Académique Documents
Professionnel Documents
Culture Documents
Problems in Analysis
by Bernard Gelbaum
1982. vii, 228 pages. 9 illus.
A Problem Seminar
by Donald J. Newman
1982. viii, 113 pages.
Exercises in Integration
by Claude George
1984. x. 550 pages. 6 illus.
Problems in Geometry
by Marcel Berger et al.
1984. viii. 266 pages. 244 illus.
Claude George
Exercises in
Integration
With 6 Illustrations
Springer-Verlag
New York Berlin Heidelberg Tokyo
Claude George Translator
University de Nancy I J.M. Cole
UER Sciences Mathematiques 17 St. Mary's Mount
Boite Postale 239 Leybum, North Yorkshire DL8 5JB
54506 Vandoeuvre les Nancy Cedex U.K.
France
Editor
Paul R. Halmos
Department of Mathematics
Indiana University
Bloomington, IN 47405
U.S.A.
987654321
ISBN 0-387-96060-0 Springer-Verlag New York Berlin Heidelberg Tokyo
ISBN 3-540-96060-0 Springer-Verlag Berlin Heidelberg New York Tokyo
Introduction
V
vi INTRODUCTION
CHAPTER 6: THE LP SPACES ... ... ... ... ... ... 225
(Exercises 6.100 - 6.125)
CHAPTER 7: THE SPACE L2. ... ... ... ... ... ... 285
(Exercises 7.126 - 7.137)
ix
x TABLE OF CONTENTS
CHAPTER 11: TRIGONOMETRIC SERIES ... ... ... ... ... 451
(Exercises 11.185 - 11.230)
BIBLIOGRAPHY ... ... ... ... ... ... ... ... 547
NAME INDEX.. ... ... ... ... ... ... ... ... 549
CHAPTER 0
Outline of the Course
From this it follows that the set X itself belongs to the a-alge-
bra A, and that the a-algebra A is closed under countable intersec-
tion. For two sets A,B e A let us denote A - B = {x:x a A,x $ B};
then we have (A - B)e A.
The smallest a-algebra containing the open sets of ]R is the
a-algebra of BOREL SETS of ]R ; this a-algebra is also the small-
est a-algebra which contains the closed (resp. open) rectangles
of]R .
1
2 CHAPTER 0: OUTLINE
There exists one and only one positive measure v on the a-alge-
bra of Borel sets of ]R such that, for every rectangle P, its
measure v(P) is equal to the volume of P.
If f and g are two mappings from z into zzq (or m) such that
f is measurable and f = g almost everywhere, then g is measurable.
4 CHAPTER 0: OUTLINE
q,(x)dx = 9 _ aimeas(Ai),
I
n i=1
OF THE COURSE 5
J f(x)dx = if = limf . .
This element of [0,+-]=]K +, which does not depend upon the sequence
(Ti) selected, is called the (LEBESGUE) INTEGRAL of f on7Rn.
Jlimf
n
= limif-'
n n
6 CHAPTER 0: OUTLINE
then
Jf , A.
fEf(x)dx=JE =fiv,
so that
u = U+ - u_,
lu l = u+ + u-,
u+u- = 0.
11fl <
Jf =
Ju+ - Ju + iJv - iJv,
(which has a meaning, because u+,u_,v+,v_ are majorised by lfl).
ff1 <J1f11
Jf= lim
n Jfn.
I JUj
<
n
u(x) = I un(x)
n
Ju =
fUn
n
OF THE COURSE 9
JE = JLEf
fE
f = n
L JE f.
n
J f = l nimfE f.
E
n
we write
M
J_
f
M}oj0
f
when this limit exists. In this respect let us note the follow-
ing Proposition:
JJ1If(x,y)ldy <
Spherical Coordinates in ]R
x1 = rsinen_2 sin82sino1coscp,
x2 = rsinOn_2 sinO2sin81sin9,
x4 = rsin4n_2 coso2,
xn = rcos8n_2,
OF THE COURSE 13
where r > 0, 0 < of < it, ITI < R. The formula for the change of
variables in this case comes down to replacing the xi by their
expression as a function of r, the oils, and cp, and dx = dx1dx2...
dxn by
rn-lsinn-ton-2
... sin2o2sinoldrdSn-2...de2doldq.
2
Let us also recall that the volume of the ball x2 + +X <, 1
n
is
n/2
v_ n
n r 2 + 1
1
p1
+q =1,
[JfPJ [JgPJ'/P.
If (f + g)PJ 1/p < 1/p +
DEFINITION: With the preceding notations, for 0 < p << - one de-
fines the following functional spaces:
G iiunilp<
n
u(x) = I u(x)
n
= Jf0 2(fig)
f,g e L.
16 CHAPTER 0: OUTLINE
0.7 CONVOLUTION
We have:
f*g = g*f.
[JKIf1i'/p
HAIp,k = <
_-
IIfIIp,K = IIfIIp,N,
n
and the formula
18 CHAPTER 0: OUTLINE
W
IIf - gIIP,N
d(f,g) = 2- 1 +
1
N=1 If - gIIp,N
Isl = s1 + .. + sn,
and we define
ISI
Ds f=
D
,f feEk, IsI << k.
axs i...axnn '
1
and ifk= -,
pN(f) = sup{IDsf(x)I:Isj < N,jxj 4< N}.
W -N pN(f - g)
d(f,g) = 2- 1 + f -
N11 pN g
IIfL = suplf(x)I
It is a Banach space.
If E and F are two of the spaces that have just been defined
we shall write E - F if E C F and if fi - f in E implies that
fi -> f in F. It is clear that E - F and F -> G implies E -* G.
We have the following diagram, where 0 < k < - and 1 < p <
Fk
E' --a -p C - + L
loc -' LPloc - Lbe
T
t
UCW LW Lp L1
D"->Dk--* K ->LWc c c
It will be noted that D -> E for any space E from the table
above, and also E -> Lloc for every E; note also that the Lp spaces
are not mutually comparable.
IIffgII <
(ii): Lp*Lq C C0 (continiously) if 1 < p,q < -, p + q = 1; fur-
thermore,
IIf*gL s IIfIIpIIgIIq;
1 oo),Cm,UC
(iv): L operates continuously in Lp (1 4 p , and CO;
DS(f*g) = ff:Dsg,
Furthermore :
4
(g,h) = Jgh = (g*)(O)
If 11p p
= [ J2nIfIP)1/P
0
2n
(f*g)(x) = 2nJ f(x - y)g(y)dy.
0
as do the inequalities:
21 2n
(f,9) = 2nJ f9, (.fig) = f1J f9
0 0
one obtains the same formulae as above. Let us note that in the
formulae defining f*g,(f,g) and (fig), one can replace the range
of integration (0,2n) by any interval of length 2n.
Clearly one can consider functions having an arbitrary period
T > 0, it is then just a matter of replacing it by T/2 everywhere.
(i): There exists a constant M such that 11,Pi11l < M for all i;
(i):II(piII1,M;
(
(ii) : lim 2l -71 i = 1;
tm
E f(y) e-2%ixy f(x)dx,
=- J-W
y e3R.
Upon setting
26 CHAPTER 0: OUTLINE
e2itiax
e (x) =
a a eat, x eat.
f'(y) = 2uiyf(y).
f' (y).
g(y) = 2ni
M
e-2nixy f(x)dx,
f (Y)
JM
M 2nixy-f(y)db,
f(x) = lim e
M}-m-M
F(fg) = F f*Fg
L2.
6( )#O
28 CHAPTER 0: OUTLINE
n-1
V(f;a,b) = sup E If(xi+1) - f(xi)I <
A i=0
b
If'(x)Idx V(f;a,b).
a
b
f(b) - f(a) = J.f'(x)dx,
a
b
V(f;a,b) = J If'(x)Idx.
a
f(x) = JF(t)dt,
x a < x < b,
a
30 CHAPTER 0: OUTLINE
+W W N
- un = u0 + I
n=1
(un + u_n) = lim
N- n=-N
E un,
when this last limit exists. When luni < - one also has
I00
M
u
n
=lim
N-
E
n=-N
u.n
M-
tm a m
c e
n
inx
= 2 + (ancosnx + bnsinnx),
n=1
an = cn + c-n, C
n = Y(a n
- ib ),
n
n > 0.
b
n
= i(c n - c -n ), c-n = '(an + ibn),
(2n e-inxf(x)dx.
?(n) = I
0
inx
f(x) ti L oneinx
-W
12n r
1
2n
an f(x)cosnxdx, bn = f(x)sinnxdx.
n n
0
.'(n) _ (f l en),
fsaen = f(n)en,
f*g(n) = f(n)g(n),
ina
in?(n) = [f'](n) + 2n {f(as + o) - f(as - o)}e S.
s=1
(-1)n+1 sinnx
f(x) ,, 2
n
n=1
OF THE COURSE 33
NOTATION: We set:
N
SN(f;x) = I f(n)einx (FOURIER SUMS),
n=-N
then:
with
34 CHAPTER 0: OUTLINE
d 1
sin(Nt+ zt
(X(t) dt + r, (X,6).,
IO
with
1r2n
1 f(x)g(x)dx = f(n)g(-n)
2n
0 n=-W
.(n)(2neinxg(x)dx.
= E
n=-0 0
THEOREM: (Fejer-Lebesgue):
E If(n)I2 <
27E if(x)I2dx
2n0 = L I?(n)I2.
NOTE
meas(E.
P Z1< < p 11 1P
n (-1)p-1aP
meas(E1 U U En) = 1 (POINCARE'S FORMULA).
p=1
n ( l
meas(G) = I (-1)P-5 l s la .
p=s l 11 p
37
38 CHAPTER 1:
n
(_1)p-s(s-11
meas(HS) = I Qp.
P=S l
EA = n Ei, EA = EA - U Ei.
ieA i+A
E = U EB',
A
BDA
I
(_1)CardAmeas(EA).
(-1)po =
P CardA=p
n
(-1)pa = (-1)CardA meas(EA) _ (-1)CardA L meas(EB)
p=1 p A A BD A
I meas(EB) (_1)CardA.
B ACB
x
(_1)CardA =
(_1) r(P 1 = - 1.
ACB r=1 l J
n
(-1)Pa = - meas(EB) meas(E1 U UEn.
p=1 p B
MEASURABLE SETS 39
n ( l
(-1)p 8Ja = (-1)CardA(CarsdA lmeas(EA)
p I
p=s l CardA3s J
(-1)CardA(CardA 1 I meas(E')
Card4 s l s J BDA B
meas(EB) (-1)CardA(CardA)
E
CardB;s ACB ll J
CardA=s
=
l s ) I ( -1)r (r - s )
P-8
(-1)s -
(-1)rlprs1
s)
0 ifp> s,
(-1)S if p = a.
n
(-l)PI p)Q = (-1)S meas(E') = (-1)smeas(G ).
s s s
p=s , CardB=s
HSt1 = HS - GS,
40 CHAPTER 1:
n (-i)P-s-1 j(
meas(Hs+l) = I l -1
p=s lsJ [P-111ap
n
(-1)P-s-1(p $ 1 l
op.
p=s+1 l J
e2-n
meas(U) S 1 = e.
n=1
E meas(E <
n
A n [UEJ
p
and consequently
(pan
lim I meas(E ) = 0.
n pan P
G meas(En) <
n
meas(Hs) 4 G meas(En).
n
s
42 CHAPTER 1:
smeas(Hs) = E smeas(EA)
CardA>,s
n
CardAmeas(E') _ meas(E.).
A
CardA>,1 i=1 1
In the general case one considers, for n > s, the sets Hs n formed
by the points which belong to at least s of the sets E1,.. -,n-
It is clear that Hs and that
n C Hs n+l
Hs = U Hs,n.
n=s
Consequently
n
<< Jim meas(E.) = s meas(E.).
n i=1 i=1
s
n l
n E1l
D (E1,...,En) = Eil - I F] ,
J
meas(D(E1,...,En)
a(E1,...,En) =
meas
a(E1,...,E1) E 1 1
n 1
i<j
meas(D(E1,...,En) n 1 L meas(D(Ei,Ej)).
i<<j
Since
1
C meas(D(E.,E
a(E1,"'' n n - 1 i<J meas E1U UEn
4 n 1 E
1
i<j
44 CHAPTER 1:
mead( U E n) <
n
and
SOLUTION: We have
A=ni(UEsl
n=1 s=n
m
mead U ES] . meas(En) > a.
s=n
EXERCISE 1.7: Let A be the set of real numbers x for which there
MEASURABLE SETS 45
x P
q
SOLUTION: Since
x + n -ng =x -q ,
by setting
B = An [0,1]
A = U (B + n).
n=-w
Ip,q = LP-
q
- 1q3 P- + 1J
' q q3 '
Bq = [0,1[n (U Ip,q).
-2. p. q + 21
1
q q
0<p q.
Consequently
BP
U>p A P,q
Everything reduces to proving that the B , which form a de-
P
creasing sequence, have a non-empty intersection. The union of
the radii of the disc with center 0 and radius q that meet A
P,q
covers (v(q) - v(p)) discs of unit diameter and mutually disjoint.
From this it follows that if meas(A ) denotes the Lebesgue mea-
P,q
sure (evaluate in radians) of A the unit circle, one has
p ,q
,q
2 it
iq meas(AP,q)
(v(q) - v(p)).
v(q)
2 liminf
q q2 v(p) 2
From this it follows that
n + E at-p,
p
p=1
where n ea and ap = 0 or 1.
(b): From this deduce that there exist two negligeable sets,
the sum of which is fit.
A = fg, (B + n).
n=-w
ap(x) = u(2p-1 x)
has measure 2-P. From this it follows that the set Bp of the x's
such that 0 4 X < 1 and
MEASURABLE SETS 49
Y a 2-(2p+1) a p = 0 or 1.
p=0 p
n+ ap2p, ap=0or1,
p
+co
F= (E t n),
n=-co
as well as
50 CHAPTER 1:
- f(x)
if xeF.
It is clear that g has all the properties desired.
Prove that
1 meas(A)
liminfndP
n n < ap Qp
where
(
1 p
cp = 2 -P{ 1 + p10 ( 1 + t
l dt}
For this one will consider a number y such that ndP > y for n
n
sufficiently large, as well as the balls B. with center xi and
radius ri, where
yl/pn-1/P.
ri t rj =
- 1/p,
IIxi - xi II > dj > yl/Pj
ri + r. yl/Pn- 1/P
+ zyl/P(2j-1/p - n-1/P) yl/Pj -1/p.
=
Pn
urn - 1 + 2C (2i-1/P
- n 1/P)Pl < meas(A(e )),
2 LL n J n
i==n
where
e = 2yl/Pn - 1/p
n
We have
2Pn
1 Pn [2rnl -1/p - 1]p =
2C
n-1/p)P
lnm (2i-'/P - = lim n
i=n i==n Il J
(Contd)
52 CHAPTER 1:
r2P
(Contd) (2,-l /P - 1)Pdx
JI
(1 - t)P
= p1
1tt dt,
meas(A) 1
St c
p P
1 meas(A)
liminfndp 4
n n c SZ
P P
1 f n(x)dx=1.
E
GO = E,
n u_P(E),
Gn = E - U n . 1,
p=1
En=Gn-1Gn, 14n<W,
E = o GG.
n-1
un(Gn) = un(E) - U up(E).
P=O
Since u(y) C -y and meas(y) > meas(E) > 0, the second hypothesis
54 CHAPTER 1:
Go m
1 = E measun(G ) = E meas(G.).
n=O n n=0 n
N
n(x)dx = nmeas(E ) = lim X n{meas(G meas(G )}
n n-1 n
E n=1 N n=1
(N-1 N l
rN-1 l
implies
Nmeas(GN) -} 0.
Thus
I n(x)dx = 1.
J
E
We shall briefly recall the proof: let c > 0 and let p be such
that
liminfnun s E. QED
n
(c): Deduce from (b) that f is measurable and that there ex-
ists a sequence of continuous functions fn which is convergent at
every point of R and whose limit is almost everywhere equal to f.
SOLUTION: (c): For every x and all c > 0 there exists a > 0 such
that
g(x') . g(x) + e
{w> 0} = U JW >n}
n=l JJJJ
But then:
]x,x+ax[(nA=0.
rx < y < r y ,
which proves that x e A - rx is an injection of A into Q, and con
sequently that A is denumerable.
and consequently
MEASURABLE SETS 59
Ar)]-.,x[ = U {Afl]-W,xn]},
n
EXERCISE 1.16: Let 0 < A < 1. For any measurable sets A,B C[0,1]
of positive measure, do there exist 0 < x < y < 1 such that
meas(Ar)[x,y]) = Ameas(A),
meas(Br)[x,y]) = Ameas(B)?
f(ts) = meas(A).
n
60 CHAPTER 1:
E f(ts+1).
n
MEASURABLE SETS 61
(n+1)a+ns= 1,
and let us decompose [0,1] into 2n + 1 contiguous intervals, al-
ternatively of length a and B, the two extreme intervals having
length a. Let A be the union of intervals of length a, and B
that of intervals of length S. If
But then,
K R In
n
al/n(n+l),
(d): If Bn = 1 - 0 < a < 1, show that meas(X) = a
(which proves the existence in [0,1] of nowhere dense compact sets
whose measure is arbitrarily close to 1).
MEASURABLE SETS 63
A U An
n=1
(f) : Let
E
nU-0 A2nt1'
(g): From part (f) above deduce the existence of Borel sets
E C IR such that for every interval I of non-zero length one has
0 < meas(E(lI) < meas(I). Can one have meas(E) < m for such sets?
( ln
meas(K) = liml3J = 0.
n
x = O.ala2...an...'
a1/n(n+l),
a
meas(K) = lima n,
n
with
1 _ 1
(n n+1
whence meas(K) = a.
n 2-i
meas(Kn) _ = 1 - 2-n,
i=1
whence:
A U An
n=1
W
meas(A) = 1 2-n = 1.
n=1
E A2n+1' F i A2n'
n
l
If I is an interval of length greater than zero-contained in
[0,1], its intersection with E or F is of measure greater than
zero, since meas(E U F) = 1. Let us assume, for example, that
meas(If1E) > 0. Then ICE contains at least two points x < y.
Let n be such that these two points belong to Al U A2 U U A2n+1'
MEASURABLE SETS 67
Since
E= U (E + n)
nea
EN N
n
A2n+1'
If
E = U (EINI + N),
ne2Z
and furthermore,
W m m
2-(2n+1) + 2 2-(2n+1) = 10
meas(E) _
n=0 N=1 n=N 9
68 CHAPTER 1:
h - g
ms
<1s - fmsons
gmS
Thus
MEASURABLE SETS 69
Ih- fm
,n I < on E,
s s s
which proves that fm n - h on E.
' s
We are going to show that for all e > 0 there exists such a
set E, which, moreover, is measurable and satisfies meas(X - E)
< S. In fact, by Egoroff's Theorem there exist measurable sets
EO,E12... such that
On setting
E=nm
m=O
h on E2.
f1Plo(P 2(n),91082(n) -
E = U r,
r
that
-> h on E.
fm in
s s
t - n Fn(t)
n
is also measurable.
1V = VV =Ova=V V =AVA
SOLUTION: (a): Let Bn be the closed ball with center 0 and radius
n. We have:
Co
n
{t: ( n F (t))nK + y} = n {t:(n Fr(t))nK + 0},
n=1 n=1 r=1
n
n (Fr (t)nK)
r=1
(c): From this deduce that there exists a mapping 9:F ->iRp
P
such that:
{t:g0(F(t))nK # 0}
is equal to
(a): Show that there exists a set E C [0,1] such that for
every x eR there exists an unique y e E such that x - y is rational.
[0,1] C S C [-1,2],
and that if r,s are two distinct rational numbers, then E + r and
MEASURABLE SETS 77
E + s are disjoint.
z =y1+r=y2+s,
and consequently y1 4 y2, which contradicts z being equivalent to
a single element of E.
V(E) = I m..
x.eE
i
(b): Show that if {xi}e C for all i then one has C,1 = P(X),
and conversely.
N
11 (E1U...UEN) = 1 ir(En) (E),
n=1
79
80 CHAPTER 2: a-ALGEBRAS
whence
E u(En) u(E).
n=1
E m = G I m. 4 1 u(En
X eE 1 n=1 x.eE 1 n=1
i i n
ieA ieA
whence
M
u(E) = sup X m. < G U(E ) .< u(E).
n
A x ieE 1 n=1
ieA
M C X, set:
CM = (Mf)E;E a C}.
PM(MnE) = u(E)
(X - F)U(EfF) > M,
whence
1-u(E)+u(EnF)=1,
1 - u(F) + u(E n F) = 1,
Moreover, if E e C, F e C, and
one has:
u(EnF) = 0,
whence
= 1M(MfE) + uM(Mf)F).
Mfl En C Mf1En+1
= M(U (MnFn)).
n
u(nO = n1 .
SOLUTION: Since {0} C n N for all n , 1 one would have 1j({0})4 1/n,
and consequently p({0}) = 0. Furthermore, since 0 is the unique
integer that is divisible by an infinite number of prime numbers,
one would have:
{o} {piNJ
r=1 i=r
1
u(p. Nn ... np. i) = pi , ..pi
11 la
1 a
s l
u(U pi1VJ p1a
i=r )) r4i1<...<1 8 pil
-
s 1
=1- ir (i -
Now, it is known (see the end of this solution) that
s
lim 1 1 = 0,
s i=1 pi
and therefore that for r >, 1
s
lim IT 1 - l = 0.
s- i=r pi
Therefore for all r > 1 one would have
ll (( ll
i1VJ = o[ piJ = 1,
ti-r s->- i=r
s
iT 1
1
=IT{E k,
i=1 1 - i=1 n=0 pi keAs
Pi
AND POSITIVE MEASURES 87
whence:
s
1im 1 1 = lim I _ I k= .
s i=1 1 - pi
s- keA S k=1
EXERCISE 2.28: Show that there does not exist a a-algebra having
a countably infinite number of elements.
{Ex}xeX - {EiIieI'
EA = U B. e C,
ieA
(1
xnlogxdx
0
1
x dx,
10
1
given that
1
12
1 6
n=1 n2
89
90 CHAPTER 2: THE
(Contd) _- 1
2J te-tdt = - 1 2
(n + 1) 0 (n + 1)
OD
logx I - xnlogx.
1-x _
n=0
l logx dx = C - 1 = - n2
J O 1 x 1)2 6
n==O (n +
-s x-1
s
n =rs x dx.
n=1 J 0e -1
Ix se-axl = xRe(s)e-ax
xs-1 xs-1
Since
C I xRe(s)-le-nxdx
n=1 J0 n=1 0
n-Re(s) < w
F(Re(s)) 00
E
n=1
2
and from this deduce that for all z e c the function t e-t coszt
is integrable on [0,Co], and calculate its integral.
t2ne-t2dt
= otn-e-tdt = r(n + ) _ (Contd)
0
J'o Jo
92 CHAPTER 3: THE
2 2n 2
t2ne-t ,
e t coszt = (-1)n (2n !
n=0
and
2n 2 Cw I 12n
t2ne-t
G J (-1)n dt = n G 2n+1
(2n)!
n-0 0 n=O 2 n!
Thus,
T 2 2n °°
!t2ne-t
2
e-t cosztdt n (2n dt
n=O 0
2 n
(z l
n4 '
= 2 E (-1)n
n0
-z2/4
e
2
snax a
dx
J Oex-1 n=1n+a 2 2
SOLUTION: If x > 0,
Co
s in= e I's inax .
= E (*)
ex - 1 n=1
Furthermore, by using the inequality Isinul < lul, one has for
n 1
rix
J le nxsinaxldx < aJ xe- dx = a
0 0 n2
so that:
a
Joe-nxsinaxdx =
0 0 n2 + a 2
c (-1) n+1
n
n=1
n+1 W 1 1 °C° 1
( -1)
nl n n 0 [T -+1 2n+ 2, =n 0 2n+1 2n+2
Furthermore, for 0 4 x < 1
1 1+x x2n+1
21og1-x
and consequently
1 +x dx = c 1
21J1l0
0 gl-x n=0 2n+1)(2n 7-2
n+1
(-1) = log2.
n
n=1
s i n a x
dx = (1 - e-a).
f 0x(x + 1) 2
cosax I 1
x2 +1 x +1
the function (p is continuously differentiable on Ft and
xsinax 2
JAx2+1 Aa0
1)-1
(by the second mean-value formula, since x(x2 + is decreas-
ing for x z 1). This proves the uniform convergence of the im-
proper integral
xsinax
dx
1 0x2+1
on [a0,=[. Consequently 9 is twice continuously differentiable
for a > 0, and
f0mxsinax
fW xsinax rW sinax dx +
dx dx
Ox2+1 0
x (x2+1)
9(0) = 0, 9'(0) = 2
96 CHAPTER 3: THE
(1-e-a).
cp(a)=2
2e-2nx
fn(x) = e- T'x -
Show that the series with the general term fn(x) is convergent
for all x > 0, and calculate its sum f(x). Next, show that each
fn, as well as f, is integrable on (0,co), and compare
ff(x)dx and
J:ffl(X)d.X.
n1
-x -2x
f (x)= e
-2 e
L
n=1 n 1 - e -x 1- e-2x
1 _ 2
ex-1 e2x-1
1
ex + 1
10fn(x)dx = - 2 2n = 0,
n
so that:
0.
n=1 JJ 0
However,
Jm x
dx ex
= [log = log2.
Oex +1 e
x
+1x=0
J
00,
n=1 0
Jm 1
olfn(x)Idx=nJol1-2uldu=n
n x
2
if 04x4n
fn(x) =
- n21x - if n s< x s n,
n
0 ifx>' 2n '
Calculate the four numbers:
(b): The same question for the sequence (gn)n>0 defined by:
Jfn = 1,
(The inequality A' 4 A is Fatou's Lemma, the two others are clear)
FUNDAMENTAL THEOREMS 99
9- 4f n for all n,
jg_ < .
Jg_
Jg = -
which is an element of and that
g < h, fg-
< - imply Jh_ < - and Jg < Jh).
jg+ < W,
then
and
lim lf(x)ldx = 0.
zI I- J K+z
lim f(x) = 0.
IIxII-)W
SOLUTION: (a):
liml If(x)Idx = 0,
r-*
since
0 and
fr Ifrl IfI
If(xn)I >. e
s
If(y) - f(x)I < if y e B + x.
2
P
J IfIP > (2) meas(B) > 0,
B+x l
n
j mG(f(x))dx < -,
then
1im f(x) = 0.
IIxlIH
IlxnII - -, If(xn)I z E.
24IfISM
on the balls B + xn. By virtue of the assumptions made about G,
one has
JG(f(x))dx 3 umeas(B),
B+x
n
Show that
2nmeas(2n-1
I f(x)Idx = 0,
a-J
lim lfl >a
An =
(2n-1
< Ifl 6 2n), B = (f = 0), C=(IfI
These sets are mutually disjoint, with union ]R , and C has meas-
ure zero. Then
J IfI = T J" If
R n=-W A
n
whence
104 CHAPTER 3: THE
(
+m
2 2nmeas(An) 1 1A s I 2nmeas(An),
n=-m IR n=-m.
SOLUTION: Observe that if for all t elR+ and n e2z one sets:
t 2nan(t).
n=-
(Note that an(t) = 0 whenever 2n > t; the an(t) are simply the
digits of the binary expansion of t). In particular,
f(x) _ 2nan(f(x)).
n=-m
Also,
f(x)dx = 0
JX
if and only if
JIi + zf(x)Idx 1
11 + peiof(x)I - 1
dx 3 0,
J X p
106 CHAPTER 3: THE
Moreover,
I1 + pe10f(x)I - 1
limJ dx
p->0 X p
>. 0,
el$Jxf(x)dx = - Jf(x)dxl
X
one obtains
f f(x)dx = 0.
JX
+W
n=- f[.+nI
(*)
I J:If l dx
n=-w C'fxI =a .f (x) I dx
fa
i F(x)dx = J(x)dx.
0
limn af(nx) = 0.
n-*-
108 CHAPTER 3: THE
SOLUTION:
r( t
(x)Idx,
so
in-CL
f(nx) I dx <
n=1 I m
I n-ajf(nx)I
<
n=1
n-af(nx) - 0.
T
A= I f(x)l dx < m
J 0
limn-2f(nx) = 0.
n
limlcosnxll/n = 1.
n
FUNDAMENTAL THEOREMS 109
ppn(x) = 2 f(nx).
n
Then:
T 1T 1 nT
Jk(x)dx = 2J If(nx)Idx = 3f
n 0
If(x)Idx
0 n 0
A
= n2
Hence
T
Ipn(x)Idx <
n 0
(log1cosxl)2.
f(x) _
lim(n 1loglcosnxl)2 = 0,
n
i.e.,
110 CHAPTER 3: THE
limlcosnxll/n = 1
n
1 N
-1
J f(x)dx = 1imN Y f(xn); (*)
0 N-?- n=1
-1 N 2nipxn
limN I e = 0.
N- n=1
xn = na - [na] (a irrational)
and
x = logn - [logn]
n
SOLUTION: We shall prove part (b) first of all, which will also
prove the step (i) => (ii) of part (a). We may assume that f is
real. Let us note that if (xn) is equidistributed then (*) holds
for every characteristic function of an interval [a,b], 0 4a 4b4 1.
This formula therefore also holds, by linearity, for every step
function. Now for every e > 0 there exists a step function 9 such
that:
1
f5 cp, J TSe+
J0 J0
f
From this one deduces that:
N 1
-1 N
f (x n )
-1 f,
l im supN E .< l imN I 9_(X )1 S E + J
N n=1 N n=1 n 0
and consequently
N 1
1 im supN t f (Xn ) < J f.
N- n=1 0
N 1
lim infN 1 f(x ) > J f,
n
N-- n=1 0
N.
Finally, let us show that (iii) => (i). We first show that
112 CHAPTER 3: THE
(iii) => (ii)t. Now, if (iii) holds, then (*) is true for every
trigonometric polynomial. If f is continuous on [0,1], and if
e > 0, there exists such a trigonometric polynomial 9 for which
If - 91 E (the Stone-Weierstrass Theorem). To simplify the
ensuing calculation, set
((1
uN(f) = N-
CN
f(xn) - J f
=1 0
Then
which proves that (ii) is satisfied. Now let 0 < a < b 4 1 and
let be the characteristic function of [a,b]. For every E > 0
there exist two continuous functions gE and fE such that
0< f E
. <g E
< 1,
Then
(
1 N
limN-X 1 N
infN-X
b - a - 2e < Jf =
n=1
f (x ) < him
e n
(xn
E
N- N-)oo n=1
-1 N -1 N
b - a + 2E Jg = limN X g E(x n) > lim supN
n=1 I
n=1
',(x ).
n
N-° N-
N 1 N e2nipxn e21Eipa
= N1 N e2nipan = N le2nipa 1
e2nipa
n=1 n=1 1 -
N 2nipx
1 n
1imN e =0
N n=1
N 1 N e21iplogn = N1 N n2nip.
n=1 n==1
Now
N n
1 N
2nip
= N 1fNt2nipdt CJn_i t2nipd(t
- n + 1),
= N-1
0 n=1
N
n2nip 2nip N jn
N1 C = N
2n p + 2nipN 1 I (t - n + 1)t2nip-ldt.
1
n=1 n=1 n-1
+
When n a 2
In
t-idt
(t - n + 1)t2nip-ldtl = log[;, n 1, .
fn-1 Jn
n-1
114 CHAPTER 3: THE
N2nip rlogN)
N 1 c n2nip
n=1
=
1+2npol{N +
1
J f(x)dx = 0,
0
but
N
N1 E f(xn) = 1 for all n.
n=1
(x) = e2ni(px+gy)
P,q
(+W T r (+0
11 f(x)g(nx + an)dx = {Jgx)dx] Jf(x)dxJ (* )
0
(AJER'S FORMULA)
AVA = VAV - AVA - VAV - AVA
116 CHAPTER 3: THE
B 1 nB+a
Jflg(x)dx.
Jf(x)g(na + an)dx = 1 g(nx + an)dx =
a na+a
a
n
gT
rn(B a)1(
nB+ang(x)dx . (x)dx + en
Jna+an
lL T JJ 0
((+m T
B
f(x)g(nx + an)dx = a1 g(x)dx,
lim T
0
E I f - hI < E.
Setting
TI g(x)dx,
0
((+00 +m
II f(x)[g(nx + an) - y]dxl < Jh(x)[g(nx+a)Y]dxI+2Mc ,
and consequently,
FUNDAMENTAL THEOREMS 117
tW
lnmssupl l f(x)[g(nx + an) - Y]dxI 4 We,
_W
REMARK: When g(x) = eix and an = 0 for all n, one obtains the
Riemann-Lebesgue Lemma:
+m
lima
f(x)e1nx
dx = 0
n-. _,
if f is integrable.
EXERCISE 3.48: Let (pn) and (an) be two sequences of real numbers
such that
L Ipnl <
n
SOLUTION: For every integer n let AN be the set of x's such that
I Ipncos(nx t an)I 4 N.
n
rn
2
cosx dx =
n
0
EIpnI<
n
S ° Cos2P(nx
S + S)dx
S = 2-2p1 p)meas(E).
E l
liSmssuplcos(nsx t 8s)I = 1.
Furthermore,
and the two latter integrals tend towards zero by virtue of the
Riemann-Lebesgue Theorem.
(b): Let 0 < a < 1 and let Ea be the set of x's (0 4 x < 2n)
such that
Note that
(2p)2p+1/2e-2p
22p 2p lI -2p(2p)1 ti 1 2-2p
( P J = 2 r2 -,E p2p+1e-2p
(p!)2
ti
120 CHAPTER 3: THE
limsupjcos(n x + )I 4 1
34w s S
An n) (n = 0,1,2,... ),
fn + 1 < f 4
and
o).
2-n
Vn D A_, f(x)dx <
JV
n
JW -A .
n n
The function
0
g= E ILV + (n + 1)ILW
n=0 n n=0 n
FUNDAMENTAL THEOREMS 121
JX
°
n=0
J
V
f+
n=0
(n+ 1)fW
-A
f+
n=0
(n1)JA f
n n n n
C n + l
4+ f + n==1
meas(A)
n <
JA
0
for I meas(An) < -, since the An's are mutually disjoint and
meas(X) < -.
1
f= IL
n=1 n Xn
J1fg ==n=1
X n=1
n
I tnfn(x) = E fn cx),
n s s
2-s
and for all x e A one will have Ifn (x) I < whenever s 3 s0,
s
where s0 is chosen so that x e AS . This proves (ii) and (iii)
hold.
(ii) => (i): If limsuptn > 0 there exists a > 0 and a se-
quence n1 < n2 < such that Itn I > a for all s. If for al-
s
most all x the series E tnfn(x) converges, then in particular
to fn (x) -> 0, and consequently fn (x) -> 0.
s s s
liminfJA Ifn1 = 0.
n-o
S
FUNDAMENTAL THEOREMS 123
<
2-s
IfJARs I
I fn
S1 fA a s
(a): Show that for all c > 0 there exists a measurable set
with finite measure such that f is bounded on A and
lim Ifl = 0
E
(which means that for all E > 0 there exists S > 0 such that if
E is measurable and meas(E) < d then
Ifl < E.
jE
Since Al C A2 C . and as
F1 (X - An) = A0UA.,
n=1
ni lX-Anf = 1
A0
f+ JAmf=0.
1
meas(A) ' f <
n0 JA
+Mmeas(E)
JE fl_/ + JEfl/
FUNDAMENTAL THEOREMS 125
JXn - Ix f,
then for all e > 0 there exists a measurable set A with finite
measure, an integrable function g > 0, and an integer n0, such
that for all n >, n0
X-AfnI< e and I g on A.
(b): Now assume that for all e > 0 there exists a measur-
able set A, an integrable function g > 0, and an integer not such
that for all n >, n0 one has:
fX fn -r fX f .
there exists a measurable set B with finite measure such that IfI
is bounded on B by a constant M. and
JXBIfI < E.
By writing
JX-A
fn=J
B-A f+JX-B f +JX (fn - fJAn -f),
if n n0 one obtains:
Furthermore, if
JX_AIfI < E,
JA If - fnI } 0.
FUNDAMENTAL THEOREMS 127
one concludes
J/n J[
SOLUTION: (c): Here are three counter-examples where all the
functions considered are zero outside [0,1], which comes down to
taking this interval as X.
(i): Let
fn = 0.
X-A
Nevertheless,
J/n = 2
(ii): Let
Since
lint 1 sin 1 dx
X
a->O ax
a>0
IJX-Afnl < E.
n
Then fn - 0 and
JX f
n
- 0.
2n
lin
IsixnxI
dx = lim=
fXlfnl
0.
n
= 1 sinxdxl Xk = 21o g2
0 1
FUNDAMENTAL THEOREMS 129
n-'° X if
l imJ - fn I= 0
(d): Assume that meas(X) < -. Show that if the fn's con-
verge towards f almost everywhere, they also converge towards f
in measure. Give an example showing that the condition meas(X)
< - is indispensible.
130 CHAPTER 3: THE
measlIf - fn I
> < 2-s,
ll
s)
S
g<E.
X-A
Set An = (If - f > E/meas(A)) and note that by part (a) one has
I
n
IfI < g almost everywhere; then, by decomposing the integral on X
into a sum of integrals on X - A, A - A n, and Af1An, one obtains:
7
Now, meas(An) - 0 by hypothesis, so (cf., Exercise 3.5M ,
FUNDAMENTAL THEOREMS 131
1imJ 2g = 0,
n-'°° A
n
EXERCISE 3.55: Let a > 0. Prove that if f and the fi's are pos-
itive measurable functions and if fi -; f in measure, then fi - fa
in measure.
where f V fi denotes the maximum of f and fi. Let e > 0 and let
M > 0. Then
l 1)
fi = f + (fi - f) < zM + ,M = M.
Thus:
so that
p(f) = j +
Ifl)-1
(c): Show that p is not a norm on M, but that the sum and
the product of two functions of M are continuous operations in
the metric defined by p.
SOLUTION: (a): This follows from the fact that if Ac = (IfI > c)
1 + If g 1+ f+ 1+ g
one deduces that p(f + g) < p(f) + p(g). From this it results
that p(f - g) satisfies the Triangle Inequality. Furthermore,
if p(f - g) = 0 then f - g = 0 almost everywhere, that is to say
that f = g in M.
so f + g -> fn + gn in measure.
Assume now that fn - 0 in measure and that for all c > 0 there
exists a > 0 such that
m meas(Ig
limsup I > a) < c
n
By the inequality from the answer to part (a), for all s one has:
2s)4-s.
meas(Ifn - fn I > 2-s) s (1 +
s+1 S
2-s
Ifn (x) - fn W1 <
s+1 s
Show that if
n Jfn = Jf
then
n-Jlf-fnl=0.
(See Exercise 6.114 for a generalisation of this result to LP
spaces).
0 4 min(f,fn) f.
'<
Then
so that :
136 CHAPTER 3: THE
and
then
SOLUTION: (a): (i) => (ii): For every set E of finite measure,
1 f 4 J f+ cmeas(E).
E (f>c)
For every E > 0 one can choose c such that the integral of f on
(f > c) is less than e/2 for all f e H. Then, if meas(E) < £/2c,
f < 1.
JE
138 CHAPTER 3: THE
For every k one can find Ek C (fk > ek) such that:
meas(Ek) = min(a,s).
which contradicts ek -> Now let e > 0 and 6 > 0 be such that
meas(E) < 6 implies
f<E
jE
f < E.
(PC)
sup{ f<
feH X-A
sup f = .
feHfX
sup f = co
fEH J Xi
meas(En) = a2-n,
feHJ En f
But this manifestly contradicts Condition (ii).
M = sup J G(f).
feH X
There exists c0 such that t-1G(t) > M/E if t > c0. Then for all
c > c0 and all fEH
J f s L G(f) s E.
(f>c) Mx
rt
G(t) = g(u)du,
J 0
limt-1G(t)
t
Define
one has:
N N
G(n + 1)(an(f) gnan(f) - G(N + 1)a (f),
- an+1(f)) = N+1
n=1
Li n Li
and consequently:
G(f) (2)
J
X n=1 gnan(f)
f s 2-S
1
Then
FUNDAMENTAL THEOREMS 141
I
f f
s=1 (f>cs) s-1 n=cs (n<-4n+1)
00 W
X X n(an(f) - an+l(f)).
s=1 n=c
s
W 00
I E a (f) 1,
s=1 n=c s n
W
G gan(f) r 1. (3)
n=1
fn = nIl [0,/n]'
the sequence (f ) is not uniformly integrable, since
n
Jf = 1
[0,1/n] n
uniformly in n, because if e > 0 and c > 1/e one has meas(fn > c)
= 0 if n .<< c, and when n > c
meas(fn>c)=n<I<e.
I fi=0, 1< i4 n.
A
JB1 =
0, 1.<i4n.
(i): A0,0 = A;
It
- [t
2
ttii
2
Ap(t) = U It Ap,q'
p, q C
A(t) = U A (t).
P>1 p
We are going to show that the A(t) realise some sort of "homotopy"
between A10 and Al in the sense that they satisfy the follow-
,
1,1
,
ing properties:
(3): f1 . = 0, 1 4 i F n;
JAW
(4): For every function f integrable on A the function
e+
t A(t)f
is continuous.
144 CHAPTER 3: THE
This will imply the property P+1, for by virtue of the Intermed-
iate Value Theorem there exists a t (0 4 t S 1) such that
J
A(t)fntl = 2( JA(O)f'+l + JA(l)fn+l)
= (/ fn+1 + f fn+1)
A1,0 Al,l
JAfn+l = 0.
A1(0) = Al 0,
,
and if p >. 1,
A
p+l
(0) = U A p+3-,q,q = U
05q<2p_1
(A
p+1,2q
UA
p+1,2q+1
06q<2p
Ap(0),
U OGq<2p-1
Ap,q =
fi= I 1 fi 0,
JAp(t) Ip,gClt Ap,q
FUNDAMENTAL THEOREMS 145
whence
fi=limJ fi=0.
J A(t) P Ap(t)
It remains to prove (4). Let e > 0; there exists 6 > 0 such that
E C A and meas(E) < 6 imply
where E (t,T) is the union of the A is for the q's such that
p p,q
I C It or I but I C I
(t It ('I
T
,The sum of the lengths .
p , q p,q
p , q T
of these intervals is less than T - t + 2 -P, so that:
meas(E(t,T)) 4 (T - t + 21-p)meas(A).
21-p
If p0 is the smallest integer such that meas(A) < 6/2, then
for T -- t< 6/2meas(A) and p > p0
e,
IfA (T)f - fA (t)fl ¢
p p
and on taking p -+ w,
IJA(T)f-1A(t)fl<e.
whence:
ay
B((%)f
is continuous.
Yi=1
E.
fi (i=1,2),
FUNDAMENTAL THEOREMS 147
yi) = o,
which implies
JE(lEf=ayi.
i
.
=ay1+ (1-a)y2.
A(x)dx
JX
f(x)dx,
JX
yi= J Xfi
JX (fi - yi) = 0,
Yi) =
JE' - 0
JE1 = ayi
Set
fl(x) x e E,
f(x) -
f2(x) x e X - E.
X = J f l + J Xf2 - J E 2 = ayl + (1 - a )y 2.
J
Adx
fX
r(A) C 1 Adx.
X
1AdxCTA.
X
y=1 f(x)dx6FA)-P(A).
150 CHAPTER 3: THE
.2(y) = JxI(f(x))d3 = 0,
and 1(a) > . 0 for all a e t A , and in particular t(f(x)) >. 0 for
all x e X. By modifying f on a negligeable set, one would then
have 1(f(x)) =.O for all x e X, or f(x)e.2-1(0)n A, so then y 4
r(.2-1(0)nA). Since it can always be assumed that A affinely
generates R , and consequently that the affine dimension of
.1-1(o)n A is strictly less than that of A, one is led to a con-
tradiction.
Jg(x)ds = 1,
1
mess E
f(x)dx,
and consequently,
f 1(V) = U f 1(Va
n=1 n
has positive measure; since f 1(Vf) has measure zero, meas(f 1(V))
> 0. Conversely, if y4 Af and V = Vf one has meas(f 1(V)) = 0.
1
yi mess TE-i7f
E.f
I
(f - yi) = 0.
JEEI
Ft C E, F0 = O, F1=E,
t * meas(Eif1Ft) is continuous,
and
JF IlE1(f - yi) = 0,
f = meas(EifFt)y,.
J Ei f1 Ft
Set:
then
FUNDAMENTAL THEOREMS 153
Now note that Gt D EOflE1, so that meas(Gt) > 0 for all t; the
function
meas(E0 - Ft)
t .-
meas( t)
f = ay0 + (1 - a)y1.
mess Gt
G
1 tmeas(E0)y0 + (1 - t)meas(E1)yl
meas Gt JG f tmeas EO + (1 - t meas E1
t
rso that the first member runs over the segment [y1,y0] when t
varies from 0 to 1. We have thus proved that If is convex.
Let us now show that Af C If. If y e Af and c > 0, the set
{x:IIf(x) - y11 < c} does not have measure zero; it therefore con-
tains a measurable set E such that 0 < meas(E) < m. But then
154 CHAPTER 3: THE
C.
Ily meas E Jfll Ilmeas E JE(f Y) 11 4
UP=o,
JE
one would be able to assume that Af C £-1(0), and one would end
by arguing by induction on the affine dimension of Af. Let us
note that if f is bounded, then Af is compact, and consequently
If = r(Af).
U Va =U V
a neJ an
FUNDAMENTAL THEOREMS 155
the Theorem. n
b
f(x)dx = 0
ffo that :
=limfV
n
f=0.
jE
n
1
meas E
Jf '
0 < meas(E) < 0,
E
nVJtn-lf(t)dt
=
f1f(p(x))dx
12 0
if f 3 0, or if
J0t1t)dt <
rb
meas(a 6 p . b) = V(bn - an) = nVJ to-ldt,
a
nVJtn-lf(t)dt
= (1)
f'f(p(x))dx
3t 0
FUNDAMENTAL THEOREMS 157
tn-1f(t)dt)
supJf.op = sup(nVTO
i 1 i 1
and fiop -> m on {x:p(x)e E}, which proves that this set has meas-
ure zero. If E has measure zero without being bounded one has
(p e E) = U (p e E fl [0,Ri] ) if Ri
i
and therefore has measure zero. Hence if fi - f almost everywhere
am ]R+, then fiop + fop almost everywhere on ]Rn.
Now assume that
Jt'If(t)Idt < m
iad let (fi) be a sequence of step functions such that
lien
tn-1If.(t)
lim J nIfiop - f.opl = nV lim - f (t)Idt + 0,
i, j-t°° IR i,j mJ 0 1
158 CHAPTER 3: THE
J ifIg < -,
Jgh = 0
L2, and note that if f >. 0 then g e L2. If g were not to be-
long to the closed vector subspace generated by the functions
IlP/, the projection theorem would assure us of the existence of
a function he L2 such that
If - filg < II ( -
f f = u + iv, u = u+ - u-, V = v+ - V_.
This result generalises to the case of a function f such that:
Jlflpg P % 1,
(If p(x) = inf(X > O:x e X(V - V)) is the gauge of the symmetric
160 CHAPTER 3: THE
(1 - p(x))V + y C Vn (V + x)
X= p(x)(z1 - z2 ), z1 a V, z2 a V.
(1 - p(x))v + y c vn (v + x).
Denoting the characteristic function of V by cp one obtains:
s meas(Vr(V + x)) =
whence
meas(V)I
11 V-V
(1 - p(x))ndx c
tf n(q)*)(x) = meas(V) 2,
FUNDAMENTAL THEOREMS 161
(1
f (1 - p(x))ndx = nmeas(V - V)J to-1(1 - t)ndt
V-V 0
_ nr(n)r(n + 1)
meas(V - V)
r(2n + 1
(n!)2
meas(V - V),
(2n)!
rhence, at last,
+0 +W
f- p(t)dt = 1, f- t2p(t)dt = 02 <
00
(a): Set:
I G(x)dx.
0
00
162 CHAPTER 3: THE
p(t)dt < 12
tI,acr 2a
1
2a2
p(t)dt
1 23 3
if 0< a<
AVA = DADA= AVA = VAV= AVA
X Er J p(t)dt
x
r
G(x)dx = 2J dxl p(t)dt = 2 p(t)J
J2
dx
j0 o V"X 0 0
2J0t2p(t)dt = a2.
0
and consequently:
xy =
b - (Q - K)
2]
[1
whence:
,,,.2 2, 1
2 '
2A
(d) : Since the function A 'r G(A2a2) is convex and G(O) = 1, for
O<a 0
l 121l
G(A2a2) < 1 - + 1 - 1 -
0J 0 2A0))) X0
164 CHAPTER 3: THE
1
u=
2A2
- 2(1- 12J =- 1
A
0 2a X
0 0
A0 - 2
v= J f2<M, f=0.
X JX
SOLUTION: Now,
e-2(1 - 1+
fa= 1 + af - 1+
Jx JX 2
FUNDAMENTAL THEOREMS 165
Note that
ml1 + au - 11 + u' al
l = Ya(1 - a),
U- *O ll
u2 J
Jim
(1+au- 11+u1al -0
J
lul- U2
Then
(1 + aef - 2 l l + of lal
liml JI = la(1 - a)f2,
E-}0 E
11 + aef - l1 + of f a l Bf2.
<
2
One can therefore apply Lebesque's Theorem, which gives the result.
f g(x)dx = 1,
X
( b): Further assume that g(x) > 0 for all x e X and that p
is strictly convex.
Show that the inequality in Jensen's Formula is strict unless
f is equal to a constant almost everywhere.
W(-) = N(y)).
if
Jh=Ih+_1h_e
then
Jh1, Jh2.
y0 = 1 f(x)g(x)dx
In particular,
X
.2= £(y0) = 0,
f(f(X))9(x)dx
Jf(x)(x)dx = +m.
X
J (p(f(x))g(x)dx Jag(x)dx = a.
X X
J = {(y,t):yeI, t 3 (,(y)}.
It is easy to see that J is convex, and that if y0 a I, the point
(y0'(P(y0)) belongs to the boundary of J. Hence there exists an
affine function on Y X1R which vanishes at this point, is positive
on J, and is not constant. This is expressed by the existence of
a linear form u on Y and of two numbers a,B such that u and a are
not simultaneously zero, and
u(y0) + a9(y0) + B = 0,
Orall0<a<1,
acp(y0) + (1 - M)9(y) cp(ay0 + (1 - a)y)
so
A = Jf(x)dx.
X
then:
1 + A` 1 + f (x)dx < 1 + A.
JX
1 + A` 5 J 1 +
f2. (*)
1 1 +f2E1+A
Y
follows from 1
-+f2 .< 1 + f. If Fca
I,-+-f
of is the length of the
arc AC (cf. the Figure).
Inequalities (*) and (**) ex-
press that:
AC S AC F AB t BC.
J/1+f2=
1
1 + A2,
0
172 CHAPTER 3: THE
1 t f`>. 1 +A`+AA ,
1 + A`
and on the other hand the integrals of both sides are equal; since
f is continuous, it folllows that f = A. Similarly, if
J1
=1+A=1 (1+f),
0 0
1 +
f2 = 1 + f,
and so
f = 0.
Jx f.1 g >. 1.
X
f > 0, g > 0.
Jx Jx
If
T gI s M, n = 0,1,2,...
E
then:
= 0, n = 0,1,2,... .
JE
ifaentg(t)dtl
.< M, n = 0,1,2,...
0
xn
eXf =
n-0 n
174 CHAPTER 3: THE
and consequently
n
fexf91 = IX
n niJf I ¢ x.
JeZFg.
4)(z) = (2)
IeZFgl
Then < eBx IgI, which proves that ¢ is an entire function
and that
IIgII1eBx+
(3)
1 ) n
n >, 0.
J F
FUNDAMENTAL THEOREMS 175
But
fn
Cn(f - 1)S
s
s
so that by (5):
L$ince -an/4 < a(8 - n/4) < an/4 < n/2, one deduces from (8) that:
Re(z) > 0, Im(z) > 0, IzI < p0, by the Maximum Principle one also
has IGI .< C in this domain. Finally, if z = pelf, p - 0, 0 < 0
< n/2,
J0tgt)dt = 0, n - 0.
a
e
J P(u)g(logu ) uu = 0.
1
a
e
c(u)g(logu) du = 0.
1
3
EXERCISE 4.7,#'' Let cp be a continuous real function on [O,a].
'Assume that W(x) > 0 if 0 < x -< a and that p(x) ti Axr as x ; 0
(A > O,r , 0). For all t > 0 set
a
dx
F(t) = ot+W(x)
f
ra
dx
(t) q(x)
t->0 0
F(t) ti
IT 1
rAl rsin(n/r) tl - 1 r
F(t) ti log 1
A
178 CHAPTER 4: ASYMPTOTIC
(For part (c), show that this can be reduced to the study of the
integral
a
(t + (p(x))-1dx,
at
Set:
y
1/r-1
if 0 < y 4 ar/t,
1 + tqt1/ryl/r)
ft(y)
0 if y > ar/t.
Ay,
Since, for y > 0 fixed and t - 0, t 1p(t1/ry1/r) -
1/r - 1
lTO mft(y) =
1y+ Ay
1/r - 1
ft(y 1+By .
EVALUATION OF INTEGRALS 179
( lY
1 it
t-r0
+o ft(y)db = 1 + dy = l r sin(77r)
0 J 0 A
SOLUTION: (c): Note that the relation proved in part (a) is true
for all r 0, so that lim F(t) = W if r >, 1. Now,
t+0
10t
< B log(1 + Ba),
j0 t +d-p(x t d+xBx =
So that
F(t) v a dx
f at t + wp x
Y
x
fat t+cx =alogt
0t+9(aty)
Men 0 < y < 1 and t -> 0 one has 9(aty) v aAty and t = o(ty).
Consequently
lim ty -
t->0 t + p(aty)
< aB
t + P(aty) t + aBty
1
t}'
lim f dy = - .
t->0 0t+ip(ats')
fxP dx < 0.
+0 1 + xqjsinxxjr
n=0
where
I = fn+1
xp _
J1 (x + n)p
dx.
n n 1 + xq, sin= Ir o 1 + (x + n)q(sinnx)r'
dx
(p(t) _
0 t + (sinnx) r
then
p(t)tinlog1 if r = 1,
1
1 - 1/r
(P(t) if r > 1.
ti rsin(7 r)(t)
Consequently, if q > 0,
Inti2Anpq
r if04r<1,
ti 21 np glogn if r = 1,
n
2
n
p - q/r if r > 1.
rsin(irt/r)
n
It follows that if q > 0 the integral converges if and only if
> max(r,l).
p
(a
l9(x)le
-f(x) dx <
(iii): J
0
E82 CHAPTER 4: ASYMPTOTIC
I
J0 g(x)e-tf(x)dx ' i
a rP-1-)(At)
+ 1l -(S+1)/a
r(p)r(g)
B(p,q) =
r(p+q)
p q+ 4 B(p + 1,q).)
B(p,q) =
SOLUTION: Let 0 < B < A, y > 1; there exists b such that 0 < b
< a and
a
f(x))3 Bx , 0 4 g(x) 4 yx if 0 < x 4 b. (*)
Since f(x) > f(b) > 0 for b 4 x < a, when t > 1 one has
t(S+1)/alag(x)e-tf(x)dx
1b
t(0+1)/a j bg(x)e-tf(x)dx
= JF(x)dx
0
where
'EVALUATION OF INTEGRALS 183
is/°g(xt-1/a)e-tf(xt-1/a)
if 0 < x bt1/a
Ft(x) =
0 if x > btl/a.
By (i) ,
a
xse-Ax
t-
limFt() =
and by (*) ,
a
0 < Ft(x) < yxse-B"
limt(S+1)/a(a
g(x)e-tf(x)
10 dx = J-X'e -Axadx
t 0
1 A-(Otl)/ar(B + 11
a a
aa+1e-a(W e-a(u-log(u+l))du.
r(a + 1) = J xae-xdx =
0 1
0
The above result can be applied to each of the integrals and
with a = 2, A = , B = 0, giving
0
F e-a(u-log(u+1))du 2xr()(2)- =
ti
1 l an
184 CHAPTER 4: ASYMPTOTIC
r(a + 1) ,
aa+gi
e-aV
(1
B(p + l,q) = J p(l - x)p+q-1dx
o 17X---X)
(1 p-1
x dx
=
p + q 0 (1 - x)p+q 1 x (1 - x)2
= B(p,q).
p
a(a + n) na
nl - r(a) ,
So
(p + q + n) r(p) nq
,,
p...(p + n) r(p + q)
x)penlog(1-x)
So
EVALUATIONS OF INTEGRALS 185
B(p,q) = r(p)r(q)
r(p + q)
,
n -x`` sinx4dx
x e = 0,
0
f eitx-x4sinx"dx r()ein/8t-5/4
ti 4
J0
F(z) = x4n+3e-zxdx
J 0
(4n + 3)!
F(z) =
4n+4
z
J x4n+3e-xsinxdx
= 0,
0
186 CHAPTER 4: ASYMPTOTIC
1
(Wn-xIT ,
J x e sinx'dx = 0.
0
Now set:
1 1
which is valid for y > 0, then for R > 0 and t > 0 one has:
I
I < exp[- tRsina - R4(cos 4 - sin-!)),
4
J f(x,t)dx = if f(iy,t)dy
0 0
limt e-in/8f(it-1y,t) = y e y
by (2)
ieln/BrWy4e-ydy
limt5/4JWf(x,t)dx =
t-3 0 0
n
exp(ez - 1).
n=0 n n, =
(b): Deduce from this that for every real number u > 0:
188 CHAPTER 4: ASYMPTOTIC
i u+2co
do
= 2nieJu_i z-(ntl)exp(eZ)dz.
u U
d , n' exp(e n - u e nlogu - lu
n eJ n n n
n
n
do+1 (1)
p0 lP ,dn-p
do+l n
(n + 1)
- 1 d -p .
(n + 1)! p)!
p-_0 p! (n -
If the series
n
f(z) _ do n, (2)
n=0
L
EVALUATIONS OF INTEGRALS 189
f'(z) = ezf(z),
z-(n+1)exp(ez)dz IN
C B
n = 2nieI
- M O
where r denotes the rectangle indi-
cated by the figure. Note that, on D A
-IN
this rectangle, if z = x + iy,
and consequently
u 2u + 4M
.< exp(e ) 0+1
JBCDA
= n1
2x_f+ (u + iy)-(n+l)exp(eu+iy)dy, (3)
do
190 CHAPTER 4: ASYMPTOTIC
+W
do exp(eu - ueulogu)J- g(y,u)dy, (4)
n!
where
Note that
((
2
l g(y,u) I = exp{ - eu[2sin2 + 2 log I 1 + 2 ] }.
U,
111111
Therefore
u
9(y,u)dy1 < 2J (1 + u 2y2)-jue dy
if ly l >7E n
dy
2
Jn 1 + -u Ie u y 2
whence
+x xeu/2
e-u/2J-xeu/2
g(y,u)dy g(te,u)dt,
-u/2
f-x =
2
limg(te-u/2, u) = e-t /2
2 2
a- 2t /x
Ig(te-u/2,u)I .<
tie-4/2Je_t2/2dt
= e-4/2 x.
Jx g(y,u)dy
x
u
d ti n! exp(e n- u eunlogu - u ).
n e/27 n n n
Bence
u u
logdn = logn! - (une nlogun - e n + jun) + 0(1).
u u
u e nlogu - e n + '2u ' nlogu = o(n(u + logu ))
n n n n n n
= o(nlogn),
EXERCISE 4.78: Show that for every real number t the integral
(
33Jdx
1
fi(t) _J cos)tx +
0 l
4,"-to=0.
Next prove that as t -> =
1 t3/2 (1
4,(t) = expi- 3 + 0(t-3/4)),
7.7
2
0(-t) = 01*J
t1T
SOLUTION: Note
J°°expi{tx 3 Jdx.
If z =x+iy,
( 3
Iexpiltz + 3 II = exp( - ty -
x2p + 1 y3),
3l 2
iexpi(tz + 3 JI < Be -X y,
x+ia 3l (a 2
expi+
Itz zJdzl 4 BJ a-x ydy 4 B
fx+i0 3 0 x2
ll
2/(t) = tz + 3 Jdz.
JL{ a
(1)
3
2/0"(t) z 2explltz + Idz
JL )
a
( 3 2
)e-ax
Iz2expi ltz + 3JI 6 B(x2 + a2
Hence
( 3
2,r7,-(,"(t) - ti(t)) (t + z)expiltz
2 + 3Jdz
JL l
3l1II
( -+2'.a
[iexpiltz + 3J] 0.
z=-°+ia =
( 3
2 4(t) = J expixlz + -) dz.
LS l
194 CHAPTER 4: ASYMPTOTIC
+m ( 3l
2
'r O(t) = e-2X/3+- expal- u2 + 23 Idu
1t l 111
1 e -2a/3 (+mexp
I- u2 + 3 -Idu.
V 1, + J+We-CO
- 1)du.
2
JIuI3e_u du.
-m
(
3
(-t) = J )dx.
rit"
0
(
Re{e-2ia/3T (
3
(-t) = 1- xpi 1u2 /T) d.)
+ 3
EVALUATIONS OF INTEGRALS 195
Set
((
f(z) = expila2 + 3x) = Rei*,
where, as z = pale,
3
R = exp( - p2sin28 - - _ sin30),
3
= p2cos20 +
p COS33.
2
e-'p
3A l 2 3/A J
J (u)du = f
f(z)dz,
"f t+L
196 CHAPTER 4: ASYMPTOTIC
J
rf(z )dz J Q
4e-ax8dO
=O
lfJ
Note next that for 0 = n/4, _ -p3/3 I. Then (2) and (3) show
that
in/4) IRe"
If(pe - e -p2 I
< - RI + IR - e-pI
2
+le-anp2/4
3 3X
Consequently:
e_p 2
ix/4 f
J f(z)dz = e
L J
f dp + H,
with
H S constant 7 p
I3e-anp /4dp = 01J
Since, furthermore
r-F_e p2dp e
-A
"I F
2- '
one has
- = e1rz/4J,r-
E + O{
1 J
Finally,
x n -2iA/3 in/4 1
Re e
EVALUATIONS OF INTEGRALS 197
- l lI
Jdxf1(x,y)dy, lf(x,y)!dxdy.
0 0 JpyJf(x,y)dx,
0 Ii. O<x,y61
2 2
f(x,y) = x - y 2
(x2 + y2)
(b):
f(x,y) = x - y
(x2 +
y2)3/2 ,
(d) :
199
200 CHAPTER 5:
SOLUTION: (a):
Y0
dy I
dx [-x2+ 1
J0 J0 (x2 + y2)2 = _
fl fl x2 2 1
x x=1 1 (_d )
n
0 (x2 + y2)2 0 i
x2 + 2]= JO
b _-4
x 2 -Y 2 1 :2 _ 2
dxdy = 2 2dxJ dy
y2 ) 2
JJ Otx,y<1 (x +Y 2)2 0 0 (x2 +
1 y =x 1
y2]y=0 = dx
2 fo [x2 + 0 x = W.
SOLUTION: (b):
1
x
J f(x,y)dy = (x - )3
0
1 1
dyj0 f(x,y)dx = 0.
J0
Finally,
1
f0lf(x,y)ldy = Ix -
I-2,
FUBINI'S THEOREM 201
so
If(x,y)Idxdy = -.
J 0)
if 0
SOLUTION: (c):
1 1 x_ b 1 1 +x y=l
f
f 0 x(x2
y =
fo + y2)3/2 d 0 Lx (x2 + y2)Jy=0
Clearly
1
dy1
(1 x-y dx = - log2,
y2)3/2
0 0 0 (x2 +
and
1 dxJ1(1 _ 1 P _ 1 dx,
1 1( (1 - x
- xy) Pdy =
fo o p o
JJxydxdy,
0<x<1
a,cysb
prove that:
J 1 xb - xa ax = log 1 + b
0 logx 1+a
b ('1 b
dyJ xydx = J yyl = log 1+b '
fa O a
10dxfab 1 xy y-b 1 b a
xydy = JOdx [l ]y
f a = J0 ogx
xdx.
D(u,y)
- - 2(x2 + y2
D(x,y)
so
b 1 b
fadvf0uvdu = fa v + 1
(x2 - y 2)Xy(x2 + y 2 )dxdy =
ffR J
'lo l+b-21og1+a
2gIta l+b
SOLUTION: We have
= X2YdXAdYAdZ,
204 CHAPTER 5:
so
xyz(1 - x - y - z )dxdydz =
fffT
r(6)r(2)r(4)r(2)r(2)r(2) =
1
r(8)r(6)r(L) 7!.
dxdy
x>O,y>O (1 + y)(1 + x2y)
f x2 - 1
low' dx0
00A = 000 - A0A = V V = AOA
f
dy
JL>0- f0 1 +y 1 +x2y 2J0 1y+ y
2
it it n
2 sin(n/2) 2
2 r
n
2 = J o dx 0 (1 + y)(1
dy
+ x2y)
=
(Contd)
FUBINI'S THEOREM 205
jm
2
dx
-f-1-)dy
Jo 1 o`1 + x2y
=J° dx Ilog11ylyco
2
logx dx,
0 x2 - 1
so
2
l ,x dx=n
J00 0 x 2- 1 4
3ydxdy
.
Jx>0,b>0 y)3
1 + (x +
x+y<a
ra du (u
J (u - v)dv = 2 ra
'
u2du
3
0 1 + u3 -u 0 1 + u3
1 + a3 - 1.
206 CHAPTER 5:
dxdydz
fffo<X,Y<l (1 + x222)(1 + y2z2)
z>O
(tan izJ2dz.
Il
To
SOLUTION:
dxdydz
JO<x,y<1 (1 + x222)(1 + y2z2)
z>0
a
dz
dxdYJ 22 22
'1O<x,y< 1 0 (1 + xz)(1 + yz)
2 2
+yy2z2Jdz
I JO<x,y<1 1 +xx2z2 1
dxdy
=
=nff
2 0<x,y<1
x+ y 2 0 0 x+ y
1(log(x
n
+ 1) - logx)dx = nlog2.
2
0
It follows that
dxdy
nlog2 = Fdzff
O<x,y<l (1 + x2z2)(1 + y2z2)
FUBINI'S THEOREM 207
1
1+dxz
2 2 2dz
-Jo{J
= r0(tan-1z12dz.
JO z J
n/2
it
q(x)dx
Then
p = a1...anp
so
+e-asu2du n
e q(x) =
7n7-
= R X n/2
JIltn 81 J s i" S
EXERCISE 5.87: Determine the values of a for which:
cosx
sinxsiny)a
dxdy <
I J0<x,y<n/21 -
a
sinx
I =
ffO<x,y<7c/2
(1 - cosxcosy) day'
sinx 2x + 0(x3 )
cosxcosy = 2 - 22x(1 + 0(r2) ),
4
r + 0(r ) r
whence:
a = 2axa + xa
sinx
(1 - cosxcosy) 2a 0( 2a-21
r r
11O<rl xardxdy =
(1
rdr.J n/2 cosxdx,
JO 0
x>O,y>O
IP(a) = xlx2...xpdx1dx2...dxp
<a
P
I(a) = a2pI(1).
Ifp32:
(0 1
rxi
x2...xpdx2...dxP
IP(1) = J x1dx1J
0 x2+...
p
<1-xl
(1
= Ip-1(1)J x1(1 - xl)2P-2dx1
0
r(2)r(2p - 1)
= Ip-1(1) r(2p + 1)
Since
r1
I1(1) = xdx
J 0
it follows that
r(3) 1
I(1)
p =
2 r(2p + 1) - (2p)!
so that
a2p
T(a)=
p (2p) 1 .
al-1 ari 1
xl ... xn dx1...dxn
Pn
1l P1+...+ rn
xi>0,
J
n
J sl a
a11...ann
a
r 11
all
... r (Yanl
1n .p a
1+...+ a
n+ll
1 pn
210 CHAPTER 5:
pi
SOLUTION: The change of variable Xi = (xi/ai) transforms the
integral into
a al . ..a an
1 n_ Xlp1
a-1
- n-1
pl...pn Jf dXl...dXn
JX.>0
i
X1+...+n
l
x.+---+x 4X
I S
a
It is clear that IS(A) = A SIs(1). Consequently,
1 a -i a
n-1dx
In(1) = In-1(1) xnn (1 - xn) 1
0
r(an)r(a1 + + an-1 + 1)
- In-1(1) r a1 + + an + 1)
Since
(l a -1
Ii(1) = J x11 dxl = a ,
it follows that
1 r(an)...r(a2)r(al + 1)
In(1) = a1 r(a1 + ... + an + 1)
FUBINI'S THEOREM 211
r(al)r(a2)...r(an)
r(a1 + + an + 1)
dxdydz
Jx>0 1 + xa + ys + zy
y>O
z>O
dx
J O1+xa+ys+zy
2/a 2/$
x = u , Y = v , z=w 2/y
The integral may be written
_ S
u2/a - lv2/S - lw2/y - 1
dudvdw.
I aOy 111u>0 l+u +v +w 2 2 2
v>o
w>o
which yields
212 CHAPTER 5:
I-a/2sin2(1/a
+ 1/B)-l6cos2/Y
- 13d8
0
Y)
Thus if
1
a + a + < 1
y
then
1 = n rlalr[oil r[1]
aay
r {cx + B + sinn l a+ 0+ Y)
yj
which can also be written
Y)
aBy r (a) r (S) r Y) I a
dxdydz
fff 0<x,y,z « 1 - cosxcosycosz
SOLUTION: Set
FUBINI'S THEOREM 213
u = tanix, V = z = taniz,
dudydw
I 4
JJJ>0 u2 + v2+ W2 + u2v2'W2
v>o
w>0
/2d3f/2dTf 0
I =
J0 2
0 1 + r4sin4Ocos28cos2
Finally, setting
we have
Noticing that
r(;)r(4) = nom,
sin 4n =
I = 4x(4)44
J+_
f(x) 2 dxfg(x)2dx - lIJ f(x)g(x)dxl2
-00
2JJ2(f(x)g(y) - f(y)g(x))2dxdy
= loo,
JJJ'ydz
dxdydz
'SIC
r 1 2
JJC dxdy = s0,
1 0J
z
JJ xdxdy = 3 V0,
ydxdy
Jcz
1 3 n0S0,
11 zdxdy = zICoJS0
2 = 1 0Jc6S0,
3
JJ z ll
so that
NZ-013 C0S0dz 3
J 00
=E ,
4 0
etc.....
2 S0dz
J00N-E-O)
V = 6 Sl + S2 + 4S3
x = a(t) + zb(t),
rz {l
y = c(t) + zd(t),
h
V = I S(z)dz = 6 (S1 + S2 + 4 S3).
11
0
0 0
FUBINI'S THEOREM 217
afm
J0(y)
I sinax
1 +x`
dx =
0 a2+y2
d y,
y,
2 rn/2
J0(y) = nJ cos(ycose)d8.
0
Jsinaxdxjf(y)e-x3'dy = JW f(y)dyje-x3'sinaxdx,
E 0 0 E
since
x r
dxJ Isinaxf(y)e-xyldy < (X - E) Je- Eylf(y) dys -.
fE 0 0
r
x
2
e-xysinaxdx
J
a
e
X
J_XYid
E Y
Consequently,
X
f(y)f e-xysinaxdx
4 a Il(0,1)(y)If(y)I + an(11-)(y)lf(y)Iy-2
E
218 CHAPTER 5:
Now,
e-xysinaxdx = a ,
0 y 2 + a2
so
f (y)
of dy = sinaxdxJf(y)e-xydy.
o a2 + y2 0 0
n/2 W
= nd8J e-xycos(ycose)dy
fo 0
=
2 (n/2 xd8 - 2x rm dt
n
0 x2 + cos28 0 1 + x2 + x212
t
2 1 W 11
f
tan
`n 1 + x t=0
7,77j=:
xt 1 + x-
W n/2
2J dyJ e-xyjcos(ycos8)Id8 < x < ").
n n
FUBINI'S THEOREM 219
show that if
Joe-H'(x)
<
T 0
then
m(r) dr <
T 1 + r2
AVA - OAV m AVA - 0A0 - AVA
I (t)dt < -,
0
(x
u(r) = supJ log(rcp(t))dt,
x>.0 0
then
Notice that u(r) > 0 for all r > 0. On the other hand,
I x
log(rp(t))dt < J1og+(rq(t))dt,
0 0
220 CHAPTER 5:
so
Consequently,
log(r(p(t))
f r2 0 0 r
(J'OW(t
(x
xlogr - H(x) H(o) + J log(rpp(t))dt,
0
so
JlogIl + zfj = 0,
Show that
Jlot'l = 0
2n
Jo logll + zeitldt = log+lzl.
2n
dtJ log 11 + Peitf(x)Idx = 0.
f0, X
2n
0 dx log 11 + peitf(x)ldt = 21 log+lpfl.
X JO X
From this it follows that IfI < p-1 almost everywhere, hence
that f = 0 almost everywhere. It remains to be seen that Fu-
bini's Theorem actually can be applied. By hypothesis
4 4n(1 t p JX If 1) < m
222 CHAPTER 5:
J fP = pJEdx(0(x)tp-ldt = pJOtp-1dtJ(flt)dx
E
M
= pJtP-1meas(f > t)dt.
0
SOLUTION: (c); One can show, as in part (a), that the set D'f
FUBINI'S THEOREM 223
f f = meas(Df) = meas(D'f).
J
k
meas(Lxf1Dn ) < meas(Lxf)D) (1)
i=1 1
JP(x)dy > 0,
224 CHAPTER 5: FUBINI'S THEOREM
and consequently:
jgq 1,
JfP - =
1
.where q is such that + 1 = . Replacing f in (*) by fgl q and
p q
g by gq yields:
Jfg' 1.
225
226 CHAPTER 6: THE
(1 - 1/PlP-1
l )
x l/p
Jg = 1,
then
J (f + op = 1.
x1/P 1-p
I
`1 + xl/PJ
Jg = 1,
then
JgP 1,
=
fPg-P
then replacing g by gp in (*) and f by 0 when g = 0 and by
otherwise, yields
1+ (Jfp)1/p
< (J(f + op) 1/p.
x->xp+1- (x+l)p.
From this it follows that if f,g,h e L
= d(f,h) + d(h,g).
Furthermore, d(f,g) = 0 implies that If - gIP = 0, that is to say
f = g almost everywhere.
To show that LP is complete it suffices to prove that if fn e LP
and
gN =
NW
then fn converges to an element of L.
E
n=1
Ifn - fn+1l, g = E
n=1
Set
Ifn - fn+ll'
Then gN - g and
JgN A
By Fatou's Lemma
JgP < A
i (fn - fn+1)
n
LP-SPACES 229
n=r
f I p
Ilfr - < d(fn'fn+1),
n=r
EXERCISE 6.104: Let K be a compact set of IItn, e > 0, and 1 , <p <
Show that there exists f e Lp(ntn) such that f 3 0, 11f11p = 1,
and:
Ilfa - flip<e, a e K.
1
f XR.
meas(BR)1/p
meas(BRtr) 1/p
Ilfa - flip E 2 ( meas(BR)
- 1)
which gives
Now let us consider the case a > 1. By the Mean Value Theorem,
fi)a-1,
IJ - Til If - fiI (f V f V fi = max(f,fi),
1)-1
and by HUlder's Inequality for the pair a,a(a -
LP-SPACES 231
if
pn
q/r(Jfpgq)1/r
Jfg < IjfIj1 p/rIIgIIq
if
t + f(t)g(x - t)
is such that
1 = C 1
PO ieI pi
It may therefore be assumed that pi < m for all i. First consider
the case where n = 2. Then
pOp0
p1 p2
which gives the formula in this case. When n > 2 one proceeds by
induction; in fact, if
1 1 1
W p2 pn
then
1 1 1
PO p1 W
whence
IIf1f2...fn11PO E IIf1IIp1IIf2...ffII,
E
IIf1IIp1IIf2IIp2...IIffIIPn
p1
_ r-p
pr , P2
=
r-q ,
gr
P3=r.
Then
1+ 1+ 1= 1+ 1- 1= 1,
p1 p2 p3 p q r
1 p1 3
2 2 3
Jh1h2h3 (Jhl ) (Jh2 ) (Jh3 ) .
If
h33
h11 = fP, h22 = gq = fpgq,
that is to say, if
234 CHAPTER 6: THE
91 - 1/p(fPg)1/P fg.
SOLUTION: (c): Assume for now that p < m, q < -, and that
p + q - 1 > 0,
so that
Jdx(Jff(t)g(x - t)Idt)r
= IIfIIPIIgIIr <
one has
h()(rdx , IIfIIPIIgIIq,
J
Ih(x)I , IIfIIpIIgIIgi
i.e.
Ilhll , Ilfllpllgllq
JfI M
Tx)If(x)Ip-2 if f(x) 0,
g0(x) _
to if f(x) = 0,
Set
1
meas E f x
if x e E,
g(x) _
0 if x4E.
Then IIgIIl = 1 and:
1 E JEIf(x)Idx 3 M.
Jfg = meas
IIJfgl
< 1, otherwise there would exist a e a such that Ial = 1 and
whence:
that is to say:
J(1 - f)gi 6 0,
By what has preceded, there exists gn > 0 such that II9n11q < 1
and Jfngn >- IIfnIIP - 1. Since Jfgn > Jfgn and IIfnIIP -; IIfIIP
the result is again true in this case.
The functions g nE are simple, II9iIlE Iiq < 1, and f(g IlE ) is in-
n n n
tegrable. From this it follows that
IJfn9I < M.
By part (a) this implies that IIfnIIP < M, and therefore that
IIfIIP = 1nmllfnllP < M.
The un's are continuous linear forms on Lq. Moreover, for g e Lq,
I Jfhl , JIfIi
and, furthermore, if f and h are real:
Jfh< IJfhI.
240 CHAPTER 6: THE
All of this shows that the left side of each of the equalities to
be proved is greater than or equal to the right side.
Let c > 0, and let fl a En be such that
J If - f1I9 < E.
fl
hl = If g 1
and if f is real
f1+
=
h
2 f 1
9
Jflhl = JIf1I 9,
and if f is real,
Jf1h2 ffig
-
=
When f is real,
IJf9l <M
for every step function g such that I9Ilq < 1, then IlfAIp < M.
SOLUTION: Set
g(y) = JXf(x,y)dx.
IIgIILp(Y) <
J(Y)P(Y)dY = JdxJf(xY)cP(Y)dY,
Jg(P <
m
EXERCISE 6.111: For every i , 1 < i < n, let X. = i 1, and let fi
be a positive measurable function on X1 x ... x Xi x ... x Xn (where
the hat indicates the term that must be omitted in the product).
Set
J_i,
Ii = fi1...dx...dx
^
(I1...n)1/(n-1)
vol(V) c S12 .
I = f=
Writing, for 2 .< i 4 no
Jf1dx2...dxnJf2...fndx1.
CHAPTER 6: THE
gi = Jf11_1dxl,
1/(n-1) 1/(n-1)
Jf1 2 .gn 2.. n'
Now, using the induction hypothesis shows that the integral on the
right side is majorised by
.dxn)1/(n-2)
(gidx2. .- 'a1..
i=2 J
Ii = Jgidx2...dxi...dxn,
apmeas(En) <
JE
If - fnlp < 'If - fnIIP,
n
lflp < e2 p,
JX_B
IfnIp<C if1<n<N,
X-B
n
which proves that the sequence (If) conserves mass and is uni-
formly integrable.
Now assume that fn - f in measure, and that the sequence (fn)
satisfies Condition (ii). For every c > 0 there exist a set A of
finite measure and a number 6 > 0 such that the conditions (2)
and (3) are satisfied for every integer n. By Fatou's Lemma the
function f also satisfies these conditions. Set
p e
En
I'f - fnI > meacA }nA.
LP-SPACES 247
There exists an integer N such that meas( n) < 6 for n > N, and
consequently
+ (J IfI )1/P +
JE
n
Ifl)n
+ (J If - fn1P)1/P
A-E
n
< 5c 1/p
limllfn
n-
II
P
= Ilf P
J X
IfnIp = j Iflp,
X
this yields
limsupfnIP t C.
n- J
X-A
Furthermore:
LP-SPACES 249
If - ffIIP 4 IfIP)11P + (
LA ffIP)1/P
( I
1 X-A
+ suplf - fnl.meas(A)1/P,
A
whence
JX-A IfPI4E
implies that
limsup1 IfnIP 6 e,
n'°° X-A
meas(Ek) -* 0,
(1)
JEk
If IP > a.
nk
250 CHAPTER 6: THE
nil IRe(fn)I = 1,
11lX
limmJ I1-I.fnlI=0,
X
then
n IIm(fn)I = 0.
JX
IfnIi
Show that
lim
n JX
Ii - fnl = 0.
LP-SPACES 251
en = JXI1 - IffII
Now,
and consequently
JXIVn1 4 1 + en.
= l YmssupJx IV n S 1.
I
Assume that 0 < L 6 1 and choose a A such that 0 < A < £C. By
considering a subsequence, one may assume that
JXIvnI > A
for all n. Now let a,3 > 0 be numbers which we will specify
how to choose later. Set
Bn = (IvnI 1 $).
252 CHAPTER 6: THE
Then
lira meas(An) = 0.
n-
Furthermore
X Ivn' J (1
B
+
X-B
+ n
n n
1 - A + E
n
meas(B) <
1 - B
Hence
6 + (1 - 6)meas(Dn ) + c n .
The inequalities
imply
lima lu = 1. (2)
n Xn I
With E = ±1,
ix(1
+ ElfnIj)2 = 1 + 2EJXIfn1 + JXIfnI,
(3)
X
n-oJXIvnI = 0. (4)
rl-imm1X I 1 - =fn
0. I
(d): Deduce from the preceding that if 0 < r < p < s <
then
SOLUTION: (a): If r < p < s and r e If, s e If, when s < m one has
fP < fs on A and fP < fr on x - A, which proves that fP is inte-
grable, and therefore that p e If. If s = -, fP << fr still holds
on X - A and furthermore fP < IIfIIp; since f is integrable we
have meas(A) < W and it again follows that f is integrable.
Assume that X = [0,W[. If f(x) = e-X, then If = ]0,W]. If
and
then:
f(1-A)p
If r,s are finite, it suffices to make h = fAp g = in
Holder's Inequality
c ( J
JX X JX
IIf(r/A)-rllWI ,
JX fr =
llfllp <
IlfurAllfIIP(1-A)
JAf p
-I JX-Af p ' JX-Afr
whence
lipnfllfllp Y,
limiflIfllp IIfiL.
On the other hand, if p > r, with r e if, when II f 1l m < W one has
limspllfllp 4 IIfil
f1r, g = f(1-a)s
It suffices to make h = in Holder's Inequality:
LrnLS C Lp.
limjjfjjP = exp(JXlogIfl ),
a(l ollflip) = J (f )
From this it follows that there exist three constants al,a2,b such
that:
Consequently:
O(x) = aloge + b.
(
= a1 logf + b = (alogf + b)
X 1
X
Ilfll
1f 11
IIfII,y = sup < °°
First, let us show that L provided with the norm f > IIf11 is
complete. For this, consider a sequence (fn) of elements of L
such that:
E Ilfnli <
n
262 CHAPTER 6: THE
For p > 1,
g(x) = E fn(x)
n
Ilf1IP
Ilfll ' = sup
p>.1 *'p ¢ Ilfi
the norms 11.11, and 11.11- would be equivalent by virtue of a
theorem of Banach; that is to say, there would exist a constant
M such that for every f e L
f=AIL -P
[O,A 0l
one has
and:
EXERCISE 6.120: (a) : Let p . 1 and f e LP(R). For all x e]R set
F(x) = I
f(t)dt.
0
lim f(x) = 0.
IXI-
SOLUTION: (a): Let e > 0. There exists d > 0 such that meas(E)
< d implies
jE
(1 FP)l/P ` (1
fP)1/P
p 1
X X
(
I Fn - p1(n0 AP-lmeas(F
n > a)da
p(nAp-2da1
4< f(x)dx
J 0 (F>A)
= pJ f(x)dxl
X
(min(n,F(x))
0
aP-2da
=1 X
f(x)Fn(x)p-ldx.
(J FP)1/P 6 1 (J fP)1/P,
X n p - 1 X
The desired result is now obtained by making n -> and using the
Monotonic Convergence Theorem.
(i) : f e Lp;
(ii): There exists a constant M such that for every sequence
P1,...,Pn of disjoint rectangles that are of non-zero
measure
n
1 meas(P.)p_111PfIP E M.
(*)
i=1 JP.
meas(P.)p-11JP.fIp
< JP.IfIp < IIfIIP (1)
i 1 .=i
JfI = Ii ciJ fl
P.
i
Comparing (1) and (2) yields the last part of the problem.
1 1 - t t 1 1- t t
0< t c 1,
9
P P0 + P1 ' q q0 + q1
LP-SPACES 267
U i8
kIl
f= E rke ,
k=1 Ak
v ip
g= I PRe nB ,
R=1
a(z)/a l0kb
fz = rk e Ak,
P(1-B(z))/(1-B) 1Tp
gz k
e IlBR
F(z) = RmT(fz)gz.
AzAz
0 1
2
By considering for e > 0 the functions G (z) = eEz F(z) to which
e
268 CHAPTER 6: THE
rka (z)/ap(9
1-S(z))/(1-S)ekk
F(z) = T(IlAk),
Al zAz
kLk
0 1 ' 1
Ilflyllpo = Ilfl+iyllpl =
rk/ameas(Ak) = IIfAIP = 1,
k
that is to say:
LP-SPACES 269
IF(iy)I A0IIT(fiy)IIg0llgiyllgo 1
2
eez
If e > 0 and G(z) = F(z), if 0 <, x <, 1, y e ]R, then
ee(x2-y2)IF(z)I e6(1-y2)M,
IG(x + iy)I = <
IG(iy)I : eE,
IF(t)I = 1 JT(f)gl : 1.
AltA
0
Al-tAt
IIT(f)IIq <,
IIT(f)IIq , A1-tA4IIfIIP
IITfIIq , AO-tAlIIfIIW.
F(z) =
zAzJRmT(fz)g
Al
0 1
is again entire and bounded on the strip 0 <, Re(z) <, 1, and
Tg(f) = fig.
If 1 + -r = 1 then
q q
IIT9(f)II , I191IgIIfIIgi
Then if
r=q
1 t
P 1-4t +t,
1
O,t 1, (3)
we have
f - f almost everywhere,
n
IIf-fnllpi0
As the space Lr is complete one deduces from (4) the existence
of a function h e Lr such that f *g -; h in Lr. Also
Ilhllr ` IIfIIpII9IIq.
272 CHAPTER 6: THE
From this it follows that f and g are convolvable and that f*g < h
almost everywhere, whence
if lf(x)l a A,
fA = f - fa
Furthermore, let 1 , p0 < p1 , w and E be a vector subspace of
Lp0(X)C Lp1(X) such that f e E implies that f' e E for A > 0. Fin-
ally, let T be a mapping of E into the set of measurable functions
on Y such that
for any f e E, g e E.
LP-SPACES 273
(a): Assume that p1 < m and that there exist numbers AO > 0,
Al > 0 such that for all f e E and all A > 0
AiIIfIIP Pi
meas(ITfI > A) < 1 , i = 0,1. (2)
Show that for all p,p0 < p < p1, there exists a constant Ap,
that depends only upon pO9p19p,A09A1, such that
f(x) if I f(x)I A,
fi(x) _
0 otherwise,
then fA e E. Set f = f
Show that if 1 = p0 < p1 t m and (2) is satisfied (and (2') if
p1 = co), then if C < p < 1 there exists a constant Ap such that
(d): The hypotheses are as for Part (c), and assume further
than meas(X) < oo.
VT
(Use the decomposition f = + , then evaluate the inte-
grals:
1 rm
meas(ITfI > A)da, meas(ITfI > A)da.
f0 1
f
u 6 M(1 + ulog+u)
and consequently:
LP-SPACES 275
Furthermore, by (2)
2A0IIfAIIP PO
O
meas(IT? I > a ,
P1 (7)
111faIIP
1
meas(ITfXI > A)
IITfIIP =
0
(W 1
+ p(2A1)Pll 1P P1 daj IfA(x)IPIdx. (8)
0 X
and that If"(x)I > t > 0 if and only if If(x)I > A, and
If(x)I(1-AIf(x)I-1)>t,
or in other words
(III>t)_(IfI>t+A).
276 CHAPTER 6: THE
Thus
meas(IfA I > t)
- (p(t) if0<t<A,
(9)
0 ift'> A.
i p- 1 M P-1
OJ 0
IITfIIP 5 PPO(2A0) dAJ t 0 q(t + A)dt
JO o
p1 W p-pl-1 A p1-1
+ ppl(2A dAJ t p(t)dt. (10)
J A
0 0
+ A)dt 5
1 t 1
= J WtPO q,(t)dt Ap PO dA
0 Jo
= 1 JtP_1(t)dt
p-p00
1
=p(p-P 0 ) IIfIIP.
AP-pl-1dA(otPl-1q(t)dt pl-1,(t)dtJtAp-pl-idA
J
= =
J- t
0
LP-SPACES 277
1 tP-1cp(t)dt
pl pJO
p0 Pi
P0(2A 0)
IITfIIP . I P - P0
+ pp,lp ) IIflIP.
1
A/2A
meas(ITfI > A) -,-c meas(ITf mI > -)L) + meas(ITfa/2AWI
> #L)'
Now the set (ITf1AAml > A/2) has measure zero, since almost every-
IITfIIP : 'A)dX
0
p0p(2A0)poJXp-pO-1 dXJ0tP0
= O l ( P
+ 2Am)dt
0 o Ilt
(Contd)
278 CHAPTER 6: THE
p0 p-p0-1 p0- 1
(Contd) , pop(2A0) 1 A dxJW t Q(t)dt
o A/2AW
_
p po 0
- p0(2A0)P0(2Am)P-P0
HA P.
P p0 P
P
meas(ITfI > A) S lr ITfI
xPl ITfI>A
IITfIIP1
- S P;
P1 Pl APl
a 1
J0PAP1AdA meas(Y)J padA = ameas(Y). (13)
0
2A0I
2A
1 Il a II Pi
II 1 Pi
whence:
pAP-1iP(A)da 2pA0JaXp-2dXjIf(x)I>aIf(x)Idx
ja
a
Pi °° P-P1 1 P1
+ p(2A1) 1 A
dA J If(x) I dx
a
1
In the last integral If(x)Ipi 4 API- If(x)I, and consequently
JpAPhlp(A)dX ap-2da
< 2pA0 J If(x)Idxj
a X a<A<If(x)I
AP-2da
+ p(2A1)PhIIf II1 F
a
2pA
1
If(x)I(aP-1
_ If(x)IP-1)+dx
pj X
P
1
p[2A0 + (2AI) 1] (14)
j PaP ,y (a)da : 1-p IIfII pi
a
pl
(
IITfIIp <, {meascY) + - 1 l) }HAP.
p
SOLUTION: (d): On one hand
1
J *(A)da : meas(Y), (15)
0
[2AiII./rIIJP1
+
A A
so
2AOJW A I f(x)Idx
1 If(x)I>,T
If(x)Ipldx.
da
+ (2A )P1
l
1 xp1fIf(x)I5y1-x-
(pi +1)/2
+ (2A 1 X- dA
)p1
2(2A
_o x
If(x)Ilog+I f(x)I2dx +
p1
1 1 11f1I1. (16)
j
LP-SPACES
whence:
Pi
2M(2A1)
B = meas(Y) + P - 1 meas(X),
1
2M(2A1)p1
C=4Ao+ pi - 1
J1 i f x < e,
u =
logx if x >, e,
J 2
Go,
<
Jf2u(f) < u(g)
l imJ u (9n) = 0 .
n
Show that
limJ fgn = 0.
n
lnmJ fnu(fn) = 0.
Show that
limJfng = 0.
n
2
xy <, 2x2u(x) + u(y) (1)
y <`
2
"
e logy `e
< 2 x mi x,
2
( (( 9
Jfgn 5 2+2u(f) + rju(g
n
so
limsupJfgn 2rJf2u(f),
and on making r 0
lim1fgn = 0.
n
( 2
Jfg 2r1fnu(fn) + rl 9 1
1u(r g)
so
2
limsup jf g 1
g (2)
n n r u(r-1g)
If r > 1
2 22
((
X22 logg
= Jgre u(g) + Jg>re u(g) logg - logr
1u( U-`-g
5 log(re)
and consequently:
284 CHAPTER 6: THE LP-SPACES
2
lim 1g
r-).. rJ = 0,
u (r-lg )
limJf g = 0.
n
CHAPTER 7
The Space L2
Show that if
nc3e>,C(1-c)
n-1
there exist two indices i,j (i < j) such that
meas(AiflAi ) 3 c2 - E.
285
286 CHAPTER 7
and setting
n
A = meas(Ai)
i=1
yields
A2 < A + 2 E meas(Ain
i<j
If one had
meas(Aif)Ai ) < c2 - e
2 2
n c - no < A2 - A < 2 .n(n 2 1) (c2 - e)
so
< c(1 - c)
n-1 '
contrary to the hypothesis.
00A=Vt0=MA=VAV-MA
THE SPACE L2 287
SOLUTION: Let
t
u(t) = J lf'(8)lde.
0
Now,
Consequently,
Furthermore,
alf'(t)Idt2
u2(a) = IJo
t J
IJ0f'(s)d8I = J0If'(s)Ids,
Then
lan12 lanfnl2 +
lanfnl2 fnl2 s (Contd)
= J = fA
X JX_A'
289
.< en + M21an126,
2
2 En
lanl 2
1-Md
L Ib f (x)I2 < 1
n n
n>n0
1=JAIfnl2+JX-AIfn 12
so
12J
1 3 meas(A) >, lb
X
n>,n0 n A Ifn 12 % (1 - M26) n,n0
E lbn1i2,
which is absurd.
290 CHAPTER 7:
G(f1,...,fn) = detll(.fil.fj)II1,<i,j,<n'
Prove that:
G(f1,...,fn) = n!JUnIdetll.fi(xj)II1<i,j.<n12dx1dx2...dxn.
SOLUTION: We have
n
Idetllfi(xj)IIl2 = £(To) f fT(i)(xi)fo(i ),
T,a i=1
n n
n L £(Ta)IT
aT(i) , a(i)
e(Ta 2 ) II aT6(1)
X a(i),
T,a i=1 T,a i=1
n
C C T1
= nt G G e(T) a
a T i=1
= det llaij iI
THE SPACE L2 291
A2 JJ 21K(x,y)I2dxdy <
=
K"(x,y) = JK'(x,z)K(zY)dz.
IK(x,y)I2dy)i
X * (
x y ILE(x)Tf(x) = J1EK(x.Y)f(Y)dy
T'Tf(x) = JX'(S.z)dzJK(z.v)f(Y)dY
= Jf(Y)dYJX'(xz)K(zY)dz.
and
= JIK(x,2)I2dsdzJIX(z.Y)l2dzdY.
21xex-tf(t)dt.
Wf(x) = f(x) + (*)
0
Also let
x
e-x + 21 ex-2tdt
0
X
2J0f(x)dxlex-t7htldt
(U*fih) = (fIUh) =
0 0
2J1 TFtTdtf
0 t
Hence
211et-xf(t)dt.
U*f(x) =
X
1
(flkl) = J e-tf(t)dt = 0:
0
41et-xdtJtet-uf(u)du
U*Uf(x) =
Jx 0
e- (u+x4
X Ixe f(u)du
0
x J2tdt
x
1e-(u+x)f(u)duJ1e2tdt
+ 41
x u
1
2e2-x
= a uf(u)du
JO
(1
2xex-uf(u)du eu-xf(u)du
- - 2J
J0 x
= - Uf(x) - U*f(x).
W*Wf = (1 + U + U* + U*U) f = f .
and
4rxex-tdt
UU*g(x) = J0
X rleu-tg(u)du
t
= 4 rxex+ug(u)du
JO
Te-2tdt
l rx
x+u
+4 e g(u)du a-2tdt
x J o
(1 x 1
2ex-ug(u)du eu-xg(u)du
2exI eug(u)du - - 2j
in Jr
= - Ug(x) - U*g(x).
then
and consequently
f + ail = 0
because
Ilf112= IIWf112=0.
(gk(P2) (gIT2)
g = 9 (p 2 + (2I2)
p2.
O2
Setting
(g1ro2 (g1W2)
f = W (g - T(p 2 ,p 2 W2} + W1,
'p2 (p2
so that
Now,
2 2
e e 2 1
(c) llcpl) - 2 , ('21 2) -
2e
whence
f=W*g-2(glcp2)cpl.
2Jxet-xg(t)dt,
f(x) = g(x) -
0
Jx2ffxI2
xlf(x)I2 F Lt2t)l2dt)Lh1(t)I2dt)
a2 = dt, a2 = JWIf'(t)I2dt.
x x
298 CHAPTER 7:
x+t
f(x + t) - f(x) = f'(u)du,
fx
a>x/(f0-B1)
Choose the value of V that maximizes the right side of this in-
equality, i.e. i = f0/2S. This yields
lim xIf(x)I2 = 0.
IxI 4M
tW +
If(x)I2dx = - x(f'(x)f(x) + f(x)f'(x))dx,
+W J+w
JIf(x)12dx . 2 xIfI I f' I
w
2(J+
x2If12)'(J+WIf'I2)'.
co
N
KN(x,y) = G fn(x)fn(7,
n=1
LN(x) = fX IKN(x,y)Idy.
(a): Assume that all the functions fn are real and that
there exists a constant A such that ILN(x)I < A for every integer
N>, 1 and all xeX.
Show that if (an)n31 is a sequence of complex numbers such
that I Ian12 < -, and that if
N
C
sN(x) = G anfn(x),
n=1
supIsN(x)I < m.
N
(Assume first that the an's are real and introduce a measurable
mapping v of X into {l,2,...,N} such that
300 CHAPTER 7:
sN(x) = maxs(x)
n = s()(x),
V
ncN
sN(x)dx
1 X
is bounded by a constant).
Show that if
.f L anfn
n=1
W
f(x) = I anfn(x).
n=1
N
lim G (flfn)fn(x) = f(x)
N n=1
uniformly for x e X.
THE SPACE L2 301
m/2 2k + 1 2k + 2
fn(x) 2 if 2m+1 x < 2m+1
0 otherwise
0=x0<xl<...
n <xn+1
n
sgn(sin2n+1%x),
rn(x) = n 0.
1
J (x)dx = 0.
0 p
W
f(x) = E anrn(x).
n=0
where
1
exp(ulf(x)l2)dx <
f0
(d): Show that the functions 1 and r r r (0 6 n 1 <
nl n2 np
n2 < < n ) form an orthonormal basis of L2 (X) (the WALSCH
p
BASIS).
SOLUTION (I):(a): One can clearly reduce to the case where the
an's are real. Let f= E anfn in L2(X). Then an = (flfn), so
aN(x) = JXxN(XIY)f(Y)dY.
For 0 4 n 4 N let
An = {x:a (x) < a(x) if 0 t p < n and s(x) 4 s(x)
if n 4 p 4 N}.
8h(x) = maxs(x)
n = sv(x)(a,)
n6N
= JXxv(x)(x,p)f(p)dy.
304 CHAPTER 7:
Note that
N
C
Kv(x)(x,y)
= G IlA (x)Kn(x,y),
n=0 n
JxsNz"(x)dx = fxf(Y)dyfxKV(x)(x,y)dx
{J dy(1 xv(x)(x,y)dx)2}Z.
x x
Now,
and furthermore,
v( a) v(R) (
I fr(a)fs(R)J fr(y)fs(y)dy.
v(a)(a'y)Kv(B)(S'y)dy =
X r=0 s=0 X
+ ffv (s)<v(a)Kv(S)(a,$)dad6.
The first integral of the right side is bounded in modulus by
SOLUTION (I):(b): Here again it may be assumed that the an's are
real. Let n1 < n2 < be a sequence of integers such that
2
a- 4 k-4
n>nk
2 2
w a = G a2 + k2 a2
n=0 n n n<nl n k=1 nk4n<nk+l
n
CM
a2 +
E
k-2 < °°
n<n1 n k=1
N
M = supl I wnanfn(x)I < (4)
N n=0
18N(X) - sN (x)I = I
E wnlwnanfn(x)I
k Nk<n4N
-1
b 2M w
X Nk'
N 1
3N(f;x) = E (flfn)fn(x) = JKN(x,Y)f(Y)dY.
n=1 0
sup{ 1 JKN(x,y)f(y)dyI
1 : feC(X), 11f11_ < 1}
0
= fIXNXYIdY = LN(x).
0
(This flows from exercise 6.108 on observing that one can replace
the space of step functions by the space of continuous functions).
Assume that
THE SPACE L2 307
supLN(x)
N,x
LNk(xk) = IIBNk(.;xk)II -, °°
SUP I (f;xk) I = °D
SNk
f
f
1 2
0n =
1, n30.
1 1
J0
fnl fn2
f0
fn2 = 0.
ni+Ci
h
j=i
1
J (f - h)$n = 0, 1 5 i 4 n+ 1,
0
n,n
and = 0 if i 4 j, while
n
A.
i f(x)dx.
xi - xi-1 xi+1
N
sN(x) = E (flfn)fn(W)
n=0
2-m},
em = sup{If(x) - f(y)I:Ix - yI .<
THE SPACE Lz 309
then
N
rx .
If (x) - sN(x)I 1 N (f(x) - f(y))dy
= I N N
S .
x - x i-1 x i-1
f(x) = E (fIfn)fn(x)
n=0
uniformly in x.
As the continuous functions are dense in L2(X), and as every
continuous function is the uniform limit, and hence also the lim-
it in L2(X), of its orthogonal projections onto Vn, it follows
that the fn form an orthonormal basis of L2(X). Finally, by
parts (I)(c) and (I)(b) formula (1) holds almost everywhere when-
ever f e L2(X)
n
f- I (f l fi )fi
Z=0
n n+1
(flfi)fi = ai(f)o'
2= 0 2= 1
with
n
Mn
A.(f) _ jxn f(x)dx
_n
i
310 CHAPTER 7:
1
still holds for f e L W. This and Lebesgue's theorem on differ-
entiation show that in this case (1) is still true almost every-
where.
--n1
1 n1 2
Ornlrn2...rn = 2 rnlrn2...rn
P
J
0 p
-(n1+1) -n1
2n1 l112
r_ r
= J2_(:l+l)rfl2.
0 2 np np
2
-n -(n +1) n -n -1 -n
1 2 1
Since rn rn has period 2 2, and since 2 = 2 2 2
2p
the term in parentheses equals zero. In particular, r and r
n n2
r = 2n/2
n L f..
2n,i<2n+1
It follows that
n0ann
THE SPACE L2 311
N
lim E an n(x) = f(x).
N+°' n=0
N
anr n
C 2k (2k)! ON 00... ON
l B0! ... a ! a0 0... aN r0 rN
I n=0 = 00+ .. +0n_ 2k
f01 250
N 2k = C (2k)! 2SN
..aN
I InO an nI +SN_k 200 ,... 20 N a0
00+.. !
N k! 200 2RN
( E an)k = ! a0 ...aN
x i...
0
and that for every integer n >, 0, 2n 4 (2n)!/n! .< 2nnn, so that
00l...0N!
(2k)! 2kkk
= kk
Ti- (2a 0 !... 2BN)! a0+-+aN
2
It follows that
By the preceding part, the function under the summation sign tends
almost everywhere to f2k as N -' . Fatou's Lemma then gives
312 CHAPTER 7:
1[f 112
ilfliPplIfllw(1-A)
But by (5)
so
Since
this yields
J exp(uf2) <
0
C 2 1
(6)
an ` 2eu '
n3N
and let
h= I an r,
n g= I anrn
n<N n>N
fo
exp(2ug2
n=0
(2n 1)n lg2n
J0 n=0
)n n
(2u II9I12
(1 + rn)0n,k g(1
= ctn+1,2k' -
ant = W.
( c ) : liman = co and
G(x,xl,...,xn)
G 1n
x...,x
,
G(tal,...,tan) = detlla.
+ a. + lP
THE SPACE L2 315
Now,
if (ai - ai )(bi - bi
detIla-III 2'j
(ai + b.)
i j
IT (a -
'<j
G(t1,...,tn)
a a =
(ai + a + 1)
Equation (*) then gives for the distance from tp to the vector
Ip - 01I...Ip - anI
d=
2(p + al + 1)
-
l in * p + a1 + 1 p+ a n +1 - 0'
IP - anI
for all n,
P + a + 1 ` A
n
316 CHAPTER 7:
IP 2 c n
-
p+an+1 p+En+i .
Since:
2E 2E
p + En + ry P +
IP - all...IP - anI
(p + al + 1 ...(p + an + 1
En(an+I)=w.
Lastly, if an then for large enough
IP - anI 2p + 1
p+an+1 =1- p+an+1
2p + 1 ,L2p+1
P t an t 1 a n
Cntan-l12dt
J1IP'(t) - I < e2.
a
IP(x) - P(0) - Cnan x nI < e, 0 < x < 1.
x=z+aixi,
(zlxi) = 0, 1 4 i n.
318 CHAPTER 7:
d2 + I X;(xlxi) = (xjx).
i
R
det
ai +- 1<i,j<n (ai + bi)
det III
ai + bj 1<i,j4n
=C
n
Tf.
Z<
(ai - aj)(bi - b.)
TI.
Z,7
(ai + aj)
THE SPACE L2 319
1 1
lim lim aldet a. = det a 1
2<i, j<n'
and that
n nlcnl2.
n=o
1 2%
meas(f(u)) = rdrJ If'(reio)l2d8.
f0 0
Furthermore, if 0 E r < 1,
nrn-1c e
f'(rei8) _
n
n=1
n2r2n-2IdnI2
If'(reio)I2d8 = 2n I
J2t
0 n=1
so
CO 1
r2n-1dr
meas(f(U)) = 2x n2Icn I2fo
n=1
W
= n I nlcnl2.
n=1
Show that if
2n
If(els)lds < 1 +lall2,
2n f0
then the function f possesses at least one root in the disclzl <1.
THE SPACE L2 321
SOLUTION: If f were never zero in the disc Izi < 1 there would
exist a function g, holomorphic in this disc, such that g(0) = 1
and f = g2. Let
a
g(z) = 1 + 2 z + bnzn
n=2
27E lg(rel-5)I2d3
Ibn12r2n
1 + -ja112r2 + G _ 1
n=2 0
1 2n
2aJ
If(relNd9.
0
Since
lim f(elo
r+1-
r2n
If(ei9)IdO > 1 + 41a112.
2n
0
fn(E)
fn(z) = -rJ
1
2n2 E - z d
1E-z01=p
r2n i8
a
fn(z0 + pelf) i0 dO,
2n
0 z0+pe -z
and consequently
r 2n i8
e
fn(z) = 27r1- r J 2pdp J0 fn(z0 + pelf) dO
2 1 ri z0+pe i8 -z
1
-z 0
2(r2 - ri -z01<r2
f w C - z0 E - z
dudv,
Jrill
where E = u + iv.
The function equal to ( - z0)1k - z01-1(g - z)-1 if ri
1E - z01 t r2 and to zero otherwise belongs to L2(U), and thus
lmfn(z) = 277r 1-
ri 11r1 1-z01<r2f
() E - z0 -z dudv.
2
-z0
-z0 g -z1 6ri-r' 1
THE SPACE L2 323
the function
g(z) = limfn(z)
n
325
326 CHAPTER 8: CONVOLUTION PRODUCTS
limIIf - 0,
t4 0 t
then there exists a function g that is uniformly continuous on 3R
such that f = g almost everywhere.
AND FOURIER TRANSFORMS 327
j(Pn
= 1, and q) n(x) = 0 when IxI a en (with cn + 0), then
= J gn(y)dyflfy(x) - f(x)IIVi(x)Idx
It follows from (*) that 9tif - f in L'. But the 9 ief are uni-
formly continuous, and since for such functions the "sup" and
the "ess sup" coincide, the (p n*f form a Cauchy sequence in UC,
and consequently Tn*f + g in UC . Clearly f = g almost every-
where.
p e F such that (*) holds (and consequently Jgg = 0 for all g e G).
V = wou.
P J' Jf P
p eF:
ff9 =
= (g,v(9)) = Jg9.
9 µ (X. =
J9)11.
This mapping is surjective, otherwise there would exist complex
numbers a1,...,an, not all zero, such that for all g e D"(I)
0= aiJgi9 = f (I aigi)cp.
1g.. = ij, 1 c i c n.
Then let
C. = Ifc., 16,j c n,
Ji(P = Jg,
which will prove the property. Note that if
Jg%p0 0,
=
and consequently
Jf0 0.
Jf = CJg.p Jg.
=
1 t)a-lf(t)dt
fa(x) = j (x -
r a
0
(a x ra lf(t)IdtJra
t)a-1dx
dx1 (x - t)(%-'If(t)Jdt = I (x -
0 0 0 t
= J0t)a - dt <
t)"-llf(t)Idt
Jx(x - <
0
SOLUTION (b):
x r t
t)0-1dtJ u)m-lf(u)du
(fa)s(x) = P (x - (t -
a P S J
0 0
)S-1(t u)a-1dt.
= r a 1r S j0f(u)du JUx(x - t -
332 CHAPTER 8: CONVOLUTIONS PRODUCTS
fa+s(x).
a-1
ga(x) = r a , 0 < x .< a,
fa(x) = 9a*fl(x).
nn[-1/2n,1/2n]'
gn
and
nx
1/2n sin
fn(x)
=
Vn-1_
1/2n
e-2nixydy
= 1
Vn
nx n
n
111
and it is clear that ?fn (x) -> 1 uniformly on every compact set
of m. When p > 1 it suffices to consider the functions
fn(x1)fn(x2)...fn(xp)'
REMARK: To determine fn one may draw some insight from the pro-
nrt- t[x
n
2 = 12
sin ax = 1.
n
fx
K
+ h) - f(x)Idx = o(IhI).
fn = f*'Pn' If x e V,
then
I fn(x + h) - fn(x) I
Vk is a connected component of n ,
k
n k C Vk+l'
It is clear that the Vk satisfy the required conditions.
336 CHAPTER 8: CONVOLUTION PRODUCTS
b
lf(x + h) - f(x)ldx = 0 lhl as h -r 0.
a
b (b+h (a+h
J lf(x + h) - f(x)ldx = J f(x)dx - f(x)dx < 2Mh,
a b a
where M denotes the upper bound of lfl on [a,b + 1]. The case
where h is negative is treated analogously. As every function
with bounded variation is a linear combination of increasing
functions, the implication (i) => (ii) is demonstrated.
In order to prove (ii) => (i) one can clearly assume that f
is real. Let [a,b] C V. There exists 6 > 0 such that
[a - 26,b + 26] C V.
b+6
J _ lf(x + h) - f(x)ldx < Mlhl if lhl < 6.
a 6
frblfi(x + h)
- fi(x)I
JI dx
a
+fb dxfd
+f-d
6 b+a
I,.(y)ldyfa-d lf(x + h) - f(x)ldx
<
JJ
< M.
b
V(fi;a,b) _ Ifi(x)Idx < M.
a
n-1
I Ifi(xk+l) - fi(xk)l '< M.
k=O
n--1
f(xk)1 < M.
kI0 If(xk+1) -
n-1
P(A) = E (f(xk+l) - f(xk)+,
k=0
n-1
N(A) = E (f(xk+l) - f(xk)_
k=0
By the preceding,
P(x) = supP(A),
A<x
N(x) = supN(A).
A<x
The numbers P(x) and N(x) are finite and increase with x. When
x 4E we need only consider the sequences A for which xn = x. In
this case
so
+m 1 +-
I
f(2nn) = I f(n). (POISSON'S FORMULA)
2n
n= n=-w
+-
f4(x) = E f(x + 2nn)
n=-w
SOLUTION: Let
V(x) = V(f;-=,x),
limV(x)=V<W.
x-).-
2n tm
4
f (x)dx = j_ f(x)dx. (1)
f
340 CHAPTER 8: CONVOLUTION PRODUCTS
then
and furthermore
k-1
If"(xi+l) - fa(xi)I
i=0
k-1 +-
= I I f(xi+l + 2xn) - f(xi + 2nn)
I
i=0 n=
k-1 +-
I I If(xi+1 + 2xn) - f(xi + 21En)I
i=0 n=--
4 V.
21E
+e-inxf(x)dx
e-inxfl(x)dx = = 2n ?(n). (2)
1 I
0
+m +W 2n
f(2un) = fa(0) = J0e_1f'(m)
X G
n-m L =
n=--
(Y
f(y) - f(x) = J g(t)dt, a < x < y < b;
x
DW(I):
(ii) : For every function q) e
rb b
f(x)4)'(x)dx + J g(x)cp(x)dx = 0.
J
a a
Next, show that if the pair (f,gl) has the same properties,
then g = gl almost everywhere. We then make the convention g = Df.
If g is continuous it will be noted that Df = f'.
(b): Let
H2(I) = {f:feL2(I),DfeL2(I)}.
(b
(fig) = (f(x)g Tx7 + Df(x).Dg x )dx
J a
b b
f(c) = f(x)* c(x)dx +f e H2(I).
JaC
Determine *c explicitly.
Show that D°°(I) is dense in H2(I), and that for a < c < b
there exists 8c a H2(I) such that
b
f(c) = 1 Df(x).DBc(x)dx, f e H2(I).
a
J2{faL+ ax.
=o, i=1,2.
SOLUTION (a): If f,g satisfy (i) and p e DP(I), let a,$ be such
that a < a < 8 < b and p(x) = 0 if x4 [a,8]. Then
fb rx 8 s
g(x)(p(x)dx = Jg(x)dxl W'(t)dt = q,'(t)dtg(x)dx
a a a fa ft
B B
= f(S)Irq,'(t)dt - I f(t)cp'(t)dt
a JJJa
b
f(t)p'(t)dt.
1
a
rb rb
I fl9'+J gro=0, 9 e D-(-T).
f a a
rb
I (f - f1)p' = 0, 'p e DP(I).
JJJa
When 9 runs over 9' runs over the set of functions *e D-(I)
b
(g - g1)T = 0, 9 e D(I),
a
2
remains to prove that H (I) is complete in the norm associated
2
with this scalar product. If (fi) is a Cauchy sequence in H (I),
(f ) and (Df ) are Cauchy sequences in L2(I). Hence there exist
f,g e L2(I) such that fi - f, Dfi - g in L2(I). If p e DP(I), then
b b
Jafi.cp' + Dfi.W = 0,
fa
b rb
fp' + gcp = 0.
J 1
a a
+0 r+0 +W
JP?(Y)TTdY
_ = 1 _ Df(x)Zxsdx =- Jf(x)c'(x)dx
J](Y)iy(y)d J = J00Y(Y)EYYdY.
_
If (y ) - f(x) I 4 A.
If f(x) were not to tend to zero as x -> there would exist E > 0
and a sequence of points xn such that
2
E
xn+l > xII
+ 2 , If(xn)I 3 E.
4A
W x +c 2 4A
2
E
2
E
2
J If(x)I2dx > I j n If(x)I2dx = W.
a n xn n 4 4A2
Second Proof: We can reduce to the case where I =]R by the fol-
artifice: if a > -m, f is extended by continuity at a (cf. (i)
a
0(t)dt = f(a).
fa
(c
f(c) = f(x) + 1 Df(t)dt, a 4 x < b,
x
so
or
12
I f(c) , 2(b - a) 112
+
b-a
112
L2(I)'
which proves our assertion.
The existence and uniqueness of i results from this. In par-
c
ticular, if W e V (I),
c b b
cp'(x)dx = J(p(x)*c(x)dx +
a a a
b b
Jaq,'(x){Dhc - Xc(x)}dx + JaV(x)*c(x)dx = 0.
(1) is continuous;
c
and consequently,
0 = aJ(Y - x)*c(x)dx - j
Y
c(Y) - c(a) = Ja - x)*(x)dx.
Y
IVC(Y) - c(a)I < MJ (Y - x)dx < 'M(Y - a)2,
a
(this is in the case where a < c < b). From (2) and (4) one ob-
tains
Acosh(c - a) - Bcosh(b - c) = 0,
Asinh(c - a) + Bsinh(b - c) = 1.
cosh(b - c)cosh(x - a)
sinh(b - a)
a 4 x 4 c,
c (x) _
cosh(c - a)cosh(b - x)
sinh(b - a
c<x4b.
This expression for *c is still valid when c = a or c = b. (If
To prove that
VI(I) is dense in H2 (I) it suffices to prove that every f e H2(I)
orthogonal to r(I) is a linear combination of Pa and 'b. Now,
the condition:
(b rb
1 f(x)p(x)dx + I Df(x)p'(x)dx = 0, T e 'O-(I),
a a
b-c
b
Xc(x) - b l a JXc(t)dt =
ba a4x4c,
a c - a
and consequently
(b - c)(x - a) a 4x4c ,
- a)
(c - a)(b - x)
c..<x .<b ,
-a
b
f -* (If(a)1 2 +J
a
Then
350 CHAPTER 8: CONVOLUTION PRODUCTS
r'E
1
I f(x)I2dx = 2n r(loglog1 )2dr <
xl<z o
Ix-L
of
axi
f'(r) '( If'(r)I = 1
r logy'
whence
2
a
(x)I dx
Jlxl ax.
(
dr
2t grl2 (BERTRAND'S INTEGRAL!)
fn
O rllo
then
dw = l ax cp + f x1 dx1Adx23
1 1J
JO<CIXkR of
(P + f x1
11p'= J f -J xI='
fw
2L (P
fa )ax
f la +
1 (2n
- eloglog1 cp(ecos ,tsine)cosed8.
0
AND FOURIER TRANSFORMS
I 2
1 (P + f 11c = o.
L1,
ab < alog(1 + a) + eb - 1,
G(x) = logl 1
cosx '
352 CHAPTER 8: CONVOLUTION PRODUCTS
f*G = hn + fn*G.
Consequently, if A > 1
Now
so
G(x) ti log
Ix -ix01
AND FOURIER TRANSFORMS 353
and consequently
eA-1G(x)
Ix 0I-1/A
ti - x
IIf*G - hnil.
1
G
< AlogAJlfnl + Xflfnllog(l + Ifnl) + J(ea_ - 1).
-1
limsupllf.G - hnll 4 J(eXG
n--
Making A } W shows that hn -> f*G uniformly, and therefore that
f*G is continuous.
EXERCISE 8.149: Prove that the algebra L1(Rn) does not have a
unit element.
JIf(x)Idx -1 1.
a contradiction.
+M
T
a
= I an T
nn (1)
n=-m
AND FOURIER TRANSFORMS 355
TOT0 = 0, a 4 0, S 4 0. (3)
If m e L1, then
(TQ*Ta9) = (ToT$)((p*c) = 0.
then:
uv = 0, (4)
(n n
en = 1,R1 _ e -ice l sinx ldx = nJ cosnxsinxdx.
n 0
2 1
c2n+1
=
0,
C2n-n
But then
+co
(eix - e-ix) + cneinx
(sinx)+
= 4 n=--
(5)
+m
(e ix - e-ix) - C cneinx
- (sinx)_ = 42 z
n=G-w
1
a1
= s1 =1,
a-1 = 0-1 4 ,
1 1
a2n - s2n n
1 - 4n2
a2n+1 = 82n+1 = 0
if n + 0 and -1,
satisfy (3). One can, if one wants, obtain real sequences on re-
placing x by x + 1/2 in (5), which gives
a1=S1=a-1=S-1,
1 ( - 1)n if n4 0 and -1
a2 n - - stn n
1 - 4n2 a2n+1 = 02n+1 = 0
AND FOURIER TRANSFORMS 357
(a): Show that if f,g a L+, then they are convolvable, and
fiegeL+.
a enxf(a
- x)dxI2 4 If(a - u)f(a - v)Idudv.
a JJ
u>-a
v>-a
u+v>O
f*gl + f1*g = 0,
SOLUTION (a): Let M > 0 and let fM(x) = f(x) if x < M, fM(x) = 0
if x > M, gM being define analogously.
Then if x < M,
tom x
-WIf(y)g(x - y)Idy = j If(y)g(x - y)Idy = (Contd)
J 0
358 CHAPTER 8: CONVOLUTION PRODUCTS
(Contd) = Jx fM(y)gM(x - y) dy
0
= f fM(y)gM(x - y) dy.
x
(f*g)(x) = f(y)g(x - y)dy.
0
(frcg)(x) = (fM*gM)(x),
SOLUTION (b):
a
fJ en(u+v)f(a
( enxf(a - x)dx)2 = - u)f(a - v)dudv
-a (ul'a
IV1<a
= n(u+v)
JJ e f(a - u)f(a - v)dudv
u>,-a
v>,-a
u+v50
n(u+v)
+ ff e f(a - u)f(a - v)dudv.
u.<<a
vsa
u+v>,0
AND FOURIER TRANSFORMS 359
en(2a-x)f(y)f(x
- y)dxdy
if 0sy#xs2a
r2aen(2a-x)(f*f)(x)dx
= 0.
0
a
enxf(a
- x)dx12 S - Of (a - v)Idudv. (1)
IJ-a
J
U -a
JJ3If(a
v a-a
u1VS<0
to 2a
IJ- ef(a - x)dxl c f If(x)Idx. (2)
a a
supiJenxf(a
a - x)dxl s M < (3)
n 0
REMARK: For use in the rest of the exercise, note that this shows
that f*f = 0 implies f = 0.
(x
= xff(y)g(x - y)dy = 0.
But then
(f*gl)*(f*gl) (f*gl)*(fltg)
(f*g)*(fl*gl) = 0,
f*gl = 0. (4)
f'tign = 0, n 3 0.
+m x
If(x - y)g(y)I dy = J If(x - y)g(y)Idy = 0.
0
By Fubini's Theorem
REMARK: (5) implies that for almost all x there exists a set
x
with measure zero such that
(f*rpi)ic(g*(Pi) = 0;
M
J If - f*(PiI } 0.
0
EXERCISE 8.152: Let 1 < p < - and f E LP(0,co). For all x > 0 set
F(x) = x1lf(t)dt.
0
1
(d): Show that if f e L1(0,m), f > 0, then F 4L (0,-).
I
F(x) = G(x,t)dt
J
0
By exercise 6.110,
IIFIIP E IIPdt
0
F(x) = lrf
00
x 0
F(x)p-1
IIFIIP "
(J-F(x)(p-1) 1dx)1/q, where + = 1.
p p l Ilfii
0
p q
That is
p
IIFIIP , 1 Ilfllp. (2)
p
IF(x)I , xjxlf(t)ldt,
0
so
Fi(x) = L xfj(t)dt
0
then
By Fatou's Lemma,
i
such that f + f in Lp. Then F. -' F in LP, and so by exercises
1
6.105 FP-1 + FP-tin Lp/(p-1) = Lq. Replacing F by Fi in (1) and
passing to the limit, shows that this formula is still valid for
f. As (2) results from Holder's Inequality applied to the right
AND FOURIER TRANSFORMS 365
FP = F(p-1)q = A/p,
Bxf'(x) = (1 - B)f(x),
that is to say:
1 - B
f(x) = Cxa, C > 0, a =
B '
G(x) = xf lf(t)Idt
0
yields
REMARK: The second proof given for part (a) above is more ele-
mentary than the first proof, which uses the generalised Minkow-
ski inequality. Further, the proof of part (b) is based upon
formula (1), which is the essential point of the second solution.
One can avoid using exercise 6.105 to establish the validity of
(1) whenever f e LP(O,m) is only assumed to be positive, by pro-
366 CHAPTER 8: CONVOLUTION PRODUCTS
(P(x) = eX/Pf(eX),
e-x/q
I if x 0,
0 ifx<0,
It is not hard to verify that
N 11 If 11
X%tc(x) = ex/PF(ex),
11XII
= (rll/r
l J1 r 0,
so that
= 4iIfti
Lp(0,co)
1 (X*c)(x) = 0,
1x1-
which implies that
xF(x)P-1F'(x)dx = - 11 F(x) dx
J 0 p0
r 3 1, + >, 1, and 8= -p t - 1,
p
r r
we have (cf. exercise 6.106)
r
--
1
s q +
1
9(x) = x1/Pf(x),
and
368 CHAPTER 8: CONVOLUTION PRODUCTS
0 if 0<x<1,
then
X*g(x) = xl/PF(x),
xlkg(x) =
J'X(t)g(xt-1)
0
tt = Ot .
0 if 0 < x 4 1,
F(x) = qx 1(x1/q - 1) if 1 4 x 4 A,
IIfIIP = (logA)1/P
Since
we have
and consequently
IIFII
asA+oo.
ITf1
P 'q
SOLUTION (d): In this case
(m x
dx
x (
J F(x)dx = f(t)dt = J f(t)dt J =
0 J Ox J 0 0 tx
1/p
IIf1ILp(t-1dt) U0lf(t)tP tt)
Also, set
L-(t-1dt) = LW(O,W).
Show that if
Jf(t)g(xt') tt
0
Q
limf*g(x) = lim(ffg)(x) = 0.
X-+0 x-
IIFIILp R) = IIf1ILp(t-1dt)
F(x)=Tf(t) sinxtdt.
to
=o(xa-1+1/p)
F(x) asx -> 0 or x
Deduce from this that
o(Ihla-1 + 1/p)
suplF(x + h) - F(x)I =
X
Let
tl -1/P -a
W(t) = tl/pf(t), X(t) = sint,
so that
- xl/p +a-1
tt .
in
x1/p+a-1('*X)(x).
F(x) =
IN 11 -1 =11w1I =Ilf1I p
Lp(t dt) Lp(t-1dt)
372 CHAPTER 8: CONVOLUTION PRODUCTS
and that
r 0<r4 m,
(t -1 dt)
UXII
L
t-1iX(t)Ir _
1
tr 1/p+a-2 +1 as t; 0 and p+a - 2 < 0.
if
8 1, p+8al, r=1+1-1,
that is to say, if
then
F(x) = 0 (2-1+1/P).
Finally,
1/qr (411/S
(f - 1I F(x) Irdx)1/r < 11f 11p,
J
0 l-I Il J
p+q= 1,
Show also that
F(x) = O(x-1/q) as x -* 0 or x -* -.
tl/qe-t,
then
x-1/q(,*f)(x).
F(x) =
Furthermore, if a > 0
IILS(t-tat) =
374 CHAPTER 8: CONVOLUTION PRODUCTS
IIf1Ip-C M;
<
JnIf
-K
;
Ilfa - fllp¢e.
J]Rfl-KSIp 4 ep, 1 4 s 4 r.
AND FOURIER TRANSFORMS 375
and then if f e H
since
p q
4 3e. (4)
This being so, let us choose fi,..,,fr e H such that (1) is satis-
fied with a = e(meas(K + A))-1/P. Writing
a
n
limsup = 1.
an+1
(ii): hIf*qnII 1;
Jf(x)dx > 0.
Ixk2p
(f*8n)(xn) I.1 -
1a.
Show that if
hn(x) = a-1 t xn
378 CHAPTER 8: CONVOLUTION PRODUCTS
( 11 an
(f*h ) (x) a I1 -
n nJJan+1
then
Jg(x)dx 3 0.
SOLUTION (a): If
a
n
limsup
n an+1
one would have, starting from some n0, an+l > )an, and conse-
n-n
quently an > an0 x 0 when n 3 n0, which contradicts the hypo-
thesis an = 0(na).
(")
0n(x) < 0n+1(x + z).
and that, denoting the volume of the ball IxI < 1 by Vp,
1 a
a 1 jf(y)e (x - y)dy >. I1 - 11 n
n n ` nJJa
an+l n+l
By (**) and part (a), for all e > 0 there exists an integer n0
such that
a
n
(1- it 0 >1-E.
n0 an0+1
1 - e .< (f*q,)(z) 4 1
Jg = fg = lim(g,f*p n) = lim(g*f,p
n} n
)
n
where the infimum is taken over all the systems of elements a1,
...,an e]R and positive numbers ai,...,A such that Al + tan = 1.
(Use the result proved in the preceding exercise).
I(f) = Jf(x)dx.
E A(a)=1.
aeatp
If A e P, set:
Al*(A2*f) (al*A2)*.f,
=
where
(A1*A2)(a) = I Aba2(a
1 - b)
bdRP
(the reader well versed in measure theory will recognise the no-
tion of the convolution product of f with discrete probability
measures). If A e P, f e L1, it is clear that
I(f) = p(f) if f 3 0,
p(af) = Ialp(f),
and the last from the fact that for all c > 0 there exist X1,a2
e P such that
so that
p(f + g) < II(xl*x2)*(f + g)111
11X1*fI11 + Ila2*9111
Now assume that one had proved that f e L1, f real, and I(f)= 0
implies p(f) = 0. Then if f e L1 is real and if cp >, 0, I((p) = 1,
one would have
which would prove (1) in this case. To pass to the general case
observe that there exists a e Q, lad = 1, such that II(f)l = I(af),
if of = u + iv, where u,v are real, then
p(v) = II(V) I = 0,
AND FOURIER TRANSFORMS 383
and consequently:
for all a e]R . Since I(f) = 0 the left side of (2) is not changed
by adding a constant to g, which allows us to assume that g > 0.
Let h(x) = g(-x); then h > 0, h e Lm, and (2) can be written
(3)
(f - M q,)*h > a - M M = 0.
I(f-M9)=I(f)-M> 0,
which is absurd.
384 CHAPTER 8: CONVOLUTION PRODUCTS
b is x
f(x)e n dx = 0
f a
b
F(z) = 1 f(x)eizxd
a
is x
lime n
n-9
exists for all x's belonging to a measurable set A of 3R with
meas(A) > 0.
Show that the sequence (an) is convergent.
is x
lime e n
n- n
AND FOURIER TRANSFORMS 385
lime = 0,
n-,m n
or
is nx
SOLUTION (a) : The set of the x eat for which lime exists is
an additive subgroup that is measurable and of strictly positive
measure. From this it follows that
is x
g(x) = lime n (*)
n-),-
exists for all x eat (cf. exercise 8.138). Then for every inte-
grable function f on at,
+W ( -+Q is x
f(x)g(x)dx = limJ f(x)e n dx
f_ n-t°° m
E 0
eiax - eiBx
is x
icne n
I
= Icn1,
we have that
IimIc I = p
n-),w Ti
is (x-y)
x,y e A, the sequence e n is therefore convergent. Because
meas(A - A) > 0 it follows from part (a) above that a = lima ex-
n-).w n
is x is x
ists. Finally, if x e A the sequences ene n and e n are con-
vergent, which proves that lime exists.
n-).w n
Ln
meas{ (f+ > A) n [-n, n] } < 11f111, (*)
n
(ii): A = sup 2nj Hn(x)dx <
-n
(+W
(iii): B = sup IxH'(x)Idx <
2n
1 n
If f e L1(T) set:
Show that
and if N is the integer such that 2n(N - 1) < h < 20, then be-
cause f has period 2n,
and consequently h < 2n. From this it follows that for A > 211f1I1,
z
(A) = meas(x:lxl < n, 1 xJ IfI > A for a z such that
z
x x < z <x+2n}
z
< meas(x:-n < x < 3n, 1 xJ IfI > A for a z such that
z
x
x < z < 3n}.
x
F(x) = J lf(u)ldu - Ax
0
where
V = {x:-n < x < 3n,F(z) > F(x) for a z such that x < z < 30
b lfI
J n
f
i
AND.FOURIER TRANSFORMS 389
so
h h
suph 11 gi(x + u)du = h-11 If(x + u)ldu,
i o 0
so
390 CHAPTER 8: CONVOLUTION PRODUCTS
supgi(x) = f(x).
2
supgi() = ?(X),
i
and therefore
Ilf lip = Sup Ilg" IIp < Apsup 1191 lip = AP Ilf IIp
i 2
((n
Writing
7c x
JE(Y)If(x - y)Idy = Hn(n)JOIf(x - y)Idy
(n u
- J H'(u)duJ If(x - y)Idy.
0n 0
Foru>.0
u
J If(x - y)Idy 6 of (x),
0
so
(JH(y)I
Similarly,
0
_x n f(x - y)Idy (iH(-n)
n
+
Y
n )du + Jb'uIc(u)du
yH (y) = fo H(u
x n
s J:Hfl(U)dU + f0juU1(u)jdu,
392 CHAPTER 8: CONVOLUTION PRODUCTS
r0 0
n n( - n) 4 I H du + I JuH'(u)ldu,
-n m _n m
One can then deduce for the operator f + K*f the same properties
as for f f".
y 1 x
P(x,y) = n 2 f Q(x,y) = f
2
(Use part (b) above). Generalise this result to the case where
f e LP(R), 1 s p<
Qf(z) = 1mQ(x
+ - t,y)f(t)dt.
(g) : Let f e L1Ot), f 3 0. Show that for all e > 0 and all
y > 0
(b
P(x,y)dx = 1 (tan-1b
- tan-1 a ). (1)
a
SOLUTION (b) : Note that P( ,y) a Lq(,t) for 1 4 q < which as-
AND FOURIER TRANSFORMS 395
ifa<x<b,
lim?f(x,y) i if x = a or x = b,
0 otherwise.
SOLUTION (c): First note that -PX(u,y) > 0 if u >. 0, and that
u2du 1
I -uP'(u,y)du = 2(
-_
2
0 X 0 (u2 + 1)2 .
= J0_uPX(u,y)IuJOIf(x ± t)Idt]du
< if-(x).
Since
it follows that
from which the desired inequality follows. This being so, let
OR-4)
(Tk) be a sequence of step functions such that II(Pk111 = and
f = I
Tk
in LThere exists a set N1 of measure zero such
that
k
Furthermore, for all z e R,
Since
OR-2)
I meas{x:suplPcpk(x + iy)I > k-2} = I <
k y>o k
for almost all x e]-N,N[, and consequently for almost all x e]R.
SOLUTION (d): For c > 0 and R > 0 let rc R be the loop formed by
the segment [-R + ic,R + ic] and the semi-circle E * is + Rein
0 < 8 t n. For all z e n, whenever 1/c and R are large enough,
g(z) = 2n1iJ
g()
- l Jr
0 = 2ni d
r c,R e,R
g(z) = r
c,R
- z- z d.
(t + ic)
g(z) _ V-
n E . t + is - z t + 2c -
dt. (5)
a subsequence, one can assume that the functions t ' g(t + ick)
converge weakly in L OR) to a function g. One can also assume
that co < y. Since
8)-1
t '* (t + iEk - Z)-1(t + ick -
y2)-1,
t y ((x - t)2 +
g(z) = Pg(z).
Note that by the preceding part of the exercise, for the sequence
under consideration,
1
z(t)
ReKf = Pf < 0,
expiQf = gexp(-Pf),
limQf(x + iy) = Hf
y40
- -in-M xsin2nlx
M
r -2nil; as M 3 m.
E e Q(x,y)dx dx
J -M
M x2+y2
f+C0
Q(E,y) = - i xsin2niax ax = - if+" xsin2nyEx d,.
_ ,Xx2 + y2 n1 x2 + 1
+W
1 cosux dx = e-u.
nJ -m x2 + 1
400 CHAPTER 8: CONVOLUTION PRODUCTS
(-E,y) = - (E,y),
yields
(E,y) isign(E)e-2t[yIEI.
Let
Y(E) = - isign(E).
and therefore
Q(-'Y?'
Hf almost everywhere,
IK(x,y)I s n
y
it is clear that
AND FOURIER TRANSFORMS 401
ReKf = Pf > 0,
(since logll + pKf(z)I > 0). For all e > 0 the function
+W
log(1 + pKf(t + ie)
YJ- dt = log(1 + pKf(z + ie)),
n
(x-t)2+g
and on taking real parts,
+W
If_C0 logll + pKf(t + ie)I
dt = logll + pKf(z + ie)I.
2 + y2
(x - t)
so that
2
+b2log+IuQf(t+iE)ldt<uy+E 11f 111.
(x-t) 2
(+m
log+lpQf(t + ie)Idt < ullflll.
IIf112,
and consequently
A) 4 a2 11/112.
AND FOURIER TRANSFORMS 403
E) Ilp , Ap Ilfllp
= sup{I(flQg(,t))I:geE,IIg II q. 1}
AgIIfIIp
IIQf(,E)IIp c Apilf1Ip
limQf(t,c) = Hf(t),
E-ro
404 CHAPTER 8
pMp-1121 - z21.
1Iz11p - Iz21p1 4
The function
405
406 CHAPTER 9:
g : In IVn
is integrable. Let
f(x) = x
Jg(t)dt.
0
For every integer N there exists a > 0 such that Iy - xj < a im-
plies y e V1 if 1 4 n<, N. Then
f(y) - f(x)
>, N,
y - x
OD
n=1 n
(b): With the In's the same as in part (a), consider the
function f such that f(xn) = cn and f(x) = 0 if x is distinct from
all the x 's.
Show that if limcn = 0 f is differentiable at no point of iR.
L c2n > p
npcn<np+1
y yx
liminfIf(y) - f(x) = 0.
l y p f(yy - X(x ) I
l
C
I n
some a > 0
-aaJf(x)dx ? a.
1
b
n = 1,2,...
`pn = AIL [0,1/n] I .
(x
f*cpn(x) = n1 f(y)dy >, a.
x-1/n
From this one concludes, as above, that f(x) = 1 for almost all
the x e [0,1].
W
f(x) = I fn(x)
n=1
BOUNDED VARIATION etc. 409
sn = f1 + ... + fn.
and consequently
fn(x) <
n=1
8,(x) - f'(x)
limm*(EnLx2 -h,x+h]
h;0
and that
BOUNDED VARIATION etc. 411
In fact, for all e > 0 there exists an open set V such that A U B
C V and
Let
V = Vn]-'.,e + c[,
V+ = vn ]c - c,+oo[.
Then
and consequently
6 m*(AUB) + 3c,
I(fn - f)
n
fn(a) - f'(x) - 0
n--i
L If(xitl - f(xi)I > A (* )
i=0
if
p-1
A = E If(yi+l) - f(yi)I A.
>
i=0
A)
If (x) - f(y)I < (A 2p
(A X)
If(yj+1) - f(yj)I -
P
n--1
+W
V(f;a,b) = J- n(t)dt,
p-C1
A(f) = L lf(xi+l) - f(xi)1,
i=0
JnA(t)dt (1)
p-1 p-1
n(t) = I v.(t), nA(t) = I pi(t),
i=0 i=0
Let D be the union of the f(A )'s and of the maxima and minima of
f on the subintervals of an arbitrary AS. In other words, if t 4
D, and if a,B are two consecutive points of a subdivision AS, then
t is distinct from f(a) and f(B) and from the maximum and minimum
of f on [a,B]. The set D is denumerable, and we are going to show
that
limn (t) = n(t) if t 4D. (5)
s S
X
F(x) = F(a) + J f(t)dt, a 4 x 4 b, (*)
a
n-1 IF(x. -
F(x i )IP
sup I _i < W, (**)
A i=O (x i+1 - xi)p
p a
F1I =
xitlfl C of/q((xi+llfiP)1/p'
x. x.
BOUNDED VARIATION etc. 417
Next assume that (**) is satisfied and let M be the value of the
left side of this relation. Let us first show that F is absolute-
ly continuous. For that, consider a finite sequence (ai,bi) of
intervals contained in [a,b] and mutually disjoint. Using Hol-
der's inequality for finite sequences this yields
IF(bi) - F(ai)Ipli/p
C
IF(bi) - F(ai) 1 a1 /q
( (bi - ai)) /q
i 1
i (bi - ai)p i
so that
nIl
InFilpn(p-1)
jblfolp = 6 M.
a i=o
bP < M,
JIfI
which, taking (1) into account, completes the proof.
p
Show, using the preceding exercise, that for every contin-
uous linear function u on LP(R) there exists g e Lq(R) such that
SOLUTION: Let
(O,xl ) if x 0,
F(x) _
{u1
- u(Il(X,01 ) if x 0,
so that
Let A > 0 and A = (-A = x0 < xi < < xn = A). Observe that if
n-1
g = e11(Xlpxl+il
i0
IIuII(n11
r Ur-1 IAF,"Iq-2
if F1 $ 0,
Aq-1
i
c
i =
0 otherwise
This yields
n-l q 1/p
= IIu II
JAF
i l
i=0 I
so
n-1 I AF i1
.
q
11
1 4 Ilullg.
i=0 Aq
i
+A
J-AIg(t)Igdt < IIu1Iq.
Let
Ug(f) = Jf(t)g(t)dt,
+
f eLY(it).
u(Il[a
sl) = ug(n[a $] ), a c a.
EXERCISE 9.173: Two functions f,g that are positive and measur-
able on ]0,1[ are called EQUIMEASURABLE FUNCTIONS if
1 1
J f(x)dx = g(x)dx.
0 J0
J f(x)dx 4 J
E To
BOUNDED VARIATION etc. 421
1 1
f(x)cp(x)dx 4 f*(x)q(x)dx.
1 0 J 0
Show that:
1 1 1
J s(3f(x))dx 4 sl I
0 jo ` x0 1
II0fIIp < (
(i): Show that for all c > 0 there exists a number A depend-
ing only upon c such that
422 CHAPTER 9:
(1 1 1
J Of(x)dx 4 2 f(x)log+f(x)dx + AJ f(x)dx + E.
0 0 0
SOLUTION (a): This property follows from the equation (cf. exer-
cise 5.98)
1 W
f(x)dx = J meas(f > y)dy.
f0 0
meas(g-1(s-1(]y,°[))),
meas(f 1(s-1(]y,='[))) =
(m>x)= U (c>xn),
n
i.e.
]0,cp*(x)[ = U ]0,p*(xn)[,
n
which means that T*(x n) -> cp*(x). Finally, if one had T* > c > 0
then, for 0 < y < e, one would have p(y) 3 x for all x > 0, which
is absurd. Note that (1) can be expressed in the following man-
ner: for x > 0 and y > 0, the conditions p(y) > x and c*(x) > y
are equivalent. From this it follows that
SOLUTION (d):
JE = J0E.f
31 JQE*.
=
424 CHAPTER 9:
and consequently
Also,
so that
(if)* < P.
Therefore
eas(E) fineas(E)
JJ E f- Ef)ot J 0 f
0
f O E JO rt
From this it is immediately deduced that
1 1
J fW< J0 f*, (3)
0
X
F(x) = Jf(t)dt, 0 < x < 1.
0
The set (9f > y) is formed of the x's for which there exists u
such that 0 < u < x and F(u) - yu < F(x) - yx. By the "Setting
Sun lemma", this set is the disjoint union of a sequence of in-
tervals ]an,bn[ such that
Thus
meas(of > y) = E (b - a )
n n
n
1b
y yJ(8f>y)f
an =
1 meas(8f>y)
I f*.
y 0
1 rmeas(8f>y)
(4)
meas 8f > y J0
a
Note that since f* is decreasing, a-1J decreases as a increas-
0
es. Consider, then, an x such that (8f)*(x) > 0 (otherwise there
would be nothing to prove). Since
(0 x
y < (8f)*(x) implies y .< _- f*,
x1
xrx
(8f)*(x) s f*
0
JosoBf = Joso(8f)* S oo
Js{.Jf*}dx .
Jx
G(x) = x-1 f*,
0
SOLUTION (h): If 0 < p < 1 then Holder's inequality for the pair
1
T-1 p1 applied to (5) yields
p
BOUNDED VARIATION etc. 427
rr10 -11-PLJ0
xPJp J
so
o =
t
Let e > 0 and choose a such that 0 < a < 1 and
f0
log1 s < .
Then
1 a 1 1 1
of f*(t)log t dt + log a 1 f(t)dt.
f0 1 0 0
Let E be the set of t 's such that 0 < t < a and f*(t) 4 t
Then
1
(Contd) e + 2Jf(t)log+f(t)dt,
0
whence, at last,
1 1 + 1 1
J Of 2 flog f + log -JO
a
f+
0 Jo
CHAPTER 10
Summation Processes:
Trigonometric Polynomials
W
Iun
n=0
(1)
converges, and if
0
(2)
Ixn - An+ll <
n0
I
00
(3)
L Anun
n=o
converges.
429
430 CHAPTER 10: SUMMATION PROCESSES
IIslI = suplsnl.
N-1
fN(s) = I ()n - An+1)sn t ANsN.
n=0
N-1
IlfN ll = E l An - Xntl I + I AN1'
n=o
Since
N
fN(s) = A0s0 t =an(sn - sn-1),
n1
and as the series I (sn - 8 n-1) is convergent, limfN(s) exists
N_
for every s e E. By the Banach-Steinhaus Theorem one therefore
has
Su "fNII<
sn+rO+r1+ +rn En
U 0 = 3 0 , un = sn - sn-1 if n >. 1.
If it is assumed that
G Ixn - xn+1I G rn
n=0 n=0
NC N-1
Anun = aNsN + (an - an+1)sn nI.
0 n I0
N-1 r
C
n-01+r0+ n
432 CHAPTER 10: SUMMATION PROCESSES
then
un = of n).
SOLUTION: Let
EN
u
n=N+1
If p > N then
and consequently
li ppuP EN,
_ 2 _
fan 1 a = a 4a n+1'
= an - an+1, n =
TRIGONOMETRIC POLYNOMIALS 433
(b): Show that if the sequence (an )n>0 is convex and bound-
ed, then
(i): It is decreasing;
2
an = an - 2a n+1 + a
n+2
SOLUTION (b): The condition A2a 0 for all n means that the
sequence (da ) is decreasing.
n
If, for some integer r Aa
r < 0,
n-1
an = ar - E Aas ar + (n - r)( Aar),
s=r
434 CHAPTER 10: SUMMATION PROCESSES
N-1
E Aan = a0 - aN,
n=0
the series I Aan is therefore convergent, and since the Aan's are
positive and decreasing, (ii) follows by the preceding exercise.
Finally,
N-1 2 N-1
(n + 1)A an = Aan - NAan = a0 - aN - NAaN,
n I=O n I=O
which implies (iii), since by (ii) the last term tends to zero.
if k 0, q)
1 = c0, -+ 0. (1)
9k ' qk+l lpk
< (P (4)
k+l'
enk
W
(i): X lu 1 4 M for all p >. 0;
n=0 n,P
W
(ii): l}i.)m
I un = 1;
00
P+00 n=0 ,P
lim I un p8n = s.
P" n=0 '
11C11 = sup f cn I
n
cc
fp(e) _ I un Pen.
n=0 '
cc
N
C
I1c - cnen II = sup I en I
n=0 n>N
one deduces that the eats form a total set in c0. By (1) it
therefore suffices to prove that for all n
limfp(en) = 0.
f (en) = u
P n,p
TRIGONOMETRIC POLYNGd'4IALS 437
80+81+. . .+8N
CC
sN Ini<N c n' GN N + 1
tN InI`N Inlcn
EXERCISE 10.17,: (a): Show that limaN exists if and only if the
N
series
438 CHAPTER 10: SUMMATION PROCESSES
W
I N 2tN
N=1
(c): Again assume that aN -)- s. Show that if for some real
number p , 1
Inlp-llcnIP
<
G
then sN -), s.
tN +v >' 2 BNS if 0 4 v 4 Ns
S
2A
where A is a constant such that Icn + c_nI . A/n for n >, 1).
SOLUTIONS (a);(b):
N N
tN = n cn = I n(c + c_n) = E n(sn sn-1
InI,N n=1 n=1
sN = (N + 1)GN - NoN-1'
tN = AN + 1)(aN ). (2)
- 'N-1
(a N - a N-1 ) N=i AN + 1)
NL
N-1
aN = N(N + 1) -1 and aN = (N + 1)
satisfy
<
N=1 IAN - N+1
L N 2t,
It
Inll/P InI1 - 1/pcnl
N < NI I nlcnl + NI
Inl-4r r<lnl.<N
N
I
N InI<r
cIn + 1(2
N
n=1
nq/P ) l/q (
lnl>r
Inl
p -l
I I pl/p.
n
)
Since
NC
it follows that
t
N ?11/q N + 1( C lnlp-11 nlp)1/P
4 L cnI +
N N I
I ni4r Ini>r
and consequently
t 1/q
N N
limssupii
N
<
[2]
4 InI>r InIP-1IC lp)1/p
n
Icn + c-nI E n , n
1
Let us show that N - 0 under these conditions, which by part
t
N
(b) will prove that sN - s. Otherwise there would exist a B > 0
and an increasing sequence of integers N such that
s
TRIGONOMETRIC POLYNOMIALS 441
tN > BNs.
S
= tN + I n(cn + c-n
tN +v
s s N <n.<N +v
s S
>. BN - vA
S
BN BN
>
BN
s 2-2 S s
7 2 4
L N
N ' N<N +v
s S S
BNs N+v
s -2 B vs B2
2
x 2 Ns +'v s 2B + 4A > 0'
jN S
a contradiction.
8P + 8 + +
p+1 8p+N
=
ap,N N + 1
SOLUTION (a):
(p + N + 1)ap+N
pop-1
ap,N N + 1
(1)
ap+N + NN p+N - ap-1
N< a< 1Q
SOLUTION (b):
p+N
(N + I)ap,N = (N + 1)8 + G
(N + p + 1 - n)(cn + c-n),
n=p+l
so that
p+N
G (N + p + 1 - n)(cn + c-n
Iop,N - spI N + 11
n=p+1
p N2+ 1 r=1
r= p
N 4pp<NP+1.
Then by equation (1) proved in part (a)
IoP,Np - op+Npl p
IaN
P, - 81p < A p
P
p+l>, q > 1.
n
p
SOLUTION:
n
-E >, qp-r if p >, r,
n
r
2 a nr 2 ¶ a qr-p.
<
r=1 r np r=1 r
q
l im I
Per=l qr-p =
q-1'
TRIGONOMETRIC POLYNOMIALS 445
limgr-P
= 0.
P4°°
lim L argr P = 0,
P-'°° r=1
1
and therefore that N t -> 0. Then by exercise 10.179 sN -> s if
N
aN ->s.
'ON' < A, SN = 0
[a N)
+00
alnlgn
is convergent.
N N
I alnlcn = a0c0 + I an(cn + c-n
n=-N n=1
N-i
(Contd) = a 8 + I Aa.8
N N
n=0
N-i
= a 8 + E Aan((n + 1)an - no n-1)
N N
n=0
N-2
= a a + NAaN-1aN-1 + E (n + 1)A2anan
N N
n=0
00
N
lien a c = (n + i)A2a a
n n
.
N-
where
Set
Prove that
i - 21 1
inf M
fmf :fe
(1
E'J < l2 + 11 1
I n n+ 1 J n+ 1
cp + c1 + + cn = 1.
We then have
B y noticing that
it
CO - c1 ;., (2 - iJcp 3 12 - 1)n + 1
and that :
we obtain:
21 1
m(f) Ii
l nJn+1
On the other hand, if:
P x) = 1 + cosx n+ + 1 + cosnx
448 CHAPTER 10: SUMMATION PROCESSES
sin(n + Z)x
j
1 1
f(x) =n+12+ 2s in x -
A = IlfII = f(x0).
f(x0 + x) 3 AcosNx.
SOLUTION (a): If
f(x) _ cne1
lnlkN
P(z) = zN C zn
InI,N
m = 11f,11. = f'(x0).
By part (b)
n/2N (Contd)
211f 11 > f (x0 + 2N) - f (x0 2N, = J- f' (x0 + x )dx >.
n/2N
450 CHAPTER 10
rn/2N 2m
(Contd) mJ_ cosNxdx = ,
N
n/ 2N
ezx =
e2nz - 1(2z 1 + zcosnx - nsinnxl
X 0 < x < 21;
n
n=1 z2 + n2 ) I
e
zx _ eTIz - 1 +
2 [(-1)nenz - 1] z2osn2
, 0 < x < n;
nz n
n=1 z +n
W
(-1)nenz] nsinnx 0 < x < R.
ezX = n E [1 -
n=1 z +n
What are the values of these series when x = 0?
in f at 0 is 1 - e21z). Therefore
451
452 CHAPTER 11:
e2nz - 1
1
(n) = 2n z n n $ 0.
Since
2tcz
(0) = 2nz
W
2z
ncothnz = 1 + V
z n=1 z2 + n2
so
ing(n) = A(n),
nn).
inTi(n) = zg(n) + 2n (2 - 2e1Ze-
Therefore if n 4 0 then
(-1)nenz - 1 z
g(n) _ ,
n
z2 + n2
di(n) =
1-(1-(T n
Zit
z2 + n2
TRIGONOMETRIC SERIES 453
Moreover
nz
e
7Z(0) = 0,
9(0) = nz 1 ,
which proves the two other formulas. In fact the second is valid
for x = 0. From this it is easily deduced that
m
C (_1) n
sinhnz z +
n=1 z n2
+
EXERCISE 11.186: Find the sums of the following series:
I acosnx nsinnx
n=1 n+ a2 2 c
n=1 n2 + a
2, (a real, a + 0).
SOLUTION: Let
W
g(x) _ acosnx
I (1)
n=1 n2 + a
inz
G(z) 2 z = x + iy, y >, 0. (2)
2e
n=1 n + a
iz
ae
G" - a2G = Re(z) > 0. (3)
e - 1
eix a
2
g,, - a g = aRe
le ix
2
Hence
for 0 < x < 2n, and hence also, by continuity, for 0 < x < 2n.
Since g(0) = g(20, 0 = 0.
Now,
r212asinhan it
so that
acosnx = ncosha(n - x)
2sinhna
- 1
2a '
0<x< 2n.
n=1 a2 + n2
00
nsinnx = nsinha(n - x)
C
2sinhna
0<x<2n.
n-1 a2 + n2
[f ](n) = - -r , n f 0,
and
in71(n) in - 27,
so
2 2n
?(n) = -
2 zn
n
472
3 '
2
2 = 47 + 4 c cosnx - nnsinnx
n=1 n2 0 < x < 2n.
3
C 1 i2
n==1 n2 - 6.
hence if n # 0
(_1)n+i 2
[g'](n) _
in '
and consequently
Furthermore,
Thus
x2 = n2 + 4 c (-1)3n cosnx
,
n=i n2
1)n+in-2 = n2
(
12
n=i
TRIGONOMETRIC SERIES 457
[h'](x) = 21sT,
and
ne-inn
inle(n) = [h ](n) -
we have
2 1 - (_1)n it
lt(n) _ - + (-1)n+l
It . 3 in
zn
and consequently
x2 =2n E(-1)
n+l sinnx
n
-n8 sin(2n + 1)x
(2n + 1)
3 0<x<1.
n=1 n=1
REMARK: It is known that the first series has 2'x as its sum;
therefore
+m
C einx (- 0, 0 < x < 21.
n=-m
458 CHAPTER 11:
(The (c) over the equals sign indicates that the equation holds
when the sum is taken in the sense of Cesaro).
sN =
N _ cosjx - cos(N + ')x
Il sinnx 2sin2x
n
and consequently
N
aN = cot'x - 2(N + 1 sinT-
x
I
n= cos(n + 2)x
=0
cot ix -
sin(N + 1)x
_
4(N + 1)sin2ix
am -> -fcotix.
and
_ 1 sin(N + 1);x 2
FN(x)
oN = N -+I ( sine,
so aN -* 0. Not that if x = 0, then
sN= 2N + 1,
so
TRIGONOMETRIC SERIES 459
aN=N+1+,,.
SOLUTION: (i) =a (ii): For all 0 < r < 1 and all n > 0,
2nei(n+1)xF(reix)dx.
0 znF(z)dz = irn+l
- 1lzl=r 10
?( - n - 1) = 0, n 0.
Therefore
+00
?(n)einz
(*)
n=-oo
SOLUTION: Let
?(-n)e-inz
an(z) = ?(n)einz +
Then
so that if n j 0
-n61 n62
+ e e - 0(e-elnl),
1?(n)l < M
le2n62 - e-2n611
_6
e < J k I < e6 ,
such that
F(z) = if = eiz
anEn
462 CHAPTER 11:
+m
E lanl < W,
+cm
f(x) = G anel , x R.
0(InI1/20'e-(InI/R)1/a).
?(n) =
O(e-elnl)
If(n)I = for some e > 0.
IS
Il
lnlr(as + 1) < (as)-asr(as + 1) ti
e-as
2.
Since
1/a /a
I I - a < as < [J..L]
J
SOLUTION: (b):
+00
f(s)(x) = E (in)sf(n)e11"`, s > 0,
-00
and consequently
1/a
IIf(s) II00 = 0( 1 ns +1/2a e- (n/R) ).
n=1
xae-(x/R)1/a
(x) = x 0.
a
xo = R(aa)a.
Now,
Ra(aa)aae-aa
cpa(xa) =
aRa+1r(aa + a) = 0(aaRar(aa)).
Furthermore,
Ra(aa)aae-aa
= O(a'Rar(aa))
Therefore
00
so
IIf(s)II Co
= O(RSal), R > 0. (1)
0(e-£InI)
I?(n)I = 0(InI2e-InI/R) = if 0 < e < R
IJ(n)I = O(e-EInI),
IIf(s)II0o
= 0(s2E-SS!) = O(Rss!) when R > 1g .
Co
f(n)einx
then
?(n)elnx
= s. (DINT'S TEST)
n=-
SOLUTION: In fact
r n
sin(Nt+ J)t
a (f;x) - s =1 (f(x + t) + f(x - t) - 2s) dt
J
0
+ eN(x),
?(n) = 0(n-a).
SOLUTION: In fact
TRIGONOMETRIC SERIES 467
2n (
)e-inxdx,
= 4nj OX) - flx + n,
0 l J
so
?(n)I 2(rc)a.
It is known that
+ eN(x),
468 CHAPTER 11:
where
lime (x) = 0.
N
EXERCISE 11.196: If
0(na-1).
I?(n)I =
11
Now,
n -: nx
e dx = na-l(G(nn) - G(in)),
J
C Ix la
TRIGONOMETRIC SERIES 469
where
rx -it
G(x) = e
0 0 Itla
dt.
I
Mo(n) na-1
If(n)I .<
P(f) _II.fkII/k
P
exists.
SOLUTION: (a): It follows from what was noted above, that, set-
ting
ak=IIfk11pI
470 CHAPTER 11:
one has
p(f) = limak/k
k
exists.
27E
0(Ihla).
If(x t h) - f(x)Idx =
0
f1 = f and fk = f"fk-1' k 3 2,
then
2n
f(n) = 1 {f(x) - f+ n)}e-inxdx,
0 IX J
so
0(Inl-a).
If(n)I = (1)
0(Inl-ak),
Ifk(n)I =
G I?k(n)I <
n
a > 0.
Show that if p is an integer such that a - p > 2, then f(p)
is defined almost everywhere and belongs to L2(T).
lnlp-ilf(n)I
E
< -,
If(P-1)(n)I 0(InI-(0'-p+1))
= with (a - p + 1) - 1 >
Assume, therefore, that a > 3/2, and let us prove that f' exists
almost everywhere and belongs to L2(T). We have
Y Inf(n)12 < M.
rx
g(t)dt = (f(n)el"x - f(n)).
0
Since
Y If(n)I < -,
TRIGONOMETRIC SERIES 473
the right side is equal to f(x) - f(0), which proves that f'(x)=
g(x) almost everywhere.
L2,
t Ilf' 112
ExERCISE 11.201: For every finite set AC 2Z and all e > 0 show
that there exists f e L1 such that
1 if Inj 4 r,
211
2N+2r+1
( 2N+ 1
ILrII <1+e.
2 I(fIki)I2.
I
if112 =I
2
-Ti(n) 12 = 1, n ea.
TRIGONOMETRIC SERIES 475
c n > 0, E cn = W c2 <
on
n n
and let g eL2(T) be such that g(n) = cn. If I I1pi111 < °° were to
hold, then we could write
E cn = I I
n n i
_
i n
_ (g*pilgi)
= G (g I Ai,ec"pi)
Z
I19112 l1q'i111,
which is absurd.
N inx
p
fN(x) = e
P=1
N
limfN(x) = 0.
N
Cm 21[
2n L
m=1
J If
2(x)I2dx = 2 < °°,
0 m m=1 m
2 N in x
fN(x) - N f 2(x) = I e p
m NI
p=m+1
2
(m+1)2-1- m2` 2
N rN-
limfN(x) = 0
N
limaNf(x) _ .
N
r2n1 1/p
IIkWk IIp 4 k2-k/p <
k k
liminfaNf(x) > k,
N
so
lima f(x)
N
SOLUTION: (a): If f e C(T) let Ti(f) = w.*f. The Ti's are con-
tinuous linear operators on C(T), and
T.(en) = r. Wen }
en
SOLUTION: (b): Let 0 < a < n, and let f be the function in C(T)
that is zero on [-Ja,'-za], equal to unity on [-n,-a] and [a,n],
and linear on [-a,-'-2a] and ['za,a]. Then
2nf(0) = 0.
ExERCISE 11.206: Let (an)n0 be such that an > 0 for all n, and
A } W. i
n
Show that there exists a continuous function f such that
in x
(Consider a series I An'Ie s , where the integers ns increase
s
quickly enough).
The series
1 in x
s
f(x) _ ante
s
`(ns) = A.
Therefore
E a1n1If(n)12 = I X"
I?(ns)I2
=E1 =.
n s S s
Show that
f = (anf(n))nea'
Yn = limfi(n) = limxnfi(n) = 0.
Z 1.
I1 - N+ 1, lanl < A,
lnl<N l
L Ian1 < A.
n
Inlm IunI = 0,
Hull = S1PIunl .
u (x) '\
anune",
and
g112
2 = E anlunl2 < IIuI12 E an
n n
is continuous on c0 and
11U(r)
11 F 1,
and by (1)
2n
u 2nI
z7(x)f(-x)dx
0
((2n
By taking for u the sequence whose n-th term equals l and all of
whose other terms are zero, one obtains
an = anf(n),
and consequently
X andf(n) <
n
and thus would hold for all f e L1(T). By the preceding exercise
this would imply -
IX <
n
n
E = nk Ek
u e L2(T),
N
I c sinnxl < 2Av.
I
n
n=1
W
f(x) = E cnsinnx
n=1
belongs to L1(T), and that the right side is its Fourier series.
SOLUTION: (a): One can assume that 0 < x < it. Let M be an ipte-
ger between 1 and N - 1. Noting that
1
sinnxI < nx and Isinpx + + singxl <
si- x < x 9
TRIGONOMETRIC SERIES 485
one obtains
N M N
An
1 csinnx1.1 111+1 E 15AMx+ M+1x.
n=1 n=1 n=M+1
I ensinnxi 4
x aN+l'
n=N+l
M
g(x) = E Cnsinnx,
n=1
W
ansinnx, (*)
n=1
486 CHAPTER 11:
(iii): an = o(1/n).
N
E ansinnx I c M, x eat, N>, 1;
n=1
(iii): an = 0(1/n).
Show that in this case the partial sums of the series (*)
converge in L1(T).
N
cos(N - ')x - cos(M + })x 1
D (x) = sinnx =
2sin ( )
N,M n=N -
IDN,M(x)I
< since
i T , 0 < IxI 5 n. 2)
M M-1
I a sinnx = aMDN M(x) + E Aan.DN n(x), (3)
n
n=N n=N '
where
4an 3 0.
= an - anti
Mp naN
ansinnxl < 'T , 0 < IxI < it. (4)
InLN
SOLUTION: (a): (4) shows that the series (*) is uniformly con-
vergent for 0 < d s IxI < it. It clearly converges for x = 0.
Thus its sum is an odd function, continuous for 0 < IxI 4 it.
lim{supl I ansinnxl} = 0.
N-' x gN+1,<n4N
In particular,
nn
lim a sin = 0.
N- 2N+14n4N n 2N
Now,
Na
I n
asin 2N N (5)
ZN+itn<N 2V2-
which proves that NaN - 0, and thus that (i) => (iii).
Let
eN = supnan.
n>,N
it
FT-1
I I ansinnxl 4 1
1 ansinnxl + I
: ansinnxI
n>.N N4n<N+p n>'N+p
naN+
IxI I na +
Non<N+p n ix
.pIXICN+(p+1)xN+
P
so that, because
TRIGONOMETRIC SERIES 489
we have at last
N02N(2'N) = 0.
If x = 7t/2N then nx < it, hence sinnx 3 0, when n < 2N and nx < 7E12,
2nx
hence sinnx a = when n < N.
N, Thus
a2N(77E
N) ;' n1N (1
n1
2N + 11 N 2N n
2N n<N
I 1 ) a
SOLUTION: (c): Inequality (5) proved above again shows that (i)
=> (iii), and in the same way (iii) => (i) follows from (6).
Furthermore, (i) implies that the sum of the series (*) belongs
to LW(T), and Lebesgue's Theorem justifies obtaining the Fourier
490 CHAPTER 11:
g(x) = Dansin2'nx
tanx
n=1
h(x) = 2 1 ansinnx.
n=1
Since
g >. 0 and
(n cWc ' n
Og(x)dx = ax (7)
°anj0 tan x
n=1
J71 lrn
f(x)sinnxdx = Da (x)sinnxdx = Aa = a .
n n k nJI 1,k k=n k n'
k=1 n
1 2
tan x - X
ji7c
JO sinntx
z
dx = 2J1 s x dx + 0(1) = + 0(1)
n logn,
492 CHAPTER 11:
and because
lognti1+2+ to,
(ii) is therefore equivalent to
Since
N
I n= I1 + + ... + N)aNtl + 1 I1 + + ... + nlpan
n=1 2 n=1 l
2
11
I1 t + +
NJaN+l =
[i++ ... + tan
n:Ntl
2
1 {1 + + + nldan.
n>.N+1
2
Finally, if sN is the n-th partial sum of the series (*), then
hence by (8)
IIf-SNII130.
00
Co
Co
L1,
a N N
8N(x) = 2 + E ancosnx = an( (x) - Dn_1(x))
2 n=0
= n
n=1
N-Ci
2 aNDN(x) + G AanDn(x).
n=o
2
Since
D0 + + Dn = (n + i)Fn,
this yields
N-1
sN(x) = 'aNDN(x) + jN aN_1FN-1(x) + E (n + i)A2an.Fn(x).
n=0
IDN(x) 1
sin x , IF (x)l <
n
(n +
1
-12X
SOLUTION: (b): If
an - 0, G
(n + 1)IA2anI < °°,
2f(k) _ (n + 1)62anFn(k)
n=0
m
= I (n + i - k)t2an.
n=k
TRIGONOMETRIC SERIES 495
Now,
N
E (n + 1 - k)a2an = ak + (N + 1 - k)DaN+1'
- aN+l
n=k
NAa -> 0
n
we obtain that
?(k) = jak,
r n a
f(t)dt <
1
1
- it aJ-f(t)dt.
a
+m a
E ccp(-n)
n = 71-h(x)(p(x)dx (*)
n=-m S
aJ a
C2
< Ih(x)I2dx.
n=-w S
a
0<W< Zn and W(x)=0 ifa< Ixl <n.
Furthermore,
Therefore
it +W
I8(0)I2n
2nf(0)$(0) = f(t)dt
fn
2 n
(2 n) f _nf(t)dt,
(f19) = J_1r.
lim85(n) = 1, (1)
S
I en1;S(n)I2 <
n
an1SS(n)I2einx.
gs(x) = I (2)
n
= (h* *9sk,P)
Therefore
gs = h*e JS on ]-Y,Y].
S
Y
112
IIgs 2 YJ-YIh*Os*8sI2 Y Ilh*ss*;s 1122 < Y Ilhll2,
and on making y - 6:
Icn12 Icn12
I = sup E = sup(lim X (0 s(n))2)
n N InkkN N s Inl4N
< anIIh112.
I -isign(n)f(n)e1RX,
where
-1 if n s -1,
sign(n) = 0 if n = 0,
1 ifn1.
The ADJOINT FOURIER SUMS of f are the numbers
I
-isign(n)f(n)eln".
SN(x) =
InI,<N
s*N = f*D*N',
ti 2logN,
J D(u)du
0
lime(u) = 0,
u+0
show that
J e(u)DN(u)du = o(logN).
0
(Set
then
3N(x)
lim togN
N
If(n)I = ° (1)
n
then f is continuous.
N
DN(u) = 2 1 sinnu. (1)
n=1
it ["(N-1)] 1
JD(u)du=2 -n-.72
0N n=0
ti 2logN.
AN(u)>.0 if 0su4 R.
show that
n r n n
iJ eA*1 < a A* + J e(u)cotiudu + o(1).
N
0 J
oN d
TRIGONOMETRIC SERIES 501
Since
ti 2logN,
n
limsup(logN)-11E J E 2a,
N fo
EAN = 0(logN).
J
0
n
J0CD = o (logN).
n
N(x) _ - 2nJ (f(x + u) - f(x - u))DN(u)du,
0
If
lim(logN)-1sN(x) _ - n (4)
N
502 CHAPTER 11:
SOLUTION: (e): If f has left and right hand limits at all points
then £ exists at every point x and is equal to
Further, if
f(n) = a
11n J '
then
A
U c
n n
Set:
u1 + 2u2 + + nun
an = n ,
2 2
u2 + 4u2 + + n u
bn = n ,
yn = n L ussin2
2n
s=1
n
2
vb +A
2
Yn 4 nv v
where xr = nr/n.
Show that if
(c): Calculate:
r2n
J pn(u)du,
0
(d): With the aid of what has gone before, prove the fol-
504 CHAPTER 11:
2 2
SOLUTION: (a): Since s us .< Asus, it is clear that b n .< Aan.
The Cauchy-Schwarz Inequality shows that
yn < n X
us l(2n) 2+ A2n L
s=1 s>nv s 2
"4
2
bnv + A2nl
r
1nv x
2
a2 _ n4v bnv + A
2
2
l i nsupyn - A
and consequently yn + 0.
TRIGONOMETRIC SERIES 505
SOLUTION: (b): If
2n-1
qpn(u) < e I If(u + xr+1) - f(u + xr)J e EV(f;0,2n).
r=0
506 CHAPTER 11:
SOLUTION: (c):
(eisxr+1 - eisxr)f(s)eisu
f(u + xr+l) - f(u + xr) ti
s
is(x + n/2n) (( ll
sinl2nJf(s)eisu.
ti 2ie r
s ll
m
(2
gn(u)du = 8n Y IJ'(s)I2sin212nJ
2nl 0 s= l
M
(27[
2j pn()du = 16n = Ij(s)I2sin2[22n-
0 s=1
n If(s)I2sin2[In- 0
s=1 JJJJ
(4P-1)/3
_ I ascosst. (1)
S=O
Show that
yptl - yp
= 3P.
x
f(x) x + limJ (1 + cost)(1 + t cos4p-1t)dt.
p-0 0
(4)
(d): Show that f has bounded variation and that its Four-
ier coefficients are not o(1/n). Calculate f(4n)).
(1 + cost)(1 t cos4p-1t)
(4p - 1) < 4p + 3
3
which shows that there exists a sequence (as) satisfying (1) for
all p > 1. By what has gone before it is clear that as = 2-u
with u eJi, for all s such that as # 0; in particular, 0 < as < 1.
yp = 2(yp - 1) + 1 = 2yp - 1,
yptl
and consequently:
3p.
yptl - yp
a
IFp+1(x) - Fp(x)I .< .. < (Contd)
(4p-1)/3<s4(4p+1-1)/3
TRIGONOMETRIC SERIES 509
Yp+1 - yp = 3p+1
(Contd)
(4p-1) 4p-1
3
- 1 + (1 + cost)(1 + cos4p-1t)
SOLUTION: (d): For all p,FF is increasing; thus the same is true
for its limit. But then f has bounded variation because it is
the difference of two increasing functions. The Fourier coeffic-
ients of f are the limit of those of -x + F (x), so
aSsinsx
f(x) s
s=1
It is clear that:
4-s
2if(4s) =
where A,B are constants depending only upon a,a, and where [u]
denotes the integral part of u.
Deduce from this that f e Lip(a).
(b) : Set
sinTrx
F(x) = if x # 0, F(0) = 1.
Prove that
(1-a)n
f(s) _ a n Flan - s). (*)
n=1 a + s
(c): Let 0 < A < loga, and let n0 be an integer such that
n0 1
a >2loga-A
For every integer v > n0 denote by sv the integer such that
v- 1-zsav<sv+
Show that
6 > 0
Vf(sv
TRIGONOMETRIC SERIES 511
SOLUTION: (a):
so
[u]-1
lf(x + a-u) - f(x - a -u ) <2 E a (1-a)n-u +2 a -na
n=1 n= [u]
2a(1-a)[u]-u + 2a-a[u]
a1-a n-a
-1 1-
Setting:
A B= ,
a 1-a2
-1 1- 2 a_a
the desired inequality is obtained. Then
a-na lI (cosanx)(cossx)dx.
n
n=0 0
On observing that
n
F(an + s) + Flan - s) = 2a F(an - s),
(an + s)
this yields
(1-a)n
(s) _ a F(an - s).
n=0 an + s
F(x)I <
n x if x # 0,
17
and that if n > v
Since
n n
avioga - 1 > a ologa - z > as 0 > 0,
it follows that
IF(an-8 1
if n > v.
V
)I <
n(avloga - 1)
TRIGONOMETJ?IC SERIES 513
sv - an > av - an - '7 > (v - n)anloga - ' >. anloga - 'z > aan,
because
n n0
a >, a > 2 logo - A
Thus
SOLUTION: (d): Taking into account part (c) and the inequalities
IF(x)l < 1 for all x and IF(x)l >. 2/n if jxj < 'zf we obtain
In0 1 and-a)n
F(an
- sv)I . _n r 1
a(l-a)n
- 0(a-v),
n=0 a+ s v v n=0
v-1
a-an
= 0(a-v),
n s
n=n0 v n=0
1 a-an = 0(a-v),
s
n=v+1 n(avloga - Z) n=0
and
(1-a)v (1-a)v
a F(av - s v )>,2a
71 2av+
av+sv
ti
1 -av
7Ca
av1f(s)I
> a
for v large enough, which proves that the .W's are not o(s-a).
Pr(x) = I rlle
n inX
1 - r2
1 - 2rcosx + r2
+00
Ar(f;x) _ L
rlnl?(n)e"x.
lim A(f;x)
r = '2(f(x + 0) + f(x - 0)).
r-*1-
(c) : Set
(x
F(x) = f(t)dt.
J n
n
( x) = 1 F(x + t) - F(x - t) Q (t)dt
r f'
A
2nj_n 2sint r '
SOLUTION: (a):
n
1
Pr > 0, J- P (x)dx = 1,
2n
n
r
Set
(n
A(f;x)
r
- s = 2n J QtP tdt.
0
then
IAr(f;a) - sI . e + Pr(d)IIQIII,
limAr (f;x) = s.
r+1-
it n
1 1
Ar(f;x) = 2J- f(t)Pr(x - t)dt - 2nJ- F(t)Pr(x - t)dt
n n
n n
= 2nJ- F(x - t)P'r (t)dt = - 2nJ- F(x + t)P'(t)dt,
r
n n
so
TRIGONOMETRIC SERIES 517
n
Ar(f;x) = - 2n j(F(x + t) - F(x - t))P'(t)dt
n
setting
we have
F(x + - t) Qr(t)dt.
Ar(f;x) = 2If 2sintF(x (1)
n
It is clear that
n
Qr(t)dt = r.
2nI
n
( 2
sint
Qr(t) = 2r(1 - r2 )I
2)
l1 - 2rcost + r )
2r
cost =
1 + r2
as long as
r2)-1
cosd < 2r(1 +
we have
n
Ar(f;x) - D(x) = 2nJ (t)Qr(t)dt + (r - 1)DF(x),
-n
since i(t) -> 0 as t - 0, one can show as in (b) that the integral
on the right side tends to 0 as r - 1, from which the result fol-
lows. (Note that there is no problem at ±n since at these points
Ar(f;x) + f(x)
rh
limh_1
h-*O 0 + u) - f(x))du = 0.
h
limb-11 If(x + U) - f(x)ldu = 0.
h->0 J0
Borel (written
(B)
c
n
s)
_m
if
n
r s r
r-
lime
n=0
ni = s
where
sn = IpI e
<n
z n (B) 1 1 z
if Re(z) < 1.
n=0
n
Br(f;x) = e-r 8 (f;x) x,
n30
n n'
Show that
Br(f;x) = (f*Br)(x),
where
rx
J(x) = J Is u du tin logx as x
0 u
(e): Prove that
r
116111ti 21ogr, r
R
1
lim(f(x + t) - f(x))log = 0
t-r0 I I
TRIGONOMETRIC SERIES 521
at a point x, then
?(n)einx
(B) f(x).
n
-r. r.
un,i = e n! '
then:
(i): U >. 0;
n,i
w
(ii) : E u
n,i = 1;
n=0
(iii): limu = 0.
ti
ni '
s rn
X1->-
limer
ni s ifsn->s.
n=0
1-zn+1
sn 1-z
and consequently
522 CHAPTER 11:
s rn
e -r L n
n=0 n! = 1 1 (1 - ze
z
Br(f; ) = f*Br,
where
n
-r C
Dr
n
= e
Br n=O nt
.
Recall that
i(n+ )x
Dn(x) = ImIe si-fix z
0 < lxI
and consequently
=
e-r(1-cosx) sin(x + rsinx)
sinix
n-1 n
J(nn) =
sinx dx.
J X sn
8=0 0
J(ma) ti loge.
and consequently
J(x) ti n logx.
SOLUTION: (e):
U
that as
2 n/2 e-2rsin2u _
this yields
524 CHAPTER 11:
_ e-2ru2) du rn/2u3e-8ru2/n2du = 01
lx/2(e-2rsin2u
0
u < 23-f
O r
Consequently
2
2 n/2 a-2ru
IIBr1I1 Isin(u + rsin2u)ldu + o(i).
= 'If 0 U
Applying the mean value theorem once again, and using the inequal-
ity
3
u- sinu F 6, u 3 0,
it follows that
3
4r3u
Isin(u + 2ru) - sin(u + rsin2u)l t
and since
43 f
o/2u2e-2ru2du
= Cl rJ
2
2(( n/2 a-2ru
IIBrlll = 0 u Isin(2r + 1)uldu + o(1).
,if
-2d
e J((2r + 1)r-'6) t 4 J((2r + 1)r-16) logr,
fo it
TRIGONOMETRIC SERIES 525
2
n/2 fm e-2u
du.
U
611F S
Then
2 IB
2 e-2 r
< liminf logr
ill
limsup
JIB rill 2
1E
- loy 1<
10r II1 _ 2
logy n2
f - Br(f;x)
SuPIBr(f;x) I = 00
if feG6.
n e-r n n
71
1Br = n>,0 n! 21J-1Dn
=
e-r
I
Ln = 1.
n.
n>.O
Br(t)
Qr(t) =
log l t l
then
U
Br(f;x) - f(x) = 2nf-X cp(t)Qr(t)dt. (1)
-2rsin2'z6
IQr(t)1 log2 sin 6
2
n/2 a-2ru
IIQr1I1 = jcJ Isin(2r + 1)uldu + o(1).
0 ulogu
a2u2
J((2r + 1-)r-') + 1
du.
logn log2 T u
But then if E > 0 and if 6 > 0 is such that ItP(t)I < e for 1t1.< 6.
TRIGONOMETRIC SERIES 527
- 2rsin2d
IB(f;x)
r - f(x)I < eA + Icp(t)Idt,
l0 2 sin d.l
d4Itkit
which proves that
limB(f;x)
r = f(x).
r-+W
An = a0 + +a .
Suppose that
a
An n-> 0.
A
n
The series
+W
L en
+w
(N)
E Cn - 8,
_m
if
a s + ... + a 8
n 0 0 n
vn = A
n
where
528 CHAPTER 11:
8 = X C .
P
IP14n
a0>a1>,.., >,an>,**, 9
an-*0. (*)
For f e L1 set
n
Nn(f) = A 1 apse-p(f),
n p=0
where
Un(x) = A
sinsn x X acos(p
p + )x, (1)
n p=0
n
1 cos(n + 1)x
V a sin(p + 7)x. (2)
= A sin ix P
n p=0
Show that
1/n
n(x)Idx
J
TRIGONOMETRIC SERIES 529
(II): (b): Show that there exist constants B,C,D such that:
n
Isin(p t 1)x dx 4 B(p + 1)log p + C(p + 1), 14 p < n,
J1/n 2sin2ix
n
slnxI
dx < Dlogn, n > 2.
1/n 2sin2Ix
and:
where E is a constant.
SnPIIUnII,
71 n-l
x )x
JO(sin2(n + 1)W){cossnnt + An Asin(p + 1)xfdx
l p=
nL 2 JJJ
530 CHAPTER 11:
7E sin2(n
f0 s infix
+ 1)x.cos(n + z) x
dx'
lTE
deduce that
1 nA
sup
n A I --
n p=1 P
< .
inx (N)
f(n)e = f(x)
n A
P < .
sup A
n p=1 P
V,
(III): For every real number4 set
na = (a + n)
n n! -
a _
Aa-1
n n
TRIGONOMETRIC SERIES 531
A = Aa
n n
a g0 An1 + ... + S n A
a =
n Aa
n
is the CESARO SUM OF ORDER a of the series I en, and that the
latter is (C,a)-CONVERGENT if 6n is (a > 0).
Prove that for every a > 0 the Fourier series of a function
f e L1(T) is (C,a)-convergent to f at every point where f is con-
tinuous.
vn P10 un,pap,
=
where
a
if 0 < p 6 n,
A-p
n
U =
n,p
0 if p > n.
532 CHAPTER 11:
It is clear that
u = 1.
p>0 n,P
Moreover,
a
n_p
un,P { An-p
and consequently
limu = 0.
n n,P
SOLUTION: (I): (b): Writing an for the n-th Cesaro sum of the
series X Cn,
n
vn an-p((p + 1)ap -
n p0 PCP-1
(n + 1)a n-1
A 0 an + A (p + 1)(an-
n n p=0 P - an- P 1)a p
p=O an,paP,
where
(P (an-
n P - an-P 1) if 0 4 p < n - 1,
(n + 1)a0
n,p A
if p = n,
n
0 ifp>n.
TRIGONOMETRIC SERIES 533
lima = 0.
n n,P
Furthermore, when all the sn's are equal to one, the same holds
for all the a 's and v 's. so
n n
A = 1.
p=0 n,P
n
=
1 sin(n ± 1)x
apcos(p + )x
n p=0
A
n
cossn 1)x
An -7-
1
i IX 0 apsin(p + J)x,
p
sin(n + 1)x
2(n + 1) ,
s in x
and consequently
so
534 CHAPTER 11:
1/n ( l
JU 4 211 + l 4.
f0
A = ap - ap+i'
we obtain
n-1
1
V (x) = cos(n + 1)x{(n + 1)anFn(x) + (p + 1)h F (x)}.
An p=0
E
P P
(1)
Since
it follows that
n-1
11711 A {(n + 1)an + E p + i)AP} = 1.
n p=0
n
sin(n + 1)x
cos(p + ')x =
I
p=o
2sinx
we obtain
fit 2
sinx dx 4 (n xdx = it
log(nn) ` Dlogn.
1/n 2sin tx 1/n (2x )
11
IU
n( C [(n + 1)a n +
p=1 P
n-1
+ [D0ogn + B (p + 1)log
p P1
n p=1
n-1 n
E (p + 1)A log 2allogn + X alog
P
P p
p=1 p p=2
n (( l
+ E palogl1
P - .
P=2 I p J)
n n-1 l
n ((
1
a log = I A logl1 + - Alogn.
l
p p=1 P t P1
p=2 P
536 CHAPTER 11:
Therefore, writing
E = D00 + 2Ba1,
we have
J7E nIl
IUnI < C + A (Elogn + B Alog1i
p + pJ).
1/n n p=1
1 sin(n + 1)x
U (x) =
n A s inmx
n
r( n-1
x 1Ancos(n + )x + 2sinZx A sin(p + 1)x}
P=0 P J
= sin(n + 1)x.Gn(x),
where
n-1
Gn(x) = cossin
xT- + A2- 1 A sin(p + 1)x.
n p=o p
Since
we have
rn
supJ sin2(n + 1)xlGn(x)Idx <
0
Let
TRIGONOMETRIC SERIES 537
2
M= sup I cot''x - x
Otxtn
Since
J
it sin2(n + 1)xcos(n +
0 s ink )x
dx
jn(n+l )
it(M + 1) + 2sin2u.cosu du
0
sin2u.cosu du
u
I
n-1 rn
1 E A s i n (n + 1)x.sin(p + 1)xdx
A n p=0 p1 0
Since
538 CHAPTER 11:
Since
n-1
E A 6 n,
An p=0 P
it follows that:
A n1 A 1 sin(p + i)xdx =
- it
2
A
p
(2p + 1)
Ti p=0 p 0 n 062p6n-1
1
nA 1 r1 + 1 l
A
2p
An
g am= An
C
pJ 2p + 1
p=1 p 162psn
1 2 A2p-1
l 1+ 2p - 2p + 1)
G
A262pcn+1 11
+ Ti
1 n A A
nt2 A
p=1 -k6
An p
s A A
2
2p + 1
n n+2 062pEn+1
Since
An+ a t a a
A
1 t
n+1
A
n+2
41+2 n+1
A ,
n n n
at last we obtain
n A
sup - < m.
A
n An p=1 p E
1 n
On = Un - n = A I aDn_p
P
n p=0
It is clear that
n
1 _
2itj On
1 [sin(n t 1)x12,
Fn (x) _
nt1 sinx J
n-1 a
su IV 1(x) I < 1 2i 2 .
An
An sin 16 lan + p0 APl n sin 16
su Un(x) < An 12
6<IxT<n n 2sin Zd
Thus
M(6)=
n
su IO(x)I-*0
On asn -}w.
6<IX
If
n A
sup
n
A E < (3)
n p=1 p
then
540 CHAPTER 11:
n
sup A <
n n p=1
i.e.
1
sup A loge <
n
By the inequalities
rl/n
llnII1<1, I JUnI <4,
0
J171 n-l
C + An (Elogn + B Aplog1 + p)
/nl nI <
pI1
log11 + pl
`
p
-(4)'
sup lln ll l
shows that
because
IIVnII1 4 1.
W
Anxn = (1 -
x)-a,l
-1<x<1,
n=0
nC Aa-i = Aa.
L
P=0 P n
a a
An = An, limAn
n
lim-=lnma+n=0.
A
n
Furthermore,
an+1 (a - i)
an = 1 + n + 1 '
from Fejer's Theorem and from part (I)(b) above. If 0 < a < 1
note that
a n
a
A
n'Pa+1
so
a
A
c na a
p=1
p aMa+1
L n
+W
E I f(n) I q IIfIIp.
n=
+00
E Ic n IP<
n=-w
+00
I einx
n=-m
+W
Ilfllqq < 1 ICn Ip.
n=-W
as
If0<t<1and
1 t
p= 1 1 t+ t
2
1 1- t + 2t = t2
q - W L 2 ,
11q
IIT(f) < IIf1IP (3)
Note that p runs over the segment [1,2] when t varies from 0
to 1, and that + = 1. Formula (3), proved for f e LW(T), is
all the terms of which are zero except for a finite number, and
let
+W
$(c)(x) =
V.
cneln"A x e3R.
n=- ,
Then
(4)
k(c)112 = Iic 112.
so
THE PROOF THAT (iii) => (ii) IS VALID ONLY IF f(O) = f(l)
and write
n - rl
n 1 .f(x.) 1f
ti=1 1 0
n
+ n -1 (1 - e (xi))f(xi)
2=1
n
In-1 I
In 1 f(xi) - J 'A < -, E(xi)f(xi) - Jl0Efl +
ti=1 0 i=1 0
(Contd)
545
546 ERRATUM TO EXERCISE 3.45
(1 -1 n
(Contd) + M1 (1 - 9 ) + Mn (1 - 8e(xi)),
e
0 i=1
and therefore
n 1 2Mf1(1 E
- fl
limsupn-1 f(x .) J fl < - 8 ) < 4eM.
2=1 0 0
Bibliography