Fourier series converge in the mean square
ProvedFamousTheorems.hasSum_fourier_series_L2Convergence of Fourier series in (the Riesz–Fischer theorem).
For every square-integrable function on the circle , the Fourier series
converges to in the norm.
The exponentials form a complete orthonormal basis of of the
circle, so this is the statement that the Fourier expansion loses nothing. Convergence is
unconditional — HasSum quantifies over finite partial sums with no ordering — which is what makes
the natural home for Fourier analysis: pointwise convergence is delicate (du Bois-Reymond
constructed a continuous function whose Fourier series diverges at a point, and Carleson's theorem
that it converges almost everywhere took until 1966), while mean-square convergence is clean and
holds for every .
Fourier asserted in 1807 that every function admits such an expansion, which was not believed; Dirichlet gave the first convergence proof under restrictive hypotheses in 1829. The definitive statement is the Riesz–Fischer theorem of 1907, and with it Parseval's identity .
Formalization note. haarAddCircle is the normalized Haar probability measure on the circle, and
fourierLp 2 i is the -th exponential viewed as an element of . The result is Mathlib's
MeasureTheory.hasSum_fourier_series_L2.
import Mathlib
namespace FamousTheorems
open MeasureTheory ProbabilityTheory Filter Set intervalIntegral
open scoped Real Topology ENNReal
theorem hasSum_fourier_series_L2 {T : ℝ} [hT : Fact (0 < T)]
(f : Lp ℂ 2 (@AddCircle.haarAddCircle T hT)) :
HasSum (fun i => fourierCoeff (f : AddCircle T → ℂ) i • fourierLp 2 i) f := by sorry
end FamousTheorems