In mathematics and analytic number theory, Vaughan's identity is an identity found by that can be used to simplify Vinogradov's work on trigonometric sums. It can be used to estimate summatory functions of the form
\sumnf(n)Λ(n)
The motivation for Vaughan's construction of his identity is briefly discussed at the beginning of Chapter 24 in Davenport. For now, we will skip over most of the technical details motivating the identity and its usage in applications, and instead focus on the setup of its construction by parts. Following from the reference, we construct four distinct sums based on the expansion of the logarithmic derivative of the Riemann zeta function in terms of functions which are partial Dirichlet series respectively truncated at the upper bounds of
U
V
F(s)=\summΛ(m)m-s
G(s)=\sumd\mu(d)d-s
- | \zeta\prime(s) |
\zeta(s) |
=F(s)-\zeta(s)F(s)G(s)-\zeta\prime(s)G(s)+\left(-
\zeta\prime(s) | |
\zeta(s) |
-F(s)\right)(1-\zeta(s)G(s)).
This last expansion implies that we can write
Λ(n)=a1(n)+a2(n)+a3(n)+a4(n),
where the component functions are defined to be
\begin{align}a1(n)&:=l\{\begin{matrix}Λ(n),&ifn\leqU;\ 0,&ifn>U\end{matrix}\ a2(n)&:=-\sum\stackrel{mdr{\stackrel{m\leqU}{d\leqV}}}Λ(m)\mu(d)\ a3(n)&:=\sum\stackrel{hd=n{d\leqV}}\mu(d)log(h)\ a4(n)&:=-\sum\stackrel{mk=n{\stackrel{m>U}{k>1}}}Λ(m)\left(\sum\stackrel{d|k{d\leqV}}\mu(d)\right).\end{align}
We then define the corresponding summatory functions for
1\leqi\leq4
Si(N):=\sumnf(n)ai(n),
so that we can write
\sumnf(n)Λ(n)=S1(N)+S2(N)+S3(N)+S4(N).
Finally, at the conclusion of a multi-page argument of technical and at times delicate estimations of these sums,[1] we obtain the following form of Vaughan's identity when we assume that
|f(n)|\leq1, \foralln
U,V\geq2
UV\leqN
\sumnf(n)Λ(n)\llU+(logN) x \sumt\leq\left(maxw
\left|\sum | ||||||
|
f(rt)\right|\right)+\sqrt{N}(logN)3 x maxUmaxV\left(\sumV\left|\sum\stackrel{M{\stackrel{m\leqN/k}{m\leqN/j}}}f(mj)\bar{f(mk)}\right|\right)1/2(V1).
It is remarked that in some instances sharper estimates can be obtained from Vaughan's identity by treating the component sum
S2
S2=\sumt\longmapsto\sumt+\sumU=:
\prime | |
S | |
2 |
+
\prime\prime | |
S | |
2 |
.
The optimality of the upper bound obtained by applying Vaughan's identity appears to be application-dependent with respect to the best functions
U=fU(N)
V=fV(N)
S(\alpha):=\sumnΛ(n)e\left(n\alpha\right).
\alpha\inR\setminusQ
\left|\alpha-
a | |
q |
\right|\leq
1 | |
q2 |
,(a,q)=1,
of the form
S(\alpha)\ll\left(
N | |
\sqrt{q |
The argument for this estimate follows from Vaughan's identity by proving by a somewhat intricate argument that
S(\alpha)\ll\left(UV+q+
N | |
\sqrt{U |
and then deducing the first formula above in the non-trivial cases when
q\leqN
U=V=N2/5
Vaughan's identity was generalized by .