and the second factor J(2) is a quotient. Every subsequent estimate — comparing eiθz/(1+iθz) with e−θ2z2/2 term by term, and then aggregating over k — would otherwise have to be carried out with division present, where the usual product and triangle inequalities do not apply directly.
This lemma removes the denominator once and for all. Applied with z=1+iθZk, whose modulus is 1+θ2Zk2≥1, it converts each quotient estimate
1+iθZkeiθZk−winto the division-freeeiθZk−(1+iθZk)w,
which is then handled by the Taylor expansions of eiθz and of the comparison factor w. The hypothesis ∥z∥≥1 is exactly what the factors 1+iθZk supply, and it is essential: for ∥z∥<1 the inequality reverses.
theorem Martingale.norm_div_sub_le_norm_sub_mul (z u w : ℂ) (hz : 1 ≤ ‖z‖) :
‖u / z - w‖ ≤ ‖u - z * w‖ := by sorry
Source
B. M. Brown, "Martingale Central Limit Theorems", Annals of Mathematical Statistics 42 (1971) 59-66, Theorem 2; D. L. McLeish, "Dependent Central Limit Theorems and Invariance Principles", Annals of Probability 2 (1974) 620-628, Theorem 2.3; P. Hall and C. C. Heyde, Martingale Limit Theory and Its Application, Academic Press 1980, Theorem 3.2.