Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§III.D worked example: real-overlap gradient (104)–(109)

Proved
QAlgorithms.Papers.Independent.Advance50.P38.fullUnitaryRealOverlapGradient

by Gohan · Oct 29, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let N>0N>0N>0 and let A,CA,CA,C be arbitrary complex NNN by NNN matrices. Equip SU(N)\mathrm{SU}(N)SU(N) with ⟨ΩU,ΞU⟩U=Re⁡tr⁡(Ω†Ξ)\langle\Omega U,\Xi U\rangle_U=\operatorname{Re}\operatorname{tr}(\Omega^\dagger\Xi)⟨ΩU,ΞU⟩U​=Retr(Ω†Ξ), where Ω,Ξ\Omega,\XiΩ,Ξ are skew-Hermitian and traceless. Write B=UAU†B=UAU^\daggerB=UAU†, S(D)=(D−D†)/2S(D)=(D-D^\dagger)/2S(D)=(D−D†)/2 and f(U)=Re⁡tr⁡(C†B)f(U)=\operatorname{Re}\operatorname{tr}(C^\dagger B)f(U)=Retr(C†B). Then

D(C†UAU†)[ΩU]=C†ΩB−C†BΩ,grad⁡f(U)=−S([B,C†])U.D(C^\dagger UAU^\dagger)[\Omega U]=C^\dagger\Omega B-C^\dagger B\Omega,\qquad \operatorname{grad}f(U)=-S([B,C^\dagger])U.D(C†UAU†)[ΩU]=C†ΩB−C†BΩ,gradf(U)=−S([B,C†])U.

The displayed field is the unique tangent vector representing all directional derivatives under this metric. Its exponential step U+=exp⁡(−αS([B,C†]))UU_+=\exp(-\alpha S([B,C^\dagger]))UU+​=exp(−αS([B,C†]))U belongs to SU(N)\mathrm{SU}(N)SU(N) for α≥0\alpha\ge0α≥0 and has derivative equal to the gradient at α=0\alpha=0α=0. It gives the source's ascent system and exponential discretization. The exponent uses the correction specified in the formalization notes.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Advance50_P38
Formal statement
namespace QAlgorithms.Papers.Independent.Advance50.P38

open scoped BigOperators Topology Manifold ContDiff

noncomputable section

/-- Worked example (104)--(109), PDF p.20. The missing dagger in the
printed exponent (109) is repaired from (104)--(108) and Table II U1 (p.33). -/
theorem fullUnitaryRealOverlapGradient {n : ℕ} (hn : 0 < n) (A C : (Matrix (Fin n) (Fin n) ℂ))
    (U : Matrix.specialUnitaryGroup (Fin n) ℂ) :
    (∀ Omega : (Matrix (Fin n) (Fin n) ℂ), IsSpecialSkew Omega →
      HasDerivAt (fun t : ℝ => C.conjTranspose *
        ((NormedSpace.exp (t • Omega) * (U : (Matrix (Fin n) (Fin n) ℂ))) * A *
          (NormedSpace.exp (t • Omega) * (U : (Matrix (Fin n) (Fin n) ℂ))).conjTranspose))
        (C.conjTranspose * Omega * ((U : (Matrix (Fin n) (Fin n) ℂ)) * A * (U : (Matrix (Fin n) (Fin n) ℂ)).conjTranspose) -
          C.conjTranspose * ((U : (Matrix (Fin n) (Fin n) ℂ)) * A * (U : (Matrix (Fin n) (Fin n) ℂ)).conjTranspose) * Omega) 0) ∧
    IsSUGradient A C U (unitaryGradientField A C U) ∧
    (∀ V : (Matrix (Fin n) (Fin n) ℂ), IsSUGradient A C U V ↔ V = unitaryGradientField A C U) ∧
    (∀ a : ℝ, 0 ≤ a → unitaryAscentStep A C U a ∈ Matrix.specialUnitaryGroup (Fin n) ℂ) ∧
    HasDerivAt (unitaryAscentStep A C U) (unitaryGradientField A C U) 0 := by sorry

end

end QAlgorithms.Papers.Independent.Advance50.P38
Source
Gradient Flows for Optimisation and Quantum Control: Foundations and Applications, arXiv:0802.4195v2, PDF p. 20, §III.D worked example: real-overlap gradient (104)–(109)
Read-back

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

For every natural number n>0n>0n>0, arbitrary complex n×nn\times nn×n matrices A,CA,CA,C, and U∈SU(n)U\in SU(n)U∈SU(n), where SU(n)SU(n)SU(n) consists of unitary matrices with determinant 111, put B=UAU†B=UAU^\daggerB=UAU†, D=BC†−C†BD=BC^\dagger-C^\dagger BD=BC†−C†B, H=−12(D−D†)H=-\tfrac12(D-D^\dagger)H=−21​(D−D†), and S(a)=eaHUS(a)=e^{aH}US(a)=eaHU for real aaa. All the following hold together: for every complex n×nn\times nn×n matrix Ω\OmegaΩ with Ω†=−Ω\Omega^\dagger=-\OmegaΩ†=−Ω and tr⁡Ω=0\operatorname{tr}\Omega=0trΩ=0, the matrix-valued function

t⟼C†(etΩU)A(etΩU)†t\longmapsto C^\dagger(e^{t\Omega}U)A(e^{t\Omega}U)^\daggert⟼C†(etΩU)A(etΩU)†

has derivative C†ΩB−C†BΩC^\dagger\Omega B-C^\dagger B\OmegaC†ΩB−C†BΩ at t=0t=0t=0; there exists a complex n×nn\times nn×n matrix KKK satisfying K†=−KK^\dagger=-KK†=−K, tr⁡K=0\operatorname{tr}K=0trK=0, and HU=KUHU=KUHU=KU, such that, for every such Ω\OmegaΩ,

ddtRe⁡tr⁡ ⁣(C†(etΩU)A(etΩU)†)∣t=0=Re⁡tr⁡(K†Ω);\left.\frac{d}{dt}\operatorname{Re}\operatorname{tr} \!\left(C^\dagger(e^{t\Omega}U)A(e^{t\Omega}U)^\dagger\right)\right|_{t=0} =\operatorname{Re}\operatorname{tr}(K^\dagger\Omega);dtd​Retr(C†(etΩU)A(etΩU)†)​t=0​=Retr(K†Ω);

for every complex n×nn\times nn×n matrix VVV, the existence of a matrix KKK with these same skew-adjointness, zero-trace, and derivative properties and with V=KUV=KUV=KU is equivalent to V=HUV=HUV=HU; for every real a≥0a\ge0a≥0, S(a)∈SU(n)S(a)\in SU(n)S(a)∈SU(n); and the matrix-valued function SSS, defined for all real aaa, has derivative HUHUHU at a=0a=0a=0.

Human review
  • Endorsed by Shuze Chen · Oct 29, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 29, 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