Report

321601366 Made Benny Prasetya Wiranata

mathbenten@yahoo.com

July 25, 2017

1
1 Excercise and Extension in Chapter 2
1.1 Exercise 2.3.3
Theorem 1.1 (Theorem 2.3.2). Let a1 ≥ a2 ≥ · · · ≥ an ≥ 0 and b ∈ Cn with
k
X k
X
b∗i ≤ ai
i=1 i=1

for k ∈ {1, ..., n}. Then there exist a(1) , a(2) , ..., a(m) ∈ Cn with (a(l) )∗ = a and {λl } ⊂ [0, 1]
Pm
satisfying λl = 1 such that
l=1
m
X
b= λl a(l) .
l=1
In addition, if Φ is a real valued function on [0, ∞)n and if the function φ : Cn → R+ ,
defined by φ(c) := Φ(c∗1 , ..., c∗n ), is convex on Cn , then

φ(b) ≤ φ(a).

Proof. Let H be the convex hull of the set of points x ∈ Cn such that x∗ = a = (a1 , ..., an ).
We only need to show that b lies in H. Let f ∈ (Cn )∗ , linear functional on Cn . Since H
is bounded we write
β = max Re(f (c)).
c∈H

The result follows if we prove that |f (b)| ≤ β by the geometric version of Hahn-Banach
theorem on Cn (a point outside the convex hull can be separated by a hyperplane), Pne.g. [4].
Now since (CnP )∗ = Cn , there exists α ∈ Cn such that P n , f (c) =
for each c ∈ CP j=1 αj cj .
Let f ∗ (c) := nj=1 αj∗ cj . Then by (2.13), |f (b)| = | nj=1 αj bj | ≤ nj=1 αj∗ b∗j = f ∗ (b∗ ).
For each i ∈ {1, ..., n}, there exists j ∈ {1, .., n} such that αj∗ = |αi |. Let us define
a0i := aj e−iarg(ai ) for each i, then (a0 )∗ = (a01 , ..., a0n )∗ = a and f (a0 ) = f ∗ (a). Thus
f ∗ (a) ≤ β and
n n−1
f ∗ (b∗ ) = αn∗ b∗j + (αn−1
∗ − αn∗ ) b∗ + · · · + (α1∗ − α2∗ )b∗1
P P
j=1 j=1
n
αn∗ (α1∗ − α2∗ )a1 = f ∗ (a).
P
≤ aj + · · · +
j=1

Therefore |f (b)| < β by the previous inequality. The last inequality of Theorem 1.1 is
directly obtained by
m m m
!
X X X
φ(b) = φ λl a(l) ≤ λl φ(a(l) ) = λl φ(a) = φ(a).
l=1 l=1 l=1

1.2 Exercise 2.3.5

Corollary 1.2 (Corollary 2.3.4). Let a, b ∈ Rn be positive and ordered, and suppose that
k
Y k
Y
bj ≤ aj for any k ∈ {1, ..., n}.
j=1 j=1

2
Then, for any continuous, monotone increasing function g : [0, ∞) → R with t 7→ g(et )
convex, we have that
k
X k
X
g(bj ) ≤ g(aj ) for any k ∈ {1, ..., n}.
j=1 j=1

In particular, (2.14) can be obtained by taking g(x) = x.

Proof. We can assume that all aj are not-zero (otherwise the result easily follows). We
also assume that all bj are not-zero, otherwise we can take the value  small enough for
the zero term and take the limit as  → 0. Set ãj := γaj and b̃j := γbj for γ large enough
so that ãj , b̃j ≥ 1 for all j.
Let g̃(x) := g(γ −1 x). It is true that g̃ is continuous and increasing. We now show that
the map t 7→ g̃(et ) is convex. Indeed, for a, b ∈ [0, ∞) and s ∈ [0, 1] we have

g̃(eas+(1−s)b ) = g(γ −1 eas+(1−s)b) )

−1 −1
= g(e(a+ln γ )s+(1−s)(b+ln γ )) )
−1 −1
≤ sg(e(a+ln γ ) ) + (1 − s)g(e(b+ln γ )) )
= sg(γ −1 ea ) + (1 − s)g(γ −1 eb ) = sg̃(ea ) + (1 − s)g̃(eb ).

Next we define c̄j := ln c̃j forPevery j (c := a, b), f (x) := g̃(ex ), x ∈ R+ (f is convex

and increasing) and Φ(x) := nj=1 f (xj ) for every x ∈ [0, ∞)n . Then the hypotheses of
Theorem 1.1 are satisfied. Therefore for ā = (ā1 , .., ān ) and b̄ = (b̄1 , .., b̄n ) we get
n
X n
X
Φ(b̄∗ ) ≤ Φ(ā∗ ) ⇔ g(bj ) ≤ g(aj ).
j=1 j=1

Last, if we only consider the first k ≤ n terms, the result follows.

1.3 Exercise 2.3.13

Theorem 1.3 (Theorem 2.3.11). Let Φ be a symmetric norm on cc , then

(a) If a ∈ JΦ and lim aj = 0, then Φ(a) = Φ(a∗ ),

j→∞

n n
b∗j ≤ a∗j for any n ∈ N,
P P
(b) If a, b ∈ JΦ with lim aj = 0 and lim bj = 0, and if
j→∞ j→∞ j=1 j=1
then Φ(b) ≤ Φ(a).

(c) If Φ((1, 0, 0, ...)) = c, then ckak∞ ≤ Φ(a) ≤ ckak1 for any a ∈ JΦ ,

(0)
(d) Both JΦ and JΦ are Banach spaces,
(0)
(e) If α is a substochastic matrix and a ∈ JΦ , resp. a ∈ JΦ , then αa is in JΦ , resp.
(0)
αa ∈ JΦ , and Φ(αa) ≤ Φ(a),

(f ) If Φ is inequivalent to k · k∞ , then JΦ consists only of sequences which vanish at

infinity,

(g) If JΦ = JΨ , then Φ and Ψ are equivalent norms.

3
Proof. (a.) (?) For finite sequence a, b, assertion (b) follows from Theorem 2.3.2.
Since a ∈ JΦ , lim Φ((a1 , a2 , ..., an , 0, ..)) exists and is finite. Let n ∈ N, there exists
n→∞
m > n such that a∗1 , ..., a∗n are among |a1 |, ..., |am |. Then by (?) one gets

Φ((a∗1 , ..., a∗n , 0, ..)) ≤ Φ((a1 , ..., am , 0, ..)).

By taking the limit as n, m → ∞, one obtains Φ(a∗ ) ≤ Φ(a). Moreover, for each
n0 ∈ N, (?) implies Φ((a∗1 , ..., a∗n0 , 0, ..)) ≥ Φ((a1 , ..., an0 , 0, ..)). Hence Φ(a∗ ) ≥ Φ(a)
and it implies Φ(a∗ ) = Φ(a).

(b.) Since a, b ∈ JΦ , then a∗ , b∗ ∈ JΦ . For n, m ∈ N with n < m, we have Φ((b∗1 , ..., b∗n , 0, ..)) ≤
Φ((a∗1 , ..., a∗m , 0, ..)). Therefore Φ(b∗ ) ≤ Φ(a∗ ). Finally by (a), Φ(b) ≤ Φ(a).

(c.) We have Φ((0, .., 0, an , 0, ..)) ≤ Φ(a) ≤ ∞

P
n=1 Φ((0, .., 0, an , 0, ..)). The first inequality
is obtained by (b) and second one by the triangle inequality. Hence,

ckak∞ ≤ Φ(a) ≤ ckak1 .

The last inequality is true if a ∈ `1 .

(0) (0)
(d.) It is clear that Φ is a norm in JΦ and JΦ . We need only consider JΦ since JΦ
is closed in JΦ . Let {a(m) }m≥1 be a Cauchy sequence in JΦ . By assertion (c), we
(m)
obtain that {an }m≥1 is a Cauchy sequence in C. Then there exists a sequence
(m)
a = (an ) such that an → an for each n. Next by applying (b) we get
(m) (m) (n) (m) (n) (m)
Φ((a1 − a1 , .., aN − aN , 0, .)) = lim Φ((a1 − a1 , .., aN − aN , 0, .))
n→∞
≤ lim Φ(a(n) − a(m) ).
n→∞

Taking the limit as N → ∞ we discover that Φ(a − a(m) ) → 0 as m → ∞. Therefore

a ∈ JΦ and JΦ is complete.
(0)
(e.) First define b := αa. Since both JΦ and JΦ are Banach spaces, we only prove for
(0)
JΦ (a similar argument will apply to JΦ ). Now let bN := (b1 , ..., bN , 0, ..) for N ∈ N.
(0)
By Proposition 2.3.7 and assertion (b), we obtain Φ(bN ) ≤ Φ(a∗ ). Since JΦ is the
(0)
closure of cc , we only need to show that αa ∈ JΦ for a ∈ cc . Indeed if a ∈ cc ,
a = (a1 , ..., am , 0, ..), only
P m first columns of α count for αa. Given  > 0, we can
choose p ∈ N so that i>p |αij | <  for j = 1, .., m. Then

Φ((0, 0, ..., 0, (αa)p+1 , ..)) ≤ Φ(a) by (b).

Therefore Φ(αa) ≤ Φ(((αa)1 , .., (αa)p , 0, ..)) + Φ(a) < ∞ and so αa ∈ JΦ0 .

(f.) Suppose a ∈ JΦ and an 9 0. Then there exists a subsequence ā = (āj ) of a such

that āj → d 6= 0 for some d ∈ C and |āj − d| ≤ 21j . Let us write 1 := (1, 1, ...).
It is clear that ā ∈ JΦ and we will show that ā − d1 ∈ JΦ . Indeed by assertion
(b) and (c), we obtain Φ(ā − d1) ≤ Φ(( 12 , .., 21j , ...)) ≤ Φ((1, 0, ...)) < ∞. Therefore
|d|Φ(1) ≤ Φ(ā − d1) + Φ(ā) < ∞ and it implies kak∞ Φ(1) ≥ Φ(a). This inequality
and (c) show that Φ is equivalent to k · k∞ .

(g.) By the first inequality of assertion (c), one can shows that the identity map from JΦ
to JΨ is closed. Then the closed graph theorem implies that Φ and Ψ are equivalent.

4
 As usual, let us define Φ0 on cc as the conjugate norm of Φ by setting
 
 X 
Φ0 (b) := sup 

aj bj  | a ∈ cc , Φ(a) ≤ 1
 
j

for any b ∈ cc . One can show that the conjugate norm of Φ0 is Φ, (Φ00 = Φ).
Theorem 1.4 (Theorem 2.3.12). Let Φ be a symmetric norm on cc . Then
P
(a) |aj bj | ≤ Φ(a)Φ(b),
j

(0)
∗ (0)
∗
(b) JΦ = JΦ0 ( JΦ0 = JΦ ) in the sense that any continuous linear functional on
(0) P
JΦ has the form a 7→ aj bj for some b ∈ JΦ0 ,
j

(0)
(c) JΦ , resp. JΦ , is reflexive if and only if both Φ and Φ0 are regular.
Proof. (a.) Let a, b ∈ cc . WLOG assume a 6= 0. Let N be the biggest integer such that
aN and bN are not zero, then by (2.13) and theorem 2.3.11 (b),
N N
a∗i b∗i
P P
|ai bi | ≤
i=1 i=1
N
a∗i
≤ Φ(a∗ ) ∗
P
Φ(a∗ ) bi
i=1
≤ Φ(a∗ )Φ0 (b∗ ) = Φ(a)Φ0 (b).

(0)
(b.) By assertion (a), every b ∈ JΦ0 defines aPlinear functional on JΦ . For each m,
there is a ∈ cc so that Φ(a) = 1 and | an bn | = Φ0 (b1 , ..., bm , 0, ..). It implies
that Φ0 (b) is the norm of b as a linear functional. Thus, we only need to show
(0)
that any f ∈ (JΦ )∗ is of the P requisite form. Since (Cn )∗ = Cn , there exists a
sequence {bn }n≥1 with f (a) = n an bn for all a ∈ cc . Moreover, by definition of
Φ0 , Φ0 (b1 , .., bn , 0, ..) = sup{f (a) | a = (a1 , .., an , 0, ..), Φ(a) = 1} ≤ kf k. Therefore
(0)
b ∈ JΦ0 . Since b ∈ JΦ0 and f = b on cc , f = b on JΦ .
(0) (0)
(c.) (⇐) If Φ and Φ0 are regular then (b) implies (JΦ )∗∗ = JΦ .
(0) (0)
(⇒) If Φ is not regular, JΦ ⊂ JΦ ⊆ (JΦ )∗∗ , so JΦ0 is not reflexive. If Φ0 is
(0) (0)
not regular, (JΦ )∗∗ = (JΦ0 )∗ is strictly bigger than JΦ = (JΦ0 )∗ by the corollary of
(0) (0)
Hahn-Banach theorem. Therefore JΦ is not reflexive. Finally, if JΦ is not reflexive,
(0)
then JΦ = (JΦ0 )∗ is not reflexive.

1.4 Extension 2.5.8

Definition 1. A Banach space X is called uniformly convex if and only if for all 0 <  < 2,
the modulus convexity,
 
x + y 
δ() ≡ inf 1 −
2 kxk = kyk = 1, kx − yk ≥ 


is strictly positive.

5
Theorem 1.5. Let X be a uniformly convex Banach space and suppose that xn converges
to x weakly and that kxn k → kxk. Then kxn − xk → 0 ( xn strongly converges to x ).

Proof. We only consider the case x 6= 0. It suffices to show that kyn − yk → 0 as n → ∞

for yn := kxxnn k and y := kxk
x
(since lim kyn − yk implies lim kxn − xk). We claim that
n→∞ n→∞
k yn2+y k → 1. It implies kyn − yk → 0 as the consequence of the assumption. Indeed by
the Hahn-Banach theorem, there exists g ∈ X ∗ such that kgk = g(y) = 1. Then we obtain
g( yn2+y ) → g(y) = 1 since yn converges weakly to y and
  
yn + y  y n + y 
1≥ 2 ≥ g
  → 1.
2 

The squeeze theorem implies k yn2+y k → 1.

Here we state some known spaces which have the unifom convexity property.

Example 1.

(1.) Every Hilbert space is uniformly convex space.

(2.) Every closed subspace of a uniformly convex Banach space is uniformly convex.

(3.) Lp space (1 < p < ∞) are uniformly convex.

Theorem 1.6. For 1 ≤ p < ∞, Jp is a linear norm space. Moreover Jp is complete and
we have the following inequality :

(i) For 2 ≤ p < ∞, kA + Bkpp + kA − Bkpp ≤ 2p−1 (kAkpp + kBkpp ),

0 0 p0
(ii) For 1 ≤ p ≤ 2, kA + Bkpp + kA − Bkpp ≤ 2(kAkpp + kBkpp ) p ,
1 1
for A, B ∈ Jp and p + p0 = 1.

By the theorem 1.8, we see that Jp is a Banach space.

Theorem 1.7. For 1 < p < ∞, Jp is uniformly convex.

Proof. Let 0 <  < 2. For kAkp = kBkp = 1 and kA − Bkp ≥ ,

(i) (1 < p ≤ 2) Applying Theorem 1.8 (ii) we obtain,

0 0 0 0 0
kA + Bkpp + ≤ 2p − kA − Bkpp ≤ 2p − p .

0 0 1
Therefore 2δ() ≥ 2 − (2p − p ) p0 > 0.
1
(ii) (2 < p < ∞) Applying Theorem 1.8 (i) we obtain 2δ() ≥ 2 − (2p − p ) p > 0.

Hence Jp is uniformly convex.

Corollary 1.8 (Theorem 2.5.7). Let p ∈ [1, ∞) and {An } ⊂ Jp and A ∈ Jp . If

w − lim An = A and kAn k → kAk, then kAn − Ak → 0.
n→∞

The Corollary above is direct consequence of Theorem 1.8 and Theorem 1.6 for p ∈
(1, ∞).

6
References
[1] Barry Simon, trace Ideals and Their Applications second edition, AMS, 2005.

[2] C.A. McCarthy, cp , Israel J. Math. 5(1967), 249-271.

[3] M. Reed and B. Simon, Methods of Modern Mathematical Physics. I. Functional Anal-
ysis, Academic Press, New York, 1972.

[4] F.R. Deutsch and P.H. Maserick, Applications of the Hahn-Banach theorem in approx-
imation theory, SIAM Review 9 (1967), 516-530.