§4.2, side-note p. 38 — the Fourier matrix is unitary, Φ*Φ = I
ProvedGeomDL.Grids.fourierMatrix_unitarydiscrete-fourier-transformgeometric-deep-learningp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1unitary-matrix
Let and let be the Fourier matrix with columns . Writing for the conjugate transpose,
So the columns of are orthonormal for the Hermitian inner product and multiplication by (the DFT) is a change of coordinates whose inverse is multiplication by (the inverse DFT).
Formalization Note The side-note calls "an orthogonal matrix"; since is complex, the property it writes, with the conjugate transpose (the book's own convention, side-note p. 38), is unitarity. The Lean statement uses conjTranspose.
Preamble
import Mathlib import Definitions.Def_GeomDL_Grids_Defs open Matrix
Formal statement
namespace GeomDL.Grids
/-- Bronstein et al. §4.2, side-note p. 38: the Fourier matrix satisfies `Φ*Φ = I`, where `Φ*` is
the conjugate transpose (the side-note calls `Φ` orthogonal; over `ℂ` this is unitarity). -/
theorem fourierMatrix_unitary {n : ℕ} [NeZero n] :
(fourierMatrix n)ᴴ * fourierMatrix n = 1 := by sorry
end GeomDL.Grids
Source
Bronstein et al., Geometric Deep Learning (arXiv:2104.13478v2), §4.2, side-note p. 38, "Since the Fourier transform is an orthogonal matrix (Φ∗Φ = I)"
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.