STUDTA TJNIV. MB A BEŞ-B O L Y A I ” , M A T H E M A TI C A , Volume XLIV , Number 3, September 1999
S O M E A P P R O X I M A T I O N ID E A L S
N I C O L A E T I T A
A b s tr a c t. We consider some approximation ideals of operators on oper
ator spaces. The method used is similar to that from [8], [10], [11], in the case of the classical Banach spaces, or [2], [7] for the case of Hilbert spaces.
1. In tr o d u ctio n
The theory of the approximation ideals is well known for the case of linear and bounded operators on Hilbert, or Banach, spaces [2], [6], [7], [8], [11].
Here we consider the special case of the completely bounded operators on operator spaces. For these notions it can be seen [1], [3], [5].
We begin by recalling some definitions.
An operator space
E
, in short O.S, is a Banach space, or a normed space before completion, given with an isometric embeddingJ
:E
—►L(H),
whereL(H)
is the space of all linear and bounded operatorsT : H FI, H
being a Hilbert space.We shall identify often
E
withJ(E)
and so we shall say that an O.S is a (closed) snbspace ofL(FF).
If
E
CL(H)
is an operator space thenMn
OE
can be identified with the space of alln
xn
matrices having entries inE,
that it will be denoted byMn (E) .
Clearly
Mn (E)
can be seen as an o.s. embedded inL(Hn),
whereHn = H
0 ... ©H
(number ofH
is n ) .Let us denote by || ||n the norm induced by
L(Ifn)
onMn (E)
, in the particular case n — 1 we get the norm of
E.
Taking the natural embeddingMn (E)
—»A/n + i (i?) we can consider
Mn {E)
included in Mn+i(E) ,
and || ||ri induced by |H|n+1 . Thus we may consider(E)
a normed space equipped with it’s naturaln n orn l ll'llco •
109
NICOLAE TI TA
We denote by
K [E]
the completion of\jMn
(E
). If we denote byn
A'o =
\jMn,
the caseE =
C, then the completion ofK
o coincides isometri-n
cally with the C*-algebra,
C(l2),
of all compact operators on the spacel2.
It is easy to check that (jA /n
(E)
can be identified isometrically withK
qÇ)E.
n
The basic idea of o.s. is that the norm of the Banach space
E
is replaced by a sequence of norms {|| ||ril} n on{ Mn ( E) } n
or by a single norm H-H^ on the space A'[E],
D efin ition 1. Let
E\
CL( H
i) andEn
CL(H
2) be operator spaces,u
:E
i—> E2
be a linear map andUn
: (xij) G Mn(E\) —y
( u( xj j) ) GMn (E2) .
We say that, u is com pletely b o u n d e d , c.b ., if sup ||w„|| < oo and we define\\u\\c b
sup||nn||.n n
D efin ition 2. (equ ivalen t)
u
is co m p le te ly b o u n d e d if the mapsun
can be extended to a single bounded map
Ur<>
:K
[AJK [E2]
and we have ||t/.||cb
= . D efin ition 3.c.b
.(E\, E2)
{?/ :E\
—)•E2
:u
is c.b.} . We shall consider thec.b.(E\<E2)
equipped with IHL.6. •Remark L
The similar definition of the uniform norm for the bounded operators can be written as follows:IMIc.6. = sup {||«oo|| ■ x e K [A ], 11*11 <
1
} .Remark 2.
Likewise the case of an isomorphism between two Banach spaces, we say that two o.s.E
i ,E2
are c o m p le te ly isom orph ic, co m p le te ly isom etric, if there is an c.b. isomorphismu
:E\ —y E2
with c.b. inverse and in addition |M|r.& —||u- 1 |«(El)l|c.fr. = 1
Let.
Ei
CL(FI
i) andE2
C be operator spaces. There is an embedding J :Ei
OE2 —y L(H
i 0Hn)
defined byJ
(x\0 x 2)(h\
0h2)
= xi(h.\)
0 x 2(h2) .
We denote by
E\
0 minE2
the completion ofE\
0E2
equipped with the norm1 1 0
SOME A P P R O X IM A TIO N IDEALS
O b viou sly
J
can be extend to an isom etric em bedding. So we can seeE\
® mjn En as an o.s. em bedded into L(H\ ® a Hn). T his space is called the m i n i m a l, s p a t ia l, t e n s o r p r o d u c t o f E\and En. (Hi ®a Hn is the hilbertian tensor p rod u ct, [7], [] 1].)If E C L(H ) is an o.s. then Mn 0 min E can be identified with the space Mn (E) and K [E] can be identified isom etrically with K ©min F. Thus, for any linear m ap u : E\ —►En we have ||t/.||c6 = ||/® u : K ® min E\ —ï K © mjn En\\ = Il/ O V ' K Omin E\ -> K ® min En\\c b . M ore generaly it can be shown that, for any o.s. F C L (H), we have ||7> ® u : F ® mjn E\ -> F ®min En\\ < ||u||c 6 . Further on, if v : F\ Fn is another c.b . m ap, we obtain
Il
U
0v
:Fl
® minEl
- >F2
©minEn\\cb
< IMIc.6. * ll^llc.6. • T h is relation will be very useful in the sequel.For others properties o f the m inim al tensor product it can be seen the papers [1]. [!)], etc.
2. A p p r o x i m a t i o n n u m b e r s o f c o m p l e t e l y b o u n d e d o p e r a t o r s
D e f i n i t i o n 4 . Let u : E -» F be a com pletely bounded m ap, u E c.b. (E , F ). T he a p p r o x i m a t i o n n u m b e r s , acnb( « ) will be defined as follow s
acnb (u) := in f {\\u— a.||c b : a E c.b. (E , F ), rank (a) < n) , n — 1 ,2 ,....
Remark 3. From this definition it results that ||w||c 6 = b (u) > a ţ b (u ) > ... > 0.
P r o p o s i t i o n 1. The approximation numbers a £ b (u) verify the following inequali
ties:
k k
1. acnb- (ui + u2) < 2 • ( rt" fc ( U l) + a" b (U2)) ,fo r k = l ’ 2 ’ •••
n = l n = l
k k
2.
^ 2 anb
(wi 0 ^2) < 2 •^ 2
(ani> (wi) * anb (ul)) i for fc = F 2, ...77 = 1 77 = 1
NICO LAE T IT A
Proof. 1) Let. e > 0. T here are a ,, i = 1 ,2 , such that rank (a*) < n and ll«i - rt>llc 6. < a n b ( « « • ) + I -
W e ob ta in :
n2b n -i
(«1 + «2) < ||(«1 + « 2) -(ai
+ a2)||c.6. << ll« l - “ lllc.b. + llu 2 - 0 2 ||c.6. <
< < b ( « 1 ) + « n b ( U 2 ) + S.
Since e is arbitrary it follow s that:
a - ï n- 1( « 1 + «2) < < 6 ( «1) + ( «2) •
Further on it results:
E < 6 (Ml + «2) < E «2 n - l (Ml + «2) + E «2-n (« 1 + U2) <
ri = 1 n = 1 n = 1
< 2 • E « 2 ' n - l ( M l + M2 ) < 2 • E ( " n b ( « l ) + M *'6 ( l / 2 ) ) .
n = l n ~ 1
2) W e consider also a ,, ?’ = 1 ,2 , such that, ran/c (a,-) < n and
||«i - a i||c ,fc. < a n 6 ' ( M . ) + §•
W e obtain :
« n b ( « 1 ° M2 ) < 11 ( w 1 O m2 ) - [ « 1 O
a2
+ a y o ( u 2 - a 2 )]||c 6 == | | ( « i ~ M l ) O ( w 2 - a 2 )||c b < { a cn b ( « ! ) + § ) • ( a ^ b ( t * 2 ) + § ) .
Since e is arbitrary it follow s that:
«2 n - l (Ml o Wo) < acnh (w i) • acnb- (w2) •
Likewise the 1) results 2). □
Remark f For the case o f the linear and bounded operators between Banach spaces the above inequalities are known, [8], [11].
In the sequel we deduce an inequality for the case o f the c.b . operator u\ Q mm Î/-2 using a sim ilar m eth od with that from [9], [10], used for the classical case o f the bou nded operators on B anach spaces.
P r o p o s i t i o n 2 . The approximation numbers acnb (iii ® mm w.o) verify the inequalities:
L
an
(ul
0minUrl)
< 6
k ncb
£ —
n = l
(
m, H M I
c.b.+ «n 6' (M2)I k I L
for fc = 1 , 2 , . . .1 1 2
SOME AP PR O XI MA TI ON IDEALS
Proof.
Let€
> 0. There are a,-,i =
1, 2, such thatrank
(a,) <n
andI K - a.ilc.6. < anb' (««O + §•
We obtain:
('M l © m i n d o) ^ ||m i © m i n w 2 ~ a l © m i n ^*21|c .6. =
= | | («1 - « l ) © m i n U2 ~ d\ © m i n ( « 2 - d 2 )\\c 6 <
< ll «l - « i l k • I k l k . + H«1 Hr.6. • ||«2 — «2IL.6. <
< («0 + §) • I k l k . + ||«1 ~ «1 + «iile
6.• («n6 («s) + §) <
< ( « n 6 (m i) 4- § ) • I M L f t . + ( | k - m i||cb + | k l k . ) • ( < 6 (l<2) + I ) <
< ( < b ( «1) + § ) • I k l k + 2 • I k I k • ( < b (t»a) + 1 ) .
Since
e
is arbitrary we obtain:aCJ (M ) ® m i n M2 ) < 2 • (acn b ( «1) • | k l k + acn b (u 2) • ||m i||cb ) .
T a k in g account th at the sequence o f the aproxim ation num bers is decreasing we can w ritte:
^2a n
b'{"iv>m\n'U
(** 1 <8> m in'** 2 ) 2)
< ÿ2 (2 •n
+ 1)n = 1 11
N ow we o b ta in : k
3
n = 1E 3 E n — I
a% (wl<8>ininW2)
, where j 2 < k < ( j + l ) 2
^
Y
2 (2 • n -J- 1 ) fl»2‘^tl®minU2^ < 3 .Y l n '
°n2 ^tl®minti2^ <n- 1 n = 1 n = l
0 ^2 a n h(U l) llU3||e.b + a n b (t<2)-||»l||c.fc: < 6 ‘ ^ fln * ' (Ui H IM Ic . b . + fln * ' (“ - H I ” dlc.b.
n = l H n = l
T h is finishes the proof. □
Remark 5. B y m ea n s o f these ap p roxim ation num bers we can define special approxi
m ation ideals in e.b. (E, F ) .
3. S p e c i a l a p p r o x i m a t i o n i d e a l s
D e f i n i t i o n 5 . Let x = {a?i, a ? o , b e a real sequence and let card (# ) be card {i ÇL N : Xi 0} .
Let K be the set o f all real sequences x G /no having the follow in g two properties:
1.
card
(j?) <n
(x),n
(a;) is a natural numberN ÎCOLAE T IŢ A
2 . X \ > X 2 > . . . > æ n (ar) > 0 .
A function $ : K —> R is called a sy m m etric n orm in g fu n ctio n if:
1. $ (.r) > 0 if x G K and x 0;
2. $ ( o r • x ) = a • $ ( x ) , for every o > 0 and x G A";
3. O (.r 4- y) < $ (x ) + $ (y) , for every x, y G A ';
4 . $ ( { 1 , 0 , 0 . . . } ) = 1 ;
k k
5. If x ,y G A' and ]T]x* < f ° r everY ^ = 1 , 2 , then $ (:r) < $ (y) .
j = l i = l
Remark 6. T h e above definition can be extend on the whole space taking
$ ( . r ) := lim $ ({x ^ , ..., x* , 0, 0 ...}) , where x* — { x * } if;/v is the sequence {k ?'| }?€;v rearranged in decreasing order.
D e f i n i t i o n 6. In the sequel we shall consider a subclass o f c.b. (E, F) which is defined as follow s:
4> - c.b. (E, F) := { « € c.b. (E, F) : |M|c* fc := $ ( R " ( « ) } „ ) < .
Remark 7. W e prove that this class has similar properties with the sim ilar classes defined for linear and b ou nded operators. (For the case o f the Hilbert spaces it can be seen [2], [7] and for the case o f the Banach spaces it can be seen [8], [9], [10], [11].)
P r o p o s i t i o n 3. — c.b., ||.||^6 ^ is a quasi-normed operator ideal.
Proof. 1. Any unidim ensional operator , a G c.b. (F , F) , belongs to
$ —c. 6. (F , F) because, in this case, the sequence { a £ 6- ( « ) } = {||w||c6 , 0 , 0 , . . . } and hence $ ( R 6 ( « ) } „ ) = I M R <
2. If u1,^2 G c .b .(E ,F ) then ui + un G c .b .(E ,F ). This results from the p rop osition 5 (1). and from the properties o f 3>, as follow s:
* ( R * R + u 2 ) } J < 2 • $ ( R b R ) + < 0 R ) } „ ) <
< 2 - ( * ( R 6 ( « , ) } „ ) + * ( { < * • R ) } J ) .
3. If v G c.b. (E, E) , u G $ — c.b. (E, F) and w G c.b. (F, F ) then 114
iv
o « o ţ) G $ -c.b. (E
,F )
.From the proposition 5 (2) and from the definition of a£6
(u)
it follows that0C nb' (<<’ ° « ° »') < ll« ’tle 6 • anb ( « ) ’ I M U . a n d h e n c e
$ ( { < 6' ( l | ,0 « 0 t ’) } „ ) < ll«ilc.6. ( « ) } „ ) ll'ilc.fc - D
Remark
8. We present now some properties similar to the properties of the classical approximation ideals L<j>, [8], [11].Lem m a 1.
The approximation numbers acnh (u) verify the inequalities:
SOME A P P R O X IM A TIO N IDEALS
k Y A ' b n -l
n — 1k k
(«) < («) < 2 • E « 5 - n - l (W) =
l ' 2’ -■
n = 1 n = 1
Proof.
The first, inequality is a consequence of the fact that the sequence{a^b
(u ) } n is decreasing.The second inequality results as follows:
E ° n b ( « ) < E anb ( « ) < E « 2 n - 1 ( « ) + E « 2 n ( « ) <
n ~ 1 n — 1 n = 1 n = 1
5: 2 * E a 2 n - l ( îf ) •
n = l
□
C orolla ry 1. ||u||^6' ( { é t - i (w )}n) a
quasi-norm equivalent
with i m i; 6- = $ ( { < 6 ( w ) } J -
k k
/Remark 9.
Since ^ (a£6- (u ))P < E ( » E < ' 6' («) i=1P
<
<
r ( p )
•E
( < 6’ (W))P . 1 < P < = 1,2,..., n = lsee the Hardy inequality [11], it follows that
* ( { K ‘ < » » ' } „ ) < * ( { ( ; î > ‘ w ) } J < ' ( » ) • * ( { « * < « » ' } „ )
and
hence
||«||$bp) is equivalent, with |M|*bp) := $ (P) ^ j j j • E « f 6' ( « ) } ) -N ICOLAE T IŢA
where <ï>(r) ( { * ; } ) = ^ } ) p is a sym m etric norm ing fu n ction , [2], [6], [7], [11], for 1 < p < oo.
Because acnb (u ) < ( a f 6 (u) • ... •acnb (u )) n < £ • J2ai'b' (n) l{ follow s> also, n
2 = 1
that
i a n b ( « ) } „ € l * (r) if 9 n («) := { ( « î fc («) • ••• • < 6 («)) " } n €
{•^n} £ ^(p) ^ ^ ^(p) ^
Remark 10. If we consider the special case o f the function
K 6 N ) p p
^ ( p ) : ( K b ( « ) } ) - > *
n
where 1 < p < oo, the classes <£(p) — c.6. (F , F ) are tensor product stable.
P r o p o s itio n 4. If G $ ( p) — c.6. (F * , F^) , fc = 1,2., then
U l
®min ^2 £ ^(p) C.6. (F^ ®min F2, Fţ ®min ^2) •
P roof It is sim ilar to that for the classical approxim ation ideals [10], [11].
First we remark that, using the relation
for A* = 1 , 2 ,..., see P rop osition 6 for p = 1.
Now taking into account the properties o f the functions it follow s that we can obtain
c (/>) • $
« * • («0 • ll«allc.fr.)P | («;
n
,c.b.
n
( « 2 ) ' I M I C b)
n
r
116
SOME A PPR O X IM A TIO N IDEALS
Hence
$(r) ( { ° n 6' (U1 ®min «2) } ) <
< O
(P) '(*(p) ( R * («!)}) • I H U + *(P) ( R 6 («2)}) • IKIU.) <
< 2 - 0 (p) . ? (p) ({o*-4- («!)}) R P) ( R fc M } ) ,
C i ( p ) = c ( p ) r and |KI|c 6 < $(P) ( R 6' («*)}).* = 1,2.
The proof is fulfiled.
□
Rem ark ÎL
The above result remains true if we consider the m axim al ten sor p r o d uct.R eferen ces
[1] D.P. Blecher, V .I. Paulsen, Tensor products of operator spaces, J. Funct. Anal. 99 (1991) 262-292.
[2] 1. Gohberg, M . Krein, Introduction on the theory of the linear, non-adjoint operators on Hilbert, spaces, Nauka, Moscow, 1965. (in Russian)
[3] V . I. Paulsen, Completely bounded maps and dilations, Pitman Research Notes in Mathematics, Longmans, London, 1986.
[4] V . I . Paulsen, Representation of function algebras, abstract operator spaces and Banach space geometry, J. Funct. Anal. 109 (1992) 113-129.
[5] G . Pisier, Operator spaces and similarity problems, Documenta Math. (I.C .M . 1998) 1-429-452.
[6] N. Salinas, Symmetric norm ideals and relative conjugate ideals, Trans. A .M .S . 188 (1974) 213-240.
[7] R. Schatten, Norm ideals of completely continuous operators, Berlin-Gottingen- Heidelberg, 1960.
[8] N. T iţa, Operators of <rp—class on non-Hilbert spaces, Studii Cercet. M at. 23 (1971) 4 6 7 -4 8 7 .(in Romanian)
[9] N. Ti(,a, On a class of operators, Collect. M ath. 32 (1981) 257-279.
[10] N. T ita, Some inequalities for the approximation numbers of tensor product operators, Anal. Şt. Univ. ” A1. I. Cuza” Iaşi 40 (1994) 329-331.
[11] N. Ti^a, S-numbers operator ideals, Ed. Univ. Transilvania, Braşov, 1998.
De p a r t m e n to f Ma t h., Fa c. of Sc i e n c e, ” Transilvani a” Un i v., 2200 Br a ş o v, Ro m a n ia
117