You are on page 1of 19

The Prehistory of the Hardy Inequality

Author(s): Alois Kufner, Lech Maligranda and Lars-Erik Persson


Reviewed work(s):
Source: The American Mathematical Monthly, Vol. 113, No. 8 (Oct., 2006), pp. 715-732
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/27642033 .
Accessed: 18/12/2012 02:30
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

of the Hardy

The Prehistory
Alois Kufner,

Lech Maligranda,

Inequality

and Lars-Erik

Persson

The development
of the famous Hardy inequality (in both its
1. INTRODUCTION.
has its own history or, as
discrete and continuous forms) during the period 1906-1928
we have called it, prehistory. Contributions
of mathematicians
other than G. H. Hardy,
such as E. Landau, G. P?lya, I. Schur, and M. Riesz, are important here. In this article
we describe some of those contributions. We also include and comment upon several
facts and early proofs that are not available inmany references on this subject.
We consider the following statement of theHardy inequality-, the discrete inequality
asserts that if p > 1 and {ak}^? is a sequence of nonnegative
real numbers, then
oo

(1)
the continuous
inequality informs us that if p > 1 and / is a nonnegative p-integrable
function on (0, oo), then / is integrable over the interval (0, x) for each positive x and

F
Several

(x~f f(t)dt) dx-(~~[) f

introductory

f^)Pdx'

(2)

remarks are in order:

(1) and (2) are the standard forms of the Hardy inequalities that can
textbooks on analysis and were highlighted first in the famous book
and P?lya [16].
Inequalities by Hardy, Littlewood,
to
the
class
of step functions one proves easily that (2) implies
(ii) By restricting (2)
seems
to
been mentioned
fact
have
first by Landau (see [11, p. 154]
This
(1).
important
and section 8 of the present paper).
?
l))p in both (1) and (2) is sharp: it cannot be replaced
(iii) The constant (p/(p
with a smaller number such that (1) and (2) remain true for all r?v?lant sequences and
(i) Inequalities
be found inmany

functions,

respectively.

(iv) Inequalities

(1) and (2) imply

the following

>
?^=i an < ?? and an 0, thenJ27=\ (?ELi

weak forms

of (1) and

(2): if

<
< oo
ak)P ??> and if f0??f(x)p dx

ter
dx < oo, respectively. We adopt Hardy's
and f(x) > 0, then
/0?? Q f* f(t)dt)p
in the historical development
because
it has been significant
that we are
minology
going to describe (see, for example, the comments at the end of Hardy's paper [9]).
(1) and (2) together with statement (iii) imply the important in
(v) Inequalities
formation that the discrete Hardy operator h and the continuous Hardy operator H,
defined by

n
h({an})=\-Yjak\,
{ k=\ J

Hf(x) = -f
x Jo f(t)dt,

=
the spaces lp into lp and Lp into Lp, respectively
(p > 1), and each has norm p!
?
spaces consisting of
1). Here, as usual, the spaces lp and Lp are the Lebesgue
p/(p

map

October

2006]

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

715

all sequences a ?
almost everywhere

Ml/, :=
We mention
Riesz [32].

classes modulo equality


[an} of real numbers and all (equivalence
functions / on (0, oo), respectively,
such that
of) measurable

< oo, ||/||Lp:=


(j

(?>?r)

that the spaces

were
lp and Lp

\f(x)\?dx

introduced

and investigated

<

oo.

in 1910 by F.

and applied in analysis and in


(vi) Inequalities
(1) and (2) have been generalized
the theory of differential equations. A lot of these developments,
and
generalizations,
in
have
been
discussed
more
books
[16],
and,
[22],
[29])
applications
(e.g.,
recently,
in the historical survey paper [21].
The aim of this paper is to complement
the existing
to this
literature devoted
describ
instance, what is described in, say, [16], [21], [22], or [29]?by
subject?for
that finally led G. H. Hardy
ing some important steps in the scientific developments
to (2) and, subsequently,
to the proof of (1) in his famous 1925 paper [11]. The story
that we are going to relate ismuch more dramatic and intricate then we ever imagined.
We hope that the reader will find it equally captivating. The tale begins over a hundred
years ago and plays itself out in the period 1906-1928.

2. HARDY'S ORIGINAL MOTIVATION: THE HILBERT INEQUALITY. In


this section we briefly discuss an inequality that was discovered
in the early 1900s
by David Hilbert (see his paper [18] from 1906) and is closely related to the discrete
Hardy inequality (1). Moreover, we confirm that this inequality was, in fact, the main
source of motivation
for Hardy when he started the research described in this paper.
In itsmost basic form theHilbert
inequality reads: ifYlm=\ am < ?? anc^Yl =\ b\ <
>
>
0 and bn
oo, where am
0, then the double series
am on

EE

n=\

converges

(the weak

form). More
GO

n?\

.m +
m= \

the inequality

precisely,
i OO

m? \

/OO

\ra=l

'

OO

\?=1

'

of the sharp constant tz, as well


holds, with re as the sharp constant. The determination
as the integral analogue of (3), are due to Schur [35] (in Hilbert's version of (3) the
constant 2tt appears in place of n). We remark that the following more general form
of (3) is sometimes referred to in the literature as the Hilbert inequality:
oo

oo

EL?t^i?)
,=,- 1 ni= \

fe<
\m

oo

= \

M
/

{/P

\n

oo

? \

XIP'

?
where p > 1 and p' ? p/(p
1). Hilbert himself was not even close to considering
this case (the //?-spaces appeared only in 1910). It was M. Riesz and G. H. Hardy who
took the first steps towards a proof of (4). In fact, Hardy acknowledged
in [12] that
Riesz had pointed out to him that his result in the paper [10] (see Theorem 3) actually
716

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

implied the weak form of (4). In any event, the Hilbert inequality and its evaluation
into what today are called Hilbert-type
inequalities have their own interesting history,
of which we recall here only a few basic facts from the beginning of the story.
In his research on solutions to certain integral equations Hilbert was led to study
special bilinear forms with sequences of real or complex numbers as entries. For details
we refer the reader to the book [19] or to the section on integral equations edited by
in Hilbert's Collected Works [20]. Moreover,
several ideas
[17, pp. 94-145]
Hellinger
can be found inHermann Weyl's
from Hilbert's
lectures around the period 1906-1908
1908 dissertation
the
[39]. In particular, on page 83 Weyl presented and discussed
following remarkable formula that had been discovered by Hilbert:
Nyv

Nyv

i
1
+
n +

= 1 ra=l

1r

'

271J-n

(ak sinkt

y^(?1)*

N N n h
umun

to be zero). This

formula

1/2

\~~^ \~^

^*

m + n

E*?

yv i '

?-s n
m
?-?
n? \ m = \ +

implies

the finite

1/2

(6)

E^2

The derivation of (6) from (5) is found in [16, pp. 235-236].


Other proofs of the Hilbert inequality were given by Wiener
(the latter proved it with the help of the so-called Schur test). We
Toeplitz's method, which is based on the identity
Nyv

(5)

dt

bk coskt)

k=\

?
= m, then
m) is understood
(if n
l/(n
version of the Hilbert inequality

LE"?
=1 m =

ambn

[38] and Schur [35]


should also mention

r 271

~
2n J0

dt

(t-Ji)

?aBe/n'X>me''m'
n=\

([37]; see also [36, p. 165]). In addition, Fej?r and Riesz [5] (see also [16, p. 235]) and
P?lya and Szeg? [31] gave proofs that exploited the theory of analytic functions. We
are most interested, however, inHardy's method for proving (3).
Note that (6) implies both (3) and a weaker cousin that was of special interest to
> 0), then the double series
Hardy: if^2^=x a2 < oo (an
am au

m + n

EE

converges.
It was
Y^n/=\
which

In fact, Hardy wrote

proved

by Hilbert,
(a" ?

]L/T=i ^7T?7
is one of the

in the introduction

in the course
0)

of his

^s convergent

theory
whenever

to [10]:

of integral
Y^=\
in the

an

that the double


series
equations,
^s convergent.
Of this theorem,

and most
beautiful
series of positive
simplest
theory of double
own proof, which
at least five essentially
different
Hilbert's
proofs have been published.
is outlined
in his Inaugural-Dissertation
series,
upon the theory of Fourier's
by Weyl
depends
[38], and two more by Schur
[35]; but none of these
[39]. Another
proof was given by Wiener
as
as
and
be
desired.
is
might
simple
elementary
proofs
terms,

October

2006]

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

717

To

these

four

a fifth,
recently
that Hubert's
theorem

I added

proofs

I observed
simplicity.
seems of some
which

first

which

to me

seemed

is an immediate

to lack nothing
of another

corollary

of

this

theorem

in itself.

interest

(See Theorem A in [9] but note that this result was already proved in the 1915 paper
[8]; compare with Theorem 2 in the next section.)
in some of the other papers that we are now
Hardy made similar pronouncements
seems
to
it
discuss.
Therefore
going
completely clear that Hardy's original motivation
when he began the research that culminated
in his discovery of inequalities
(1) and
was
to
weak
the
Hilbert
(2)
prove (the
form of)
inequality.
In [8] Hardy stated and applied
to the Hilbert theorem:

3. THE 1915 PAPER.


has obvious connections
Theorem

1. Let an > 0 and An = YH=\


4

oo

(l) >

\ 2

?1

>
(ll) *?

^?f

ak- The convergence

oo

the following

of any of the three series

oooo

!\ n

theorem, which

(ill) >

>

? ,n + m

implies that of the others.


Hardy

actually proved

the equivalence

of the convergence

?a

Ja
/W/W,

-n

a > 0, / is nonnegative
the estimates
establishing

-dx

dy,

on (a, oo), and F(x)

and integrable

where

x+ y

Ja

Ja

/?OO

OO

of the three integrals

fx f(t)

dt, by

and observing that I\ < h < 2IX. In the proof of the sequence case he said only that
"the proof of this theorem is much like that for integrals" [8, p. 164], but he probably
realized later that the sequence case is more delicate, because he considered
it again
in the 1919 paper [9]. However,
the estimation of (iii) by (i) can be done, in a manner
similar to what Hardy did in [8] for integrals, as follows:
N
^-\

N
^-^v

n= \
?-!^m

? ? N
aman

n
^-?\

^?\

+ m n~~=

?-??-im
n?\ m ?
N

aman
\

+
n n~?

^?\
m?1
^?n

<2)an-)am

the following
Hardy also established
= 2:
case
of
for
the
p
precursor
(1)

Moreover,

718

?-f m -f

N
?

an

^?>

2)an

result, which may

be regarded

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

as a

113

J27=\

(An/n)2,

Proof.

Let

w^tn an > 0

convergence
of the series YH?=\?yi
=
where An
ak.
??=1

2. The

Theorem

ELi

the nonincreasing

{a*} denote

at- SinceK

of the sequence

rearrangement

- An andE^li

al

implies

that of

{an} and put

xt is sufficient to prove

L^i?)2>

that J27=\ (A*Jn)2 is convergent. Thus, we may assume without loss of generality that
the sequence
1, it is enough
according to Theorem
{an} is nonincreasing. Moreover,
is convergent. The last series can be written in the form
to prove that Y^L\ anAn/n
^ ? Y1T=\ Ylm=\ aman/n, and itwill be convergent ifY1T=\ ^ < ??' wnere

Moreover,
find that

by making

obvious

??
1 i1 ^r-^\
am

aman

! oo

\m

= l

k<lL<k+X

! / oo

\m =1

ra=l

inequality, we

^?a

m= \

k<^<k+\

n(x>?)

the Cauchy-Schwarz

and using

estimates

i \"^v v~a

(k=l,2,...).

S*=EE~
k<%<k+\

1/2

/ oo

\ra=l

J/2

(?[x>?) =?2>
m? \

ra=l

Therefore
OO

/:=1

so S < co and the proof

OO

m= \

k=\

00

2.61, where

is not sufficient

to deliver

a proof of

inequality
??

n=\

n?\

?(3/2)

m?\

is complete.

1. We see that the foregoing argument


Remark
=
2 directly. It gives only the (weaker)
(1) for p

with C =

oo

oo

? denotes

the Riemann

zeta-function.

him to do this work,


2. Hardy also pointed out the fact that had motivated
theorem (Theorem 1), Theorem 2 is essen
namely, that, by virtue of the equivalence
theorem in the case p = 2.
tially equivalent to (the weak form of) Hilbert's

Remark

3. In this paper, as we have already mentioned, Hardy also stated and proved
Remark
1 and 2. However, he formulated his results for in
continuous versions of Theorems
or
>
a
He continued this
or
of in the final form
0
instead
with
tegrals f
f*.
f0??
/*
we
no
but
have
clear
and
[10],
explanation for why he first
practice in the papers [9]
in
formulated his result in the
case.
he
In
the
1925
studied only this
paper
any event,
to
the
earlier
formulation.
connection
the
natural way and also explained

October

2006]

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

719

4. THE 1919 PAPER.


Probably the most important contribution of [9] was the new
2
it contained.
In fact, the paper also includes a first proof of
Theorem
that
of
proof
? 2 and even
case
the
for
p
(1)
gives the best constant 4 (see Remark 4),
inequality
not
did
make
mention
of this fact.
although Hardy
explicit
2.

New proof of Theorem

An\2

It is clear that

?an -\-

An -

\2
an\

= Aa +2

+ 2l

<2al

(? An

x2
-an

-4

Hence,

/ A \2

E(?)
for

each

N.

Moreover,

N n A

A2n_{)

(A2 -

A2

(A2n

A2n_{),

so

A2
N

(N-l)-N

1
A2

V???

w=1 n(n +

N ? A \2

<

A2

2-3

this estimate

a2

A2

1-2

substituting

(7)

n
A2

By

-iv

(A2n

/ A \2

?4E*^?(^
-2anAn

l)

into (7) we obtain

N /A \2

-j

-?E^E;^:.
which

yields

/7=1

Obviously

(8) implies

the statement

X'

in Theorem

/2==1

2.

Remark
4. Hardy had plainly not yet realized that inequality (8) can be used to derive
the discrete Hardy inequality (1) for p = 2 and do so with the sharp constant 4 (more
on this in section 6).

720

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

Remark
5. In [9] Hardy also stated some results for the continuous case (e.g., a gen
the most important point for the history of the
eralization of Theorem
1). However,
was
his claim that
Hardy inequality

dx<A i f2(x)dx
j (- j f(t)dt\
a > 0 and that 4 is the best constant. In fact, it can be shown that this estimate
? 2.
Hardy did not give a proof of this assertion but only referred to
implies (2) for p
the proof in the discrete case that we have just presented. This can seem a little curious
if we take into account the information in Remark 4.

when

5. AN IMPORTANT CONTRIBUTION FROM RIESZ. In his paper [10] from


1920 Hardy
Dr. M.

referred to a letter from Marcel


to whom

Riesz

which

is equal
His
preferable.

Hardy

I recently
in simplicity

to mine
proof

proceeded

communicated

to formulate

Theorem
seems

and which

an interesting

suggests

naturally

the Riesz

and wrote

Riesz

to both

En=\/<7
\
= 1 ^
//
(v)P

found

of us more

natural

weak

and

proof,
therefore

.Then,

form:
convergent,

Y1T an

(n + 2)~p +

another

viz.

result in the following

> 0, and
3 (M. Riesz).
If p > I, an
=
(An/n)p is convergent, where An
YH=\ ak

=
Proof. Let <$>n n~p + (n + l)~p +
we have for each N (with A0 = 0)

2 at once

generalization,

Theorem
Y^

[10, p. 315]:

by partial

then

summation,

- *?+ >=
<-.)*?
A^+i
E<A?
n=\
n=\

= a?
En=\n=\
N

n=\

n=\

<

Moreover,
oo

4>n < n~p +

From

these estimates

/
Jn

x~p dx = n~p H

n-(p-(P-D

and the Holder


Up

T-

< -t?n-i'-?.

inequality,1
i m

i/<?

(P:)

t^-iP)

("l-J+H-

and referred to a paper [23] by


'in [101 Hardy remarked that inequality
(9) was probably due to Holder
book it is said that "Holder states the theorem in a less symmetrical
In the Hardy-Littlewood-P?lya
form given a little earlier by Rogers"
[16, p. 25] and uses the name Holder
inequality. We believe that these
later began to call (9) the Holder
words by Hardy could have been important when mathematicians
inequality.
Landau.

in 1888, one
second author pointed out in [28] that an equivalent variant of (9) had been proved by Rogers
is again only an equivalent variant of (9) and is different
year earlier than Holder produced his version, which
from Rogers's.
this classical
[32]. Therefore
Inequality (9) in this precise form was proved in 1910 by F. Riesz
inequality or the H?lder-Rogers-Riesz
inequality.
inequality could have been called the Holder-Rogers

The

October

2006]

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

721

it follows

that

EM

?tt?E?.(v

Hence

(10)

tm*fc)'?<
argument actually yields more than what Hardy
(10) implies the correctness of (1) with the constant
l))p.

6. Riesz's
Remark
Theorem 3; namely,
in place of (p/(p -

in
formulated
?
l))p
(p2/(p

In [10] Hardy observed that estimates in the proof of (10) were


1920 PAPER.
?
constant Cp =
that
the
could be improved merely by
and
fairly rough
(p2/(p
l))P
out
In
he
the
Riesz
estimates.
that Cp could be replaced
particular,
pointed
refining
The argument that confirms this fact he
with the strictly smaller constant (p?(p))p.
received in a letter from Schur [34]. Obviously, Hardy already believed by then that
?
\))p was the sharp constant, even if he did not claim so explicitly (see [11, p.
(pl(p
reason for this was surely Schur's remark in the same letter that this was at
One
154]).
6. THE

true

least

for

2.

Hardy himself did not directly observe that inequality (8) from his 1919 paper could
?
2 with the best constant C == 4. We
actually be used to derive inequality (1) for p
don't know for certain what Schur's argument was, but the information in the next
as follows: Let cn ? 1 ? 2/(n + 1)
section convinces us that it went approximatively
and

for m

2,

ai

...

3,

?2

let
*

am

<2(/V-l)m + l

Then

am+i

(Cl+---+Cm)iyj

YH=\

bk- Dividing

a^m

a2m

b2,

..

bN.

in place of N we obtain

N
/B \2

{C(N-\)m+\ +

am+2
=

?(/V-l)m+2

>

Bn

(8) with Nm

from inequality

4m
J2bl

where

b\,

+(cm +

/B \2

+'"

l+---+C2w)iyJ
''*
+ CNm) ( "TT ) ,

-+ oo, we find that


by m and letting m
(ci +c2-\-\-cm)/m

(cm+1 + cm+2 H-h

-+

c2m)/m ->

1,

1,

etc.

722

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

Accordingly,

= 2 with the best constant.


which, in particular, implies (1) for p
A significant element in the paper is the formulation of the following
version of (2) (see also [11, p. 150]):
4. If p >

Theorem

and the constant

> 0, and
f

1, a > 0, f(x)

(p/(p

l))pis

is convergent,

f(x)pdx

preliminary

then

sharp.

he pointed out the


Hardy did not prove inequality (11) in the 1920 paper. However,
fact that from consideration of the function f(x) = x~(1//?)~6, where 6 is a sufficiently
?
small positive constant, it follows that the constant (p/(p
l))p is sharp. He also
best constant in the discrete case
claimed that he could prove that the corresponding
?
to assert
could not be strictly less than (p/(p
l))p, but he hesitated for the moment
that (1) held with the constant (p/(p
l))p.
The letter [24] from Landau to
TO HARDY.
FROM LANDAU
7. A LETTER
that
this letter was officially pub
It
is
1921.
June
from
dates
21,
surprising
Hardy
lished in [26] five years later than the letter of Landau to Schur (see also Remark 8).
The reason for this long delay is not obvious, but it is clear that the contents of the let
?
ter [24] are of interest, for it gives a proof of (1) with the sharp constant (p/(p
l))p
and this had not been published prior to that time. The main result proved in this letter
reads:

5. Let p >

Theorem

holds for all N


Inequality
Proof

=
1, an > 0, andAn
YH=\

inN or N =

oo. Moreover,

the constant

the Hardy-Landau

called

(12) is sometimes

ak- Then

the inequality

is sharp for N =
inequality

oo.

(see [27, p. 188]).

For y > 0 we have


yp

py + p

i > o.

This is seen to be the so-called Bernoulli


inequality when we write it in the form
?
=
> 1+
yx/y2 we find that
1). By using this elementary inequality with y
yp
p(y

yf
October

2006]

pyy2~x + i.p- ^2

the prehistory

of the hardy

> o.

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

723

set yi ?

We

=
bn and y2

(p

\)Bn/(pn),

where

Bn

JZ/Ui ^'

anc^ ^earn tnat

so

??-?^rE^?fr?.f^yE^r.o.
\

H=l

Moreover,

pbnBp~l

?=1

??_!)

pBp~x(Bn

>

??

#?_,,

whence

by

summa

partial

tion
N

/ f? \ p~1

E*(t)

^k-c,)?
/v
>
E'K^-srrW)^'-"^'^
we discover

the two inequalities

Combining

?
where cn = p(\ +
p +
^)~p
the previous section, putting
b\

b2

bm

1 ->

that

n -> oo. Next we use the argument

1when

au

bm+i

?
b(N-\)m+[

and replacing

TVwith

^(W-l)m+2

m to conclude
TV
N

bm+2

?>^m

=
Z?2m

a2,

...

from

<2/v

that
A \ P

(^iySo,i(c,+C2...+0(^y
+

(cm+i + cw+2 H-h

'

(C(7V-l)m+l

?
c2m) (

C(#-l)m+2

-l2

"' "
+

??
<WVm) (

+ cm)/m ?> 1,
Dividing
by m and then letting m ?> oo we note that (cj + c2 +
+ c2m)/m -> 1, and so forth, which means that (12) holds for all
(cm+i + cm+2 +
finite N (hence remains valid when N -> oo).
In order to prove that (p/(p ? l))p is the sharp constant for N = oo we consider
? 1
=
n~]/p~? (0 < s < 1
an
/p). For this choice of an
n nn

An=?Jk-^-s>
fr?1
k=\

x-1,p-edx

^1

'

l-l/p-eV

724

.(?l-l/^_1)>_P_(nl-l/P-i_1))

p-\

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

implying

that
p

Ar.

>

n J

\p

I)

I n~{-ep

( 1

n{-]/P-?

nl-l/p-e)

>

(n-l-ep

pn~2+i/"+s-sp).

Furthermore,

.,4?)'(t-:-^).
where CNi? -> C as N -> oo for any s > 0 because

2?

l/p

e+

ep >

I. Thus

?(^/?<>(^)'(-^/?<)-(^)'
since

a?

5Zn=i

Note

established.

tion is thereby
e =

~>

^~1_e/?

5Z?=i

ooasN-^

oo

and

that the foregoing

0+.

The

calculation

sharpness

still works

asser

when

0.

Remark

7.

sequently,
n.

con
In his letter [24] Landau also mentioned
that equality in (12)?and,
if and only if an = 0 for all
in the discrete Hardy inequality (1)?occurs

8. It is not evident how the letter [24] from Landau to Hardy (dated June 21,
and
the letter [26] from Landau to Schur (dated June 22, 1921) are related, but
1921)
to judge by the information that Hardy provides in [11] and the published form that
appears in [26], they must be very similar, perhaps even the same.

Remark

8. THE 1925 PAPER.


and wrote the following:

In the introduction

to [11] Hardy

first formulated

Theorem

the corresponding
with
theorem
for infinite
I did not give a proof, being
occupied
primarily
are
sense
E.
Prof.
Landau
state
in
what
effected.
did
not
the
has
and
series,
recently
integrations
to this note and I give here a proof of the theorem
form.
in a more precise
recalled my attention

6). Later on in the paper he also volunteered

(see Theorem
[11, p. 154]:
In a letter

[24]

(1), which

dated

October

2006]

information

to me a direct proof of
Prof. E. Landau
communicated
?
constant.
the
He
out that if the
of
also
\))p
(p/(p
pointed
to the case a = 0, then the theorem
for series, with the correct
extended
?
=
<
at once by taking /(jc)
a2,
\, f(x)
ai,0<je
may be deduced
21

June

the correct

gives
integral theorem were
value of the constant,
....
1 <x
<2,

the following

1921,

value

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

725

that we are dealing with a decreasing


loss of generality,
> a2 >
a decreasing
and observing
that
sequence ax

Indeed, assuming, without


function / or, equivalently,
the function

x Jo
on [n ?

is decreasing

+ 1)

= r^ak+a?X-n

f(t)dt

1, n], we obtain
-n-\

/Tl=\ak

XLi^V^v^r

, 1x\

+ ??(^-w +1),

sE/

E(%*)

=r(?i/<')'i')<"?(^T)'r/wp<"
00
PHardy

adds a comment

In a more
a=0
series

E<

1/

a different

concerning

recent

letter (13 Dec.


[25], he shows
1924)
that for a > 0 (by a method
resembling
to dependence
of the latter.
theorem
from

With this information inmind, Hardy


in the following form:
Theorem
f* f(t)dt

letter [11, p. 154]:

6. Let p > 1 and let f(x)


< 00 for every x > 0 and

The constant

(p/(p

l))pis

formulated

how

to deduce

that of Prof.

the integral
theorem
so
and
reduce
P?lya)

and proved his famous

> 0 be
p-integrable

for
the

inequality

on (0, 00). Then F(x)

(2)

sharp.

= 0 when x < a, then his


Hardy himself remarked in [11, p. 150] that if f(x)
in
follows
from
Theorem
6. Hardy's original proof
version
[10] (Theorem 4)
previous
to the
but in a postscript
contained fairly many technical details and explanations,
a
an
out
We
here
essential
he
suggested by P?lya.
present
simplification
pointed
proof
ideas
but
avoids
technical
details
follows
that
many
by
Hardy's original
closely
proof
appealing to P?lya's simplification.
=
(F(x)p)
Proof
By partial integration and the identity d/dx
pF(x)p~x f(x), which
holds for almost all x in (0, 00), we obtain for arbitrary a and A with 0 < a < A < 00:

Ja V X

) P
P

- 1
Ja

ax-p

1AF(a)p-^-F(A)p
1
P

--F(a)p
P-l

726

dx

+ -t--\
P-l

+ -^?
P

fA/F(x)Y~x
* J
Ja \(-^)

f
1 Ja

xl-p^-(F(x)p)dx
dx

f(x)dx.

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

Moreover,

invoking

the continuous

version

of the Holder

<
< A and
?
? such that a
applying
Choosing
? F
F(x)
(a) instead of F(x), we find that

'F(x)-F(a)Y

inequality

the preceding

(9), we

see that

two inequalities

to

a
.FM-W
P-

mdK

1 Ja

Hence

"'W-??)V,\"'S

_?_///"

f(
and a fortiori

In this inequality we first let a -> 0+ and observe


To finish the proof we let A -> oo and ? -? 0+.

that F(x)

? F
(a) increases

to F(x).

9. Hardy had already drawn attention to the sharpness of the constant


it as part
\))p in his 1920 paper. In that paper he decided not to mention
(p/(p
of the theorem that he stated, but he did include the details of the proof of the sharp
ness (by using a modified
form of the example he had considered
in the 1920 paper).

Remark

9. FURTHER CONTRIBUTIONS FROM THE 1925 PAPER. In addition to the


main result (Theorem 6), the 1925 paper included a number of interesting results that
In
have exerted significant influence on research related to Hardy-type
inequalities.
particular, Hardy proved that the following variant of (1) holds even if the standard
arithmetic mean (l/n) Yll=\ ak is replaced with amore general arithmetic means (with
general weights):
7. Suppose

Theorem
and

An

X\ +

X2 +

that an > 0 and kn > 0, that An =


+

kn for

n =

1,2,

...,

and

that

Xxa\ +
Y^=\

A.2a2+
^n^n

Then

October

2006]

the prehistory

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

727

+ Xnan

*5 convergent.

6 to suitable step functions


(see
Hardy proved this theorem by applying Theorem
that was mentioned
Landau's observation
just prior to Theorem 6). Hardy also no
ticed that if one replaces an with axJp in the inequality (14) and lets p -* oo, then
l))p -> e and
(p/(p

+ ' ' '+


+
Ka?,P\P
lim ?Xiax?P X2a2?P

p^oo \

xx + x2 -\-h

8. IfY^Lx

^nan is convergent,

constant

Therefore

Hardy

arrived

oo

...axn"y/A"<e^2xnan,
n?\

J^Ki^'^2
n=\
the

mean.

then

oo

and

-*-VMfl

^y=(?N22-..^)'

to the geometric
The scale of power means decreases
at the following
limit result of his Theorem 7:
Theorem

(^ ^

(15)

is sharp.

In the standard case where

each \n

1 (15) becomes

(axa2...an)x,n

<
ejT^an,

(16)

which is the natural limit inequality of (1). Inequality (16) was first proved by Car
its name: the Carleman
leman in 1922 (see [1]), whence
inequality. This inequality
and applied in several ways and has its own interesting history.
has been generalized
It
Carleman's
original proof, which was quite long, involved Lagrange multipliers.
must have come as a big surprise for him to see the simple proof derived from the
Hardy inequalities. He no doubt learned of it quickly, because Hardy was engaged in
a collaboration with Carleman at the time (see, for example, their joint paper [2]).
By carrying out a similar limiting process in (2) we obtain the following
inequality:

exp(-

i Inf(t)dt\dx <e
J f(x)dx,

(17)

on each finite interval (0, x). We should


is strictly positive and measurable
where f(x)
in
that
the
mention
original paper [11] inequality (17) appears without the constant e
in place of f(x).
and with exp f(x)
that it was P?lya who made him aware of the elegant lim
Hardy acknowledged
that
(15), (16), and (17). Sometimes
directly implied the inequalities
iting argument
called
the
is
(17)
Knopp inequality, although Knopp's paper on the subject dates from
in the
1928. The name P?lya-Knopp
inequality seems now to have gained acceptance
literature (see [22] and the references given there).

10. ELLIOT'S AND INGHAM'S PROOFS; THE COPSON INEQUALITY. In


1926 Elliott
[6, p. 50]):

[4] gave a simple and very elegant proof of (1) (see also [16, p. 240] and
if we set an = An/n and a0 = 0, then from the Young inequality

uv

<-1

p'

728

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

we obtain
-

ol?

[nan

7*Pn-l?n=*Pn

P-i

(n

l)a?_i]o^

p-\

p-\

.tfn-jL)^"-1^-.
p-i/

^'('-^)

a?-i

<x?

p-i

K'-"?-'+?'-'l

7^r
-??;].

T[(n-lX,
from 1 to Af yields

Summing

E^y-^Ei^r?^.o,
so from Holder's

inequality

(9) we

infer that

'-'-7
^?*nm

i^E(v)

E(t)

by the last factor leads to (1).

Division

later Grandjot

Two years

r*
v^

~
Ell

n?\

^???

V^/

AN

A\ (

?-t
v
/n?L

~~~?~
\

?-flT
n=7

x
,\n)

the identity

[7] derived

An-\

\n~

n-l)

that

by observing

2anAn

i An

A1
-'

Al
n

A-y

,A"

+(?-!)

n > 2. This gives rise to another proof of (1) in the case p ? 2.


Ingham also found a simple proof of (2) (see [16, p. 243]). Since

when

Hf(x) = x f f(t)dt= [ f(tx)dt,


Jo
Jo
it follows

from theMinkowski

integral inequality

that

i/p

f Jo

(Hf(x))p dx\

=||#/||,=

jo
/'JO

f(tx)dt
/?l

<J

-jior^v-October

2006]

= ?\U
\\f(tx)\\pdt
J
Up

the prehistory

11

of the hardy

roo

1//?

f(tx)"dx\ dt

'
p-\

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

mpds
or

In 1927 Copson
[3] proved Hardy's Theorem 7 by adapting Elliott's proof and by
bringing into play the dual to Hardy's
inequality, a result now known as the Copson
> 0, kn > 0, and
*s convergent,
then
inequality, if p > 1, an
J27=\ ^nan
OO

"\ s*

/OO

EME^rV
n= \

2^m = \ ^m /

\k=n

and the constant pp is best possible.


for each n asserts that

E Ef
\&=?

??

n= \

inequality

\ P
/

(20)

^PE?>

The Copson

OO / OO
n=l

^'E<

in the case where

Xn =

OO

pu
??1

p > 1 and that the constant pp is sharp. Hardy


[9] had earlier stated a weak
version of (21) in the case p = 2, and therefore (21) is sometimes called the Copson
Hardy inequality. Hardy [13] was the first to remark upon the duality between (20) and
(14).

when

11. FINAL DISCUSSION.


present results in a clear and
Usually mathematicians
and further research. Nonetheless,
it
polished form that is well suited for applications
is also well known that a lot of creative work, questening, collaboration,
and sometimes
even failure arise during the process of coming to the final formulations of results. The
history that we have described of the evolution of Hardy's famous inequalities can
serve as an unusually good illustration of this fact. We ourselves were quite surprised
and fascinated by several of the details we have presented here. In particular, we feel
that as a by-product
of writing
this article we have acquired improved insight into
of
and
Hardy's way
thinking
acting, and we hope that we have been able to communi
cate to the reader at least some flavor of this experience.
this understanding
it has become clear to us that G. H. Hardy (a) had
Through
contacts
with
other
mathematicians
who were interested in the subject and who
good
to
him
in various ways (e.g., via private or formal
information
passed along
significant
a
was
master
at
real
letters), (b)
developing
important parts of the theory himself,
(c) was very good at cleverly synthesizing knowledge
gained from numerous sources,
and (d) played a central role in the developments
described in this paper. In particular,
it is totally appropriate that his name adorns inequalities
it has
(1) and (2). However,
to be said that other mathematicians
also made very important contributions
to this
(e.g., E. Landau, G. P?lya, M. Riesz, and I. Schur). For example, if the
development
results of these individuals had been published in a different way, we might today refer
to the discrete inequality (1) as the Riesz or the Landau-Riesz
or the Hardy-Landau
Riesz

inequality.

ACKNOWLEDGMENTS.
The first named
Sciences

We

author has been

of the Czech

dedicate
supported

this paper to the memory


by the grant # A1019305

of Godfrey Harold Hardy


(1877-1947).
of the Grant Agency
of
of the Academy

Republic.

REFERENCES
1.
2.

730

T. Carleman,

Sur les fonctions


in Fifth Scandinavian
quasi-analytiques,
1923, pp. 181-196.
(Helsinki,
1922), Akadem. Buchh., Helsinki,
T. Carleman
and and G. H. Hardy, Fourier series and analytic functions,
124-133.

Congress
Proc. Royal

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

of Mathematicians
Soc. A 101 (1922)

[Monthly

113

3.
4.
5.
6.

E. T. Copson, Note on series of positive


terms, J. London Math. Soc. 2 (1927) 9-12.
J. London Math.
of some recently proved facts as to convergency,
E. B. Elliott, A simple exposition
1(1926)93-96.
L. Fej?r and F. Riesz, ?ber einige
E. C. Francis and J. E. Littlewood,
some

On

identities

Math. Z. 11 (1925) 305-314.


Ungleichungen,
in Infinite Series, Deighton
1928.
Bell, Cambridge,
Examples
theorem, J. London Math. Soc. 3 (1928)
relating to Hardy's convergence
funktionaltheoretische

7.

K. Grandjot,
114-117.

8.

G. H. Hardy, Notes on some points in the integral calculus, XLI. On the convergence
and series, Messenger
of Math. 45 (1915) 163-166.
double-series
Notes on some points in the integral calculus, LI. On Hubert's

-,9.

theorems

connected
48(1919)

concerning

10.
-,

Notes

on a theorem

Notes

on some points
(1925) 150-156.

54
Note

12.
-,

of

the convergence

infinite

series

and

of certain

integrals

theorem,

and some
of Math.

integrals, Messenger

107-112.

11.
-,
of Math.

of Hubert, Math.
in the integral

on a theorem

23(1925)45-46.
Remarks

of Hubert

Z. 6(1920)
314-317.
LX. An inequality

calculus,

concerning

series

of positive

between

integrals, Messenger

terms, Proc.

London

Math.

Soc.

on three recent notes

in the journal, J. London Math. Soc. 3 (1928) 166-169.


to a chapter on inequalities,
J. London Math. Soc. 4 (1929) 61-78.
Prolegomena
and G. P?lya, The maximum
of a certain bilinear form, Proc.
15. G. H. Hardy, J. E. Littlewood,
265-282.
Math. Soc. 25(1926)

13.
-,

Soc.

14.
-,

16.
-,
17.
18.

London

1967.
2nd ed., Cambridge University
Press, Cambridge,
Inequalities,
in
und unendliche
?ber Integralgleichungssysteme
E. Hellinger,
Huberts Arbeiten
Gleichungssysteme,
vol. 3, Springer-Verlag,
David Hubert, Gesammelte
Berlin,
1979, pp. 94-145.
Abhandlungen,
Nachr.
der linearen Integralgleichungen,
einer allgemeinen
Theorie
D. Hubert, Grundz?ge
G?ttingen
(1906)157-227.

19.
-,

Grundz?ge

einer

Theorie

allgemeinen

der

linearen

Teubner,

Integralgleichungen,

Leipzig,

1912.
20.
-,
21.
22.
23.

1979.
vol. 3, Springer-Verlag,
Gesammelte
Berlin,
Abhandlungen,
its history and some related
and L. E. Persson, The Hardy inequality-About
A. Kufner, L. Maligranda,
of Technology,
Lule?, 2006.
results, research report, Lule? University
2003.
A. Kufner and L. E. Persson, Weighted
of Hardy Type, World Scientific,
Inequalities
Singapore,
E. Landau, ?ber einen Konvergenzsatz,
G?ttingen Nachr. (1907) 25-21.
Letter

24.
-,
25.
-,
26.
-,
27.

to G. H. Hardy, June 21,


to G. H. Hardy, December

1921.

13, 1924.
series of positive terms: Extract from a letter of Prof. E. Landau
concerning
to Prof. I. Schur, J. London Math. Soc. 1 (1926) 38-39.
in Real Analy
Selected Problems
B. M. Makarov, M. G. Goluzina, A. A. Lodkin, and A. N. Podkorytov,
American Mathematical
1992.
sis, Transi, of Math. Monographs,
Society, Providence,
Letter

A note on a theorem

Holder's

should

28.

L. Maligranda,

29.

1(1998)69-83.
B. Opic and A. Kufner, Hardy-Type
Inequalities,
man Scientific & Technical, Harlow,
1990.

30.
31.
32.
33.

36.
37.
38.
39.

inequality

be called

Rogers'

Pitman Research

inequality,

Notes

Math.

inMathematics,

Inequal.

Appl.

no. 219, Long

G. P?lya, Proof of an inequality, Proc. London Math. Soc. (2) 24 (1926) 57.
und Lehrs?tze aus der Analysis,
1925.
G. P?lya and G. Szeg?, Aufgaben
Berlin,
Springer-Verlag,
?ber Systeme
F. Riesz, Untersuchungen
integrierbarer Funktionen, Math. Ann. 69 (1910) 449^4-97.
M. Riesz, Letter to G. H. Hardy, 1919 or 1920.
I. Schur, Letter

34.
35.
-,

Why

to G. H. Hardy,
zur

Bemerkungen
Math.
Ver?nderlichen,/.

1918 or 1919.
der

Theorie

140(1911)

beschr?nkten

Bilinearformen

mit

unendlich

vielen

1-28.

to the Art of Mathematical


Master Class. An Introduction
J.M. Steele, The Cauchy-Schwarz
Inequalities,
2004.
Press, Cambridge,
Cambridge University
und bilinearen Formen von unendlich vielen Ver?nderlichen,
O. Toeplitz, Zur Theorie der quadratischen
G?ttingen Nachr. (1910) 489-506.
von Herrn Hubert, Math. Ann. 68 (1910)
eines Reihensatzes
Beweis
F. Wiener, Elementarer
mit besonderer Ber?cksichtigung
des Fourierschen
H. Weyl, Singul?re
Integralgleichungen
1908.
orems, Ph.D. dissertation, G?ttingen,

ALOIS

KUFNER

Institute

October

received

his Ph. D. from

of the former Czechoslovak

2006]

the Czech

(now Czech)

the prehistory

361-366.
Integralthe

in 1965. He has worked at the Mathematical


Academy
of Sciences,
serving as deputy director ( 1974?

Academy

of the hardy

inequality

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

731

he taught at the Charles University


institute. Simultaneously,
and the
ofthat
1978) and director (1978-1990)
of West Bohemia
in Pilsen, where he was
both in Prague, and at the University
Czech Technical
University,
In 1983 he received the Doctor of Sciences
head of the Department
of Mathematics
(1977-1990).
degree at
the Academy
was

a full professor
at the University
of West Bohemia
in 1985. In 1988, he
of Sciences. He was awarded the National
of the Czechoslovak
Academy
in 1985. His main fields of interests are differential
and theory of
equations

and became

of Sciences

member

elected

Prize

of

function

corresponding
the Czech Republic
spaces. He is the author or coauthor

them, Hardy-Type
2003).
Mathematical

(Longman,

Inequalities

of about

of Sciences

Institute, Academy

120 mathematical

1990) and Weighted

papers

Inequalities

of the Czech Republic,

Zitn?

and eight monographs,

of Hardy

Type

25, 115 67 Praha

among

(World Scientific,
1, Czech Republic

kufner@math.cas.cz

in Poznan
received his Ph. D. from the Adam Mickiewicz
(Poland) in
University
at the Polish Academy
Orlicz. He has held appointments
of Sciences
of Wladyslaw
the supervision
in Caracas,
the Venezuelan
Institute for Scientific Research
Central of Venezuela
in Poznan, University
(IVIC)
at Lule?
he is in the Department
of Mathematics
and Claremont McKenna
in Caracas,
Currently
College.
in 1991 he completed
the Doctor
of Sciences
of Technology
(Sweden), where
degree (=Docent
University

LECH

MALIGRANDA

1979 under

He received the Stefan Banach Prize of the Polish


became professor of mathematics.
in Venezuela
in 1990, and the Ad?lie Prize
in 1982, the State Prize from CONICIT
Society
in 2001. His research interests include functional
of Technology
for the Best Teacher at the Lule? University
He is the author or coauthor of
and sometimes
the history of mathematics.
real analysis,
inequalities,
analysis,

in Sweden)
Mathematical

and in 2000

one hundred mathematical

papers

and the monograph

Orlicz

Spaces

and Interpolation

SE-971

87 Lule?,

(Campinas

SP Brasil,

1989).
Department
of Mathematics,
lech@sm.luth.se

Lule?

University

of Technology,

Sweden

in 1974 under the super


received his Ph. D. from the Urnea University
PERSSON
(Sweden)
at Ume? University,
the Lule? University
of Technology,
of Ingemar Wik. He has held appointments
at the Lule?
Since 1994 he has been a full professor of mathematics
and Uppsala University.
Narvik University,
a member
of
He was president of the Swedish Mathematical
of Technology.
(1996-1998),
Society
University
at the Royal Academy
of Sciences
from 1995, and director of
of Mathematics
the Swedish National Committe
at Lule? University
of Technology
from 1995. His research interests in
the Center of Applied Mathematics
LARS-ERIK
vision

and homogenization
the interpolation of operators, Fourier analysis, function spaces, inequalities,
theory.
The Homogenization
Method
is the author or coauthor of 120 mathematical
papers and three monographs:
Lund 1993), Weighted
An Introduction
2003),
(Studentlitteratur,
Inequalities
of Hardy Type (World Scientific,
clude

He

and Convex

and their Applications


Lule? University

Functions

Department
of Mathematics,
larserik@ sm.luth.se

732

(Springer,

2006).

of Technology,

SE-971

87 Lule?,

Sweden

THE MATHEMATICAL ASSOCIATION OF AMERICA

This content downloaded on Tue, 18 Dec 2012 02:30:26 AM


All use subject to JSTOR Terms and Conditions

[Monthly

113

You might also like