Documentation

LeanPool.PhaseRetrieval.Constant.Internal.HighFreqBandEstimate

HighFreqBandEstimate #

Slow factor and factorization #

Pointwise bounds #

Interval partition quantities #

Infrastructure: Haar-to-interval decomposition #

The key identity relating Haar integrals on AddCircle T to interval integrals on ℝ is:

∫ f ∂haarAddCircle = (1/T) ∑_{k=0}^{N-1} ∫_{I_k} f(↑t) dt

This follows from:

Intermediate lemmas (4.1a–d) #

Parseval identity for finite Fourier sums indexed by Fin L #

Auxiliary #

Core lower bound #

Combines assembly + Poincaré to get: ∫ ρ(P)² d(haar) ≥ circleNormSq(Q) / 32.

The arithmetic reduces to: 136π²(L−1)² ≤ N² (under 1343L² ≤ N²).

Since 136π² ≈ 1342.27, we need ⌈136π²⌉ = 1343. The hypothesis 1343 * L^2 ≤ N^2 implies 136π²(L-1)² ≤ 136π²L² ≤ 1343L² ≤ N².

Main theorem #

theorem FockSPR.high_freq_band_estimate {N L : } (hN : 1 N) (hL : 1 L) (hNL : 1343 * L ^ 2 N ^ 2) (b : Fin L) (P : AddCircle T) (hP : P = fun (t : AddCircle T) => m : Fin L, b m * (fourier ↑(N + m)) t) :

Theorem 4.1: High-frequency band estimate. For N ≥ 1, L ≥ 1, 1343·L² ≤ N², and P(t) = ∑ b(m)·fourier(N+m)(t): circleNormSq(P) ≤ 32 · ∫ ρ(P)² d(haar).