Spectral decomposition of power iterates
ProvedPowerMethod.iterate_eq_sumFor a self-adjoint operator T on a finite-dimensional inner product space E with eigenbasis (e_i) and eigenvalues (lambda_i), expanding x0 = sum c_i e_i in the eigenbasis gives T^k x0 = sum (c_i lambda_i^k) e_i for every k.
import Mathlib
namespace PowerMethod
/-- **Spectral decomposition of iterates.** For a self-adjoint operator `T` on a finite-dimensional
inner product space `E`, expanding a vector `x0` in the orthonormal eigenbasis of `T` shows that
the `k`-th iterate `T^k x0` is obtained by raising each eigenvalue coefficient to the `k`-th
power. -/
theorem iterate_eq_sum
{𝕜 : Type*} [RCLike 𝕜] {E : Type*} [NormedAddCommGroup E] [InnerProductSpace 𝕜 E]
[FiniteDimensional 𝕜 E] {T : E →ₗ[𝕜] E} (hT : T.IsSymmetric) {n : ℕ}
(hn : Module.finrank 𝕜 E = n) (x0 : E) (k : ℕ) :
(T ^ k) x0 = ∑ i, ((hT.eigenvectorBasis hn).repr x0 i * (hT.eigenvalues hn i : 𝕜) ^ k) •
hT.eigenvectorBasis hn i := by
sorry
end PowerMethod
Read-back
What the Lean code literally says, in plain math · claude-sonnet-5
This declaration states, for a scalar field that is either the real or complex numbers, a finite-dimensional inner-product space over , a natural number equal to the dimension of , and a linear operator that is self-adjoint (i.e. for all ) — all of these being fixed background data (implicit types/instances plus the explicit hypothesis that is self-adjoint and the explicit hypothesis that ) — that there is an associated orthonormal basis of consisting of eigenvectors of , with corresponding real eigenvalues satisfying for each index (with understood as embedded into where needed), and that for every vector and every natural number (so ranges over , with no positivity assumption), writing for the unique coordinates of in this basis, the -fold self-composition of applied to equals , i.e. . Both and are universally quantified, so the claim covers the degenerate case , where is the identity map and the asserted equation reduces (since ) to the tautological basis expansion ; it also covers for some or all , in which case those terms vanish for ; and it covers , where is the trivial space, is forced to be , and both sides of the equation are the empty sum, . The statement itself asserts nothing about the eigenvalues being sorted, distinct, or nonzero, nor about or being nonzero. Finally, the proof of this theorem is literally the placeholder sorry, meaning no actual proof has been supplied — the equality is asserted in the code but not established by it.
Confirmed by the mission captain (proposal self-audit).