Small-batch Schmidt projection yields less than Procrustean filtering
ProvedQAlgorithms.Papers.Independent.Support50.P42.small_batch_comparisonprove2quantum
Let be a batch size and let be the common Schmidt bias. Let be the expected concentrated entropy from Schmidt projection, before any standardization loss. Then
The right side is the one-pair Procrustean singlet yield. The comparison uses entropy per input pair, as in Figure 1; product endpoints are excluded because both yields vanish there. This pins the source's strict small-batch efficiency comparison.
Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P42
Formal statement
noncomputable section
namespace QAlgorithms.Papers.Independent.Support50.P42
open scoped BigOperators Matrix ComplexConjugate
open MeasureTheory ProbabilityTheory
attribute [local instance] Classical.propDecidable Classical.decEq
-- Reuse the exact accepted objects, without introducing same-name aliases.
open Information100.P21 (Op Vec FiniteLOCC)
open Dependency50.P42 (dimensionedPower singlets)
open Dependency50.P48 (pureFidelity)
open Gap48.P32 (pairOfState)
open Next46.P39 (rectAct)
/-- PDF p.9 and Figure 1, pp.20--21: entropy yield per input pair, before
standardization. Product endpoints are excluded from strict comparison. -/
theorem small_batch_comparison (n : ℕ) (hn : 1≤n) (hn5 : n<5)
(p : ℝ) (hp : 0<p) (hp1 : p<1) :
projectedYield n p/(n : ℝ)<procrusteanYield p := by sorry
end QAlgorithms.Papers.Independent.Support50.P42
end
Source
Concentrating Partial Entanglement by Local Operations, arXiv:quant-ph/9511030v1, PDF p. 9, §III and Figure 1: Procrustean versus small-batch Schmidt yields
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every natural number with , and every real number with , the following strict inequality holds:
Thus the statement includes exactly the batch sizes ; in particular, the sum on the left is zero when .
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.