On the Distribution of Values of Trigonometric Sums with Linearly Independent Frequencies
Author(s): M. Kac
Source: American Journal of Mathematics, Vol. 65, No. 4 (Oct., 1943), pp. 609-615
Published by: The Johns Hopkins University Press
Stable URL: [Link] .
Accessed: 07/12/2014 21:26
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
[Link]
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@[Link].
The Johns Hopkins University Press is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Mathematics.
[Link]
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
ON THE DISTRIBUTION OF VALUES OF TRIGONOMETRIC SUMS
WITH LINEARLY INDEPENDENT FREQUENCIES.*
By M. KAC.
1. Let
( 1) f (t) aakcos 27rXkt,
kC=1
where the a's are real and the A's real and linearly independent. Denote by
N,((a) the number of t's in (- T, T) for which f(t) - a and consider the
ratio NT (a)/2T. The problemiof finding
(2) E(a) lim-NT(a)
T->oo2T
was considered by E. R. valn Kampen, A. Wintner and the present author
in an evenimore general case than (1). However,with the exception of the case
= 2, a, = a2- 1 no explicit formula for E(a) was found and it seemed
worthwhile to provide such a formula. In wlhat follows we shall. prove that
n n
+00
+00+?? 1XJo(I ak I t)-]fJ(I ak IV$2+4,r2xk212)
(3) E(a)= ddfJ cosat k=1 2 d .
-00 -00
2. Let J1be an arbitrary positive number and put
F=F(t;P)==F(t;41* *.,(,,)-=--a+?akos27r(k7ct+k) (O_ k <1)
k=I
F'= F/81.
Denote by Ga(0Zy
,,,) the inumber of t's in (0,Ju) for which F = 0.
Then by section 2 of loc. cit.' if G0(4o1< ,n4) is Riemann integrable the
limit (2) exists and
(4) E(a) = . . Ga(cki qon)do,t12
We first prove that Ga is Riemann integrable.
* Received July 16, 1942.
' American Journal of Mathematics, vol. 61 (1939), pp. 985-991, in particular,
sections 2 and 3. The (Xt + 0 )'s in formula (12) should be replaced by 2ir (Xt + 5) 's.
2 The righlt side of (4) does not depend oni ,. In loc. cit.' we put A =I/Xi
(assuming X, > 0) in which case the integral (4) can be reduced to an (n-1)-
dimensional one.
609
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
610 M. KAC.
I. Ga(4tn On)- Ga(P) (O ? pjc< 1) is bounded. If Ga(PP) were
not bounded we could find such a sequence {Pn} of points in the n-dimensional
cube that G(P,) -> oo as n - co. {Pn} obviously contains a conivergentsub-
sequence which, without loss of generality, may be assumed to be the sequence
itself. Let PO be the limit of {Pn}. The functions F(z; Pn) and F(z; Po)
(as functions of the complex variable z) are analytic and F(z; Pn) eonverges
uniformly to F(z; PO) in every bounded region of the complex z-plane.
Since F(z; PO) is not identically zero it is a well-known theorem that
the number of complex roots of F(z; Pn) in every fixed region approaches the
number of complex roots of F(z; Po) in the same region. This will contra-
clict our assumption that G(z(P.) -> co, if we fix the region in such a way as
to colmtainthe interval (0, ,u).
II. The set of points of discontinuity of Ga(P) has n-dimensional
Lebesgute measure 0. It is easily seen that if P' is such that F(t; P1)
(O ? t ? y) has single roots only, Ga(P) is continuous at P - P. Hence,
the set of points of discontinuity of G,a(P) is the set H of those P's for whicl
there exists at least one t (O ? t uu) such that simultaneously
F(t P) O and F'(t;P) -O.
Let E > 0; iotice that
nn
! (; P)
!F'(t; P) I -27r
=2_
_ I zakXk sin 27r(Ak7t+
nadkil7(k?~) Ok) |, C5 E I akkk
? 92 it,aXI ~I
- t
F"(t; P) |- akXk2 COS2ir(At + 0k) 472 | C aJ | M2
h-l k=1
and put M Max (Ml1,112)
= Put furthermore-tj jE=/lI (j-O, 0, l * s),
where s [1 + (J1p,/E)] anidobserve that 0 =- to < t1 < t2 <
= < t, ? U.
Let now Ei be the set of those P's for which
F(tj; P) ! < E and lF'(tj; P) I < E,
and let E Ej.
j=O
I assert that II C E. In fact, let P0 E11. Then there exists a T (O ? r ? )
such that F(T;PO) =F'(T;PO) =0 and we caln find such a j that tj < T
? tj+l (ts+l ) f. [Hence,
=
F (t ; Po) | F (tj; PO) - F (T; PO) ?1l I T--tj I < E
and
F'(tj;P)!=P F'(tj; Po) -F'(T; PO) |?c 2I T - tjI < E.
This means that Poc Ej and therefore POE E. Thus H C E and
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
THE DISTRIBUTION OF VALUES OF TRIGONOMETRIC SIJMS. 611
meas. H ? E meas. Ej.
j=1
In order to find meas. Ej let us recall that
400
o 00
1 r (J sine sinH +)q {
2 ,~ exp({i(y +&q)ded-q
-00 -00
is equal to 1 if both !y ! and 8 are less than E, 0 if either y or 8 is
greater than E, 1/2 if either I 8 < E or vyi< ,,1 and
1/4 if Y= I8 . This leads to the formula
+00 +00
m Ei -
meas. ffsin sinEqi
-00 -00
X J *** J exp {i (eF(tj;P + -qF'(tj; P) 3,d) ded;d
dq)n]
if one notices that the measure of the set of those P's for which F(tj; P) = c
and the measure of the set of P's for which F'(t; P) =- E are equal to O.'
Now, we note that -forevery t
fz>exp{i(sF?.+ *d.
F') }dPi*
k= 0
(5)
=-exp (--iat ) Ill exp{1(a k COS27r(Xkt+ Pik) ? sin27r(Xkt + 4>k)) } d4>
2l7rakXk?)
n
exp(- iae) iI J(I ak V2 +47r2XA2i2),
k=1
and hence
+00 +00
J fJ"sin
m1as. { c sin E
meas. Ej 7r2=
n
X exp( i4) H1Jo( ak I 2/2 + 47r2Xk y2
)2dedr1.
k=1
3The interchange of the order of integration can be justified ill a way exactly
analogous to that followed by Kac and Steilnhaus in " Sur les fonctionis ind6pendantes
III," Studia Mathematica, vol. 6 (1936), pp. p3-94.
4This statemelnt can be easily proved. It suffices, however, for our purposes to
recall that every measurable funftion - (P) has the property that for "inearly all"
C's (i. e. except perhaps a denumerable set) the measure of the set of P's for which
< (P) =e is equal to zero. In viewr of this fact we may restrict ourselves to e's lot
belonging to the exceptional set tand save ourselves the trouble of proving that in our
particular case the exceptional set is enipty.
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
612 AI. KAC.
JO clenotes here, as usual, the Bessel function of order 0. Sinice Jo (I u )
(IIt 12-1/2)as Iit oowe see that for n ? 3
K(ep,) =-exp (- iae) HJ,(+ a
k=l
V$2
? 4w2Ae2B2)=?I e3/41 D K311)
so that K(d, ) EL3/'.
E -ence, by applying the Hl61der-Rieszinequality we have
[Link] < 1 (4 S $ C7I da )
-oo) -00
+00 +W +cO +o0
x IfJ Ks(ep q) 13/2 dedrq2/ A(J2
e
in-ll ded)q
-O) -0 -0on_ 00
BE4/3.
Finallv,
meas. H ?< mieas.E ? mneas.E= BsE413 - B[1 + (Ml/lE)]E4/3 (,(E1/,)
j=1
ucidsince E can be mlladearbitrarily small we have
mneas.H 0.
The above reasoning b)reaksdown for m 2. But in this case the proof that
meas. H 0 follows from quite elementary geometric considerations. I andl
=
11 imply, in view of a well-known theorem, the Riemann integrability of G".
3. We shall base our future considerations on the following formula
+00
(6) Ga(01c ,' = f [JF C
cos(F F!! dt] d.
==ir-00
To prove (6) let the a's, 0 < a,<a, < . . .< 1' <i, be the turninig poilnts
of F so that
(7) cos($E) F' I d ? ([sin 8j+Pp-sinfljf/),
j=o
where fo =F(O), i = F(/1), p --F(aj) (i? j? r') and a + or-sign
is assigned according as Bj+l > /j or 13i+l < 13j. By a well-known formula
of Dirichlet
+00
[ ,J
{ sin /j+e -sin kij de
-00
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
THE DISTRIBITTIONOF VALUES OF TRIGONOMETRICSUMS. 613
is equal to 2, 1 or 0 according as /j,8f%+j
< 0, Thj,/j3= 0 or /3jSi., > 0.5 The
signs in (6) are such that
+00
(8) 27r 4
[Jfos($E) F' d dg
-00
+00
1r 1 ('sin ,Sje- Pje
sin/$
2 =O 7r
-00
anid the right side is easily seen to be equal to the number of zeros of F in
(0, pu). This proves (6). Let us notice thai; r (the number of turning points
of F in (0, f)) although depending on 01, * , (, remailns boundled. In
fact r is the number of zeros of F' in (O, t) and the reasoning given in 2
I is applicable to this case. Thus there exists a number C such that
(9) < C.
4.6 Let
Ga ?l J(P)= U0 cos($f) F' dtj de.
It is well knownithat for every m and b
S < 1
27rJ
dett
anld hence by (7) and (9) we have
Win})(P) < C.
Now, for every P, G(16(n)
(p) Ga(P) as m - oo and since the Ga(mt)(P) are
->
uniformly bounded we have bv a well-known theorem that
(10) lim ... G(m)(P)d p dq)n ..f J Ga(P)dht d4n.
On the other hand
4'.. G'G'(1t)(P)do,.
dp1 .
__
1 J:rn4' . f. [f cos(s-) F!I dt] do,
. dqnde
5 It is obvious that 6j i pj_l and therefore f3j and j+, cannot be simultaneously 0.
If 0 or ALis a root of F = 0 it is countecl as 1 in formula (6). Since the set of those
P's for which either 0 or A is a root, is of measure zero the followinig considerations
are not affected by this peculiarity of formula (6).
6 This section employs exactly the same reasoninig as the one in loc. cit.3
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
6t14 M. KAC.
W-hichtogether with (10) implies that
La(e) . f [f cos(eF) F' dt] d .
do#n
is integrable in (- oo, oo) and that
+00
.1 ~ ~
(~~~ . Ga(P) d(p .. do?t- ~~~~~~~~~~~~2
? .O - .. La()de.
7r
-00
5. It remains now to compute La(a). It is well known that for every
real y,
+00
yIyl= ! flo -OydqCsy-q 2
-00
and therefore
+00
-
! I lf C-cos (F')d
-00
Hlence
+00
D
La() 1 _(_) d_
-00
wThere
*D(t, ) = j co (IF)
[cos cos(qF)cos(71')] do dO,ndt
and the interchange of the order of iintegration is justified by the fact that
the onily infinite integral occulrring in the formula is absolutely conivergent.
Now,
cos ($F) cos(QqF') =
exp {i (eF + F') } + -1exp {i ($F -') }
I F + yE')) + j1-exp {i(-
+ exp {i(- eF-Y')}
and it follows (see (5)) that
a1 5
101 . I cos (F) cos (yF') do, - don
11 Jo(!k
=cos a k=l V2 +47r2Xkk22)-
Putting -= 0 we get
, . . . . .c d6b,
cos (eF) dP *. cos JJo(1 ak I e).
O L #_~~~~~~~~~~~k1
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
THE DISTRIBUTION OF VALUES OF TRIGONOMETRIC SUMS. 615
Thus
-cc nn
La(t)IL cosat,f 4 KJlo(J k ) H Jo(I ak
|V 44
e2+ r
2Ak2q d
7r +CN q
and finally by using (4) and (11)
n n
+00+00
rnJO HJ(ae-IJ(aIV2
(| a7cI ge) IIJ o( ak ! +47w2Xii2)2
+4T &
1
E(a) = -2 f cosae k=1 _=_ drde.
-00 -00
6. Since Jo( ) ( u -1/2) as I1 ->oo we see that for n ? 5
the integrand of (12) is absolutely integrable. It may also be noted that if
a1 = a2 *an I a simple asvmptotic formula can be obtained if the
=
A's satisfy certain conditions. In fact, we have in this case
+00 +Co In
rr ~ ~Jon( 1 JO( -\/.62 + 42AkV
,2X
E(a) Z2' 2 daqd4
cosad ) k \,,
27r2 _00 _0 7q-
8nd letting it
u/\n, = v/27r/n, a =bV/n we obtain
+oo +0 _
E (bV-/n)
\/- | +00
f
SX
cos byv
Jon(u/Vn) IIJo(Vu2 + 21
~i2n
/n)
=-
dvdd.
-00 -00 V
Assuming now that
n n
lim n1 zX k2 Ax lim n
r -2 Ak4 ?
,n-> o k=1 n-->C1 k=l1
we have
Jon (u/ n) -) exp (- u2/2)
17 Jo ( Vu2 - Ak2v2/n)
>, exp (- '2/2) exp (- Av2/2)
k=_1
anid the convergence is uniform in every bounded region of the (u, v) plane.
Thus, since all the integrals involved are absolutely convergent,
+00 +00
1 C_CV1
b-\
E(bVn) n cosbu exp (-u2/2) exp (-Av9/2) dvdu.
-00-0v
The last integral can be easily evaluated and we obtain the formula
2
E(bV/n) - - V/Aexp (-b 2/2).
7r
CORNELLUNIVERSITY,
ITHACA,N. Y.
This content downloaded from [Link] on Sun, 7 Dec 2014 [Link] PM
All use subject to JSTOR Terms and Conditions
