Functional Equations With Multiple Gamma Factors and the Average Order of

Arithmetical Functions
Author(s): K. Chandrasekharan and Raghavan Narasimhan
Source: Annals of Mathematics , Jul., 1962, Second Series, Vol. 76, No. 1 (Jul., 1962), pp.
93-136
Published by: Mathematics Department, Princeton University

Stable URL: https://www.jstor.org/stable/1970267

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to Annals of Mathematics

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
ANNALS OF NIATHEMATICS

Vol. 76, No. 1, July, 1962

Printed in Japan

FUNCTIONAL EQUATIONS WITH MULTIPLE GAMMA FACTORS

AND THE AVERAGE ORDER OF ARITHMETICAL FUNCTIONS

BY K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

(Received October 11, 1961)

1. Introduction

The determination of the exact order of magnitude of a given arithmeti-

cal function is, in general, a difficult problem. Striking instances are pro-
vided by the problem of lattice points of a circle, or of a rectangular hy-
perbola, and by Ramanujan's function r(n) [5]. If r2(n) denotes the num-
ber of representations of n as a sum of two squares, and

R(x) = Ex o r2(n) = wrx + P(x),

then, as is well known, it is an exceedingly difficult problem to determine
0, the smallest value of 5 such that

P(X) = O(x +8)

for every s > 0. Various upper bounds for 0 have been found, beginning
with Sierpinski's result that 0 ? 1/3. The method of exponential sums
gives better estimates, for example 0 ? 13/40. On the other hand an
early result of Hardy, and of Landau, gives 0 > 1/4 [6]. Actually it is
known, after Hardy [6] and Ingham [9], that

(1.1) lim supsM. P(x) + +co lim inf P(x) =-co.

A similar situation prevails in the case of divisor problems [14]. Our object
is to consider estimates of this type for a general class of arithmetical
functions. We are led to such a class by the observation that the gener-
ating function of r2(n) is Epstein's zeta-function ,2(s), which has the repr
sentation

U2(S) = n r2(n) Res > 1,

n=1n

and satisfies the functional equation

(1.2) wSsr)T (s) = s:f(1 - S)l2(1 - 8)

We start here with a general functional equation of the form

A(s)p(s) - A(8 -SM8 s),

where 8 is real, /\(s) = F f(acs + d), ra, is positive, a, > 1, and
93

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
94 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

/S is complex. We consider the solutions of equation (1.2) in Dirichlet

series, namely

p(S) = Ad an , Re s > y anacomplex,

K

*(s) = E ,'In bn y Re s > I, bn complex,

and the averages of their coefficients, namely

AP (x) 1_ r a~(
AA(X) -I(p LAx n(X iXn
BP(x) _ 1 + b P _O.
M F(p + 5i)Vn9 ( J
and prove two main theorems on th
may be). The first theorem, which we
the functional equation is satisfied,
of n, then

Re [AP(x) - Qp(x)] = n+(xO), p > 0

where 0 = (1/2A)[A8 + p(2A - 1) - 1/2]. A similar result holds for the
imaginary part. The function Qp(x) arises from the singularities of 9(s)
and of F(s), and can be explicitly written down in any given instance of
the functional equation. The exponent 0 is the best possible for p large
enough. Under further assumptions on (bn, "na p), it can be proved that

lim sups~. Re [AP(x) - Qp(x)] +

lim infg. Re [AA(x) - Qp(x)]

Xe

in analogy with (1.1). A special case of this f2-theorem in which /\(s) =

F(s) has been considered by us in a recent paper [3]. We give here also
a second general theorem, which we call an 0-theorem, in which we esti-
mate the order of A' (x) - Q(x). The proof rests on an identity of the
form

(1.3) AA(x) - Qp(x) = LnO bnIP(PnX)n y

where

4() 1 -~O ( s)A(s) XS?Ps d

2I X i C-i r(P + 8 + 1-S)A(8 - S)
which is obtained as a consequence of the general functional equation,
and on estimates for Ip(x) together with its derivatives, or differences,

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 95

the case may be. When the gamma factor A(s) is of a special type, as
for instance A(s) F F(s), F2(s/2), or F2((s + 1)/2), one can explicitly evaluate
I,(x) in terms of Bessel functions. The resulting series on the right-han
side of (1.3) is absolutely convergent for large p, but non-absolutely con-
vergent for smaller values; and the method of equi-convergent trigono-
metric integrals, which we recently applied [2] to the study of such identi-
ties, gives us precise results.
Among the solutions of the general functional equation are the zeta-
functions, and L-series, of algebraic number fields, and our results there-
fore apply to the character-sums associated with such functions. Our
results seem to yield more than what is known so far about such sums.
In the classical lattice-point problems and divisor problems, our 0-results
give less than what has been obtained by special methods, although even
there, our estimates are decidedly non-trivial. Our f2-results, even in the
classical cases, seem generally to be as strong as any so far known.

2. Definitions and Formulas

We begin with the definition of a general functional equation.

DEFINITION 2.1. Let {an}, {bn} be two sequences of complex numbers
not all the terms of which are zero. Let {Xv}, {Jn} be two sequences of
(strictly) positive numbers, strictly increasing to infinity. Let 8 be a real
number, and s a complex number with s = a + it. Let

(2.1) '\(s) ==J= rF(a?s + 8S,)

where N ? 1, 8, is an arbitrary complex n

E a=1 > I. We say that the functional equa
(2.2) A(s)p(s)- A(-(8 s)'(8 - S)
holds, if q and V can be represented by the

(P(S) = E i(S) = *bn

each of which converges absolutely in some half-plane, and there exists

in the s-plane a domain ?D, which is the exterior of a bounded, closed set
S, in which there exists a holomorphic function X with the property

(2.3) limjt1"oX(a + it) = 0.

uniformly in every interval - co < al ? a ?2
X(S) = A(s)9(s), for a > ao,
X(8) =: /(8 - M8) - 8), for a < K ,

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
96 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

where a and S are some constants.

Condition (2.3) could be replaced by an apparently weaker condition,
such as I x(a + it) = O(el'4t' ) uniformly in al _< a _ 2, but that is
little interest to us here.
We shall require several formulas involving the Bessel functions Ju, Yo,
and the modified Bessel functions I,, K,. We have the expansions

(2.4) J2(z) = ;: o (_1)m(z/2)2m1'f+[F(m + 1)F(m + 2 + 1)]f' [17, p. 40].

(2.5) 2I(z) = exp (-ivi/2) J2(z exp (iw/2))

= E=o (z/2)2-+',j[v(m + l)F(m + V + 1)]-l [17, p. 77].

If n is a positive integer, we also have

w Yn(z) = 2 +log - In-i ( Z ) F(- m)

L2 o JM() 2~ F(m + 1)
(2.6) zn+2mn
(265=o [(l)(Z) (h +n+
[17, pp.
wheren= 1,2,3, *.,m = 1
and Y is Euler's constant. If v is not an integer

(2.7) Y,(z) = [sin (vz)]-4[J2(z) cos (w) - J-,(z)] [17, p. 64].

If n is a positive integer, we have

Kn(Z) = (-1)n+lIn(z)log Iz + I jn- (_1)mnkZm F(n

\2/ 2 \1O 2/ F(m + i)
I 21n+n ]r
(2.8) + (1 )n =0 Z) (n + m +1)

+ -(m + 1) [F(m + 1)r(n + m + 1)L1 [17, p. 80],

F

and for v not an integer,

(2.9) KV(z) = Lw[ sin (vz)] j[L_,(z) - I,(z)] [17, p. 78].

2

We also have the asymptotic expansions

(2.10) K2(x) = ( 1) e-[i + 0 [17, p. 202],

(2.11) Y2(x)Q =( ) [Sin (x- z - 1 z) + O( 1)] [17, p. 199],

(2.12) J2(x) = ( /2 [Cos (x - -1 V- -!7w)O +(x)] [17, p. 199],

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 97

as x -+ co, valid for all complex v.

For any constant a, we have

(2.13) log F(z + a) = z + a - - )log z - z + 1 log 27c + 0

[18, p. 251],

as I z c x, uniformly for -w + 8 ? arg z <_ -8y 8 > 0. We frequent-

ly use the estimate

as I y co , uniformly in - cox X1 < x x2 < + co. From standard re-

(2.14) f U(x + iY) exp (- ? y) y 1 x12(2wr)l2

sults on Laplace transforms, we have

(2.15) e-StV1-l sin (at)dt - f(>) [(s - ia)-v - (s + ia)-v] ,

Jo 21

for Rev > -1, and Res > IIm a , and

(2.16) e-sttv1 cos (at) dt - F(>) [(s + ia)va + (s - ia)a]

for Re v > 0, and Re s > I Ima 1.

1 c+ 2s-a-lr(s/2) x-sds J-(X)
(2.17) 2W c-i- F(2 - 8/2 + 1) Xv
0 < c ? Re (v + 1), Re v > 0
[17, p. 192].
For a > 0, and k > -1, we have

1 __ - t)a-ltk/2 Yk{(t~af)"2}dt

(2.18) = a- 2J /2x(k+a)/2 Yk?I{(xIaJ)1/2} + F(k + 1) 2k?2x(k?~)I7Jy I2

n Yk +a I~~F(a)wuP
X Soa-k-y Xt+ k {(X
[Tables of Integral Transforms, vol. 2, p. 196],

where S stands for Lommel's function.

Lommel's function has the properties

(2.19) / d SX k k k+,xX()] = 2(a - l)x+ Sak2, (+k1(X)

[17, p. 348]

(2.20) S-k-l, ca+k(X) = O(I x jc-k-2) [17, p. 351],

uniformly for Re x > 8 > 0.

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
98 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

3. The main f2-theorem

For p > 0, let

(3.1) AA(X)-r( S ) ,n~XA")(x) 1

the dash denoting that the last term has to be multiplied by 1/2 if p = 0
and x = >~ Let F(Po + i) Exn= a(
and x = xn. Let

(3.2) Q 2x 1 x s+Pds P + 0
2W'i iCp F(s +p + i)
where p(s) =X aX s as in Definition 2.1, and Cp encloses all the singu-
larities of the integrand to the right of a = -p - 1 - k, where k is such
that k > 1(8/2) - (1/4A) 1, and all the singularities of q lie in a > -k.
Note that if p is integral, Cp encloses all the singularities of the integrand.
Our object, in this section, is to prove an 72-theorem for AP (x) - Qp(x).
The proof uses an idea of Ingham [9].

THEOREM 3.1. Suppose that the equation

A(s)cp(s) = A(8- s)(8 - s)

is satisfied as in Definition 2.1, and that {1a"} contains a subset {1an}

such that no number ,4/(2A) is representable as a linear combination of
the numbers 'l/(2A) with coefficients +1, unless '"1/(2A) = "1/(2A) for som
r, in which case '"I, (2Al has no other representation. Suppose, in addi-
tion, that

(3.3) ,?? Rebnk = +

n1(A8+p+l/2)/2A

Then we have

(3.4) lim supX~. Re [APXx) - Qp(x)] +

(3.5) 1iminfX~. Re [AP (x) - Q(x)]

Xe

where

(3.6) 0 =_ A8 + p(2A -1) -1/2

2A

If in assumption (3.3) we replace Re bnk by Im bnkl then in conclusions (3.4

and (3.5) we should have Im [Ag(x) - Q (x)] in place of Re [Ag(x) - Qp(x
THEOREM 3.2. Suppose that the equation

A(s)(P(s) - zA(8 - s)*(8 - s)

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 99

is satisfied. If Re be # 0 for at least one value of n, then

(3.7) Re [AA(x) - Qp(x)] = f2?(x0), p ? 0 s

and if Im by # 0 for at least one value of n, then

(3.8) Im [AP(x) - Qp(x)] = ,2+(xl) p ? 0 9

with the same Qp(x) as in Theorem 3.1. The exponent 0 is the "best pos-
sible" for p large enough, in the sense that if p > 2AS - AS - 1/2, where
,8 is such that En 1 b. p ,P < c? , then
(3.9) AP(x) - Qp(x) = O(xo).

PROOF OF THEOREMS 3.1 AND 3.2. Given p 0, define

(3.10) ap= p- AA
2A

Let vp be the integer defined by the inequalities: ap _ vp < 1 + ap. Then,

either vp > ap + (1/2A), or vp < ap + (1/2A) = ap+1, in which case vP+1 >
ap+1 + (1/2A), for then vp+l = 1 + vp ? 1 + ap ? (1/A) + ap = (1/2A) +
ap+,. Thus, given p _ 0, we can choose an integer m which is 'arbitrari
large', and which is such that

(3.11) 2P+sm >- ap+m + (1/2A).

If a and 8 are such that a, I X-; < co, and E

such that c > 0, c > a, c > A, then by a standard formula, we have

(3.12) AP+m(x) = 1 Fr(s)p(s) xs+P+rnds

2WiT c-i- F(s + p + 1 + m)

By a familiar argument (see for instance [2]), we shift the line of inte-
gration from a = c to a = 8- c, change the variable s -+ - s, and apply
the functional equation, to obtain

AP+in(x) = S:c+m(X)

(3.13) +1 C+i(0 F - s)(s)IMs) xS+P+m-sds

22r ic- '(p + m + 1 + 8 -s)A(8-s)
where

(3.14) SPC+m(X)= 1 F (s)cp(s) xs+P+Mnds

2Wi CP+m F(s + P + 1 + m)

CP+m being a curve which encloses all the singularities of the integrand in
the strip 8 - c < a < c, in which we may assume all the singularities of
cp(s) to be located. These operations are justified if c and m are large
enough. For the convergence of the integral in (3.13), it is sufficient, in

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
100 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

view of (2.14), that

<~pm AS +p +m
2A

Let

(3.15) CP+m = + p + < <_.

2A 4A

Since m is at our disposal, this choice of c ensures that Cp+m encloses all
the singularities of the function (F(s)/I(s + p + 1)) cp(s)xS+P to the
of a = -p - 1 - k.

Let

f(x) . Ci G(s)x'+P+-sds
where

G(s) =(5 - s)A(s)

F(p + m + 1 + 8 - s)A(8 - s)

Because of (2.13), we have

log Fr(8-s) = -(p + m + 1)log(-s) + O(ls-

F(p + m+ 1 +8cS-s) \jsj
and

log rF(as + S,)

r(a,8 - a~s + )

= (a s + ,- I ) log s - -o 8 -as + - log (-s)

+ (log aor)(2as - a,8) - 2a.s + O

so that

(3.16) log G(s) - log F(As + ) es}= B +0

where

=1/2 + - 1 (S- 1/2),

(3.17) X\=,u+A'+[p+m+1,
(= 2{ f1faoga,- AlogA},
B = -8 Ev a. log av + (A8 + p + m + 1) log A.

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 101

Hence

f(x) 5 G()X+P+m-sds
(3.18) 2 C+
= 5 H(s)x8?+P+-sds + {G(s) - H(s)}x1+1+ff--ds
2 ri c-ioo 2Wri c-i_

where

H(s) rF(As + a) eB+?s

F1(X - As)
and

G(s) -H(s) = H(s) * (11)

so that

{G(s) - H(s)}xs+P+m-sds
27ri c-iwO
1 r +(1/2A) +i/o 1 \

2W7ri c+(1/2A)-iw H(s) I + R

where R = O(xP+m-VPd-Am) is the residue of (G(s) -H(s))x8+P+m-s at a possi
pole at 5 = 8 ? 2p+rn Because of (3.11), p)p+m _ ((p + m) - AS + 1)/2A,
and we have

(3.19) R = O(x fA8+ (2A-1) (p+ m)-1Il /2A)

while
1 (c+ (1/2A)+i' 1 N
. H(s) * t J*X8+p+m-sd = (8+- c- (1/2A))
27Ui C+?(1/2A)-i o( ds | |(x8?P-m
(3 . 20) = O(x[AS+ (2A-1) (p +m)1-3/41/2A)

On the other hand, we can explicitly evaluate the integral

L C~i0H(s)xsl P+m Sds = - . e B0( Ci+P+m) |i 1F(As + i) (xe-('))s I-P+m

27ri c-i- 2i 3 c-i- r(X - As)

A1 5 F r(z + j) 8+p+m-(z/A)
A, 2Zi AC-i-r(X - Z) Y Adz As zxe y.
I Acq--tj+ io rPts 8++m (iA-s
(3.21) = A1 |s +P+m+ (1A) -(SIA)ds, Z + f= s
27r, JAc+lj,-i-F(X + t - s)

- A1. jyI-?p+m+(ji/A) . 1 5 i F(s) (?J )s

217r? AC41-i' 0+ P - s)
- A1 yl?+p+m l-(rQ/A) .J2tk+AS+P+m(2Y) [cf. 2. 17]
(y1/2A)2ji1A+-A8 1-P - 2m

- A1. (Y1/2A)AS+(2A-1)(P-m)Jh,?AS +P+m(2yl/2A) , where A1 > 0 .

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
102 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

On putting together (3.18)-(3.21), we obtain

(3.22) f(x) = A2. (Xl/2A)A +(2A-1) (p+m) J2p+AS+P+m(hxlI2A

+ O[(xlI2A)A84 (2A-1)(p+m)-3/4]

where h = 2e-12A. If we set

(3.23) o = AA + (2A-l)(p + m),

we have, on using (3.22) together with (2.12) in (3.13),

AP+m(x) - SP+m(X)

(3.24) A3. bn=l b [(Xjn)1/2A]caI/2 COS (hxl/(2A)/l/(2A) + D)

+ O(X(1/2A) (o-3/4))

Let r be an integer sufficiently large, depending on m and large when

compared to m. Replacing x by X2A in (3.24), and multiplying it through-
out by

(3.25) a'/ = r + AA + (2A - 1)p + -

and integrating it relative to x from 0 to co, we find that the left-hand

side gives

UY d i) (e-sxXr)[APm(x2A) - Sp+m(X2)]dx

(3.26) = (-1)maTY e exr(>) d [AP+m(X2A) - Scm(X2A )]dx

00x

= (2A)m(-1)aY e-sxXr[AP(X2A) _ S;(X2A)]dx

since AA(x) = O(x0), where C = C(p) is a suitable constant. On the other

hand, since

(--) (e- sxr) = (o~r-2AM?+m e, 82Am-r+vXv )e-sx

e X2A-1)m n t r -Am e of(3.24),

we have from the 0-term on the right-hand side of (3.24),

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 103

a X?7( 4ix xi)rn(e-sxxr) O(x

= q Ouy . s2Am-r+v e-sxxv

(3.27) r -2Am ? ? r - 2Am + m,

- .E q'YS2Am-r+[ v 0 ? 14)

If we now consider the first term on the right hand side of (3.24), we ha

S(a) aY (e-sbxr) E' bn l121c2Ax' os (hx/(2 + D)dx

(3.28) = 0 e- 2Ammr+mxv -lI2esx E bn cos (hf

0 71=r-2Am n=1 p5A8+(p+m)+l/2j/2A
o~~~~~~~~~~~~~~~~

= =
Eo bn Ev______
______ e,s8 . "Y| e-SXXV8+c-1/2
a ecos (h~uljA + D)dx CO ^1
n=Al p[A8+(p+m)+l/21!2A no

where er-(2A_1)m = (-l)m.

For 0 < a < 1, we have

1On ?bn) 5,(p+mn)+1/2}/2A

n=l p{A8+ e-sxxv+1-112 cos |J
(h/'12Ajx
n + D)dx
? K. qy-v-o-l/2

where

K = bnjfr(i+ + a +1/2)
p(A8+p+m+1/2)/2A

Since v < r - (2A - 1)m, it is clear that - -o - 1/2 > 0 so that

this term is bounded in 0 < a < 1. Hence, we conclude that

lim">+o S(a)

= (Ab+p+m+1/2)/2A -vlim F 0 aY| esxxv?-1IO cos (h/Ax + D)dx

Set
V(x) = (2 + eix + e-x)/2 = 2 cos2 (x/2),
and
W(x) = - V(h 1aAX - Ok),

where N is a positive integer, and Ok is a real number to be a

chosen. We see that

W(x) = 1 + i-E 1 [exp (ihyak - iOk) + exp (-ihfi x + ijk]

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
104 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

where U(x) is a trigonometric polynomial with exponents

(3.29) h(r1fat + r2/112A + * N + rNllaA)

where rv = 0, or + 1, or -1, and no combination is such that all the r,'s
are zero, or all but one are zero and the remaining one is ?11. By hy-
pothesis, none of the exponents in (3.29) equals hAI2A for any n.
Let us now consider the limit

(3.30) L _ lim_-+ | e-fxxr{Al(x2A)-SP(x2A)} W(x)dx

This requires that we consider (3.26), (3.27) and (3.23) with s = a ih

and take their limits as a +0. It is clear that (3.27) contributes zero.
To consider the contribution of (3.28) we first observe that if

I-5e-Sxx
00 cos (Ex + D)dx,

then

I 2 {e (_E) + e-e (s + 1E) } [cf. (2.15) - (2.16)1

so that

(0, if s a ? iE,
(3.31) lime +0 aP~'I =IP(p + 1)e+iD1/2 if s = a ? iE.
If we set E - h[4!'A s = a ? ihpt4,A, and use (3.31) in (3.28), we see that
the only contribution is from the term corresponding to v- r - 2Am +
m, and hence

(3.32) L = A4*EN bnk -sin (D + t + Ok)

and =1/2 or 0 according as m is even or odd. Let D + =d + id'.

Since W(x) _ 0, we have

lim 0 P(Y| Re [AP(x2l)-SC(x2A)Ie-x W(x)xrdx

(3.33) < lim supx J [ Re A+A )p(x2)} ]

x lim", 67 ) 0 eOxl -1W(x)dx .

We have, however,

limzr 6 e-Gxx-l W(x)dx=1,

17(Y)

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 105

because W(x) is of the form 1 + (1/

(3.30) with (3.32) we obtain

lim supx" Re [AP(x) -S(xA)]

X A + (2A-1)p-1/2

2 A5[? Re bfk sin (d + Ok) cosh d'

- k=l a(AS +P+/2) /2A

- N Im bnk * cos (d + Ok)s

k =1 y (AS +p+1/2) /2A

Here cosh d' ? 1, and sinh d' = 0 o-ily if d' = 0. Now if

we can always choose 0 such that a cos 0 + b sin 0 = (1/
Hence, by proper choice of Ok, we have

lim supx b Re [AP, A) - Sp(xA)]

? A6?3? IJRebn j coshd' + I Imbjk II sinhd'I

p(AS +p+1/2)/2A

Similarly

lim infxO Re XAS

[AA(Xl ) s: (X2A)I
? (2A-1) p-11/2

?-A) ? I Re b j cosh d' + j Im bnk sinh d'I

6 1(At-p+1/2)/2A

Since

O" |Rebnk + 00
n=1 t(AS?+p+1/2)/ (2A)

by hypothesis, we obtain

lim Re [AP(X2A) - S(x2A) = ?

XA8? (2A-1)p-112 -

Since APA(X2A) - Qp(X2A) = AP(X2A) _ Sc(X2A) + S c(X2A) Q (X2A) and

Spc(X2A) _ Q (X2A) = O(XA8+(2A-1)P-112) by the definition of Q,, we have

lim Re [AP(X2A) - QP(x2A) = 00

_AS+ ?(2A-l)p-1/2

and an analogous result holds for the imaginary part, thus provin
rem 3.1. If the subset {,"n k} consists of a single element '"nr for
Re bnr # 0, we get the Q2-conclusions of Theorem 3.2. The proof

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
106 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

0-conclusion is more conveniently handled along with the proof of the

next theorem.

4. The main 0-theorem

Theorem 3.2 contained a statement of an upper estimate for the order

of magnitude of AA(x) - Qp(x) for large p. In this section we give an
estimate for A?(x) - Q,(x).
THEOREM 4.1. Suppose that the functional equation

A(s)P(s) = A(8- s)-8 - s)

is satisfied with 8 > 0, and that the only singularities of the function
q are poles. Then we have

AA(x) - Q0(x) = O(X8/2-1/(4A)+2Aqu)

(4.1) + O(Xq-11 (2A) -n logr-1 x)

for every rY > 0, where x' = x + O(xl1-I1(2A)), q - maximum of the real

parts of the singularities of Ap r the maximum order of a pole with real
part q, and u = 8 -8/2-1/4A, where8 lis such that l b I Pn_ < ??
If, in addition, an > 0, then we have

(4.2) A0(x) _ QO(X) = O(XI/21/(4A)+2Anu) + O(xq-1/(2A)-- logr-lx)

PROOF. If y > 0, and p is an integer, we define the 'pth finite differ-

ence' of the function F(x) as

APYF(x) = EP=O (-1)P-v(",)F(x + vy)

If F has p derivatives,

APF(x) = x dt, + dt2 * * |. . FP)(tp)dt

z tj tP -1

where F(P) is the ptI derivative of F.

Let p be a sufficiently large integer, and c cp = ((AA + p)/2A) -a
0 < e < 1/4A. Then, as in the proof of Theorem 3.1, we have

1 FQ3- sA-s
(4.3) AA(x) - SPf(x) = .s J F( + )AQ3 ) q(5)X?d5
[cf. (3.13)]
where

Sp"(x)= 1N )F(S)P(s) -xs+pds,

2W't JCF(S + p + i)

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 107

C being a curve which encloses all the singularities of the integrand in

the strip 8 - c < a < c, in which we may assume all the singularities of
q(s) located.

If we write

(4.4) 1(X) = 1 Fcp+i (8 - S)A(s) x8+P-sds

2WZ Cp-i- F(rp + 1 + S3 - - s)
and assume, as we may, that c # 8 + n for any integer n, that cp > c',
where c' = max (-Re p/A, -Re 3,/,a), v = 1, 2, *. *, n, then, as before,
we have

(4.5) I(X) = 0(X[A8- (2A-1)P-1/2]/2A) + O(XP-VP)

and we suppose, replacing p by p + 1 if necessary, that the second term

is of lower order than the first. Then (4.3) can be written as

(4.6) AA(x) - Sp((x) bI(PX) WP(X)

say, for suitable p large enough. Now in (4.4) we shift the line of inte-
gration from a = cp to a curve C' consisting of the lines a = c0 + i
I t I > R for a suitable R, together with three sides of the rectang
the vertices c0 - iR, cO + r - iR, cO + r + iR, and c0 + iR. We assume
that r and R are so chosen that all the poles of the integrand in (4.4) to
the left of CP are, in fact, to the left of C'. (This shift is justified by a
familiar argument using Stirling's formula for F.) We can then differ-
entiate I(x) inside the integral and obtain

(4.7) I'P'(X) = 21 5g(s)x8-sds g(s) - - s)A(s)

27ci C, 1~~~PQS + 1 -s)AQS - s)
As in Theorem 3.1 we then have

J( ) .1 g(s)xsd
(4.8) 2Wli C'

=2 2 \sh(s)xl-sds
7Ci C' 2 7C C' + 2 .S[g(s) - h(

where

h(s) F(As + A) eB+?S + A8 + 1

(x - As)
1 N i
22

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
108 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

and

g(s) -h(s) = h(s) * 0

so that

(4 9) - [g(1 )_-_h()]x-sds = i h(s). Oy) * x --3ds

(4.9) 2 7 [s - 27i C'+112A |s|
=O(x 81-/2 )4 O(x 8/- 4),

esince theOnpoint
< 1/4A. -fe/A
the other is we
hand, to have,
the left of the
as before curve ((C' + (1/2A)), and 0 <
1 h(s)xl-sds =A, P(As +? ) x'S-Sds
2 7Zi C' c, F(p + AS + 1 -As)
= Al 1 r(s) x ( sIA)+(YiA)ds
AC4-} r(2p + A* + 1 -
- -Aix1
xa'2J2
2S, t+A (2A)
AS(x' ) t

since all the points s = -n lie to the left of the path of integration, so
that

(4.10) h(s)xl 8ds = O(X(112)-(114A)

On combining (4.8)-(4.10), we get

(4.11) I(p)(X) = O(X(8-2)-(114A)

We now apply the operator AP to the equation (4.6). On the left hand
side we have

AyAP(x) = LA 5$ an A-(x

+ 1 r(p 1i) (-1)Pv( <) :!,,X+,,,a,(X + VY-X,)P

and since {r(p + 1)}-'AP(x - X)P = yP, we obtain

(4.12) AIJAP(x) = AA(x)yP + O(yP !x<Xfx?PYn a)

Again

A Py Sp"()=| dtidt2 . 5 Qo(tp)dtp y

ac tj tp -1

since the pt1 derivative

Q,(x) = 2Wi .|(s) xsds xI logrt-lx

Q.(x) = s~

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 109

where r, is the order of the pole at s

(4.13) Ap&Sc(x) = Q0(x) * yP + O(xq l(logr x)yP?1)
Thus (4.12) and (4.13) give

(4.14) AP[AP(x) - Sc(x)] = yP[AA(x) - Q,(x)] + O(yP lxql log 'x)

+ ?(YP Jx<Xn!z+py I an.)
On the other hand, if an > 0, then A?(x) is monotone, and we have

(4.15) yPAI(x + py) > APAP(x) ? yPA'(x).

Because of (4.5) and (4.11) we have

_ ) (( I(x) I) = O(X-1/2)-(1/4A)+P(1-(1/2A))

APIAx) = O(yp I I~p)(X) () = O(ypX(812)-(114A)

Thus

(4.16) ?pI(ttnx) = O[X(8/2)- (l/4A) (8/2)-(1/4A)+p(l-(1/2A)) min (Xp(1-(1/2A)) pp/2AyP)]

Further, if v _ /3, we have

(4.17) zI bn I = E I b. - O(z V)
while if v < 8,

(4.18) Eunz | e g n | = ZzzI b. I I=bI - O(z-V)

From (4.6) and (4.16) we see that

A W(X) = ((5 /2)+(14A))Y

+ O (X(812)-(1 /4A) +p(1-(1/2A) ) I b.

- (xn>z / A(?/2) +(1/4A) + (p/2A)

If we write

u=S_ 8 _ 1
2 4A

(note that u > 0, cf. Remark 5.1) and use the estimates (4.17)-(4.18), we
have

(4.19) zP WV(X) = O(ypx(1'2)-(1/4A)zU) + O(x '2 1/4A)+p(1-(1/2A))ZU-(p/2A))

On setting
=x -(1/2A) 2A

Z- =

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
110 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

we have

(4.20) p Y4WP(X) = O(yP-2AUX(8/2)-(1/4A)+(2A-1)u)

In view of (4.6) we compare (4.14) and (4.20) and obtain

AO(x) - QO(x) = O(y-2AUX(8/2)-(1/4A)+(2A-l)u) + O(yXq-1 logr-lX)

+ O(x<xn:x+Py I an 1)
Now we choose

y = Xl-(112A)--q with C > 0

so that

(4.21) AOA(x) - QO(X) = O(X(8/2)(1I4A) ) + O(X 2) 71 logrx)

+ 0(1:Z<xn: +P a.
which proves (4.1).
If, in addition, an _ 0, then we combine (4.13) and (4.15) to conclude
that

AO(x) - QO(x) y-PAP W(x) + O(YXq- 1ogr-1X)

= O(x(8/2)(1/4A)?2A-qu) + O(xq(1/2A)- logrix)
? A?(X + PY) - QO(X)
Since

QO(X + py) - QO(X) = O(Xqly logrlX)

= O(Xq (1/2A)-n logr-lx)

we have

(4.22) AO(x) - QO(x) = O(X(8/2) (1/4A)+2A7u) + O(Xq-(1/2A)- log-'x

which proves (4.2).

In view of (4.5) and (4.6), we observe that

AP-4m(x) - Spc+m(X) = b
W 8+P+M 1PIpM(/JnX)
and proceeding as in the proof of (4.11), we obtain

IpM) (,eex) = 1anCapX)[A8+P(2A-1)-1/2]/2A)

and the series

rnoi=e tha+ n ?23 A3 < Hec

provided that p > 2A,8 -A8 -1/2. Hence

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 111

AA(x) - Sp (x) = (lunx)

(4.23)Ju
O(X[A8+ (2A-l)p-(1/

and since Sp(x) - Qp(x) - 0(X

O(XfA8+(2A-1)P-(12)}/2A) fo
The case p < 2AS - AS - 1/2 is covered by the following:

If

no"= 1 bnl = co
?(8/2) + (1/4A) + (p/2A)

then

(4.24) AA(x) - Qp(x) = O(X(8/2)-(1/4A) +p(l-(1/2A))+2A77(U-(p/2A)))

? O(xq+p (1/2A)-77 logr-lx)

? 0(x((12A)-)P / x<AX<x' Ian |)

The idea of using finite differences in this fashion is due to Landau.

5. Some remarks on Theorems 3.1, 3.2, and 4.1

(5.1). We observe that the functional equation implies that a > 8/2 +
1/4A. For identity (4.23) cannot hold for p = 0 since the left hand side
is then discontinuous while the right hand side is not.

(5.2). If an 2 0 for all large n, and p(s) is an entire function, then g

cannot satisfy the functional equation. For then Q0(x) = p(O) = L'
cc, and AA(x) - Qo(x) = >-E x an, and this cannot change sign infin
often.

(5.3). If, instead of functional equation (2.2), we have the more gener-
al equation A,(s)p(s) = A2(8 - )*(8- s), where
A1(S) = IIN Ir(as + lvy), A2(S) = I r( Fas + f8')
with ReS,, - Re,8', then the analysis of our previous theorems goes
through with few changes. If La ' = ,1 (fl - 1/2) + 1/2, then B -- B'
B + a - A, and J2tk+A?+P+m Jt+k'+A?pP+m -

(5.4). We have assumed that A > 1. Analogous theorems could clearly

be proved if we had assumed only that a,, > 0.
(5.5). In theorem 4.1, if we assume that

(5.6) E Jn< I bn I= O(xa logblx)

then in estimates (4.21) and (4.22) we can replace Xu by E(x) logb-lx,

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
112 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

where E(x) = max (Xa-(8I2) -l4A) log x). If be _ 0, then the analogue of
the estimate (4.22) for B,,(x) gives us (5.6), with a = the abscissa of co
vergence of E boseus, and b = maximum order of a pole of p on a =
which exists since b >? 0.

6. Functional equation (2.2) with A < 0.

The above considerations lead us to ask if the functional equation (2.2)

can hold with A < 0. We observe that it cannot, for, as before, for a
sufficiently large c, with c # 8 + n for any integer n, we have

AA(x) - Sc(x) = i("nX)

where

1(X) - I(X) = 1 F8 - 8)A(8) x Psds

2Wi J c-i- F(W + P +1-s)(i-)
and p is a large enough integer. Since A < 0 we can shift the line of
integration from a = c to a = c + M, where M is large compared to p,
and obtain

AA(x) - Sc(x) = IC+?l[(fnX) + En<c?-s anxP

and since Ic1P) (x) = O(x?+P-c`P), we can differentiate the above identity,
and obtain a contradiction from the fact that A?(x) is discontinuous.
One proves in the same way that the functional equation cannot be
satisfied by a function p all of whose singularities lie in a horizontal str
A corollary to this remark is that, if the functional equation is satisfied
by p, *, with a, > 0, and if 1/R, 1/* can be represented by absolutely
convergent Dirichlet series in some halfplane, then p, r have infinitely
many non-real zeros. It is clear that such zeros must lie in a vertical strip.
Our proof of the main f2-theorem was based on identity (4.23). Setting
up a corresponding identity when A < 0 becomes a delicate problem, as
is illustrated by the example of p(s) = 1/(U(s)) = E (,L(n))/ns. p satisfies
the equation

(s1)12F(1 -)9(S) - -S12F(8 )R(1 - s)

We have

M(~ ~~~(
MP(x)= - (a-)P- 1c+i-
<' 1a(n) 1s~xs' r(s) ()5 e> 1.
Pds),
n=X F(p + i) 2wriCic-i 1?(s +P+1

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 113

We can no longer move the line of integration arbitrarily to the left.

However, one can show that there exists a sequence T, -+oo such that
I T(s) I> T;-B for -1 < a _ 2, t = T>, B being an absolute constant [14,
p. 185]. If z runs over the non-trivial zeros of ?(s), when p is large, we
obtain

Mp(x) = lim_. E. Imz R p(x, z) + Fp(x)

where Fp(x) is of the form xP times a power series in 1/x, and Rp(x, z) is
the residue of P(s)q)(s)xs+P/P(s + p + 1) at s = z. However, one cannot
show that the series

E IRp(x, z) I < so
when p is large, unless one knew, for example, that the orders of the
zeros of C(s) are bounded. Even if one knew this, if we set Qp(x) =
A Rp(x, z), Mp(x) - Qp(x) = Fp(x) no longer changes sign infinitely often.
However, it is possible to show that Mp(x) itself changes sign infinitely
often. Our next theorems no longer need a functional equation. The
relationship with the equation is that if A < 0, then the solution has infi-
nitely many complex singularities.

(6.1). Let pp(s) = E' l anX;8 have a finite abscissa of absolute con-
vergence. Suppose that all the singularities of p(s) on the real axis are
poles, and that (p has a non-real singularity. Let K be the upper bound of
the real parts of the non-real singularities of Ap, and let S0(x) be the sum
of the residues of (p(s)/s)xs at the real poles > K. Then we have

A' (x) - S0(x) = f2?(XK-l), for every e > 0 .

We use the following theorem of Landau [10, p. 88]. If a(x) is a function
integrable in any finite interval in x > 0, and g(s) = (a(x)Ixs)dx has a

half-plane of convergence, and if a(x) ? 0 for all large x, then if c

abscissa of convergence of the integral, then s = a, is a singularity
Suppose that at a pole s = u, p(s)/s has the expansion qp(s)/s =
1:=, (c,,j(s - u)V) + a regular function. Then the residue of (R(s)xs)Is at
x = U is

1r cVxu(log x)v
Ev-1 (V 1 )!
Now, we may clearly suppose that X, = 1. Then

(S) = 1 d a > a.
From the fuXl+1

From the formula

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
114 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

y(log X)v' dx = V!_ _

Xs 'l~$' d=(s_ 1)v+1 '
we conclude that

f(s) = yAA(Z)- SO(x) dx

has no real singularity r.

Let Kr < x and a(x) = (AA(x) - SO(x) - XKl)/x. Then

XS)=f (s) - 1 _ _ x)
S-1 1 is

Suppose if possible that a(x) < 0 for all large x. Let o,, be the abscissa
of convergence of the above integral. Since f (s) has a (non-real) singu-
larity with real part > K - s, it follows that oo0 _ K. If a(x) < 0 for all
large x, then s = ao is a singularity of g(s), which however is not the case
since f (s) has no real singularity >K and K1 < K. Hence A(x) - SO(x) =
Q+(xKJ), and similarly it is 2_(xK1). This proves (6.1).

(6.2). If in addition to the hypotheses of (6.1), we suppose that p(s)

has a pole of order v > 0 at s = K + iz-, r # 0 , then

AO(x) - SO(X) = ?(XK log"-1 X) .

Let a,(x) = x-'{A?(x) -SO(x) - cxK logv-l x}. Then, we have

h(s) =f(s) c/(Lv - i)! - ac(x) dx .

(5 -K)v I~ Xs

Let ao be the abscissa of convergence of this integral

real singularity >K, h(s) has a real singularity at s = K, but none >K.
Hence, if a,(x) < 0, for all large x (say x > Y) we conclude, by Landau's
theorem that ao =K x. Hence, for a > K ,

Ih(a +it) I <A |a)dx - ja z)dX? B-h(a) .

If R = limlK+O (a- K)vh(ai + iz), then R # 0, and we conclude that

I R I < lim-K+O ((i - )v(B - h(a)) c

()- i)!

If we choose c with 0 < c < (v-)! - R 1, we conclude that ac(x) must be

>0 for some arbitrarily large values of x; i.e.,

AA(x) - SO(x) = 72+(XK Jog'1- x).

Similarly, we prove the other part of (6.2).

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 115

We remark that we could prove in the same way that, under the hy-
potheses of (6.1)(resp. 6.2), we have

AP(x) _ SP(X) = G,(XK+P-L)

(resp. f2?(XK+p logvl x)), where Sp(x) is the sum of the residues of
(r(s)/F(s + p + 1))q(s)xs+P on the real axis >? . For the idea of the proof
of (6.1) and (6.2) see Ingham [10. pp. 90-91].
It should be noted that the difference between the Theorems (6.1), (6.2)
and the main f2-theorem is that SO does not include the residues at all the
poles of p(s) in (6.1), and the result is in fact false if we replaced SO by
QO = sum of the residues of (qp(s)xs)Is to the right of a suitable line (even
if this sum had a meaning).
Finally, if A > 0 (not necessarily a, > 0), and we do not suppose that
the singularities of T lie in a compact set, we can obtain f2-theorems co
responding to Theorems 3.2 and 6.1, assuming that the "sum of the res-
idues in any vertical strip" has a meaning. The case A = 0 is similar to
the case A < 0.

7. Special functional equations and the corresponding identities

As was already observed, functional equation (2.2.) generally implies
an identity of the form (4.6), namely

(7.1) AP(x) - Sp(x) = E" bn -+I(PnX),

for sufficiently large p. Here

I(X) =1 Fcp+ r(8 - S)A(s) x?+p-sds

27J Cp-i F(p + 1 + a - s)A(3 - s)
For proving the f2-theorem and the 0-theorem, we required a knowledge
of the asymptotic behaviour of I(x). We actually used the fact that [cf.
(3.22)]

I(x) = A1(x (2A))A+? (2A-1)PJ2,,+ 8+p(hx11 (2A)) + 0 ([X1/(2A)]AS+(2A-1)p-(3/4))

together with the asymptotic expansion of the Bessel function J,(x). It

happens, however, that for some special cases of the functional equation
(2.2), I(x) can be explicitly evaluated and the range for which (7.1) is
valid can be precisely determined. The case A(s) r(s) has already been
dealt with in [2]. Some other cases are

(7.2) p2Q 1 s) (s) =P2( - Sj(1 - s)

(7.3) (i 1 ) (- P)P(S) = SIP(1 + i - + p - S)

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
116 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

(7.4) r2(1 + s)9(s) = r2(2 - s> (- s)

(Some results on equation 7.2 were announced by us in [4].)1

Since our approach to all these three equations is formally the same,
we shall deal in detail with only one of them, for instance (7.2). Equation
(7.3) is more general than (7.2) but the evaluation of I(x) for (7.3) when
p is not an even integer is simpler than when it is.

THEOREM 7.1.

(a) Functional equation (7.2) implies the identity

(7.5) A,(x) - P,(x) = -2-PU .'= b(- (p) F/+p24(X#)1/2

where x > 0, F,(x) = Y(x) + (- 1)Y-1(2fi)K,(x), p is an integer such that

p ? 2fl - 3/2, where f8 is a number with the property 7 I bn pI (e; < o
and

P7(x) = r(s)I(s) xs+rds

where C7 is a curve which encloses all the singularities of the integrand.

(b) Functional equation (7.2) implies the identity

(7.6) AA(x) - Rp(x) = c E:= b0(?) +p)/2 /2}

where x > ,O p = k + a, where k is an integer and 0 < a < 1, k _ 2fl -
3/2, and Rp(x) is a C--function in x > 0.
(c) If (7.2) holds, and

supo019| Em2<Mu<(m+A)2 z be - (1), as m - oo

then the series of Bessel functions on the right hand side of (7.6) con-
verges for p ? 2,3 - 5/2, uniformly in any interval in x > 0 in which
the function on the left hand side is continuous, and boundedly in any
interval 0 < xl ? x ? X2 < ?? if p = 0-.
For the proof of Theorem 7.1 we first require a lemma in which a contour
integral is evaluated.

LEMMA 7.1. Let x > 0, k a positive even integer, and 0 < c < k/2.
Then we have

1 On p. 1334 of [4], in the statement of Th6orem I, lines 2-5, replace i by p, and

puisse Otre by ne puisse etre.

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 117

1 PC+i r2(s/2)xk-Rds

27ri Jr-i r2(i - s)(1 ) -s).(k -s)

(7.7) 2~-m
=-2 X x2Fk(4Vx) -41-'-1 E02m<k-, (4x)2m + 1)
PROOF. Consider the integral

(7.8) I= 1 r2(S/2)Xk-sds
2z 7r Rr2( -S(1 - s) * * (k - s

taken along a rectangle ?R with vertices at c - iR, c + iR, - (2m + 1) +

iR, - (2m + 1) - iR, where R > 0, and m is a positive integer. The
condition 0 < c < k/2 secures the convergence of the integral. The inte-
grand has poles at s = -n, where n is a positive, even integer, with n < c.
In order to compute the residues it will be convenient to rewrite (7.8) as

(7.9) I = 1 41-kr(2S) cot (ZS)(4X)k-2sds

2Wi r(k + 1 - 28)

by first changing the variable s -+2s, and then using the well-known
formulae r(1 - 2s) = 2-2SZ-12r(1/2 - s)r(1 - s), and r(s)r(1 - s)
z/(sin zs). Near s = -n we have the expansions

F(2s) = 1 Hn)[1 + r'() + h2,}2(s + n) + *..]

2r'(2fl + 1)(s + {n(1

cot Zs) 1 1 +
cot(s) 7r (s +) J

and

r( +) k-2__ = rk-4X)k+2n [1 + (s + n){2I(k + 1 + 2n)

L'(k+ 1- 2S) F(k + 1+2n)
-2 log 4x} + *]
where we have

= F'(x) = 1 + ,.+ +n
F(x) 2 fl

so that *(n + 1) = F'

grand in (7.9) at the poles s = -n, n = O 2, ..* 2m, equals

41-k {, (2n + 1) + *(k + 1 + 2n)}(4X)k+2,

7r n~o(2n)!(2n + k)!

-log (4x)
no(2n)! Enm
(2n + k)! (4XY'(2+

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
118 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

If M -+ 0c, this equals

{2k-2xk 2 { Yk(4-v/) - 2Kk(4V/)}

(4x)2 (k- - 2n)!1

8052n~k-1 (2n)!
The sum of the residues of the integrand at the poles s=+2m, where
0 < 2m < c is given by

r2(m)x k2m
1:0<2m<c r2(1/2 - m)r(k + 1 - 2m)r(2m)
- 2<2m8 C r(2M)2-2(2m-l)xk-2m
r(k + 1 - 2M)w

41-k r(k - 2p)(4X)2P

'k-c<2P<k F(2p + 1)

since the residue of r(s) at s = n is (-1)"/n!. Hence the sum of all the
residues equals

(7.10) _2 1kXk2Fk(4/Vx)- 4 (2X<k-()2,r(k - 2n)

Z ]['~~(2n + i)
Now if we let R oo, and use the asymptotic formula (2.14) for r(x), we
obtain

(7.11) I = 1 XC+iW _ 11(2m+l) + ioo ]2(S/2)xk-sds

2Wr?, c-iOO 2Z -(2m+1)-io r2(1 - S)(i - S) .(k - s)

since c < k/2. Equation (7.11) holds for each fixed m. If we now let
m -* + cc, we see that the second integral tends to zero, for we have

r(-m - 1/2 + it) = F(1/2 + it)

r - - 1/ + it = 1/2 + it)( - 1/2 - 1 - it) ... ( - 1
so that

I r(-m - 1/2 + it) I ? A1l exp(-(1/2)wr t 1).{1/2(1/2+ 1) ... (1/2 + m)}-

< A1 * exp(-(1/2)w It 1) *2m+l(m!)-l

while r(m + 1 -it) = (m - it)(m-1-it) .. (1-it)r(1-it), so that

F(m + 1 - it) I _ A2 * (m!)(1 + It 1)1/2 exp(-(1/2)w I t 1), the

r(-1/2-m+it) < B._ 2"

r(m + 1 - it) = (m!)2(1 + jt J)112

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 119

Hence if

j = I cP( rn( -1/2 + (1/2)it) $k+2m+l-it dt,

27J -- F2(rM + 1 - (1/2) it) (-2M-2 +it) * (- 2M- 1-k + it)
we have

(7.12) IJI B .4x x~ ~ t0 asrn -+~oo.

(r!m)4 3+ (i + t2)1/2 ... (k + t2)1/2 a

On combining (7.12) with (7.11) and (7.10) we get the lemma.

PROOF OF THEOREM 7.1.
(a) If c is sufficiently large, we have, as in (3.13),

(7.13) AA(x) = S,(x) + 1 c F_ P(i - S) p2(S/2) )$-S+P*(s)ds

FZ 1(2 + P- S)F][2(1 -8)

where

SP(x) = F,(s +p + i) 'p(s)xs+Pds,

9 being a curve which encircles the singularities of the integrand which

lie in the strip 1 - c < a < c. Now let p be a sufficiently large, odd inte-
ger, large when compared to c. We may rewrite (7.13) as

AP(x) = Sp(x) + EnO bn.e-l-P

(7.14) 1 _ _ _ _2 __(S _2)(_ ,x_)l+P-sds
27Ti ____1-
2zlt -i r(l s(1 - S') . .. (1 + p S')
2

We now employ (7.7) with k = p + 1, to obtain from (7.14) the identity

(7.15) Ag(x) - Pp(x) =-2-P E,,, x F,+pl4(M x)121

where

P~(x)= S~x) -~C<2~p?1 x1+p-2inX(I- 2rn)

PPp() =SP(X) - C<2m:PA-l F(2 + p - 2M)r2(1/2 - M)r(2M)

- 27ri
Fs Fr(s)X(s)x'S~ ds = 1 F r(s)P(s) xs+Pds
C r(S + p + 1)r2 (S12) 27ri Cp r(S + p + 1
Cp being a curve which encloses all the singularities of the integrand
p being a sufficiently large odd integer. Since [17, p. 66, p. 79]

(xdx)( xK,,(x)) = - x-'KK,(x), d )(XV Yv(x)) = XV-l YOAx)I

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
120 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

d [xv/2F{4(flx)l/2}] = 2-1/2 * x(v-l)/'F,_lf4(pnx)l/'

dx

the form of the identity in (7.15) remains unchanged by differentiation,

provided that the resulting series on the right hand side is absolutely con-
vergent, which will be ensured if bn I beI pe(+p'2-114 < 0, or p > 2fl -
3/2, because of (2.10) and (2.11). Hence (7.15) holds for all integral p >
2, - 3/2 as claimed in Theorem 7.1 (a). The series on the right hand side
of (7.15) converges uniformly in 0 < xl < x <x2 < 0o if p > 2,l - 3/2,
irrespective of whether p is an integer or not, and therefore defines a con-
tinuous function.
To prove Theorem 7.1 (b), we observe that fractional integration of
order a of the identity (7.5) yields the relation

AP(x) - P. (x) =c1 fl1k/2

Ew bl P(a)
1

-(X t)&-lt(k+l)/2Fl+k{4(tpjl)l/2}dt

(7.16) C2 En=1 b , (l+p)/2 Y1+,f4(xpn)l/2j

+ C3 E" bn />tIlb X1 So-Sk-2,l+p{4(x/n)l/2}

+ C4 (x - t)O-lEk(t)dt
0

where S stands for Lommel's function, and

\(k+l) /2

(7.17) Ek(t) - E b t ) Kl+kf4(tafn)1/2}

is a C--function in t > 0. This follows from (2.18). Because of (2.20)

the series of Lommel functions in (7.16) is absolutely and uniformly con-
vergent in 0 < xl _ x _ x2 < ca, and is actually a C--function in that
interval for k > 2,8 - 3/2. Further the integral

(x - t)O-lEk(t)dt = [5 + X(5 - t)w-lEk(t)dt

where each of the integrals is a C --function in 0 < xl < x < x2 < co*
The first because Ek(t) is Co in t > 0, and I Ek(t) I < ME' (I bn 1)/(pk +1
where M = sup,>0 y(k+l)/2 I Kl+k(4Vy) I < ca, while the second is equal t
x-8

yc-i Ek(x - y)dy

which is obviously a C--function. If we now set

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 121

R,(x) = P,(x) -, C3E , b( ) S { (1/2) (++p)

- c(x- t)1-'Ek(t)dt
0

in (7.16), we obtain (7.6) as claimed in Theorem 7.1 (b).

The proof of Theorem 7.1 (c) is accomplished by the method of equi-
convergent trignometric integrals employed in a similar situation in a
previous paper of ours [2, Th. III].

THEOREM 7.2. (a) Functional equation (7.3) implies the identity

(7.18) AA(x)A n=~~~~~~~~~~

- PP(x) = 2PE 1 bn ) Fl+p, _4(xp,)
where x > 0, p is an integer such that p > 2,8 - p - 3/2,

Fl+p,,f4(xpn)l12
= cos {(1/2)(1 + p)7c}Jl+p+?{4(xj e)1/2}

-sin {(1/2)(1 + p)7}[ Y1+p+,{4(xf2)1'2}] + 2 COS PZ Kp{4(xp )1/2}

and

PP'(x)= 1 I F(s)9(s) x8pds

2Z Cp' P1(s + p ? i
where C' is a curve which encloses all the singularities of the integrand.
Further the functional equation implies that f > p/2 + 3/4.
(b) Functional equation (7.3) implies the identity

A() - RP(x) = cl E' I bn _ Yl+p+,f4(xp,)1/2}

+ c2 ( z (1/2)(1+ P ?l+p+14(xpn) /2

where x > 0, p = k + a, k is an integer, and 0 < a < 1, k > 2,8 - p -

3/2 and RPP is a C--function in x > 0.
(c) If (7.3) holds and

sup0:A |m2<Ig <&)2 n< A)= o(1), as m -+ oo

the series of Bessel functions on the right hand side of (7.19) converge
for p> 2,8 - p - 5/2, uniformly in any interval in x > 0 in which the
function on the left hand side of (7.19) is continuous and boundedly in
any interval 0 < x1 _ x _ x2 < ?? if p = 0.
The proof is parallel to that of Theorem 7.1. The integral to be evalu-

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
122 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

ated here is

f(x) = . r+ )sXl+P-S+Pds
( 8() cosec (-p Z
27N, c- s~ - S'(2 + P+ P S)

If p is an even integer,

p2( S xl+P+P-s
f(x) = 5 l)P12 ds
27C c-i- o2 1

which is of the same

p replaced by p + p. If p is non-integral, we evaluate f (x) anew.

THEOREM 7.3.

(a) Functional equation (7.4) implies the identity

APA(x) - P.'(x) = 2-P (12) /+P) Y 4(x )1/2

+ (- 1)P+1 2 K1+{4(Xa )"12}

where x > 0, p is an integer such that p ? 2,9 - 3/2, and

P (x) = 27C1 CpF(1 (s) q (s)xs+Pds ,

2 Z C" ~l+ s + p)

where C." encloses all the singularities of the integrand.

(b) If p = k + a, where k is an integer, k _ 2 -3/2 and 0< < 1,
then equation (7.4) implies the identity

(7.20) APg(x) - Rp"(x) = C1 , b( )2]

where R." is a C--function in x > 0.

(c) If (7.4) holds, and if

5UPo?,h?1 | m2<In<(m+?)2 nis-1/2 = (1) , as m o

the series of Bessel functions in (7.20) converges for p ? 2,9 - 5/2, uni-
formly in any interval in x > 0 in which the function on the left hand
side of (7.20) is continuous, and boundedly in any interval 0 < x, ? x
X2 < Co.

REMARK. Apropos part (c) of Theorems 7.1, 7.2 and 7.3, it may be ob-
served that if p < 2,3 - 5/2, the series of Bessel functions in question are

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 123

summable by Cesaro's method, and the correct order of summability can

be obtained as in [2, Th. IV].

8. Applications to zeta-functions, and L-series,

of algebraic number fields

We now deal with the applications of the above theorems to ?-functions,

and L-series, of algebraic number fields. We need some preliminaries.
Let K be an algebraic number field of degree n over the field P of
rational numbers with discriminant A. Let r1 be the number of real con-
jugates of K, and 2r2 the number of imaginary conjugates of K. Let (E
denote an ideal class in K. For any ideal a, a will denote the comple-
mentary ideal. The unit ideal is denoted by a, and b = 1/o is the different
of K. For any a, a = 1/ab. If a e (, the class of a[ will be denoted by E.
Let f be an integral ideal of K. Let Gf be the group of ideals a = b/c
with b, c co-prime to f. Let CQ be the quotient of Gf by the group of
principal ideals (a) with a 1 mod f and a totally positive. CQ is called
the 'ray class group' mod f.
Let 6 be an ideal class of K. We define the Dedekind zeta-function of
class E by the relation

(8.1) 1K(s, () = Eaec(N)s

the sum being taken over the non-zero integral ideals in E. We define
the Dedekind zeta-function by the relation

(8.2) ?X(s) = a 1
~K()= a (Nat)8s
the sum being taken over all non-zero integral ideals of K. Clearly

(8.3) ?K(s) = fly ?K(S9 (i i

It is known, after Hecke [11, p. 67], tha

with a simple pole at s = 1, with residue

2rl+r2;r2R

WVJ~ AJ

where w is the number of roots of unity in K and R is the regulator of K.

It satisfies the functional equation

(8.4) (s, () = 4(1 -, (i)

where

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
124 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

(s, (i) = Bsrrl( S )r2(s),K(S, E), B = 2-r2Z-7nl/2ViA

The function UKf(s) has a simple pole at s = 1 with residue xh, where h is
the class number of K, and satisfies a similar functional equation.
Let X be a character of Cf. Define

(8.5) L(s, X) = E X(a)

(Ncr),s
where the sum is over all non-zero, integral ideals of K. Suppose that X
is a proper character (that is to say it does not have a "period" f1, where
f1 is a proper divisor of f).
Let Y # 0 be a number of K, such that (Y) = a/bf, where a is an integral
ideal with (a, bf) = 1. Set

T(X) = T = amodf X(a)e 0

the sum being taken over a complete residue system mod f, and let C(X) =
(Nf)-112T(X). Then I C(X) I = 1. When a = (a) is a principal ideal, we have
X((Q)) = X(a)v(a) ,
where X(a) is a 'residue class character' mod f, and v(a) is a 'character
of signature', that is if a,, ..., a,, are the real conjugates of a, then
v(aY) = ( CY ) ri (ta

where the ai's are 0 or 1. If v(a) _ 1, we have the 'class characters'

mod f. Define

(8.6) W(X) = X(O i)( apC(X)

Let us first consider quadratic fields, imaginary and real, and the as-
sociated zeta-functions and L-series. In these cases we can give identities
of the type considered in Theorems 7.1-7.3.

Imaginary quadratic field. In this case K = P(V-/D), and ?ffJs

satisfies the functional equation

(8.7) (VD i)sP(s)W(S E) = (VDh)1sP(l - S)?ff(1 - s, E)

In the notation of our theorems, we have

Q,(x) = 2E-s( F(s) (V'D') f ()x+Pds

2Z PS+ p + i)\27

- K( E) XP + XXP+1 V/D
r(p +) F(p+2) 27r

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 125

with X = 2;/wVfD.
Let X be a proper character mod f (v(a) =1 always). Then L(s, X) satis-
fies the equation

(8.8) 4(s, X) = W(X)W(l -, X),

where

t(s, X) = [(DNf)12 ]P(s)L(s. X)

In this case Q,(x) = 0.

The identities corresponding to (8.7) and (8.8) are as follows [2]:

1 - ~:5. (X - Ncr)P
F(p + 1) aeg_
(8.9) ) +( ) ( )(1/2) (1+p){41(Na)12}

where

QP'(x) = IK"(0 (i) P + 'T X+

F(p + 1) I7(p +)

(8.10) r(P1ENalxX(CQ)(X
+ 1) EAY()XN)- Nat)P
( (DNf)12 ( (X)(1/2) (l+P){ 42(xNa)lI2}
-L 27C a NXataJ j+Pj -l(DNf)r
By [2, Theorem III] the series converge for p > -1/2.
Equations of the type (8.7) were considered by us in [2].

Real Quadratic field. In this case K = P(1VD), and CK(s, 1g) satisfies
the equation

(8.11) 5(S, (9) =W ( s9 E),

where

f(s, U) = ('/i)82(IS ),K(s, j)

Here

QP(x = 152 1:P K(-22 - 1, )Xp-2v-11VLvh2

= 2v+1<p F(2V + 2)F(p - 2P))

+ ?K(O, E) XP + XXP+1 (VD

P(p+i) r(p+2Y Z
with X = (2 log s)/VD, e > 1 being the fundamental un

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
126 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

Let X be a proper ray class character. Then for the L-series there
are three cases to consider.
(i) v(a) -=1, in which case we have

(8.12) t(S, X) = W(x)(1 - S, X),

where t(s, X) = [(DNf)912/1z]P2(s/2)L(s, X), and

(ii) v(a) = (a/jl a )al(a'/l a' )a2, ala2 = 0, in which case we have

(8.13) t(s,tx) = W(X)M(1 S-, %)

where

t(s, X) = L (DNf)l"2 1sF( S )r(S + 1)L(sq X)

and Q,(x) = 0.
(iii) v(a) = Na/l Na 1, in which case we have

(8.14) t(S, X) = W(X)(1 S- s,%),

with

~(s, X) = (DNf)"21' 2 s + 1)L( X)

In each of these cases, there is an identity of the type given in (8.9) and
(8.10). In the case of (8.11) and (8.12) the function J1+P has to be replaced
by F1?+[cf. (7.5)], in case (8.13) no change is necessary, and in case (8.14),
J1+P has to be replaced by Y1+p + (-l)P+1(2/z)K1+,. We next consider
Zeta-functions with Grossencharacters. Let A be a Grossencharacter
on ideals mod f. When a is a principal ideal (a), then

A(ca) = A((a)) = X(a)X(a)v(a) ,

where X(a) is a Grossencharacter mod f on numbers, X(a) is a residue class

character, and v(a) is a character of signature of the form given before.
If X(a) 1, we get ray class characters as a special case.
Let eq = 1, q = 1, 2, ... , ri and eq = 2, q = ri + 1, * * , r + 1, w
r = r, + r2 -1. Then there exist -/(A) with I YI = 1, real numbers
a1, .**, ar+l, and non-negative rational integers ap, a' (same for A as
for A) such that if the associated residue class character X of A is a
proper character,

(8.15) t(s, A) = W(A)t(1-s, A),

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 127

where W is a "Gauss sum" of absolute value 1, and

t(s, A) = 7(A) .+ll2 P(-)B81(s, A),

where B = 2-r2Z-f(l A I Nf)1/2 z = e (s + (a. + a')/2) + ix,

C'(s, A) - A(cr)

If A is not the principal character, then '(s, A) and t(s, A) are entire
functions. We set PA(S) = 11' F(z,,/2). We write the equation for
'(s, A)-=1 Cmm_8 Cm = cA(m) =NW m A(f), as

B81PA(s)t'(S, A) = C * B1-8Lr'(1 - s)(1 - s )

and assert that, for some integer m > 1, and v = (n + 1)/2n,

(8.16) E Re(C. c ,)j co

and if cm is not always real, we have, in addition,

(8.17) EIm(C * cpm) co

the summation being over the rational primes p.

PROOF. By Remark (5.1), we have I cm I m-v = co. Further it ca
be seen that CA(M) is multiplicative, that is CA(m) CA,,(k) = CA(mk), if
(m, k) = 1, and that Cm = 0(m8), for s > 0. Using these facts, it can be
proved that if w > j and , IC I m-w - co, then , I c I ppw - co
-,, I c = I p-v = co. If cm is real for all m, this gives (8.16). If not,
a, + i90, l,0 s 0 for some m. If, in addition, , IjReCc, j p-v <
follows that E J Im(C. c,)I pVv = co. This gives E IRe(C. cmp)Ip-v
since Re (cm * C * cp) = aRe (C * c,) - I0Jm (C * c,), and cmp = c
p > m.
Let us now consider the application of Theorems 3.1, 3.2 to the functions
fK(Si E), fK(S), and '(s, A).
In the first case we have a. = bn = 1, Xn = Sn = B`cNa, Gp(s) = ? K(S E
+s= (Si ), 8 = 1, i = 1 + s > 1. For the set {1ak} we take again
the set {B-11}, where 1 runs through the positive square-free integers.
We then have

[{F(P + 1)}'1 ENacx (x - Nc)P QP(x)]

(8.18) limx ___ aE_ - + co
(1-1/n) (p+l/2)

for 0 < p < (n - 1)/2, and

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
128 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

f2 (x(1-1/n) (p+1/2)\

(8.19) [{P(P + 1)}1 ag (-Na)-Q()] = O((-/

for p > (n - 1)/2.

An application of Theorem 4.1 with 8 = 1, A = n/2, 7 =

(n - 1)/(n(1 + n)) yields, in view of Remark 5.5,

(8.20) ENasx, 1 = Xx + O(x'n-"/( f+1

aeg

In the case of ?K(s), the results (8.19)-(8.20) hold with X replaced by

xh. The O-results here are the same as Landau's [11, p. 135]. Landau's
i2-result, however, is

ENCa~z, 1 - Qo(x) = f2(xE'-(1/n)1(12) 7q)

ceg

for every )2 > 0, which is weaker than (8.18).

In the case of ?(s, A) we conclude that if E I cA(m) I = 00, then

(8.21) Re[1( +l) EcNa? A(a)(x - Na)P - Q(x)]

x(1-1/n) (p+1/2)

and, if cA(m) is not real, the result also holds with Im for Re. If
X I CA(m) I MVPn < oo 0 (in particular, if p > (n - 1)/2),

1 O(x (1-1/n) (p+1/2))

(8.22) P(p + i) EfCtz A(a)(

the (2: applying to the real and

Theorem 4.1 with 8 = 1, A = n/2, Y2 = (n - 1)/(n(n + 1)) gives

N A(cz) = O((n-lMn+l)) + x<NCx' I A(a) I

where x' = x + O(x'n-1')/n+')). It follows therefore from (8.20), that

(8.23) E?at!x A(a) = O(x(n-l)/(n+l))

9. The Dedekind-Epstein zeta-function

Let K be a totally real algebraic number field of degree n, and let Q be

a totally positive quadratic form in m variables with integral coefficients
in K (every conjugate is positive definite).
Let a be any ideal in K and define, for t = (t1, ..., tj) with ti > 0

E3(Q, a, t) = Exie e-7r(tQ[x])

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 129

X1

where x is a column ( of elements of a and Q[x] = x'Qx, x' being the

transpose of x. Let 01(Q, a, t) = @(Q, a, t) - 1. The following transfor-

mation formula holds:

e(Q, a, t) = (A(C[))-mI2(TN( Q 1))-1/2(1v(t))--M/26(Q-1, a, t-1)

where I Q I is the determinant of Q and A(a) is the absolute value of th

discriminant of a. Let 9? be the cartesian product of the positive real
axis by itself n times and dv = (dt1 ... dtQ)/(t1 ... tQ) the invariant
element of -T. The group U of units s e K acts on 9? by the rule

t = (t1, *. . ,t) +(tJ I|* * .. t I En 12) 1

where se is the V)t conjugate of s. Let F be a fundamental region of U
in 2?. Let F, be the part of F with t1 ... tn < 1, F2 that with t1 * - - tn > 1.
Finally, let

(9.1) C(Q, ac, s) = {x}ea (NQLx])

the sum being over a complete system of columns x e a which are non-
associate, x and y being associate if there is a unit s e K with x = sy.
Let &(Q, a, s) = Fn(s)z-ns8(Q, a, s). Then one shows that

&(Q, Q, s) -21f- RsS ~m/2

+ ( /)(N(2
- s- D)
F } + (1/24) {(21(Q, c, t)(Nt

+ A(Ca)-iy2(N(I Q I))-1/2(e1(

where Nt t1 ... ti, and R is the regulator. It follows that

(9.2) S(Q, ac s) = iA(Q) m/2(N(I Q ))-1/2 (Q-1, , ay -S

Moreover, C(Q, a, s) has a simple pole at s = m/2 with residue

wn2n-1R(a)-mI2(TN( Q 1))-1/2 = X,
say. C(Q, a, s) has a zero at s = -p, p integral and p ? 1 (and at s = 0
if n > 1).

XXp+mI/2rp + mk)

(9.3) QP(x) + m2 + 1 (Qc + 0)1)

r P M + F(p +i)

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
130 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

Let a(Q, t) be the number of representations of t by Q. We conclude that

(9.4) ENtSx t5ate a(Q, t) (x - Nt)P - QP(x) = 2+(X(mI4)-(lI4n)+p[l-(l2n)])

E~tX'_-1_2t aQt)F[(p + 1

(and is moreover O(X(m/4)-(l/4n)+p[l-(l!2n]) for p > (mn - 1)/2). Finally,

Theorem 4.1 gives us, with = (mn - 1)/(2n(mn + 1)),

ENt<x, a(Q, t) = * Xm/2 + O(X(m2)(n-l)/(mn i1))

t5__':2t'

10. Some classical arithmetical functions

We give here a selection of examples to illustrate the effect of the

theorems obtained. Obviously no attempt is made to be exhaustive.

The divisor function d(n). If d(n) denotes the number of divisors of

n, we take an = bn= d(n), and Xn= = = wn, and p(s) = A(s) = Z-sU
r-S E d(n)n-", for Re s > 1. The functional equation is

F2(s/2)p(s) = F2{(1/2)(1 -- s)}*(1 - s),

so that 8 = 1, A = 1. For the subset {jen} specified in Theorem 3.1, we
take the sequence {{wZ1}12}, where 1 runs through the positive square-free
integers. That {111/2} have the required property follows from a theorem
of Besicovitch [1]. It is also easy to see that E, d(l)- 12)p-(3/4) < + co fo
p> 1/2, while E d(l)-(1/2'p-(314) = + oo for p ? 1/2. Further

F zp?2(0) XwP p 7 1p-2v-1 2(-2 -1)

QP(x) = 2(P + 1) ) ZVZ J (2- 1)F(2V + 2)

+ j 2 (X )pl[y _ !1(2 + p) + log x]

If we write Dp(x) = {r(p + 1)1-1 US d(n)(x - n)P, then by Theorems

3.1, 3.2 we have

lim supz>. DP(x) - Qp(x) + cc

(10.1)lim inf D(p/2) + (1/4) = ?

for 0 ? p ? 1/2, while

nQ+(X (p/2)+(1/4))
(10.2) DP(x) - QP(x) = (

for p > 1/2.

If p = 0, then QO(x) = (1/4) + x[(2y - 1) + log x], where -y is Euler'

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 131

constant, since

- (n + 1) = r'(1) + En r-1 = - + nLr-1, C(0) = -I/2,

hence (10.1) gives the results of Hardy and of Ingham referred to in ? 1.

In order to apply Theorem 4.1, we note that 3 = 1 + s > 1, d(n) > O0
and if we choose C = 1/6, we get, in view of Remark 5.5,

(10.3) En<x d(n) = x[(27 - 1) + log x] + O(x113 log x).

The error term is not essentially different from the one originally obtained
by Voronoi [16].
Theorem 7.1 enables us to write down the identity for DA(x), namely

(10.4) DA(x) - Rp(x) = c , d(n)( ) Yl+p{4w(nx)1'2},

for p > -1/2. In the special case in which p = 0, we have

F d(n) = 1/4 + x[(2-Y - 1) + log x]

E - J1 d(n)( - FL1{4z(nx)1l2}

for x > 0. [12].

The function ak(n). If uk(n) denotes the sum of the kth powers of di-
visors of n, the case k = 0 corresponds to d(n), so that we may now assume
k :t 0. If we take an = bn = Uk(n), Xn = wn = il then p(s) = A(s) =
7sT(s)?(s - k) = En ak(n)n-r, for Re s > max (1, k + 1). The function
al equation is

r(s/2)F{(1/2)(s - k)}p(s)
= F{(1/2)(1 + k - s)}F{(1/2)(1 - s)}1{(1/2)(1 + k - s)}
so that 8 = k + 1, A = 1. For the subset {felk} in Theorem 3.1, we may
again take the sequence {Z1/211/2} as in the case of d(n). If k > 0, it is k
that Eq (k(l)l-a < + ?? for a > k + 1, while Ok(l)h-a = + ?? for a ?
k + 1. The function QP(x) is easily computed.
If we write SkP(x) ={P(p + 1)}-1 E', Jk(n)(x - n)P, we have for k > 0
fS2 (X(1/2)[k+p+(1/2)])
(10.6) Sk,(x) -Q(x) {O( (1=2) k+?(1/2)];

if p > k + 1/2, while

limlim
supX__)'. SkP(x) - QP(x
supers(1/2)[k+p+(1/2)] = + cc
=+?
(10.7)

lim infX-0S (1/2)-k+ +(1/2)] = - 0

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
132 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

if p ? k + 1/2.

If k < 0, we have Uqak(l)1-a < + ?? if a > 1, andE Ukk(l)1-a = + cc

if a ? 1. Hence

vQ((1/2) [k-i- p+(1/2)]

Sk(x) - QP(x) = 4(x-lI2rktP(
( O(X xl1/2[k?pl1/2])

if p > I k I + 1/2, while

rim SUPx,+(, SA) = + cc , lim inf S0 SkP(X - Q+(x) C

00 (1/2)[k~~~~~p+(1/2)] ~~~(1/2)Ek+p+(1/2)]

if p ? IkI + 1/2.
Theorem 7.2 enables us to write down an identity for SP(x) (cf. Oppen-
heim [12]), and the series of Bessel functions converges for p > I k -I 1

The function dk(n). Let dk(n) denote the number of ways of expressing
n as a product of k factors. We take k > 2, (since k = 2 corresponds to
d(n)), a. = bn= dk(n), Xn = -ln = r+kI2n, and p(s) = *(s) = 7w-ks/2Ck
,-ks/2 E7 dk(n)n-s for Re s > 1. The functional equation is

Fk(s/2)p(s) = Fk{(1/2)(1 - s)}*(1 - s)

so that 8 = 1, A = k/2. For the subset {,ae} specified in Theorem 3.1, we
take the sequence {(wZk121)lIk}, where 1 runs through the positive square-
integers. Further E dk(l)1-[(P/k)+(1/2k)+(1/2)] < + c0 for p > (k - 1)/2
dk (1)1-[(p/k)+(1/2k)+(1/2)] = + oo for p < (k - 1)/2. We have also

Q (x) - 7r d12)kp-k(O) (X) + 7rkp/ (/2() Pkflog k/2)

(1/2)kpSo(->-) X \-2,i-l +-p

-Z d1<2V+1-~-p 1'(p - 22V)r(22- + 2 ) \2k/'

where Pk_1(log X/(;'k'2)) is a polynomial in log X/(;'k'2) with the leading coef-
ficient 1/((k - 1)!r(2 + p)),
If we write DAk(x) = {P(p + 1)}'1 nx dk(n)(x- n)P, then by Theorems

3.1, 3.2, we have

lim supX DO k(x)QP( - o+?

(10.8)

lim infX-. DN,k~x - Q~x CIO

l [1- (1/k)][p+ (1/2)]

for 0 ? p ? (1/2)(k - 1), while

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 133

f 2+( E[- (llk)]Lp+ (1/2)]

(10.9) DPk(x) Qp(x) = - 2(-Xl/]EP?

for p > (1/2)(k - 1).

In order to apply Theorem 4.1, we note that , = 1 + s > 1, and dk(n) >
0, and if we choose C = (k - l)/(k(k + 1)), we get, in view of Remark 5.5,

(10.10) EnS dk(n) = xPk-1(log x) + O(x(-l)I(kxl) logy1 x)

Product of functional equations. The main theorems enable us to
obtain results on the asymptotic behaviour of the coefficient-sum of the
product of two Dirichlet series which satisfy functional equations with
the same 8. We give an illustration. Let k > 0 be an integer, and X a
real, non-principal character mod k. Suppose that X is a proper character,
and put

p(s) = C(s) L(s, X) =ns Ean

where L(s, X) = E(X(n))/n-". One sees at once that

n an = En<zn X)L I
and since

(k1/2)r 2 ) 2)9s

= W(X)( -z )_( -s)1(1-s + a)qs(I -

where a = 0 or 1 according as X(-1) = 1 or -1, our theorems give

En< X(n)[~
5?n~~z n) -x__
X _X (% = +(X 1/4)
which can be written as

En<xX(nJx{ 0(13)
-x x {n1 = n2 (x14) ,
where {x/n} denotes the fractional part of x/n. This shows that the
results one might get by using the equations separately may be weaker
than those obtained by using the equation for the product.

Miscellaneous examples. We mention finally a few examples for which

the results given in ? 6 are applicable.
(1) Let e(n) denote the Mobius function. Then E (u(n))/ns = 1/C

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
134 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

If 0 is the upper bound of the real parts of the zeros of C(s), then 1/C(s)
has no pole on the line u = 0 only if 0 > 1/2, and it follows from (6.1)
that

M(x) = I j(n) = i2,(x0`).

If 0 = 1/2, then if N denotes the upper bound of the orders of the zeros
of C(s), we have, by (6.2), if N < oo,

M(x) = 2? (X12(log X)N-1)

while, if N = 00, M(x) = &2_(X112(log x),') for every '7 > 0.

(2 ) (C(s))/(C(2s)) = E1 (I j(n) 1)/nS. If Q(x) = I (n) 1, then it is
known that Q(x) - 6x/Zr2 = R(x) satisfies

R(x) = O(X112)

As in (1) above, we have

R(x) = &i?(X114)

(3) ( 1(s -))/(C(s)) = (ep(n))/ns, where q' is the Euler function.

We have, as in (1),

3X2 -2(I2
En~x 9p(n) -$=Q(l2

(4) (?4(s))/(C(2s)) = E7 (d(n))2/ns. Then, we have

Enx (d(n))2 - XP3'(log X) =- 2_(xl/4)

where P3' is a polynomial of the third degree. Examples (1), (3) above
satisfy a functional equation with A =-1/2, A 0 0 respectively, (2) and
(4) do not.
The method of proof of the main theorems is applicable even for
functional equations of certain other types. An example is provided by
the zeta-function of an indefinite quadratic form [13] with negative de-
terminant. (The case of positive determinant falls directly under the
main theorem and was dealt with in [2,3].)
( 5 ) Let S be a quadratic form in k variables with integral coefficients
and negative determinant and j(S, t) the 'measure of representation of
the number t by S'. Let

?(S, S) = Et>O S, t)
ts

Then ~(S, s) satisfies the equation

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
 ARITHMETICAL FUNCTIONS 135

-Sr(S)M(S, s)
=-[-S/2)-sl pk -S jSj-112(_1)(k-n+1)/2 cot ZS *(S-', k _-s

coseczs-C( S-1, k 2 s

where (n, k - n) is the signature of S (see [15]). The "identity" now

takes the form

.M, (SyX) , I rt e (S, t)(x -t)P = apxP+l + bpxP+kI2 + C xP

F+d dt?>O
+ L'o (S-',
0i) S1t)(t t)
(p/2)+(k/4)-
Yp+(k,2)(2cV tx) + R(x) ,

where R(x) is the sum of three absolutely convergent series and R(x) =
O(x(P/2)+[(k-l)4]-8) for some a > 0. The method of proof of the S2-theorem
is then applicable, and gives

Mp(S, x) - (apxP+l + bpxP+kI2 + C XP) = S2(X(1/2)[[(k-1)/2]+p)

and for 0 ? p ? (k - 1)/2,

Mp(S, x) - (apxP+l + bpxP+k/2 + c xp)

li x??4 1l/2[p+ (k-1) /2]_

Moreover, the method of equi-convergent trigonometric integrals ap-

plies, as before, and shows that

Et>/ MS1, t X (p/2) + (k/4) YPk,2(2zVfx)

converges for p > (1/2)(k - 3), and is summable (C, p) for p > (1/2)(k - 3) -
p. The details are the same as in [2].

TATA INSTITUTE OF FUNDAMENTAL RESEARCH

REFERENCES

1. A. S. BESICOVITCH, On the linear independence of fractional powers of integers, J.

London Math. Soc., 15 (1940), 3-6.
2. K. CHANDRASEKHARAN and RAGHAVAN NARASIMHAN, Hecke's functional equation
and arithmetical identities, Ann. of Math., 73 (1961), 1-23.
3. , Hecke's functional equation and the average order of arithmetical functions,
Acta Arithmetica, 6 (1961), 487-503.
4. , Sur l'ordre moyen de quelques fonctions arithmdtiques, C. R. Acad. Sci.
Paris, 251 (1960), 1333-1335.
5. G. H. HARDY, Ramanujan, Cambridge, 1940.
6. , The average order of the arithmetical functions P(x) and A(x), Proc. London
Math. Soc. (2), 15 (1916), 192-213.

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms
136 K. CHANDRASEKHARAN AND RAGHAVAN NARASIMHAN

7. , and J. E. LITTLEWOOD, The approximate functional equation in the theory

of the zeta function, with applications to the divisor problems of Dirichlet and
Piltz, Proc. London Math. Soc. (2), 21 (1922), 39-74.
8. E. HECKE, Eine neue Art von Zetafunktionen und ihre Beziehungen zur Verteilung
der Primzahlen, I, II, Math. Z., 1 (1918), 357-376, 6 (1920), 11-51; Werke, 215-
234, 249-289.
9. A. E. INGHAM, On two classical lattice point problems, Proc. Cambridge Phil. Soc.,
36 (1940), 131-138.
10. , The distribution of prime numbers, Cambridge Tract No. 30, 1932.
11. E. LANDAU, (i) Einftihrung in die elementare und analytische Theorie der algebrai-
schen Zahlen und der Ideale, Teubner, 1927. (ii) Uber die Anzahl der gitter-
punkte in gewissen Bereichen, Gott. Nachr., (1912), 687-771.
12. A. OPPENHEIM, Some identities in the theory of numbers, Proc. London Math. Soc.,
(2), 26 (1927), 295-350.
13. C. L. SIEGEL, Uber die Zetafunktionen indefiniter quadratischen Formen, I, II, Math.
Z., 43 (1938), 682-708; 44 (1939), 398-426.
14. E. C. TITCHMARSH, The Theory of the Riemann Zeta-Function, Oxford, 1951.
15. V. VENUGOPAL RAO, The lattice point problem for indefinite quadratic forms with
rational coefficients, J. Indian Math. Soc., 21 (1957), 1-40.
16. G. VORONOI, Sur une fonction transcendante et ses applications a la sommation de
quelques series, Ann. Sci. Ecole Nor. Sup., (3), 21 (1904), 207-267; 459-533.
17. G. N. WATSON, A Treatise on the Theory of Bessel Functions, 2nd Ed., Cambridge,
1948.
18. E. T. WHITTAKER and G. N. WATSON, A Course of Modern Analysis, 4tA Ed., Cam-
bridge, 1958.

This content downloaded from

105.156.110.139 on Thu, 02 Nov 2023 12:36:56 +00:00
All use subject to https://about.jstor.org/terms