6

Riesz's Lemma says that for a Banach space $X$, a proper closed subspace $Y\subset X$, and $\epsilon\gt 0$ there exist $x\in X$ with $\|x\|=1$ and $d(x,Y)\ge 1-\epsilon$.

I have to find a closed subset in $l_\infty$ that demonstrates that the above result is not true for $\epsilon=0$. That is, construct $Y$ such that for all $x\in l^\infty$ with $\|x\|_{l^\infty}=1$ it holds $d(x,Y)<1$.

The problem is that I'm not suposed to use weak topology, so I can't use Jame's Theorem (I've read a solution using it).

hbghlyj
  • 5,361
Navareño
  • 154
  • I think I didn't explain myself well, the subset $Y$ has to verify the Riesz's Lemma strictly with $\epsilon>0$, I mean, for every $x\in X$ with $||x||=1$ is $d(x,Y)<1$. – Navareño Jun 21 '17 at 22:31
  • see also https://math.stackexchange.com/questions/905087/rieszs-lemma-for-l-infty-and-alpha-1 – daw Jun 23 '17 at 10:25

1 Answers1

7

We proceed in two steps: first show that we can construct such a subspace using a non-norm attaining functional, second construct such a functional for $l^\infty$.


Assume that $g\in X^*$ is a functional with $\|g\|_{X^*}=1$ such that for all $x$ with $\|x\|_{X}=1$ it holds $|g(x)|<1$. I.e., the supremum in the definition of the functional is not attained.

Define $Y:=\ker g$. Let $x$ with $\|x\|_X=1$ be given. Assume wlog $g(x)>0$. Then there is $z$ with $g(x)<g(z)<1$. Define $y:=x - g(x)g(z)^{-1}z$. Then $g(y)=0$, thus $y\in Y$, and $$ \|y-x\| = g(x)g(z)^{-1} \|z\|<1. $$ This shows $d(x,Y)<1$ for all $x$ with $\|x\|_X=1$.


It remains to construct such a functional $g\in (l^\infty)^*$. Let $\phi\in (l^\infty)^*$ be an extension of the limit functional. Let $(a_k)$ be a sequence of positive numbers such that $\sum_{k=1}^\infty a_k=1$. Define $$ g(x)=\sum_{k=1}^\infty a_kx_k - \phi(x_k). $$ First let us show $\|g\|=2$. For given $n$ set $$ x_k = \begin{cases} 1 & k\le n\\ -1 & k >n \end{cases} $$ Then $$ g(x)=\left(\sum_{k=1}^n a_k - \sum_{k=n+1}^\infty a_k\right) +1 \to\left( \sum_{k=1}^\infty a_k\right)+1 = 2\quad \text{ for } n\to\infty. $$ If $x$ is constant one, then $g(x)=0$. Let $x$ be of unit norm with $g(x)>0$, but $x_k<1$ for some $k$, then $g(x) < g(x + (1-x_k)e_k)$. This shows that $|g(x)|<\|g\|$ for all $x$ with $\|x\|_{l^\infty}=1$.


So we do not use James' theorem, but rather construct a not norm-attaining functional explicitly. The construction in James original paper uses Banach limits as well, there I got the inspiration from.

daw
  • 54,637
  • 2
  • 44
  • 85