Adaptive single-copy schemes with randomized post-processing are collective POVMs on
ProvedBosonicLearn.adaptive_scheme_collectiveEvery adaptive single-copy learning scheme, followed by any randomized classical post-processing, is realized by one collective measurement on all the copies.
Let be the -mode Fock space and let be the set of density operators on , with the Borel -algebra of the trace-norm topology. An adaptive single-copy scheme on copies with outcomes in a measurable space assigns to every round and every record of earlier outcomes a POVM on with outcomes in . The POVM is measured on a fresh copy, no quantum system is kept between rounds, and the outcome law of round depends measurably on . For a state , let be the law of the transcript when every round measures a fresh copy of . A randomized estimator is a Markov kernel from to ; is the probability that the estimate lies in given the transcript .
For every and , every measurable space , every adaptive single-copy scheme and every estimator , there is a POVM on the -copy space , with outcomes in , such that for every state and every measurable
In words, the Born law of on is the law of the estimate that the scheme and its estimator produce from copies of .
Role. Single-copy schemes are thereby a special case of collective (entangled) schemes, so a lower bound on the number of copies proved for collective measurements also holds for adaptive single-copy schemes. The proof of Theorem 3 uses this for , where the collective lower bound of Theorem 9 is applied to single-copy learners.
Formalization Note The scheme is an AdaptiveScheme n N O, its transcript law is transcriptLaw, and the estimator is a Markov kernel into State. The POVM is a FockPOVM on the basis of ; by definition its Born law is a probability measure on every density operator of the -copy space, entangled ones included. The conclusion is the equality of measures E.law (ρ^{⊗N}) = (transcriptLaw S ρ).bind est, where has the entries (WildeQIT.Schumacher.tensorPow). The paper uses this inclusion of measurement classes without stating it as a separate result.
import Mathlib import Definitions.Def_BosonicLearn_transcriptLaw import Definitions.Def_BosonicLearn_FockPOVM import Definitions.Def_WildeQIT_Schumacher_tensorPow open QAlgorithms.Papers.Independent.Dependency50.P33 (State traceSize) open QAlgorithms.Papers.Independent.Claimed50.P47 (Occupation) open MeasureTheory ProbabilityTheory open scoped ENNReal Matrix
namespace BosonicLearn
/-- **Adaptive single-copy schemes are collective schemes.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, proof of Theorem 3, p. 67: "When ν > 1, Theorem 9 gives Ω(n²/ε²), even for arbitrary collective
measurements"; single-copy model, §4, p. 19 ("the choice of POVM may depend on all previous classical outcomes. No
quantum system is retained from one round to the next") and §5.1, p. 30 ("the t-th POVM can depend on the first
t − 1 measurement outcomes x_{<t}").
For every adaptive single-copy scheme `S` on `N` copies of an `n`-mode system, with outcomes in any measurable space
`O`, and every randomized estimator `est` (a Markov kernel from transcripts to density operators), there is one POVM
`E` on the `N`-copy Fock space `ℓ²(ℕⁿ)^{⊗N}`, with outcomes in the state space, whose Born law on every product state
`ρ^{⊗N}` is the law of the estimate `ρ̂` produced by running `S` on `N` fresh copies of `ρ` and then applying `est`:
`E.law(ρ^{⊗N}) = ∫ est(x, ·) dT_ρ(x)`, with `T_ρ` the transcript law.
**Formalization Note.** The law of the estimate is `Measure.bind` of the transcript law with the estimator kernel.
`E` is a `FockPOVM`, so its Born law is a probability measure on every density of the `N`-copy space, entangled ones
included; the identity is asserted on product states `ρ^{⊗N}` (`WildeQIT.Schumacher.tensorPow`). The paper uses this
inclusion of measurement classes without stating it. -/
theorem adaptive_scheme_collective {n N : ℕ} {O : Type} [MeasurableSpace O]
(S : AdaptiveScheme n N O) (est : Kernel (Fin N → O) (State (Occupation (Fin n))))
(hest : IsMarkovKernel est) :
∃ E : FockPOVM (Fin N → Occupation (Fin n)) (State (Occupation (Fin n))),
∀ ρ : State (Occupation (Fin n)),
E.law (WildeQIT.Schumacher.tensorPow N ρ.coefficients) =
(transcriptLaw S ρ.coefficients).bind est := by sorry
end BosonicLearn