Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Note this is a homework problem so I am looking for a hint not a solution:

For normed linear spaces $X$ and $Y$, I'm trying to show that $K(X,Y)$, the set of compact operators $X\to Y$ is a closed subset of $B(X,Y)$ the set of bounded operators $X\to Y$.

At first I thought it might be similar to showing that $c_{0}$ is a closed subspace of $c$. But the standard argument for that (if I am not mistaken) relies on the fact that the scalar field is complete.

Note: It turns out that $Y$ must be complete in order for the result to be true.

I start by assuming $f_{n}\in K(X,Y)$ is compact, and that $f_{n}\to f$ for some $f\in B(X,Y)$.

I want to show $f\in K(X,Y)$ using the criterion that for every sequence $x_{n}\in B_{X}$, $f(x_{n})$ has a convergent subsequence.

For each $m \geq 1$, by the compactness of $f_{m}$, there is a subsequence $x_{n_{k}}$ such that $f_{m}(x_{n_{k}})$ is convergent to some value in $Y$, say $y_{m}$.

If the same subsequence served as an appropriate witness for each $m\geq 1$, I think I may be able to get somewhere by changing orders of limits using an upper bound for the sequence $(f_{n})$. But I doubt this is the case, and thus I am stuck.

share|cite|improve this question
I think it would be easier to show that $f(B_X)$ is totally bounded. To do this, let $\epsilon>0$, then choose $N$ so that $\Vert f_N-f\Vert<\epsilon$. Choose a finite $\epsilon$-net, $\{f_N(x_1),\ldots , f_N(x_n)\}$ for $f_N(B_X)$. Then show this is a $3\epsilon$-net for $f(B_X)$. – David Mitra Jan 10 '13 at 18:12
Thank you. I will try this. – roo Jan 10 '13 at 18:26
I didnt use the concept of totally bounded, but I applied your choice of $N$ and immediately got the convergent subsequence I require. Thanks again! – roo Jan 10 '13 at 18:47
If you define a compact operator as one that maps bounded sets to sets with compact closure, then I'm not sure this is true. With $Y$ Banach, this is a standard result. But for arbitrary $Y$ I don't see how to get anything stronger than the result that $f$ maps bounded sets to totally bounded sets. What is your definition of a compact operator? – David Mitra Jan 10 '13 at 21:25
As you suggested, in my course an operator $T:X\to Y$ is compact if $\overline{T(B_{X})}$ is norm compact in $Y$. The issue you mention is actually why I avoided the totally bounded approach. It seems easy to go from compact to totally bounded, but not the other way around. Some google-browsing reveals that the other direction definitely holds in metric spaces but these facts are not available to me. – roo Jan 10 '13 at 21:30
up vote 1 down vote accepted

A sequential argument that works when $Y$ is a Banach space is given below:

Hint: We take take subsequences of subsequences and diagonalize.

Below are the details:

Let $X_1=(x_n)$ be a sequence in $B_X$. Choose a subsequence $X_2=(x_n^1)_n$ of $X_1$ such that $f_1(x_n^1)$ is convergent. Now choose a subsequence $X_3=(x_n^2)_n$ of $X_2$ such that $f_2(x_n^2)$ converges. Continue in this manner...

We thus have subsequences $$ (x_n^1)\supset (x_n^2)\supset (x_n^3)\supset\cdots $$so that $(f_m x_n^m)_n$ is convergent for each $m$.

Now set $y_n=x_n^n$. Then $(f_m y_n)_n$ is a convergent sequence for each $m$.

We also have $$\eqalign{\Vert f(y_n)-f(y_l)\Vert &\le\Vert (f-f_m)(y_n)\Vert +\Vert f_m(y_n-y_l)\Vert+\Vert (f_m-f)(y_l)\Vert \cr &\le 2\Vert f-f_m\Vert +\Vert f_m(y_n-y_l)\Vert } $$ It follows from the above that $(f(y_n))_n$ is a Cauchy sequence and, thus, convergent.

I'm not sure how to make the above argument go through when $Y$ is only assumed to be a normed space; however, you can show that $f(B_X)$ is totally bounded as in my comment above.

share|cite|improve this answer
This looks like the proof of Arzela-Ascoli (or some version of it I saw in an ODE's course). $Y$ is only a Normed space, not a Banach space, so this may not work in general. But I used your first hint in the comment above and the problem worked out fine. – roo Jan 10 '13 at 18:52
Unaccepted as you asked, but deleting it doesn't seem right as it is still valuable. – roo Jan 10 '13 at 21:25
Reaccepting this as it turns out completeness of $Y$ was a missing hypothesis. – roo Jan 12 '13 at 19:37

The following is incorrect:

Let $x_{n}$ be a sequence in $B_{X}$. Choose $N\geq 1$ so that $\|f_{N} - f\| < \epsilon$.

Choose a subsequence $x_{n_{k}}$ such that $f_{N}(x_{n_{k}})\to y$ for some $y\in Y$, and then $K\geq 1$ such that $\|f_{N}(x_{n_{K}}) - y\| < \epsilon$.

Then $\|f(x_{n_{K}}) - y\| \leq \|f(x_{n_{K}}) - f_{N}(x_{n_{K}})\| + \|f_{N}(x_{n_{K}}) - y\| < 2\epsilon$.

This allows arbitrarily close approximation of $y$ by elements of the sequence $f(x_{n})$, which gives the subsequence which is needed.

share|cite|improve this answer
But, for different $\epsilon$, and thus different $N$, how do you know you can use the same limit $y$? Or am I missing something? (Also, would you unaccept my answer? I'd like to delete it as I misread (or rather didn't read carefully) your question.) – David Mitra Jan 10 '13 at 19:20
You're absolutely right. I was jumping to conclusions. Thanks for pointing out my error. – roo Jan 10 '13 at 21:25
I think I can take smaller and smaller values of $\epsilon$, and form a sequence from the resulting $y$ values, and then cross my fingers that it has a convergent subsequence. I think this will fix it. – roo Jan 10 '13 at 21:49

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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