Analytic consequences of regular variation of the normalizing sequence #
Consequences of regular variation of (a_n) used in the source (lines 1015–1050, 1080–1110,
1136–1251, 1747–1790, 2233–2298), for a monotone a : ℕ → ℝ, positive at n ≥ 1, with
RegularlyVaryingSeq ρ a:
tendsto_ratio_of_tendsto_div: ifk n / n → c > 0thena (k n) / a n → c ^ ρ(coversa_{⌊εn⌋}/a_n → ε^ρ,a_{n-o(n)}/a_n → 1,a_{n-1}/a_n → 1).eventually_rpow_le_le:a_n = n^{ρ+o(1)}.tendsto_div_self,tendsto_atTop,exists_le_mul,exists_le_mul_of_le_mul:a_n = o(n),a_n → ∞,a_n ≤ C n, anda_y ≤ C a_xwhenevery ≤ K x.tendsto_ratio_zero_of_isLittleO:M_n = o(n)givesa_{M_n} / a_n → 0.tendsto_sub_ratio,tendsto_add_ratio,eventually_uniform_sub_add:a_{n ∓ m_n} / a_n → 1form_n = o(n), uniformly overm ≤ M_n.tendsto_sqrt_div: forr_n = √(a_{M_n} a_n),r_n / a_{M_n} → ∞andr_n / a_n → 0.eventually_floor_rpow_bounds,eventually_floor_rpow_div_le,eventually_div_floor_rpow_le: polynomial cutoffsM_n = ⌊n^θ⌋₊.tendsto_mul_div_floor_rpow,tendsto_div_floor_rpow: the two limits in the proof of Lemmalem:critM(source lines 1087–1105), for ann^{o(1)}numerator.
Adapted from PoissonDeclumping/RegularVariation/Consequences.lean in
igor-kortchemski/rare-subtree-patterns, commit
fc5ef722392b1af41e8b9d5fc45d5a521b7823dd (Apache-2.0).
If 0 ≤ f n ≤ C n ^ γ eventually along l ≤ atTop, with γ < 0, then f → 0 along l.
If k n / n → c > 0, then a (k n) / a n → c ^ ρ.
For fixed ε > 0, a ⌊ε n⌋₊ / a n → ε ^ ρ (the definition, restated).
a n → ∞ (positive index).
a_n = o(n) (index ρ < 1): a n / n → 0.
Comparability: for every K, there is C > 0 with a y ≤ C a x whenever 1 ≤ x and
y ≤ K x (e.g. a_q ≍ a_n for n/2 ≤ q ≤ n).
If M_n = o(n) then a (M n) / a n → 0 (positive index).
a_{n - m_n} / a_n → 1 for m_n = o(n) (natural subtraction).
a_{n + m_n} / a_n → 1 for m_n = o(n).
Uniform version: for M_n = o(n) and ε > 0, eventually for all j ≤ M n,
(1-ε) a n ≤ a (n-j) ≤ a n ≤ a (n+j) ≤ (1+ε) a n.
Source lines 1747–1750: with r_n = √(a_{M_n} a_n), M_n → ∞ and M_n = o(n),
r_n / a_{M_n} → ∞ and r_n / a_n → 0.
Polynomial cutoffs: for θ > 0 and δ > 0, eventually
n^(θρ - δ) ≤ a ⌊n^θ⌋₊ ≤ n^(θρ + δ), i.e. a_{⌊n^θ⌋} = n^{θρ + o(1)}.
Polynomial cutoff M_n = ⌊n^(1-η)⌋₊, η < 1: for every δ > 0, eventually
a_{M_n} / a_n ≤ n^(-ρη + δ).
Polynomial cutoff M_n = ⌊n^θ⌋₊, θ > 0: for every δ > 0, eventually
a_n / M_n ≤ n^(ρ - θ + δ).
Lemma lem:critM (source lines 1090–1097): if η < 1 - ρ, ε > 0, and e_n ≥ 0 is
n^{o(1)} along a filter l ≤ atTop, then a_n e_n / (ε ⌊n^(1-η)⌋₊) → 0 along l.
Lemma lem:critM (source lines 1099–1105): if η < 1 and e_n ≥ 0 is n^{o(1)} along
a filter l ≤ atTop, then e_n / a_{⌊n^(1-η)⌋} → 0 along l.