Scalar facts: character orthogonality for a primitive root of unity #
Reindexing a sum over range N by k ↦ k + 1 for an N-periodic summand.
Character orthogonality: ∑_{k<N} ζ^{dk} = N if N ∣ d and 0 otherwise.
Discrete Fourier coefficients in a star-algebra over ℂ #
The j-th discrete Fourier coefficient N⁻¹ ∑_{l<N} ζ^{-jl} d l of a family
d : ℕ → A (only the values d 0, …, d (N-1) matter).
Equations
- Orthogonalization.fourierCoeff ζ N d j = (↑N)⁻¹ • ∑ l ∈ Finset.range N, ζ ^ (-(j * ↑l)) • d l
Instances For
Two Fourier transforms in independent variables commute.
The orthogonality kernel of the Fourier transform: transforming twice, in the
variables l and l', with opposite frequencies and summing over the frequency
recovers the diagonal.
The adjoint of a Fourier coefficient is the (-j)-th coefficient of the adjoints.
Parseval's identity in the algebra: for the Fourier coefficients
Y j = N⁻¹ ∑_{l<N} ζ^{-jl} d l, ∑_j (Y j)* (Y j) = N⁻¹ ∑_{l<N} (d l)* (d l).
The dictionary: spectral projections of a unitary of finite order #
The Fourier coefficient of an eigenfamily: if x l * p = ζ ^ (e l) • p for every l,
then (N⁻¹ ∑_{l<N} ζ^{-jl} x l) * p is p when j ≡ e (mod N) and 0 otherwise.
The spectral projections of a unitary w of order N (the paper's
p_j = n⁻¹ ∑_k e^{-2ijkπ/n} u^k, line 149): P i = N⁻¹ ∑_{k<N} ζ^{-ik} w^k.
Equations
- Orthogonalization.spectralProj ζ N w i = Orthogonalization.fourierCoeff ζ N (fun (k : ℕ) => w ^ k) ↑↑i
Instances For
The spectral projections sum to 1.
The spectral projections are (star) projections.
The inverse dictionary: the unitary of an orthogonal family of projections #
The unitary ∑_j ζ^j r j of a family r of projections (the paper's
u = ∑_k e^{2ikπ/n} p_k, line 149).
Equations
- Orthogonalization.fourierUnitary ζ r = ∑ j : Fin N, ζ ^ ↑↑j • r j
Instances For
The powers of the reconstructed unitary: (∑_j ζ^j r j)^l = ∑_j ζ^{jl} r j.
The reconstructed unitary has order dividing N.
The spectral projections of the reconstructed unitary are the original projections.
Operators on H: PVMs, membership in M, and Parseval for ‖·‖_φ #
The projections of a PVM are pairwise orthogonal: q j' * q j = δ_{j' j} q j.
(From q j = q j (∑ k, q k) q j = q j + ∑_{k ≠ j} (q k q j)* (q k q j) and positivity.)
The spectral projections of a unitary w ∈ M of order N form a PVM in M.
‖-x‖_φ² = ‖x‖_φ².
Parseval for the seminorm ‖·‖_φ: ∑_j ‖Y j‖_φ² = N⁻¹ ∑_{l<N} ‖d l‖_φ² for the
Fourier coefficients Y j = N⁻¹ ∑_{l<N} ζ^{-jl} d l (only linearity of φ is used).
The hypothesis of Corollary 1.5 is the hypothesis of Theorem 1.4 for the spectral
PVMs: ∑_{i,j} ‖p_i q_j − q_j p_i‖_φ² = (nm)⁻¹ ∑_{k<n} ∑_{l<m} ‖u^{k+1} v^{l+1} − v^{l+1} u^{k+1}‖_φ².
Corollary 1.5 in finite dimension #
Corollary 1.5 (cor:almost_commuting) in finite dimension.