Rényi-2 overlap of with a thermal state:
ProvedBosonicLearn.fugacity_renyi2_tensorPowbosonic-gaussian-statesquantum-informationquantum-learning
A closed form for when is copies of a passive Gaussian state and is copies of a thermal state.
Let be a fugacity matrix () with , , let and (the diagonal of the thermal state of fugacity ). Then is a density operator, the weights on are positive with sum , , and
Proof idea: is block diagonal in the photon number, , is multiplicative and ; sums over copies factorize.
Formalization Note fugacityCoeff R is ; the tensor power is WildeQIT.Schumacher.tensorPow. Real parts of the (positive real) determinants are used.
Preamble
import Mathlib import Definitions.Def_BosonicLearn_fugacityCoeff import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P33 import Definitions.Def_WildeQIT_Schumacher_tensorPow open QAlgorithms.Papers.Independent.Dependency50.P33 (State IsDensity traceSize) open QAlgorithms.Papers.Independent.Claimed50.P47 (Occupation) open scoped ENNReal Matrix ComplexOrder
Formal statement
namespace BosonicLearn
/-- **Rényi-2 overlap of `N` copies of a passive Gaussian state with a thermal reference state.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §8, proof of Theorem 9, Eqs. (230)–(234), pp. 66–67 (the Holevo quantity of the ensemble `{ρ_{U_a}^{⊗N}}` is bounded
through the pairwise divergences of single copies); here the divergence is the Rényi-2 quantity
`Tr(ρ σ^{−1/2} ρ σ^{−1/2})` relative to the thermal state `σ = (1 − r₀)ⁿ Γ(r₀·1)`.
Let `R` be a fugacity matrix (`0 ⪯ R ≺ 1`) with `R² ≺ r₀·1`, `0 < r₀ < 1`, and let `p₀(u) = (1 − r₀)ⁿ r₀^{|u|}` be the
diagonal of the thermal state. Then `ρ_R^{⊗N}` is a density operator, the weights `∏ₜ p₀(x_t)` on `(ℕⁿ)^N` are positive
with sum `1`, and
`∑_{x,y} |(ρ_R^{⊗N})_{xy}|² / √(∏ₜ p₀(x_t) ∏ₜ p₀(y_t)) = ( det(1 − R)² / ((1 − r₀)ⁿ det(1 − R²/r₀)) )^N`.
**Formalization Note.** `ρ_R = det(1 − R)Γ(R)` is `fugacityCoeff R`; `ρ^{⊗N}` is `WildeQIT.Schumacher.tensorPow`. For one
copy, `Γ(R)` is block diagonal in the photon number, `Γ(r₀^{−1/2}R)_{uv} = r₀^{−|u|/2}Γ(R)_{uv}`, `Γ` is multiplicative and
`Tr Γ(A) = 1/det(1 − A)`. Both determinants are positive reals; their real parts are used. -/
theorem fugacity_renyi2_tensorPow {n : ℕ} (R : Matrix (Fin n) (Fin n) ℂ) (hR : R.PosSemidef)
(hR1 : (1 - R).PosDef) (r₀ : ℝ) (h0 : 0 < r₀) (h1 : r₀ < 1)
(hR2 : (((r₀ : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) - R * R).PosDef) (N : ℕ) :
IsDensity (WildeQIT.Schumacher.tensorPow N (fugacityCoeff R)) ∧
HasSum (fun x : Fin N → Occupation (Fin n) => ∏ t, ((1 - r₀) ^ n * r₀ ^ (∑ i, x t i))) 1 ∧
0 < (1 - ((r₀⁻¹ : ℝ) : ℂ) • (R * R)).det.re ∧
HasSum (fun x : (Fin N → Occupation (Fin n)) × (Fin N → Occupation (Fin n)) =>
‖WildeQIT.Schumacher.tensorPow N (fugacityCoeff R) x.1 x.2‖ ^ 2 /
Real.sqrt ((∏ t, ((1 - r₀) ^ n * r₀ ^ (∑ i, x.1 t i))) *
∏ t, ((1 - r₀) ^ n * r₀ ^ (∑ i, x.2 t i))))
(((1 - R).det.re ^ 2 /
((1 - r₀) ^ n * (1 - ((r₀⁻¹ : ℝ) : ℂ) • (R * R)).det.re)) ^ N) := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, §8, proof of Theorem 9, pp. 65–67 (Eqs. (230)–(234): the divergence of the ensemble is bounded through single copies)