The following looks like a strengthening of the approximation property, but I don't know, maybe this is equivalent. I would be grateful if somebody could explain this.

Let $X$ be a Banach space with the approximation property, and $K$ a compact set in $X$.

Is it true that for each $\varepsilon>0$ there exists an (bounded) operator of finite rank $T:X\to X$ such that

1) $T$ approximates the identity operator on $K$: $$ \forall x\in K\qquad ||Tx-x||<\varepsilon, $$ and

2) the range of $T$ belongs to the linear span of $K$ $$ T(X)\subseteq\text{span}\ K $$ ?

EDIT. Yes, I see that this is not equivalent, because otherwise we obtain that the (usual) approximation property is inherited by closed subspaces. OK, I change a bit my question:

Is anything known about this variant of the approximation property?

  • $\begingroup$ This sounds to me as this is mainly or possibly about retraction(?). $\endgroup$ – Włodzimierz Holsztyński Nov 12 '16 at 5:20

Since every separable closed subspace of $X$ is the closed linear span of a compact set, your property is equivalent to the HAP (hereditary approximation property).

Approximate $K$ by a finite rank bounded linear operator $T$ on the closed linear span of $K$. A small perturbation argument shows that WLOG $T$ ranges in the linear span of $K$. Since $T$ has finite rank, $T$ can be extended to bounded linear operator on $X$ that has the same range.

If $X$ is non separable, use the fact that if every separable subspace of $X$ has the AP (or is even contained in a subspace that has the AP), then $X$ has the AP.

Spaces with the HAP are rare but there are some that are not isomorphic to Hilbert spaces. I constructed the first one in the 1970s and Pisier's weak Hilbert spaces are a nice class of such spaces. More recent work on them is contained in my papers with Figiel and with Szankowski.


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.