Let $X$ be the interval $[0,1]$ with Lebesgue measure. Is there a function $f\in L^p(X)$ for all $p\in[1,\infty)$ that is not $\in L^\infty(X)$? If so, what is an example?

Motivation: In a course on measure theory this fall, I've learned proofs that $L^p(X)\supset L^q(X)$ if $q>p$ and that if $f\in L^\infty(X)$, then $\|f\|_\infty = \lim \limits_{p\to\infty} \|f\|_p$. This prompted me to wonder if $L^\infty(X) = \bigcap _p L^p(X)$. A classmate gave me a general theoretical reason to believe the contrary: $L^p(X)$ is a reflexive space for $1 \lt p \lt \infty$ but not for $p=1,\infty$; but intersections of reflexive spaces are reflexive. This logic seems sound to me; but it implies the containment $L^\infty(X) \subset \bigcap_p L^p(X)$ is strict. If so, there must be a function that is $L^p$ for all $p$ but not a.e. bounded. What is it?

  • 3
    $\begingroup$ There is a flaw in your classmate's theoretical argument: If you have closed reflexive subspaces of a Banach space then their intersection is reflexive (simply because closed subspaces of reflexive spaces are reflexive). However, $L^q(X)$ is dense in $L^p(X)$ for $ \infty \geq q \gt p \geq 1$, so you cannot conclude. $\endgroup$ – t.b. Dec 12 '11 at 3:27
  • $\begingroup$ To complement Jonas's answer below: if you merely want a concrete example on some probability space $(\Omega,\mu)$, then a Gaussian random variable, viewed as a function $\Omega\to{\mathbb R}$, is in every $L^p(\Omega,\mu)$ -- it ``has moments of all orders'' -- but is not a.s. bounded. $\endgroup$ – user16299 Apr 4 '12 at 6:40

The logarithm is an example.

To see that $\log$ is in $L^p$ for each $p>0$, it suffices to see that $|\log(t)|^p<t^{-1/2}$ for $t$ sufficiently small. This follows from the fact that $\lim\limits_{t\searrow0}\;|\log(t)|t^{1/(2p)}=0$, which is a quick application of l'Hôpital's rule.

The $L^p$ norm of $\log$ happens to be $\Gamma(p+1)^{1/p}$. This can be seen by making the change of variables $t=-\log(u)$ in the integral $\Gamma(x)=\int\limits_{0}^\infty t^{x-1}e^{-t}\:dt$. So by Stirling's approximation, $\|\log\|_p$ is close to $p/e$ when $p$ is large.


For another example, observe that the series $$\sum_{n=1}^\infty n^p\frac{1}{2^n}$$ converges for any $p$ from root test since $\limsup_{n\rightarrow \infty} \left(\frac{n^p}{2^n}\right)^{1/n} = \frac{1}{2} < 1$.

You can define the function $$f(x) = \sum_{n=1}^\infty n\chi_{(2^{-n}, 2^{-n+1}]}(x)$$ as an example.

  • 2
    $\begingroup$ +1 but note that morally this is the same example as Jonas' - observe that your $f$ is bounded by constant multiples of $-\log x$. $\endgroup$ – Ben Blum-Smith Jan 23 '16 at 22:52
  • $\begingroup$ But how do we know that $\displaystyle \left(\sum_{n=1}^\infty \frac{n}{2^n} \right)^p= \sum_{n=1}^\infty \frac{n^p}{2^n}$ $\endgroup$ – Twink Apr 11 '20 at 23:03
  • $\begingroup$ @Twink the two terms are not equal, if you compute the $L^p$ norm of $f$, you get $\sum_{n=1}^\infty \frac{n^p}{2^n}$. $\endgroup$ – Xiao Apr 12 '20 at 16:06

Xiao and Jonas Meyer give good answers, but there maybe somethings wrong with your motivation:

$L^p(X)\supset L^q(X)$ if $q<p$.

Because I saw a proposition in Folland's Real Analysis which on P186

Suppose $0<p<q<\infty$. Then $L^p \nsubseteq L^q $ iff $X$ contains sets of arbitrarily small positivite measure, and $L^q \nsubseteq L^p $ iff $X$ contains sets of arbitrarily large finite measure.

So, in general we have $L^p \nsubseteq L^q$ for all $p\neq q$.

  • $\begingroup$ This should be left as a comment to the original post, not an answer. $\endgroup$ – Alex Dec 19 '19 at 11:27

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.