A Generalization of Special Atom Spaces with Arbitrary Measure
by
Paul Alfonso, Jr.
A dissertation submitted to the Graduate Faculty of
Auburn University
in partial ful llment of the
requirements for the Degree of
Doctor of Philosophy
Auburn, Alabama
December 13, 2010
Keywords: Weighted Generalized Special Atoms, Weighted Metric Space,
A ( ; ), B ( ; )
Copyright 2010 by Paul Alfonso, Jr.
Approved by
Geraldo Soares de Souza, Chair, Professor of Mathematics and Statistics
Narendra Govil, Professor of Mathematics and Statistics
Asheber Abebe, Associate Professor of Mathematics and Statistics
Abstract
A brief historical account of the development of special atom spaces is presented followed
by the introduction of two new function spaces, A ( ; ) and B ( ; ), which are general-
izations of previous special atom spaces utilizing arbitrary measures rather than Lebesgue
measure of intervals. Known de nitions relating to normed vector spaces are extended to
apply to the new function spaces of arbitrary measure. The properties of the new function
spaces are discussed including the relationship between the spaces as well as the relationship
of the spaces with well known function spaces such as Lebesgue spaces, Lp, Lip( ; ) and
( ; ).
Major results include H older-type inequalities for both A ( ; ) and B ( ; ). In the
case of B ( ; ), the dual of B ( ; ) is determined and a Representation Theorem for the
weighted bounded linear functionals of B ( ; ) is presented in detail. However, for A ( ; )
we mention that the dual follows the same idea of the theorem for B ( ; ). That is, that
we only need to estimate k AkA( ; ) for a -measurable set A. Indeed we show there is
a positive constant M such that k AkA( ; ) M (A). The duality and representation
theorems for A ( ; ) follow easily. Interpolation of Operators Theorems are presented
on sublinear operators which map B( ; 1p) into weak Lp and A( ; 1p) into weak Lp spaces.
Finally, we present the multiplication operator on A ( ; ) and B ( ; ) for (t) = t, and
show under some conditions this operator is bounded on those spaces.
ii
Acknowledgments
I have many people to thank for their support and help with this dissertation. The
people listed in the following paragraphs are at the top of a very long list of people who
supported my e orts in some way.
First, I would like to thank God. God has blessed me in more ways than I can list and
I strive to be his servant.
I am especially thankful for my advisor, Geraldo de Souza. Without his guidance and
understanding I doubt I would have nished my PhD. His sage advice helped me deal with
my absence from high level mathematics for so many years. I am grateful for the work of
Professor Abebe and Professor Govil, the remaining members of my committee. All of my
professors have been understanding in dealing with an Air Force pilot who still wanted to
work on a PhD in mathematics. Thank you to Professor Khodadadi for volunteering to act
as the outside reader for my dissertation.
Eddy Kwessi, Huybrechts Bindele, and Dan Brauss provided useful insight on many
occasions and were good friends during my time at Auburn.
A very special thank you to my parents, Paul and Dolores Alfonso, whom I can never
repay for all of their love and support over the years. Their constant reinforcement of the
importance of an education has driven me to achieve. I would also like to thank my sister
Sheryl, and my brother Dean as well as my extended family who have always supported my
endeavors.
Finally, and most importantly, I am thankful for my lovely wife, Kristal, and our son
Tony. Both have had to endure many nights without me as I struggled with some problem
or issue. Your love, patience, and understanding made this dissertation possible.
iii
Table of Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
1 Introduction and Historical Background . . . . . . . . . . . . . . . . . . . . . . 1
1.1 A Major Question in the 20th Century . . . . . . . . . . . . . . . . . . . . . 1
1.2 Special Atom Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Generalizations of Special Atom Space . . . . . . . . . . . . . . . . . . . . . 5
2 De nitions and Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1 Basic De nitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 De nitions of A ( ; ) and B ( ; ) . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Functional and Operator type De nitions . . . . . . . . . . . . . . . . . . . . 12
2.4 De nitions of Lip( ; ), ( ; ), Lorentz spaces and some results . . . . . . 14
3 Properties of A ( ; ) and B ( ; ) . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Completeness and Lp Inclusion . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Relationships within B ( ; ) and A ( ; ) . . . . . . . . . . . . . . . . . . . 26
4 Major Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1 H older-Type Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Duality and Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Interpolation of Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.4 Multiplication Operator on L(p;1) . . . . . . . . . . . . . . . . . . . . . . . 43
5 Comments on the dual of A ( ; ) . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.1 Relationship between Lip( ; ) and ( ; ) . . . . . . . . . . . . . . . . . . 47
5.2 Closing Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
iv
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
A Vector Space proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
B Veri cation of minimum in proof of Theorem 4.4 . . . . . . . . . . . . . . . . . 55
v
Chapter 1
Introduction and Historical Background
This dissertation will introduce and explore the properties and applications of two new
function spaces, denoted as A ( ; ) and B ( ; ). This dissertation begins with historical
background motivating these spaces, followed by several de nitions which streamline the
introduction of the new function spaces. We will be presenting several theorems regarding
the properties of the new function spaces as well as the relationship with other known
function spaces.
1.1 A Major Question in the 20th Century
In 1923, Frigyes Riesz introduced a new function space he named after G.H. Hardy
following a paper written by Hardy in 1915, see [32],[26]. The space is de ned as:
De nition 1.1 (Hardy?s Space Hp(D);0
0, we are only concerned
with the case p 1.
1
During the rst half of the twentieth century, a well known fact was that for 1
1. Note f 2L(p;1)
if kfkL(p;1) = R2 0 f (t)t1p 1dt < 1, where f is the decreasing rearrangement of f, see
comments following De nition 2.10. Indeed, de Souza showed that for p > 1, A( ;1=p) =
B( ;1=p) = L(p;1) with equivalent norms, see [9]. The main result in this dissertation is a
further generalization of A( ; ) and B( ; ), utilizing di erent \norms"which are de ned
as weighted metrics. The remainder of this dissertation will de ne and explore A ( ; ) and
B ( ; ), beginning with several useful de nitions, including the de nitions of A ( ; ) and
B ( ; ).
7
Chapter 2
De nitions and Comments
In order to concisely de ne our new function spaces, we will introduce several new de -
nitions in this section. Many of the following de nitions are extensions of known de nitions
and are noted as such. We will include examples and comments for clari cation as appro-
priate. For the remainder of this dissertation, we shall assume that any function denoted by
the symbol is de ned and nite for real numbers in its given domain.
2.1 Basic De nitions
The rst de nition provided below de nes a class of functions which we will utilize
throughout this dissertation.
De nition 2.1 (Class C functions) We de ne C to be a class of functions
: [0;1)![0;1) satisfying the following conditions:
(0) = 0; is strictly increasing and continuous (2.1)
( x) ( ) (x) for some function : R+ 7!R+ (2.2)
(x+y) k ( (x) + (y)) for some constant k 1 (2.3)
(x)!1 as x!1 (2.4)
In order to illustrate Class C functions we present the following Lemma.
Lemma 2.1 (Class C is not empty) For 2 (0;1] the real functions 1(t) and 2(t)
de ned by 1(t) = t and 2(t) = ln (t+ 1) on [0;1) are in the Class C functions.
8
Proof. Let 2 (0;1]. First, consider 1. 1(0) = 0 and 1 is clearly continuous. Now
01(t) = t 1 > 0 so 1 is strictly increasing and property (2.1) is satis ed. For property
(2.2) of Class C functions let 2R+, then 1( t) = ( t) = t = 1(t) and we have
1( ) = . Property (2.3) follows directly from the inequality (t + s) t + s , since
1(t+s) = (t+s) t +s = 1(t) + 1(s). This inequality is simple to prove:
(t+s) = (t+s)(t+s) 1 = t(t+s) 1 +s(t+s) 1 t t 1 +s s 1
since (t + s) 1 t 1 and (t + s) 1 s 1. Thus, (t + s) t + s . The nal property
is clearly true and 1 2C .
The proof for 2 is slightly more involved. Property (2.1) is clear since natural log
is strictly increasing and continuous with 2(0) = ln (1) = 0. In order to prove prop-
erty (2.2) We break into two cases with = 1. First consider the case > 1. Let
g(t) = ln(t + 1) ln( t + 1), then g0(t) = t+1 t+1 and g0(t) = 0 occurs only at t = 0.
Since g0(t) > 0, g(t) is strictly increasing and g(0) is a minimum we conclude g(0) 0. Thus, we have g(0) = 0 1.
Combining the two cases and the trivial case = 1 we conclude ln( t+ 1) ( + 1) ln(t+ 1)
for all 0, that is 2( ) = + 1. Finally consider 6= 1. We now have 2( t) =
ln (t + 1) (( + 1) ln( t + 1)) = ( + 1) ln (t + 1). Setting 2( ) = ( + 1) property
(2.2) is proved. Since property (2.4) is clearly true, all that remains is to prove property
(2.3). Since 2(t + s) = ln (t + s + 1), letting = 1 we have 2(t + s) = ln(t + s + 1).
Now utilizing logarithmic properties we see 2(t) + 2(s) = ln(t + 1) + ln(s + 1) = ln((t +
1)(s + 1)) = ln(ts + t + s + 1) ln(t + s + 1) = 2(t + s) since ts 0. Letting 6= 1,
2(t+s) = ln (t+s+ 1) (ln(t+ 1) + ln(s+ 1)) ln (t+ 1) + ln (s+ 1) = 2(t) + 2(s)
9
by the third property proved above for 1(t) = t :2
The following de nition is an extension of a normed vector space which is applicable to
the spaces introduced in this dissertation.
De nition 2.2 (Weighted Metric Space) Let X be a vector space. X is said to be a
weighted metric space if there is a given real valued function k kX called a weighted metric
on X satisfying:
kxkX > 0 if x6= 0 (2.5)
kxkX = 0 if and only if x = 0 (2.6)
k xkX w(j j)kxkX for all scalars , w a function on R+ (2.7)
kx+ykX kw (kxkX +kykX) for all x;y2X; kw 1, a scalar (2.8)
Note if one replaces (2.7) and (2.8) by (2.7)?: k xkX = j jkxkX and (2.8)?: kx + ykX
kxkX +kykX for all x;y2X, then k kX is a norm in the usual sense.
Note: for convenience, we will use the typical symbol for norm throughout this disser-
tation although in many cases our representation is not actually a norm. This distinction
will be pointed out where appropriate. Wherever a metric satis es the four above restric-
tions, we will refer to this metric as a weighted metric. The next de nition de nes a useful
relationship between weighted metric spaces.
De nition 2.3 (Weighted Continuous Space Inclusion) Let X and Y be two weighted
metric spaces, one says that X is weighted continuously contained in Y if:
1. X Y and
2. There are constants M;k > 0 and a function : [0;1) ! [0;1); (0) = 0 such that
kfkY M (kkfkX).
10
Note if (t) = t, then the above de nition is the usual de nition for continuously con-
tained normed spaces. For convenience, we will now de ne Weighted Generalized Special
Atoms. This de nition provides us a type of shorthand notation which we utilize throughout
the remainder of this dissertation.
De nition 2.4 (Weighted Generalized Special Atoms for arbitrary measure)
Given a nite measure space ([ ; ];A; ), for n2N, let An, Bn, Xn, and E be measurable
sets in [ ; ] such that AnSBn = Xn, AnTBn =;and (An) = (Bn). For 2(0;1], we
de ne the weighted generalized special atoms of Type I and II, bI;IIn : [ ; ]!R as follows:
bI;II1 (t) = 1 ([ ; ]); bIn(t) = 1 (X
n)
[ An(t) Bn(t)];bIIn (t) = 1 (B
n)
[ Bn(t)]; n6= 1
where E is the characteristic function of E.
De nition 2.5 (Completeness) A Weighted Metric Space X is said to be complete if and
only if every Cauchy sequence in X converges to an element in X.
2.2 De nitions of A ( ; ) and B ( ; )
Armed now with the above de nitions, we can now de ne the two new spaces A ( ; )
and B ( ; ), which are the foundation of this dissertation.
De nition 2.6 (A ( ; )) For 2(0;1], let (bIn)n 1 be weighted generalized special atoms
of Type I and (cn)n 1 be a sequence of real numbers, a nite measure on sets in a algebra
of [ ; ] and 2C . We de ne the space A ( ; ) as:
A ( ; ) =
(
f : [ ; ]!Rjf(t) =
1X
n=1
cnbIn(t);
1X
n=1
(jcnj) <1
)
For f2A ( ; ) we de ne a \norm" as
kfkA ( ; ) = inf
1X
n=1
(jcnj)
11
where the in mum is taken over all possible representations of f.
De nition 2.7 (B ( ; )) For 2(0;1], let (bIIn )n 1 be weighted generalized special atoms
of Type II and (cn)n 1 be a sequence of real numbers, a nite measure on sets in a
algebra of [ ; ] and 2C . We de ne the space B ( ; ) as
B ( ; ) =
(
f : [ ; ]!Rjf(t) =
1X
n=1
cnbIIn (t);
1X
n=1
(jcnj) <1
)
For f2B ( ; ) we de ne a \norm" as
kfkB ( ; ) = inf
1X
n=1
(jcnj)
where the in mum is taken over all possible representations of f.
The word \norm" in the previous two de nitions is in quotations since in many in-
stances,k kA ( ; ) andk kB ( ; ) will not turn out to be norms. However, we will show that
this \norm" is, in fact, at all times at least a weighted metric.
We point out here for (t) = t, A ( ; ) and B ( ; ) reduce to the spaces A( ; )
and B( ; ), which de Souza introduced in 2006, see [9]. Thus, considering the previous
discussion, we see that A ( ; ) and B ( ; ) are indeed generalizations of all the spaces
derived from the original de Souza space. The next de nitions further generalize linear
functionals and operator norms to meet the needs of subsequent proofs.
2.3 Functional and Operator type De nitions
A few operator-type de nitions need extension for the purposes of this dissertation.
Namely, we extend the de nitions of bounded linear operators , operator norms, and dual
spaces to suit our purposes.
12
De nition 2.8 (Weighted Bounded Linear Functional) Let X be a weighted metric
space, we say that is a weighted bounded linear functional on X if : X!R such that:
1. is linear
2. There are constants M;k > 0 and a continuous function : [0;1) ! [0;1); (0) = 0
such that for all f2X, j (f)j M (kkfkX).
Note: If (t) = t then the above de nition is the usual de nition for bounded linear func-
tionals.
De nition 2.9 (Weighted Operator Norm) Let X be either A ( ; ) or B ( ; ) and
? be a weighted bounded linear functional on X. We de ne the weighted operator norm, k?k
as follows for f2X:
k?k= sup
f6=0
j?(f)j
1(k kfkX)
Again, note that if (t) = t and k = 1, the weighted operator norm reduces to the usual
de nition of operator norms (or equivalent characterization of operator norms.) We now will
extend the traditional de nition of the dual space of a normed space to the weighted metric
space equivalent.
De nition 2.10 (Dual Space of a Weighted Metric Space) The space of all weighted
bounded linear functionals on a weighted metric space X is called the dual of X and is denoted
by X .
The nal de nition of this section is a special case of a de nition given by de Souza and
Bloom in [3] relating to operators. In order to formulate this de nition we must rst de ne
the decreasing rearrangement of a real valued, measurable function f. Let f be a real valued
measurable function on T, then for y> 0 let
m(f;y) = m(jfj;y) =jfx2T;jf(x)j>ygj
13
wherej jdenotes the Lebesgue measure on T. m(f;y) is known as the distribution function
of f. Now we can de ne the decreasing arrangement of f, denoted by f , as
f (t) = inffyjm(f;y) tg:
Armed with the previous de nition we can now de ne restricted weak type r operators.
De nition 2.11 (Restricted Weak Type r Operators) Let be a nite measure. We
say that an operator T is restricted weak type r if for any -measurable set A in [ ; ]
t1r (T A) (t) M 1r(A)
where M is an absolute constant and is the decreasing rearrangement of T A(t).
2.4 De nitions of Lip( ; ), ( ; ), Lorentz spaces and some results
Here are the de nitions of three well known spaces for later reference. We will see that
Lip( ; ) and ( ; ) are not only related to each other but have relationships with A ( ; )
and B ( ; ) as well as their duals. The Lorentz spaces are useful in interpolation of operator
theorems.
De nition 2.12 (Lip( ; )) For 2 (0;1] and a nite measure on sets of [ ; ] the
space Lip( ; ) is de ned as
Lip( ; ) =
g : [ ; ]!Rj 1 (B)
Z
B
g(t)d (t)
><
>>:
cum
2
; m even
cvm+1
2
; m odd
bIIm(t) =
8>
><
>>:
bIIum
2
(t); m even
bIIvm+1
2
(t); m odd
Then u + v = P1m=1 dmbIIm(t). Clearly all that remains to show is that P1m=1 (jdmj) <1.
This is clear sinceP1m=1 (jdmj) = P1n=1 (jcunj)+P1n=1 (jcvnj) <1. Thusu+v2B ( ; )
and B ( ; ) is closed under addition.2
Our next theorem states the property mentioned earlier that the \norms" of the spaces
are weighted metrics.
Theorem 3.2 (Weighted Metric) The functions k kB ( ; ) and k kA ( ; ) are weighted
metrics on B ( ; ) and A ( ; ), respectively.
We provide the proof for B ( ; ). The proof for A ( ; ) is analogous.
Proof. Proof of (2.6) (kxkX = 0 if and only if x = 0):
(() Since 2C we know is non-negative which implieskfkB ( ; ) 0. Let f2B ( ; )
such that f = 0. Then one possible representation of f is letting all coe cients cn be zero.
Hence, (jcnj) = 0 for all cn?s so P1n=1 (jcnj) = 0. This implies kfkB ( ; ) = 0.
()) Let f2B ( ; ) such thatkfkB ( ; ) = 0. Let fk be a representation of f, then fk(t) =
P1
n=1cnkb
II
nk(t). De ne dk =
P1
n=1 (jcnkj). Now either f = 0 and the proof is complete or
there exists a sequence of representations of f,ffkg1k=1 such that dk!0 as k!1. Consider
the second case where f 6= 0 and such a sequence of representations of f exists. Then for
all n2N, jcnkj! 0 as k!1: Else, assume for some m2N, jcmkj is bounded below by
some > 0 and (jcmkj) > 0. Let = (jcmkj2 , then P1n=1 (jcnkj) (jcmkj) > > 0 for all
k. This contradicts our original assumption thatkfkB ( ; ) = 0. Thus the coe cients in the
sequence of representations of f all converge to zero and hence f = 0.
18
Proof of (2.5) (kxkX > 0 if x6= 0):
This follows directly from (2.6) and the fact that kfkB ( ; ) 0 for all f2B ( ; ).
Proof of (2.7) (k xkX w(j j)kxkX for all scalars , w real-valued function):
Let 2R, and f2B ( ; ) then f(t) = P1n=1 cnbIIn (t) and
f(t) =
1X
n=1
cnbIIn (t) =
1X
n=1
cnbIIn (t)
so
k fkB ( ; ) =
1X
n=1
cnbIIn
B ( ; )
1X
n=1
(j cnj)
applying property (2.2) of we have
1X
n=1
(j cnj) =
1X
n=1
(j j) (jcnj) = (j j)
1X
n=1
(jcnj):
Thus, taking the in mum over all representations of f in the above inequality we conclude
k fkB ( ; ) (j j)kfkB ( ; ) :
Proof of (2.8) (kx+ykX kw (kxkX +kykX) for all x;y2X; kw 1, a scalar):
Let f;g2B ( ; ). Then (f + g)(t) = P1n=1c(f+g)nbII(f+g)n(t) where c(f+g)n 2R. Now let
> 0. By de nition ofk kB ( ; ) there are sequences of real numbers cfn, cgn and atoms bfn,
bgn such that
f(t) =
1X
n=1
cfnbIIfn(t) and g(t) =
1X
n=1
cgnbIIgn(t)
and for the constant k given in (2.3) we have
1X
n=1
(jcfnj) kfkB ( ; ) + 2k
and
1X
n=1
(jcgnj) kgkB ( ; ) + 2k
:
19
Combining the above equations and multiplying by k :
k
1X
n=1
(jcfnj) +
1X
n=1
(jcgnj)
!
k kfkB ( ; ) +kgkB ( ; ) + :
Notice now that (f +g)(t) = P1n=1cfnbIIfn(t) +P1n=1cgnbIIgn(t) which can be re-written as
(f +g)(t) =
1X
n=1
8
>><
>>:
cfn
2
bIIfn
2
(t); if n even
cgn+1
2
bIIgn+1
2
(t); if n odd:
Therefore, taking the in mum below over all representations of f +g we have
kf+gkB ( ; ) = inf
1X
n=1
jc(f+g)nj
1X
n=1
8>
><
>>:
cfn
2
; n even
cgn+1
2
; n odd
=
1X
n=1
(jcfnj)+
1X
n=1
(jcgnj):
Applying property (2.3) of and continuing above
1X
n=1
(jcfnj)+
1X
n=1
(jcgnj) k
1X
n=1
(jcfnj) +
1X
n=1
(jcgnj)
!
k kfkB ( ; ) +kgkB ( ; ) + :
Since was arbitrary we conclude
kf +gkB ( ; ) k kfkB ( ; ) +kgkB ( ; ) :2
The next theorem introduces the interesting result A ( ; ) B ( ; ).
Theorem 3.3 (Inclusion) For all 2C , A ( ; ) B ( ; ). Moreover, jjfjjB ( ; )
2 12 jjfjjA ( ; ).
Proof. Let f2A ( ; ), then for all n2N there are cn2R such that
f(t) =
1X
n=1
cnbIn(t); where
1X
n=1
(jcnj) <1
20
Re-writing f(t):
f(t) =
1X
n=1
cn
(Xn) ( An(t) Bn(t))
=
1X
n=1
cn An(t)
(Xn)
1X
n=1
cn Bn(t)
(Xn) :
Now, since An\Bn =;and (An) = (Bn) we have (Xn) = (AnSBn) = (An)+ (Bn) =
2 (An) = 2 (Bn). Substituting into previous equation we have:
f(t) =
1X
n=1
cn An(t)
(2 (An))
1X
n=1
cn Bn(t)
(2 (Bn))
= 12
1X
n=1
cn An(t)
( (An))
1X
n=1
cn Bn(t)
( (Bn))
!
:
We introduce two new variables to re-index the above equation:
Dm =
8>
><
>>:
Bm2 ; m even
Am+1
2
; m odd
pm =
8>
><
>>:
cm2 ; m even
cm+1
2
; m odd
Substituting and noting that Dm are measurable sets we have:
f(t) =
1X
m=1
pm Dm(t)
2 (Dm) =
1X
m=1
pm
2 b
II
m(t)
Finally, we have
1X
m=1
(
pm2
) =
1X
n=1
(
cn2
) +
1X
n=1
(
cn2
) = 2
1X
n=1
(
cn2
)
21
Applying property (2.2) of C functions
2
1X
n=1
(
cn2
) 2
1X
n=1
1
2
(jcnj) = 2
1
2
1X
n=1
(jcnj) <1
So, we have f(t)2B ( ; ) and A ( ; ) B ( ; ).
Taking the in mum over all representations of f of both sides of the above inequality gives
jjfjjB ( ; ) 2 12 jjfjjA ( ; ):2
3.2 Completeness and Lp Inclusion
In this section, we will provide a proof that Weighted Metric Spaces are complete in
the sense of De nition 2.5 if and only if every absolutely summable series in the space
is summable. This property of Weighted Metric Spaces is very useful in several proofs
presented. The relation of the spaces A ( ; ) and B ( ; ) to the Lebesgue spaces, Lp is
then examined. This relationship with Lp spaces is vital to the last theorem presented in
this section which demonstrates the completeness of A ( ; ) and B ( ; ).
Theorem 3.4 (Completeness of Weighted Metric Spaces) Let X be a weighted met-
ric space. Then X is complete if and only if every absolutely summable series is summable.
The proof given below is an adaptation of a proof given by Royden, see [34].
Proof. ()) Let X be complete andhfiibe an absolutely summable series of elements of X.
Then P1i=1kfikX = M <1. Let > 0 then there is an N2N such that P1i=NkfikX < kw
where kw is the constant from De nition 2.2. Let sn = Pni=1kfikX be the partial sum of
hfii, then for n m N, we have
ksn smkX =
nX
i=m
fi
X
kw
1X
i=m
kfikX kw
1X
i=N
kfikX <
22
and hence hsii is a Cauchy sequence in X and must converge to an element f in X since X
is complete. Therefore, limn!1Pni=1 fi = f in X.
(() Let hfni be a Cauchy sequence in X. Then for each integer k there is an integer nk
such that kfn fmkX < 2 k for all n;m > nk. Hence, we choose nk0s such that nk+1 > nk
and obtain the subsequence hfnki of hfni. Let g1 = fn1 and gk = fnk fnk 1, then for k> 1
we have a series hgki whose kth partial sum is fnk but kgkkX 2 k+1. Therefore, we have
1X
k=1
kgkkX kg1kX +
1X
k=2
2 k+1 =kg1kX + 1
and hgki is absolutely summable. Thus, there exists f 2 X such that hfnki! f in X.
Finally, we show limn!1fn = f. Since hfni is a Cauchy sequence, for > 0 there is an N2N
such that kfn fmkX 2kw for all n;m > N. Since fnk !f, there is a K2N such that
for all k K we have kfnk fkX 2kw. Take k so large that k>K and nk >N, then
kfn fkX kfn fnk +fnk fkX kw(kfn fnkkX +kfnk fkX) kw( 2k
w
+ 2k
w
) = :
Since was arbitrary, we have limn!1fn = f and X is complete based on De nition 2.5 :2
The Lemma below will be used in several subsequent proofs.
Lemma 3.1 For n2N, let cn be real numbers and 2C such that P1n=1 (jcnj) <1 ,
and let k be the constant from property (2.3) of C functions, then
1X
n=1
jcnj 1
k
1X
n=1
(jcnj)
!
where 1 is the inverse of .
23
Proof. Let n, cn, , and k be as above in the statement of the lemma. Choose N 2 N.
Then, since is continuous, well de ned, nite, strictly increasing, and (0) = 0, we have
NX
n=1
jcnj= 1
NX
n=1
jcnj
!!
1
k
NX
n=1
(jcnj)
!
:
Taking the limit as N!1 of the above inequality gives the desired result.2
It is interesting to note that if one chooses (t) = tp for p2(0;1), then 1(t) = t1p and
k = 1 (see Lemma 2.1), applying Lemma 3.1 gives a well known inequality
1X
n=1
jcnj
1X
n=1
jcnjp
!1
p
The following theorem presents the aforementioned relationships between A ( ; ), B ( ; ),
and Lp spaces.
Theorem 3.5 (Lp Inclusion) For p 1 and < 1p, A ( ; ) and B ( ; ) are weighted
continuously contained in Lp[ ; ].
Proof. First, consider one weighted generalized special atom of Type II (where B is a
measurable set in [ ; ]):
jjbIIjjpLp =
Z
B(t)
(B)
p
d (t) = 1 p (B)
Z
B
B(t)pd (t)
= (B) p (B) = 1 p (B) 1 p ([ ; ]):
Let M ;p = ( 1 p ([ ; ]))1=p, then for any weighted generalized special atom of Type II
we have jjbIIjjLp M ;p.
Now let f2B ( ; ), then for n2N there are cn2R such that
f(t) =
1X
n=1
cnbIIn (t);
1X
n=1
(jcnj) <1:
24
Taking the Lp norm of f we have:
jjfjjLp =
1X
n=1
cnbIIn
Lp
1X
n=1
jjcnbIInjjLp =
1X
n=1
jcnjkbIInkLp
1X
n=1
jcnjM ;p = M ;p
1X
n=1
jcnj:
Now, by Lemma 3.1 we know
1X
n=1
jcnj 1
k
1X
n=1
(jcnj)
!
:
Combining this with our previous inequality,
jjfjjLp M ;p
1X
n=1
jcnj)jjfjjLp M ;p 1
k
1X
n=1
(jcnj)
!
:
Finally taking the in mum of the above inequality over all representations of f we have:
jjfjjLp M ;p 1
k kfkB ( ; )
:
In other words, B ( ; ) is weighted continuously contained in Lp[ ; ]. Now consider g2
A ( ; ). Using Theorem 3.3, g2B ( ; ) such that jjgjjB ( ; ) 2 12 jjgjjA ( ; ), and
we have A ( ; ) weighted continuously contained in Lp[ ; ]: Indeed, there is a constant
M ;p such thatjjgjjLp M ;p 1
k kgkB ( ; )
and we have the following relation between
Lp and A ( ; ):
jjgjjLp M ;p 1
k kgkB ( ; )
M ;p 1
k 2
1
2
jjgjjA ( ; )
:2
We now have all of the tools needed to demonstrate that A ( ; ) and B ( ; ) are
complete weighted metric spaces. We will provide the proof for the B ( ; ) case since the
A ( ; ) case is similar.
25
Theorem 3.6 (Completeness of A ( ; ) and B ( ; )) For 2(0;1) the spaces A ( ; )
and B ( ; ) are complete.
Proof. In order to demonstrate the completeness of B ( ; ), we apply Theorem 3.4. To this
end, let hfni be an absolutely summable sequence in B ( ; ), then P1n=1kfnkB ( ; ) <1.
Let cnm 2R such that fn(t) = P1m=1cnmbIInm(t) with P1m=1 (jcnmj) <1 for each n2N.
We must show that P1n=1fn converges in B ( ; ). By de nition of kkB ( ; ), for all > 0
there is a representation of fn such that
1X
m=1
(jcnmj) kfnkB ( ; ) + 2n
for each n. Let f(t) = P1n=1fn(t), then f(t) = P1n=1P1m=1cnmbIInm(t) and
kfkB ( ; )
1X
n=1
1X
m=1
(jcnmj)
1X
n=1
kfnkB ( ; ) +
1X
n=1
2n =
1X
n=1
kfnkB ( ; ) + <1:
Applying Theorem 3.4, we conclude that B ( ; ) is complete.2
3.3 Relationships within B ( ; ) and A ( ; )
The following discussion presents two theorems regarding the relation of these spaces
for di erent measures, alpha values and base functions . Theorems for both spaces are
presented below while the proof is given for only the B ( ; ) case. The proof for the
A ( ; ) is again analogous.
Theorem 3.7 (Relationship of A spaces) Given A 1( 1; 1) and A 2( 2; 2), if 2
1 and there exists k1;k2 2 R+ such that 1(t) k1 2(t), and 1(E) k2 2(E) for all
1; 2 measurable sets E [ ; ] then A 2( 2; 2) A 1( 1; 1). Moreover, there exists
a constant M2R+ such that kfkA 1( 1; 1) MkfkA 2( 2; 2).
26
Theorem 3.8 (Relationship of B spaces) Given B 1( 1; 1) and B 2( 2; 2), if 2
1 and there exists k1;k2 2 R+ such that 1(t) k1 2(t), and 1(E) k2 2(E) for all
1; 2 measurable sets E [ ; ] then B 2( 2; 2) B 1( 1; 1). Moreover, there exists
a constant M2R+ such that kfkB 1( 1; 1) MkfkB 2( 2; 2)
Proof of Theorem 3.8.
Let B 1( 1; 1), B 2( 2; 2) and k1;k2 2R+ be given such that 2 1, 1(t) k1 2(t),
and 1(E) k2 2(E). Let f2B 2( 2; 2) then
f(t) =
1X
n=1
cnbIIn (t) and
1X
n=1
2(jcnj) <1
Substituting the de nition of weighted generalized special atom of Type II we have:
f(t) =
1X
n=1
cn Bn(t)
22 (Bn)
=
1X
n=1
cn
1
1 (Bn)
22 (Bn)
Bn(t)
11 (Bn):
Now, we have:
11 (Bn)
22 (Bn)
k 12 12 (Bn)
22 (Bn) = k
1
2
( 1 2)
2 (Bn) k
1
2 ( 2)
( 1 2)([ ; ]):
Let M1 = k 12 ( 2)( 1 2)([ ; ]).
Utilizing conditions (2.1) and (2.2) on C :
1X
n=1
1(
cn 11 (Bn)
22 (Bn)
) =
1X
n=1
1(jcnj
1
1 (Bn)
22 (Bn))
1X
n=1
1(M1jcnj)
1X
n=1
1(M1) 1(jcnj):
27
Continuing the above inequality and using the fact that 1 k1 2,
1X
n=1
1(M1) 1(jcnj) = 1(M1)
1X
n=1
1(jcnj) 1(M1)k1
1X
n=1
2(jcnj):
So,
1X
n=1
1(
cn 11 (Bn)
22 (Bn)
) <1
Thus, f 2B 1( 1; 1) and B 2( 2; 2) B 1( 1; 1) Finally, letting M = k1 1(M1) and
taking the in mum of both sides of the above inequality over all representations of f we
conclude kfkB 1( 1; 1) MkfkB 2( 2; 2):2
28
Chapter 4
Major Results
Armed with the basic properties of A ( ; ) and B ( ; ) we am now able to prove
deeper results. This chapter discusses the duality of the new spaces as well as interpolation
of operator theorems.
4.1 H older-Type Inequalities
To nd the dual spaces of A ( ; ) and B ( ; ), a rst step is to nd a H older-type
inequality for each space. Essentially, we desire to nd a function space X associated with
B ( ; ) such that for g2X and f2B ( ; ) we have a result similar to:
Z
g(t)f(t)d (t)
kgkXkfkB
( ; )
Similarly, we would like to nd a function space Y to couple with A ( ; ) for a H older-type
inequality.
Consider the case for B ( ; ). As in previous proofs, we rst consider one weighted
generalized special atom of Type II. For g in our arbitrary function space X we then have:
Z
g(t)bII(t)d (t)
=
Z
g(t) B(t) (B)d (t)
= 1
(B)
Z
B
g(t)d (t)
:
Thus, in order to establish our desired inequality we would like a function space X for g which
will provide the bound above. Recalling the de nition of Lip( ; ) given in De nition 2.12,
we see that Lip( ; ) is the desired candidate space to couple with B ( ; ). Similarly we nd
that ( ; ) given in De nition 2.13 is the desired candidate space to couple with A ( ; ).
The next two theorems and proofs validate the above choices for the desired inequalities.
29
Theorem 4.1 (H older?s-type inequality for B ( ; )) For f2B ( ; ) and g2Lip( ; )
the following inequality holds for 2(0;1):
Z
g(t)f(t)d (t)
jjgjjLip( ; ) 1 k jjfjjB
( ; )
where k is the constant given in condition (2.3) of Class C functions.
Proof. Let f 2B ( ; ) be a weighted generalized special atom of Type II, that is f(t) =
bII(t), and let g 2 Lip( ; ). Then, by the argument in the previous paragraph and the
de nition of Lip( ; ), we have:
Z
g(t)f(t)d (t)
=
Z
g(t)bII(t)d (t)
= 1
(B)
Z
B
g(t)d (t)
jjgjjLip( ; ):
Now, let N 2 N and fN be a nite combination of Type II weighted generalized special
atoms. Then, fN(t) = PNn=1 cnbIIn (t) and we have:
Z
g(t)fN(t)d (t)
=
Z
g(t)
NX
n=1
cnbIIn (t)d (t)
=
NX
n=1
cn
Z
g(t)bIIn (t)d (t)
NX
n=1
jcnj
Z
g(t)bIIn (t)d (t)
NX
n=1
jcnjjjgjjLip( ; ) =jjgjjLip( ; )
NX
n=1
jcnj:
Applying Lemma 3.1 gives the result:
Z
g(t)fN(t)d (t)
jjgjjLip( ; ) 1
k
NX
n=1
(jcnj)
!
:
Taking the in mum of the above inequality over all representations of fN, we conclude
Z
g(t)fN(t)d (t)
jjgjjLip( ; ) 1 k kfNkB
( ; )
: (4.1)
30
In order to extend equation (4.1) asN!1takef2B ( ; ), then there are real coe cients
cn such that f(t) = P1n=1cnbIIn (t) with P1n=1 (jcnj) 1. Now for N2N de ne fN(t) =
PN
n=1cnb
II
n (t), then fN 2B ( ; ) and it is clear that limn!1fN(t) = f(t).
Let rN = R g(t)fN(t)d (t), then rN 2R since from equation (4.1) we know
jrNj jjgjjLip( ; ) 1
k kfNkB ( ; )
:
Consider the sequence hrNi. Let M2N such that N >M, then
rN rM =
Z
(fN(t) fM(t))g(t)d (t)
and
jrN rMj jjgjjLip( ; ) 1
k kfN fMkB ( ; )
:
Also,
kfN fMkB ( ; ) =
NX
n=M+1
cnbIIn
B ( ; )
NX
n=M+1
(jcnj)
SinceP1n=1 (jcnj) <1, we knowPNn=M+1 (jcnj)!0 asN;M!1, and hencekfN fMkB ( ; ) !
0 as N;M!1. Having 2C , we know (0) = 0 and is strictly increasing and it follows
that:
jrN rMj jjgjjLip( ; ) 1
k kfN fMkB ( ; )
!0 as N;M!1:
Thus,hrNiis a Cauchy sequence in R and convergent in R since R is complete. In other words,
lim
N!1
rN = lim
N!1
R
g(t)fN(t)d (t) exists. We now demonstrate that this limit is independent
of the sequence hfNi converging to f. To this end, let hhNi be a sequence of functions in
B ( ; ) that converges to f. Then we must show:
lim
N!1
Z
g(t)hN(t)d (t) = lim
N!1
Z
g(t)fN(t)d (t):
31
Consider then,
lim
N!1
Z
(fN(t) hN(t))g(t)d (t)
lim
N!1
jjgjjLip( ; ) 1
k kfN hNkB ( ; )
:
We next examine kfN hNkB ( ; ):
kfN hNkB ( ; ) =kfN f +f hNkB ( ; ) kw
kfN fkB ( ; ) +kf hNkB ( ; )
:
Since kfN fkB ( ; ) !0 and kf hNkB ( ; ) !0 as N!1, we have
lim
N!1
Z
(fN(t) hN(t))g(t)d (t)
!0 as N!1:
and we conclude that the limit as N!1 is independent of the sequence used to converge
to f in B ( ; ). Hence, we de ne
lim
N!1
Z
fN(t)g(t)d (t) :=
Z
f(t)g(t)d (t):
With this result, we can now take the limit as N!1 of equation (4.1) to obtain our nal
result:
lim
N!1
Z
fN(t)g(t)d (t)
=
Z
f(t)g(t)d (t)
lim
N!1
jjgjjLip( ; ) 1
k kfNkB ( ; )
=jjgjjLip( ; ) 1
k kfkB ( ; )
:2
Theorem 4.2 (H older?s-type inequality for A ( ; )) For f2A ( ; ) and g2 ( ; )
the following inequality holds for 2(0;1):
Z
g(t)f(t)d (t)
jjgjj ( ; ) 1 k jjfjjA
( ; )
where k is the constant given in condition (2.3) of Class C functions.
32
Proof. Let f 2A ( ; ) be a weighted generalized special atom of Type I, that is f(t) =
bI(t), and let g 2 ( ; ). Then there are -measurable sets X;A;B 2 [ ; ] such that
X = ASB, ATB =;, (A) = (B), and we have
Z
g(t)f(t)d (t)
=
Z
g(t)bI(t)d (t)
=
Z
1
(Xn)[ An(t) Bn(t)]g(t)d (t)
=
1
(X)
Z
A
g(t)d (t)
Z
B
g(t)d (t)
jjgjj ( ; ):
Now, let N 2 N and fN be a nite combination of Type I weighted generalized special
atoms. Then fN(t) = PNn=1cnbIn(t), and we have:
Z
g(t)fN(t)d (t)
=
Z
g(t)
NX
n=1
cnbIn(t)d (t)
=
NX
n=1
cn
Z
g(t)bIn(t)d (t)
NX
n=1
jcnj
Z
g(t)bIn(t)d (t)
NX
n=1
jcnjjjgjj ( ; ) =jjgjj ( ; )
NX
n=1
jcnj:
Applying Lemma 3.1 and taking the in mum over all representations of fN gives the result
Z
g(t)fN(t)d (t)
jjgjj ( ; ) 1 k kfNkB
( ; )
: (4.2)
In a similar fashion to the previous proof, we can show that if a sequence hfNi converges
to f in A ( ; ), then lim
N!1
R
g(t)fN(t)d (t) exists and is independent of the choice of the
sequence used to converge to f in A ( ; ). Thus, for such a sequencehfNiand g2 ( ; )
we de ne:
lim
N!1
Z
fN(t)g(t)d (t) :=
Z
f(t)g(t)d (t):
Taking the limit as N!1 in equation (4.2) above provides our desired result:
Z
g(t)f(t)d (t)
jjgjj ( ; ) 1 k jjfjjA
( ; )
:2
33
The following Lemma is useful in applying the H older?s-type inequality in Theorem 4.1
in the search for the dual of B ( ; ).
Lemma 4.1 Let bII(t) be a weighted generalized special atom of Type II, then for 2(0;1)
bII
B ( ; )
= (1):
Proof. Let 2 (0;1) and p = 1 , then for f 2 B ( ; ) by Theorem 3.5, f 2 Lp with
jjfjjLp 1
k kfkB ( ; )
since M ;p = 1 in this case. Now let f be a weighted generalized
special atom of Type II, then
kfkpLp = bII 1 L
1
=
Z
B(t)
(B)
1
d (t) = 1 (B)
Z
B
d (t) = 1:
By the de nition of kkB ( ; ) we have
1 = bII L
1
1
k bII B
( ; )
1 (k (1)):
Taking of both sides of the above inequality we obtain
(1) k bII B
( ; )
k (1)) (1)k
bII B
( ; )
(1)
In other words, kbIIkB ( ; ) = (1).2
We can now examine the duality of A ( ; ) and B ( ; ).
4.2 Duality and Representation
Perhaps the most signi cant result of this dissertation is the following theorem regarding
the dual of B ( ; ). In fact, the dual of B ( ; ) is a characterization of Lip( ; ).
34
Theorem 4.3 (Duality and Representation for B ( ; )) ? is a weighted bounded lin-
ear functional on B ( ; ) if and only if there is a unique g2Lip( ; ) so that for all f 2
B ( ; ) we have ?(f) = R f(t)g(t)d (t). Moreover, k?k is equivalent to kgkLip( ; ). That
is there are absolute real constants c1 and c2 such that c2kgkLip( ; ) k?k c1kgkLip( ; ).
In other words, the dual space of B ( ; ) is equivalent to Lip( ; ). That is B ( ; ) =
Lip( ; ).
Proof. Let g 2 Lip( ; ). De ne ?g : B ( ; ) ! R by ?g(f) := R f(t)g(t)d (t) for
f 2 B ( ; ). By Theorem 4.1, ?g is a weighted bounded linear functional since j?gj
jjgjjLip( ; ) 1 k jjfjjB ( ; ) and the integral operator is linear. Now, de ne as follows:
: Lip( ; )!B ( ; ); g7! (g) = ?g:
Then is one-to-one:
Let g1;g2 2Lip( ; ) such that (g1) = (g2), then for f2B ( ; ) we have
Z
f(t)g1(t)d (t) =
Z
f(t)g2(t)d (t))
Z
f(t) (g1(t) g2(t))d (t) = 0
Hence, we conclude that g1(t) = g2(t) almost everywhere which is su cient for our purposes
by the de nition of kkLip( ; ).
It remains to show that is onto:
Let ?2B ( ; ), then we must show that there is a unique g2Lip( ; ) such that (g) = ?
or in other words, = ?g. For any -measurable set E in [ ; ], de ne (E) := ?( E).
Note that by this de nition is a signed measure, see [34]. We show that << . That
is, we show that if (E) = 0 then (E) = 0. By the de nition of weighted bounded
linear functionals, De nition 2.8, there are constants M;k> 0 and a continuous real valued
function with (0) = 0 such that
j (E)j=j?( E)j M
kk EkB ( ; )
:
35
Since E(t) = (E) E(t) (E), we have k EkB ( ; ) ( (E)). Now due to the continuity of
and if (E)!0 then k ( (E))!0 which combined with the above inequality implies
j (E)j M
kk EkB ( ; )
M (k ( (E)))!0 as (E)!0:
Therefore, if (E) = 0 then (E) = 0 and we have << . By the Radon-Nikodym
Theorem, there is a -measurable function g such that (E) = REg(t)d (t). Thus,
?( E) =
Z
E
g(t)d (t) =
Z
E(t)g(t)d (t) and ?
E
(E)
=
Z
E(t)
(E)g(t)d (t)
by the linearity of ?. Hence, ?(bII) = R bII(t)g(t)d (t) for any weighted generalized special
atom of Type II. Now let fN be a nite linear combination of Type II atoms, then there
are real coe cients cn such that fN(t) = PNn=1cnbIIn (t) and we have
?(fN) = ?
NX
n=1
cnbIIn (t)
!
=
NX
n=1
cn? bIIn (t) =
NX
n=1
cn
Z
bIIn (t)g(t)d (t)
=
Z
NX
n=1
cnbIIn (t)g(t)d (t) =
Z
fN(t)g(t)d (t)
so
lim
N!1
?(fN) = lim
N!1
Z
fN(t)g(t)d (t):
Now, let f2B ( ; ). Then f(t) = P1n=1cnbIIn (t) and similar to previous proofs, we de ne
the sequence hfNi by fN(t) = PNn=1cnbIIn (t). It is clear that hfNi!f as N !1. Since
? is a weighted bounded linear functional, we have lim
N!1
?(fN) = ?(f). So, by a similar
argument to the proof of Theorem 4.1 we conclude:
?(f) = lim
N!1
Z
fN(t)g(t)d (t) =
Z
f(t)g(t)d (t):
36
Thus, all that remains to show is thatg2Lip( ; ). That is, we must show 1 (E) REg(t)d t <
K for some constant K. Consider then
1
(E)
Z
E
g(t)d (t)
= 1
(E)
Z
E(t)g(t)d (t)
=
Z
E(t)
(E)g(t)d (t)
= ?
E(t)
(E)
:
Again, using the de nition of weighted bounded linear functional, we can continue the above
equation with constants M;k> 0 such that
?
E(t)
(E)
M
k
E(t)
(E)
B ( ; )
!
:
Finally, applying Lemma 4.1 we have
1
(E)
Z
E
g(t)d (t)
M
k
E(t)
(E)
B ( ; )
!
M (k2 (1)) 1 with equivalent norms,
see [9]. We use this fact to characterize all functions g so that the multiplication operator
Tg de ned by Tgf := g f maps L(p;1) into L(p;1) and is bounded.
We rst review the motivation for this result. Given the linear multiplication operator
Tg de ned above, we would like to characterize all functions g so that for f 2 L(p;1),
kTgfkL(p;1) CkfkL(p;1) where C is a constant. To this end, recall that for f2L(p;1), then
there are constants cn and -measurable sets Bn in [ ; ] such that f(t) = P1n=1cn Bn(t)
withP1n=1jcnj 1p(Bn) <1since L(p;1) is equivalent to B( ; 1p). Thus we consider a generic
43
-measurable set B in [ ; ]. We note rst that for a function g we have:
B(t)g(t) =
8
>><
>>:
g(t); if t2B
0; if t is not in B
and thus ( B(t)g(t)) =
8
>><
>>:
g (t); if t2[0; (B)]
0; otherwise
where denotes the decreasing rearrangement of a function. As a result, for Tg B = B g
we have:
Z
(Tg B(t)) t1p 1d (t) =
Z
( B(t)g(t)) t1p 1d (t) =
Z (B)
0
(g(t)) t1p 1d (t) =
1p(B) 1
1p(B)
Z (B)
0
(g(t)) t1p 1d (t):
The above equation leads to our next de nition which provides the means to solve the
characterization required.
De nition 4.1 (Xp) Given a nite measure space ([ ; ];A; ), for B 2A, de ne the
space Xp as:
Xp =
(
g : [ ; ]!Rj 1
1p(B)
Z (B)
0
(g(t)) t1p 1d (t) 0 at C =
M1
M2
p1p2
p2 p1 as desired we would have
Mq1
1
A1 1
C 1A1 2 >Mq2
1A
2
1
C 1A2 2 )
M
1
M2
q 1
A1 1
>
1A
2
1
C 1A2 2+2 1A1 =
1A
2
1
C 1A2 1A1 =
1A
2
1
M
1
M2
q
:
So we must show 1A1 1 > 1A2 1 or 1A1 > 1A2 . If we show this inequality then one can
work backwards through the above inequalities showing g00(C) > 0 for C =
M1
M2
p1p2
p2 p1 which
guarantees this value of C is the minimum of g(C). Now p1 pp1 > p2 pp2 since p2 q(p2 p)pp2 which is equivalent to 1A1 > 1A2 . Therefore C =
M1
M2
p1p2
p2 p1 is
the minimum of g(C).2
56