Achievable copy cap for simultaneous signed estimation
DefinitionPersistQMem_Supplement_AchievableFix a finite family of Hermitian observables on a -dimensional source. A worst-case copy cap is achievable with retained quantum dimension at most and error if there is a source-independent persistent-memory protocol using accesses, followed by a measurable terminal POVM whose real-vector output is , such that for every density matrix ,
The protocol may use an arbitrary measurable classical record space, adaptive complete instruments, and unlimited classical records. Every quantum register retained between accesses counts toward the single dimension cap .
Formalization Note The published PersistQMem.Protocol supplies the adaptive receiver. As for the round instruments, the final POVM is represented by a jointly measurable positive density against an s-finite reference kernel, now on the full real-vector output space. Integrability is required explicitly, preventing a Bochner integral from defaulting to zero.
import Mathlib
import Definitions.Def_PersistQMem_Protocol
set_option autoImplicit false
open MeasureTheory ProbabilityTheory
open scoped Matrix ComplexOrder
namespace PersistQMem
namespace Supplement
/-- A measurable real-vector estimator attached to a published persistent-memory protocol.
After `N` source accesses, the classical history and retained memory are measured with a POVM
density `Q`; the outcome `muHat : J -> Real` is the requested real output vector. This is
Definition 5.1 in Tao--Shen, *Structural Lower Bounds for Quantum Property Estimation with
Persistent Memory*, STOC 2027 submission (manuscript of 2026-11-03, sha256 6d48ad24...), p.15
(PDF p.18).
**Formalization Note.** The terminal POVM is represented, just like the round instruments in
`Protocol`, by a jointly measurable positive density against an s-finite reference kernel. Taking
the output space itself to be `J -> Real` retains Definition 5.1's unrestricted real outputs rather
than imposing a finite-output restriction. The integrability fields prevent Lean's Bochner integral
from silently taking the value zero. -/
structure SignedEstimator (J X L : Type) [Fintype J] [MeasurableSpace X] [Fintype L]
(K D N : Nat) where
protocol : Protocol X L K D
terminalKernel : Kernel (Fin N -> X) (J -> Real)
terminal_sfinite : IsSFiniteKernel terminalKernel
Q : (Fin N -> X) -> (J -> Real) -> Matrix (Fin K) (Fin K) Complex
Q_measurable : forall i j,
Measurable (fun q : (Fin N -> X) × (J -> Real) => Q q.1 q.2 i j)
Q_effect : forall h muHat, (Q h muHat).PosSemidef
Q_complete_integrable : forall h i j,
Integrable (fun muHat => Q h muHat i j) (terminalKernel h)
Q_complete : forall h i j,
∫ muHat, Q h muHat i j ∂(terminalKernel h) =
(1 : Matrix (Fin K) (Fin K) Complex) i j
Q_integrable : forall rho : Matrix (Fin D) (Fin D) Complex,
WildeQIT.IsDensityOperator rho ->
Integrable
(fun q : (Fin N -> X) × (J -> Real) =>
(Q q.1 q.2 * branch protocol rho N q.1).trace.re)
((refMeasure protocol N) ⊗ₘ (terminalKernel))
variable {J X L : Type} [Fintype J] [MeasurableSpace X] [Fintype L]
variable {K D N : Nat}
/-- Probability that every requested expectation is estimated to additive error at most `epsilon`. -/
noncomputable def simultaneousSuccess (E : SignedEstimator J X L K D N)
(G : J -> Matrix (Fin D) (Fin D) Complex) (epsilon : Real)
(rho : Matrix (Fin D) (Fin D) Complex) : Real :=
∫ q, if (forall j, abs (q.2 j - (rho * G j).trace.re) <= epsilon)
then (E.Q q.1 q.2 * branch E.protocol rho N q.1).trace.re else 0
∂((refMeasure E.protocol N) ⊗ₘ (E.terminalKernel))
/-- The estimator solves the simultaneous signed task with success probability at least `2/3` for
every density matrix, as in Definition 5.1 and the standing convention of Sections 2.2--2.4. -/
def SolvesSignedTask (E : SignedEstimator J X L K D N)
(G : J -> Matrix (Fin D) (Fin D) Complex) (epsilon : Real) : Prop :=
forall rho, WildeQIT.IsDensityOperator rho ->
(2 : Real) / 3 <= simultaneousSuccess E G epsilon rho
/-- A worst-case copy cap `N` is achievable for the simultaneous signed-estimation task with retained
dimension at most `K`. The existential record space, adaptive instrument, terminal POVM and decoder
are all chosen before the unknown source state.
**Formalization Note.** `Protocol` uses exactly `N` rounds; early stopping is represented by idle
rounds, as on p.15. A receiver of smaller retained dimension embeds into `C^K`; theorem statements
also record the dimension-one upper-bound witnesses explicitly. -/
def Achievable (G : J -> Matrix (Fin D) (Fin D) Complex) (K : Nat) (epsilon : Real)
(N : Nat) : Prop :=
exists (X : Type) (mX : MeasurableSpace X) (L : Type) (fL : Fintype L),
exists E : @SignedEstimator J X L inferInstance mX fL K D N,
@SolvesSignedTask J X L inferInstance mX fL K D N E G epsilon
end Supplement
end PersistQMem