Professional Documents
Culture Documents
[manna]
= f. Sw) Ky.8) dy.
By the bounded convergence of (2.12.2), the left-hand side tends
to that of (2.12.3) as Yoo. Also since K(y,8) is periodic, with
period a,
es eg mtb
{ $oXe8) ay = (f+ 3, J, Janka ay
tents
‘norms ay + $ f'na+marny6 dy
a
®
= Froxw.a dy + | xWKly,3) dy,
by the dominated convergence of the series, K(y,3) being bounded
for a fixed 6. The result now follows from Theorem 17, with x = 0,
a= —}a, 6 = da, and K(0,y,3) = Kiy,8)-mW
TRANSFORMS OF THE CLASS 1*
3.1. Plancherel’s theory of Fourier transforms. Tux formulae
(1.2.1), (1.2.2), connecting a pair of Fourier cosine transforms f(z),
(2), express a relation between these functions which is formally
symmetrical. But in all the theorems which we have proved so far,
the two functions satisfy quite different conditions, so that the
symmetry is only formal.
A theory of the reciprocity which is completely symmetrical was
first given by Plancherel.} It depends, not on ordinary convergence
or summability, but on mean convergence.
For complex transforms Plancherel’s theorem is
‘Tuxonem 48. Let f(x) be a (real or complex) function of the class
IX —co,00), and let
Feaa) = Tas | Sorter dy. 1.1)
Then, a2 a> 0, F(x, a) converges in mean over (—00,00) to @ function
F(z) of LX—co,00); and reciprocally
.
$0) = F555 | Podeovdy 12)
converges in mean to f(z). na
‘The transforms f(z), F(z) are connected by the formulae
a1 a f ped
Fe) = Jey a] Hy) ay, (3.1.3)
(3.1.4)
‘for almost all values of x.
It will be seen from the proof that we might replace F(z,a) by
:
F(e,a,b) = ea J soe ay,
where @->00, 600, in any manner.
4 Planchorl (1), (3), (@) (4).70 TRANSFORMS OF THE CLASS 1+ Chap. 11
At the same time we obtain
‘Tuuore 49. If f(x), F(x) and g(x), G(x) are Fourier transforms
as in the above theorem, (2.1.1.), (2.1.2), and (2.1.3) hold.
For cosine and sine transforms the theory is as follows.
‘Turonm 50. Let f(z) belong to LX0,00), and let
Fee,a) = | () f fiyeoszy dy.
Then, a8 a-> co, F(z,a) converges in mean over (0,00) to a function
E(x) of LX0,00); and reciprocally
t4e0)= ,/(2) f rlavonsy ay
converges in mean to f(z). We have almost everywhere
‘TamonuM 51. The analogue of Theorem 50 for sine transforms holds,
with cosy replaced by sinzy, and sin zy by 1—cos.zy.
‘TreoneM 52. (2.1.4), (2.1.5), (2.1.6), and (2.1.7) hold for transforms
of Lt.
‘The cosine and sine theorems may be obtained by taking f(z) even
or odd in the ‘complex’ theorem.
We shall give several different proofs of these theorems.
3.2. Fourier transforms, first method.} This is suggested by
Fourier’s formal process (§ 1.1). Let.
cepa
fia)de (v= 0,41,..),
an
and O,(2) = 3 a,c,
Then if 6 > 0 and n = [s]—
=e
Tim ,(c) = f flyer dy
= +
4 Titchmarsh (1), (2)32 TRANSFORMS OF THE CLASS 1* n
uniformly in any finite interval. For
. avy
oer f soe a| [ Feryermn—een dy —
4 A
o
»
oan
7 aT ane
ed domraa — [so ay)
.
: om
0,
Fe) = a5 J toerae (4.1.1)
converges in mean with exponent p'. The mean limit F(z), called the
tranaform of f(z), satisfies
° Lie 0
J Fer de 0o, to F(z) say, with ex-
ponent p’. The remainder of the proof is the same as in the special
case where "= 2k.
Still another proof of Theorem 74 can be obtained from a general
theorem of M. Riesz on functional operations. See Zygmund, § 9.2.
4.6. The Parseval formula.
‘Taxonem 75. If fle) and G(x) belong to L*(—o0,00),1
[ Fe, Ke) de = ae ay J aya f pena
2 7
1 a" f _
= af soe | Gla)e de = J Sal—t,8) at.
‘Making a 00, and applying (4.6.1) to the left-hand side, we obtain
: i
f F(e)G(e) de = f fldg(—t6) de.
4 4
Making 6->0o, and applying (4.6.1) to the right-hand side, we
obtain (2.1.1).
There are also obvious extensions of Theorems 58-62.
4} Titchmarsh, Theory of Functions, § 12.53.108 TRANSFORMS OF OTHER L-CLASSES Chap. IV
4.7. Theorems on resultants.
‘Tunonem 76. If f(x), Fle) are transforms of L?, LY, and g(2),
G(x) of L’, L?, then (2.1.9) are transforms in the sense that (2.1.8)
holds for all values of 2.
Proof similar to that of Theorem 64.
Tanonma 77. If fla), F(a) are transforms of L?, L*, and g(x) is L,
then (2.1.9) are tranaforms of L?, L?’.
Proof similar to that of Theorem 65.
‘Tuxonna 78. Let f(z), F(z) be transforms of Le, L*, and g(z),
O(c) of Le, Le, where
tH >i. (47)
Then F(x)O(z), Fea f orte—nev
are transforms of classes L”, LP renpectoely, where
Po ae
‘That the resultant of f and g belongs to L? follows from Lemma B
of §4.2, with 1-A = I/p, 1—» = 1/g. That FG belongs to L*
follows at once from Holder's inequality in the form
fire ds<(f LF dz)? ( f 1@ie ae).
‘The condition (4.7.1) implies that p 2. That the
‘Young-Hausdorff theorem fails for p > 2 follows incidentally from
‘Theorem 80. For, if/(2) belongs to L?, not only is |F(2)/*' integrable,
but so is [2*(2)|?|2|?-%; and hence so is
IF@)rer* (p 0,
feo) = 0, and let {f(z)}>x°-* (1 < p < 2) belong to L(0,0). Then
F(z) belongs to 2.
Since f(z) in non-increasing, and f(co) = 0, the integral
Fe) = He) = |(2) f soreoseyay
converges for every x > 0. Let
re /() [ape ay+/@) Jones ay
= F@+he).
By the second mean-value theorem
11a JE) fomnar~ Jee
Hence mei <2/2) 20),
and f Ua) de | Pow-tde > Preyer te,
Sie) < Kr-o-si,
[flair < Kifle)px?-*.
‘THnonEM 83. Let f(x) be the integral of order (2—p)|p of a function
$(x) of 12, Then Fix) exists and belongs to L?.
Let oye) = /| () f $sin(zt-}-m/p) at.
‘Then, by Theorem 80, 2'-#°®,(z) converges in mean (p) to g(x) say.
Let G(x) be the cosine transform of g(z). Then G,(z) belongs to L?’.
= lim2 f Sn ae f $(t)sin(at+ n/p) dt
= im? { gcyae f Seven tel) a,
the inversion being justified by uniform convergence.
+ See Titchmarsh, Theory of Functions, p. 396,412,413 TRANSFORMS OF OTHER L-CLASSES 15
The inner integral is
api" 0. Then F(z) belongs to LP for
P <2.
rapt Azh for 2 < 1/h, the left-hand side is greater than
ap
A few irey dz,
»
Hence J a?" |P(2)|P" de = O(We-tr'),
‘
Let #6) = f nF (@)/P de.
See Titchmarsh (12).ne TRANSFORMS OF OTHER Z-CLASSES —chap.1V
‘Then, if B <0 if 1—B—aB-+A/p < 0, ie. if
P
> ryap—T
Similarly for the integral over (—£,—1). This proves the theorem.
A particular case, corresponding to the original theorem of Bemn-
stein, is that if the condition is satisfied with a> 1/p, then F(x)
belongs to 1(0,c0), so that the Fourier integral
J F(det dt
is absolutely convergent for all values of x.
To show that the range for f in the above theorem cannot be
extended, consider the even function
i
So = ae
where 0 2h
Vla+h)—fle—)| = hf e+an)| (—1<0-<1)
<2hife—W)| < if Ge),
‘since |f'(z)| is positive and steadily decreasing. Hence
Fite —pee—myp ae = ofte F irae ae)
a a
(> 0),
e( f esr de + fes a) = O-),
ms i
Also
Piter-ste—nirae = of f iz) = O(b-2"),43, TRANSFORMS OF OTHER L.CLASSES ur
‘The conditions of the theorem are therefore satisfied with a = 1/p—a.
Hence F(z) belongs to ZA for B > 1/(1—a). Butt F(z) ~ Ke? as
2 +00, 80 that F(z) does not belong to Z3-0,
In the case a <1, p = 2 it is possible to put the theorem into a
form in which it is reversible.
‘Tunonem 85. If fz) belongs to L, the conditions
fte+ay—ste—myp ae = Olh}*) (0 1/(a-+4), the case
p =2 of the above theorem. But this last step is, of course, not
reversible.
4.14, Mellin transforms of the class L”. Let us denote by 2”
the class of functions f(z) such that
fre 0, and that
fle) is not null. Let
a1 f Man
= [ Pa=vsin w>o,
as before. Then eee Ih
uF ie)
ven =% [A aa>o
if a is large enough. Let 0 0, (Vi >1,
var
winery =|p f de] < PUI < 0,
‘
while if U > 0, [P| <2,
(W+iV PCV] < (+ UP +1 < 42,5.14, 5.15 HILBERT TRANSFORMS MMs
f Catitey—ney 4.
tT
= froasen< fronds (6144)
From (5.14.8) and (5.14. oy it follows that
F one 4.
aH
(ite? (ety"+- eos bpm Soy sin bpm al?
R =i pliant A Hite fk “leet Ae? al ta
(eh) = var Sawer > Mai
for sufficiently small y and all z. Hence
WP
[ie A); g(z,A),
defined as in (3.1.2), are transforms of 4; and the former is also L?.
‘They are therefore transforms of L”, L?’, and Theorem 75 gives
a
[ Peete) de = f fle)a(—z,») de.
A se
‘As in the previous section
Lim .(p'g(—z,0) = o(—2),
and the result follows.
Tanonzm 110. If f is L? (1 A)
is the transform of
1 2 f ao tinaaaicantene 2) inde 2) ay,
‘The proof now goes as before, but to complete it we want
— t For ti28—2) gy
tim f serde Foe ZO= a
Now
| fasta
a8
<3 [wie Ae + Jt ) eeiau0
as 50, for any fixed x; and, as before, its mean p'th power is
bounded, The result therefore follows from the lemma of § 5.12.VI
UNIQUENESS AND MISCELLANEOUS THEOREMS
6.1, Uniqueness of trigonometrical integrals. Tux classical
uniqueness problem for trigonometrical series is to show that if
day t+ 5 cosnz-+b, sin nx) =
for all values of « in (0, 2n), or all values with some exceptions, then
4a, = 0, by = 0 for all values of m,
‘The corresponding problem for integrals is to show that if
{ falueosay-+b(y)sin zy} dy = 0 (6.1.1)
in some sense or other for all values of x, possibly with some excep-
tions, then a(y) = 0, 6(y) = 0 almost everywhere. A more general
problem is to show that if a given function f(x) is represented by a
trigonometrical integral,
J {alyoosay-+b(y)sin ay} dy = fle), (6.1.2)
in some sense, then the integral is necessarily of the Fourier form,
i.e. in some sense
ay) =2 f fereoszyde, O19)
Owing to the symmetry of the Fourier integral formula between a
function and its transform, in the integral case this is not, formally,
anew problem. It simply amounts to the question whether a(z) and
0(z) are representable by Fourier integrals; and in some cases the
answer follows from theorems which we have already proved.
Suppose, for example, that a(x) and 6(z) belong to L(0,<0), and
that ©
{ falyeos2y-+b(y)sin2y} dy = 0
for almost all values of x. Adding and subtracting the formulae with
and —z, it follows that both
fatyeosay dy = 0, { y)sinzy dy =61,62 MISCELLANEOUS THEOREMS 153
for almost all x, By Theorem 14 (for an even function)
a 7
2 a FH
tim? (- iooseu au f aly)cosuy dy = a(z)
for almost all x, and in this case the left-hand side is 0 for allz. Hence
a(z) = 0 almost everywhere. Similarly B(z) — 0 almost everywhere.
‘Theorem 22 can be used to give the same result.
Suppose, again, that a(y) belongs to 1°(0,0o), and that
f ayjeosay dy = 0
for almost all values of x. Since the limit and mean limit of
sequence, if they both exist, are equal almost everywhere, the cosine
transform of a(z), in the sense of Theorem 48, is null, and hence a(x)
jis the mean limit of a sequence of null functions, and is therefore null.
‘The uniqueness theory of Fourier series suggests a different type
of theorem, in which the possible values of x for which (6.1.1) fails
are much more restricted, but in which a(x) and 6(z) do not neces-
sarily belong to L-classes. The main difference between the theory
for series and that for integrals is that the convergence of 5 a, cosnz,
for example, in a set of positive measure, implies that a, > 0; but
the convergence of ~-
f ajeosay dy
does not imply that a(z) > 0 as x-> oo; for example, the integral is
convergent if a(z) = e*cose**.
6.2. The expression
Tim SE LA)+O(
mo
is called the generalized second derivative of (z). The uniqueness
theory of Fourier series depends on the theorem of Schwarz (see
‘Titchmarsh, Theory of Functions, § 13.84), that if (2) is continuous,
and has at all points of an interval’ the generalized second derivative 0,
then ©(z) is a linear function in the interval. Here we shall proceed at
‘once to the general problem with f(z), atid use
TuxoreM 111. Let ©(z) be continuous in (a,b), and have at every
‘point of this interval a finite generalized second derivative f(x), which
+f De ta Valido-Pounsin, Coure danalyee infnitésimale, 1014 e.
L
—20(2)
(6.2.1)184 UNIQUENESS AND Chap. VI
belongs to L(a,b). Then
2) = fw frre dvtataz (@ co, in any manner.
4.2. The most obvious source of such results is in the formulae
(2.1.11), If & is an integer, the transform of {F(z)}* is formally
@ny-*44,(2), where
Bale) =f flea daca f fleet thy) dy
‘We can deal with such integrals by means of the following lemmas.
4} Titehmarsh (2). $ See Hardy, Littlewood, and Polya, Inequalitia, pp. 198-203,42 TRANSFORMS OF OTHER L-CLASSES 2
Lewaa a. (Young’s inequality.) Uf f(z) and g(z) belong to LMt-»
and I~ reapectively, where d > 0,» > 0, A+p <1, then
| fre
< (fergie as)" F ypsovae)'( Ff igre-w ae)
Holder's inequality for three fonctions is :
[fone ( wma | wwe ra
where a-+B+y = 1,a>0,B>0,y>0. Putting
WRixPy, wp If, Ix = gl,
, the result follows.
Luana p. Let 9,(/) = (f ute)? a)”
4 He) = f Mote a,
then San-r-0(8) < Syn-w/)9u (0
‘Young's inequality gives
° a4
wwe < ( F yoo mae—ow a).
Sia DSi).
Hence
F eomrode< f yonende F gte—v 0-0 dex
Su al POMS (GOALS
: itt
ON ee ec
and the result follows.
Luanna y. If f(x) belongs to L*%-», then gy(z) belonge to L?, and
F tosertae <( f itereme-van)*28 TRANSFORMS OF OTHER L-CLASSES —— Ohap.1¥
We have x(t) = I HObe-ale—t) at,
and, applying the previous lemma repeatedly,
Sa) e) < Sy/(0—AGY) Peay (o-gy)
< Sy tgy Gea)? < ~~ < Pya- yoy},
the result stated.
4.3. Prooft of Theorem 74 for p = 2k/(2k—1). Suppose first
that f(z) and g(z) belong to L*, and are zero outside a finite interval.
Then °
em J autem) dy
satisfies the same conditions, and (e.g. by Theorem 64) its transform
is F()G(e).
Repeating the argument k—1 times, and making all the functions
equal, we see that the functions (F(x)}* and (2n)-##¥4,(z) are trans-
forms of the class L, Hence
f LF(@) de =
This proves (4.1.2) for the special class of functions considered, and
p= %/(2k—1).
Now let f(z) be any function of the class Z**-, ‘The function
equal to fiz) if a < |x| 00, we obtain
f FeO) Flea) de < 7) anl(f + i ) eset» ae)
‘The right-hand side tends to 0 as aco, b->co. Hence F(z,a) con-
+ The argument is analogous to that of W. H. Young for Fourier series. Fora list
‘of Young's papers see Zygmund’s bibliography.43 TRANSFORMS OF OTHER L-CLASSES cy
verges in mean, to F(z) say. We have
fre ae— iim f teeee
co of
Lf Sate f semen ay
ae J iu
(since _f dy = lim. f)
=z f sw) af uy) Ea,
iw
and by the Parseval formula (for g, 4’) the inner integral is equal tor
(2m) f (ve dv.
Hence we obtain a
Wie : ff *
= ay [ eve dv = so0 a
We, J O) vf Ko) do = 55 J Ge) av J Stuye dy
> { G(X) F(v) dv,
since G(v) belongs to L* over (0,x). Hence the result,
Ifp = 1 and q = 1, then P = 1; see Theorem 41.
4.8. Another extension of Plancherel’s theorem. We shall
next obtain a generalization of Plancherel’s theorem in a different
direction, due to Hardy and Littlewood.
+f Hardy and Littlewood (1).108 TRANSFORMS OF OTHER L.CLASSES Chap. 1V
‘Tunonsa 79. If |f(2)|?xt-* (q > 2) belongs to L{—co,00), then F(z),
the transform of f(z), exists, and belongs to IA; and
J Feeds < K@) f Yleyieie-*ae.
(i) Consider the case g = 4.
Suppose first that f(z) belongs to L*, and vanishes outside a finite
interval. Then F(z) is L* and bounded, and ,(2){¥(z)}* is the trans-
form of ©
He) = f fudfe—v) dy,
which also belongs to L*. Hence
an fF e@)itde = f pte) de.
Now f(z) |-ig(z), where g(z) belongs to L‘. Hence
_ f woe, fF wey) 1g
v= | yee! J iwi eu lew
iene < fm Mv ay f
irae
A f mle lo@e—y)P dy.
jet J “weit
‘Hence
f nwpiiote—wi8
f ee Jf veiigiemuk
J
J fi Ae ae
J
i
=A
A
f {locniivr tote—y)ite—yi
40S aS aaa | ee
=A wontar f ai gee = 4 f Iow)I* dy.48 TRANSFORMS OF OTHER 1-CLASSES 108
Hence firwacca f yeiterae
The proof now follows the usual lines. Let f(z) be any function
such that {f(z)}42* is L. Approximating to f(z) over (a,b) by a
sequence of functions of the special type, we prove as in §4.3 that
f rea—Re, outa 0o, we obtain
J ren—rearrde
y2
5.8. Taxon 98, Let f(z) belong to L*(—20,00). Then
Se) = f.@)4f-@),
where f,(2) belongs to L%(—c0,00), and is the mean-square limit of an
analytic function f,(z), regular for I(z) > 0; and similarly f_(z) is the
mean-square limit of f-(2), regular for I(z) <0.
Let F(z) be the transform of f(z), and
f@)
e ¢
1 teu du, @) = a
Ts J Flue du, fl) gen J Poet du.
Plainly f(z) and f_(z) are regular for y > 0, y < 0, respectively. The
rest of the theorem follows as in § 5.4.
‘Tueonen 99. Let f(z) belong to L%(0,00). Then
F@) = fs) Ho@),
where fis(z) belongs to L*(0,00), and is the mean-square limit as
argz—>-+0 of an analytic function fis(2), regular for argz > 0; and
similarly f.(2) is the mean-square limit of f..(2), regular for arg <0.ase CONJUGATE INTEGRALS Chap. V
‘This may be deduced from the previous theorem by putting z (of
the present theorem) — ef; or deduced directly from Mellin trans-
forms. In fact
itis
,
fol= se [ Borrd, foe= tf Berra
5.9. THronEm 100f. If f(z) belongs to L(—co,c0), then
f fet0—fle—b)
7
exists for almost alt values of x.
‘We may suppose without loss of generality that f(z) > 0. Define
U(z,y) and V(z,y) by (5.3.5), (5.3.6), and let
90) = Veta == f f0a y>o,.
From its definition it is clear that U(z,y) > 0
Let We) = et we Ues-WFeew),
Since U(x,y) > 0, |¥(2)| <1. Hence, as y-> 0, ¥(2) tends to a finite
limit for almost all x (Theorem 94); and this limit can be 0 in a set
of measure 0 only, since U(z,y) tends to the finite limit f(z) almost:
everywhere, Hence (2) tends to a finite non-zero limit almost
everywhere, Hence (2) tends to a finite limit almost everywhere.
Hence V(z, y) tends to a finite limit almost everywhere. ‘The result
thon follows from Theorem 92.
5.10. Hilbert transforms of the class L?.
‘TuronEs 101. Let f(z) belong to L?(—00,00), where p > 1. Then
the formula
(6.10.1)
defines almost everywhere a function 9(2), also belonging to L?(—co,0).
The reciprocal formula
joa —} f Settee (5.10.2)
Plessner (1). T believe that thie version of the argument is due to Little.
‘wood.510 HILBERT TRANSFORMS 133
also holds almost everywhere; and
Fie ae
You might also like