Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 18 for Σ⪰12I\Sigma\succeq\tfrac12IΣ⪰21​I with constant 1/21/\sqrt21/2​: 12∥ρ(μ′,Σ)−ρ(μ,Σ)∥1≤12∥Σ−1/2(μ′−μ)∥2\tfrac12\|\rho(\mu',\Sigma)-\rho(\mu,\Sigma)\|_1\le\tfrac1{\sqrt2}\|\Sigma^{-1/2}(\mu'-\mu)\|_221​∥ρ(μ′,Σ)−ρ(μ,Σ)∥1​≤2​1​∥Σ−1/2(μ′−μ)∥2​

Proved
BosonicLearn.lemma_18_classical

by Senrui Chen · Nov 13, 2026 · Mathlib 0df444a (Lean v4.33.1)

bosonic-gaussian-statesquantum-informationquantum-learning

A first-moment perturbation bound for Gaussian states with classical covariance, proved without the Gaussian fidelity formula.

Let Σ\SigmaΣ be a covariance matrix with Σ⪰121\Sigma\succeq\tfrac12\mathbf 1Σ⪰21​1 and let ρ(μ,Σ)\rho(\mu,\Sigma)ρ(μ,Σ), ρ(μ′,Σ)\rho(\mu',\Sigma)ρ(μ′,Σ) be Gaussian states. Then

12∥ρ(μ′,Σ)−ρ(μ,Σ)∥1 ≤ 12(μ′−μ)TΣ−1(μ′−μ).\tfrac12\big\|\rho(\mu',\Sigma)-\rho(\mu,\Sigma)\big\|_1\ \le\ \frac1{\sqrt2}\sqrt{(\mu'-\mu)^T\Sigma^{-1}(\mu'-\mu)} .21​​ρ(μ′,Σ)−ρ(μ,Σ)​1​ ≤ 2​1​(μ′−μ)TΣ−1(μ′−μ)​.

Comparison with Lemma 18 of the paper (BosonicLearn.lemma_18): same right-hand side up to the constant (12\tfrac1{\sqrt2}2​1​ instead of 12\tfrac1221​), under the extra hypothesis Σ⪰121\Sigma\succeq\tfrac12\mathbf 1Σ⪰21​1. Both tomography theorems (Theorem 8 and [BMEM25, Theorem 10]) assume Σ⪰121\Sigma\succeq\tfrac12\mathbf 1Σ⪰21​1 and have unspecified constants, so this version suffices there.

Proof idea. ρ(μ,Σ)\rho(\mu,\Sigma)ρ(μ,Σ) is a mixture of coherent states over N(μ,Σ−121)\mathcal N(\mu,\Sigma-\tfrac12\mathbf 1)N(μ,Σ−21​1). The shift δ=μ′−μ\delta=\mu'-\muδ=μ′−μ is split as (Σ−12)Σ−1δ(\Sigma-\tfrac12)\Sigma^{-1}\delta(Σ−21​)Σ−1δ, absorbed in the mixing measure at total-variation cost ≤A/2π\le\sqrt A/\sqrt{2\pi}≤A​/2π​, plus 12Σ−1δ\tfrac12\Sigma^{-1}\delta21​Σ−1δ, a displacement of the coherent states at cost ≤B/2\le\sqrt B/2≤B​/2, where A+B=δTΣ−1δA+B=\delta^T\Sigma^{-1}\deltaA+B=δTΣ−1δ.

Formalization Note Same formal shape as BosonicLearn.lemma_18, with the hypothesis (S - (1/2) • 1).PosSemidef and the bound ENNReal.ofReal (Real.sqrt (…) / Real.sqrt 2).

Preamble
import Mathlib
import Definitions.Def_BosonicLearn_traceDist
import Definitions.Def_BosonicLearn_IsGaussianState

open QAlgorithms.Papers.Independent.Dependency50.P33 (State traceSize)
open QAlgorithms.Papers.Independent.Claimed50.P47 (Occupation)
open MeasureTheory
open scoped ENNReal Matrix
Formal statement
namespace BosonicLearn

/-- **Lemma 18 for classical covariances, with constant `1/√2`.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, Lemma 18 (first-moment perturbation), Eq. (204), p. 60: "For every valid covariance matrix Σ and all µ, µ′ ∈ ℝ^{2n},
½‖ρ(µ′, Σ) − ρ(µ, Σ)‖₁ ≤ ½‖Σ^{−1/2}(µ′ − µ)‖₂."

If `Σ ⪰ ½I`, then for Gaussian states `ρ(µ, Σ)` and `ρ(µ′, Σ)`,
`½‖ρ(µ′, Σ) − ρ(µ, Σ)‖₁ ≤ (1/√2) √((µ′ − µ)ᵀ Σ⁻¹ (µ′ − µ))`.

**Formalization Note.** Same statement as `lemma_18` except for the hypothesis `Σ − ½I ⪰ 0` and the constant `1/√2`
in place of `½`. It is proved without the Gaussian fidelity formula: `ρ(µ, Σ)` is a mixture of coherent states over
`N(µ, Σ − ½I)`; the shift `δ = µ′ − µ` is split as `(Σ − ½I)Σ⁻¹δ` (absorbed in the mixing measure, total variation
cost) plus `½Σ⁻¹δ` (a displacement of the coherent states). The promise `Σ ⪰ ½I` holds in Theorem 8 and in
[BMEM25, Theorem 10], and the weaker constant is absorbed in their constants. -/
theorem lemma_18_classical {n : ℕ} (ρ ρ' : State (Occupation (Fin n))) (μ μ' : Fin n ⊕ Fin n → ℝ)
    (S : Matrix (Fin n ⊕ Fin n) (Fin n ⊕ Fin n) ℝ)
    (hS : (S - (1 / 2 : ℝ) • (1 : Matrix (Fin n ⊕ Fin n) (Fin n ⊕ Fin n) ℝ)).PosSemidef)
    (h : IsGaussianState ρ μ S) (h' : IsGaussianState ρ' μ' S) :
    traceDist ρ'.coefficients ρ.coefficients ≤
      ENNReal.ofReal (Real.sqrt ((μ' - μ) ⬝ᵥ (S⁻¹ *ᵥ (μ' - μ))) / Real.sqrt 2) := by sorry

end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, Lemma 18 (first-moment perturbation), Eq. (204), p. 60

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me