Let $f_k: \mathbb{R} \rightarrow \mathbb{R} , f_k(x) = \frac{1}{\sqrt{x}}\chi_{\left[\frac{1}{2^{k+1}},\frac{1}{2^k}\right]}(x)$

For $k \rightarrow \infty $ the interval $[\frac{1}{2^{k+1}},\frac{1}{2^k}]$ will decrease it's range, but I would say that for every $k \in \mathbb{N}$ there exists an $x \in \mathbb{R}$ so that $x \in [\frac{1}{2^{k+1}},\frac{1}{2^k}]$.

Can anyone tell me the function $f \leftarrow f_k$ to which the sequence converges and how? Pointwise of course, even of course not. But in measure or in $L^1, L^2 ,...$?


You need to fix $x>0$. Then there is $K$ such that $\frac{1}{2^k}<x$, for $k>K$. Then $f_k(x)=0$, for $k>K$. Then $f(x)=0$.

For $x\leq0$, $f_k(x)=0$, for all $k$. Then $f(x)=0$.

Now, for $L^1$ we can integrate.


This tells us that $||f_k-0||_1\to0$.

We can do the computation in general, in $L^p$:

$$||f_k||_p^p=\int_{1/2^{k+1}}^{1/2^{k}}x^{-p/2}=\begin{cases}\left.\frac{x^{1-p/2}}{1-p/2}\right|_{1/2^{k+1}}^{1/2^k}=\frac{1}{1-p/2}\left(\frac{1}{2^{k(1-p/2)}}-\frac{1}{2^{(k+1)(1-p/2)}}\right)&\text{ for }p\neq2\\\log(x)|_{1/2^{k+1}}^{1/2^k}=\log(1/2^k)-\log(1/2^{k+1})=\log(2)&\text{ for }p=2\end{cases}$$

The difference is now, that if $1-p/2\leq0$ then the limit doesn't go to zero.

We know that for a convergent sequence in $L^p$ we can always extract a subsequence that converges pointwise a.e. Therefore if it were to converge it would converge to $f=0$. Therefore, for $1-p/2\leq0$ it doesn't converge. For $1-p/2>0$ it tends to $f=0$ as before.

  • $\begingroup$ Wow, that was a quick sufficiant answer, thank you very much! And I guess I can assume that $f_k$ converges in measure, because $||f_k - f|| = ||f_k|| \rightarrow 0$, right? Or does that follow directly out of $L^1$ convergence? $\endgroup$ – RoyPJ Jan 28 '15 at 14:49
  • 1
    $\begingroup$ The set of points where $|f_k(x)-f(x)|>\epsilon$ is the interval $[1/2^{k+1},1/2^k]$. Its size is $\frac{1}{2^k}-\frac{1}{2^{k+1}}\to0$. $\endgroup$ – Pp.. Jan 28 '15 at 14:51

Your sequence is going to zero pointwise. On the one hand, $f_k(0) \equiv 0$. On the other hand, if $x>0$ and $k>-\log_2(x)$, then $f_k(x)=0$.

Additionally, your sequence is dominated by $f(x) = \frac{1}{\sqrt{x}} \chi_{[0,1]}(x)$, which is in $L^p$ for $p \in [1,2)$. So it is definitely going to zero in $L^p$ for $p \in [1,2)$. It does not go to zero in $L^2$: by direct calculation we have $\| f_k \|_2^2=\int_{2^{-k-1}}^{2^-k} \frac{1}{x} dx=\ln(2^{-k})-\ln(2^{-k-1})=\ln(2)$. This means it does not go to zero in any $L^p$ for $p>2$ either (because on a finite measure space, convergence in $L^{p_1}$ implies convergence in $L^{p_2}$ if $p_2 \leq p_1$).


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.