I recently found out that $f \in L^{p}(\mathbb R)$ does not necessarily imply that $\lim\limits_{\vert x \vert \to \infty} f(x)=0$. Take for example, $f(x)=\begin{cases} n, \text{if } x \in [n,n+\frac{1}{n^{3}}]\\ 0, \text{if } x \in ]n+\frac{1}{n^{3}},n+1[\end{cases}$, $n \in \mathbb N$ it is clear that $\lim\limits_{\vert x \vert \to \infty} f(x)$ does not exist and further $\vert\vert f\vert \vert_{p}=\sum\limits_{n \in \mathbb N}\frac{1}{n^{2}}<\infty$.

But in my proof to show that $f_{n}\xrightarrow{L^{p}([-N,N])} f$ on every $N \in \mathbb N$ and $\vert \vert f_{n} \vert \vert_{p}\xrightarrow{n \to \infty}\vert\vert f\vert \vert$ implies that $f_{n}\xrightarrow{L^{p}(\mathbb R)} f$

In this proof I used the fact that since $\vert \vert f_{n} \vert \vert_{L^{p}([-N,N])}\xrightarrow{n \to \infty}\vert \vert f \vert \vert_{L^{p}([-N,N])}$ and $\vert \vert f_{n} \vert \vert_{p}\xrightarrow{n \to \infty}\vert\vert f\vert \vert$ then for $A_{N}:=\mathbb R\setminus[-N,N]$: $\vert \vert f_{n} \vert \vert_{L^{p}(A_{N})}\xrightarrow{n \to \infty}\vert \vert f \vert \vert_{L^{p}(A_{N})}$

This is all logical, but now I see a flaw in my following argument:

Let $\epsilon > 0$. We can thus choose an $N \in \mathbb N$, so that $\int_{A_{N}}\vert f_{n}\vert^{p}d\mu<\epsilon$. I used ths path in my solution, and it was marked as correct, but surely this cannot be true by the paragraph I highlighted above? Am I missing something in this proof?

  • $\begingroup$ Is the inequality supposed to hold for all $n \in \mathbb N$ simultaneously or for some fixed $n$? $\endgroup$ – user159517 Jul 9 at 21:43
  • $\begingroup$ I think you are confusing $f_n$ and $f$. There is no $n$ is $f$ $\endgroup$ – MPW Jul 9 at 22:03

Fix $p\in ]1,\infty]$. Then $f\in L^p(\Bbb R)$ doesn't imply $\lim_{|x|\to\infty} f(x)=0$, but it does imply that $\lim_{N\to\infty} \|f_n |_{A_N}\|_p=0$. But the $L^p$ norms of "tail of the function" does converge to $0$:

Define the sequence $a_N:=\int_0^N |f_n|^p \,\mathrm d\mu$. By definition, this sequence converges and $$\lim_{N\to\infty} a_N = \underbrace{\int_0^\infty |f_n|^p \,\mathrm d\mu}_{\displaystyle=:l}.$$

So (see remark) $$\lim_{N\to\infty} \int_{N}^\infty |f_n|^p \,\mathrm d\mu = \lim_{N\to\infty} \int_{N+1}^\infty |f_n|^p \,\mathrm d\mu = \lim_{N\to\infty} l -a_N = l -\lim_{N\to\infty} a_N = \color{orange}0.$$

By the same argumentation, we have $$\lim_{N\to-\infty} \int_{-\infty}^N |f_n|^p \,\mathrm d\mu= \color{blue}0.$$

It follows that $$\lim_{N\to\infty} \int_{A_N} |f_n|^p \,\mathrm d\mu = \color{orange}0+\color{blue}0=0.$$

Remark. In order to use $\int_N^\infty |f_n|^p \,\mathrm d\mu = \lim_{N\to\infty} l -a_N$, I would first have to prove that $\int_N^\infty |f_n|^p \,\mathrm d\mu$ converges and then use the fact that $a_N+\int_{N+1}^\infty |f_n|^p \,\mathrm d\mu = l$. The proof of convergence is very similar to the proof for series.


For $f \in L^p$, define $(g_N)_{N\geq 1}$ as $$g_N (x) := |f(x)| 1_{\{|f(x)| \leq N\}}.$$ Clearly, it holds that $$\lim_{N\to\infty} g_N(x) = |f(x)|, \quad x \in \mathbb R, $$ so the monotone convergence theorem implies $$\lim_{N\to\infty} \|g_N\|_p = \|f\|_p,$$ which in turn gives $$\lim_{N\to\infty} (\|f\|_p^p - \|g_N\|_p^p) = \lim_{N\to\infty} \int_{A_N} |f(x)|^p dx = 0.$$

To show the "fact" you want to use more directly, consider that $$ \|f_n\|_{L^p (A_N)}^p = \|f_n\|_{L^p (\mathbb R)}^p -\|f_n\|_{L^p (-N,N)}^p \to \|f\|_{L^p (\mathbb R)}^p -\|f\|_{L^p (-N,N)}^p = \|f\|_{L^p (A_N)}^p$$


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.