• Aucun résultat trouvé

SOME DENSITY THEOREMS FOR COMPLEX LOCALLY CONVEX LATTICES

N/A
N/A
Protected

Academic year: 2022

Partager "SOME DENSITY THEOREMS FOR COMPLEX LOCALLY CONVEX LATTICES"

Copied!
13
0
0

Texte intégral

(1)

FOR COMPLEX LOCALLY CONVEX LATTICES

DIMITRIE KRAVVARITIS and GAVRIIL P ˘ALTINEANU

In this paper we continue the study began in [3] of real locally convex lattices.

Thus, some Bishop type approximation theorems for complex locally convex lat- tices of (M)-type are given. We also show how various approximation theorems, known for weighted spaces, follows from this results.

AMS 2000 Subject Classification: Primary 41A65; Secondary 46A40.

Key words: complex lattice, complex locally convex lattice of (M)-type, antisym- metric ideal, antisymmetric set, weighted space.

1. INTRODUCTION

LetE be a complex locally convex lattice of (M)-type,Aa subset of the center of E, and F a (complex) vector subspace ofE.

First, we introduce the notion of antisymmetric ideal with respect to the pair (A, F). This notion is a generalization, for locally convex lattices, of the notion of antisymmetric set from the theory of function algebras.

Further, we study the main properties of the familyAF,A of all (A, F)- antisymmetric ideals and state some density theorems of the type

x∈F iffπI(x)∈πI(F), ∀I ∈ AF,A, I-minimal, where πI is the canonical mappingE →E/I.

An abstract version of the Stone-Weierstrass theorem for complex locally convex lattices is also given.

2. PRELIMINARIES

According to [9] a complex vector lattice is a pair (E, m), whereE is a complex vector space and m a modulus mapping, i.e. m : E → E with the properties:

i)m(m(x)) =m(x),∀x∈E;

ii) m(αx) =|α|m(x),∀α∈C,∀x∈E;

iii)m(m(m(x) +m(y))−m(x+y)) =m(x) +m(y)−m(x+y),∀x, y∈E;

REV. ROUMAINE MATH. PURES APPL.,53(2008),2–3, 167–179

(2)

iv)E= Sp(m(E)) (the (complex) subspace spanned bym(E)).

Now, we recall some results and definitions concerning complex vector lattices. The reader is referred to [4], [8] and [9].

Proposition2.1.If E is a complex lattice, then m(E) is a convex cone and ER =m(E)−m(E)is a real vector lattice with respect to the order relation

x6yiffy−x∈m(E).

Moreover, x>0 iff x=m(x) and for each x∈ER we have m(x) =|x|= (−x)∨x.

Proposition2.2. Every x∈E can be written in a unique way as a sum x=x1+ ix2, xi∈ER, i= 1,2.

We shall denote Rex=x1 and Imx=x2. So, for every x ∈E we have x= Rex+ i Imx, and x = Rex−i Imx.

Definition 2.1. Let E be a complex vector lattice (c.v.l.). A complex vector sublattice ofE, is a complex vector subspaceF ofEwith the following property: x ∈F implies Rex∈F and m(x) ∈F. A subset S of E is said to be solid if m(y) 6m(x),∀x ∈S,∀y ∈E impliesy ∈S. An ideal I ofE is a complex vector subspace I of E which is a solid set.

Proposition2.3.Let Ebe a c.v.l.,I ⊂Ean ideal andπI:E →E/I the canonical mapping. The modulus on the quotient vector space E/I is defined by

mcII(x)) =πI(m(x)), ∀x∈E.

The pair (E/I,mcI) is a complex vector lattice.

Definition 2.2. A complex vector lattice E is called archimedean ifER

is archimedean (i.e. x60 wheneverαx6y for somey∈E+ and allα >0).

Definition 2.3. Let E be a c.v.l., and let r ∈ E+. A sequence {xn} is called r-uniformly convergent to x if for every ε > 0 there exists a natural number nε such thatm(xn−x)6εrfor all n>nε.

A sequence {xn} is called r-uniformly fundamental if for every ε > 0, there exists a natural number nε such that m(xn+p−xn)6εr for all n>nε and all p∈N.

A complex vector lattice E is called relatively uniformly complete if for everyr ∈E+ and everyr-fundamental sequence {xn}there existsx∈E such that {xn} ber-uniformly convergent to x.

Proposition2.4. Every archimedean and relatively uniformly complete c.v.l., has the Riesz decomposition property (i.e. for each x∈E , y1, y2 ∈E+

satisfying m(x) 6 y1+y2, there exist x1, x2 ∈E such that x =x1+x2 and m(x1)6y1, m(x2)6y2).

(3)

See [4], Theorem 3.3.

Proposition2.5. Let E be a c.v.l. Then for every x∈E we have m(x) = sup{Reax; a∈C, |a|= 1}.

Definition 2.4. LetE1 and E2 be two c.v.l. A (complex) linear operator T : E1 → E2 is called order bounded, if T maps order bounded sets into order bounded sets (i.e. for each x∈(E1)+ there existsu ∈(E2)+ such that m1(y)6x impliesm2[T(y)]6u).

Proposition 2.6. Let E be an archimedean, relatively uniformly com- plete c.v.l. and let F be an order complete c.v.l. Then every order bounded complex linear operator T :E → F has a linear modulus |T| that is defined for each x∈E+ by

|T|x= sup{m2(T(y)); m1(y)6x}= sup{Re(λT)(x); λ∈C, |λ|= 1}

(|T|is additive on E+ and can be extended as a positive linear operator from E to F).

Definition 2.5. A complex locally convex lattice of (M)-type (or of (AM)- type) is an archimedian relatively uniformy complete c.v.l. endowed with a linear topology such that there exists a neighborhood basis of the origin con- sisting of sublattices which are solid and convex.

3. ANTISYMMETRIC IDEALS

In the sequel, E denotes a complex, locally convex lattice of (M)-type.

The center Z(E) of E is the algebra of all order bounded endomorphisms on E, that is, those U ∈ L(E, E) for which there exists λU > 0 such that m(U(x))6λUm(x) for allx∈E. The real part of the center is

ReZ(E) =Z(E)+−Z(E)+.

Definition 3.1. For every closed ideal I of E and every U ∈ Z(E), we define UI :E/I →E/I by

(3.1) UII(x)) =πI(U(x)), x∈E.

It is easily seen that the operator UI is well defined and that UI ∈ Z(E/I).

Definition 3.2. Let I and J be two closed ideals of E such that I ⊂J. Define the mappings

πIJ :E/I →E/J, πIJI(x)] =πJ(x), x∈E,

(4)

and

MIJ :Z(E/I)→Z(E/J), MIJ(T)(πJ(x) =πIJ(T(πI(x))), T ∈Z(E/I).

Lemma3.1. The mappings πIJ and MIJ have the properties below:

1)πIJ is a lattice homomorphism;

2)MIJ(T)∈Z(E/J) for every T ∈Z(E/I);

3)MIJ(T)∈Z(E/J)+ for every T ∈Z(E/I)+. Proof. 1) For every x∈E we have

mbJIJI(x))) =mbJJ(x)) =πJ(m(x)) =πIJI(m(x))) =πIJ(mbII(x))), hence

(3.2) mbJIJI(x))) =πIJ(mbII(x))).

2) For everyT ∈Z(E/I) we have

mbJ(MIJ(T))(πJ(x)) =mbJIJ(T(πI(x)))) =πIJ(mbI(T(πI(x))) 6πIJTmbII(x))) =λTπIJ(mbII(x)))

TπIJI(m(x))) =λTπJ(m(x)) =λTmbJJ(x)), hence MIJ(T)∈Z(E/J).

3) If πJ(x)>0 then πJ(x) = mcJJ(x)] =πJ[m(x)]. Further, for every T ∈Z(E/I)+ we have

mbJ[MIJ(T)(πJ(x))] =mbJ[MIJ(T)(πJ(m(x)))] =mbJIJ(T(πI(m(x))))]

IJ[mbI(T(πI(m(x))))] =πIJ[T(πI(m(x)))]

=MIJ(T)(πJ(m(x))) =MIJ(T)(πJ(x)), hence MIJ(T)>0.

Definition 3.3. Let A be a subset of Z(E) containing 0, and let F be a complex vector subspace ofE. A closed idealIofEis said to be antisymmetric with respect to the pair (A, F) (or (A, F)-antisymmetric) if for every U ∈A with the properties

i)UI ∈ReZ(E/I);

ii) UII(F)]⊂πI(F),

there exists α∈Csuch thatUI =α1E/I, where1E/I is the identity operator on E/I. Of course, E itself is antisymmetric with respect to any pair (A, F).

Further, we denote by AA,F(E) the family of all (A, F)-antisymmetric ideals of E.

Now, we consider the particular caseE=C(X,C), whereXis a compact Hausdorff space. Then the center of C(X,C) is C(X,C) itself, namely, if U :C(X,C) → C(X,C) belongs to the center of C(X,C), then there exists

(5)

a function aU ∈ C(X,C) such that U(f)(x) = aU(x)f(x), ∀f ∈ C(X,C),

∀x∈X.

The mapping U → aU : Z[C(X,C)] → C(X,C) is an isomorphism of Banach algebras, hence Z[C(X,C)]∼=C(X,C).

On the other hand, it is known that for every closed subset S of X, the set IS={f ∈C(X,C); f|S = 0} is a closed ideal ofC(X,C), and every closed ideal ofC(X,C) has this form.

Definition 3.4. LetAbe a subset ofC(X,C) containing 0, and letF be a complex vector subspace of C(X,C). A closed subset S of X is said to be (A, F)-antisymmetric, if everya∈A, such that

i)a(x)∈R,∀x∈S;

ii) af|S ∈F|S,∀f ∈F, is constant on S.

Remark 3.1. A closed subsetSofXis (A, F)-antisymmetric if and only if the corresponding idealIS={f ∈C(X,C); f|S= 0}is (A, F)-antisymmetric in the sense of Definition 3.3. Indeed, it is sufficient to observe that πIS(g) = g|S , for everyg∈C(X,C).

Remark 3.2. IfF =Ais a subalgebra ofC(X,C), then a closed subsetS ofX is antisymmetric with respect toAif everya∈Athat is real valued onS is constant on S. Thus, we retrieve the original concept of antisymmetric set with respect to a subalgebra of C(X,C) introduced by E. Bishop. (See [1].)

Remark 3.3. In the context of weighted spaces, the concept of antisym- metric set was introduced by J.B. Prolla in 1971.

Remark 3.4. If E is a real locally convex lattice of (M)-type, then Defi- nition 3.3 coincides with Definition 3.1 of [3].

Theorem3.1. Let E be a complex locally convex lattice of (M)-type, A a subset of the center Z(E) containing 0, and F a complex vector subspace of E. If (Iα) is a family of (A, F)-antisymmetric ideals of E such that J =P

α

Iα6=E, then I =T

α

Iα is an (A, F)-antisymmetric ideal.

Proof. Let U ∈A such that i) UI ∈ReZ(E/I);

ii) UII(F)]⊂πI(F).

According to Lemma 3.1, MIIα(UI)∈ReZ(E/Iα). Further, we have MIIα(UI)(πIα(F)) =πIIα[UII(F))]⊂πIIαI(F)] =πIα(F).

Since Iα is (A, F)-antisymmetric, we deduce that there isaα ∈Csuch that (3.3) MIIα(UI) =aα ·1E/Iα.

(6)

On the other hand, we have

MIIα(UI)(πIα(x)) =πIIα[UII(x))] =πIIαI(U(x))]

Iα[U(x)] =UIαIα(x)], ∀x∈E, hence

(3.4) MIIα(UI) =UIα.

It follows from (3.3) and (3.4) that

(3.5) UIα =aα ·1E/Iα.

Now, it follows from the commutative diagram ReZ(E/I) M−→IIα ReZ(E/Iα)

MIJ& .MIαJ ReZ(E/J)

that

(3.6) MIJ(UI) =MIαJ(MIIα(UI)) =MIαJ(UIα) =aαMIαJ(1E/Iα).

Next, we have

(3.7) MIαJ(1E/Iα)(πJ(x)) =πIαJ(1E/Iα)(πIα(x)) =

IαJIα(x)) =πJ(x) =1E/JJ(x)).

From (3.6) and (3.7) we deduce that

(3.8) MIJ(UI) =aα·1E/J, ∀α.

Since 1E/J is not constant on E/J (from the hypothesis E 6= J, hence E/J 6={0}), we have aα =a (constant). Therefore, from (3.4) and (3.5) we deduce that

(3.9) UIα =a·1E/Iα =MIIα(UI).

Now, from (3.9) and from the fact that MIIα(1E/I) = 1E/Iα, we de- duce that

(3.10) MIIα(UI−a·1E/I) = 0, ∀α.

SinceMIIα(UI−a·1E/I)(πIα(x)) =πIα(U(x)−ax), ∀x∈I, ∀α we have U(x)−ax∈Iα, ∀α, hence

(3.11) U(x)−ax∈I, ∀x∈E.

From (3.11) we deduce that UI =a·1E/I, thus I is an (A, F)-antisymmetric ideal.

Corollary 3.1. Every (A, F)-antisymmetric ideal contains a unique minimal (A, F)-antisymmetric ideal.

(7)

Proof. Let I ∈ AA,F(E) be such that I 6= E and let Ie= T

{J|J ∈ AA,F(E), J ⊂I}.According to Theorem 3.1, Ie∈ AA,F(E). Obviously,Ie⊂I and Ieis minimal.

4. A GENERALIZATION OF DE BRANGE’S LEMMA We start with two remarks.

Remark 4.1. Let E be a complex vector lattice and let U :E → E be a positive linear operator. Then m(U(x)) 6 U(m(x)), ∀x ∈ E. Indeed, if x∈ER thenU(x)∈ER and

m(U(x)) =U(x)∨ −U(x)6U(m(x)).

For an arbitrary x∈E we have m(U(x)) = sup

a∈C

|a|=1

ReaU(x) = sup

θ∈[0,2π]

(cosθReU(x)−sinθImU(x))

= sup

θ∈[0,2π]

(cosθU(Rex)−sinθU(Imx)) = sup

a∈C

|a|=1

U(Re(ax)),

hence m(U(x))6U(m(x)).

Remark 4.2. If g : E → C is an order bounded linear functional and U :E →E is a positive linear operator, then|g(U(x))|6|g|(U(m(x))).This is a consequence of Remark 4.1 and of the formula

|g|(U(m(x))) = sup{|g(z)|; m(z)6U(m(x))}.

Lemma 4.1. Let E be a complex locally convex lattice of (M)-type, A a subset of Z(E) containing 0, F a vector subspace of E and V a con- vex and solid neighborhood of the origin of E, which also is a sublattice. If f ∈ Ext

V0∩F0 and I = {x∈E; |f|(m(x)) = 0}, then I is an (A, F)- antisymmetric ideal.

Proof. Let U ∈Asuch that i)UI ∈ReZ(E/I);

ii) UII(F)]⊂πI(F).

According to Definition 3.3, we must prove that there exists a∈Csuch that UI=a·1E/I.AsUI ∈ReZ(E/I), we can suppose that

(4.1) 06UI 61E/I.

(8)

Indeed, let S, T ∈ Z(E/I)+ such that UI =S−T. SinceS ∈ Z(E/I)+, one can find λS>0 with the propertym[S(πb I(x))]6λSm(πb I(x)), ∀x∈E,hence (4.2) S(πI(x))6λSπI(x), ∀πI(x)>0.

If we replace S by λ1

SS, we get 0 6 S 6 1E/I. On the other hand, it is clear that S[πI(F)] ⊂ πI(F). Now, if we prove that S = α·1E/I and T = β ·1E/I, then it will follow that UI = a·1E/I, a ∈ C; so that our supposition (4.1) is completely justified. Further, let UI ∈ ReZ(E/I)+ with the properties 0 6UI 61E/I andUII(F)]⊂πI(F). Asf ∈I0, there exists g∈(E/I)0 such thatf =πI0g. Obviously,g∈Ext

πI(V)0∩πI(F)0 . Denote g1 =UI0g, g2= (1E/I−UI)0g and

ai= inf

λ >0|gi∈λ[πI(V)]0 = sup{|giI(x))|; x∈V}, i= 1,2.

Since g=g1+g2 ∈(a1+a2I(V)0,we get

(4.3) a1+a2 >1.

On the other hand, for everyx1, x2∈V we have

|g1I(x1))|+|g2I(x2))|=|g(UII(x1)))|+

g(1E/I−UI)(πI(x2)) . If we denote y =mcII(x1))∨mcII(x2)), and take into account that πI(V) is a solid set and a sublattice, we get y ∈πI(V).

Now, according to Remark 4.2 we have

|g1I(x1))|+|g2I(x2))|6|g|(UI(y)) +|g|(1E/I−UI)(y)

=|g|(UI(y) +y−UI(y)) =|g|(y)61.

Therefore,|g1I(x1))|+|g2I(x2))|61, for anyx1, x2∈V, hencea1+a261.

From (4.3) we deduce that

(4.4) a1+a2 = 1.

Let us note that

(4.5)

UI0g

=UI0(|g|).

Indeed, as 0 6 UI0 6 1(E/I)0, we have UI0(g+) = (UI0g)+ and UI0(g) = (UI0g) thus,

UI0g

=UI0(g+) +UI0(g) =UI0(|g|).

Further, from (4.5) and Remark (4.1) we have (4.6) |f|(m(U(x)))6|f|(U(m(x))) =

πI0g

(U(m(x)))

I0 |g|(U(m(x))) =|g|(UII(m(x)))) = UI0g

I(m(x))).

If we now suppose that g1= 0, then it follows from (4.6) that|f|(m(U(x))) = 0, ∀x ∈ E, hence UII(x)] = πI[U(x)] = 0, ∀x ∈ E. Thus, g1 = 0 implies

(9)

UI = 0 = 0·1E/I and the proof is complete. Using a similar argument, we deduce thatg2 = 0 implies 1E/I−UI = 0, hence UI =1E/I.

Therefore, from now on we can suppose that gi 6= 0 for i = 1,2; thus ai > 0 for i = 1,2. On the other hand, it follows from the assumption UII(F)] ⊂ πI(F), that gi ∈πI(F)0, i= 1,2, hence gai

i ∈ πI(F)0∩πI(V)0, for i = 1,2. Since g is an extremal element of the set πI(F)0∩πI(V)0, and g = a1ag1

1 +a2ga2

2, either g = ag1

1 or g = ag2

2. On the other hand, the equality g1 = a1g implies (a1 ·1E/I−UI)0g = 0, hence UI = a1 ·1E/I. By a similar argument, g2 =a2g involvesUI =a2·1E/I and this concludes the proof.

5. AN APPROXIMATION THEOREM The main result of the paper is as follows.

Theorem5.1. Let E be a complex locally convex lattice of (M)-type, A a subset of Z(E) containing 0, and F a vector subspace of E. If we denote by AeF,A the family of all minimal (A, F)-antisymmetric ideals of E, then for every x∈E we have

x∈F ⇐⇒ πI(x)∈πI(F), ∀I ∈AeF,A.

Proof. The necessity is clear. Conversely, if we suppose that πI(x) ∈ πI(F), ∀I ∈ AeF,A and x /∈ F, then we can find f ∈ E0 such that f(x) 6= 0 and f(y) = 0, ∀y ∈F. Let V be a solid, convex neighborhood of the origin of E which is a sublatice of E . Obviously we can suppose thatf ∈V0.Since f ∈ F0, we deduce that f ∈ F0 ∩V0 and by the Krein-Milman theorem we may assume that f ∈Ext

F0∩V0 .If we putJ ={z∈E; |f|(m(z)) = 0}

then, according to Lemma 4.1, that J is an (A, F)-antisymmetric ideal. On the other hand, by Corollary 3.1, there exists a minimal (A, F)-antisymmetric ideal J0 ⊂J. Since f ∈J00∩F0 and by the hypothesis πJ0(x) ∈πJ0(F), we have f(x) = 0, and this contradicts the choice off.

Remark 5.1. LetE, AandF be as in Definition 3.3, with the additional hypothesis AF ⊂F. Then a closed ideal I ofE is antisymmetric with respect to (A, F) (or A-antisymmetric) if for every U ∈A for whichUI ∈ReZ(E/I) there exists α∈Csuch thatUI=α·1E/I.

Let us denote byAeA the family of all minimalA-antisymmetric ideal of E. From Theorem 5.1 we deduce

Theorem5.2. Let E, Aand F as in Theorem5.1, with the additional hypothesis AF ⊂F. Then for any x∈E we have

x∈F iff πI(x)∈πI(F), ∀I ∈AeA.

(10)

Remark 5.2. Theorem 5.2 is a generalization of Prolla’s approximation theorem for weighted spaces (see Theorem 6.2 below).

Remark 5.3. IfEis a real locally convex lattice of (M)-type, Theorem 5.1 coincides with Theorem 3.1 of [3].

Remark 5.4. We recall that ifJ is a maximal ideal of a real vector lattice E, then E/J has dimension 1. (See [7], pp. 66.)

On the other hand, if I is an ideal in a complex vector lattice E and J =I∩ER, thenJ is an ideal in the real latticeERandI =J+ iJ. Moreover, ifI is maximal in E,thenJ is maximal in ER, hence E/I has dimension 2.

The next result is an abstract version of the Stone-Weierstrass theorem for locally convex lattices.

Theorem5.3. Let E, Aand F as in Theorem5.1, with the additional properties:

i)Ais a (complex) selfadjoint vector subspace of Z(E) (i.e. U = ReU+ i ImU ∈A implies U = ReU −i ImU ∈A).

ii) AF ⊂F;

iii) F is not included in any maximal ideal of E;

iv) every closed ideal I of E such that {UI|U ∈A} ⊂ C·1E/I is a maximal ideal.

Then F =E.

Proof. Let x ∈ E and I ∈ AeF,A. Hypothesis i) implies that ReU ∈ A and ImU ∈A for every U ∈A. It follows from ii) that ReUI =a·1E/I and ImUI =b·1E/I, for somea∈R and b∈R, henceUI= (a+ ib)·1E/I.

According to hypothesis iv),I is a maximal ideal, thus,E/I has dimen- sion 2.

As{0} ⊂πI(F)⊂πI(E) =E/I,we have eitherπI(F) ={0}orπI(F) = πI(E). It follows from (iii) thatπI(F)6={0}. Therefore,πI(F) =πI(E), thus πI(x)∈πI(F) for everyI ∈AeF,A. According to Theorem 5.1, we have x∈F.

6. SOME APPROXIMATION THEOREMS FOR WEIGHTED SPACES

The aim of this section is to give some applications of Theorem 5.1.

LetX be a locally compact Hausdorff space andV a Nachbin family on X, i.e., a set of nonnegative upper semicontinuous functions on X such that for every v1, v2∈V and every λ >0 there isv∈V such thatv1, v26λv. We shall denote the corresponding weighted space by

CV0(X,C) ={f ∈C(X,C); f v vanishes at infinity, ∀v∈V}.

(11)

The weighted topology on CV0(X,C) is denoted by ωV and is determined by the seminorms {pv}v∈V, where

pv(f) = sup{|f(x)|v(x); x∈X} for any f ∈CV0(X,C).

Obviously, CV0(X,C) is a complex locally convex lattice of (M)-type with respect to the weighted topology and to the modulus mapping m(f) = |f|,

∀f ∈CV0(X,C).

Remark 6.1. For every closed idealI of CV0(X,C) there exists a closed subset SI of X such that

I ={f ∈CV0(X,C); f|SI = 0}.

Indeed, if we putJ =I∩CV0(X,R), thenJ is a closed ideal ofCV0(X,R) and according to a result of Portenier (see [2], Lemma 3.8), there is a closed subset SI of X such that I = {f ∈CV0(X,R); f|SI = 0}. On the other hand, we have I =J + iJ, hence I ={f ∈CV0(X,C); f|SI = 0}.

Remark 6.2. The center of CV0(X,C) is Cb(X,C) – the Banach space of all bounded continuous complex valued function on X. Indeed, it follows from Proposition 2 of [6] that if U belongs to the center of CV0(X,C). Then there exists a bounded continuous function aU :X→Csuch that

U(f)(x) =aU(x)f(x), ∀f ∈CV0(X,C), ∀x∈X.

The mapping U → aU : Z(CV0(X,C)) → Cb(X,C) is an isomorphism of Banach algebras.

Definition 6.1. Let A be a subset of Cb(X,C) containing 0, and let F be a vector subspace of CV0(X,C). A closed subset S of X is said to be (A, F)-antisymmetric if everya∈A, real valued onS, such thataf|S ∈F|S,

∀f ∈F, is constant onS.

Clearly, a closed subsetSofX is (A, F)-antisymmetric if and only if the corresponding ideal IS ={f ∈CV0(X,C); f|S = 0} is (A, F)-antisymmetric in the sense of Definition 3.3.

We shall denote by SF,A the family of all (A, F)-antisymmetric subsets of X. Theorem 3.1 has the following dual version.

Theorem6.1. Let A and F be as in Definition 6.1. If Sα ∈ SF,A, ∀α and T

α

Sα 6=∅, then S

α

Sα ∈ SF,A.

Corollary6.1. For every x ∈X there exists a maximal closed subset Sx of X, antisymmetric with respect to the pair (A, F), such that x ∈ Sx. Moreover, if x6=y then Sx and Sy are either equal or disjoint.

See Theorem 3 of [5].

From Theorem 5.1 we deduce

(12)

Theorem6.2. Let A be a subset of Cb(X,C) containing 0and let F be a vector subspace of CV0(X,C). Then, for every f ∈CV0(X,C), we have

f ∈F if and only if f|Sx ∈F|Sx, ∀x∈X,

where (Sx)x∈X is the family of all maximal (A, F)-antisymmetric subsets of X.

Remark 6.3. Theorem 6.2 coincides with Corollary 1 of [5].

Remark6.4. IfAF ⊂F, then a closed subsetSofXis (A, F)-antisymmetric if any a∈A, real valued onS, is constant onS.

We shall denote bySeA the family of all maximal antisymmetric subsets of X.

From Theorem 6.1 we deduce

Theorem6.3 (Prolla). Let E, A and F be as in the Theorem 6.1, with the additional property AF ⊂F. Then, for every f ∈CV0(X,C), we have

f ∈F ifff|S ∈F|S , ∀S∈SeA.

Nachbin’s density theorem below follows from Theorem 5.3.

Theorem6.4. Let A be a self adjoint subalgebra of Cb(X,C) and let F be a vector subspace of CV0(X,C). Assume that

i)AF ⊂F;

ii) Aseparates the points of X;

iii) for every x∈X there is f ∈F such that f(x)6= 0.

Then F =CV0(X).

Proof. It follows from (iii) thatF is not included in any maximal ideal.

Since AF ⊂ F and A separates the points of X, every (A, F)-antisymmetric subsetSofXis a singleton, thus the corresponding idealIS is a maximal ideal.

So, the hypotheses of Theorem 5.3 are fulfilled. Consequently, F = CV0(X).

REFERENCES

[1] E. Bishop, A generalization of the Stone-Weierstrass theorem. Pacific J. Math. 11 (1961), 777–783.

[2] A. Goullet de Rugy, Espaces de fonctions ponderables. Israel J. Math. 12 (1972), 147–160.

[3] D. Kravvaritis and G. P˘altineanu,A density theorem for locally convex lattices.Abstract and Applied Analysis5(2004), 387–393.

[4] W.A. Luxemburg and A.C. Zaanen, The linear modulus of an order bounded linear transformation.I. Indag. Math.33(1971), 422–434.

(13)

[5] G. P˘altineanu,A generalization of the Stone-Weierstrass theorem for weighted spaces.

Rev. Roumaine Math. Pures Appl.7(1978), 1065–1068.

[6] G. P˘altineanu and D.T. Vuza,A generalization of the Bishop approximation theorem for locally convex lattices of(AM)-type. Rend. Circ. Mat. Palermo (2) Suppl. No. 52, Vol. II (1998), 687–694.

[7] H.H. Schaefer, Banach Lattices and Positive Operators. Springer-Verlag, Berlin–

Heidelberg–New York, 1974.

[8] W.J. De Schipper,A note on the modulus of an order bounded linear operator between complex vector lattices.Indag. Math.41(1979), 38–53.

[9] D.T. Vuza,Elements of the theory of modules over ordered rings.In: Order Structures in Functional Analysis, Vol. 2, pp. 175–283. Ed. Academiei Romˆane, Bucure¸sti, 1989.

(Romanian)

Received 17 July 2007 National Technical University of Athens Department of Mathematics TK 157-80, Zografou Campus

Athens, Greece [email protected]

and Technical University of Civil Engineering Bucharest

Department of Mathematics Bl. Lacul Tei 124, Sector 2 020396, Bucharest 38, Romania

[email protected]

Références

Documents relatifs

We observe that, in the case when X is a linear space and the maps u 1 , u 2 are linear, Corollary 3.4 extends to the locally convex spaces an im- portant theorem which was proved

and ~et A carry the Allan boundedness. In this section we restrict our interest to commutative topolo- gical algebras endowed with a Fréchet topology and to the

— It is an immediate consequence of the former theorem and from the fact that every barrelled space whose completion is Baire is a Baire-like, [6].. — It is immediate that our Theorem

Conversely, given a Kolmogoroff linear topological space, there exists a topologically equivalent system of neighborhoods (s) such that the neighborhoods U of o satisfy v. Hence we

We define in §3 a set of equivariant elementary surgeries on the links, and prove in §4 (Theorem 4.8) that performing a flip on a simple convex polytope means performing an

We can push the technique much farther and it is quite possible that one can use techniques relating affine geometry and complex analysis to complete the proof,

tially complete, infrabarreled locally convex space is barreled. The purpose of this note is to prove the corresponding result in the polynomial context ; that is, we

L’accès aux archives de la revue « Annali della Scuola Normale Superiore di Pisa, Classe di Scienze » ( http://www.sns.it/it/edizioni/riviste/annaliscienze/ ) implique l’accord