Please help me to solve the following problem that is in the Lebesgue integral discussion

Give an example of a sequence $\,\,f_n : [0, 1] \to \Bbb R$ of continuous functions such that $\,\,\|f_n\|_\infty \to \infty$ but $\int_0^1\lvert\, f_n\rvert\,d\lambda \to 0$ as $n\to\infty$.

Original screenshot

  • $\begingroup$ Think of straight line segments with big $y$-intercept and very small $x$-intercept. $\endgroup$ – David Mitra Dec 27 '13 at 13:34
  • 1
    $\begingroup$ I converted the image to $\TeX$. Please do it yourself next time. Here is a tutorial. $\endgroup$ – Ayman Hourieh Dec 27 '13 at 13:37

Let $$ f_n(x)=\max\left\{n,n^{-1}x^{-1/2}\right\}. $$

Then $\|\,f_n\|_\infty=n$ while $\|\,f\|_1<\dfrac{1}{2n}$.


Here's another: $f_n(x) = nx^{n^2}$


Below I have given a sufficient condition for your problem to have a positive solution in a rather general measure space, afterwhich I give an example in the case of $[0,1]$ with its usual Euclidean topology to help illustrate my construction.

A general solution:

Proposition: More generally, let $X$ be a locally-compact Hausdorff space, $<X,\mathscr{F},\mu>$ be a non-trivial, regular (ie.: $\mu \neq 0$) $\sigma$-finite Borel meaure space which is not purely atomic. Then for any real number $r$, there exists a sequence of measurable functions $f_n$ satisfying: \begin{equation} f_n\overset{L^{\infty}_m}{\rightarrow} \infty \mbox{ and } f_n\overset{L^{p}_m}{\rightarrow} r , \end{equation} for every $p\in [1,\infty)$.

Proof: Since the measure space is $\sigma$-finite and non-trivial, there exists some $A\in \mathscr{F}$ with $\mu(A)\leq \infty$. Moreover, since every measure which is not purely atomic is the sum of a purely atomic $\mu_a$ and a purely nonatomic (or diffuse) measure $\mu_d$ (see this article), we may without loss of generality assume $\mu$ to be purely nonatomic (by only considering its nonatomic part).

Now consider $A$ as a disjoint union of two sets $B$ and $C$, both of positive $\mu$-measure.

Now therefore, since $B$ must also be diffuse, for every sufficiently large $k\in \mathbb{Z}^+$ there exists $B_k\in \mathscr{F}$ such that $B_{k+1}\subseteq B_k$ and $\mu(B_k)=\frac{1}{k}.

Therefore the sequence of measurable functions: \begin{equation} g_k:=kI_{A_k} + \frac{r}{\mu(C)}I_{C}, \end{equation} must have the desired properties (for large enough $k$).

However, the sequence $g_k$ is not continous but $C_c(X,\mathbb{R})$ is dense in $L_{\mu}^p$, then for every $k \in \mathbb{Z}^+$, there exists some continous function $f_k^p$ with compact support such that $\|f_k^p-g_k\|_{L^p_{mu}} = o_{\mu}( 2^{-k})$.

Therefore $f_k$ converge in $L_{\mu}^p$ to r, and in $L^{\infty}_{\mu}$ to $\infty$, as desired.

A particular case for the Lebesgue measure space

Corollary: In the case of the Lebesgue measure space on $[0,2]^d$ we may take: \begin{equation} B_k:=[0,\frac{1}{\sqrt[d]{k}}]^d \mbox{ and } C:=[1,2]^d \end{equation}

Then \begin{equation} f_k(x):= k\prod_{i=1}^d (1-\frac{x-(\frac{1}{\sqrt[d]{k}}+2^{-k})}{2^{-k}})I_{[0,\frac{1}{\sqrt[d]{k}}]}(x_i) + rI_{[1,2]^d} \end{equation}.

Example: More particularly for $\mathbb{R}$ we may take: \begin{equation} A_k:=[0,\frac{1}{k}] \end{equation}

Then $f_k(x):= k(1-\frac{x-(\frac{1}{k}+2^{-k})}{2^{-k}})I_{[0,\frac{1}{k}]}(x_i)$ solves your problme :).


\begin{align*} f_n(x)=\begin{cases}n&\text{if $x\in\left[0,\frac1 {2n^2}\right]$;}\\ \text{[linear descent; see below]}&\text{if $x\in\left(\frac{1}{2n^2},\frac{1}{n^2}\right]$;}\\0&\text{if $x\in\left(\frac1{n^2},1\right]$.}\end{cases} \end{align*} $\|f_n\|_{\infty}=n,$


The linear descent is needed to make the function continuous. It can be computed to be $x\mapsto-2n^3x+2n$.

[Figure below is not to scale.]

Note: not to scale


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.