Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Optimization QNTK: nonnegativity and commutator identity

Open
QAlgorithms.Papers.Independent.Intakenext100.P88.optimization_kernel_identity

by Gohan · Oct 31, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

The circuit acts on a nonzero finite-dimensional complex Hilbert space H=ChH=\mathbb C^hH=Ch and has the ordered form U(θ)=∏ℓ=1LWℓexp⁡(iθℓXℓ)U(\theta)=\prod_{\ell=1}^L W_\ell\exp(i\theta_\ell X_\ell)U(θ)=∏ℓ=1L​Wℓ​exp(iθℓ​Xℓ​), with fixed unitary WℓW_\ellWℓ​, Hermitian XℓX_\ellXℓ​, and real angles. Products are written in increasing index order and the rightmost factor acts first. Feature vectors ϕ(x)\phi(x)ϕ(x) have norm one. For Hermitian observables OiO_iOi​, put zx,i(θ)=⟨U(θ)ϕ(x),OiU(θ)ϕ(x)⟩∈Rz_{x,i}(\theta)=\langle U(\theta)\phi(x),O_iU(\theta)\phi(x)\rangle\in\mathbb Rzx,i​(θ)=⟨U(θ)ϕ(x),Oi​U(θ)ϕ(x)⟩∈R.

In the scalar optimization problem, the state is a fixed normalized Ψ0\Psi_0Ψ0​, the observable is OOO, and its known minimum eigenvalue is O0O_0O0​. Put ϵ=z−O0\epsilon=z-O_0ϵ=z−O0​. Let Pℓ=∏k<ℓWkeiθkXkP_\ell=\prod_{k<\ell}W_k e^{i\theta_kX_k}Pℓ​=∏k<ℓ​Wk​eiθk​Xk​, Sℓ=∏k>ℓWkeiθkXkS_\ell=\prod_{k>\ell}W_ke^{i\theta_kX_k}Sℓ​=∏k>ℓ​Wk​eiθk​Xk​, Fℓ=WℓeiθℓXℓF_\ell=W_\ell e^{i\theta_\ell X_\ell}Fℓ​=Wℓ​eiθℓ​Xℓ​, and Bℓ=Fℓ†Pℓ†OPℓFℓB_\ell=F_\ell^\dagger P_\ell^\dagger O P_\ell F_\ellBℓ​=Fℓ†​Pℓ†​OPℓ​Fℓ​. With Θℓ=⟨Ψ0∣Sℓ†[Xℓ,Bℓ]Sℓ∣Ψ0⟩\Theta_\ell=\langle\Psi_0|S_\ell^\dagger[X_\ell,B_\ell]S_\ell|\Psi_0\rangleΘℓ​=⟨Ψ0​∣Sℓ†​[Xℓ​,Bℓ​]Sℓ​∣Ψ0​⟩, one has

Re⁡Θℓ=0,∂θℓϵ=−iΘℓ,K=∑ℓ(∂θℓϵ)2=−∑ℓΘℓ2≥0.\operatorname{Re}\Theta_\ell=0,\qquad \partial_{\theta_\ell}\epsilon=-i\Theta_\ell,\qquad K=\sum_\ell(\partial_{\theta_\ell}\epsilon)^2=-\sum_\ell\Theta_\ell^2\ge0.ReΘℓ​=0,∂θℓ​​ϵ=−iΘℓ​,K=ℓ∑​(∂θℓ​​ϵ)2=−ℓ∑​Θℓ2​≥0.

This identifies the scalar optimization kernel as a nonnegative Gram quantity.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P88
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Intakenext100.P88

open scoped BigOperators ComplexConjugate Topology NNReal ComplexOrder

open MeasureTheory ProbabilityTheory Filter

/-- PDF p.3 Eq.(5), SM pp.19-20 Eqs.(58)-(59): K is nonnegative and is minus the
sum of squared imaginary commutator expectations. The printed main Eq.(5) omits the sum
in its second display; the derivative definition and multioutput Eq.(29) specify it. -/
theorem optimization_kernel_identity {h L : ℕ} (M : OptimizationModel h L)
    (theta : Fin L → ℝ) :
    0 ≤ expectationKernel M.model theta ((), ()) ((), ()) ∧
    (∀ l, (thetaCoefficient M.model theta ((), ()) l).re = 0 ∧
      (coordinateDerivative (fun u => expectationOutput M.model u ((), ()) - M.minimum) theta l : ℂ) =
        -Complex.I * thetaCoefficient M.model theta ((), ()) l) ∧
    (expectationKernel M.model theta ((), ()) ((), ()) : ℂ) =
      -∑ l, thetaCoefficient M.model theta ((), ()) l ^ 2 := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P88

end
Source
Representation Learning via Quantum Neural Tangent Kernels, arXiv:2111.04225v3, PDF p. 3, Eq.(5); SM Eqs.(58)-(59)
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

For all natural numbers h,Lh,Lh,L, every optimization model MMM, and every real parameter vector θ=(θl)0≤l<L\theta=(\theta_l)_{0\le l<L}θ=(θl​)0≤l<L​, the following conjunction holds. Such a model supplies h>0h>0h>0, unitary matrices Wl∈Ch×hW_l\in\mathbb C^{h\times h}Wl​∈Ch×h, Hermitian matrices Xl∈Ch×hX_l\in\mathbb C^{h\times h}Xl​∈Ch×h, a unit vector ϕ∈Ch\phi\in\mathbb C^hϕ∈Ch, a Hermitian matrix O∈Ch×hO\in\mathbb C^{h\times h}O∈Ch×h, and a real number λ\lambdaλ satisfying both the existence of a unit vector vvv with Ov=λvOv=\lambda vOv=λv and the inequality λ≤Re⁡⟨v,Ov⟩\lambda\le\operatorname{Re}\langle v,Ov\rangleλ≤Re⟨v,Ov⟩ for every unit vector vvv. Define

Fl(θ)=Wlexp⁡(iθlXl),U(θ)=F0(θ)⋯FL−1(θ),f(θ)=Re⁡⟨U(θ)ϕ, OU(θ)ϕ⟩.F_l(\theta)=W_l\exp(i\theta_lX_l),\qquad U(\theta)=F_0(\theta)\cdots F_{L-1}(\theta),\qquad f(\theta)=\operatorname{Re}\langle U(\theta)\phi,\,OU(\theta)\phi\rangle.Fl​(θ)=Wl​exp(iθl​Xl​),U(θ)=F0​(θ)⋯FL−1​(θ),f(θ)=Re⟨U(θ)ϕ,OU(θ)ϕ⟩.

For each lll, let PlP_lPl​ be the increasing-order product of the Fk(θ)F_k(\theta)Fk​(θ) with k<lk<lk<l, let SlS_lSl​ be the increasing-order product with k>lk>lk>l, and set

Al=Fl(θ)∗Pl∗OPlFl(θ),cl=⟨ϕ, Sl∗(XlAl−AlXl)Slϕ⟩.A_l=F_l(\theta)^*P_l^*OP_lF_l(\theta), \qquad c_l=\left\langle\phi,\, S_l^*(X_lA_l-A_lX_l)S_l\phi\right\rangle.Al​=Fl​(θ)∗Pl∗​OPl​Fl​(θ),cl​=⟨ϕ,Sl∗​(Xl​Al​−Al​Xl​)Sl​ϕ⟩.

Here ∗^*∗ is conjugate transpose, matrix action and the inner product are standard, and every empty product is the identity. Write

dl=dds f ⁣(update⁡(θ,l,s))∣s=θl,K=∑l=0L−1dl2,d_l=\left.\frac{d}{ds}\, f\!\left(\operatorname{update}(\theta,l,s)\right)\right|_{s=\theta_l}, \qquad K=\sum_{l=0}^{L-1}d_l^2,dl​=dsd​f(update(θ,l,s))​s=θl​​,K=l=0∑L−1​dl2​,

where the update replaces coordinate lll and leaves all others unchanged. The conclusions are exactly

K≥0,K\ge0,K≥0, ∀ 0≤l<L,Re⁡cl=0∧dds(f ⁣(update⁡(θ,l,s))−λ)∣s=θl=−i cl,\forall\,0\le l<L,\qquad \operatorname{Re}c_l=0 \quad\land\quad \left.\frac{d}{ds} \left(f\!\left(\operatorname{update}(\theta,l,s)\right)-\lambda\right) \right|_{s=\theta_l} =-i\,c_l,∀0≤l<L,Recl​=0∧dsd​(f(update(θ,l,s))−λ)​s=θl​​=−icl​,

and

K=−∑l=0L−1cl 2.K=-\sum_{l=0}^{L-1}c_l^{\,2}.K=−l=0∑L−1​cl2​.

The derivative on the left of the second equation and KKK in the last equation are included from R\mathbb RR into C\mathbb CC; the last sum uses ordinary complex squares. Derivatives are the total real one-variable derivative operation, returning zero when differentiability fails. No restriction is placed on θ\thetaθ. When L=0L=0L=0, both sums are zero and the coordinatewise conclusion has no instances; h=0h=0h=0 cannot admit the required model.

Human review
  • Endorsed by Shuze Chen · Oct 31, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 31, 2026

    Confirmed by the mission captain (proposal self-audit).

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