Lemma 8: whitening a nonorthogonal basis gives an orthonormal basis
OpenQAlgorithms.Papers.Independent.Gap48.P45.whiteningBasis_orthonormalBasisAll physical Hilbert spaces are nonzero complex separable Hilbert spaces, with finite dimensions included. States are positive trace-class operators of trace one, channels are completely positive and trace preserving on the entire trace class, and logarithms are natural. Inner products are conjugate-linear in the first argument. Let be a normalized finite or countable basis of in the sense of unique convergent ordered expansions. For any strictly positive probability distribution , compute . The inverse-square-root images exist and
This result constructs the orthogonal family used in the reversibility normal form.
import Definitions.Def_QAlgorithms_Papers_Independent_Gap48_P45
noncomputable section
open Filter MeasureTheory
open scoped BigOperators ENNReal Topology
attribute [local instance] Classical.propDecidable Classical.decEq
attribute [local instance] IsStarNormal.instContinuousFunctionalCalculus
namespace QAlgorithms.Papers.Independent.Gap48.P45
open QAlgorithms.RealRegisters (Hilbert)
open QAlgorithms.Papers.Independent.Dependency50.P33
(State IsTraceClass IsTraceClassChannel traceSize traceClosure rankOne productCoeff tensorApply)
open QAlgorithms.Papers.Independent.Frontier18.P01 (Channel Complementary traceFirst traceSecond)
open QAlgorithms.Papers.Independent.Frontier18.P07
(GeneralizedEnsemble averageCoefficients extendedEntropyNats extendedRelativeEntropyNats ensembleHolevo pureStates finiteEnsembleCapacity)
open QAlgorithms.Papers.Independent.Frontier18.P18
(operatorCoeff boundedOperator conjugateCoeff ReversibleOn IsometricallyEquivalent
IsParsevalFrame IsKrausRepresentation)
open QAlgorithms.Papers.Independent.Claimed50.P16 (tensorRankLE)
open QAlgorithms.Papers.Independent.Next46.P31 (positiveSquareRoot)
open QAlgorithms.Papers.Independent.Next46.P42 (minimumOutputEntropy)
/-- Lemma 8, PDF pp.25–26 with (9), p.8: whitening a normalized unique
convergent-expansion basis gives an ONB for every positive law. The p.26 proof
says apply rho_bar; its displayed equation instead uses sqrt(rho_bar).
The statement keeps inverse exponent -1/2 and explicitly supplies the domain equation. -/
theorem whiteningBasis_orthonormalBasis
{I : Type} [Countable I] [Nonempty I]
(n : ℕ∞) (v : BoundIndex n → Hilbert I)
(hUnit : ∀ i, ‖v i‖ = 1) (hBasis : UniqueConvergentBasis v)
(pi : BoundIndex n → ℝ) (hLaw : PositiveLaw pi) :
(∀ i, positiveSquareRoot (boundedOperator (vectorAverage pi v)) (whitenedVector pi v i) =
(Real.sqrt (pi i) : ℂ) • v i) ∧ OrthonormalBasisFamily (whitenedVector pi v) := by sorry
end QAlgorithms.Papers.Independent.Gap48.P45
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be a countable, nonempty type, put , let , and set . Let satisfy for every , and assume that every has exactly one coefficient family for which the ordered partial sums converge in norm to as . Let satisfy for every and , where this last equality requires convergence of the family of finite subsums. Define the matrix by . Let be the bounded complex linear operator chosen to satisfy , if such an operator exists, and otherwise, where is the coordinate unit vector. Define to be the continuous-functional-calculus value of the function , viewed as complex valued, at . For each , define to be a chosen vector satisfying and , if such a vector exists, and define otherwise. Then for every , the family is orthonormal, and for every the family has sum , with convergence over finite subsets of . The index parameter permits finite values and ; at , the probability-law hypothesis is impossible because its empty sum would have to equal .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.