Invent. math.

75, 123-141 (1984) Iyl lden ~ione$

mathematicae
9 Springer-Verlag 1984

Mean values of the Riemann zeta-function

and its derivatives

S.M. Gonek
Department of Mathematics, University of Rochester, Rochester, NY 14627, USA

w 1. Introduction and statement of results

In 1918 Hardy and Littlewood [-2] proved that as T-~ov

T
1~(89 ~ Tlog T. (1)
1

In 1928 Ingham [3] generalized this considerably by showing that as T ~

r T
~(u)(89 it)~(~)(89 it)dt ~ (log T) "+v+ ', (2)
1 ~ -t- V "}- s

wher ((")(s) denotes the ~th derivative of ((s) (=((~ Since ((u)( 89
=((,1(89 it), it follows in particular that
T T
t 1~(")(89 it)lZdt ~2/2 + 1 (log T) 2u+ 1, (3)

which gives (1) when # = 0 . Our object in this paper is to prove some new types
of mean value formulae which are, when the Riemann hypothesis is assumed,
discrete analogues of (1)-(3).
We denote the non-trivial zeros of ~(s) by P = f l + i 7 and we set L
1 T
= 2-~-log ~ - . Our main result is the following

Theorem. I f T is sufficiently large and ~ is any real number satisfying tctl < 89
then

~(U)(p + i~L-1)~(v)(1- p - i e t L - x)
I<v<T

= ( _ 1)u+~ ( 1 -H(l~,v, 2 ~ ) - H ( v , # , - 2no0) 2 ~ (log T)"+~+ 2

/~+v+ 1
+ O(T(log T y + ~+ '), (4)
124 S.M. Gonek

where
(2~i) l
H(//, v, 2~z~)=/~! ~=o
~ (l+#+l)!(l+#+v+2)"

The constant implicit in the O-term is independent of ~.

As an immediate consequence we have
Corollary 1. Suppose that the Riemann hypothesis is true. I f T, ~, and H are as
in the Theorem, then

~)( 89+ i(Y+c~L-1))~)( 89 + c~L-1))

I~?-<T

=(-1)"+~ ( / / + v_1+ l H(g,v, 2 r c e ) - H ( v , # , - 2 z c c 0 ) 2 ~ (log T)U+v+2

+ O (T(log T) u+'+ 1 (5)

In particular,

1~r189 + i(7 + c~g- 1))12

l<_-7<T

= ( 2 ~ -1 H ( / / ' g ' 2 n c 0 - H ( # ' # ' - 2rt~)). ~-~

T (l~

+ O(T(log T) 2"+ a). (6)

The constants implicit in the O-terms are independent of ~.

In (5) and (6) we have discrete analogues of (2) and (3). Note that there are
T
~ - log T terms in the sum in (6) and that the right-hand side of (6) is
/~z T
(log T) 2"+ 2
(2//+ 1)(//+ 1)2 2~

when e = 0 a n d / / > 1. Thus, comparing (3) and (6), we see that on the Riemann
hypothesis the average of [~(")(89+ i?)[2 over those zeros with 1 <7 < T is smaller
by a factor of than the average of IEr189 2 over all points with
l <_t<_T.
The case l / = 0 of (6) is a discrete analogue (1) and is of interest in its own
right so we state it as
Corollary 2. Assume the Riemann hypothesis is true. I f T is sufficiently large and
o~ is a real number such that I~[ < L/2, then

I~(g+
1 i(7 + ~L- 1))12= ( 1_ (sinnc~] 2] 2--~log2T+O(TlogT). (7)
I~7~_T \ gc~ l/

The constant implicit in the O-term is independent of c~.

Mean values of the Riemann zeta-function and its derivatives 125

J. Mueller (see [6] and [7]) has recently found an interesting application of
Corollary 2. D e n o t e by 0 < 7 2 <72 <--- the imaginary parts of the zeros of ((s)
in the upper half-plane and set
log 7,
2=limsup(7"-7"-l) 2n '
n

log 7,
/~ = l i m i n f ( 7 " - 7"- 1) 2n
n

A. Selberg [8] has remarked that ~ t < l and 2 > 1, and H. M o n t g o m e r y [5],
assuming the R i e m a n n hypothesis, has shown that # < 0 . 6 8 . Mueller's result is
Corollary 3. If the Riemann hypothesis is true 2 > 1.9.
As the p r o o f of this is brief, we give it in Sect. 6.

This paper is, in revised form, a part of my doctoral thesis written under the direction of
Professor H.L. Montgomery. He suggested the problem which led to this investigation and made
many helpful comments concerning the form and content of my thesis. It gives me gerat pleasure
to express my gratitude to him here. I would also like to thank Professors D.J. Lewis, G. Piranian,
and J. Ullman for carefully reading my thesis, and the referee for his very helpful suggestions.

w2. Some formulae and estimates

Before we develop the basic idea of the p r o o f of the Theorem, it will be useful
to set down certain formulae and estimates.
T h r o u g h o u t this paper s = a + it denotes a complex variable.
Let

z ( 1 - s) = n ~ - ~ (8)

F(s) being the gamma-function. The unsymmetric form of the functional

equation for ~(s) is

(1 - s) = X(1 - s) ~(s). (9)

We also require the symmetric form of the functional equation. Set

~(s)= 89 (2) ~(s). (10)

Then
~(s) = 4(1 - s). (11)

The function ~(s) is entire of order one and its only zeros are the non-trivial
zeros of ((s).
126 S.M. Gonek

We write Stirling's formula for F(s) in the form

logF(s)=(s-89189 (1)
+O ~ , (12)

This is valid for Isl~ and ] a r g s [ < r t - 6 , where 6 > 0 is arbitrary but fixed (see
Whittaker and Watson [10; Chaps. 12, 13]). Using this, it is not difficult to
show that

X(1-s)=e Z \~] exp[itlogt/2ne] (t3)

for a fixed and t > 1, say.

Euler's psi-function is defined by

r'(s)
~O( s) - (14)
r(s) "

When l a r g s l < r t - 6 and Isl>89 we have

O(s)=l~ (~s[)" (15)

This may be derived from (12) by means of Cauchy's estimate for analytic
functions.
As is well known, in each intervaI ( n , n + l ) (n=2,3,...) we can select a
number T, such that if ~ is the ordinate of any zero of ~(s), then IT,-,/]
1
>-- In this way we obtain a sequence J" which will be fixed throughout
log T,"
this paper.
Recall that if T is large and does not coincide with the ordinate of any zero
of ~(s), then
1
~-(cr+iT)= ~ t-O(log T)
Iv-TI< 1 S--p

uniformly for - l < c r < 2 (cf. Davenport [1; p. 99]). There are <{log T terms in
this sum and if Te~-, each term is ,~logT. Thus, for each large T e 3 - and
uniformly for - 1 < cr _<2,

~-(a+ iT),~tog 2 T. (16)

r

By logarithmic differentiation of (10) we have

2s-1
~-~'(s)=~-(' (s) + 89~ (s/2) - 89log n - t - -(s- s- 1) (17)
Mean values of the Rlemann zeta-function and its derivatives 127

We deduce from this, (15), and (16) that

~- ( 6 + i T ) ~ l o g 2 T (18)

for all large T~Y;, uniformly for - 1 <a_<2.

Similarly, we may combine the estimate

-(o+i0 l (a>a>l)

with (15) and (17) to obtain

~ (a + it) ~ l o g 2[tl (19)

for a > a > l and I t l > l , say.

Finally, we need the estimates

[ [t]~-~+~ if or<0
((~)(o+it)~{tt[ ~(1-~)+~ if 0 - < a - < l (20)
/

[It] ~ if a > l ,

where e > 0 is arbitrary, It[>_-89 and v = 0 , 1,2 ..... These may be derived from
the case v = 0 , It] >88 (for which see Titchmarsh [9; pp. 81-82]) by applying
Cauchy's estimate for the derivatives of analytic functions to ~(s) in a small
disc centered at s=a+it.

w3. Beginning of the proof

We can now begin the proof of the Theorem, although we will require a
section of lemmas (Sect. 4 below) to complete it.
Let 1 < a < 2 and let R denote the closed rectangle in the complex plane
with vertices at a+i, a+iT, 1 - a + i T , 1 - a + i , where T is large. We define
!

I=I(#, v, 6 ) = - - ~ ~'(s)~(U)(s+i6)~(v)(1-s-i6)ds,
2hi on
where @R is the boundary of R and the integral is taken in the counterclock-
wise sense. Also, we assume @is real and 161-589 By the theory of residues

I(#,v, 6)= ~ ~(")(p+i6)~(~)(1-p--i6) (21)

1 <,2<T

provided that no zero p lies on @R. Since the ordinate of the first zero of ~(s)
above the real axis is > 14 and no zeros lie on the vertical edges of R, we need
only insure that T is not the ordinate of a zero. This will be the case if T e J - ,
the set constructed in Sec. 2. F r o m now on we assume T e g - ; at the end of the
proof this restriction will be removed.
128 S.M. Gonek

To prove the Theorem we must estimate the integral I(#, v, 6). To do this
we first split it into four parts corresponding to the four sides of R. We write
4
I(#, v, 6)= Z Ij(#, v, 6),
j=l
where 11 is the integral over [a+i,a+iT), I z is over [a+iT, 1 - a + i T ) , 13 is
over [ 1 - a + i T , l - a + / ) , and 14 is over [1-a+i,a+i). Since Ill< 89 the in-
tegral in 14 is bounded, i.e. 14,~ 1. Next

I2 '~ ~-a=<o.-<amaX~ ( a + i T ) ~ ( " ) ( a + i T + i 6 ) ~ ) ( 1 - a - i T - i 6 )

~logZT max ]~")(a+iT+if)~v)(1-a-iT-i6)l

1--a<-a<-a
by 08). The last line is

__<logZT( max + max + m a x ) l ( w ) ( a + i T + i 6 ) ( t ~ ) ( 1 - a - i T - i 6 ) [ ,

1--a<=a<=O 0 5 o ' - - < 1 1 =<o'=<a

which by (20) is
=<log2 T(T"-~+ 2~+ T-~+ 2~+ T~-~+ 2~).

Since a > 1 and e > 0 is arbitrary, we obtain

12 '~ T~-~ +~.

This and the estimate for 14 lead to

I(#, v, 6) =I~(#, v, 6)+ I3( #, v, 6)+O(T"-~+~). (22)

We now consider 13. The logarithmic derivative of (i1) is

S ~ - --S.

Using this and the fact that both ((~)(s) and ~-(s) satisfy the reflection principle,
we get

I3(#, v, 6)= - - - 1 1-.+i

~ (1-s)(W)(s+if)((~)(1-s-if)ds
2nit - .+ iT g

-2nil 1 i~'~(a-it)~(u)( 1 - a + i t +ib)~(V)(a-it-ic~)idt

l i~'
-27ri ~ ~ (a+it)~")(1 -a-it-ifi)~(~)(a+it +ib)idt
1 a+iT~
- (s) ((")(1 - s - i 6) ((~)(s + i 6) ds
2rti .+i
=Is(v, #, (5).
M e a n values of the R i e m a n n zeta-function a n d its derivatives 129

This and (22) yield

I(#, v, 8) =I1(#, v, 8)+I1(v , #, 8) +O(T"-~+~). (23)

Our problem is now reduced to estimating I~(#, v, 6).

w3. Lemmas

Our first two lemmas are modified versions of Lemmas 3.2 and 3.3 of
N. Levinson [4].
Lemma 1. There is a small c >0 such that

I o= ~ exp it log ~r ~2~! dt

r 1 -c)
=(2rc)l-,r%-i, +~i/4 + O(r,- ~)

for large r and a arbitrary but fixed.

Proof. This result follows from the usual stationary phase techniques. If we set
t = r(1 + x) then we can write

I o =(27t) 89
where
It = i exp(irf(x))(1 +x)"-~dx
--c

with
f ( x ) = (1 + x ) l o g ( 1 + x ) - (1 +x).
Z2
Now f ( z ) = ~ - + . . , is holomorphic in a neighborhood of z = 0 with only a

double zero at z = 0 . Thus u ( z ) = ~ = z + . . . is holomorphic and u'(z)+-O in

a neighborhood of z=0. We make the change of variables u(z)=u and obtain,
if c is sufficiently small,
u(c)
11= ~ eiru2/Zg(u)du,
u(-c)

where g(u)= 1 + ... is holomorphic in a neighborhood of u=0. Now integration

by parts yields
I~= ~ e~'"~/2du+O ,
u(-c)
and
S e iru2/2du = e~I/4 + 0
u(-c)

by the method of stationary phase; this proves Lemma 1.

130 [Link]

Lemma 2. For large A and A <r<B <2A

~ exp [itl~ ( t ) ] \2hi dt=(2n)l-"r~e-i'+~i/4+E(r'A'B)' (24)

where a is fixed and where

Aa+89 ( Ba+89
E(r,A,B)=O(A"-')+O (IA_rI+A 89 \IB_-~]~B~]. (25)

For r<=A or r>B,

B
! exp [it l~ ( t ) ] (\2n!
t~-~"-~dt=E(r, A, B).

Proof If A<r<=A+A ~ or if B-B~<r<__B the integral is O(A") by Titchmarsh

[9], Ch. IV, 4.5, Lemma 4.5. If instead A +A ~ < r < B - B ~, we have

! exp [it l~ (r@) ] \2-n~! dt=Io+Jl+J2,

where
J~= S exp it log ~e \2n ! dt
A

and similarly for J2. Now integration by parts shows that

and of course a similar estimate holds for J2. Finally, if either r < A - A ~ or
r>B+B 89 the required estimate follows by integration by parts. In view of
Lemma 1, this completes the proof.
Lemma 3. For m-O, 1,2, ..., A large, and A <r<B<__2A,

iexp[itlog(t)]\2n] (log~-) dt
A
(r) m
=(2n)l-"rae -ir+~i/4 l o g ~ n- +E(r,A,B)(logA) m,

while for r ~ A or r > B,

t (~) ( l o g ~ ) dt=E(r,A,B)(logA)~,

where E(r, A, B) is (25).

Proof The proof of Lemma 3 is easily obtained from Lemma 2 and integration
by parts and therefore we omit it.
Mean values of the Riemann zeta-function and its derivatives 131

L e m m a 4. Let E(r,A,B) be as in (25), where A is large and A<B<_2A. Assume

{b,},~= 1 is a sequence of complex numbers such that b, ~ n ~ for any ~ > O. Then if
a>l,
~ b"E(2nn, A , B ) ~ A " - ~ .
n=l na

Proof Choose e so that 0 < e < a - 1 . By (25)

b
E ~",E(2nn, A , B ) ~ ~ n-"+~E(27rn, A,B)
n=l n n=l

.= 1 .= 1 n"-~(lA - 2 n n l + A ~')
1
+ B"+89 ~ I n~-~(lB- 2nnl + B~)"

The p r o o f of the l e m m a is completed by noting that

n_a+~ 1
n=l

and

~C -1.
.= 1 n"-~(] C - 2nnl + C ~)

indeed the last inequality is easily established by considering separately the

ranges I C - 2 n n [ < C ~ and I C - 2 n n [ > C ~.
L e m m a 5. Let {b.}.~ 1 be a sequence of complex numbers such that for any ~ > O,
b . ~ n ~. Let a> 1 and let m be a non-negative integer. Then for T sufficiently
large,

1 i(~,lb.n_a_~,))~(l_a_it)(log~_~)"dt
2n 1 =
= ~ b,(log n) m + O ( T " - ~ ( l o g T)m). (26)
l <.n<_ T / 2 n

Proof By (13) we have

1 T H -a-it t n

-- ao . t
1 i ( ~ b.n-"-i~)e-~i/4exp[ztl~189176 '~dt
2n r/2 . ~ \2n!

+0 ~ ]b,ln -a ta-~(logt)"dt . (27)

2 n=l
132 S.M. Gonek

Since b. ~ n~, •
n=l
[ b , [
n - " ~ 1 for a > 1. The error term is therefore

T ~ ~ (log T) m. (28)

To treat the main term on the right-hand side of (27) we write it as

n=l 2

the inversion of summation and integration being justified by absolute con-

vergence. Now the integral in (29) is of the form estimable by Lemma 3 with A
T
= ~ , B=T, and r=2nn. Thus (29) is equal to

~" b,(logn)"+ log~- b,n-aE 2nn,-- T

T/4n<n<=T/21z n= 1 2'

for large T. By Lemma 4 the second term is

<~T"-89 T) m.

Hence (29) is equal to

E
T/4n<n<-_ T/2n
b,(log n)" + O (T"- ~(log T)").

Using this and (28) in (27) we obtain

1 bnn -a-it Z(1-a-it) log dt

2n T/2 n 1
= ~ b.(log n)" + O ( T a- ~(log T)") (30)
T/4n<n<=T/2~

T
for T > To, say. Now let l be the unique integer such that TO< ~ - < 2 To. Adding

~o~et~er the r~sult o~,~0)~or~.e ranges I;' ~ ] ~--~'""~' wefind that

1 i(~bnH_a_it))~(l__a__i~)(log~)md~
2 ~ T/21 n= 1
= ~ b,(logn)m+O(T~-~(logT)").
T/2 ~+ ln <n~ T/2 n

Noting that

( )
1 v~2' ~ b,,n-"-a Z ( 1 - a - i O (~)m
log~- dt~l
2~ 1 n=l
Mean values of the Riemann zeta-function and its derivatives 133

and
b, (log n)m~ 1,
l <=n<=T/21+ln

we obtain the result.

Lemma 6. For a fixed, v>=O, and It]>l we have

Z ( ' ) ( I - s ) = x ( 1 - s ) ( - l o g 2rc
\
Itl ]/ ~+O(Itl"-~(l~ (31)

Proof We proceed by induction on v. The case v = 0 is obviously true. Now

suppose the lemma proved for v = 0 ..... # - 1 . We differentiate the identity

x'(1 - s) = z ( l - s ) - - (1 - s )
Z
and obtain
(32)
v=O

We have

X
and by (15) and Cauchy's estimate for the derivatives of an analytic function
applied to a small disc centered at s, we find that

-(l-s)= -
Z

\ I
Also
g(1-s)=O(ltl ~-~) for [tl>l. (35)

The required result now follows from (32)-(35) and the induction hypothesis,
oo

Lemma 7. Let ~(")(s) ~(V)(s)=,~= 1 A.(#,n~ v___~)

(Re s > l), where #, v _>-0. Then for x _>1,

#!v!
~, A,(#,v)=(-1)u+v(#+v+l)! x(log x)U+v+ 1 +O(x(logx)U+~)" (40)
n<x

Proof Lemma 7 is a simple exercise but we give a proof for completeness. We

have

( - 1)"+~ ~ A,(#, v ) - ~ (log d)U(log r) v

n<__x dr_-<x
=( ~ ~ + Z ~ - ~ ~ )(l~176 ~"
d<=VTr<=x/a r<-_CYa<=x/r a_~ffYr<=l,~
134 S.M. G o n e k

Since
(log d)" = z(log z) a + 0 (z(log z)"- 1),
d_-<z
our sum is
a<=
~v~ (l~ x ~ (log r) ~rX log +O(x(logx) "+~).

N o w we can replace the last two sums by integrals, again introducing a

remainder term O(x(logx)"+~), and we have to deal with

x ~ (logt)" log ~ (logt) ~ log --

1 t 1 t

! xx~dt
: (log (log 7
1

If we make the change of variable t = exp (u log x), we see that

1
~ dt x)U+ ~+ 1
i (logt)" log -7-=(lo8 S u"(1 -u)Vdu
1 0

#!v! (logx).+~+ 1
-(#+v+l)!

by the well known beta integral, and our lemma follows.

L e m m a 8 . Let ((")(s)((~)(s) (s-iS)= B.(#,v,6)

nS (a > 1), where #, v >=0 and (5
n=l
is a real number. Then for x > 1
(icS log x) l
B.(#, v, 5)=(-1)u+~+ l #Iv!x(logx) "+~+ 2
.-<_x t=o(l+#+v+2)!
+ O(x(log x) u+ ~+ 1).
co A.(#, Y) ~( ~ A(n)
Proof. We write ((")(s)~(~)(s)= - - as in L e m m a 7 and s) = - ns
n= 1 ns n=l

Then
B.(#, v, 6)= - ~ ~, A(d)di~A./a(#, v)
n<=x n<xaln
=-~,, A(d)d ia ~ A,(#,v).
d<x e <=x/d

Using L e m m a 7 to estimate the inner sum, we find that

y. nn(#,v,,5)=(_l).+~+l #!v! A(d) / x \ ~'+~+l

,,<=~, (#+ v+ l)! X d ~ x ~ z - ~ tl~
M e a n values of the R i e m a n n zeta-function and its derivatives 135

or

.=<~B.(/~, v, ~) = ( _ 1)u+~+ 1 (#+#!v!

v+ 1)! xL.+,,+ I(X, 6)+O(xLu+.(x, 0)), (36)

where
L,,(x,~)=a~(lOgd)".

To estimate L~ let O(x)= ~ A(d) and write

d<=x

L~(x, 6 ) - 1 u-~-zT~dO(u).

By the prime number theorem with remainder, O(x)=x+E(x), where

E ( x ) ~ x e x p ( - c l ~ x ) f o r s o m e f i x e d c > O a n d x = > l . Thus

L,,(x, (5) - ul -i~ 1

- bfl- -i,5 dE(u).

The second integral on the right is

=~l,~logx~+ i E~u, (~ (,og~)~ 1+~1_i~t (log~)~)du

(log x) ~ + (log x) ~ i exp ( - c lo]/i~) du
1 U
(log x) ~.

In the first integral we replace logu by ( - 1)Z(log x) ~-z(log u) ~, u i~ by

~ (i6 log u)t and change the order of summation and integration to obtain
l=o l!

i l~ du ~ ~i~tz~ (~) ~-l~logx~-~logu/~+zdu

x u
1 =

t~o_ l! (l~ ~z=0 /+2+1"

The innermost sum equals
l!~c!
xl(1-x)"dX-(l+~+l)!,
0
136 S.M. Gonek

SO the entire expression is equal to

(i~ log x) l
x!(l~ ~ (l+~c+l)!"
l=0
This gives

L~ (x, 6) = ~c! (log x) ~+ 1 ~ (i~ log x) l t- O ((log x)K).

l=o ( / + ~ c + l ) !
Using this in (36) we easily find that

~" B~(#, v, c~)=(-1)"+~+luTvlx(lo~x~ "+~+z ~ (i6 logx) t

,=<x ~" " " '~ " ,~o(l~+v+-2)!
+ O (x(log x) "+ ~+ 1).

This proves the lemma.

L e m m a 9. Suppose that for a fixed 2 > 1,

a.=x(logx)~+O(x(logx) a-1 ) (x > 2).

n<x

Then if • > 1 is fixed,

a,(logn)~=x(logx)~+a+O(x(logx) ~+~-1 ) (x >2).

n<x

Proof Trivial, by partial summation.

L e m m a 10. Let 2, v be integers with 2 > 1 and v > O. Then

~=o ().+ ~)! (v+).)(2-1)!

Proof The sum equals

= 9 (_1)~
,,= o ( - 1 ) " (v -- ~c)! (2 + K) T. (v+2)! ,,= o \v--lc!

(v+2)! ~=o
The last sum is the coefficient of x ~ in ( 1 - x ) ~ + ~ ( 1 - x ) -1 and is therefore equal
to the coefficient o f x ~ in (1-x)~+~-~, i.e. ( - 1 ) ~ ( v + ~~- l ) . -- So the above is

_ v' /v+2-1, (v+2-1), 1

(v+2)l k v ] -(v+2)!(2- 1)!-(v+2)(2-1)!"

w5. Completion of the proof

We are now in a position to estimate the integral Ii(/z, v, 6) and thereby to

complete the proof of the Theorem. By (21) and (23) we have
Mean values of the Riemannzeta-functionand its derivatives 137

~(U)(p+if)~(~(1--p--if)=I~(l~,V,f)+Ix(v,l~,f)+O(Ta-~+~), (37)
l<=7<T

where 1 < a < 2 , [31<89 T~Y-, and

1 a~ir ~,
I1(/~, v, 6 ) = ~ ,+i ~ (s)~j")(s+i6)~(~(1-s-i6)ds.

A simple change of variable gives

i1(1~,v, 6) - l a+i~+~)
.+.r+a)~g, (s -i6)~W)(s)~v)(1 -s)ds.
21ri
Now for a fixed a > l , the integrand is bounded over the interval [a+i, a+i(1
+3)]. Also, by (19) and (20), the part of the integral along [a+iT, a+i(T+6)]
is
<{log T. T ~/3. T "-~+~/3 ,~ T "-~+~.
Thus
1 a+iT t~,
I1(/~, v, 6 ) = ~ ,!i ~ (s-i6)~w'(s)~)(1 -s)ds+O(T"-~+~)"

Taking the vth derivative of (9) according to Leibniz's rule, we find that

~'~'(1-s)= ~ (;)(-1)~)(s)z~-~)(1-s).
Hence
I~(tt, v, 6)= ~ (;) (-1)~ll~(#, v, 6)+O(T~-~+~), (38)
K=O
where
] a + i T ~r

By Lemma 6, (19), and (20) it is not difficult to see that

1~(/~, v, 6) = ( - 2rt
1)v-~ i - ~ (a+it-i6)~W)(a+it)~)(a+it)

9)~(1-a--it) log 2re ] dt+O(T"-~+~).

By (15) and (17)

_~'
~(a+it-i6) =~(a+it_i6)+_~log~+7+O
, t rci (~)

for t > 1, say. Hence

138 S.M. Gonek

- 2~z 1 ~ (a+it-i(5)((")(a+it)((~)(a+it)Z(1-a-it) l~ dt

-F (--1)~-~ r~89 log~/

t ]v_~+, dt
2x 1
( - 1)~-~ ~ rci / t X~-~
4 27z 1 --4 ((")(a+it)~(~)(a+it)x(1 -a--it) (log ~ ) dt
T

+o(r"-~+~).
The next-to-last error term is O(T "-89 by (13) and (20) so we may write

I1 ~(#, v, 3 ) = ( - - 1)v-~(It~ 1 +II~2+I1K3)+O(T"-i+~ ). (39)

To treat I~ ~1 write

(s- i3)(t")(s)((~)(s)= ,= t B.(#, s~, (a>l)

as in Lemma 8. Then
1 T( ) / t \v-K
!
Since the B~'s are easily seen to be ~ n ~ for any e > 0, we have by Lemma 5 that
for T sufficiently large

I1~1 = ~ Bjt~,tc, 3)(logn)~-~+o(r"-~(logT)~-~).

n< T/2r~
It now follows from Lemmas 8 and 9 that

i3 log ~-~ T T ,+~+2

+ O (T (log r)" +~ + ~) + O (T"- 89(log T) ~- ~) (40)

for all large T.

Next, writing
((.)(s)((~)(s) = ~ A.(/I, ~c) (a> 1)
n=l ns
as in Lemma 7, we have
( ) / t \~-~+~
1 i 89 ~ A.(#,~c)n-"-" X ( 1 - a - i t ) t l o g ~ x )
11K2=~-~ 1 n= 1
Mean values of the Riemann zeta-function and its derivatives 139

The A.'s are ,~ n~ for any e > 0, hence by Lemma 5

I1~2-~
1
~ A.(#,~c)(logn)V-~+l+O(Y"-~(logT)~-~+1)
n<=T/2rr

for sufficiently large T. By Lemmas 7 and 9 we then find that

__ . . { V ~#+v+2
11K2 (-1)u+~#T~ T T log +O(T(logT)
"+v+l)
2(#+K+l)! 2~ \ ~1
+o(r"-89 (41)
The treatment of I1~ 3 is analogous to that of Ix~ 2 and leads to
(-1)"+~p!~c!r~i r ( r~u+~+l
IlK3 = (#+K+I)! 4 2~ . l ~ +O(T(logT)"§
q- O ( T a - 89 T) v - r), (42)
for all large T.
Combining (39)-(42), we see that

T ( T ]u+v+2 / 1
( T) /
its log
Ii~(#,v,a)=(-1)"+~#!~c!~ log ~- ! 2(p+~c+ 1)!
l=0
+ O(T(log T)u+~+1)+o(r~-~+").
Hence, by (38) and Lemma 10,

I1(#, v, 6) = ( - I)"+ v#! ~ - klog ~ ) (-1)~

(:t ( p + ~ + 1)!
- ialog~ (-1)~ (I+/x+K+2)(
1=0 ~c=O

+ O (V(log T) "+ v+ 1) .oFO(T.-}+ ~)

T !/log T ],+~+2 1 )~~176 #!i61og

= ( - 1 ) " + ~ - ~ \ ~-~I 2(#+v+l) ,=o(l+#+l)!(l+l~+v+2)
+ 0 (g(log T)"+ ~+ 1) + 0 (T"- ~+ 9.

It follows from this and (37) (with a= 88 say) that

T / T \~.+v+2
l=_<~__<T
9 {/~+1
v+l H(#,v, 61ogf---~)-H(v,#,-61og~--~)}
+ O ( T 0 o g T) "+ v+ t),
140 S.M. G o n e k

where T is a sufficiently large element of ~--, 161<89 and

(ic) l
HOt, V, r
/~!~__L'
ox~( l + y + 1)! ( l + y + v+2)'

Taking 8 = e L -1, where L = l o g ~ - and c~ is a real number satisfying

I~1 <L/2, we obtain
~ ~(,)(o+ic~L_l)~(~)(l_p_ic~L_~)=(_l),+~ r log

{1
" g+v+l H(y,v, 2 n ~ ) - H ( v , # , - 2 n e )
) +o(r(logT)U+~+~). (43)

This is clearly equivalent to (4) when TsJ-. To remove the restriction on T

note that increasing T by a bounded amount introduces O(log T) terms into
the sum in (43), and by (20) these are no larger than O(T~+~). Moreover, the
right-hand side of (43) changes by at most O((logT)"+~+2). Since these errors
are smaller than the O-term in (43), (43) holds for all large T within O(1) of an
element of ~--, that is, for all large T. The proof of the Theorem is now
complete.

w6. Proof of corollary 3

Assume the Riemann hypothesis is true and let 0<71~72~... denote the
ordinates of the zeros of ~(s) in the upper half-plane. Integrating both sides of
(7) with respect to 9 over the interval [ - f l / 2 , fl/2], we have
yn+ fl/2L
E I l~(89+ it)]2tit ~ F (fl) Tlog
1 = < 7 . < T 7n--fl/2L
T.
where
a/2 (sinn~]~
F(fl)= ~ 1 - d~.
-/7/2 \ 7~ /
Now if we choose
(7, - 7.- 0 2n
fl > 2 = lim sup
. log 7.

it is clear that the left-hand side above will be greater than

T
]~(89+ it)[Z dt,
1

which is ~ T l o g T by (1); that is, F(fl)>l. But a machine calculation shows

that F(1.9)=0.997 .... Hence 2 > 1.9.
The same argument could of course be based on a comparison of (3) and
(6) with # > 0 . But as y increases this seems to lead to progressively worse
lower bounds for 2.
Mean values of the Riemann zeta-function and its derivatives 141

References

1. Davenport, H.: Multiplicative number theory. 2nd edition. Berlin-Heidelberg-NewYork:

Springer 1980
2. Hardy, G.H., Littlewood, J.E.: Contributions to the theory of the Riemann zeta-function and
the theory of the distribution of primes. Acta Math. 41, 119-196 (1918)
3. Ingham, A.E.: Mean-value theorems in the theory of the Riemann zeta-function. Proc. London
Math. Soc. (2) 27, 273-300 (1928)
4. Levinson, N.: More than one third of the zeros of Riemann's zeta-function are on a= 89
Advances in Math. 13, 383-436 (1974)
5. Montgomery, H.L.: The pair correlation of zeros of the zeta function. Proc. Symp. Pure Math.
A.M.S. Providence 24, 181-193 (1973)
6. Mueller, J.: Gaps between consecutive zeta zeros. Analytic number theory. Lecture Notes in
Math., vol. 899. Berlin-Heidelberg-New York: Springer 1981
7. Mueller, J.: On the difference between consecutive zeros of the Riemann zeta function. J. of
Number Theory 14, 327-331 (1982)
8. Selberg, A.: The zeta-function and the Riemann hypothesis. Skandinaviske Mathematikerkon-
gres 10, 187-200 (1946)
9. Titchmarsh, E.C.: The theory of the Riemann zeta-function. Oxford: Oxford University Press
1951
10. Whittaker, E.T., Watson, G.: A course of modern analysis. Cambridge: Cambridge University
Press 1969

Oblatum 11-IX-1982 & 29-IX-1983