BULLETIN OF THE AMERICAN MATHEMATICAL SOCIETY Volume 81, Number 6, November 1975 APPLICATIONS OF BANACH IDEALS OF OPERATORS BY J. R. RETHERFORD1'2 For Stacy and Dana 1. Apologies. Since much of the recent work in the Banach space aspects of Functional Analysis, especially the geometry of Banach spaces, could, by a bit of chicanery, be construed as applications of Banach ideals, the title does not indicate a complete survey. This work is surely not exhaustive of the subject matter. Thus, many good papers are totally ignored. This is somewhat compensated for by the monograph [1.1] of Lindenstrauss and Tzafriri on the geometry of the classical Banach spaces and the "pre-book" [1.2] of A. Pietsch on the general theory of ideals of operators. Since a lecture should have a central theme, I have chosen a fundamental result of Grothendieck which asserts that there are Banach spaces E and F for which every bounded linear operator from E to F is 2-absolutely summing. (Definitions will be forthcoming.) This result and the local struc ture of Banach spaces are the unifying topics of this paper. For the numerous topics this unification omits, again, apologies. I have addressed myself to the material at hand twice before [1.3], [1.4], [1.5], the latter in collaboration with Y. Gordon and D. R. Lewis. I apologize for mentioning, again, the beautiful result of Stegall and Lewis [1.6] and tramping once again over ground covered in [1.3]-[1.5]. However, I feel, perhaps with prejudice, that these results are worthy of further discussion. An additional apology of sorts is needed. I have included many definitions which are old hat to experts in Banach space theory. It is hoped that the material will thus be accessible to a larger audience, perhaps even to some persons completely outside Functional Analysis. Finally, many of the results stated have their natural setting in Probability Theory. I have avoided the probabilistic language entirely. Thus "random variable" becomes "measurable function" etc. This is an editorial judgment on my part, and apologies are extended to anyone this may offend. This is an expanded version of an invited address given at the meeting of the Society in Nashville, Tennessee, November 8, 1974. AMS (MOS) subject classifications (1970). Primary 46-02, 47-02, 46C05, 46B10, 46B15, 46E05, 46E15, 46E30, 47B10. Key words and phrases. Banach ideals of operators, type, cotype, stability, LUST, p- absolutely summing, nuclear, integral, S-, £)- and ^-spaces. p p p 1 This paper was prepared under grant GP-34193 of the National Science Foundation (U.S.A.). 2 Portions of this paper were written during a visit to the Institut fur Angewandte Mathematik der Universitat Bonn. The author gratefully acknowledges partial support from the Sonderforschungsbereich 72. Copyright © American Mathematical Society 1975. 978 APPLICATIONS OF BANACH IDEALS OF OPERATORS 979 A remark on the Bibliography is in order. Instead of listing bibliographi cal references alphabetically we have listed the results as they appear in the text, e.g. [5.3] means reference [3] of §5. Thus, one interested in the proof of results stated in a given section can go to the source immediately. Due to the length of the paper almost all proofs are omitted; hopefully, this unorthodox bibliography will help compensate for that. 2. Notation. Most of our notation is standard. All spaces considered are Banach spaces. The word operator means bounded linear transformation. By an isomorphism we mean a one-to-one, open operator. The Banach-Mazur distance, d(E,F), between Banach spaces E and F, is given by d{E,F) = inf||T||||T-1|| where the infimum is taken over all isomorphisms between E and F If E and F are not isomorphic set d(E, F)=+°°. A projection P is an operator from £ to E with P2=P. If A is a subspace (=closed linear manifold) then A is complemented in E if there is a projection P with P(E)=A. We will denote the identity operator on a Banach space E by id . E A sequence (x ) in a Banach space E is a Schauder basis for E (uncon n ditional Schauder basis for E) if for each xeE there is a unique sequence of scalars (a„) such that £n=i a x converges to x in norm (such that £n=i e„a„x„ n n converges for all choices of signs e = ±l). The functionals f defined by n h fi(x)=ai, are called the coefficient functionals of the basis (x ). 4 For l^p^oo we denote by Ip the Banach space of scalar sequences (a») with l|a||=(ï|a|p)1"' if lëp<oo, i = sup|ai| if p = o°. i Similarly, for l (F) where T is any discrete set. In particular, we denote by lp p the space of n-tuples with the above norm. We denote by c the closed 0 subspace of L consisting of those sequences which tend to 0. Given p in [l,o°) we will always denote by p' the number satisfying l/p+l/p'=l. If E' denotes the conjugate of a Banach space E then (l )'=I '. p P We will use tensor notation in §§11, 12, and 14. By J ®i<, we mean the p closure of the finite rank operators T : l >-> l in the norm p q inf{£r=i ||/i|| ||xi|| : Tx = £"=i /t(x)Xi}. By I ® l we mean the closure of the same p q finite rank operators in the usual operator norm. A similar statement holds for ÇÔi; and JJjôi;. If (S, X, jut) is a measure space then as usual Lp(S, 2, /x) or Lp(jLi) denotes the Banach space of equivalence classes of almost everywhere equal func tions under the norm \\fl = (\\m\"n(ds)yP for lëp<°o, s = ess sup |/(s)| for p = «. 980 J. R. RETHERFORD [November By a probability measure space (il, JLL) we mean a positive measure with ix(il) = 1. For jut Lebesgue measure on [0, 1] we will suppress jLt(or dt) and write L [0, 1]. p By C(K) we mean the Banach space of continuous scalar valued functions under the sup norm. If (E ) is a sequence of Banach spaces then n (© E )l = {(xn), x eE \\\(x )\\p = £ IW|P < +*>}• n p n n n For a Banach space E let S (0 = inf{l-e||x+y||:||x|| = ||y||=l,||x-y||^t>0}. E The function 8 is called the modulus of convexity of E. If 8 (t)>0 for E E 0<8^2 then E is said to be uniformly convex. The spaces Lp(jLt), l<p<&, are uniformly convex. Let A^l and l^p^oo. A Banach space E is an S£ -space if for each finite PtX dimensional subspace F^E there is a finite dimensional subspace B with FciBcE such that d(B, ln )=\ where n=dim B. A space E is an i£ -space if P p it is an ££ , -space for some A^l. These spaces, introduced in [2.1], px generalize and include the L (S,X, JU) and C(X)-spaces above. P An operator T from E to F is compact if the image of the unit ball of E is relatively compact in F. A space E has the approximation property if every compact operator from F to E is the limit of finite rank operators. The space E has the bounded approximation property if there is a constant OO such that if B is a finite dimensional subspace of E there is an operator on E with finite dimensional range such that ||T||^C and T restricted to B is the identity. If C can be taken to be 1 then E is said to have the metric approximation property. A remarkable result of Enflo [2.2] (see also [2.3]) asserts that not every Banach space has the approximation property. A sequence of subspaces (E„) in a Banach space E is uniformly com plemented if there is an M>0 and a sequence of projections (P ) with n P (E) = E and ||P ||^M for all positive integers n. n n n Following [2.4] we will say that a Banach space E is an £f -space if for pA each n E contains subspaces E„ which are uniformly complemented and d(E , ip)=A; and E is a 2> , -space if for each n E contains E with n px n d(En, lp)^\. Finally, E is an Sf - or a 2> -space if it is a 3),\- or 9^-space p p p for some A^l. For the relationships between ifp1-, Sfp- and 2> -spaces p we refer the reader to [2.4]. 3. History. An ideal in the ring 5£(H) of all bounded linear operators on a separable infinite dimensional Hilbert space is a subset A(H) with the properties: if S, Si, S eA(H) and R, TeS£(H) then Si + S eA(H) and 2 2 RSTeA(H). The oldest ideal known to the author is the ideal cr (H) of "Hilbert- 2 Schmidt" operators. This ideal originated with the work of D. Hilbert [3.1] and E. Schmidt [3.2]. In considering the question "What operators on Hilbert space have a 1975] APPLICATIONS OF BANACH IDEALS OF OPERATORS 981 trace?", F. J. Murray and J. von Neumann [3.3] found the ideal cri(H) of "trace class" operators. Later J. von Neumann and R. Schatten [3.4] generalized the Hilbert-Schmidt and trace class operators to the ideals o-p(H) (0<p«x>). We will have more to say about these important ideals later. The final work in the ideal theory in i£(H), in the sense that a "complete" characterization of all two-sided ideals was given, was done by J. W. Calkin [3.5]. In particular, Calkin showed that there is a one-to-one correspon dence between the ideals A(H) and the permutation invariant ideals in the ring L of bounded sequences. (Further results along these lines were obtained by Schatten [3.6] and Gohberg and Krein [3.7]. More recently, the situation on nonseparable Hilbert spaces has been considered by Ooster brink [3.8] and his colleagues.) On the other hand, as we will see, to obtain an ideal theory on Banach spaces that is suitable for applications, it is not sufficient to consider only the ring 5£(E) of bounded linear operators on a Banach space E. We must consider the space «SP(E, F) of bounded linear operators between arbitrary Banach spaces E and F. Roughly speaking, a subset si of the class £ of all bounded linear operators between all Banach spaces is an ideal if whenever S, Si, S esi 2 and R,Te<£, then Si+S e^ and RSTesi (whenever Si+S and RST are 2 2 defined). We will give precise definitions later. The ideal of compact operators was introduced by F. Riesz [3.9] in 1918. This is the first example of a "Banach" ideal known to the author. (Recall that Banach spaces were not introduced until circa 1932!) Still other examples are the ideals of weakly compact operators (S. Kakutani [3.10]), nuclear operators (A. F. Ruston [3.11], A. Grothendieck [3.12]), and the strictly singular operators (T. Kato [3.13]). Other special classes of operators were considered by numerous authors. However, the general theory of Banach ideals of operators began, I believe, with the fundamental work on tensor products of Schatten [3.14] and Grothendieck [3.12]. However, the nontrivial translation from the language of tensor products to operators on Banach spaces was accomplished in a series of papers by A. Pietsch [3.15]-[3.21] and, in particular, [3.22] which influenced much of the subsequent work in the area. 4. Diversity of applications. The applications of the theory of Banach ideals have been numerous but mainly in three directions: I—Classifying types of locally convex spaces (e.g. Schwartz spaces, nuclear spaces); papers related to applications of type I include [4.1]-[4.4] and the numerous references given in [4.3] and [4.4]. II—Measure theory on Banach spaces (linear stochastic processes); here the work is mainly by L. Schwartz and the French school. Principal works are [4.5]-[4.7]. See also [4.8] and the Séminaire Maurey-Schwartz 1972-1973, 1973-1974. We will have some thing to say about the important Schwartz duality theorem later. Ill—The structure theory of Banach spaces; applications of type III will be our 982 J. R. RETHERFORD [November concern in this paper. Appropriate references will be given in the subse quent sections. 5. Banach ideals. Throughout the remainder of the paper SB denotes the class of all bounded linear operators between arbitrary Banach spaces and i£(E, F) the set of all such operators between specific Banach spaces E and F. We now define an ideal in the sense of A. Pietsch [5.1]. We say that a class A of bounded linear operators is an ideal if for each set A(E, F)= AH^(£,F) one has (a) if x'e£', yeF then x'®yeA(E,F) (x'®y denotes the rank one operator given by x'®y(x)=(x, x')y. Clearly, every rank one operator has this form.); (b) A(E,F) is a linear subset of S£(E,F) for each E and F; and (c) if Ue^(X,E), TeA(E,F\ Ve^(F, Y), then VTUeA(X, Y). The finite rank operators ^obviously form the smallest ideal. A function a on the operators T in an ideal A to the nonnegative real numbers is an ideal norm if (d) x'eE', yeF then a(x'®yH|x'||||y||; (e) S, TeA(E,F) then a(S+T)^a(S)+a(T); and (f) if Ue£(X,E);TeA(E,F) and Ve<£(F, Y), then a(VTU)^\\V\\a(T)\\U\\. An ideal A with norm a, [A, a], is a Banach ideal if each component A(E,F) is a Banach space under a. To any linear normed ideal [A, a] one can associate three normed ideals in a more or less natural fashion: (I) The dual ideal [A', a']: An operator T is in A'(E,F) if and only if TeA(F\E'). Here a'(T)=a(T'); (II) The conjugate ideal [AA, aA]: AA(E,F) is the class of all operators Te£(E, F) for which there is a p>0 such that for any L e &(F, E) |traceLT|^pa(L). The norm aA(T) is defined by infp, p satisfying the above inequality. [AA, aA] is always a Banach ideal. (III) The adjoint ideal [A*, a*]: A*(E,F) is the class of all Te£{E,F) for which there is a p > 0 such that for all finite dimensional Banach spaces X, Y and for all Ve#(X,E), UeA(Y,X), WeX(F, Y), |trace VUWT| ^p\\W\\ \\V\\ a(U). Here the norm a* is also given by inf p, p satisfying the above inequality. The ideal [A*, a*] is also always a Banach ideal. The ideals AA and A* are intimately related. Indeed, for any Te!£(E, F), a*(T)^aA(T) and equality holds if both E and F have the metric approxi mation property. We now give a few examples of Banach ideals. These ideals are due to several different authors. Bibliographical information and a table showing the relationships between various ideals is given in [5.2]. 1975] APPLICATIONS OF BANACH IDEALS OF OPERATORS 983 (1) Let C(E, F) denote the closure of &(E, F) in 2(E, F), and K(E, F) the compact operators from E to F. Then [££, ||-||], [K, ||-||] and [C, ||-||] are Banach ideals. For a finite or denumerably infinite set {xi, • • •, x } in a Banach space E, N let eP({xJ = sup{(|i|(xi,/)|p)1/P:||/||=l}, if l^p < +oo? e-({x.}) = supjsup |<x,, />| : ||/|| = 1 j ; / £, \ 1/p M{*}) = (XJWIP) if I^P<+^, aoc({Xi}) = sup ||xi||; and ({x,}) = sup{ | f <x ƒ,) | : s ({ƒ,}) ^ l}, i9 P P P (2) [up, 7T ] denotes the ideal of p-absolutely summing operators: Te P n (E, F) if there is a p>0 such that a ({TXi})^pe ({Xi}), for all finite sets p p p {xi,- • •, x } in E. The norm TT is given by 7r (T)=inf p, p as above. N P p The ideal [n , 7r ] will be extensively used throughout the remainder of p p the paper. (3) [Dp, dp] denotes the ideal of strongly p-summing operators: Te D (E, F) if there is a p>0 such that a- [{TXi})^pa ({Xi}) for all finite sets P p p {xi, • • •, x } in E ; d (T) = inf p. N p (4) [Ip, i ] denotes the ideal of p-integral operators: T e I (E, F) if there is a p P probability measure /m and operators Ve&(E, LOO(JUL)) and We«S?(Loo(jui), F"), F" the bidual of F, such that WjV=iT, where j is the canonical in jection of LOO(JUL) into Lp(jui) and i the canonical injection of F into F", i.e., £_J_> j <—^->F" M/ui)—< •Lpüo The norm ip is given by i (T) = inf||V|| ||W||, the infimum taken over all p probability measures p, and operators V, W. (5) [N , v ] denotes the ideal of p-nuclear operators: TeN (E, F) if T has p p p a representation T=£rU/i®yi, freE', yeF and a ({/i})<+°°, and t P £p'({yi})<+00 (l/p+l/p' = 1). If p=°° there is the additional requirement that fi-*0 as i-»oo. The p-nuclear norm is given by i> (T) = inf a ({/i})e ({yi}), p P P where the infimum is taken over all such representations of T. An operator T in the class Ni will be called a nuclear operator. This class of course generalizes the "trace class operators" on Hubert space. (6) [Q, c ] denotes the ideal of operators factoring compactly through p t : T€ Q(E, F) if there are A e C(E, l ), B e C(l , F) with T = BA. The norm p p p 984 J. R. RETHERFORD [November c is given by c (T) = inf||A||||B||, where the infimum is over all such p p factorizations of T. (7) [Jp,q> ipq]> l=q=p=°°, denotes the ideal of operators factoring through a diagonal Be££(L (|w), L (ii)): TeI , (E, F) if for some positive measure /x p q pq there are operators A eS£(E, Lp(n)), B €^(L (ft), L (jLt)), where B is of the p q form Bf=f-g for some fixed geL(|m), l/r = l/q-l/p, and Ce££(L (fx), ƒ"), r q such that iT=CBA, where i is the canonical injection of F into F". The norm ^ is given by ipc(T) = inf||^V|| ||J3|| ||C||. Observe that I (E, F)=Lo, (E, F) with equality of norms. q q (8) [J ,q, jVq], l=q=p=°°, denotes the ideal of operators factoring through P D °n : Te J (E, F) if iT admits a factorization as follows: q p pq where UeU (E,G) and VeD (G,F"). Here j (T)=inf 7r (U)d (V), the p q M p q infimum taken over all U, V, G. The last ideal we will discuss is a generalization of an ideal introduced by Kwapien. (9) [r , ,7pq] is defined as in [I , , i ], the difference being that B ranges pq pq pq over all members of i£(L (jn), L (\x)). p q The adjoints, conjugates and duals of these ideals (and several others) are computed in [5.2]. We remark that in the ideals (8) and (9) whenever p = q we will index the ideal and its norm by p alone. 6. Ideal characterizations of ^ -spaces. We first present some charac p terizations of $£ -spaces via Banach ideals. We begin with a characterization p of ^«-spaces. THEOREM 1 [6.1]. The following assertions are equivalent: (a) n (E,F)=I (E,F) for all F; and 1 1 (b) E is an ^Eœ-space. Using this result Lewis and Stegall [6.2] proved THEOREM 2. Let E be a Banach space. Then IIi(E, F)=Ni(E, F) for all F if and only if E' is isomorphic to Ji(0 for suitable T. There are some beautiful applications of Theorem 2. APPLICATIONS. (1) If E is a complemented subspace of Li[0, 1], isomorphic to a subspace of a separable conjugate space, it is isomorphic to V. In particular. (2) Li[0,1] is not isomorphic to a subspace of a separable conjugate space (Gel'fand [6.3], Pelczynski [6.4]. (3) Any separable i£i-space which is isomorphic to a conjugate space is isomorphic to U. (We remark that there are many separable ^i-spaces. Indeed let <ï> :Zi—»Li[0,1] be a 0 surjection and let Xi=<&o1(0). Let $i be a surjection from h onto Xi and 1975] APPLICATIONS OF BANACH IDEALS OF OPERATORS 985 X =®ï1(0). In general let X = 4>n-i(0). Then all of the X are separable, 2 n n nonisomorphic ££i-spaces [6.5].) We now give an omnibus result which includes results of Cohen [6.6], Holub [6.7], Johnson [6.8], Kwapien [6.9], Lewis [6.10] and Persson [6.11], as well as some new results. This result was first proved in [6.12]. THEOREM 3. The following are equivalent (l^p^o°): (a) I , the identity on E, factors through L ; E p (b) r (F,E)=>C(F,E) for all F; p (c) r <E',F')=>C(E',F') for all F; p (d) r (E,E)2C(E,E) and E has the metric approximation property; p (e) T*(E, F) = ME, F) for all F; (f) T*(F, E) = h(F, E) for all F; (g) r*(E, E) = Ii(E, E) and E has the metric approximation property; (h) for every Banach space G, and every adjoint operator, if We n (E',G') then WeI (G,E); P p (i) if Ven (E,G) then V'eI >(G\E'); and P p (j) T (E,F)=>C(E,F)forallK p We mention that some other characterizations of c , Ji, i£«,-spaces and 0 ^i-spaces are given in [6.13]. It would be of considerable interest to know the situation whenever the range space in Theorem 2 is fixed. We conjecture that if IL(F, E) = Ni(F, E) for all Banach space F then E must be finite dimensional. 7. More on ^ -spaces. In [7.1] Grothendieck outlined the theory of p tensor products of Banach spaces. This was, in fact, the "beginning" of the theory of ideals of operators on Banach spaces. Indeed, Grothendieck showed the importance of factoring techniques which will be emphasized over and over in this paper. The crowning achievement of this work of Grothendieck was called by him "the fundamental theorem of tensor products." In terms of matrices this theorem can be stated as follows: Let (%)?,,=! be a finite matrix of real numbers such that EM=I o ^ l ^l whenever |fc|^il, |SJ|=1. Then, for every set of unit vectors (xO?=i and (y,-)?=i in a Hubert space X Oi,-(Xi, y,)pK , G where K is an absolute constant. Here (•, •) denotes the inner product in G the Hubert space. We remark that the exact value of K is not known. G Surprisingly this fundamental paper of Grothendieck lay dormant for many years but was finally taken up again by Lindenstrauss and Pelczynski in 1968 [7.2]. Lindenstrauss and Pekzynski were persuaded to write their paper avoiding the notion of tensor products because "the paper of Grothendieck was quite hard to read and its results were not generally known even to experts in Banach space theory." This remarkable paper of Lindenstrauss and Pefczyiiski contains the seeds of the application of 986 J. R. RETHERFORD [November Banach ideals of operators. (Earlier [7.3] Grothendieck essentially used the theory to obtain the Dvoretzky-Rogers theorem.) Before stating a few of their results let us mention that a new proof of the Grothendieck inequality has recently been given by Maurey [7.4]. One of the achievements of the Lindenstrauss-Pefczynski paper was the introduction of the classes of spaces called the i£ -spaces defined in the introduction. The proofs of the theorems p below depend heavily on the Grothendieck inequality. THEOREM 1. Let X be an ^Ei-space and H an S£2-space (=isomorph of a Hubert space). Then 2(X, H)=IIi(X, H). As remarked by Lindenstrauss and' Pefczynski it is conceivable that Theorem 1 actually characterizes i£i and i£ -spaces. Indeed they obtained 2 the following partial converse to Theorem 1. THEOREM 2. Let X and Y be Banach spaces such that X has an uncon ditional basis and such that i£(X, Y)=IIi(X, Y). Then X is isomorphic to h and Y is isomorphic to a Hubert space. THEOREM 3. Let X be an !£œ-space and Y an %v-space, l^p^2. Then S(X, Y) = n (X, Y). 2 Of the numerous applications of these results the following are striking examples. APPLICATION 1. Let X be a complemented subspace of an i£i(i£oo)-space Y and let (xi) be a normalized unconditional basis in X. Then the basis (x) is t equivalent to the unit vector basis (e*) in li(c ), i.e. the operator T defined 0 by TXi — d is an isomorphism. For the next application we consider the complex Banach space Li(fx), where JUL is Haar measure on {z:|z| = l}. Let Hi be the closure of the polynomials Y*=oakzk in LI(JUL). PROPOSITION. £(H l) * IIi(Hi, I). U 2 2 We thus obtain the classic result of D. J. Newman (see [7.5, p. 154]). APPLICATION 2. Every isomorphic image of Hi in an arbitrary i£i-space X is uncomplemented. Another application is Grothendieck's characterization of Hubert spaces. APPLICATION 3. A Banach space X is isomorphic to a Hubert space if and only if it is isomorphic to a subspace of an ££i -space and to a quotient of an ^oc-space. Of course, if the roles of the i£i- and i£œ-spaces are interchanged, every Banach space meets the requirement. Finally if ££(X, Y) = Ili(X, Y) there is a bit one can say about X. More precisely, PROPOSITION. If S£(X, Y) = IL(X, Y) then l02(X,I )=n,(X,I ); 2 2 2° if Xn=i *n is unconditionally convergent in X then £n=i ||xn||2<+00; and, 3° if Z is any espace, 2(Z, X) = n (Z, X). 2 For a detailed study of this proposition, see [7.6]. 1975] APPLICATIONS OF BANACH IDEALS OF OPERATORS 987 8. Schwartz duality theorem. Following Kwapien [8.1] we present the Schwartz duality theorem without the theory of cylindrical measures and radonifying operators. We first need to extend the notion of p-absolutely summing operator to include the interval [0,1). For 0<p<+o° define a p-absolutely summing operator in the obvious way, i.e., T is p-absolutely summing if there is a C such that for Xi • • • Xn e E î||Tx|rëCsupÉ|<x />|'\ i i) i = l 1|/||=1 i = l For p=0 we say that T is 0-absolutely summing (and write Tello(E, F), if for each e>0 there is a ô>0 such that for Xi • • • Xn e E and n ï sup £ - min[l, |<x /)|] < Ô, i? II/II=I i-i.n it follows that £"=i n1 min[l, ||TXi||]<e. This last definition is a reformulation of Schwartz's definition of a radonifying operator. Now let /UL be a probability measure on a Hausdorff space Ci and 0^p^o°. A linear operator v:E^>L (tl, n) is p-decomposable if there is a <p :il-+E' p such that (a) for each xeE, (x, <p(-)) is JUL-measurable and equal to v(x) n—a.e.; and (b) there is ƒ eL (fl, /x) such that ||<p(o>)||^/(û>) JUL—a.e. p THEOREM (THE SCHWARTZ DUALITY THEOREM). Let E be a Banach space and 0^p<+o°. If v:E-*L (ïî, /x) is p-decomposable then v is p-absolutely p summing. For 0^q<p<2 there is an isomorphic embedding of Lp[0,1] into L (H, fi). If p = 2 the same is true for all 0^q<+oo. We denote such an q isomorphism by v . p THEOREM 2. Let Kp^oo and T:Lp[0,1]->L2[0,1]. The following are equivalent: (a) v T is q-decomposable for all q«x>; 2 (b) v T is 0-decomposable; 2 (c) T is 0-absolutely summing; and, (d) T is p'-absolutely summing. THEOREM 3. Let either l^p<2 and 0^q<p or p = 2 and 0^q<+o°. Let F be an £ -space. If Te£(E,F) and T'eII (F',E') then TeU (E,F). p q 0 APPLICATION 1. Let F be an i£r-space where l<r^2. (a) If 0^p^2 then II (F, E)=U (F, E); P 0 (b) If 2^p<oo then n (E,F)=II (E,F). p 2 In particular for an isomorph of a Hilbert space (i.e., an £6 -space), H, an 2 operator is p-absolutely summing for some p, 0^=p<o° if and only if it is 2-absolutely summing (=Hilbert-Schmidt).