Refined bilinear lattice-cluster bound (Lemma 5.14 of arXiv:2607.10105, eq. (5.58)) #
Paper: L-5.14 — (5.58) — the refined bilinear lattice bound
This file completes the PAPER_MAP node L-5.14: the bilinear cluster bound (5.58)
with the refining factor min (1, (Q/P)^{1/4}), P = |A|·M⁴, Q = |B|·N⁴.
Contents:
single_y_bound— the pointwise (iny) estimate, paper (5.63)/(5.64);bilinear_cluster_S_bound— theQ ≤ P(S-)bound∑∑ ≲ |A|^{1/4} M⁻³ |B|^{3/4} N⁻¹, paper (5.65)–(5.67);bilinear_cluster_bound— the fully refined (5.58), combining theS-bound withbilinear_cluster_bound_weakfromAnderson4D.PermSum.Local.
Quarter powers are Real.rpow (· ^ (1/4 : ℝ)); indicator constraints are
Finset.filter; constants are explicit in private lemmas and existential in the two
public theorems. The dyadic engine of Local.lean is private there, so the small
parametric geometric-sum lemmas are (re)proved here in ratio-generic form.
rpow helpers for quarter powers #
Dyadic bracketing and per-term bounds (threshold-T versions) #
The three cases of the pointwise estimate (5.63) #
Pointwise (in y) bilinear estimate, paper (5.63). For M-separated A
(M ≥ 1), any u, y and any N,
∑_{x ∈ A, |u-x| ≥ M, |x-y| ≥ max(M,N)} |u-x|⁻²|x-y|⁻² ≤ 524288·M⁻⁴·(1 + P^{1/4}/max(|u-y|, M, N)) with P = |A|·M⁴.
Stage 2: summation in y, small-y case (paper (5.66)–(5.67)) #
Stage 2 (paper (5.66)–(5.67)). Summing the pointwise bound (5.63) over the y
with max(|u-y|, M, N) = max(M, N), under Q ≤ P, meets the S-target
|A|^{1/4}·M⁻³·|B|^{3/4}·N⁻¹.
Stage 3: summation in y, case max = |u-y| (paper (5.65)) #
Stage 3 (paper (5.65)). Summing the pointwise bound (5.63) over the y with
max(|u-y|, M, N) = |u-y|, under Q ≤ P, meets the S-target.
Stage 4: assembly of the refined (5.58) #
Bilinear cluster S-bound (the Q ≤ P case of (5.58)). For M-separated A,
N-separated B, M, N ≥ 1, any u, under Q = |B|N⁴ ≤ P = |A|M⁴:
∑∑ 𝟙·𝟙·|u-x|⁻²|x-y|⁻² ≤ C·|A|^{1/4}·M⁻³·|B|^{3/4}·N⁻¹.
Refined bilinear cluster bound, paper (5.58) (Lemma 5.14). For M-separated
A, N-separated B (M, N ≥ 1) and any u,
∑∑ 𝟙_{|u-x| ≥ M}·𝟙_{|x-y| ≥ max(M,N)}·|u-x|⁻²|x-y|⁻² ≤ C·|A|^{1/2}M⁻²·|B|^{1/2}N⁻²·min(1, (Q/P)^{1/4}),
P = |A|M⁴, Q = |B|N⁴ (with Lean's x/0 = 0 convention when A = ∅).