Consider a complex Banach space $X$ with a real subspace isometric to $\ell^1_{\mathbb R}$. What is the best constant $c$ such that $X$ contains a complex subspace $c$-isometric to $\ell^1_{\mathbb C}$?

I guess this is a very classical question, but I could not find an answer.

In this question, $\ell^1_{\mathbb R}$ (respectively $\ell^1_{\mathbb C}$) is the real (resp. complex) Banach space of summable sequences of real (resp. complex) numbers. And two Banach spaces $Y,Z$ are called $c$-isometric if there is a linear isomorphism $T:Y\to Z$ such that $\|T\| \|T^{-1}\| \leq c$.

The motivation is that I want to understand the complex Banach spaces of trivial (Rademacher) type. A famous theorem of Maurey and Pisier, usually stated for real Banach spaces, shows that $\ell^1_{\mathbb R}$ is finitely representable in any (real) space of trivial type $\mathbb R$. My question is essentially equivalent to "what happens when the field of scalars is $\mathbb C$?".


Schechtman and I discussed your question this morning and have these comments.

You can get $1+\epsilon$. The usual argument for improving the constant works in the complex case as well as the real case; i.e., if a complex Banach space contains a subspace complex isomorphic to $\ell_1$, then for all $\epsilon >0$ it contains a subspace $1+\epsilon$-complex isomorphic to $\ell_1$.

To see that a complex Banach space that contains a real subspace isomorphic to $\ell_1$ also contains a complex subspace that is complex isomorphic to $\ell_1$, apply Dor's extension to the complex case of Rosenthal's $\ell_1$ theorem. The full theorem says that a (real or complex) Banach space contains an isomorphic copy of $\ell_1$ iff the space contains a bounded sequence that has no weakly Cauchy subsequence.

For other values of $p$ the situation is different. By the Odell-Schlumprecht distortion theorem, for every $M$ there is a an equivalent renorming of $\ell_p$, $1<p<\infty$, s.t. no subspace of the resulting space is $M$-isomorphic to $\ell_p$.

Something we don't see is the following (not that we spent much time thinking about it). Suppose a complex Banach space contains a real subspace isomorphic to real $\ell_p$, $1<p<\infty$. Does it contain a complex subspace isomorphic to complex $\ell_p$? However, the local version looks OK; i.e., if real $\ell_p$ is $C$-finitely representable in a complex space $X$ (and hence $1+\epsilon$-finitely representable by Krivine's theorem) then complex $\ell_p$ is something like $4+\epsilon$-finitely representable in $X$. For that you use spreading model theory to get a suppression $1$ unconditional basis in a spreading model s.t. the basis is isometrically $\ell_p$ for real coefficients. It looks like with a bit of work you can change the $4+\epsilon$ to $1+\epsilon$, but we did not check it to the end. It looks like you can make the norm invariant for multiplication of expansions in every coordinate by $\pm 1$ (different $\pm$ in the various coordinates) by passing to a block basis, and multiplication of coordinates by complex numbers having modulus one should also work by passing to a block basis.

  • $\begingroup$ Thank you Bill. Do you have a reference for the usual argument for improving the constants? $\endgroup$ – Mikael de la Salle Feb 11 '14 at 19:46
  • $\begingroup$ R. J. James, Uniformly non-square Banach spaces, Annals of Math., 80 (1964), 542-550. $\endgroup$ – Bill Johnson Feb 11 '14 at 20:09
  • $\begingroup$ Or look at the argument on page 4 of math.tamu.edu/~bill.johnson/CJZh10.pdf $\endgroup$ – Bill Johnson Feb 11 '14 at 20:10

I think that $c$ arbitrarily close to $1$ can be realized. From Dvoretzky's theorem we now that given any $\varepsilon>0$, there is some finite $N$ such that the $N$-dimensional space with the $\ell^1$ norm, denote it by $\ell^1_{\mathbb{R}}(N)$, contains a $2$-plane $H$ which is $(1+\varepsilon)$-isometric to Euclidean plane. Write $$\ell^1_{\mathbb{R}} = \ell^1_{\mathbb{R}}(N) \oplus \ell^1_{\mathbb{R}}(N) \oplus \dots $$ where $\oplus$ is a "Banach sum" with the norm on the sum being the sum of the norms. Then the subspace obtained by taking a copy of $H$ in each copy of $\ell^1_{\mathbb{R}}(N)$ should be $(1+\varepsilon)$-isometric to $\ell^1_{\mathbb{C}}$.

  • $\begingroup$ Thank you. If I understand correctly, your construction gives an $\mathbb R$-linear $(1+\varepsilon)$-isometry. I was probably not clear in my question, but what I am looking for is a $\mathbb C$-linear isomorphism between $\ell^1_{\mathbb C}$ and a subspace of $X$. $\endgroup$ – Mikael de la Salle Feb 10 '14 at 22:15
  • $\begingroup$ Ok, I missed the word "complex" in your question, sorry. $\endgroup$ – Benoît Kloeckner Feb 11 '14 at 15:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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