THE
ROOTS OFTRIGONOMETRIC INTEGRALS
BY N. G.
DEBRUIJN
1. Introduction. Concerning the roots of trigonometric integrals
G.
P61ya(see
references at the end of the paper) has proved a number of results whichhederivedfrom properties of the roots of polynomials.
He
proved,forinstance, the reality of all the rootsofthe followingfunctions:(1.1)
f
e-t2e
’ztdt
(n
1, 2, 3,...);
(1.2)
C(t)e
’
dt(h
>
0),
where
C(t)
exp (-), cosht),
and(1.3)
f_
exp(--at
4"-
bt2"ct
)
exp iztdr,wherea
>
0, breal, c>_
0, n 1, 2, 3,....
(See
concerning(1.1), [7], [8];
concerning
(1.2), [6],
[8];
concerning(1.3), [8].)
Another important result of P61ya is the following one
(see [8]):
Suppose that the functionF(t)
of the realvariable satisfies(1.4)
F(t)
integrable over<
<
F(t)
(F(-t))*,
F(t)
O(e
-Itlb)
for
--
,
b>
2.(The
* indicates the conjugate imaginary.)Let (t)
be an integral function of genus 0 or 1, withreal roots only, and let the number,
be>_
0. If the functionF(t)
is such that all the roots of the integral(1.5)
F(t)e
’ztdt
arereal, then thesameholds for the function
f:
F(t)q(it)e*t’e
t dt.The function (it)e
"
is easily seen to be the limit of a sequence of poly-nomials, all of whoserootsarepurelyimaginary. P61ya’s result, statedinotherwords,
is that these functions are universalfactors:
which conserve the reality of the roots of any trigonometric integral whose integrandsatisfies(1.4).
P61ya also proved that the functions (it)e"
indicated above arethe only analytical functionswith this property. The latterresult will not beused in the present paper.ReceivedJuly 16,1948.
198 N.G. DE BRUIJN
In
thesequelwecontinueP61ya’s researches.Our
mainresults areTEORE
1.Let
f(t)
be an integralfunction
of
and such that its derivative$’(t) is the limit (uniformlyin any bounded domain
of
the t-plane)of
a sequenceof
polynomials, allof
whoseroots lie onthe imaginaryaxis.Suppose
furthermore
thatf(t)
is notaconstant,
and thatf(t)
f(-
t),
](t)
>_
0for
real valuesof
t. Then theintegral_(R)
e-teTM dt has realroots only.(The
conditions(f(t)
f(-t), f(t)
>
0)
may be replaced by weaker ones, namely,"f(t)
(f(-t))*, Ref(t)
>
0for o<
<
oo",
iff(z)isapolynomialorafunction of thetype
(1.6) (see
Theorems 19 and 20 respectively).It
isnot easy toseewhetherthe latterset of conditionsissufficientinthe generalcase.)
P61ya’s results(1.1)
and(1.2)
are special cases of this one, but(1.3)
is not.TEORE
2.Let
N
beapositive integer andput(1.6)
P(t)
p.e"
(Re
pN>
0; p.* p_.,n 0, 1, 2,---).
Let
thefunction
q(x) beregular in the sector-r/2N
N
-1arg p
<
arg x<
r/2N
N
-
arg prand on its boundary, with possible exceptionof
x 0 and x whichmay bepoles(of
arbitraryfinite
order)
for
q(x). Furthermore suppose(1.7)
(q(x))* q(1/x*)in thissector(inother
words, q(x)
isrealfor
x 1. Then all butafinite
numberof
rootsof
thefunction
(1.8)
(z)
f:o
e-’("Q(t)e
’’tdt
(Q(t)
q(e’))
arereal.
It
may be remarked that our method fails to give any useful information concerning the number and location of the non-real roots of(1.8)
in thegeneral case,sothat this very peculiar result may be ofverylittle practical importance.The special functions
(C(t)
exp( -hcosht))
(1.9)
(z)
C(t)
a.e
"’
e"’
dt(
>
O,
a*.
a_.),
which have
N
pairs of non-real roots at most(Theorem
21),
may be of someinterest since the Riemann
-funetion
can be approximated by functions of this type(see [8]).
It
will be worthwhile to determine classes of functions of this type with the property thut all the roots are real.We
shall study these questionsin6.
(Those
readers who are mainly interested in considerations concerning the /-funetion may omit the proof of Theorem I and the relatedresults in5,
andinTheorem 2 need to considerthe ease
P(t)
(e’
+
e-t)
only.In
that easethe complicated
4
is superfluous since the results of that section then reduce to well-known asymptotic formulas concerning the F-function.)THE ROOTS OF TRIGONOMETRIC INTEGRALS 199
In
7
we expose what progress has been made in this paper in the direction ofthe Riemann hypothesis, and also how small this progressis.An
outline oftheproofs of Theorems 1 and2concludesourintroduction. Sections2and3 will furnishfunctionsS(t)
which arespecial universalfactorsin PSlya’s sense but which have stronger properties than those stated above.
A
functionS(t)
of the realvariable t, satisfyingS(t)
(S(--t))*
will becalleda stronguniversal
factor
if itjoins properties(a)
and(t) below,
foranyfunctionF(t)
satisfying(1.4).
(a)
If the roots of(1.5)
lie in a stripIm
z_<
A(A
>
0),
then those off:.
F(t)S(t)e
‘tdt lie in a strip
[Im
z<
A1
whereA1
<
A,
A
independent ofF(t).
(t)
IfF(t)
is such that, for any>
0, all but a finite number of roots of(1.5)
lie in the stripJim
z_<
,
then thefunctionf_ F(t)S(t)e
itdthas only
a finitenumber of non-real roots.
It
will be evident from(a)
that any strong universal factor is a universal factorinPSlya’s sense.A
functionS(t)
of thetype(1.10)
S(t)
a,e’)’t(k
>
O,
a,,a*,,)
-N
is astronguniversalfactorifall itsrootslie onthe imaginaryaxis. (Conversely,
if
S(t)
is auniversalfactor and ifit isof thetype(1.10),
thenits roots lieonthe imaginary axis. This follows from PSlya’s result on universalfactors.)
Thisresultisobtainedby generalizingatheorem of
J. L. W. V. Jensen
onthe location of the roots of the derivative of a polynomial with real coefficients(2)
and applyingitto integral functions(3).
The functionse
’,
,
>
O,
alsoturn out tohave property(a),
butit isdoubtful whether they have property(t).
The functions
(1.8)
willbe shownto have butafinitenumber of roots outsideanystrip
Jim
z[
_<
e, e>
0. This will becarried out by proving asymptotic formulas for(z),
depending on the expansion(5.7).
In
that formula anauxiliary function
H(s)
occurs which is a generalization of the F-function. Asymptotic formulas forH(s)
willbe derivedin4.
Now
letP(t)
and Q(t) satisfy the conditions of Theorem 2; then alsoP(t)
and Q(t)/(e
-t-
2+
e-t)
satisfy these.conditions.From
what is said above itis evident that the function
f_
e-()Q(t)(e
+
2+
e-t)-eit
dt has but a finite number of roots outside any stripJim
z_<
,.
Now
applying property()
withS(t)
e* -t-
2+.
e-t we obtainTheorem2.In
order to sketch the proof of Theorem 1, letP(t)
satisfy the conditions of Theorem 2 and suppose thatP’(t)
has purely imaginary roots only.Let
A be the smallest numberwiththepropertythatthe roots ofO(z)
f:
e-P(t)e dr, which has but a finite number of non-real roots by virtue of Theorem 2, lie inthe strip
Ira
z_<
A,
andsuppose A>
0. The functionP’(t)
is a strong200 N. G. DE BRUIJN
Im
z[
_<
41
A1
<
A.But
itis easilyseen from partial i,tegration that the latter integral equals-z(z). It
follows that the roots ofh(z)
alsolie in the stripIm
z_< A
This contradicts the minimum property of A.Hence
A 0and all theroots of
(z)
arereal.It
will be relatively easy to extend thisto the integrals of Theorem 1 oncon-sidering
F(t)
asthelimitofa sequence of functionsP(t).
The followingnotations areusedthroughout the paper.
Re
aandImadenote the real and imaginary parts ofa:aRe
+
iIm
a;*
denotes the conjugate ofa. Iff(z)
isafunction of the complex variablez,
thenf*(z)
is defined byf*(z)
(f(z*))*. A
polynomial or integral functionf(z)
is calledrealiff(z)
=-
f*(z),
thatisto say iff(z)
isrealforreal values ofz.All the trigonometric integrals considered in this paper are real integral
functions ofz.
2. Theorems on polynomials.
We
shall deal with linear combinationsof the type(2.3)
for a given polynomialf(z)
with real coefficients; the simplest case isf
(z)
f(z
+
i)f(z
i). Several propertiesoftheroots offl
(z)
areknown;
they all express in some way that the roots of
f(z)
lie closer to the real axisthan those of
f(z).
1. The number of non-real roots of
fl
(z)
does not exceedthat off(z).
(Thisis a special caseof Poulain’stheorem.
See [11;
Abschn.VI,
Aufg.63].)
2. If the roots of
f(z)
lie in the stripIm
z_<
1, thenfi(z) has real roots only. Namely,f(z
+
i) #f(z
/)]for
Im
z # 0. (Properties 1 and 2 arecontained asspecial cases inTheorem9a.)
3. If a an are the roots of
f(z)
andf
n
those off(z),
then1
Im
fl,--<
i
Im
a,].
(See [1;
Theroem5].)
These propertiesaremeant for illustration andwillnot beusedinthepresent paper.
We
shallnowderive a newresult ofthis type, Theorem3, whichforms the base of ourpaper.It
is ageneralization of thesecond propertyabove.We
first prove
LEMMA
1.Put
z x iy(x
andyreal), andf(z)
z+
A where A>_
0;let Xbea positivenumber.
If
isdefined
by(2
X2)i
(A
>
X)
and 0(4
_
X),
then we havef(z+iX)]
>
f(z
ih) if
(Izl
2)y
>
0 andlf(z
+
iX)[
<
f(z
iX) [/j"(I
z[
2)y
<
0.Proof.
We
evaluatef(z
+
iX)12
f(z
iX)12
{x
+
i(y+
X)
}2
+
A212
{x
+
i(YX)}2
+
5212
8yX(x+
y2
+
XA2).
The assertiondirectly follows.
Now
consider an arbitrary real polynomialf(z)
of degree>
0.It
can bewritteninthe form
(2.1)
f(z)
A
{(z
a,)
+
A}
l-I
(z-
b),THE ROOTS OF TRIGONOMETRIC INTEGRALS 201 where
a
b; real;A
0,A
0. Again, let Xbea positive number.To
anyA
exceeding X we construct the circular regionC
defined by(x
a)
y
_< /
X2;ifA
_<
Xwe takeC
to beempty.ByS
S(f)
wedenotethesum ofall
C
and thereal axis.We
now showTHEOREM 3.
If
f(z)
isOf
the type(2.1)
and X>
O,
a complex numberO,
thenall the rootsof
thepolynomial(2.2)
f
(z
-
iX)+
*f
(z
iX)(whichhasrealcoecients) lie in S.
Proof.
We
suppose to lie in the upper half-plane and outsideS.
Abbre-viating(2.1)
we writef(z)
A
I-I
g(z)I
hi(z).
Trivially hi(i"+
ix)[
>
[hi(
iX)I,
anditfollows fromLemma
1that also g,(+
iX)>
[g(
iX)].
Hence f("
-t-
iX)>
f("
iX)I.
If-
liesin the lowerhalf-planeand outsideS,
thenf(
-t-
iX)<
f(
i},)[.
In
both cases we conclude that i" isnota root of
(2.2).
We
remarkthat the limit caseof Theorem 3 forX--
0leads toawell-known theorem ofJ. L. W. V. Jensen
on the roots of the derivative of a polynomial.(See [3]
and[10;
Abschn.III,
Aufg.35].)
We
want to iterate the result of Theorem 3. Therefore, we firstdefine a setS
S(f), N
1, 2, which isthesum of the real axisand the regionsC,i 1, ,n. If/
>
XN
we takeforCN
the regionN-l(x
a)
y
_<
A
NX
",
which isbounded byan ellipse;ifA _<
XN1/2,
C
isempty.It
is readily deducedfrom Theorem 3 that if the roots of the realpolynomial g(z) lieinS(f)
thenthoseof g(z+
iX)+
*g(z iX),O,
lieinS+l(f).
THEOREM 4.Suppose
that all the rootsof
the polynomial q(u)o
au
,
aN
O,
lie on the unit circleul
1, thatf(z)
is areal polynomial, and thatX
>
O. Then therootsof
N
(2.3)
T-(T")I(z)
af{z
+
(2k
N)iX}
k--O
are contained in
SN(f).
HereT
represents a translation operatordefined
byT"f(z)
f(z
+
i).Proof.
The functionu-(u)
canbe writtenin the formN
U-%(U2)
Oly
(k
u+
@k u-l)
(Ol
z
O, k
0)
kl
By
Theorem3, the real polynomial(IT
*T-X)f(z)
has its roots inS(f).
A
second application shows that the roots of(T
-
*T-X)(IT
-
*T-X)f(z)
lie in
S2(f),
etc.,so that theroots ofT-Nq(T)f(z)
turntolie inSN(f).
THEOREM 5.Let f(z)
be areal polynomial whoseroots lie in thestrip202 N.G.DE BRUIJN
of
the polynomial(2.3)
satisfyIm
z[ <_
A2N
] if
A>
hN,
Im
z 0 if 4<
N
.
Proof.
Follows directly from Theorem 4andfrom the definition ofS(f).
3. Application to integral functions. Strong universal factors.
Let
a real integral function be given of thetype(3.1)
f(z)
Az"e
[I
(1
z/p,)ez/p’,
where
A
isrealand 0,misanaturalnumber,aisreal, p 0,Im
p-
4, p -2<
and therootspand
p* have thesamemultiplicity.It
ispossible to construct a sequence of polynomialsfl(z), f2(z),
all having their roots in the stripIm
z_<
4, converging uniformly tof(z)
in any bounded region. Since the product(3.1)
converges uniformly in any bounded region, it is ob-viously sufficient to prove it for the functions e(a
real),(1
z/p)e/"if
p is real and
(1
zips)(1 z/p*) exp (z/p z/p*) ifp is notreal.In
the latter casep71
W
p.-1
is real, and thus it only remains to be proved that e(a real)
istheuniform limitofasequence of polynomialswith roots onlyin the stripIm
z]
<
A.We
have indeede lim,(1
W
az/n)’,
converginguni-formlyin anyfinite region. Since also e-b"
(b
>_
0)
is the limit of a sequence of polynomials with realroots, the
same applies to the functione-b’f(z),
b>_
O,
iff(z)
satisfiesthecon-ditionsmentioned above.
Conversely, it seems probable that, if a sequence of real polynomials with roots in the strip
Im
z_<
A converges, uniformlyin any bounded region, toan integral function, then this function will be of the type
e-bf(z),
where the genus off(z)
is either 0 or 1.(The
corresponding problem for functions with real rootswas solved by Phlya[4].) We
do not need the solution of this problemforourpresentpurposes. (Afterthispaperwas writtenthe conjecture stated above has been proved byMr. J. Korevaar.)
Namely, we are able torestrict ourselves to integral functions of order
<
2; these functions satisfyIf(z)
<
exp(I
zi),
p<
2,for z sufficientlylarge. According toHadamard’s theory such a function can be expanded into a product ofthe type(3.1)
withor
1-2
<
o:
Ifwenowsuppose that the roots off(z)
lieinthe stripIm
z-<:
/xand that
f(z)
isreal for realz,
itimmediately follows thatf(z)
has all the prop-ertiesmentioned inthe beginning of thissection.We
thusobtainTHEOREM 6.
If
the orderof
the real integralfunction
f(z)
is<
2 andif
the rootsof
f(z)
lie in thestripIm
z-
4, A>_
O,
then there existsasequenceof
realpolynomials
f(z)
whoserootslie alsoin that strip, such thatf(z)
f(z)
uniformly in anyfinite
region.For
convenience we explicitly formulate the following well-known result(see
[10;
123, Abschn. 3, Aufg.201]).
THE ROOTS OF TRIGONOMETRIC INTEGRALS 203 THEOREM7.
Iff
(Z),
f
(Z),
f
(Z),
areintegralfunctions,f(z)
not identicallyO,
withf.(z)
f(z)
uniformly inanyfinite
region, andif
therootsof
f
(z), f(z),
allbelong to a givenclosedpoint-set
S,
then therootsof
f(z)
also lie inS.
Theorems4and 5 cn nowbe applied to sequences of polynomials.
THEOREM 8.
If
f(z)
satisfies
the conditionsof
Theorem 6 and q(u) thoseof
Theorem4,then therootsof
(3.2)
T-(T)f(z)
(
>
O, Tf(z)
f(z
+
i))satisfy
IImzl
_<
(A
-
N)
1/2/fA >
N
,Imz
0/f0_<: A_<)N.
Proof.
Let f(z)
f(z)
according to Theorem6.It
is easilyseenthat(3.3)
T-(TX)f(z)
--
T-(TX)f(z)
uniformlyin anyfinite region.
By
Theorem5, the polynomialsontheleft have their rootsinthe stripIm
zz
{Max (A
Nh,
0)
}t.
Now
the desired result follows from Theorem 7.THEOREM9.
Let
therealintegralfunction
f(z)
beof
order<
2and supposethatf(z)
has butafinite
numberof
roots outside the stripIm
z-
A./f
furthermore
q(u)
satisfies
the conditionsof
Theorem4, then all but afinite
numberof
rootsof
(3.2)
satisfylIm
z_
{Mx
(A
Nk,
0)}
t.
Proof.
We
putf(z)
g(z)h(z), where g(z) is a polynomial, and the roots of the integral functionh(z)
liein the strip]Im
z_<:
A.It
iseasily seenfrom the arguments usedin the beginning of thissectionthatf
isthe limit ofasequence of polynomials of thetypef(z)
g(z)h(z), whereh.(z)
hs no roots outsideIm
z_<
5.We
may actually tke forh(z)
polynomials which have, apart fromunumber ofreulones, only rootswhich reroots ofh(z)
also.According to Theorem 4, the roots of
T-X(TX)f(z)
lie inS(f.).
It
fol-lows from the definition ofS
thatS(f)
S(h.)
+
S(g)_
R
W
S(g), whereR
represents the set Iraz[
_<
{Max (A
N),,
0)}t.
Now (3.3)
and Theorem 7 show that the roots of(3.2)
belong toR
-
S(g). S(g) consistsof a finite number of ellipses. Since each ellipse contains but a finite number of roots of
(3.2),
our proofiscompleted.In
the special caseN
1, ),>_
A we can obtain more complete information on the number of non-real roots.THEOREM
9.If
thereal integralfunction
f(z)
of
order<
2hasexactly 2 rootsoutside the strip
Im
z_
,
andif
k>_
A>_
O,
O,
then thefunction
fi
(z)
f(z
+
ih)*f(z
ih) has2 non-realrootsat most.Proof.
The functionW(z)
f(z
iX) has not more than ] roots in thelower half-plane. The theorem nowfollows by
Lemma
2,6.
In
the general case ),<
A of Theorem 9 the analogue of Theorem 9a is not true.Even
the following statement is false’ iff(z)
is a real polynomial with204 N.G. DE BRUIJN
2k roots outside the strip
Im
z_
1, thenfl(z),
0<
},<
1, has at most 2/roots outside the same strip. Taking i, ),
--
0, we should inferthatif(z)
has at most 2/ roots outside that strip. This is incorrect, for instance, for
f(z)
(z
+
4)(z q-
121/2),
f’(z)
6z(z
-
-(12))
.
In
order to be able to apply the preceding results to trigonometric integralswefirst state
THEOREM
10.Let
b beanumber>
2, and lot thereal orcomplexfunction
F(t)
be integrableover
<
<
and satisfy(3.4)
F(t)
(F(- t))
*for
allrealvaluesof
t,(3.5)
Then the trigonometric integral
F(t)
O(e
-’)
(t
----
).
(3.6)
f(z)
F(t)e
’
dtrepresentsareal integral
function
of
order<
2.A
simpleproofcanbe foundinPSlya[8].
THEOREM
11.Let F(t)
satisfy the conditionsof
the preceding theorem and suppose that the rootsof
thefunction
S(t)
EM-M
aeXt,
a*
a_ aMO,
>
O,
lieonthe imaginary axis. Then we have"If
theroots(all
butafinite
numberof
theroots)
of
(3.6)
lie in thestripIm
z_
/thentheroots(all
butafinite
numberof
theroots)
of
the real integralfunction
(3.7)
F(t)S(t)e
’z’dt
lie in the strip
Jim
z]
_<
{A
1/2Mh}
/fA >
(1/2M)
1/2,
and are realira
<_
h(1/2M)
.
Proof.
Since the roots ofS(t)
re purely imaginary, the roots of thepoly-nomial
(u)
_M
au-
lieon theunit circleu]
1; hence(u)
satisfiestheconditionsofTheorem4
(2M
N). Now
ourtheorem immediately follows from Theorem 8 (Theorem9)
and from the fact thatT
-
_
F(t)e
dtF(t)e’e
TMdr,
whencef_
f_
M ae*Xte
T-x/(Tx/)
F(t)e’"
dtF(t)
E
dt. --UA
slightly better result can be obtained ifS(t)
containsfactorse
x -[-*e
TM.
For
instance, the functionS(t)
e
x*-
2-q-
*e
-
gives rise to the strip[Im
z_
{A
1/2},}1/2,
butS(t)
e
x’-
*e
-x gives the stripJim
z_
{/
),}t,
which follows from Theorem 8(Theorem 9)
on taking(u)
THEROOTS OF TRIGONOMETRIC INTEGRALS 205 THEOREM 12.
If
S(t)
I]
(k
e’
+
*e-X),
where 1,>
O,
k 1, 2,
N,
and the roots(all
but afinite
numberof
theroots)
of
(3.6)
lie in the stripJim
z<_
/, then the roots(all
but afinite
numberof
theroots)
of
(3.7)
lie in thestripJim
z[
_<
{Max (A
h,
0)}
.
Theorem 11 proves the statements
(a)
and()
madein the introductioncon-cerning strong universal factors. Although it will not be used in this paper,
we shall prove here that also the functions e
1/2x2t2,
),>
0, have property(a).
We
donotyetknow whether theyhaveorhavenot property().
THEOREM 13.
If
F(t)
satisfies
the conditionsof
Theorem 10, andif
all therootsof
(3.6)
lie in the strip[Im
z<-
A,
then alltherootsof
g(z)f: F(t)e1/2X2’e
TMdt
lie in thestrip
(3.8)
}Imz] _<
{Max
(A:
,0)}
.
Proof.
By
Theorem 12, the roots of g(z)f_ F(t) (cosh
kt/N)N’.e
TMdt lie in the strip
(3.8).
Owing to Theorem 7 it is now sufficient to prove that g(z) --+ g(z) uniformly in any finite region. Now this follows from(3.5)
and from the fact that for>
},wehave(3.9)
e-t’2’ (coshkt/N)
----)e-t’’+t’,
uniformly in
-
<
<
o.(3.9)
results from the inequality cosh y_<
et’,
<
y<
,
whence(cosh kt/N)
<_
e1/2’.
For completeness we mention the following theorem, a slight extension of Pdlya’s result on universal factors
(see
[8])
which dealswith thecase A 0.THEOREM
14.Let F(t)
satisfy the conditionsof
Theorem 10 and suppose that the rootsof
(3.6)
lie in the stripJim
z]
_<
A.Let (z)
bea real integralfunction
of
genus 0 or 1(that
is, afunction
of
the type(3.1)),
with real roots only. Then the rootsof
f
F(
t)(it)e’’
dt(3.10)
liein the strip
Ira
z_
A also.A
proof can be given by introducing quite trivial modifications in PSlya’s proof for the case A 0.It
is,however,
also possible to deduce Theorem 14fromTheorem 12.
It
ishardly necessary to say that the function (it)e’,
a>_
O,
satisfies as a"universal factor" aswellas (it), that isto say thattheroots of
(3.11)
F(t)(it)e""e
dtlie in
Jim
z[
_
A for anyF(t)
satisfying the conditions of Theorem 14.But
the roots of
(3.11)
evenlie inanarrowerstrip, which canbe shown by applying Theorem 13 to(3.10).
20 N. G. DE BRUIJN
4. Application of the saddle-point method. siderations will beplayedby the function
An
important part in ourcon-(4.1)
H(s)
e-g(U)u"du,where
Re
s>
0 and(4.2)
g(u) u+
0/1u(N-1)/jV+
0/2U(N-2)/N+
+
We
are interested in the asymptotic behavior ofH(s)
forRe
s>
0,Is
large; the natural numberN
and the coefficients al 0/N remaining constant.The 0/’sneed not be real.
For
0/1 0/N 0 we haveH(s)
sF(s),
sothat asymptotic formulas forthe F-functionwillappearasspecial cases.
In
the sequel, positive constants al a2 will occur, chosen sufficientlylarge to suit some special purpose. These numbers may depend on
N,
0/1av, butnot on s.
The integrand of
(4.1) has,
fors large,just one saddle-point in the domain argu<
r satisfying(4.3)
;g’() s.Namely, puttings z
,
w(w
andzarepositive forsandu positive), weobtainthe equation
wN
-t-
(N-
1)N-’0/,w
N-+
(N
2)N
-’a2w-
-1-
-t-
N-10/-lw
zr,
whence, for
w[
>
al the function1/z
can be expanded into a convergentpower series z
-
w-
(N
1)N-0/lw
-2q-
....
Solving this equation by the Biirmann-Lagrange inversion formula we obtain w
-
z-
-1-
z
-
-b/3z-3
-
whence wNz(1
+
2
z-1
+
3
z-2
+
o)--N
z(1
_
lZ--1
+
2Z-2
+
"’’),
convergentfor]z]
>
a. It
followsthat, forIs]
>
aa, theonlysolutions of
(4.3)
are(4.4)
where s
-lv,
s-3/’r,
are derived from one and the same branch of thefunc-tions-i/r.
We
shallrestrictourselves to the regions args<
r,1 [arg
u<
r,andhencewe only have to consider thecasewheres/Nispositive fors 0.
We
henceforthdivide intotwocases,Casea"
4r/9
<
[args
<
1Case#:
args]
_<
4r/9,
andwe put
L
1/2
[
inCase
0/,L
1/2
[
1/2NinCase
f,M
I
1/8N’
qL
in both cases.(The
constant4r/9
is ofcourse not essential.In Case
it may bereplaced byany other number
<
inCase
0/however, itcannot be replacedby arbitrary small positivenumbers.)
Our
integration contour for the integral(4.1)
will beI.
A
straight line from 0 toq.II.
Thestraight lineuq-
y,
-L
_<
y<
o we notice that argt[ <
THE ROOTS OFTRIGONOMETRIC INTEGRALS 2O7 Themajor contribution to
H(s)
is furnished bythe integralalongII
passing throughthe saddle-point.
Its
valueis(4.5)
e-aC")u’du1
e-x(’) dy,L
where
K(u)
g(u)
sloguanduT
yt.
We
havedK/dy
{’(u)
su-
}
u-
lug’(u)
g’(O(4.6)
u-11
t(dlug’(u)l/dy)
dyu-11;;
(d{ug’(u)l/du)
dy.It
is easilyseenfrom(4.4) that,
forIs
>
a4,Re
s>
0,y>_ --L,
wehave(4.7)
(4.8)
arg(u/})
<
3r/8.
I
follows from(4.6),
(4.7), (4.8) that,
if!1 >
a
Re K(u)
decreases from y-L
toy--M. Furthermore,
for y>
M
wehave,
againby(4.6), (4.7)
and(4.8),
forI1
>
aT,(4.9)
I
follows that(4.10)
Re
(dK/dy)>
-
$/’u y>
ly/(1
-]-y)>
1-.
e-x(’’ dy
<
1/2l/i
1
exp(--K(/jM/Ii))
I,
(4.11)
e-KC’) dy<
81exp (-K(/I
+
M$t))
l-We
now consider the interval-M
_<
y_<
M.
SinceM
/jl/s,
wehaveu-’
1q--
Y-i
<
asY1-1,
Is]
>
ag and by(4.6)
and(4.7)
wehave dg/dy y+
y%-][
<
alo([
y-’/
+
Y%-’
l)
([
S>
an).
Hence,
for--M yM,
(4 12)
K(u)
K(O
YW
Ya-i
<
a,o(I
y-’/
I+
y’C
I),
(4.13)
-K()
208
.
G. DEBRUIJNFrom (4.12)
we easily infer that the right members of(4.10)
and(4.11)
are0(-’/e-(),
sothat(4.5),
(4.10),
(4.11)
and(4.13)
give(4.14)
e-()u"du(2r)1/2e-()
{1
-t-
O(-I/N)
In
the secondplacewe considerthe integral(4.15)
e-()u" dufor which the
Cases
aand have to betreated separately.-r
itiseasilyseen, onCasea.
Sinceq-
1/2]lttand4r/9
<
args<
drawing a figure and carrying out some elementary trigonometric calculations, that arg q
>
arg(e=’/
e
,1)
e>
5r/9
forIsl
large.It
follows that, ifIs
>
a.
andif u runsthrough the straight linefrom 0 to q, the maxi-mum value of e-(") is attained at u q. The same being true for(since
Re
s>
0),
we obtain(4:.16)
e-’
d<lee
-"
I.
It
wasnoticedbefore that, ifI1
>
,
Re K()
decreasesfrom u q to M(t,
soghagRe
K(q)>_
Re K(
M(1/2),
whence gheleft side .of(4:.16)
isO(-/%
-/’)
(see (4:.12)).
Case.
argsl
_< 4r/9.
We
putu qt, O<_
<_
1, q--
i/
Then wehave
(see (4.6))
(4.17)
dK/dt
q{g’(u)
s/ul
t-{ug’(u)
g’()},
and ugh(u) g’(ti)
f’
{ug’(u)}’
du,{ug’(u)}’
1-+-O(U--I/N).
Hence aa
and
a
canbechosen such that for u>
a,,Is
>
al,(4.18)
ug’(u) g’()-(1
+
)(
u)
(I
l<
sin/20).
Now
it is easily seen from a figure that forIs
>
a
arg(
u)
arg
$1
<
9r/20.
So
itfollows from(4.17)
and(4.18)
that, forsl >
a,,Re
K(u)
decreases if u runs through the straight line segment from u Uoa,q/I
q tou q,whence e-("’_<
e-() foru on thatsegment.Let
u bethe maximumvalueof e-(") on the remaining segmentfrom 0 toUo. Ifu liesbetween0 anduo, 0
_<
_<
aa
q-,
wehaveI-’"’
_<
1
‘(u/uo).-K‘.>l
<_
I’"’>
i
-‘>!.
Sincetheconstant
a
tz e"(") isindependent ofs, we obtain(.)
e-’u"
&
<
(1
+
.)I
e
-’’
THE ROOTS OF TRIGONOMETRIC INTEGRALS 209 THEOREM15.
If
bisapositive constant andH(s)
isgivenby(4.1),
wehave(4.20)
H(s)
(2r)1/2e-()’{
1+
O(S
-l/N)
},
uniformly
for
Re
s>
-b, s]--oo.
Here s+
"rs-i/r-k
"rs-/+
(absolutelyconvergent
for
slarge)satisfies
g() s.(We
noticethat forN
iweobtaing(u) 1, sand then(4.20)
becomes the familiarStirling formula forF(s
+
1).)
Proof.
ForRe
s>
0 the result follows from(4.14), (4.16),
and(4.19).
In
order to be able to consider values ofs in the left half-plane, we continue
H(s)
bythe formula
(s
+
1)H(s)
H(s
+
1)
+
al(N
1)N-H(s
+
1N
-i)
(4.21)
.+-
-
aN_N-1H(s
+
N-),
which canbefound bypartial integration:
(s
+
1)H(s)
f:
e-(")g’(u).u
+du. Itfollows from(4.21)
thatH(s)
canbe continued overthe whole plane but forthe points s -1, -1
l/N,
-12IN,
which arepossible poles ofH(s). It
alsofollowsthat(4.21)
holds for all values ofsexcept for these points. The functionh(s)
(2r)te-()
’,
s+
/1sl-1/N+
is regular andsatisfies
(4.22)
sh(s
p)/h(s) 1+
O(s
-/v)
uniformly for0_
p_
b+
1,[args]
<r,
sl >
a.
This can be shown bysome elementary calculations.
Now
suppose that(4.20)
is true for the regionRe
s>
-k/N.
Then,
by(4.21)
and(4.22),
it can be verified forRe
s>
-(k
+
1)IN. Hence
the theorem follows by induction.In
the following Theorems 16 and 17 we shallcollect someresults concerningH(s)
to be usedinthe nextsection.THEOREM 16.
Put
s r+
iv(r
and real),andlet b andcbepositive constants. There are constantsA
and C(A
and C may depend on aar)
such that,for
-b<
a<
c"
I,
r>
A
andanypositivenumber pwehave(4.23)
[U(s
p)[
< Cl/!s
[-
U(s
[.
Proof.
TakeC
{1 +io/1 [(1
1/N)
+
+lav-
I/N}(1
+
c)
1/2.It
follows from Theorem 15 and from(4.22),
thatA,
A
>
C,
canbe determined such that for 0_
p 1, a_
-b,[r
>
A,
we have[sH(s
p)/H(s)<
(1
+
c)
<
C
+.
Now
the casep 1 can beproved by induction.Suppose
that(4.23)
is true for 0_
p_
k/N
(k
an integer, k>_
N),
and thatk/N
<
pl
_
(k
+
1)IN.
Then we have[U(s
pl[o-t-_
1)[
<
C’
[H(s)[/[
s[o,-1,
H(s
p, / 11/N)
<
C’//H(s)
I/[
s-/’/",
and so on, andhence,
since
[rl
>
A
>
C,
IH(s
p,+
1k/N)
<
C
’
[sl
’-’ [g(s)[,k
1,2,N
1.It
nowfollows from(4.21)
that210
.
G. DE BRUIJN_<
(1
q-
a,I(N
1)N-’ -t-
-t-
a-N-’)c’ls
I’-’
H(s)
I.
Since
<
c r and p,>
1, wehaveis
p,-t-
1>-
Is
I(1
q-
c)
-
and(4.23)
follows forp m.THEOREM 17.
If
b, cand aregiven positivenumbers, thenpositivenumbersA
and
C
canbefound
suchthatH(s)
>
Ce
-’+)1+’"s a
-k
it, inthe regiona
>
-b,
Isi >
A.
Thisfollows from Theorem 15bysomesimple calculations.
5. Proof of Theorems 1 and2.
In
ordertocompletetheproof forTheorem2, which was outlined in theintroduction,
it is sufficient to show(Theorem 18)
that the integral
(1.8)
has butafinitenumber ofroots outside any stripIIm
z<
e,
>
0. Theorem2follows(for
Q(t)) by applying Theorem 18 to the functionQ(t)
Q(t)/(e
-t-
2-1-
e-)
whichalso satisfies the conditions of Theorem2,
andusing Theorem 12or 11.We
shall establishaseriesexpansion (formula(5.7))
for(z)
whichgeneralizesaformula ofP61ya
(see [6;
formula(11)])
for thefunction(1.2).
Theoccurrenceof Q(t) in our
(z)
makes it very difficult to carry out P61ya’s method in the present case.We
therefore develop a new method which uses contour inte-gration.Let
thenumbers a a a be defined bypy
-t-
pr-yr-1-t-
-t-1/N N 1/N N--1
1/2po
=--
(Pr Y)-t-
a(pr y)-I-
-t-
a and define g(u) by(4.2).
Thenwe
have,
by(1.6),
(5.1)
P(t)
g(per’)
-t- g*(p*e-’).
Let B
>
0besuchthatQ(t) isregularforRe
>_
B.
Considerthe following pathsW,
W,
W.
inthecomplext-plane.W
consists of ahalf line from 2riq-
up to the point 2ri -{-B,
the line segmentfrom2ri--
B
toB,
and the real axis fromB
to-t-W.
isthe contour ofarectangle, taken in positivedirection, withverticesB,
2ri
+
B,
2riB,
--B.Wa
consists of the segments -o to-B,
-B to 2ri B and 2riB
to2ri
Considering
e-’(*)Q(t)e
’
as ourintegrand, we immediately find that(5.2)
,q-L.q-L.
f:i’_+:
(1
e-2,’))(z).
For
onW1,
andBsufficientlylarge,we havean expansionTHE ROOTS OF TRIGONOMETRIC INTEGRALS 211
(5.4)
convergingabsolutely and
_
uniformlywithrespecttot.Hence
,
(exp(-(r’)))e
-’’/’"’d.
The integral
f.
(exp(--g(pe’)))e
’
dt(5.5)
convergesforany complex value ofz.
To
evaluate it, firstsupposeIm
z<
0,
then shift thevertical part of the path
W
infinitely to the left.It
follows that(5.5)
equals(1
e-)
f(R)
(exp(--
g(pve)))e
’
dt andhence(5.6)
(exp(--g(pe2C)))e
’’
dtN-(1
e-")pT’/nH(izN
-
1),
w
where
H
is the function introduced in(4.1).
SinceH(s)
is regular over the whole plane, with exception of simple poles at s-k/N,
k 1, 2, theright side of
(5.6)
isan integralfunction, and(5.6)
holdsfor allz.We
candealwith.
in thesame way,and after that(5.2)
and(5.3)
leadto expansion 1,,+,,/H(.iz--v
)
(5.7)
+
#’*P*(’
N
--1T
1--efw.
e-"t)Q(t)e
’t dt---l(z)
+
3(z)
-
.(z).
Thisformulaholds for all values of
z,
with theexception of the points z ik, k 0, :1, :t:2,We
are nowable to proveTHEOREM
18.If
P(t)
andQ(t)
satisfy theconditionsof
Theorem2,andif
weput(5.8)
(z)
N-l_:p(’/K)lNH((iz
+
K)N
-
1),
thenwehave,
for
O,
(5.9)
(z)
(z)(1 -t-
o(1)}
uniformlyinthe half-plane
Im
z_
-,
and(z)
hasbutafinite
numberof
roots212 N. G. DE BRUIJN
Proof.
We
first consider the point-setR
defined by the inequalities(5.10)
-IRe
z<
Im
z<
-e(I
z>
A).
(On
generalizing Theorem 16 it is possible toextend our considerations to the regions defined by the set ofinelualities
IRe
z>
1,Im
z<
-e, z>
A.
The power of the Phragmt!n-Lindelhf theorem applied below however enables
us to restrict ourselves to the smaller regions indicated
above.)
Here A
ischosen sufficiently large to suit some conditions indicated in the followihg.
By
Theorem 16, thefirst seriesof(5.7)
satisfies(5.11)
l(z)
(z)
_<
"P.-KPN-I
(K+)/iVfl+v/1VZ--(K+v)X.Zl
/(Z)
-K+I
<
C
[z
1-1/2V
((z)I,
if
A
is sufficiently large and zR.
(C1
C2,
may depend onN,
K,
p;,
but not onz.)
Analogously(5.12)
a(z)
<
zFurthermore wehave. again by Theorem 16, taking p (iz
iz*)/N,
N
1H
N
1<C4]z[
-2’/N
H
iz+K
N
-1for z
R
andA
sufficiently large.Hence,
by(5.11)
and(5.12)
(5.13)
&(z)
+
&3(z)
(z){1
+
o(1)}.
We
now turn to the third part of(5.7),
i.e., the function&2(z).
It
follows from the regularity properties ofQ(t)(see
Theorem2)
that the pathW2
canbe reducedto the pathW:,
consisting ofarectanglewithre,icesB
W
(
T
3 argpN)i/N;B
+
2i+
(--
arg pN)i/N;--B
+
2i+
(--argp)i/N;B
(
+
argpN)i/N for a certain positive number The rectangle being independent ofz,
wefind, forIm
z-,
_<
C
exp{B[Imz[
(1/2r
+
i)[Re z/N[
+
argpRe z/N}.
From
Theorem17,
with c>
BN,
<
,
we noweasily derivethat(see (5.8))
(5.14)
(z)
(z)o(1)
([
z[--)
uniformlyinthe region considered.
It
follows from(5.13)
and(5.14)
that(5.9)
holdsuniformly for zR.
In
the regionR, defined byIm
z[
<
[Re zl,
zl >
A,
we have, by Theorem 10,(z)
0(exp[z
Ix),
<
2.On
the boundary ofR,
we haveTHE ROOTS OF TRIGONOMETRIC INTEGRALS 213
(z)
(z){1
-t-
o(z)]
andfurthermoreI(z)
>
1 for z inR
providedthatA
has been chosen sufficiently large.It
follows, by a well-known theorem of Phragmdn-LindelSf, that(5.9)
holdsuniformlyinR
Since
H(s)
has only a finite number of roots forRe
s>
-b, which follows from Theorem 17, now Theorem 18 is completely proved.Hence
Theorem 2istrue
(see
Introduction).We
nowturn to the proof of Theorem 1.In
the first place we deduce from Theorem2and Theorem 11, by an argumentexplained in theintroduction,THEOREM
19./f
P(t)
_
pe(N
>
O, Re
p>
O,
p* p_) andif
all the.rootsof
itsderivativeP’(t)
arepurely imaginary,then thefunction
dthas real roots only.
We
easily inferTHEOREM
20.If
all the rootsof
the derivativeof
the polynomialf(t)
qt
(N
positive andeven)
are purely imaginary, andif
qzv>
0(so
thatf(t)
isreal
for
purely imaginary), then thefunction
(z)
f_
e-()e’
dt has real roots only.Proof.
We
putf’(t)
NqN
I-
(t
ip,), p,real, and for>
0,N-1
Nqz
I
ksinh(t-
ip,))-’x(t),
f
Cx(r)dr
hx(t).
zo
cx(t)
hasthe formN-1
(5.15)
ce’/x.
--N+I
Since
N
iseven wehaveCo 0;henceCx(t)
also has the form(5.15). It
iseasily verified thatCx(t)
is realfor purely imaginary vlues oft, and furthermorex(ht)
2N(N
1)-l(1/2}k) Nque-"’/x
e.v-,)t+
....
hr-1
It
follows,that,
for>
ho 2r-
I,
p.1,
the functionPx(t)
satisfies the conditions of Theorem 19, so that
x(z)
f-
e-t()e"
dt has real roots only.It
remains to beproved that(see
Theorem7)
(5.16)
limx(z)
(z)
,-+
uniformly in anyboundeddomainofthe z-plane.
We
candetermine positiveconstants i<
r/2N
andi
suchthat(5.17)
(sinhw)/w
>
1/2,
arg (sinhw)/w
<
1/2N,
for any value of w satisfying argw
<
i andIm
w<
i. Furthermore wecanfix positive numbersA
andA
such that the numbers w(t
ip)/h lie in that region for all vlues of>
A,
k> A
and u, 1_
u_
N
1.It
214 N, G. DE BRUIJN
nowfollows from
(5.17)
that a positive constant c existssuch thatRe (t)
>
ctN-lfort>A1,X
>
A.. Hence
(5.18)
lim lim supe-*(t)e
ztdt =0
uniformly in any bounded domain ofthe z-plane. The same
holds,
of course, forf:.
Since qn
>
0,N
even, wealsofindthag(5.19)
lirae-te"
de 0uniformlyin any bounded
domain,
andtheameforJ’;.
For
any fixedvalue ofA
wehave(5.20)
lime-*X(’)e
’’
dt e-f(’)ez’dt-co A A
uniformly in any bounded domain, because
(t)
--
f(t)
uniformly in--A
<
t<A.
From (5.18), (5.19)
and(5.20)
we infer(5.16)
andourtheoremisproved.Proof
of
Theorem 1. Suppose thatf(t)
satisfies the conditions of Theorem 1. These conditions can also be expressed in the following form(see [4])"
f’(t)
is of thetype
f’(t)
aet*t*+
H:I
(1+
2vt2
)
wherea>
0,b>
0, kaninteger>
0,.
_>
0, r 1, 2,---,
0
<
o.Now
letf.(t)
be defined byf’(t)
a(1
A- bt2/n)"t
+’I-’
(1
-I-
it,t2),
f,(0)
f(0)
andput(z)
f:
e-’(t, eTM dr,n(z)
f:(R)
eTM eTMdr.It
iseasy tofindpositive numbersA,
and no, such thatforn>no,t
>
Aort
<
-Awehavef(t)
>
dandfn(t)
>
d.
Since
f,(t)
f(t)
forn--.
,
uniformlyinanyfinite t-interval, we noweasily inferthat(see
proof of Theorem20)
(5.21)
lim4,=(z)
q,(z)uniformlyin anybounded domain of thez-plane.
It
is easily seenthat the polynomialsf=(z)
satisfy the conditions of Theorem 20,whence itfollowsthat(I).(z)
has real roots only. Application of(5.21)
and Theorem 7completestheproofofTheorem 1.6. Functions of the type
(1.9). We
considerthe functions of the type(1.9).
In
connection with the Riemann hypothesis(see 7)
it may be important to investigateclassesof functions of thistype.In
the following we shall prove some results concerning these functions.Very
little of the preceding sections will be neededhere,
since the reality of the roots off:
C(t)e
TM dr,C(t)
exp(-h
cosht),
(special case of Theorem 1 or19)
was already proved by P61ya(see (1.2)).
Only Theorem 25 requires asymptotic expressions and stronguniversal factorsas itstools.THE ROOTS OF TRIGONOMETRICINTEGRALS 215
LEMA
2.If
U(z)
andV(z)
are real polynomials such thatW(z)
U(z)
"F
iV(z) hasnrootsinthelowerhalf-plane,then
U(z)
hasnpairsof
conjugatecomplex roots at most.(The
case n 0 is the well-known theorem of Hermite-Biehler(see [10;
Abschn. 3, Aufg.25]).)
Proof.
We
may assume thatU(z)
andV(z)
have no real roots in common,so that
W(z)
has no real roots.We
also assume that the degree m ofW(z)
satisfiesm
>_
2n, for otherwise the result is trivial.Now
if zrunsthrough the real axis from -oo to oo, the argument ofW(z)
increases by an amount ofr(m
2n). Hence
there areat least m 2n different points on the real axis whereW(z)
ispurely imaginary. If z 0o isnotone of these points, wethusfind m 2n real roots of
U(z)
at least, and otherwise at leastm 2n 1.But
in thelastcase thedegreeofU(z)
must be<
m, so thatU(z)
has atmost 2ncomplexrootsinbothcases.THEOREM 21. The
function
(1.9)
(which has but afinite
numberof
non-real rootsonaccountof
Theorem2)
hasN
pairsof
conjugatecomplex roots at most.Proof.
Thefunctions, whereC(t)
exp(-
cosht),
>
0,(6.1)
(z)
C(t)e
’z’dt
(6.2)
q(z)(C(t)
cosht)e
dr, q(z)(C(t)i
sinht)e
’
dt have real roots only (Theorems 1 and11),
whereas the roots of q(z)
ql(Z) iq.(z) lie in the upper half-plane.
By
partial integrationit iseasilyseenthat, for k 0, 1, 2, the functions zq
(z)
andz q.(z)
are of the formk+l
z(z)
C(t)
=.e
e dt(1
1,2)
-k--1 ()
1, 2,are 0. Further-where
a
().
a()_.,
andthehighest coefficientsa+.,-()
isrealifk isodd,and purely imaginaryifk
+
iseven.It
followsmore an,
thatthe function
(z),
given by(1.9),
canbeexpressed as alinear combination of(z),
i(z),z(z),
z-(z),
(z),
z-(z),
with real coeffi-cients. Sincez(z)
-h(z)
wehave(6.3)
z(z)
A (z) (z)
+
B(z) (z),
where
A (z)
andB(z)
arereal polynomials ofdegreeN
at most.Now
ifF(z)
andG(z)
are real polynomials, of arbitrary degree, with the property that the roots ofF(z)
iG(z) all lie in the upper half-plane, thenA(z)F(z)
+
B(z)G(z)
has at mostN
pairs of conjugate complex roots. This follows by application ofLemma
2 to the functionU(z)
iV(z)(A(z)
+
216 N.G. DE BRUIJN
iB(z))(F(z)
iG(z)).On
approximating(z
i)l(z)
i2(z)
by poly-nomialsF(z
i) whose roots also lie in the upper half-plane, we find that(6.3)
hasN
pairs of conjugate complexroots at most.In
connection with the Riemann hypothesis it appears to be important to findlarge classes of functionsN
(6.4)
ane"
(a*
a-n)
-N
withthepropertythatthe integral
(1.9)
has real rootsonly.For
instance, this followsfrom Theorem,ll, ifall the roots of(6.4)
lie on the imaginary axis; but this result does not help much in the direction of the Riemann problem, which seems to berelatedto functions of thetype(6.5)
C(t)
(,
+
cosht)e
’
dr,
n=l
where p are real and 1.
In
the following we establish some newresults, which do not help either, but which may serve s material for obser-vation.
THEOnnM
22.If
>
O,
O,
theintegrals(z)
f
C(t)(
+
cosht)e
TMdt and
(z)
f:
C(t)(
+
cosht)e
dthave realroots only.Proof.
(The
reality of the roots of(z)
can also be proved as follows.It
follows from Theorem 21 that(z)
has t most one root in the upper half-plane. SinceT
(z)
is real andeven, this possibleonemust be purely imaginary.But
since the integrand is positive forz purely imaginary sucha root does not exist. The same argument is used in the proof of Theorem23.)
By
partial integration we easily expressT
and in terms of the functions(z)
and(z)
definedby(6.2)"
(6.6)
z,(z)
z,(z)
x,(z),
(6.7)
z2(z)
z(z)
+
{z
2(1
+
)}2(z).
The roots of the polynomials z
ih
andz+
i{z
h(1
+
t)}
liein the upper half-plane and hence, by the lemma,(6.6)
and(6.7)
have real rootsonly(see
the proof of Theorem21). We
noticethat the same can be said if -1_
p
_
0, for thenp cosh and-
cosht
areuniversal factors.On the other hand it is not difficult toshow that both integrals have a pair ofpurely imaginary roots if
-
ispositive and sufficiently large.The following theoremisobtainedbyageneralization of themethod employed above. Theresult, however,seemsto be too complicated for application to any wide class of polynomials.
THEOREM23.
Let
be>
0 andletf(y) beareal polynomial, suchthat~(z)
:
C(t)f (cosh t)e
iTHE ROOTS OF TRIGONOMETRIC INTEGRALS 217
fl(Y)
Yf(Y)
T
f(Y)
X-’f’(y)
+
X-2f"(y)
-
if
>_
O.<
O,
providedthatfl
(y) does not change signfor
y>_
1.Proof.
Puttingq,(e)
(C()f(eosh )eosh
)e’"
d,
It
isalso truefor
(C(t)f(cosh
t)isinht)e’"
dr,we findby partial integration
(6.8)
zC(t)fl(cosh t)e
’’
dtz4(z)
X(z).
Furthermore, wenoticethat the roots of
-(z
i)(z)
i:(z) all have imaginary part 1.For
>_
0 the functionA (z)
iB(z)z
iX has noroots inthe lower half-plane sothat theargument used inthe proof of the pre-ceding theorem shows that the function
(6.8)
has real roots only.If
<
0 we inferthat(6.8)
has onepair of conjugate complex roots at most. Since that function isan odd function ofz,
these possible roots must bepurely imaginary.But
iff
(y) does not change sign for y>_
1 and if z is purely imagi-nary, thenC(t)f(cosh t)e
TM doesnot change sign for<
<
.
It
follows that the integral does not vanish.The following applicationmay be of some interest. The functionf(y) satisfies the conditions of Theorem 23, for
(6.1)
has real roots only andyN
(cosh t)
Nisa universalfactor. ConsequentlyTheorem 23 showsthat, forX
>
0,(6.9)
C(t)
1+
X cosh+
2
+
+
N
hasreal roots only.
THEOREM 24.
Let
the polynomialf(y)of
degreeN
have negative roots only, and letX beanumber>_
1/2N.
Then thefunction
~(z)
has realroots only.It
may be surmised that the condition X>_
1/2N
can be replaced by a much weakerone.Proof.
We
may assume thatf(y)has noroots for -1_<
y_<
1, the factors y a cosh a(-1
_<
a_<
1)
being universal factors(Theorem 11). We
now proceed by double induction, in the first place with respect to the degreeN
off(y)(the
theorem is true forN
0)
and secondlywithrespecttoanumber n,the smallestpositive integer with thepropertythatf(y) has atleast oneroot exceeding -1n/X. We
shall reduce thecase(N, n)
either tolowerN
or to lowern;the case(N,
0)
will always be reduced to lowerN.
Suppose
thatf(y)(of
degreeN)
has negative roots<:
-1 only, the largest of which, p say, satisfies-1-n/X
<
p<_
-1-(n- 1)/X,
n>_
1.Now
218 N.G. DE BRUIJN
consider the polynomialfl(y) f’(y) Xf(y). Sincef(y) has real roots only, the same applies tof(y).
Now
f(y) has no roots for y>
1 and exactly onerootpsatisfyingp
-t- -1
<
p_<
1. Namely,f’(y)/f(y) decreasesmonotonically for y>_
p.For
y p-t-
h-i
we havef’(y)/f(y)>
(yp)-I
h, and, sincetheroots off(y) aresupposed to be
<
-1, theinequalityf’(1)/f(1)
<
1/2N
_
holds.Now
ifp<
-1,the polynomialf
(y) belongstoacaseoflowernand hence(6.10)
f_
c(t) f(cosh t)e
’ztdt
issupposedto have real rootsonly. If,
however,
-1_<
p_<
1, the polynomial f.(y)f(y)/(y
p)is ofdegreeN
1, anditfollowsthat therootsof(6.10),
with
f2
instead off, arereal.But
sinceinthat case cosh p isa universal factorthesameapplies to(6.10)
itself.Again applyingauniversal
factor,
we findthati
C(t)f(cosh
t)sinh e’zt dt zC(t)f(cosh
t)e
’ztdt
hasreal rootsonly. This completes ourinduction.
We
have notyetbeen able to generalizeTheorem 24 to cases wheref(cosh t)
isaninfiniteproduct of factors 1 c,cosh
t,.
c
>
0. Suchanextensionmight be given perhaps by carrying outasuitable reductionprocessand using Theorem 25below.In
Theorem 26 weshallusesuchamethodin adifferentcase, whereareduction processcanbe found indeed. Theorem 26 may be of some interest sinceitgivesa result ofa typeweshould like tohave for functions of the form
(6.5).
THEOREM25.
If
O,
0 8.,1
andif
n is anatural number, then thereexistsapositive number
A(h,
8,n)
withtheproperty that therootsof
(6.11)
(z)
f_
C(t) f(cosh t)e
’ztdt
lie inthestrip
Im
z<-
A(,
8,n)
for
any real polynomialf(y)of
degreenwhose roots lie inthesector1/2
<_
arg y_<
i.(It
is possible toprovethesame for the region consisting of the real axisand circlezl
_<
A(),,
8,n).
This can be done by introducing a denominator(e’ -t-
2-
e-t)
(see
theproofofTheorem2).)
Proof.
Thefunction(6.1)
canbedevelopedasfollows:(z)
(- 1)
r(iz)
+
2.(- 1)
r(-iz)
,--0