The Vaughan Type I coefficient is bounded by
ProvedTaoFivePrimes.vaughan_typeI_coeff_le_logarithmetic-functionsdirichlet-convolutionnumber-theoryvaughan-identityvon-mangoldt
In Vaughan's identity the second term is , and summing it against a test function produces the divisor coefficient
For every pair of real cutoffs and every natural number ,
Here is the Möbius function, the von Mangoldt function, and the truncation of to arguments at most .
This pointwise bound is what makes the second term of Vaughan's identity a harmless Type I contribution: its coefficients are of divisor size, no larger than uniformly in the cutoffs, so the term can be handled by the same estimates as the first one.
Formalization Note At and both sides vanish, so no positivity hypothesis on is required.
Preamble
import Mathlib import Definitions.Def_TaoFivePrimes_VaughanTruncation
Formal statement
theorem TaoFivePrimes.vaughan_typeI_coeff_le_log (U V : ℝ) (d : ℕ) :
|(TaoFivePrimes.truncLe U TaoFivePrimes.moebiusR *
TaoFivePrimes.truncLe V ArithmeticFunction.vonMangoldt) d| ≤ Real.log d := by sorrySource
Terence Tao, Every odd number greater than 1 is the sum of at most five primes, Mathematics of Computation 83 (2014), 997-1038, https://arxiv.org/abs/1201.6656, Section 4, proof of Lemma 4.11: "where f(d) := sum_{b|d: d/U <= b <= V} mu(d/b) Lambda(b)" and "Note that |f(d)| <= sum_{b|d} Lambda(b) = log d."