Trace pairing: for a bounded operator and a nuclear expansion
ProvedBosonicLearn.effect_trace_pairingThe trace of a bounded operator against a trace-class matrix can be computed along any nuclear expansion, and all the series involved converge.
Let be any index set, a complex matrix and a bounded operator on whose coordinates are given by the rows of : for every and every , the series converges to . Let be a matrix with a nuclear expansion
Then
- for every the series converges;
- the series converges;
- the series converges;
- the two sums are equal:
Role. The left-hand side of clause 4 is the quantity that appears in the Born rule of a POVM in this mission, written as an iterated series. The statement shows that it is a genuine convergent sum and rewrites it as an absolutely convergent series of inner products, which is the form in which one can pass to limits, integrate over a parameter, or compare two effects. It is a foundation for BosonicLearn.fockPOVM_postprocess and BosonicLearn.adaptive_scheme_product_povm.
Formalization Note Convergence is Summable, and the two sides of clause 4 are tsums; by clauses 1–3 they are genuine sums and not the default value that Lean assigns to a divergent series. The hypothesis on is clause 3 of BosonicLearn.effect_matrix_operator. The nuclear expansion is a NuclearExpansion X of Dependency50.P33. The index set is arbitrary and need not be positive.
Related platform material The space used here (Hilbert A) is the same Lean type as OAI.TensorLp.H A, and is the same type as OAI.EntropyPhotonNumber.Fock n. The published definitions OAI_Analysis_TensorLp (tensor products, slices and operator lifts on ) and OAI_MathematicalPhysics_Fock_Operators (OAI.CrossEnsemble.operator, the bounded operator 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) and BosonicLearn.effect_matrix_operator (effect matrices as bounded positive operators).
import Mathlib import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P33 open QAlgorithms.Papers.Independent.Dependency50.P33 (NuclearExpansion) open QAlgorithms.RealRegisters (Hilbert) open scoped ENNReal
namespace BosonicLearn
/-- **Trace pairing of a bounded operator with a trace-class matrix.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §5.4, Eq. (63), p. 35 (outcome likelihoods are Born probabilities `⟨ϕ|ρ|ϕ⟩ = Tr(|ϕ⟩⟨ϕ| ρ)`). This makes
precise the remark in the definition `BosonicLearn_FockPOVM` that "`Tr(E ρ)` is the iterated series
`∑ᵢ ∑ⱼ E_ij ρ_ji`, which converges absolutely for bounded `E` and trace-class `ρ`".
Let `F` be a bounded operator on `ℓ²(I)` whose coordinates are given by the rows of a matrix `E`,
`(F z)_i = ∑_j E_{ij} z_j`, and let `X` have a nuclear expansion `X_{ij} = ∑_a L_a(i) conj(R_a(j))` with
`∑_a ‖L_a‖‖R_a‖ < ∞`. Then
1. for every `i` the series `∑_j E_{ij} X_{ji}` converges;
2. the series `∑_i (∑_j E_{ij} X_{ji})` converges;
3. the series `∑_a ⟨R_a, F L_a⟩` converges;
4. `∑_i ∑_j E_{ij} X_{ji} = ∑_a ⟨R_a, F L_a⟩`.
**Formalization Note.** Convergence is `Summable` (unconditional); the two sides of clause 4 are `tsum`s, which
by clauses 1–3 are genuine sums and not the default value `0`. The hypothesis on `F` is clause 3 of
`BosonicLearn.effect_matrix_operator`. The left-hand side of clause 4 is the expression in the Born rule
`FockPOVM.law_born`. The index set is arbitrary. -/
theorem effect_trace_pairing {I : Type} (E : Matrix I I ℂ) (F : Hilbert I →L[ℂ] Hilbert I)
(hF : ∀ (z : Hilbert I) (i : I), HasSum (fun j => E i j * z j) (F z i))
(X : Matrix I I ℂ) (e : NuclearExpansion X) :
(∀ i, Summable fun j => E i j * X j i) ∧
Summable (fun i => ∑' j, E i j * X j i) ∧
Summable (fun a => inner ℂ (e.right a) (F (e.left a))) ∧
∑' i, ∑' j, E i j * X j i = ∑' a, inner ℂ (e.right a) (F (e.left a)) := by sorry
end BosonicLearn