4
$\begingroup$

Consider the hyperplane $H=\{f\in L^\infty: \int f = 0\}$ of $L^\infty = L^\infty[0,1]$. My question is:

1. What is the Banach-Mazur distance between $H$ and $L^\infty$? Are there "natural" isomorphisms between these two spaces?

Since $L^\infty$ is 1-injective, we have the lower bound $$ inf\{\|P\|: \mbox{$P$ is a projection from $L^\infty$ onto $H$}\}\leq d_{BM}(H,L^\infty). $$ Recalling that the canonical projection $P:L^\infty\rightarrow H$ given by $Pf(x)= f(x) - \int f$ has norm $2$, we can pose the following related question:

2. Given a (linear, continuous) projection $P:L^\infty\rightarrow H$, do we always have $\|P\|\geq 2$?

$\endgroup$
1
  • $\begingroup$ Do you know whether the B-M distance is strictly larger than 2? My guess is that the exact value is not known. $\endgroup$ Apr 3, 2013 at 15:37

2 Answers 2

5
$\begingroup$

IIRC, if $K$ is a compact Hausdorff space that has no isolated points, then every projection from $C(K)$ onto a hyperplane has norm at least two. Maybe Dan Amir proved this? Anyway, in your situation the proof goes like this. A projection $P$ from $L^\infty$ onto $H$ has the form $Pf = f -(\int f) g$, where $\int g = 1$. Let $E=[g\ge 1]$ and observe that $g$ has positive measure. Take a subset $F$ of $E$ that has measure $\epsilon > 0$ and let $h=1_F - 1_{F^c}$. Then $H$ has norm one in $L^\infty$ and $\int h = -1 + 2\epsilon$. Thus $Ph$ is at least $2 - 2\epsilon$ on $F$.

$\endgroup$
1
$\begingroup$

For a pointed metric space $(M,d,0)$, we denote by $Lip_0(M)$ the Banach space of all real-valued Lipschitz functions $f$ defined on $M$ and such that $f(0)=0$. Recall that $Lip_0 [0,1] = L^\infty$ (see, for example, N. Weaver's book Lipschitz Algebras). It's not hard to see that the hyperplane $H$ defined above is isometrically isomorphic to $Lip_0([0,1]/\{0,1\})$, where $[0,1]/\{0,1\}$ the metric space obtained by collapsing $\{0,1\}$ to one point (see again Weaver for quotient metric spaces), that is, $[0,1]/\{0,1\}$ is the unit circle with the length distance. $Lip_0 ([0,1]/\{0,1\})$ can be seen as the subspace of $Lip_0 [0,1]$ formed by the functions $f$ such that $f(1)=0$. Take the obvious linear Lipschitz extension operator $T$ from $Lip_0 \{0,1\}$ to $Lip_0 [0,1]$ and define $\Phi: Lip_0 \{0,1\} \oplus_\infty Lip_0 ([0,1]/\{0,1\}) \rightarrow Lip_0 [0,1]$ by $\Phi(f,g)\doteq T(f)+g$. It is straightforward that $\Phi$ is an onto isomorphism with $\|\Phi\|=2$ and $\|\Phi^{-1}\|=2$. Since $Lip_0\{0,1\} \oplus_\infty Lip_0 ([0,1]/\{0,1\})$ is isometrically isomorphic to $Lip_0 ([0,1]/\{0,1\})$, it follows that $d_{BM}(H,L^\infty)\leq 4$.

Combining with Bill Johnson's answer we then have $2\leq d_{BM}(H,L^\infty) \leq 4$, and the search for the exact value continues.

From the above construction we get an explicit isomorphism between $H$ and $L^\infty$; not a very natural one though.

$\endgroup$
3
  • $\begingroup$ Is the exact value known for the analogous problem for $C[0,1]$? BTW: It is considered impolite by many users, including myself, to ask a question on MO and then vanish for days. $\endgroup$ Apr 8, 2013 at 17:15
  • $\begingroup$ I don't know about the C[0,1] problem either... if I find something about it, I'll post. Thank you for the etiquette advice. I'm a new user and I should learn this. After reading your answer and checking it, I felt like thinking about the upper bound before giving a feedback. I understand that that could be interpreted as impolite; and definitely I should clicked on the 'accept answer' button as soon as I checked it, so I apologize! $\endgroup$ Apr 9, 2013 at 8:19
  • $\begingroup$ Accepting an answer, especially an incomplete one, is not necessary; even undesirable, because a better answer might be posted. Just show that you are watching for information on the question you asked and respond to questions. $\endgroup$ Apr 10, 2013 at 2:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.