Occupation matrices of two far-apart members of the hard family: compressions with
ProvedBosonicLearn.fugacityFamily_occupation_splitThe matrix-analytic part of the proof of Lemma 12.
Setting (§5.1): , , with , for (Hermitian, traceless, ). The photon occupation matrix is .
There are absolute constants such that the following holds. Let , and with . Then, calling the two fugacity matrices in a suitable order , there exist a unitary matrix and a set of modes such that the compressions
( the coordinate projection on ) satisfy
In the paper: diagonalizes , indexes the eigenvalues of the sign for which the positive part has trace at least (), , and the bounds are Eqs. (128), (129), (133), (134), (144), (145).
Formalization Note The paper's is comparable to , so the bounds are stated with . are kept as matrices supported on . The trace norm is WHMultiplicativity.schattenNorm 1, and is Gsupp n. The order is expressed by a disjunction of the two possibilities.
import Mathlib import Definitions.Def_BosonicLearn_fugacityFamily import Definitions.Def_BosonicLearn_Gsupp open scoped ENNReal Matrix ComplexOrder
namespace BosonicLearn
/-- **Occupation matrices of two far-apart members of the hard family (matrix-analytic part of Lemma 12).**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §5.7, proof of Lemma 12, Eqs. (124)–(134): the photon occupation matrix `N_X = R_X(I − R_X)^{−1}`, `Δ = N_{X₁} − N_{X₂}`,
the resolvent identity, `‖Δ‖_tr ≥ c ν̄ ξ ‖X₁ − X₂‖_tr`, `‖Δ‖_op ≤ C ν̄ ξ`, `c ν̄ I ≤ N_X ≤ C ν̄ I`, the projector `Π₊` onto the
positive part of `Δ`, and the compressions `B_± = Π₊ N Π₊` with `B₊ − B₋ = Δ₊ ≥ 0`, `δ_ph = Tr Δ₊`.
There are absolute constants `c, C > 0` such that for `n ≥ 1`, `1/n ≤ ν ≤ 1`, `0 < ξ < 1/32`, `X₁, X₂ ∈ G_supp` and
`a > 0` with `‖X₂ − X₁‖_tr ≥ a n`: after ordering the two fugacity matrices `R_{X₁}, R_{X₂}` suitably as `(R₊, R₋)`,
there are a unitary `U` (diagonalizing `N₊ − N₋`) and a set of modes `A` (its positive eigen-directions) such that
the compressions `B_± = P_A U N_± U† P_A` (`N_± = R_±(1 − R_±)⁻¹`) satisfy `0 ⪯ B₋ ⪯ B₊ ⪯ C·1`,
`c a n ν ξ ≤ Tr(B₊ − B₋) ≤ C |A| ν ξ`, `c |A| ν ≤ Tr B₋`, `Tr(B₋ + B₋²) ≤ C |A| ν`.
**Formalization Note.** `ν̄ = r/(1 − r)` of the paper is comparable to `ν` (`ν/2 ≤ ν̄ ≤ 4ν/3`·const), so the bounds are
stated with `ν`. `B_±` are `n × n` matrices supported on `A × A`. The trace norm is `WHMultiplicativity.schattenNorm 1`. -/
theorem fugacityFamily_occupation_split :
∃ c : ℝ, 0 < c ∧ ∃ C : ℝ, 0 < C ∧
∀ n : ℕ, 1 ≤ n → ∀ ν : ℝ, 1 / (n : ℝ) ≤ ν → ν ≤ 1 → ∀ ξ : ℝ, 0 < ξ → ξ < 1 / 32 →
∀ X₁ ∈ Gsupp n, ∀ X₂ ∈ Gsupp n, ∀ a : ℝ, 0 < a →
a * n ≤ WHMultiplicativity.schattenNorm 1 ((X₂ - X₁ : Herm0 n) : Matrix (Fin n) (Fin n) ℂ) →
∃ Rp Rm : Matrix (Fin n) (Fin n) ℂ,
((Rp = fugacityFamily ν ξ X₁ ∧ Rm = fugacityFamily ν ξ X₂) ∨
(Rp = fugacityFamily ν ξ X₂ ∧ Rm = fugacityFamily ν ξ X₁)) ∧
∃ (U : Matrix (Fin n) (Fin n) ℂ) (A : Finset (Fin n)), Uᴴ * U = 1 ∧
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).PosSemidef ∧
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rp * (1 - Rp)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) -
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).PosSemidef ∧
(((C : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) -
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rp * (1 - Rp)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).PosSemidef ∧
c * a * n * ν * ξ ≤
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rp * (1 - Rp)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) -
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).trace.re ∧
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rp * (1 - Rp)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) -
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).trace.re ≤ C * A.card * ν * ξ ∧
c * A.card * ν ≤
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)).trace.re ∧
((Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)) +
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0)) *
(Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0) * (U * (Rm * (1 - Rm)⁻¹) * Uᴴ) *
Matrix.diagonal (fun i => if i ∈ A then (1 : ℂ) else 0))).trace.re ≤ C * A.card * ν := by sorry
end BosonicLearn