Finite scale transfer for exponential sums #
This module gives two ways to rewrite an exponential sum at a different length scale. Finite
Fourier inversion dilates a sum from scale N to q * N, while decomposition into residue
classes produces sums at scales approximately N / q. These identities are used to compare
the exponential-sum growth exponent at different values of α.
Finite Fourier orthogonality #
Reindexing exponential sums #
theorem
Expdb.norm_exponentialSumAt_dilate_le
(F : ℝ → ℝ)
(T N : ℝ)
(a b q : ℕ)
[NeZero q]
(hq : 0 < q)
(hT : T ≠ 0)
(hN : N ≠ 0)
:
Triangle-inequality form of exponentialSumAt_dilate.
The scale induced by restricting a sum to the residue class r mod q.
Equations
- Expdb.residueScale N q r = (N - ↑r) / ↑q
Instances For
theorem
Expdb.exponentialSumAt_residue
(F : ℝ → ℝ)
(T N : ℝ)
(a b q r : ℕ)
(hq : 0 < q)
(hr : r < q)
(hra : r ≤ a)
(hab : a ≤ b)
(hN : N ≠ 0)
(hrN : ↑r < N)
:
∑ n ∈ Finset.Icc a b with n % q = r, oscillatory F T N ↑n = exponentialSumAt (residuePhase F N r) T (residueScale N q r) ((a - r) ⌈/⌉ q) ((b - r) / q)
Reindex one residue class as an exponential sum with residuePhase and residueScale.
theorem
Expdb.norm_exponentialSumAt_le_sum_residues
(F : ℝ → ℝ)
(T N : ℝ)
(a b q : ℕ)
(hq : 0 < q)
(hqN : ↑q ≤ N)
(hN : N ≠ 0)
(ha : N ≤ ↑a)
(hb : ↑b ≤ 2 * N)
:
‖exponentialSumAt F T N a b‖ ≤ ∑ r ∈ Finset.range q,
(‖exponentialSumAt (residuePhase F N r) T (residueScale N q r) ((a - r) ⌈/⌉ q)
(min ((b - r) / q) ⌊2 * residueScale N q r⌋₊)‖ + 1)
Split an exponential sum into residue classes, trimming each upper endpoint at a cost of at most one term.