Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 2: orthogonal Lie-algebra factors

Proved
QAlgorithms.Papers.Independent.Advance50.P50.proposition2

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

prove2quantum

An nnn-qubit input ρ\rhoρ is a positive semidefinite, trace-one matrix on the fixed computational basis, of dimension d=2nd=2^nd=2n. A kkk-copy model has a fixed input-independent unitary UUU, a Hermitian observable OOO, effective observable O~=U†OU\widetilde O=U^\dagger O UO=U†OU, and exact response h(ρ)=Tr⁡[Uρ⊗kU†O]h(\rho)=\operatorname{Tr}[U\rho^{\otimes k}U^\dagger O]h(ρ)=Tr[Uρ⊗kU†O]. Here k≥1k\ge1k≥1. The declared domain RRR consists of density matrices, and its symmetry groups preserve RRR and its binary labels. Invariance means equality of exact expectations on every ρ∈R\rho\in Rρ∈R under the specified conjugation action. Let GGG be a represented unitary Lie group, g\mathfrak gg its actual real tangent Lie algebra, and iI∈giI\in\mathfrak giI∈g. Its Hilbert–Schmidt complement g⊥\mathfrak g^\perpg⊥ is taken inside the skew-Hermitian matrices. Suppose R⊆igR\subseteq i\mathfrak gR⊆ig. For every copy jjj, fix a Hermitian Aj∈ig⊥A_j\in i\mathfrak g^\perpAj​∈ig⊥ and an arbitrary operator BjB_jBj​ on the other copies. Insert AjA_jAj​ in copy jjj, preserving all other copy indices. If O~\widetilde OO belongs to the complex span of these specified separated terms, then

h(ρ)=h(VρV†)=0(ρ∈R,V∈G).h(\rho)=h(V\rho V^\dagger)=0\quad(\rho\in R,V\in G).h(ρ)=h(VρV†)=0(ρ∈R,V∈G).

Thus the model is invariant by orthogonality, independently of commutant membership.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Advance50_P50
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Advance50.P50

open MeasureTheory ProbabilityTheory

open scoped ComplexConjugate ENNReal

-- Exact accepted primitives; no paper-local redeclarations.

open QAlgorithms.Papers.Independent.Information100.P63 (Mat tensorPower)

open QAlgorithms.Papers.Independent.Gap48.P20 (Copies)

open QAlgorithms.Papers.Independent.Gap48.P38 (subsystemReduction)

open QAlgorithms.Papers.Independent.Information100.P32 (purity)

open QAlgorithms.Papers.Independent.Information100.P08 (IsPure)

-- Matrices carry the associative commutator bracket X*Y-Y*X.

attribute [local instance 100] LieRing.ofAssociativeRing

/-- PDF p.5, Proposition 2, Appendix B pp.23–24: factors orthogonal to the
actual Lie algebra force zero predictions, hence invariance.
Printed: PDF p.5 describes u(d) as skew-symmetric and writes exp(g)=G and g={X:exp(X) in G}.
Changed: Use the actual real skew-Hermitian tangent Lie algebra of the represented unitary group; keep iI in g, the stated orthogonal spaces, separated factors and the zero-output conclusion from Appendix B.
Why: The source means the unitary Lie algebra. A reflection is in O(d) but not exp(o(d)); the printed exponential rule cannot define the Lie algebra for the disconnected groups the paper uses. Tangents retain the adjoint-invariance argument without an exponential-surjectivity premise. -/
theorem proposition2 {n k : ℕ} (hk : 0 < k) (M : ExpectationModel (Qubits n) k)
    (L : MatrixLieData (Qubits n))
    (hidentity : Complex.I • (1 : Mat (Qubits n)) ∈ L.algebra)
    (R : Set (Mat (Qubits n))) (label : Mat (Qubits n) → Bool)
    (hdata : DatasetSymmetry (groupMatrices L.group) R label)
    (hinput : R ⊆ timesI (L.algebra : Set (Mat (Qubits n))))
    (A : Fin k → Mat (Qubits n))
    (B : (j : Fin k) → Mat ({c : Fin k // c ≠ j} → Qubits n))
    (hA : ∀ j, A j ∈ timesI (lieComplement L) ∧ (A j).IsHermitian)
    (hobs : M.effective ∈ separatedSpan A B) :
    DomainInvariant (groupMatrices L.group) R M.response ∧
      ∀ rho ∈ R, M.response rho = 0 := by sorry

end QAlgorithms.Papers.Independent.Advance50.P50

end
Source
Group-Invariant Quantum Machine Learning, arXiv:2205.02261v2, PDF p. 5, Proposition 2: orthogonal Lie-algebra factors
Read-back

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

Let n,kn,kn,k be natural numbers with k>0k>0k>0, and let H\mathcal HH be the complex Euclidean space indexed by nnn-bit strings. An expectation model supplies a unitary UUU and a Hermitian matrix OOO on H⊗k\mathcal H^{\otimes k}H⊗k; put Heff=U†OUH_{\mathrm{eff}}=U^\dagger O UHeff​=U†OU and h(ρ)=Re⁡Tr⁡(Uρ⊗kU†O)h(\rho)=\operatorname{Re}\operatorname{Tr}(U\rho^{\otimes k}U^\dagger O)h(ρ)=ReTr(Uρ⊗kU†O), with tensor-power entries ∏cρxc,yc\prod_c\rho_{x_c,y_c}∏c​ρxc​,yc​​. Let LLL consist of a unitary subgroup GGG and a real Lie subalgebra g\mathfrak gg of matrices on H\mathcal HH, all of whose elements satisfy X†=−XX^\dagger=-XX†=−X, such that g\mathfrak gg is exactly the set of derivatives c′(0)c'(0)c′(0) of curves c:R→Mat⁡(H)c:\mathbb R\to\operatorname{Mat}(\mathcal H)c:R→Mat(H) satisfying c(0)=Ic(0)=Ic(0)=I and c(t)∈Gc(t)\in Gc(t)∈G for every real ttt. Define

gskew⊥={X:X†=−X, Tr⁡(X†Y)=0 for every Y∈g},iS={iX:X∈S}.\mathfrak g^\perp_{\mathrm{skew}} =\{X:X^\dagger=-X,\ \operatorname{Tr}(X^\dagger Y)=0\text{ for every }Y\in\mathfrak g\}, \qquad iS=\{iX:X\in S\}.gskew⊥​={X:X†=−X, Tr(X†Y)=0 for every Y∈g},iS={iX:X∈S}.

Assume iI∈giI\in\mathfrak giI∈g. Let RRR be a set of matrices and ℓ\ellℓ a Boolean-valued function such that every ρ∈R\rho\in Rρ∈R is positive semidefinite with trace 111, and for every V∈GV\in GV∈G and ρ∈R\rho\in Rρ∈R, VρV†∈RV\rho V^\dagger\in RVρV†∈R and ℓ(VρV†)=ℓ(ρ)\ell(V\rho V^\dagger)=\ell(\rho)ℓ(VρV†)=ℓ(ρ). Assume R⊆igR\subseteq i\mathfrak gR⊆ig. For each j<kj<kj<k, let AjA_jAj​ be a Hermitian matrix in igskew⊥i\mathfrak g^\perp_{\mathrm{skew}}igskew⊥​, and let BjB_jBj​ be any complex matrix indexed by assignments of nnn-bit strings to the copy positions other than jjj. Assume HeffH_{\mathrm{eff}}Heff​ lies in the complex span of the kkk matrices whose entries are

(Sj)x,y=(Aj)xj,yj(Bj)x∣{c:c≠j}, y∣{c:c≠j}.(S_j)_{x,y}=(A_j)_{x_j,y_j} (B_j)_{x|_{\{c:c\ne j\}},\,y|_{\{c:c\ne j\}}}.(Sj​)x,y​=(Aj​)xj​,yj​​(Bj​)x∣{c:c=j}​,y∣{c:c=j}​​.

Then h(VρV†)=h(ρ)h(V\rho V^\dagger)=h(\rho)h(VρV†)=h(ρ) for every V∈GV\in GV∈G and ρ∈R\rho\in Rρ∈R, and h(ρ)=0h(\rho)=0h(ρ)=0 for every ρ∈R\rho\in Rρ∈R. Here n=0n=0n=0 is allowed, the BjB_jBj​ have no Hermiticity requirement, and RRR may be empty, making the conclusions vacuous.

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