Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Finite scalar dQNTK coefficient response

Proved
QAlgorithms.Papers.Independent.Intakenext100.P88.optimization_response_finite

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.

Absorb the reference angles into the fixed gates and set θ=δφ\theta=\delta\varphiθ=δφ, where δ>0\delta>0δ>0. Write Jaℓ=∂φℓza(0)J_{a\ell}=\partial_{\varphi_\ell}z_a(0)Jaℓ​=∂φℓ​​za​(0) and Ha,kℓ=∂φk∂φℓza(0)H_{a,k\ell}=\partial_{\varphi_k}\partial_{\varphi_\ell}z_a(0)Ha,kℓ​=∂φk​​∂φℓ​​za​(0) for derivatives of the actual circuit. The frozen kernel is Kab=∑ℓJaℓJbℓK_{ab}=\sum_\ell J_{a\ell}J_{b\ell}Kab​=∑ℓ​Jaℓ​Jbℓ​.

In scalar optimization with known minimum eigenvalue O0O_0O0​, let e=z(0)+∑ℓJℓvℓ+12∑k,ℓHkℓvkvℓ−O0e=z(0)+\sum_\ell J_\ell v_\ell+\frac12\sum_{k,\ell}H_{k\ell}v_kv_\ell-O_0e=z(0)+∑ℓ​Jℓ​vℓ​+21​∑k,ℓ​Hkℓ​vk​vℓ​−O0​ and D=2∑k,ℓJℓHkℓvkD=2\sum_{k,\ell}J_\ell H_{k\ell}v_kD=2∑k,ℓ​Jℓ​Hkℓ​vk​. For η>0\eta>0η>0, define the retained coefficient dynamics by F0=eF_0=eF0​=e, I0=0I_0=0I0​=0, Ft+1=(1−ηK)FtF_{t+1}=(1-\eta K)F_tFt+1​=(1−ηK)Ft​ and It+1=(1−ηK)It−ηDFtI_{t+1}=(1-\eta K)I_t-\eta DF_tIt+1​=(1−ηK)It​−ηDFt​. Then

Ft=(1−ηK)te,I0=0,It+1=−η(t+1)(1−ηK)tDe.F_t=(1-\eta K)^t e,\qquad I_0=0,\qquad I_{t+1}=-\eta(t+1)(1-\eta K)^tDe.Ft​=(1−ηK)te,I0​=0,It+1​=−η(t+1)(1−ηK)tDe.

This gives the fixed-time leading perturbative response without asserting convergence of exact nonlinear training or a uniform-in-time remainder.

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 pp.4-5 Eqs.(19)-(21), SM p.22 Eqs.(78)-(81): fixed-time retained scalar
response. K and D are computed from the circuit jets. The t=0 interacting part is separate;
there is no truncated natural exponent and no asserted all-time nonlinear error estimate. -/
theorem optimization_response_finite {h L : ℕ} (M : OptimizationModel h L)
    (delta eta : ℝ) (hdelta : 0 < delta) (heta : 0 < eta) (v : Fin L → ℝ) :
    let K := frozenKernel M.model delta ((), ()) ((), ())
    let D := deltaKernel M.model delta v ((), ()) ((), ())
    let e := quadraticOutput M.model delta v ((), ()) - M.minimum
    let R := retainedResidual (fun _ _ : Unit => K) (fun _ _ : Unit => D)
      eta (WithLp.toLp 2 (fun _ : Unit => e))
    (∀ t, (R t).1 () = (1 - eta * K) ^ t * e) ∧
    (R 0).2 = 0 ∧
    ∀ t : ℕ, (R (t + 1)).2 () =
      -eta * ((t : ℝ) + 1) * (1 - eta * K) ^ t * D * e := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P88

end
Source
Representation Learning via Quantum Neural Tangent Kernels, arXiv:2111.04225v3, PDF p. 22, SM Eqs.(78)-(81); main Eqs.(19)-(21)
Read-back

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

Let h,Lh,Lh,L be natural numbers and let MMM consist of 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\phiϕ, a Hermitian observable OOO, and a real number m∗m_*m∗​ such that Ow=m∗wOw=m_*wOw=m∗​w for some unit vector www, and m∗≤Re⁡⟨w,Ow⟩m_*\le\operatorname{Re}\langle w,Ow\ranglem∗​≤Re⟨w,Ow⟩ for every unit vector www. Define f(θ)=Re⁡⟨U(θ)ϕ,OU(θ)ϕ⟩f(\theta)=\operatorname{Re}\langle U(\theta)\phi,OU(\theta)\phi\ranglef(θ)=Re⟨U(θ)ϕ,OU(θ)ϕ⟩, with U(θ)=∏l=0L−1Wlexp⁡(iθlXl)U(\theta)=\prod_{l=0}^{L-1}W_l\exp(i\theta_lX_l)U(θ)=∏l=0L−1​Wl​exp(iθl​Xl​) in increasing index order. Fix δ,η>0\delta,\eta>0δ,η>0 and v∈RLv\in\mathbb R^Lv∈RL, and put g(u)=f(δu)g(u)=f(\delta u)g(u)=f(δu), Jl=∂lg(0)J_l=\partial_lg(0)Jl​=∂l​g(0), and Hkl=∂k(∂lg)(0)H_{kl}=\partial_k(\partial_lg)(0)Hkl​=∂k​(∂l​g)(0), using real coordinate derivatives. Define K=∑lJl2K=\sum_lJ_l^2K=∑l​Jl2​, D=∑l(Jl∑kHklvk+(∑kHklvk)Jl)D=\sum_l\bigl(J_l\sum_kH_{kl}v_k+(\sum_kH_{kl}v_k)J_l\bigr)D=∑l​(Jl​∑k​Hkl​vk​+(∑k​Hkl​vk​)Jl​), and e=f(0)+∑lJlvl+12∑k,lHklvkvl−m∗e=f(0)+\sum_lJ_lv_l+\frac12\sum_{k,l}H_{kl}v_kv_l-m_*e=f(0)+∑l​Jl​vl​+21​∑k,l​Hkl​vk​vl​−m∗​. Let (pt,qt)(p_t,q_t)(pt​,qt​) be vectors on a one-point index set, identified with their real coordinates, defined by p0=ep_0=ep0​=e, q0=0q_0=0q0​=0, pt+1=(1−ηK)ptp_{t+1}=(1-\eta K)p_tpt+1​=(1−ηK)pt​, and qt+1=(1−ηK)qt−ηDptq_{t+1}=(1-\eta K)q_t-\eta Dp_tqt+1​=(1−ηK)qt​−ηDpt​. Then pt=(1−ηK)tep_t=(1-\eta K)^t ept​=(1−ηK)te for every t∈Nt\in\mathbb Nt∈N, the initial second vector is zero, and qt+1=−η(t+1)(1−ηK)tDeq_{t+1}=-\eta(t+1)(1-\eta K)^tDeqt+1​=−η(t+1)(1−ηK)tDe for every t∈Nt\in\mathbb Nt∈N. These identities include t=0t=0t=0 and allow L=0L=0L=0, with empty sums equal to zero.

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