You are on page 1of 67

Jay Jorgenson Serge Lang

Basic Analysis
of Regularized Series
and Products

Springer-Verlag
Berlin Wdelbup NewYork
--=w
Hong Koq Barcelona
Budapest
Authors
Jay A. Jorgenson
Serge Lang
Department of Mathematics Foreword
Yale University
Box 2155 Yale Station
New Haven. CT 06520. USA The two papers contained in this volume provide results on
which a series of subsequent papers will be based, starting with
[JoL 92b], [JoL 92d) and [JoL 931. Each of the two papers contains
an introduction dealing at greater length with the mathematics
involved.
The two papers were &st submitted in 1992 for publication in
J. mine angew. Math. A referee emitted the opinion: "While such
generalized products are of interest, they are not of such central
interest as to justify a series of long papers in expensive journals."
The referee was cautious, stating that this "view is subjective" , and
adding that he "will leave to the jud ement of the editors whether
%
to pass on this recommendation to t e authors". The recornmen-
dation, in addition not to publish "in expensive journals", urged
US to publish a monograph instead. In any case, the editors took
full responsibility for the opinion about the publication of our se-
ries "in expensive journalsn. We disagree very strongly with this
opinion. In fact, one of the applications of the complex analytic
properties of r e e i z e d products contained in our first paper is
to a generalization of Crank's theorem, which we prove in great
generality, and which appears in Math. Annalen [JoL 92bj. The
MathematicsSubjectClassification(l991): 1lM35.l lM41,llM99,30BSO,30D15, referee for Math. Ann. characterized this result as "important and
35P99,35899,39B99,42A99 basic in the field".
(Authors' Note: then is no MSC number for regularized products, but there should be.)
Our papers were written in a self-contained way, to provide a
suitable background for an open-ended series. Thus we always con-
sidered the possible alternative to put them in a Sprin er Lecture
ISBN 3-540-57488-3 Springer-Verlag Berlin Heidelberg New York
ISBN 0-387-57488-3 Springer-Verlag New York Berlin Heidelberg B
Note, and we are very grateful to the SLN editors and pringer for
publishing them.
This work is subject to copyright. All rights are reserved, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, re-use
of illustrations. recitation, broadcasting, reproductionon microfilms or in any other Acknowledgement: During the preparation of these papers,
way. and storage in data banks. Duplication of this publication or puts thereof is the first author received support from the NSF Postdoctoral Fel-
permitted only under the provisions ofthe GermPn Copyri%ltLaw of Ssptember 9, 1owsMg WS-S-65681 and from FBF grant DMS-93-07023.The
1965. in its cumnt version, and permission for use must dways be obtained from second author benefited from his visits at the Max Planck Institut
Springer-Verlag. Violations are liable for prosecution under the German Copyright in Bonn.
Law.
Q Springer-Verlag Berlin Heidelberg 1993
Printed in Germany
214613140-543210 - Printed on acid-free paper
Part I

Some Complex Analytic Properties


of Regularized Products and Series

Content

Introduction . . .... .. . . . .. ... ... . . . ...


.. . .. .. . .. . . . .. . .. 1
. ... .. . . . . .. . .
1. Laplace-Mellin Transforms .. . . . . ... . .. . . . 9
2. Laurent Expansion at s = 0,
. ..
Weierstrass Product, and Lerch Formula .. . ........ 3 0
. .. . . . . . .. . . . .. . . .
3. Expressions at s = 1 .. .. . . .. . . . . . .. .38
4. Gauss Formula . .... ..... .. .. . . .. . ...
. . . . .... .. ..... . . 4 8
.. . . . . . . . . .. . . . ..
5. Stirlin Formula . . . . . ... .. . ... . . . . . . .53
k
6. Hanke Formula . .... ... ... ... . . ... .. .. .
. .. .. .. . . . . . .. 65
7. Mellin Inversion Formula .................. ........
. .. 73
. .. ... . .. . .. . . . . .
Table of Notation . . . .. .. . . . . . . .. .... .85

Part I1

A Parseval Formula for Functions with a


Singular Asymptotic Expansion at the Origin
Content

. . . ..... . .. . .. . .. . ... .. .
Introduction . . . .. . . . . . . . . 9 1
. . .....
1. A Theorern on Fourier Integrals .... ... .. .. 93
2. A P a n e d Formula . ............. . ..... .....
.. . . 105
. .. . . ... . ...
3. The General P a t 4 Formula . . . . . . 109
. .
4. The ParseVal Formula for I,(a+ it) . . .... .. ... 114

Bibliography .................. .................. 119


Part I

Some Complex Analytic Properties


of Regularized Products and Series
Introduction
We shall describe how parts of analytic number theory and parts
of the spectral theory of certain operators (differential, pseudo-
differential, elliptic, etc.) are being merged under a more general
analytic theory of regularized products of certain sequences satis-
fying a few basic axioms. The most basic examples consist of the
sequence of natural numbers, the sequence of zeros with positive
imaginary part of the Riemann zeta function, and the sequence
of eigenvalues, say of a positive Laplacian on a compact manifold.
The resulting theory applies to the zeta and L-functions of number
theory, or representation theory and modular forms, to Selberg-like
zeta functions in spectral theory, and to the theory of regularized
determinants familiar in physics and other parts of mathematics.
Let {Ak) be a sequence of distinct complex numbers, tending to
infinity in a sector contained in the right half plane. We always
put Xo = 0 and Xk # 0 for k. 2 1. We are also given a sequence
{ak) of complex numbers. We assume the routine conditions that
the Dirichlet series

converge absolutely for some a > 0. If a, E Z>o for all k, we view


ak as a multiplicity of Xk, and we call this thespectral case. We
may form other functions, namely:
00
The theta series 8 ( t ) = C ake'x" for t > 0;
k=O
00
The neta function ((s) = akACs;
k=1
00
The Hurwitz zeta function C(s, z ) =
k=O
+
ak(z Xk)-';

The xi function ((8, z) = r(s)C(s,z), which can be written as


product will be characterized explicitly below in terms of the
the Laplace-Mellin transform of the theta function, that is sequence { Ak ) and appropriate conditions.
We keep in mind the following four basic examples of the spectral
case.
Example 1. T h e gamma function. Let Xk = k range over
the natural numbers. Then the zeta function is the Ftiemann zeta
function CQ, and the Hurwitz zeta function is the classical one
For the sequence { A k ) with ak = 1 for all k, consider the deriv- (whence the name we have given in the general case). The theta
ative function is simply

Putting s = 0 formally, as Euler would do (cf. [Ha 49]), we find


The corresponding Weierstrass product is that of the gamma func-
C'(0) = C - log A*. tion.
Example 2. The Dedekind zeta function. Let F be an
algebraic number field. The Dedekind zeta function is defined
Therefore if the zeta function has an analytic continuation at s = 0, for Re(s) > 1 by the series
then 00

may be viewed a s giving a value for the meaningless infinite product where a ranges over the (non-zero) ideals of the ring of algebraic
on the right. Similarly, using the sequence { A k + z ) instead of { A k } , integers of F, and Na is the absolute norm, in other words, the
we would obtain a value for the meaningless infinite product index Na = (a : a). Then

D(Z) = exp(-C(O, z)) = XI + I),


k=l

where the derivative here is the partial derivative with respect to where a, is the number of ideals a such that Na = k. This theta
the variable s. To make sense of this procedure in the spectral function is different from the one which occurs in Hecke's classical
case, under certain conditions one shows that the sequence { A k ) proof of the functional equation of the Dedekind zeta functions (cf.
-
-
also determines: [La 701, Chapter XIII). Of ~ m cthis , example extends in a nat-
ural way to Dirichkt, Hedp, Artin, and other Gaeries classically
- The regularized product associated to number fields. In these extesIlSions, ak is usually not
an integer.
D(z) = cP(') ~ ( z ) ,
Zaa functions a r k q frola representation theory and the theory
where P is a normalizing polynomial, and E(z) is a standard of automorphic ibctiew comtatutc au &BILlljOn of the present
,but we omit hss fmtba meattaat of them to avoid having
Weierstrass product ha zeros at the numbers -4 with mul-
3
tipliuty a,. The degree P and the order of the Weierstrass ex-g
to el rate on their more complicated definitions.
Example 3. Regularized determinant of a n operator. In The multiplication formula may be viewed as a special case of
this case, we let {Ak} be the sequence of eigenvalues of an operator. the Artin formalism treated in [JoL 92dJ. The Parseval formula,
In the most classical case, the operator is the positive Laplacian on which determines the Fourier transform of r'/r as a distribution
a compact Riemannian manifold, but other much more complicated on a vertical line will be addressed in the context of a general re-
examples also arise naturally, involving possibly non-compact man- sult in Fourier analysis in [JoL 92~1.Here we show that the other
ifolds or pseudo differential operators. Suitably normalized, the formulas can be expressed and proved in a general context, under
function D(z) is viewed as a regularized determinant (generalizing certain axioms (covering all four examples and many more com-
the characteristic polynomial in fhite dimensions). plicated analogues). We shall find systematically how the simple
expression 1/(1 - e-') is replaced by theta functions throughout
Example 4. Zeros of t h e zeta function. Let {pk}range over the formalism developed in this part. More generally, whenever the
the zeros of the Riemann zeta function with positive imaginary above expression occurs in mathematics, one should be on the look-
part. Let a t be the multiplicity of pk, conjecturally equal to 1. out for a similar more general structure involving a theta function
Put A; = pk/i. The sequence {Xi} is thus obtained by rotating associated to a sequence having a regularized product.
the vertical strip to the right, so that it becomes a horizontal strip.
A theorem of Cram& [Cr 191 gives a meromorphic continuation T h e Asymptotic Expansion Axiom. The main axiom is a
(with a logarithmic singularity at the origin) for the function certain asymptotic expansion of the theta function at the origin,
given a s AS 2 in 1; namely, we assume that there exists a se-
quence of complex numbers {p) whose real parts tend to infinity,
and polynomials Bp such that
which amounts to a theta function in this case (after the change of
variables z = it).
This fourth example generalizes as follows. Suppose given a se-
quence {Ak)such that the corresponding zeta function has an Euler The presence of log terms is essential for some applications.
product and functional equation whose fudge factors are of regular-
ized product type. (These notions will be defined quite generally In Example 1, the asymptotic expansion of the theta function is
in [JoL 92bl.) We are then led to consider the sequence {A;} de- immediate, since B(t) = 1/(1 - e-'). In Example 2, this expansion
fined as above. From $7, one sees that a regularized product exists follows from the consideration of $7 of the present part. In both
for the sequence {Ak}. We will show in [JoL 92b] that a regular- Examples 1 and 2, Bp is constant for all p, so we say that there are
ized product also exists for the sequence {AJ}. Passing from {Ak} no log terms.
to {A;} will be called climbing t h e ladder in the hierarchy of In the spectral theory of Example 3, Minakshisundaram-Pleijel
regularized products. (MP 491 introduced the zeta function formed with the sequence of
Basic Formulas. The example of the gamma function provides eigenvalues Ak, and Ray-Singer introduced the so-called analytic
a basic table of properties which can be formulated and proved torsion [RS 731, namely ('(0). Voros [Vo 871 and Cartier-Voros
under some additional basic conditions which we shall list in a [CaV 901 gave further examples and raults, dealing with a se-
=
' moment . The table ineludes quence of n u m h whose real part ten& to inffnity. We have found
The multiplication formula their axio91atizatian caPlcernigg the oomsponding theta function
useful. However, both artides [Vo 871 w d [CaV 901 leave some
-
* The Lerch formula
The (other) Gauss formula basic questions open in laying doan t h foundations of regularized
The Stirling formula products. V o w himself states: "In the present work, we shall
The Hankel formula not be concerned with rigorous p r d , which certainly imply addi-
The Mellin inversion formula tional regularity pmpcrtiea for the q w n c e (Ak)." Furthermore,
The Parseval formula. Cartier-Vorw have only certain spkiflc and special applications in
mind (the Poisson summation formula and the Selberg trace for- types, the formula is new as far as we know. In Example 3, the for-
mula in the case of compact Riemann surfaces). Because of our mula occurs in many special cases of the theory of analytic torsion
more general asymptotic expansion for the theta function, the the- of Ray-Singer and in Vome [Vo 871, formula (4.1). In Example 4,
ory becomes applicable to arbitrary compact manifolds with arbi- the formula specializes to a formula discovered by Deninger for the
trary Riemannian metrics and elliptic pseudwdifferential operators Riemann zeta function Cg (see [De 921, Theorem 3.3).
where the log terms appear starting with [DUG 751, and continu- Applications. Aside from developing a formalism which we
ing with [BrS 851, [Gr 861 and [Ku 881 for the spectral theory. In find interesting for its own sake, we also give systematically funda-
this case, the theta function is the trace of the heat kernel, and its mental analytic results which are used in the subsequent series of
asymptotic expansion is proved as a consequence of an asymptotic parts, including not only the generalization of Crarnhr's theorem
expansion for the heat kernel itself. mentioned above, but for instance our formulation of general ex-
In Example 4 for the Riemann zeta function, the asymptotic plicit formulas analogous to those of analytic number theory (see
expansion follows as a corollary of Cramhr's theorem. A log term [JoL 931). These particular applications deal with cases when the
appears in the expansion. The generalization in [JoL 92b] involves zeta function has an Euler product and functional equation. Such
some extra work. Some of the arguments used to prove the asymp- cases may arise from a regularized product by a change of variables
totic expansion AS 2 are given in 55 of the present part (especially I = s(s - 1). The Selberg zeta function itself falls in this category.
, Theorem 5.11), because they are directly related to those used to However, so far the Euler product does not play a role. A prema-
prove Stirling's formula. Indeed, the Stirling formula gives an as- ture change of variables z = s(s - 1) obscures the basic properties
ymptotic expansion for the log of a regularized product at infinity; of the regularized product and Dirichlet series which do not depend
in [JoL 92b] we require in addition an asymptotic expansion for on the Euler product.
the Laplace transform of the log of the regularized product in a
neighborhood of zero. Although, as we have pointed out, some special cases of our
formulas are known, many others are new. Our results and formulas
Normalization of t h e Weierstrass Product by t h e Lerch concerning regularized products are proved in sufficient generality
Formula. We may now return to describe more accurately our to apply in several areas of mathematics where zeta functions occur,
normalization of the Weierstrass product. When the Hurwitz zeta e.g. analytic number theory, representation theory, spectral theory,
function ((s, z) is holomorphic at s = 0, and all numbers ak are ergodic theory and dynamical systems, etc. For example, our Lerch
positive integers, there is a unique entire functions D(z) whose formula is seen to apply to Selberg type zeta functions not only for
zeros are the numbers -Ak with multiplicities ah, with a normalized Riemann
-- surfaces but for certain higher
- dimensional manifolds as
Weierstrass product such that the Lerch formula is valid, namely well.
log D(z) = -('(0, z), Furthermore, our general principle of climbing the ladder of reg-
ularized products applies to the scattering determinant associated
where the derivative on the right is with respect to the variable s.
As to the Weierstrass order of D, let po be a leading exponent
<
in the asymptotic expansion for B(t), i.e. Re(po) Re(p) for all p
such that Bp # 0. Let M be the largest integer < -Re(po). Then
-
-

+
M 1 is the order of D.
In the general case with the log terms present in the asymptotic previously.
expansion, the Hurwitz zeta function ((s, z ) may be meromorphic
at s = 0 instead of being holomorphic, and we show in 53 how to Our theory also applies to R u d e type zeta functions arising in
make the appropriate definitions so that a similar formula is valid. ergodic theory and dynamicd systems (for example, see [Fr 861 and
references given in that part).
In Example 1, this formula is the classical Lerch formula. In
Example 2, and various generalizations to E h c t i o n s of various Therefore, we feel that it is timely to deal systematically with
51. Laplace-Mellin Transforms
the theory of regularized products, which we find central in math-
ematics.
We first recall some standard results concerning Laplace-Mellin
For the convenience of the reader, a table of notation is included transforms. The Mellin transform of a measurable function f on
at the end of this part. (0,m) is defined by

0 .
The Laplace transform is defined to be

0
The Laplace-Mellin transform combines both, with the defini-
tion

We now worry about the convergence conditions. The next lemma


is standard and elementary.
Lemma 1.1. Let I be an interval of real numbers, possibly
infinite. Let U be an open set of complex numbers, and let
f = f (t, z ) be a continuous function on I x U. Assume:
(a) For each compact subset K of U the integral

1 f 4 dt
0 9

is uniformly convergent for z E Ii .


(b) For each t the function z H f(t, z ) is holomorphic.
Let
F(z)= 1 f(t, 2) dt.
Then the second partid &f satisfies the same two conditions as
f , the function F is holomorphic an U,and
F ( z )= 1 &f ( t ,); dt.
For a proof, see [La 851, Chapter XII, 81. We then have immedi-
ately:
Lemma 1.2. Special Case. Let z be such that Re(z) > 0 also provides a meromorphic continuation of the given integral.
and let bp E C. Then for Re(s) > 0 we have The next lemma shows how an asymptotic expression for a func-
0
tion f (t) near t = 0 gives a meromorphic continuation and an as-
ymptotic expansion at infinity for the Laplace-Mellin transform of
f (t). We first consider a special case.
Lemma 1.3. Special Case. Let f be piecewise continuous on
the integral being absolutely convergent, uniformly for (0, m). Assume:
(a) f (t) is bounded for t -+ m.
Re(z) 1 61 > 0 and Re(s) > h2 > 0.
+
(b) f(t) = bptP O(tRe(q))for some bp E C, p, q E C such that
General Case. For any polynomial B, let B(&) be the asso- Re(p) < Re(q), and t + 0.
ciated constant coefficient differentid operator. For Re(%)> 0, Then for Re(s) = a > -Re(p) and Re(%)> 0 the Laplaee-Mellin
p E C andRe(s+p) > O wehave integral
-
LM f (s, I) = dw (t)e-" ta
dt
-
t
converges absolutely, and for %(s) > -Re(q) the function LM f
The proof of Lemma 1.2 follows directly from an interchange of has a meromorphic continuation such that
differentiation and integration, which is valid for z and s in the
above stated region.
At this point let us record several very useful formulas. For any
z with Re(%)> 0 and any s with Re(s) > 0,
where for fixed z, s I+ g(s, r ) is holomorphic for Re(s) > -Re(q).
The only possible poles in s of LM f when Re(%)> 0 and when
+
Re(s q) > 0 are at s = -p - n with n E Z>O. - All poles are
simple, and the residue at s = -p is bp.
Proof. We decompose the integral into a sum:
where the path of integration in the second integral is such that
u/z is real. By expanding e-" in a power series about the origin,
we have
LM f (s, Z) = 1-(f (t) - bptp)e-" ta
:+d"
- e-"bpt8+P -
dt
t
+
which shows that the given integral can be meromorphically con-
! tinued to dl s E C and all r C. The integration by parts formula

(f(t) - b,tp)e-" t'


dt
7 + bp r(s + P).
The second integral, from 1 to oo, is entire in s, and converges Then for Re(s) = a > -Re(p) and &(I) > 0 the Laplace-Mellin
>
uniformly for all z such that Re(z) 6 > 0, and for all s such that integral
Re(s) is in a finite interval of R. Also, we have LM f(s, z) = lm- dt
f(t)c-zt ta
t
converges absolutely, and for Re(s) > -Re(q) the function LM f
has a meromorphic continuation such that

since

where for fixed z, s o g(s, z) is holomorphic for Re($) > -Re(q).


The only possible singularity of LM f when Re(z) > 0 and
So there remains to analyze the first integral, from 0 to 1. More +
Re(s q) > 0 are poles of order at most deg Bp + 1 at s = - p - n
generally, let q E C, and let g be piecewise continuous on (0,1] with n E Z>O.
-
satisfying
g(t) = O(tRe(q)) for t -+ 0. The proof is the same as in the special case invoking Lemma 1.1.
Then the integral
In brief, the presence of the logarithmic term bp(t) introduces
+
a pole at s = -p of order deg Bp 1, and Lemma 1.3 explicitly
describes the polar part of the expansion of the Laplace-Mellin
integral near s = -p.

is obviously holomorphic in z E C and Re(s) > -Re(q), by Lemma The preceding lemmas give us information for
1.1. This concludes the proof of Lemma 1.3. 0
g(t) = o(tRe(9)llog tim).
We are now going to show the effect of introducing logarithmic
terms. For p E C, we let Bp denote a polynomial with complex We end our sequence of lemmas by describing more precisely the
coefficients and we put behavior due to the term bp(t)tP.
bp(t) = Bp(logt). Lemma 1.4. Let f be piecewise continuous on ( 0 , ~ ) Assume:
.
We then let B(d, ) be the corresponding constant coefficient partial (a) There is some c E R such that f (t) = O(ect) for t -,oo.
differential operator.
- Lemma 1.3. General Case. Let f be piecewise continuous +
(b) f(t) = bp(t)tp 0(tRe(4jlog tim) ~ i t hbp(t) = B,(IO~t),
on (0, oo ) . Assume: k ( p ) < W q ) , m f Z>O
- and t 4 0.
A -

L
I
(a) f (t) is bounded for t oo. Then LM f (s, 2) is meromorphic in each variable for
p (b) f (t) = bp(t)tP + O(tRe(q)1 log t lm) for some function Re(z) > e and Re(s) > -rRe(q)
bp(t) = Bp(logt) E C[logt],
-
such that Re(p) < Re(q), m f Z>O,and t + 0.
when necessary, the horizontal half line going from -00 to 0. We
Proof. We split the integral: define the open set:
U L = the complement of the above half lines in C.
If all Xk are real and positive, then we note that U L is simply C
minus the negative real axis Rco.
-
where We also suppose given a sequence A = {ak) of distinct complex
numbers. With L and A, we form the asymptotic exponential

I1(s,p, Z) = I(
0
dt
f (t) - bp(t)tp)e-zita-t '
polynomials for integers N 2 1:

Throughout we shall also write

Similarly, we are given a sequence of complex numbers

The specified poles in s are going to come only from 12.That is:
I1(s,p,z) is holornorphic for z E C and Re($)> -Re(q). with
12(s,p,z) has a meromorphic continuation given by expanding
e-" in its Taylor series and integrating term by term to get
increasing to infinity. To every p in this sequence, we associate a
polynomial Bp and, as before, we set

bp(t ) = Bp(log t ).
I3 (s, 2) is holomorphic for Re(z) > c and all s E C. The theorem We then define the asymptotic polynomials at 0 to be
follows. 0
We now consider an i&te sequence L = f Xk) of distinct corn-
plex numbers satisfying:
DIR 1. For every positive real number c, there is only a finite
number of k such that &(Ak) 5 c. " m a t Perhaps most, applications the exponents p are real. Be-
cause t ere are significant cases when the exponents are not nec-
We use the convention that Xo = 0 and Xk # 0 for k 2 1. essarily real, we lay the foundations in appropriate generality. We
Under condition DIR 1 we delete from the complex plane C the define
horizontal half lines going from -00 to -Ak for each k, together, m(q) = max deg Bp for Re(p) = R.e(q).
Let C ( T ) be the algebra of polynomials in TP with arbitrary Define n(q) to be
complex powers p E C. Then, with this notation,

Throughout we will use p' to denote an element in the sequence {p}


We let f be a piecewise continuous function on (0,oo) satisfying with next largest real part for which BPt is not zero. In particular,
the following asymptotic conditions at infinity and zero. this means that
AS 1. Given a positive number C and to > 0, there exists N
and 'h > 0 such that n ( d >= max deg Bp.
Re(p)<Re(,)

I f(t) - Q ~ ( t ) <l ~ e for -t 2 to.


~ ~

AS 2. For every q, we have -


Theorem 1.5. Let f sat* AS 1 and AS 2. Then LMf
has a meromorphic continuation for s E C and z E UL. For
eacb z, the h c t i o n s LMf (s,z) has poles o d y at the points
+
-(p n) with b, # 0 in the asymptotic expansion off at 0. A
+
pole at - ( p n) has order at most n(p) + 1. In the s~ecialcase
f(t) - PJt) = O,(tRe(d(logtlffl(q))
for t + 0, -.
when the asymptotic expaasion at 0 h& no log terms, the poles
where, as indicated, the implied constant depends on q. are simple.
Often we will write AS 2 as Proof. We first do the analytic continuation in z, for Re(s) large.
We subtract an exponential polynomial QN from f(t), using the
asymptotic axiom AS 1,to write

Also, we will write PJt) = P, f (t) to denote the dependence on f .


The crucial condition is AS 2 and, in practice, is the most difficult
to verify.
Let po be an exponent of the asymptotic expansion AS 2 with
smallest (negative) value of Re(po). Let M be the largest integer
< -Re(po ). We call M the reduced order of the sequence {Ak).
- has reduced order 0.
For instance, the sequence Z>o
The case when all ttk are ~on-neg&i\rektegers will be called the
spectral case. In such a situation one can view the coefficients
ak as determining a multiplicity in which the element Xk appears
in the sequence L.
The case when all the polynomials B are either 0 or constants,
denoted by bp, will be cailed the speed case. In such a situation
P, is a polynomial with complex exponents and without the log
terms.
It is immediate that the terms involving the gamma function have Theorem 1.6. Let f satisfy AS 1 and AS 2. Then for every
the above stated meromorphy properties. In particular, note that z E U Land s near 0, the function LM f (s, z) has a pole at s = 0
the appearance of the term +
of order at most n(0') 1, and the function LM f (s, z ) has the
Laurent expansion

requires us to restrict z to UL.So, at this time, it remains to study


the integral involving f - QN. Since f satisfies AS 2, then so does
f - QN , and the integral where, for each j < 0, Rj(O; z) E C is a polynomial of degree
s -Re(po).
Proof. For any C > 0, choose N aa in AS 1so that f - QN is
bounded as t + oo. Since

is absolutely convergent for b ( s ) > -Re(po) and Re(z) > -C if


f(t) - QN(t) = ~ ( e - ~for~t )-+ m.
I
By taking N sufficiently large, one can make C arbitrarily large,
so this process shows how to meromorphically continue LM f (s, z) it suffices to prove the theorem for f -QN, or, equivalently, assume
as a function of z. Next we show how to continue meromorphically that f(t) is bounded as t + oo.

For this we can apply Lemma 1.4 to f - QN, which shows that
for &(I) > -C the function

is meromor hic in C. In addition, Lemma 1.4 shows that the only so we include Re(p) = 0 in the sum. We decompose the integral
P
possible po es are as described in the statement of the theorem.
Note that the set of these poles is discrete, because the values p + n
into a sum:

tend to infinity. This completes the proof of the theorem. 0

We shall use a systematic notation for the coefficients of the Lau-


rent expansion of LM f (s, I) near s = so. Namely we let Rj(so;I)
-
be the coefficient of (s so)j, so that

We shall be particularly interested when so = 0 or so = 1. Also,


' when necessary, we will express the dependence of the coefficients
on the function f by writing

The first two integrals are holomorphic at s = 0 and Re(%) > 0.


By Lemma 1.2, the third integral is simply Theorem 1.8. The function LM f satisfies the equation
I

Proof. This is immediate by differentiating under the integral


sign for &(s) and Re(%)sdlciently large, and follows otherwise
by analytic continuation. 0

Corollary 1.9. For any integer j, we have


where h o ( t )is entire in z. This proves the theorem, and (4) g'ives us
an explicit determination of the polynomials R,(O; Z)for j < 0. O
In particular, for the constant terms, we have
The constant term &(so; z ) is so important that we give it a
special notation; namely, for a meromorphic function G ( s ) we let
CT,,,,G(s) = constant term in the Laurent expansion
Corollary 1.10. The Mellin transform
' That is,
I
Corollary 1.7. Define has a meromorphic continuation to s E C whose only possible
1 poles are at s = - p such that bp # 0.
Cr(s,t )= -LM f ( s ,Z) and e j ( s ,z ) = LM f (s,t ) .
r(s)
Then, in the special case, Cs(s, z ) is holomorphic at s = 0 for + -
Proof. Write f = fo f l with fo = Q N and f l = f Q N with
z E U L and N sufficiently large so that we can apply AS 1 to fl with C > 0.
Then M fo is entire in s. As for f l , we have, for any q,
Ci(0,I ) = CT8=oF~(s; 2) +TR-IJ(O;2 ) .
Proof. In the special case n(O1)= 0 , so LM f ( s ,z) has a pole at
s = 0 of order at most 1. Since

we have

The first integral in (5) is hdomorphic hrr Re($) > -Re(q), by


Lemma 1.4. By the construction of fi, the second integral in (5)
from which the corollary follows. 0 is entire in s. The sum in (6) is memmorphic for d s 6 C with
possible poles at s = -p. With ail this, the proof is complete. 0
Given the sequences A and L as defined in AS 1,one can con- AS 3. Given 6 > 0, there exists an a > 0 and a constant C > 0
sider: such that for all N and 0 < t 5 6 we have
a Dirichlet series

We shall see that the three conditions AS 1, AS 2 and AS 3


on the theta series correspond to conditions on the zeta function
a theta series describing the growth of the sequence {Ak) and also the sequence
{ak). The condition DIR 1 simply states that the sequence {AL)
converges to infinity, in some weak sense. The following condition
gives a slightly stronger convergence requirement.
DIR 2.
a reduced theta series (a) The Dirichlet series

and, more generally, for each positive integer N,


converges absolutely for some real a. Equivalently, we can I
I
a truncated theta series say that there exists some a 0 E R>osuch that I

lakl = O(lxklUo)for k -+ 00.


I
(b) The Dirichlet series
We shall assume throughout that the theta series converges abso-
lutely for t > 0. From DIR 1it follows that the convergence of the
theta series is uniform for t 2 6 > 0 for every 6. We shall apply the
above results to f = 6 with the associated sequence of exponential
polynomials QN6 being the natural ones, namely
N-I
converges absolutely for some real a. Specifically, let a1 be
a real number for which
k= 1
-
A

Note that the above notation leads to the formula

Theorem 1.11. Assume that the theta series


The absolute convergence of the theta series B(t) describes a type of
convergence of 6(N, new infinity that is uniform for all N. The fol-
lowing condition describes a type of uniformity of the asymptotics
of 8 ( t ) near t = 0.
is uniformly bounded for all N. By letting N approach oo, we can
satisfies AS 1, AS 2 and AS 3, and assume that Re(Xk) > 0 interchange limit and integral, by dominated convergence, to show
for d l k > 0. Then for Re(s) > a , +
that for a 2 cr e > a ,

in the sense that the series on the right converges absolutely, to which completes the proof of the theorem. 0
M6(s). L particular, the convergence condition DIR 2(a) is
satisfied for the Dirichlet series +
Remark 1. If -Re(po) is an integer, then M 2 is the smallest
integer m for which the Dirichlet series
00

k=l

Proof. For a = Re(s) > a,we have +


converges. If -&(PO) is not an integer, then M 1 is the smallest
such m. In any event, we let mo be the smallest such m. The
exponent &(po), which comes from AS 2, is not independent of
the integer mo, which comes from DIR 2. In fact

Indeed
From AS 1we have

which goes to zero as C + os. Similarly, by AS 3 we have that


The second integral on the right is holomorphic for

+
The first integral on the right has its &st pole at s po = 0, so
: which is uniformly bounded if u > a. Therefore, we have that for at -pe, whence me > -F&(B). The first inequality in (6) follows
+
a 3 a e > a, the difference from the minimality of mo since Re(po) 0.
The following axidition on the v e n c o L uires that, beyond
"PI.
what is stated in the convergence condition DI 1, the sequence
X k approaches infinity in a sector.
DIR 3. There is a fixed r > 0 such that for all k sufficiently Let us apply this inequality to x = c2lAklt and then sum for k 2 N
large, we have to obtain, for any t > 0,

Equivalently, there exists positive constants Cl and C2 such that


for all k sufficiently large,

Now if we let
Q=uo+ul,
Theorem 1.12. Let (L, A) be sequences for which the associ-
and
a ted Dirichlet series
e$
k= 1

then (7) becomes


00

k=l

satisfies the three convergence conditions DIR 1, DIR 2 and


T)IR 3. Then the theta series

which establishes the asymptotic condition AS 3.


satisfies AS 1 and AS 3. In order to establish AS 1, we need same preliminary calcula-
tions. First, note that by choosing ca < c2, we can write
Proof. Let us first show how AS 3 follows. Directly from DIR 2
and DIR 3 we have, for some constants cl and c2, the inequalities

Now let
CN = min(c3IXk 1) for k 2 N,
and

r*

"Wote that for any x > 0, there is a constant c = c(uo + 01) such By DIR 1and DIR 3 we have
-that
Z ~ ~ + ue-'l < c. lim CN = oo,
N-roo
which gives the bound
and since
1
I
1
-CN 5 -c3(XkJ for k
2 2
>N,
we can write

Therefore, which shows that AS 1holds, and completes the proof of the the-
orem. 0

Remark 2. The convergence condition DIR 1 is assumed as


part of the asymptotic condition AS 1. Theorem 1.11 asserts that
By combining (8) and (9) we conclude that if the theta series 00

I satisfies the asymptotic conditions AS 1,AS 2 and AS 3, then


so, in particular,
lim tLN' = 0.
N+oo

Therefore, there exists to < oo such that


Further, the sequences (L, A) satisfies the convergence condition
tkN' 5 to for all N. DIR 2(a) and, by Corollary 1.10, ((s) has a meromorphic contin-
uation to all s E C. Theorem 1.12 states that if the two sequences
(L, A) satisfy the three convergenceconditionsDIR 1,DIR 2 and
Note that for t > to and any k 2 N we have DIR 3, then the corresponding theta series satisfies the asymptotic
conditions AS 1 and AS 3. As previously stated, the condition
1 1 AS 2 is quite delicate and, in practice, is the most difficult to ver-
>
(c31XkI - f ~ ) t (c31Xtl- Z C ~ ) t o ify. In 57 we will show, under additional meromorphy and growth
condition hypothesis on (, AS 2 follows from the three convergence
conditions DIR 1,DIR 2 and DIR 3.
CalXk] -~ C N
2 log IXk la1
c31.\rl - f C N

Therefore, for t > to,


$2. Laurent expansion at s = 0, Weierstrass product, Let D(z) be an entire function of strict order 5 m with the
a n d t h e Lerch formula same zeros as Dm,L (z), counting multiplicities. Then there exists
<
a polynomial PD(z) of degree m such that
In this section, we consider the case when ak is a non-negative
integer for all k, which we define to be t h e spectral case. The
corresponding Dirichlet series is then also called spectral, or t h e
spectral zeta function. We develop some ideas of Voros [Vo 871, We shall describe conditions that determine the polynomial PD
especially about his formula (4.1) which we formulate in a gen- uniquely.
eral context below as Theorem 2.1. Our arguments are somewhat
different from those of Voros, who makes "no pretence of rigour". Suppose the three convergence conditions DIR 1, DIR 2 and
Basically, we want to make sense out of an infinite product of a DIR 3 are satisfied, so we have the spectral zeta function and the
sequence of complex numbers L = {Ak), counted with multiplic- numbers
C$
00
ities A = {ak), which satisfies certain conditions. The results in
$1 and the above definitions establish a line of investigation via a C(n) = for n 2 m.
zeta function. In this section we indicate the line along Weierstrass k=l
products, and we show how the two approaches connect. We then have the power series of D(z) at the origin coming from
the expansion
We recall the construction of a Weierstrass product. Let X be a
non-zero complex number, let m be an integer 2 mo, and let
(1) log D(z) = a. log z + PD(z) + C (-1)"-'~(n) ;
00
zn

Em(z,A) = (1 - ): exp 00
n=m

= ao log z + co + x ( - l ) n - ' c n -
zn
n
or also n=1

where the coefficients cn satisfy

We shall work under the assumptions DIR 2 and DIR 3, and, for
>
m mo, we define the Weierstrass product If D(0) # 0, meaning a0 = 0,then D(0) = eCO,and we have
00
- log [D(z)/D(O)] = c ( - 1 ) " ~ . -.zn
n=l
n
From this we see that the coefficients of PD(z) can be determined
By the elementary theory of Weierstrass products, Dm,~(z)is an from the expansion (2) and the numbers
- ?tire function of strict order 5 m, that is

log JDm,L(z)l= O(lzlm) for IzI + 00.


Since D m a , L ( ~is)an entire function, the logarithmic derivative

w e use the adjective "strictn to avoid putting an e in the exponent


on the right hand side.
is memmorphic, and further derivatives (dldz)' log Dm,,,~ ( z are
) If we let
also meromorphic, immediately expressible as sums .that are of
>
Mittag-Leffler type as follows. For an integer r 1 let us define

we can write Tr(z) as a Mellin transform:


We have trivially
d
-T,(Z) = -rTr+l (z).
dz
Since the logarithmic derivative transforms products to sums, as
part of the standard elementary theory of Weierstrass products we The convergence of the integral and sum is uniform for Re(%)suf-
obtain for each r 2 1 the expansion: ficiently large.
We shall ap ly Theorem 1.8 with f = 0 and LM f = [ to obtain
%
the following t eorem, which we call the Lerch formula.
Theorem 2.1. In the spectral case, assume that the theta func-
tion 8 satisfies the asymptotic conditions AS 1, AS 2 and AS 3
with the natural sequence of exponential polynomials (4). Then
there exists a unique polynomial PL(z) ofdegree 5 mo - 1 such
that if we define

where N = mo - r - 1. In (3) we have the usual binomial coefficient


then for aJI z E C with Re(z) sufjliciently large, we have

where Hence exp(-CT,,o[(s,z)) has an analytic continuation to all


IIn(r) = (-1Inr(r + 1). ..(r + n - 1). z E C to the entire function DL(z). In particular,
Note that for n > 1we have
D;/Dt(z) = -&CT,,o((s, z) = CTaZl((s, z) = &(l;z).

Let us define IIn(r) for negative n through this recursive relation. Proof. If we count Xo = 0 with multiplicity ao, we find for Re(s)
-
- large and r >
mo,
-
Because of the absolute convergence of the theta series qt),we
have, by Theorem 1.8, for all z such that Re(z+Ak) > Ofor all k,
n
abdanyr2m0,
where, as before, conditions AS 1, AS 2 and AS 3. Then L" and OL1l(t)necessarily
satisfy these conditions and

We now look at the constant term in a Laurent expansion at s = 0. From this, we immediately have
If r > n o , then the series

and
DL(%)= DL' (%)DL"( z ) .

converges. At s = 0 the gamma function has a first order pole with A particular example of such a decomposition is the case when L'
residue 1. Note that the n,(O) = 0. So, Theorem 1.6 gives us a +
is a finite subset of L, in which case DL.(%)= n [ ( z X k ) e T ] ; see
bound mo - 1 for the degrees of the polynomials in z occuring as Remark 2 below.
coefficients in the negative powers in s of the Laurent expansion It now becomes of interest to determine in some fashion the coef-
of t ( s ,z ) . From this, we conclude that (&)'((s, z ) is holomorphic ficients of the polynomial PL( z ) ,and for this it suffices to determine
near s = 0. With all this, we can set s = 0 and obtain the equality
P ~ ) ( o )We
. define the reduced sequence Lo to be the sequence
which is the same as L except that we delete Xo = 0. Then

@LO(t)= @(t)- a0
and, as discussed in Fkmark 1, we have
from which we obtain, using (3),
Dmo , L O ( % ) = z-'ODmo ,L(%)-

Theorem 2.2. The polynomials PL and PL,,as defined in The-


Hence, there is a polynomial P L ( z )of degree < r - 1 such that orem 2.1, are equal. If we let & be the partial derivative with
respect to the second variable, then

P , ! ~ ( o=) -B~CT.=o[(s,0 )
for Re(%)sufficiently large, which completes the proof of the theo- for0 5 r 5 mo - 1. That is
rem upon setting r = mo. 0
-
-

We call D ( z ) = DL(%)the regularised product associated to


the sequence L + z . In particular, D(0) is the r e p l a r i d product
" of the sequence L.
Remark 1. Assume L can be written as the disjoint union of the
sequences L' and L" where L' satisfies the convergence conditions
DIR 1, DIR 2 and DIR 3, and Ov(t) satisfies the asymptotic
The canonical product of D m o , L o (is~ )such that the Taylor series Indeed, we have the immediate relation
of log D m o , L o (at
~ )the origin begins with powers of z which are at
l&t rmo. Hence,
coming from the definitions and the expansion

This proves the theorem. 0


Remark 2. The normalization DL(z) that we have given is
the most convenient one for the formalism we are developing. One Example 2. Let CQ be the Rianann zeta function, and let {pk}
may also define the characteristic determinant DL(z) by the be the sequence of zeros in the critical strip with Im(pk) > 0. Let
condition
X i = pkli. It is a corollary of a theorem of Cram& [Cr 191 that
the theta function
O(t) = C ale-":'

so Rl ,((O; Z)is the coefficient of s in the Laurent expansion of ((s, z) satisfies AS 2. The Lerch formula for the sequence {Xi} (with mul-
at s = 0. In the special case we have that tiplicities ak) then specializes to a formula discovered by Deninger
[De 921, Theorem 3.3. In [JoL 92bJwe extend C r a d r ' s theorem
to a wide class of functions having an Euler product and func-
tional equation, including the L-series of a number field (Hecke and
Note that log D ~ ( zand
) log DL(z) differ by an obvious polynomial Artin), L-functions arising in representation theory and modular
in z coming from the Laurent expansion of r ( s ) at s = 0. For forms, Selberg-type zeta funtions and L-functions for Riemann sur-
+
example, if L is a finite sequence, then DL(z) = n ( z At). Also, faces and certain higher dimensional manifolds, etc. The present
let us record the formula section therefore applies to these functions as well, and thus the
Lerch formula is valid for them. Note that Deninger's method did
not give him the analytic continuation for the expressions in his
formula, but such continuation occurs naturally in our approach.

which, again, holds only in the special case.


Example 1. Theorems 2.1 and 2.2 provide a general setting for
the some classical formulas. Voros [Vo 871 gives examples, including
the simplest case which concerns the sequence Xk = k and the
- gamma function (see his Example c). Theorems 2.1 and 2.2 contain
as a special case the classical Lach hrmula, which states that

and 00
$3. Expressions at s = 1. Proof. Let us write
In this section we return to general Dirichlet series, as in $1,
meaning we are given sequences (L, A) whose Dirichlet series that
satisfy the convergence conditions DIR 1, DIR 2 and DIR 3, with
associated theta series eL,A= 8 that satisfies the three asymptotic
conditions AS 1, AS 2, and AS 3. With this, we will study the
xi-function

near s = 1. Referring to AS 2, let us define the principal part


of t h e t h e t a function Poe(t) by

Recall that, by definition, m(0) = max deg Bp for Re(p) = 0 so Using AS 2, we have that the integral in (3) is holomorphic for all
z E C and Re(s) > 0, as is the integral in (5), and the integral in
(4) is holomorphic for all s E C and Re(%) > 4.Combining this,
the stated claim has been proved. 0
(2) e(t) - Poe(t) = { O(l log t lrn(0)) inin the special case
the general case, Corollary 3.2. The function

as t approaches zero.
Theorem 3.1. Let C and N be related as in AS 1. Then the
function is meromorphic at s = 1 for all z with singularities that. are
simple poles at z = -Ak. Also, the residue at z = -Ak is equal
to ak.

Pmf. bx~&Weikmt&epfoofofTheorem 3.1 by taking C,con-


sequently N, sufficiently large. 0
We now study the singular term
has a holomorphic continuation to the region
40

that appears in Corollary 3.2. To do so, we expand e-zt in a power Theorem 3.4. Near s = 1, the Hurwitz xi function ((s, z) has
series and apply the following lemma. the expansion
R-n(l)-l(l; 2)
Lemma 3.3. Given p, there is an entire function hp(z) such
that as s approaches 1, such that: ((~9t) = ( 8 - 1)nW+l +-..+R-l(l;Z)
(5 - 1)
+ &(l; z) + O(s - 1),
(a) Special Case: where:
(a) For j < 0, Rj(l; z) is a polynomial of degree < -Re(po ); in
fact, the polar part of ((s, z) near s = 1 is expressed by

(b) &(l; z) = CT8,1((s, z) is a meromorphic function in z for


all z E C whose singulrzrities are simple poles at z = -X
with residue equal to ak. Furthermore,

(b) General Case:


In the special case, the expansion of as,z) near s = 1 simplifies
to

with

Let (L, A) be sequences of complex numbers that satisfy the


From Lemma 3.3, we can assert the existence of an entire func- three convergence conditions DIR 1,DIR 2 and DIR 3 and such
tion h(z) such that the singular term can be written as that the associated theta function

satisfies the three asymptotic conditions AS 1,AS 2 and AS 3.


We define the regularized harmonic series R(=) associated t.o
The followin theorem, which is the main result of this section, (L, A) to be
then follows %rectly hTheorem 3.1 and Lemma 3.3.
Remark 1. Theorem 3.4 states that the regularized harmonic Theorem 3.5. The function Fq(s,z) has a meromorphic con-
series associated to (L, A) is a meromorphicfunction in z whose sin- tinuation to the region
gularities are simple poles at z = -Ak and corresponding residues
equal to ak. Further, by DIR 2 and DIR 3, Theorem 1.8 and
Theorem 1.12, we have, for any integer n 2 mo, the expression
Proof. Note that we can write (8) as

F&, = region of meromorphy

If all numbers a k are integers, then one can assert the existence of
a meromorphic function D(z), unique up to constant factor, which
satisfies the relation
(9) j
0
je(t) - Pqe(t)l e-"t8
dt
-
t
Re@) > -Re(q), all z

(6) D1/D(z) = R(z).


T h e Spectral Case. In this case, with ak E Z>o for all k,
Theorem 3.4 asserts the existence of an entire function D(z), unique
(10) + i
1
[8(t) - 'QN8(t)]e"tt'-
dt
t
all s, Re(%)> -CI

up to constant factor, such that (11) + J [ Q ~ 8 ( t-) Pq8(t)]e-"V- dtt all s, z E UL.
1
(7) D1/D(z) = CT,,l((s, z).
For (9), the asserrtion of meromorphy on the right follows from
Further, we have, by the Lerch formula (Theorem 2.1), condition AS 2 and Lemma 1.3; for (10) the assertion follows by
Lemma 1.4, picking N and C as in AS .l. As for (ll), the integral
irhe Basic Identity: is a sum of integrals of elementary functions which we now recall.
First we have
R(z) = CT8,1(~(s, z) = -dzCT8=o(~(s,z) = D',/DL(~).
We shall normalize the constant factor in (7) so that

Next we shall give another type of expression for the singular so that
term, which also gives an expression for CT8,1(L(s, z) leading into
the Gauss formula of the next section.
Consider the function
M

For the special case, when bp is constant, we have

d set F = Fo,which is especially important among the Laplace-


-
ellin transforms of the functions 8 Pq8. Following the results
d techniques in $1, we shall study am analytic continuation of
(3, z) and then compare Fg.(s,z ) with ((9, z).
add for the general case we have the expansion replacing bp by Corollary 3.7. The constant term of ((s, z) at s = 1 is given
B#(&), namely: by

where 00

F(1,z) = 1
0
[e(t) - Po8(t)] e-"dt.

From this, the conclusion stated above follows since the integral in The constant term on the right is obtained simply by multiply-
(11) is simply a sum of terms of the form given in (12), (13) and ing the Laurent expansions of r ( s +p) and z-"-p and is a universal
(14). 0 expression, meaning an expression that depends solely only on Bp
for Re(p) < 0. A direct calculation shows that there exists a poly-
Theorem 3.6. For any q and for all (s, z) in the region of +
nomial gp of degree at most deg Bp 1 such that
meromorphy, we have:
a) In the special case,
The possible pole of r(s+p) at s = 1accounts for the possibility of
deg gp exceeding deg Bps For convenience of the reader, we present
these calculations explicitly in the special case.
b) in the general case, Let p E C and write

and

Proof. For Re(s) large, one can interchange the sum and integral
in the definition of Fq(s,z) and use that I=O
..
Then

from which we obtain the equation


.. The rest follows from the definition of [(s, z) and analytic contin-
$
! uation. 0

By taking q = 0 in Theorem 3.6 and writing Fo(s,z) = F(s, z),


we obtain the following corollary.
Corollary 3.8. In the special case, & ( l ; z ) has the integral hyperbolic Riemann surface X. For convenience, let us briefly recall
the definition of the Selberg zeta function.
expression Let e x ( t ) be the trace of the heat kernel corresponding to the
hyperbolic Laplacian that acts on Cw functions on X; hence, Bx(t)
is the theta function associated to the sequence L of eigenvalues of
the Laplacian. The sequence A counts multiplicities of the eigen-
values. It is well known that

where
/
w

~ ( 1Z ), = [ e ( t )- ~ ~ e ( e-"dt.
t)]
where b-1 is a constant that depends solely on the genus g of X,
0
namely
Ll = 27r(2g - 2).
Example 1. Suppose we are in the spectral case and also that
B(t) is such that the principal part PoO(t) is simply L i l t . Then The iogarithmic derivative of the Selberg zeta function is defined
the equations in Theorem 3.6 become by the equation

Z i / Z x ( s ) = (2s - 1) [Ox ( t ) - b-l k(t)]e-'("-')'dt


0
where
where k ( t ) is a universal function, independent of X (see [Sa 871).
From Theorem 3.6, the definition of the Selberg zeta function, and
the example above, we have the relation

we can write

This relation was uscd in [JoL 92d) to prove that the Selberg zeta
function satisifies Artin's formalism ([La 701).
+
P In this case, the integral expression for &(l; z ) reminds one of
7 tHe Gauss formula for the 1 rrithmic derivative of the gamma
"d
function, which will be studie in the next section.
Example 2. The regularid harmonic series is simply related
to the classical Selberg zeta function Z x ( s ) associated to a compact
48

$4. Gauss Formula Theorem 4.1. There is a polynomial Sw(z) of degree in z


Next we show that a classical formula of Gauss for I"/r can
be formulated and proved more generally for the regularized har-
monic series. As before, we let Po8 denote the principal part of an with coefficients in C[log w ](w) such that for any w E C with
as~~mptotic expansion at 0. Define Re(w) > 0 and Re(w) > ma{-&(Ar
k
+z)),

If B(t) satisfies the three asymptotic conditions, then it is immediate


that e-"B(t) also satisfies the three asymptotic conditions. As
before, if the principal part of the theta function is
+
In the special case, Sw(z)E C(w)[z] C(w) log w[z].
Proof. From Theorem 1.8, we have for sufficiently large Re(s)
and Re(z), while viewing w as fixed, the equalities
then

Recall from Theorem 1.8 that

To compute the constant term in the expansion at s = 0, we can


substitute s = 0 in the Laplace-Mellin integral of 8, - PoB., thus
so, in particular, we have getting the desired integral on the right hand side of the formula in
the theorem. As for the integral of POezwe can use the expression
(1) for the principal part of 8, to get
In the spectral case, we have, by the Lerch formula,
-
- R(z + w ) = D;/DL(z + w ) .
Finally, recall that C ( T ) is the algebra of polynomiah in TP with
"r arbitrary complex powers p E C. In fact, we obtain an explicit f o d a for Sw(r),namely
With all this, we can follow the development leading to Corollary
3.8 and state the following theorem, which we call the general
Gauss formula.
This expression shows that Sw(z) is a polynomial in z, of degree In the case when ak E Z for all k, this allows us to write the
< -Re(po ), with coefficients in C (w)[log w]. Recall. that in the integral Iw(z) from Corollary 4.2 as
special case, all B, are constants, and, for any p, the expression

where Hw is a meromorphic function on C, uniquely defined up to


a constant factor. Following Theorem 2.1, we define S t ( % )to be
lies in C(w) + C(w) log w. With this, the proof of the theorem is the integral of S,(z) with zero constant term, so Sw(z) = &S$(I).
complete. 0 Define HW(z)by the relation

Remark. A direct calculation shows that there exists a poly-


+
nomial B; of degree at most deg B, 1 such that
Then Hw(z) is the unique meromorphic function of z such that

and
The possible pole of I'(s + p) at s = 0 accounts for the possibility Hw(0) = D(w).
of deg B; exceeding deg B,. Using the relation
Furthermore, we have

In the spectral case, meaning ak E Z>O for dl k, Hw(z) is entire.


one can hope to express Sw(z) in terms of the polynomials Bi. The function I,(%) will be studied in $3of [JoL 92~1,from a Fourier
H~wever,even in the special case, such an expression is quite in- theoretic point of view.
volved.
Example 1. We show here how the classical Gauss formula is
Corollary 4.2. For fixed w E C with Re(w) > 0 and a special case of Theorem 4.1. Let

and for z E C with Re(z) sufficientlylarge, the integral and


a(n) = 1 for all n.
The theta function can be written as

as a function of z, is holomorpbc in z and has a meromorphic


continuation to all z E C with poles at the points Xk w in+ The principal part is simply
+
L w and corresponding residues equal to ak.

Proof. Immediate from Theorem 3.4(b) and Theorem 4.1. 0


SO no = - 1 and pl = 0. This allows us to compute the polynomial 55. Stirling Formula
Sw(z) and obtain the equation As in previous sections, we work with sequences (L, A) that sat-
isfy the three convergence conditions DIR 1, DIR 2 and DIR 3,
and such that the associated theta series satisfies the three asymp-
totic conditions AS 1, AS 2 and AS 3. With this, we consider
the asymptotic behavior of certain functions when Re(z) + oo, by
Therefore, by combining the Example from $2 and Theorem 4.1, which we mean that we allow Re(%)-,oo in some sector in the right
we have half plane. The point of such a restriction is that in such a sector,
Re(z) and It.1 have the same order of magnitude, asymptotically.
Specifically, we shall determine the asymptotics of
-CT,,o((s, z) as x = Re(a) + oo.
- 11~ - ~ ' d-t7 - log w. These asymptotics apply in the spectral case to
1-e-' t
0

Now set w = 1 to get and, therefore, can be viewed as a generalization of the classical
Stirling formula.
To begin, we will study the asymptotics of
00

~ ( sz), = J ~(t)e-"t8-
dt
t
for Re(z) oo.
which is the classical Gauss formula. 0
Fix Re(q) > 0 and, as before, let Pq8 be as in AS 2. Let us write
Example 2. Let (L, A) be the sequences of eigenvalues and
multiplicities associated to the Laplacian that acts on smooth sec-
tions of a power of the canonical sheaf over a compact hyperbolic
Riemann surface. Then one can combine Example 2 from $3, The-
orem 4.1 and the Lerch formula to establish the main theorem of
[DP 861 and [Sa 871, without using the Selberg trace formula. In
brief, this theorem states that the Selberg zeta function Zx(s) is
expressible, up to universal gamma-like functions, in terms of the
regularized product DL(s(s - 1)).

Recall that
m(q) = max deg Bp for Re(p) = Re(q).
The terms in (1) and (2) are handled in the two following lemmas.
54

Lemma 5.1. Let x = Re(%)and fix q with Re(q) > 0. Then Lemma 5.2. We have
Jz(0, z) = O(e-"1s) for x + w.

Jl(O, z) = ~ ( x - ~ ~ ( ~ ' ( l o for


~ xx )+
~ m.
~~)) Proof. From AS 1 , J2(s,z) is holomorphic at s = 0 and
00

Proof. Recall that J2(0,Z) = J e(t)e-" dt


-
t '
e(t) - pqe(t) = o(tRe(q)l1ogtlm(q))for t -+ O.
from which the lemma follows, since e(t) = O(ect)for some c > 0
and so, for x sufficiently large,
Since Re(q) > 0, Jl(s, z) is holomorphic at s = 0 and 00

1
yielding the stated estimate. O
0
As for J ~ ( sz),
, recall that
For any complex number so, we have the power series expansion

r(s)
-x8 = (kE - so,k)
C.(SO,~S
e
l=O
,! x)' (S - So)'
(- log
) Lemma 5.3. We have

from which we have, by AS 2,


Proof. Simply write

and as in Lemma 5.2, note that, for any s,

Combining the above three lemmas, we have established the


following theorem, which we refer to as the generalized Stirling's
as x -, w. This completes the proof of the lemma. 0 formula.
Theorem 5.4. Suppose q is such that &(q) > 0. Let Lemma 5.5. For any p E C, we have

Thenforx 4 00, ,

+
CTa=o<(s,I) = Bq(r) ~ ( x - ~ ~ ( q )x)"(q)).
(lo~

It is im~ortantto note that Theorem 5.4 shows that the asymp- ~ - ~ (= (-1)" and ~ ( n=) -(-7).
n )-
totics of ~ T , = ~ ( ( S , Zas) & ( z ) -+ oo are governed by the as>m& n! n!
totics of 9(t) as t -+ 0. Also, in the spectral case, the Lerch formula
(Theorem 2.1) applies to give One can prove Lemma 5.5 by using the power series expansion
(4) and the formula
- log D L ( ~=) CTa=o(~(s,4,
and, hence, Theorem 5.4 determines the asymptotics of the reg-
r(s + 1) = s ~ ( s ) and r t ( i ) = -7,
ularized product log DL(%)as Re(z) -+ oo and, consequently, the which comes from the expansion
asymptotics of the the characteristic determinant
log D L(z) as &(z) -+ oo.

Remark 1. In Remark 1 of 34 we defined the polynomials B,+ Proposition 5.6. In the special case, we have, for &(q) > 0,
by the formula -

CTa=, Bp(&) [r!ap


:)] = z-P8;(log r).

Using these polynomials, one can write and so


Bq(z)= z-PBi(logr),
Re(p)<Re(q)
so Theorem 5.4 becomes the statement that
I) =
CTa=o<(~, C I-PB;(log t ) + o ( x - ~ ~ ( ' ) ( x)"")
~o~ )
-
-- WP)
<Re( IT)

as x = Re(%)-+ oo. This restatement emphasizes the point that for Re(z) + m.
:' the asymptotics of 9(t) as t -+ 0 determine the asymptotics 0
CT,&(s, z) as Re(z) oo. -+ Example 1. Assume L is such that
To further develop the asymptotics given in Theorem 5.4, one
can use the following lemma.
Remark 2. Lemma 5.7 will be applied to an integral of the which is defined for Re(z) > 0, has a holomorphic extension to
form include a neighborhood of z = 0.

Proof. We have

which is a product of e-' and an integral of exactly the type con-


sidered in Lemma 5.7.
Lemma 5.8. Let [q] = [Re(q)].Then the function is of class from which the stated result immediately follows. 0
~ [ g l -and
~ has the Taylor expansion
.Having studied the transforms of J1 and J2, we now deal with
the transform of J3.
Lemma 5.10. Given p and Bp, there is a function hi, mero-
morphic in s and entire in z, to be given explicitly below, such
that for Re(z) > 0, we have
with the Taylor polynomial g, of degree < Re(q) - 3.

Proof. From (1)we have, by interchanging the order of integra-


tion,

and, in particular,
We now apply Lemma 5.7 to the function

noting that, by AS 2,

where
-- -
hp(z) = CT8=0hp(s, z).
-
as t -,0. With this, the proof of the lemma is complete. D
+
Proof. It suffices to assume that -Re(s p) is a large, non-
integral, real number, from which the result follows by analytic
Lemma 5.9. The function continuation in s. For this, note that
From this we have are given by universal formulas, depending linearly on El;, hence
on Bp.
As is shown in the proof of Lemma 5.3, we can write

where an

The lemma follows by putting


1
By interchanging order of integration, we can write

and Therefore, Lemma 5.9 applies to imply the existence of a function


fa, holomorphic in a neighborhood of 0, such that

so we can write
R e m a r k 3. A direct calculation shows that there exists a poly-
+
nomial Bp# of degree at most deg Bp 1 such that

(7) CTa=0Bp(a.) [ Ata+~-l


sin[n(s +p)] I = zp-l B? (log z). With all this, we can combine Lemmas 5.8,5.9, and 5.10 and obtain
the following theorem.

The possible zero of sin[?r(s+p)j at s = 0 accounts for the possibility Theorem 5.11. Let {gg), h, {B?), {f,} and {h,} be the
of deg Bf exceeding deg Bp. Further, by combining Remark 1 and above sequences of polynomials and entire functions for those p
Lemma 5.10, we arrive at the formula for which Bp # 0. Then for each q with Re(q) > 3
-

i
4
e-zuu-pB;(log u)du = zp-' ~ f ( l zo) ~ hp(z). +
As is clear from (6), the function hp doeti depend on the choice of
a. We note that the power series hp(z) and the polynomials B f
$6. Hankel Formula
as the real variable t approaches zero from the right.
The gamma function and the classical zeta function are well
- we have that
Example 2. In the case L = Z>O known to satisfy a Hankel formula, that is they are representable
as an integral over a Hankel contour. The existence of such a for-
mula depends on the integrand having at least an analytic contin-
uation over the Hankel contour, and expecially having an analytic
continuation around 0. Indeed, the integrands in these classical
Hankel transforms are essentially theta functions. Unfortunately,
as t approaches zem. In particular, the sequence {p} is simply theta functions cannot always be analytically continued around 0.
Z>- and all polynomials Bp have degree zero. Since For instance, C c-"" cannot, although c~~~ can. Thus the ana-
logues of the classical Hankel transform representations are missing

I =(4)"
log t
- t '
in general.
However, we shall give here one possible Hankel type formula ex-
pressing the Hurwitz xi function associated to the sequences (L, A)
as a complex integral of the regularized harmonic series. Observe
we have, by using the Lerch formula (Theorem 2.1), the equation
that the Hankel formula which we prove here is different from the
classical Hankel representation of the gamma function or the zeta
function. We note that such a formula was used by Deninger for
the Riemann zeta function (see 93 of [De 921).
As in previous sections, let (L, A) be sequences d o s e Dirichlet
series satisfy the three cower ence conditions DIR 1,DIR 2 and
7
Let q -+ co in Theorem 5.11. Usin the absolute convergence of
(8), we get (9). In order to deduce (9 from Theorem 5.11, we have
used that, upon letting q approach infinity, the power series (8)
8
DIR 3, and whose associate theta function satisfies the asymp-
totic conditions AS 1, AS 2 and AS 3, We suppose that the
converges in a neighborhood of the origin. In general, questions sequence L = {Ak}is such that
concerning the convergence of the above stated power series, which
are necessarily questions concerning the growth of the coefficients
b,, must be addressed. For now, let us complete this example by
combining the above results to conclude that Since we deal with arbitrary Dirichlet series ((9) = C akA;,' it is

i" - - 1 + qt).
e - t u log I'(u)du = logt
t 1-et
not the case in general that there is a regularized product whose
logarithmic deriviative gives the constant term CTaZI[(s, 2 ) ; in-
deed, this exists only in the case when ak Z for all k. However,
important a plications will be made to the spectral case, meaning
when ak E $>o for all k, in which case such a regularized product
1

This formula verifies calculations that appear in [Cr 191 (see also DL exists. T h ,for this case, we record here once more the
- f3oL 92b)). Bask Mentity:
Remark 4. The formal power series arising from the asymp-
$ totic expansion in Theorems 5.4 and 5.11 (letting Re(() -+ oo) are
interesting beyond their truncations mod ~ ( t l q l - ' ) . In important
applications, and notably to itive elliptic operators, these power All the formulas of this section involving
series are convergent, and del? e entire functions. This is part of the
theory of Voltma operators, c.f. [Di 781, Chapter XXIII, (23.6.5.3).
from z to any -Ak is at least cm-O1 with a suitable small constant
may then be used with R(z) replaced by D;/DL(z) in the appli- c. From D I R 2(a) we have that
cations to the spectral case.
Recall from D I R 2, D I R 3 and Theorem 1.8 that for any integer
n > uo we have the formula
so, if IAkI < 2m, we have for those k for which IXJ < 2m, the bound
lak( = CmeO,for a suitable large constant C. With all this, the
first sum can be bounded by

To begin, let us establish the following general lemma.


Lemma 6.1. With assumptions as above, for every sufjticiently If lzl = Tm, we can write this bound as stating
large positive integer m, there exists a real number Tm with
m 5 Tm 5 m + 1 and a real number u2 such that

This establishes that O:R(z) hrrs polynomial growth on the (in-


Proof. For every sufficiently large positive integer m, let us write creasing) circles centered at the origin with radius T,. From the
expression
m

one has that 8:R(t) is bounded for r E R>owith r sufficiently


large. Upon integrating the function 8:R(z) n times dong a path
consisting of R>o and the circle 11 = Tm, we conclude that R(r)
itself has polynomial growth on the circles Iz 1 = Tm. Indeed, from
If'we restrict z to the annulus m < 1
11 < m + 1, we have the (2) and the fact that a0 2 -Re(po), we have
estimate
IR(z)I = O(lzlu~++'n+l)(~~+l)
) for IzI =Tm, and as T, -+ oo.
This completes the proof of the lemma. 0
which is bounded independent of m provided n 2 uo. It should noted that the proof of Lemma 6.1 explicitly constructs
-.

By the convergence condition DIR 2(b), we have that


+
the &e of u2. Indeed, since one can take n < uo 1, 0 2 can be
written in terms of 00 and u1.
Having established this pr+minary leama, we can now proceed
with our Hankel formula for the regularized harmonic series R. Let
6 be a fixed positive number and assume
so there are O(m"+l) terms in the first sum. Also, we conclude
the existence a sequence {T,) of real numbers, tending to infinity
+
and with m < Tm < m 1, such that if lzl= Tm,then the distance
Combining the residue theorem with Theorem 3.4(b), we get
Let C6 denote the contour consisting of:
- the lower edge of the cut from -a, to -6 in the cut plane
u L;
- the circle S6, given by w = 6ei4 for 4 ranging from -a to
a;
- and the upper edge of the cut from -6 to -co in the cut Now let T -+ oo along the sequence {T,) that was constructed
planeUL. in Lemma 6.1. With this, the integral over ST will go to zero as
If G(z) is a meromorphic function in U L ,
then +
T approaches infinity if s is such that &(s) > 6 2 1. For these
values of s, the limit of the sum above is the Hurwitz zeta function,
and the theorem is proved. O

For the remainder of this section we will study the analytic con-
tinuation of the integral given in Theorem 6.2. The proof of Theo-
rem 6.2 shows that the problem in extending the region for which
Symbolically, let us set the integral conver es arises because of the asymptotic behavior of
R(r - t) for -Re($? large. In any case, the following lemma takes
care of the integral over the circle S6.
Lemma 6.3. For any fixed 6 sufticiently small, and any z such
that &(z) is sufficiently large, the integral

When taking the sum of the two integrals on the negative real axis,
it is of course understood that for the second integral, we deal with
the analytic continuation of G(w) over the circle S6.
is holomorphic for all s E C.
We call the result of the following theorem the Hankel formula. Proof. Since Re(z) is sufficiently large, the integrand is abso-
Theorem 6.2. Let s E C be such that &(s) > a 2 + 1. Then lutely bounded on S6.
for any z with Re(%)sufficiently large, we have
Lemma 6.3 implies that the analytic continuation of the Hankel
< +
formula to any s for which Re($) 6 2 1 must take account the
asymptotic behavior of R(z - w) for fixed z and as -&(w) -t oo.
We can use the eneralized Stirling's formula, Theorem 5.4, to
f
extend the Hanke formula in much the same way AS 2 is used to
extend the Hurwitz xi function (see,in particular, Lemma 1.3 and
- PT'OOf. k t T 6 R>o be SU& that 1 41# T for d l k, and let C6.T Theorem 1.5).
denote the contour consisting of: Lemma 6.4. For fixed 6 > 0, for any z with Fk(z) sufficiently
- the lower edge of the cut from -T to -6 in UL; large and for any q and p for which &(q) > Re(p), the integral
- the circle Sa, given by w = 6ei4 for 4 ranging from -a to
a;
- the upper edge of the cut from -6 to -2' in UL;
- and the circle ST,given by w = ~ e ' 4for ) ranging from a
to -7r.
is holomorphic for Re(s) = a > -RR(q) + 1. we have, by Lemma 6.3,

Proof. For any T > 1, the integral over the segments from -T
to -6, both the lower cut and the upper cut, is finite for all s since
the integrand is absolutely bounded. By Theorem 5.4, we have

& [+ 11 w z - t ) - Bq(z - t)] t-'dt


Lemma 6.4 yields a similar result for the integral of R - B,. 0
Next, we let 6 approach zero and show that the integral over Sa
will go to zero.
Lemma 6.6. For asy q aad p for which Re(q) > Re(p), and for
any t. such that &(z) is sufficiently large, we have

which converges if -a - Re(q) < -1, as asserted. 0

Proposition 6.5. Let q be such that Re(() > 1, and Jet 6 be


a suffjciently small positive number s u d that lXk 1 > 6 for dl k.
Then for any z with Re(z) sufficiently large,
and

lim
6+0
[([+ n -
I
[R(z - t) - Bq(z t)] log t dt

Proof. If Re(z) is sufficiently large, the functions

R(z - t) and R(z - t) - B,(z - t)


are bounded as t approaches zero. The second term is bounded
by t-Re(q) as t 0. Therefore, the first integral is bounded by a
multiple of
-

Proof. Write the integral over C6 as the sum of integrals over


S6and the line segments between -00 and -6. Since
which approaches zero as 6 approaches zero. The same estimate
proves the second assertion. O
Combining Lemma 6.4, Proposition 6.5 and Lemma 6.6, we have
Theorem 6.7. With notation as above, 97. Mellin Inversion Formula
So far we have considered sequences (L, A) that satisfy the three
convergence conditions DIR 1, DIR 2 and DIR 3 and whose
associated theta function 9 satisfies the three asymptotic conditions
AS 1,AS 2 and AS 3. Fkom these assumptions we then derived
properties concerning various transforms. We now want to perform
other operations, so we reconsider these axioms ab ovo. Especially,
we shall consider the inverse Mellin transform which gives 8 in
terms of ('. Throughout this section we shall assume
+ 6lim4 0 CTaSo-
2x2

Recall that Theorem 1.12 proved that the convergence condi-


One can view the content of Theorem 6.7 as a type of regularized tions DIR 1, DIR 2 and DIR 3 imply the asymptotic conditions
form of the Fundamental Theorem of Calculus. One should note AS 1 and AS 3. With this, Theorem 1.11 applies to show that for
the presence of the term -7 - logt in Theorem 6.7, which is a Re(s) sufficiently large we have
function that also appeared in our general Stirling formula; see
Proposition 5.6 of the previous section.
To conclude, let us note that if instead of considering the Hur-
witz xi function we would have studied the Hurwitz zeta function
we would have obtained the following result.
As previously stated, the asymptotic condition AS 2 is quite del-
Theorem 6.8. With notation as above, icate. In this section, we will show how, by imposing additional
assumptions of meromorphy and certain growth conditions on <,
r
the asymptotic condition AS 2 follows. To be 'n, we need the fol-
lowing lemma which addresses the question o convergence of the

( 1 + 1 ) [R(z- t ) - B,(z - t)] (- logt) dt


partial series
N-1
21ri
k=l
to ('(s) as we let N -+ oo.
Lemma 7.1. Assume that the sequences (L, A) satisty the con-
vergence conditions DIR 1, DIR 2 and DIR 3, and let a0 be
as in DIR 2(a). For each N 2 1 let
-
The pmof of TBeoren 6.8 f o b w s thst of Theorem 6.7 with the
only change being the use of the formula
$N = suplarg(Xk)l for all k 2 N,
so $ N < for ad N sutficiently large.
in place of (3). (a) If s = u + it is such that
74

then for N sufEcientJy large then we have shown that

which establishes part (a). Note that if a > 00, then, by the
(b) For all s in any fixed compact subset of the half plane convergence condition DIR 2(a),
Re(s) = a > ao,
the convergence
which shows that. the upper bound in (a) approaches zero a. N
approaches m, for fixed s with Re($) sufficiently large. As for (b),
N -+00
I
lim ((5) -
k=l
akX;' = 0, since +N is bounded, if s lies in a compact subset K of the half
plane Re(s) > 00, then l t l is
~ bounded
~ independent of N for all
s E K. This completes the proof of the lemma. 0
is uniform.
Let cp be a suitable function, which will be appropriately char-
acterized below. For any a E R, let &(a) be the vertical line
Proof. If we write logXk = loglXkl + iarg(Xk), we have the Re($) = o in C. Under suitable conditions on cp which guarantee
bound the absolute convergence of the following integral, we define the
vertical transform V,cp of cp to be

From Stirling's formula (see Theorem 5.4 and, specifically, Example


1 of 55), one sees that on vertical lines L(u), the gamma function
has the decaying behavior

( s ) = 0 ( e C ) for every c with 0 c c < */2,


If we define the absolute zeta to be
(2) asd (s/+ oo,

where, aa indicated the implied wmtant de on c. Previously


we studied the ~ e d transform
n frds
and, from eorem 1.10, we have
and N-1
where For N sufficiently large, take t,$N as in Lemma 7.1, and let us write
00 +
s = a iu. With this we have the bound

We shall now study the inversion formula


I

which is valid if
a > -Re(Xk) for all k.
The inversion formula (3) is essentially a standard elementary in- +
is the L1-norm of eIu1*~I'(u iu) on the vertical line L(u) By the
version obtained for each individual term from the relation convergence condition DIR 3 and (2), we conclude there exists a
constant C, independent of N ,such that

The classical proof of (4) comes from an elementary contour in-


tegration along a large rectangle going to the left, using the fact If t > 6 > 0, (6) can be written as

The relation (4) is then applied by putting u = Xkt and summing


over k. More precisely:
Proposition 7.2. Assume that C satisfies DIR 1, DIR 2 and If o > 0 0 , ~ g ) ( utends
) to zero when N -. oo. This shows that
DIR 3, and let a0 be a s defined in DIR 2(a). Then for ev- the theta senes
ery 6 > 0, the associated theta series converges absolutely and
>
uniformly for t 6 > 0 and

converges absolutely and uniformly for t 2 6 > 0, as asserted in


Proof. Let g = V,C be the vertical transform of C with a > 0 0 . the statement of the proposition. 0
By (4) we have
Remark 1. Proposition 7.2, in particular the inequality (6),
shows that the three convergence conditioxis DIR 1,DIR 2 and
DIR 3 implies the asymptotic condition AS 3. This provides
another proof of the first part of Theorem 1.12.
Since AS 2 was used to show that MO(s) has a meromorphic more difficult to prove this polynomid growth, and there are cases
continuation, it is natural to expect some type of meromorphy con- where it remains to be determined exactly what is the order of
dition on ( in order to prove that the vertical transform V,( sat- growth in vertical strips. The proof of polynomial growth depends
isfies the asymptotic condition AS 2. Independently of any a , we on functional equations and Euler products in the classical cases.
define the domain of V, and denote it by Dom(V), to be the We shall determine conditions under which we get AS 2 for
space of functions 9 satisfying the following conditions: V,cp. Immediately from the restricted owth of cpr on vertical
V 1. y is meromorphic and has only a finite number of poles and Cauchy's formula, we have:
I"
lines V 1, the asymptotic behavior of (p on horizontal link V 3,
in every right half plane.
Lemma 7.3. Let cp E Dom(V), and l e i d < a be such that yl?
V 2. vl? is L1-integrableon every vertical line where vl? has has no pole on L ( d ) and L(a). Let R ( d , u) be the strip (the
no pole. infinite rectangle) defined by the inequalities

V 3. Let a1 < a 2 be real numbers. There exists a sequence


{Tm),with m E Z and
and let {-PI be the sequence of poles of (pr.Then
Tm + oo if m-, oo and T, -, -oo i f m - t -w,
such that uniformly for u E [al,u2], we have

To continue, let us analyze the sum

Very often one has the simpler condition:


V 3'. y(s)l?(s) 0 for Is( -, w and s lying within any
vertical strip of finite width. which was obtained in Lemma 7.3. For this, let us write
Even though the condition V 3' is satisfied in many cases, it is
necessary to have V 3 as stated. The reason for condition V 3 is
that we shall let a rectangle of integration tend to infinity, and we
need that the integral of cpr on the top and bottom of the rectangle
tends to 0. Throughout, it is understood that when we take the
vertical transform, we select a such that cpr has no pole on L(u). Let dp = -ord-,[cpr] and consider the Laurent expansion
The vertical transform V,cp of cp depends on the choice of a, of S = -p:
- course, and we shall determine this dependence in a moment.
Remark 2. In the more standard cases, such as zeta functions
, of number fields or modular forms, or easier kinds of Selberg zeta
functions, there is no difficulty in verifying the three conditions
V 1, V 2, V 3, or, usually, V 3'. In fact, functions cp of this
kind have usually at most polynomial growth in vertical stri s,
70 their product with the gamma function decreases exponenti ly S
in vertical strips. In certain other interesting cases, it may be
Then from (7) and (8) we see immediately that there exists a poly- Theorem 7.5. Let (L,A) be sequences that satisfy the three
nomial Bp of degree dp such that convergence conditions DIR 1,DIR 2 and DIR 3, and assume
that the associated zeta function

As in $1, given a sequence {p) of complex numbers, ordered by


increasing real parts that tend to infinity, and given a sequence of
polynomials {BPIfor every p in the sequence, we define is in Dom(V). Then the associated theta series

As before, we put
satisfies the asymptotic conditions AS 1,AS 2 and AS 3. The
m(q) = maxdeg Bp for Re(p) = Re(q). asymptotic expansion of AS 2 is given by (9) and Theorem 7.4
with cp = C.
With this, we can combine Lemma 7.3 and (9) to obtain: Proof. he or em 1.12 shows that if (L, A) satisifies the three con-
vergence conditions, then the associated theta series satisfies the
Theorem 7.4. Let cp E Dom(V), and let {-p) be the sequence asymptotic conditions AS 1 and AS 3. Theorem 7.4 shows that
of poles of yI'. Let u be a positive, sufficientlylarge, real number the meromorphy assumption on C, namely that C is in Dom(V),
such that neither cp nor I' have poles for Re(s) 2 u. Then the implies AS 2.
vertical transform V,cp of cp satisfies the asymptotic condition
AS 2, or, brieffy stated,
To conclude this section, we will use Theorem 7.4 to show how,
given sequences whose theta series satisfies the asymptotic condi-
tions, one can construct a new sequence with the same property.
The following theorem follows immediately from the definitions and
the holomorphy of the exponential function.
+
Proof. If we let s = ut it with u' a large negative number, then Theorem 7.6. Let z E C be fixed and assume that (L, A) is
we have the estimate such that the associated theta series satisifies the three asymp-
totic conditions AS 1, AS 2 and AS 3. Then the sequence
+
(L z, A) satisfies the three asymptotic conditions AS 1,AS 2
and AS 3.
-
-
where IlyI'lll,,~ is the L1-norm of cpr on the line L(u1). By V 2,
Pmej Tke theta k b i o r t asm&kd to (L + z,A) is simply
this integral is finite. Now simply combine Lemma 7.3 and (9), and e-"&,,~(t), hence the proof follows.
let t 0 to conclude the proof. 0
Theorem 7.7. Let r > 0 and assume that (L, A) is such that
With all this, we can now summarize the connection between the associated theta -es s a t i d e s the three apymptotic condi-
the convergence conditions that apply to sequences (L, A) and the tions AS 1, AS 2 and AS 3. Then the theta series associated
asymptotic conditions that apply to the associated theta series. to
({A:''}, A) = (L~I', A)
Theorem 7.8. Let (Ll, A , ) and (L2, A2) be sequences such
satisfies satisfies the three asymptotic conditions AS 1, AS 2 that the associated theta series satisify the three asymptotic
and AS 3. If the exponents {p} in the asymptotic expansion conditions AS 1, AS 2 and AS 3. Assume there exists an
of BrtA near t = 0 are real, then so are the exponents in the E > 0 such that for all j and k, we have
asymptotic expansion of eLIIr,A near t = 0.

Proof. Since C(s) is meromorphic as a function of s, so is C(rs),


and we can apply Theorem 7.4. The reality statement is immediate or, in other words, the tensor product series satisfies DIR 3.
from the proof of Lemma 7.3. Further, assume 'there exists some N such that the truncated
zeta functions
Let us consider two zeta functions (1 and (2 corresponding to
the sequences (LI ,Al ) and ( 4 , A2), respectively. Let (L3, A3) be
the tensor product, which we define so that L3 is the family of are in Dom(V). Then the theta series associated to the tensor
all products
product (L3,A3) satisfies satisfies the three asymptotic condi-
tions AS 1,AS 2 and AS 3. If the exponents in the asymptotic
L3 = {Ak.\>}with Xk E L1 and A> E k ; expansion AS 2 are real for O1 and 02, then they are also real
for 03 = @ 82.
while A3 is the family of all products {aka$), so then the zeta
function C3 is simply the product Proof. Since B1 and 82 satisify AS 1,it is immediate that O3 sat-
isfies AS 1. Also, it is immediate that the tensor product (L3, A3)
satisifes the convergence conditions DIR 1 and DIR 2 since the
product of two absolutely convergent Dirichlet series is absolutely
convergent. Therefore, by Corollary 1.10, we know that the zeta
When written in full, the zeta function (3 is functions (1 and axe meromorphic in C. By assuming that the
tensor product series satisifies DIR 3, Proposition 7.2 applies to
show that 03 = 01 882 satisifes AS 3. Finally, by assuming that the
truncated zeta functions are in Dom(V), we may apply Theorem
7.4 to conclude that 83 satisfies AS 2.
In the case of real exponents for the asymptotic expansions of 81
and 82, we know, again by Corollary 1.10, that the poles of (1 and
and the associated theta series 03 reads C2 are of the form -(p + n) with real p and n E Z. Hence, by the
proof of Lemma 7.3, the exponents of the asymptotic expansion for
83, which are the poles of C3, are also real, thus concluding proof
of the theorem. 0

.= To study 6 and 83, we shdt apply Lcmma 7.1. For this, at d&e
Example 1. Let L be the sequence of eigenvalues associated
the truncated zeta function to be to a Laplacian that acts on Coofunctions on a compact hyperbolic
Riemann surface X. The parametrix construction of the heat ker-
N-1 nel shows that L satisfies the three asymptotic conditions AS 1,
00
-
AS 2 and AS 3. From Theorem 7.6 we have that L 1/4 satisfies
the three asymptotic conditions, and Theorem 7.7 implies that the
sequence
4
~
- = {,/-I ={~k}
also satisfies the three asymptotic conditions. The sequence TABLE OF NOTATION
Because of an accumulation of conditions and notation, and their
112 + ,Ki - ,/zip = (112 + irk} use in the current series of papers, we tabulate here the main ob-
jects and conditions that we shall consider.
is1 precisely the set of zeros { p } with Im(p) > 0 for the Selberg We let L = {Ak} and A = {ak} be sequences of complex num-
zeta function associated to X. Note that the general Cram& the- bers. To these sequences we associate various objects.
orem proved in [JoL 92b] also proves that the sequence d m
satisfies the three asymptotic conditions AS 1, AS 2 and AS 3. A Dirichlet series or zeta function:
Essentially the same argument, applied in reverse, holds for gen-
eral zeta functions, such as those associated to the theta function
as in CramCr's theorem ([JoL 92bl).
and, more generally, for each positive integer N, the truncated
Example 2. The Dedekind zeta function of a number field Dirichlet series

The sequences L and A (or Dirichlet series) may be subject to the


satisfies all three conditions DIR 1, DIR 2, and D I R 3, and following conditions:
lies in Dom(V). Here ak is the number of ideals a with Na = k.
Therefore, by Theorem 7.5, the associated theta function D I R 1. For every positive real number c, there is only a finite
number of k such that Re(Ak) c.<
DIR 2.
(a) The Dirichlet series

satisfies the three asymptotic conditions, especially AS 2. Note


that this theta function is different from the theta function used
in the classical (Hecke) proof of the functional equation. A similar converges absolutely for some real a. Equivalently, we
remark of course holds for the L-series. can say that there exists some a 0 E R>o such that
Iakl = 0(lAklu0) for k -, oo.
(b) The Dirichlet series

converges absolutely for some real a. Specifically, let


u1 be a real number for which
We define mo to be the smallest integer 2 1 such that We are also given a sequence of complex numbers

DIR 3. There is a fixed c > 0 such that for all L sufficiently increasing to infinity. To every p in this sequence, we associate a
large, we have polynomial Bp and we set

bp(t) = Bp(logt).

We then define:
Equivalently, there exists positive constants Cland C2
such that for all k, The asymptotic polynomials at 0:

A theta series or theta function:


We define

m(q) = max deg Bp for Re(p) = Re(q)

a reduced theta series and


n(q) = maxdegB, for Re(p) < Re(q).
00

Let C ( T )be the algebra of polynomials in TP with arbitrary com-


plex powers p C. Then, with this notation, Pq(t)E C[log t ](t).
and, more generally, for each positive integer N, the truncated A function f on (0,oo) = R>o may be subject to asymptotic
theta series conditions:
AS 1. Given a positive number C and to > 0, there exists N
and K > 0 such that

< KC-"' for t 2 to.


If(t) - Q ~ ( t ) l
&
9 The asymptotic exponential polynomials for integers N 2 1:
AS 2. For every q, we have

f ( t )- P'(t) = 0(tR.(~)l
logtlm(q') for t -+ 0.
We will write AS 2 as

AS 3. Given 6 > 0, there exists an a > 0 and aconstant C >0


such that for all N and 0 < t 5 6 we have

Part I1
Given a polynomial Bp a direct calculation shows that there
exist polynomials B;, $ and B f of degree 5 degBp 1 such +
that: A Parseval Formula for Functions with a
Singular Asymptotic Expansion at the Origin

= 9 - I B: (log z )

The possible pole of I'(s + p) at s = 0 or s = 1 and the possible


zero of sin[x(s + p)] at s = 0 accounts for the possibility of deg B;,
deg g p , or deg B f exceeding deg Bp.
Introduction
We shall determine the Fourier transform of a fairly general type
of function cp which away from the origin has derivatives that are
of bounded variation on R and are in L1(R), but at the origin the
function has a principal part which is a generalized polynomial in
the variable s and log x, namely

where {p) ranges ovei a fmite number of complex numbers with


Re(p) < 0, for all p, Bp is a polynomial, and rn is some positive
integer. Thus the Fourier transform is determined as a distribution,
and more generally as a functional on a large space of test functions
also to be described explicitly. Alternatively, we may say that we
are proving the P a r s e d fonnula for such a function cp and its
Fourier transform.
Aside from the P a r s e d formula havihg intrinsic interest in pure
Fourier analysis, it arises in a natural way in analytic number the-
ory, in the theory of differential and pseudo differential operators,
and more generally in the theory of regularized products as devel-
oped in [JoL 92aJ.
In the so-called "explicit formula2 of number theory, one proves
essentially that the sum of a suitable function taken wer the primes
is equal to the sum of the Fourier-Mellin transform taken over the
zeros of the zeta function. Cbically, only very special cases were
given (see Ingham [In321), and Weil was the fkst to observe that
the formula was valid on a rather large space of test functions, and
could be expressed as an equ functionals. The sum over
theprimgoiackteerrah(Lt whit& ~BBCR&S b an inte-
gral of the teat function against the logarithmic derivative of the
gamma function, taken over a vertical line Re(s) = a [We 52). Weil
proved whet amounted to a Parsed formula, by determining what
amounted to the Fourier t r d o r m of the Iogarithmic derivative of
the gamma function on a vertical line in an explicit form. Weil's
functional was reproduced with some additional details (making
use of some general results concerning gened Schwarz distribu-
tions) in [La 701. However, the form in which Weil (and [La 701)
left the functional at infinity still required what appeared as too $1. A Theorem on Fourier Integrals
complicated arguments to identify it with the classical forms in This section is preliminary, and proves some lemmas on Fourier
the classical special cases. Barner reformulated the Weil functional integrals which will be used in the next section.
in a more practical form [Ba 811, [Ba 901 and also extended the We recall the Fourier transform
domain of validity of the formula. Our P a r s e d formula includes
as a special case the formulas of Weil and Barner for the gamma
function.
In spectral theory or in the theory of regularized series and prod-
u c t ~there
, arises a regularized harmonic series R and a regularized
product (or regularized determinant) D. As a corollary of the Par- We shall be concerned with Fourier inversion. For f E L1(R)and
s e d formula, we determine the Fourier transform of R, and of the A > 0 we define
+
logarithmic derivative D1/D(a it) on vertical lines of the form
+
a it, where D is the regularized product from $3 of [JoL 92a].
This determination is applied in our general version of explicit for-
mulas [JoL 931. Several other applications will also be given in
subsequent papers, including for instance functional equations for
f ~ ( x=
) lr /
-00
sin A(x - y)
f ( ~ ) x-Y dy = -
6- A
general theta functions.
As to the proof, we shall first give a special case which covers the The middle expression with the sine comes from the last expression
classical case of number theory and Barner's formulation. In this and the definition of f A , after an application of Fubini's theorem
special case, the P a r s e d formula involves only the Dirac functional and the evaluation of a simple integral of elementary calculus. Let
applied to the test function, whereas the general case concerning
arbitrary regularized determinants involves higher derivatives of
the test function. These higher derivatives come from the polar
part in the asymptotic expansion of the theta function at 0. The We are interested in seeing how fA converges to f , that is we want
special case occurs when this polar part consists only of l/x. the inversion formula f = f - in the form
Our proof in the special case is more direct than Barner's or
Weil's, and our formulation of the result already exhibits some f (3) = lim fA(x).
A -+06
general principles which were not immediately apparent in previ-
ous proofs. In particular, we avoid what now appear as detours by We shall give conditions under which the inversion formula is true.
formulatin lemmas in pure Fourier analysis showing how the sin-
f,
gularity be aves under Fourier transform. These lemmas are used We let the Schwartz space Sch(R) be the vector space of func-
in the special and eneral case. We are indebted to Peter Jones for tions which are infinitely differentiable, and such that the function
fl
a proof of one of t ese lemmas, which has independent interest in and all its derivatives tend rapidly to 0 at infinity. That f tends
general Fourier analysis. rapidly t o 0 at infinity means that for all polynomials P the
frurchn Pf is hcwdd. T k tk Sehwa&z space is self Ad,that
is
Sch(R)" = Sch(R).
An elementary result of analysis am&s that the formula f "" = f -
is true for f in the Schwartz space. We sha.ll assume such an
elementary result, and extend it to functions which are less smooth,
namely we shall deal with functiom of bounded variation. All the
background material needed is contained in F a 931. We let BV(R)
denote the space of complex valued functions of bounded variation The desired bound follows by the standard absolute estimates.
on R , i.e. the space of functions of bounded variation on each We have used here that over a finite interval [a, b], the two terms
finite interval [a, b], and such that the total variations are uniformly f(b)S(Ab) and f(a)S(Aa) tend to 0 as a -, -00, b 4 oo because
bounded for all [a, b], in other words there exists B > 0 such that S is bounded, and f tends to 0. This concludes the proof. 0
V(f, a, b) 5 B for all [a, b]. We recall a classical theorem from Fourier anal sis giving nat-
We let ural conditions under which Fourier inversion hold;l, especially at
a discontinuity. For this purpose, we shall say that a function f is
normalized at a point x if

Remark. If f E BV(R) f l L1(R) then f(x) 4 0 ao x 4 foo.


>
Indeed, if 1 f(x)l c > 0 for infinitely many x tending to infinity,
since f E L1(R) there are infinitely many y such that 1f (y)l c/2 < Thus the right and left limits off exist at x, and the value of f at
(say), and so the function could not be of bounded variation. x is the midpoint. We say that f is normalized iff is normalized
We let dpf be the Riemann-Stieltjes measure associated with f , at. every x E R.
and sometimes abbreviate dpf by df. Theorem 1.2. Let f E BV(R) n L1(R), and suppose f is
The function f A exhibits different behavior near 0 and at infinity. normalized. Then fA is bounded independently of A and
Its behavior will be described in part in the lemmas below. We shall
need the function z

- S(X) = - 1
0
sin t
lim ~ A ( x=) f (x) for all x E R.
A-00

For a proof, see Titchmarsh [Ti 481.


so S is continuous and bounded on R. In fact, S is bounded by the The remainder of this section is devoted to examining the uni-
area under the first arch of (sint)/t (between 0 and T),as follows formity of the convergence in Theorem 1.2. To begin, we mention
at once by the alternating nature of the integrand. a very special case.
Lemma 1.1. Let f E BV(R) n L1(R). Then for A > 0 the Lemma 1.3. There exists a function a E Sch(R) such that Q"
function f A is bounded, and in fact there is a uniform bound, has compact support, and a(0) # 0. For such a function, we
independent of A: have
QA = Q

for all A suiffciently large.


where (1 11 is the sup norm.
Proof. Let @ E C,OO(R) be an even function 2 0 with p(0) > 0.
.
.
- Proof. Integrating by parts yields Let a = PA. Tfien @ = a" has compad support, and the direct
evaluation of the Fourier integral together with the definition of QA
shows that the other conditions are satisfied. 0

=- 7 S[At)a'f [t - x).
The following quantitative formulationof the Riemann-Lebesgue
lemma is proved by Barner in fBa 901, Sat8 82,in $21, to which the
reader is referred for a proof.
Lemma 1.4. Assume: Corollary 1.7. Assume g has M derivatives and all the func-
tions g , g('), . .. ,g ( M ) are in BV(R) n L1( R ) .Assume that
(a) s E BV(R).
g ( x ) = O(lxlM+') for 1x1 -,0.
( b ) g ( x ) = O ( ( x l t )for some e > 0 as x 4 0.
Then:
Then the improper integral that follows exists for A > 0 and
satisfies the bound (a) The function g~ has M derivatives ( g A ) ( l ),. ..,( g A ) ( M ) and

( g ~ ) (=~ ()g ( * ) ) A for L = 1,.. .M,


so, without ambiguity, we can write the derivatives of g~
(MI
a s g ~, . . . , S A .
(1)

Also, we need the following elementary lemma. (b) We have


Lemma 1.5. F o r d l 0 < a < b a n d A > 0 we have

the estimate on the right being independent of A.


PTOO~. Let h be a differentiable function on R, with derivative
Proof. This follows from the change of variables u = At, the h', such that h, h' E BV(R) n L1(R). Since d/du((sinAu)/u) is
alternating nature of the integrand, and an elementary area esti- bounded, one can interchange derivative and integral (see page 357
mate. 0 of [La 831). These calculations yield that
The main result of this section is the following uniform version
of Theorem 1.2.
Theorem 1.6. Let g E BV(R) n L 1 ( R ) and assume

g ( x ) = O(lxlt) for 1x1 0.

Let 6 = min(i, &). Then for all A > 1,


00
-
the estimate on the rigbt being independent of A.
= / h(~) (-g)[ sin A(y
(y - r ) x ) ] d~

Theorem 1.6 will be proved using a ,series of lemmas. Before


mntinuin ,let us state the following mmllaqr of Theorem 1.6 that
further r&e8 the uniformity of the pointvase convergence result
stated in Theorem 1.2.
The integration by parts step is valid since (sinAu)/u is bounded Case I. Suppose 1x1 < A-4 and, say, x > 0. Then
and h(x) approaches zero as x approaches infinity. This proves (a)
by induction, letting h = g(k). As to (b), if we apply Theorem 1.6 sin Ay sin A(x - y)
to the function gLM)(x)we get Y X,Y
0

Proof. Since

By repeatedly integrating this equation from 0 to x and applying


(a), we get (b), thus proving the corollary. we use the Mean Value Theorem and the hypothesis x < A*' to
get the bound
, The remainder of this section is devoted to the proof of Theorem
1.6. To do so, we will write
sin Ay sin A(x - y)
x-Y
1
4c A ~ X5 z1I2,

and then we estimate the desired inte a1 in the coarsest way with
f
the sup norm to conclude the proof o the present case. 0

Case 11. Suppose 1x1 2 A-4, so A-I _< lx11/4.


In this case, we will bound each separate integral without using
and investigate the finite and infinite intervals separately. For no- the difference of sines. Lemma 1.4 takes care of the term with
tational simplicity, we will consider the integrals over the intervals (sin Ay)/y, giving a bound of x ' / ~ . The next lemma takes care of
[O, 11 and [I, m), with the analysis over the intervals (-00,-1]and the term with (sin A(x - y))/(x - y).
[- 1,O] being identical to that over the corresponding positive in-
tervals. We begin with the analysis over the finite intervals.
Proposition 1.8. Let g E BV(R)nL1(R). Assume in addition
that there exists e > 0 such that
!

g(x) = O(lxy) for x 4 0.


the estimate on the right independent of A.
Then there is 6 > 0 such that for all A 2 1 we have Proof. We split the integral
- y) 1

--
-

0
sin Ay sin A(x
3-Y I dy = og(1xI6)for x 4 0,

the estimate on the right being independent of A.


Proof. We owe the proof of this proposition to Peter Jones, who
showed that one can take 6 = min(in(t,&), as stated in Theorem
1.6. We need to s lit the integrd over m o u a intervals, depending
on A and x. We bt settle the easiest case.
For the first integral 11, we change variable putting u = x - y. and we estimate Jl,J2 successively. For the integral J1, using
Then the limits of integration for u are lul 5 A-', and we get a Lemma 1.5, we find:
bound

For the second integral 12, again with u = z - y, we find that (5)
For th.e integral J: let xarbe the characteristic function of [O,Y),
that is put

If g is continuous at y, then
where the interval over which we take the sup norm of g is

20
Using the growth estimate g(x) = O(lxlf) for x near zero and the
This is a convenient expression to plug into an application of Fu-
assumption A'l _< $'I4, we get bini's theorem which gives
1 1

~2 = J / x(y, t ) sin x-Y


A(x - y)
dcrg(t)dy
to I0

For the third integral 13, since the set of discontinuities of a


function of bounded variation is countable, we can select xo such
that g is continuous at xo and

>
For simplicity, we just look at y xo. The other piece for I3is
done in the same way. We then decompose the integral over y 2 xo
into a sum:

Hence by Lemma 1.5, we find


which concludes the estimate of the second integral J2, and there- The point evaluations of g(y)SA(y) will be zero as b approaches
fore of the integral 13. Thus (3), (4), (5), and (6) conclude the infinity, since g(t) = 0, g(b) approaches zero, and SA(y)is bounded.
proof of Lemma 1.9. 0 Also, by Lemma 1.5
Having taken care of all cases, we have completed the proof of
Proposition 1.8.
To finish the proof of Theorem 1.6, we need the following propo-
sition.
Proposition 1.10. Let g E BV(R)nL1(R). Then for all A 2 1 Therefore, we have shown, after letting t approach 1and b approach
we have oo that

I[ sin Ay
dY)ydYl < ;vR(g),
the estimate on the right being independent of A. as asserted. 0
The proof of Proposition 1.10, which is much easier than that of
Proposition 1.8, will be given through the following lemmas. We now consider the integral in Proposition 1.10 in two separate
cases, when x 5 1/A4 aad when x > l/A4.
Lemma 1.11. Assume g E BV(R) n L1(R). Then
00
Case I. Assume x < l/A4. Then there is a universal constant
sin Ay 3 C such that
/g(Y)TdY < ;iv~(g).
1

Proof. Extend g to [0, m) by defining g(t) = 0 if 0 5 t < 1. For


fixed t E ( 0 , l ) and b > 1, consider the integral
the estimate on the right independent of A.
Proof. By the Mean Value Theorem we have

E So, we can bound the integral in question by


" Using integration by parts we have
CA241glll 5 (~11911d~1/2.0
52. A Parseval formula
Case 11. Assume x > 1/A4 and, without loss of generality, we
can also assume that x 5 112. Then We recall the definition of the hermitian product

sin Ay sin A(x - y)

1
Y . 2 - Y ] dy
5 c,xli4,

where the constant C, is independent of A, and depends only For f,g in the Schwartz space, we assume the elementary Parseval
on g. Form ula
Proof. Let us estimate the integrals separately. In fact, write (f,!l) = (fA,gA).
the integrals as The point of this section is to prove this formula under less restric-
tive conditions. We shall extend conditions of Barner [Bg 901, for
00

T g ( sin A(Y- sin Ay which the formula is true. The basic facts from real analysis (func-
tions of bounded variation and Stieltjes integral, Fourier transforms
y - 2 ")dar - Jg(y)Tdy under smooth conditions) used here are contained in [La 831.
1 1
We call the following conditions from [Ba 901 the basic condi-
tions on a function f:
Condition 1. f E BV(R) n L1(R).
sin Au
sin Au
du+
J g(u+x)-
U
du Condition 2. There exists e > 0 such that
f(x) = f(O)+O(lxle) for x -,0.

Condition 3. f is normalized, as defined in 91.


In addition to functions satisfying the basic conditions, we shall
The first integral is bounded by llgllooxsince (sin Au)/u is bounded deal with a special pair of functions, arising as follows. We suppose
on [1/2,1]. Lemma 1.11 applies to bound the second integral, in- given:
dependent of x, as well as the third integral. The bound achieved
is - A Borel measure p on R+ such that dp(x) = $(x)dx, where
6 $ is some bounded, (Borel) measurable function.
Ibllcox + z v ~ ( ~ ) .
-.
-
- AIzmaddefbtnctioerpoaR+a&U
, Since 1/A < x ' / ~ ,Case I1 is proved. (a) The function cpo(x) = cp(x) - l / x is bounded as x
approaches zero.
"t Combining the two cases above, we have completed our proof (b) Both functions l / x and ~ ( x are
) in L1(lpI) outside a
.. of Proposition 1.10. With all this, the proof of Theorem 1.6 is neighborhood of zero.
completed by combining Proposition 1.8 and Proposition 1.10.
Thus we impose two asymptotic conditions on cp, one condition
near zero and one condition at W t y . We cdl ( p , 9)a special
pair. We define the functional By the assumptions on (p, 9 ) and f , we can interchange the inte-
grals which are absolutely convergent. We then perform the inner
integration with respect to t, and the expression becomes

+ y)
Its "Fourier transform" as a distribution is the function W,1, such
= JJ
RxR+
f (9) sin A(x
7[ ~ ( x ) + 1-
sin Ay
- xY dy4w
that sin A(x - y ) sin Ay
2-Y
R+ R

1 eit~
= -Wp,JXt) where xt(x) =
6
The following theorem generalizes the Barner-Weil formula [Ba 901. At this point, we are finished with the proof in the case f = a is
a function satisfying the conditions of Lemma 1.3, namely f2.4 = a
Theorem 2.1. Let f satisfy the three basic conditions, and let for sufficiently large A. For the general case, we write the above
(p, 9) be a specid pair. Then integral as a sum

Note that cp(x) - 1/x is in L1(Ipl) by our assumptions on (p, 9).


Our final step is to prove that we can take the limit as A -+ oo
under the integral sign in both terms for an arbitrary f satisfying
the basic conditions. In the first integral, one can apply the domi-
Proof. At a certain point in the proof, we shall need to distin- nated convergence theorem by the boundedness of fA (see Theorem
ghish two cases, but we proceed as far as we can go without such 1.2) and the assumptions on ~ ( x-) l/x. As to the second integral,
distinction, according to a rather standard pattern of proof. We we split the integral
have:

j fAct,wL,ctidt
-A

Again, we apply the dominated convergence theory to the inte-


gral over [1,rn), by Theorem 1.2, the boundedness of fA, and the
assumption that l / x is in L1(p). The more diffidt part is the
integral over the inteval [0,11.
Let a be as in Lemma 1.3 and such that a(0) = f(O), which can
be achieved after multiplying a by a constant. Let g = f - a. The
formula of Theorem 2.1 is linear in f , and it is immediately verified $3. The General Parseval Formula
that cr and hence g satisfies the basic conditions and .
Following the ideas in [JoL 92a], we now prove a general Parseval
g(x) = O(JxJt) for x -, 0. formula associated to measurable functions with arbitrary principal
part, thus generalizing the results of $2 in which the function ~ ( x )
was required to have principal part equal to l/x.
Baving proved the formula for a,we are reduced to proving it for g.
Recall that Theorem 1.6 states that for a positive 6, which depends Suppose we are given:
on e,
- A Borel measure p on R+ such that dp(x) = $(x)dx, where
t )t is some bounded (Borel) measurable function.
~ A ( x-) g ~ ( 0 =
) 0()xJ6)for x 4 O
.- A measurable function cp on RS having the following prop-
uniformly in A. Therefore, the dominated convergence theorem erties. There is a function Po(x) E Cilog s](+), which we
again applies since x-'+* is integrable over [O,l], and we get, since shall write as
d o ) = 0,
Po (x) = bp(x)xp with bp(x) = BP(logx) E C[log Z]
W P )<O
such that:
(a) There is some integer m > 0 such that
After taking the limit as A 4 oo under the integrals we see that
the final expression above becomes cp(x) - Po(x) = O( 1 log xim) for x -,0.
(b) Let M be the largest integer < -Re(po), so that

Then both functions xMPo(x) and cp(x) are in L1( 1 ~ 1 1 )


which proves the theorem. 0 outside a neighborhood of zero.
From condition (a) and the power series expansion of eitz one ob-
tains the existence of functions uk(x) such that

The functions uk(x) come from the expression


As before, the above requirements impose two asymptotic.condi- 2.1. For completeness, let us present the details. We have:
tions on 9 , one condition near zero and one condition at infinity,
and we call such a pair (p, 9 ) a special pair. We define the func-
tional

Its "Fourier transform" as a distribution is the function W,l,(t) As in the proof of Theorem 2.1, the assumptions on (p, 9 ) and f al-
such that low us to interchange the integrals which are absolutely convergent.
By integrating with respect to t, this expression becomes

sin A(s+ y)
M

RxR+ k=O

1 Continuing, we have:
= -W,,,(nt) where xt(x) = elt?
6

The following theorem generalizes Theorem 2.1.


Theorem 3.1. Assume f and its first M derivatives satisfy the
three basic conditions, and let (p, 9 ) be a special pair. Then

In the above steps we have used the differentiation formula

and the integration by parts f~rmula

Proof. The proof is essentially identical to the proof of Theorem


which is valid by Lemma 1.1 and the arguments given in the proof so C X ~ + ~ ~ ( PisOintegrable.
)+' The dominated convergence theo-
of Corollary 1.7. rem applies, and the theorem is proved.
At this point, we are finished with the proof in the case f = o is
a function satisfying the conditions of Lemma 1.3, namely c r = ~ o
for sufficiently large A. For the general case, we write the integral
as the sum

The proof of Theorem 3.1 now finishes as did the proof of The-
orem 2.1. By an appropriate extension of Lemma 1.3, choose
an o for which o~ = o for sufficiently large A and the num-
bers o(O), . . . , d M ) ( 0 ) have been chosen to agree with the first
M derivatives of f at zero, and set g = f - a. The above integrals
are linear in the function f , so proving the theorem for g will im-
ply the theorem for f , so we work with g. By Theorem 1.2, the
boundedness of g~ (as stated in Lemma 1.1), and assumption (b)
above, one can apply the dominated convergence theorem to the
first two integrals above. Using Corollary 1.7 and the definition of
the function uk(x), we can write the third integral as

By Corollary 1.7, the integrand is bounded by c ~ ~ + ~ ~ (and


P ~ ) + ~

M+Re(po)+61-1+6,
115

54. T h e Parseval Formula for Iw(a + i t ) Recall that the principal part of the theta function is
To conclude our investigation of the regularized harmonic series,
let us show how Theorem 3.1 applies to prove a Parseval formula Po+) = C bp(.)xP,
associated to the regularized harmonic series encountered in $3 of M P <o
)
[JoL 92a]. We will assume the notation defined in 53.
so
Recall as in 54 of [JoL 92a] that the classical Gauss formula e ( ~-)Pee($) = qllog as -+ 0.
states that for Fte(z) > 0,
As in [JoL 92a], let Bz(x) = e-zz6(x). By expanding e-Zz in a
power series, we see that the principal part of B,(x) is

If we let z = a + it with a > - 1 , then we get


From $4 of [JoL 92a] we recall the following result.
Theorem 4.1. For any fixed complex w with

where
eVaz
%(x) = I - e-z and dp(x) = e-'dx. the integral

One can view (1) as a type of regularized Fourier transform repre-


sentation of the gamma function. Finally, by the change of vari-
ables t to t l b then x to bx for b > 0, ( 1 ) becomes

is convergent for Re(z) > 0. Awther, Iw(z) has a meromorphic


continuation to ad z E C with simple poles at -Ak +
w with
residue a,.

Remark. In the spectral case, when ((s) = x a k A i s with


Next we will present a generalization of ( 1 ) making use of results
ak E Z>o,
- Theorem 4.1 of [JoL 92a] states that
in 55 of [JoL 92a]. That is, associated to any Dirichlet series 5
satisfying DIR 1 and DIR 2, as d e h e d in 51 of [JoL =a], and
whose associated theta function B(t) satisfies AS 1, AS 2 and
AS 3, we will realize the regularized harmonic series I,(z), whose
definition will be recalled below, as a regularized Fourier transform. where S,(z) is a polynomial in a of degree < -Re(po), with coef-
Theorem 2.1 applies to the classical Parseval formula involving the ficients whose dependence on L is through bp for Re@) < 0, and
amma function, which is used in the Barner-Weil explicit formula. whose dependence on w is through elemeats in C[bgw ] ( w ) . Also,
k this section we will use our regularized Fourier transform to
present a general Parseval formula.
in 54 of [JoL 92a] it is shown that Theorem 4.1 yields the classical
Gauss formula in the case L = Zlo.
To continue, let us work with the principal part of the theta Theorem 4.3. Assume f and its first M derivatives satisfy the
function. If we restrict the variable z in (2) to a vertical line by three basic conditions. For any w E C with
letting z = a + it we get
Re(w) > mku{-&(At)} and &(w) > 0,

and any a E R+,define

dpW(x)= e-wzdx and O,(x) = B(X)~-~'.

Then
where the coefficients

depend on the variables a, x and on (' through the coefficients of


tP for Re(p) < 0 (see AS 2). With this, the integral in Theorem
4.1 can be written as
(4)
1
- Theorem 4.3 is the classical
In the spectral case when L = Z>,J
Barner-Weil formula.

Thus, we obtain:
Corollary 4.2. For any w E C with Re(w) > maxk{-Re(Xk)}
and Re(w) > 0, and any a E R+ define

dpW(x)= e-wzdx and O,(X) = B(x)~-"".

Then

=- +
Iw(a it) =
k<-Re(po)

This is our desired generalization of (1) and will be referred to as the


regularized Fourier transform representation of a regularized
harmonic series. Then Theorem 3.1 yields:
BIBLIOGRAPHY

[Ba 811 BARNER, K.: On Weil's explicit formula. J. reine angew.


Math. 323, 139-152 (1981).
[Ba 901 BARNER, K.: Einfiihrung in die Analytische Zahlentheo-
rie. Preprint (1990).
[BeKn 861 BELAVIN, A. A., and KNIZHNIK, V. G.: Complex geom-
etry and the theory of quantum string. Sou. Phy. JETP
2 , 214228 (1986).

, m d SEELEY, R: Regular singular asymp-


[BrS 851 B R ~ N I N GJ.,
totic~.Adv. Math. 58, 133-148 (1985).
[CaV 901 CARTIER, P., and VOROS, A,: Une nouvelle interprhta-
tion de la formule des traces de Selberg. pp. 1-68, volume
87 of Progress in Mathematics , Boston: Birkhauser (1990).
[Cr 19) CRAMER, H.: Studien iiber die Nullstellen der Riemann-
schen Zetafunktion. Math. 2. 4, 104130 (1919).
[De 921 DENINGER, C.: Local L-factors of motives and regular-
ized products. Inuent. Math. 107, 135-150 (1992).
[DP 861 D'HOKER, E., and PHONG, D.: On determinants of La-
placians on Riemann surfaces. Comm. Math. Phys. 105,
537-545 (1986).
[Di 781 DIEUDONNE, J.: &menta d' Andyae, Vol. VII, Paris:
Gauthier-Cribs (1978), reprinted as 3katise on Analysis
Volume 10- VII San Diego: Academic Press (1988).
[DG 751 DUISTERMAAT, J. J. and GUILLEMIN, V. W.: The
spectrum of p i t i v e elliptic operators and periodic bichar-
acteristics. Inuent. Math. 29, 39-79 (1975).
[Fk 861 FRIED, D.: Analytic torsion and closed geodesics on hy- [La 931 LANG, S.: Real and Functional Analysis, Third Edition,
perbolic manifolds. Invent. Math. 84, 523-540 (1986). New York: Springer-Verlag (1993).
[G 77) GANGOLLI, R.: Zeta functions of Selberg's type for com- [McS 67) McKEAN, H. P., and SINGER, I. M.: Curvature and eigen-
pact space forms of symmetric space of rank one. nl. J. values of the Laplacian. J. Differ. Geom. 1, 43-70 (1967).
Math. 21, 1-42 (1977).
[Gr 861 GRUBB, G;: Functional Calculus of P~eudo-Differential [MP 491 MINAKSHISUNDARAM, S., and PLEIJEL, A.: Some pro-
perties of the eigenfunctions of the Laplace operator on Rie-
Boundary Problems. Progress in Mathematics 65 Boston: mrtnnian manifolds. Can. J. &a& l, 242-256 (1949).
Birkhauser (1986).
[Ha 491 HARDY, G. H.: Divergent Series. Oxford: Oxford Univer- [RS 731 RAY, D., and SINGER, I.: Analytic torsion for complex
sity Press (1949). manifolds. Ann. Math. 98, 154-177 (1973).
[In 321 INGHAM, A.: The Di~tributionof Prime Numberq Cam- [Sa 871 SARNAK, P.: Determinants of Laplacians. Comm. Math.
bridge University Press, Cambridge, (1932). Phys. 110, 113-120 (1987).
[JoL 92a] JORGENSON, J., and LANG, S.: Some complex analytic [Su 851 SUNADA, T.: Riemannian coverings and isospectral man-
properties of regularized products and series. This volume. ifolds. Ann. Math. 121, 169-186 (1985).
[JoL 92b] JORGENSON, J., and LANG, S.: On Cramhr's theorem [TaZ 911 TAKHTAJAN, L. A., and ZOGRAF, P. G.: A local index
for general Euler products with functional equations. To
appear in Math. Annalen. theorem for families of &operators on punctured Riemann
surfaces and a new Kiihler metric on their moduli spaces.
POL 92c) JORGENSON, J., and LANG, S.: A P a r s e d formula for Comm, Math. Phys. 137, 399-426 (1991).
functions with a singular asymptotic expansion at the ori-
gin. This volume. [Ti 481 TITCHMARSH, E. C.: Introduction to the Theory of Four-
ier-Integrab, 2nd Edition Oxford University Press, Oxford
[JoL 92dj JORGENSON, J., and LANG, S.: Artin formalism and (1948).
heat kernels. To appear in J. reine angew. Math.
[JoL 931 JORGENSON, J., and LANG, S.: Explicit formulas and [Ve 811 VENKOV, A. B., and ZOGRAF, P. G.: On analogues of
regularized products. Yale University Preprint (1993). the Artin factorization formulas in the spectral theory of
automorphic functions connected with induced representa-
[Ku 881 KUROKAWA, N.: Parabolic components of zeta functions. tions of Fuschsian groups, Soviet Math. Dokl., 21, 94-96
Proc. Japan Acad., Ser A 64, 21-24 (1988). (1981)

[La 701 LANG, S.: Algebraic Number Theory, Menlo Park: Add- [Ve 831 VENKOV, A. B., and ZOGRAF, P. G.: On analogues of
the Artin factorization formulas in the s ctral theory of
- ison-Wesley (1970), reprinted as Graduate Texts in Math-
ematics 110, New York: Springer-Verlag (1986). at? ~ i e a s Jd
witat represents-
tions of b c h s i a n groups, Math. USSR-Izu. 21, 435-443
,'a [La 831 LANG, S.: Real Analysis, Menlo Park: Addison-Wesley (1983).
(1983).
[Vo 871 VOROS, A.: Spectral functions, special hctions, and the
[La 851 LANG, S.: Complex Analysis, Graduate Texts in Mathe- Wberg zeta function. Comm. Math. Phya 110, 439-465
matics 103, New York: Springer-Verlag (1985). (1987).
Lecture Notes in Mathematics
[We 521 WEIL, A.: Sur les ''formules explicitd de la thkorie des For information about !Ms. 1-1384
please contact your bookseller or Springer-Vedag
nombres premiers, Comm. Lund (vol. dkdiC ir Marcel
Riesz), 252-265 (1952).
[We 721 WEIL, A.: Sur les formules explicites de la thCorie des nom-
bres, Im. Mat. Nauk (Ser. Mat.) 36, 3-18 (1972). Vol. 1385: A.M. Anile. Y. Cboquet-Brubrt (Edr.). V d 110(: M.-P. M8Uiavin(E!d.X W n d r e d ' A I g h Paul
Relativistic FluidDy.rmia.Seminu, 1987. V. 309pages. Dakdla khriePad Mdliavin. Roorcdiap. 1987-1988.
1989. N.410 p q a . 1989.
Vol. 1386: A. Bellem. C.W. O w . E. Russo (Edr.). VoL 1405: S. Ddrcti (Ed.). OpUmiutios. Fmcedags,
Nu~llericJMethod8 for Ckdinuy Mtkrunlrl E ~ W ~ O I U IW. V. m pga 1989. vd. 1406: L ~reobrw(FA.).
Procadiags, 1987. W.1% p g r ~1989. . Amlytk~dConUlvedPrauiowIII.~ap.
Vol. 1381: M. Pettovi'c. l t a t i v c Method8 for S i d - 1988. nI42 p8ga. 1989.
t l l l s a u I n c h u l o n d P d ~ Z P O L X 1989. ~ 6 ~ ~ Vd. 1401:W.PoWm.Roof1Lc#r. VI, 213pgw 1989.
Vol. 13883 J. Sbi.oQ. T.A. S*suq T. To@ (Ed8.X Vd. 14W W. Ukk,Tnuforrmdom Omups rod Algcbnic
Mrr&rmdul Logic and Applicuio~.M y r . 1987. K-Tbeor)..XU, 4 3 w.1989.
V. 223 pages. 1989. Vol. 1409: E WJror. Cb. Lub&b, M. M . . lEeNumaid
Vol. 1000: S d EdiUonH. Hopf, DiflenntW Oeoramy S o ~ d ~ - A l g e b n i c S y r ( e bym RungaKutta
t
in the h g e . W. 184pag48.1989. Metbodr. VII. 139 pagr~.1989.
Vol. 1410: PJ. Crmu. 0. Gi1-Medrw. A.M. Naveira
(Ms.), Difkrrntirl Ocomscry. Rocerdings. 1988. V. 308
1989.
VoL 141I: B, Jimg (Ed.). TopdoIicll Fixed Point Theory
and Appllutiolu. Rocesdings. 1981VI, 203 mu. 1989.
Vd. 1412: V.V. Kalnbdh. V.M. Zolotarev (Eds.).
Vol. 1391: S. Cmulual.. A. WQO. (Bb.). P d a M t y Stability Probbnu for Stocbutic MoQlr. Roceedings.
T h b o y o a V o c r o r S p r t r l V . ~ ~ , 1 9 8 7 . V m . 4 2 4 1987. x, 388ppl; 1989.
p a p . 1989. V d 1413: S. Wright, U.ic(oslwuof UICInjeuivelll,Factor.
V d 1 3 9 1 : R S f l b o l , R d A l p b n i c ~ . X 2 I 5 ~ I& M pap^. 1989.
1989. Vol. 1414: & Raminz & Amllrw (Ed.), Algebraic
Vd. 1393: N. Bouhrrr.D. Fey4 P. H*rL.0.MolroboQtLi Geome&y a d CospIcx M y & Rocredingr. 1987. VI.
(Ed8.X Stmidm de lWxk du Porcaticl Pmir. No. 9. ISOpyor. 1989.
Rocredinlr.W, 265 p o p . 1989. VoL 1415: M. Lanpi&M. Waldadmidt (Us.). Ciaqu.l)te
Vol. 1394 T.L. 0111, W.W. Zlcbvy (Bb.). No.lhKlr Ans de 301yn6mes. Fifty Years of Polynomials.
S e m I ~ p h ~ d ~ E q ~ a n d A m r c c Pmcecdinlr. o n . 1988.X 235 pages.1990.
Proaediyl. 1987. M,233 paga 1989. Vol. 1416: C Alkn (Bd.). G&mCllie Symplectiquc et
Vol. 1395. K. AlLdi (Ed.), NumbaTbor), Madras 1987. hwmdqw. Roerccdlq* 1988. v. 289 pga 1990.
Roaediqs. m234 wa. 1989. VoL 1411: AJ. Soamem, A. Bhcdiom, E.L. Uvorni
Vol. 13%: L Acardi, W. voa Walddela (l3b.X Que (Ed&), AIgckrk G a w r y . FVocadiny. 1988. V. 320
turn ProbrMlily and AppUeuioar IV. Rocscdtngs, 1987. page& 1990.
VI, 355 pages. 1989. Vol. 1419: M. Mi- (Ed ), Homotopy Tbtory aad
Vol. 1397: P.R. l h a (BaXNupaialh a l y d a Pa- RcLcrQTopia.R0ardl.p. 1968. V. 241 pages. 1990.
dklRoaulng. sd.y,1987. VL 264 ppr. * l. Vol. 1419: P.S. Bulh.. P.Y. Lee. J.L. Mawbin. P.
Muldowney. W.P. Plcffcr (Bds.). New Integrals.
m
Rocrcrdir,~1988. v. wa. 1990.

1
VatI4&YOllbl*LLT~~bIlalAnrlytb
Ocoagl.hoadbaI)#W.W6pages. 1990.
Vol. 1421: HA. Bin&& A Nodincar Theory of
~ ~ X H . 2 1 4 ~ 1990. p l .

v. 249 pages. 191LP.


Lecture Notes in Mathematics
[We 521 WEIL, A.: Sur les "formules explicites" de la thbrie des For information about Vols. 1-1384
please contact your bookseller or Springer-Verlag
nombres premiers, Comm. Lund (vol. dddid B Marcel

[We 72) WEIL, A.: Sur les formules explicites de la thbrie des nom-
bres, Izv. Mat. Nauk (Ser. Mat.) 36, 3-18 (1972). Vol. 1385: A.M. Anile. Y. Cboquet-Brubrt (Edr.).
Reluivirtic Fluid Dynrmiu. Scminu. 1987. V. 308pycr
1989.
Vol. 1386: A. Bellen, C.W. Gear, E. Rsuo (Us.), Vd. 1105: S. Dokdd (Ed.). Optiducion. Roceedinp.
Numeriul Mctboda f a Oldbry M f k k r d d Quaiom. 1988. V. 223 pqa. 1989. Vol. IJ06: L Jmbaen (Ed.).
1987. W, 136 1989. M y t i c Tbea), dCoMid F d o u In. Proceedings.
Vd. 1387: M. 1Wravi%, (tcnoiva McUnnl8 far Shd. 1!#8. W,142 pap. 3999.
t r n e a u I l u & d ~ ' l l a a . X M J p r p ~ 1 9 9 9 . Vol. lUn: W.PoblqRooflbxy. VI,213 pqea. 1989.
Vol. 1388: 1. Sbi.ob. T.A. S h u n , T. TtlrpC (Eds.),
Mrtbermtial Logic ud AppUcrtionr. Roaedi.y. 1987.
V, 223 paps. 1989.
Vol. 10b0:SecondWtioa.H. Hopf, Mff& Gmmet~~
in the L ~ W, E 184pyer. 1989.
Vol. 1389: E Bllliea C Wbum(Ed&). Algebraic Cuwr Vd. 14la F.J. Cmnh 0. GiI-A4edr~0,A.M. Navein
and FYojecdveGamwry.Proctdiqh 1988. V. 288 pap.
1989. (Ed&), Diffacnt*I G a w t r y . Roceed*y. 1988. V. 308
pages. 1989.
Vol. 1 3 9 0 : 0 . ~ P n t o . L . ~ ~ . ~ . S I o c b w l c p n i .VoL a).
l 1411: B. J i a q Topological FixedPoint Tbcay
Differrotid BpDItioar ud Applicrdolu If. Pmcdiny. aedpd~ppllertion~ Roacd&y. 1988. VI. 203 pages. 1989.
1918. VI. 258 pyol. 1999. Vd. 1412: V.V. wrl.lbdlor. V.M. Zolotuev (Eds.).
Vol. 1391: S. Crmh.*. A. Wcm. (Ma.), Robability Smbility Robbnu for Stocbutic Modela. Procetdinga.
T b c a y o n V ~ S p r c u I V . ~ y . 1 9 8 7 . M I . 4 2 4 1987. x 380 pqa 1 m .
pages. 1989.
Vol. 1392: R SilhoL Rul Algcbnic Surf-. X 215 pages.
1989. Vol. 1414: E. Runires dc Amllrw (Ed.). Algebraic
Vol. 1393: N. Boulmu. D. k y d , F. Hkrds0. MoWodzki Oeolnay ud Complex Anrlyais. Prooeedinga. 1987. VI.
(Eds.). StnJtuinds Tb(orio du PoccPakl M r . No. 9. 180pgu. 1989.
Roacdiyr. VI. 265 ma. 1989. Vol. 141%M. Laagwia,M. Wddadddt (Eds.). Cioqumtc
Vol. 1394: T.L. 0111, W.W. Zrhry w.). Noalillur Ana de Polynbmer. Fifty Yeur of Polynoriirls.
Semiproup. P a W Diffaw~pirlEqurtiom ud A m r t m . Roceediqs, 1988. IX. 235 pagcs.1990.
Roceedinga 1987. IX, 233 prpr. 1989.
Vol. 1416: C. Alkrc (Ed.). OColdoic Symplcctique et
Vol. 1395: K. Alldi (Ed.), Number lbxy.h d w lM7. Mbdque. Roceedinp, 1988. V. 289 pager. 1990.
Roaediyr. W, 234 paps. 1989. Vol. 1417: AJ. Sonunew A. B i d o n . EL. L i v m i
Vol. 13%: L A d . W.vor W W & (Ed&). Quu- (Ed&), Alpbnic Geometry. hxedingr. 1988. V. 320
turn Probability ud AppUcleio~N. Roceediry, 19111. pages. 1990.
VI. 355 WU 1969. Vol. 1418: M. Mialun (Ed. ). Homotopy Tbeory and
Vol. 13W P.R. 'ILwr (Ed.). NmaiaiM y & md Pa- RslrrcdTopies. h c d i y l . 19s. V. 241 pages. 1990.
nlkl Pmccning. Seminu, 1987. Vl, 264 w.1989.
Vol. 1419: P.S. Bullen. P.Y. Lee. J.L. Mrwhin. P.
Vol. 139& LC. Kim, B.H. W. (Ed&). Group - Muldowaay. W.F. Pfeffsr (Wa.). New Integrrls.
K o m I ~ ~ V . 1 8 9 p l 1989. r r -19n.v.aQZprlu. 1990.
Vot. t399: W.-t. mala, K L.tp mf. lbiUNm& d Vot. W&U.WbW, A. Togu~U@t&.), R u t Asdytlc
Compkx Manifolds. Roacdiry, 1988. V. 171 pycr Olomrc*.- 19U.N.#6pyu. 1990.
1999. Vol. 1421: H.A. B4iomi. A Nonlinear Theory of
Vol. 1400: U. J u MiadMoUve~ud Alpk.ic b ~ ~ X & 2 11990. 4 ~ .
lhlry.Ma246psp. 1990. Val. 1411: V. VHL.I ($d). Cwplcx Geometry and
Vol. 141: J. A W r W Qldr,X Set Tbmq d k J y d ~Fmmdllp.
. IW1. V. 109 p a p . 1990.
I I a ~ R P a d l l l pS l l . V , T n w ~ ~lm9. . V d lltl: U k I(.eLr* Strblr hmompy Groups of
Vd. 1 ~ . C . ~ ? . ~ , B . H w r r o g J . - I , . h l y A Caqukc-AubtdACpoub. WII. 330 pages.
E B & f . ~ t t s r ~ ~ ~1990.1 9 a
V,249pe#s. 1989. Vd. 1424: P.R bw*8. J.H. Rawdey, T w i w Theory
Printing: Wcihert-DrucL GmbH, Dumrt.dt
Vol. 140%B. Shoom (Ed.). Ceasluairl Opdmiudocl. f a ~ S p l l n a r i c S p r c s r I I L1 1 2 ~1990.
~ .
Bindiw: B u & SablUtw,
~ OrOnrtrdt S d n a , 1986. V. 314 pe#a. 1909.

You might also like