§4.2, side-note p. 37 — the eigenvalues e^{−2πik/n}, k = 0, …, n−1, of the shift S are distinct
ProvedGeomDL.Grids.shift_eigenvalues_distincteigenvaluegeometric-deep-learningp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1roots-of-unityshift-operator
Let . The eigenvalues of the right shift on the Fourier basis vectors , , are pairwise distinct: for ,
The book needs distinct eigenvalues so that the common eigenbasis of the commuting circulant matrices is determined by the shift alone; it states that the assumption "is satisfied with our choice of ".
Formalization Note and are elements of ZMod n, entering the exponent through ZMod.val. Together with the eigenvector statement for this says that has distinct eigenvalues.
Preamble
import Mathlib import Definitions.Def_GeomDL_Grids_Defs open Matrix
Formal statement
namespace GeomDL.Grids
/-- Bronstein et al. §4.2, side-note p. 37: the shift `S` has distinct eigenvalues. The eigenvalues
`λ_k = e^{-2πik/n}`, `k = 0, …, n-1`, of `S` on the Fourier basis are pairwise distinct. -/
theorem shift_eigenvalues_distinct {n : ℕ} [NeZero n] (k l : ZMod n)
(hkl : Complex.exp (-(2 * (Real.pi : ℂ) * Complex.I * (k.val : ℂ) / (n : ℂ))) =
Complex.exp (-(2 * (Real.pi : ℂ) * Complex.I * (l.val : ℂ) / (n : ℂ)))) :
k = l := by sorry
end GeomDL.Grids
Source
Bronstein et al., Geometric Deep Learning (arXiv:2104.13478v2), §4.2, side-note p. 37, "We must additionally assume distinct eigenvalues … This assumption is satisfied with our choice of S."
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.