Positivity 2: 47–75, 1998. 47 © 1998 Kluwer Academic Publishers. Printed in the Netherlands.
Symmetric Functionals and Singular Traces
1 2 3,† 1, P.G. DODDS , B. DE PAGTER , E.M. SEMENOV and F.A. SUKOCHEV ∗ 1Department of Mathematics and Statistics, The Flinders University of South Australia, GPO Box 2100, Adelaide, SA 5001, Australia; 2Department of Mathematics, Delft University of Technology, Mekelweg 4, 2628 CD Delft, The Netherlands; 3Department of Mathematics, Voronezh State University, Universitetskaya pl.1, Voronezh 394693, Russia
(Received: 11 November 1997; Accepted: 4 December 1997) Abstract. We study the construction and properties of positive linear functionals on symmetric spaces of measurable functions which are monotone with respect to submajorization. The con- struction of such functionals may be lifted to yield the existence of singular traces on certain non- commutative Marcinkiewicz spaces which generalize the notion of Dixmier trace.
Mathematics Subject Classifications (1991): 46E30, 46L50
Key words: symmetric functional, singular trace, Marcinkiewicz space, symmetric operator space
1. Introduction and Preliminaries The purpose of this paper is to study the properties of positive linear functionals on symmetric Banach function spaces which are monotone with respect to subma- jorization, in both the commutative and non-commutative setting. Such functionals are called symmetric (see Definition 2.1 below). This notion finds its roots and motivation in the work of Dixmier [7] on the existence of non-trivial traces on the factor B(H) which are singular in the sense that they vanish on all finite rank operators. Such traces have subsequently found important application in non- commutative geometry and quantum field theory [4]. The methods used by Dixmier in his construction of singular traces are based on ideas familiar in the theory of rearrangement invariant spaces; in fact, his approach can be viewed as first constructing a symmetric functional on a certain Marcinkiewicz sequence space and then “lifting” this functional to the ideal of compact operators whose singular value sequences lie in this space. Our objective therefore is to study the existence and properties of symmetric functionals, first in the setting of symmetric spaces E on the positive half-line (see sections 2,3 below) and then to lift the results (see section 4) to the more general setting of symmetric spaces of measurable operators affiliated with some semi-finite von Neumann algebra.
† Research supported by RFFI, Grant 95-01-00135 ∗ Research supported by the Australian Research Council.
PIPS No.: 159135 (postkap:mathfam) v.1.15
*159135* post56.tex; 23/04/1998; 10:32; p.1 48 P.G. DODDS ET AL.
Since any normal symmetric functional is a scalar multiple of integration with respect to Lebesgue measure (see Proposition 2.6), our principal results concern the existence of singular symmetric functionals. The familiar Lebesgue spaces p 1 1 L 0, ), 1
x∗(t) inf s 0 m( x >s ) t ,t>0. = { ≥ : {| | } ≤ } For basic properties of symmetric spaces and rearrangements, we refer to the mono- graphs [14], [3]. We note that for any symmetric space E,
1 1 L 0, ) L∞ 0, ) E L 0, ) L∞ 0, ), [ ∞ ∩ [ ∞ ⊂ ⊂ [ ∞ + [ ∞
post56.tex; 23/04/1998; 10:32; p.2 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 49 with continuous embeddings. We will denote by E0 the symmetric space associated with E. The space E0 consists of all measurable functions y for which
∞ y E sup x(t)y(t) dt x E, x E 1 < . k k 0 := | | : ∈ k k ≤ ∞ Z0 If E∗ denotes the Banach dual of E, it is known that E0 E∗ and E0 E∗ if ⊂ = and only if the norm E is order-continuous, i.e. from xn E,xn n 0, it k·k { }⊆ ↓ follows that xn E 0. We note that the norm E on the symmetric space E k k → k·k is order-continuous if and only if E is separable, in which case limt x∗(t) 0 for all x E. →∞ = ∈ 1 If x,y L 0, ) L∞ 0, ), we will say that x is submajorized by y and write x y∈if and[ only∞ + if [ ∞ ≺≺ t t x∗(s)ds y∗(s)ds for all t>0. ≤ Z0 Z0 The symmetric space E on 0, ) will be called fully symmetric if and only if [ ∞ x y, y E, implies x E and x E yE. It is well known (see, for ≺≺ ∈ ∈ k k ≤kk example, [14]) that (E, E)is fully symmetric if and only if E is an interpolation 1 k·k space for the pair (L 0, ), L∞ 0, )), with interpolation constant 1. If the norm on E is order continuous,[ ∞ then E[is fully∞ symmetric.
2. Symmetric Functionals on 0, ). [ ∞ Throughout this section, E will denote a fully symmetric Banach function space on 0, ). [ ∞ DEFINITION 2.1. A linear functional 0 φ E∗ will be called symmetric if and only if 0 f, g E and f g imply that≤ φ(f)∈ φ(g). ≤ ∈ ≺≺ ≤ If E L1 0, ), then the functional φ defined by setting φ(f) ∞ f(t)dt for ⊆ [ ∞ = 0 all f E is clearly symmetric. Recall that a functional 0 φ E∗ is called ∈ ≤ ∈R normal (or order continuous)iffn n 0a.e.inEimplies that φ(fn) n 0. For example, see [25]. ↓ ↓
1 PROPOSITION 2.2. If 0 <φ E∗is normal and symmetric, then E L 0, ) ∈ ⊆ [ ∞ and there exists 0 0 φ(χA) w(t)dt αm(A) < βm(B) w(t)dt φ(χB), = ≤ ≤ = ZA ZB post56.tex; 23/04/1998; 10:32; p.3 50 P.G. DODDS ET AL. and this is a contradiction. 2 Before presenting some non-trivial examples of symmetric functionals, we first show that symmetric functionals are dilation invariant in the sense of of the fol- lowing proposition. For s>0, the dilation operator Ds is defined by setting Dsf(t) f(t/s)for all t>0 and functions f on 0, ). The linear mapping Ds := [ ∞ is bounded on E; moreover, Ds E max 1,s ([14], Corollary 1 to Theorem II k k ≤ { } 4.5). Note that (Dsf)∗ Dsf∗. = PROPOSITION 2.3. If 0 <φ E∗ is symmetric, then φ(Dsf) sφ(f ) for all f E and all s>0. ∈ = ∈ Proof. Let 0 f E.Forn N , there exist disjointly supported 0 ≤ ∈ ∈ ≤ f ,...,fn Eand such that f ∗ f ∗ for all k 1,...,n.Leth f fn. 1 ∈ k = = := 1 +···+ It is easy to see that h∗(t) f ∗(t/n) for all t>0, that is, h∗ Dnf ∗. Hence, = = φ(Dnf) φ((Dnf)∗) φ(Dnf∗) φ(h∗) = n = n = φ(h) φ(fk) φ(f∗) = = = k k 1 k 1 X= X= nφ(f ∗) nφ(f ). = = Via a standard argument, it now follows that φ(Drf) rφ(f ) for all rationals = 0 1 = 1 ≤ 2 = 2 Q Given 0 for all n 1, 2,... which implies that φ(Dsf) sφ(f ). 2 = = Remark 2.4. It is clear from the above that non-trivial symmetric functionals fail to exist on any of the spaces E Lp 0, ), 1 0(where1(t) 1, for all t>0). = = We now present some examples of non-trivial symmetric functionals. Further constructions will be discussed in the next section. EXAMPLE 2.5. (i) (cf. [19],[1]) Let ψ 0, ) 0, )be concave and :[ ∞ →[ ∞ increasing with ψ( ) limt ψ(t) and ∞ := →∞ =∞ ψ(2t) lim 1. (2.1) t ψ(t) →∞ = post56.tex; 23/04/1998; 10:32; p.4 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 51 For example, we may take ψ(t) log(t 2), t 0. Let Mψ be the corresponding Marcinkiewicz space consisting:= of all measurable+ ≥ functions f on 0, ) for which [ ∞ 1 t sup f ∗(s)ds < ψ(t) ∞ t>0 Z0 equipped with the norm 1 t f M sup f ∗(s)ds. k k ψ := ψ(t) t>0 Z0 For basic properties of Marcinkiewicz spaces, we refer to [14]. The space (Mψ , Mψ)is a fully symmetric Banach function space. Note that ψ0 Mψ ,butthat k·k 1 1 ∈ ψ0 L 0, ) since ψ( ) . Consequently, Mψ L 0, ).For0 f 6∈ [ ∞ ∞ =∞ 6⊆ [ ∞ ≤ ∈ Mψ, we define the function Φ(f) by setting 1 t Φ(f )(t) f ∗(s)ds, t > 0. (2.2) := ψ(t) Z0 It follows that 0 Φ(f) Cb(0, ),whereCb(0, )denotes the space of all bounded (real-valued)≤ continuous∈ functions∞ on (0, )∞. We now observe that ∞ Φ(f g) Φ(f) Φ(g) wD / Φ(f g) (2.3) + ≤ + ≤ 1 2 + for all 0 f, g Mψ ,wherew(t) ψ(2t)/ψ(t), t > 0. In fact, the first ≤ ∈ = inequality is just the well known submajorization f g f ∗ g∗. The second inequality follows by noting that + ≺≺ + t t 2t f ∗(s)ds g∗(s)ds (f g)∗(s)ds + ≤ + Z0 Z0 Z0 which follows from [14], II.2.2. Now let L be a linear functional on the space Cb(0, ) with the properties ∞ ∞ (a) L (u) 0forall0 u Cb(0, ). ∞ ≥ ≤ ∈ ∞ (b) L (u) limt u(t), for all u Cb(0, ) for which lim u(t) exists. ∞ →∞ t = ∈ ∞ →∞ (c) L (Dsu) L (u) for all s>0, u Cb(0, ). ∞ = ∞ ∈ ∞ Such a functional is easily obtained from a translation invariant generalized limit at on Cb( R ). Note that if u, v Cb(0, ) and limt v(t) exists, then +∞ ∈ ∞ →∞ L (vu) (limt v(t))L (u).For0 f Mψ,wenowdefineφ(f) L∞(Φ(f )).= From→∞(2.3) it follows∞ that ≤ ∈ := ∞ φ(f g) φ(f) φ(g) L (wD1/2Φ(f g)) + ≤ + ≤ ∞ + for all 0 f, g Mψ.By(2.1), limt w(t) 1, and so the given properties of the functional≤ L∈ imply that →∞ = ∞ L (wD1/2Φ(f g)) L (Φ(f g)) φ(f g) ∞ + = ∞ + = + post56.tex; 23/04/1998; 10:32; p.5 52 P.G. DODDS ET AL. and consequently φ(f g) φ(f) φ(g) for all 0 f, g Mψ . Moreover, it + = + ≤ ∈ is clear that φ(λf) λφ(f ) for all 0 f Mψ , 0 λ R , and therefore φ = ≤ ∈ ≤ ∈ extends to a positive linear functional on Mψ , which we continue to denote by φ. From the definition it is clear that φ is symmetric; further, ψ(t) ψ(0 ) Φ(ψ0)(t) − + 1 = ψ(t) → 1 as t . Consequently, φ(ψ0) 1andso φ 1. If 0 f Mψ L 0, ), →∞ = k k= ≤ ∈ ∩ [ ∞ then limt Φ(f )(t) 0andsoφ(f) 0. This shows that φ 0onMψ 1 →∞ 1= = = ∩ L 0, ). Note that L 0, ) Mψ whenever ψ(0 )>0. [Finally,∞ we observe[ that∞ if the⊆ above construction+ is carried out in the case that 1 ψ( )< ,thenMψ L 0, )and the functional φ obtained is given by ∞ ∞ 1 ⊆ [ ∞ φ(f) ψ( )− ∞ f(s)ds since = ∞ 0 R 1 ∞ lim Φ(f )(t) f∗(s)ds t = ψ( ) →∞ ∞ Z0 for all 0 f Mψ in this case. (ii) Assume≤ ∈ now that ψ 0, ) 0, )is concave and increasing with ψ(0 ) 0, and that :[ ∞ →[ ∞ + = ψ(2t) lim 1. (2.4) t 0 ψ(t) = ↓ 1 1 For example, we may take ψ to be given by ψ(t) 2(1 log t)− for 0 post56.tex; 23/04/1998; 10:32; p.6 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 53 Our next objective is to obtain decomposition theorems for symmetric func- tionals which will give considerable insight into their structure. We recall that a functional 0 φ E∗ is called singular if whenever 0 φ0 E∗ is order ≤ ∈ ≤ ∈ continuous and satisfies 0 φ0 φ, it follows that φ0 0 (equivalently, φ is disjoint from all order continuous≤ ≤ functionals on E; see, for= example [25], Section 87). As is well known, a functional 0 φ E∗ is singular if and only if φ ≤ ∈ vanishes on some order dense ideal, [25] Section 90. Moreover, any 0 φ E∗ ≤ ∈ has a unique decomposition φ φn φs , with φn order continuous (normal) and = + φs singular. In the proposition which follows, we will show that if φ is symmetric, then both φn and φs are symmetric. We denote by Eb the norm closure in E of the 1 (order dense) ideal (L L∞) 0, ); equivalently, Eb is the norm closure in E of the space of all stepfunctions.∩ [ ∞ PROPOSITION 2.6. If 0 φ E∗ is a symmetric functional, then there exist ≤ ∈ symmetric functionals 0 φ ,φ E∗ for which φ φ φ and such that ≤ 1 2 ∈ = 1 + 2 ∞ R (i) φ1(f ) c 0 f(t)dt for some 0 c , with c>0only possible if E L1=0, ). ≤ ∈ ⊆ [ ∞R (ii) φ 0 on Eb (in particular, φ is singular). 2 = 2 The decomposition φ φ1 φ2 is the unique decomposition of φ into its normal and singular parts. = + Proof. Observe first that if φ(χA) 0 for all measurable A 0, )with = ⊆[ ∞ m(A) < ,thenφ 0onEb and so we may take φ 0andφ φ ∞ = 1 = = 2 in this case. Now assume that there exists a measurable set A0 0, )such χ χ χ ⊆[ ∞ that m(A0)< and φ( A0)>0. Since A∗ Dm(A0) 0,1 , it follows that χ ∞χ χ 0 = χ [ ] 0 φ( A0) φ( A∗ ) m(A0)φ( 0,1 ),andsoφ( 0,1 )>0. Without loss 6= = 0 = [ ] [ ] of generality, we may assume that φ(χ 0,1 ) 1. It then follows that φ(χA) [ ] = = χ ∞ m(A)φ( 0,1 ) m(A) for all measurable A 0, )so that φ(f) 0 f(t)dt for all stepfunctions[ ] = f .If0 f E, there⊆[ exists∞ a sequence of= stepfunctions ≤ ∈ R fn n∞ 1 such that 0 fn n f a.e. on 0, ).Since { } = ≤ ↑ [ ∞ ∞ 0 fn(t)dt φ(fn) φ(f), n 1,2,..., (2.5) ≤ = ≤ = Z0 it follows from the Beppo–Levi theorem that f L1 0, ). Consequently, E 1 ∈ [ ∞ ⊆ L 0, ).Ifφ(f ) ∞ f(s)ds for all f E,thenφ E∗ and from (2.5) [ ∞ 1 := 0 ∈ 1 ∈ it follows that 0 φ1 φ.Sinceφ(f) φ1(f ) for all stepfunctions f ,itis ≤ R≤ = clear that φ φ on Eb. Putting φ φ φ , it follows that 0 φ E∗ = 1 2 = − 1 ≤ 2 ∈ and φ 0onEb. It remains to show that φ is symmetric. To this end, suppose 2 = 2 that 0 post56.tex; 23/04/1998; 10:32; p.7 54 P.G. DODDS ET AL. χ If f2 f1 f1(t0) (t0,t1 ,thenf2 g1,φ1(f2) φ1(g1) and φ2(f2) φ2(f1). Since:=φ is symmetric,+ f ] g implies≺≺ that φ(f ) =φ(g ), hence = 2 ≺≺ 1 2 ≤ 1 φ (f ) φ (f ) φ(f ) φ (f ) 2 = 2 2 = 2 − 1 2 φ(g ) φ (g ) φ (g ) φ (g). ≤ 1 − 1 1 = 2 1 = 2 This shows that g is symmetric. Since the last assertion of the proposition is obvi- ous, this completes the proof of the proposition. 2 It follows in particular from the preceding proposition that any singular symmetric 1 functional vanishes on the order dense ideal Eb. Further, if E L 0, ),then any symmetric functional on E is singular. Let us note that a6⊆ result[ similar∞ to Proposition 2.6 is given in [12] Lemma 2.5, although there appears to be a gap in the proof given there. We will now give a further decomposition of singular symmetric functionals. We shall need the following terminology. DEFINITION 2.7. A symmetric functional 0 φ E∗ is said to be (i) supported ≤ ∈ 1 at 0 if φ 0 on E L∞ 0, ); (ii) supported at if φ 0 on E L 0, ). = ∩ [ ∞ ∞ = ∩ [ ∞ Remark 2.8. (i) A symmetric functional supported either at 0 or at is clearly ∞ singular. (ii) If 0 φ ,φ E∗ are symmetric functionals such that φ is supported ≤ 1 2 ∈ 1 at 0 and φ2 at ,thenφ1 φ2 0. Indeed , φ1 φ2 0onE L∞ 0, )and on 1 ∞ ∧ = 1 ∧ = ∩ [ ∞ E L 0, ), and hence on E since E (L L∞) 0, ). ∩ [ ∞ ⊆ + [ ∞ The terminology introduced in Definition 2.7 is in fact motivated by the alter- native characterizations given in the following lemma. We leave the simple proof to the reader. LEMMA 2.9. Let 0 φ E∗ be a symmetric functional. ≤ ∈ (a) φ is supported at 0 if and only if φ(f ∗χ(s, )) 0 for all f E and all ∞ = ∈ 0 0 THEOREM 2.10. If 0 φ E∗ is a singular symmetric functional, then φ has a ≤ ∈ unique decomposition φ φ1 φ2 such that 0 φ1,φ2 E∗ are symmetric, φ1 is supported at 0 and φ is= supported+ at . ≤ ∈ 2 ∞ Proof. Note first that any f E can be written as f f1 f2 with f1 1 ∈ = + ∈ E L 0, ) and f2 E L∞ 0, ) (with f1,f2 0iff 0). Suppose that f f ∩ f[ is∞ another such∈ decomposition.∩ [ ∞ Then f f≥ f ≥f (L1 L ) 0, =) ˜1 ˜2 1 ˜1 ˜2 2 ∞ and+ so φ(f f ) φ(f f ) 0. This yields− that=φ(f−) ∈φ(f )∩and φ(f[ )∞ 1 ˜1 ˜2 2 1 ˜1 2 φ(f ). We can− therefore= define− φ=,φ by setting φ (f ) φ(f= ), φ (f ) φ(f =). ˜2 1 2 1 := 1 2 := 2 It is clear that 0 φ ,φ E∗ and that φ φ φ . Moreover, φ 0on ≤ 1 2 ∈ = 1 + 2 1 = post56.tex; 23/04/1998; 10:32; p.8 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 55 1 E L∞ 0, ) and φ 0onE L 0, ). It remains to be shown that φ ,φ ∩ [ ∞ 2 = ∩ [ ∞ 1 2 are symmetric. Suppose then that 0 f, g E and that f g.Setg g1 g2 1 ≤ ∈ ≺≺ = + with g E L 0, ) and g E L∞ 0, ). From [3] Section 3.7, there exist 1 ∈ ∩ [ ∞ 2 ∈ ∩ [ ∞ 0 f1,f2 E with f1 g1,f2 g2 and f f1 f2.Sinceφis symmetric, it follows≤ that∈φ (f ) φ(f≺≺) φ(g≺≺) φ(g),= and similarly,+ φ (f ) φ (g),andit 1 = 1 ≤ 1 = 2 ≤ 2 follows that φ1,φ2 are symmetric. 2 Note that the functional constructed in Example 2.5 (i) is supported at ,andthat in Example 2.5 (ii) is supported at 0. The above decomposition is illustrated∞ in part (iii) of the same example. Note further that if E L1 0, ), then necessarily any ⊆ [ ∞ 1 singular functional is supported at 0, since E L∞ 0, ) (L L∞) 0, ) in this case. ∩ [ ∞ = ∩ [ ∞ Remark 2.11. (Symmetric functionals on 0, 1)). Given a fully symmetric Banach function space on 0, ), we may consider[ the corresponding Banach function space E on 0, 1) consisting[ ∞ of those measurable functions f on 0, 1) for which [ [ f ∗ E, with norm given by f E 0,1) f ∗ E,forallf E0,1).Itis easy∈ to see that every fully symmetrick k [ Banach:= k functionk space on∈0, 1[) is of this form. The notion of a symmetric functional on E 0, 1) is obtained[ by the obvious modification of Definition 2.1. [ Let 0 φ E 0, 1)∗ be a symmetric functional. For 0 φ0(f g) φ (f g)∗χ 0,1) + = + [ φ(f ∗χ 0,1) g∗χ 0,1)) φ0(f ) φ0(g). ≤ [ + [ = + For the converse inequality, note that f ∗ g∗ 2D / (f g)∗,andso + ≺≺ 1 2 + f∗χ0,1/2) g∗χ0,1/2) 2D1/2 (f g)∗χ 0,1) . [ + [ ≺≺ ˜ + [ post56.tex; 23/04/1998; 10:32; p.9 56 P.G. DODDS ET AL. Using the fact that φ vanishes on L∞ 0, 1),thisimpliesthat [ φ0(f ) φ0(g) φ(f ∗χ 0,1/2) g∗χ 0,1/2)) + = [ + [ 2φ(D1/2 (f g)∗χ 0,1) ) ≤ ˜ + [ φ (f g)∗χ 0,1) . = + [ This concludes the proof of the claim. Finally, we note conversely that if 0 φ E∗ is a symmetric functional and if ≤ 0 ∈ φ(f) φ (f ∗) for all 0 f E 0, 1),thenφextends to a symmetric functional := 0 ≤ ∈ [ 0 φ E 0, 1)∗ (see Theorem 4.2 below). Of course, if φ0 is supported at , then≤ φ ∈ 0.[ The preceding observations reduce the study of symmetric functionals∞ on 0, 1=) to the corresponding study on 0, ). [We conclude this section with some[ remarks∞ concerning the structure of the space of symmetric functionals on E. For this purpose, we will denote by (E∗ )sym + the collection of all symmetric functionals on E. It is clear that (E∗ )sym is a cone, + that is, if φ1,φ2 (E∗ )sym and if 0 α, β R ,thenαφ1 βφ2 (E∗ )sym. ∈ + ≤ ∈ + ∈ + PROPOSITION 2.12. With respect to the order structure induced by E∗, the cone (E∗ )sym is a lattice. More precisely: + (i) if φ1,φ2 (E∗ )sym,thenφ1 φ2 (E∗ )sym,whereφ1 φ2 denotes the ∈ + ∧ ∈ + ∧ infimum of φ1,φ2 in the vector lattice ReE∗; (ii) if φ1,φ2 (E∗ )sym, then the supremum φ1 s φ2 of φ1 and φ2 in (E∗ )sym exists and∈ is given+ by the formula ∨ + φ s φ (f ) sup φ (u) φ (v) 0 u, v E,u v f (2.6) 1 ∨ 2 = { 1 + 2 : ≤ ∈ + ≺≺ } for all 0 f E. ≤ ∈ Proof. (i) We recall that the infimum of φ1,φ2 in the vector lattice ReE∗ is given by the formula (φ φ )(f ) inf φ (u) φ (v) 0 u, v E,u v f 1 ∧ 2 = { 1 + 2 : ≤ ∈ + = } for all 0 f E. Now suppose that 0 f, g E and f g.If0 u, v E ≤ ∈ ≤ ∈ ≺≺ ≤ ∈ with u v g, then there exist 0 u1,u2 E with u1 v1 f and u1 u, v +v.Since= φ ,φ are symmetric,≤ it follows∈ that + = ≺≺ 1 ≺≺ 1 2 (φ φ )(f ) φ (u ) φ (v ) φ (u) φ (v), 1 ∧ 2 ≤ 1 1 + 2 1 ≤ 1 + 2 and hence (φ φ )(f ) (φ φ )(g). Consequently, φ φ is symmetric. 1 ∧ 2 ≤ 1 ∧ 2 1 ∧ 2 (ii) If the right hand side of (2.6) is denoted by φ0(f ), then it clearly suffices to show that φ0 is additive on E+. To this end, let 0 f, g E and suppose that 0 u, v E satisfy u v f g.There exist≤ 0 ∈f ,g E such ≤ ∈ + ≺≺ + ≤ 1 1 ∈ that u v f1 g1 and f1 f, g1 g respectively. Using the well-known Riesz decomposition+ = + property for≺≺ vector≺≺ lattices, we may write u u u and = 1 + 2 post56.tex; 23/04/1998; 10:32; p.10 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 57 v v1 v2 with 0 u1,u2,v1,v2 Eand u1 v1 f1 and u2 v2 g1.Since u= v+ f and u ≤ v g, it follows∈ that + = + = 1 + 1≺≺ 2 + 2 ≺≺ φ (u) φ (v) (φ (u ) φ (v )) (φ (u ) φ (v )) 1 + 2 = 1 1 + 2 1 + 1 2 + 2 2 φ (f ) φ (g) ≤ 0 + 0 and so φ (f g) φ (f ) φ (g). To prove the converse inequality, let >0be 0 + ≤ 0 + 0 given and assume that 0 ui ,vi E,i 1,2 satisfy u1 v1 f and u2 v2 g and ≤ ∈ = + ≺≺ + ≺≺ φ (u ) φ (v )>φ(f ) , φ (u ) φ (v )>φ(g) . (2.7) 1 1 + 2 1 0 − 1 2 + 2 2 0 − If u 1/2D (u u ) and v 1/2D (v v ),then := 2 1 + 2 := 2 1 + 2 1 1 u v D(u u ) D (v v ) + = 2 2 1 + 2 + 2 2 1 + 2 1 1 D f ∗ D g∗ ≺≺ 2 2 + 2 2 1 1 ∗ 2D / D f D g (f g)∗, ≺≺ 1 2 2 2 + 2 2 = + and hence φ1(u) φ2(v) φ0(f g). On the other hand, using Proposition 2.3 and (2.7), it follows+ that ≤ + φ (u) φ (v) φ (u u ) φ (v v ) 1 + 2 = 1 1 + 2 + 2 1 + 2 (φ (u ) φ (v )) (φ (u ) φ (v )) = 1 1 + 2 1 + 1 2 + 2 2 >φ(f ) φ (g) 2. 0 + 0 − Consequently, φ (f ) φ (g) φ (f g), and this suffices to complete the proof.2 0 + 0 ≤ 0 + 3. Existence of Symmetric Functionals As in the previous section, E will continue to denote a fully symmetric Banach function space on 0, ). In this section, we consider the existence and non- existence of (non -trivial)[ ∞ symmetric functionals on certain classes of Banach func- tion spaces. As noted in Proposition 2.6, non-zero symmetric normal functionals exist if and only if E L1 0, ), and so the primary interest lies in the existence of singular symmetric⊆ functionals.[ ∞ We note immediately that non-zero symmetric 1 1 functionals do not exist on (L L∞) 0, ) and (L L∞) 0, ). Indeed, this 1 ∩ [ ∞ + [ ∞ is clear for (L L∞) 0, ) from Proposition 2.6. Now suppose that 0 φ is a ∩ [ ∞1 ≤ symmetric functional on (L L∞) 0, ).Sinceφis singular, it follows that φ 0 1 + [ ∞ 1 1 = on (L L∞) 0, ) and hence φ 0onL 0, )as L 0, ) is contained in ∩ [ ∞ 1 = [ ∞ [ ∞ the norm closure of L L∞ 0, ). Moreover, it follows from Proposition 2.3 that ∩ [ ∞ 1 φ 0onL∞ 0, )and hence φ 0on(L L∞) 0, ). We will show first = [ ∞ = + [ ∞ post56.tex; 23/04/1998; 10:32; p.11 58 P.G. DODDS ET AL. that non-zero singular symmetric functionals do not exist on Orlicz and Lorentz spaces. For the convenience of the reader we recall some of the definitions, but for the details, we refer to any of the books [3,14,25]. In fact, we will treat the Orlicz and Lorentz spaces by considering the (so-called) Orlicz–Lorentz spaces. For more details on such spaces, we refer to [13]. Let Φ 0, ) 0, be a Young’s function, that is, Φ is convex, Φ(0) 0 and 0 <Φ(t)<:[ ∞ →[for some∞] t>0, and let ψ 0, ) 0, )be increasing,= ∞ :[ ∞ →[ ∞ concave and ψ(0) 0. We assume that if Φ jumps to at some t0 > 0, then = 0 ∞ Φ(t ) Φ(t ). The modular MΦ,ψ is defined on L 0, ) by setting 0 = 0− [ ∞ ∞ MΦ,ψ(f ) Φ(f ∗(t))dψ(t) := Z0 ∞ Φ(f ∗(t))ψ0(t)dt ψ(0 )Φ( f ). (3.1) = + + k k∞ Z0 The Orlicz–Lorentz space LΦ,ψ is defined by setting 0 1 LΦ,ψ f L 0, ) MΦ,ψ f < for some k>0 := ∈ [ ∞ : k ∞ equipped with the norm Φ,ψ given by k·k 1 f Φ,ψ inf k>0 MΦ,ψ f 1 . (3.2) k k := : k ≤ The Orlicz–Lorentz space (LΦ,ψ, Φ,ψ ) is a fully symmetric Banach function space on 0, ). In the special casek·k that ψ(t) t for all t 0, we obtain the [ ∞ = ≥ Orlicz space LΦ (with the Luxemburg norm Φ). In the case that Φ(t) t for k·k = all t 0, we obtain the Lorentz space associated with ψ, usually denoted 3ψ . ≥ THEOREM 3.1. If φ is a symmetric singular functional on the Orlicz–Lorentz space LΦ,ψ,thenφ 0. Proof. Suppose, on= the contrary, that 0 <φis a non-trivial symmetric singular functional on LΦ,ψ. We may assume that φ 1. Let 0 ∞ 1 Φ D f (s) ψ0(s)ds > 1, τ τ n Z0 post56.tex; 23/04/1998; 10:32; p.12 SYMMETRIC FUNCTIONALS AND SINGULAR TRACES 59 or equivalently ∞ 1 1 Φ f (s) ψ0(τs)ds > , τ n τ Z0 for all τ>0andalln 1,2,....Sincef fn n 0on 0, )this implies that = ≥ ↓ [ ∞ ∞ 1 Φ f(s) ψ0(τs)ds τ =∞ Z0 for all τ>0. Since ψ 0 is decreasing, it follows that ∞ 1 Φ f(s) ψ0(s)ds τ =∞ Z0 for all τ 1, and hence for all τ>0. This contradicts the fact that f LΦ,ψ. ≥ ∈1 Suppose now that φ is supported at ,thatis,thatφ 0onLΦ,ψ L 0, ). If ∞ = ∩ [ ∞ 1 LΦ,ψ then L∞ LΦ,ψ. As noted in Remark 2.4, φ 0onL∞ 0, ).Sinceφ is∈ supported at , it⊆ then follows that φ 0, contrary to= hypothesis.[ Consequently,∞ ∞ = 1 LΦ,ψ. In particular, it follows that f(t) 0ast .Iffn (f χ n, ))∗ 6∈ → →∞ := [ ∞ for n 1, 2,...,thenf fn 0 pointwise on 0, ); in fact, fn n 0. As in the preceding= paragraph,≥ but using↓ now the assumption[ ∞ that φ isk supportedk∞ ↓ at , 1 ∞ we obtain that τ − Dτ fn Φ,ψ > 1 and hence k k ∞ 1 1 Φ Dτ fn(s) ψ0(s)ds ψ(0 )Φ Dτ fn > 1 τ + + τ Z0 ∞ for all τ>0. The last term is only present if ψ( 0 )>0, in which case LΦ,ψ + ⊆ L∞ 0, ). Again via a substitution it follows that [ ∞ ∞ 1 1 1 1 Φ fn(s) ψ0(τs)ds ψ(0 )Φ fn > τ + τ + τ τ Z0 ∞ for all τ>0andn 1,2,.... For each fixed τ, it follows from fn n 0that 1 = k k∞ ↓ Φ fn n 0, and so f fn n 0on 0, ), implies that τk k∞ ↓ ≥ ↓ [ ∞