### January 23th, 2005

2000 Mathematics Subject Classification. 46B20, 47A12.Research partially supported by Spanish MCYT project no. BFM2003-01681 and Junta de Andalucıa grant FQM-185 .
<ph f="ecbx">On the intrinsic and the spatial numerical range</ph>
Miguel Martın,1 Javier Merı, and Rafael Paya
mmartins@ugr.es, jmeri@ugr.es, rpaya@ugr.es

• Abstract. For a bounded function $f$  from the unit sphere of a closed subspace $X$  of a Banach space $Y$  , we study when the closed convex hull of its spatial numerical range $W\left(f\right)$  is equal to its intrinsic numerical range $V\left(f\right)$  . We show that for every infinite-dimensional Banach space $X$  there is a superspace $Y$  and a bounded linear operator $T:X⟶Y$  such that $\overline{co}W\left(T\right)\ne V\left(T\right)$  . We also show that, up to renormig, for every non-reflexive Banach space $Y$  , one can find a closed subspace $X$  and a bounded linear operator $T\in L\left(X,Y\right)$  such that $\overline{co}W\left(T\right)\ne V\left(T\right)$  .
Finally, we introduce a sufficient condition for the closed convex hull of the spatial numerical range to be equal to the intrinsic numerical range, which we call the Bishop-Phelps-Bollobas property, and which is weaker than the uniform smoothness and the finite-dimensionality.
We characterize strong subdifferentiability and uniform smoothness in terms of this property.

1 Introduction

Given a Banach space $Y$  over $\mathbb{K}$  ( $=\mathbb{R}$  or $\mathbb{C}$  ), we write ${B}_{Y}$  for the closed unit ball and ${S}_{Y}$  for the unit sphere of $Y$  . The dual space of $Y$  will be denoted by ${Y}^{*}$  . If $Z$  is another Banach space, we write $L\left(Z,Y\right)$  for the Banach space of all bounded linear operators from $Z$  into $Y$  ; if $Z=Y$  we simply write $L\left(Y\right):=L\left(Y,Y\right)$  to denote the the Banach algebra of all bounded linear operators on $Y$  . For an element $u\in {S}_{Y}$  , we write $D\left(Y,u\right):=\left\{{y}^{*}\in {Y}^{*}:\parallel {y}^{*}\parallel ={y}^{*}\left(u\right)=1\right\},$  the ${w}^{*}$  -closed and convex set of all states of $Y$  relative to $u$  . Let us mention two facts, both consequence of the Hahn-Banach Theorem, which will be relevant to our discussion. On one hand, we have
 $\begin{array}{c}{lim}_{\alpha ↓0}\frac{\parallel u+\alpha y\parallel -1}{\alpha }=max\left\{Re{z}^{*}\left(y\right):{z}^{*}\in D\left(Z,u\right)\right\},\end{array}$ (1)
(see [10,Theorem V.9.5for a proof ). On the other hand, if $X$  is a subspace of $Y$  and $u\in X$  , then $D\left(X,u\right)$  coincides with the restriction to $X$  of the elements of $D\left(Y,u\right)$  .
If $Y$  is a Banach space, by a closed subspace of $Y$  we mean a Banach space $X$  and an inclusion operator $J:X⟶Y$  (i.e., $J$  is a linear isometry), and we also say that $Y$  is a superspace of $X$  . When no confusion is possible, we omit $J$  , but all the definitions below depend on the way that $X$  is a subspace of $Y$  . Let us fix $X$  and $Y$  as above. We write $\Pi \left(X,Y\right)$  to denote the subset of ${S}_{X}×{S}_{{Y}^{*}}$  given by $\Pi \left(X,Y\right):=\left\{\left(x,{y}^{*}\right)\in {S}_{X}×{S}_{{Y}^{*}}:{y}^{*}\in D\left(Y,Jx\right)\right\}.$  If $X=Y$  , we just write $\Pi \left(Y\right):=\Pi \left(Y,Y\right)$  . We denote by $B\left({S}_{X},Y\right)$  the Banach space of all bounded functions from ${S}_{X}$  to $Y$  , endowed with the natural supremum norm, and we write ${C}_{\text{u}}\left({S}_{X},Y\right)$  for its closed subspace consisting of all bounded and uniformly continuous functions. For $f\in B\left({S}_{X},Y\right)$  we can define two different numerical ranges, namely, the spatial numerical range defined as $W\left(f\right):=\left\{{y}^{*}\left(f\left(x\right)\right):\left(x,{y}^{*}\right)\in \Pi \left(X,Y\right)\right\},$  and the intrinsic numerical range given by $V\left(f\right):=\left\{\Phi \left(f\right):\Phi \in D\left(B\left({S}_{X},Y\right),J{|}_{{S}_{X}}\right)\right\}.$  The name of intrinsic numerical range comes from the fact that if $f$  belongs to any closed subspace $Z$  of $B\left({S}_{X},Y\right)$  , we can calculate $V\left(f\right)$  using only elements in ${Z}^{*}$  . These numerical ranges appeared in a paper by L. Harris [16for continuous functions. In the particular case when $X=Y$  and $f$  is (the restriction to ${S}_{Y}$  of ) a bounded linear operator, the spatial numerical range was introduced by F. Bauer (field of values subordinate to a norm [1), extending Toeplitz’s numerical range of matrices [25and, concerning applications, it is equivalent to Lumer’s numerical range [18. Also in this case, the intrinsic numerical range appears as the algebra numerical range in the monographs by F. Bonsall and J. Duncan [7, 8; we refer the reader to these books for general information and background. When $f$  is (the restriction to ${S}_{Y}$  of ) a uniformly continuous function from ${B}_{Y}$  to $Y$  which is holomorphic on the interior of ${B}_{Y}$  , both ranges appeared for the first time in [15, where some applications are given. Let us fix a Banach space $Y$  and a closed subspace $X$  . For every $f\in B\left({S}_{X},Y\right)$  , $V\left(f\right)$  is closed and convex, and we have
 $\begin{array}{c}\overline{co}W\left(f\right)\subseteq V\left(f\right),\end{array}$ (2)
where $\overline{co}$  means closed convex hull. (Indeed, for $x\in {S}_{X}$  and ${y}^{*}\in {S}_{{Y}^{*}}$  , the mapping $x\otimes {y}^{*}$  from $B\left({S}_{X},Y\right)$  to $\mathbb{K}$  defined by $\left[x\otimes {y}^{*}\right]\left(g\right):={y}^{*}\left(g\left(x\right)\right)\left(g\in B\left({S}_{X},Y\right)\right)$  is an element of $D\left(B\left({S}_{X},Y\right),J\right)$  .) In the case when $X=Y$  , the inclusion above is known to be an equality whenever $f$  is a uniformly continuous function [16,Theorem 1(see also [7,§9for bounded linear operators, [15for holomorphic functions, and [23for a slightly more general result). On the other hand, the equality $\overline{co}W\left(f\right)=V\left(f\right)$  for arbitrary bounded functions cannot be expected in general. Indeed, this equality holds for every $f\in B\left({S}_{Y},Y\right)$  if and only if $Y$  is uniformly smooth [22. In the general case when $X$  is a proper subspace, two sufficient conditions are given in [16,Theorems2and3for the equality in Eq.  2 , namely, such a equality holds for all $f\in {C}_{\text{u}}\left({S}_{X},Y\right)$  if either $X$  is finite-dimensional or $Y$  is uniformly smooth (see definition below). Let us mention that if $\overline{co}W\left(f\right)=V\left(f\right)$  for a bounded function $f\in B\left({S}_{X},Y\right)$  , then $maxReV\left(f\right)=supReW\left(f\right).$  Therefore, the following formulae, consequence of Eq.  1 , will be useful:
 $\begin{array}{cc}maxReV\left(f\right)& ={lim}_{\alpha ↓0}\frac{\parallel J+\alpha f\parallel -1}{\alpha }={lim}_{\alpha ↓0}{sup}_{x\in {S}_{X}}\frac{\parallel x+\alpha f\left(x\right)\parallel -1}{\alpha },\end{array}$ (3)
 $\begin{array}{cc}supReW\left(f\right)& ={sup}_{x\in {S}_{X}}{lim}_{\alpha ↓0}\frac{\parallel x+\alpha f\left(x\right)\parallel -1}{\alpha }.\end{array}$ (4)
 $\begin{array}{}\end{array}$
To state the main results of the paper, let us recall some definitions and notations.
The norm of a Banach space $Y$  is said to be smooth at $u\in {S}_{Y}$  if $D\left(Y,u\right)$  reduces to a singleton, and it is said to be Frechet-smooth or Frechet differentiable at $u$  whenever there exists
 $\begin{array}{c}{lim}_{\alpha \to 0}\frac{\parallel u+\alpha y\parallel -1}{\alpha }\end{array}$ (5)
uniformly for $y\in {B}_{Y}$  . If this happens for all $u\in {S}_{Y}$  we say that the norm of $Y$  is Frechet differentiable. If, in addition, the limit in  5 is also uniform in $u\in {S}_{X}$  , we say that the norm of $Y$  is uniformly Frechet differentiable at ${S}_{Y}$  or that $Y$  is uniformly smooth. A natural succedanea of Frechet differentiability of the norm when smoothness is not required is the following notion introduced by D. Gregory [13. The norm of $Y$  is strongly subdifferentiable (ssd in short) at $u$  whenever there exists ${lim}_{\alpha ↓0}\frac{\parallel u+\alpha y\parallel -1}{\alpha }$  uniformly for $y\in {B}_{Y}$  . If this happens for all $u\in {S}_{Y}$  , we simply say that the norm of $Y$  is ssd. Thus, the norm of $Y$  is Frechet differentiable at $u$  if and only if it is strongly subdifferentiable at $u$  , and $Y$  is smooth at $u$  . This property has been fully investigated in [11, where we refer the reader for background. It is shown in [11,Theorem 1.2that the norm of $Y$  is ssd at $u$  if and only if $D\left(Y,u\right)$  is strongly exposed by $u$  , i.e., for every $\varepsilon >0$  , there exists $\delta >0$  such that ${y}^{*}\in {B}_{{Y}^{*}},Re{y}^{*}\left(u\right)>1-\delta ⟹d\left({y}^{*},D\left(Y,u\right)\right)<\varepsilon .$  In this paper we study when the equality in Eq.  2 holds. The results of the paper can be divided in two categories.
The first category consists of negative results: we present examples of pairs of Banach spaces $Y$  and closed subspaces $X$  in which the equality in Eq.  2 fails, even for elements of $L\left(X,Y\right)$  . In section  2 we show that for every infinite-dimensional Banach space $X$  , there is a superspace $Y$  and an element $T\in L\left(X,Y\right)$  such that $\overline{co}W\left(T\right)\ne V\left(T\right)$  . In section  3 , we give concrete examples of Banach spaces $Y$  for which there is a closed subspace $X$  and an element $T\in L\left(X,Y\right)$  such that $\overline{co}W\left(T\right)\ne V\left(T\right)$  . Such examples are ${c}_{0}$  , ${\ell }_{2}{\sum }_{\infty }\left({\ell }_{2}{\oplus }_{1}{\ell }_{2}\right)$  , and, up to renorming, every non-reflexive Banach space. We will use the following notation:
a Banach space $Y$  is said to have the FR-property if for every closed subspace $X$  and every $T\in L\left(X,Y\right)$  , the equality $\overline{co}W\left(T\right)=V\left(T\right)$  holds.
The second category is that consisting of positive results. We introduce in section  4 a sufficient condition for the FR-property which covers all the previously known examples and may be interesting by itself. We use the name “Bishop-Phelps-Bollobas property” for it since it is related to the quantitative version of the Bishop-Phelps theorem [4given by B. Bollobas [6. We relate this property to the strong subdifferentiability of the norm and to the uniform smoothness.

2 When we fix the subspace

We recall that, when $X$  is finite-dimensional, for every superspace $Y$  and every (uniformly) continuous function $f:{S}_{X}⟶Y$  , the equality $\overline{co}W\left(f\right)=V\left(f\right)$  holds [16,Theorem 2. The aim of this section is to show that this fact characterizes the finite-dimensionality, even if we restrict ourselves to bounded linear operators.
Theorem 2.1. Let $X$  be an infinite-dimensional Banach space. Then, there are a superspace $Y$  and an operator $T\in L\left(X,Y\right)$  such that $\overline{co}W\left(T\right)\ne V\left(T\right)$
We need the following easy lemma.
Lemma 2.2. If $X$  is an infinite-dimensional Banach space, then there exists a norm-one operator $S\in L\left(X,{c}_{0}\right)$  which does not attain its norm.
• Proof. Since $X$  is infinite dimensional, the Josefson-Nissenzweig theorem (see [9,§XII) assures the existence of a sequence $\left\{{x}_{n}^{*}\right\}$  in ${S}_{{X}^{*}}$  ${w}^{*}$  -converging to $0$  . Now, the operator $S:X⟶{c}_{0}$  defined by $\left[Sx\right]\left(n\right)=\frac{n}{n+1}{x}_{n}^{*}\left(x\right)\left(x\in X,n\in \mathbb{N}\right),$  does not attain its norm.
• Proof of Theorem  2.1 . Let $Y=X\oplus {c}_{0}$  endowed with the norm $\parallel \left(x,t\right)\parallel =max\left\{\parallel x\parallel ,\parallel Sx{\parallel }_{\infty }+\parallel t{\parallel }_{\infty }\right\},$  where $S\in L\left(X,{c}_{0}\right)$  is a norm-one operator which does not attain its norm, and let $J:X⟶Y$  be the natural inclusion $Jx=\left(x,0\right)$  for every $x\in X$  . If we define $T\in L\left(X,Y\right)$  by $Tx=\left(0,Sx\right)$  for every $x\in X$  , it is straightforward to check that ${lim}_{\alpha ↓0}{sup}_{x\in {S}_{X}}\frac{\parallel x+\alpha Tx\parallel -1}{\alpha }=1\text{and}{sup}_{x\in {S}_{X}}{lim}_{\alpha ↓0}\frac{\parallel x+\alpha Tx\parallel -1}{\alpha }=0.$  Thus, Eq.  3 and  4 give $V\left(T\right)\ne \overline{co}W\left(T\right)$  , as desired.
Remark 2.3. With a bit more of work, one can show that the superspace $Y$  in the above theorem can be found in such a way that $Y/X$  has dimension $1$  . We divide the proof in two cases, depending on whether $X$  is reflexive or not.
Case 1: Suppose $X$  is not reflexive. Then by the James theorem, there exists ${x}^{*}\in {S}_{{X}^{*}}$  which does not attain its norm. Thus, we can define $Y=X\oplus \mathbb{K}$  endowed with the norm $\parallel \left(x,t\right)\parallel =max\left\{\parallel x\parallel ,|{x}^{*}\left(x\right)|+|t|\right\}\left(x\in X,t\in \mathbb{K}\right),$  which contains $X$  as the subspace $\left\{\left(x,0\right):x\in X\right\}$  . If we take $T\in L\left(X,Y\right)$  defined by $Tx=\left(0,{x}^{*}\left(x\right)\right)$  for every $x\in X$  , it is straightforward to show, by using Eq.  3 and  4 , that $maxReV\left(T\right)=1$  and $supReW\left(T\right)=0$  . Case 2: Suppose $X$  is reflexive. By the Elton-Odell ( $1+\varepsilon$  )-separation theorem, there are ${\varepsilon }_{0}>0$  and a sequence $\left\{{x}_{n}^{*}{\right\}}_{n⩾0}$  of elements of ${S}_{{X}^{*}}$  , satisfying $\parallel {x}_{n}^{*}-{x}_{m}^{*}\parallel ⩾1+{\varepsilon }_{0}\left(n\ne m\right)$  (see [9,§XIV). Since $X$  is reflexive, for each $n\in \mathbb{N}$  there exists ${x}_{n}\in {S}_{X}$  such that $|\left({x}_{n}^{*}-{x}_{0}^{*}\right)\left({x}_{n}\right)|=\parallel {x}_{n}^{*}-{x}_{0}^{*}\parallel ⩾1+{\varepsilon }_{0}.$  Therefore,
 $\begin{array}{c}|{x}_{0}^{*}\left({x}_{n}\right)|⩾|\left({x}_{n}^{*}-{x}_{0}^{*}\right)\left({x}_{n}\right)|-|{x}_{n}^{*}\left({x}_{n}\right)|⩾1+{\varepsilon }_{0}-1={\varepsilon }_{0}.\end{array}$ (6)
On the other hand, for each $n\in \mathbb{N}$  , we take ${y}_{n}^{*}=\frac{{x}_{n}^{*}-{x}_{0}^{*}}{\parallel {x}_{n}^{*}-{x}_{0}^{*}\parallel }\in {S}_{{X}^{*}}$  and we observe that ${y}_{n}^{*}\left({x}_{n}\right)=1$  for every $n\in \mathbb{N}$  . Since $X\mathbb{⊉}{c}_{0}$  , it can be deduced from the proof of the Elton-Odell theorem that $\left\{{x}_{n}^{*}\right\}$  is a basic sequence and so, it converges to zero in the $weak$  topology by the reflexivity of ${X}^{*}$  (see [24,Theorem II.7.2). Using this, and the fact that $\parallel {x}_{n}^{*}-{x}_{0}^{*}\parallel ⩾1+{\varepsilon }_{0}\text{and}\parallel {x}_{0}^{*}\parallel ⩽1,$  we obtain $\overline{lim}{y}_{n}^{*}\left(x\right)<1\left(x\in {B}_{X}\right).$  This clearly implies that the operator $S\in L\left(X,{\ell }_{\infty }\right)$  given by $\left[Sx\right]\left(n\right)=\frac{n}{n+1}{y}_{n}^{*}\left(x\right)\left(x\in X,n\in \mathbb{N}\right)$  does not attain its norm. Now, we take $Y=X\oplus \mathbb{K}$  with the norm given by $\parallel \left(x,t\right)\parallel =max\left\{\parallel x\parallel ,\parallel Sx{\parallel }_{\infty }+|t|\right\}\left(x\in X,t\in \mathbb{K}\right),$  we write $J\in L\left(X,Y\right)$  for the natural inclusion and, we consider the operator $T\in L\left(X,Y\right)$  defined by $Tx=\left(0,{x}_{0}^{*}\left(x\right)\right)$  for all $x\in X$  . Using Eq.  4 and the fact that $S$  does not attain its norm, we obtain $supReW\left(T\right)=0$  . To compute $maxReV\left(T\right)$  , we observe that $\parallel J+\alpha T\parallel ⩾\parallel {x}_{n}+\alpha T{x}_{n}\parallel =\parallel \left({x}_{n},\alpha {x}_{0}^{*}\left({x}_{n}\right)\parallel ⩾\parallel S{x}_{n}\parallel +\alpha |{x}_{0}^{*}\left({x}_{n}\right)|$  so, by using Eq.  6 and the fact that $\parallel S{x}_{n}\parallel ⟶1$  , we get $\parallel J+\alpha T\parallel ⩾1+\alpha {\varepsilon }_{0}\text{for every}\alpha >0\text{.}$  By just using Eq.  3 , we get $maxReV\left(T\right)⩾{\varepsilon }_{0}$  , which finishes the proof.

3 When we fix the superspace

As we commented in the introduction, the following result is a particular case of [16,Theorems2and3.
Proposition 3.1. Finite-dimensional spaces and uniformly smooth spaces have the FR-property.
In the preceding section we have constructed examples ad hoc of Banach spaces $Y$  which do not have the FR-property. The aim of this section is to present some concrete examples of this phenomenon which will also show that some natural extensions of Proposition  3.1 are not possible.
Let us give the first example.
Example 3.2. ${c}_{0}$  does not have the FR-property. Indeed, let $Y={c}_{0}\oplus {\mathbb{K}}^{2}$  endowed with the norm $\parallel \left(x,\lambda ,\mu \right)\parallel =max\left\{\parallel x{\parallel }_{\infty },|\lambda |+|\mu |\right\}\left(x\in {c}_{0},\lambda ,\mu \in \mathbb{K}\right),$  which is isometrically isomorphic to ${c}_{0}$  . We take a norm-one functional ${x}_{0}^{*}$  on ${c}_{0}$  not attaining its norm, we consider the closed subspace $X=\left\{\left(x,{x}_{0}^{*}\left(x\right),0\right):x\in {c}_{0}\right\}$  of $Y$  , and we write $J$  for the natural inclusion of $X$  into $Y$  . If we consider the operator $T:X⟶Y$  given by $T\left(x,{x}_{0}^{*}\left(x\right),0\right)=\left(0,0,{x}_{0}^{*}\left(x\right)\right)\left(x\in {c}_{0}\right),$  by using Eq.  3 , Eq.  4 , and the fact that ${x}_{0}^{*}$  does not attain its norm, it is easy to verify that $maxReV\left(T\right)=1\text{and}supReW\left(T\right)=0,$  which finish the proof.
Since the norm of ${c}_{0}$  is ssd (see [11,corollary 2.6, for instance), the above example shows that Proposition  3.1 cannot be extended to the class of Banach spaces with ssd norm.
On the other hand, using the ideas appearing in the above example, it is easy to prove the following.
Proposition 3.3. Every non-reflexive Banach space admits an equivalent norm failing the FR-property.
• Proof. Let $Z$  be a non-reflexive Banach space. Then, $Z$  is isomorphic to $Y=V{\oplus }_{\infty }\left(\mathbb{K}{\oplus }_{1}\mathbb{K}\right)$  , where $V$  is a $2$  -codimensional closed subspace of $Z$  and, therefore, it is also non-reflexive.
Then, we choose ${v}_{0}^{*}\in {S}_{{V}^{*}}$  which does not attain its norm, we define the closed subspace $X=\left\{\left(v,{v}_{0}^{*}\left(v\right),0\right):v\in V\right\},$  and we consider $J$  the natural inclusion of $X$  in $Y$  . As in the preceding example, the operator $T:X⟶Y$  given by $T\left(v,{v}_{0}^{*}\left(v\right),0\right)=\left(0,0,{v}_{0}^{*}\left(v\right)\right)\left(x\in X\right)$  satisfies $maxReV\left(T\right)=1\text{and}supReW\left(T\right)=0.\Lambda$
In view of Propositions  3.1 and  3.3 , one may wonder if reflexivity implies the FR-property. This is not the case, as the following example shows.
Example 3.4. The superreflexive space $Y={\ell }_{2}{\oplus }_{\infty }\left({\ell }_{2}{\oplus }_{1}{\ell }_{2}\right)$  does not have the FR property. Proof. First of all, it is straightforward to show that the norm-one operator $S:{\ell }_{2}⟶{\ell }_{2}$  defined by $\left[Sx\right]\left(n\right)=\frac{n}{n+1}x\left(n\right)\left(x\in {\ell }_{2},n\in \mathbb{N}\right)$  does not attain its norm. Now, we consider the closed subspace $X=\left\{\left(x,Sx,0\right):x\in {\ell }_{2}\right\}$  with its natural inclusion in $Y$  , and we define the operator $T:X⟶Y$  by $T\left(x,Sx,0\right)=\left(0,0,Sx\right)\left(x\in X\right).$  The proof will be finished if we show that $supReW\left(T\right)=0$  and $maxReV\left(T\right)⩾1$  .
For the first equality, given $x\in {S}_{{\ell }_{2}}$  we may find ${\alpha }_{x}>0$  such that $\left(1+{\alpha }_{x}\right)\parallel Sx\parallel <1$  .
Then, for each $0<\alpha <{\alpha }_{x}$  we have $\parallel \left(x,Sx,0\right)+\alpha T\left(x,Sx,0\right)\parallel =\parallel \left(x,Sx,\alpha Sx\right)\parallel =max\left\{1,\left(1+\alpha \right)\parallel Sx\parallel \right\}=1,$  and therefore ${lim}_{\alpha ↓0}\frac{\parallel \left(x,Sx,0\right)+\alpha T\left(x,Sx,0\right)\parallel -1}{\alpha }=0.$  The arbitrariness of $x\in {S}_{{\ell }_{2}}$  gives $supReW\left(T\right)=0$  . On the other hand, for each $\alpha >0$  , we observe that $\parallel J+\alpha T\parallel ⩾\left(1+\alpha \right)\frac{n}{n+1}\left(n\in \mathbb{N}\right),$  so $\parallel J+\alpha T\parallel ⩾1+\alpha$  , and $maxReV\left(T\right)={lim}_{\alpha ↓0}\frac{\parallel J+\alpha T\parallel -1}{\alpha }⩾1.$

4 A sufficient condition: The Bishop-Phelps-Bollobas property

The aim of this section is to study a sufficient condition for the FR-property which, actually, covers all the examples given previously. The motivation for this property is the quantitative version of the classical Bishop-Phelps’ Theorem [4, 5established by B. Bollobas [6(see [8,§16for the below version).
Theorem 4.1 (Bishop-Phelps-Bollobas). Let $Y$  be a Banach space and $\varepsilon >0$  .
Whenever ${y}_{0}\in {S}_{Y}$  and ${y}_{0}^{*}\in {S}_{{Y}^{*}}$  satisfy that $Re{y}_{0}^{*}\left({y}_{0}\right)>1-\frac{{\varepsilon }^{2}}{4}$  , there exists $\left(y,{y}^{*}\right)\in \Pi \left(Y\right)$  such that $\parallel y-{y}_{0}\parallel <\varepsilon \text{and}\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon .$
This theorem has played an outstanding role in some topics of geometry of Banach spaces (see [12, 20, 21, for instance), specially in the study of ssd norms [11or in the study of spatial numerical range of operators [8,§16and§17. Also, the proof of the fact that $\overline{co}W\left(f\right)=V\left(f\right)$  for every $f\in {C}_{\text{u}}\left({S}_{Y},Y\right)$  given in [16,Theorem 1uses the above result. For bounded linear operators, this equality can be also deduced from [17,Theorem 8, a result whose proof also uses the Bishop-Phelps-Bollobas theorem. Motivated by these facts, we introduce a property which will be sufficient for the FR-property and it may be of independent interest.
Definition 4.2. Let $Y$  be a Banach space and let $X$  be a closed subspace of $Y$  . We say that $\left(X,Y\right)$  is a Bishop-Phelps-Bollobas pair (BPB-pair in short) if for every $\varepsilon >0$  there exists $\delta >0$  such that whenever ${x}_{0}\in {S}_{X}$  , ${y}_{0}^{*}\in {S}_{{Y}^{*}}$  satisfy $Re{y}_{0}^{*}\left({x}_{0}\right)>1-\delta$  , there exists $\left(x,{y}^{*}\right)\in \Pi \left(X,Y\right)$  so that $\parallel {x}_{0}-x\parallel <\varepsilon \text{and}\parallel {y}_{0}^{*}-{y}^{*}\parallel <\varepsilon .$  We say that a Banach space $Y$  has the BPB property if for every closed subspace $X$  of $Y$  , $\left(X,Y\right)$  is a BPB-pair.
The next result shows that the BPB property is sufficient for the FR-property. Actually, it can be proved that the equality in Eq.  2 holds for uniformly continuous functions.
Theorem 4.3. Let $Y$  be a Banach space and $X$  a closed subspace such that $\left(X,Y\right)$  is a BPB-pair. Then, for every $f\in {C}_{\text{u}}\left({S}_{X},Y\right)$  , the equality $\overline{co}W\left(f\right)=V\left(f\right)$  holds.
• Proof. Let $J\in L\left(X,Y\right)$  be the inclusion map. Let $f\in {C}_{\text{u}}\left({S}_{X},Y\right)$  and $\Phi \in D\left({C}_{\text{u}}\left({S}_{X},Y\right),J\right)$  .
By [16,Proposition 1, it suffices to show that  $\begin{array}{c}Re\Phi \left(f\right)⩽supReW\left(f\right).\end{array}$ (7)
For each $n\in \mathbb{N}$  , by using [16,Lemma 1we may find ${x}_{n}\in {S}_{X}$  and ${y}_{n}^{*}\in {S}_{{Y}^{*}}$  such that  $\begin{array}{c}Re\Phi \left(f\right)⩽Re{y}_{n}^{*}\left(f\left({x}_{n}\right)\right)+1/n\end{array}$ (8)
and ${y}_{n}^{*}\left({x}_{n}\right)⟶1$  . Since $\left(X,Y\right)$  is a BPB-pair, it follows that there exists a sequence ${\left\{\left({\stackrel{~}{x}}_{n},{\stackrel{~}{y}}_{n}^{*}\right)\right\}}_{n\in \mathbb{N}}\subseteq \Pi \left(X,Y\right)$  such that $\left\{{x}_{n}-{\stackrel{~}{x}}_{n}{\right\}}_{n\in \mathbb{N}}\mathbb{⟶}0\text{and}\left\{{y}_{n}^{*}-{\stackrel{~}{y}}_{n}^{*}{\right\}}_{n\in \mathbb{N}}\mathbb{⟶}0.$  By Eq.  8 ,
 $\begin{array}{cc}Re\Phi \left(f\right)& ⩽Re{\stackrel{~}{y}}_{n}^{*}\left(f\left({\stackrel{~}{x}}_{n}\right)\right)+Re\left[{y}_{n}^{*}-{\stackrel{~}{y}}_{n}^{*}\right]\left(f\left({\stackrel{~}{x}}_{n}\right)\right)+Re{y}_{n}^{*}\left(f\left({x}_{n}\right)-f\left({\stackrel{~}{x}}_{n}\right)\right)+1/n\end{array}$
 $\begin{array}{cc}& ⩽supReW\left(f\right)+\parallel {y}_{n}^{*}-{\stackrel{~}{y}}_{n}^{*}\parallel \parallel f{\parallel }_{\infty }+\parallel f\left({x}_{n}\right)-f\left({\stackrel{~}{x}}_{n}\right)\parallel +1/n\end{array}$
 $\begin{array}{}\end{array}$
for all $n\in \mathbb{N}$  . Thus, Eq.  7 follows from the above and the uniform continuity of $f$  .
It is worth mentioning that the above proof follows the lines of [16,Theorem 1.
Corollary 4.4. Let $Y$  be a Banach space with the BPB property. Then, $Y$  has the FR-property.
As a consequence of the above corollary and Theorem  2.1 , we get the following.
Corollary 4.5. Let $X$  be an infinite-dimensional Banach space. Then, there is a superspace $Y$  of $X$  such that $\left(X,Y\right)$  is not a BPB-pair.
The above result implies that not every Banach space $Y$  has the BPB property. For instance, the examples given in section  3 of Banach spaces which do not have the FR-property also fail the BPB property.
Example 4.6. The spaces ${c}_{0}$  and ${\ell }_{2}{\oplus }_{\infty }\left({\ell }_{2}{\oplus }_{1}{\ell }_{2}\right)$  fail the BPB property in their canonical norms. Every non-reflexive Banach space admits an equivalent norm failing the BPB property.
On the other hand, if we restrict ourselves to finite-dimensional subspaces, we get a characterization of the ssd norms.
Proposition 4.7. Let $Y$  be a Banach space. Then, the norm of $Y$  is ssd if, and only if, for every finite-dimensional subspace $X\subseteq Y$  , the pair $\left(X,Y\right)$  is BPB.
• Proof. We suppose first that the norm of $Y$  is ssd. Let $X$  be a finite-dimensional subspace of $Y$  and let $\varepsilon >0$  be given. Since the norm of $Y$  is ssd, [11,Theorem 1.2gives us that for each $x\in {S}_{X}$  there exists ${\delta }_{x}>0$  so that ${y}^{*}\in {S}_{{Y}^{*}},Re{y}^{*}\left(x\right)>1-{\delta }_{x}⟹d\left({y}^{*},D\left(Y,x\right)\right)<\varepsilon .$  Therefore, if for each $x\in {S}_{X}$  we define ${A}_{x}=\left\{u\in {S}_{X}:\parallel u-x\parallel   the compactness of ${S}_{X}$  assures the existence of ${x}_{1},...,{x}_{n}\in {S}_{X}$  such that ${S}_{X}{=}^{n}{\bigcup }_{i=1}{A}_{{x}_{i}}.$  Then, $\delta =min\left\{\frac{{\delta }_{{x}_{i}}}{2}:i=1,...,n\right\}$  satisfies the BPB condition. Indeed, let ${x}_{0}\in {S}_{X}$  and ${y}_{0}^{*}\in {S}_{{Y}^{*}}$  be such that $Re{y}_{0}^{*}\left({x}_{0}\right)>1-\delta .$  Since ${x}_{0}\in {S}_{X}$  , there exists $j\in \left\{1,...,n\right\}$  so that ${x}_{0}\in {A}_{{x}_{j}}$  , that is $\parallel {x}_{0}-{x}_{j}\parallel   Therefore, $Re{y}_{0}^{*}\left({x}_{j}\right)>1-{\delta }_{{x}_{j}}$  which implies the existence of ${y}^{*}\in D\left(Y,{x}_{j}\right)$  such that $\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon$  .
To prove the converse, it is enough to fix ${x}_{0}\in {S}_{Y}$  and to show that ${x}_{0}$  strongly exposes $D\left(Y,{x}_{0}\right)$  [11,Theorem 1.2. To do so, let $X$  be the subspace of $Y$  generated by ${x}_{0}$  and, fixed $\varepsilon >0$  , let $\delta >0$  be given by the definition of the BPB for the pair $\left(X,Y\right)$  and $\varepsilon /2$  . Suppose now that ${y}_{0}^{*}\in {S}_{{Y}^{*}}$  is such that $Re{y}_{0}^{*}\left({x}_{0}\right)>1-\delta$  , then there exists $\left(x,{y}^{*}\right)\in \Pi \left(X,Y\right)$  so that $\parallel x-{x}_{0}\parallel <\varepsilon /2\text{and}\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon /2.$  Since $x\in \text{span}\left({x}_{0}\right)$  , there exists a modulus-one $\lambda \in \mathbb{K}$  such that $x=\lambda {x}_{0}$  . Therefore, $|\lambda -1|=\parallel \lambda {x}_{0}-{x}_{0}\parallel =\parallel x-{x}_{0}\parallel <\varepsilon /2,$  and then, $\lambda {y}^{*}\in D\left(Y,{x}_{0}\right)\text{and}\parallel \lambda {y}^{*}-{y}_{0}^{*}\parallel ⩽\parallel \lambda {y}^{*}-{y}^{*}\parallel +\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon /2+\varepsilon /2=\varepsilon ,$  which finishes the proof.
Since the norm of any finite-dimensional Banach space is ssd (see [11,pp. 48), we have the following corollary, which also implies the first part of Proposition  3.1 .
Corollary 4.8. Every finite-dimensional Banach space has the BPB property.
The other class of spaces with the FR-property given in Proposition  3.1 is the one of uniformly smooth spaces. This result can be also deduced from Corollary  4.4 , as the following proposition shows.
Proposition 4.9. Every uniformly smooth space has the BPB property.
• Proof. Let $Y$  be an uniformly smooth space. Then, ${Y}^{*}$  is uniformly convex, so, for every $\varepsilon >0$  , we may find $\delta >0$  (the modulus of convexity of ${Y}^{*}$  ) such that ${x}^{*},{y}^{*}\in {S}_{{Y}^{*}},\parallel {x}^{*}+{y}^{*}\parallel >2-\delta ⟹\parallel {x}^{*}-{y}^{*}\parallel <\varepsilon$  (see [2,Chapter IIfor instance). Let $X$  be a subspace of $Y$  , and let ${x}_{0}\in {S}_{X}$  and ${y}_{0}^{*}\in {S}_{{Y}^{*}}$  be so that $Re{y}_{0}^{*}\left({x}_{0}\right)>1-\delta$  . If we consider ${y}^{*}\in {S}_{{Y}^{*}}$  such that $Re{y}^{*}\left({x}_{0}\right)=1$  , we have $\parallel {y}^{*}+{y}_{0}^{*}\parallel ⩾Re\left({y}^{*}+{y}_{0}^{*}\right)\left({x}_{0}\right)>2-\delta$  and, therefore, $\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon ,$  which finishes the proof.
Observe that, in the above proof, the relation $\varepsilon -\delta$  does not depend on the subspace.
The next result shows that this fact actually characterizes the uniform smoothness.
Proposition 4.10. Let $Y$  be a Banach space with the BPB property in such a way that the relationship between $\varepsilon$  and $\delta$  in Definition  4.2 does not depend on the subspace $X$  . Then, $Y$  is uniformly smooth.
• Proof. In view of [11,Proposition4.1, it is enough to show that the limit ${lim}_{t↓0}\frac{\parallel u+ty\parallel -1}{t}=:\tau \left(u,y\right)$  exists uniformly for $y\in {B}_{Y}$  and $u\in {S}_{Y}$  . Given $\varepsilon >0$  , let $0<\delta <2$  be given by the “uniform” BPB property. Now, for $y\in {B}_{Y}$  , $u\in {S}_{Y}$  and $0  , we consider ${y}_{t}=\frac{u+ty}{\parallel u+ty\parallel }\in {S}_{Y}\text{and}{y}_{t}^{*}\in D\left(Y,{y}_{t}\right).$  It is immediate to check that $Re{y}_{t}^{*}\left(u\right)>1-\delta$  so, if we take $X=\text{span}\left(u\right)$  , the BPB property assures the existence of $\left(x,{z}_{t}^{*}\right)\in \Pi \left(X,Y\right)$  such that $\parallel x-u\parallel <\varepsilon$  and $\parallel {z}_{t}^{*}-{y}_{t}^{*}\parallel <\varepsilon$  . Since $x\in \text{span}\left(u\right)$  , there exists a modulus-one $\lambda \in \mathbb{K}$  such that $x=\lambda u$  . Therefore, $|\lambda -1|=\parallel \lambda u-u\parallel =\parallel x-u\parallel <\varepsilon ,$  and then, $\lambda {z}_{t}^{*}\in D\left(Y,u\right)\text{and}\parallel \lambda {z}_{t}^{*}-{y}_{t}^{*}\parallel ⩽\parallel \lambda {z}_{t}^{*}-{z}_{t}^{*}\parallel +\parallel {z}_{t}^{*}-{y}_{t}^{*}\parallel <\varepsilon +\varepsilon =2\varepsilon ,$  Now, the facts $\frac{\parallel u+ty\parallel -1}{t}=\frac{Re{y}_{t}^{*}\left(u+ty\right)-1}{t}⩽Re{y}_{t}^{*}\left(y\right)$  and $\tau \left(u,y\right)⩾Re\lambda {z}_{t}^{*}\left(y\right)$  (by Eq.  1 ), give $0⩽\frac{\parallel u+ty\parallel -1}{t}-\tau \left(u,y\right)⩽Re{y}_{t}^{*}\left(y\right)-Re\lambda {z}_{t}^{*}\left(y\right)⩽\parallel \lambda {z}_{t}^{*}-{y}_{t}^{*}\parallel <2\varepsilon ,$  and the arbitrariness of $\varepsilon >0$  finishes the proof.
We conclude the paper proving that a pair $\left(X,Y\right)$  is a BPB-pair provided that $X$  is an absolute ideal of $Y$  . Let us introduce the necessary definitions. We refer the reader to [8,§ 21, [19, and references therein for background. A closed subspace $X$  of a Banach space $Y$  is said to be an absolute summand of $Y$  if there exists another closed subspace $Z$  such that $Y=X\oplus Z$  and, for every $x\in X$  and $z\in Z$  , the norm of $x+z$  only depends on $\parallel x\parallel$  and $\parallel z\parallel$  . We also say that $Y$  is an absolute sum of $X$  and $Z$  . This implies that there exists an absolute norm on ${\mathbb{R}}^{2}$  such that $\parallel x+z\parallel =|\left(\parallel x\parallel ,\parallel z\parallel \right){|}_{a}\left(x\in X,z\in Z\right).$  By an absolute norm we mean a norm $|\cdot {|}_{a}$  on ${\mathbb{R}}^{2}$  such that $|\left(1,0\right){|}_{a}=|\left(0,1\right){|}_{a}=1$  and $|\left(a,b\right){|}_{a}=|\left(|a|,|b|\right){|}_{a}$  for every $a,b\in \mathbb{R}$  . Useful results about absolute norms are the following inequality $max\left\{|a|,|b|\right\}⩽|\left(a,b\right){|}_{a}⩽|a|+|b|a,b\in \mathbb{R},$  and the fact that absolute norms are nondecreasing and continuous in each variable. We say that $X$  is an absolute ideal of $Y$  if ${X}^{\perp }$  is an absolute summand of ${Y}^{*}$  , in which case, ${Y}^{*}$  can be identified with ${X}^{*}\oplus {X}^{\perp }$  with a convenient absolute sum. It is clear that absolute summands are absolute ideals, but the converse is not true.
Absolute summands and absolute ideals are generalizations of the well-known M-summands, L-summands, M-ideals, and the more general class of ${L}_{p}$  -summands [3, 14.
Proposition 4.11. Let $Y$  be a Banach space and let $X$  be a closed subspace. If ${X}^{*}$  is an absolute summand of ${Y}^{*}$  , then the pair $\left(X,Y\right)$  is BPB. In particular, this occurs when $X$  is an absolute ideal of $Y$  .
We need the following easy result, which we separate from the proof of the proposition for the sake of clearness.
Lemma 4.12. Let $E$  be $\left({\mathbb{R}}^{2},|\cdot {|}_{a}\right)$  where $|\cdot {|}_{a}$  is an absolute norm. We write ${b}_{0}=max\left\{b⩾0:|\left(1,b\right){|}_{a}=1\right\},$  and we define $A\left(\delta \right)=\left\{\left(a,b\right)\in {B}_{E}:a>1-\delta ,b⩾{b}_{0}\right\}\left(\delta >0\right).$  Then, for every $\varepsilon >0$  there exists $\delta >0$  such that $\text{diam}\left(A\left(\delta \right)\right)<\varepsilon .$
• Proof. Suppose, for the sake of contradiction, that the result does not hold. Then, there exists ${\varepsilon }_{0}>0$  such that for every $n\in \mathbb{N}$  , $\text{diam}\left(A\left(\frac{1}{n}\right)\right)⩾{\varepsilon }_{0}$  . So, we may find $\left({a}_{n},{b}_{n}\right)\in A\left(\frac{1}{n}\right)$  such that $|\left({a}_{n},{b}_{n}\right)-\left(1,{b}_{0}\right){|}_{a}⩾\frac{{\varepsilon }_{0}}{2}$  , and thus  $\begin{array}{c}\frac{{\varepsilon }_{0}}{2}⩽|{a}_{n}-1|+|{b}_{n}-{b}_{0}|\left(n\in \mathbb{N}\right).\end{array}$ (9)
Let $\left\{\left({a}_{{\sigma }_{n}},{b}_{{\sigma }_{n}}\right)\right\}$  be a convergent subsequence of $\left\{\left({a}_{n},{b}_{n}\right)\right\}$  , and let $\left(1,b\right)\in {S}_{E}$  be its limit.
By Eq.  9 and the fact that $\left({a}_{{\sigma }_{n}},{b}_{{\sigma }_{n}}\right)\in A\left(\frac{1}{{\sigma }_{n}}\right)$  , it is immediate to check that $\frac{{\varepsilon }_{0}}{2}⩽|b-{b}_{0}|\text{and}b⩾{b}_{0}.$  So, $b$  is strictly bigger than ${b}_{0}$  , a contradiction.
• Proof of Proposition  4.11 . There exist a subspace $Z$  of ${Y}^{*}$  and an absolute norm $|\cdot {|}_{a}$  on ${\mathbb{R}}^{2}$  so that ${Y}^{*}={X}^{*}\oplus Z$  and $\parallel \left({x}^{*},{z}^{*}\right)\parallel =|\left(\parallel {x}^{*}\parallel ,\parallel {z}^{*}\parallel \right){|}_{a}\left({x}^{*}\in {X}^{*},{z}^{*}\in Z\right).$  For $\varepsilon >0$  fixed, we take ${\delta }_{1}>0$  given by the preceding lemma applied for $\varepsilon /3$  , and we define $\delta :=min\left\{{\delta }_{1},\frac{{\varepsilon }^{2}}{36}\right\}.$  To finish the proof, for ${x}_{0}\in {S}_{X}$  and ${y}_{0}^{*}=\left({x}_{0}^{*},{z}_{0}^{*}\right)\in {S}_{{Y}^{*}}$  satisfying $Re{y}_{0}^{*}\left({x}_{0}\right)=Re{x}_{0}^{*}\left({x}_{0}\right)>1-\delta ,$  we have to find $\left(x,{y}^{*}\right)\in \Pi \left(X,Y\right)$  so that $\parallel {y}^{*}-{y}_{0}^{*}\parallel <\varepsilon \text{and}\parallel x-{x}_{0}\parallel <\varepsilon .$  To this end, since $\parallel {x}_{0}\parallel =1=‖\frac{{x}_{0}^{*}}{\parallel {x}_{0}^{*}\parallel }‖\text{and}Re\frac{{x}_{0}^{*}}{\parallel {x}_{0}^{*}\parallel }\left({x}_{0}\right)⩾Re{x}_{0}^{*}\left({x}_{0}\right)>1-\frac{{\varepsilon }^{2}}{36},$  we can apply the classical Bishop-Phelps-Bollobas Theorem ( 4.1 ) to $\left({x}_{0},\frac{{x}_{0}^{*}}{\parallel {x}_{0}^{*}\parallel }\right)\in X×{X}^{*}$  to get $\left(x,{x}^{*}\right)\in \Pi \left(X\right)$  such that  $\begin{array}{c}‖{x}^{*}-\frac{{x}_{0}^{*}}{\parallel {x}_{0}^{*}\parallel }‖<\frac{\varepsilon }{3}\text{and}\parallel x-{x}_{0}\parallel <\frac{\varepsilon }{3}.\end{array}$ (10)
Now, we distinguish two cases. Suppose first that $\parallel {z}_{0}^{*}\parallel ⩽{b}_{0}$  . Then, we take ${y}^{*}:=\left({x}^{*},{z}_{0}^{*}\right)$  , which satisfies $Re{y}^{*}\left(x\right)=1$  and $\parallel {y}^{*}\parallel =|\left(1,\parallel {z}_{0}^{*}\parallel \right){|}_{a}=1$  . Using Eq.  10 and the definition of $\delta$  , we get $\parallel {y}^{*}-{y}_{0}^{*}\parallel =\parallel {x}^{*}-{x}_{0}^{*}\parallel <\frac{\varepsilon }{3}+\delta <\varepsilon .$  So, the pair $\left(x,{y}^{*}\right)$  satisfies the desired condition.
Suppose otherwise that $\parallel {z}_{0}^{*}\parallel >{b}_{0}$  . In this case, we take ${y}^{*}:=\left({x}^{*},\frac{{b}_{0}}{\parallel {z}_{0}^{*}\parallel }{z}_{0}^{*}\right)$  , which clearly satisfies $Re{y}^{*}\left(x\right)=1=\parallel {y}^{*}\parallel$  . Now, $\left(1,{b}_{0}\right)$  and $\left(\parallel {x}_{0}^{*}\parallel ,\parallel {z}_{0}^{*}\parallel \right)$  belong to $A\left(\delta \right)$  and the diameter of this set is less than $\varepsilon /3$  by Lemma  4.12 , so we have $|\parallel {z}_{0}^{*}\parallel -{b}_{0}|⩽|\left(1,{b}_{0}\right)-\left(\parallel {x}_{0}^{*}\parallel ,\parallel {z}_{0}^{*}\parallel \right){|}_{a}<\frac{\varepsilon }{3}$  and $\parallel {y}^{*}-{y}_{0}^{*}\parallel =|\left(\parallel {x}^{*}-{x}_{0}^{*}\parallel ,\parallel {z}_{0}^{*}\parallel -{b}_{0}\right){|}_{a}⩽\parallel {x}^{*}-{x}_{0}^{*}\parallel +|\parallel {z}_{0}^{*}\parallel -{b}_{0}|<\varepsilon .\Lambda$
By just applying the above proposition and Theorem  4.3 , we get the following.
Corollary 4.13. Let $Y$  be a Banach space and let $X$  be a closed subspace of $Y$  such that ${X}^{*}$  is an absolute summand of ${Y}^{*}$  (in particular, if $X$  is an absolute ideal of $Y$  ). Then, $\overline{co}W\left(f\right)=V\left(f\right)$  for every $f\in {C}_{\text{u}}\left({S}_{X},Y\right)$  .
An interesting particular case is the case of $M$  -embedded and $L$  -embedded spaces. A Banach space $X$  is said to be $M$  -embedded if it is an $M$  -ideal of ${X}^{**}$  , and it is $L$  -embedded if ${X}^{**}=X{\oplus }_{1}Z$  for some closed subspace $Z$  of ${X}^{**}$  .
Corollary 4.14. If $X$  is an $M$  -embedded or an $L$  -embedded space, then $\left(X,{X}^{**}\right)$  is a BPB-pair.
We do not know if the assumption of being $M$  -embedded or $L$  -embedded in the above result is superabundant.
Acknowledgment: The authors would like to express their gratitude to Pradipta Bandyopadhyay, Catherine Finet, Gilles Godefroy, and Gines Lopez for their valuable suggestions and fruitful conversations about the subject of this paper.
References

1. F. L. Bauer, On the field of values subordinate to a norm, Numer. Math., 4 (1962), 103–111.
2. B. Beauzamy, Introduction to Banach spaces and their geometry, Mathematics Studies 68, North Holland, Amsterdam, 1982.
3. E. Behrends et al., ${L}^{p}$  -structure in real Banach spaces, Lecture Notes in Math. 613, Springer-Verlag, Berlin, 1977.
4. E. Bishop and R. R. Phelps, A proof that every Banach space is subreflexive, Bull. Amer. Math. Soc 67 (1961), 97–98.
5. E. Bishop and R. R. Phelps, The support functionals of a convex set, Proc. Sympos. Pure Math., Vol. VII: Convexity, pp. 27–35. Amer. Math. Soc., Providence, R. I., 1963.
6. B. Bollobas, An extension to the theorem of Bishop and Phelps, Bull. London Math. Soc. 2 (1970), 181–182.
7. F. F. Bonsall and J. Duncan, Numerical Ranges of operators on normed spaces and of elements of normed algebras, London Math. Soc. Lecture Note Ser. 2, Cambridge, 1971.
8. F. F. Bonsall and J. Duncan, Numerical Ranges II, London Math. Soc. Lecture Note Ser. 10, Cambridge, 1973.
9. J. Diestel, Sequences and Series in Banach spaces, Graduate Texts in Mathematics, 92, Springer-Verlag, New York.
10. N. Dunford and J. Schwartz, Linear Operators, Part I: General Theory, Interscience, New York, 1957.
11. C. Franchetti and R. Paya, Banach spaces with strongly subdifferentiable norm., Bolletino U. M. I. 7 (1993), 45–70.
12. J. R. Giles, Convex analysis with application in differentiation of convex functions, Research Notes in Math. 58, Pitman, Boston, 1982.
13. D. A. Gregory, Upper semi-continuity of subdifferential mappings, Canad. Math. Bull. 23 (1980), 11–19.
14. P. Harmand, D. Werner, and W. Werner, $M$  -ideals in Banach spaces and Banach algebras, Lecture Notes in Math. 1547, Springer-Verlag, Berlin, 1993.
15. L. A. Harris, The numerical range of holomorphic functions in Banach spaces, American J. Math. 93 (1971), 1005–1019.
16. L. A. Harris, The numerical range of functions and best approximation, Proc. Camb. Phil. Soc. 76 (1974), 133–141.
17. Å. Lima, The metric approximation property, norm-one projections and intersection properties of balls, Israel J. Math. 84 (1993), 451–475.
18. G. Lumer, Semi-inner-product spaces, Trans. Amer. Math. Soc. 109 (1961), 29–43.
19. R. Paya, Numerical range of operators and structure in Banach spaces, Quart. J. Math. Oxford 33 (1982), 357–364.
20. R. R. Phelps, Support cones in Banach spaces and their applications, Adv. Math. 13 (1974), 1–19.
21. R. R. Phelps, Convex functions, monotone operators and differentiability, Lecture Notes in Math. 1364, Springer-Verlag, Berlin, 1989.
22. A. Rodrıguez-Palacios, A numerical range characterization of uniformly smooth Banach spaces, Proc. Amer. Math. Soc. 129 (2001), 815–821.
23. A. Rodrıguez-Palacios, Numerical ranges of uniformly continuous functions on the unit sphere of a Banach space, J. Math. Anal. Appl. 297 (2004), 472–476.
24. I. Singer, Bases in Banach Spaces I, Die Grundlehren der mathematischen Wissenschaften, Band 154. Springer-Verlag, New York-Berlin, 1970.
25. O. Toeplitz, Das algebraische Analogon zu einem Satze von Fejer, Math. Z. 2 (1918), 187–197.

1 Corresponding author. E-mail: mmartins@ugr.es, Fax: +34958243272