Some Geometric Aspects Related to Lim’s Condition

Let Y be a closed subspace of X and 0 <\varepsilon <2. Consider \{\phi_\alpha\} to be a net in B(Y^*) and \phi \in B(Y^*) such that \phi_\alpha \xrightarrow{w^*} \phi and \|\phi_\alpha - \phi_\beta\| \geq \varepsilon for all \alpha \neq \beta. Let \overline{\phi_\alpha} be any norm-preserving extension of \phi_\alpha and assume without loss of generality that \overline{\phi_\alpha} \xrightarrow{w^*} z^* for some z^* \in B(X^*). Clearly, z^*|_Y=\phi.

Notice that, we have \| \overline{\phi_\alpha}-\overline{\phi_\beta}\| \geq \|\phi_\alpha - \phi_\beta\| \geq \varepsilon.

So, \|\phi\| \leq \|z^*\| \leq 1-\varepsilon/2.

We verify Condition (iii) of [CPW, Theorem 2], to get the required conclusion. Let x \in S({\mathcal A}) be a normal element and consider the commutative C^*-algebra generated by \{x,~x^*\}. By the Gelfand-Naimark theorem, we have the identification of this separable subalgebra as C_0(\sigma(x)). We also have that \sigma(x) is a locally compact metric space. Since (\ddag) is a hereditary property, the separable commutative C^*-subalgebra C_0(\sigma(x)) has property (\ddag). By the remarks made at the start of Section 2, we get that \sigma(x) is a discrete set with 0 as the only possible accumulation point. Hence, we have the decomposition (v) from [CPW, Theorem 2]. Now, by Corollary  5 and the example following it, we see that the Hilbert spaces under consideration are finite-dimensional.

Let 0 \leq s \leq 1 and 0 <\varepsilon <2s. Suppose \{\phi_{\alpha}\} is a net in X^* with \|\phi_{\alpha}\| \leq s for all \alpha and \phi \in B(X^*) is such that \phi_{\alpha} \xrightarrow{w^\ast} \phi and \|\phi_{\alpha}-\phi_{\beta}\| \geq \varepsilon for all \alpha \neq \beta. We have, 0 <\varepsilon/s <2. Let \psi_{\alpha}=\phi_{\alpha}/s for all \alpha and \psi=\phi/s. Then, \psi_\alpha \in B(X^*) for all \alpha, \psi_{\alpha} \xrightarrow{w^\ast} \psi and \|\psi_{\alpha}-\psi_{\beta}\| \geq \varepsilon/s. Therefore, \|\psi\| \leq 1 - \varepsilon/2s. This implies, \|\phi\| \leq s - \varepsilon/2.

Clearly, if Z has property (\dagger), then X and Y being subspaces of Z and (\dagger) being hereditary, X and Y have property (\dagger).

For the proof of the converse part, we first note that Z^* = X^* \bigoplus_{\ell_1} Y^*. We denote the norms on X and Y also by \|\cdot\|. Let 0 <\varepsilon <2, \{\phi_{\alpha}\} be a net in B(Z^*) and \phi \in B(Z^*) such that \phi_{\alpha} \xrightarrow{w^\ast} \phi and \|\phi_{\alpha}-\phi_{\beta}\| \geq \varepsilon for all \alpha \neq \beta. We now choose components from the corresponding summands in the dual and also assume, without loss of generality, that the components of \phi are non-zero. For each \alpha, let x^*,y^*,x_{\alpha}^*, y_{\alpha}^* be such that \phi_{\alpha}=x_{\alpha}^* + y_{\alpha}^* for all \alpha and \phi=x^* + y^*. We have, \|\phi_\alpha\| = \|x_{\alpha}^*\|+\|y_{\alpha}^*\| \leq 1. Passing to subnets, if necessary, we may assume that \|x_{\alpha}^*\|=s for some 0 \leq s \leq 1 and for all \alpha. So, \|y_{\alpha}^*\| \leq 1-s. We also have, x_\alpha^* \xrightarrow{w^\ast} x^* and y_\alpha^* \xrightarrow{w^*} y^*.

We have, \|\phi_\alpha - \phi_\beta\|=\|x_\alpha^*-x_\beta^*\| + \|y_\alpha^*-y^*\| \geq \varepsilon. Again, passing to subnets if necessary, let s'=\|x_\alpha^*-x_\beta^*\| for all \alpha \neq \beta. So, \|y_\alpha^*-y_\beta^*\| \geq \varepsilon-s' for all \alpha \neq \beta.

Since, X and Y have property (\dagger), by Lemma  8, \|x^*\| \leq s - (\varepsilon-s')/2 and \|y^*\| \leq (1-s) - s'/2. Thus,

Again, the hereditary nature of (\ddag) implies that X_\gamma has (\ddag) for each \gamma.

Conversely, we have X^*=\bigoplus_{\ell_1}X_\gamma^*. Again, we denote the norm on each X_\gamma as \|\cdot\|. In what follows, we use the standard identification of X^\ast, in particular, when the indexing set I is uncountable, 0 \neq \phi \in B(X^\ast) has only countably many non-zero coordinates.

Let 0 <\varepsilon <2, \{\phi_\alpha\} be a net of norm-attaining functionals in B(X^*) with \|\phi_\alpha - \phi_\beta\| \geq \varepsilon for all \alpha \neq \beta and \phi \in B(X^*) be a norm-attaining functional such that \phi_{\alpha} \xrightarrow{w^\ast} \phi. .5em We first note that any norm-attaining 0 \neq \phi \in B(X^*) has only finitely many non-zero coordinates. To see this, suppose x \in S(X) and

Thus, whenever \phi(\gamma) \neq 0, \|x(\gamma)\|=1 and it is easy to see that the corresponding \phi(\gamma) attains its norm at x(\gamma) \in S(X_{\gamma}). Since \{\|x(\gamma)\|\} vanishes at infinity, we obtain that \phi has only finitely many non-zero coordinates.

So, without loss of generality, we may assume that for each \alpha, \phi_{\alpha} has only first n coordinates non-zero.

Therefore, proceeding as in the proof of Theorem  9, we obtain that \|\phi\| \leq 1-\varepsilon/2.

It follows from the proof of Theorem  7 that the hypothesis here is sufficient to ensure the c_0-decomposition of {\mathcal A} into finite-dimensional spaces. Thus {\mathcal A} has (\ddag).

The argument follows the lines of [LM, Theorem 4]. We have, X^*=\bigoplus_{\ell_1} X_i^*.

Let each X_i satisfy Lim’s condition. Let 0 \leq s \leq 1, \{\phi_\alpha\} be a net in X^* such that \phi_\alpha \xrightarrow{w^\ast} 0 and \|\phi_\alpha\|=s for all \alpha. Let \phi \in X^*.

Clearly, \|\phi_\alpha - \phi\| \leq s + \|\phi\|.

Let \varepsilon >0 be given. For any finite subset \mathcal{F} of \mathcal {I}, \|\phi_\alpha-\phi\| \geq \|\phi_\alpha\|-\|\phi\| + 2 \sum_{i \in \mathcal{F}}(|\phi(i)|-|\phi_\alpha(i)|).

Let \mathcal{F}_0 be a finite subset of \mathcal {I} such that \sum_{i \in \mathcal{F}_0} |\phi(i)| >\|\phi\| - \varepsilon. Since, \phi_{\alpha} \xrightarrow{w^\ast} 0, so there exists \alpha_0 such that for \alpha \geq \alpha_0, \sum_{i \in \mathcal{F}_0} |\phi_\alpha (i)| <\varepsilon. Thus, for \alpha \geq \alpha_0,

So, \lim_\alpha \|\phi_\alpha - \phi\| = s + \|\phi\|.

Let \mathcal A satisfy property (\ddagger). Then, by Theorem  7, \mathcal A=\bigoplus_{c_0}\mathcal{L}(H_\alpha), where H_\alpha is a finite-dimensional Hilbert space for each \alpha.

But, every finite-dimensional space satisfies Lim’s condition. So, \mathcal{L}(H_\alpha) satisfies Lim’s condition for each \alpha. By the above theorem, \mathcal A satisfies Lim’s condition.

From the proof of [NR, Proposition 3.10 ((ii) \implies (i))], Y=Y_1 \bigoplus_{\infty}Y_2, where Y_1 is finite-dimensional and Y_2=\sum_{i \in \mathbb{N} \setminus I}\bigoplus_{c_0}X_i, where I is a finite set. By Theorem  12, Y_2 satisfies Lim’s condition. Y_1 being finite-dimensional also satisfies Lim’s condition. Hence, Y satisfies Lim’s condition.

By theorem 7, we have that \mathcal{A}=\bigoplus_{c_0}\mathcal{L}(H_\alpha), where \mathcal{L}(H_\alpha) is the space of bounded operators on a finite-dimensional Hilbert space H_\alpha. Hence, the conclusion follows from Theorem  12. We also get from [LM, Lemma 4] that {\mathcal A} has the w^*-normal structure. Therefore, by [Lim, Theorem 1], any non-expansive map T: K \rightarrow K has a fixed point for every w^*-compact convex set K in {\mathcal A}^\ast.

We have that Y^\perp \cong (X/Y)^* via the canonical isometry. Let 0 <\varepsilon <2, \{\phi_\alpha\} be a net of norm-attaining functionals in B((X/Y)^*) such that \|\phi_\alpha - \phi_\beta\| \geq \varepsilon and \phi \in B((X/Y)^*) be a norm-attaining functional with \phi_\alpha \xrightarrow{w^\ast} \phi. We can identify \{\phi_\alpha\} and \phi as elements in Y^\perp.

Now, \phi attains norm on B(X/Y). So, there exists x_0 \in X such that \phi(x_0)=\|\phi\|=d(x_0, Y). But, Y is proximinal, thus, there is y_0 in Y such that \phi(x_0)=\phi(x_0-y_0)=\|\phi\|=\|x_0-y_0\|. So, \phi is norm-attaining as an element of Y^\perp. Similarly, for each \alpha, \phi_\alpha is norm-attaining as an element of Y^\perp.

Now, (\ddag) is hereditary. So, Y^\perp has (\ddag). Thus, (X/Y) has (\ddag).

Let Y be a separable subspace of X such that \ell_1 \not\hookrightarrow Y. By a result in [Di, Pg. 215], every element of S(Y^{**}) is a w^*-sequential limit of elements of S(Y). Let \phi \in S(Y^{**}) be a norm-attaining functional and \{y_n\} be a sequence in S(Y) such that y_n \xrightarrow{w^\ast} \phi. If \{y_n\} does not converge to \phi in the norm, then \{y_n\} is not Cauchy. Then there is a subsequence, again denoted by \{y_n\} such that for some \varepsilon >0, this subsequence is \varepsilon-separated, i.e., \|y_n-y_m\| \geq \varepsilon for all n \neq m. Clearly y_n’s are norm-attaining on Y^\ast and y_n \rightarrow \phi in the w^*-topology.

Now, Y^\perp \subset X^* is a w^*-closed and hence, a proximinal subspace. Thus Y^* (being isometrically isomorphic to X^*/Y^\perp) has property (\ddag). So, \|\phi\| \leq 1-\varepsilon/2, a contradiction. Thus, we get that \phi \in Y. Now by the Bishop-Phelps theorem, we conclude that Y is reflexive and hence, as reflexivity is a separably determined property, X is a reflexive space.

By an application of Krein-Milman theorem, any element x^* of \partial_e B(X^\ast) can be extended to an element of \partial_e C(\Omega)^\ast. So, |x^*({\bf 1})|=1. Now, w^*-limit of such a point also takes absolute value 1 at the constant function {\bf 1}. Thus, \overline{\partial_e B(X^\ast)}^{w^*} \subset S(X^\ast). All the functionals in the w^*-closure attain their norm at {\bf 1}.

Since X^\ast = span\{x^\ast\} \bigoplus_1 N, we note that if y^\ast \in S(X^\ast) is norm-attaining and y^\ast = \alpha x^\ast+n^\ast for some n^\ast \in N, with both the components non-zero, then it is easy to see that x^\ast also attains its norm. Now, using the Bishop-Phelps theorem on the denseness of norm-attaining functionals ([Ho, Pg. 169]), we get that there is a norm-attaining functional y^\ast for which the components in the decomposition are non-zero. Thus x^\ast attains its norm.

Let x^\ast be a w^*-accumulation point of D. Let \{x^\ast_{\alpha}\} \subset D be a net such that x^\ast_{\alpha} \rightarrow x^\ast in the w^* topology. Since any two independent points in D can not belong to the same one dimensional L-summand, we may and do assume that \|x^\ast_{\alpha}-x^\ast_{\beta}\| =2 for \alpha \neq \beta. Since \overline {\partial_e B(X^*)}^{w^*} \subset S(X^\ast), we get \|x^\ast\|=1. This contradicts property (\dag).

This shows that D, after identifying vectors that are multiples of scalars of modulus 1, is a finite discrete set. Thus X is isometric to \ell^\infty(k). In the case of a uniform algebra, we note, by Lemma  19, that extreme points of the dual unit ball as well as their w^*-accumulation points are norm-attaining functionals. Therefore, any uniform algebra satisfying property (\ddag) is finite-dimensional.

Let x_0 \in S(X). Denote by S_{x_0}:=\{f \in S(X^*) : f(x_0)=1\}. Since X is an L^1-predual space and as S_{x_0} is a w^*-closed extreme convex subset (face) of B(X^*), it follows from [E, Theorem 3.3] that span\{S_{x_0}\} is a w^*-closed subspace and X^\ast = span\{S_{x_0}\} \bigoplus_1 N for a closed subspace N \subset X^\ast.

Now, if J = \{x \in X: x(S_{x_0})=0\}, then J^\bot = span\{S_{x_0}\}. Let \pi: X \rightarrow X/J be the quotient map and \Phi: X/J \rightarrow A(S_{x_0}) be defined as \Phi(x)(x^\ast)= x^\ast(x_0), where x^\ast \in S_{x_0} and x \in X. This is a linear map taking \pi(x_0) to 1 and it is easy to see that \Phi is a surjective isometry.

It now follows from [L, Theorem 2, Section 19] and its complex versions from [L, Section 20], that S_{x_0} is a Choquet simplex. Since X/J has (\ddag), we obtain from Theorem 21 that X/J is isometric to \ell^\infty(k). Consequently x is a k-smooth point.