Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

The χ2\chi^2χ2-kernel of displaced thermal states: Tr[τzτ−1/2τz′τ−1/2]=exp⁡(2κ Re⟨z,z′⟩)\mathrm{Tr}[\tau_z\tau^{-1/2}\tau_{z'}\tau^{-1/2}]=\exp(2\kappa\,\mathrm{Re}\langle z,z'\rangle)Tr[τz​τ−1/2τz′​τ−1/2]=exp(2κRe⟨z,z′⟩), κ=(1−r)/r\kappa=(1-r)/\sqrt rκ=(1−r)/r​

Proved
BosonicLearn.thermal_displaced_kernel

by Senrui Chen · Nov 14, 2026 · Mathlib 0df444a (Lean v4.33.1)

bosonic-gaussian-statesquantum-informationquantum-learning

Displaced thermal states are "coherent vectors" for the inner product ⟨X,Y⟩τ=Tr⁡[Xτ−1/2Yτ−1/2]\langle X,Y\rangle_\tau=\operatorname{Tr}[X\tau^{-1/2}Y\tau^{-1/2}]⟨X,Y⟩τ​=Tr[Xτ−1/2Yτ−1/2]. This identity is the key computation in a proof of the one-sided relative covariance bound (Theorem 7) through a χ2\chi^2χ2-divergence relative to a thermal state.

Let τ=((1−r) ra†a)⊗n\tau=((1-r)\,r^{a^\dagger a})^{\otimes n}τ=((1−r)ra†a)⊗n with 0<r<10<r<10<r<1, Fock weights pw=(1−r)nr∣w∣p_w=(1-r)^nr^{|w|}pw​=(1−r)nr∣w∣, and τz=D(z)τD(z)†\tau_z=D(z)\tau D(z)^\daggerτz​=D(z)τD(z)† for z∈Cnz\in\mathbb C^nz∈Cn, i.e. (τz)uv=∑wD(z)uw pw D(z)vw‾(\tau_z)_{uv}=\sum_wD(z)_{uw}\,p_w\,\overline{D(z)_{vw}}(τz​)uv​=∑w​D(z)uw​pw​D(z)vw​​. Then for all z,z′∈Cnz,z'\in\mathbb C^nz,z′∈Cn

∑u,v(τz)uv (τz′)uv‾pupv=exp⁡(κ∑j(zˉjzj′+zjzˉj′)),κ=1−rr,\sum_{u,v}\frac{(\tau_z)_{uv}\,\overline{(\tau_{z'})_{uv}}}{\sqrt{p_up_v}}=\exp\Big(\kappa\sum_j(\bar z_jz'_j+z_j\bar z'_j)\Big),\qquad\kappa=\frac{1-r}{\sqrt r},u,v∑​pu​pv​​(τz​)uv​(τz′​)uv​​​=exp(κj∑​(zˉj​zj′​+zj​zˉj′​)),κ=r​1−r​,

and the double series converges absolutely. For r=ν/(1+ν)r=\nu/(1+\nu)r=ν/(1+ν) (covariance (12+ν)I(\frac12+\nu)I(21​+ν)I), κ=1/ν(1+ν)\kappa=1/\sqrt{\nu(1+\nu)}κ=1/ν(1+ν)​.

Proof idea: with W(x,y)=exa†eyaW(x,y)=e^{xa^\dagger}e^{ya}W(x,y)=exa†eya one has W(x,y)W(x′,y′)=eyx′W(x+x′,y+y′)W(x,y)W(x',y')=e^{yx'}W(x+x',y+y')W(x,y)W(x′,y′)=eyx′W(x+x′,y+y′), D(z)=e−∣z∣2/2W(z,−zˉ)D(z)=e^{-|z|^2/2}W(z,-\bar z)D(z)=e−∣z∣2/2W(z,−zˉ), rsNW(x,y)r−sN=W(rsx,r−sy)r^{sN}W(x,y)r^{-sN}=W(r^sx,r^{-s}y)rsNW(x,y)r−sN=W(rsx,r−sy) and Tr⁡[rNW(x,y)]=(1−r)−1erxy/(1−r)\operatorname{Tr}[r^NW(x,y)]=(1-r)^{-1}e^{rxy/(1-r)}Tr[rNW(x,y)]=(1−r)−1erxy/(1−r). Checked numerically (error <10−11<10^{-11}<10−11).

Formalization Note ⟨m∣D(z)∣k⟩\langle m|D(z)|k\rangle⟨m∣D(z)∣k⟩ is the published displacementMatrix z m k; the weights and the arrays τz\tau_zτz​ are introduced by equations; the first conclusion is the absolute convergence of the series defining (τz)uv(\tau_z)_{uv}(τz​)uv​.

Preamble
import Mathlib
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P39

open QAlgorithms.Papers.Independent.Dependency50.P39 (displacementMatrix)
Formal statement
namespace BosonicLearn

/-- **The `χ²`-kernel of displaced thermal states: `Tr[τ_z τ^{−1/2} τ_{z′} τ^{−1/2}] = exp(2κ Re⟨z, z′⟩)`.**
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 (one-sided relative covariance bound); this identity is the key
computation of a proof of that bound through a `χ²`-divergence relative to a thermal state.

Let `τ = ((1 − r) r^{a†a})^{⊗n}`, `0 < r < 1`, with Fock weights `p_w = (1 − r)ⁿ r^{|w|}`, and
`τ_z = D(z) τ D(z)†` the displaced thermal state, `(τ_z)_{uv} = ∑_w D(z)_{uw} p_w \overline{D(z)_{vw}}`
(`D(z)_{uw} = ∏ⱼ ⟨u_j|D(z_j)|w_j⟩`). Then, with `κ = (1 − r)/√r`,

`∑_{u,v} (τ_z)_{uv} \overline{(τ_{z′})_{uv}} / √(p_u p_v) = exp(κ ∑ⱼ (z̄ⱼ z′ⱼ + zⱼ z̄′ⱼ))`,

and the double series converges absolutely.

**Formalization Note.** `⟨m|D(z)|k⟩` is the published `displacementMatrix z m k`. The weights and the arrays `τ_z`
are introduced by equations. The first conclusion is the absolute convergence of the series defining `(τ_z)_{uv}`.
For `r = ν/(1 + ν)` (covariance `(½ + ν)I`), `κ = 1/√(ν(1 + ν))`. -/
theorem thermal_displaced_kernel {n : ℕ} (r : ℝ) (hr0 : 0 < r) (hr1 : r < 1) (z z' : Fin n → ℂ)
    (p : (Fin n → ℕ) → ℝ) (hp : p = fun w => (1 - r) ^ n * r ^ (∑ i, w i))
    (T : (Fin n → ℂ) → (Fin n → ℕ) → (Fin n → ℕ) → ℂ)
    (hT : T = fun z u v => ∑' w : Fin n → ℕ,
      (∏ j, displacementMatrix (z j) (u j) (w j)) * ((p w : ℝ) : ℂ) *
        star (∏ j, displacementMatrix (z j) (v j) (w j))) :
    (∀ (y : Fin n → ℂ) (u v : Fin n → ℕ), Summable fun w : Fin n → ℕ =>
      ‖(∏ j, displacementMatrix (y j) (u j) (w j)) * ((p w : ℝ) : ℂ) *
        star (∏ j, displacementMatrix (y j) (v j) (w j))‖) ∧
    Summable (fun x : (Fin n → ℕ) × (Fin n → ℕ) =>
      ‖T z x.1 x.2‖ * ‖T z' x.1 x.2‖ / Real.sqrt (p x.1 * p x.2)) ∧
    HasSum (fun x : (Fin n → ℕ) × (Fin n → ℕ) =>
      T z x.1 x.2 * star (T z' x.1 x.2) / ((Real.sqrt (p x.1 * p x.2) : ℝ) : ℂ))
      (Complex.exp ((((1 - r) / Real.sqrt r : ℝ) : ℂ) *
        ∑ j, (star (z j) * z' j + z j * star (z' j)))) := 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

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