Analysis Problem of the Day 6

Today’s problem is a commonly appearing exercise in the theory of L^p spaces.

Problem: Let X be a finite measure space, and let f \in L^p(X) for all 1 \leq p \leq \infty. Show that \lim_{p \to \infty} \|f\|_p = \|f\|_\infty.


Solution: Let’s first understand intuitively why this might be true. As p \to \infty, the parts where |f|>1 start contributing more to the L^p-norm of f (since a^p gets very large for large p and |a|>1,) while the parts where |f|<1 contribute less and less to the norm. Thus, we would expect that in the limit as p \to \infty, the part that will contribute to the norm is the largest value of f, or in other words, the essential supremum \|f\|_\infty. It also suggests that for this problem, instead of thinking horizontally, that it is instead helpful to think vertically, i.e. in terms of the sets \{x \in X: |f(x)|>\lambda\}.

Now for a formal proof. We prove the existence and value of the limit by showing both \liminf_{p \to \infty} \|f\|_p \leq \|f\|_\infty and \limsup_{p \to \infty} \|f\|_p \geq \|f\|_\infty. By Hölder’s inequality with f and 1_X, we have

    \[\|f\|_p \leq \mu(X)^{\frac{1}{p}-\frac{1}{q}} \|f\|_q\]

for 1 \leq p < q \leq \infty. In particular, for q = \infty, we get that \|f\|_p \leq \mu(X)^{\frac{1}{p}}\|f\|_\infty. Taking the limsup on both sides yields \limsup_{p \to \infty} \|f\|_p \leq \|f\|_\infty.

For the other direction, we recall the decomposition

    \[\int |f|^p dx = \int_{|f|>\lambda} |f|^p dx + \int_{|f| \leq \lambda} |f|^p dx \geq \lambda^p \mu(\{x:|f|>\lambda\}).\]

Now, the key here is that by definition, \|f\|_\infty = \sup\{\lambda: \mu(|\{x:|f|>\lambda\}) >0\}. Thus, for any \epsilon>0, let \lambda = \|f\|_\infty - \epsilon. Taking p-th roots on both sides yields

    \[\|f\|_p \geq (\|f\|_\infty-\epsilon)\mu(\{x:|f|>\|f\|_\infty-\epsilon\})^{\frac{1}{p}}.\]

Taking the \liminf as p \to \infty on both sides yields \liminf_{p \to \infty}\|f\|_p \geq \|f\|_\infty - \epsilon, and since \epsilon>0 is arbitrary, we obtain

    \[\limsup_{p \to \infty} \|f\|_p \leq \|f\|_\infty \leq \liminf_{p \to \infty} \|f\|_p,\]

so we conclude that \lim_{p \to \infty} \|f\|_p = \|f\|_\infty.

Remark 1: By interpolation theorems, it suffices to take f \in L^p \cap L^\infty for any p<\infty for the statement of the problem to hold, since then f \in L^q for any q>p.

Remark 2: This result still holds true if \mu(X)=\infty. The only issue in our argument is now the application of Hölder’s, which only works on finite measure spaces. In this case, we can instead use the following trick, taking r to be large enough so that f \in L^r:

    \[\|f\|_p^p = \|f^r f^{p-r}\|_1 \leq \|f^{p-r}\|_\infty \|f^{r}\|_1 = \|f\|_\infty^{p-r} \|f^r\|_1,\]

so taking p-th roots on both sides and taking the \limsup as p \to \infty yields

    \[\limsup_{p \to \infty} \|f\|_p \leq \|f\|_\infty.\]

Leave a Reply

Your email address will not be published. Required fields are marked *