Consistency Defect Le Of Recovery Bound
ProvedMIPStarRE.QPBT.consistencyDefect_le_of_recoveryBoundThe averaging estimate underlying the recovery transport, stated for the Born weights alone. The weights w x b g of a pair of outcomes have fiber sums W g independent of the question x, and the two diagonal overlaps differ only on the discrepancy event, whose conditional probability is bounded by F x g; averaging that bound against the fiber weights gives η.
All objects, parameters, typeclass assumptions, and hypotheses are quantified in the accompanying formal statement; each symbol denotes the object of the displayed Lean type with that name. The conclusion is the proposition
This result is an importable reduction step in the formal dependency graph for the quantum Pauli basis test completeness or soundness argument.
Formalization Credit. This Lean declaration is part of the Quantum Pauli basis test formalization by Ruixuan Deng. The source code was generated by AI agents; Ruixuan Deng is the responsible maintainer.
Mathematical Source. The mathematical source is Ji, Natarajan, Vidick, Wright, and Yuen, MIP*=RE, arXiv:2001.04383v3.
import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_AxisParallelLine import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_DiagonalLine import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_Distribution import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_LinePolynomials import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_LowDegreePolynomial import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_OpFamily import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_ParametersBase import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_ParametersFiniteAnswers import Definitions.Def_LIDT_MIPStarRE_LDT_Basic_QuantumState import Definitions.Def_LIDT_MIPStarRE_LDT_CommutativityPoints_Defs import Definitions.Def_LIDT_MIPStarRE_LDT_ExpansionHypercubeGraph_Defs_Core import Definitions.Def_LIDT_MIPStarRE_LDT_GlobalVariance_Defs_Core import Definitions.Def_LIDT_MIPStarRE_LDT_MakingMeasurementsProjective_Defs import Definitions.Def_LIDT_MIPStarRE_LDT_Pasting_Defs_Families import Definitions.Def_LIDT_MIPStarRE_LDT_Pasting_Defs_Interpolation import Definitions.Def_LIDT_MIPStarRE_LDT_Tactic_AvgCongr import Definitions.Def_LIDT_MIPStarRE_LDT_Test_StrategyCore import Definitions.Def_LIDT_MIPStarRE_Quantum_FiniteMatrix_Basic import Definitions.Def_LIDT_MIPStarRE_Quantum_Measurement import Definitions.Def_QPBT_QPBT_Games_Consistency import Definitions.Def_QPBT_QPBT_Games_Defs import Definitions.Def_QPBT_QPBT_Games_DistanceTheorems_Support import Definitions.Def_QPBT_QPBT_Games_DistanceTheorems_TensorSupport import Mathlib namespace LIDT.MIPStarRE.LDT end LIDT.MIPStarRE.LDT namespace LIDT.MIPStarRE.Quantum end LIDT.MIPStarRE.Quantum namespace MIPStarRE.QPBT end MIPStarRE.QPBT namespace MIPStarRE.QPBT.DistanceCalculus end MIPStarRE.QPBT.DistanceCalculus namespace MIPStarRE.QPBT.RecoveryInternal end MIPStarRE.QPBT.RecoveryInternal namespace MIPStarRE.QPBT.SandwichProduct end MIPStarRE.QPBT.SandwichProduct /-! # Transporting a consistency relation along the recovery step The recovery step of `lem:ld-combining-recovery` replaces a relation between a point measurement of the original strategy read through a combination and a measurement of the combined game read at a point of the combined space by a relation between the point measurement itself and the same measurement read through the recovered polynomial tuple. This module isolates the consistency calculus of that replacement, in a form which does not mention the low-degree game. Both relations are diagonal overlaps of the same pair of measurements placed on opposite tensor factors, and the two relabellings differ only on the event that the recovered outcome disagrees with the outcome of the point measurement while the two relabellings agree. The probability of that event, averaged over the auxiliary sample, is bounded at each question by a quantity depending only on the outcome of the second measurement; since the first measurement is complete, the total Born weight of an outcome of the second measurement does not depend on the question, and the bound may therefore be averaged over the question against that weight. The two conclusions of `lem:ld-soundness` transported here place the point measurement on opposite tensor factors, so the calculus is recorded twice, once for each placement; the two statements share the averaging estimate, which is insensitive to the placement. ## Main statements * `consistencyDefect_le_of_recoveryBound` — the transport of a consistency relation along a relabelling which is correct off a rare event, with the point measurement on the first tensor factor. * `consistencyDefect_le_of_recoveryBound_right` — the same transport with the point measurement on the second tensor factor. ## References * `references/neexp-paper/05_quantum_preliminaries.tex:1440-1503` * `references/qpbt-paper/06_nonlocal_games_and_mipstar.tex:219-288` * `blueprint/src/chapter/ch13_qpbt_test.tex:617-680` -/ open scoped BigOperators MatrixOrder Matrix ComplexOrder open MIPStarRE.QPBT open LIDT.MIPStarRE.LDT hiding Measurement open LIDT.MIPStarRE.Quantum open MIPStarRE.QPBT.DistanceCalculus open MIPStarRE.QPBT.SandwichProduct open MIPStarRE.QPBT.RecoveryInternal /-! ## Pair expansions of the diagonal overlap -/ /-! ## The averaging estimate -/ /-! ## The transport of a consistency relation -/ open MIPStarRE.QPBT.RecoveryInternal
set_option maxHeartbeats 1600000 in
theorem MIPStarRE.QPBT.consistencyDefect_le_of_recoveryBound
{X Y β Γ R ιA ιB : Type*}
[Fintype X] [DecidableEq X] [Nonempty X]
[Fintype Y] [DecidableEq Y] [Nonempty Y]
[Fintype β] [DecidableEq β] [Fintype Γ] [DecidableEq Γ]
[Fintype R] [DecidableEq R]
[Fintype ιA] [DecidableEq ιA] [Fintype ιB] [DecidableEq ιB]
(A : X → _root_.LIDT.MIPStarRE.Quantum.Measurement β ιA) (B : _root_.LIDT.MIPStarRE.Quantum.Measurement Γ ιB)
(ψ : EuclideanSpace ℂ (ιA × ιB)) (hψ : ‖ψ‖ = 1)
(f : X → Γ → β) (e₁ : X → Y → β → R) (f₁ : X → Y → Γ → R)
(F : X → Γ → ℝ) (η : ℝ)
(hFbound : ∀ x b g, _root_.LIDT.MIPStarRE.LDT.avgOver (_root_.LIDT.MIPStarRE.LDT.uniformDistribution Y)
(fun y => if f x g ≠ b ∧ e₁ x y b = f₁ x y g then (1 : ℝ) else 0) ≤ F x g)
(hFavg : ∀ g, _root_.LIDT.MIPStarRE.LDT.avgOver (_root_.LIDT.MIPStarRE.LDT.uniformDistribution X) (fun x => F x g) ≤ η) :
_root_.MIPStarRE.QPBT.consistencyDefect (_root_.LIDT.MIPStarRE.LDT.uniformDistribution X)
(fun x b => _root_.MIPStarRE.QPBT.heteroKron ((A x).effect b) 1)
(fun x b => _root_.MIPStarRE.QPBT.heteroKron 1 ((B.postprocess (f x)).effect b)) ψ ≤
_root_.MIPStarRE.QPBT.consistencyDefect (_root_.LIDT.MIPStarRE.LDT.uniformDistribution (X × Y))
(fun z r => _root_.MIPStarRE.QPBT.heteroKron (((A z.1).postprocess (e₁ z.1 z.2)).effect r) 1)
(fun z r => _root_.MIPStarRE.QPBT.heteroKron 1 ((B.postprocess (f₁ z.1 z.2)).effect r)) ψ + η := by sorry