• 検索結果がありません。

Strong subdifferentiability of norms and geometry of Banach spaces

N/A
N/A
Protected

Academic year: 2022

シェア "Strong subdifferentiability of norms and geometry of Banach spaces"

Copied!
10
0
0

読み込み中.... (全文を見る)

全文

(1)

Strong subdifferentiability of norms and geometry of Banach spaces

G. Godefroy, V. Montesinos, V. Zizler

Dedicated to the memory of Josef Kolom´y

Abstract. The strong subdifferentiability of norms (i.e. one-sided differentiability uni- form in directions) is studied in connection with some structural properties of Banach spaces. It is shown that every separable Banach space with nonseparable dual admits a norm that is nowhere strongly subdifferentiable except at the origin. On the other hand, every Banach space with a strongly subdifferentiable norm is Asplund.

Keywords: strong subdifferentiability of norms, Asplund spaces, renormings, weak com- pact generating

Classification: 46B03

1. Introduction

Let X be a real Banach space and x ∈ X. A norm k · k on X is said to be strongly subdifferentiable(SSD)at xif the one-sided limit

t→0lim+

1

t (kx+thk − kxk)

exists uniformly onhin the unit sphereSX ofX ([8], [4], [5], [1], [6]). If a norm k · k on X is strongly subdifferentiable at every point of X we say that k · k is an SSD norm on X. Similarly, if k · k is Gˆateaux or Fr´echet differentiable on X\{0} we say thatk · kis a Gˆateaux orFr´echet differentiable normon X. Note that a normk · k on a Banach spaceX is Fr´echet differentiable at x∈X if and only ifk · kis both Gˆateaux differentiable and SSD atx. From the monotonicity of the differentiation quotient for convex functions and from the classical Dini theorem it follows that every norm on a finite dimensional space is SSD at every point of the space. It can be shown (see e.g. [4]) that the canonical sup-norm k · k on ℓ is SSD at x = {xn} if and only if kxk ∈ {|x/ n|;|xn| 6= kxk}, where {·} denotes the derived set of {·}. Therefore the sup-norm onc0 is SSD at every point of c0. On the other hand, we show in Proposition 5 below that

Supported by NSERC (Canada), Generalitat Valencia and DGICYT PB91-0326 and PB91-0538 (Spain)

(2)

the set S of all SSD points of the sup-norm of the space ℓ is not a Gδ set in ℓ(although Scontains a denseGδ set of points of Fr´echet differentiability (see e.g. [2, p. 26, 29])). It is known that various classes of Banach spaces can be characterized by various differentiability properties of norms (see e.g. [2]). The class of separable Banach spaces with nonseparable dual is strictly larger than the class of separable Banach spaces with an equivalent Gˆateaux differentiable norm which is nowhere Fr´echet differentiable (see e.g. [2, p. 101, 104]). In the main result in this paper, Theorem 1, the separable spaces with nonseparable dual are characterized as those separable Banach spaces that admit norms which are nowhere strongly subdifferentiable except at the origin. In the third part of this paper some applications of the strong subdifferentiability of norms in the geometry of Banach spaces are shown. For example, it is proved that a Banach space with an SSD norm is Asplund.

We use a standard notation in this note. We denote by BX the closed unit ball ofX and bySX its unit sphere. Forx∈SX we let

J(x) ={x∈SX; x(x) = 1}.

We will use the following analogue of the ˇSmulyan lemma: The normk · kis SSD atx∈SX if and only if for everyε >0 there isδ >0 such that dist x, J(x)

< ε wheneverx∈BX is such thatx(x)>1−δ(see [5]). We recall that a subspace Y of the dual Banach space (X,k · k) is called 1-norming if for everyx∈X we have

kxk= sup{x(x); x∈BX∩Y}.

Theball topologybX on a Banach spaceXis the weakest topology onX in which the ballsB(x, δ) ={y∈X; ky−xk ≤δ},x∈X andδ >0, are closed. We refer to [7] for more on this subject. Furthermore, recall that a biorthogonal system {xα, xα}α∈Λ ⊂X ×X is a shrinking Markuˇseviˇc basis for a Banach space X if span{xα} = X and spank·k{xα} = X. Note that the set {xα} ∪ {0} is then a weak compact set that generates X. We refer to [14] and [17] for more on Markuˇseviˇc bases and to [2] for the definition and basic properties of the projectional resolutions of the identity on nonseparable Banach spaces as well as for some unexplained notions and results used in this paper.

2. SSD characterization of separable Asplund spaces The following theorem is a main result in this paper.

Theorem 1. Any separable Banach space with nonseparable dual admits an equivalent norm that is nowhere strongly subdifferentiable except at the origin.

Proof: If a separable Banach spaceX contains an isomorphic copy ofℓ1, then X admits an equivalent Gˆateaux differentiable norm that is nowhere Fr´echet differentiable ([3], see also [2, p. 101, 104]). Such a norm is nowhere SSD except at the origin. We will therefore assume thatX is a separable Banach space that

(3)

does not contain an isomorphic copy of ℓ1 and such that X is nonseparable.

Without loss of generality we assume that the original norm| · |ofX is Gˆateaux differentiable (see e.g. [2, Theorem II.3.1]).

There is a subsetK:={xε, ε∈ {0,1}N} ofSX that isw-homeomorphic to the Cantor set in [0,1] and a subset K∗∗ :={x∗∗ε ; ε∈ {0,1}N} of 54BX∗∗ such that x∗∗ε (xγ) =δεγ (the Kronecker delta) for ε, γ ∈ {0,1}N ([18]). Denote by L:=K∪(−K), C:= convw(L) andB :=BX+C, whereBXis the dual unit ball for the norm| · |onX. Clearly,Bis the dual unit ball for an equivalent normk · konX. Choose an arbitrary elementx∈K and letH:={x}⊂X. Using the fact thatX contains a countable separating set forX we let{Hn} be a sequence of closed hyperplanes inX with H1 =H and such that∩Hn={0}.

Furthermore, we letT1be the identity map onXand forn= 2,3, . . . we letTnbe an isomorphism ofX onto itself such thatTn(Hn) =H. We assume without loss of generality that the norms ofTnare uniformly bounded. We let {Mn} be the collection of all elementary clopen (i.e. closed and open) subsets ofK or−K.

Forn∈Nwe letk · knbe a seminorm defined on X by kxkn= sup{fn(t)hx, ti: t∈Mn}, x∈X,

wherefn∈C(Mn) is such that 0≤fnand supMnfn= 1 is attained at a unique pointan∈Mn.

Finally, we define an equivalent norm||| · |||onX by

|||x|||=

X

n=1

2−nkTnxk+

X

n,m=1

2−(n+m)kTnxkm.

We will show that||| · ||| is SSD only at the origin. In doing so, we will use the following fact that directly follows from the monotonicity of the differentiation quotient for convex functions: Ifϕ1 and ϕ2 are two convex functions andϕ1 is not SSD atx, thenϕ12 is not SSD atx.

Hence to finish the proof we only need to show that givenx∈X\{0}, at least one of the component norms or seminorms in the definition of||| · ||| is not SSD atx.

Assume first thatx /∈H and let the face ofC determined byxbe denoted by F(C, x), i.e.

F(C, x) ={x ∈C; x(x) = sup{y(x); y∈C}}.

x is not identically zero on K, hence sup {y(x); y ∈ C} > 0. Note that ExtF(C, x)⊂ ExtC⊂Lby Milman’s theorem, where ExtC denotes the set of all extreme points ofC. Thus,L∩F(C, x) is a nonemptyw-closed subset ofL.

We consider two cases: first assume that x is constant on no set Mn that intersects L∩F(C, x). Then, using the fact that x is a positive constant on

(4)

F(C, x), we can find a sequence {xn} ⊂L\F(C, x) that isw-convergent to an x ∈ L∩F(C, x). We may and do assume that the sequence {xn} and x are inK. We show that

| · |− dist xn, F(C, x)

k · k− dist xn, F(C, x)

25 .

To see this, denoting the corresponding biorthogonal functionals to{xn}by{x∗∗n } we have

x∗∗n(xn−e)≥1 for everye∈ ExtF(C, x).

Since X does not contain an isomorphic copy of ℓ1, every x∗∗ ∈ X∗∗ is the w-limit of a sequence in X ([15]) hence it satisfies the barycentric calculus and thus, by the Lebesgue Dominated Convergence Theorem, we have

x∗∗n(xn−x)≥1 for everyx∈F(C, x).

Hence for everyx∈F(C, x) and everyn∈Nwe have

kxn−xk≥2−1|xn−x| ≥2−1·45 x∗∗n (xn−x)≥ 25 .

Note that the Gˆateaux differentiability of the norm| · |implies thatF(BX, x) is a singleton, sayu0. Therefore

F(B, x) =F(BX, x) +F(C, x) =u0+F(C, x).

Hence for alln∈Nwe have

k · k−dist u0+xn, F(B, x)

=k · k−dist u0+xn, u0+F(C, x)

=k · k−dist xn, F(C, x)

25 . We haveku0+xnk≤1 for alln∈Nand

lim (u0+xn)(x) = (u0+x)(x)

= sup {x(x); x ∈BX}+ sup{x(x); x∈C}

=kxk= 1.

By the variant of ˇSmulyan’s lemma stated in Introduction, we have that k · k is not SSD atx.

Second, assumex /∈Hand there is ann∈Nsuch thatMnintersectsL∩F(C, x) andxis a (positive) constant onMn. We shall show thatk·knis not SSD atx. We assume without loss of generality thatkxkn= 1. We observe that the seminorm

(5)

k · kn is Gˆateaux differentiable at x(see e.g. [2, p. 5]). Therefore to show that k · knis not SSD atxis equivalent to show thatk · knis not Fr´echet differentiable atx. To see this we lettk andskbe points ofMndifferent fromansuch that (as elements in [0,1])tk< sk and

min{fn(tk), fn(sk)}>1−k−2 for allk∈N.

Furthermore, fork∈N, we letx∗∗k ∈X∗∗ be defined by x∗∗k =x∗∗t

k −x∗∗s

k, where x∗∗t

k and x∗∗s

k denote now the corresponding biorthogonal functionals in the set K∗∗. From Goldstine’s theorem, for k ∈ N we find yk ∈ X such that kykk ≤ kx∗∗k k and

hyk, tk−ski>hx∗∗k , tk−ski −1 2

=3 2

. We thus have

x+1

kyk

n≥fn(tk)D x+1

kyk, tkE

>

1− 1 k2

1 + 1 k

yk, tk ,

x−1

kyk

n≥fn(sk)D x−1

kyk, skE

>

1− 1 k2

1− 1 k

yk, sk . So

k

x+1

kyk

n+

x−1

kyk

n−2kxkn

>

> k 1− 1

k2

2 + 1 k

Dtk−sk, yk

E−2k >

>

k−1 k

2 + 3 2k

−2k=3 2 −2

k− 3 2k2 ≥ 1

8

for allk= 2,3, . . .. Hencek · knis not Fr´echet differentiable atxand this finishes the proof ifx /∈H.

In general, given x ∈ X\{0}, there is n ∈ N such that x /∈ Hn and thus Tnx /∈H. By the argument above, one of the norms or seminormsk · kork · km

is not SSD atTnx. Therefore the norm||| · |||is not SSD atx. This concludes the

proof of Theorem 1.

Note that the statement in Theorem 1 actually provides for a characterization of separable spaces with nonseparable dual (cf. e.g. [2, Theorem I.5.7]).

(6)

3. Applications of the strong subdifferentiability

The following result shows a few applications of the strong subdifferentiability in the geometry of Banach spaces.

Theorem 2. Let the normk·kof a Banach spaceXbe strongly subdifferentiable.

Then

(i) The spaceX is an Asplund space (see[5], [1], [6]).

(ii) IfX hask · k-norm1projectional resolution of the identity{Pα}α≤µ, then {Pα}α≤µ is a projectional resolution of the identity onX. If moreover X as well as every complemented subspaceY ofX with densY less than densX has ak · k-norm1 projectional resolution of the identity, then X has a shrinking Markuˇseviˇc basis; in particular, this applies to X with densX =ℵ1.

(iii) Any weakly closed bounded subset ofX is an intersection of finite unions of balls in X ([6]).

In the proof of Theorem 2 as well as in many other results on SSD norms, the following result plays a crucial rˆole.

Lemma 3([6]). Let the norm of a Banach spaceX be strongly subdifferentiable.

ThenXcontains no proper norm closed1-norming subspace.

Proof: We give a proof that is slightly different from that in [6]. Assume first that X is separable and k · k is an SSD norm on X. Suppose that N ⊂X is a norm closed 1-norming proper subspace ofX with respect to k · k. Let{xn} be a w-dense sequence in BX ∩N (which is then w-dense in BX as N is 1-norming). Since the normk · k is SSD, from the variant of ˇSmulyan’s lemma discussed in Introduction it follows that

dist {xn}, J(x)

= 0 for everyx∈SX. Let

B=SX∩ ∪nB(xn,12) ,

whereB(xn, 12) denotes the ball of radius 1/2 centered atxn.

ThenB is a subset ofSX on which anyx∈X attains its norm. SinceN is a proper subspace ofX, we can choosex∗∗∈N⊂X∗∗with kx∗∗k∗∗= 1 and x0 ∈BXsuch thatx∗∗(x0)>4/5. From Goldstine’s theorem we find a sequence {xk} ⊂BX such that for alln∈ {0,1,2, . . .} we have

limk xn(xk) =x∗∗(xn).

We may and do assume that x0(xk)>4/5 for all k∈N. By Simons’ inequality ([16], see e.g. [2, Lemma I.3.7]) we have

inf

kyk; y∈ conv{xk} ≤ sup

x∈B

{limsupx(xk)}.

(7)

Sincex∗∗(xn) = 0 for alln∈Nandkxkk ≤1 for allk∈N, from the definition of the setB it follows that

limsup

k

|x(xk)| ≤ 12

for all x ∈ B. Hence from Simons’ inequality it follows that there is y ∈ conv{xk} such thatkyk <3/4 . However, since x0(xk)>4/5 for every k∈N, we havex0(y)≥4/5. Thuskyk ≥x0(y)≥4/5. This contradiction completes the

proof ifX is separable.

The proof for the general case follows from the first part of this proof and from the following variant of the Mazur separable exhaustion argument.

Lemma 4. LetX be a Banach space. Assume thatXcontains a proper norm- closed 1-norming subspace. Then there is a separable subspace Y of X and a proper norm-closed1-norming subspace of Y.

Proof: LetN be a proper norm-closed 1-norming subspace ofX and letx0 ∈ X be such that dist (x0, N) = 1. For a subspaceG⊂N we say that a subspace F ⊂X isG-good if for everyg∈ span{G, x0}we have

kgk= sup{g(f); f ∈SX∩F}.

Given a subspaceF ⊂ X we say that a subspace H ⊂X is F-norming if for everyf ∈F we have

kfk= sup{h(f); h∈H∩SX}.

We construct a sequence {Gn} of separable subspaces of N and a sequence {Fn} of separable subspaces of X as follows: Choose g1 ∈ N arbitrarily, put G1 = span {g1} and let F1 be a separable subspace of X that is G1-good. If G1, G2, . . . , Gn and F1, F2, . . . , Fn have been chosen, pick a separable subspace Gn+1⊂N such thatGn+1⊃Gnand thatGn+1 isFn-norming and then choose a separable subspaceFn+1 ofX such thatFn+1⊃Fn andFn+1 isGn+1-good.

Put Y :=∪Fn ⊂X and G :=∪Gn k·k ⊂N. Denote by Re the restriction map of X to Y. Then Re is an isometry of a subspace Gof N onto a norm closed subspaceN1:= ReGofY which is clearly 1-norming inY. Ifg∈Gn, then

kRex0− RegkY= sup{|x0(y)−g(y)|; y∈SX ∩Y}

≥ sup{|x0(f)−g(f)|; f ∈SX ∩Fn}

=kx0−gkX≥1.

From this it follows that dist (N1, Re x0) ≥ 1 in Y. Hence N1 is a proper subspace ofY. This completes the proof of Lemma 4 and thus finishes the proof

of Lemma 3.

(8)

Proof of Theorem 2: (i) LetY be a separable subspace ofX and let{xn}be aw-dense sequence inBY. Then the subspaceN = spank·k{xn}is 1-norming and the restriction of the normk · ktoY is SSD. Therefore, by Lemma 3 we have N =Y. This shows thatY is separable. ThereforeX is an Asplund space.

(ii) The only thing we need to prove is that for every limit ordinalγ≤µwe have

α<γ∪ Pα(X)k·k=Pγ(X) (see e.g. [12] for details).

To show the latter we first notice that since Pγx = lim

α→γ Pαx in the w- topology, for all x ∈ X, and that kPαk = 1 for all α ≤ µ, the restriction

α<γ∪ PαX∗ k·k

↾PγX is a 1-norming subspace in (PγX). By Lemma 3 we thus have

α<γ∪ PαX∗ k·k

↿PγX = (PγX). Letx ∈PγX. There exists y ∈ ∪

α <γPαX∗ k·k such that y↿P

γX = x↿P

γX. Then x = Pγx = Pγy = y. We obtain that

α<γ∪ Pα(X)k·k=Pγ(X).

(iii) We only need to show that the ball topologybX coincides on bounded sets with the weak topology ([7]). For this, it is enough to show that ifY is a separable subspace ofX andx ∈X, then the restrictionx↿Y isbY-continuous onBY ([7, Proposition 2.5]). For this, in turn is sufficient to prove that x↿Y belongs to all 1-norming norm closed subspaces ofY ([7, Theorem 2.4]). This is clearly so by Lemma 3. We refer to [6] for a selfcontained proof of (iii).

The following proposition shows that the set of all SSD points of a norm on an Asplund space may in general be a non-Gδ set in the space.

Proposition 5. The set of all SSD points of any given norm on any Banach spaceX is anFσδ set in the spaceX. The set of allSSDpoints of the sup-norm onℓis not aGδ set inℓ.

Proof: Letk · k be the norm of a Banach spaceX. For k, n∈Nwe denote by Fk,n the set of allx∈X such that

sup

h∈SX

{|t1

1 (kx+t1hk − kxk)−t1

2 (kx+t2hk − kxk)|} ≤ k1 whenever 0< t1 ≤t2<1/n.

Fork, n ∈N, the setsFk,n are closed and for the set S of all SSD points of the normk · kwe have

S=∩

k(∪

nFk,n).

LetX beℓendowed with the sup-norm| · |. Define the map Ψ : {0,1}N→X by

Ψ(ε) =

X

i=1

2−iε(i)χ[i,[, ε∈ {0,1}N,

(9)

whereχ[i,[ denotes the characteristic function of [i,∞[ inN. It follows that Ψ(ε) is an SSD point for the norm| · | if and only ifε∈Q, where

Q={ε∈ {0,1}N; card{i; ε(i) = 1}<∞}

(cf. e.g. [4]).

SinceQis a countable dense subset of a perfect compact metric space{0,1}N, Qis not aGδ set in{0,1}N. Moreover, denoting by S the set of SSD points, we have Ψ−1(S) = Qand Ψ is a continuous map of{0,1}N into X. ThereforeS is

not aGδset inX.

We finish this paper with a list of a few open problems in this area.

Problems 6.

(i) Does every Asplund space admit an equivalent SSD norm? In particular, does the spaceC(K) admit an equivalent SSD norm wheneverKis a tree space?

Recall that R. Haydon found an example of a tree such that the space C(K) does not admit any equivalent Gˆateaux differentiable norm. He also found many other connections between properties of trees and renormings ([9], see also [2, Chapter VII]).

(ii) Assume that a Banach space X admits an equivalent Gˆateaux differen- tiable norm and that X admits also an equivalent SSD norm. Does X admit an equivalent Fr´echet differentiable norm?

Recall that S. Troyanski showed thatX admits an equivalent locally uni- formly rotund norm provided thatX admits both strictly convex norms and norms with the Kadec-Klee property ([19], see also [2, Corollary IV.3.6]).

(iii) Assume that the norm of a separable Banach space X has the property that its restriction to every infinite dimensional closed subspaceY ⊂Xhas a point of Fr´echet differentiability onY. Is thenXnecessarily separable?

The statement might be considered as a sort of linearization in the argu- ment in the Baire Great Theorem (cf. the Jayne-Rogers Baire 1 selectors [11], see also [2, Chapter I.4]).

(iv) Assume thatX is separable andX is nonseparable. Does there exist on X an equivalent norm that is “uniformly non SSD” in the sense analogous to the roughness of norms?

Note that the answer to this question is positive if, for instance,Xcontains an isomorphic copy ofℓ1 ([3], see e.g. [2, p. 101 and 104]).

(v) Assume thatX is (in general nonseparable) non Asplund space. DoesX admit an equivalent norm that is nowhere SSD except at the origin?

Acknowledgement. A part of the research for this paper was done while the two first named authors were visiting the Department of Mathematics, University of Alberta in Edmonton. They thank this Department for excellent working conditions and hospitality. Thanks are due also to the referee for his suggestions.

(10)

References

[1] Contreras M.D., Pay´a R.,On upper semicontinuity of duality mapping, Proc. Amer. Math.

Soc.121(1994), 451–459.

[2] Deville R., Godefroy G., Zizler V.,Smoothness and renormings in Banach spaces, Pitman Monographs and Surveys in Pure and Applied Mathematics64(1993).

[3] Deville R., Godefroy G., Hare D., Zizler V.,Differentiability of convex functions and the convex point of continuity property in Banach spaces, Israel J. Math.59(1987), 245–255.

[4] Franchetti C.,Lipschitz maps and the geometry of the unit ball in normed spaces, Archiv.

Math.46(1986), 76–84.

[5] Franchetti C., Pay´a R.,Banach spaces with strongly subdifferentiable norm, Boll. Uni. Mat.

Ital.VII-B(1993), 45–70.

[6] Godefroy G.,Some applications of Simons’ inequality, Seminar of Functional Analysis II, Univ. of Murcia, to appear.

[7] Godefroy G., Kalton N.J.,The ball topology and its applications, Contemporary Math.85 (1989), 195–238.

[8] Gregory D.A.,Upper semicontinuity of subdifferential mappings, Canad. Math. Bull.23 (1980), 11–19.

[9] Haydon R.,Trees in renorming theory, to appear.

[10] James R.C.,Weakly compact sets, Trans. Amer. Math. Soc.113(1964), 129–140.

[11] Jayne J.E., Rogers C.A.,Borel selectors for upper semicontinuous set valued mappings, Acta Math.155(1985), 41–79.

[12] John K., Zizler V.,Smoothness and its equivalents in weakly compactly generated Banach spaces, J. Funct. Anal.15(1974), 161–166.

[13] ,Projections in dual weakly compactly generated Banach spaces, Studia Math.49 (1973), 41–50.

[14] Lindenstrauss J., Tzafriri L.,Classical Banach spaces I, Sequence Spaces, Springer-Verlag, 1977.

[15] Odell E., Rosenthal H.P.,A double dual characterization of separable Banach spaces con- taining1, Israel J. Math.20(1975), 375–384.

[16] Simons S.,A convergence theorem with boundary, Pacific J. Math.40(1972), 703–708.

[17] Singer I.,Bases in Banach Spaces II, Springer-Verlag, 1981.

[18] Stegall C.,The Radon-Nikodym property in conjugate Banach spaces, Trans. Amer. Math.

Soc.206(1975), 213–223.

[19] Troyanski S.L.,On a property of the norm which is close to local uniform convexity, Math.

Ann.27(1985), 305–313.

Equipe d’Analyse Universit´e Paris VI, Tour 46-0, 4 ´Etage, 4 Place Jussieu, 75252 Paris Cedex 05, France

and

Department of Mathematics, University of Missouri-Columbia, Columbia, Missouri 65211, USA

Departamento de Matem´aticas, Universidad Polit´ecnica de Valencia, c/Vera, s/n. 46071 - Valencia, Spain

Department of Mathematics, University of Alberta, Edmonton, Alberta T6G 2G1, Canada

(Received August 2, 1994)

参照

関連したドキュメント

In this paper, we show the triangle inequality and its reverse inequality in quasi- Banach spaces.. Key words and phrases: Triangle inequality,

Abstract: In this paper, we show the triangle inequality and its reverse inequality in quasi- Banach spaces.... Triangle Inequality in Quasi-Banach Spaces Cong Wu and Yongjin

Keywords: nonlinear operator equations, Banach spaces, Halley type method, Ostrowski- Kantorovich convergence theorem, Ostrowski-Kantorovich assumptions, optimal error bound, S-order

We introduce an implicit iterative process for a nonexpansive semigroup and then we prove a weak convergence theorem for the nonexpansive semigroup in a uniformly con- vex Banach

We study a sequence of generalized projections in a reflexive, smooth, and strictly convex Banach space1. Our result shows that Mosco convergence of their ranges implies their

In Section I we shall prove the theorems of integral operators that are bounded and completely continuous in H¨older spaces with weight.. Section II will contain the proof of

[18] Mitchell A.R., Smith Ch., An existence theorem for weak solutions of differential equations in Banach spaces, in Nonlinear Equations in Abstract

Our approach follows essentially the pattern introduced by Filippov [4] and developed by Frankowska [5], Tolstonogov [16], and Papageorgiou [13], however with the basic difference