natural logarithm. For A⊆SIA\subseteq\mathcal S_IA⊆SI​ and ω∈SI\omega\in\mathcal S_Iω∈SI​, the defined property means that D(ρ,ω)<∞D(\rho,\omega)<\inftyD(ρ,ω)<∞ for every ρ∈A\rho\in Aρ∈A and that ρ↦D(ρ,ω)\rho\mapsto D(\rho,\omega)ρ↦D(ρ,ω), valued in [0,∞][0,\infty][0,∞], is continuous on AAA. Membership ω∈A\omega\in Aω∈A and finiteness of the entropy of ω\omegaω are not required; for empty AAA, the property is vacuous.

QAlgorithms.Papers.Independent.Gap48.P18.StateChannel

For arbitrary index types I,JI,JI,J, call a complex array XXX on an index type KKK trace class when it admits sequences un,vn∈ℓ2(K;C)u_n,v_n\in\ell^2(K;\mathbb C)un​,vn​∈ℓ2(K;C) such that ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞ and Xij=∑nun(i)vn(j)‾X_{ij}=\sum_nu_n(i)\overline{v_n(j)}Xij​=∑n​un​(i)vn​(j)​. A state on KKK is such an array for which every finite quadratic form ∑i,j∈Fzi‾Xijzj\sum_{i,j\in F}\overline{z_i}X_{ij}z_j∑i,j∈F​zi​​Xij​zj​, for finite F⊆KF\subseteq KF⊆K and z:K→Cz:K\to\mathbb Cz:K→C, is real and nonnegative, and whose real diagonal is summable with sum 111. The structure consists of a function Φ:CI×I→CJ×J\Phi:\mathbb C^{I\times I}\to\mathbb C^{J\times J}Φ:CI×I→CJ×J, proofs that it maps every trace-class array to a trace-class array, satisfies Φ(X+Y)=Φ(X)+Φ(Y)\Phi(X+Y)=\Phi(X)+\Phi(Y)Φ(X+Y)=Φ(X)+Φ(Y) for trace-class X,YX,YX,Y, and satisfies Φ(cX)=cΦ(X)\Phi(cX)=c\Phi(X)Φ(cX)=cΦ(X) for every c∈Cc\in\mathbb Cc∈C and trace-class XXX, together with the requirement that for every n∈Nn\in\mathbb Nn∈N and every state RRR on I×{0,…,n−1}I\times\{0,\ldots,n-1\}I×{0,…,n−1}, the array

T(j,a),(j′,b)=Φ((R(i,a),(i′,b))i,i′∈I)j,j′T_{(j,a),(j',b)} =\Phi\bigl((R_{(i,a),(i',b)})_{i,i'\in I}\bigr)_{j,j'}T(j,a),(j′,b)​=Φ((R(i,a),(i′,b)​)i,i′∈I​)j,j′​

is a state on J×{0,…,n−1}J\times\{0,\ldots,n-1\}J×{0,…,n−1}; it also contains a function a:SI→SJa:\mathcal S_I\to\mathcal S_Ja:SI​→SJ​ and a proof that the coefficient array of a(ρ)a(\rho)a(ρ) equals Φ(ρ)\Phi(\rho)Φ(ρ) for every state ρ\rhoρ. The additivity and homogeneity requirements apply only to trace-class inputs. No countability or nonemptiness of I,JI,JI,J is assumed, and the n=0n=0n=0 requirement is vacuous because there are no states on the empty index type.

QAlgorithms.Papers.Independent.Gap48.P18.compressedCoefficients

For any index type III, bounded complex-linear operator PPP on ℓ2(I;C)\ell^2(I;\mathbb C)ℓ2(I;C) with P2=P=P∗P^2=P=P^*P2=P=P∗, and state ρ\rhoρ with coefficient matrix RRR, define

Kij=∑a∈I′∑b∈I′(Pea)(i)Rab(Peb)(j)‾,p=∑i∈I′Re⁡Kii.K_{ij}=\sum_{a\in I}'\sum_{b\in I}'(Pe_a)(i)R_{ab}\overline{(Pe_b)(j)}, \qquad p=\sum_{i\in I}'\operatorname{Re}K_{ii}.Kij​=a∈I∑′​b∈I∑′​(Pea​)(i)Rab​(Peb​)(j)​,p=i∈I∑′​ReKii​.

This declaration returns the complex matrix with entries p−1Kijp^{-1}K_{ij}p−1Kij​, where the real inverse is then embedded into C\mathbb CC, and eae_aea​ is the coordinate unit vector at aaa. A state RRR admits an expansion Rij=∑nun(i)vn(j)‾R_{ij}=\sum_n u_n(i)\overline{v_n(j)}Rij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, has real nonnegative quadratic forms on every finite coordinate subset, and has summable real diagonal entries of sum 111. The definition requires neither p>0p>0p>0 nor p≠0p\ne0p=0; under the total inverse convention 0−1=00^{-1}=00−1=0, it returns the zero matrix when p=0p=0p=0. It returns a matrix without separately asserting that this matrix is a state. Infinite sums are total sums, equal to zero for nonsummable families, and no countability assumption on III is imposed.

QAlgorithms.Papers.Independent.Gap48.P18.countableLaw

For any type III, any sequence of finite nonnegative real weights (wn)n∈N(w_n)_{n\in\mathbb N}(wn​)n∈N​, and any sequence of states (ρn)n∈N(\rho_n)_{n\in\mathbb N}(ρn​)n∈N​ on III, this defines the measure

μ=∑n=0∞wnδρn.\mu=\sum_{n=0}^{\infty}w_n\delta_{\rho_n}.μ=n=0∑∞​wn​δρn​​.

Thus, for every measurable set EEE of states, μ(E)=∑nwn1E(ρn)\mu(E)=\sum_n w_n\mathbf 1_E(\rho_n)μ(E)=∑n​wn​1E​(ρn​), with the sum taken in [0,∞][0,\infty][0,∞]; repeated states contribute repeatedly. A state is a complex matrix XXX admitting an expansion Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​ with am,bm∈ℓ2(I;C)a_m,b_m\in\ell^2(I;\mathbb C)am​,bm​∈ℓ2(I;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, having real nonnegative quadratic forms on every finite coordinate set, and having summable real diagonal of sum 111. The measurable structure is Borel for the topology generated by {X:τ(X−R)<ε}\{X:\tau(X-R)<\varepsilon\}{X:τ(X−R)<ε}, where RRR is a state, ε>0\varepsilon>0ε>0, and τ(Z)\tau(Z)τ(Z) is the infimum of the expansion costs ∑m∥am∥∥bm∥\sum_m\|a_m\|\|b_m\|∑m​∥am​∥∥bm​∥ for ZZZ, or +∞+\infty+∞ if there is no expansion. The weights are not required to be summable or to sum to 111: the total mass is ∑nwn\sum_n w_n∑n​wn​, which may be infinite, and all-zero weights give the zero measure.

QAlgorithms.Papers.Independent.Gap48.P18.couplingSet

For any types I,JI,JI,J and states ρ\rhoρ on III and σ\sigmaσ on JJJ, this defines the set of states ω\omegaω on I×JI\times JI×J satisfying exactly

∑r∈Jω(i,r),(j,r)=ρij(i,j∈I),∑r∈Iω(r,a),(r,b)=σab(a,b∈J).\sum_{r\in J}\omega_{(i,r),(j,r)}=\rho_{ij} \quad(i,j\in I), \qquad \sum_{r\in I}\omega_{(r,a),(r,b)}=\sigma_{ab} \quad(a,b\in J).r∈J∑​ω(i,r),(j,r)​=ρij​(i,j∈I),r∈I∑​ω(r,a),(r,b)​=σab​(a,b∈J).

The sums use the total complex infinite-sum operation, with value 000 on nonsummable families, and no additional summability assumptions are stated. A state on any index type TTT is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​ for sequences am,bm∈ℓ2(T;C)a_m,b_m\in\ell^2(T;\mathbb C)am​,bm​∈ℓ2(T;C) with ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, having real nonnegative quadratic forms on every finite coordinate set, and having summable real diagonal of sum 111. The definition makes no countability assumption and does not assert uniqueness of ω\omegaω.

QAlgorithms.Papers.Independent.Gap48.P18.decreaseCoefficient

For any index type III and state σ\sigmaσ on III, this defines

inf⁡{t∈(0,∞):sup⁡s⊆I, s finitemax⁡ ⁣{Re⁡tr⁡(ft(σ∣s×s)),0}<∞}\inf\left\{t\in(0,\infty): \sup_{s\subseteq I,\ s\text{ finite}} \max\!\left\{\operatorname{Re}\operatorname{tr}\bigl(f_t(\sigma|_{s\times s})\bigr),0\right\}<\infty\right\}inf{t∈(0,∞):s⊆I, s finitesup​max{Retr(ft​(σ∣s×s​)),0}<∞}

as an element of [0,∞][0,\infty][0,∞], where ft(0)=0f_t(0)=0ft​(0)=0, ft(x)=xtf_t(x)=x^tft​(x)=xt for x≠0x\ne0x=0, and the matrix function is obtained by continuous functional calculus. A state is a complex matrix RRR admitting Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, having real nonnegative finite quadratic forms, and having summable real diagonal entries with sum 111. Each eligible ttt is embedded into [0,∞][0,\infty][0,∞] by taking its nonnegative part, which is just ttt because t>0t>0t>0. If there are no eligible exponents, the infimum is ∞\infty∞; an infimum of eligible positive exponents can be 000 without 000 itself being eligible. The finite subsets include the empty subset, and no countability assumption on III is imposed.

QAlgorithms.Papers.Independent.Gap48.P18.decreasingFamily

For any state σ\sigmaσ indexed by N={0,1,…}\mathbb N=\{0,1,\ldots\}N={0,1,…}, any finite nonnegative real number ccc, and any natural number nnn, this defines the set of states ρ\rhoρ satisfying all three conditions

ρii=σii(i∈N),ρij=0if i≠j and (i<n or j<n),D(ρ,σ)≤c.\rho_{ii}=\sigma_{ii}\quad(i\in\mathbb N),\qquad \rho_{ij}=0\quad\text{if }i\ne j\text{ and }(i<n\text{ or }j<n),\qquad D(\rho,\sigma)\le c.ρii​=σii​(i∈N),ρij​=0if i=j and (i<n or j<n),D(ρ,σ)≤c.

A state is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, with am,bm∈ℓ2(N;C)a_m,b_m\in\ell^2(\mathbb N;\mathbb C)am​,bm​∈ℓ2(N;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, having real nonnegative quadratic forms on every finite coordinate set, and having summable real diagonal of sum 111. The extended nonnegative quantity D(ρ,σ)D(\rho,\sigma)D(ρ,σ) is the supremum over every finite s⊆Ns\subseteq\mathbb Ns⊆N, including s=∅s=\varnothings=∅, of

{max⁡ ⁣{Re⁡tr⁡(ρslog⁡ρs−ρslog⁡σs+σs−ρs),0},ker⁡σs⊆ker⁡ρs,+∞,otherwise,\begin{cases} \max\!\left\{\operatorname{Re}\operatorname{tr} (\rho_s\log\rho_s-\rho_s\log\sigma_s+\sigma_s-\rho_s),0\right\}, &\ker\sigma_s\subseteq\ker\rho_s,\\ +\infty,&\text{otherwise}, \end{cases}{max{Retr(ρs​logρs​−ρs​logσs​+σs​−ρs​),0},+∞,​kerσs​⊆kerρs​,otherwise,​

where ρs,σs\rho_s,\sigma_sρs​,σs​ are the coordinate restrictions and logarithms use continuous functional calculus with the total real logarithm, including log⁡0=0\log0=0log0=0. Because ccc is finite, the inequality excludes D(ρ,σ)=+∞D(\rho,\sigma)=+\inftyD(ρ,σ)=+∞. At n=0n=0n=0, the off-diagonal condition is vacuous, leaving equality of the diagonals and the entropy bound.

QAlgorithms.Papers.Independent.Gap48.P18.distanceSquared

For any index type III and states ρ,σ\rho,\sigmaρ,σ with coefficient matrices R,SR,SR,S, the definition gives τ(R−S)2∈[0,+∞]\tau(R-S)^2\in[0,+\infty]τ(R−S)2∈[0,+∞], where

τ(X)=inf⁡an,bn∈ℓ2(I;C)∑n∥an∥∥bn∥<∞Xij=∑nan(i)bn(j)‾ for all i,j∑n=0∞∥an∥∥bn∥.\tau(X)= \inf_{\substack{a_n,b_n\in\ell^2(I;\mathbb C)\\ \sum_n\|a_n\|\|b_n\|<\infty\\ X_{ij}=\sum_n a_n(i)\overline{b_n(j)}\ \text{for all }i,j}} \sum_{n=0}^{\infty}\|a_n\|\|b_n\|.τ(X)=an​,bn​∈ℓ2(I;C)∑n​∥an​∥∥bn​∥<∞Xij​=∑n​an​(i)bn​(j)​ for all i,j​inf​n=0∑∞​∥an​∥∥bn​∥.

The infimum is over countably indexed expansions, and an empty family of admissible expansions has infimum +∞+\infty+∞. A state matrix itself must admit such an expansion, must have every finite quadratic form real and nonnegative, and must have summable real diagonal entries with sum 111. The square is taken in the extended nonnegative reals, so +∞+\infty+∞ squares to +∞+\infty+∞. The definition includes coinciding states, for which the value is 000, and imposes no countability assumption on III; there are no state arguments when III is empty.

QAlgorithms.Papers.Independent.Gap48.P18.entropySup

For any index type III and any subset AAA of its state space, this definition assigns the extended nonnegative real supremum

sup⁡R∈A sup⁡s⊆Is finitemax⁡ ⁣{Re⁡tr⁡(Rs) log⁡ ⁣(Re⁡tr⁡(Rs))−Re⁡tr⁡ ⁣(Rs(log⁡)(Rs)), 0},\sup_{R\in A}\ \sup_{\substack{s\subseteq I\\s\text{ finite}}} \max\!\left\{ \operatorname{Re}\operatorname{tr}(R_s)\, \log\!\bigl(\operatorname{Re}\operatorname{tr}(R_s)\bigr) -\operatorname{Re}\operatorname{tr}\!\bigl(R_s(\log)(R_s)\bigr), \,0 \right\},R∈Asup​ s⊆Is finite​sup​max{Retr(Rs​)log(Retr(Rs​))−Retr(Rs​(log)(Rs​)),0},

where RsR_sRs​ is the finite coordinate restriction and the matrix logarithm is obtained by continuous functional calculus for the total real logarithm, with log⁡0=0\log 0=0log0=0. A state matrix RRR admits an expansion Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, has real nonnegative finite quadratic forms, and has summable real diagonal of sum 111. No countability, compactness, measurability, or nonemptiness assumption is imposed on III or AAA. The empty coordinate restriction contributes 000, the value may be +∞+\infty+∞, and A=∅A=\varnothingA=∅ gives value 000.

QAlgorithms.Papers.Independent.Gap48.P18.extremeClosure

For any index type III, let SI\mathcal S_ISI​ consist of complex arrays ρ\rhoρ admitting expansions ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​, where un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, with every finite quadratic form real and nonnegative and with summable real diagonal of sum 111. For any complex array XXX, let τ(X)\tau(X)τ(X) be the infimum of these expansion costs for XXX, with empty infimum ∞\infty∞, and equip SI\mathcal S_ISI​ with the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε} for ε>0\varepsilon>0ε>0. For a subset A⊆SIA\subseteq\mathcal S_IA⊆SI​, form the set of its coefficient arrays, view that set in the real vector space of complex arrays, and take its extreme points: those members that do not lie strictly between two distinct members of that set. The defined set is the closure, in SI\mathcal S_ISI​, of the states whose coefficient arrays are such extreme points. No convexity, closedness, or nonemptiness of AAA is assumed; for empty AAA, the resulting set is empty.

QAlgorithms.Papers.Independent.Gap48.P18.finiteLaw

For any type III, any natural number nnn, any family of finite nonnegative real weights (wk)0≤k<n(w_k)_{0\le k<n}(wk​)0≤k<n​, and any family of states (ρk)0≤k<n(\rho_k)_{0\le k<n}(ρk​)0≤k<n​ on III, this defines the measure

μ=∑k=0n−1wkδρk.\mu=\sum_{k=0}^{n-1}w_k\delta_{\rho_k}.μ=k=0∑n−1​wk​δρk​​.

For every measurable set EEE, its mass is ∑k<nwk1E(ρk)\sum_{k<n}w_k\mathbf 1_E(\rho_k)∑k<n​wk​1E​(ρk​); repeated states contribute repeatedly. A state is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​ with am,bm∈ℓ2(I;C)a_m,b_m\in\ell^2(I;\mathbb C)am​,bm​∈ℓ2(I;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, having real nonnegative quadratic forms on every finite coordinate set, and having summable real diagonal of sum 111. States carry the Borel measurable structure of the topology generated by {X:τ(X−R)<ε}\{X:\tau(X-R)<\varepsilon\}{X:τ(X−R)<ε}, where RRR is a state, ε>0\varepsilon>0ε>0, and τ(Z)\tau(Z)τ(Z) is the infimum of the stated expansion costs for ZZZ, with value +∞+\infty+∞ if there is no expansion. No normalization of the weights is required: the measure has finite total mass ∑k<nwk\sum_{k<n}w_k∑k<n​wk​, and for n=0n=0n=0 it is the zero measure.

QAlgorithms.Papers.Independent.Gap48.P18.jointMarginalConstraintSet

For any types I,JI,JI,J, any set AAA of states on III, and any set BBB of states on JJJ, this defines the set of states ω\omegaω on I×JI\times JI×J for which there exist ρ∈A\rho\in Aρ∈A and σ∈B\sigma\in Bσ∈B satisfying

∑r∈Jω(i,r),(j,r)=ρij(i,j∈I),∑r∈Iω(r,a),(r,b)=σab(a,b∈J).\sum_{r\in J}\omega_{(i,r),(j,r)}=\rho_{ij} \quad(i,j\in I), \qquad \sum_{r\in I}\omega_{(r,a),(r,b)}=\sigma_{ab} \quad(a,b\in J).r∈J∑​ω(i,r),(j,r)​=ρij​(i,j∈I),r∈I∑​ω(r,a),(r,b)​=σab​(a,b∈J).

Every displayed sum is the total complex infinite-sum operation, which returns 000 for a nonsummable family; no separate summability hypotheses on these sums are included. Here a state on any index type TTT is a complex matrix XXX admitting an expansion Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, with am,bm∈ℓ2(T;C)a_m,b_m\in\ell^2(T;\mathbb C)am​,bm​∈ℓ2(T;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, whose quadratic forms ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ are real and nonnegative for every finite s⊆Ts\subseteq Ts⊆T and every z:T→Cz:T\to\mathbb Cz:T→C, and whose real diagonal is summable with sum 111. No countability assumptions are imposed. The defined set is empty if either AAA or BBB is empty.

QAlgorithms.Papers.Independent.Gap48.P18.layer

For any type III and any state σ\sigmaσ on III, this defines the set of all states ρ\rhoρ on III satisfying ρii=σii\rho_{ii}=\sigma_{ii}ρii​=σii​ as complex numbers for every i∈Ii\in Ii∈I, with no further equality required between off-diagonal entries. Here a state is a complex matrix XXX admitting an expansion Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, where am,bm∈ℓ2(I;C)a_m,b_m\in\ell^2(I;\mathbb C)am​,bm​∈ℓ2(I;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, such that ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ is real and nonnegative for every finite s⊆Is\subseteq Is⊆I and every z:I→Cz:I\to\mathbb Cz:I→C, and such that its real diagonal entries are summable with sum 111. No countability assumption is imposed; if III is empty, there is no state σ\sigmaσ satisfying this diagonal-sum requirement.

QAlgorithms.Papers.Independent.Gap48.P18.meanCoefficients

For any index type III and any measure μ\muμ on its state space, the definition produces the complex matrix MμM_\muMμ​ given entrywise by

(Mμ)ij=∫Rij dμ(R).(M_\mu)_{ij}=\int R_{ij}\,d\mu(R).(Mμ​)ij​=∫Rij​dμ(R).

The state space consists of matrices RRR admitting expansions Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and summable ∥an∥∥bn∥\|a_n\|\|b_n\|∥an​∥∥bn​∥, having real nonnegative finite quadratic forms, and having summable real diagonal entries with sum 111. Its measurable sets are the Borel sets for the topology generated by the sets {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, for states QQQ and real ε>0\varepsilon>0ε>0, where τ(X)\tau(X)τ(X) is the infimum of ∑n∥an∥∥bn∥\sum_n\|a_n\|\|b_n\|∑n​∥an​∥∥bn​∥ over all such expansions of XXX, with empty infimum +∞+\infty+∞. The integral is the complex Bochner integral. No probability, finiteness, or entrywise integrability hypothesis is imposed on μ\muμ; a nonintegrable entry has integral 000 under the total integral convention. The resulting matrix is not required by this definition to be a state. In particular, the zero measure gives the zero matrix.

QAlgorithms.Papers.Independent.Gap48.P18.measureHolevo

For any index type III and any Borel measure μ\muμ on its state space, the definition gives the extended nonnegative integral

∫−D(R,Mμ) dμ(R),(Mμ)ij=∫Rij dμ(R),D(X,Y)=sup⁡s⊆I, s finiteDs(X,Y),\int^- D(R,M_\mu)\,d\mu(R), \qquad (M_\mu)_{ij}=\int R_{ij}\,d\mu(R), \qquad D(X,Y)=\sup_{s\subseteq I,\ s\text{ finite}}D_s(X,Y),∫−D(R,Mμ​)dμ(R),(Mμ​)ij​=∫Rij​dμ(R),D(X,Y)=s⊆I, s finitesup​Ds​(X,Y),

where, writing Xs,YsX_s,Y_sXs​,Ys​ for finite coordinate restrictions, Ds(X,Y)=+∞D_s(X,Y)=+\inftyDs​(X,Y)=+∞ if ker⁡Ys⊈ker⁡Xs\ker Y_s\not\subseteq\ker X_skerYs​⊆kerXs​, and otherwise

Ds(X,Y)=max⁡ ⁣{Re⁡tr⁡(Xs(log⁡)(Xs)−Xs(log⁡)(Ys)+Ys−Xs),0}.D_s(X,Y)= \max\!\left\{\operatorname{Re}\operatorname{tr} \bigl(X_s(\log)(X_s)-X_s(\log)(Y_s)+Y_s-X_s\bigr),0\right\}.Ds​(X,Y)=max{Retr(Xs​(log)(Xs​)−Xs​(log)(Ys​)+Ys​−Xs​),0}.

The logarithms use continuous functional calculus for the total real logarithm, with log⁡0=0\log 0=0log0=0, and the empty restriction contributes 000. States are matrices with summable countable expansions Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​, an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C), ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, real nonnegative finite quadratic forms, and summable real diagonal of sum 111. The state topology is generated by {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, where QQQ is a state, ε>0\varepsilon>0ε>0, and τ\tauτ is the infimum of the expansion sums just described; the measure uses its Borel sets. No probability or integrability assumption is imposed on μ\muμ, and MμM_\muMμ​ need not be a state; its nonintegrable entries are defined to be 000. The outer integral is the lower Lebesgue integral and may be +∞+\infty+∞; the zero measure gives value 000.

QAlgorithms.Papers.Independent.Gap48.P18.optimalAverage

For any index type III, states are complex arrays ρ\rhoρ with an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​, where un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, all finite quadratic forms are real and nonnegative, and the real diagonal is summable with sum 111. Define τ(X)\tau(X)τ(X) as the infimum of these expansion costs, with empty infimum ∞\infty∞, and equip the state space SI\mathcal S_ISI​ with the topology generated by {σ:τ(σ−ρ)<ε}\{\sigma:\tau(\sigma-\rho)<\varepsilon\}{σ:τ(σ−ρ)<ε}, ε>0\varepsilon>0ε>0, and its Borel measurable structure. For arrays X,YX,YX,Y, let D(X,Y)D(X,Y)D(X,Y) be the supremum over finite coordinate sets F⊆IF\subseteq IF⊆I of [Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+[\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​ if ker⁡YF⊆ker⁡XF\ker Y_F\subseteq\ker X_FkerYF​⊆kerXF​, and ∞\infty∞ otherwise, using natural logarithms through continuous functional calculus and [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0). For a probability measure μ\muμ, set mμ,ij=∫σij dμ(σ)m_{\mu,ij}=\int\sigma_{ij}\,d\mu(\sigma)mμ,ij​=∫σij​dμ(σ), and define

C(A)=sup⁡μ probability on SIsupp⁡μ finite, supp⁡μ⊆A∫SI−D(σ,mμ) dμ(σ).C(A)=\sup_{\substack{\mu\text{ probability on }\mathcal S_I\\ \operatorname{supp}\mu\text{ finite},\ \operatorname{supp}\mu\subseteq A}} \int_{\mathcal S_I}^{-}D(\sigma,m_\mu)\,d\mu(\sigma).C(A)=μ probability on SI​suppμ finite, suppμ⊆A​sup​∫SI​−​D(σ,mμ​)dμ(σ).

Given A⊆SIA\subseteq\mathcal S_IA⊆SI​ and a supplied proof that A≠∅A\ne\varnothingA=∅, the defined state is an arbitrary classically chosen ω∈SI\omega\in\mathcal S_Iω∈SI​ satisfying D(ρ,ω)≤C(A)D(\rho,\omega)\le C(A)D(ρ,ω)≤C(A) for every ρ∈A\rho\in Aρ∈A, if such a state exists; otherwise it is an arbitrary element of AAA, chosen using the supplied nonemptiness proof. In the first branch, membership in AAA is not required, and neither branch asserts uniqueness.

QAlgorithms.Papers.Independent.Gap48.P18.probabilityAverage

For any countable index type III and any Borel probability measure μ\muμ on its state space, the definition returns a(μ)a(\mu)a(μ), where aaa is one continuous map chosen from the existence assertion that a continuous state-valued averaging map exists for all such probability measures. Specifically, for every probability measure ν\nuν, every coefficient function R↦RijR\mapsto R_{ij}R↦Rij​ is integrable and

(a(ν))ij=∫Rij dν(R).\bigl(a(\nu)\bigr)_{ij}=\int R_{ij}\,d\nu(R).(a(ν))ij​=∫Rij​dν(R).

Here states are complex matrices admitting expansions Rij=∑nun(i)vn(j)‾R_{ij}=\sum_n u_n(i)\overline{v_n(j)}Rij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, having real nonnegative finite quadratic forms, and having summable real diagonal of sum 111. The state topology is generated by the sets {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, where τ(X)\tau(X)τ(X) is the infimum of those expansion sums for XXX, QQQ is a state, and ε>0\varepsilon>0ε>0; probability measures carry their usual weak topology. The choice is of the whole continuous averaging map and is not an additional uniqueness assertion. Countable III includes finite and empty types, although an empty III has neither states nor probability measures on its state space.

QAlgorithms.Papers.Independent.Gap48.P18.projectedAverage

For any index type III, bounded complex-linear operator PPP on ℓ2(I;C)\ell^2(I;\mathbb C)ℓ2(I;C) with P2=P=P∗P^2=P=P^*P2=P=P∗, set AAA of states on III, and supplied witness that A≠∅A\ne\varnothingA=∅, this defines a state by the following choices. For a state matrix RRR, put KijR=∑a′∑b′(Pea)(i)Rab(Peb)(j)‾K^R_{ij}=\sum_a'\sum_b'(Pe_a)(i)R_{ab}\overline{(Pe_b)(j)}KijR​=∑a′​∑b′​(Pea​)(i)Rab​(Peb​)(j)​ and pR=∑i′Re⁡KiiRp_R=\sum_i'\operatorname{Re}K^R_{ii}pR​=∑i′​ReKiiR​, where eae_aea​ is the coordinate unit vector; let BBB be the set of states with coefficients pR−1KRp_R^{-1}K^RpR−1​KR for some R∈AR\in AR∈A with pR>0p_R>0pR​>0. If B≠∅B\ne\varnothingB=∅, apply the following selection rule to BBB; otherwise apply it to AAA. For a nonempty set EEE, if there exists a state ω\omegaω such that D(R,ω)≤C(E)D(R,\omega)\le C(E)D(R,ω)≤C(E) for every R∈ER\in ER∈E, choose one such ω\omegaω; if none exists, choose a member of EEE using its nonemptiness witness. The first choice is not required by this definition to belong to EEE. Here states are complex matrices admitting countable expansions Rij=∑nun(i)vn(j)‾R_{ij}=\sum_n u_n(i)\overline{v_n(j)}Rij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, having real nonnegative finite-coordinate quadratic forms and summable real diagonal entries of sum 111. For any matrix XXX, let τ(X)\tau(X)τ(X) be the infimum of ∑n∥un∥∥vn∥\sum_n\|u_n\|\|v_n\|∑n​∥un​∥∥vn​∥ over such expansions of XXX, with value ∞\infty∞ when there is none; the state topology is generated by {S:τ(S−R)<ε}\{S:\tau(S-R)<\varepsilon\}{S:τ(S−R)<ε} for states RRR and ε>0\varepsilon>0ε>0. Define

D(X,Y)=sup⁡s⊆I, s finite{max⁡{Re⁡tr⁡(Xslog⁡Xs−Xslog⁡Ys+Ys−Xs),0},ker⁡Ys⊆ker⁡Xs,∞,otherwise,D(X,Y)=\sup_{s\subseteq I,\ s\text{ finite}} \begin{cases} \max\{\operatorname{Re}\operatorname{tr}(X_s\log X_s-X_s\log Y_s+Y_s-X_s),0\}, &\ker Y_s\subseteq\ker X_s,\\ \infty,&\text{otherwise}, \end{cases}D(X,Y)=s⊆I, s finitesup​{max{Retr(Xs​logXs​−Xs​logYs​+Ys​−Xs​),0},∞,​kerYs​⊆kerXs​,otherwise,​

where Xs,YsX_s,Y_sXs​,Ys​ are coordinate compressions and logarithms use continuous functional calculus with log⁡0=0\log0=0log0=0. The number C(E)C(E)C(E) is the supremum of ∫−D(R,R‾μ) dμ(R)\int^-D(R,\overline R_\mu)\,d\mu(R)∫−D(R,Rμ​)dμ(R) over Borel probability measures μ\muμ with finite topological support contained in EEE, where (R‾μ)ij=∫Rij dμ(R)(\overline R_\mu)_{ij}=\int R_{ij}\,d\mu(R)(Rμ​)ij​=∫Rij​dμ(R). All these infinite sums and coordinate integrals use total conventions, assigning zero to nonsummable families or nonintegrable coordinate functions, respectively. No countability, finite-capacity, or existence-of-center hypothesis is imposed; if P=0P=0P=0, then BBB is empty and the selection rule is applied to AAA.

QAlgorithms.Papers.Independent.Gap48.P18.projectionImage

For any index type III, bounded complex-linear operator PPP on ℓ2(I;C)\ell^2(I;\mathbb C)ℓ2(I;C) satisfying P2=P=P∗P^2=P=P^*P2=P=P∗, and set AAA of states on III, this defines the set of states σ\sigmaσ for which there exists ρ∈A\rho\in Aρ∈A such that, writing RRR for ρ\rhoρ's coefficient matrix,

Kij=∑a∈I′∑b∈I′(Pea)(i)Rab(Peb)(j)‾,p=∑i∈I′Re⁡Kii,K_{ij}=\sum_{a\in I}'\sum_{b\in I}'(Pe_a)(i)R_{ab}\overline{(Pe_b)(j)}, \qquad p=\sum_{i\in I}'\operatorname{Re}K_{ii},Kij​=a∈I∑′​b∈I∑′​(Pea​)(i)Rab​(Peb​)(j)​,p=i∈I∑′​ReKii​,

one has p>0p>0p>0 and σij=p−1Kij\sigma_{ij}=p^{-1}K_{ij}σij​=p−1Kij​ for every i,ji,ji,j. Here eae_aea​ is the coordinate unit vector, and a state is a complex matrix admitting an expansion ∑nun(i)vn(j)‾\sum_n u_n(i)\overline{v_n(j)}∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, having real nonnegative finite-coordinate quadratic forms, and having summable real diagonal entries of sum 111. Infinite sums use their total conventions, with nonsummable families assigned sum zero. Only inputs of strictly positive mass are admitted; the set is empty when AAA is empty or P=0P=0P=0. No countability assumption on III is made.

QAlgorithms.Papers.Independent.Gap48.P18.projectionMass

For any index type III, bounded complex-linear operator P:ℓ2(I;C)→ℓ2(I;C)P:\ell^2(I;\mathbb C)\to\ell^2(I;\mathbb C)P:ℓ2(I;C)→ℓ2(I;C) satisfying P2=P=P∗P^2=P=P^*P2=P=P∗, and state ρ\rhoρ with coefficient matrix RRR, this defines the real number

∑i∈I′Re⁡(∑a∈I′∑b∈I′(Pea)(i) Rab (Peb)(i)‾),\sum_{i\in I}'\operatorname{Re}\left( \sum_{a\in I}'\sum_{b\in I}' (Pe_a)(i)\,R_{ab}\,\overline{(Pe_b)(i)} \right),i∈I∑′​Re(a∈I∑′​b∈I∑′​(Pea​)(i)Rab​(Peb​)(i)​),

where eae_aea​ has coordinate 111 at aaa and 000 elsewhere. The state condition means that Rab=∑nun(a)vn(b)‾R_{ab}=\sum_n u_n(a)\overline{v_n(b)}Rab​=∑n​un​(a)vn​(b)​ for vectors un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) with ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, every finite-coordinate quadratic form of RRR is real and nonnegative, and its real diagonal entries are summable with sum 111. The displayed sums are taken in their specified nested order and use total infinite-sum conventions: a nonsummable real or complex family has total sum 000. The definition imposes no additional positive-mass hypothesis; for P=0P=0P=0 its value is 000. No countability assumption on III is made.

QAlgorithms.Papers.Independent.Gap48.P18.projectionSuccessInfimum

For any index type III, bounded complex-linear operator PPP on ℓ2(I;C)\ell^2(I;\mathbb C)ℓ2(I;C) satisfying P2=P=P∗P^2=P=P^*P2=P=P∗, and set AAA of states on III, this defines

inf⁡ρ∈Amax⁡ ⁣{∑i∈I′Re⁡(∑a∈I′∑b∈I′(Pea)(i)ρab(Peb)(i)‾), 0}\inf_{\rho\in A} \max\!\left\{ \sum_{i\in I}'\operatorname{Re}\left( \sum_{a\in I}'\sum_{b\in I}'(Pe_a)(i)\rho_{ab}\overline{(Pe_b)(i)} \right),\,0\right\}ρ∈Ainf​max{i∈I∑′​Re(a∈I∑′​b∈I∑′​(Pea​)(i)ρab​(Peb​)(i)​),0}

in [0,∞][0,\infty][0,∞], where eae_aea​ is the coordinate unit vector. A state matrix ρ\rhoρ has an expansion ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_n u_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, has real nonnegative finite-coordinate quadratic forms, and has summable real diagonal entries totaling 111. The infimum ranges over every member of AAA, without restricting to states of positive mass. The maximum with zero is the conversion from a real number to an extended nonnegative number. If A=∅A=\varnothingA=∅, the value is ∞\infty∞; if P=0P=0P=0 and A≠∅A\ne\varnothingA=∅, it is 000. The infinite sums are total sums, zero for nonsummable families, and III need not be countable.

QAlgorithms.Papers.Independent.Gap48.P18.radius

For any index type III, states are complex arrays ρ\rhoρ admitting expansions ρij=∑nun(i)vn(j)‾\rho_{ij}=\sum_nu_n(i)\overline{v_n(j)}ρij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(I;C)u_n,v_n\in\ell^2(I;\mathbb C)un​,vn​∈ℓ2(I;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞, whose finite quadratic forms are real and nonnegative and whose real diagonal is summable with sum 111. For a finite F⊆IF\subseteq IF⊆I, denote coordinate compression by XFX_FXF​, and define

D(X,Y)=sup⁡F⊆IF finite{[Re⁡tr⁡(XFlog⁡XF−XFlog⁡YF+YF−XF)]+,ker⁡YF⊆ker⁡XF,∞,otherwise,D(X,Y)=\sup_{\substack{F\subseteq I\\F\text{ finite}}} \begin{cases} [\operatorname{Re}\operatorname{tr}(X_F\log X_F-X_F\log Y_F+Y_F-X_F)]_+, &\ker Y_F\subseteq\ker X_F,\\ \infty,&\text{otherwise}, \end{cases}D(X,Y)=F⊆IF finite​sup​{[Retr(XF​logXF​−XF​logYF​+YF​−XF​)]+​,∞,​kerYF​⊆kerXF​,otherwise,​

where [r]+=max⁡(r,0)[r]_+=\max(r,0)[r]+​=max(r,0) and the logarithms are given by continuous functional calculus of the natural logarithm. For a set AAA of states and a state σ\sigmaσ, the defined value is sup⁡ρ∈AD(ρ,σ)∈[0,∞]\sup_{\rho\in A}D(\rho,\sigma)\in[0,\infty]supρ∈A​D(ρ,σ)∈[0,∞]. The state σ\sigmaσ need not belong to AAA; the value may be infinite, and for A=∅A=\varnothingA=∅ the supremum is 000.

QAlgorithms.Papers.Independent.Gap48.P18.setChiCapacity

For any index type III and any subset AAA of its state space, this definition assigns

sup⁡μ a Borel probability measuresupp⁡μ finite, supp⁡μ⊆A∫−D(R,Mμ) dμ(R),(Mμ)ij=∫Rij dμ(R).\sup_{\substack{\mu\text{ a Borel probability measure}\\ \operatorname{supp}\mu\text{ finite},\ \operatorname{supp}\mu\subseteq A}} \int^- D(R,M_\mu)\,d\mu(R), \qquad (M_\mu)_{ij}=\int R_{ij}\,d\mu(R).μ a Borel probability measuresuppμ finite, suppμ⊆A​sup​∫−D(R,Mμ​)dμ(R),(Mμ​)ij​=∫Rij​dμ(R).

Here D(X,Y)=sup⁡s⊆I, s finiteDs(X,Y)D(X,Y)=\sup_{s\subseteq I,\ s\text{ finite}}D_s(X,Y)D(X,Y)=sups⊆I, s finite​Ds​(X,Y); Ds(X,Y)=+∞D_s(X,Y)=+\inftyDs​(X,Y)=+∞ when ker⁡Ys⊈ker⁡Xs\ker Y_s\not\subseteq\ker X_skerYs​⊆kerXs​, and otherwise

Ds(X,Y)=max⁡ ⁣{Re⁡tr⁡(Xs(log⁡)(Xs)−Xs(log⁡)(Ys)+Ys−Xs),0}.D_s(X,Y)= \max\!\left\{\operatorname{Re}\operatorname{tr} \bigl(X_s(\log)(X_s)-X_s(\log)(Y_s)+Y_s-X_s\bigr),0\right\}.Ds​(X,Y)=max{Retr(Xs​(log)(Xs​)−Xs​(log)(Ys​)+Ys​−Xs​),0}.

The restrictions are coordinate restrictions, logarithms use continuous functional calculus for the total real logarithm with log⁡0=0\log 0=0log0=0, and the empty restriction contributes 000. States are matrices admitting expansions Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and summable ∥an∥∥bn∥\|a_n\|\|b_n\|∥an​∥∥bn​∥, having real nonnegative finite quadratic forms and summable real diagonal of sum 111. Their topology is generated by {R:τ(R−Q)<ε}\{R:\tau(R-Q)<\varepsilon\}{R:τ(R−Q)<ε}, where τ\tauτ is the infimum of such expansion sums, QQQ is a state, and ε>0\varepsilon>0ε>0; support consists of points whose every open neighborhood has positive measure. The averages use Bochner integrals, with nonintegrable entries assigned 000, and the displayed outer integral is the lower Lebesgue integral. There is no countability assumption on III, measurability assumption on AAA, or requirement that an average lie in AAA. The value belongs to [0,+∞][0,+\infty][0,+∞]; if there are no eligible probability measures, the supremum is 000.

QAlgorithms.Papers.Independent.Gap48.P18.spectralPowerTrace

For any index type III, a state σ\sigmaσ on III, and any real number ttt, this defines the extended nonnegative number

sup⁡s⊆I, s finitemax⁡ ⁣{Re⁡tr⁡(ft(σ∣s×s)),0},ft(x)={0,x=0,xt,x≠0,\sup_{s\subseteq I,\ s\text{ finite}} \max\!\left\{\operatorname{Re}\operatorname{tr}\bigl(f_t(\sigma|_{s\times s})\bigr),0\right\}, \qquad f_t(x)= \begin{cases} 0,&x=0,\\ x^t,&x\ne0, \end{cases}s⊆I, s finitesup​max{Retr(ft​(σ∣s×s​)),0},ft​(x)={0,xt,​x=0,x=0,​

where ftf_tft​ is applied by the continuous functional calculus. A state here is a complex matrix RRR with an expansion Rij=∑nan(i)bn(j)‾R_{ij}=\sum_n a_n(i)\overline{b_n(j)}Rij​=∑n​an​(i)bn​(j)​, where an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, whose finite quadratic forms ∑i,j∈szi‾Rijzj\sum_{i,j\in s}\overline{z_i}R_{ij}z_j∑i,j∈s​zi​​Rij​zj​ are real and nonnegative, and whose real diagonal entries are summable with sum 111. The supremum includes the empty compression, which contributes 000, and may be infinite. There is no restriction t>0t>0t>0: at t=0t=0t=0 the scalar function still takes value 000 at zero and value 111 on positive eigenvalues; for negative ttt, zero eigenvalues also contribute 000. No countability assumption on III is made; if III is empty, no state σ\sigmaσ satisfies the stated state conditions.

QAlgorithms.Papers.Independent.Gap48.P18.stateClosedConvexHull

For any index type III and any subset AAA of its state space, this definition gives the set of states ρ\rhoρ, with coefficient matrix RRR, such that for every real ε>0\varepsilon>0ε>0 there is a matrix YYY in the real convex hull of the coefficient matrices of AAA satisfying

τ(R−Y)<ε.\tau(R-Y)<\varepsilon.τ(R−Y)<ε.

That real convex hull consists of finite convex combinations of matrices from AAA. For any matrix XXX, τ(X)\tau(X)τ(X) is the infimum of ∑n∥an∥∥bn∥\sum_n\|a_n\|\|b_n\|∑n​∥an​∥∥bn​∥ over all representations Xij=∑nan(i)bn(j)‾X_{ij}=\sum_n a_n(i)\overline{b_n(j)}Xij​=∑n​an​(i)bn​(j)​, with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and summable norm products; an empty infimum is +∞+\infty+∞. A state matrix must itself admit such a representation, have every finite quadratic form real and nonnegative, and have summable real diagonal entries with sum 111. The approximation inequality is strict, and no countability, nonemptiness, closedness, or convexity assumption is imposed on AAA or III. For A=∅A=\varnothingA=∅, the resulting set is empty.

QAlgorithms.Papers.Independent.Gap48.P18.stateEntropyNats

For any index type III and state ρ\rhoρ, write RRR for its complex coefficient matrix. Here a state means that RRR admits a representation Rij=∑n=0∞an(i)bn(j)‾R_{ij}=\sum_{n=0}^{\infty}a_n(i)\overline{b_n(j)}Rij​=∑n=0∞​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, every finite quadratic form ∑i,j∈szi‾Rijzj\sum_{i,j\in s}\overline{z_i}R_{ij}z_j∑i,j∈s​zi​​Rij​zj​ is real and nonnegative, and the real diagonal entries are summable with sum 111. The definition assigns the extended nonnegative real number

sup⁡s⊆Is finitemax⁡ ⁣{Re⁡tr⁡(Rs) log⁡ ⁣(Re⁡tr⁡(Rs))−Re⁡tr⁡ ⁣(Rs (log⁡)(Rs)), 0},\sup_{\substack{s\subseteq I\\s\text{ finite}}} \max\!\left\{ \operatorname{Re}\operatorname{tr}(R_s)\, \log\!\bigl(\operatorname{Re}\operatorname{tr}(R_s)\bigr) -\operatorname{Re}\operatorname{tr}\!\bigl(R_s\,(\log)(R_s)\bigr), \,0 \right\},s⊆Is finite​sup​max{Retr(Rs​)log(Retr(Rs​))−Retr(Rs​(log)(Rs​)),0},

where Rs=(Rij)i,j∈sR_s=(R_{ij})_{i,j\in s}Rs​=(Rij​)i,j∈s​, and (log⁡)(Rs)(\log)(R_s)(log)(Rs​) denotes continuous functional calculus applied to the real logarithm. The logarithm is the total real logarithm, with log⁡0=0\log 0=0log0=0. The empty finite subset is included and contributes 000; the supremum may be +∞+\infty+∞. No countability or nonemptiness assumption on III is imposed, although no state exists when III is empty.

QAlgorithms.Papers.Independent.Gap48.P18.stateRelativeEntropyNats

For any index type III and states ρ,σ\rho,\sigmaρ,σ with complex coefficient matrices R,SR,SR,S, the definition assigns sup⁡s⊆I, s finiteDs(R,S)\sup_{s\subseteq I,\ s\text{ finite}}D_s(R,S)sups⊆I, s finite​Ds​(R,S), valued in [0,+∞][0,+\infty][0,+∞]. A state matrix XXX here admits an expansion Xij=∑nan(i)bn(j)‾X_{ij}=\sum_n a_n(i)\overline{b_n(j)}Xij​=∑n​an​(i)bn​(j)​ with an,bn∈ℓ2(I;C)a_n,b_n\in\ell^2(I;\mathbb C)an​,bn​∈ℓ2(I;C) and ∑n∥an∥∥bn∥<∞\sum_n\|a_n\|\|b_n\|<\infty∑n​∥an​∥∥bn​∥<∞, has real nonnegative finite quadratic forms, and has summable real diagonal entries whose sum is 111. For the finite restrictions Rs,SsR_s,S_sRs​,Ss​, the quantity in the supremum is

Ds(R,S)={max⁡ ⁣{Re⁡tr⁡(Rs(log⁡)(Rs)−Rs(log⁡)(Ss)+Ss−Rs),0},ker⁡Ss⊆ker⁡Rs,+∞,otherwise.D_s(R,S)= \begin{cases} \max\!\left\{\operatorname{Re}\operatorname{tr} \bigl(R_s(\log)(R_s)-R_s(\log)(S_s)+S_s-R_s\bigr),0\right\}, &\ker S_s\subseteq\ker R_s,\\ +\infty,&\text{otherwise}. \end{cases}Ds​(R,S)={max{Retr(Rs​(log)(Rs​)−Rs​(log)(Ss​)+Ss​−Rs​),0},+∞,​kerSs​⊆kerRs​,otherwise.​

The matrix logarithms use continuous functional calculus for the total real logarithm, including log⁡0=0\log 0=0log0=0. The kernel condition is tested separately on every finite restriction. The empty restriction contributes 000; the result may be infinite even though each finite value is finite. There is no countability assumption on III; an empty III admits no state arguments.

QAlgorithms.Papers.Independent.Gap48.P18.tailLayer

For any state σ\sigmaσ indexed by N={0,1,…}\mathbb N=\{0,1,\ldots\}N={0,1,…} and any natural number nnn, this defines the set of states ρ\rhoρ satisfying ρii=σii\rho_{ii}=\sigma_{ii}ρii​=σii​ for every i∈Ni\in\mathbb Ni∈N, together with

i≠j and (i+1<n or j+1<n)⟹ρij=0(i,j∈N).i\ne j\ \text{and}\ (i+1<n\ \text{or}\ j+1<n) \quad\Longrightarrow\quad \rho_{ij}=0 \qquad(i,j\in\mathbb N).i=j and (i+1<n or j+1<n)⟹ρij​=0(i,j∈N).

Here a state is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, where am,bm∈ℓ2(N;C)a_m,b_m\in\ell^2(\mathbb N;\mathbb C)am​,bm​∈ℓ2(N;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, whose quadratic form ∑i,j∈szi‾Xijzj\sum_{i,j\in s}\overline{z_i}X_{ij}z_j∑i,j∈s​zi​​Xij​zj​ is real and nonnegative for every finite s⊆Ns\subseteq\mathbb Ns⊆N and every z:N→Cz:\mathbb N\to\mathbb Cz:N→C, and whose real diagonal is summable with sum 111. For n=0n=0n=0 or n=1n=1n=1, the additional off-diagonal condition is vacuous, so only equality of the diagonals is required.

QAlgorithms.Papers.Independent.Gap48.P18.unionObjective

For any type III, any natural number nnn, any family of sets AkA_kAk​ of states on III, supplied with the assumption that every AkA_kAk​ is nonempty, and any finite nonnegative real weights wkw_kwk​, 0≤k<n0\le k<n0≤k<n, this defines the extended nonnegative number

∑k<nwkC(Ak)+∫D(X,Mμ) dμ(X),μ=∑k<nwkδωk,(Mμ)ij=∫Xij dμ(X).\sum_{k<n}w_k C(A_k)+\int D(X,M_\mu)\,d\mu(X), \qquad \mu=\sum_{k<n}w_k\delta_{\omega_k}, \qquad (M_\mu)_{ij}=\int X_{ij}\,d\mu(X).k<n∑​wk​C(Ak​)+∫D(X,Mμ​)dμ(X),μ=k<n∑​wk​δωk​​,(Mμ​)ij​=∫Xij​dμ(X).

The outer integral is the nonnegative extended integral, and each coordinate integral is the complex Bochner integral. Here a state is a complex matrix XXX admitting Xij=∑mam(i)bm(j)‾X_{ij}=\sum_m a_m(i)\overline{b_m(j)}Xij​=∑m​am​(i)bm​(j)​, with am,bm∈ℓ2(I;C)a_m,b_m\in\ell^2(I;\mathbb C)am​,bm​∈ℓ2(I;C) and ∑m∥am∥∥bm∥<∞\sum_m\|a_m\|\|b_m\|<\infty∑m​∥am​∥∥bm​∥<∞, having real nonnegative quadratic forms on every finite coordinate set, and having summable real diagonal of sum 111. States carry the Borel structure of the topology generated by {X:τ(X−R)<ε}\{X:\tau(X-R)<\varepsilon\}{X:τ(X−R)<ε}, where τ\tauτ is the infimum of these expansion costs, or +∞+\infty+∞ when no expansion exists. For arbitrary matrices X,YX,YX,Y, define D(X,Y)=sup⁡s⊆I finited(Xs,Ys)D(X,Y)=\sup_{s\subseteq I\text{ finite}}d(X_s,Y_s)D(X,Y)=sups⊆I finite​d(Xs​,Ys​), including the empty restriction, where

d(U,V)={max⁡ ⁣{Re⁡tr⁡(Ulog⁡U−Ulog⁡V+V−U),0},ker⁡V⊆ker⁡U,+∞,otherwise.d(U,V)= \begin{cases} \max\!\left\{\operatorname{Re}\operatorname{tr} (U\log U-U\log V+V-U),0\right\}, &\ker V\subseteq\ker U,\\ +\infty,&\text{otherwise}. \end{cases}d(U,V)={max{Retr(UlogU−UlogV+V−U),0},+∞,​kerV⊆kerU,otherwise.​

Matrix logarithms use continuous functional calculus with the total real logarithm, including log⁡0=0\log0=0log0=0. The quantity C(B)C(B)C(B) is the supremum of ∫D(X,Mν) dν(X)\int D(X,M_\nu)\,d\nu(X)∫D(X,Mν​)dν(X) over probability measures ν\nuν whose topological support is finite and contained in BBB, with MνM_\nuMν​ defined by the same coordinate integrals; an empty supremum is 000. For each kkk, if there exists a state ω\omegaω satisfying D(ρ,ω)≤C(Ak)D(\rho,\omega)\le C(A_k)D(ρ,ω)≤C(Ak​) for every ρ∈Ak\rho\in A_kρ∈Ak​, then ωk\omega_kωk​ is a classically chosen such state, without an additional requirement that it belong to AkA_kAk​; otherwise, ωk\omega_kωk​ is a classically chosen member of AkA_kAk​, using the supplied nonemptiness assumption. The weights need not sum to 111, capacities may be infinite, and extended arithmetic uses 0⋅∞=00\cdot\infty=00⋅∞=0. For n=0n=0n=0, the nonemptiness assumption is vacuous and the defined value is 000.

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

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 22, 2026

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

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