Let $1<p<\infty$. Recall that a Banach space $X$ is $l_{p}$-saturated if every infinite-dimensional subspace of $X$ contains a subspace isomorphic to $l_{p}$. I have a seemingly stronger notion of $l_{p}$-saturated Banach spaces. We say that a Banach space $X$ is strongly $l_{p}$-saturated if for every infinite-dimensional subspace $M$ and every $\epsilon>0$, there exists an infinite-dimensional subspace $N$ of $M$ such that $N$ is $(1+\epsilon)$-isomorphic to $l_{p}$. I have the following two questions:

Question 1. If a Banach space $X$ is $l_{p}$-saturated, is $X$ strongly $l_{p}$-saturated?

Question 2. Are there other examples of strongly $l_{p}$-saturated spaces besides $l_{p}$? Is $L_{p}$ strongly $l_{p}$-saturated?

Thank you!

  • $\begingroup$ $L_p$ contains copies of $\ell_2$; it is not $\ell_p$-saturated. $\endgroup$ Dec 8, 2016 at 15:12
  • $\begingroup$ The Lorentz and Garling sequence spaces and certain Orlicz sequence spaces are strongly $\ell_p$-saturated. Probably so are the Lorentz-Orlicz and Musielak-Orlicz sequence spaces. Other spaces which I don't remember off the top of my head but are worth checking, include the $(p,q,\xi)$-Schreier spaces, the James spaces $J_p$ and spaces of the form $(\oplus X_n)_p$ for dim$(X_n)<\infty$. Note that the Lorentz sequence spaces are saturated with $(1+\epsilon)$-complemented copies of $\ell_p$, and probably also some Orlicz. It is an open question whether the same goes for Garling sequence spaces. $\endgroup$
    – Ben W
    Dec 8, 2016 at 15:54
  • $\begingroup$ @BenWallis, Could you check that $(\oplus X_{n})_{p}$ for $dim(X_{n})<\infty$ is strongly $l_{p}$-saturated? I can check that the James $p$-space is strongly $l_{p}$-saturated. $\endgroup$ Dec 9, 2016 at 12:08
  • $\begingroup$ For $(\oplus X_n)_p$, as we may assume $1<p<\infty$, this is just a gliding hump argument. $\endgroup$
    – Ben W
    Dec 9, 2016 at 12:31
  • $\begingroup$ One condition which guarantees that an $\ell_p$ saturated space be strongly $\ell_p$ saturated would be that the space has a Schauder basis $(e_i)_{i=1}^\infty$ (or even just an FDD) such that either $(1)$ for any $n\in\mathbb{N}$, $0=k_0<\ldots <k_n$, and $x_i=\sum_{j=k_{i-1}+1}^{k_i} a_je_j$, $$\|\sum_{i=1}^n x_i\|^p \leqslant \sum_{i=1}^n \|x_i\|^p,$$ or $(2)$ for any $n\in\mathbb{N}$, $0=k_0<\ldots <k_n$, and $x_i=\sum_{j=k_{i-1}+1}^{k_i} a_je_j$, $$\|\sum_{i=1}^n x_i\|^p \geqslant \sum_{i=1}^n \|x_i\|^p.$$ It would also be true if we only assumed such an inequality for skipped blocks. $\endgroup$
    – user114263
    Jul 31, 2018 at 17:55

1 Answer 1


As for question 1; this is the famous distortion problem. Your question has negative answer already for some renorming of $\ell_p$:

E. Odell, Th. Schlumprecht, The distortion problem of Hilbert space, Geom. Funct. Anal., 3, 201–207.

As for your second question, $L_p$ is not $\ell_p$-saturaded as the Rademachers span a copy of $\ell_2$.

  • $\begingroup$ Thanks, Ben and Tomek. $L_{p}$ is not $l_{p}$-saturated. But I have another question on $L_{p}$. Given any infinite-dimensional closed subspace $M$ of $L_{p}(1<p<\infty)$ and any $\epsilon>0$. Do there exist an infinite-dimensional closed subspace $N$ of $M$ and $1<q<\infty$ such that $N$ is $(1+\epsilon)$-isomorphic to $l_{q}$? $\endgroup$ Dec 9, 2016 at 9:10
  • $\begingroup$ I'm not completely sure, but my guess is that probably yes. I would look at chapter 6 from the Albiac/Kalton book to see if the isomorphism constants can be made that small. At least for $L_p$, $2\leq p<\infty$, it "should" be doable. $\endgroup$
    – Ben W
    Dec 9, 2016 at 10:48

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.