Download as pdf or txt
Download as pdf or txt
You are on page 1of 41

6

Schauder Bases

Definition 6.1
Let X be an infinite-dimensional normed linear space. A sequence {ed~l
in X is called a Schauder basis of X if for every x E X there is a unique se-
00
quence of scalars (ad~l' called the coordinates of x, such that x = L ajej .
j=l

Clearly, [e.} is a linearly independent set in X.


If a Banach space X has a Schauder basis, then it is separable because
finite rational combinations form a countable dense set.
If X is n-dimensional, the notion of Schauder basis coincides with the
algebraic basis . We will write {ed, span l e.}, etc ., to cover both the finite-
and infinite-dimensional cases.
The condition in the definition can be weakened. If every vector x E X
n
has a unique decomposition L aje j such that L ajej ~ x as n --t 00, it
j=l
already implies that [e.] is a Schauder basis (see, e.g., [Sinl]) .
If {e j} is a Schauder basis of a normed space X, then the canonical
projections Pn : X --t X are defined for n E N by Pn (~ aje j) = j~ ajej .
Lemma 6.2
Let {ej} be a Schauder basis of a normed space X. The canonical projections
P; satisfy:
(i) dim(Pn(X)) = n;
(ii) PnPm = PmPn = Pmin(m,n) ;
(iii) Pn(x) --t X in X for every x EX.
M. Fabian et al., Functional Analysis and Infinite-Dimensional Geometry
© Springer Science+Business Media New York 2001
162 6. Schauder Bases

Conversely, if bounded linear projections {P n};:"=1 in a normed space X


satisfy (i) -( iii), then Pn are canonical projections associated with some
Schauder basis of X .
PROOF : The set {ed is a linearly independent set in X . Thus (i) follows,
and (ii) is obvious. The property (iii) follows directly from the definition of
the Schauder basis.
Conversely, if projections Pn satisfy (i)-(iii), put Po = 0 and choose
e; E Pi(X) n Ker(Pi-d . Then
n
X = lim (Pn(x)) = lim (Pn(x) - Po(x)) = lim L(Pi(X) - Pi-1(X))
n~oo n~oo n~oo

i=1
00 00

L(Pi(X) - Pi-1(X)) = L (kiei


i=1 i=1
for some scalars (kj because dim(Pn(X)f Pn- 1(X)) = 1. The uniqueness of
00
(k j for x E X followsfrom the fact that if x = L fJj ej, then by the continuity
j=1
n
of Pn we get Pn(x) = L fJjej, and hence fJjei = Pj(x) - Pi-1(X) = (kjei .
j=1
Thus, {ed is a Schauder basis of X and Pn are projections associated with
{e.].
o
Fact 6.3
Let {ej} be a Schauder basis of a normed linear space X with canonical
projections Pn. IfsuPn IlFnl! < 00 (we say that Pn are uniformly bounded),
then {ed is also a Schauder basis of the completion X of X .
PROOF: We will show that the extensions Pn
of Pn on X satisfy (i)-(iii) of
Lemma 6.2. Since Pn(X) is finite-dimensional, it is closed in X and thus
Pn (X) = Pn(X), so (i) follows. (ii) is extended from Pn to Pn by the
continuity of Pn . Since Pn(x) ~ x for all x in a dense set X and Pn are
uniformly bounded, we have Pn(x) ~ x in X, so (iii) is also true. Since
en E Pn(X) n Ker(P n- 1), we get en E Pn(X) n Ker(Pn_ 1) for every n.
Therefore, Pn are canonical projections associated with the Schauder basis
{ed of X.
o
Lemma 6.4
Let {ed be a Schauder basis of a Banach space (X , 11 ·11). Define III· lion X
by Illxlll = SUPn II taje j I for x = f aie c. Then:
1=1 1=1
(i) I I · I I is a norm on X , {ei} is a Schauder basis of (X , I I · III), and the
canonical projections r; are uniformly bounded by 1 in III· I I ;
6. Schauder Bases 163

(ii) III . III is an equivalent norm on X .


PROOF : (i) : The triangle inequality and homogeneity of [[I · 111 are simple to
check. Since for every x E X we have IIxll = lim
n-oo i=l
It aiei II by Lemma 6.2,
we obtain that Illxli 2: IIxll for every x E X . This in particular means that
111·111 is a norm on X .
To show that {e;} is a Schauder basis of (X, III . III), we use Lemma 6.2.
The properties (i) and (ii) are straightforward . To check (iii), we note that
for x E X we have

IIIx - Pm(x)111 = sup IIPn(x) - PnPm(x)1I = sup IlPn(x) - Pm(x)lI- 0


n n~m

as m - 00. Finally, for mEN we estimate

IIlPmlll sup ![IPm(x)1II = sup sup IIPnPm(x)1I = sup sup IIPnPm(x)1I


mxm~l mxm~l n n mxm~l

sUP{ sup{IIPnPm(x)lI ; x with


n
s~p IIPi(x)1I ~
,
1}} ~ 1.
(ii) : We will show that X is complete in the norm III . III; that is, that
X C X, where X is the completion of X in 1[1 ·111 . By (i), we already know
that {ei} is a Schauder basis of X. Given x EX, there is a unique sequence
of scalars Q'i such that x = L Q'iei , where the convergence is in the norm
111 ·111· Since 111 ·1112: 11 ·11 on X, we get that L Q'iei is Cauchy in II ·11 and thus
convergent to some element x' E X in the norm 11 ·11. As shown in part
(i) , LQ'iei then converges to x' in the norm 111 ·111 . Thus x = x' E X . This
means that X is complete in I I · 1[1 .
The formal identity map Ix: (X, 111 ·111) - (X, II ·11) is a linear bijection of
a Banach space (X, 1 1 ·11) onto a Banach space (X, II · 11) , which is continuous
since III . III 2: II , 11· By the Banach open mapping principle, it follows that
I XI is continuous, which means that III . III is an equivalent norm on X .
o
Theorem 6.5 (Banach)
Let {ei} be a Schauder basis of a Banach space X. The canonical projections
Pn associated with {ei} are uniformly bounded.
The value bc{e;} = SUPn IlPnll is called the basis constant of {e;}.
PROOF: Define 111 ·111 as in Lemma 6.4. Then IllPnlll ~ 1 for every n , and since
III . III is an equivalent norm , the result follows.
o
Considering the vectors en, we see that IlPn II 2: 1, in particular bc{ e;} 2:
1. A Schauder basis {ei} is called normalized if lIeiII = 1 for every n . It
is called monotone if bc{ e;} = 1; that is, its associated projections satisfy
IlPnll = 1 for every n .
164 6. Schauder Bases

Let {e.} be a Schauder basis of a Banach space X. For j E Nand


=L = aj. Then II Pj (x ) - = IIh(x)ejll =
00
x aiei, denote h(x) Pj-1(x)1I
i=l
Ih(x)I ·lIejll and thus

Therefore, h E X* for every j . The functionals {td are called the asso-
ciated biorthogonal functionals (or coordinate functionals) to {e.}, and we
00
have x = L fi(x)e i for every x EX.
i=l
We will denote the biorthogonal functionals fi by ei and say that {ei ;ei}
is a Schauder basis of a Banach space X . It is a biorthogonal system , and
we have just proved that lIedllleill:S 2bc{ed . Note that {en is separating
for X.

Examples
(i) Any linear basis of a finite-dimensional Banach space X is a Schauder
basis of X . In particular, any Auerbach basis is a Schauder basis.
(ii) Any orthogonal basis of a Hilbert space H is a Schauder basis of H .
(iii) If X =Co or X =
lp for p E [1 ,(0) , then the sequence {ed of the
i
standard unit vectors ei = (0, . .. , 0, 1, 0, ...) is a Schauder basis of X .
All these statements are easy to verify.
(iv) Let {tj }];:1 be a sequence of distinct points in [0,1] such that t 1 = 0,
t2 = 1, and {tj} = [0,1]. Define projections r; from C[O, 1] into C[O, 1] by
P1(J) = 1(0) and Pn(J) to be the piecewise linear function with nodes at
tj, j = =
1, . . . , n and such that Pn(J)(tj) l(tj) for j =
1, . . . , n .
By Lemma 6.2 and the uniform continuity of continuous functions on
[0, 1], it follows that the projections Pn determine a monotone Schauder
basis of C[O, 1]. This basis is called the Faber-Schauder basis of C[O, 1].

(v) Let the functions hi be defined on [0,1] as follows: ho(x) = 1 for x E


[0 ,1]; h 1(x) = 1 for x E [0, t) and h 1(x) = -1 for x E [t, 1]; h2(x) = 1
for x E [0, ~) , h 2(x) = -1 for x E [~ ' t),
h2(x) = a for x E (t ,I] ;
h3(x) = 1 for x E [t , ~) , h3(x) =
-1 for x E (~ ' 1), and h3(x) a =
elsewhere, etc . Fix p E [1,(0) . The set {hd is linearly independent. Since
H = span {hd contains the characteristic functions of dyadic intervals, we
-L
get H P = Lp[O , 1] .
m n
For x = L (Xihi E H, define Pn(x) = L (Xihi (we can always assume
i=O i=O
that m ~ n by adding Ohj to the sum) . These projections satisfy (i)-(iii) of
Lemma 6.2, so by Fact 6.3, to show that {hd is a Schauder basis of Lp[O, 1]
we must only prove that Pn are uniformly bounded on (H,II ·ll p ) .
6. Schauder Bases 165

n n+1
Assume that I = L a; h; and 9 = L a; h; for some real numbers a; .
;=1 ;= 1
Then I and 9 differ only on some dyadic interval I , where I has a constant
value, say b, and 9 has value b + a n +1 on the first half of I and the value
b - a n +1 on the second half of I. Since for every p E [1, 00) we have Ibl P =
I~ (b + an+d + Hb - an-diP :s ~Ib + an+dP + ~Ib - an+l! P by convexity of
IxlP, we get 11/11 :s 11911 in Lp and it follows that the projections Pn have
norm at most 1. Thus {hd is a monotone Schaud er basis of Lp[O , 1]. By
inspection, we see that {hd is an orthonormal basis of L 2[0 , 1].
(vi) On the space c, define proj ections Pn for x = (x;) E c by P1(x) =
(X1 ,X1," ')' P2(X) = (Xl ,X2,X2, . . .), ... , Pn(x) = (X1 ,X2, ' '''X n ,xn, . .. ).
Then, for x E c and n EN ,

IIX - Pn(x)11 = 11(0, .. . , 0, xn+l - xn, Xn+2 - Xn , .. .)11 = max IXj - xnl--
J>n
°
as n -- 00. By Lemma 6.2, Pn generate a Schauder basis Xl = (1,1 , ...),
n
X2 = (0,1,1 , ), .. . , Xn = (0, .. . , 0,1, 1, .. .). Given Z = (Zl,Z2, " ') E c
and (0:'1 ,0:'2 , , O:'n) such that (Zl' Z2," " Zn , Zn, · .. ) = 0:'1X1 + .. .+ O:'nXn,
= = =
we calculate that Zl 0:'1 , Z2 0:'1 + 0:'2 , . . . , Zn 0:'1 +...+ O:'n. Therefore,
Pn(z) =0:'1 X1 + ... + O:'nXn and IIPn( z)II qIaxl zjl = JS:n
max{IO:'d ; 10:'1 + =
0:'2 1, .. . ,1 0:'1 + ...+ O:'nl} . The basis {x;} is called th e summing basis of c.

Let {e;} be a Schauder basis of X . Considering e, E X** we see that


{ei ;e;} is a biorthogonal system in X * .
Fact 6.6
Let [ e., ei } be a Schauder basis of a Banach space X with the canonical
projections Pn .
n n
(i) For n E N , P~(f) =L I( e;)ei =L e;(f)ei for every f E X * .
;= 1 ;= 1
(ii) P~(f) ~ f in X * for every I E X * .
(iii) {ei; ed is a Schauder basis ofspan{ei} with the canonical projections
P~. In particular, P~(f) -. I for every I E span{ei} .

00
PROOF: (i) : For n E N , IE X* , and x = L ei(x)e;, we have
;=1
n n

P~(f)(x) = f(Pn(x)) = I(Lei(x)e;) = LI(e;)ei(x) .


;= 1 ;= 1

(ii) : By the continuity of IE X * ,


n n

lim P~(f)(x)
n - oo
= n-+
lim'"
oo L....t
eiCx )/ (e;) = I( n-
lim'"
oo L....J
ei (x )e;) = I(x).
;= 1 ;= 1
166 6. Schauder Bases

(iii) : We easily check that P~P;;' = P~in(n ,m)' If f E span{ei}, then


P~(J) = f for n large enough and thus lim IIP~(J) - fll O. Since =
IlPnll = IIP~II, {P~} are uniformly bounded and we can apply Lemma 6.2
and Fact 6.3.
o
Definition 6.7
Let {ei; ei} be a Schauder basis of a Banach space X. It is called shrinking
00
if span {en = X* . It is called boundedly complete if L aiei converges

I
i= 1

whenever the scalars ai are such that SUPn Il i~ aiei < 00.
Every normalized shrinking basis {ei} has the property that ei ~ 0
because .lim (ek( ed) = 0 for every kEN . However, there exists a Banach
1->00

space with a normalized non-shrinking Schauder basis {ei} such that e, ~ 0


in X ([PeSz]) .
Fact 6.8
Let {e.; en be a Schauder basis of a Banach space X with the canonical
projections Pn . The following are equivalent:
(i) {e.; ei} is shrinking.
(ii) {ei} is a Schauder basis of X* .
lim II fl-{
(iii) n-+oo .}. II = 0 for every f E X* .
span e , s>n

PROOF: (i) => (ii) : Fact 6.6.


(ii) => (i) : If projections {P~} generate a Schauder basis of X * , then
P~ (J) -> f for every f and thus X* = span{ei}·
(i) ¢::::} (iii): Note that if P is a bounded linear projection of a Banach
space X onto P(X), then sup (J) sup(P*(J)) = =
IIP*(J)II and
P(Bx) Bx
Bp(x) C P(Bx) C IIPllBx n P(X) C IIPIIBp(x) .
Thus we have, for every f E X* and n ,

IIfl(Ix-Pn)(X)1I sup{f(x); x E B(Ix-Pn)(X)}


< sup{f(x) ; x E (Ix - Pn)(Bx)}
< sup{f(x); x E (11Pnll + I)B(Ix-Pn)(X)}'
Hence IIfl(Ix- Pn)(X)1I s
IIf - P~(J)II ::; (IIPnll + 1)llfl(Ix - Pn)(X )II . Thus,
{Ed is shrinking if and only if II fl(Ix- Pn)(X)II--- 0 for every f E X*.
o
The standard unit-vector basis of Co and £p , p E (1,00) , is shrinking,
whereas the standard un it-vector basis of £1 is not. The standard unit-
6. Schauder Bases 167

vector basis of i p is boundedly complete for p E [1,00), whereas the


n
standard unit-vector basis of Co is not, as the vectors (1, .. . , 1, 0, . .. ) show .

Proposition 6.9
Let {ei ;ei} be a Schaude r basis of a Banach space X. If {ej} is shrinking,
then the mapping T( x**) = (x**(en) is an isomorphism of X** onto the
space of all sequences (a j) such that 111(aj)1II = SUPnlli~ a ie ill < 00.
Moreover, if {ed is monotone, then T is an isometry.
PROOF : It is routine to check that 111·11 defines a norm on the vector space of
(ad such that 111(ai)111 < 00. Denote J{ =
bc{ed , and let Pn be the canonical
projections associated with {ei} . For x E X, x* E X* , and x** E X** , we
n n
have P~(x*) = L x*(ej)ei and P~*(x**)(x*) = L x**(ei)x*(ei) , so we
i=l j=l
n
can write P~*(x**) = L x**(ei)ej. Thus
j=l

III: x**(ei)ejll =
n

IIIT(x**)1II = sup sup 11P;*(x**)11 ~ I<lIx**1I


n j=l n

and T is a bounded linear operator with IITII ~ I< .


Now consider (ai)~l such that 111(ai) III < 00. Since X* is separable ({ed
is shrinking) and t~ ajej} is bounded in X** , there exists a w* -cluster

point x** of {.f:


,=1
ajei} , which satisfies x**(en = a j . Moreover , II x**1I ~

limsupllj~ a.e, II :::; 111(ai)lll · Thus, T(x**) = (a j) and IIT(x**)11 > Ilx**IL
which completes the proof.
D

Theorem 6.10
Let {e.: ei} be a Schauder basis of a Banach space X . If {ej} is boundedly
complete, then X is isomorphic to (span { en r.
PROOF : Let Pn be the canonical projections associated with {ed. Denote
Z = span{en and define J :X ---+ Z* by J(x):z 1-+ z(x) ; then J is a
bounded linear operator. We will show that J is an isomorphism of X onto
Z*.
Let x EX. Then, for every z E Z , we have IJ(x)(z)1 = Iz(x)1 ~ Ilzllllxll,
so IIJ(x)1I ~ IIxli. On the other hand, for n E N, find x* E Sx' such that
x*(Pn(x)) = IIPn(x)lI . Since P;(X*) = span{en?=l ' we have P~(x*) E
Z , and IIP~(x*)1I :::; J{ = bc{ed . By definition, J(Pn(x))(P~(x*)) =
(P~(X*))(Pn(X)) = x*(P;(x)) = x*(Pn(x)) = IlPn(x)1I and therefore
168 6. Schauder Bases

P;(x' ) _ 1 1
IIJ(Pn(x ))lI z ' ~ J(Pn(X))(IIP~ (x' )II) - 11P~ (x')lIllPn(x)1I ~ :K llPn(x)lI · By
the continuity of J, we have kllxll ~ IIJ(x)lIz· ~ Ilxll for every x EX.
We will now show that J maps X onto Z*. To th is end , observe first
that {ei, J(ei)} is a Schauder basis of Z , and let Pn denote its canonical
projections. Then P~(z*) ~ z* in Z* and SUPn II P~ II = SUPn II Pnll ~ K <
00; hence, for every z" E Z* and n E N , we have

II J (t
i=1
z*(en ei ) liz. = lit z*(enJ(e;)liz. = IIP~(z*)1I ~
i=1
K ·llz*lI·

Thus we have Il i~ z*(ei)eill ~ K · II J (~ z*(ei)ei) liz. ~ K 2 ·llz*ll· Since


00

the basis {ei} is boundedly complete, the series L z* (ei)ei is convergent


i=1
in X to some x EX. Because J is a continuous map,
n n

lim J(""' z*(enei) = lim""' z*(enJ(e;) = lim (P~(z*))


J(x) = n-+oo L....J n-+ oo L....J n-+oo
i=1 i=1
in th e norm topology of Z*. But P~(z*) ~ z* in Z*, so z" = J(x) and J
is an onto map .
D

Theorem 6.11 (James; see, e.g., [LiT2])


Let X be a Banach space with a Schauder basis {e;} . X is reflexive if and
only if {ei} is both shrinking and boundedly complet e.
PROOF : Let X be reflexive. By Fact 6.6, for every f E X *, we have
P~ (I) ~ f and thus P~ (I) s; f in X * since X is reflexive. Therefore
X* = spanW{e;} = span{ei} by Mazur's theorem, and {e;} is a shrinking
basis of X .
By Proposition 6.9, X ** is isomorphic to the space Y = {(a i) ; 111(ai)1II =
m;xlli~ aieill < oo} . Under this correspondence, X C X" corresponds to

Y1 = {(a i);fa iei converges} . Since X is reflexive, we get Y Y1 and =


.=1
{ei} is boundedly complete.
Conversely, if {e;} is shrinking and boundedly complete , then we have
= =
the above identification and Y1 Y; thus X X **, and X is reflexive.
D

Definition 6.12
A sequence {e;} in a Banach space X is called a basic sequence if {e;} is
a Schauder basis of span{e;}.
6. Schauder Bases 169

A basic sequence {ei} is called shrinking if it is a shrinking basis of


span{ed·
Proposition 6.13 (Banach)

°
Let {ed be a sequence in a Banach space X . {e.} is a basic sequence if and
only if there is K > such that for all n < m and scalars ai, . . . , am we
have

Moreover, the smallest such K is equal to bc{ ed.


PROOF : One implication is clear from

On the other hand, suppose that K satisfies Ili~ aieill ~ Klli~ aieill for

all ai and n ~ m . We define projections P« on span Ie.} by r; (f aie i) =


~=1
n
L: aiei for m > n and scalars ai, and observe that Pn have norm at most
i=1
K. We check that Pn satisfy (i)-(iii) of Lemma 6.2 on spanj e.} , so by
Fact 6.3, {ed is a Schauder basis ofspan{ed and bc{ed ~ K .
o
Not every separable Banach space admits a Schauder basis (Enflo ; see,
e.g., [LiT2]) . But we have the following statement .
Theorem 6.14 (Mazur)
Every infinite-dimensional Banach space contains a basic sequence. If X·
is separable, then X contains a shrinking basic sequence.
In the proof, we use the following lemma.

Lemma 6.15
Let Y be a finite-dimensional subspace of an infinite-dimensional Banach
space X . For every e > 0, there is x E Sx such that lIylI ~ (1+c:)IIY+Axll
for every y E Y and every scalar A.
PROOF : Let e E(O,1) . Let {yd~1 be an ~-net in Sy . For i E {1, .. . ,m},

°
choose yi E Sx- with y;(y;) = 1. Since X is infinite-dimensional , there is
x E Sx such that y;(x) = for every i = 1, . . . , m . We claim that x has
the desired property. Indeed, let y E Sy . Choose i E {1, .. . , m } such that
IIYi - yll < ~ ' Let A be a scalar . Then

lIy+ Axil ~ IIYi + >.xll- ~ ~ yi(Yi + AX) - ~ = 1- ~ ~ 1~£ '


170 6. Schauder Bases

Thus, given y E Y \ {O} and a scalar ,x, we have IIrrfu + rr%1r x ll ~ l~£'
o

n (1 + cn) :s
00

PROOF OF THEOREM 6.14: Given c > 0, find Cn > 0 with


n=l
1 + c. Let Xl E Sx be an arbitrary element. Using Lemma 6.15 , construct
inductively a sequence {Xn}~=2 in Sx such that , for every n ~ 1,
lIyll :s (1 + cn)lIy + 'xxnHIl for all y E span{xl ," " x n } .
By induction, for n < m and scalars al,"" am, we have

Iltaixill:s (l+cn) .. . (l+cm-dllI:aixill:s (1 + c)/II:aiXi/l.


i=l i=l i=l
By Proposition 6.13, {xd is a basic sequence and bc{ xd :s 1 + c.
Note that IlPnll :s n(1 + ci) -. 1 as n -.
00

i=n
00 . If X* is separable,
we may assume without loss of generality that the norm of X is Frechet
differentiable (Theorem 8.19). By Proposition 8.34, the basic sequence that
we have constructed is shrinking.
o
It is not known whether every separable Banach space X contains a
closed subspace Y such that both Y and XjY have a Schauder basis. We
also mention another open problem: If X is nonseparable, is there a closed
subspace Y of X such that X/Y is separable and infinite-dimensional? We
note that if X is separable and nonreflexive, there is a nonreflexive closed
subspace Y of X such that Y has a Schauder basis (Pelczyriski; see, e.g.,
[Dis2]).
Definition 6.16
Let [e,} be a basic sequence in a Banach space X , and let Ud be a basic
sequence in a Banach space Y. We say that {ed is equivalent to Ud if
for all sequences of scalars (ad, L aiei converges if and only if L adi
converges.
Fact 6.17
Let {ed be a basic sequence in a Banach space X, and let {td be a sequence
in a Banach space Y. The following are equivalent:
(i) {td is a basic sequence equivalent to {e.},
(ii) There is an isomorphism T of span{ed onto spanUd such that
T(ei) = Ii for every i.
(iii) There are C ll C 2 > 0 such that for all scalars al , ... , an we have
6. Schauder Bases 171

PROOF: (i) => (ii): Define a map T from span{ ed into spanUi} by
T(~ aie i) = i~ aik From the equivalence of {ed and {xd , we have
that T is well defined, one-to-one, and onto spanUi} .
00
We will now show that T has a closed graph. Indeed , if x k = L a fe i
i=l
=L =L
00 00 00

converge to x ai ei and T(x k


) af Ii converge to L cdi , then
i=l i=l i=l
by the continuity of the coordinate functionals we have af -+ ai , and in
the same way af -+ Ci for every i. Hence ai = Ci for every i, and thus
T(x k ) -+ Ladi = T(x) . By the closed graph theorem , T is continuous,
and by the open mapping theorem , T- 1 is continuous as well.
(ii) => (iii) : This follows easily with C 2 = liT/I, C 1 = liT-III .
(iii) => (i) : For n < m and scalars a1 , · · ·,a m , we have Ili~ adill <
C 1C2 bc{edlli~ adill, so by Proposition 6.13, Ud is a basic sequence . By
(iii) , we also have that L adi is Cauchy if and only if L aie i is Cauchy.
o
The following result is sometimes called the small perturbation lemma.
Theorem 6.18 (Krein, Milman , Rutman [LiT2], Valdivia [Va14])
Let {e.} be a basic sequence in a Banach space X , and let {en be the
coefficient funct ionals of th e basis {ei} of span{ed . A ssum e that Ud is a
00

sequenc e in X such that L Il ei - f d i liei II = C < 1. Th en:


i=l
(i) Ud is a basic sequence in X equivalent to {ed.
(i i) If span{ ed is complem ented in X , then so is spanUd.
(i ii) If {ed is a Schau der basis of X , then so is U i} . Moreo ver, let un
be the coefficient funct ionals of the basis Ud of X. Th en span {en =
spanun ·
PROOF : (i) : Ext end ei to functionals on X of the same norm . For x E X,
00 00

we have L Ilei(x)(e i - /;)11 ~ Ilxll L lI ei ll llei - fdl = CllxlJ, so S(x) =


i=l i=l
00
L ei (x)( e, - f i) defines a bounded linear operator from X into X with
i=l
IISII ~ C < 1. Let T =
lx - S . We have Ilx - T(x)11 IIS(x)1I ~ CllxlJ, =
and hence IIT(x)1I 2: (1 - C)llxll · Since 1 - C > 0, by Exercise 1.27, T is
an isomorphism into ; in particular , T(X) is closed.
We claim that T(X) = X . Assume T(X) =1= X . Since C < 1, by
Lemma 1.23 there is x E Sx such that dist(x, T(X)) > C, which con-
tradicts IIx - T(x) 1I ~ C; so T is an isomorphism of X onto X . Then,
using T( e;) = fi , we obtain that T maps span{ ed onto spanUd and (i) is
proved.
172 6. Schauder Bases

(ii): It follows from the proof of (i) using T and Fact 5.4.
(iii): Let T be the isomorphism of X onto X from (i). Since [e.} is a
Schauder basis of X, we get X = T(X) = T(span{ei}) = spanUd , so Ui}
is a Schauder basis of X.
k
Fix some i E N and denote x k = 2:
fi (ei )ej for kEN. Then xk E
i=1
span{ej} and, by Fact 6.6, xk ~ Ii. For x E Bx , we then have fi(x) =
00

2: fi(ei)ej(x) , so for k ~ i we estimate


i =1
00 00

lUi - xk)(x)1 = I I: fi(ei )ej(x)1 = I I: ft(ei -/j )ej(x)1


i=k+1 i=k+1
00

< IIfili I : lI ei -/jllllejil ~ 0 as k ~ 00 .


i=k+1
Since the estimate is independent of x E Bx, we get Ilfi - x kII ~ 0 as
k ~ 00 and thus fi E span {en · Therefore span c span {en ·un
Note that (T-I)*(et) = fi, so IIfill ~ II(T-I)*lllIeill for i E N . Thus ,
the series 2: IIfi IIl1ei - fill converges and we can reverse the roles of ei and
fi in the last paragraph, obtaining span{en C spanUt} .
o

Bases in Classical Spaces


Definition 6.19
Let {ed be a basic sequence in a Banach space X . A sequence of nonzero
Pj+!
vectors {ui} in X of the form Ui = 2: aiei with scalars ai and PI <
i=pj+1
P2 < ... is called a block basic sequence of {ed .
Note that a block basic sequence of {ei} is a basic sequence with basis
constant not greater than bc{e.]. Indeed, for k ~ I , we have
k Pj+l k Pj+l

III:ai I : aieill = III: I : aiaieill


i=1 i=pj+1 i=1 i=pj+1
I P j+l I

s bc{edllI: I : aiaiei ll = bc{edllI:aiuill ·


i =l i = p j + l i=l

The following result is often used when investigating subspaces of a given


space.
6. Schauder Bases 173

Theorem 6.20 (Pelczyriski [Pel2])


Let X be a Banach space with a Schauder basis {e;}. If Y is an infin ite-
dimensional closed subspace of X , then Y contains an infinite-dimensional
closed subspace Z with a Schauder basis that is equivalent to a block basic
sequence of {e;} .

PROOF: Let K = bc{e;} . Given pEN, let Wp be the finite-codimensional


subspace of X defined by

{x E X ; x= L
00

Wp = aiei} = span{ e;}i>p .


i=p+l

Then Wp n Y is infinite-dimensional , so there is Y E Sv n Wp . We will


inductively construct the two equivalent basic sequences .
00

Choose an arbitrary Yl = I: a} e, E Y with IIYll1 = 1. Find PI E N


i=1
PI
such that for Ul = I: a}ei E X we have I/Yl - ull1 < 4k ' Choose Y2 =
i=1
00 P2
I: alei E Sy n Wpl , and fix P2 E N such that for U2 = I: al ei we
i=PI+l i=PI+l
have II Y2 - u211 < 2 . ~2 K ' Continue in this manner. Then {ui} is a block
00

basic sequence of{ e;} . Since I: IIYi -ui II <


2k and lI u; II :S 2 bc{ ui} :S 2K,
i=1
by Theorem 6.18, {Yi} is a basic sequence in Y equivalent to {ui} , so the
subspace Z = span{Yi} has the desired property.
o
This procedure of finding vectors with almost successive supports is
called the "sliding hump argument." An easy modification provides the
following corollary.

Corollary 6.21 (Bessaga-Pelceyriski selection principle; see, e.g., [LiT2])

°
Let X be a Banach space with a Schauder basis {e;} . If a sequence {x n }
satisfies inf IIxnll > and X n ~ 0, then some subsequence {x n k } of {x n }
is a basic sequence equivalent to a block basic sequence of {e;}.

We will now investigate the structure of subspaces of lp spaces .

Proposition 6.22
Let X be Co or lp with P E [1 ,(0). If {uj} is a normalized block basic
sequence of the standard unit-vector basis {e;} , then {Ui} is equivalent to
{e;} , span{ u;} is isom etri c to X , and there is a projection of norm one of
X onto span{ui}.
174 6. Schauder Bases

Pi+i Pi+i
PROOF : We present the prooffor f p . Let Uj = I: Aiei with I: IAilP =
i=Pi+1 i=Pi +1
1 for j EN . Then
m Pi+i m Pi+i
(L L (L lajlP L
1 1

lajlPIAdPY = IAilPr
j=l i=Pi +1 j=l i=Pi+1

(tlajIP)t = lit ajej II,


j=l j=l
so {ed and {ud are equivalent and T(I:ajuj) = I:ajej is the isometry.
To find a projection, for every j E N choose uj E span{ e, }f~;~+l c for f;
which lIuj II = uj (Uj) = 1. Then uj (Uk) = 0 for k # j and the operator P
00
from X into span{Uj} defined by P( x) = I: uj (x )Uj is a linear projection
j=l
PHi
of X onto span{uj} . For x = I:aiei E X , we have /uj(x)/P:s I: lad P
i=Pi+ 1
for every j since lIujll = 1 and thus
Pi+i
= L L =L
00 00

IIP(x)IIP luj(x)IP · IAd P luj(x)IP


j=l i=Pi +1 j=l
00 Pi+i
< L L lad
j=l i=Pi +1
P
= IlxliP .
This shows that IIPI/ = 1.
o
Combining Theorem 6.20 and Proposition 6.22, we have the following
theorem.
Theorem 6.23
Let X be Co or f p with p E [1, 00). If Y is an infinite-dimensional closed
subspace of X, then Y contains a subspace Z that is isomorphic to X and
complemented in X .
The method of the proof of the following result is called the Pe/czynski
decomposition method.
Theorem 6.24 (Pelczyriski; see, e.g., [LiT2])
Let X be Co or f p , p E [1,00). If Y is an infinite-dimensional complemented
subspace of X then Y is isomorphic to X .
I

PROOF: Let Xl be a complement ofY in X ; then X is isomorphic to Y67X1 .


Denote by '" the equivalence of being isomorphic. By Theorem 6.23, there
is an infinite-dimensional subspace Z of Y such that Z '" X and Z is
complemented in X, hence also in Y. Let Y '" Z 67 Y1 . We will use the fact
6. Schauder Bases 175

that both X EB X and (2: EBX )x ar e isomorphic to X (Exercise 6.1 5) . Thus

X EB Y '"'" X EB (Z EB Yl ) '"'" (X EB Z) EB Yl '"'" (X EB X) EB Yl


...... X EBY1'""'Z EBY1'""' Y.
On the other hand ,

(X EB X EB .. ')x EB Y '"'" ((Y EB Xl ) EB (Y EB Xd + .. -Ix EB Y


(Xl EB X l EB .. ')x EB (Y EB Y EB · · ')x EB Y
(X l EB Xl EB · · ·)x EB (Y EB Y EB . . ')x
((Y EB Xl) + (Y EB X l) + .. ·)x '"'" (X EB X EB· · ·)x '"'" X.
Thus X '"'" Y .
o
We note that for p E (1 ,00) , p f. 2, there are closed subspaces of f p
that are isomorphic to f p and not complemented in f p . On the other hand ,
if a subspace of f p is isome tric t o f p , th en it is necessarily complemented
([PeI2]). Also , if p E [1, 00) \ {2} , then th ere ar e infinite-dimensional closed
subspaces of f p t hat are not isomo rp hic to f p . A referen ce for results in t his
area is , for exam ple, [PeBe].
When comparing t he structure of Banach sp aces, an import ant rol e is
played by cer t ain class es of op erators . Let X, Y be Banach spaces and
T E B( X , Y ). We say that T is st rictl y singular if there is no infinite-
dimensi onal subspace Z of X such that TI
z is an isomorphism into Y . We
say that T is a compa ct operator if T( B x) is compact in Y . Not e that every
compact op erator is stri ctl y singul ar (Exercise 6.21) .

Proposition 6.25 (P itt ; see, e.g., [LiT2])


Let 1 ~ p < r < 00 . E very bound ed lin ear operator T f rom t ; into f p or
from Co into f p is compact.
PROO F: Assume th at T : f r ...... f p is not com pact . Sin ce f r is reflexive, th ere
is a seque nce {x n} in i ; such t ha t Xn ~ 0 in f r and IIT(xn)1I ~ E: for
some E: > 0 (Exercise 7.8). Then inf Ilxnll > 0, so using Coro llary 6.21 and
Proposition 6.22 first in f r and th en in f p , we find a subsequen ce {x nk} of
{x n} su ch that {x nk} is equivalent to the canonical basis oUr and {T( xnJ }
is equivalent to the cano nical basis of f p .
If (ak) E f r \ f p , then 2:ak xnk is convergent; hen ce T(2: akXn k)
2: akT( Xnk) E i; Thus 2: lak IP< 00, which is a contradiction .
The Co case follows by Theorem 7.7.
o
We t hus obtain th at f p , f q ar e not isomorphic for p f. q. Usin g The-
orems 6.23 and 6.25, we deduce that if p, q E [1, 00) and p f. q, then
the spaces f p and f q do not have isomorphic infinite-dimens ional closed
subspaces. Such spaces ar e called totally incomparable.
176 6. Schauder Bases

Theorem 6.26 (Dunford, Pettis, Pelczyriski; see, e.g., [DiUh])


Let X be a Banach space and T E B(co, X) . If T is not compact, then
there is a subspace Z of Co such that Z is isomorphic to Co and is an Tlz
isomorphism into X.
In particular, X contains a subspace isomorphic to Co . Before proving
t his t heorem, we fix some notation.
Let {Xi} be a bounded sequence in a Banach space X. A sequence {Yn}
is called an iI-average of {Xi} if it is of the form Yn = L aiXi, where
i EA n
{An} is a sequence of pairwise disjoint finite subsets of Nand L la;! = 1
iEA n
for all n. Note that if {x;} is equivalent to the standard unit-vector basis
of iI, then so is {Yn} .
Lemma 6.27
Let [e., en
be the canonical basis of co. Let V be a bounded subset of iI -
If V is not relatively compact, then there exist e > 0, sequences {x n } and
{Yn} in V, a sequence of positive numbers {An}, and an iI-average {zn} of
{en such that for every n E N , Ilxn -Ynll ~ e and II zn -An(x n -Yn)II S ~ -

Note that then An S 23£ for every n. Indeed, since II zn II = 1, the second
inequality implies IIA n(xn - Yn)1I S ~ ' while II xn - Ynll ~ c.
PROOF: Since V is not relatively compact, there is c > 0 and an infinite
subset W of V such that II x - yll > e whenever x, YEW , x # y. We
proc eed by induction . Suppose that xi , Yi ' Zj, Aj have been constructed
for j S n - 1. Let A j =
supp(Zj) and set B n =
U A j . Consider the map
j<n

T(f aiei) = .L =
aiei acting from i l into u; span{e;}iEBn -
.=1 'EBn
Since Un is a finite-dimensional subspace of Co, we can cover Tn(V) with
finitely many balls of rad ius t2 _Since W is an infinite set in V, we can
choose X n and Yn in W such that T(x n) and T(Yn) are in the same ball
and thus IIT(xn) - T(Yn)II S ~ -
Find N > max(Bn) such that II(Il. -PN)(x n)II < t2 ' II(It. -PN)(Yn)II <
tr Let c; = =
{i; i > N} and An N \ (B n U Cn). Decompose X n =
X n + x; + x~ and Yn = Y~ + Y; + Y~ so that supp(x~),supp(y~) CAn,
supp(x;) ,supp(y;) C n.; and IIx~II < t2 ' IIy~1I < t2 · Note that IIx~­
y~11 = IIT(xn) - T(Yn) II S ~ and

IIx~ - y~II ~ IIxn - Ynll-Il x~ - y~ II - IIx~II - IIy~II ~ 2c/3 .


Define An = IIx; - y;II- l and Zn = An(x; - Y;). Then eupp[z.,') CAn,
II znll = 1, and
II zn - An (xn - Yn)1I Anll(x~ - Y~) - (x n - Yn) II
< An(/Ix~ - y~II + IIx~II + IIYnIl 3 )
6. Schauder Bases 177

o
PROOF OF THEOREM 6.26 : (Bator [Bao]) 1fT is not compact, then also T*
is not a compact operator (Theorem 7.7). Because T*(Bx.) is not relatively
compact in.f 1 , from Lemma 6.27 we obtain e > 0, sequences {un} and {vn}
in Bx" a sequence of positive numbers {An}, and an .f1-average {zn} of
{en such that for every n E N we have IIT*(u n) - T*(vn )1I 2:: e and
II Zn - An (T* (Un) - T*(Vn)) II ~ ~ .
For Zn = 2: aieL put Pn = {i E An; ai > O} and N n = {i E An; aj <
iEA n
O}. Define Xn E Co by Xn = XPn - XNn for n E N . Then
if i = i.
if i =I j .
The space Z = {2:tnxn; {tn} E co} is isometric to Co because the xn's
have disjoint supports. Moreover , if {tn} E Co and lI{t n }1
1 Itjl 1 for = =
some i , then

sup IX*(T(L tnxn))1


x'EB x *

> ~I((uj - Vj)(T(LtnX n)) I


2t IAj (T*(Uj) - T*(Vj ))(L tjx;)1
IZj(2:tnxn)I-I(Zj - Aj(T*(uj) -T*(Vj)))(2: tnXn)1
>
2Aj
> 2t (1 - ~) = 4t 2:: ~ ,
so II T (X)II 2:: ~llxll for x E Z and T is an isomorphism into.
o
We will now show that the structure of subspaces of L p is different from
that of .fp .
Theorem 6.28 (Khintchine; see, e.g. , [LiT2])
If p E (1, (0), then Lp[O, 1] contains a complemented subspace isomorphic
to .f2 • LIlO, 1] contains a subspace isomorphic to .f2 •

Note that L1[0, 1] has no complemented subspace isomorphic to .f2


(Exercise 11.35).
In the proof of Theorem 6.28, we will use Rademacher's functions defined
by rn(t) = sign (sin(2 n 1rt)) for t E [0,1] and n E N. Clearly, rn E SLp [O,I)
for every p 2:: 1.
1
It is easy to observe that fo rl(t)rk(t) dt = Okl o Consequently, {rn} is
an orthonormal set in L 2[0 , 1]. In particular, II f anrnll = II(an )lIl2;
n;;:1 L2
178 6. Schauder Bases

that is, {rn} considered in L 2 is a basic sequence that is equivalent (with


constant 1) to the canonical basis of £2 . The behavior of {rn} in L p spaces
is described in the following result.
Lemma 6.29 (Khintchine inequality)
Let rnbe the Rademacher functions on [0,1] . For every p E [1,00), there
exist positive constants A p and Bp such that, for every aI, . . . , am,

By A p and Bp we denote the best possible constants in this inequality.


They are called Khintchine's constants and their values are known. We

r.
observed that A 2 = B 2 = 1. By the Holder inequality, it follows that if
2:: (f01Ifl
1 1

P > r, then (JOI Ifl


P dt) P
r
dt Consequently , A r ::; A p and
s; ::; s;
PROOF : By the last remark, it is enough to show that there exist Al > 0
and B 2 k < 00 for all kEN. We start with B 2k •

1
1 m 2k
(I>nrn(t)) dt
o

1r~:
n=l

L
1
Aal " .. ,aja~~ a~~
.. . (t) , .. r~~ (t) dt,
j
here the summation runs through all multiindices (aI, ' , , , aj) with L ai =
i =l
2k and 1 ::; nl ::; .. . ::; nj ::; m. By elementary combinatorics, Aa1". "aj =
;1)!..~j2:)!" Observe that f; nnl< ... <nk
r~~(t)dt = 1 if all ai are even, and it
is equal to 0 otherwise. Using {3i = a;/2 , we therefore write

where the summation runs through subsets ({31, " " (3i) of N such that
i
L {3i = k and 1 ::; nl ::; . , . ::; ni ::; m. Thus, we have
i=l
6. Schauder Bases 179

j
Since I: f3i = k, the minimum above is a minimum of a finite set of
i=l
positive numbers and so it is positive. Hence B 2 k exists finite .
m
To prove the existence of A 1, put f(t) = I: anrn(t). By Holder's in-
n= 1
equality used for p =
~ and q =
3, by the first part of the proof and our
observation before this lemma, we have
1
1 If(t)1 2dt 1 1
If(t)1 ~ If(t)lt dt ~ (1 1
If(t)1 dt) ~ (1 1
If(tW dt) t

< (1 1
If(t)1 dt) ~ B! (~ Ian 12 ) ~
1
(1 If(t)1 dt) ~ B! (1 1
If(tW dt) ~ .

Therefore (J01 If(t)1 dt r 2

2: B; t (J; If(tW dt r;
1

that is,

l 1 If (t )ldt 2: B;;2(1 1 If(t


Wdt)! = B;;2(~ lanI2)! .

o
PROOF OF THEOREM 6.28: Define a map T from £2 into L p[O ,l] by
00
T((a n )) = L: anrn, where {r n} are the Rademacher functions. By
n=l
Lemma 6.29, it follows that T is an isomorphism from £2 into L p[O,l].
If p 2: 2, then Lp[O , 1] is a subspace of L2[0, 1] (Exercise 1.8) . Let P be the
restriction to Lp [0,1] of the orthogonal projection of L 2[0, 1] onto span{rn } .
We have P(J)(t) = f: Uo1 f(s)rn(s) ds)rn(t). This projection is bounded
n=1
in Lp[O, 1] since IIP(J)IILp ~ BpIlP(J)liL2 ~ BpllfllL2 ~ BpllfllLp by the
Khintchine and Holder inequalities .
t
If p E (1 ,2) and q > 2 satisfies ~ + = 1, there is a subspace Y of
Lq[O, 1] isomorphic to £2 complemented by a projection P . Then P*(Y*) is
a complemented subspace of Lp[O, 1] isomorphic to £2 (Exercise 5.10).
o
One of the consequences is that £p is not isomorphic to Lp[O, 1] for p i=
2,00 . Indeed, by duality, it is enough to show that £p is not isomorphic
to Lp[O , 1] for p < 2. Suppose that £p is isomorphic to Lp[O, 1] for some
p < 2. By Theorem 6.28, £2 must be isomorphic to a subspace of £p , which
180 6. Schauder Bases

is a contradiction with Pitt's theorem. The situation is different if p = 00


because loo is isomorphic to Loo (Exercise 6.18). Also, L 2[O , 1] is isometric
to l2 by Theorem 1.38.
The spaces L p and L q are not isomorphic if p 1= q, p, q E [1,00). This
follows for instance using the notion of type and cotype. If 1 ~ p < q ~ 2,
then lp is not isomorphic to a subspace of L q (this can be proved for instance
using the notion of type). However, L p contains a subspace isometric to L q .
We refer to [PeBe] for a survey in this area .

Unconditional Bases
Recall that a series L: Xi in a Banach space X is unconditionally convergent
if L: CiXi converges for all choices of signs Ci = ±1 (Exercises in Chapter 1).
Definition 6.30
A Schauder basis {ed of a Banach space X is said to be unconditional if,
for every X E X, its expansion x = L: aiei converges unconditionally.
A sequence [e.] in a Banach space X is called an unconditional basic
sequence if it is an unconditional basis of span [e.}.

It is straightforward to check that the canonical basis {ei} of Co or lp,


p E [1,00), is an unconditional basis.
Also, as we saw in the notes following Theorem 1.38, every orthonormal
basis of a separable Hilbert space is unconditional. Note that every basis
equivalent to an unconditional basis is also unconditional.
Example
n
Let {ed be the standard unit-vector basis of co . For n E N , set Xn = L: ej.
i=l
00 00
We check that if x = L: Q:iei, then z = L: 13n xn , where 13n = Q:n - Q:n+1 .
i=l n=l
00
It follows that {13n}~=l forms a convergent series with L: 13n = Q:1 and
n=l

I/xl/oo = SUPk L~k 13n On the other hand, for every convergent series L: 13k,

we have x = f: 13k xk = f: (f: 13k) ej E co ·


k=l i=l k=i
We just showed that (co, 1/·1/00) is isometric to the space of convergent

series {13n} with the norm 1/(13n)1/ = SUPk L~k 13k The uniqueness of expan-
00
sion x = L: 13n xn follows by induction, using the fact that L: 13k = Q:i, where
i= i
Q:i is the standard i-th coordinate of x (13n is then necessarily Q: n - Q:n+d.
It follows that {x n } is a Schauder basis of co , which is not unconditional.
6. Schauder Bases 181

Indeed, L: (-~r Xn E Co , while L: ~Xn rt Co . The basis {x n } is called the


summing basis of Co .
Proposition 6.31
Let {ed be a sequence in a Banach space X. The following are equivalent:
(i) {ei} is an unconditional basic sequence.
(ii) There is a constant J{ such that for all scalars a1 , . . . , am and signs
Ci = ±l, we have

(iii) There is a constant L such that for all scalars a1, . . . , am and every
subset o of {I , .. . , m} we have

lII:aieili ~ Lllfaieill·
iEI1 i=1

We claim that the condition (iii) can be equivalently stated as


/Ii~ aiei/l ~ Lll i~ aieill whenever a eN.
Indeed, let a be any (even infinite) subset of N. Assuming (iii), we show
that L: aiei is Cauchy : Given e > 0, there is no such that for m > n> no
iEI1

we have Ili~ aieill < f · Considering a' = un {n , . .. , m} and b, = ai


for n ~ i ~ m , b, = °otherwise, we use the condition (iii) to see that
I .EI1,n<.<m
. L:. aiei" < c. Thus ~ a.e, is convergent . Passing to the limit, we
'EI1
get th~ claim.
However, in most applications, the finite-sum statement is easier to
handle because one does not need to discuss the convergence . A sim ilar
observation can be made about the condition (ii).
PROOF: (i) ==::} (iii) : Let Y = span{ed . Given a C N, define an operator
PI1 from Y into Y by PI1 (x) = L: a.e, for x = L:aiei . The operator PI1 is
iEI1
well defined since L: aiei converges whenever E a.e; converges (use (i) and
i E I1
Exercise 1.36). We now check that PI1 has a closed graph . Indeed, let x k -+ x
in Y for x k = L:afei, x = L:aiei , and PI1 (x k ) = L: afei -+ Y = L:biei.
i i EI1
From the continuity of the biorthogonal funetionals in Schauder bases, we
have af -+ ai for every i, and for the same reason, af -+ b, for every i.
Thus, b, =
ai for every i and hence PI1 (x) =
Y, meaning that PI1 has a
closed graph and is thus continuous.
Consider now the family of operators PI1 , o running through all subsets
ofN . We claim that for every fixed x = L:aiei E X , the family {PI1(x)} is
bounded . Indeed, from th e unconditionality we get that , given e > 0, there
182 6. Schauder Bases

is a finite set FeN such that I


II;~ a;e; < e whenever An F = 0 (see the
exercises in Chapter 1). From this, the boundedness of {Pu(x)} for every
x E X follows. The Banach-Steinhaus uniform boundedness principle gives
that the operators P, are uniformly bounded by some L .
(iii) ===> (ii): Given scalars al, "" am and signs e, = ±1, we define
a = {i; e, = I} and a' = {I, . .. , m} \ a , Then
III:a;e; - I: a;e;11
;Eu ;Eu'

< III: a;e; II + III: a;e;II ~ 2Lllf a;e; II ·


;Eu ;Eu' ;=1

(ii) ===> (iii): Given a1, oo.,a m and a C {1,oo.,m}, we define c; = 1 if


i E a and e, = -1 for i E {I, ... , m} \ a . Then
1 m
211I:(c;a;ei+ a;e;)11
;=1
1 1
211I: CiaieiII + 211I: aieil ~ KIII:aieill ·
m m m
<
i=l i=l ;=1
(ii) and (iii) ===> (i): Given n < m and scalars al,' . . , am , we use (iii) with
a = {I, . .. , n} to see that, by Proposition 6.13, {ei} is a basic sequence with
bc{e;} S; L . Now let
00
L: aiei
i=l
be a convergent series. Given C::i = ±1 , using
(ii) we show thatlIi~ C::ia ieill ~ 2Klli~ aieill· Thus 2:C::i a;ei is Cauchy,
and hence convergent. This shows that 2: aiei converges unconditionally.
o
The best possible constant K from the condition (ii) in Proposition 6.31
is called the unconditional basis constant of {ei} and is denoted by ubc{ e.}.
In the preceding proof we have shown that L ~ K and bc{e;} ~ ubc{ e;}.
A natural question is whether every Banach space contains an uncondi-
tional basic sequence (cf., Theorem 6.14). This long-standing problem was
answered in the negative by Gowers and Maurey ([GoMa]).
Theorem 6.32 (James; see, e.g., [LiT2])
Let X be a separable Banach space. If X has an unconditional Schauder
basis that is not boundedly complete, then X contains an isomorphic copy
of co·
In the proof, we will use the following statement.
Lemma 6.33
Let {e;} be an unconditional basic sequence in a Banach space X. Then,
6. Schauder Bas es 183

for all scalars (aj) such that L a jej converges and all bounded sequ ences of
scalars {Aj} , we have

IIfAja ieili ~ ubc{ej}(s~pIAjl)lIfaiejll ·


i=1 • j=1

PROOF: Gi ven mEN , pick x* E Sx' so t hat x* (~Ajajej) = lIi~ Aja ieill
and define e, by e; = 1 if ajx*(ei ) ~ 0 and e, = -1 if ajx*(ej) < O. Then
m m
li t Ajaiei II ~ L lAd lajx'( ei)1 ~ ( sup lAd) L E: jajx*(ej)
i=1 j=1 l $i:5 m i=1

o
PROOF OF THEOREM 6.32: Let {e;} be an unconditional basis of X t hat is
not boundedly com plete . Then t here are scalars (ai) such that Il j~ aiejll ~
00
1 for every nand L aiei does not converge. By the Cauchy crite rion, there
j=1
ar e E: > 0 and natural numbers PI < ql < P2 < q2 ... such that for Uj =
.f: aj ej we have lI ujll ~ E: for every i , yet Ilf
J =Pi .=1
Ui// ~ Kilt a jejll ~ K ,
.= 1
where K = ub c{ e.}.
By the pr evious lemma , for every sequence {Aj }j=1 of scalars, we ha ve

< K(s~p l\l)lIf Ujll


J j =1

< J{2( sup IAjl) = K 211(Aj)lIoo .


j

On t he ot her hand , from t he un conditionality of {ej} we have for each

i E {l , . . . , m }: IIj~IAjUj l ~ kllA ju jll ~ KIAd ; that is, ~11(Aj)"oo <


Ilj~1 Aj Uj I · T hus, {Uj} is equivale nt to the canonical basis of co·
o
Corollary 6.34
Th e space £ 1[0 ,1] does not have an unconditional Schauder basis.
184 6. Schauder Bases

PROOF : By the classical Steinhaus theorem, LdO, 1] is weakly sequentially


complete (see, e.g., [Woj]). Let {eil be the standard unit-vector basis of
n
Co. Then the sequence Xn = L: e, is weakly Cauchy but not weakly con-
i= l
vergent , therefore Co is not isomorphic to a subspace of L1[0, 1]. If LdO, 1]
had an unconditional basis {Xi}, {Xi} would be boundedly complete by
Theorem 6.32 and thus L 1[0 , 1] would be isomorphic to a dual space by
Theorem 6.10. But this is not possible : in a separable dual space , every
closed convex bounded set is the closed convex hull of its extreme points,
while the unit ball of L 1[0, 1] has no extreme points (Exercise 3.79) .
o
Theorem 6.35 (James; see, e.g., [LiT2])
Let X be a separable Banach space. If X has an unconditional Schauder
basis that is not shrinking, then X contains an isomorphic copy of £1 .
In particular, a separable Banach space with an unconditional Schauder
basis contains an isomorphic copy of £1 if X* is nonseparable.
PROOF : Let {eil be the assumed basis; set K = bc{eil . Since {e.} is not

°
shrinking, there is f E Sx· such that sup {J(z) ; z E Ex nspan{eili~n} f+
0. This means that there is e > and a sequence nk < nk+l such that
sup{J(z) ; z E Ex nspan{e nk,e nk+1, . ..}} 2: e
for every k. Using the sliding hump technique, we can construct a normal-
ized block basic sequence Uj of {eil such that f( Uj) 2: e/2 for every j .
Let mEN and a1, · · ·, am be scalars. Assume L ai 2: L -ai .
i S m, a i ~ O iSm,ai<O
Then

> ; II I: aiUill2:; ' f( L aiUi)


iSm,ai~O i S m> a i ~ O

> 2~ ~ a' > ~ ~ a' + ~ ~ -a,'


K z:
iSm ,ai~O
'-K z:
i:5m ,a;~O
' K c:
i:5m,a i<O
m

> ~ I:lail.
i= l

If L: ai < L: -ai, we consider -ai to obtain the same in-


i:5m ,ai<0
I
i:5m>ai~O

equality. On the other hand, since {Uj} is normalized, we have Il i~ aiUi ~


i~ lail· Therefore, 4~II(ai)lIl1 ~ Ili~ aiu ill ~ lI(ai)lIl1; that is, {Uj} is
equivalent to the canonical basis of £1 .
To see the second statement, note that if {er; ei} is shrinking, then X* =
span{e.] and thus X* is separable.
o
6. Schauder Bases 185

Corollary 6.36 (James; see, e.g., [LiT2])


Let X be a Banach space with an unconditional basis. Then X is reflexive
if and only if X contains no isomorphic copy of £1 or co.
PROOF : It follows directly from Theorem 6.11 and Theorems 6.35 and 6.32.
o
Definition 6.37

°
The James space J consists of all sequences (O'i) of real numbers such that
lim(O'i) = and II(ai)II < 00 , where the norm II(ai)II is defined by
I

II(ai)11 = sup ((O'n , - O'n2)Z + (O'n 2 - O'n3)Z + ... + (O'nk _1 - O'nk)Z) 2 .


n,< ...<nk .

It is standard to verify that (J, II ·IDis a Banach space . Moreover, the


n
sequence {ed of the standard unit vectors en = (0, .. . ,0,1 ,0, . . .) is a
monotone Schauder basis of J .
Claim
{ed is a shrinking Schauder basis of J and J** = J $ span{(l, 1, 1, . . .)},-
in particular, J is not reflexive.
PROOF: Assume that the basis is not shrinking. As in the proof of Propo-
sit ion 6.35, we find x* E J* , e > 0, and a normalized block basic sequence
{ Uj} of {ed such that x* (Uj) > e for every j. Consider the vector U =
00
L ~Uj . By Holder's inequality applied to the norm of J and by considering
J
j=l
the division points of Uj 's, we find that IIuliz ~ 2 IIUj liZ ::; 2 L /2
< 00. L /2
On the other hand, x*(u) = L x* (Uj ) Ii is not finite, a contradiction.
By Proposition 6.9, J** is identified with (f3i) such that SUPk Ili~ f3i ei II <
00 via the map T(x**) = (x**(ei)) . From the definition ofthe norm II . II of

J and from the fact that SUPk Il it f3i ei II < 00, it follows that i~~ (f3;) = f3
exists . Thus (f3i - (3)i corresponds to an element of J since .lim (f3i - (3) = 0.
• -00
Therefore J** = J $span{(l , 1, .. .)} .
o
Since J** is separable, J does not contain an isomorphic copy of Co or
£1 . Indeed , if J contained an isomorphic copy of co, then J* would have a
quotient isomorphic to c~ , and J** would have a subspace isomorphic to
c~* = £00 ' A similar argument shows that J does not contain an isomorphic
copy of £1 .
By Corollary 6.36, J does not have an unconditional basis . How-
ever, J contains an isomorphic copy of £z; in particular, it contains an
unconditional basic sequence .
186 6. Schauder Bases

Bessaga and Pelczyriski proved (see, e.g., [LiT2]) that if a Banach space
X has an unconditional basis and Y is a nonreflexive closed subspace of X ,
then Y contains an isomorphic copy of Co or fl. Using this result and the
fact that J (like every separable Banach space) is isometric to a subspace
of C[O, 1], we see that C[O, 1] does not admit any unconditional basis.
It was a long-standing problem whether every Banach space contains ei-
ther a reflexive subspace or an isomorphic copy of Co or fl. This problem was
answered in the negative by Gowers. In fact, Gowers showed ([Gow2]) that
there is a separable Banach space X such that every infinite-dimensional
closed subspace of X has a nonseparable dual and yet X contains no
isomorphic copy of fl.

Theorem 6.38 (Bessaga, Pelczyriski; see, e.g., [LiT2J)


Let X be a Banach space with no subspace isomorphic to co. If for a se-
quence {x n } ~=l in X there exists M >0 such that t~l enX n II ~ M for
every mEN and every choice of signs en = ±1 , i = 1, ... , m, then L Xn
is unconditionally convergent.
00
The condition is often equivalently stated as L Ix*(xn)1 ~ M for every
n=l
x* E Sx* .
Indeed, given x* E Sx" set en = 1 if x*(xn) 2: 0 and en = -1 if
x*(x n) < O. Then , for mEN ,
m m k m

L Ix*(xn)1 = Lx*(enXn) = x*(LenXn) ~ IILenXnll ·


n=l n=l n=l n=l
On the other hand , assume the condition with x* . Given en = ±1, m E
N, we find x* E Sx' such that lL~l enXnl/ = x*C~l enXn) and then

m m

ilL enXnl1 s L Ix*(xn)1 s M.


n=l n=l

PROOF : We will show that L X n is convergent. Since we can repeat the


proof for every Lenxn, en = ±1, we get that LX n is unconditionally
convergent . Assume that L X n does not converge. Then, by the Cauchy
criterion, there is e > 0 and integers PI < ql < P2 < q2 ' " such that
qk
L Xn II > e for every k. Put Yk = Ls» 00
n for every k. Then L If(Yk)1
II n=Pk n=Pk
X
k=l
~
00

L If(xn)1 < Mllfll for every f E X* . In particular, f(Yk) -+ 0; that is,


k=l
Yk ~ O. Moreover , e ~ IIYkll ~ 2M for all k; hence there is a subsequence
(denoted again {Yk}) that is a basic sequence (Corollary 6.21). Consider
6. Schauder Bases 187

m
L akYk · Then
k=l
£II( ak )1100
bc{Yd < Ilf>kYkll
k=l
= sup{lfak!(Yk)l ; f
k=l
E Ex-}
m

< (m:x/akl) sup{:L: If(Yk)l ; f E Ex- } :S MII(ak)lIoo .


k=l
Hence, {Yk} is equivalent to the canonical basis of co , a contradiction.
o
Note that the canonical basis of Co satisfies the condition in Theorem 6.38
and is not unconditionally convergent .
Theorem 6.39 (Bessaga, Pelczyriski; see, e.g., [LiT2])
Let X be a Banach space. If X* has a subspace isomorphic to co , then
X has a complemented subspace isomorphic to 11 • In particular, X* has a
subspace isomorphic to f oo .
PROOF : Let T be an isomorphism from Co into X* . Then T* maps X** onto
f 1 (Exercise 2.39). Since Ex is w* -dense in Ex-- by Goldstine's theorem ,
using Corollary 2.25 we find K > 0 and X n E X for n E N such that
n-1
Ilxnll :S K, T*(xn)(e n) = 1 and L IT*(xn)(e;)1 < ~ , where e; are the
;=1
standard unit vectors in co. By Theorem 6.38 and Proposition 6.22, it
follows that {T*(x n)} has a subsequence {T*(x nk)} that is equivalent to the
canonical basis of f 1 and whose span is complemented in f 1 by a projection
P. Therefore, for some constant M > 0 and every choice of scalars {ak} k::1
satisfying L lakl < 00, we have

< KMIII=akT*(xnk)11
k=l
00

< KMIIT*IIII:L: akxnkll·


k=l
Hence T* is an isomorphism of Y = span{xnJ onto span{T*(x nk)} . Thus,
Y is isomorphic to f 1 and P = (T*)-lPT* is a projection of X onto Y.
Finally, P*(X*) is isomorphic to P(X)* = fi = foo (Exercise 5.10) .
o
Given a bounded sequence in a Banach space with a separable dual, we
can extract a weakly Cauchy subsequence. Rosenthal characterized spaces
sharing this property as spaces not containing an isomorphic copy of fl'
Precisely, he proved that if {x n } is a bounded sequence in a Banach space
X, then either {x n } has a weak Cauchy subsequence or contains a subse-
quence that is equivalent to the standard unit-vector basis of f 1 (see, e.g.,
188 6. Schauder Bases

[LiT2]). There are separable spaces whose dual is nonseparable and yet
they do not contain an isomorphic copy of £1 (e.g., the space JT discussed
in the exercises) .

Markushevich Bases
Definition 6.40
Let X be a Banach space. A biorthogona/ system {Xa;!a}aEr in X is called
a Markushevich basis of X if span{xa}aEr = X and {fa}aEr separates
the points of X .
A Markushevich basis {x a; la}aEr is called shrinking if span{fa} = X* .
Clearly, every Schauder basis of a Banach space X is a Markushevich ba-
sis of X. An example of a Markushevich basis that is not a a Schauder basis
is the sequence of trigonometric polynomials {ei21fnt; n = 0, ±1 , ±2, ...} in

°
the space (7[0,1] of complex continuous functions on [0,1] whose values at
and 1 are equal, with the sup-norm.
Theorem 6.41 (Markushevich; see, e.g., [LiT2])
Let X be a separable Banach space. If { zih c X satisfies span{ Zi h = X
and {gih C X* separates points of X , then there is a Markushevich basis
=
{Xi; I;} of X such that span{xi} span{z;} and span{fi} span{g;} . =
PROOF: Define Xl = Zl and h = 9kr!9k 1(Zl), where k 1 EN is such that
gk 1(Zl) :f= O. Then find the smallest integer h 2 such that gh 2 rt spanUd .
Define h = gh2 - gh2(X1)h· Find an index k 2 such that h(Zk 2 ) :f= 0,
and set X2 = (Zk 2 - h (Zk 2 )xI) / h (Zk 2 ) ' Let h3 be the smallest integer
such that Zh3 rt span{x1 ,x2}' Put X3 = Zh 3 - h( Zh 3)X1 - h(Zh 3)X2 and
h = (gk3 - gk3 ( X t } h - gk3 ( X 2 ) h ) / g k 3(X3) , where k 3 is an index such that
gk3 (X3) :f= 0. Continue by induction. At the step 2n , we construct hn
first ; at the step 2n + 1, we start by constructing X2n+ 1. It follows that
span {zd; C span {xd ~n and span {gd ~ c span {Jd ~n. Clearly, /;(Xj) =
8ij, spantXi} C span{z;} and span{/;} C span{g;}.
o
It is an open problem whether every separable Banach space X admits
a Markushevich basis {Xi; f;}~l with Ilxili = 11/;11 = 1 for all i. It is known
that, given a separable space X and £ > 1, a Markushevich basis of X
exists so that sup IIxillll/dl < 1+£ (Ovsepian-Pelcsyriski; see, e.g., [LiT2]).
Theorem 6.42 (Gurarii, Kadec [GuKa])
Let Z be a closed subspace of a separable Banach space X. Any M arku-
shevich basis {Xi; Ii} of Z can be extended to a Markushevich basis of
X.
Precisely, there are Zj EX, <Pj E X*, and extensions of Ii to 1fJi E X* such
that {{Xi} U {Zj}; {1fJi} U {<pj}} is a Markushevich basis of X .
6. Schauder Bases 189

PROOF : Extend all Ii onto X and denote these extensions by f;, . Let {Yj, ¢Jj}
be a Markushevich basis of X/Z (it is separable) . For all i, choose Yj E Yj
=
and define <Pj(x) ¢Jj(x) for x E X ; note that <Pj(x;) =
0 for all i. We have
span { {Xi} U {Yj}} = X and {f;,} U {<pj } is a family separating points of X .
j _ i _
Put Zj = Yj - L: Aij Xi and "pi = fi -L: Aij<Pj , where Aij = Ii (Yj) for
i= l j=l
t
i i= j and Aii = Ii (Yi) 0 Then {{ xd U {Zj}; Nil U{<Pj }} is a Markushevich
basis of X that extends {Xi, fd . Indeed , from the definition of Zj and "pi , it
is clear that span {Xi, Zj} = X, {"pi} U {<pj} is separating points of X , and
"pi extend Ii onto X . It is routine to check that the system is biorthogonal.
o

Theorem 6.43 (Johnson , [Johl])


The space £00 does not admit a Markushevich basis.

We will need the following lemma.

Lemma 6.44
Let X be a Banach space. Every reflexive subspace of X* is w*-closed in
X*.

PROOF : Let Y be a reflexive subspace of X* . By the Banach-Dieudonne


theorem, it suffices to show that By is w*-closed in X* . Since By is w-
compact in Y , it is w-compact as a subset of X* . Therefore, By is w*-
compact in X* and thus w*-closed in X* .
o

PROOF OF THEOREM 6.43 : Assume that {xa ; fa} aEr is a Markushevich


basis of £00 ' Put Y = span {fa} aEr o We claim that Y is reflexive. It is
enough to prove that By is weakly sequentially compact.
Let {Yn} be a sequence in By . Since every element of span{ja} has a
countable support over {x a } (as a limit in norm of a sequence of points in
span{ja} that have finite support) , there is a countable subset N of I' such
that for Q' E r \ Nand n E N we have Yn(x a) = O.
By the Cantor diagonal argument, let {Ynk} be a subsequence of {Yn}
such that lim (Ynk(X a)) exists for every Q' E N . Since Yn(x a) = 0 for every
k-oo
Q' E r\ Nand {Yn} is bounded, we have that lim (Ynk(X)) exists and is
k-oo
finite for every X E X Denote this limit by y( x). Then Y is a bounded
n:
0

linear functional on £00 and Ynk ~ Y in By the Grothendieck property


of £00 (see Exercises 6.30 and 6.31) , this means that Ynk ~ Y in £~ and
Y E By since By is w-closed in £~ . Therefore, By is w-compact in £~ and
thus also w-compact in Y . Consequently, Y is reflexive.
190 6. Schauder Bases

By Lemma 6.44, Y is w*-closed in l~ , and since Y is separating for


loo, it is w* -dense in l~ . Thus Y = l~ . Consequently, l~ and thus t; is
reflexive, a contradiction.
o
Theorem 6.45 (Plichko)
Let X be a Banach space. If X is separable, then X* is a complemented
subspace of a Banach space with Markushevich basis.
In particular, loo is a complemented subspace of a Banach space with
Markushevich basis.
PROOF: Let I be a set with card(I) = dens(X*) . Define U = (2:co)
I £1(1)
and Z = (X* EB U)1' Let {en;fn} be a Markushevich basis of X scaled so
that lIe nll $ 1 for every n. Choose a dense set {yi};EI in Bx-. For i E I
and n E N, let
n
y~ = yi _ Lyi(ej)/j.
j=1
For a fixed n E N , define hn =
{{yi(ennd i' We have yi(en ) = 0 for k ~ n
and lyi(en)1 $ 1 for k < n , and hence

hn E (Ll1) = U* .
I loo (r)

Let u~ be the standard unit vector in U and g~ be th e standard unit


vector in the space (2:l1) = U* . For n E Nand i E I, put x~ =
I loo(1 )
(y~, u~) E Z and e~ = (en, -h n ) E Z *. We claim that {{X~}nEN,iEI U
{Um, Onn,mEN ; {e~}mEN U {(O, g~nnEN,iEd is a Markushevich basis of
Z.
It is easy to see that this is a biorthogonal system and that th e functionals
separate points of Z (e~ = en on X*).
It remains to show that the span of vectors {x~ , Ii }n,jEN,iEI is dense in
Z. To see this, we note that for n EN and i E I we have
n n
u~ + yi = u~ + y~ + Lyi(ej)1i = x~ + Lyi(ej)1i E span{ x~,fj };:1'
j=1 j=1
Also, for every i E I , we have
k . .

II L
Uln k+yl - yl'11 = k
1 --+ 0 as k --+ 00 .
n=1
Therefore yi E span { x~ , Ii} . Furthermore, for every n E Nand i E I , we
n
have u~ = x~ - y~ = x~ - yi + 2: yi (ej)1i E span {x~ , Ii }.
j=1
6. Schauder Bases 191

Thus span{x~,xj}n ,J' EN , 1' EI = Z, and the proof is complete .


o

Exercises
6.1 Let {e-y} be a Hamel (i.e., algebraic) basis of an infinite-dimensional
Banach space X . Show that some of the coordinate functionals associated
with this basis are not continuous .
Hint : Pick an infinite sequence {enJ in {e-y} . Consider the vector x =
f
.=1
2-illen i" o Since {e-y} is a Hamel basis of X, we have x = LXjej ,
en i F
where F is a finite set . Let np be such that np rt F , i.e., x n p = O. For
m . en - .
every m, i~ 2-'llen:1I has the np-coordmate equal to 2- n p /llenpll· If the n p-
coordinate functional were continuous, we would have x n p = 2- n p /ll enpII =1=
0, a contradiction.
6.2 Why do we not use (iii) in Lemma 6.2 and the Banach-Steinhaus
theorem to conclude that Pn are uniformly bounded?
Hint : Do we know that they are bounded operators?
6.3 Show that the canonical projections of a Schauder basis of a normed
space X need not be uniformly bounded if X is not a Banach space .
Hint : Consider the trigonometric polynomials in the space of continuous
functions on [0,271"] .
6.4 Use the not ion of basic sequence to prove that a Hamel basis of an
infinite-dimensional Banach space has cardinality at least continuum.
Hint : Take any basic sequence {x n } in X. Let {NaJa E r be a collection of
infinite subsets of N such that NO' n N{3 is finite if a =1= (3 and card(r) ~ c
(Lemma 5.16). Define Yo' = L 2- i X i . Then {Yo'} is a linearly independent
iEN"
set of cardinality at least c.
6.5 Let {e.} be a Schauder basis of a Banach space X . Prove that there is
an equivalent norm on X in which {ei} is monotone.
Hint : Put IIIx~1 = sup., IlPn(x)lI· Then
IllPm(x)111 = sup II r;r; (x)1I = sup II Pn(x)1I :s sup IIPn(x)lI·
n n::;m n

6.6 Let {e;} be a Schauder basis. For n:S mEN , define Tn,m (f
.=1
aiei) =
m
L aiei · We say that {ei} is bimonotone if IITn,m II = 1 for all n , m.
i= n
192 6. Schauder Bases

(i) Show that IITn ,m II ~ 2 bc{ e.]:


(ii) Show that there is an equivalent norm 11/ . 11/ on X such that {ej} is a
bimonotone basis of (X, 111 ·111) .

Hint: (i) : Tn ,m = Pm - Pn- 1 .


(ii) : I I x II = sup IITn ,m(x)lI·
n$m

6.7 Let {e.} be a Schauder basis of a Banach space X . Show that {ei} is
monotone if and only ifIlPn+1(x)1I ~ IIPn(x)1I for every x E X and n E N.

Hint: If IIPnll = 1 for all n, use Pn(x) = PnPn+l(x) . If the condition holds,
prove first IIPk(x)1I ~ IlPn(x)11 for all k ~ n and then take the limit for
k -+ 00 .

6.8 Show that C[O, 1] has a Schauder basis consisting of polynomials.

Hint : Consider any Schauder basis {e;} of C[O, 1] and approximate {e;}
by polynomials using the Stone-Weierstrass theorem. Use the stability
theorem for Schauder bases .

6.9 Let X, Y be closed subspaces of a separable Banach space Z such that


X nY = {O}. Assume that the algebraic sum X +Y is not closed in Z . Show
that there exist a basic sequence {x;} C X and a basic sequence {y;} C Y
such that Ilxill = IIYjll = 1 and Ilxi - yill < 4- i for n E N. In particular, X
and Y have infinite-dimensional closed subspaces that are isomorphic.
Thus, if X and Yare totally incomparable spaces (for example, £p, £q for
P i= q), t hen X + Y is closed in every overspace.
Hint: If X + Y is not closed, then (Exercise 5.15) there exists z EX , Y E Y ,
=
with Ilxll lIyll =
1 and IIx - yll arbitrarily small. Then use the proof of
Proposition 6.14 and Proposition 6.18.

6.10 A Banach space X is said to have the approximation property if for


every compact set K in X and every e > 0 there is a bounded linear
finite-rank operator T from X into X such that IIx - T(x)1I < e for every
x E K . Show that every Banach space X with Schauder basis has the
approximation property.
Note that a different definition of this property is given in Chapter 7
after Proposition 7.4 . The proof that they are equivalent can be found, for
example, in [LiT3].

Hint: Let {Pn } be the canonical projections for some Schauder basis of X .
Then we have lim IlPn(x) - xii = 0 for x E X. The limit is uniform on
n-+oo
compact sets since SUPn IIPnl1 < 00. See the proof of Proposition 7.4.

6 .11 Let [e. ; en be a Schauder basis of a Banach space X . Show that if


{e.] is shrinking, t hen {ei} is a boundedly complete bas is of X· .
6. Schauder Bases 193

Hint: {ei} is a Schauder basis of X* . Consider ai such that sup II i~ aiei I <
00 . X is separable, so there is a sequence {nd C Nand x* E X* such that
nk •
X~k = Laiei ~ x* (Exercise 3.111). Since {en is a basis of X*, we
i=l
00
have x* = L f3iei · Fix j E N. Then lim(x~k(ej)) = x*(ej) = f3j and
i=l k
00
x~k(ej) = aj for nk ~ i . hence aj = f3j. Thus L aiei = x*.
i=l
6.12 Let {e. : en be a Schauder basis of a Banach space X . Show that
span{ ei} is a norming subspace of X* .

Hint : Let Illx!1=sup IlPn(x)ll · Then IIIPnll1 =


1 for every n, so IIIP~III 1. =
Given f E Sx·, we have III P; (J)III ~ 1, P;(J) E span{ei, . .. ,e~}, and
P;(J) ~ f. Hence span{en n B'x. w· = B'x. , where B'x. is the unit ball
in the dual norm to 111 ·111 . Thus, span{en is l-norrning for (X , 111 ·111) . Since
III . III is an equivalent norm on X, span{en is a norming set .
6.13 Let X = (L£~)2 ' Show that every infinite-dimensional closed
subspace of X contains an isomorphic copy of £2 .
Hint : Given a closed subspace Z of X , there is a subsequence in Z that is
equivalent to a block basic sequence of the canonical basis of X such that
the "nodes" are at th e ends of the canonical blocks. This block basis is
equivalent to the standard unit-vector basis of £2 .
6.14 Show that (L Lp[O , IJ)t p
is isometric to Lp[O , 1] .

Hint : Put An = [ni1'~] and identify f E Lp[O , 1] with the sequence UlAn}'
Use th e isometry of Lp[O , 1] and Lp(A n ) .
6.15 Show that (L£p). <-p
is isometric to £p and (LCo) Co
is isometric to Co .

Hint : Dir ect computation .


6.16 Show that (I: C[O, l]t is isomorphic to C[O, 1].
Hint : Use Pelczyriski's decomposition method . Represent (L C[O , 1]) Co as
a complemented subspace of C[O , 1] using an infinite number of nodes in
[0,1] and the closed subspace of C[O , 1] of functions that vanish at these
nodes .
6.17 Let p E (1,00). Show that Lp[O , 1] EB £2 is isomorphic to Lp[O , 1] .
Hint : Use that £2 is isomorphic to a complemented subspace of Lp[O , 1] .
6 .18 Use Akilov's result that Loo[O, 1] is complemented in every overspace
to show that £00 is isomorphic to Loo[O , 1] ([Pell]) . Because £1 is not iso-
morphic to L 1 [0, 1], this provides an example of two non-isomorphic spaces
194 6. Schauder Bases

whose duals are isomorphic. This also shows that no isomorphism of loo
and Loo[O , 1] can be w*-w*-continuous (Exercises 2.41 and 3.20) .

Hint : As the dual of a separable space, L oo is isomorphic to a subspace


of loo. loo is complemented in every overspace, so we use the Pelczyriski
decomposition method.

6.19 Show that l2 is isomorphic to a quotient of l oo . Compare this result


with Exercise 11.33.

Hint : The space l2 is isomorphic to a subspace of L 1 [0, 1] by Theorem 6.28.


Therefore, l ; ~ l2 is isomorphic to a quotient of Li = L oo , and then use
the previous exercise.
6.20 Assume that T is a bounded linear operator from a Banach space
X into X such that T(X) is not closed. Show that there is no finite-
codimensional subspace Y of X such that T would be an isomorphism
from Y into X.

Hint: Assume that T is an isomorphism from a finite-codimensional Y


into X. We have X = Y $ Z since Y is complemented. Then T(X) =
T(Y) + T(Z) . If T is an isomorph ism on Y , then T(Y ) is closed. Since Z
is finite-dimensional, we have that T(Y) + T(Z) is closed.

6.21 Let X , Y be Banach spaces. Show that every compact operator from
X into Y is strictly singular.
Give an example of a strictly singular operator that is not compact.
Hint: Let Z be a subspace of X such that T is an isomorphism of Z
onto T(Z) . Then T(Bz) is a closed subset of T(Bx ); hence it is com-
pact. Because Tlz is an isomorphism, Bz must be compact; hence Z is
finite-d imensional. .
For the second question , consider the formal ident ity map from l 1 into
l2.

6.22 (Kato) Let X be an infinite-dimensional Banach space, and let T be


a bounded linear operator from X into X such that the restriction to every
infinite-dimensional closed subspace of X is not compact . Show that there
is a finite-co dimensional subspace Z of X such that the restriction of T on
Z is an isomorphism.

Hint: Assume that the restriction of T to every finite-codimensional sub-


space is not an isomorphism . Then , for every fJ > 0 and {xn~1 C X*,
there is x E Sx with IIT(x)11 :s
fJ and xi(x) =0 for i = 1, .. . , m . By
the proof of Mazur's theorem, there is a basic sequence {x;} C X with
bc{xi} :s 2 so that IIT(xi)1I < 8- i . Let Z = span{x;} . Then T(Bz) is
contained in the set n:::aiT(xd; lad :s 2}, which is totally bounded due
:s
to IIT(xi)1I 8- i (see Exercise 1.48) .
6. Schauder Bases 195

6.23 Show that the sum of two strictly singular operators T, S from a
Banach space X into a Banach space Y is strictly singular.
Hint : By repeated use of the previous exercise, first for T and then for S,
get an infinite-dimensional closed subspace Z of X on which both T and
S are compact . Then T + S is compact on Z and thus not an isomorphism
on Z .
6.24 Let T be a strictly singular operator from a Banach space X into a
Banach space Y. Is T* necessarily strictly singular?
Hint : No. Consider a bounded linear operator T from £1 onto £2 (Theo-
rem 5.9). T is strictly singular because £2 is not isomorphic to any subspace
of £1 (Pitt 's theorem) . However, T* is an isomorphism into (Exercise 2.39)
and thus T* is not strictly singular .
6.25 Let r be uncountable and 1 S p < q < 00. Show that there is no
bounded linear one-to-one operator from £q(f) into ip(f). Similarly, there
is no bounded linear one-to-one operator from co(f) into ip(f), p E [1,00) .

Hint : Assume that such an operator T :iq(f) -+ ip(f) exists . Let e > a and
an uncountable set r 1 c I' be such that IIT(e,) II ~ e for l' E I' 1. An infinite
sequence in {e, },Er 1 tends weakly to zero, and we use Pitt 's theorem for
its closed span .
6.26 Show that every strictly singular operator T from i p into i p is
necessarily compact for p E [1,00) .
Hint : Examine the proof of Pitt 's theorem.
6.27 Does there exist a bounded linear operator from C[O, 1] onto i 2?
Note that there is no isomorphic copy of £2 complemented in C[O, 1] (this
follows using the Dunford-Pettis property; see Chapter 11).
Hint : Yes. £; is isomorphic to a subspace of £1 [0,1], which is isomorphic to
a subspace of C[O , 1]* by the Riesz representation theorem. By Lemma 6.44,
£; is w*-closed in C[O, 1]* .
6.28 Does there exist a bounded linear operator from £p onto i q, p =f. q E
[1 ,00) ?
Hint : No. Pitt's theorem , duality.
6.29 Does there exist a bounded linear operator from i p onto co?
Hint : If and only if p = 1. For p = 1, see Theorem 5.9. For p > 1, use the
reflexivity of £p .
6.30 A Banach space X is said to have the Grothendieck property if for all
{x~} C X*, x~ ~ a implies x~ ~ O.
Show that if X has the Grothendieck property and T is a bounded linear
operator onto Y, then Y has the Grothendieck property.
196 6. Schauder Bases

Hint: Direct proof, use the fact that T* is w* -w* -continuous and T** IS
onto (Exercise 2.39).
6.31 Does there exist a bounded linear operator from loo onto co? Does
there exist a bounded linear operator from loo onto ll?
Hint: No. It is known that loo has the Grothendieck property ([Dis2]) and
Co does not.
Assume there was a bounded linear operator from loo onto ll. By The-
orem 5.9, Co = q(ll)' so we would get a bounded linear operator from 100
onto Co.
6.32 Show that a block basic sequence of an unconditional basis is
unconditional.
6.33 Show that every normalized unconditional basis of a Hilbert space is
equivalent to the canonical basis of l2.
Hint : Let 2: Xj be unconditionally convergent. For C > 0, there is no
such that IIj~ eix, II < C for every n , m 2: no and e, = ±1. The gener-
alized parallelogram equality says l:i~Jn~1 CjXjl12 = 2m - n +1 n~l I/X jW,

2: I/xdl2 < c2. Thus, if {e;}


m
so is a normalized unconditional basis of a
i ==n
Hilbert space H , then 2:ajej converges only if 2: ar converges.
On th e other hand , if I: ar
< 00 , then from unconditionality and the
parallelogram equality we get that 2: ajej is unconditionally convergent.
Note that an analogous result holds for Co and ll . However, in these cases
the proof is more difficult (see, e.g., [LiT2]).
6.34 Let X be a Banach space. Let {x n } be a bounded sequence in X such
that {x n } has no weakly Cauchy subsequence. Prove that X contains an
isomorphic copy of II (which follows from Rosenthal's theorem mentioned
earlier in the text) assuming that X has an unconditional basis .
Hint: X* must be nonseparable; otherwise we use the Cantor diagonal
procedure to produce a weak Cauchy subsequence. If X* is nonseparable,
use Theorem 6.36.
6.35 A Banach space X is called weakly sequentially complete if every
weakly Cauchy sequence is weakly convergent in X . An example of a weakly
sequentially complete space is any reflexive space (show this) and any gen-
eral L 1 (Jl) space (this is a classical Steinhaus theorem) . Show that if X
is weakly sequentially complete and nonreflexive, then X must contain an
isomorphic copy of ll.
Hint: The first problem : use the Eberlein-Smulian th eorem . The second
problem: If X does not contain an isomorphic copy of l l, every bounded
6. Schauder Bases 197

sequence {x n } in X has a subsequence {x n k } that is weakly Cauchy by


Rosenthal's theorem. Since X is weakly sequentially complete, {x n k } is
weakly convergent in X . Therefore, X is reflexive by the Eberlein-Smulian
theorem .
6.36 Let X be a Banach space (not necessarily separable) . A family
{e-y hEr in X is called an unconditional Schauder basis of X if for ev-
ery x E X there is a unique family of real numbers {a-y }-YEr such that
x = I: a-y e-y , where the summation is meant in the sense that for every
e > 0 there is a finite set Fer such that IIx - -y~, a-ye-y II :s e for ev-
ery F' :J F Note that for every x E X only count ably many coordinates
0

a-y are nonzero . Indeed, given n E N, there is a finite set Fer such
that 11-yE., a-ye-y II :s ~ for every finite F' disjoint from F 0 Applying this to

F' = h} and assuming lI e-y11 = 1, we get Ir: la-yl ~ ~} C F .


Similarly as for the case of countable Schauder bases, it can be shown
that the coordinate functionals a~ are in fact in X* .
Show that the closed linear span of the coordinate functionals of an
unconditional basis is a norming subspace in X* .

Hint: Use P~ .

6.37 Let X be a Banach space. Show that if X* is separable, then X admits


a shrinking Markushevich basis.
Hint : The proof of Theorem 6.41.
6.38 Let X, Y be separable Banach spaces. Show that there is a bounded
linear operator mapping X onto a dense subset in Y .
Hint : Let {Xi;!;} and {Zi; gil be Markushevich bases of X and Y,
respectively, with {t;} and {zd bounded. Put T( x) = I: 2- i fi (x )Zi '
6.39 Let X be an infinite-dimensional separable Banach space . Show that
there is a biorthogonal system {Xi; fd such that span {xd = X and {td is
not separating.
Note that this cannot be done in finite-dimens ional spaces .

Hint: Let {Yi} be a linearly independent sequence such that span{Yi} = X.


Pick x E X \ span{yd (infinite-dimensional spaces cannot have a countable
Hamel basis ; Exercise 1.64). Since {x, Yl, .. o} is a linearly independent set,
=
using Hahn-Banach we find gi E Sx· such that gi(X) 0, gi(Yi) 1, and =
9i(Yj) = 0 for j =
1, . . 0' i-I. Then {gd separates points of span{yd
and as in Theorem 6.41 we find a biorthogonal system {xi;!d so that
span{x;} =
span{yd and span{t;} =
span{g;}. Then span{xd X and =
=
{td does not separate points . Indeed, gi(X) 0 for all i ; hence also f(x) 0 =
for all f E span {gi} '
198 6. Schauder Bases

6.40 Show that James's space J (Definition 6.37) contains an isomorphic


copy of £2 .
Hint : Let H denote the subspace of J consisting of vectors with even coor-
dinates equal to zero. Show that on H, James's norm is equivalent to the
£2 norm.
6.41 Let Ui = ei - ei+l, where {ei} is the shrinking basis of J . Show that
{Ui} is a boundedly complete basis of J, which is not shrinking, and

6.42 Let {Ui} be the boundedly complete basis of J from the previous
exercise. Let {vd be the biorthogonal functionals to {ud , which span the
predual space J. to J by Theorem 6.10. Define 9 E r by g(L eiUi) = Lei.
Show that s:
= span( {vd u {g}) = span(J. U {g}).
Hint : J. =span{vd . Clearly, 9 E r \ J. ; then use dim(r fJ.) = 1.
6.43 Assume that we replace the exponent 2 in the definition of James 's
space J by 1. Is the resulting space isomorphic to £1 ? Can Co be obtained
in a similar way?
Hint : Yes. Yes.
6.44 Is the following an equivalent norm on J ames's space J ?

Hint: Yes.
6.45 Define a norm I I . m on J ames's space J by

Illxll = sup ~((xnl - x n2)2 + ... + (x nm_ 1 - x nm)2 + (x nm - xnY) t.

Show that III . III is an equivalent norm on J and , in this norm, T" is
isomet ric to J .
Hint : Consider the map U: F" --+ J defined by
U(x··) = (-A ,x··(et) - A,x··(e2) - A, . . .),
where A = lim(x··(e;)) .
6.46 Find a Schauder basis {ed of a Banach space X with separable dual
such that [e.] is not shrinking.
Hint: Consider the biorthogonal functionals Ud to the standard unit-
vector basis {e.] in James's space J . Since the standard basis is shrinking,
6. Schauder Bases 199

{fd is a basis of J* . However , {fd cannot be shrinking because its


biort hogonal fun ctionals are {e.} in J and span {e.} = J i= J* *.
6.47 Does there exist a Ban ach space X not isomorphic to a Hilbert space
and such that X * is isom orphic to X?
Hint : Cons ider (f p EB f q )2' where p E (1,00) and'!' + .!. = 1. To obtain a
nonr eflexive examp le, consider J EB J* , where J is am~s's space. J
6.48 Does t here exist a Ban ach spa ce X such t hat X EB X is no t isomorphic
to X ?
Hint : Consider James's space. What is the codimension of J EB J in (J EBJ )**?

In the remaining exercises , we define and investigate the James tree space
JT. All notations and definitions made in one exercis e carry to the following
ones. For more information , see [FeGa] .
6.49 Let T = {(n , i); 0::; i ::; 2n - 1, n E No} be equipped with a partial
ord ering < determined by the relation (n, i) < (n + 1, j) iff j E {2i , 2i + I} .
Then (T , <) is a binary tree. By a segm ent we mean a sub set S = {t E
T ; (n, i) ::; t ::; (m,j)} for som e (n, i) ,(m ,j) E T . A maximal linearly
ord ered subset of T is called a branch. Let r denote t he set of all branches
in T . Show t hat card t l") = c, t he cardinality of t he cont inuum .
6.50 The J ames t ree space JT consists of all real functions defined on T
wit h the norm

IIxll = sup(L:(
k

j= l
L:
(n, i) ESj
x(n, i) ) r,
2 ~

where t he sup remum is taken over all finit e sets of pairwise disjo int seg-
ments in T . Verify t hat JT is a Banach space wit h a bo undedly complete
bas is { e (n ,i)} ord ered lexicograph ically; that is,
(0, 0) :5 (1,0 ) :5 (1,1 ) :5 . . . :5 (n , i) :5 (n , i + 1) :5 .. . :5 (n , 2n - 1)
:5 (n+l,O):5 . . .
and the fun ctions e (n ,i) are defined by e(n,i)( m , j) =
1 if (n, i) (m, j) and =
zero othe rwise (i.e. , e( n ,i) ( m , j) = 8(m ,j)(n ,i )) '
In particular , there exists a pr edual JT* to JT with a shrinking basis
{f(n ,i)} with e(n,i) (J( m,j) ) = 8(m ,j ),(n ,i )' Thus, JT* can be represented as a
00 2R _ l
space of real functions F on T sa tisfying F = w* - lim L L F(n , i )f(n.i)'
n=O i = O

Hint : By cont radict ion. Consider x E JT and e > 0 such t hat for ev-
ery n E N t here exists a finite set {Sj}j=l of pairwise disjoint segments

i)) )
k 2 ~
satisfying (L ( L x(n , 2 ~ e. Then there exists a fast-growing
j= l (n,i )ES j
200 6. Schauder Bases

sequence {nil of integers such that S~'l n S~12 0 for /1 =I- /2, Thus
I: ( I: x(n, i)) 2 = 00 , a contradiction.
I (n,i) Es t '

6.51 For every branch I E I', let Y-y = {x E JT; supp( x) C I}' Show that
Y-y is a subspace of JT isomorphic to James 's space J .
Show that the linear operator P-y: JT - JT defined by
00 2 R_1

r; (I: I: a(n,i )e(n ,i») = I: ate t


n=O i = O t E-y

is a norm-one projection of JT onto Y.


Hint : See Exercise 6.41.
6.52 For F E JT* , define S(F) E l 2(r) by S(F) = ( lim . F(n , i))
n-oo ,( n,, )E-y -y
for I E r . Show that S :JT* - l2(r) is a bound ed linear mapping.
Hint: IIFII = sup{F(x) ; x E EJT} . Using this , one can get IISII = 1. To
show that S is onto , given {Jdi= 1 C r and scalars a1 , . ' " an, find F E JT*
n
such that S( F ) = I: a n i and IIFII = vI: al·
i=1
6.53 Show that Ker(S) = JT*.
Hint: For m = 0,1 ,2 , .. ., define for j E {O , . . . , 2m - 1}
00 2 R_1
00

Pm,j (2: 2: t(n,i)e(n ,i») = 2: 2: t(n,i)e(n,i)


n=O i=O n=m (m,j):S;(n,i»
2 m _1
and Pm = E Pm ,j. Then Pm are proj ections on JT. First , show that, for
j =O
x* E Ker(S), lim ( ~ax IIP~ i(x *)II ) = 0. If it were not th e case, there
n-oo 0<1< 2 R - 1 '
°
would be a sequence -{ (~k, ik)h EN and c > such that II Ptnk,ik) (x*)1I > c.
One can show that there exists only a limited numb er of incomparable
elements in {(nk ' ik) hEN, and thu s there exists a branch I E r con-
taining infinitely many elements from {( nk , ik)} kEN. We may assume that
{(n k' ik)hEN C I and IIP(*n Ie, l' Ie )(x*) - p(*n k+l , l' k+l )(x*)11> s for all k.
By the previous exercise, we have P-y(x*) E span{f(n ,i); (n , i ) E I} and ,
for k large enough, II(P~k - P~k + JP-y( x *)1I < ~ c . Define for kEN maps
U*k -- P* . - P*
(n k,lk) .
(nk+l,lk+l) - (Pnk
* - P*
nk+t )P*
"t '

Then Uk is a dual projection to some Uk and IlUk(x*)1I > ~€. The


supports of the subspaces Uk (JT) are mutually disjoint and no branch I ' E

II t1U; (x*)lr
r
k
int ersects more than one ofthem. Thus
and for j large enough we obtain a contradiction.
= kt1 11U;(x *)1I 2 ,
6. Schauder Bases 201

Now assume that there is v* E Ker(S) \ JT* , and let dist(v* , JT*) >
(1- 8) II v*11 for some 8. To get a contradiction, we find x , yESJT such that
= =
v(x) > 1- 8, v(y) > 1- 8, and Pm1(x) 0, Pm2(y) y for some m2 much
larger than mI. For 8 small enough , we use our limit formula and obtain
by calculating the norm of x + y from definition that IIx + yll < 3.5. Details
may be found in [LiSt].
6.54 Show that JT is a separable space with nonseparable dual containing
no isomorphic copy of £1.
Hint : Using Dixmier 's projection, from the preceding exercise we obtain
that JT** ""' JT $ £2(r) . Thus card(JT**) = c. If there was a copy of £1
in JT, we would find a copy of £l(C) in JT* by Exercise 5.51 and thus
£00 (c) would be isomorphic to some quotient of JT** . This would imply
that card( JT**) > 2c , which is a contradiction.

You might also like