Exceptional: On The Set in Goldbach'S Problem
Exceptional: On The Set in Goldbach'S Problem
Exceptional: On The Set in Goldbach'S Problem
By
Hiroshi MIKAWA
1. Introduction.
denote the number of even integers not exceeding
Let $E(x)$ that are not $x$
authors have unconditionally given the non-trivial bounds for $E(x)$ . The best
one of these is due to H. L. Montgomery and R. C. Vaughan. In 1975 they [13]
showed that there exists a positive constant such that $\delta$
$E(x)\ll x^{1-\delta}$ .
Chen J.-r. [2] gave an explicit value of , which is very small. $\delta$
providing
(3) $\frac{7}{12}<\theta\leqq 1$
.
This bound 7/12 comes from a zero density estimate for the Dirichlet L-series.
In 1981 Lou S.-t. and Yao Q. [9] attempted to sharpen the inequality (2). Later
Yao [20] replaced, in the same range of as (3), the right hand side of (2) $\theta$
by with some .
$x^{\theta(1-\delta)}$
$\delta>0$
write $\psi(x)=[x]-x+1/2$ , $e(x)=e^{2\pi ix}$ and $\Vert x\Vert=\min_{n\in Z}|x-n|$ . The convention
$n\sim N$ means that $N<n\leqq N^{\prime}\leqq 2N$ . $\rho$ in $\sum_{\rho(\chi)}$ runs through the set of non-trivial
zeros of .
denote certain positive absolute constants. For simplicity,
$L(s, \chi)$ $c_{i}\prime s$
2. Lemmas.
LEMMA 1. Let $l\leqq a\leqq q,$ The Hurwitz zeta-function $\zeta(s, a/q)$ is
$(a, q)=1$ .
regular, except for a simple pole at $s=1$ , of residue 1. Also, it satisfies the
growth conditions:
$|{\rm Im}(s)|\leqq x$
$c_{3}\leqq|{\rm Im}(s)|\leqq x$ .
LEMMA 2. Let $N(\sigma, \tau, x)$ denote the number of zeros of $L(s, \chi)$ in the
rectangle; ${\rm Re}(s)\geqq\sigma,$ $|{\rm Im}(s)|\leqq T$ . Let $A>0$ be given. If $q\leqq(\log T)^{A}$ and $\sigma\geqq$
.
LEMMA 4. For $k=1,2$ ,
$\sum_{\chi(mod q)}$
$\sum_{\rho(\chi),{\rm Re}(\rho)\geq_{1}\sigma_{\leq T}}|\sum_{n\sim N}\chi(n)\Lambda(n)n^{-\rho}|^{2k}\ll(qT+N^{k})N^{k(1-2\sigma)}(\log qN)^{12}|{\rm Im}(\rho)$
$\mathcal{P}$
where
$P(z)=$ $\prod_{p<z,p\in \mathcal{P}}p$
.
where is an approximation of
$X$
and $\Sigma\Omega(a)$ $\omega(d)$
is a multiplicative function
which satisfies some conditions of regularity. Write
$V(z)=\prod_{p<z,p\in \mathcal{P}}(1-\frac{\omega(p)}{p})$ .
LEMMA 5. Let $z,$ $D>2$ and $s=\log D/\log z$ . For $2\leqq s\leqq 4$
we have
$S(\Omega, \mathcal{P}, z)\geqq V(z)X\{\frac{2e^{\gamma}Iog(s-1)}{s}+O((\log\log D)^{-1})\}+R$ ,
$|\lambda_{d}|\leqq\mu^{2}(d)$
.
Moreover, on writing $z_{i}=\exp((\log D)(\log\log D)^{-i/10})$ , if $\lambda_{d}\neq 0$
for $d>z_{1}z$ then $\lambda_{d}$
where
$z_{2}\leqq P_{h}<D^{1/4}$
, $P_{h^{\prime}}=P_{h}z_{9}$
and $|\alpha(h, m)|\leqq 1$ .
This is the Rosser-Iwaniec linear sieve $[6, 14]$ . We explain the decomposi-
tion (2.1). It follows from $[6, 14]$ that if $\lambda_{d}\neq 0$
then is the form $d$
$\nu|P(z_{2})$ $p_{i}|P(z)/P(z_{2})$ ,
where
$K(y, \Delta)=(\frac{\sin\pi\Delta y}{\pi y})^{2}$ .
Put
$\mathcal{D}=\sum_{x<m,n}\sum_{\leqq 2x}a_{m}\overline{a}_{n}\max(O, \Delta-|m-n|)$ ,
$S=\int_{x}^{2x}|\sum_{l<n\leqq l+\Delta}a_{n}|^{2}dt$
,
and
$R=\Delta^{3}(\sup_{x<n\leqq 2x}|a_{n}|)^{2}$ .
We shall show $S\geqq \mathcal{D}+O(R)$ and $\mathcal{D}\geqq \mathcal{F}$
, from which Lemmas follow immediately.
Now,
$S\geqq\int_{x}^{2x-\Delta}=\sum_{x<m.n}\sum_{\leqq 2x}a_{m}\overline{a}_{n}$
meas. $\left\{\begin{array}{ll} & x\leqq t\leqq 2x-\Delta\\ t. & m-\Delta\leqq t<m\\ & n-\Delta\leqq t<n\end{array}\right\}$
$S\geqq\sum_{x+\Delta<m.n\leqq}\sum_{2x-\Delta}a_{m}\overline{a}_{n}(\Delta-|m-n|)+O(R)$
$|m-n|\leqq\Delta$
$=\mathcal{D}+O(R)$ .
Moreover,
$=\mathcal{F}$
,
since for .
$K(y, \Delta)\geqq(\Delta/2)^{2}$ $|y|\leqq 1/2\Delta$
3. Proof of Theorem.
Let be a sufficiently large parameter. We take $y=Y^{\theta_{1}}(1/4<\theta_{1}<1/2)$ and
$x$
$(2Y, 4Y]$ and $\mathscr{D}=(x-3Y, x-2Y$ ]. Our aim is to give a lower bound of the sum
$\mathcal{G}=\mathcal{G}(2k)=$
$\sum_{2k=p+p^{l}}$
$\log p$ ,
$p\in B$
for $ 2k\in_{t}\chi$
. Moreover, define
$S=S(2k, z)=$ $\sum_{2k=p+n}$
$\log p$
$p\in 9$
$(n.\mathcal{P}(2k.t))=1$
where
$\mathcal{P}(2k, z)=\prod_{p<z}p$ .
$p\mathcal{X}2k$
Suppose that
(3.1) $(4Y)^{1/3}<z<Y^{1/2}$ .
Since and $p\in B$ imply $n=2k-p\in_{-}A,$
$ 2k\in j\zeta$ $n$
counted by $S$ has at most two
prime factors . We thus have $\geqq z$
$p\in B$
where
$\rho^{\prime}(n)=$
$\sum_{n=pp^{i},z\leq p\leq p^{l}}1$
.
For any $l\in-A$ ,
$\Lambda(m)\Lambda(n)$
$\rho^{\prime}(l)\leqq$
$\sum_{l=mn,z\leqq m\leq n}\overline{\log}\overline{m\log n}$
$\leqq$
$\sum_{l=mn_{Y/z},\sqrt 2\overline{Y}<n\leq 4}1\frac{\Lambda(m)\Lambda(n)}{ogm\log n}$
$\leqq\frac{1}{\log(Y/z)1}\sum_{l\overline{ogz}=mn}\Lambda(m)\Lambda(n)$
$=_{-\cdot\rho(l)}-$ say.
p\in $
$=\Omega$
, say.
We proceed to evaluate $\Omega$
by means of Circle method. Define
$Q=yQ_{1}^{-2}$
On the Exceptional Set in Goldbach’s Problem 519
$Q_{1}=(\log Y)^{D}$
$m=[1/Q, 1+1/Q]\backslash M$ ,
where $D$ is a constant specified later, and $*stands$ for the restriction $(a, q)=1$ .
Furtheremore,
$T(\alpha)=--\sum_{n\in\Lambda}\rho(n)e(\alpha n)$ ,
where
$\Gamma=\frac{1}{2}\log(\frac{8Y}{z^{2}})$
.
Then,
$\Omega=\int_{1/Q}^{1+1/Q}T(\alpha)W(\alpha)e(-2k\alpha)d\alpha$
$=\int_{M}(T(\alpha)-V(\alpha))W(\alpha)e(-2k\alpha)d\alpha$
$+\int_{M}V(\alpha)W(\alpha)e(-2k\alpha)d\alpha$
$+\int_{m}T(\alpha)W(\alpha)e(-2k\alpha)d\alpha$
(3.3) $=\Omega_{0}-\Re_{1}-\Re_{2}$
, say.
We first consider $\Omega_{0}$
.
$\Omega_{0}=\sum_{q\leqq Q_{1}}\sum_{a=1}^{q}*\int_{-}^{1/_{1}q_{/}Q_{qQ}}-\cdot-\Gamma\frac{\mu(q)}{\varphi(q)}(\sum_{\circ}e(\beta n))(\sum_{P\in B}(\log p)e((\frac{a}{q}+\beta)p))$
$\times e(-2k(\frac{a}{q}+\beta))d\beta$
$\sum_{n\in\Lambda}\int_{-1^{2}/2}^{1/}e((n+p-2k)\beta)d\beta\equiv 1$ .
$=Y_{-}^{-}\Gamma\sum_{q\leqq Q_{1}}\frac{\mu^{2}(q)}{\varphi^{2}(q)}c_{q}(2k)$
(3.7) $\Re_{0}=\Re_{0}(2k)=Y_{-}^{-}\Gamma\sum_{q>Q_{1}}\frac{\mu^{2}(q)}{\varphi^{2}(q)}c_{q}(2k)$ .
where
$\mathfrak{S}(2k)=2\prod_{p>2}(1-\frac{1}{(p-1)^{2}})\prod_{p1,p>^{k_{2}}}(\frac{p-1}{p-2})$ .
$l_{q,a}$
then we find
On the Exceptional Set in Goldbach’s Problem 521
$\Vert\alpha^{\prime}-\alpha\Vert\geqq\Vert\frac{a^{\prime}}{q^{\prime}}-\frac{a}{q}\Vert-\Vert\alpha^{\prime}-\frac{a^{\prime}}{q^{\prime}}\Vert-\Vert\alpha-\frac{a}{q}\Vert$
$\geqq\frac{1}{q\prime q}-\frac{1}{q^{\prime}Q}-\frac{1}{qQ}$
$\geqq(2Q_{1})^{-2}$
.
We therefore have
$\sum_{2k\in J\zeta}\Re_{1}(2k)^{2}$
$\ll\int_{M}\int_{M}|V(\alpha^{\prime})W(\alpha^{\prime})||V(\alpha)W(\alpha)|\min(y,$ $\frac{1}{\Vert\alpha\prime-\alpha\Vert})d\alpha^{\prime}d\alpha$
$\ll Q_{1}^{2}\int_{M}|V(\alpha)|^{2}d\alpha\int_{M}|W(\alpha)|^{2}d\alpha$
$+y\sup_{(\alpha,q)=}\int_{I_{q,a}}|V(\alpha)W(\alpha)|d\alpha\cdot\int_{M}|V(\alpha)W(\alpha)|d\alpha$
then
$\Delta^{2}\int_{-1/2\Delta}^{1/2\Delta}|V(\frac{a}{q}+\beta)|^{2}d\beta\ll\Delta^{2}Y\mathcal{L}^{-E}$
uniformly for $(a, q)=1$ and $q\leqq(\log Y)^{B}$ . Here the implied constant depends only
on $\epsilon,$
$B$
and $E$ .
$2k\in\sum_{\zeta}\Re_{2}(2k)^{2}\ll\int_{m}\int_{m}|T(\alpha^{\prime})W(\alpha^{\prime})||T(\alpha)W(\alpha)|\min(y,$ $\frac{1}{\Vert\alpha-\alpha\Vert})d\alpha^{\prime}d\alpha$
$\ll\Delta(\int_{m}|T(\alpha)W(\alpha)|d\alpha)^{2}$
$+y\int_{m}|T(\alpha)W(\alpha)|(\int_{m_{\alpha^{\prime}-\alpha^{1}|\leq 1/2\Delta}}|T(\alpha^{\prime})W(\alpha^{\prime})|d\alpha^{\prime})d\alpha$
$\ll\Delta\int_{m}|T(\alpha)|^{2}d\alpha\int_{m}|W(\alpha)|^{2}d\alpha$
LEMMA 10. Let $1<\Delta<Y^{1/2}$ . Suppose that $|\alpha-a/q|\leqq 1/q^{2}$ with $(a, q)=1$ .
Let $E,$ $\epsilon>0$
be given. If $4Y/z\leqq Y^{8/15-\epsilon}$
then
$\sup_{a\in m}\Delta^{2}\int_{-1/\Delta}^{1/2\Delta_{2}}/$
$\ll\Delta^{2}Y\mathcal{L}^{F}Q_{1}^{-1/2}$
.
Combining this with (3.12) we obtain
(3.13) $\sum_{2k\in j\zeta}\Re_{2}(2k)^{2}\ll yY^{2}\mathcal{L}^{F}Q_{1}^{-1/4}$ .
For any given $E>0$ , we take $D=4(E+F)$ . Then (3.8), (3.11) and (3.13) become
$\leqq\frac{(1/2)\cdot.(1/15)}{(8/15)(7/15)}+\delta_{1}$
(3.15) $<\frac{1}{7}+\delta_{1}$
,
PROPOSITION 1.
$\Omega\leqq(\frac{1}{7}+\delta_{1})\mathfrak{S}(2k)\frac{Y}{\log Y}-\sum_{i=0}^{2}\Re_{i}+O(Y(\log Y)^{-3})$
.
Here, $\Re_{i}s$ satisfy (3.14).
For $d$
with $(d, 2k)=1$ , put
$=\sum_{d<D}\lambda_{d_{2k_{d|}}}\sum_{-p+n ,p^{-}\in_{n}B}$
iog $p-Y\sum_{d<D}\frac{\lambda_{d}}{d}\sum_{q|d}\frac{\mu(q)}{\varphi(q)}c_{q}(2k)$
$+O(D\mathcal{L}^{2}+Y\mathcal{L}\sum_{Q_{1}<q<D}\frac{|\mu(q)|}{\varphi(q)q}|c_{q}(2k)|)$
], as before, we see
(3.18) $\sum_{2k\in\$}\Re_{3}(2k)^{2}\ll yY^{2}\mathcal{L}^{4}Q_{1}^{-1}$ .
We next appeal to Circle method. Put
$S(\alpha)=\Sigma(\Sigma\lambda_{d})e(\alpha n)$
$n\in dd|n$
.
$d<D$
For $\alpha=a/q+\beta\in l_{q.a}$ , we write
$U(\alpha)=\sum_{n\in k}.((\sum_{d|n_{D}}\lambda_{d})e(\frac{a}{q}n)-(\sum_{d<_{d^{D}}}\frac{\lambda_{d}}{d}))e(\beta n)$ .
We then have
$ Z=\int_{1/Q}^{1+1/Q}S(\alpha)W(\alpha)e(-2k\alpha)d\alpha$
$=\int_{M}(S(\alpha)-U(\alpha))W(\alpha)e(-2k\alpha)d\alpha$
$+\int_{M}U(\alpha)W(\alpha)e(-2k\alpha)da$
$+\int_{m}S(\alpha)W(\alpha)e(-2k\alpha)d\alpha$
, we infer that
(3.20) $Z_{0}=Z_{1}+O(Y\mathcal{L}^{-3})$ ,
LEMMA 11. In Lemma 9, replace the sum and the condition $4Y/z\leqq Y^{8/15}$ $V$
by $U$
and $D\leqq Y^{1-\epsilon}$
, respectively. Then, the resulting assertion holds true.
by $S$
and $D\leqq Y^{1-\epsilon}$
, respectively. Then, the resulting assertion holds true.
(3.22) $\Sigma=\sum_{i=3}^{5}\Re_{i}+O(Y\mathcal{L}^{-3})$ .
We turn now to $\Pi$ . By Mertens’ theorem,
$\prod_{p^{p}x^{<_{2}z_{k}}}(1-\frac{1}{\varphi(p)})=\prod_{2<p<z}(\frac{p-1}{p}\frac{p(p-2)}{(p-1)^{2}})\cdot\prod_{pk}(\frac{p-1}{p-2})$
$\geqq 2\log(\frac{15}{7}-1)-\delta_{2}$
$>\frac{1}{4}-\delta_{2}$
,
$\Pi>(\frac{1}{4}-\delta_{2})\frac{\mathfrak{S}(2k)Y}{\log Y}$
.
Combining this with (3.16) and (3.22) we therefore obtain
PROPOSITION 2.
526 Hiroshi MIKAWA
problem. For all , the left hand side of (3.22) is zero. If $|R(2k)|\leqq$
$2k\in \mathcal{E}$
$\mathfrak{S}(2k)Y/11\log Y$ for some $2k\in X=(x, x+y$ ], then the right hand side of (3.22)
$\leqq\sum_{2k\in J\zeta}|R(2k)|^{2}$
$\ll yY^{2}\mathcal{L}^{-(A+2)}$
or
$E(x+y)-E(x)=\#\mathcal{E}\cap j\zeta\ll y\mathcal{L}^{-A}$ ,
as required.
This completes our proof of Theorem, apart from the verification of Lemmas
9, 10, 11 and 12.
4. Major arc.
Let $\epsilon^{\prime},$
$|f|_{2}\ll\Delta^{2}x\mathcal{L}^{-E}$
where the implied constant depends only on and $E$ . An admissible func- $\epsilon^{\prime},$ $B$
$v=\frac{\mu(q)}{\varphi(q)}\Delta\log(\frac{N}{M})+A$ . R. .
$\ll(\Delta x^{-\epsilon}+1)\mathcal{L}^{3}$
,
$v=\frac{1}{\varphi(q)}\sum_{\chi(q)}\chi(a)\tau(\overline{\chi})_{l<mn}\sum_{n\in^{\leq}J^{t+\Delta}}\chi(mn)\Lambda(m)\Lambda(n)+A$ . R..
Here $d=(M, N$ ]. By the explicit formula [15, Kap. V1I Satz 4.4.],
$\sum_{t/n<m\leq(t+\Delta)/n}\chi(m)\Lambda(m)=E_{0}\frac{\Delta}{n}$
$v=\frac{\mu(q)}{\varphi(q)}\Delta\sum_{(n^{n\in_{q)}J_{=1}}}\frac{\Lambda(n)}{n}$
$=\frac{\mu(q)}{\varphi(q)}\Delta\log(\frac{N}{M})$
$-\frac{1}{\varphi(q)}\sum_{\chi(q)}\chi(a)\tau(\overline{\chi})($
$\sum_{\rho(\chi),x<|1m(\rho)|\leqq x^{2}})\theta(\rho)N(\rho, \chi)$
$\sum_{\rho(\chi),|{\rm Im}(\rho)|\leqq x}+$
$\theta(s)=\theta(s, t, \Delta)=\frac{(t+\Delta)^{s}-t^{s}}{s}$
.
First we consider $v_{3}$ .
$|v_{3}|_{1}\ll\sqrt{q}\mathcal{L}^{2}\sum_{n\in 3}\int_{x}^{3x}\min(1,$ $\frac{1}{nx\Vert t/n\Vert})dt$
$\ll\sqrt{q}\mathcal{L}^{4}$
.
Since $v_{3}\ll\sqrt{q}\mathcal{L}^{2}N$
trivially, we have
(4.3) $|v_{3}|_{2}\ll qN\mathcal{L}^{6}$
.
Hence is admissible.
$v_{3}$
$|v_{2}|_{2}\ll \mathcal{L}\sum_{\chi(q)}\int_{x}^{2x}|$
$\sum_{\rho(X),x<|{\rm Im}(\rho)|\leqq x^{2}}\frac{t^{\rho}}{\rho}N(\rho, x)|^{2}dt$
$|v_{1}|_{2}\leqq\frac{q}{\varphi(q)}\sum_{\chi(q)}|\sum_{\rho(\chi)}\theta N|_{2}$
1 ${\rm Im}(\rho)|\leqq U$
$\ll\Delta^{2}x\mathcal{L}^{F}(\frac{q}{\Delta}+\frac{N}{x})$
(4.6) $\ll\Delta^{2}x^{3/4}$
.
In another case $1/2\leqq\sigma\leqq 1$
Lemma 2 yields that the supremum may be taken
,
over $1/2\leqq a\leqq 1-\eta(x)$ only, where $\eta(x)=c_{4}(\log x)^{-4/5}$ . Furtheremore, by Lemmas
3 and 4, we have
$K_{\sigma}(U)\leqq(\sum_{\chi(q)}$
$\sum_{\rho(\chi),R_{m^{e}}(\rho)\geqq\sigma}1)^{1/2}(\sum_{\chi(q)}$
$\sum_{\rho(\chi),R(\rho)\geqq\sigma}|N(\rho|I(\rho)|\leqq U|{\rm Im}^{e}(\rho)|\leqq U\chi)|^{4})^{1/2}$
Here $k>12/5$ . We see the supremum over $U$ is attained at $ U=x/\Delta$ again.
Then, since and
$x/\Delta\leqq M^{2}$ , we have that the contribution to
$1-\sigma\geqq 0$ is $|v_{1}|_{2}$
at most
$(1-\frac{1}{4})\frac{12}{5}\frac{1}{4}+(\frac{8}{15}-1)=-\frac{1}{60}$
.
Hence, there exists a constant $\xi>0$ such that
$(\frac{x}{\Delta})^{k/4}\frac{N}{x}\leqq x^{-\xi}$
.
Thus (4.7) becomes
$\ll\Delta^{2}xq\mathcal{L}^{F}\sup_{I/2\leq\sigma\leq 1-\eta(x)}x^{-2\xi(1-\sigma)}$
$\ll\Delta^{2}xq\mathcal{L}^{F}\exp(-c_{6}(\log x)^{-1/6})$
(4.8) $\ll\Delta^{2}x\mathcal{L}^{-E}$
,
with (4.2), (4.3) and (4.4), we conclude that $v-v_{0}$ is also admissible, as required.
$u=(\sum_{d<_{d^{D}}}\frac{\lambda_{d}}{d})\Delta+A$ . R..
PROOF. Put
$f(s)=\sum_{n=1}^{\infty}(\sum_{a1,a<^{n_{D}}}\lambda_{d})e(\frac{a}{q}n)n^{-S}$ ,
$=\frac{1}{q^{\iota}}\sum_{r|q}\sum_{b\Leftarrow 1}^{r}*(\sum_{d<D}\lambda_{d}e(\frac{abd}{r})d^{-S})\gamma^{\$}($
$\sum_{\iota>0,l\equiv b(r)}l^{-s})$
$j=1.2$
$\ll\sup_{1/2\leq\sigma\leq c}q^{2}\mathcal{L}^{2}(Dx^{-2/3})^{1-\sigma}$
$\ll q^{2}x^{1/6}$ ,
$\frac{b}{r})|^{2})\int_{1/2-iU}^{1/2+iU}|D_{1}(s,$ $\frac{b}{r})|^{2}|ds|$
$\ll \mathcal{L}^{2}\sup_{x/\Delta\leqq U\leq x}(\frac{x}{U})^{2}\sum_{r|q}\sum_{b=1}^{r}*(\sup_{|{\rm Im} ts)|\leqq U}|\zeta(s,$
Since is admissible.
$\Delta\geqq x^{1/4+\epsilon^{\prime}},$
$u_{1}$
5. Auxiliary results.
In this section we provide for the proof of Lemmas 16 and 17 in the next
section. For real numbers and , we set $a,$
$b$
$c,$ $a\leqq b$
$\Phi(a, b;c)=\psi(b-c)-\psi(a-c)$
$=[b-c]-[a-c]-(b-a)$ .
Let and be arbitrary sequences with modulus
$\alpha,$ $\beta,$ $\gamma$
. Moreover, let 1
$\delta$
$\leqq 1$
Here, $1\ll K,$ $D,$ $M,$ $N\ll x,$ $MN\leqq t$ and $s\leqq 2x/D,$ $2x/MN$.
Our aim is to prove $F\ll Kx^{1-\epsilon},$ $\epsilon>0$
. Since $F\ll KDMN$ trivially, we assume
(5.1) $DMN>x^{1-2\epsilon}$ .
LEMMA 13. For any $\epsilon>0$
, if $MN,$ $DM\leqq x^{2/3-4\epsilon}$
then
$F_{1}\ll Kx^{1-\epsilon}$
.
PROOF. We appeal to the Erdos-Tur\’an inequality [8, Lemme 2], which
runs as follows. For arbitrary real numbers and $H>2$ , $x_{n}$
$\ll \mathcal{L}$
$\sup_{y\leqq 2x/DM,\sigma\subset[N.2N]}(\frac{N}{H}+\sum_{0<h\leqq H}\frac{1}{h}|,\sum_{d(n^{n\in t_{=}})^{\llcorner}1}e(\frac{hy}{n})e(-hk\overline{\frac{mn}{d}})|)$ .
$\ll \mathcal{L}\frac{KDMN}{H}+\mathcal{L}^{4}x^{\epsilon}(1+\frac{Hx}{DMN})Kx^{1-6\epsilon}$
.
Also,
where
$|A_{h}|\ll\min(\frac{\log H}{H},\frac{H}{h^{2}})$
.
in is expressed as a sum of the main part of Fourier series,
Now, $\Phi$ $F_{2}$
$F_{2}=F_{2^{\prime}}+F_{2^{\prime\prime}}$
.
First we consider $F_{2^{\prime\prime}}$
$=\sum_{h}|A_{h}|S(h)$ , say.
By the same argument as that in $[11, $(17)(18) $]$ with $q=1$ , we then obtain
$F_{2^{\prime\prime}}\ll Kx^{1-2\epsilon}+\mathcal{L}^{s}x^{4\epsilon}K\{(MN)^{S/2}+D(MN)^{1/2}\}$ .
Here we have taken $H=DMNx^{3e-1}>2$ , because of (5.1). Since $D,$ $MN<x^{2/3-4\epsilon}$ ,
we have
$F_{2^{\prime\prime}}\ll Kx^{1-\epsilon}$
.
We turn now to $F_{2}$
‘. By (5.2),
$F_{2}^{\prime}=$ $\sum$ $\Sigma\sum$ $\sum\alpha_{d}\beta_{m}\gamma_{n}\delta_{k}$
$0<k\leq Kd\sim_{(d.\tilde{m}n)=}DmNn_{1}\sim N$
$mn\leq t$
where the supremum is over all arithmetical functions $B$ with $|B|\leqq 1$ . To
deal with the condition $mn\leqq t$ , we use a lemma in Fourier analysis [1, Lemma
2.2.], which runs as follows. Let be arbitrary finite sequence. If $X\leqq Y$ $a_{n}$
then
$|\sum_{n\leq X}a_{n}|\leqq\int_{-\infty}^{\infty}f(\theta, Y)|\sum_{n}a_{n}e(\theta n)|d\theta$
.
Here, the kernel $f\geqq 0$
satisfies
$\int_{-\infty}^{\infty}f(\theta, Y)d\theta\ll\log Y$ .
Now, applying this lemma with $Y=t/M$, we may remove the condition
$mn\leqq t$ in (5.4) with the cost of $(\log x)$ time. Thus, the method in [11, p. 38,
39] is applicable. We therefore have
$F_{2^{\prime}}\ll \mathcal{L}x^{3\epsilon}\{(KDM)^{1/2}x^{1/2-\epsilon}+Kx^{6\epsilon/2}(D^{1/2}M’/4N^{s/2}+DM^{3/4}N‘/2)\}$ .
Since $DM<x^{8/16-4\epsilon}\cdot x^{7/15-4\epsilon}=x^{1-8\epsilon}$
, the above first term is
$\ll \mathcal{L}x^{3e}K^{1/2}x^{1-5\epsilon}$
$\ll Kx^{1\leftarrow\epsilon}$
.
We may assume $M\geqq N$, because of the symmetry of $F_{2}$
, so that $M\geqq(MN)^{1/2}$ .
Hence, the second term is
$\ll Kx^{6\epsilon}\{D^{1/2}(MN)^{3/2-1/8}+D(MN)^{1/2}M^{1/4}\}$
$\ll Kx^{6\epsilon}\{(x^{8/15-4\epsilon})^{1f/8}+(x^{8/15-4\epsilon})^{3/2}(x^{7/16-4\epsilon})^{1/4}\}$
$\ll Kx^{1-\epsilon}$
.
We thus obtain Lemma 14.
$j$
II $(h;D, V)=$ $\sum_{h=lmn}1(l)a(m)1(n)$ ,
$mn\leq D$
$m\leqq V$
536 Hiroshi MIKAWA
$im(h;D, U)=\sum_{h=lmn}1(l)a(m)b(n)$ ,
$mn\leq D$
$m.n\geqq U$
$ Ji(*)=Ji(\alpha, x, \Delta;*)=\Delta^{2}\int_{-1/}^{1/2\Delta_{2\Delta}}J*)e((\alpha+\beta)h)|^{2}d\beta$
(61). $|\alpha-\frac{a}{q}|\leqq\frac{1}{q^{2}}$
with $(a, q)=1$ and $1<q<\Delta<x/2$ .
where
The above simultaneous congruences are soluble if and only if $(d_{1}, d_{2})|r$ . We
write and $r^{*}=r/(d_{1}, d_{2})$ . We then have, with the notation in
$d_{i^{*}}=d_{i}/(d_{1}, d_{2})$
section 5, that
$\#\{n\}=\#\{l:\underline{\max}([\frac{x,d_{1}s,d_{2}s)}{d_{1},d_{2}]}<\gamma^{*}\overline{\frac{d_{1}^{*}}{d_{2}^{*}}}+l\leqq\frac{2x}{[d_{1},d_{2}]}\}$
$=\frac{2x-\max(x,d_{1}s,d_{2}s)}{[d_{1},d_{2}]}+\Phi(\frac{2x}{[d_{1}}d_{2}^{-}\overline{]}\frac{\max(x,d_{1}s,d_{2}s)}{[d_{1},d_{2}]}$ ; $\gamma^{*}\overline{\frac{d_{1}^{*}}{d_{2}^{*}}})$
$=\Phi_{0}+\Phi$ , say.
If $d_{1}$
or $d_{2}>2x/s$ then $\#\{n\}=0$ . Hence
$\Sigma=\sum_{0<r\leqq t}e(\alpha r)\sum_{d_{1}.d_{2^{\xi}}}\sum_{2x/s}a(d_{1})a(d_{2})(\Phi_{0}+\Phi)+O(\Delta^{2}x^{\epsilon})$
$(d_{1}.d_{2})|r$
$\frac{1}{\Vert\alpha\delta\Vert})$
$\ll x\sum_{0<\delta\leqq\Delta}(\delta\sum_{d\leqq 2_{1}x/s}\frac{|a(d)|}{d})^{2}\delta^{-1}\min(\frac{\Delta}{\delta},$
(6.5) $\ll\Delta(\frac{x}{s})^{2}\mathcal{L}^{F}$
.
In conjunction with $(6.2)-(6.5)$ we obtain Lemma 15.
$+\Delta^{2}\mathcal{L}^{F}x^{1-2\epsilon}+\sup$ $|\sum_{\iota,t}|$
,
$s\leq x^{1/}t\leqq\Delta_{2+\epsilon}$
where
$\sum=\sum_{r0<\leqq t}e(\alpha r)\sum_{d_{1}.d_{2}\leqq\min t2}\sum_{x/s,D)}w(d_{1})w(d_{2})$
$(d_{1}, d_{2})|r$
with
$w(d)=\sum_{d=,mn^{\prime}}a(m^{\prime})1(n^{\prime})$
.
Write
$(d_{1}, d_{2})=\delta$ ,
$d_{2}=\delta d$ ,
$r^{*}=k$ .
We then find
538 Hiroshi MIKAWA
$\sum\ll x^{\epsilon}\delta\leq|$
where $F_{1}$
is defined in section 5. Since
$\frac{DV}{\delta}\frac{D}{\delta}\leqq(\frac{x}{\delta})^{2/3-8\epsilon}$
Lemma 13 yields
$\Sigma\ll x^{\epsilon}\sum_{\delta\leq\Delta}(\frac{\Delta}{\delta})(\frac{x}{\delta})^{1-2\epsilon}$
$\ll\Delta x^{1-\epsilon}$
.
Combining this with (6.6) we obtain Lemma 16.
When $\delta>D^{2}x^{2\epsilon-1}$
, we use the trivial bound
$F_{2}\ll\frac{\Delta}{\delta}(\frac{D}{\delta})^{2}$
.
Otherwise, we make use of Lemma 14. Since
$\frac{D}{\delta}\leqq(\frac{x}{\delta})^{8/15-12\epsilon}$
and $\frac{U}{\delta}\geqq(\frac{x}{\delta})^{\iota/15}$
Lemma 14 yields
$F_{2}\ll(\frac{\Delta}{\delta})x^{1-3\epsilon}$
.
On the Exceptional Set in Goldbach’s Problem 539
We therefore have
$\ll\Delta x^{1-2\epsilon}$
,
as required.
LEMMA 18.
the ranges for $m,$ and into $0((\log x)^{3})$ intervals of the form $(M, 2M$ ],
$n$ $k$
$k.k^{\prime}\sim K$
$\ll\Delta \mathcal{L}^{F}MNK$
$m\geq M$
$ 0<mr\leqq\Delta$
$\ll\Delta(\frac{\Delta}{q}+\frac{\Delta}{M}+q)$
,
by [15, Kap. VI Lemma 6.3.]. Combining this with (6.7) and (6.8), we finally
540 Hiroshi MIKAWA
obtain
$\ll\Delta x\mathcal{L}^{F}(U+\frac{\Delta}{q}+\frac{\Delta}{V}+q)$
.
7. Minor arc.
In this section we prove Lemmas 10\dagger and 12\dagger below, which cover Lemmas
10 and 12, respectively. Throughout this section we assume that
$|\alpha-\frac{a}{q}|\leqq\frac{1}{q^{2}}$
with $(a, q)=1$ , and $1<\Delta<x^{2/3}$ .
LEMMA $ 10\dagger$
. Put
$\xi\Gamma=\sigma\tau(\alpha, x, \Delta, N)=\Delta^{2}\int_{-1/2}^{1/2\Delta_{\Delta}}|_{x<m}\sum_{x<n^{n}\leqq^{\leqq}N^{2x}}\Lambda(m)\Lambda(n)e((\alpha+\beta)mn)|^{2}d\beta$ .
$\rho(h)=\sum_{h=m}\Lambda(m)\Lambda(n)$ , $L\sigma=(\sqrt{x}, N$ ].
We shall decompose into $O(1)$ sums of the ji-type weight, which is defined
$\rho$
$v=Nx^{-1/4}$
$w=x^{\xi}$ , $\epsilon=20\xi$
.
Then, for $x<mn\leqq 2x$ and $n\in d$ ,
$m^{\prime}>u$
$n^{\prime}\succ 0$
Let $\rho_{2}$
and $\rho_{3}$
be the sums corresponding to
$\delta=2$ $\delta=2$
$\epsilon=2$ $\epsilon=2,3$
and
$\mu=1$ $\mu=1$
$\nu=1$ $\nu=4$
therefore have that is splitted into 21 sums of the $jIV(h;u, w, )$ -type sum,
$\rho_{4}$
with or $(0,2x$ ].
$\mathscr{S}=_{c}9$
$d\leqq w$
$m.m>wnn^{\prime,}\in J$
$eg\in J$
Thus, $\rho_{2}(h)$
is divided into 4 sums of the $j$
II $(h;w^{2}N, w^{4})$ -type sum.
We turn now to $\rho_{3}$
542 Hiroshi MIKAWA
since $nn^{\prime}\in d$
and $u<n^{\prime}\leqq v$
imply $n>1$ . Hence, $\rho_{3}(h)$
is a sum of two $1m(h$ ;
$w^{2}N,$ $u$ )-type sums.
We therefore conclude that
$\sigma\tau\ll J$
II $(w^{2}N, w^{4})+Jm(w^{2}N, u)+JIV(u, w, \mathscr{S})$ .
We note that
$w^{6}N\leqq x^{8/15-\epsilon+12\xi}\leqq x^{2/3-8\xi}$
,
$u=2x^{1/4}\geqq x^{1/5}$ .
Thus, Lemmas 16, 17 and 18 yield
$f\ll\Delta x\mathcal{L}^{F}\{\Delta q^{-1/2}+(q\Delta)^{1/2}\}+\Delta^{2}x^{1-\xi}+\Delta x\mathcal{L}^{F}(u+\frac{\Delta}{q}+\frac{\Delta}{w}+q)$
as required.
LEMMA $ 12\uparrow$
. Let $(\lambda_{d})=(\lambda_{d}(D, z))$ be the weights introduced in Lemma 5. Put
$+\Delta x\mathcal{L}^{F}(D^{1/4}z_{9}+\frac{\Delta}{q}+\frac{\Delta}{z_{2}}+q)$
On the Exceptional Set in Goldbach’s Problem 543
$\ll\Delta x\mathcal{L}^{F}\{D^{1/4}z_{9}+\Delta q^{-1/2}+(q\Delta)^{1/2}\}+\Delta^{2}X^{F}z_{1}^{2}D+\Delta^{2}x\mathcal{L}^{F}z_{2}^{-1}$
References
[1] Bombieri, E. and Iwaniec, H., On the order of $\zeta((1/2)+it)$ . Ann. Scuola Norm.
Sup. Pisa 13 (1986), 449-472.
[2] Chen, J.-r., The exceptional set of Goldbach number, Sci. Sinica 26 (1983), 714-731.
[3] Gallagher, P. X., A large sieve density estimate near , Invent. Math. 11 (1970), $\sigma=1$
329-339.
[4] Hardy, G. H. and Littlewood, J. E., Some problems of ‘Partitio numerorum’; (V).,
Proc. London Math. Soc. (2) 22 (1923), 46-56.
[5] Huxley, M. N., Large values of Dirichlet polynomials III, Acta Arith. 23 (1975),
435-444.
[6] Iwaniec, H., A new form of the error term in the linear sieve, Acta Arith. 37
(1980), 307-320.
[7] Iwaniec, H. and Jutila, M., Primes in short intervals, Ark. Mat. 17 (1979), 167-178.
[8] Laborde, M., Nombres presque-premiers dans de petits intervalles, S\’eminaire de
Th\’eorie des Nombres, Bordeaux I, Ann\’ee 1977-78, expos\’e 15. $n^{o}$
[9] Lou, S. -t. and Yao, Q., The exceptional set of Goldbach numbers in a short in-
terval, Acta Math. Sinica 24 (1981), 269-282.
[10] Mikawa, H., On prime twins, Tsukuba J. Math. 15 (1991), 19-29.
[11] Mikawa, H., On the Brun-Titchmarsh theorem, Tsukuba J. Math. 15 (1991), 31-40.
[12] Montgomery, H. L., Topics in multiplicative number theory, Springer, L. N. Math.
227, 1971.
[13] Montgomery, H. L. and Vaughan, R. C., The exceptional set in Goldbach’s problem,
Acta Arith. 27 (1975), 353-370.
[14] Motohashi, Y., Sieve methods and prime number theory, Tata Inst. Fund. Res.,
Lect. Math. Phys. 72, 1983.
[15] Pracher, K., Primzahlverteilung, Springer, 1957.
[16] Ramachandra, K., On the number of Goldbach numbers in small intervals, J. Indian
Math. Soc. 37 (1973), 157-170.
[17] Vaughan, R. C., Sommes trigonom\’etriques sur les nombles premiers, C. R. Acad.
Sci. Paris S\’er. A 286 (1977), 981-983.
[18] Vinogradov, I. M., Representation of an odd number as a sum of three primes,
Dokl. Akad. Nauk SSSR 4 (1934), 185-187.
[19] Wolke, D., \"Uber das Primzahl-Zwillingsproblem, Math. Ann. 283 (1989), 529-537.
[20] Yao, Q., The exceptional set of Goldbach numbers in a short interval, Acta Math.
Sinica 26 (1982), 315-322.
Institute of Mathematics
University of Tsukuba