Recently, I bumped into the class of operators that factor through $\ell_{1}(X)$ for some set $X$. For now, $X$ is a set with arbitrary cardinality but if it leads to a more concrete answer to my questions below, feel free to restrict $X$ to be countable. The restriction makes little to no difference to what it follows. I should also warn that I am fairly ignorant of operator ideals and Banach space theory, so please be gentle.

First some definitions. Define,

$$ \|T\|_{\ell_{1}}= \inf\{\|R\|\|S\|\} $$

where the infimum is taken over all the factorizations of $T$ as $\xrightarrow{S}\ell_{1}(X)\xrightarrow{R}$. Obviously, $\|T\|\leq \|T\|_{\ell_{1}}$. Define $\mathcal{L}(A, B)$ to be the linear space of operators that factor through some $\ell_{1}(X)$ with the above norm. The little bit of thought that I have dedicated to this has produced the following up to now (and please correct me, if I have fumbled somewhere):

  1. We have $\|RTS\|_{\ell_{1}}\leq \|R\|\|T\|_{ell_{1}}\|S\|$. Sketch: obvious.

  2. The normed space $\mathcal{L}(A, B)$ is complete. Sketch: If $(T_{n})$ is $\|,\|_{\ell_{1}}$-Cauchy then it has a uniform limit. To prove that this limit factors through some $\ell_{1}(X)$ note two things. First, if you have a factorization through $\ell_{1}(X)$ as $RS$ and $X\subseteq Y$ then, since $\ell_{1}(X)$ is a norm-1 complemented subspace of $\ell_{1}(Y)$, you can make the factorization to pass through the larger $\ell_{1}(Y)$ without altering $\|R\|\|S\|$. Second, one has the isometric isomorphism, $$ \sum_{n}\ell_{1}(X_{n})\cong \ell_{1}(\coprod_{n} X_{n}) $$ which allows to take a sequence of factorizations and push them all to a common space $\ell_{1}(X)$. Thus the uniform limit factors through some $\ell_{1}(X)$.

  3. Finite-rank operators factor through $\ell_{1}(X)$. Sketch: all finite-dimensional spaces are linearly homeomorphic to $\ell_{1}(n)$. These first three conditions taken together mean that $(A, B)\mapsto \mathcal{L}(A, B)$ is an operator ideal (or Banach ideal, I am uncertain of the official terminology).

  4. Each $T$ is completely continuous. Sketch: a sequence in $\ell_{1}(X)$ lives inside a copy of $\ell_{1}$. The Schur property of $\ell_{1}$ gives the result.

Now for my first batch of questions: can this class of operators be characterized? Any more salient properties of these operators? And what about the norm $\|,\|_{\ell_{1}}$, is there some other more enlightening description of it? How far is it from the operator norm?

The second batch of questions is related to what are the properties required of a full subcategory $C$ of the category of Banach spaces so that one obtains an operator ideal by factorizing operators through it. An obvious example is the ideal of weakly compact operators that by Davis-Figiel-Johnson-Pelczynski is the class of operators that factor through reflexive spaces. My guess is that something like $\omega_{1}$-filteredness of $C$ with $\omega_{1}$ the first uncountable ordinal, is enough for the argument to go through, but I am sure someone smarter and more knowledgeable has already thought about this.

If you have appropriate references, that would be great; extra kudos if available online. Next September I will have access to a library and plan to get my hands on the Defant, Floret monograph Tensor norms and operators ideals -- not a very cheerful prospect actually, as the book looks rather daunting. The book Absolutely summing operators by Diestel, Jarchow and Tonge should also be useful, but alas, last time I checked it was not available.

Regards, TIA, G. Rodrigues

  • $\begingroup$ It is "Davis", not "Davies". $\endgroup$ Commented Aug 10, 2010 at 15:52

2 Answers 2


I think the most user friendly introduction to the topic is the book of Diestel, Jarchow, and Tonge, so I suggest you locate that book. There you will find the description of the dual ideal to the operators that factor through some $L_1$ space. (Actually it is better to look at operators $T:X\to Y$ such that $i_YT$ factors through some $L_1$ space, where $i_Y$ is the canonical embedding of $Y$ into $Y^{**}$. This operator ideal is called $\Gamma_1$.) You can also find the basic things in N. Tomczak-Jaegermann's book ``Banach-Mazur distances and finite-dimensional operator ideals", published by Longman.

A subproblem to characterizing the operators which factor through $\ell_1$ is to characterize the operators into $L_1$ that factor through $\ell_1$. I think this has not been done. The operators FROM $L_1$ that factor through $\ell_1$ are the differentiable operators (D.R. Lewis and C. Stegall, JFA 12 (1973)).

  • $\begingroup$ Since you mention the Lewis-Stegall paper, I presume that what you call "differentiable operators" are the representable operators, that is, those $T:L_{1}(\Omega)\to B$ for which there is a measurable a.e.-bounded $g:\Omega\to B$ such that $Tf= \int gf$, correct? $\endgroup$ Commented Aug 10, 2010 at 16:11
  • 1
    $\begingroup$ Well, yes, but differentiable usually refers to an equivalent form. One of the other nice equivalences is that every set of positive measure contains a smaller set of positive measure s.t. the restriction of the operator to the functions supported on the smaller set is a compact operator. $\endgroup$ Commented Aug 10, 2010 at 18:13
  • $\begingroup$ Just thought I'd mention that if the OP doesn't have access to a library carrying a hardcopy, or know someone who owns one, then (s)he's unlikely to find N. Tomczak-Jaegermann's book, as it's near impossible to get hold of a copy any other way. But obviously that won't matter if (s)he can get hold of Diestel, Jarchow and Tonge's book, much of which can be viewed on Google Books: books.google.com.au/… $\endgroup$ Commented Aug 10, 2010 at 21:09

In the paper R. Cilia and J. M. Gutierrez, Left $\ell_1$-factorable polynomials, Glasgow Math. J. 51 (2009), 631-649 you may find some results on operators factoring through $\ell_1$


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.