A Gaussian chaos bound: for independent standard Gaussian vectors when
ProvedBosonicLearn.gaussian_bilinear_exp_boundbosonic-gaussian-statesquantum-informationquantum-learning
The classical Gaussian integral that produces the square of the relative covariance error in a proof of Theorem 7.
Let be independent standard Gaussian vectors in and a real symmetric matrix with . Then is integrable and
The exact value is . Proof idea: integrating out gives ; writing with unit vectors and , Jensen's inequality gives .
Formalization Note The measure is stdGaussian (EuclideanSpace ℝ ι) and its product with itself.
Preamble
import Mathlib open MeasureTheory ProbabilityTheory open scoped Matrix
Formal statement
namespace BosonicLearn
/-- **A Gaussian chaos bound: `𝔼 exp(gᵀ M g′) ≤ 1 + Tr M²` when `Tr M² ≤ ½`.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §7.1, Theorem 7, Eqs. (184)–(185), p. 55 (the bound is quadratic in the relative covariance error `Δ_rel`); this is the
classical Gaussian integral that produces `Δ_rel²` in a `χ²` proof of that bound.
Let `g, g′` be independent standard Gaussian vectors in `ℝ^ι` and `M` a real symmetric matrix with
`Tr M² ≤ ½`. Then `exp(gᵀ M g′)` is integrable and `𝔼 exp(gᵀ M g′) ≤ 1 + Tr M²`.
**Formalization Note.** The exact value is `det(1 − M²)^{−1/2}`. Integrating out `g′` gives `exp(½|Mg|²)`; writing
`|Mg|² = ∑ₖ t_k ⟨o_k, g⟩²` with unit vectors `o_k` (the normalized rows) and `∑ t_k = Tr M² =: T`, Jensen's
inequality gives `𝔼 exp(½|Mg|²) ≤ (1 − T)^{−1/2} ≤ 1 + T`. -/
theorem gaussian_bilinear_exp_bound {ι : Type} [Fintype ι] [DecidableEq ι] (M : Matrix ι ι ℝ)
(hM : M.IsSymm) (hT : (M * M).trace ≤ 1 / 2) :
Integrable (fun q : EuclideanSpace ℝ ι × EuclideanSpace ℝ ι =>
Real.exp (∑ i, ∑ j, q.1 i * M i j * q.2 j))
((stdGaussian (EuclideanSpace ℝ ι)).prod (stdGaussian (EuclideanSpace ℝ ι))) ∧
∫ q : EuclideanSpace ℝ ι × EuclideanSpace ℝ ι, Real.exp (∑ i, ∑ j, q.1 i * M i j * q.2 j)
∂((stdGaussian (EuclideanSpace ℝ ι)).prod (stdGaussian (EuclideanSpace ℝ ι))) ≤
1 + (M * M).trace := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, §7.1, Theorem 7, Eqs. (184)–(185), p. 55