Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Measuring one system, then a second one with a POVM chosen from the first outcome, is a single POVM on both

Proved
BosonicLearn.fockPOVM_compose

by Senrui Chen · Nov 13, 2026 · Mathlib 0df444a (Lean v4.33.1)

bosonic-gaussian-statesquantum-informationquantum-learning

Measuring a first system and then measuring a second system with a POVM chosen according to the first outcome is a single measurement on the joint system.

Let MMM be a POVM on ℓ2(J)\ell^2(J)ℓ2(J) with outcomes in a measurable space Ω\OmegaΩ. For every x∈Ωx\in\Omegax∈Ω let QxQ_xQx​ be a POVM on ℓ2(I′)\ell^2(I')ℓ2(I′) with outcomes in a measurable space OOO, depending measurably on xxx in the following sense: for every density operator τ\tauτ on ℓ2(I′)\ell^2(I')ℓ2(I′) there is a Markov kernel κτ\kappa_\tauκτ​ from Ω\OmegaΩ to OOO with κτ(x,⋅)\kappa_\tau(x,\cdot)κτ​(x,⋅) equal to the outcome law of QxQ_xQx​ on τ\tauτ.

Then there is a POVM M′M'M′ on ℓ2(J×I′)=ℓ2(J)⊗ℓ2(I′)\ell^2(J\times I')=\ell^2(J)\otimes\ell^2(I')ℓ2(J×I′)=ℓ2(J)⊗ℓ2(I′) with outcomes in Ω×O\Omega\times OΩ×O such that for all density operators ρ\rhoρ on ℓ2(J)\ell^2(J)ℓ2(J) and τ\tauτ on ℓ2(I′)\ell^2(I')ℓ2(I′) and every measurable C⊆Ω×OC\subseteq\Omega\times OC⊆Ω×O

Mρ⊗τ′(C)=∫Ωκτ(x,{o:(x,o)∈C}) dMρ(x),M'_{\rho\otimes\tau}(C)=\int_\Omega\kappa_\tau\big(x,\{o:(x,o)\in C\}\big)\,\mathrm dM_\rho(x),Mρ⊗τ′​(C)=∫Ω​κτ​(x,{o:(x,o)∈C})dMρ​(x),

where MρM_\rhoMρ​ and Mρ⊗τ′M'_{\rho\otimes\tau}Mρ⊗τ′​ denote outcome laws. In terms of effects, E′(A×B)=∫AE(dx)⊗Gx(B)E'(A\times B)=\int_AE(\mathrm dx)\otimes G_x(B)E′(A×B)=∫A​E(dx)⊗Gx​(B), with EEE and GxG_xGx​ the effects of MMM and QxQ_xQx​.

Role. This is one round of an adaptive protocol: the first system carries the copies already measured, with the transcript so far as outcome, and the second system is the next copy. Iterating it gives BosonicLearn.adaptive_scheme_product_povm.

Formalization Note The operator ρ⊗τ\rho\otimes\tauρ⊗τ has entries ρ(u,v) τ(a,b)\rho(u,v)\,\tau(a,b)ρ(u,v)τ(a,b) at ((u,a),(v,b))((u,a),(v,b))((u,a),(v,b)), and the right-hand side is the composition-product (M.law ρ) ⊗ₘ (κ τ). The measurable dependence on xxx is given as data: a kernel κ τ for every coefficient array τ, Markov, and equal to (Q x).law τ on densities; this is how AdaptiveScheme provides its round laws. M′M'M′ is a FockPOVM, so its effects are bounded between 000 and 1\mathbf 11 on all finitely supported vectors of the joint system and its Born law is a probability measure on every density operator of the joint system, entangled ones included.

Related platform material The space ℓ2(A)\ell^2(A)ℓ2(A) used here (Hilbert A) is the same Lean type as OAI.TensorLp.H A, and ℓ2(Nn)\ell^2(\mathbb N^n)ℓ2(Nn) is the same type as OAI.EntropyPhotonNumber.Fock n. The published definitions OAI_Analysis_TensorLp (tensor products, slices and operator lifts on ℓ2(A×B)\ell^2(A\times B)ℓ2(A×B)) and OAI_MathematicalPhysics_Fock_Operators (OAI.CrossEnsemble.operator, the bounded operator ∑i∣ui⟩⟨vi∣\sum_i|u_i\rangle\langle v_i|∑i​∣ui​⟩⟨vi​∣ of a family, with hasSum_operator and entry_hasSum) are sorry-free and can be imported. Proved lemmas of this mission that can be imported: BosonicLearn.traceDist_le_one (nuclear expansions, truncation to finite index sets) BosonicLearn.effect_matrix_operator (effect matrices as bounded positive operators), BosonicLearn.effect_trace_pairing (Born sums along nuclear expansions), BosonicLearn.fockPOVM_form_measure (effects are weakly σ\sigmaσ-additive) and BosonicLearn.fockPOVM_postprocess.

Preamble
import Mathlib
import Definitions.Def_BosonicLearn_FockPOVM

open QAlgorithms.Papers.Independent.Dependency50.P33 (IsDensity)
open MeasureTheory ProbabilityTheory
open scoped ENNReal Matrix
Formal statement
namespace BosonicLearn

/-- **Sequential composition of a POVM with an outcome-dependent POVM on a second system.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §5.1, p. 30: "A non-entangled adaptive learning protocol on N copies can be specified by a family of
sequences of POVMs {M_{x_1}}, …, {M^{(x_{<t})}_{x_t}}, … Here, the t-th POVM can depend on the first t − 1
measurement outcomes x_{<t}"; also §4, p. 19 ("the choice of POVM may depend on all previous classical outcomes").

Let `M` be a POVM on `ℓ²(J)` with outcomes in `Ω`, and for every `x ∈ Ω` let `Q_x` be a POVM on `ℓ²(I')` with
outcomes in `O`, depending measurably on `x`: for every density operator `τ` on `ℓ²(I')` the map
`x ↦ (outcome law of Q_x on τ)` is a Markov kernel `κ_τ` from `Ω` to `O`. Then there is a POVM `M'` on
`ℓ²(J × I') = ℓ²(J) ⊗ ℓ²(I')` with outcomes in `Ω × O` such that for all density operators `ρ` on `ℓ²(J)` and `τ` on
`ℓ²(I')` the outcome law of `M'` on `ρ ⊗ τ` is the law of the pair `(x, o)` where `x` has the law of `M` on `ρ` and,
given `x`, `o` has the law of `Q_x` on `τ`.

**Formalization Note.** `ρ ⊗ τ` has entries `ρ(u, v) τ(a, b)` at `((u, a), (v, b))`. The joint law is the
composition-product `(M.law ρ) ⊗ₘ κ_τ`. The measurable dependence on `x` is given as data: kernels `κ τ` for
every coefficient array `τ`, required to be Markov and to agree with `(Q x).law τ` on densities; this is the form
in which `AdaptiveScheme` provides its round laws. `M'` is a `FockPOVM`, so its Born law is a probability measure
on every density of the two systems, entangled ones included. -/
theorem fockPOVM_compose {J I' Ω O : Type} [DecidableEq J] [DecidableEq I'] [MeasurableSpace Ω]
    [MeasurableSpace O] (M : FockPOVM J Ω) (Q : Ω → FockPOVM I' O)
    (κ : Matrix I' I' ℂ → Kernel Ω O) (hκ : ∀ τ, IsMarkovKernel (κ τ))
    (hκQ : ∀ τ, IsDensity τ → ∀ x, κ τ x = (Q x).law τ) :
    ∃ M' : FockPOVM (J × I') (Ω × O), ∀ (ρ : Matrix J J ℂ) (τ : Matrix I' I' ℂ),
      IsDensity ρ → IsDensity τ →
        M'.law (fun p q => ρ p.1 q.1 * τ p.2 q.2) = (M.law ρ) ⊗ₘ (κ τ) := by sorry

end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, §5.1, p. 30 ("the t-th POVM can depend on the first t − 1 measurement outcomes x_{<t}"); §4, p. 19 ("the choice of POVM may depend on all previous classical outcomes")

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me