3

Let $p\in[1,\infty]$, $a\in\ell^\infty$ and $$T:\ell^p\to\ell^p\;,\;\;\;x\mapsto ax.$$ Note that $\left\|T\right\|=\left\|a\right\|_{\ell^\infty}$. It's easy to see that if $a\in\ell^1$, then $T$ is nuclear and hence compact.

Can we more generally conclude that $T$ is compact?

Assume $p<\infty$. The intuitive approach would be to define $$T_n:\ell^p\to\ell^p\;,\;\;\;x\mapsto(a_1x_1,\ldots,a_nx_n,0,\ldots)$$ for $n\in\mathbb N$ and to show that $\left\|T_n-T\right\|\to0$. Now, $$\left\|(T_n-T)x\right\|^p=\sum_{i=n+1}^\infty|a_ix_i|^p\tag1$$ for all $x\in\ell^p$ and $n\in\mathbb N$. The right-hand side of $(1)$ clearly converges to $0$ as $n\to\infty$, but this only yields convergence in the strong operator topology.

On the other hand, if $(a_n)_{n\in\mathbb N}$ would be nonincreasing and $a_n\to0$ (e.g. $a_n=\frac1n$), then $$\left\|(T_n-T)x\right\|^p\le|a_{n+1}|^p\left\|x\right\|_{\ell^p}^p\tag2$$ for all $x\in\ell^p$ and $n\in\mathbb N$ and we could immediately conclude $\left\|T_n-T\right\|\to0$.

EDIT: Maybe the general result can be obtained from the convergence in strong operator topology (which holds by $(1)$) by noting that $\ell^p$ has the approximation property; but I'm not sure whether this is true.

0xbadf00d
  • 13,422
  • Take $a$ to the be the sequence $a_{n} = 1$. Then $T$ is the identity, which cannot be a compact operator unless the underlying space is finite dimensional. –  Feb 05 '21 at 12:58

3 Answers3

3

First of all, you cannot conclude that an operator is compact if it is the strong limit of finite rank operators, for example in any Hilbert space the identity operator is the strong limit of a net of finite-rank projections, but of course it is compact if and only if the dimension of our space is finite.

As Timur Bakiev says, the multiplication operator $T:\ell^p\to\ell^p$ is compact if and only if the sequence of $\ell^\infty$ that acts on $\ell^p$ belongs to $c_0$.

You already showed the converse: In your notation, $$\|(T_n-T)x\|^p=\sum_{k=n+1}^\infty|a_k|^p\cdot|x_k|^p\leq\sup_{k\geq n+1}|a_k|^p\cdot\|x\|^p$$ so $\|T_n-T\|\leq\sup_{k\geq n+1}|a_k|\to0$.

Conversely, suppose that $T$ is compact. We will use the following lemma from real analysis: if $(t_n)$ is a sequence of real numbers and every subsequence of $(t_n)$ has a further subsequence converging to $t$, then $(t_n)$ converges to $t$. See this post for example: Every subsequence of $x_n$ has a further subsequence which converges to $x$. Then the sequence $x_n$ converges to $x$.

By definition of compactness, if $(x_n)\subset\ell^p$ is a bounded sequence, $(Tx_n)$ has a convergent subsequence in $\ell^p$. Fix indexes $n_1<n_2<n_3<\dots$ and consider the elements $e_k\in\ell^p$ having $0$ everywhere and $1$ in the $n_k$ position. Then $\|e_k\|=1$, so this sequence is bounded and therefore $(Te_k)$ has a convergent subsequence in $\ell^p$. To simplify my notation, I will assume that $n_1=1, n_2=2,\dots$ but the method below can be applied to arbitrary subsequences too, you can verify this yourself easily.

So as I said, let $e_n$ be the sequence having $0$ everywhere and $1$ in the $n$-th position, so $\|e_n\|=1$ for all $n$. Note that $Te_n$ is the sequence that has $0$ everywhere and $a_n$ in the $n$-th position, so $\|Te_n\|_p=|a_n|$. Now $(Te_n)$ has a convergent subsequence, say $(Te_{n_k})$ and say that $Te_{n_k}\to y\in\ell^p$, thus $\|Te_{n_k}-y\|_p^p\to0$. If $y=(y_i)$, then $$\|Te_{n_k}-y\|_p^p=\sum_{i=1}^{n_k-1}|y_i|^p+|a_{n_k}-y_{n_k}|^p+\sum_{i=n_k+1}^\infty|y_i|^p$$ Fix an integer $m$. For $\varepsilon>0$ choose $k>m$ so large so that $\|Te_{n_k}-y\|_p^p<\varepsilon$.

Then $|y_m|^p\leq\sum_{i\neq n_k}|y_i|^p+|a_{n_k}-y_{n_k}|^p=\|Te_{n_k}-y\|_p^p<\varepsilon$ and as $\varepsilon>0$ was arbitrary, we conclude that $y_m=0$, so $y=0$ since $m$ was arbitrary before fixing. We conclude that $|a_{n_k}|\to0$.

Now as I said, this can be done for any subsequence (this is why I fixed $n_1<n_2<\dots$ in the beginning, nevermind that I eventually went wLOG to the case $n_1=1,n_2=2\dots$), so the lemma will yield that $|a_n|\to0$, i.e. $a_n\in c_0$.

  • Thank you for your answer. You've only consider the case $p<\infty$. Does the same claim hold for $p=\infty$? – 0xbadf00d Feb 06 '21 at 14:30
  • @0xbadf00d I think so. I think the same argument can apply in the same way, of course you must change sums by suprema. If there is a point where it goes wrong, let me know and I can add some details. I believe this answers the question entirely. – Just dropped in Feb 06 '21 at 14:39
  • Yes, I didn't had the time to check the answer before. – 0xbadf00d Feb 07 '21 at 05:26
1

The set of all compact operators between Banach spaces is closed in norm topology.

A multiplication operator on $l^p$ is compact if and only if the corresponding sequence is an element of $c_0$.

Timur Bakiev
  • 2,215
1

Here's another proof of necessity. Let $T_\alpha$ be a multiplication operator and $\alpha = \{\alpha_n\} \in l^\infty$ be the corresponding sequence.

Let $\alpha \notin c_0$. Then there exists $\varepsilon > 0$ s.t. for any $N \in \mathbb{N}$ we can find $n>N$ for which $|a_n| > \varepsilon$. Hence we can assume $|a_n| > \varepsilon$ for all $n \in \mathbb{N}$. This assumption is made for simplicity, but if we take the subspace of $l^p$ corresponding to the chosen subsequence, it will be isometrically isomorphic to $l^p$. A restriction of $T_\alpha$ to this subspace is again a compact operator. Then $T_\alpha(\mathbb{B}_1) \supset \mathbb{B}_\varepsilon$ ($\mathbb{B}_1$ is a unit ball).

Indeed, let $x \in l^p$, $||x|| \geq 1$. Then $$ ||T_\alpha x||_p = \sqrt[p]{\sum_{n=1}^\infty |\alpha_n x_n|^p} > \sqrt[p]{\sum_{n=1}^\infty \varepsilon^p |x_n|^p} = \varepsilon ||x||_p \geq \varepsilon, $$ and the statement follows since our operator is surjective.

But $\mathbb{B}_\varepsilon$ is not relatively compact in infinite dimensional space. Hence $T_\alpha$ is not compact.

Timur Bakiev
  • 2,215
  • I don't think you can assume that, consider $a_{2n+1}=1$, $a_{2n}=1/n$. – Just dropped in Feb 05 '21 at 18:23
  • @JustDroppedIn The main idea is the image is infinite dimensional. This assumption is made for simplicity, but if we take the subspace of $l^p$ corresponding to the chosen subsequence, it will be isomerically isomorphic to $l^p$. So, what’s the problem? – Timur Bakiev Feb 05 '21 at 18:32
  • No problem. I am just saying that we cannot assume that without harm to the generality. I understand that the main point is elsewhere, but I still dont see how the ball of radius epsilon is contained in the image. – Just dropped in Feb 05 '21 at 18:36
  • @JustDroppedIn Well, it’s a good question. I will show it it my answer explicitly. The key idea is that our operator is bounded below. – Timur Bakiev Feb 05 '21 at 18:39
  • So now you are showing that $T_\alpha(\ell^p\setminus B_1)\subset \ell^p\setminus B_\varepsilon$, right? – Just dropped in Feb 05 '21 at 19:51
  • @JustDroppedIn yes, exactly – Timur Bakiev Feb 05 '21 at 19:53
  • and why is this equal to $\ell^p\setminus T_a(B_1)$? Sorry, I can't see it and I can't see how we can move to the general case that we talkeed about earlier. Anyway, maybe I am too tired today! – Just dropped in Feb 05 '21 at 19:58
  • JustDroppedIn Sorry, I don't quite understand the first question, but to clarify the second one, I've made a small edit. – Timur Bakiev Feb 05 '21 at 20:10