I am trying to understand a step in the following proof of completeness of $L^p$ in Stein-Shakarchi's Functional Analysis. (See the proof on page 5 of the link or at the end of this post.)

At the beginning of the proof, it is said that

Let $\{f_n\}_{n=1}^\infty$ be a Cauchy sequence in $L^p$, and consider a subsequence $\{f_{n_k}\}_{k=1}^\infty$ of $\{f_n\}_{n=1}^\infty$ with the following property $\|f_{n_{k+1}}-f_{n_k}\|\le 2^{-k}$ for all $k\geq 1$.

Question: Why can the sequence be considered as it is?

On a YouTube video, it explains about a similar subsequence.

I still don't understand why for $n,m>n_k$ $\Vert f_{n}-f_{m}\Vert_p\implies \Vert f_{n_k}-f_{n_{k+1}}\Vert_p$, thus an increasing subsequence. Why is it justified to make $n$ to depend on $k$, $n_k$?

enter image description here enter image description here enter image description here

  • $\begingroup$ What happened to all the $\sum$s? $\endgroup$ – Lord Shark the Unknown May 19 at 18:22
  • $\begingroup$ I don't know, I don't know why the author omitted them all; $\endgroup$ – galleta May 19 at 18:23
  • 1
    $\begingroup$ Basically the trick here is to replace the Cauchy sequence by a subsequence that converges really quickly. $\endgroup$ – Lord Shark the Unknown May 19 at 18:23
  • $\begingroup$ @LordSharktheUnknown What kind of convergence? $\endgroup$ – galleta May 19 at 18:55

The authors mention at the beginning of the proof that

The argument is essentially the same as for $L^1$ (or $L^2$); see Section 2, Chapter 2 and Section 1, Chapter 4 in Book III.

It is said clearly there (see also a snapshot at the end) that

The existence of such subsequence is guaranteed by the fact that $\|f_{n}-f_{m}\|\leq \epsilon$ whenever $n,m\geq N(\epsilon)$, so that it suffices to take $n_k=N(2^{-k})$.

[Added for elaboration.]

There exists an integer $N(2^{-1})>0$ such that for all $n,m\geq N(2^{-1})$, $$ \|f_{n}-f_{m}\|\leq 2^{-1}\tag{1}. $$ There exists an integer $N(2^{-2})>N(2^{-1})$ such that for all $n,m\geq N(2^{-2})$, $$ \|f_{n}-f_{m}\|\leq 2^{-2}\tag{2}. $$ There exists an integer $N(2^{-3})>N(2^{-2})$ such that for all $n,m\geq N(2^{-3})$, $$ \|f_{n}-f_{m}\|\leq 2^{-3}\tag{2}. $$ ... so on and so forth.

Now, let $n_1=N(2^{-1})$, $n_2=N(2^{-2})$, $n_3=N(2^{-3})$, $\cdots$.

Since $n_1,n_2\geq N(2^{-1})$, we have by (1) $$ \|f_{n_2}-f_{n_1}\|\leq 2^{-1}. $$

Since $n_2,n_3\geq N(2^{-2})$, we have by (2) $$ \|f_{n_3}-f_{n_2}\|\leq 2^{-2}. $$

Since $n_3,n_4\geq N(2^{-3})$, we have by (3) $$ \|f_{n_4}-f_{n_3}\|\leq 2^{-3}. $$

... so on and so forth.

The following is a snapshot of the beginning of the proof for completeness of $L^1$ in Stein-Shakarchi's Book III (page 70 Theorem 2.2).

enter image description here

  • $\begingroup$ Why can we take the sequence as an increasing sequence? $\endgroup$ – galleta Jun 8 at 16:40
  • $\begingroup$ What do you mean by "increasing sequence"? Are you asking why $n_1<n_2<n_3<\cdots$? $\endgroup$ – Jack Jun 8 at 17:51
  • $\begingroup$ See my edit. Not that $\{f_{n_k}\}_{k=1}^\infty$ is NOT necessarily an increasing sequence. In fact one may talk about a sequence of complex-valued functions and thus no such thing as "increasing sequence" here. One can say that the subscripts $\{n_k\}_{k=1}^\infty$ form an increasing sequence though. $\endgroup$ – Jack Jun 8 at 18:07
  • $\begingroup$ Got it Jack thanks. Do we take this W.L.O.G $N(2^{-2})>N(2^{-1})$? Because $2^{-2}<2^{-1}$ actually $\endgroup$ – galleta Jun 8 at 19:44
  • 1
    $\begingroup$ Well, I would not use "WLOG" there. Since $2^{-2}<2^{-1}$, we are taking a smaller $\epsilon$. On the other hand, we want to pick one positive integer $N(2^{-2})$ such that (i) $N(2^{-2})>N(2^{-1})$ (ii) for all $n,m\geq N(2^{-2})$, the inequality (2) holds. There are of course lots of other choices of $N(2^{-2})$ that would work, but we only need one such $N(2^{-2})$. $\endgroup$ – Jack Jun 8 at 20:00

If $\{f_n\}_n$ is Cauchy, for each $k\in\Bbb N$ we can find an $N_k$ (depending on the $k$ we just chose) such that

$$ \|f_n - f_m\| < 2^{-k} \quad\text{ whenever } n,m\geq N_k.$$

For $k+1$, we likewise get

$$ \|f_n - f_m\| < 2^{-k-1} \quad\text{ whenever } n,m\geq N_{k+1}.$$

Now, both inequalities will hold for $n,m \geq \max\{N_k,N_{k+1}\}$, so we can choose $N_{k+1} > N_k$, and similarly we can choose an $N_{k+2} > N_{k+1} > N_{k}$ for $2^{-k-2}$ so that

$$\|f_{N_k} - f_{N_{k+1}}\| < 2^{-k}\quad\text{and}\quad \|f_{N_{k+1}} - f_{N_{k+2}}\| < 2^{-k-1}$$

by choosing $n = N_k$ and $m = N_{k+1}$, etc in the first inequalities above.

  • $\begingroup$ Thank you. Does the subsequence $f_{N_k}$ converge pointwise? $\endgroup$ – galleta May 19 at 23:24
  • 1
    $\begingroup$ You're welcome :) And no, $L^p$ convergence does not imply pointwise convergence in general. $\endgroup$ – ryan221b May 19 at 23:26
  • $\begingroup$ it does not hold in general then there exist the possibility of $f_{N_k}$ to converge pointwise? $\endgroup$ – galleta May 19 at 23:40
  • $\begingroup$ Yes, if you take all the functions to be equal, for example. Or, for continuous functions in $L^1([0,1])$, uniform convergence implies both pointwise and $L^1$ convergence, so both can occur simultaneously. $\endgroup$ – ryan221b May 20 at 0:00
  • $\begingroup$ On this $\|f_{N_k} - f_{N_{k+1}}\| < 2^{-k}\quad\text{and}\quad \|f_{N_{k+1}} - f_{N_{k+2}}\| < 2^{-k-1}$ what is the whenever condition? $\endgroup$ – galleta Jun 6 at 19:47

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.