Truncation of an arithmetic function at a real cutoff
DefinitionTaoFivePrimes_VaughanTruncationTao's proof of his variant of Vaughan's identity (Lemma 4.11) begins by splitting both the Möbius function and the von Mangoldt function at a real cutoff:
where is when and otherwise. This file provides those two truncation operators for an arbitrary real-valued arithmetic function.
For a real cutoff and an arithmetic function , truncLe U f is the arithmetic function
and truncGt U f is its complement. Both are again arithmetic functions — they vanish at because does — and the file records the two projection simp lemmas together with the only structural fact needed downstream, namely that the two pieces reconstruct the original:
The cutoff is a real number rather than a natural one, matching the paper: the parameters in Vaughan's identity are optimised over the reals, and no integrality is ever assumed of them.
Also provided are zetaR and moebiusR, the real-valued incarnations of the constant-one function and of , so that all four factors of Vaughan's identity live in the single Dirichlet-convolution ring ArithmeticFunction ℝ, in which the identity is an equation between ring elements.
import Mathlib.NumberTheory.ArithmeticFunction.VonMangoldt
import Mathlib.NumberTheory.ArithmeticFunction.Moebius
/-!
# Truncation of arithmetic functions at a real cutoff
Source: Terence Tao, *Every odd number greater than 1 is the sum of at most five primes*,
https://arxiv.org/abs/1201.6656, Section 4, proof of Lemma 4.11 (the variant of Vaughan's
identity), where the Moebius and von Mangoldt functions are split as
`mu = mu * 1_{<= U} + mu * 1_{> U}` and `Lambda = Lambda * 1_{<= V} + Lambda * 1_{> V}`.
Only the two truncation operators and their structural lemmas live here; the Vaughan
identity itself is a separate theorem.
-/
namespace TaoFivePrimes
/-- `truncLe U f` is the arithmetic function agreeing with `f` on arguments `n` with
`(n : ℝ) ≤ U`, and vanishing elsewhere. Tao writes this as `f 1_{≤ U}`. -/
noncomputable def truncLe (U : ℝ) (f : ArithmeticFunction ℝ) : ArithmeticFunction ℝ :=
⟨fun n => if (n : ℝ) ≤ U then f n else 0, by simp⟩
/-- `truncGt U f` is the arithmetic function agreeing with `f` on arguments `n` with
`U < (n : ℝ)`, and vanishing elsewhere. Tao writes this as `f 1_{> U}`. -/
noncomputable def truncGt (U : ℝ) (f : ArithmeticFunction ℝ) : ArithmeticFunction ℝ :=
⟨fun n => if U < (n : ℝ) then f n else 0, by simp⟩
@[simp] theorem truncLe_apply (U : ℝ) (f : ArithmeticFunction ℝ) (n : ℕ) :
truncLe U f n = if (n : ℝ) ≤ U then f n else 0 := rfl
@[simp] theorem truncGt_apply (U : ℝ) (f : ArithmeticFunction ℝ) (n : ℕ) :
truncGt U f n = if U < (n : ℝ) then f n else 0 := rfl
/-- The two truncations of `f` at `U` add back up to `f`. -/
theorem truncLe_add_truncGt (U : ℝ) (f : ArithmeticFunction ℝ) :
truncLe U f + truncGt U f = f := by
ext n
rw [ArithmeticFunction.add_apply, truncLe_apply, truncGt_apply]
by_cases h : (n : ℝ) ≤ U
· rw [if_pos h, if_neg (not_lt.mpr h), add_zero]
· rw [if_neg h, if_pos (not_le.mp h), zero_add]
/-- The real-valued zeta arithmetic function `ζ(n) = 1` for `n ≥ 1`. -/
noncomputable abbrev zetaR : ArithmeticFunction ℝ :=
((ArithmeticFunction.zeta : ArithmeticFunction ℕ) : ArithmeticFunction ℝ)
/-- The real-valued Moebius arithmetic function. -/
noncomputable abbrev moebiusR : ArithmeticFunction ℝ :=
((ArithmeticFunction.moebius : ArithmeticFunction ℤ) : ArithmeticFunction ℝ)
end TaoFivePrimes