9
$\begingroup$

We know that for a sequence of measurable functions, $L^p$ convergence implies convergence in measure. However, the converse is false. I'm having trouble coming up with a counter-example, please help! Also, does convergence in measure differ from a.e. pointwise convergence?

$\endgroup$
  • 2
    $\begingroup$ Convergence in measure does not imply pointwise convergence almost everywhere. One counterexample is a sequence of characteristic functions of intervals of length converging to 0 such that every point is both in and out of infinitely many intervals. For instance, the intervals could be $[0,1/2],[1/2,1],[0,1/4],[1/4,1/2],...$. $\endgroup$ – Kevin Carlson Oct 3 '12 at 22:15
8
$\begingroup$

Convergence almost everywhere implies convergence in measure for finite measure space. You can see this by applying the dominated convergence theorem to the integral of the function $\chi_{\{|f_n - f|>\epsilon\}}$ (which is the measure of the set $\{|f_n - f|>\epsilon\}$), which is dominated by $1$ on a finite measure space. On an infinite measure space, this need not hold. An example where it fails is $f_n (x) = \chi_{[n,\infty)}(x)$ which converges to zero pointwise everywhere, but for $\epsilon < 1$ we have $\mu\{x: |f_n(x)|>\epsilon\} = \infty$ for any $n$, so it cannot converge in measure.

An example of a function which converges to zero in measure but not in $L^1$ is $f_n = n \chi_{[0,\frac{1}{n}]}$, since $\int |f_n| = 1$ for each $n$, but $\mu\{|f_n| > \epsilon\} = \frac{1}{n}$ for $\epsilon$ small.

$\endgroup$
3
$\begingroup$

Here is a counterexample such that we have convergence in measure, but not in $L^p$. Take $f_n=2^{-n}1_{[2^n,2^{n+1}]}$, then $|\{f_n>\epsilon\}|\rightarrow 0$. Hence $f_n\rightarrow 0$ in measure, but $||f_n||_{L^p}=1$ for all $n$.

EDIT: My previous counterexample was not correct, and I made a false statement about convergence a.e. I have corrected my counterexample, and removed the false statement.

$\endgroup$
  • 1
    $\begingroup$ Surely $m \{x | |f_n(x) - 0| \geq \epsilon \} = 1$ for all $\epsilon > 0$? Maybe $f_n = n 1_{[n,n+\frac{1}{n}]}$? Then $f_n(x) \to 0 $ a.e., $\|f_n\|_1 = 1$, and $f_n$ converges to $0$ in measure. $\endgroup$ – copper.hat Oct 3 '12 at 22:22
  • $\begingroup$ Thank you very much. I was sloppy, and too quick to answer. I mostly work in probability spaces, hence finite measure spaces, so I did not proof my work correctly. My new answer should be correct. Sorry for the confusion. $\endgroup$ – Carl Morris Oct 3 '12 at 23:29

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.