The hard family for hot states: occupation matrices give passive Gaussian states of the promise class with Rényi-2 overlap
ProvedBosonicLearn.occupation_familybosonic-gaussian-statesquantum-informationquantum-learning
The ensemble used for Theorem 9, parametrized so that the occupation matrix (not the fugacity matrix) is linear in ; all bounds are uniform in the temperature scale .
Let be a Hermitian matrix with , , , , and . Then
- , , and ;
- for every there is a density operator with Fock coefficients which is a passive Gaussian state with ;
- .
The quantity in 3 is for the thermal state of fugacity (see BosonicLearn.fugacity_renyi2_tensorPow); in an eigenbasis of it is .
Formalization Note and are introduced by equations Nm = …, R = Nm * (1 + Nm)⁻¹. The determinants are positive reals; their real parts are used.
Preamble
import Mathlib import Definitions.Def_BosonicLearn_fugacityCoeff import Definitions.Def_BosonicLearn_IsPassive import Definitions.Def_BosonicLearn_promiseClass 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
/-- **The hard family for hot states: occupation matrices `N_X = m(1 + ξX)`.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §8, proof of Theorem 9, Eq. (222), p. 65 (the ensemble `ρ_U` of passive Gaussian states with `Σ ⪰ (½ + ν)I`), in the
parametrization of §5.1, Eqs. (30)–(31), p. 30, with the occupation matrix (rather than the fugacity matrix) linear in `X`.
Let `X` be Hermitian with `−4·1 ⪯ X ⪯ 4·1`, `m ≥ 1`, `0 < ξ ≤ 1/32`, `N = m(1 + ξX)` and `R = N(1 + N)⁻¹`. Then
1. `(7/8)m·1 ⪯ N ⪯ (9/8)m·1`, `0 ⪯ R`, `1 − R` is positive definite, and `R² ≺ r₀·1` for `r₀ = m/(1 + m)`;
2. for every `0 < ν ≤ (7/8)m` there is a state with Fock coefficients `ρ_R = det(1 − R)Γ(R)` which is a passive Gaussian
state of the promise class `Σ ⪰ (½ + ν)I`;
3. the Rényi-2 overlap with the thermal state of fugacity `r₀` satisfies
`det(1 − R)² / ((1 − r₀)ⁿ det(1 − R²/r₀)) ≤ exp(21 ξ² n)`.
**Formalization Note.** In an eigenbasis of `X` with eigenvalues `x_i`, `N_i = m(1 + ξx_i)` and the quantity in 3 is
`∏ᵢ (1 + m²ξ²x_i²/(m + 2mN_i − N_i²)) ≤ ∏ᵢ (1 + (64/49)ξ²x_i²)`. Real parts of the (positive real) determinants are used. -/
theorem occupation_family {n : ℕ} (X : Matrix (Fin n) (Fin n) ℂ) (hX : X.IsHermitian)
(h4 : (((4 : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) - X).PosSemidef)
(h4' : (((4 : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) + X).PosSemidef)
(m ξ : ℝ) (hm : 1 ≤ m) (hξ0 : 0 < ξ) (hξ : ξ ≤ 1 / 32)
(Nm R : Matrix (Fin n) (Fin n) ℂ)
(hN : Nm = ((m : ℝ) : ℂ) • ((1 : Matrix (Fin n) (Fin n) ℂ) + ((ξ : ℝ) : ℂ) • X))
(hRdef : R = Nm * (1 + Nm)⁻¹) :
(Nm - ((7 / 8 * m : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ)).PosSemidef ∧
(((9 / 8 * m : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) - Nm).PosSemidef ∧
R.PosSemidef ∧ (1 - R).PosDef ∧
(((m / (1 + m) : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) - R * R).PosDef ∧
(∀ ν : ℝ, 0 < ν → ν ≤ 7 / 8 * m →
∃ ρ : State (Occupation (Fin n)), ρ.coefficients = fugacityCoeff R ∧ IsPassive ρ ∧
ρ ∈ promiseClass n ν) ∧
(1 - R).det.re ^ 2 /
((1 - m / (1 + m)) ^ n * (1 - (((m / (1 + m))⁻¹ : ℝ) : ℂ) • (R * R)).det.re) ≤
Real.exp (21 * ξ ^ 2 * 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 (Eq. (222): the ensemble of passive Gaussian states with ); parametrization of §5.1, Eqs. (30)–(31), p. 30